Merge airlied/drm-next into drm-misc-next

Backmerge the main pull request to sync up with all the newly landed
drivers. Otherwise we'll have chaos even before 4.12 started in
earnest.

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

148 files changed, 4619 insertions, 3017 deletions

diff --git a/drivers/gpu/drm/nouveau/dispnv04/arb.c b/drivers/gpu/drm/nouveau/dispnv04/arb.c
index a555681c3096..90075b676256 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/arb.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/arb.c
@@ -198,7 +198,7 @@ nv04_update_arb(struct drm_device *dev, int VClk, int bpp,
                 int *burst, int *lwm)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         struct nv_fifo_info fifo_data;
         struct nv_sim_state sim_data;
         int MClk = nouveau_hw_get_clock(dev, PLL_MEMORY);
@@ -227,7 +227,7 @@ nv04_update_arb(struct drm_device *dev, int VClk, int bpp,
                 sim_data.mem_page_miss = ((cfg1 >> 4) & 0xf) + ((cfg1 >> 31) & 0x1);
         }
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_TNT)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_TNT)
                 nv04_calc_arb(&fifo_data, &sim_data);
         else
                 nv10_calc_arb(&fifo_data, &sim_data);
@@ -254,7 +254,7 @@ nouveau_calc_arb(struct drm_device *dev, int vclk, int bpp, int *burst, int *lwm
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
 
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_KELVIN)
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_KELVIN)
                 nv04_update_arb(dev, vclk, bpp, burst, lwm);
         else if ((dev->pdev->device & 0xfff0) == 0x0240 /*CHIPSET_C51*/ ||
                  (dev->pdev->device & 0xfff0) == 0x03d0 /*CHIPSET_C512*/) {
diff --git a/drivers/gpu/drm/nouveau/dispnv04/crtc.c b/drivers/gpu/drm/nouveau/dispnv04/crtc.c
index a72754d73c84..ab7b69c11d40 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/crtc.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/crtc.c
@@ -113,8 +113,8 @@ static void nv_crtc_calc_state_ext(struct drm_crtc *crtc, struct drm_display_mod
 {
         struct drm_device *dev = crtc->dev;
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_bios *bios = nvxx_bios(&drm->device);
-        struct nvkm_clk *clk = nvxx_clk(&drm->device);
+        struct nvkm_bios *bios = nvxx_bios(&drm->client.device);
+        struct nvkm_clk *clk = nvxx_clk(&drm->client.device);
         struct nouveau_crtc *nv_crtc = nouveau_crtc(crtc);
         struct nv04_mode_state *state = &nv04_display(dev)->mode_reg;
         struct nv04_crtc_reg *regp = &state->crtc_reg[nv_crtc->index];
@@ -138,7 +138,7 @@ static void nv_crtc_calc_state_ext(struct drm_crtc *crtc, struct drm_display_mod
          * has yet been observed in allowing the use a single stage pll on all
          * nv43 however.  the behaviour of single stage use is untested on nv40
          */
-        if (drm->device.info.chipset > 0x40 && dot_clock <= (pll_lim.vco1.max_freq / 2))
+        if (drm->client.device.info.chipset > 0x40 && dot_clock <= (pll_lim.vco1.max_freq / 2))
                 memset(&pll_lim.vco2, 0, sizeof(pll_lim.vco2));
 
 
@@ -148,10 +148,10 @@ static void nv_crtc_calc_state_ext(struct drm_crtc *crtc, struct drm_display_mod
         state->pllsel &= PLLSEL_VPLL1_MASK | PLLSEL_VPLL2_MASK | PLLSEL_TV_MASK;
 
         /* The blob uses this always, so let's do the same */
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 state->pllsel |= NV_PRAMDAC_PLL_COEFF_SELECT_USE_VPLL2_TRUE;
         /* again nv40 and some nv43 act more like nv3x as described above */
-        if (drm->device.info.chipset < 0x41)
+        if (drm->client.device.info.chipset < 0x41)
                 state->pllsel |= NV_PRAMDAC_PLL_COEFF_SELECT_SOURCE_PROG_MPLL |
                                  NV_PRAMDAC_PLL_COEFF_SELECT_SOURCE_PROG_NVPLL;
         state->pllsel |= nv_crtc->index ? PLLSEL_VPLL2_MASK : PLLSEL_VPLL1_MASK;
@@ -270,7 +270,7 @@ nv_crtc_mode_set_vga(struct drm_crtc *crtc, struct drm_display_mode *mode)
                 horizEnd = horizTotal - 2;
                 horizBlankEnd = horizTotal + 4;
 #if 0
-                if (dev->overlayAdaptor && drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
+                if (dev->overlayAdaptor && drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
                         /* This reportedly works around some video overlay bandwidth problems */
                         horizTotal += 2;
 #endif
@@ -505,7 +505,7 @@ nv_crtc_mode_set_regs(struct drm_crtc *crtc, struct drm_display_mode * mode)
         regp->cursor_cfg = NV_PCRTC_CURSOR_CONFIG_CUR_LINES_64 |
                              NV_PCRTC_CURSOR_CONFIG_CUR_PIXELS_64 |
                              NV_PCRTC_CURSOR_CONFIG_ADDRESS_SPACE_PNVM;
-        if (drm->device.info.chipset >= 0x11)
+        if (drm->client.device.info.chipset >= 0x11)
                 regp->cursor_cfg |= NV_PCRTC_CURSOR_CONFIG_CUR_BPP_32;
         if (mode->flags & DRM_MODE_FLAG_DBLSCAN)
                 regp->cursor_cfg |= NV_PCRTC_CURSOR_CONFIG_DOUBLE_SCAN_ENABLE;
@@ -546,26 +546,26 @@ nv_crtc_mode_set_regs(struct drm_crtc *crtc, struct drm_display_mode * mode)
          * 1 << 30 on 0x60.830), for no apparent reason */
         regp->CRTC[NV_CIO_CRE_59] = off_chip_digital;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
                 regp->CRTC[0x9f] = off_chip_digital ? 0x11 : 0x1;
 
         regp->crtc_830 = mode->crtc_vdisplay - 3;
         regp->crtc_834 = mode->crtc_vdisplay - 1;
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 /* This is what the blob does */
                 regp->crtc_850 = NVReadCRTC(dev, 0, NV_PCRTC_850);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
                 regp->gpio_ext = NVReadCRTC(dev, 0, NV_PCRTC_GPIO_EXT);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
                 regp->crtc_cfg = NV10_PCRTC_CONFIG_START_ADDRESS_HSYNC;
         else
                 regp->crtc_cfg = NV04_PCRTC_CONFIG_START_ADDRESS_HSYNC;
 
         /* Some misc regs */
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE) {
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE) {
                 regp->CRTC[NV_CIO_CRE_85] = 0xFF;
                 regp->CRTC[NV_CIO_CRE_86] = 0x1;
         }
@@ -577,7 +577,7 @@ nv_crtc_mode_set_regs(struct drm_crtc *crtc, struct drm_display_mode * mode)
 
         /* Generic PRAMDAC regs */
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
                 /* Only bit that bios and blob set. */
                 regp->nv10_cursync = (1 << 25);
 
@@ -586,7 +586,7 @@ nv_crtc_mode_set_regs(struct drm_crtc *crtc, struct drm_display_mode * mode)
                                 NV_PRAMDAC_GENERAL_CONTROL_PIXMIX_ON;
         if (fb->format->depth == 16)
                 regp->ramdac_gen_ctrl |= NV_PRAMDAC_GENERAL_CONTROL_ALT_MODE_SEL;
-        if (drm->device.info.chipset >= 0x11)
+        if (drm->client.device.info.chipset >= 0x11)
                 regp->ramdac_gen_ctrl |= NV_PRAMDAC_GENERAL_CONTROL_PIPE_LONG;
 
         regp->ramdac_630 = 0; /* turn off green mode (tv test pattern?) */
@@ -649,7 +649,7 @@ nv_crtc_mode_set(struct drm_crtc *crtc, struct drm_display_mode *mode,
 
         nv_crtc_mode_set_vga(crtc, adjusted_mode);
         /* calculated in nv04_dfp_prepare, nv40 needs it written before calculating PLLs */
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 NVWriteRAMDAC(dev, 0, NV_PRAMDAC_SEL_CLK, nv04_display(dev)->mode_reg.sel_clk);
         nv_crtc_mode_set_regs(crtc, adjusted_mode);
         nv_crtc_calc_state_ext(crtc, mode, adjusted_mode->clock);
@@ -710,7 +710,7 @@ static void nv_crtc_prepare(struct drm_crtc *crtc)
 
         /* Some more preparation. */
         NVWriteCRTC(dev, nv_crtc->index, NV_PCRTC_CONFIG, NV_PCRTC_CONFIG_START_ADDRESS_NON_VGA);
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE) {
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE) {
                 uint32_t reg900 = NVReadRAMDAC(dev, nv_crtc->index, NV_PRAMDAC_900);
                 NVWriteRAMDAC(dev, nv_crtc->index, NV_PRAMDAC_900, reg900 & ~0x10000);
         }
@@ -886,7 +886,7 @@ nv04_crtc_do_mode_set_base(struct drm_crtc *crtc,
         crtc_wr_cio_state(crtc, regp, NV_CIO_CRE_FF_INDEX);
         crtc_wr_cio_state(crtc, regp, NV_CIO_CRE_FFLWM__INDEX);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_KELVIN) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_KELVIN) {
                 regp->CRTC[NV_CIO_CRE_47] = arb_lwm >> 8;
                 crtc_wr_cio_state(crtc, regp, NV_CIO_CRE_47);
         }
@@ -967,7 +967,7 @@ static void nv11_cursor_upload(struct drm_device *dev, struct nouveau_bo *src,
                 {
                         struct nouveau_drm *drm = nouveau_drm(dev);
 
-                        if (drm->device.info.chipset == 0x11) {
+                        if (drm->client.device.info.chipset == 0x11) {
                                 pixel = ((pixel & 0x000000ff) << 24) |
                                         ((pixel & 0x0000ff00) << 8) |
                                         ((pixel & 0x00ff0000) >> 8) |
@@ -1008,7 +1008,7 @@ nv04_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
         if (ret)
                 goto out;
 
-        if (drm->device.info.chipset >= 0x11)
+        if (drm->client.device.info.chipset >= 0x11)
                 nv11_cursor_upload(dev, cursor, nv_crtc->cursor.nvbo);
         else
                 nv04_cursor_upload(dev, cursor, nv_crtc->cursor.nvbo);
@@ -1124,8 +1124,9 @@ nv04_crtc_create(struct drm_device *dev, int crtc_num)
         drm_crtc_helper_add(&nv_crtc->base, &nv04_crtc_helper_funcs);
         drm_mode_crtc_set_gamma_size(&nv_crtc->base, 256);
 
-        ret = nouveau_bo_new(dev, 64*64*4, 0x100, TTM_PL_FLAG_VRAM,
-                             0, 0x0000, NULL, NULL, &nv_crtc->cursor.nvbo);
+        ret = nouveau_bo_new(&nouveau_drm(dev)->client, 64*64*4, 0x100,
+                             TTM_PL_FLAG_VRAM, 0, 0x0000, NULL, NULL,
+                             &nv_crtc->cursor.nvbo);
         if (!ret) {
                 ret = nouveau_bo_pin(nv_crtc->cursor.nvbo, TTM_PL_FLAG_VRAM, false);
                 if (!ret) {
diff --git a/drivers/gpu/drm/nouveau/dispnv04/cursor.c b/drivers/gpu/drm/nouveau/dispnv04/cursor.c
index c83116a308a4..f26e44ea7389 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/cursor.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/cursor.c
@@ -55,7 +55,7 @@ nv04_cursor_set_offset(struct nouveau_crtc *nv_crtc, uint32_t offset)
         crtc_wr_cio_state(crtc, regp, NV_CIO_CRE_HCUR_ADDR0_INDEX);
         crtc_wr_cio_state(crtc, regp, NV_CIO_CRE_HCUR_ADDR1_INDEX);
         crtc_wr_cio_state(crtc, regp, NV_CIO_CRE_HCUR_ADDR2_INDEX);
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 nv_fix_nv40_hw_cursor(dev, nv_crtc->index);
 }
 
diff --git a/drivers/gpu/drm/nouveau/dispnv04/dac.c b/drivers/gpu/drm/nouveau/dispnv04/dac.c
index b6cc7766e6f7..4feab0a5419d 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/dac.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/dac.c
@@ -66,7 +66,7 @@ int nv04_dac_output_offset(struct drm_encoder *encoder)
 static int sample_load_twice(struct drm_device *dev, bool sense[2])
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         int i;
 
         for (i = 0; i < 2; i++) {
@@ -80,19 +80,19 @@ static int sample_load_twice(struct drm_device *dev, bool sense[2])
                  * use a 10ms timeout (guards against crtc being inactive, in
                  * which case blank state would never change)
                  */
-                if (nvif_msec(&drm->device, 10,
+                if (nvif_msec(&drm->client.device, 10,
                         if (!(nvif_rd32(device, NV_PRMCIO_INP0__COLOR) & 1))
                                 break;
                 ) < 0)
                         return -EBUSY;
 
-                if (nvif_msec(&drm->device, 10,
+                if (nvif_msec(&drm->client.device, 10,
                         if ( (nvif_rd32(device, NV_PRMCIO_INP0__COLOR) & 1))
                                 break;
                 ) < 0)
                         return -EBUSY;
 
-                if (nvif_msec(&drm->device, 10,
+                if (nvif_msec(&drm->client.device, 10,
                         if (!(nvif_rd32(device, NV_PRMCIO_INP0__COLOR) & 1))
                                 break;
                 ) < 0)
@@ -133,7 +133,7 @@ static enum drm_connector_status nv04_dac_detect(struct drm_encoder *encoder,
                                                  struct drm_connector *connector)
 {
         struct drm_device *dev = encoder->dev;
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         struct nouveau_drm *drm = nouveau_drm(dev);
         uint8_t saved_seq1, saved_pi, saved_rpc1, saved_cr_mode;
         uint8_t saved_palette0[3], saved_palette_mask;
@@ -236,8 +236,8 @@ uint32_t nv17_dac_sample_load(struct drm_encoder *encoder)
 {
         struct drm_device *dev = encoder->dev;
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
-        struct nvkm_gpio *gpio = nvxx_gpio(&drm->device);
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
+        struct nvkm_gpio *gpio = nvxx_gpio(&drm->client.device);
         struct dcb_output *dcb = nouveau_encoder(encoder)->dcb;
         uint32_t sample, testval, regoffset = nv04_dac_output_offset(encoder);
         uint32_t saved_powerctrl_2 = 0, saved_powerctrl_4 = 0, saved_routput,
@@ -288,7 +288,7 @@ uint32_t nv17_dac_sample_load(struct drm_encoder *encoder)
         /* nv driver and nv31 use 0xfffffeee, nv34 and 6600 use 0xfffffece */
         routput = (saved_routput & 0xfffffece) | head << 8;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CURIE) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CURIE) {
                 if (dcb->type == DCB_OUTPUT_TV)
                         routput |= 0x1a << 16;
                 else
@@ -403,7 +403,7 @@ static void nv04_dac_mode_set(struct drm_encoder *encoder,
         }
 
         /* This could use refinement for flatpanels, but it should work this way */
-        if (drm->device.info.chipset < 0x44)
+        if (drm->client.device.info.chipset < 0x44)
                 NVWriteRAMDAC(dev, 0, NV_PRAMDAC_TEST_CONTROL + nv04_dac_output_offset(encoder), 0xf0000000);
         else
                 NVWriteRAMDAC(dev, 0, NV_PRAMDAC_TEST_CONTROL + nv04_dac_output_offset(encoder), 0x00100000);
diff --git a/drivers/gpu/drm/nouveau/dispnv04/dfp.c b/drivers/gpu/drm/nouveau/dispnv04/dfp.c
index 2e5bb2afda7c..9805d2cdc1a1 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/dfp.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/dfp.c
@@ -281,7 +281,7 @@ static void nv04_dfp_mode_set(struct drm_encoder *encoder,
                               struct drm_display_mode *adjusted_mode)
 {
         struct drm_device *dev = encoder->dev;
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         struct nouveau_drm *drm = nouveau_drm(dev);
         struct nouveau_crtc *nv_crtc = nouveau_crtc(encoder->crtc);
         struct nv04_crtc_reg *regp = &nv04_display(dev)->mode_reg.crtc_reg[nv_crtc->index];
@@ -417,7 +417,7 @@ static void nv04_dfp_mode_set(struct drm_encoder *encoder,
         if ((nv_connector->dithering_mode == DITHERING_MODE_ON) ||
             (nv_connector->dithering_mode == DITHERING_MODE_AUTO &&
              fb->format->depth > connector->display_info.bpc * 3)) {
-                if (drm->device.info.chipset == 0x11)
+                if (drm->client.device.info.chipset == 0x11)
                         regp->dither = savep->dither | 0x00010000;
                 else {
                         int i;
@@ -428,7 +428,7 @@ static void nv04_dfp_mode_set(struct drm_encoder *encoder,
                         }
                 }
         } else {
-                if (drm->device.info.chipset != 0x11) {
+                if (drm->client.device.info.chipset != 0x11) {
                         /* reset them */
                         int i;
                         for (i = 0; i < 3; i++) {
@@ -464,7 +464,7 @@ static void nv04_dfp_commit(struct drm_encoder *encoder)
                 NVReadRAMDAC(dev, head, NV_PRAMDAC_FP_TG_CONTROL);
 
         /* This could use refinement for flatpanels, but it should work this way */
-        if (drm->device.info.chipset < 0x44)
+        if (drm->client.device.info.chipset < 0x44)
                 NVWriteRAMDAC(dev, 0, NV_PRAMDAC_TEST_CONTROL + nv04_dac_output_offset(encoder), 0xf0000000);
         else
                 NVWriteRAMDAC(dev, 0, NV_PRAMDAC_TEST_CONTROL + nv04_dac_output_offset(encoder), 0x00100000);
@@ -486,7 +486,7 @@ static void nv04_dfp_update_backlight(struct drm_encoder *encoder, int mode)
 {
 #ifdef __powerpc__
         struct drm_device *dev = encoder->dev;
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
 
         /* BIOS scripts usually take care of the backlight, thanks
          * Apple for your consistency.
@@ -624,7 +624,7 @@ static void nv04_tmds_slave_init(struct drm_encoder *encoder)
         struct drm_device *dev = encoder->dev;
         struct dcb_output *dcb = nouveau_encoder(encoder)->dcb;
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_i2c *i2c = nvxx_i2c(&drm->device);
+        struct nvkm_i2c *i2c = nvxx_i2c(&drm->client.device);
         struct nvkm_i2c_bus *bus = nvkm_i2c_bus_find(i2c, NVKM_I2C_BUS_PRI);
         struct nvkm_i2c_bus_probe info[] = {
                 {
diff --git a/drivers/gpu/drm/nouveau/dispnv04/disp.c b/drivers/gpu/drm/nouveau/dispnv04/disp.c
index 34c0f2f67548..5b9d549aa791 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/disp.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/disp.c
@@ -35,7 +35,7 @@ int
 nv04_display_create(struct drm_device *dev)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_i2c *i2c = nvxx_i2c(&drm->device);
+        struct nvkm_i2c *i2c = nvxx_i2c(&drm->client.device);
         struct dcb_table *dcb = &drm->vbios.dcb;
         struct drm_connector *connector, *ct;
         struct drm_encoder *encoder;
@@ -48,7 +48,7 @@ nv04_display_create(struct drm_device *dev)
         if (!disp)
                 return -ENOMEM;
 
-        nvif_object_map(&drm->device.object);
+        nvif_object_map(&drm->client.device.object);
 
         nouveau_display(dev)->priv = disp;
         nouveau_display(dev)->dtor = nv04_display_destroy;
@@ -139,7 +139,7 @@ nv04_display_destroy(struct drm_device *dev)
         nouveau_display(dev)->priv = NULL;
         kfree(disp);
 
-        nvif_object_unmap(&drm->device.object);
+        nvif_object_unmap(&drm->client.device.object);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/dispnv04/disp.h b/drivers/gpu/drm/nouveau/dispnv04/disp.h
index 7030307d2d48..bea4543554ba 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/disp.h
+++ b/drivers/gpu/drm/nouveau/dispnv04/disp.h
@@ -129,7 +129,7 @@ nv_two_heads(struct drm_device *dev)
         struct nouveau_drm *drm = nouveau_drm(dev);
         const int impl = dev->pdev->device & 0x0ff0;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS && impl != 0x0100 &&
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS && impl != 0x0100 &&
             impl != 0x0150 && impl != 0x01a0 && impl != 0x0200)
                 return true;
 
@@ -148,7 +148,7 @@ nv_two_reg_pll(struct drm_device *dev)
         struct nouveau_drm *drm = nouveau_drm(dev);
         const int impl = dev->pdev->device & 0x0ff0;
 
-        if (impl == 0x0310 || impl == 0x0340 || drm->device.info.family >= NV_DEVICE_INFO_V0_CURIE)
+        if (impl == 0x0310 || impl == 0x0340 || drm->client.device.info.family >= NV_DEVICE_INFO_V0_CURIE)
                 return true;
         return false;
 }
@@ -170,7 +170,7 @@ nouveau_bios_run_init_table(struct drm_device *dev, u16 table,
                             struct dcb_output *outp, int crtc)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_bios *bios = nvxx_bios(&drm->device);
+        struct nvkm_bios *bios = nvxx_bios(&drm->client.device);
         struct nvbios_init init = {
                 .subdev = &bios->subdev,
                 .bios = bios,
diff --git a/drivers/gpu/drm/nouveau/dispnv04/hw.c b/drivers/gpu/drm/nouveau/dispnv04/hw.c
index 74856a8b8f35..b98599002831 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/hw.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/hw.c
@@ -89,7 +89,7 @@ NVSetOwner(struct drm_device *dev, int owner)
         if (owner == 1)
                 owner *= 3;
 
-        if (drm->device.info.chipset == 0x11) {
+        if (drm->client.device.info.chipset == 0x11) {
                 /* This might seem stupid, but the blob does it and
                  * omitting it often locks the system up.
                  */
@@ -100,7 +100,7 @@ NVSetOwner(struct drm_device *dev, int owner)
         /* CR44 is always changed on CRTC0 */
         NVWriteVgaCrtc(dev, 0, NV_CIO_CRE_44, owner);
 
-        if (drm->device.info.chipset == 0x11) { /* set me harder */
+        if (drm->client.device.info.chipset == 0x11) {  /* set me harder */
                 NVWriteVgaCrtc(dev, 0, NV_CIO_CRE_2E, owner);
                 NVWriteVgaCrtc(dev, 0, NV_CIO_CRE_2E, owner);
         }
@@ -149,7 +149,7 @@ nouveau_hw_decode_pll(struct drm_device *dev, uint32_t reg1, uint32_t pll1,
                 pllvals->NM1 = pll1 & 0xffff;
                 if (nv_two_reg_pll(dev) && pll2 & NV31_RAMDAC_ENABLE_VCO2)
                         pllvals->NM2 = pll2 & 0xffff;
-                else if (drm->device.info.chipset == 0x30 || drm->device.info.chipset == 0x35) {
+                else if (drm->client.device.info.chipset == 0x30 || drm->client.device.info.chipset == 0x35) {
                         pllvals->M1 &= 0xf; /* only 4 bits */
                         if (pll1 & NV30_RAMDAC_ENABLE_VCO2) {
                                 pllvals->M2 = (pll1 >> 4) & 0x7;
@@ -165,8 +165,8 @@ nouveau_hw_get_pllvals(struct drm_device *dev, enum nvbios_pll_type plltype,
                        struct nvkm_pll_vals *pllvals)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &drm->device.object;
-        struct nvkm_bios *bios = nvxx_bios(&drm->device);
+        struct nvif_object *device = &drm->client.device.object;
+        struct nvkm_bios *bios = nvxx_bios(&drm->client.device);
         uint32_t reg1, pll1, pll2 = 0;
         struct nvbios_pll pll_lim;
         int ret;
@@ -184,7 +184,7 @@ nouveau_hw_get_pllvals(struct drm_device *dev, enum nvbios_pll_type plltype,
                 pll2 = nvif_rd32(device, reg2);
         }
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CELSIUS && reg1 >= NV_PRAMDAC_VPLL_COEFF) {
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CELSIUS && reg1 >= NV_PRAMDAC_VPLL_COEFF) {
                 uint32_t ramdac580 = NVReadRAMDAC(dev, 0, NV_PRAMDAC_580);
 
                 /* check whether vpll has been forced into single stage mode */
@@ -222,6 +222,7 @@ nouveau_hw_get_clock(struct drm_device *dev, enum nvbios_pll_type plltype)
                 uint32_t mpllP;
 
                 pci_read_config_dword(pci_get_bus_and_slot(0, 3), 0x6c, &mpllP);
+                mpllP = (mpllP >> 8) & 0xf;
                 if (!mpllP)
                         mpllP = 4;
 
@@ -232,7 +233,7 @@ nouveau_hw_get_clock(struct drm_device *dev, enum nvbios_pll_type plltype)
                 uint32_t clock;
 
                 pci_read_config_dword(pci_get_bus_and_slot(0, 5), 0x4c, &clock);
-                return clock;
+                return clock / 1000;
         }
 
         ret = nouveau_hw_get_pllvals(dev, plltype, &pllvals);
@@ -252,7 +253,7 @@ nouveau_hw_fix_bad_vpll(struct drm_device *dev, int head)
          */
 
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         struct nvkm_clk *clk = nvxx_clk(device);
         struct nvkm_bios *bios = nvxx_bios(device);
         struct nvbios_pll pll_lim;
@@ -391,21 +392,21 @@ nv_save_state_ramdac(struct drm_device *dev, int head,
         struct nv04_crtc_reg *regp = &state->crtc_reg[head];
         int i;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
                 regp->nv10_cursync = NVReadRAMDAC(dev, head, NV_RAMDAC_NV10_CURSYNC);
 
         nouveau_hw_get_pllvals(dev, head ? PLL_VPLL1 : PLL_VPLL0, &regp->pllvals);
         state->pllsel = NVReadRAMDAC(dev, 0, NV_PRAMDAC_PLL_COEFF_SELECT);
         if (nv_two_heads(dev))
                 state->sel_clk = NVReadRAMDAC(dev, 0, NV_PRAMDAC_SEL_CLK);
-        if (drm->device.info.chipset == 0x11)
+        if (drm->client.device.info.chipset == 0x11)
                 regp->dither = NVReadRAMDAC(dev, head, NV_RAMDAC_DITHER_NV11);
 
         regp->ramdac_gen_ctrl = NVReadRAMDAC(dev, head, NV_PRAMDAC_GENERAL_CONTROL);
 
         if (nv_gf4_disp_arch(dev))
                 regp->ramdac_630 = NVReadRAMDAC(dev, head, NV_PRAMDAC_630);
-        if (drm->device.info.chipset >= 0x30)
+        if (drm->client.device.info.chipset >= 0x30)
                 regp->ramdac_634 = NVReadRAMDAC(dev, head, NV_PRAMDAC_634);
 
         regp->tv_setup = NVReadRAMDAC(dev, head, NV_PRAMDAC_TV_SETUP);
@@ -447,7 +448,7 @@ nv_save_state_ramdac(struct drm_device *dev, int head,
         if (nv_gf4_disp_arch(dev))
                 regp->ramdac_8c0 = NVReadRAMDAC(dev, head, NV_PRAMDAC_8C0);
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE) {
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE) {
                 regp->ramdac_a20 = NVReadRAMDAC(dev, head, NV_PRAMDAC_A20);
                 regp->ramdac_a24 = NVReadRAMDAC(dev, head, NV_PRAMDAC_A24);
                 regp->ramdac_a34 = NVReadRAMDAC(dev, head, NV_PRAMDAC_A34);
@@ -463,26 +464,26 @@ nv_load_state_ramdac(struct drm_device *dev, int head,
                      struct nv04_mode_state *state)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_clk *clk = nvxx_clk(&drm->device);
+        struct nvkm_clk *clk = nvxx_clk(&drm->client.device);
         struct nv04_crtc_reg *regp = &state->crtc_reg[head];
         uint32_t pllreg = head ? NV_RAMDAC_VPLL2 : NV_PRAMDAC_VPLL_COEFF;
         int i;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS)
                 NVWriteRAMDAC(dev, head, NV_RAMDAC_NV10_CURSYNC, regp->nv10_cursync);
 
         clk->pll_prog(clk, pllreg, &regp->pllvals);
         NVWriteRAMDAC(dev, 0, NV_PRAMDAC_PLL_COEFF_SELECT, state->pllsel);
         if (nv_two_heads(dev))
                 NVWriteRAMDAC(dev, 0, NV_PRAMDAC_SEL_CLK, state->sel_clk);
-        if (drm->device.info.chipset == 0x11)
+        if (drm->client.device.info.chipset == 0x11)
                 NVWriteRAMDAC(dev, head, NV_RAMDAC_DITHER_NV11, regp->dither);
 
         NVWriteRAMDAC(dev, head, NV_PRAMDAC_GENERAL_CONTROL, regp->ramdac_gen_ctrl);
 
         if (nv_gf4_disp_arch(dev))
                 NVWriteRAMDAC(dev, head, NV_PRAMDAC_630, regp->ramdac_630);
-        if (drm->device.info.chipset >= 0x30)
+        if (drm->client.device.info.chipset >= 0x30)
                 NVWriteRAMDAC(dev, head, NV_PRAMDAC_634, regp->ramdac_634);
 
         NVWriteRAMDAC(dev, head, NV_PRAMDAC_TV_SETUP, regp->tv_setup);
@@ -519,7 +520,7 @@ nv_load_state_ramdac(struct drm_device *dev, int head,
         if (nv_gf4_disp_arch(dev))
                 NVWriteRAMDAC(dev, head, NV_PRAMDAC_8C0, regp->ramdac_8c0);
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE) {
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE) {
                 NVWriteRAMDAC(dev, head, NV_PRAMDAC_A20, regp->ramdac_a20);
                 NVWriteRAMDAC(dev, head, NV_PRAMDAC_A24, regp->ramdac_a24);
                 NVWriteRAMDAC(dev, head, NV_PRAMDAC_A34, regp->ramdac_a34);
@@ -600,10 +601,10 @@ nv_save_state_ext(struct drm_device *dev, int head,
         rd_cio_state(dev, head, regp, NV_CIO_CRE_FFLWM__INDEX);
         rd_cio_state(dev, head, regp, NV_CIO_CRE_21);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_KELVIN)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_KELVIN)
                 rd_cio_state(dev, head, regp, NV_CIO_CRE_47);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
                 rd_cio_state(dev, head, regp, 0x9f);
 
         rd_cio_state(dev, head, regp, NV_CIO_CRE_49);
@@ -612,14 +613,14 @@ nv_save_state_ext(struct drm_device *dev, int head,
         rd_cio_state(dev, head, regp, NV_CIO_CRE_HCUR_ADDR2_INDEX);
         rd_cio_state(dev, head, regp, NV_CIO_CRE_ILACE__INDEX);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
                 regp->crtc_830 = NVReadCRTC(dev, head, NV_PCRTC_830);
                 regp->crtc_834 = NVReadCRTC(dev, head, NV_PCRTC_834);
 
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
                         regp->gpio_ext = NVReadCRTC(dev, head, NV_PCRTC_GPIO_EXT);
 
-                if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+                if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                         regp->crtc_850 = NVReadCRTC(dev, head, NV_PCRTC_850);
 
                 if (nv_two_heads(dev))
@@ -631,7 +632,7 @@ nv_save_state_ext(struct drm_device *dev, int head,
 
         rd_cio_state(dev, head, regp, NV_CIO_CRE_SCRATCH3__INDEX);
         rd_cio_state(dev, head, regp, NV_CIO_CRE_SCRATCH4__INDEX);
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
                 rd_cio_state(dev, head, regp, NV_CIO_CRE_EBR_INDEX);
                 rd_cio_state(dev, head, regp, NV_CIO_CRE_CSB);
                 rd_cio_state(dev, head, regp, NV_CIO_CRE_4B);
@@ -660,12 +661,12 @@ nv_load_state_ext(struct drm_device *dev, int head,
                   struct nv04_mode_state *state)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         struct nv04_crtc_reg *regp = &state->crtc_reg[head];
         uint32_t reg900;
         int i;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
                 if (nv_two_heads(dev))
                         /* setting ENGINE_CTRL (EC) *must* come before
                          * CIO_CRE_LCD, as writing CRE_LCD sets bits 16 & 17 in
@@ -677,20 +678,20 @@ nv_load_state_ext(struct drm_device *dev, int head,
                 nvif_wr32(device, NV_PVIDEO_INTR_EN, 0);
                 nvif_wr32(device, NV_PVIDEO_OFFSET_BUFF(0), 0);
                 nvif_wr32(device, NV_PVIDEO_OFFSET_BUFF(1), 0);
-                nvif_wr32(device, NV_PVIDEO_LIMIT(0), drm->device.info.ram_size - 1);
-                nvif_wr32(device, NV_PVIDEO_LIMIT(1), drm->device.info.ram_size - 1);
-                nvif_wr32(device, NV_PVIDEO_UVPLANE_LIMIT(0), drm->device.info.ram_size - 1);
-                nvif_wr32(device, NV_PVIDEO_UVPLANE_LIMIT(1), drm->device.info.ram_size - 1);
+                nvif_wr32(device, NV_PVIDEO_LIMIT(0), drm->client.device.info.ram_size - 1);
+                nvif_wr32(device, NV_PVIDEO_LIMIT(1), drm->client.device.info.ram_size - 1);
+                nvif_wr32(device, NV_PVIDEO_UVPLANE_LIMIT(0), drm->client.device.info.ram_size - 1);
+                nvif_wr32(device, NV_PVIDEO_UVPLANE_LIMIT(1), drm->client.device.info.ram_size - 1);
                 nvif_wr32(device, NV_PBUS_POWERCTRL_2, 0);
 
                 NVWriteCRTC(dev, head, NV_PCRTC_CURSOR_CONFIG, regp->cursor_cfg);
                 NVWriteCRTC(dev, head, NV_PCRTC_830, regp->crtc_830);
                 NVWriteCRTC(dev, head, NV_PCRTC_834, regp->crtc_834);
 
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
                         NVWriteCRTC(dev, head, NV_PCRTC_GPIO_EXT, regp->gpio_ext);
 
-                if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE) {
+                if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE) {
                         NVWriteCRTC(dev, head, NV_PCRTC_850, regp->crtc_850);
 
                         reg900 = NVReadRAMDAC(dev, head, NV_PRAMDAC_900);
@@ -713,23 +714,23 @@ nv_load_state_ext(struct drm_device *dev, int head,
         wr_cio_state(dev, head, regp, NV_CIO_CRE_FF_INDEX);
         wr_cio_state(dev, head, regp, NV_CIO_CRE_FFLWM__INDEX);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_KELVIN)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_KELVIN)
                 wr_cio_state(dev, head, regp, NV_CIO_CRE_47);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_RANKINE)
                 wr_cio_state(dev, head, regp, 0x9f);
 
         wr_cio_state(dev, head, regp, NV_CIO_CRE_49);
         wr_cio_state(dev, head, regp, NV_CIO_CRE_HCUR_ADDR0_INDEX);
         wr_cio_state(dev, head, regp, NV_CIO_CRE_HCUR_ADDR1_INDEX);
         wr_cio_state(dev, head, regp, NV_CIO_CRE_HCUR_ADDR2_INDEX);
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 nv_fix_nv40_hw_cursor(dev, head);
         wr_cio_state(dev, head, regp, NV_CIO_CRE_ILACE__INDEX);
 
         wr_cio_state(dev, head, regp, NV_CIO_CRE_SCRATCH3__INDEX);
         wr_cio_state(dev, head, regp, NV_CIO_CRE_SCRATCH4__INDEX);
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
                 wr_cio_state(dev, head, regp, NV_CIO_CRE_EBR_INDEX);
                 wr_cio_state(dev, head, regp, NV_CIO_CRE_CSB);
                 wr_cio_state(dev, head, regp, NV_CIO_CRE_4B);
@@ -737,14 +738,14 @@ nv_load_state_ext(struct drm_device *dev, int head,
         }
         /* NV11 and NV20 stop at 0x52. */
         if (nv_gf4_disp_arch(dev)) {
-                if (drm->device.info.family < NV_DEVICE_INFO_V0_KELVIN) {
+                if (drm->client.device.info.family < NV_DEVICE_INFO_V0_KELVIN) {
                         /* Not waiting for vertical retrace before modifying
                            CRE_53/CRE_54 causes lockups. */
-                        nvif_msec(&drm->device, 650,
+                        nvif_msec(&drm->client.device, 650,
                                 if ( (nvif_rd32(device, NV_PRMCIO_INP0__COLOR) & 8))
                                         break;
                         );
-                        nvif_msec(&drm->device, 650,
+                        nvif_msec(&drm->client.device, 650,
                                 if (!(nvif_rd32(device, NV_PRMCIO_INP0__COLOR) & 8))
                                         break;
                         );
@@ -770,7 +771,7 @@ static void
 nv_save_state_palette(struct drm_device *dev, int head,
                       struct nv04_mode_state *state)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         int head_offset = head * NV_PRMDIO_SIZE, i;
 
         nvif_wr08(device, NV_PRMDIO_PIXEL_MASK + head_offset,
@@ -789,7 +790,7 @@ void
 nouveau_hw_load_state_palette(struct drm_device *dev, int head,
                               struct nv04_mode_state *state)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         int head_offset = head * NV_PRMDIO_SIZE, i;
 
         nvif_wr08(device, NV_PRMDIO_PIXEL_MASK + head_offset,
@@ -809,7 +810,7 @@ void nouveau_hw_save_state(struct drm_device *dev, int head,
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
 
-        if (drm->device.info.chipset == 0x11)
+        if (drm->client.device.info.chipset == 0x11)
                 /* NB: no attempt is made to restore the bad pll later on */
                 nouveau_hw_fix_bad_vpll(dev, head);
         nv_save_state_ramdac(dev, head, state);
diff --git a/drivers/gpu/drm/nouveau/dispnv04/hw.h b/drivers/gpu/drm/nouveau/dispnv04/hw.h
index 3bded60c5596..3a2be47fb4f1 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/hw.h
+++ b/drivers/gpu/drm/nouveau/dispnv04/hw.h
@@ -60,7 +60,7 @@ extern void nouveau_calc_arb(struct drm_device *, int vclk, int bpp,
 static inline uint32_t NVReadCRTC(struct drm_device *dev,
                                         int head, uint32_t reg)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         uint32_t val;
         if (head)
                 reg += NV_PCRTC0_SIZE;
@@ -71,7 +71,7 @@ static inline uint32_t NVReadCRTC(struct drm_device *dev,
 static inline void NVWriteCRTC(struct drm_device *dev,
                                         int head, uint32_t reg, uint32_t val)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         if (head)
                 reg += NV_PCRTC0_SIZE;
         nvif_wr32(device, reg, val);
@@ -80,7 +80,7 @@ static inline void NVWriteCRTC(struct drm_device *dev,
 static inline uint32_t NVReadRAMDAC(struct drm_device *dev,
                                         int head, uint32_t reg)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         uint32_t val;
         if (head)
                 reg += NV_PRAMDAC0_SIZE;
@@ -91,7 +91,7 @@ static inline uint32_t NVReadRAMDAC(struct drm_device *dev,
 static inline void NVWriteRAMDAC(struct drm_device *dev,
                                         int head, uint32_t reg, uint32_t val)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         if (head)
                 reg += NV_PRAMDAC0_SIZE;
         nvif_wr32(device, reg, val);
@@ -120,7 +120,7 @@ static inline void nv_write_tmds(struct drm_device *dev,
 static inline void NVWriteVgaCrtc(struct drm_device *dev,
                                         int head, uint8_t index, uint8_t value)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         nvif_wr08(device, NV_PRMCIO_CRX__COLOR + head * NV_PRMCIO_SIZE, index);
         nvif_wr08(device, NV_PRMCIO_CR__COLOR + head * NV_PRMCIO_SIZE, value);
 }
@@ -128,7 +128,7 @@ static inline void NVWriteVgaCrtc(struct drm_device *dev,
 static inline uint8_t NVReadVgaCrtc(struct drm_device *dev,
                                         int head, uint8_t index)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         uint8_t val;
         nvif_wr08(device, NV_PRMCIO_CRX__COLOR + head * NV_PRMCIO_SIZE, index);
         val = nvif_rd08(device, NV_PRMCIO_CR__COLOR + head * NV_PRMCIO_SIZE);
@@ -165,13 +165,13 @@ static inline uint8_t NVReadVgaCrtc5758(struct drm_device *dev, int head, uint8_
 static inline uint8_t NVReadPRMVIO(struct drm_device *dev,
                                         int head, uint32_t reg)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         struct nouveau_drm *drm = nouveau_drm(dev);
         uint8_t val;
 
         /* Only NV4x have two pvio ranges; other twoHeads cards MUST call
          * NVSetOwner for the relevant head to be programmed */
-        if (head && drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (head && drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 reg += NV_PRMVIO_SIZE;
 
         val = nvif_rd08(device, reg);
@@ -181,12 +181,12 @@ static inline uint8_t NVReadPRMVIO(struct drm_device *dev,
 static inline void NVWritePRMVIO(struct drm_device *dev,
                                         int head, uint32_t reg, uint8_t value)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         struct nouveau_drm *drm = nouveau_drm(dev);
 
         /* Only NV4x have two pvio ranges; other twoHeads cards MUST call
          * NVSetOwner for the relevant head to be programmed */
-        if (head && drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (head && drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 reg += NV_PRMVIO_SIZE;
 
         nvif_wr08(device, reg, value);
@@ -194,14 +194,14 @@ static inline void NVWritePRMVIO(struct drm_device *dev,
 
 static inline void NVSetEnablePalette(struct drm_device *dev, int head, bool enable)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         nvif_rd08(device, NV_PRMCIO_INP0__COLOR + head * NV_PRMCIO_SIZE);
         nvif_wr08(device, NV_PRMCIO_ARX + head * NV_PRMCIO_SIZE, enable ? 0 : 0x20);
 }
 
 static inline bool NVGetEnablePalette(struct drm_device *dev, int head)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         nvif_rd08(device, NV_PRMCIO_INP0__COLOR + head * NV_PRMCIO_SIZE);
         return !(nvif_rd08(device, NV_PRMCIO_ARX + head * NV_PRMCIO_SIZE) & 0x20);
 }
@@ -209,7 +209,7 @@ static inline bool NVGetEnablePalette(struct drm_device *dev, int head)
 static inline void NVWriteVgaAttr(struct drm_device *dev,
                                         int head, uint8_t index, uint8_t value)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         if (NVGetEnablePalette(dev, head))
                 index &= ~0x20;
         else
@@ -223,7 +223,7 @@ static inline void NVWriteVgaAttr(struct drm_device *dev,
 static inline uint8_t NVReadVgaAttr(struct drm_device *dev,
                                         int head, uint8_t index)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         uint8_t val;
         if (NVGetEnablePalette(dev, head))
                 index &= ~0x20;
@@ -259,10 +259,10 @@ static inline void NVVgaProtect(struct drm_device *dev, int head, bool protect)
 static inline bool
 nv_heads_tied(struct drm_device *dev)
 {
-        struct nvif_object *device = &nouveau_drm(dev)->device.object;
+        struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
         struct nouveau_drm *drm = nouveau_drm(dev);
 
-        if (drm->device.info.chipset == 0x11)
+        if (drm->client.device.info.chipset == 0x11)
                 return !!(nvif_rd32(device, NV_PBUS_DEBUG_1) & (1 << 28));
 
         return NVReadVgaCrtc(dev, 0, NV_CIO_CRE_44) & 0x4;
@@ -318,7 +318,7 @@ NVLockVgaCrtcs(struct drm_device *dev, bool lock)
         NVWriteVgaCrtc(dev, 0, NV_CIO_SR_LOCK_INDEX,
                        lock ? NV_CIO_SR_LOCK_VALUE : NV_CIO_SR_UNLOCK_RW_VALUE);
         /* NV11 has independently lockable extended crtcs, except when tied */
-        if (drm->device.info.chipset == 0x11 && !nv_heads_tied(dev))
+        if (drm->client.device.info.chipset == 0x11 && !nv_heads_tied(dev))
                 NVWriteVgaCrtc(dev, 1, NV_CIO_SR_LOCK_INDEX,
                                lock ? NV_CIO_SR_LOCK_VALUE :
                                       NV_CIO_SR_UNLOCK_RW_VALUE);
@@ -335,7 +335,7 @@ static inline int nv_cursor_width(struct drm_device *dev)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
 
-        return drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS ? NV10_CURSOR_SIZE : NV04_CURSOR_SIZE;
+        return drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS ? NV10_CURSOR_SIZE : NV04_CURSOR_SIZE;
 }
 
 static inline void
@@ -357,7 +357,7 @@ nv_set_crtc_base(struct drm_device *dev, int head, uint32_t offset)
 
         NVWriteCRTC(dev, head, NV_PCRTC_START, offset);
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_TNT) {
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_TNT) {
                 /*
                  * Hilarious, the 24th bit doesn't want to stick to
                  * PCRTC_START...
@@ -382,7 +382,7 @@ nv_show_cursor(struct drm_device *dev, int head, bool show)
                 *curctl1 &= ~MASK(NV_CIO_CRE_HCUR_ADDR1_ENABLE);
         NVWriteVgaCrtc(dev, head, NV_CIO_CRE_HCUR_ADDR1_INDEX, *curctl1);
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 nv_fix_nv40_hw_cursor(dev, head);
 }
 
@@ -398,7 +398,7 @@ nv_pitch_align(struct drm_device *dev, uint32_t width, int bpp)
                 bpp = 8;
 
         /* Alignment requirements taken from the Haiku driver */
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_TNT)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_TNT)
                 mask = 128 / bpp - 1;
         else
                 mask = 512 / bpp - 1;
diff --git a/drivers/gpu/drm/nouveau/dispnv04/overlay.c b/drivers/gpu/drm/nouveau/dispnv04/overlay.c
index 6275c270df25..5319f2a7f24d 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/overlay.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/overlay.c
@@ -97,7 +97,7 @@ nv10_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
                   uint32_t src_w, uint32_t src_h)
 {
         struct nouveau_drm *drm = nouveau_drm(plane->dev);
-        struct nvif_object *dev = &drm->device.object;
+        struct nvif_object *dev = &drm->client.device.object;
         struct nouveau_plane *nv_plane =
                 container_of(plane, struct nouveau_plane, base);
         struct nouveau_framebuffer *nv_fb = nouveau_framebuffer(fb);
@@ -119,7 +119,7 @@ nv10_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
         if (format > 0xffff)
                 return -ERANGE;
 
-        if (drm->device.info.chipset >= 0x30) {
+        if (drm->client.device.info.chipset >= 0x30) {
                 if (crtc_w < (src_w >> 1) || crtc_h < (src_h >> 1))
                         return -ERANGE;
         } else {
@@ -174,7 +174,7 @@ nv10_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
 static int
 nv10_disable_plane(struct drm_plane *plane)
 {
-        struct nvif_object *dev = &nouveau_drm(plane->dev)->device.object;
+        struct nvif_object *dev = &nouveau_drm(plane->dev)->client.device.object;
         struct nouveau_plane *nv_plane =
                 container_of(plane, struct nouveau_plane, base);
 
@@ -198,7 +198,7 @@ nv_destroy_plane(struct drm_plane *plane)
 static void
 nv10_set_params(struct nouveau_plane *plane)
 {
-        struct nvif_object *dev = &nouveau_drm(plane->base.dev)->device.object;
+        struct nvif_object *dev = &nouveau_drm(plane->base.dev)->client.device.object;
         u32 luma = (plane->brightness - 512) << 16 | plane->contrast;
         u32 chroma = ((sin_mul(plane->hue, plane->saturation) & 0xffff) << 16) |
                 (cos_mul(plane->hue, plane->saturation) & 0xffff);
@@ -268,7 +268,7 @@ nv10_overlay_init(struct drm_device *device)
         if (!plane)
                 return;
 
-        switch (drm->device.info.chipset) {
+        switch (drm->client.device.info.chipset) {
         case 0x10:
         case 0x11:
         case 0x15:
@@ -347,7 +347,7 @@ nv04_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
                   uint32_t src_x, uint32_t src_y,
                   uint32_t src_w, uint32_t src_h)
 {
-        struct nvif_object *dev = &nouveau_drm(plane->dev)->device.object;
+        struct nvif_object *dev = &nouveau_drm(plane->dev)->client.device.object;
         struct nouveau_plane *nv_plane =
                 container_of(plane, struct nouveau_plane, base);
         struct nouveau_framebuffer *nv_fb = nouveau_framebuffer(fb);
@@ -427,7 +427,7 @@ nv04_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
 static int
 nv04_disable_plane(struct drm_plane *plane)
 {
-        struct nvif_object *dev = &nouveau_drm(plane->dev)->device.object;
+        struct nvif_object *dev = &nouveau_drm(plane->dev)->client.device.object;
         struct nouveau_plane *nv_plane =
                 container_of(plane, struct nouveau_plane, base);
 
@@ -495,7 +495,7 @@ err:
 void
 nouveau_overlay_init(struct drm_device *device)
 {
-        struct nvif_device *dev = &nouveau_drm(device)->device;
+        struct nvif_device *dev = &nouveau_drm(device)->client.device;
         if (dev->info.chipset < 0x10)
                 nv04_overlay_init(device);
         else if (dev->info.chipset <= 0x40)
diff --git a/drivers/gpu/drm/nouveau/dispnv04/tvnv04.c b/drivers/gpu/drm/nouveau/dispnv04/tvnv04.c
index 477a8d072af4..01664357d3e1 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/tvnv04.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/tvnv04.c
@@ -54,7 +54,7 @@ static struct nvkm_i2c_bus_probe nv04_tv_encoder_info[] = {
 int nv04_tv_identify(struct drm_device *dev, int i2c_index)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_i2c *i2c = nvxx_i2c(&drm->device);
+        struct nvkm_i2c *i2c = nvxx_i2c(&drm->client.device);
         struct nvkm_i2c_bus *bus = nvkm_i2c_bus_find(i2c, i2c_index);
         if (bus) {
                 return nvkm_i2c_bus_probe(bus, "TV encoder",
@@ -206,7 +206,7 @@ nv04_tv_create(struct drm_connector *connector, struct dcb_output *entry)
         struct drm_encoder *encoder;
         struct drm_device *dev = connector->dev;
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_i2c *i2c = nvxx_i2c(&drm->device);
+        struct nvkm_i2c *i2c = nvxx_i2c(&drm->client.device);
         struct nvkm_i2c_bus *bus = nvkm_i2c_bus_find(i2c, entry->i2c_index);
         int type, ret;
 
diff --git a/drivers/gpu/drm/nouveau/dispnv04/tvnv17.c b/drivers/gpu/drm/nouveau/dispnv04/tvnv17.c
index 434d1e29f279..6d99f11fee4e 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/tvnv17.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/tvnv17.c
@@ -46,7 +46,7 @@ static uint32_t nv42_tv_sample_load(struct drm_encoder *encoder)
 {
         struct drm_device *dev = encoder->dev;
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_gpio *gpio = nvxx_gpio(&drm->device);
+        struct nvkm_gpio *gpio = nvxx_gpio(&drm->client.device);
         uint32_t testval, regoffset = nv04_dac_output_offset(encoder);
         uint32_t gpio0, gpio1, fp_htotal, fp_hsync_start, fp_hsync_end,
                 fp_control, test_ctrl, dacclk, ctv_14, ctv_1c, ctv_6c;
@@ -130,7 +130,7 @@ static bool
 get_tv_detect_quirks(struct drm_device *dev, uint32_t *pin_mask)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
 
         if (device->quirk && device->quirk->tv_pin_mask) {
                 *pin_mask = device->quirk->tv_pin_mask;
@@ -154,8 +154,8 @@ nv17_tv_detect(struct drm_encoder *encoder, struct drm_connector *connector)
                 return connector_status_disconnected;
 
         if (reliable) {
-                if (drm->device.info.chipset == 0x42 ||
-                    drm->device.info.chipset == 0x43)
+                if (drm->client.device.info.chipset == 0x42 ||
+                    drm->client.device.info.chipset == 0x43)
                         tv_enc->pin_mask =
                                 nv42_tv_sample_load(encoder) >> 28 & 0xe;
                 else
@@ -362,7 +362,7 @@ static void  nv17_tv_dpms(struct drm_encoder *encoder, int mode)
 {
         struct drm_device *dev = encoder->dev;
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_gpio *gpio = nvxx_gpio(&drm->device);
+        struct nvkm_gpio *gpio = nvxx_gpio(&drm->client.device);
         struct nv17_tv_state *regs = &to_tv_enc(encoder)->state;
         struct nv17_tv_norm_params *tv_norm = get_tv_norm(encoder);
 
@@ -435,7 +435,7 @@ static void nv17_tv_prepare(struct drm_encoder *encoder)
         /* Set the DACCLK register */
         dacclk = (NVReadRAMDAC(dev, 0, dacclk_off) & ~0x30) | 0x1;
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE)
                 dacclk |= 0x1a << 16;
 
         if (tv_norm->kind == CTV_ENC_MODE) {
@@ -492,7 +492,7 @@ static void nv17_tv_mode_set(struct drm_encoder *encoder,
                         tv_regs->ptv_614 = 0x13;
                 }
 
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_RANKINE) {
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_RANKINE) {
                         tv_regs->ptv_500 = 0xe8e0;
                         tv_regs->ptv_504 = 0x1710;
                         tv_regs->ptv_604 = 0x0;
@@ -587,7 +587,7 @@ static void nv17_tv_commit(struct drm_encoder *encoder)
         nv17_tv_state_load(dev, &to_tv_enc(encoder)->state);
 
         /* This could use refinement for flatpanels, but it should work */
-        if (drm->device.info.chipset < 0x44)
+        if (drm->client.device.info.chipset < 0x44)
                 NVWriteRAMDAC(dev, 0, NV_PRAMDAC_TEST_CONTROL +
                                         nv04_dac_output_offset(encoder),
                                         0xf0000000);
diff --git a/drivers/gpu/drm/nouveau/dispnv04/tvnv17.h b/drivers/gpu/drm/nouveau/dispnv04/tvnv17.h
index 1b07521cde0d..29773b325bd9 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/tvnv17.h
+++ b/drivers/gpu/drm/nouveau/dispnv04/tvnv17.h
@@ -130,13 +130,13 @@ void nv17_ctv_update_rescaler(struct drm_encoder *encoder);
 static inline void nv_write_ptv(struct drm_device *dev, uint32_t reg,
                                 uint32_t val)
 {
-        struct nvif_device *device = &nouveau_drm(dev)->device;
+        struct nvif_device *device = &nouveau_drm(dev)->client.device;
         nvif_wr32(&device->object, reg, val);
 }
 
 static inline uint32_t nv_read_ptv(struct drm_device *dev, uint32_t reg)
 {
-        struct nvif_device *device = &nouveau_drm(dev)->device;
+        struct nvif_device *device = &nouveau_drm(dev)->client.device;
         return nvif_rd32(&device->object, reg);
 }
 
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl826e.h b/drivers/gpu/drm/nouveau/include/nvif/cl826e.h
index 05e6ef7cd190..91e33db21a2f 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/cl826e.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/cl826e.h
@@ -10,5 +10,5 @@ struct g82_channel_dma_v0 {
         __u64 offset;
 };
 
-#define G82_CHANNEL_DMA_V0_NTFY_UEVENT                                     0x00
+#define NV826E_V0_NTFY_NON_STALL_INTERRUPT                                 0x00
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl826f.h b/drivers/gpu/drm/nouveau/include/nvif/cl826f.h
index cecafcb1e954..e34efd4ec537 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/cl826f.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/cl826f.h
@@ -11,5 +11,5 @@ struct g82_channel_gpfifo_v0 {
         __u64 vm;
 };
 
-#define G82_CHANNEL_GPFIFO_V0_NTFY_UEVENT                                  0x00
+#define NV826F_V0_NTFY_NON_STALL_INTERRUPT                                 0x00
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl906f.h b/drivers/gpu/drm/nouveau/include/nvif/cl906f.h
index 2caf0838fcfd..a2d5410a491b 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/cl906f.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/cl906f.h
@@ -10,5 +10,6 @@ struct fermi_channel_gpfifo_v0 {
         __u64 vm;
 };
 
-#define FERMI_CHANNEL_GPFIFO_V0_NTFY_UEVENT                                0x00
+#define NV906F_V0_NTFY_NON_STALL_INTERRUPT                                 0x00
+#define NV906F_V0_NTFY_KILLED                                              0x01
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cla06f.h b/drivers/gpu/drm/nouveau/include/nvif/cla06f.h
index 46301ec018ce..2efa3d048bb9 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/cla06f.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/cla06f.h
@@ -25,5 +25,6 @@ struct kepler_channel_gpfifo_a_v0 {
         __u64 vm;
 };
 
-#define NVA06F_V0_NTFY_UEVENT                                              0x00
+#define NVA06F_V0_NTFY_NON_STALL_INTERRUPT                                 0x00
+#define NVA06F_V0_NTFY_KILLED                                              0x01
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/class.h b/drivers/gpu/drm/nouveau/include/nvif/class.h
index 82235f30277c..3a2c0137d4b4 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/class.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/class.h
@@ -2,23 +2,31 @@
 #define __NVIF_CLASS_H__
 
 /* these class numbers are made up by us, and not nvidia-assigned */
-#define NVIF_CLASS_CONTROL                                    /* if0001.h */ -1
-#define NVIF_CLASS_PERFMON                                    /* if0002.h */ -2
-#define NVIF_CLASS_PERFDOM                                    /* if0003.h */ -3
-#define NVIF_CLASS_SW_NV04                                    /* if0004.h */ -4
-#define NVIF_CLASS_SW_NV10                                    /* if0005.h */ -5
-#define NVIF_CLASS_SW_NV50                                    /* if0005.h */ -6
-#define NVIF_CLASS_SW_GF100                                   /* if0005.h */ -7
+#define NVIF_CLASS_CLIENT                            /* if0000.h */ -0x00000000
+
+#define NVIF_CLASS_CONTROL                           /* if0001.h */ -0x00000001
+
+#define NVIF_CLASS_PERFMON                           /* if0002.h */ -0x00000002
+#define NVIF_CLASS_PERFDOM                           /* if0003.h */ -0x00000003
+
+#define NVIF_CLASS_SW_NV04                           /* if0004.h */ -0x00000004
+#define NVIF_CLASS_SW_NV10                           /* if0005.h */ -0x00000005
+#define NVIF_CLASS_SW_NV50                           /* if0005.h */ -0x00000006
+#define NVIF_CLASS_SW_GF100                          /* if0005.h */ -0x00000007
 
 /* the below match nvidia-assigned (either in hw, or sw) class numbers */
+#define NV_NULL_CLASS                                                0x00000030
+
 #define NV_DEVICE                                     /* cl0080.h */ 0x00000080
 
 #define NV_DMA_FROM_MEMORY                            /* cl0002.h */ 0x00000002
 #define NV_DMA_TO_MEMORY                              /* cl0002.h */ 0x00000003
 #define NV_DMA_IN_MEMORY                              /* cl0002.h */ 0x0000003d
 
+#define NV50_TWOD                                                    0x0000502d
 #define FERMI_TWOD_A                                                 0x0000902d
 
+#define NV50_MEMORY_TO_MEMORY_FORMAT                                 0x00005039
 #define FERMI_MEMORY_TO_MEMORY_FORMAT_A                              0x00009039
 
 #define KEPLER_INLINE_TO_MEMORY_A                                    0x0000a040
@@ -99,6 +107,12 @@
 #define GF110_DISP_OVERLAY_CONTROL_DMA                /* cl507e.h */ 0x0000907e
 #define GK104_DISP_OVERLAY_CONTROL_DMA                /* cl507e.h */ 0x0000917e
 
+#define NV50_TESLA                                                   0x00005097
+#define G82_TESLA                                                    0x00008297
+#define GT200_TESLA                                                  0x00008397
+#define GT214_TESLA                                                  0x00008597
+#define GT21A_TESLA                                                  0x00008697
+
 #define FERMI_A                                       /* cl9097.h */ 0x00009097
 #define FERMI_B                                       /* cl9097.h */ 0x00009197
 #define FERMI_C                                       /* cl9097.h */ 0x00009297
@@ -140,6 +154,8 @@
 
 #define FERMI_DECOMPRESS                                             0x000090b8
 
+#define NV50_COMPUTE                                                 0x000050c0
+#define GT214_COMPUTE                                                0x000085c0
 #define FERMI_COMPUTE_A                                              0x000090c0
 #define FERMI_COMPUTE_B                                              0x000091c0
 #define KEPLER_COMPUTE_A                                             0x0000a0c0
diff --git a/drivers/gpu/drm/nouveau/include/nvif/client.h b/drivers/gpu/drm/nouveau/include/nvif/client.h
index 4a7f6f7b836d..b52a8eadce01 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/client.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/client.h
@@ -11,8 +11,7 @@ struct nvif_client {
         bool super;
 };
 
-int  nvif_client_init(const char *drv, const char *name, u64 device,
-                      const char *cfg, const char *dbg,
+int  nvif_client_init(struct nvif_client *parent, const char *name, u64 device,
                       struct nvif_client *);
 void nvif_client_fini(struct nvif_client *);
 int  nvif_client_ioctl(struct nvif_client *, void *, u32);
diff --git a/drivers/gpu/drm/nouveau/include/nvif/driver.h b/drivers/gpu/drm/nouveau/include/nvif/driver.h
index 8bd39e69229c..0c6f48d8140a 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/driver.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/driver.h
@@ -1,5 +1,7 @@
 #ifndef __NVIF_DRIVER_H__
 #define __NVIF_DRIVER_H__
+#include <nvif/os.h>
+struct nvif_client;
 
 struct nvif_driver {
         const char *name;
@@ -14,9 +16,11 @@ struct nvif_driver {
         bool keep;
 };
 
+int nvif_driver_init(const char *drv, const char *cfg, const char *dbg,
+                     const char *name, u64 device, struct nvif_client *);
+
 extern const struct nvif_driver nvif_driver_nvkm;
 extern const struct nvif_driver nvif_driver_drm;
 extern const struct nvif_driver nvif_driver_lib;
 extern const struct nvif_driver nvif_driver_null;
-
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/if0000.h b/drivers/gpu/drm/nouveau/include/nvif/if0000.h
index 85c44e8a1201..c2c0fc41e017 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/if0000.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/if0000.h
@@ -1,9 +1,16 @@
 #ifndef __NVIF_IF0000_H__
 #define __NVIF_IF0000_H__
 
-#define NV_CLIENT_DEVLIST                                                  0x00
+struct nvif_client_v0 {
+        __u8  version;
+        __u8  pad01[7];
+        __u64 device;
+        char  name[32];
+};
+
+#define NVIF_CLIENT_V0_DEVLIST                                             0x00
 
-struct nv_client_devlist_v0 {
+struct nvif_client_devlist_v0 {
         __u8  version;
         __u8  count;
         __u8  pad02[6];
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/client.h b/drivers/gpu/drm/nouveau/include/nvkm/core/client.h
index eaf5905a87a3..e876634da10a 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/client.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/client.h
@@ -1,5 +1,6 @@
 #ifndef __NVKM_CLIENT_H__
 #define __NVKM_CLIENT_H__
+#define nvkm_client(p) container_of((p), struct nvkm_client, object)
 #include <core/object.h>
 
 struct nvkm_client {
@@ -8,9 +9,8 @@ struct nvkm_client {
         u64 device;
         u32 debug;
 
-        struct nvkm_client_notify *notify[16];
+        struct nvkm_client_notify *notify[32];
         struct rb_root objroot;
-        struct rb_root dmaroot;
 
         bool super;
         void *data;
@@ -19,15 +19,11 @@ struct nvkm_client {
         struct nvkm_vm *vm;
 };
 
-bool nvkm_client_insert(struct nvkm_client *, struct nvkm_object *);
-void nvkm_client_remove(struct nvkm_client *, struct nvkm_object *);
-struct nvkm_object *nvkm_client_search(struct nvkm_client *, u64 object);
-
 int  nvkm_client_new(const char *name, u64 device, const char *cfg,
-                     const char *dbg, struct nvkm_client **);
-void nvkm_client_del(struct nvkm_client **);
-int  nvkm_client_init(struct nvkm_client *);
-int  nvkm_client_fini(struct nvkm_client *, bool suspend);
+                     const char *dbg,
+                     int (*)(const void *, u32, const void *, u32),
+                     struct nvkm_client **);
+struct nvkm_client *nvkm_client_search(struct nvkm_client *, u64 handle);
 
 int nvkm_client_notify_new(struct nvkm_object *, struct nvkm_event *,
                            void *data, u32 size);
@@ -37,8 +33,8 @@ int nvkm_client_notify_put(struct nvkm_client *, int index);
 
 /* logging for client-facing objects */
 #define nvif_printk(o,l,p,f,a...) do {                                         \
-        struct nvkm_object *_object = (o);                                     \
-        struct nvkm_client *_client = _object->client;                         \
+        const struct nvkm_object *_object = (o);                               \
+        const struct nvkm_client *_client = _object->client;                   \
         if (_client->debug >= NV_DBG_##l)                                      \
                 printk(KERN_##p "nouveau: %s:%08x:%08x: "f, _client->name,     \
                        _object->handle, _object->oclass, ##a);                 \
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/device.h b/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
index 6bc712f32c8b..d426b86e2712 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
@@ -262,7 +262,7 @@ extern const struct nvkm_sclass nvkm_udevice_sclass;
 
 /* device logging */
 #define nvdev_printk_(d,l,p,f,a...) do {                                       \
-        struct nvkm_device *_device = (d);                                     \
+        const struct nvkm_device *_device = (d);                               \
         if (_device->debug >= (l))                                             \
                 dev_##p(_device->dev, f, ##a);                                 \
 } while(0)
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h b/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h
index 9ebfd8782366..d4cd2fbfde88 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h
@@ -20,6 +20,7 @@ struct nvkm_engine_func {
         int (*fini)(struct nvkm_engine *, bool suspend);
         void (*intr)(struct nvkm_engine *);
         void (*tile)(struct nvkm_engine *, int region, struct nvkm_fb_tile *);
+        bool (*chsw_load)(struct nvkm_engine *);
 
         struct {
                 int (*sclass)(struct nvkm_oclass *, int index,
@@ -44,4 +45,5 @@ int nvkm_engine_new_(const struct nvkm_engine_func *, struct nvkm_device *,
 struct nvkm_engine *nvkm_engine_ref(struct nvkm_engine *);
 void nvkm_engine_unref(struct nvkm_engine **);
 void nvkm_engine_tile(struct nvkm_engine *, int region);
+bool nvkm_engine_chsw_load(struct nvkm_engine *);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/memory.h b/drivers/gpu/drm/nouveau/include/nvkm/core/memory.h
index 9363b839a9da..33ca6769266a 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/memory.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/memory.h
@@ -6,9 +6,10 @@ struct nvkm_vma;
 struct nvkm_vm;
 
 enum nvkm_memory_target {
-        NVKM_MEM_TARGET_INST,
-        NVKM_MEM_TARGET_VRAM,
-        NVKM_MEM_TARGET_HOST,
+        NVKM_MEM_TARGET_INST, /* instance memory */
+        NVKM_MEM_TARGET_VRAM, /* video memory */
+        NVKM_MEM_TARGET_HOST, /* coherent system memory */
+        NVKM_MEM_TARGET_NCOH, /* non-coherent system memory */
 };
 
 struct nvkm_memory {
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h b/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h
index d92fd41e4056..7bd4897a8a2a 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h
@@ -5,7 +5,7 @@
 struct nvkm_mm_node {
         struct list_head nl_entry;
         struct list_head fl_entry;
-        struct list_head rl_entry;
+        struct nvkm_mm_node *next;
 
 #define NVKM_MM_HEAP_ANY 0x00
         u8  heap;
@@ -38,4 +38,10 @@ int  nvkm_mm_tail(struct nvkm_mm *, u8 heap, u8 type, u32 size_max,
                   u32 size_min, u32 align, struct nvkm_mm_node **);
 void nvkm_mm_free(struct nvkm_mm *, struct nvkm_mm_node **);
 void nvkm_mm_dump(struct nvkm_mm *, const char *);
+
+static inline bool
+nvkm_mm_contiguous(struct nvkm_mm_node *node)
+{
+        return !node->next;
+}
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/object.h b/drivers/gpu/drm/nouveau/include/nvkm/core/object.h
index dcd048b91fac..96dda350ada3 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/object.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/object.h
@@ -62,6 +62,11 @@ int nvkm_object_wr32(struct nvkm_object *, u64 addr, u32  data);
 int nvkm_object_bind(struct nvkm_object *, struct nvkm_gpuobj *, int align,
                      struct nvkm_gpuobj **);
 
+bool nvkm_object_insert(struct nvkm_object *);
+void nvkm_object_remove(struct nvkm_object *);
+struct nvkm_object *nvkm_object_search(struct nvkm_client *, u64 object,
+                                       const struct nvkm_object_func *);
+
 struct nvkm_sclass {
         int minver;
         int maxver;
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h b/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h
index 57adefa8b08e..ca9ed3d68f44 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h
@@ -32,7 +32,7 @@ void nvkm_subdev_intr(struct nvkm_subdev *);
 
 /* subdev logging */
 #define nvkm_printk_(s,l,p,f,a...) do {                                        \
-        struct nvkm_subdev *_subdev = (s);                                     \
+        const struct nvkm_subdev *_subdev = (s);                               \
         if (_subdev->debug >= (l)) {                                           \
                 dev_##p(_subdev->device->dev, "%s: "f,                         \
                         nvkm_subdev_name[_subdev->index], ##a);                \
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/dma.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/dma.h
index 114bfb737a81..d2a6532ce3b9 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/dma.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/dma.h
@@ -12,9 +12,6 @@ struct nvkm_dmaobj {
         u32 access;
         u64 start;
         u64 limit;
-
-        struct rb_node rb;
-        u64 handle; /*XXX HANDLE MERGE */
 };
 
 struct nvkm_dma {
@@ -22,8 +19,7 @@ struct nvkm_dma {
         struct nvkm_engine engine;
 };
 
-struct nvkm_dmaobj *
-nvkm_dma_search(struct nvkm_dma *, struct nvkm_client *, u64 object);
+struct nvkm_dmaobj *nvkm_dmaobj_search(struct nvkm_client *, u64 object);
 
 int nv04_dma_new(struct nvkm_device *, int, struct nvkm_dma **);
 int nv50_dma_new(struct nvkm_device *, int, struct nvkm_dma **);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h
index e6baf039c269..7e498e65b1e8 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h
@@ -4,13 +4,26 @@
 #include <core/engine.h>
 struct nvkm_fifo_chan;
 
+enum nvkm_falcon_dmaidx {
+        FALCON_DMAIDX_UCODE             = 0,
+        FALCON_DMAIDX_VIRT              = 1,
+        FALCON_DMAIDX_PHYS_VID          = 2,
+        FALCON_DMAIDX_PHYS_SYS_COH      = 3,
+        FALCON_DMAIDX_PHYS_SYS_NCOH     = 4,
+};
+
 struct nvkm_falcon {
         const struct nvkm_falcon_func *func;
-        struct nvkm_engine engine;
-
+        const struct nvkm_subdev *owner;
+        const char *name;
         u32 addr;
-        u8  version;
-        u8  secret;
+
+        struct mutex mutex;
+        const struct nvkm_subdev *user;
+
+        u8 version;
+        u8 secret;
+        bool debug;
 
         struct nvkm_memory *core;
         bool external;
@@ -19,15 +32,25 @@ struct nvkm_falcon {
                 u32 limit;
                 u32 *data;
                 u32  size;
+                u8 ports;
         } code;
 
         struct {
                 u32 limit;
                 u32 *data;
                 u32  size;
+                u8 ports;
         } data;
+
+        struct nvkm_engine engine;
 };
 
+int nvkm_falcon_v1_new(struct nvkm_subdev *owner, const char *name, u32 addr,
+                       struct nvkm_falcon **);
+void nvkm_falcon_del(struct nvkm_falcon **);
+int nvkm_falcon_get(struct nvkm_falcon *, const struct nvkm_subdev *);
+void nvkm_falcon_put(struct nvkm_falcon *, const struct nvkm_subdev *);
+
 int nvkm_falcon_new_(const struct nvkm_falcon_func *, struct nvkm_device *,
                      int index, bool enable, u32 addr, struct nvkm_engine **);
 
@@ -42,6 +65,51 @@ struct nvkm_falcon_func {
         } data;
         void (*init)(struct nvkm_falcon *);
         void (*intr)(struct nvkm_falcon *, struct nvkm_fifo_chan *);
+        void (*load_imem)(struct nvkm_falcon *, void *, u32, u32, u16, u8, bool);
+        void (*load_dmem)(struct nvkm_falcon *, void *, u32, u32, u8);
+        void (*read_dmem)(struct nvkm_falcon *, u32, u32, u8, void *);
+        void (*bind_context)(struct nvkm_falcon *, struct nvkm_gpuobj *);
+        int (*wait_for_halt)(struct nvkm_falcon *, u32);
+        int (*clear_interrupt)(struct nvkm_falcon *, u32);
+        void (*set_start_addr)(struct nvkm_falcon *, u32 start_addr);
+        void (*start)(struct nvkm_falcon *);
+        int (*enable)(struct nvkm_falcon *falcon);
+        void (*disable)(struct nvkm_falcon *falcon);
+
         struct nvkm_sclass sclass[];
 };
+
+static inline u32
+nvkm_falcon_rd32(struct nvkm_falcon *falcon, u32 addr)
+{
+        return nvkm_rd32(falcon->owner->device, falcon->addr + addr);
+}
+
+static inline void
+nvkm_falcon_wr32(struct nvkm_falcon *falcon, u32 addr, u32 data)
+{
+        nvkm_wr32(falcon->owner->device, falcon->addr + addr, data);
+}
+
+static inline u32
+nvkm_falcon_mask(struct nvkm_falcon *falcon, u32 addr, u32 mask, u32 val)
+{
+        struct nvkm_device *device = falcon->owner->device;
+
+        return nvkm_mask(device, falcon->addr + addr, mask, val);
+}
+
+void nvkm_falcon_load_imem(struct nvkm_falcon *, void *, u32, u32, u16, u8,
+                           bool);
+void nvkm_falcon_load_dmem(struct nvkm_falcon *, void *, u32, u32, u8);
+void nvkm_falcon_read_dmem(struct nvkm_falcon *, u32, u32, u8, void *);
+void nvkm_falcon_bind_context(struct nvkm_falcon *, struct nvkm_gpuobj *);
+void nvkm_falcon_set_start_addr(struct nvkm_falcon *, u32);
+void nvkm_falcon_start(struct nvkm_falcon *);
+int nvkm_falcon_wait_for_halt(struct nvkm_falcon *, u32);
+int nvkm_falcon_clear_interrupt(struct nvkm_falcon *, u32);
+int nvkm_falcon_enable(struct nvkm_falcon *);
+void nvkm_falcon_disable(struct nvkm_falcon *);
+int nvkm_falcon_reset(struct nvkm_falcon *);
+
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
index ed92fec5292c..24efa900d8ca 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
@@ -40,6 +40,7 @@ struct nvkm_fifo {
 
         struct nvkm_event uevent; /* async user trigger */
         struct nvkm_event cevent; /* channel creation event */
+        struct nvkm_event kevent; /* channel killed */
 };
 
 void nvkm_fifo_pause(struct nvkm_fifo *, unsigned long *);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/power_budget.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/power_budget.h
new file mode 100644
index 000000000000..f5f4a14c4030
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/power_budget.h
@@ -0,0 +1,26 @@
+#ifndef __NVBIOS_POWER_BUDGET_H__
+#define __NVBIOS_POWER_BUDGET_H__
+
+#include <nvkm/subdev/bios.h>
+
+struct nvbios_power_budget_entry {
+        u32 min_w;
+        u32 avg_w;
+        u32 max_w;
+};
+
+struct nvbios_power_budget {
+        u32 offset;
+        u8  ver;
+        u8  hlen;
+        u8  elen;
+        u8  ecount;
+        u8  cap_entry;
+};
+
+int nvbios_power_budget_header(struct nvkm_bios *,
+                               struct nvbios_power_budget *);
+int nvbios_power_budget_entry(struct nvkm_bios *, struct nvbios_power_budget *,
+                              u8 idx, struct nvbios_power_budget_entry *);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h
index 794e432578b2..0b26a4c860ec 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h
@@ -29,7 +29,7 @@ struct nvkm_mem {
         u8  page_shift;
 
         struct nvkm_mm_node *tag;
-        struct list_head regions;
+        struct nvkm_mm_node *mem;
         dma_addr_t *pages;
         u32 memtype;
         u64 offset;
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/iccsense.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/iccsense.h
index 3c2ddd975273..b7a9b041e130 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/iccsense.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/iccsense.h
@@ -8,6 +8,9 @@ struct nvkm_iccsense {
         bool data_valid;
         struct list_head sensors;
         struct list_head rails;
+
+        u32 power_w_max;
+        u32 power_w_crit;
 };
 
 int gf100_iccsense_new(struct nvkm_device *, int index, struct nvkm_iccsense **);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h
index 27d25b18d85c..e68ba636741b 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h
@@ -9,6 +9,7 @@ struct nvkm_mc {
 
 void nvkm_mc_enable(struct nvkm_device *, enum nvkm_devidx);
 void nvkm_mc_disable(struct nvkm_device *, enum nvkm_devidx);
+bool nvkm_mc_enabled(struct nvkm_device *, enum nvkm_devidx);
 void nvkm_mc_reset(struct nvkm_device *, enum nvkm_devidx);
 void nvkm_mc_intr(struct nvkm_device *, bool *handled);
 void nvkm_mc_intr_unarm(struct nvkm_device *);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h
index e6523e2cea9f..ac2a695963c1 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h
@@ -43,6 +43,7 @@ int nv40_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int nv46_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int nv4c_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int g84_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
+int g92_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int g94_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int gf100_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int gf106_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/pmu.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/pmu.h
index f37538eb1fe5..179b6ed3f595 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/pmu.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/pmu.h
@@ -1,10 +1,12 @@
 #ifndef __NVKM_PMU_H__
 #define __NVKM_PMU_H__
 #include <core/subdev.h>
+#include <engine/falcon.h>
 
 struct nvkm_pmu {
         const struct nvkm_pmu_func *func;
         struct nvkm_subdev subdev;
+        struct nvkm_falcon *falcon;
 
         struct {
                 u32 base;
@@ -35,6 +37,7 @@ int gk110_pmu_new(struct nvkm_device *, int, struct nvkm_pmu **);
 int gk208_pmu_new(struct nvkm_device *, int, struct nvkm_pmu **);
 int gk20a_pmu_new(struct nvkm_device *, int, struct nvkm_pmu **);
 int gm107_pmu_new(struct nvkm_device *, int, struct nvkm_pmu **);
+int gm20b_pmu_new(struct nvkm_device *, int, struct nvkm_pmu **);
 int gp100_pmu_new(struct nvkm_device *, int, struct nvkm_pmu **);
 int gp102_pmu_new(struct nvkm_device *, int, struct nvkm_pmu **);
 
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h
index b04c38c07761..5dbd8aa4f8c2 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h
@@ -26,7 +26,7 @@
 #include <core/subdev.h>
 
 enum nvkm_secboot_falcon {
-        NVKM_SECBOOT_FALCON_PMU = 0,
+        NVKM_SECBOOT_FALCON_PMU = 0,
         NVKM_SECBOOT_FALCON_RESERVED = 1,
         NVKM_SECBOOT_FALCON_FECS = 2,
         NVKM_SECBOOT_FALCON_GPCCS = 3,
@@ -35,22 +35,23 @@ enum nvkm_secboot_falcon {
 };
 
 /**
- * @base:               base IO address of the falcon performing secure boot
- * @irq_mask:           IRQ mask of the falcon performing secure boot
- * @enable_mask:        enable mask of the falcon performing secure boot
+ * @wpr_set: whether the WPR region is currently set
 */
 struct nvkm_secboot {
         const struct nvkm_secboot_func *func;
+        struct nvkm_acr *acr;
         struct nvkm_subdev subdev;
+        struct nvkm_falcon *boot_falcon;
 
-        enum nvkm_devidx devidx;
-        u32 base;
+        u64 wpr_addr;
+        u32 wpr_size;
+
+        bool wpr_set;
 };
 #define nvkm_secboot(p) container_of((p), struct nvkm_secboot, subdev)
 
 bool nvkm_secboot_is_managed(struct nvkm_secboot *, enum nvkm_secboot_falcon);
-int nvkm_secboot_reset(struct nvkm_secboot *, u32 falcon);
-int nvkm_secboot_start(struct nvkm_secboot *, u32 falcon);
+int nvkm_secboot_reset(struct nvkm_secboot *, enum nvkm_secboot_falcon);
 
 int gm200_secboot_new(struct nvkm_device *, int, struct nvkm_secboot **);
 int gm20b_secboot_new(struct nvkm_device *, int, struct nvkm_secboot **);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/timer.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/timer.h
index 82d3e28918fd..6a567fe347b3 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/timer.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/timer.h
@@ -48,10 +48,8 @@ void nvkm_timer_alarm_cancel(struct nvkm_timer *, struct nvkm_alarm *);
         } while (_taken = nvkm_timer_read(_tmr) - _time0, _taken < _nsecs);    \
                                                                                \
         if (_taken >= _nsecs) {                                                \
-                if (_warn) {                                                   \
-                        dev_warn(_device->dev, "timeout at %s:%d/%s()!\n",     \
-                                 __FILE__, __LINE__, __func__);                \
-                }                                                              \
+                if (_warn)                                                     \
+                        dev_WARN(_device->dev, "timeout\n");                   \
                 _taken = -ETIMEDOUT;                                           \
         }                                                                      \
         _taken;                                                                \
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h
index 71ebbfd4484f..d23209b62c25 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h
@@ -11,6 +11,7 @@ struct nvkm_top {
 u32 nvkm_top_reset(struct nvkm_device *, enum nvkm_devidx);
 u32 nvkm_top_intr(struct nvkm_device *, u32 intr, u64 *subdevs);
 u32 nvkm_top_intr_mask(struct nvkm_device *, enum nvkm_devidx);
+int nvkm_top_fault_id(struct nvkm_device *, enum nvkm_devidx);
 enum nvkm_devidx nvkm_top_fault(struct nvkm_device *, int fault);
 enum nvkm_devidx nvkm_top_engine(struct nvkm_device *, int, int *runl, int *engn);
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c
index 4df4f6ed4886..f98f800cc011 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
@@ -87,7 +87,7 @@ nouveau_abi16_put(struct nouveau_abi16 *abi16, int ret)
 s32
 nouveau_abi16_swclass(struct nouveau_drm *drm)
 {
-        switch (drm->device.info.family) {
+        switch (drm->client.device.info.family) {
         case NV_DEVICE_INFO_V0_TNT:
                 return NVIF_CLASS_SW_NV04;
         case NV_DEVICE_INFO_V0_CELSIUS:
@@ -175,7 +175,7 @@ nouveau_abi16_ioctl_getparam(ABI16_IOCTL_ARGS)
 {
         struct nouveau_cli *cli = nouveau_cli(file_priv);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         struct nvkm_gr *gr = nvxx_gr(device);
         struct drm_nouveau_getparam *getparam = data;
 
@@ -321,7 +321,7 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
         }
 
         /* Named memory object area */
-        ret = nouveau_gem_new(dev, PAGE_SIZE, 0, NOUVEAU_GEM_DOMAIN_GART,
+        ret = nouveau_gem_new(cli, PAGE_SIZE, 0, NOUVEAU_GEM_DOMAIN_GART,
                               0, 0, &chan->ntfy);
         if (ret == 0)
                 ret = nouveau_bo_pin(chan->ntfy, TTM_PL_FLAG_TT, false);
diff --git a/drivers/gpu/drm/nouveau/nouveau_backlight.c b/drivers/gpu/drm/nouveau/nouveau_backlight.c
index 8b1ca4add2ed..380f340204e8 100644
--- a/drivers/gpu/drm/nouveau/nouveau_backlight.c
+++ b/drivers/gpu/drm/nouveau/nouveau_backlight.c
@@ -65,7 +65,7 @@ static int
 nv40_get_intensity(struct backlight_device *bd)
 {
         struct nouveau_drm *drm = bl_get_data(bd);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         int val = (nvif_rd32(device, NV40_PMC_BACKLIGHT) &
                                    NV40_PMC_BACKLIGHT_MASK) >> 16;
 
@@ -76,7 +76,7 @@ static int
 nv40_set_intensity(struct backlight_device *bd)
 {
         struct nouveau_drm *drm = bl_get_data(bd);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         int val = bd->props.brightness;
         int reg = nvif_rd32(device, NV40_PMC_BACKLIGHT);
 
@@ -96,7 +96,7 @@ static int
 nv40_backlight_init(struct drm_connector *connector)
 {
         struct nouveau_drm *drm = nouveau_drm(connector->dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         struct backlight_properties props;
         struct backlight_device *bd;
         struct backlight_connector bl_connector;
@@ -133,7 +133,7 @@ nv50_get_intensity(struct backlight_device *bd)
 {
         struct nouveau_encoder *nv_encoder = bl_get_data(bd);
         struct nouveau_drm *drm = nouveau_drm(nv_encoder->base.base.dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         int or = nv_encoder->or;
         u32 div = 1025;
         u32 val;
@@ -148,7 +148,7 @@ nv50_set_intensity(struct backlight_device *bd)
 {
         struct nouveau_encoder *nv_encoder = bl_get_data(bd);
         struct nouveau_drm *drm = nouveau_drm(nv_encoder->base.base.dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         int or = nv_encoder->or;
         u32 div = 1025;
         u32 val = (bd->props.brightness * div) / 100;
@@ -169,7 +169,7 @@ nva3_get_intensity(struct backlight_device *bd)
 {
         struct nouveau_encoder *nv_encoder = bl_get_data(bd);
         struct nouveau_drm *drm = nouveau_drm(nv_encoder->base.base.dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         int or = nv_encoder->or;
         u32 div, val;
 
@@ -187,7 +187,7 @@ nva3_set_intensity(struct backlight_device *bd)
 {
         struct nouveau_encoder *nv_encoder = bl_get_data(bd);
         struct nouveau_drm *drm = nouveau_drm(nv_encoder->base.base.dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         int or = nv_encoder->or;
         u32 div, val;
 
@@ -213,7 +213,7 @@ static int
 nv50_backlight_init(struct drm_connector *connector)
 {
         struct nouveau_drm *drm = nouveau_drm(connector->dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         struct nouveau_encoder *nv_encoder;
         struct backlight_properties props;
         struct backlight_device *bd;
@@ -231,9 +231,9 @@ nv50_backlight_init(struct drm_connector *connector)
         if (!nvif_rd32(device, NV50_PDISP_SOR_PWM_CTL(nv_encoder->or)))
                 return 0;
 
-        if (drm->device.info.chipset <= 0xa0 ||
-            drm->device.info.chipset == 0xaa ||
-            drm->device.info.chipset == 0xac)
+        if (drm->client.device.info.chipset <= 0xa0 ||
+            drm->client.device.info.chipset == 0xaa ||
+            drm->client.device.info.chipset == 0xac)
                 ops = &nv50_bl_ops;
         else
                 ops = &nva3_bl_ops;
@@ -265,7 +265,7 @@ int
 nouveau_backlight_init(struct drm_device *dev)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         struct drm_connector *connector;
 
         if (apple_gmux_present()) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.c b/drivers/gpu/drm/nouveau/nouveau_bios.c
index 23ffe8571a99..9a0772ad495a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.c
@@ -215,7 +215,7 @@ int call_lvds_script(struct drm_device *dev, struct dcb_output *dcbent, int head
          */
 
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         struct nvbios *bios = &drm->vbios;
         uint8_t lvds_ver = bios->data[bios->fp.lvdsmanufacturerpointer];
         uint32_t sel_clk_binding, sel_clk;
@@ -319,7 +319,7 @@ static int
 get_fp_strap(struct drm_device *dev, struct nvbios *bios)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
 
         /*
          * The fp strap is normally dictated by the "User Strap" in
@@ -333,10 +333,10 @@ get_fp_strap(struct drm_device *dev, struct nvbios *bios)
         if (bios->major_version < 5 && bios->data[0x48] & 0x4)
                 return NVReadVgaCrtc5758(dev, 0, 0xf) & 0xf;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_MAXWELL)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_MAXWELL)
                 return nvif_rd32(device, 0x001800) & 0x0000000f;
         else
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA)
                 return (nvif_rd32(device, NV_PEXTDEV_BOOT_0) >> 24) & 0xf;
         else
                 return (nvif_rd32(device, NV_PEXTDEV_BOOT_0) >> 16) & 0xf;
@@ -638,7 +638,7 @@ int run_tmds_table(struct drm_device *dev, struct dcb_output *dcbent, int head,
          */
 
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         struct nvbios *bios = &drm->vbios;
         int cv = bios->chip_version;
         uint16_t clktable = 0, scriptptr;
@@ -1255,7 +1255,7 @@ olddcb_table(struct drm_device *dev)
         struct nouveau_drm *drm = nouveau_drm(dev);
         u8 *dcb = NULL;
 
-        if (drm->device.info.family > NV_DEVICE_INFO_V0_TNT)
+        if (drm->client.device.info.family > NV_DEVICE_INFO_V0_TNT)
                 dcb = ROMPTR(dev, drm->vbios.data[0x36]);
         if (!dcb) {
                 NV_WARN(drm, "No DCB data found in VBIOS\n");
@@ -1918,7 +1918,7 @@ static int load_nv17_hwsq_ucode_entry(struct drm_device *dev, struct nvbios *bio
          */
 
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         uint8_t bytes_to_write;
         uint16_t hwsq_entry_offset;
         int i;
@@ -2012,7 +2012,7 @@ uint8_t *nouveau_bios_embedded_edid(struct drm_device *dev)
 static bool NVInitVBIOS(struct drm_device *dev)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_bios *bios = nvxx_bios(&drm->device);
+        struct nvkm_bios *bios = nvxx_bios(&drm->client.device);
         struct nvbios *legacy = &drm->vbios;
 
         memset(legacy, 0, sizeof(struct nvbios));
@@ -2064,7 +2064,7 @@ nouveau_bios_posted(struct drm_device *dev)
         struct nouveau_drm *drm = nouveau_drm(dev);
         unsigned htotal;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA)
                 return true;
 
         htotal  = NVReadVgaCrtc(dev, 0, 0x06);
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 8a528ebe30f3..548f36d33924 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -48,7 +48,7 @@ nv10_bo_update_tile_region(struct drm_device *dev, struct nouveau_drm_tile *reg,
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
         int i = reg - drm->tile.reg;
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
         struct nvkm_fb *fb = device->fb;
         struct nvkm_fb_tile *tile = &fb->tile.region[i];
 
@@ -100,7 +100,7 @@ nv10_bo_set_tiling(struct drm_device *dev, u32 addr,
                    u32 size, u32 pitch, u32 flags)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_fb *fb = nvxx_fb(&drm->device);
+        struct nvkm_fb *fb = nvxx_fb(&drm->client.device);
         struct nouveau_drm_tile *tile, *found = NULL;
         int i;
 
@@ -139,60 +139,62 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
         kfree(nvbo);
 }
 
+static inline u64
+roundup_64(u64 x, u32 y)
+{
+        x += y - 1;
+        do_div(x, y);
+        return x * y;
+}
+
 static void
 nouveau_bo_fixup_align(struct nouveau_bo *nvbo, u32 flags,
-                       int *align, int *size)
+                       int *align, u64 *size)
 {
         struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
 
         if (device->info.family < NV_DEVICE_INFO_V0_TESLA) {
                 if (nvbo->tile_mode) {
                         if (device->info.chipset >= 0x40) {
                                 *align = 65536;
-                                *size = roundup(*size, 64 * nvbo->tile_mode);
+                                *size = roundup_64(*size, 64 * nvbo->tile_mode);
 
                         } else if (device->info.chipset >= 0x30) {
                                 *align = 32768;
-                                *size = roundup(*size, 64 * nvbo->tile_mode);
+                                *size = roundup_64(*size, 64 * nvbo->tile_mode);
 
                         } else if (device->info.chipset >= 0x20) {
                                 *align = 16384;
-                                *size = roundup(*size, 64 * nvbo->tile_mode);
+                                *size = roundup_64(*size, 64 * nvbo->tile_mode);
 
                         } else if (device->info.chipset >= 0x10) {
                                 *align = 16384;
-                                *size = roundup(*size, 32 * nvbo->tile_mode);
+                                *size = roundup_64(*size, 32 * nvbo->tile_mode);
                         }
                 }
         } else {
-                *size = roundup(*size, (1 << nvbo->page_shift));
+                *size = roundup_64(*size, (1 << nvbo->page_shift));
                 *align = max((1 <<  nvbo->page_shift), *align);
         }
 
-        *size = roundup(*size, PAGE_SIZE);
+        *size = roundup_64(*size, PAGE_SIZE);
 }
 
 int
-nouveau_bo_new(struct drm_device *dev, int size, int align,
+nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
                uint32_t flags, uint32_t tile_mode, uint32_t tile_flags,
                struct sg_table *sg, struct reservation_object *robj,
                struct nouveau_bo **pnvbo)
 {
-        struct nouveau_drm *drm = nouveau_drm(dev);
+        struct nouveau_drm *drm = nouveau_drm(cli->dev);
         struct nouveau_bo *nvbo;
         size_t acc_size;
         int ret;
         int type = ttm_bo_type_device;
-        int lpg_shift = 12;
-        int max_size;
-
-        if (drm->client.vm)
-                lpg_shift = drm->client.vm->mmu->lpg_shift;
-        max_size = INT_MAX & ~((1 << lpg_shift) - 1);
 
-        if (size <= 0 || size > max_size) {
-                NV_WARN(drm, "skipped size %x\n", (u32)size);
+        if (!size) {
+                NV_WARN(drm, "skipped size %016llx\n", size);
                 return -EINVAL;
         }
 
@@ -208,8 +210,9 @@ nouveau_bo_new(struct drm_device *dev, int size, int align,
         nvbo->tile_mode = tile_mode;
         nvbo->tile_flags = tile_flags;
         nvbo->bo.bdev = &drm->ttm.bdev;
+        nvbo->cli = cli;
 
-        if (!nvxx_device(&drm->device)->func->cpu_coherent)
+        if (!nvxx_device(&drm->client.device)->func->cpu_coherent)
                 nvbo->force_coherent = flags & TTM_PL_FLAG_UNCACHED;
 
         nvbo->page_shift = 12;
@@ -255,10 +258,10 @@ static void
 set_placement_range(struct nouveau_bo *nvbo, uint32_t type)
 {
         struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);
-        u32 vram_pages = drm->device.info.ram_size >> PAGE_SHIFT;
+        u32 vram_pages = drm->client.device.info.ram_size >> PAGE_SHIFT;
         unsigned i, fpfn, lpfn;
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CELSIUS &&
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CELSIUS &&
             nvbo->tile_mode && (type & TTM_PL_FLAG_VRAM) &&
             nvbo->bo.mem.num_pages < vram_pages / 4) {
                 /*
@@ -316,12 +319,12 @@ nouveau_bo_pin(struct nouveau_bo *nvbo, uint32_t memtype, bool contig)
         if (ret)
                 return ret;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA &&
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA &&
             memtype == TTM_PL_FLAG_VRAM && contig) {
                 if (nvbo->tile_flags & NOUVEAU_GEM_TILE_NONCONTIG) {
                         if (bo->mem.mem_type == TTM_PL_VRAM) {
                                 struct nvkm_mem *mem = bo->mem.mm_node;
-                                if (!list_is_singular(&mem->regions))
+                                if (!nvkm_mm_contiguous(mem->mem))
                                         evict = true;
                         }
                         nvbo->tile_flags &= ~NOUVEAU_GEM_TILE_NONCONTIG;
@@ -443,7 +446,7 @@ void
 nouveau_bo_sync_for_device(struct nouveau_bo *nvbo)
 {
         struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
         struct ttm_dma_tt *ttm_dma = (struct ttm_dma_tt *)nvbo->bo.ttm;
         int i;
 
@@ -463,7 +466,7 @@ void
 nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo)
 {
         struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
         struct ttm_dma_tt *ttm_dma = (struct ttm_dma_tt *)nvbo->bo.ttm;
         int i;
 
@@ -579,9 +582,9 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
                                          TTM_PL_FLAG_WC;
                 man->default_caching = TTM_PL_FLAG_WC;
 
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
                         /* Some BARs do not support being ioremapped WC */
-                        if (nvxx_bar(&drm->device)->iomap_uncached) {
+                        if (nvxx_bar(&drm->client.device)->iomap_uncached) {
                                 man->available_caching = TTM_PL_FLAG_UNCACHED;
                                 man->default_caching = TTM_PL_FLAG_UNCACHED;
                         }
@@ -594,7 +597,7 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
                 }
                 break;
         case TTM_PL_TT:
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA)
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA)
                         man->func = &nouveau_gart_manager;
                 else
                 if (!drm->agp.bridge)
@@ -654,20 +657,20 @@ nve0_bo_move_init(struct nouveau_channel *chan, u32 handle)
 
 static int
 nve0_bo_move_copy(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        struct nvkm_mem *node = old_mem->mm_node;
+        struct nvkm_mem *mem = old_reg->mm_node;
         int ret = RING_SPACE(chan, 10);
         if (ret == 0) {
                 BEGIN_NVC0(chan, NvSubCopy, 0x0400, 8);
-                OUT_RING  (chan, upper_32_bits(node->vma[0].offset));
-                OUT_RING  (chan, lower_32_bits(node->vma[0].offset));
-                OUT_RING  (chan, upper_32_bits(node->vma[1].offset));
-                OUT_RING  (chan, lower_32_bits(node->vma[1].offset));
+                OUT_RING  (chan, upper_32_bits(mem->vma[0].offset));
+                OUT_RING  (chan, lower_32_bits(mem->vma[0].offset));
+                OUT_RING  (chan, upper_32_bits(mem->vma[1].offset));
+                OUT_RING  (chan, lower_32_bits(mem->vma[1].offset));
                 OUT_RING  (chan, PAGE_SIZE);
                 OUT_RING  (chan, PAGE_SIZE);
                 OUT_RING  (chan, PAGE_SIZE);
-                OUT_RING  (chan, new_mem->num_pages);
+                OUT_RING  (chan, new_reg->num_pages);
                 BEGIN_IMC0(chan, NvSubCopy, 0x0300, 0x0386);
         }
         return ret;
@@ -686,15 +689,15 @@ nvc0_bo_move_init(struct nouveau_channel *chan, u32 handle)
 
 static int
 nvc0_bo_move_copy(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        struct nvkm_mem *node = old_mem->mm_node;
-        u64 src_offset = node->vma[0].offset;
-        u64 dst_offset = node->vma[1].offset;
-        u32 page_count = new_mem->num_pages;
+        struct nvkm_mem *mem = old_reg->mm_node;
+        u64 src_offset = mem->vma[0].offset;
+        u64 dst_offset = mem->vma[1].offset;
+        u32 page_count = new_reg->num_pages;
         int ret;
 
-        page_count = new_mem->num_pages;
+        page_count = new_reg->num_pages;
         while (page_count) {
                 int line_count = (page_count > 8191) ? 8191 : page_count;
 
@@ -724,15 +727,15 @@ nvc0_bo_move_copy(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 
 static int
 nvc0_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        struct nvkm_mem *node = old_mem->mm_node;
-        u64 src_offset = node->vma[0].offset;
-        u64 dst_offset = node->vma[1].offset;
-        u32 page_count = new_mem->num_pages;
+        struct nvkm_mem *mem = old_reg->mm_node;
+        u64 src_offset = mem->vma[0].offset;
+        u64 dst_offset = mem->vma[1].offset;
+        u32 page_count = new_reg->num_pages;
         int ret;
 
-        page_count = new_mem->num_pages;
+        page_count = new_reg->num_pages;
         while (page_count) {
                 int line_count = (page_count > 2047) ? 2047 : page_count;
 
@@ -763,15 +766,15 @@ nvc0_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 
 static int
 nva3_bo_move_copy(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        struct nvkm_mem *node = old_mem->mm_node;
-        u64 src_offset = node->vma[0].offset;
-        u64 dst_offset = node->vma[1].offset;
-        u32 page_count = new_mem->num_pages;
+        struct nvkm_mem *mem = old_reg->mm_node;
+        u64 src_offset = mem->vma[0].offset;
+        u64 dst_offset = mem->vma[1].offset;
+        u32 page_count = new_reg->num_pages;
         int ret;
 
-        page_count = new_mem->num_pages;
+        page_count = new_reg->num_pages;
         while (page_count) {
                 int line_count = (page_count > 8191) ? 8191 : page_count;
 
@@ -801,35 +804,35 @@ nva3_bo_move_copy(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 
 static int
 nv98_bo_move_exec(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        struct nvkm_mem *node = old_mem->mm_node;
+        struct nvkm_mem *mem = old_reg->mm_node;
         int ret = RING_SPACE(chan, 7);
         if (ret == 0) {
                 BEGIN_NV04(chan, NvSubCopy, 0x0320, 6);
-                OUT_RING  (chan, upper_32_bits(node->vma[0].offset));
-                OUT_RING  (chan, lower_32_bits(node->vma[0].offset));
-                OUT_RING  (chan, upper_32_bits(node->vma[1].offset));
-                OUT_RING  (chan, lower_32_bits(node->vma[1].offset));
+                OUT_RING  (chan, upper_32_bits(mem->vma[0].offset));
+                OUT_RING  (chan, lower_32_bits(mem->vma[0].offset));
+                OUT_RING  (chan, upper_32_bits(mem->vma[1].offset));
+                OUT_RING  (chan, lower_32_bits(mem->vma[1].offset));
                 OUT_RING  (chan, 0x00000000 /* COPY */);
-                OUT_RING  (chan, new_mem->num_pages << PAGE_SHIFT);
+                OUT_RING  (chan, new_reg->num_pages << PAGE_SHIFT);
         }
         return ret;
 }
 
 static int
 nv84_bo_move_exec(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        struct nvkm_mem *node = old_mem->mm_node;
+        struct nvkm_mem *mem = old_reg->mm_node;
         int ret = RING_SPACE(chan, 7);
         if (ret == 0) {
                 BEGIN_NV04(chan, NvSubCopy, 0x0304, 6);
-                OUT_RING  (chan, new_mem->num_pages << PAGE_SHIFT);
-                OUT_RING  (chan, upper_32_bits(node->vma[0].offset));
-                OUT_RING  (chan, lower_32_bits(node->vma[0].offset));
-                OUT_RING  (chan, upper_32_bits(node->vma[1].offset));
-                OUT_RING  (chan, lower_32_bits(node->vma[1].offset));
+                OUT_RING  (chan, new_reg->num_pages << PAGE_SHIFT);
+                OUT_RING  (chan, upper_32_bits(mem->vma[0].offset));
+                OUT_RING  (chan, lower_32_bits(mem->vma[0].offset));
+                OUT_RING  (chan, upper_32_bits(mem->vma[1].offset));
+                OUT_RING  (chan, lower_32_bits(mem->vma[1].offset));
                 OUT_RING  (chan, 0x00000000 /* MODE_COPY, QUERY_NONE */);
         }
         return ret;
@@ -853,14 +856,14 @@ nv50_bo_move_init(struct nouveau_channel *chan, u32 handle)
 
 static int
 nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        struct nvkm_mem *node = old_mem->mm_node;
-        u64 length = (new_mem->num_pages << PAGE_SHIFT);
-        u64 src_offset = node->vma[0].offset;
-        u64 dst_offset = node->vma[1].offset;
-        int src_tiled = !!node->memtype;
-        int dst_tiled = !!((struct nvkm_mem *)new_mem->mm_node)->memtype;
+        struct nvkm_mem *mem = old_reg->mm_node;
+        u64 length = (new_reg->num_pages << PAGE_SHIFT);
+        u64 src_offset = mem->vma[0].offset;
+        u64 dst_offset = mem->vma[1].offset;
+        int src_tiled = !!mem->memtype;
+        int dst_tiled = !!((struct nvkm_mem *)new_reg->mm_node)->memtype;
         int ret;
 
         while (length) {
@@ -940,20 +943,20 @@ nv04_bo_move_init(struct nouveau_channel *chan, u32 handle)
 
 static inline uint32_t
 nouveau_bo_mem_ctxdma(struct ttm_buffer_object *bo,
-                      struct nouveau_channel *chan, struct ttm_mem_reg *mem)
+                      struct nouveau_channel *chan, struct ttm_mem_reg *reg)
 {
-        if (mem->mem_type == TTM_PL_TT)
+        if (reg->mem_type == TTM_PL_TT)
                 return NvDmaTT;
         return chan->vram.handle;
 }
 
 static int
 nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
-                  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
+                  struct ttm_mem_reg *old_reg, struct ttm_mem_reg *new_reg)
 {
-        u32 src_offset = old_mem->start << PAGE_SHIFT;
-        u32 dst_offset = new_mem->start << PAGE_SHIFT;
-        u32 page_count = new_mem->num_pages;
+        u32 src_offset = old_reg->start << PAGE_SHIFT;
+        u32 dst_offset = new_reg->start << PAGE_SHIFT;
+        u32 page_count = new_reg->num_pages;
         int ret;
 
         ret = RING_SPACE(chan, 3);
@@ -961,10 +964,10 @@ nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
                 return ret;
 
         BEGIN_NV04(chan, NvSubCopy, NV_MEMORY_TO_MEMORY_FORMAT_DMA_SOURCE, 2);
-        OUT_RING  (chan, nouveau_bo_mem_ctxdma(bo, chan, old_mem));
-        OUT_RING  (chan, nouveau_bo_mem_ctxdma(bo, chan, new_mem));
+        OUT_RING  (chan, nouveau_bo_mem_ctxdma(bo, chan, old_reg));
+        OUT_RING  (chan, nouveau_bo_mem_ctxdma(bo, chan, new_reg));
 
-        page_count = new_mem->num_pages;
+        page_count = new_reg->num_pages;
         while (page_count) {
                 int line_count = (page_count > 2047) ? 2047 : page_count;
 
@@ -995,33 +998,33 @@ nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 
 static int
 nouveau_bo_move_prep(struct nouveau_drm *drm, struct ttm_buffer_object *bo,
-                     struct ttm_mem_reg *mem)
+                     struct ttm_mem_reg *reg)
 {
-        struct nvkm_mem *old_node = bo->mem.mm_node;
-        struct nvkm_mem *new_node = mem->mm_node;
-        u64 size = (u64)mem->num_pages << PAGE_SHIFT;
+        struct nvkm_mem *old_mem = bo->mem.mm_node;
+        struct nvkm_mem *new_mem = reg->mm_node;
+        u64 size = (u64)reg->num_pages << PAGE_SHIFT;
         int ret;
 
-        ret = nvkm_vm_get(drm->client.vm, size, old_node->page_shift,
-                          NV_MEM_ACCESS_RW, &old_node->vma[0]);
+        ret = nvkm_vm_get(drm->client.vm, size, old_mem->page_shift,
+                          NV_MEM_ACCESS_RW, &old_mem->vma[0]);
         if (ret)
                 return ret;
 
-        ret = nvkm_vm_get(drm->client.vm, size, new_node->page_shift,
-                          NV_MEM_ACCESS_RW, &old_node->vma[1]);
+        ret = nvkm_vm_get(drm->client.vm, size, new_mem->page_shift,
+                          NV_MEM_ACCESS_RW, &old_mem->vma[1]);
         if (ret) {
-                nvkm_vm_put(&old_node->vma[0]);
+                nvkm_vm_put(&old_mem->vma[0]);
                 return ret;
         }
 
-        nvkm_vm_map(&old_node->vma[0], old_node);
-        nvkm_vm_map(&old_node->vma[1], new_node);
+        nvkm_vm_map(&old_mem->vma[0], old_mem);
+        nvkm_vm_map(&old_mem->vma[1], new_mem);
         return 0;
 }
 
 static int
 nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
-                     bool no_wait_gpu, struct ttm_mem_reg *new_mem)
+                     bool no_wait_gpu, struct ttm_mem_reg *new_reg)
 {
         struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
         struct nouveau_channel *chan = drm->ttm.chan;
@@ -1033,8 +1036,8 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
          * old nvkm_mem node, these will get cleaned up after ttm has
          * destroyed the ttm_mem_reg
          */
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
-                ret = nouveau_bo_move_prep(drm, bo, new_mem);
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+                ret = nouveau_bo_move_prep(drm, bo, new_reg);
                 if (ret)
                         return ret;
         }
@@ -1042,14 +1045,14 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
         mutex_lock_nested(&cli->mutex, SINGLE_DEPTH_NESTING);
         ret = nouveau_fence_sync(nouveau_bo(bo), chan, true, intr);
         if (ret == 0) {
-                ret = drm->ttm.move(chan, bo, &bo->mem, new_mem);
+                ret = drm->ttm.move(chan, bo, &bo->mem, new_reg);
                 if (ret == 0) {
                         ret = nouveau_fence_new(chan, false, &fence);
                         if (ret == 0) {
                                 ret = ttm_bo_move_accel_cleanup(bo,
                                                                 &fence->base,
                                                                 evict,
-                                                                new_mem);
+                                                                new_reg);
                                 nouveau_fence_unref(&fence);
                         }
                 }
@@ -1124,7 +1127,7 @@ nouveau_bo_move_init(struct nouveau_drm *drm)
 
 static int
 nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
-                      bool no_wait_gpu, struct ttm_mem_reg *new_mem)
+                      bool no_wait_gpu, struct ttm_mem_reg *new_reg)
 {
         struct ttm_place placement_memtype = {
                 .fpfn = 0,
@@ -1132,35 +1135,35 @@ nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
                 .flags = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING
         };
         struct ttm_placement placement;
-        struct ttm_mem_reg tmp_mem;
+        struct ttm_mem_reg tmp_reg;
         int ret;
 
         placement.num_placement = placement.num_busy_placement = 1;
         placement.placement = placement.busy_placement = &placement_memtype;
 
-        tmp_mem = *new_mem;
-        tmp_mem.mm_node = NULL;
-        ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait_gpu);
+        tmp_reg = *new_reg;
+        tmp_reg.mm_node = NULL;
+        ret = ttm_bo_mem_space(bo, &placement, &tmp_reg, intr, no_wait_gpu);
         if (ret)
                 return ret;
 
-        ret = ttm_tt_bind(bo->ttm, &tmp_mem);
+        ret = ttm_tt_bind(bo->ttm, &tmp_reg);
         if (ret)
                 goto out;
 
-        ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, &tmp_mem);
+        ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, &tmp_reg);
         if (ret)
                 goto out;
 
-        ret = ttm_bo_move_ttm(bo, intr, no_wait_gpu, new_mem);
+        ret = ttm_bo_move_ttm(bo, intr, no_wait_gpu, new_reg);
 out:
-        ttm_bo_mem_put(bo, &tmp_mem);
+        ttm_bo_mem_put(bo, &tmp_reg);
         return ret;
 }
 
 static int
 nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
-                      bool no_wait_gpu, struct ttm_mem_reg *new_mem)
+                      bool no_wait_gpu, struct ttm_mem_reg *new_reg)
 {
         struct ttm_place placement_memtype = {
                 .fpfn = 0,
@@ -1168,34 +1171,34 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
                 .flags = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING
         };
         struct ttm_placement placement;
-        struct ttm_mem_reg tmp_mem;
+        struct ttm_mem_reg tmp_reg;
         int ret;
 
         placement.num_placement = placement.num_busy_placement = 1;
         placement.placement = placement.busy_placement = &placement_memtype;
 
-        tmp_mem = *new_mem;
-        tmp_mem.mm_node = NULL;
-        ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait_gpu);
+        tmp_reg = *new_reg;
+        tmp_reg.mm_node = NULL;
+        ret = ttm_bo_mem_space(bo, &placement, &tmp_reg, intr, no_wait_gpu);
         if (ret)
                 return ret;
 
-        ret = ttm_bo_move_ttm(bo, intr, no_wait_gpu, &tmp_mem);
+        ret = ttm_bo_move_ttm(bo, intr, no_wait_gpu, &tmp_reg);
         if (ret)
                 goto out;
 
-        ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
+        ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_reg);
         if (ret)
                 goto out;
 
 out:
-        ttm_bo_mem_put(bo, &tmp_mem);
+        ttm_bo_mem_put(bo, &tmp_reg);
         return ret;
 }
 
 static void
 nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
-                     struct ttm_mem_reg *new_mem)
+                     struct ttm_mem_reg *new_reg)
 {
         struct nouveau_bo *nvbo = nouveau_bo(bo);
         struct nvkm_vma *vma;
@@ -1205,10 +1208,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
                 return;
 
         list_for_each_entry(vma, &nvbo->vma_list, head) {
-                if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
-                              (new_mem->mem_type == TTM_PL_VRAM ||
+                if (new_reg && new_reg->mem_type != TTM_PL_SYSTEM &&
+                              (new_reg->mem_type == TTM_PL_VRAM ||
                                nvbo->page_shift != vma->vm->mmu->lpg_shift)) {
-                        nvkm_vm_map(vma, new_mem->mm_node);
+                        nvkm_vm_map(vma, new_reg->mm_node);
                 } else {
                         WARN_ON(ttm_bo_wait(bo, false, false));
                         nvkm_vm_unmap(vma);
@@ -1217,20 +1220,20 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
 }
 
 static int
-nouveau_bo_vm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem,
+nouveau_bo_vm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_reg,
                    struct nouveau_drm_tile **new_tile)
 {
         struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
         struct drm_device *dev = drm->dev;
         struct nouveau_bo *nvbo = nouveau_bo(bo);
-        u64 offset = new_mem->start << PAGE_SHIFT;
+        u64 offset = new_reg->start << PAGE_SHIFT;
 
         *new_tile = NULL;
-        if (new_mem->mem_type != TTM_PL_VRAM)
+        if (new_reg->mem_type != TTM_PL_VRAM)
                 return 0;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
-                *new_tile = nv10_bo_set_tiling(dev, offset, new_mem->size,
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_CELSIUS) {
+                *new_tile = nv10_bo_set_tiling(dev, offset, new_reg->size,
                                                 nvbo->tile_mode,
                                                 nvbo->tile_flags);
         }
@@ -1253,11 +1256,11 @@ nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo,
 
 static int
 nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
-                bool no_wait_gpu, struct ttm_mem_reg *new_mem)
+                bool no_wait_gpu, struct ttm_mem_reg *new_reg)
 {
         struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
         struct nouveau_bo *nvbo = nouveau_bo(bo);
-        struct ttm_mem_reg *old_mem = &bo->mem;
+        struct ttm_mem_reg *old_reg = &bo->mem;
         struct nouveau_drm_tile *new_tile = NULL;
         int ret = 0;
 
@@ -1268,31 +1271,31 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
         if (nvbo->pin_refcnt)
                 NV_WARN(drm, "Moving pinned object %p!\n", nvbo);
 
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA) {
-                ret = nouveau_bo_vm_bind(bo, new_mem, &new_tile);
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA) {
+                ret = nouveau_bo_vm_bind(bo, new_reg, &new_tile);
                 if (ret)
                         return ret;
         }
 
         /* Fake bo copy. */
-        if (old_mem->mem_type == TTM_PL_SYSTEM && !bo->ttm) {
+        if (old_reg->mem_type == TTM_PL_SYSTEM && !bo->ttm) {
                 BUG_ON(bo->mem.mm_node != NULL);
-                bo->mem = *new_mem;
-                new_mem->mm_node = NULL;
+                bo->mem = *new_reg;
+                new_reg->mm_node = NULL;
                 goto out;
         }
 
         /* Hardware assisted copy. */
         if (drm->ttm.move) {
-                if (new_mem->mem_type == TTM_PL_SYSTEM)
+                if (new_reg->mem_type == TTM_PL_SYSTEM)
                         ret = nouveau_bo_move_flipd(bo, evict, intr,
-                                                    no_wait_gpu, new_mem);
-                else if (old_mem->mem_type == TTM_PL_SYSTEM)
+                                                    no_wait_gpu, new_reg);
+                else if (old_reg->mem_type == TTM_PL_SYSTEM)
                         ret = nouveau_bo_move_flips(bo, evict, intr,
-                                                    no_wait_gpu, new_mem);
+                                                    no_wait_gpu, new_reg);
                 else
                         ret = nouveau_bo_move_m2mf(bo, evict, intr,
-                                                   no_wait_gpu, new_mem);
+                                                   no_wait_gpu, new_reg);
                 if (!ret)
                         goto out;
         }
@@ -1300,10 +1303,10 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
         /* Fallback to software copy. */
         ret = ttm_bo_wait(bo, intr, no_wait_gpu);
         if (ret == 0)
-                ret = ttm_bo_move_memcpy(bo, intr, no_wait_gpu, new_mem);
+                ret = ttm_bo_move_memcpy(bo, intr, no_wait_gpu, new_reg);
 
 out:
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA) {
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA) {
                 if (ret)
                         nouveau_bo_vm_cleanup(bo, NULL, &new_tile);
                 else
@@ -1323,54 +1326,54 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp)
 }
 
 static int
-nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
 {
-        struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+        struct ttm_mem_type_manager *man = &bdev->man[reg->mem_type];
         struct nouveau_drm *drm = nouveau_bdev(bdev);
-        struct nvkm_device *device = nvxx_device(&drm->device);
-        struct nvkm_mem *node = mem->mm_node;
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
+        struct nvkm_mem *mem = reg->mm_node;
         int ret;
 
-        mem->bus.addr = NULL;
-        mem->bus.offset = 0;
-        mem->bus.size = mem->num_pages << PAGE_SHIFT;
-        mem->bus.base = 0;
-        mem->bus.is_iomem = false;
+        reg->bus.addr = NULL;
+        reg->bus.offset = 0;
+        reg->bus.size = reg->num_pages << PAGE_SHIFT;
+        reg->bus.base = 0;
+        reg->bus.is_iomem = false;
         if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
                 return -EINVAL;
-        switch (mem->mem_type) {
+        switch (reg->mem_type) {
         case TTM_PL_SYSTEM:
                 /* System memory */
                 return 0;
         case TTM_PL_TT:
 #if IS_ENABLED(CONFIG_AGP)
                 if (drm->agp.bridge) {
-                        mem->bus.offset = mem->start << PAGE_SHIFT;
-                        mem->bus.base = drm->agp.base;
-                        mem->bus.is_iomem = !drm->agp.cma;
+                        reg->bus.offset = reg->start << PAGE_SHIFT;
+                        reg->bus.base = drm->agp.base;
+                        reg->bus.is_iomem = !drm->agp.cma;
                 }
 #endif
-                if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA || !node->memtype)
+                if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA || !mem->memtype)
                         /* untiled */
                         break;
                 /* fallthrough, tiled memory */
         case TTM_PL_VRAM:
-                mem->bus.offset = mem->start << PAGE_SHIFT;
-                mem->bus.base = device->func->resource_addr(device, 1);
-                mem->bus.is_iomem = true;
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
-                        struct nvkm_bar *bar = nvxx_bar(&drm->device);
+                reg->bus.offset = reg->start << PAGE_SHIFT;
+                reg->bus.base = device->func->resource_addr(device, 1);
+                reg->bus.is_iomem = true;
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+                        struct nvkm_bar *bar = nvxx_bar(&drm->client.device);
                         int page_shift = 12;
-                        if (drm->device.info.family >= NV_DEVICE_INFO_V0_FERMI)
-                                page_shift = node->page_shift;
+                        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_FERMI)
+                                page_shift = mem->page_shift;
 
-                        ret = nvkm_bar_umap(bar, node->size << 12, page_shift,
-                                            &node->bar_vma);
+                        ret = nvkm_bar_umap(bar, mem->size << 12, page_shift,
+                                            &mem->bar_vma);
                         if (ret)
                                 return ret;
 
-                        nvkm_vm_map(&node->bar_vma, node);
-                        mem->bus.offset = node->bar_vma.offset;
+                        nvkm_vm_map(&mem->bar_vma, mem);
+                        reg->bus.offset = mem->bar_vma.offset;
                 }
                 break;
         default:
@@ -1380,15 +1383,15 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 }
 
 static void
-nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg)
 {
-        struct nvkm_mem *node = mem->mm_node;
+        struct nvkm_mem *mem = reg->mm_node;
 
-        if (!node->bar_vma.node)
+        if (!mem->bar_vma.node)
                 return;
 
-        nvkm_vm_unmap(&node->bar_vma);
-        nvkm_vm_put(&node->bar_vma);
+        nvkm_vm_unmap(&mem->bar_vma);
+        nvkm_vm_put(&mem->bar_vma);
 }
 
 static int
@@ -1396,7 +1399,7 @@ nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
 {
         struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
         struct nouveau_bo *nvbo = nouveau_bo(bo);
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
         u32 mappable = device->func->resource_size(device, 1) >> PAGE_SHIFT;
         int i, ret;
 
@@ -1404,7 +1407,7 @@ nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
          * nothing to do here.
          */
         if (bo->mem.mem_type != TTM_PL_VRAM) {
-                if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA ||
+                if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA ||
                     !nouveau_bo_tile_layout(nvbo))
                         return 0;
 
@@ -1419,7 +1422,7 @@ nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
         }
 
         /* make sure bo is in mappable vram */
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA ||
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA ||
             bo->mem.start + bo->mem.num_pages < mappable)
                 return 0;
 
@@ -1461,7 +1464,7 @@ nouveau_ttm_tt_populate(struct ttm_tt *ttm)
         }
 
         drm = nouveau_bdev(ttm->bdev);
-        device = nvxx_device(&drm->device);
+        device = nvxx_device(&drm->client.device);
         dev = drm->dev;
         pdev = device->dev;
 
@@ -1518,7 +1521,7 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
                 return;
 
         drm = nouveau_bdev(ttm->bdev);
-        device = nvxx_device(&drm->device);
+        device = nvxx_device(&drm->client.device);
         dev = drm->dev;
         pdev = device->dev;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index e42360983229..b06a5385d6dd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -26,6 +26,8 @@ struct nouveau_bo {
         struct list_head vma_list;
         unsigned page_shift;
 
+        struct nouveau_cli *cli;
+
         u32 tile_mode;
         u32 tile_flags;
         struct nouveau_drm_tile *tile;
@@ -69,7 +71,7 @@ nouveau_bo_ref(struct nouveau_bo *ref, struct nouveau_bo **pnvbo)
 extern struct ttm_bo_driver nouveau_bo_driver;
 
 void nouveau_bo_move_init(struct nouveau_drm *);
-int  nouveau_bo_new(struct drm_device *, int size, int align, u32 flags,
+int  nouveau_bo_new(struct nouveau_cli *, u64 size, int align, u32 flags,
                     u32 tile_mode, u32 tile_flags, struct sg_table *sg,
                     struct reservation_object *robj,
                     struct nouveau_bo **);
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index f9b3c811187e..dbc41fa86ee8 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -45,10 +45,20 @@ MODULE_PARM_DESC(vram_pushbuf, "Create DMA push buffers in VRAM");
 int nouveau_vram_pushbuf;
 module_param_named(vram_pushbuf, nouveau_vram_pushbuf, int, 0400);
 
+static int
+nouveau_channel_killed(struct nvif_notify *ntfy)
+{
+        struct nouveau_channel *chan = container_of(ntfy, typeof(*chan), kill);
+        struct nouveau_cli *cli = (void *)chan->user.client;
+        NV_PRINTK(warn, cli, "channel %d killed!\n", chan->chid);
+        atomic_set(&chan->killed, 1);
+        return NVIF_NOTIFY_DROP;
+}
+
 int
 nouveau_channel_idle(struct nouveau_channel *chan)
 {
-        if (likely(chan && chan->fence)) {
+        if (likely(chan && chan->fence && !atomic_read(&chan->killed))) {
                 struct nouveau_cli *cli = (void *)chan->user.client;
                 struct nouveau_fence *fence = NULL;
                 int ret;
@@ -78,6 +88,7 @@ nouveau_channel_del(struct nouveau_channel **pchan)
                 nvif_object_fini(&chan->nvsw);
                 nvif_object_fini(&chan->gart);
                 nvif_object_fini(&chan->vram);
+                nvif_notify_fini(&chan->kill);
                 nvif_object_fini(&chan->user);
                 nvif_object_fini(&chan->push.ctxdma);
                 nouveau_bo_vma_del(chan->push.buffer, &chan->push.vma);
@@ -107,13 +118,14 @@ nouveau_channel_prep(struct nouveau_drm *drm, struct nvif_device *device,
 
         chan->device = device;
         chan->drm = drm;
+        atomic_set(&chan->killed, 0);
 
         /* allocate memory for dma push buffer */
         target = TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED;
         if (nouveau_vram_pushbuf)
                 target = TTM_PL_FLAG_VRAM;
 
-        ret = nouveau_bo_new(drm->dev, size, 0, target, 0, 0, NULL, NULL,
+        ret = nouveau_bo_new(cli, size, 0, target, 0, 0, NULL, NULL,
                             &chan->push.buffer);
         if (ret == 0) {
                 ret = nouveau_bo_pin(chan->push.buffer, target, false);
@@ -301,12 +313,26 @@ nouveau_channel_init(struct nouveau_channel *chan, u32 vram, u32 gart)
 {
         struct nvif_device *device = chan->device;
         struct nouveau_cli *cli = (void *)chan->user.client;
+        struct nouveau_drm *drm = chan->drm;
         struct nvkm_mmu *mmu = nvxx_mmu(device);
         struct nv_dma_v0 args = {};
         int ret, i;
 
         nvif_object_map(&chan->user);
 
+        if (chan->user.oclass >= FERMI_CHANNEL_GPFIFO) {
+                ret = nvif_notify_init(&chan->user, nouveau_channel_killed,
+                                       true, NV906F_V0_NTFY_KILLED,
+                                       NULL, 0, 0, &chan->kill);
+                if (ret == 0)
+                        ret = nvif_notify_get(&chan->kill);
+                if (ret) {
+                        NV_ERROR(drm, "Failed to request channel kill "
+                                      "notification: %d\n", ret);
+                        return ret;
+                }
+        }
+
         /* allocate dma objects to cover all allowed vram, and gart */
         if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
                 if (device->info.family >= NV_DEVICE_INFO_V0_TESLA) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.h b/drivers/gpu/drm/nouveau/nouveau_chan.h
index 48062c94f36d..46b947ba1cf4 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.h
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.h
@@ -1,7 +1,7 @@
 #ifndef __NOUVEAU_CHAN_H__
 #define __NOUVEAU_CHAN_H__
-
 #include <nvif/object.h>
+#include <nvif/notify.h>
 struct nvif_device;
 
 struct nouveau_channel {
@@ -38,6 +38,9 @@ struct nouveau_channel {
         u32 user_put;
 
         struct nvif_object user;
+
+        struct nvif_notify kill;
+        atomic_t killed;
 };
 
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_connector.c b/drivers/gpu/drm/nouveau/nouveau_connector.c
index 966d20ab4de4..f5add64c093f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_connector.c
+++ b/drivers/gpu/drm/nouveau/nouveau_connector.c
@@ -419,7 +419,7 @@ nouveau_connector_ddc_detect(struct drm_connector *connector)
         struct drm_device *dev = connector->dev;
         struct nouveau_connector *nv_connector = nouveau_connector(connector);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_gpio *gpio = nvxx_gpio(&drm->device);
+        struct nvkm_gpio *gpio = nvxx_gpio(&drm->client.device);
         struct nouveau_encoder *nv_encoder;
         struct drm_encoder *encoder;
         int i, panel = -ENODEV;
@@ -521,7 +521,7 @@ nouveau_connector_set_encoder(struct drm_connector *connector,
                 return;
         nv_connector->detected_encoder = nv_encoder;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
                 connector->interlace_allowed = true;
                 connector->doublescan_allowed = true;
         } else
@@ -531,8 +531,8 @@ nouveau_connector_set_encoder(struct drm_connector *connector,
                 connector->interlace_allowed = false;
         } else {
                 connector->doublescan_allowed = true;
-                if (drm->device.info.family == NV_DEVICE_INFO_V0_KELVIN ||
-                    (drm->device.info.family == NV_DEVICE_INFO_V0_CELSIUS &&
+                if (drm->client.device.info.family == NV_DEVICE_INFO_V0_KELVIN ||
+                    (drm->client.device.info.family == NV_DEVICE_INFO_V0_CELSIUS &&
                      (dev->pdev->device & 0x0ff0) != 0x0100 &&
                      (dev->pdev->device & 0x0ff0) != 0x0150))
                         /* HW is broken */
@@ -984,17 +984,17 @@ get_tmds_link_bandwidth(struct drm_connector *connector, bool hdmi)
                 /* Note: these limits are conservative, some Fermi's
                  * can do 297 MHz. Unclear how this can be determined.
                  */
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_KEPLER)
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_KEPLER)
                         return 297000;
-                if (drm->device.info.family >= NV_DEVICE_INFO_V0_FERMI)
+                if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_FERMI)
                         return 225000;
         }
         if (dcb->location != DCB_LOC_ON_CHIP ||
-            drm->device.info.chipset >= 0x46)
+            drm->client.device.info.chipset >= 0x46)
                 return 165000;
-        else if (drm->device.info.chipset >= 0x40)
+        else if (drm->client.device.info.chipset >= 0x40)
                 return 155000;
-        else if (drm->device.info.chipset >= 0x18)
+        else if (drm->client.device.info.chipset >= 0x18)
                 return 135000;
         else
                 return 112000;
@@ -1041,7 +1041,7 @@ nouveau_connector_mode_valid(struct drm_connector *connector,
                 clock = clock * (connector->display_info.bpc * 3) / 10;
                 break;
         default:
-                BUG_ON(1);
+                BUG();
                 return MODE_BAD;
         }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_debugfs.c b/drivers/gpu/drm/nouveau/nouveau_debugfs.c
index 411c12cdb249..fd64dfdc7d4f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_debugfs.c
+++ b/drivers/gpu/drm/nouveau/nouveau_debugfs.c
@@ -259,8 +259,9 @@ nouveau_debugfs_init(struct nouveau_drm *drm)
         if (!drm->debugfs)
                 return -ENOMEM;
 
-        ret = nvif_object_init(&drm->device.object, 0, NVIF_CLASS_CONTROL,
-                               NULL, 0, &drm->debugfs->ctrl);
+        ret = nvif_object_init(&drm->client.device.object, 0,
+                               NVIF_CLASS_CONTROL, NULL, 0,
+                               &drm->debugfs->ctrl);
         if (ret)
                 return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
index 6b570079d185..72fdba1a1c5d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -414,7 +414,8 @@ nouveau_display_init(struct drm_device *dev)
                 return ret;
 
         /* enable polling for external displays */
-        drm_kms_helper_poll_enable(dev);
+        if (!dev->mode_config.poll_enabled)
+                drm_kms_helper_poll_enable(dev);
 
         /* enable hotplug interrupts */
         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
@@ -495,7 +496,7 @@ int
 nouveau_display_create(struct drm_device *dev)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
         struct nouveau_display *disp;
         int ret;
 
@@ -512,15 +513,15 @@ nouveau_display_create(struct drm_device *dev)
 
         dev->mode_config.min_width = 0;
         dev->mode_config.min_height = 0;
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_CELSIUS) {
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_CELSIUS) {
                 dev->mode_config.max_width = 2048;
                 dev->mode_config.max_height = 2048;
         } else
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA) {
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA) {
                 dev->mode_config.max_width = 4096;
                 dev->mode_config.max_height = 4096;
         } else
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_FERMI) {
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_FERMI) {
                 dev->mode_config.max_width = 8192;
                 dev->mode_config.max_height = 8192;
         } else {
@@ -531,7 +532,7 @@ nouveau_display_create(struct drm_device *dev)
         dev->mode_config.preferred_depth = 24;
         dev->mode_config.prefer_shadow = 1;
 
-        if (drm->device.info.chipset < 0x11)
+        if (drm->client.device.info.chipset < 0x11)
                 dev->mode_config.async_page_flip = false;
         else
                 dev->mode_config.async_page_flip = true;
@@ -558,7 +559,7 @@ nouveau_display_create(struct drm_device *dev)
                 int i;
 
                 for (i = 0, ret = -ENODEV; ret && i < ARRAY_SIZE(oclass); i++) {
-                        ret = nvif_object_init(&drm->device.object, 0,
+                        ret = nvif_object_init(&drm->client.device.object, 0,
                                                oclass[i], NULL, 0, &disp->disp);
                 }
 
@@ -1057,6 +1058,7 @@ int
 nouveau_display_dumb_create(struct drm_file *file_priv, struct drm_device *dev,
                             struct drm_mode_create_dumb *args)
 {
+        struct nouveau_cli *cli = nouveau_cli(file_priv);
         struct nouveau_bo *bo;
         uint32_t domain;
         int ret;
@@ -1066,12 +1068,12 @@ nouveau_display_dumb_create(struct drm_file *file_priv, struct drm_device *dev,
         args->size = roundup(args->size, PAGE_SIZE);
 
         /* Use VRAM if there is any ; otherwise fallback to system memory */
-        if (nouveau_drm(dev)->device.info.ram_size != 0)
+        if (nouveau_drm(dev)->client.device.info.ram_size != 0)
                 domain = NOUVEAU_GEM_DOMAIN_VRAM;
         else
                 domain = NOUVEAU_GEM_DOMAIN_GART;
 
-        ret = nouveau_gem_new(dev, args->size, 0, domain, 0, 0, &bo);
+        ret = nouveau_gem_new(cli, args->size, 0, domain, 0, 0, &bo);
         if (ret)
                 return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index 279497b15e7b..d234a3b70bad 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -37,6 +37,8 @@
 #include <core/pci.h>
 #include <core/tegra.h>
 
+#include <nvif/driver.h>
+
 #include <nvif/class.h>
 #include <nvif/cl0002.h>
 #include <nvif/cla06f.h>
@@ -109,35 +111,53 @@ nouveau_name(struct drm_device *dev)
                 return nouveau_platform_name(dev->platformdev);
 }
 
+static void
+nouveau_cli_fini(struct nouveau_cli *cli)
+{
+        nvkm_vm_ref(NULL, &nvxx_client(&cli->base)->vm, NULL);
+        usif_client_fini(cli);
+        nvif_device_fini(&cli->device);
+        nvif_client_fini(&cli->base);
+}
+
 static int
-nouveau_cli_create(struct drm_device *dev, const char *sname,
-                   int size, void **pcli)
+nouveau_cli_init(struct nouveau_drm *drm, const char *sname,
+                 struct nouveau_cli *cli)
 {
-        struct nouveau_cli *cli = *pcli = kzalloc(size, GFP_KERNEL);
+        u64 device = nouveau_name(drm->dev);
         int ret;
-        if (cli) {
-                snprintf(cli->name, sizeof(cli->name), "%s", sname);
-                cli->dev = dev;
 
-                ret = nvif_client_init(NULL, cli->name, nouveau_name(dev),
-                                       nouveau_config, nouveau_debug,
+        snprintf(cli->name, sizeof(cli->name), "%s", sname);
+        cli->dev = drm->dev;
+        mutex_init(&cli->mutex);
+        usif_client_init(cli);
+
+        if (cli == &drm->client) {
+                ret = nvif_driver_init(NULL, nouveau_config, nouveau_debug,
+                                       cli->name, device, &cli->base);
+        } else {
+                ret = nvif_client_init(&drm->client.base, cli->name, device,
                                        &cli->base);
-                if (ret == 0) {
-                        mutex_init(&cli->mutex);
-                        usif_client_init(cli);
-                }
-                return ret;
         }
-        return -ENOMEM;
-}
+        if (ret) {
+                NV_ERROR(drm, "Client allocation failed: %d\n", ret);
+                goto done;
+        }
 
-static void
-nouveau_cli_destroy(struct nouveau_cli *cli)
-{
-        nvkm_vm_ref(NULL, &nvxx_client(&cli->base)->vm, NULL);
-        nvif_client_fini(&cli->base);
-        usif_client_fini(cli);
-        kfree(cli);
+        ret = nvif_device_init(&cli->base.object, 0, NV_DEVICE,
+                               &(struct nv_device_v0) {
+                                        .device = ~0,
+                               }, sizeof(struct nv_device_v0),
+                               &cli->device);
+        if (ret) {
+                NV_ERROR(drm, "Device allocation failed: %d\n", ret);
+                goto done;
+        }
+
+done:
+        if (ret)
+                nouveau_cli_fini(cli);
+        return ret;
 }
 
 static void
@@ -161,7 +181,7 @@ nouveau_accel_fini(struct nouveau_drm *drm)
 static void
 nouveau_accel_init(struct nouveau_drm *drm)
 {
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         struct nvif_sclass *sclass;
         u32 arg0, arg1;
         int ret, i, n;
@@ -215,7 +235,7 @@ nouveau_accel_init(struct nouveau_drm *drm)
         }
 
         if (device->info.family >= NV_DEVICE_INFO_V0_KEPLER) {
-                ret = nouveau_channel_new(drm, &drm->device,
+                ret = nouveau_channel_new(drm, &drm->client.device,
                                           NVA06F_V0_ENGINE_CE0 |
                                           NVA06F_V0_ENGINE_CE1,
                                           0, &drm->cechan);
@@ -228,7 +248,7 @@ nouveau_accel_init(struct nouveau_drm *drm)
         if (device->info.chipset >= 0xa3 &&
             device->info.chipset != 0xaa &&
             device->info.chipset != 0xac) {
-                ret = nouveau_channel_new(drm, &drm->device,
+                ret = nouveau_channel_new(drm, &drm->client.device,
                                           NvDmaFB, NvDmaTT, &drm->cechan);
                 if (ret)
                         NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
@@ -240,7 +260,8 @@ nouveau_accel_init(struct nouveau_drm *drm)
                 arg1 = NvDmaTT;
         }
 
-        ret = nouveau_channel_new(drm, &drm->device, arg0, arg1, &drm->channel);
+        ret = nouveau_channel_new(drm, &drm->client.device,
+                                  arg0, arg1, &drm->channel);
         if (ret) {
                 NV_ERROR(drm, "failed to create kernel channel, %d\n", ret);
                 nouveau_accel_fini(drm);
@@ -280,8 +301,8 @@ nouveau_accel_init(struct nouveau_drm *drm)
         }
 
         if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
-                ret = nvkm_gpuobj_new(nvxx_device(&drm->device), 32, 0, false,
-                                      NULL, &drm->notify);
+                ret = nvkm_gpuobj_new(nvxx_device(&drm->client.device), 32, 0,
+                                      false, NULL, &drm->notify);
                 if (ret) {
                         NV_ERROR(drm, "failed to allocate notifier, %d\n", ret);
                         nouveau_accel_fini(drm);
@@ -407,12 +428,17 @@ nouveau_drm_load(struct drm_device *dev, unsigned long flags)
         struct nouveau_drm *drm;
         int ret;
 
-        ret = nouveau_cli_create(dev, "DRM", sizeof(*drm), (void **)&drm);
+        if (!(drm = kzalloc(sizeof(*drm), GFP_KERNEL)))
+                return -ENOMEM;
+        dev->dev_private = drm;
+        drm->dev = dev;
+
+        ret = nouveau_cli_init(drm, "DRM", &drm->client);
         if (ret)
                 return ret;
 
-        dev->dev_private = drm;
-        drm->dev = dev;
+        dev->irq_enabled = true;
+
         nvxx_client(&drm->client.base)->debug =
                 nvkm_dbgopt(nouveau_debug, "DRM");
 
@@ -421,33 +447,24 @@ nouveau_drm_load(struct drm_device *dev, unsigned long flags)
 
         nouveau_get_hdmi_dev(drm);
 
-        ret = nvif_device_init(&drm->client.base.object, 0, NV_DEVICE,
-                               &(struct nv_device_v0) {
-                                        .device = ~0,
-                               }, sizeof(struct nv_device_v0),
-                               &drm->device);
-        if (ret)
-                goto fail_device;
-
-        dev->irq_enabled = true;
-
         /* workaround an odd issue on nvc1 by disabling the device's
          * nosnoop capability.  hopefully won't cause issues until a
          * better fix is found - assuming there is one...
          */
-        if (drm->device.info.chipset == 0xc1)
-                nvif_mask(&drm->device.object, 0x00088080, 0x00000800, 0x00000000);
+        if (drm->client.device.info.chipset == 0xc1)
+                nvif_mask(&drm->client.device.object, 0x00088080, 0x00000800, 0x00000000);
 
         nouveau_vga_init(drm);
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
-                if (!nvxx_device(&drm->device)->mmu) {
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+                if (!nvxx_device(&drm->client.device)->mmu) {
                         ret = -ENOSYS;
                         goto fail_device;
                 }
 
-                ret = nvkm_vm_new(nvxx_device(&drm->device), 0, (1ULL << 40),
-                                  0x1000, NULL, &drm->client.vm);
+                ret = nvkm_vm_new(nvxx_device(&drm->client.device),
+                                  0, (1ULL << 40), 0x1000, NULL,
+                                  &drm->client.vm);
                 if (ret)
                         goto fail_device;
 
@@ -497,8 +514,8 @@ fail_bios:
 fail_ttm:
         nouveau_vga_fini(drm);
 fail_device:
-        nvif_device_fini(&drm->device);
-        nouveau_cli_destroy(&drm->client);
+        nouveau_cli_fini(&drm->client);
+        kfree(drm);
         return ret;
 }
 
@@ -527,10 +544,10 @@ nouveau_drm_unload(struct drm_device *dev)
         nouveau_ttm_fini(drm);
         nouveau_vga_fini(drm);
 
-        nvif_device_fini(&drm->device);
         if (drm->hdmi_device)
                 pci_dev_put(drm->hdmi_device);
-        nouveau_cli_destroy(&drm->client);
+        nouveau_cli_fini(&drm->client);
+        kfree(drm);
 }
 
 void
@@ -560,7 +577,6 @@ static int
 nouveau_do_suspend(struct drm_device *dev, bool runtime)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nouveau_cli *cli;
         int ret;
 
         nouveau_led_suspend(dev);
@@ -590,7 +606,7 @@ nouveau_do_suspend(struct drm_device *dev, bool runtime)
                         goto fail_display;
         }
 
-        NV_INFO(drm, "suspending client object trees...\n");
+        NV_INFO(drm, "suspending fence...\n");
         if (drm->fence && nouveau_fence(drm)->suspend) {
                 if (!nouveau_fence(drm)->suspend(drm)) {
                         ret = -ENOMEM;
@@ -598,13 +614,7 @@ nouveau_do_suspend(struct drm_device *dev, bool runtime)
                 }
         }
 
-        list_for_each_entry(cli, &drm->clients, head) {
-                ret = nvif_client_suspend(&cli->base);
-                if (ret)
-                        goto fail_client;
-        }
-
-        NV_INFO(drm, "suspending kernel object tree...\n");
+        NV_INFO(drm, "suspending object tree...\n");
         ret = nvif_client_suspend(&drm->client.base);
         if (ret)
                 goto fail_client;
@@ -612,10 +622,6 @@ nouveau_do_suspend(struct drm_device *dev, bool runtime)
         return 0;
 
 fail_client:
-        list_for_each_entry_continue_reverse(cli, &drm->clients, head) {
-                nvif_client_resume(&cli->base);
-        }
-
         if (drm->fence && nouveau_fence(drm)->resume)
                 nouveau_fence(drm)->resume(drm);
 
@@ -631,19 +637,14 @@ static int
 nouveau_do_resume(struct drm_device *dev, bool runtime)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nouveau_cli *cli;
 
-        NV_INFO(drm, "resuming kernel object tree...\n");
+        NV_INFO(drm, "resuming object tree...\n");
         nvif_client_resume(&drm->client.base);
 
-        NV_INFO(drm, "resuming client object trees...\n");
+        NV_INFO(drm, "resuming fence...\n");
         if (drm->fence && nouveau_fence(drm)->resume)
                 nouveau_fence(drm)->resume(drm);
 
-        list_for_each_entry(cli, &drm->clients, head) {
-                nvif_client_resume(&cli->base);
-        }
-
         nouveau_run_vbios_init(dev);
 
         if (dev->mode_config.num_crtc) {
@@ -758,7 +759,7 @@ nouveau_pmops_runtime_resume(struct device *dev)
 {
         struct pci_dev *pdev = to_pci_dev(dev);
         struct drm_device *drm_dev = pci_get_drvdata(pdev);
-        struct nvif_device *device = &nouveau_drm(drm_dev)->device;
+        struct nvif_device *device = &nouveau_drm(drm_dev)->client.device;
         int ret;
 
         if (nouveau_runtime_pm == 0)
@@ -772,7 +773,10 @@ nouveau_pmops_runtime_resume(struct device *dev)
         pci_set_master(pdev);
 
         ret = nouveau_do_resume(drm_dev, true);
-        drm_kms_helper_poll_enable(drm_dev);
+
+        if (!drm_dev->mode_config.poll_enabled)
+                drm_kms_helper_poll_enable(drm_dev);
+
         /* do magic */
         nvif_mask(&device->object, 0x088488, (1 << 25), (1 << 25));
         vga_switcheroo_set_dynamic_switch(pdev, VGA_SWITCHEROO_ON);
@@ -841,20 +845,20 @@ nouveau_drm_open(struct drm_device *dev, struct drm_file *fpriv)
         get_task_comm(tmpname, current);
         snprintf(name, sizeof(name), "%s[%d]", tmpname, pid_nr(fpriv->pid));
 
-        ret = nouveau_cli_create(dev, name, sizeof(*cli), (void **)&cli);
+        if (!(cli = kzalloc(sizeof(*cli), GFP_KERNEL)))
+                return ret;
 
+        ret = nouveau_cli_init(drm, name, cli);
         if (ret)
-                goto out_suspend;
+                goto done;
 
         cli->base.super = false;
 
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
-                ret = nvkm_vm_new(nvxx_device(&drm->device), 0, (1ULL << 40),
-                                  0x1000, NULL, &cli->vm);
-                if (ret) {
-                        nouveau_cli_destroy(cli);
-                        goto out_suspend;
-                }
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+                ret = nvkm_vm_new(nvxx_device(&drm->client.device), 0,
+                                  (1ULL << 40), 0x1000, NULL, &cli->vm);
+                if (ret)
+                        goto done;
 
                 nvxx_client(&cli->base)->vm = cli->vm;
         }
@@ -865,10 +869,14 @@ nouveau_drm_open(struct drm_device *dev, struct drm_file *fpriv)
         list_add(&cli->head, &drm->clients);
         mutex_unlock(&drm->client.mutex);
 
-out_suspend:
+done:
+        if (ret && cli) {
+                nouveau_cli_fini(cli);
+                kfree(cli);
+        }
+
         pm_runtime_mark_last_busy(dev->dev);
         pm_runtime_put_autosuspend(dev->dev);
-
         return ret;
 }
 
@@ -895,7 +903,8 @@ static void
 nouveau_drm_postclose(struct drm_device *dev, struct drm_file *fpriv)
 {
         struct nouveau_cli *cli = nouveau_cli(fpriv);
-        nouveau_cli_destroy(cli);
+        nouveau_cli_fini(cli);
+        kfree(cli);
         pm_runtime_mark_last_busy(dev->dev);
         pm_runtime_put_autosuspend(dev->dev);
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 8d5ed5bfdacb..eadec2f49ad3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -86,14 +86,17 @@ enum nouveau_drm_handle {
 
 struct nouveau_cli {
         struct nvif_client base;
+        struct drm_device *dev;
+        struct mutex mutex;
+
+        struct nvif_device device;
+
         struct nvkm_vm *vm; /*XXX*/
         struct list_head head;
-        struct mutex mutex;
         void *abi16;
         struct list_head objects;
         struct list_head notifys;
         char name[32];
-        struct drm_device *dev;
 };
 
 static inline struct nouveau_cli *
@@ -111,7 +114,6 @@ struct nouveau_drm {
         struct nouveau_cli client;
         struct drm_device *dev;
 
-        struct nvif_device device;
         struct list_head clients;
 
         struct {
@@ -165,6 +167,8 @@ struct nouveau_drm {
         struct backlight_device *backlight;
         struct list_head bl_connectors;
         struct work_struct hpd_work;
+        struct work_struct fbcon_work;
+        int fbcon_new_state;
 #ifdef CONFIG_ACPI
         struct notifier_block acpi_nb;
 #endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
index a9d48100e74f..2665a078b6da 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
@@ -60,7 +60,7 @@ nouveau_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
 {
         struct nouveau_fbdev *fbcon = info->par;
         struct nouveau_drm *drm = nouveau_drm(fbcon->helper.dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         int ret;
 
         if (info->state != FBINFO_STATE_RUNNING)
@@ -92,7 +92,7 @@ nouveau_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *image)
 {
         struct nouveau_fbdev *fbcon = info->par;
         struct nouveau_drm *drm = nouveau_drm(fbcon->helper.dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         int ret;
 
         if (info->state != FBINFO_STATE_RUNNING)
@@ -124,7 +124,7 @@ nouveau_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 {
         struct nouveau_fbdev *fbcon = info->par;
         struct nouveau_drm *drm = nouveau_drm(fbcon->helper.dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         int ret;
 
         if (info->state != FBINFO_STATE_RUNNING)
@@ -266,10 +266,10 @@ nouveau_fbcon_accel_init(struct drm_device *dev)
         struct fb_info *info = fbcon->helper.fbdev;
         int ret;
 
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA)
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA)
                 ret = nv04_fbcon_accel_init(info);
         else
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_FERMI)
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_FERMI)
                 ret = nv50_fbcon_accel_init(info);
         else
                 ret = nvc0_fbcon_accel_init(info);
@@ -324,7 +324,7 @@ nouveau_fbcon_create(struct drm_fb_helper *helper,
                 container_of(helper, struct nouveau_fbdev, helper);
         struct drm_device *dev = fbcon->helper.dev;
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         struct fb_info *info;
         struct nouveau_framebuffer *fb;
         struct nouveau_channel *chan;
@@ -341,8 +341,9 @@ nouveau_fbcon_create(struct drm_fb_helper *helper,
         mode_cmd.pixel_format = drm_mode_legacy_fb_format(sizes->surface_bpp,
                                                           sizes->surface_depth);
 
-        ret = nouveau_gem_new(dev, mode_cmd.pitches[0] * mode_cmd.height,
-                              0, NOUVEAU_GEM_DOMAIN_VRAM, 0, 0x0000, &nvbo);
+        ret = nouveau_gem_new(&drm->client, mode_cmd.pitches[0] *
+                              mode_cmd.height, 0, NOUVEAU_GEM_DOMAIN_VRAM,
+                              0, 0x0000, &nvbo);
         if (ret) {
                 NV_ERROR(drm, "failed to allocate framebuffer\n");
                 goto out;
@@ -471,19 +472,43 @@ static const struct drm_fb_helper_funcs nouveau_fbcon_helper_funcs = {
         .fb_probe = nouveau_fbcon_create,
 };
 
+static void
+nouveau_fbcon_set_suspend_work(struct work_struct *work)
+{
+        struct nouveau_drm *drm = container_of(work, typeof(*drm), fbcon_work);
+        int state = READ_ONCE(drm->fbcon_new_state);
+
+        if (state == FBINFO_STATE_RUNNING)
+                pm_runtime_get_sync(drm->dev->dev);
+
+        console_lock();
+        if (state == FBINFO_STATE_RUNNING)
+                nouveau_fbcon_accel_restore(drm->dev);
+        drm_fb_helper_set_suspend(&drm->fbcon->helper, state);
+        if (state != FBINFO_STATE_RUNNING)
+                nouveau_fbcon_accel_save_disable(drm->dev);
+        console_unlock();
+
+        if (state == FBINFO_STATE_RUNNING) {
+                pm_runtime_mark_last_busy(drm->dev->dev);
+                pm_runtime_put_sync(drm->dev->dev);
+        }
+}
+
 void
 nouveau_fbcon_set_suspend(struct drm_device *dev, int state)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        if (drm->fbcon) {
-                console_lock();
-                if (state == FBINFO_STATE_RUNNING)
-                        nouveau_fbcon_accel_restore(dev);
-                drm_fb_helper_set_suspend(&drm->fbcon->helper, state);
-                if (state != FBINFO_STATE_RUNNING)
-                        nouveau_fbcon_accel_save_disable(dev);
-                console_unlock();
-        }
+
+        if (!drm->fbcon)
+                return;
+
+        drm->fbcon_new_state = state;
+        /* Since runtime resume can happen as a result of a sysfs operation,
+         * it's possible we already have the console locked. So handle fbcon
+         * init/deinit from a seperate work thread
+         */
+        schedule_work(&drm->fbcon_work);
 }
 
 int
@@ -503,6 +528,7 @@ nouveau_fbcon_init(struct drm_device *dev)
                 return -ENOMEM;
 
         drm->fbcon = fbcon;
+        INIT_WORK(&drm->fbcon_work, nouveau_fbcon_set_suspend_work);
 
         drm_fb_helper_prepare(dev, &fbcon->helper, &nouveau_fbcon_helper_funcs);
 
@@ -514,10 +540,10 @@ nouveau_fbcon_init(struct drm_device *dev)
         if (ret)
                 goto fini;
 
-        if (drm->device.info.ram_size <= 32 * 1024 * 1024)
+        if (drm->client.device.info.ram_size <= 32 * 1024 * 1024)
                 preferred_bpp = 8;
         else
-        if (drm->device.info.ram_size <= 64 * 1024 * 1024)
+        if (drm->client.device.info.ram_size <= 64 * 1024 * 1024)
                 preferred_bpp = 16;
         else
                 preferred_bpp = 32;
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index a6126c93f215..f3e551f1aa46 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -190,7 +190,7 @@ nouveau_fence_context_new(struct nouveau_channel *chan, struct nouveau_fence_cha
                 return;
 
         ret = nvif_notify_init(&chan->user, nouveau_fence_wait_uevent_handler,
-                               false, G82_CHANNEL_DMA_V0_NTFY_UEVENT,
+                               false, NV826E_V0_NTFY_NON_STALL_INTERRUPT,
                                &(struct nvif_notify_uevent_req) { },
                                sizeof(struct nvif_notify_uevent_req),
                                sizeof(struct nvif_notify_uevent_rep),
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.h b/drivers/gpu/drm/nouveau/nouveau_fence.h
index ccdce1b4eec4..d5e58a38f160 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.h
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.h
@@ -99,6 +99,7 @@ struct nv84_fence_priv {
         struct nouveau_bo *bo;
         struct nouveau_bo *bo_gart;
         u32 *suspend;
+        struct mutex mutex;
 };
 
 int  nv84_fence_context_new(struct nouveau_channel *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 201b52b750dd..ca5397beb357 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -175,11 +175,11 @@ nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 }
 
 int
-nouveau_gem_new(struct drm_device *dev, int size, int align, uint32_t domain,
+nouveau_gem_new(struct nouveau_cli *cli, u64 size, int align, uint32_t domain,
                 uint32_t tile_mode, uint32_t tile_flags,
                 struct nouveau_bo **pnvbo)
 {
-        struct nouveau_drm *drm = nouveau_drm(dev);
+        struct nouveau_drm *drm = nouveau_drm(cli->dev);
         struct nouveau_bo *nvbo;
         u32 flags = 0;
         int ret;
@@ -194,7 +194,7 @@ nouveau_gem_new(struct drm_device *dev, int size, int align, uint32_t domain,
         if (domain & NOUVEAU_GEM_DOMAIN_COHERENT)
                 flags |= TTM_PL_FLAG_UNCACHED;
 
-        ret = nouveau_bo_new(dev, size, align, flags, tile_mode,
+        ret = nouveau_bo_new(cli, size, align, flags, tile_mode,
                              tile_flags, NULL, NULL, pnvbo);
         if (ret)
                 return ret;
@@ -206,12 +206,12 @@ nouveau_gem_new(struct drm_device *dev, int size, int align, uint32_t domain,
          */
         nvbo->valid_domains = NOUVEAU_GEM_DOMAIN_VRAM |
                               NOUVEAU_GEM_DOMAIN_GART;
-        if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA)
+        if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA)
                 nvbo->valid_domains &= domain;
 
         /* Initialize the embedded gem-object. We return a single gem-reference
          * to the caller, instead of a normal nouveau_bo ttm reference. */
-        ret = drm_gem_object_init(dev, &nvbo->gem, nvbo->bo.mem.size);
+        ret = drm_gem_object_init(drm->dev, &nvbo->gem, nvbo->bo.mem.size);
         if (ret) {
                 nouveau_bo_ref(NULL, pnvbo);
                 return -ENOMEM;
@@ -257,7 +257,7 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
         struct nouveau_cli *cli = nouveau_cli(file_priv);
-        struct nvkm_fb *fb = nvxx_fb(&drm->device);
+        struct nvkm_fb *fb = nvxx_fb(&drm->client.device);
         struct drm_nouveau_gem_new *req = data;
         struct nouveau_bo *nvbo = NULL;
         int ret = 0;
@@ -267,7 +267,7 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
                 return -EINVAL;
         }
 
-        ret = nouveau_gem_new(dev, req->info.size, req->align,
+        ret = nouveau_gem_new(cli, req->info.size, req->align,
                               req->info.domain, req->info.tile_mode,
                               req->info.tile_flags, &nvbo);
         if (ret)
@@ -496,7 +496,7 @@ validate_list(struct nouveau_channel *chan, struct nouveau_cli *cli,
                         return ret;
                 }
 
-                if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA) {
+                if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA) {
                         if (nvbo->bo.offset == b->presumed.offset &&
                             ((nvbo->bo.mem.mem_type == TTM_PL_VRAM &&
                               b->presumed.domain & NOUVEAU_GEM_DOMAIN_VRAM) ||
@@ -767,7 +767,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
                                       push[i].length);
                 }
         } else
-        if (drm->device.info.chipset >= 0x25) {
+        if (drm->client.device.info.chipset >= 0x25) {
                 ret = RING_SPACE(chan, req->nr_push * 2);
                 if (ret) {
                         NV_PRINTK(err, cli, "cal_space: %d\n", ret);
@@ -840,7 +840,7 @@ out_next:
                 req->suffix0 = 0x00000000;
                 req->suffix1 = 0x00000000;
         } else
-        if (drm->device.info.chipset >= 0x25) {
+        if (drm->client.device.info.chipset >= 0x25) {
                 req->suffix0 = 0x00020000;
                 req->suffix1 = 0x00000000;
         } else {
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.h b/drivers/gpu/drm/nouveau/nouveau_gem.h
index 7e32da2e037a..8fa6ed9ddd3a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.h
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.h
@@ -16,7 +16,7 @@ nouveau_gem_object(struct drm_gem_object *gem)
 }
 
 /* nouveau_gem.c */
-extern int nouveau_gem_new(struct drm_device *, int size, int align,
+extern int nouveau_gem_new(struct nouveau_cli *, u64 size, int align,
                            uint32_t domain, uint32_t tile_mode,
                            uint32_t tile_flags, struct nouveau_bo **);
 extern void nouveau_gem_object_del(struct drm_gem_object *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_hwmon.c b/drivers/gpu/drm/nouveau/nouveau_hwmon.c
index 71f764bf4cc6..23b1670c1c2f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_hwmon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_hwmon.c
@@ -43,7 +43,7 @@ nouveau_hwmon_show_temp(struct device *d, struct device_attribute *a, char *buf)
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         int temp = nvkm_therm_temp_get(therm);
 
         if (temp < 0)
@@ -69,7 +69,7 @@ nouveau_hwmon_temp1_auto_point1_temp(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
               therm->attr_get(therm, NVKM_THERM_ATTR_THRS_FAN_BOOST) * 1000);
@@ -81,7 +81,7 @@ nouveau_hwmon_set_temp1_auto_point1_temp(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -102,7 +102,7 @@ nouveau_hwmon_temp1_auto_point1_temp_hyst(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
          therm->attr_get(therm, NVKM_THERM_ATTR_THRS_FAN_BOOST_HYST) * 1000);
@@ -114,7 +114,7 @@ nouveau_hwmon_set_temp1_auto_point1_temp_hyst(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -134,7 +134,7 @@ nouveau_hwmon_max_temp(struct device *d, struct device_attribute *a, char *buf)
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
                therm->attr_get(therm, NVKM_THERM_ATTR_THRS_DOWN_CLK) * 1000);
@@ -145,7 +145,7 @@ nouveau_hwmon_set_max_temp(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -165,7 +165,7 @@ nouveau_hwmon_max_temp_hyst(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
           therm->attr_get(therm, NVKM_THERM_ATTR_THRS_DOWN_CLK_HYST) * 1000);
@@ -176,7 +176,7 @@ nouveau_hwmon_set_max_temp_hyst(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -197,7 +197,7 @@ nouveau_hwmon_critical_temp(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
                therm->attr_get(therm, NVKM_THERM_ATTR_THRS_CRITICAL) * 1000);
@@ -209,7 +209,7 @@ nouveau_hwmon_set_critical_temp(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -230,7 +230,7 @@ nouveau_hwmon_critical_temp_hyst(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
           therm->attr_get(therm, NVKM_THERM_ATTR_THRS_CRITICAL_HYST) * 1000);
@@ -243,7 +243,7 @@ nouveau_hwmon_set_critical_temp_hyst(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -263,7 +263,7 @@ nouveau_hwmon_emergency_temp(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
                therm->attr_get(therm, NVKM_THERM_ATTR_THRS_SHUTDOWN) * 1000);
@@ -275,7 +275,7 @@ nouveau_hwmon_set_emergency_temp(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -296,7 +296,7 @@ nouveau_hwmon_emergency_temp_hyst(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n",
           therm->attr_get(therm, NVKM_THERM_ATTR_THRS_SHUTDOWN_HYST) * 1000);
@@ -309,7 +309,7 @@ nouveau_hwmon_set_emergency_temp_hyst(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
 
         if (kstrtol(buf, 10, &value) == -EINVAL)
@@ -349,7 +349,7 @@ nouveau_hwmon_show_fan1_input(struct device *d, struct device_attribute *attr,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
 
         return snprintf(buf, PAGE_SIZE, "%d\n", nvkm_therm_fan_sense(therm));
 }
@@ -362,7 +362,7 @@ nouveau_hwmon_get_pwm1_enable(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         int ret;
 
         ret = therm->attr_get(therm, NVKM_THERM_ATTR_FAN_MODE);
@@ -378,7 +378,7 @@ nouveau_hwmon_set_pwm1_enable(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
         int ret;
 
@@ -401,7 +401,7 @@ nouveau_hwmon_get_pwm1(struct device *d, struct device_attribute *a, char *buf)
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         int ret;
 
         ret = therm->fan_get(therm);
@@ -417,7 +417,7 @@ nouveau_hwmon_set_pwm1(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         int ret = -ENODEV;
         long value;
 
@@ -441,7 +441,7 @@ nouveau_hwmon_get_pwm1_min(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         int ret;
 
         ret = therm->attr_get(therm, NVKM_THERM_ATTR_FAN_MIN_DUTY);
@@ -457,7 +457,7 @@ nouveau_hwmon_set_pwm1_min(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
         int ret;
 
@@ -481,7 +481,7 @@ nouveau_hwmon_get_pwm1_max(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         int ret;
 
         ret = therm->attr_get(therm, NVKM_THERM_ATTR_FAN_MAX_DUTY);
@@ -497,7 +497,7 @@ nouveau_hwmon_set_pwm1_max(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
         long value;
         int ret;
 
@@ -521,7 +521,7 @@ nouveau_hwmon_get_in0_input(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_volt *volt = nvxx_volt(&drm->device);
+        struct nvkm_volt *volt = nvxx_volt(&drm->client.device);
         int ret;
 
         ret = nvkm_volt_get(volt);
@@ -540,7 +540,7 @@ nouveau_hwmon_get_in0_min(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_volt *volt = nvxx_volt(&drm->device);
+        struct nvkm_volt *volt = nvxx_volt(&drm->client.device);
 
         if (!volt || !volt->min_uv)
                 return -ENODEV;
@@ -557,7 +557,7 @@ nouveau_hwmon_get_in0_max(struct device *d,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_volt *volt = nvxx_volt(&drm->device);
+        struct nvkm_volt *volt = nvxx_volt(&drm->client.device);
 
         if (!volt || !volt->max_uv)
                 return -ENODEV;
@@ -584,7 +584,7 @@ nouveau_hwmon_get_power1_input(struct device *d, struct device_attribute *a,
 {
         struct drm_device *dev = dev_get_drvdata(d);
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_iccsense *iccsense = nvxx_iccsense(&drm->device);
+        struct nvkm_iccsense *iccsense = nvxx_iccsense(&drm->client.device);
         int result = nvkm_iccsense_read_all(iccsense);
 
         if (result < 0)
@@ -596,6 +596,32 @@ nouveau_hwmon_get_power1_input(struct device *d, struct device_attribute *a,
 static SENSOR_DEVICE_ATTR(power1_input, S_IRUGO,
                           nouveau_hwmon_get_power1_input, NULL, 0);
 
+static ssize_t
+nouveau_hwmon_get_power1_max(struct device *d, struct device_attribute *a,
+                             char *buf)
+{
+        struct drm_device *dev = dev_get_drvdata(d);
+        struct nouveau_drm *drm = nouveau_drm(dev);
+        struct nvkm_iccsense *iccsense = nvxx_iccsense(&drm->client.device);
+        return sprintf(buf, "%i\n", iccsense->power_w_max);
+}
+
+static SENSOR_DEVICE_ATTR(power1_max, S_IRUGO,
+                          nouveau_hwmon_get_power1_max, NULL, 0);
+
+static ssize_t
+nouveau_hwmon_get_power1_crit(struct device *d, struct device_attribute *a,
+                              char *buf)
+{
+        struct drm_device *dev = dev_get_drvdata(d);
+        struct nouveau_drm *drm = nouveau_drm(dev);
+        struct nvkm_iccsense *iccsense = nvxx_iccsense(&drm->client.device);
+        return sprintf(buf, "%i\n", iccsense->power_w_crit);
+}
+
+static SENSOR_DEVICE_ATTR(power1_crit, S_IRUGO,
+                          nouveau_hwmon_get_power1_crit, NULL, 0);
+
 static struct attribute *hwmon_default_attributes[] = {
         &sensor_dev_attr_name.dev_attr.attr,
         &sensor_dev_attr_update_rate.dev_attr.attr,
@@ -639,6 +665,12 @@ static struct attribute *hwmon_power_attributes[] = {
         NULL
 };
 
+static struct attribute *hwmon_power_caps_attributes[] = {
+        &sensor_dev_attr_power1_max.dev_attr.attr,
+        &sensor_dev_attr_power1_crit.dev_attr.attr,
+        NULL
+};
+
 static const struct attribute_group hwmon_default_attrgroup = {
         .attrs = hwmon_default_attributes,
 };
@@ -657,6 +689,9 @@ static const struct attribute_group hwmon_in0_attrgroup = {
 static const struct attribute_group hwmon_power_attrgroup = {
         .attrs = hwmon_power_attributes,
 };
+static const struct attribute_group hwmon_power_caps_attrgroup = {
+        .attrs = hwmon_power_caps_attributes,
+};
 #endif
 
 int
@@ -664,9 +699,9 @@ nouveau_hwmon_init(struct drm_device *dev)
 {
 #if defined(CONFIG_HWMON) || (defined(MODULE) && defined(CONFIG_HWMON_MODULE))
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_therm *therm = nvxx_therm(&drm->device);
-        struct nvkm_volt *volt = nvxx_volt(&drm->device);
-        struct nvkm_iccsense *iccsense = nvxx_iccsense(&drm->device);
+        struct nvkm_therm *therm = nvxx_therm(&drm->client.device);
+        struct nvkm_volt *volt = nvxx_volt(&drm->client.device);
+        struct nvkm_iccsense *iccsense = nvxx_iccsense(&drm->client.device);
         struct nouveau_hwmon *hwmon;
         struct device *hwmon_dev;
         int ret = 0;
@@ -728,8 +763,16 @@ nouveau_hwmon_init(struct drm_device *dev)
         if (iccsense && iccsense->data_valid && !list_empty(&iccsense->rails)) {
                 ret = sysfs_create_group(&hwmon_dev->kobj,
                                          &hwmon_power_attrgroup);
+
                 if (ret)
                         goto error;
+
+                if (iccsense->power_w_max && iccsense->power_w_crit) {
+                        ret = sysfs_create_group(&hwmon_dev->kobj,
+                                                 &hwmon_power_caps_attrgroup);
+                        if (ret)
+                                goto error;
+                }
         }
 
         hwmon->hwmon = hwmon_dev;
@@ -759,6 +802,7 @@ nouveau_hwmon_fini(struct drm_device *dev)
                 sysfs_remove_group(&hwmon->hwmon->kobj, &hwmon_fan_rpm_attrgroup);
                 sysfs_remove_group(&hwmon->hwmon->kobj, &hwmon_in0_attrgroup);
                 sysfs_remove_group(&hwmon->hwmon->kobj, &hwmon_power_attrgroup);
+                sysfs_remove_group(&hwmon->hwmon->kobj, &hwmon_power_caps_attrgroup);
 
                 hwmon_device_unregister(hwmon->hwmon);
         }
diff --git a/drivers/gpu/drm/nouveau/nouveau_led.c b/drivers/gpu/drm/nouveau/nouveau_led.c
index 3e2f1b6cd4df..2c5e0628da12 100644
--- a/drivers/gpu/drm/nouveau/nouveau_led.c
+++ b/drivers/gpu/drm/nouveau/nouveau_led.c
@@ -38,7 +38,7 @@ nouveau_led_get_brightness(struct led_classdev *led)
 {
         struct drm_device *drm_dev = container_of(led, struct nouveau_led, led)->dev;
         struct nouveau_drm *drm = nouveau_drm(drm_dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
         u32 div, duty;
 
         div =  nvif_rd32(device, 0x61c880) & 0x00ffffff;
@@ -55,7 +55,7 @@ nouveau_led_set_brightness(struct led_classdev *led, enum led_brightness value)
 {
         struct drm_device *drm_dev = container_of(led, struct nouveau_led, led)->dev;
         struct nouveau_drm *drm = nouveau_drm(drm_dev);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
 
         u32 input_clk = 27e6; /* PDISPLAY.SOR[1].PWM is connected to the crystal */
         u32 freq = 100; /* this is what nvidia uses and it should be good-enough */
@@ -78,7 +78,7 @@ int
 nouveau_led_init(struct drm_device *dev)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvkm_gpio *gpio = nvxx_gpio(&drm->device);
+        struct nvkm_gpio *gpio = nvxx_gpio(&drm->client.device);
         struct dcb_gpio_func logo_led;
         int ret;
 
@@ -102,6 +102,7 @@ nouveau_led_init(struct drm_device *dev)
         ret = led_classdev_register(dev->dev, &drm->led->led);
         if (ret) {
                 kfree(drm->led);
+                drm->led = NULL;
                 return ret;
         }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_led.h b/drivers/gpu/drm/nouveau/nouveau_led.h
index 187ecdb82002..21a5775028cc 100644
--- a/drivers/gpu/drm/nouveau/nouveau_led.h
+++ b/drivers/gpu/drm/nouveau/nouveau_led.h
@@ -42,7 +42,7 @@ nouveau_led(struct drm_device *dev)
 }
 
 /* nouveau_led.c */
-#if IS_ENABLED(CONFIG_LEDS_CLASS)
+#if IS_REACHABLE(CONFIG_LEDS_CLASS)
 int  nouveau_led_init(struct drm_device *dev);
 void nouveau_led_suspend(struct drm_device *dev);
 void nouveau_led_resume(struct drm_device *dev);
diff --git a/drivers/gpu/drm/nouveau/nouveau_nvif.c b/drivers/gpu/drm/nouveau/nouveau_nvif.c
index 15f0925ea13b..b3f29b1ce9ea 100644
--- a/drivers/gpu/drm/nouveau/nouveau_nvif.c
+++ b/drivers/gpu/drm/nouveau/nouveau_nvif.c
@@ -60,20 +60,15 @@ nvkm_client_ioctl(void *priv, bool super, void *data, u32 size, void **hack)
 static int
 nvkm_client_resume(void *priv)
 {
-        return nvkm_client_init(priv);
+        struct nvkm_client *client = priv;
+        return nvkm_object_init(&client->object);
 }
 
 static int
 nvkm_client_suspend(void *priv)
 {
-        return nvkm_client_fini(priv, true);
-}
-
-static void
-nvkm_client_driver_fini(void *priv)
-{
         struct nvkm_client *client = priv;
-        nvkm_client_del(&client);
+        return nvkm_object_fini(&client->object, true);
 }
 
 static int
@@ -108,23 +103,14 @@ static int
 nvkm_client_driver_init(const char *name, u64 device, const char *cfg,
                         const char *dbg, void **ppriv)
 {
-        struct nvkm_client *client;
-        int ret;
-
-        ret = nvkm_client_new(name, device, cfg, dbg, &client);
-        *ppriv = client;
-        if (ret)
-                return ret;
-
-        client->ntfy = nvkm_client_ntfy;
-        return 0;
+        return nvkm_client_new(name, device, cfg, dbg, nvkm_client_ntfy,
+                               (struct nvkm_client **)ppriv);
 }
 
 const struct nvif_driver
 nvif_driver_nvkm = {
         .name = "nvkm",
         .init = nvkm_client_driver_init,
-        .fini = nvkm_client_driver_fini,
         .suspend = nvkm_client_suspend,
         .resume = nvkm_client_resume,
         .ioctl = nvkm_client_ioctl,
diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c
index a0a9704cfe2b..1fefc93af1d7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_prime.c
+++ b/drivers/gpu/drm/nouveau/nouveau_prime.c
@@ -60,6 +60,7 @@ struct drm_gem_object *nouveau_gem_prime_import_sg_table(struct drm_device *dev,
                                                          struct dma_buf_attachment *attach,
                                                          struct sg_table *sg)
 {
+        struct nouveau_drm *drm = nouveau_drm(dev);
         struct nouveau_bo *nvbo;
         struct reservation_object *robj = attach->dmabuf->resv;
         u32 flags = 0;
@@ -68,7 +69,7 @@ struct drm_gem_object *nouveau_gem_prime_import_sg_table(struct drm_device *dev,
         flags = TTM_PL_FLAG_TT;
 
         ww_mutex_lock(&robj->lock, NULL);
-        ret = nouveau_bo_new(dev, attach->dmabuf->size, 0, flags, 0, 0,
+        ret = nouveau_bo_new(&drm->client, attach->dmabuf->size, 0, flags, 0, 0,
                              sg, robj, &nvbo);
         ww_mutex_unlock(&robj->lock);
         if (ret)
diff --git a/drivers/gpu/drm/nouveau/nouveau_sgdma.c b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
index db35ab5883ac..b7ab268f7d6f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_sgdma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
@@ -24,10 +24,10 @@ nouveau_sgdma_destroy(struct ttm_tt *ttm)
 }
 
 static int
-nv04_sgdma_bind(struct ttm_tt *ttm, struct ttm_mem_reg *mem)
+nv04_sgdma_bind(struct ttm_tt *ttm, struct ttm_mem_reg *reg)
 {
         struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)ttm;
-        struct nvkm_mem *node = mem->mm_node;
+        struct nvkm_mem *node = reg->mm_node;
 
         if (ttm->sg) {
                 node->sg    = ttm->sg;
@@ -36,7 +36,7 @@ nv04_sgdma_bind(struct ttm_tt *ttm, struct ttm_mem_reg *mem)
                 node->sg    = NULL;
                 node->pages = nvbe->ttm.dma_address;
         }
-        node->size = (mem->num_pages << PAGE_SHIFT) >> 12;
+        node->size = (reg->num_pages << PAGE_SHIFT) >> 12;
 
         nvkm_vm_map(&node->vma[0], node);
         nvbe->node = node;
@@ -58,10 +58,10 @@ static struct ttm_backend_func nv04_sgdma_backend = {
 };
 
 static int
-nv50_sgdma_bind(struct ttm_tt *ttm, struct ttm_mem_reg *mem)
+nv50_sgdma_bind(struct ttm_tt *ttm, struct ttm_mem_reg *reg)
 {
         struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)ttm;
-        struct nvkm_mem *node = mem->mm_node;
+        struct nvkm_mem *node = reg->mm_node;
 
         /* noop: bound in move_notify() */
         if (ttm->sg) {
@@ -71,7 +71,7 @@ nv50_sgdma_bind(struct ttm_tt *ttm, struct ttm_mem_reg *mem)
                 node->sg    = NULL;
                 node->pages = nvbe->ttm.dma_address;
         }
-        node->size = (mem->num_pages << PAGE_SHIFT) >> 12;
+        node->size = (reg->num_pages << PAGE_SHIFT) >> 12;
         return 0;
 }
 
@@ -100,7 +100,7 @@ nouveau_sgdma_create_ttm(struct ttm_bo_device *bdev,
         if (!nvbe)
                 return NULL;
 
-        if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA)
+        if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA)
                 nvbe->ttm.ttm.func = &nv04_sgdma_backend;
         else
                 nvbe->ttm.ttm.func = &nv50_sgdma_backend;
diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c
index ec4668a41e01..13e5cc5f07fe 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ttm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c
@@ -36,7 +36,7 @@ static int
 nouveau_vram_manager_init(struct ttm_mem_type_manager *man, unsigned long psize)
 {
         struct nouveau_drm *drm = nouveau_bdev(man->bdev);
-        struct nvkm_fb *fb = nvxx_fb(&drm->device);
+        struct nvkm_fb *fb = nvxx_fb(&drm->client.device);
         man->priv = fb;
         return 0;
 }
@@ -64,45 +64,45 @@ nvkm_mem_node_cleanup(struct nvkm_mem *node)
 
 static void
 nouveau_vram_manager_del(struct ttm_mem_type_manager *man,
-                         struct ttm_mem_reg *mem)
+                         struct ttm_mem_reg *reg)
 {
         struct nouveau_drm *drm = nouveau_bdev(man->bdev);
-        struct nvkm_ram *ram = nvxx_fb(&drm->device)->ram;
-        nvkm_mem_node_cleanup(mem->mm_node);
-        ram->func->put(ram, (struct nvkm_mem **)&mem->mm_node);
+        struct nvkm_ram *ram = nvxx_fb(&drm->client.device)->ram;
+        nvkm_mem_node_cleanup(reg->mm_node);
+        ram->func->put(ram, (struct nvkm_mem **)&reg->mm_node);
 }
 
 static int
 nouveau_vram_manager_new(struct ttm_mem_type_manager *man,
                          struct ttm_buffer_object *bo,
                          const struct ttm_place *place,
-                         struct ttm_mem_reg *mem)
+                         struct ttm_mem_reg *reg)
 {
         struct nouveau_drm *drm = nouveau_bdev(man->bdev);
-        struct nvkm_ram *ram = nvxx_fb(&drm->device)->ram;
+        struct nvkm_ram *ram = nvxx_fb(&drm->client.device)->ram;
         struct nouveau_bo *nvbo = nouveau_bo(bo);
         struct nvkm_mem *node;
         u32 size_nc = 0;
         int ret;
 
-        if (drm->device.info.ram_size == 0)
+        if (drm->client.device.info.ram_size == 0)
                 return -ENOMEM;
 
         if (nvbo->tile_flags & NOUVEAU_GEM_TILE_NONCONTIG)
                 size_nc = 1 << nvbo->page_shift;
 
-        ret = ram->func->get(ram, mem->num_pages << PAGE_SHIFT,
-                             mem->page_alignment << PAGE_SHIFT, size_nc,
+        ret = ram->func->get(ram, reg->num_pages << PAGE_SHIFT,
+                             reg->page_alignment << PAGE_SHIFT, size_nc,
                              (nvbo->tile_flags >> 8) & 0x3ff, &node);
         if (ret) {
-                mem->mm_node = NULL;
+                reg->mm_node = NULL;
                 return (ret == -ENOSPC) ? 0 : ret;
         }
 
         node->page_shift = nvbo->page_shift;
 
-        mem->mm_node = node;
-        mem->start   = node->offset >> PAGE_SHIFT;
+        reg->mm_node = node;
+        reg->start   = node->offset >> PAGE_SHIFT;
         return 0;
 }
 
@@ -127,18 +127,18 @@ nouveau_gart_manager_fini(struct ttm_mem_type_manager *man)
 
 static void
 nouveau_gart_manager_del(struct ttm_mem_type_manager *man,
-                         struct ttm_mem_reg *mem)
+                         struct ttm_mem_reg *reg)
 {
-        nvkm_mem_node_cleanup(mem->mm_node);
-        kfree(mem->mm_node);
-        mem->mm_node = NULL;
+        nvkm_mem_node_cleanup(reg->mm_node);
+        kfree(reg->mm_node);
+        reg->mm_node = NULL;
 }
 
 static int
 nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
                          struct ttm_buffer_object *bo,
                          const struct ttm_place *place,
-                         struct ttm_mem_reg *mem)
+                         struct ttm_mem_reg *reg)
 {
         struct nouveau_drm *drm = nouveau_bdev(bo->bdev);
         struct nouveau_bo *nvbo = nouveau_bo(bo);
@@ -150,7 +150,7 @@ nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
 
         node->page_shift = 12;
 
-        switch (drm->device.info.family) {
+        switch (drm->client.device.info.family) {
         case NV_DEVICE_INFO_V0_TNT:
         case NV_DEVICE_INFO_V0_CELSIUS:
         case NV_DEVICE_INFO_V0_KELVIN:
@@ -158,7 +158,7 @@ nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
         case NV_DEVICE_INFO_V0_CURIE:
                 break;
         case NV_DEVICE_INFO_V0_TESLA:
-                if (drm->device.info.chipset != 0x50)
+                if (drm->client.device.info.chipset != 0x50)
                         node->memtype = (nvbo->tile_flags & 0x7f00) >> 8;
                 break;
         case NV_DEVICE_INFO_V0_FERMI:
@@ -169,12 +169,12 @@ nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
                 break;
         default:
                 NV_WARN(drm, "%s: unhandled family type %x\n", __func__,
-                        drm->device.info.family);
+                        drm->client.device.info.family);
                 break;
         }
 
-        mem->mm_node = node;
-        mem->start   = 0;
+        reg->mm_node = node;
+        reg->start   = 0;
         return 0;
 }
 
@@ -197,7 +197,7 @@ static int
 nv04_gart_manager_init(struct ttm_mem_type_manager *man, unsigned long psize)
 {
         struct nouveau_drm *drm = nouveau_bdev(man->bdev);
-        struct nvkm_mmu *mmu = nvxx_mmu(&drm->device);
+        struct nvkm_mmu *mmu = nvxx_mmu(&drm->client.device);
         struct nv04_mmu *priv = (void *)mmu;
         struct nvkm_vm *vm = NULL;
         nvkm_vm_ref(priv->vm, &vm, NULL);
@@ -215,20 +215,20 @@ nv04_gart_manager_fini(struct ttm_mem_type_manager *man)
 }
 
 static void
-nv04_gart_manager_del(struct ttm_mem_type_manager *man, struct ttm_mem_reg *mem)
+nv04_gart_manager_del(struct ttm_mem_type_manager *man, struct ttm_mem_reg *reg)
 {
-        struct nvkm_mem *node = mem->mm_node;
+        struct nvkm_mem *node = reg->mm_node;
         if (node->vma[0].node)
                 nvkm_vm_put(&node->vma[0]);
-        kfree(mem->mm_node);
-        mem->mm_node = NULL;
+        kfree(reg->mm_node);
+        reg->mm_node = NULL;
 }
 
 static int
 nv04_gart_manager_new(struct ttm_mem_type_manager *man,
                       struct ttm_buffer_object *bo,
                       const struct ttm_place *place,
-                      struct ttm_mem_reg *mem)
+                      struct ttm_mem_reg *reg)
 {
         struct nvkm_mem *node;
         int ret;
@@ -239,15 +239,15 @@ nv04_gart_manager_new(struct ttm_mem_type_manager *man,
 
         node->page_shift = 12;
 
-        ret = nvkm_vm_get(man->priv, mem->num_pages << 12, node->page_shift,
+        ret = nvkm_vm_get(man->priv, reg->num_pages << 12, node->page_shift,
                           NV_MEM_ACCESS_RW, &node->vma[0]);
         if (ret) {
                 kfree(node);
                 return ret;
         }
 
-        mem->mm_node = node;
-        mem->start   = node->vma[0].offset >> PAGE_SHIFT;
+        reg->mm_node = node;
+        reg->start   = node->vma[0].offset >> PAGE_SHIFT;
         return 0;
 }
 
@@ -339,7 +339,7 @@ nouveau_ttm_global_release(struct nouveau_drm *drm)
 int
 nouveau_ttm_init(struct nouveau_drm *drm)
 {
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
         struct nvkm_pci *pci = device->pci;
         struct drm_device *dev = drm->dev;
         u8 bits;
@@ -352,8 +352,8 @@ nouveau_ttm_init(struct nouveau_drm *drm)
                 drm->agp.cma = pci->agp.cma;
         }
 
-        bits = nvxx_mmu(&drm->device)->dma_bits;
-        if (nvxx_device(&drm->device)->func->pci) {
+        bits = nvxx_mmu(&drm->client.device)->dma_bits;
+        if (nvxx_device(&drm->client.device)->func->pci) {
                 if (drm->agp.bridge)
                         bits = 32;
         } else if (device->func->tegra) {
@@ -396,7 +396,7 @@ nouveau_ttm_init(struct nouveau_drm *drm)
         }
 
         /* VRAM init */
-        drm->gem.vram_available = drm->device.info.ram_user;
+        drm->gem.vram_available = drm->client.device.info.ram_user;
 
         arch_io_reserve_memtype_wc(device->func->resource_addr(device, 1),
                                    device->func->resource_size(device, 1));
@@ -413,7 +413,7 @@ nouveau_ttm_init(struct nouveau_drm *drm)
 
         /* GART init */
         if (!drm->agp.bridge) {
-                drm->gem.gart_available = nvxx_mmu(&drm->device)->limit;
+                drm->gem.gart_available = nvxx_mmu(&drm->client.device)->limit;
         } else {
                 drm->gem.gart_available = drm->agp.size;
         }
@@ -433,7 +433,7 @@ nouveau_ttm_init(struct nouveau_drm *drm)
 void
 nouveau_ttm_fini(struct nouveau_drm *drm)
 {
-        struct nvkm_device *device = nvxx_device(&drm->device);
+        struct nvkm_device *device = nvxx_device(&drm->client.device);
 
         ttm_bo_clean_mm(&drm->ttm.bdev, TTM_PL_VRAM);
         ttm_bo_clean_mm(&drm->ttm.bdev, TTM_PL_TT);
diff --git a/drivers/gpu/drm/nouveau/nouveau_usif.c b/drivers/gpu/drm/nouveau/nouveau_usif.c
index 08f9c6fa0f7f..afbdbed1a690 100644
--- a/drivers/gpu/drm/nouveau/nouveau_usif.c
+++ b/drivers/gpu/drm/nouveau/nouveau_usif.c
@@ -103,7 +103,7 @@ usif_notify(const void *header, u32 length, const void *data, u32 size)
         }
                 break;
         default:
-                BUG_ON(1);
+                BUG();
                 break;
         }
 
@@ -313,7 +313,8 @@ usif_ioctl(struct drm_file *filp, void __user *user, u32 argc)
         if (!(ret = nvif_unpack(-ENOSYS, &data, &size, argv->v0, 0, 0, true))) {
                 /* block access to objects not created via this interface */
                 owner = argv->v0.owner;
-                if (argv->v0.object == 0ULL)
+                if (argv->v0.object == 0ULL &&
+                    argv->v0.type != NVIF_IOCTL_V0_DEL)
                         argv->v0.owner = NVDRM_OBJECT_ANY; /* except client */
                 else
                         argv->v0.owner = NVDRM_OBJECT_USIF;
diff --git a/drivers/gpu/drm/nouveau/nouveau_vga.c b/drivers/gpu/drm/nouveau/nouveau_vga.c
index c6a180a0c284..eef22c6b9665 100644
--- a/drivers/gpu/drm/nouveau/nouveau_vga.c
+++ b/drivers/gpu/drm/nouveau/nouveau_vga.c
@@ -13,13 +13,13 @@ static unsigned int
 nouveau_vga_set_decode(void *priv, bool state)
 {
         struct nouveau_drm *drm = nouveau_drm(priv);
-        struct nvif_object *device = &drm->device.object;
+        struct nvif_object *device = &drm->client.device.object;
 
-        if (drm->device.info.family == NV_DEVICE_INFO_V0_CURIE &&
-            drm->device.info.chipset >= 0x4c)
+        if (drm->client.device.info.family == NV_DEVICE_INFO_V0_CURIE &&
+            drm->client.device.info.chipset >= 0x4c)
                 nvif_wr32(device, 0x088060, state);
         else
-        if (drm->device.info.chipset >= 0x40)
+        if (drm->client.device.info.chipset >= 0x40)
                 nvif_wr32(device, 0x088054, state);
         else
                 nvif_wr32(device, 0x001854, state);
diff --git a/drivers/gpu/drm/nouveau/nv04_fbcon.c b/drivers/gpu/drm/nouveau/nv04_fbcon.c
index 6a2b187e3c3b..01731dbeb3d8 100644
--- a/drivers/gpu/drm/nouveau/nv04_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv04_fbcon.c
@@ -136,7 +136,7 @@ nv04_fbcon_accel_init(struct fb_info *info)
         struct drm_device *dev = nfbdev->helper.dev;
         struct nouveau_drm *drm = nouveau_drm(dev);
         struct nouveau_channel *chan = drm->channel;
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         int surface_fmt, pattern_fmt, rect_fmt;
         int ret;
 
diff --git a/drivers/gpu/drm/nouveau/nv17_fence.c b/drivers/gpu/drm/nouveau/nv17_fence.c
index 79bc01111351..6477b7069e14 100644
--- a/drivers/gpu/drm/nouveau/nv17_fence.c
+++ b/drivers/gpu/drm/nouveau/nv17_fence.c
@@ -76,9 +76,9 @@ nv17_fence_context_new(struct nouveau_channel *chan)
 {
         struct nv10_fence_priv *priv = chan->drm->fence;
         struct nv10_fence_chan *fctx;
-        struct ttm_mem_reg *mem = &priv->bo->bo.mem;
-        u32 start = mem->start * PAGE_SIZE;
-        u32 limit = start + mem->size - 1;
+        struct ttm_mem_reg *reg = &priv->bo->bo.mem;
+        u32 start = reg->start * PAGE_SIZE;
+        u32 limit = start + reg->size - 1;
         int ret = 0;
 
         fctx = chan->fence = kzalloc(sizeof(*fctx), GFP_KERNEL);
@@ -129,7 +129,7 @@ nv17_fence_create(struct nouveau_drm *drm)
         priv->base.context_base = dma_fence_context_alloc(priv->base.contexts);
         spin_lock_init(&priv->lock);
 
-        ret = nouveau_bo_new(drm->dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
+        ret = nouveau_bo_new(&drm->client, 4096, 0x1000, TTM_PL_FLAG_VRAM,
                              0, 0x0000, NULL, NULL, &priv->bo);
         if (!ret) {
                 ret = nouveau_bo_pin(priv->bo, TTM_PL_FLAG_VRAM, false);
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index 413b178c86b2..2517adbe7089 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -447,18 +447,18 @@ nv50_dmac_ctxdma_new(struct nv50_dmac *dmac, struct nouveau_framebuffer *fb)
         args.base.target = NV_DMA_V0_TARGET_VRAM;
         args.base.access = NV_DMA_V0_ACCESS_RDWR;
         args.base.start  = 0;
-        args.base.limit  = drm->device.info.ram_user - 1;
+        args.base.limit  = drm->client.device.info.ram_user - 1;
 
-        if (drm->device.info.chipset < 0x80) {
+        if (drm->client.device.info.chipset < 0x80) {
                 args.nv50.part = NV50_DMA_V0_PART_256;
                 argc += sizeof(args.nv50);
         } else
-        if (drm->device.info.chipset < 0xc0) {
+        if (drm->client.device.info.chipset < 0xc0) {
                 args.nv50.part = NV50_DMA_V0_PART_256;
                 args.nv50.kind = kind;
                 argc += sizeof(args.nv50);
         } else
-        if (drm->device.info.chipset < 0xd0) {
+        if (drm->client.device.info.chipset < 0xd0) {
                 args.gf100.kind = kind;
                 argc += sizeof(args.gf100);
         } else {
@@ -852,7 +852,7 @@ nv50_wndw_atomic_check_acquire(struct nv50_wndw *wndw,
 
         if (asyw->image.kind) {
                 asyw->image.layout = 0;
-                if (drm->device.info.chipset >= 0xc0)
+                if (drm->client.device.info.chipset >= 0xc0)
                         asyw->image.block = fb->nvbo->tile_mode >> 4;
                 else
                         asyw->image.block = fb->nvbo->tile_mode;
@@ -1404,7 +1404,7 @@ nv50_base_ntfy_wait_begun(struct nv50_wndw *wndw, struct nv50_wndw_atom *asyw)
 {
         struct nouveau_drm *drm = nouveau_drm(wndw->plane.dev);
         struct nv50_disp *disp = nv50_disp(wndw->plane.dev);
-        if (nvif_msec(&drm->device, 2000ULL,
+        if (nvif_msec(&drm->client.device, 2000ULL,
                 u32 data = nouveau_bo_rd32(disp->sync, asyw->ntfy.offset / 4);
                 if ((data & 0xc0000000) == 0x40000000)
                         break;
@@ -1529,7 +1529,7 @@ nv50_base_new(struct nouveau_drm *drm, struct nv50_head *head,
                 return ret;
         }
 
-        ret = nv50_base_create(&drm->device, disp->disp, base->id,
+        ret = nv50_base_create(&drm->client.device, disp->disp, base->id,
                                disp->sync->bo.offset, &base->chan);
         if (ret)
                 return ret;
@@ -2330,7 +2330,7 @@ static int
 nv50_head_create(struct drm_device *dev, int index)
 {
         struct nouveau_drm *drm = nouveau_drm(dev);
-        struct nvif_device *device = &drm->device;
+        struct nvif_device *device = &drm->client.device;
         struct nv50_disp *disp = nv50_disp(dev);
         struct nv50_head *head;
         struct nv50_base *base;
@@ -2364,7 +2364,7 @@ nv50_head_create(struct drm_device *dev, int index)
         drm_crtc_helper_add(crtc, &nv50_head_help);
         drm_mode_crtc_set_gamma_size(crtc, 256);
 
-        ret = nouveau_bo_new(dev, 8192, 0x100, TTM_PL_FLAG_VRAM,
+        ret = nouveau_bo_new(&drm->client, 8192, 0x100, TTM_PL_FLAG_VRAM,
                              0, 0x0000, NULL, NULL, &head->base.lut.nvbo);
         if (!ret) {
                 ret = nouveau_bo_pin(head->base.lut.nvbo, TTM_PL_FLAG_VRAM, true);
@@ -2603,7 +2603,7 @@ static int
 nv50_dac_create(struct drm_connector *connector, struct dcb_output *dcbe)
 {
         struct nouveau_drm *drm = nouveau_drm(connector->dev);
-        struct nvkm_i2c *i2c = nvxx_i2c(&drm->device);
+        struct nvkm_i2c *i2c = nvxx_i2c(&drm->client.device);
         struct nvkm_i2c_bus *bus;
         struct nouveau_encoder *nv_encoder;
         struct drm_encoder *encoder;
@@ -3559,7 +3559,7 @@ nv50_sor_enable(struct drm_encoder *encoder)
                 nv50_audio_enable(encoder, mode);
                 break;
         default:
-                BUG_ON(1);
+                BUG();
                 break;
         }
 
@@ -3593,7 +3593,7 @@ nv50_sor_create(struct drm_connector *connector, struct dcb_output *dcbe)
 {
         struct nouveau_connector *nv_connector = nouveau_connector(connector);
         struct nouveau_drm *drm = nouveau_drm(connector->dev);
-        struct nvkm_i2c *i2c = nvxx_i2c(&drm->device);
+        struct nvkm_i2c *i2c = nvxx_i2c(&drm->client.device);
         struct nouveau_encoder *nv_encoder;
         struct drm_encoder *encoder;
         int type, ret;
@@ -3732,7 +3732,7 @@ nv50_pior_enable(struct drm_encoder *encoder)
                 proto = 0x0;
                 break;
         default:
-                BUG_ON(1);
+                BUG();
                 break;
         }
 
@@ -3778,7 +3778,7 @@ static int
 nv50_pior_create(struct drm_connector *connector, struct dcb_output *dcbe)
 {
         struct nouveau_drm *drm = nouveau_drm(connector->dev);
-        struct nvkm_i2c *i2c = nvxx_i2c(&drm->device);
+        struct nvkm_i2c *i2c = nvxx_i2c(&drm->client.device);
         struct nvkm_i2c_bus *bus = NULL;
         struct nvkm_i2c_aux *aux = NULL;
         struct i2c_adapter *ddc;
@@ -3851,7 +3851,7 @@ nv50_disp_atomic_commit_core(struct nouveau_drm *drm, u32 interlock)
                 evo_data(push, 0x00000000);
                 nouveau_bo_wr32(disp->sync, 0, 0x00000000);
                 evo_kick(push, core);
-                if (nvif_msec(&drm->device, 2000ULL,
+                if (nvif_msec(&drm->client.device, 2000ULL,
                         if (nouveau_bo_rd32(disp->sync, 0))
                                 break;
                         usleep_range(1, 2);
@@ -3986,6 +3986,11 @@ nv50_disp_atomic_commit_tail(struct drm_atomic_state *state)
                 }
         }
 
+        for_each_crtc_in_state(state, crtc, crtc_state, i) {
+                if (crtc->state->event)
+                        drm_crtc_vblank_get(crtc);
+        }
+
         /* Update plane(s). */
         for_each_plane_in_state(state, plane, plane_state, i) {
                 struct nv50_wndw_atom *asyw = nv50_wndw_atom(plane->state);
@@ -4035,6 +4040,7 @@ nv50_disp_atomic_commit_tail(struct drm_atomic_state *state)
                         drm_crtc_send_vblank_event(crtc, crtc->state->event);
                         spin_unlock_irqrestore(&crtc->dev->event_lock, flags);
                         crtc->state->event = NULL;
+                        drm_crtc_vblank_put(crtc);
                 }
         }
 
@@ -4363,7 +4369,7 @@ module_param_named(atomic, nouveau_atomic, int, 0400);
 int
 nv50_display_create(struct drm_device *dev)
 {
-        struct nvif_device *device = &nouveau_drm(dev)->device;
+        struct nvif_device *device = &nouveau_drm(dev)->client.device;
         struct nouveau_drm *drm = nouveau_drm(dev);
         struct dcb_table *dcb = &drm->vbios.dcb;
         struct drm_connector *connector, *tmp;
@@ -4387,7 +4393,7 @@ nv50_display_create(struct drm_device *dev)
                 dev->driver->driver_features |= DRIVER_ATOMIC;
 
         /* small shared memory area we use for notifiers and semaphores */
-        ret = nouveau_bo_new(dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
+        ret = nouveau_bo_new(&drm->client, 4096, 0x1000, TTM_PL_FLAG_VRAM,
                              0, 0x0000, NULL, NULL, &disp->sync);
         if (!ret) {
                 ret = nouveau_bo_pin(disp->sync, TTM_PL_FLAG_VRAM, true);
diff --git a/drivers/gpu/drm/nouveau/nv50_fence.c b/drivers/gpu/drm/nouveau/nv50_fence.c
index f68c7054fd53..a369d978e267 100644
--- a/drivers/gpu/drm/nouveau/nv50_fence.c
+++ b/drivers/gpu/drm/nouveau/nv50_fence.c
@@ -37,9 +37,9 @@ nv50_fence_context_new(struct nouveau_channel *chan)
 {
         struct nv10_fence_priv *priv = chan->drm->fence;
         struct nv10_fence_chan *fctx;
-        struct ttm_mem_reg *mem = &priv->bo->bo.mem;
-        u32 start = mem->start * PAGE_SIZE;
-        u32 limit = start + mem->size - 1;
+        struct ttm_mem_reg *reg = &priv->bo->bo.mem;
+        u32 start = reg->start * PAGE_SIZE;
+        u32 limit = start + reg->size - 1;
         int ret;
 
         fctx = chan->fence = kzalloc(sizeof(*fctx), GFP_KERNEL);
@@ -82,7 +82,7 @@ nv50_fence_create(struct nouveau_drm *drm)
         priv->base.context_base = dma_fence_context_alloc(priv->base.contexts);
         spin_lock_init(&priv->lock);
 
-        ret = nouveau_bo_new(drm->dev, 4096, 0x1000, TTM_PL_FLAG_VRAM,
+        ret = nouveau_bo_new(&drm->client, 4096, 0x1000, TTM_PL_FLAG_VRAM,
                              0, 0x0000, NULL, NULL, &priv->bo);
         if (!ret) {
                 ret = nouveau_bo_pin(priv->bo, TTM_PL_FLAG_VRAM, false);
diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
index 52b87ae83e7b..bd7a8a1e4ad9 100644
--- a/drivers/gpu/drm/nouveau/nv84_fence.c
+++ b/drivers/gpu/drm/nouveau/nv84_fence.c
@@ -107,8 +107,10 @@ nv84_fence_context_del(struct nouveau_channel *chan)
         struct nv84_fence_chan *fctx = chan->fence;
 
         nouveau_bo_wr32(priv->bo, chan->chid * 16 / 4, fctx->base.sequence);
+        mutex_lock(&priv->mutex);
         nouveau_bo_vma_del(priv->bo, &fctx->vma_gart);
         nouveau_bo_vma_del(priv->bo, &fctx->vma);
+        mutex_unlock(&priv->mutex);
         nouveau_fence_context_del(&fctx->base);
         chan->fence = NULL;
         nouveau_fence_context_free(&fctx->base);
@@ -134,11 +136,13 @@ nv84_fence_context_new(struct nouveau_channel *chan)
         fctx->base.sync32 = nv84_fence_sync32;
         fctx->base.sequence = nv84_fence_read(chan);
 
+        mutex_lock(&priv->mutex);
         ret = nouveau_bo_vma_add(priv->bo, cli->vm, &fctx->vma);
         if (ret == 0) {
                 ret = nouveau_bo_vma_add(priv->bo_gart, cli->vm,
                                         &fctx->vma_gart);
         }
+        mutex_unlock(&priv->mutex);
 
         if (ret)
                 nv84_fence_context_del(chan);
@@ -193,7 +197,7 @@ nv84_fence_destroy(struct nouveau_drm *drm)
 int
 nv84_fence_create(struct nouveau_drm *drm)
 {
-        struct nvkm_fifo *fifo = nvxx_fifo(&drm->device);
+        struct nvkm_fifo *fifo = nvxx_fifo(&drm->client.device);
         struct nv84_fence_priv *priv;
         u32 domain;
         int ret;
@@ -212,15 +216,17 @@ nv84_fence_create(struct nouveau_drm *drm)
         priv->base.context_base = dma_fence_context_alloc(priv->base.contexts);
         priv->base.uevent = true;
 
+        mutex_init(&priv->mutex);
+
         /* Use VRAM if there is any ; otherwise fallback to system memory */
-        domain = drm->device.info.ram_size != 0 ? TTM_PL_FLAG_VRAM :
+        domain = drm->client.device.info.ram_size != 0 ? TTM_PL_FLAG_VRAM :
                          /*
                           * fences created in sysmem must be non-cached or we
                           * will lose CPU/GPU coherency!
                           */
                          TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED;
-        ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0, domain, 0,
-                             0, NULL, NULL, &priv->bo);
+        ret = nouveau_bo_new(&drm->client, 16 * priv->base.contexts, 0,
+                             domain, 0, 0, NULL, NULL, &priv->bo);
         if (ret == 0) {
                 ret = nouveau_bo_pin(priv->bo, domain, false);
                 if (ret == 0) {
@@ -233,7 +239,7 @@ nv84_fence_create(struct nouveau_drm *drm)
         }
 
         if (ret == 0)
-                ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0,
+                ret = nouveau_bo_new(&drm->client, 16 * priv->base.contexts, 0,
                                      TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED, 0,
                                      0, NULL, NULL, &priv->bo_gart);
         if (ret == 0) {
diff --git a/drivers/gpu/drm/nouveau/nvif/Kbuild b/drivers/gpu/drm/nouveau/nvif/Kbuild
index ff8ed3a04d06..067b5e9f5ec1 100644
--- a/drivers/gpu/drm/nouveau/nvif/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvif/Kbuild
@@ -1,4 +1,5 @@
 nvif-y := nvif/object.o
 nvif-y += nvif/client.o
 nvif-y += nvif/device.o
+nvif-y += nvif/driver.o
 nvif-y += nvif/notify.o
diff --git a/drivers/gpu/drm/nouveau/nvif/client.c b/drivers/gpu/drm/nouveau/nvif/client.c
index 29c20dfd894d..12db54965c20 100644
--- a/drivers/gpu/drm/nouveau/nvif/client.c
+++ b/drivers/gpu/drm/nouveau/nvif/client.c
@@ -26,6 +26,9 @@
 #include <nvif/driver.h>
 #include <nvif/ioctl.h>
 
+#include <nvif/class.h>
+#include <nvif/if0000.h>
+
 int
 nvif_client_ioctl(struct nvif_client *client, void *data, u32 size)
 {
@@ -47,37 +50,29 @@ nvif_client_resume(struct nvif_client *client)
 void
 nvif_client_fini(struct nvif_client *client)
 {
+        nvif_object_fini(&client->object);
         if (client->driver) {
-                client->driver->fini(client->object.priv);
+                if (client->driver->fini)
+                        client->driver->fini(client->object.priv);
                 client->driver = NULL;
-                client->object.client = NULL;
-                nvif_object_fini(&client->object);
         }
 }
 
-static const struct nvif_driver *
-nvif_drivers[] = {
-#ifdef __KERNEL__
-        &nvif_driver_nvkm,
-#else
-        &nvif_driver_drm,
-        &nvif_driver_lib,
-        &nvif_driver_null,
-#endif
-        NULL
-};
-
 int
-nvif_client_init(const char *driver, const char *name, u64 device,
-                 const char *cfg, const char *dbg, struct nvif_client *client)
+nvif_client_init(struct nvif_client *parent, const char *name, u64 device,
+                 struct nvif_client *client)
 {
+        struct nvif_client_v0 args = { .device = device };
         struct {
                 struct nvif_ioctl_v0 ioctl;
                 struct nvif_ioctl_nop_v0 nop;
-        } args = {};
-        int ret, i;
+        } nop = {};
+        int ret;
 
-        ret = nvif_object_init(NULL, 0, 0, NULL, 0, &client->object);
+        strncpy(args.name, name, sizeof(args.name));
+        ret = nvif_object_init(parent != client ? &parent->object : NULL,
+                               0, NVIF_CLASS_CLIENT, &args, sizeof(args),
+                               &client->object);
         if (ret)
                 return ret;
 
@@ -85,19 +80,11 @@ nvif_client_init(const char *driver, const char *name, u64 device,
         client->object.handle = ~0;
         client->route = NVIF_IOCTL_V0_ROUTE_NVIF;
         client->super = true;
-
-        for (i = 0, ret = -EINVAL; (client->driver = nvif_drivers[i]); i++) {
-                if (!driver || !strcmp(client->driver->name, driver)) {
-                        ret = client->driver->init(name, device, cfg, dbg,
-                                                  &client->object.priv);
-                        if (!ret || driver)
-                                break;
-                }
-        }
+        client->driver = parent->driver;
 
         if (ret == 0) {
-                ret = nvif_client_ioctl(client, &args, sizeof(args));
-                client->version = args.nop.version;
+                ret = nvif_client_ioctl(client, &nop, sizeof(nop));
+                client->version = nop.nop.version;
         }
 
         if (ret)
diff --git a/drivers/gpu/drm/nouveau/nvif/driver.c b/drivers/gpu/drm/nouveau/nvif/driver.c
new file mode 100644
index 000000000000..701330956e33
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvif/driver.c
@@ -0,0 +1,58 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include <nvif/driver.h>
+#include <nvif/client.h>
+
+static const struct nvif_driver *
+nvif_driver[] = {
+#ifdef __KERNEL__
+        &nvif_driver_nvkm,
+#else
+        &nvif_driver_drm,
+        &nvif_driver_lib,
+        &nvif_driver_null,
+#endif
+        NULL
+};
+
+int
+nvif_driver_init(const char *drv, const char *cfg, const char *dbg,
+                 const char *name, u64 device, struct nvif_client *client)
+{
+        int ret = -EINVAL, i;
+
+        for (i = 0; (client->driver = nvif_driver[i]); i++) {
+                if (!drv || !strcmp(client->driver->name, drv)) {
+                        ret = client->driver->init(name, device, cfg, dbg,
+                                                   &client->object.priv);
+                        if (ret == 0)
+                                break;
+                        client->driver->fini(client->object.priv);
+                }
+        }
+
+        if (ret == 0)
+                ret = nvif_client_init(client, name, device, client);
+        return ret;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/Kbuild b/drivers/gpu/drm/nouveau/nvkm/Kbuild
index 2832147b676c..e664378f6eda 100644
--- a/drivers/gpu/drm/nouveau/nvkm/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/Kbuild
@@ -1,3 +1,4 @@
 include $(src)/nvkm/core/Kbuild
+include $(src)/nvkm/falcon/Kbuild
 include $(src)/nvkm/subdev/Kbuild
 include $(src)/nvkm/engine/Kbuild
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/client.c b/drivers/gpu/drm/nouveau/nvkm/core/client.c
index e1943910858e..0d3a896892b4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/client.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/client.c
@@ -31,6 +31,43 @@
 #include <nvif/if0000.h>
 #include <nvif/unpack.h>
 
+static int
+nvkm_uclient_new(const struct nvkm_oclass *oclass, void *argv, u32 argc,
+                 struct nvkm_object **pobject)
+{
+        union {
+                struct nvif_client_v0 v0;
+        } *args = argv;
+        struct nvkm_client *client;
+        int ret = -ENOSYS;
+
+        if (!(ret = nvif_unpack(ret, &argv, &argc, args->v0, 0, 0, false))){
+                args->v0.name[sizeof(args->v0.name) - 1] = 0;
+                ret = nvkm_client_new(args->v0.name, args->v0.device, NULL,
+                                      NULL, oclass->client->ntfy, &client);
+                if (ret)
+                        return ret;
+        } else
+                return ret;
+
+        client->object.client = oclass->client;
+        client->object.handle = oclass->handle;
+        client->object.route  = oclass->route;
+        client->object.token  = oclass->token;
+        client->object.object = oclass->object;
+        client->debug = oclass->client->debug;
+        *pobject = &client->object;
+        return 0;
+}
+
+const struct nvkm_sclass
+nvkm_uclient_sclass = {
+        .oclass = NVIF_CLASS_CLIENT,
+        .minver = 0,
+        .maxver = 0,
+        .ctor = nvkm_uclient_new,
+};
+
 struct nvkm_client_notify {
         struct nvkm_client *client;
         struct nvkm_notify n;
@@ -138,17 +175,30 @@ nvkm_client_notify_new(struct nvkm_object *object,
         return ret;
 }
 
+static const struct nvkm_object_func nvkm_client;
+struct nvkm_client *
+nvkm_client_search(struct nvkm_client *client, u64 handle)
+{
+        struct nvkm_object *object;
+
+        object = nvkm_object_search(client, handle, &nvkm_client);
+        if (IS_ERR(object))
+                return (void *)object;
+
+        return nvkm_client(object);
+}
+
 static int
-nvkm_client_mthd_devlist(struct nvkm_object *object, void *data, u32 size)
+nvkm_client_mthd_devlist(struct nvkm_client *client, void *data, u32 size)
 {
         union {
-                struct nv_client_devlist_v0 v0;
+                struct nvif_client_devlist_v0 v0;
         } *args = data;
         int ret = -ENOSYS;
 
-        nvif_ioctl(object, "client devlist size %d\n", size);
+        nvif_ioctl(&client->object, "client devlist size %d\n", size);
         if (!(ret = nvif_unpack(ret, &data, &size, args->v0, 0, 0, true))) {
-                nvif_ioctl(object, "client devlist vers %d count %d\n",
+                nvif_ioctl(&client->object, "client devlist vers %d count %d\n",
                            args->v0.version, args->v0.count);
                 if (size == sizeof(args->v0.device[0]) * args->v0.count) {
                         ret = nvkm_device_list(args->v0.device, args->v0.count);
@@ -167,9 +217,10 @@ nvkm_client_mthd_devlist(struct nvkm_object *object, void *data, u32 size)
 static int
 nvkm_client_mthd(struct nvkm_object *object, u32 mthd, void *data, u32 size)
 {
+        struct nvkm_client *client = nvkm_client(object);
         switch (mthd) {
-        case NV_CLIENT_DEVLIST:
-                return nvkm_client_mthd_devlist(object, data, size);
+        case NVIF_CLIENT_V0_DEVLIST:
+                return nvkm_client_mthd_devlist(client, data, size);
         default:
                 break;
         }
@@ -190,7 +241,8 @@ nvkm_client_child_get(struct nvkm_object *object, int index,
         const struct nvkm_sclass *sclass;
 
         switch (index) {
-        case 0: sclass = &nvkm_udevice_sclass; break;
+        case 0: sclass = &nvkm_uclient_sclass; break;
+        case 1: sclass = &nvkm_udevice_sclass; break;
         default:
                 return -EINVAL;
         }
@@ -200,110 +252,54 @@ nvkm_client_child_get(struct nvkm_object *object, int index,
         return 0;
 }
 
-static const struct nvkm_object_func
-nvkm_client_object_func = {
-        .mthd = nvkm_client_mthd,
-        .sclass = nvkm_client_child_get,
-};
-
-void
-nvkm_client_remove(struct nvkm_client *client, struct nvkm_object *object)
-{
-        if (!RB_EMPTY_NODE(&object->node))
-                rb_erase(&object->node, &client->objroot);
-}
-
-bool
-nvkm_client_insert(struct nvkm_client *client, struct nvkm_object *object)
-{
-        struct rb_node **ptr = &client->objroot.rb_node;
-        struct rb_node *parent = NULL;
-
-        while (*ptr) {
-                struct nvkm_object *this =
-                        container_of(*ptr, typeof(*this), node);
-                parent = *ptr;
-                if (object->object < this->object)
-                        ptr = &parent->rb_left;
-                else
-                if (object->object > this->object)
-                        ptr = &parent->rb_right;
-                else
-                        return false;
-        }
-
-        rb_link_node(&object->node, parent, ptr);
-        rb_insert_color(&object->node, &client->objroot);
-        return true;
-}
-
-struct nvkm_object *
-nvkm_client_search(struct nvkm_client *client, u64 handle)
-{
-        struct rb_node *node = client->objroot.rb_node;
-        while (node) {
-                struct nvkm_object *object =
-                        container_of(node, typeof(*object), node);
-                if (handle < object->object)
-                        node = node->rb_left;
-                else
-                if (handle > object->object)
-                        node = node->rb_right;
-                else
-                        return object;
-        }
-        return NULL;
-}
-
-int
-nvkm_client_fini(struct nvkm_client *client, bool suspend)
+static int
+nvkm_client_fini(struct nvkm_object *object, bool suspend)
 {
-        struct nvkm_object *object = &client->object;
+        struct nvkm_client *client = nvkm_client(object);
         const char *name[2] = { "fini", "suspend" };
         int i;
         nvif_debug(object, "%s notify\n", name[suspend]);
         for (i = 0; i < ARRAY_SIZE(client->notify); i++)
                 nvkm_client_notify_put(client, i);
-        return nvkm_object_fini(&client->object, suspend);
-}
-
-int
-nvkm_client_init(struct nvkm_client *client)
-{
-        return nvkm_object_init(&client->object);
+        return 0;
 }
 
-void
-nvkm_client_del(struct nvkm_client **pclient)
+static void *
+nvkm_client_dtor(struct nvkm_object *object)
 {
-        struct nvkm_client *client = *pclient;
+        struct nvkm_client *client = nvkm_client(object);
         int i;
-        if (client) {
-                nvkm_client_fini(client, false);
-                for (i = 0; i < ARRAY_SIZE(client->notify); i++)
-                        nvkm_client_notify_del(client, i);
-                nvkm_object_dtor(&client->object);
-                kfree(*pclient);
-                *pclient = NULL;
-        }
+        for (i = 0; i < ARRAY_SIZE(client->notify); i++)
+                nvkm_client_notify_del(client, i);
+        return client;
 }
 
+static const struct nvkm_object_func
+nvkm_client = {
+        .dtor = nvkm_client_dtor,
+        .fini = nvkm_client_fini,
+        .mthd = nvkm_client_mthd,
+        .sclass = nvkm_client_child_get,
+};
+
 int
 nvkm_client_new(const char *name, u64 device, const char *cfg,
-                const char *dbg, struct nvkm_client **pclient)
+                const char *dbg,
+                int (*ntfy)(const void *, u32, const void *, u32),
+                struct nvkm_client **pclient)
 {
-        struct nvkm_oclass oclass = {};
+        struct nvkm_oclass oclass = { .base = nvkm_uclient_sclass };
         struct nvkm_client *client;
 
         if (!(client = *pclient = kzalloc(sizeof(*client), GFP_KERNEL)))
                 return -ENOMEM;
         oclass.client = client;
 
-        nvkm_object_ctor(&nvkm_client_object_func, &oclass, &client->object);
+        nvkm_object_ctor(&nvkm_client, &oclass, &client->object);
         snprintf(client->name, sizeof(client->name), "%s", name);
         client->device = device;
         client->debug = nvkm_dbgopt(dbg, "CLIENT");
         client->objroot = RB_ROOT;
-        client->dmaroot = RB_ROOT;
+        client->ntfy = ntfy;
         return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/engine.c b/drivers/gpu/drm/nouveau/nvkm/core/engine.c
index ee8e5831fe37..b6c916954a10 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/engine.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/engine.c
@@ -27,6 +27,14 @@
 
 #include <subdev/fb.h>
 
+bool
+nvkm_engine_chsw_load(struct nvkm_engine *engine)
+{
+        if (engine->func->chsw_load)
+                return engine->func->chsw_load(engine);
+        return false;
+}
+
 void
 nvkm_engine_unref(struct nvkm_engine **pengine)
 {
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c b/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c
index b0db51847c36..be19bbe56bba 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c
@@ -29,7 +29,8 @@
 #include <nvif/ioctl.h>
 
 static int
-nvkm_ioctl_nop(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_nop(struct nvkm_client *client,
+               struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_nop_v0 v0;
@@ -46,7 +47,8 @@ nvkm_ioctl_nop(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_sclass(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_sclass(struct nvkm_client *client,
+                  struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_sclass_v0 v0;
@@ -78,12 +80,12 @@ nvkm_ioctl_sclass(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_new(struct nvkm_object *parent, void *data, u32 size)
+nvkm_ioctl_new(struct nvkm_client *client,
+               struct nvkm_object *parent, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_new_v0 v0;
         } *args = data;
-        struct nvkm_client *client = parent->client;
         struct nvkm_object *object = NULL;
         struct nvkm_oclass oclass;
         int ret = -ENOSYS, i = 0;
@@ -104,9 +106,11 @@ nvkm_ioctl_new(struct nvkm_object *parent, void *data, u32 size)
 
         do {
                 memset(&oclass, 0x00, sizeof(oclass));
-                oclass.client = client;
                 oclass.handle = args->v0.handle;
+                oclass.route  = args->v0.route;
+                oclass.token  = args->v0.token;
                 oclass.object = args->v0.object;
+                oclass.client = client;
                 oclass.parent = parent;
                 ret = parent->func->sclass(parent, i++, &oclass);
                 if (ret)
@@ -125,10 +129,7 @@ nvkm_ioctl_new(struct nvkm_object *parent, void *data, u32 size)
                 ret = nvkm_object_init(object);
                 if (ret == 0) {
                         list_add(&object->head, &parent->tree);
-                        object->route = args->v0.route;
-                        object->token = args->v0.token;
-                        object->object = args->v0.object;
-                        if (nvkm_client_insert(client, object)) {
+                        if (nvkm_object_insert(object)) {
                                 client->data = object;
                                 return 0;
                         }
@@ -142,7 +143,8 @@ nvkm_ioctl_new(struct nvkm_object *parent, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_del(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_del(struct nvkm_client *client,
+               struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_del none;
@@ -156,11 +158,12 @@ nvkm_ioctl_del(struct nvkm_object *object, void *data, u32 size)
                 nvkm_object_del(&object);
         }
 
-        return ret;
+        return ret ? ret : 1;
 }
 
 static int
-nvkm_ioctl_mthd(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_mthd(struct nvkm_client *client,
+                struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_mthd_v0 v0;
@@ -179,7 +182,8 @@ nvkm_ioctl_mthd(struct nvkm_object *object, void *data, u32 size)
 
 
 static int
-nvkm_ioctl_rd(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_rd(struct nvkm_client *client,
+              struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_rd_v0 v0;
@@ -218,7 +222,8 @@ nvkm_ioctl_rd(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_wr(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_wr(struct nvkm_client *client,
+              struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_wr_v0 v0;
@@ -246,7 +251,8 @@ nvkm_ioctl_wr(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_map(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_map(struct nvkm_client *client,
+               struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_map_v0 v0;
@@ -264,7 +270,8 @@ nvkm_ioctl_map(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_unmap(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_unmap(struct nvkm_client *client,
+                 struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_unmap none;
@@ -280,7 +287,8 @@ nvkm_ioctl_unmap(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_ntfy_new(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_ntfy_new(struct nvkm_client *client,
+                    struct nvkm_object *object, void *data, u32 size)
 {
         union {
                 struct nvif_ioctl_ntfy_new_v0 v0;
@@ -306,9 +314,9 @@ nvkm_ioctl_ntfy_new(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_ntfy_del(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_ntfy_del(struct nvkm_client *client,
+                    struct nvkm_object *object, void *data, u32 size)
 {
-        struct nvkm_client *client = object->client;
         union {
                 struct nvif_ioctl_ntfy_del_v0 v0;
         } *args = data;
@@ -325,9 +333,9 @@ nvkm_ioctl_ntfy_del(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_ntfy_get(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_ntfy_get(struct nvkm_client *client,
+                    struct nvkm_object *object, void *data, u32 size)
 {
-        struct nvkm_client *client = object->client;
         union {
                 struct nvif_ioctl_ntfy_get_v0 v0;
         } *args = data;
@@ -344,9 +352,9 @@ nvkm_ioctl_ntfy_get(struct nvkm_object *object, void *data, u32 size)
 }
 
 static int
-nvkm_ioctl_ntfy_put(struct nvkm_object *object, void *data, u32 size)
+nvkm_ioctl_ntfy_put(struct nvkm_client *client,
+                    struct nvkm_object *object, void *data, u32 size)
 {
-        struct nvkm_client *client = object->client;
         union {
                 struct nvif_ioctl_ntfy_put_v0 v0;
         } *args = data;
@@ -364,7 +372,7 @@ nvkm_ioctl_ntfy_put(struct nvkm_object *object, void *data, u32 size)
 
 static struct {
         int version;
-        int (*func)(struct nvkm_object *, void *, u32);
+        int (*func)(struct nvkm_client *, struct nvkm_object *, void *, u32);
 }
 nvkm_ioctl_v0[] = {
         { 0x00, nvkm_ioctl_nop },
@@ -389,13 +397,10 @@ nvkm_ioctl_path(struct nvkm_client *client, u64 handle, u32 type,
         struct nvkm_object *object;
         int ret;
 
-        if (handle)
-                object = nvkm_client_search(client, handle);
-        else
-                object = &client->object;
-        if (unlikely(!object)) {
+        object = nvkm_object_search(client, handle, NULL);
+        if (IS_ERR(object)) {
                 nvif_ioctl(&client->object, "object not found\n");
-                return -ENOENT;
+                return PTR_ERR(object);
         }
 
         if (owner != NVIF_IOCTL_V0_OWNER_ANY && owner != object->route) {
@@ -407,7 +412,7 @@ nvkm_ioctl_path(struct nvkm_client *client, u64 handle, u32 type,
 
         if (ret = -EINVAL, type < ARRAY_SIZE(nvkm_ioctl_v0)) {
                 if (nvkm_ioctl_v0[type].version == 0)
-                        ret = nvkm_ioctl_v0[type].func(object, data, size);
+                        ret = nvkm_ioctl_v0[type].func(client, object, data, size);
         }
 
         return ret;
@@ -436,12 +441,13 @@ nvkm_ioctl(struct nvkm_client *client, bool supervisor,
                                       &args->v0.route, &args->v0.token);
         }
 
-        nvif_ioctl(object, "return %d\n", ret);
-        if (hack) {
-                *hack = client->data;
-                client->data = NULL;
+        if (ret != 1) {
+                nvif_ioctl(object, "return %d\n", ret);
+                if (hack) {
+                        *hack = client->data;
+                        client->data = NULL;
+                }
         }
 
-        client->super = false;
         return ret;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/mm.c b/drivers/gpu/drm/nouveau/nvkm/core/mm.c
index 09a1eee8fd33..fd19d652a7ab 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/mm.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/mm.c
@@ -147,6 +147,7 @@ nvkm_mm_head(struct nvkm_mm *mm, u8 heap, u8 type, u32 size_max, u32 size_min,
                 if (!this)
                         return -ENOMEM;
 
+                this->next = NULL;
                 this->type = type;
                 list_del(&this->fl_entry);
                 *pnode = this;
@@ -225,6 +226,7 @@ nvkm_mm_tail(struct nvkm_mm *mm, u8 heap, u8 type, u32 size_max, u32 size_min,
                 if (!this)
                         return -ENOMEM;
 
+                this->next = NULL;
                 this->type = type;
                 list_del(&this->fl_entry);
                 *pnode = this;
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/object.c b/drivers/gpu/drm/nouveau/nvkm/core/object.c
index 67aa7223dcd7..89d2e9da11c7 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/object.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/object.c
@@ -25,6 +25,65 @@
 #include <core/client.h>
 #include <core/engine.h>
 
+struct nvkm_object *
+nvkm_object_search(struct nvkm_client *client, u64 handle,
+                   const struct nvkm_object_func *func)
+{
+        struct nvkm_object *object;
+
+        if (handle) {
+                struct rb_node *node = client->objroot.rb_node;
+                while (node) {
+                        object = rb_entry(node, typeof(*object), node);
+                        if (handle < object->object)
+                                node = node->rb_left;
+                        else
+                        if (handle > object->object)
+                                node = node->rb_right;
+                        else
+                                goto done;
+                }
+                return ERR_PTR(-ENOENT);
+        } else {
+                object = &client->object;
+        }
+
+done:
+        if (unlikely(func && object->func != func))
+                return ERR_PTR(-EINVAL);
+        return object;
+}
+
+void
+nvkm_object_remove(struct nvkm_object *object)
+{
+        if (!RB_EMPTY_NODE(&object->node))
+                rb_erase(&object->node, &object->client->objroot);
+}
+
+bool
+nvkm_object_insert(struct nvkm_object *object)
+{
+        struct rb_node **ptr = &object->client->objroot.rb_node;
+        struct rb_node *parent = NULL;
+
+        while (*ptr) {
+                struct nvkm_object *this = rb_entry(*ptr, typeof(*this), node);
+                parent = *ptr;
+                if (object->object < this->object)
+                        ptr = &parent->rb_left;
+                else
+                if (object->object > this->object)
+                        ptr = &parent->rb_right;
+                else
+                        return false;
+        }
+
+        rb_link_node(&object->node, parent, ptr);
+        rb_insert_color(&object->node, &object->client->objroot);
+        return true;
+}
+
 int
 nvkm_object_mthd(struct nvkm_object *object, u32 mthd, void *data, u32 size)
 {
@@ -214,7 +273,7 @@ nvkm_object_del(struct nvkm_object **pobject)
         struct nvkm_object *object = *pobject;
         if (object && !WARN_ON(!object->func)) {
                 *pobject = nvkm_object_dtor(object);
-                nvkm_client_remove(object->client, object);
+                nvkm_object_remove(object);
                 list_del(&object->head);
                 kfree(*pobject);
                 *pobject = NULL;
@@ -230,6 +289,9 @@ nvkm_object_ctor(const struct nvkm_object_func *func,
         object->engine = nvkm_engine_ref(oclass->engine);
         object->oclass = oclass->base.oclass;
         object->handle = oclass->handle;
+        object->route  = oclass->route;
+        object->token  = oclass->token;
+        object->object = oclass->object;
         INIT_LIST_HEAD(&object->head);
         INIT_LIST_HEAD(&object->tree);
         RB_CLEAR_NODE(&object->node);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
index cceda959b47c..273562dd6bbd 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
@@ -993,7 +993,7 @@ nv92_chipset = {
         .mc = g84_mc_new,
         .mmu = nv50_mmu_new,
         .mxm = nv50_mxm_new,
-        .pci = g84_pci_new,
+        .pci = g92_pci_new,
         .therm = g84_therm_new,
         .timer = nv41_timer_new,
         .volt = nv40_volt_new,
@@ -2138,6 +2138,7 @@ nv12b_chipset = {
         .ltc = gm200_ltc_new,
         .mc = gk20a_mc_new,
         .mmu = gf100_mmu_new,
+        .pmu = gm20b_pmu_new,
         .secboot = gm20b_secboot_new,
         .timer = gk20a_timer_new,
         .top = gk104_top_new,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.c
index 0a1381a84552..070ec5e18fdb 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.c
@@ -137,7 +137,6 @@ nv50_disp_dmac_new_(const struct nv50_disp_dmac_func *func,
                     const struct nvkm_oclass *oclass,
                     struct nvkm_object **pobject)
 {
-        struct nvkm_device *device = root->disp->base.engine.subdev.device;
         struct nvkm_client *client = oclass->client;
         struct nvkm_dmaobj *dmaobj;
         struct nv50_disp_dmac *chan;
@@ -153,9 +152,9 @@ nv50_disp_dmac_new_(const struct nv50_disp_dmac_func *func,
         if (ret)
                 return ret;
 
-        dmaobj = nvkm_dma_search(device->dma, client, push);
-        if (!dmaobj)
-                return -ENOENT;
+        dmaobj = nvkm_dmaobj_search(client, push);
+        if (IS_ERR(dmaobj))
+                return PTR_ERR(dmaobj);
 
         if (dmaobj->limit - dmaobj->start != 0xfff)
                 return -EINVAL;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/hdagt215.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/hdagt215.c
index 6f0436df0219..f8f2f16c22a2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/hdagt215.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/hdagt215.c
@@ -59,7 +59,7 @@ gt215_hda_eld(NV50_DISP_MTHD_V1)
                         );
                 }
                 for (i = 0; i < size; i++)
-                        nvkm_wr32(device, 0x61c440 + soff, (i << 8) | args->v0.data[0]);
+                        nvkm_wr32(device, 0x61c440 + soff, (i << 8) | args->v0.data[i]);
                 for (; i < 0x60; i++)
                         nvkm_wr32(device, 0x61c440 + soff, (i << 8));
                 nvkm_mask(device, 0x61c448 + soff, 0x80000003, 0x80000003);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c
index 567466f93cd5..0db8efbf1c2e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c
@@ -433,8 +433,6 @@ nv50_disp_dptmds_war(struct nvkm_device *device)
         case 0x94:
         case 0x96:
         case 0x98:
-        case 0xaa:
-        case 0xac:
                 return true;
         default:
                 break;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorg94.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorg94.c
index 4510cb6e10a8..627b9ee1ddd2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorg94.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorg94.c
@@ -39,13 +39,6 @@ g94_sor_loff(struct nvkm_output_dp *outp)
 }
 
 /*******************************************************************************
- * TMDS/LVDS
- ******************************************************************************/
-static const struct nvkm_output_func
-g94_sor_output_func = {
-};
-
-/*******************************************************************************
  * DisplayPort
  ******************************************************************************/
 u32
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dma/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/dma/base.c
index f11ebdd16c77..11b7b8fd5dda 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dma/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dma/base.c
@@ -28,24 +28,6 @@
 
 #include <nvif/class.h>
 
-struct nvkm_dmaobj *
-nvkm_dma_search(struct nvkm_dma *dma, struct nvkm_client *client, u64 object)
-{
-        struct rb_node *node = client->dmaroot.rb_node;
-        while (node) {
-                struct nvkm_dmaobj *dmaobj =
-                        container_of(node, typeof(*dmaobj), rb);
-                if (object < dmaobj->handle)
-                        node = node->rb_left;
-                else
-                if (object > dmaobj->handle)
-                        node = node->rb_right;
-                else
-                        return dmaobj;
-        }
-        return NULL;
-}
-
 static int
 nvkm_dma_oclass_new(struct nvkm_device *device,
                     const struct nvkm_oclass *oclass, void *data, u32 size,
@@ -53,34 +35,12 @@ nvkm_dma_oclass_new(struct nvkm_device *device,
 {
         struct nvkm_dma *dma = nvkm_dma(oclass->engine);
         struct nvkm_dmaobj *dmaobj = NULL;
-        struct nvkm_client *client = oclass->client;
-        struct rb_node **ptr = &client->dmaroot.rb_node;
-        struct rb_node *parent = NULL;
         int ret;
 
         ret = dma->func->class_new(dma, oclass, data, size, &dmaobj);
         if (dmaobj)
                 *pobject = &dmaobj->object;
-        if (ret)
-                return ret;
-
-        dmaobj->handle = oclass->object;
-
-        while (*ptr) {
-                struct nvkm_dmaobj *obj = container_of(*ptr, typeof(*obj), rb);
-                parent = *ptr;
-                if (dmaobj->handle < obj->handle)
-                        ptr = &parent->rb_left;
-                else
-                if (dmaobj->handle > obj->handle)
-                        ptr = &parent->rb_right;
-                else
-                        return -EEXIST;
-        }
-
-        rb_link_node(&dmaobj->rb, parent, ptr);
-        rb_insert_color(&dmaobj->rb, &client->dmaroot);
-        return 0;
+        return ret;
 }
 
 static const struct nvkm_device_oclass
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/dma/user.c b/drivers/gpu/drm/nouveau/nvkm/engine/dma/user.c
index 13c661b1ef14..d20cc0681a88 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/dma/user.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/dma/user.c
@@ -31,6 +31,19 @@
 #include <nvif/cl0002.h>
 #include <nvif/unpack.h>
 
+static const struct nvkm_object_func nvkm_dmaobj_func;
+struct nvkm_dmaobj *
+nvkm_dmaobj_search(struct nvkm_client *client, u64 handle)
+{
+        struct nvkm_object *object;
+
+        object = nvkm_object_search(client, handle, &nvkm_dmaobj_func);
+        if (IS_ERR(object))
+                return (void *)object;
+
+        return nvkm_dmaobj(object);
+}
+
 static int
 nvkm_dmaobj_bind(struct nvkm_object *base, struct nvkm_gpuobj *gpuobj,
                  int align, struct nvkm_gpuobj **pgpuobj)
@@ -42,10 +55,7 @@ nvkm_dmaobj_bind(struct nvkm_object *base, struct nvkm_gpuobj *gpuobj,
 static void *
 nvkm_dmaobj_dtor(struct nvkm_object *base)
 {
-        struct nvkm_dmaobj *dmaobj = nvkm_dmaobj(base);
-        if (!RB_EMPTY_NODE(&dmaobj->rb))
-                rb_erase(&dmaobj->rb, &dmaobj->object.client->dmaroot);
-        return dmaobj;
+        return nvkm_dmaobj(base);
 }
 
 static const struct nvkm_object_func
@@ -74,7 +84,6 @@ nvkm_dmaobj_ctor(const struct nvkm_dmaobj_func *func, struct nvkm_dma *dma,
         nvkm_object_ctor(&nvkm_dmaobj_func, oclass, &dmaobj->object);
         dmaobj->func = func;
         dmaobj->dma = dma;
-        RB_CLEAR_NODE(&dmaobj->rb);
 
         nvif_ioctl(parent, "create dma size %d\n", *psize);
         if (!(ret = nvif_unpack(ret, &data, &size, args->v0, 0, 0, true))) {
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
index 1c9682ae3a6b..660ca7aa95ea 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
@@ -32,6 +32,17 @@
 #include <nvif/unpack.h>
 
 void
+nvkm_fifo_recover_chan(struct nvkm_fifo *fifo, int chid)
+{
+        unsigned long flags;
+        if (WARN_ON(!fifo->func->recover_chan))
+                return;
+        spin_lock_irqsave(&fifo->lock, flags);
+        fifo->func->recover_chan(fifo, chid);
+        spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+void
 nvkm_fifo_pause(struct nvkm_fifo *fifo, unsigned long *flags)
 {
         return fifo->func->pause(fifo, flags);
@@ -55,19 +66,29 @@ nvkm_fifo_chan_put(struct nvkm_fifo *fifo, unsigned long flags,
 }
 
 struct nvkm_fifo_chan *
-nvkm_fifo_chan_inst(struct nvkm_fifo *fifo, u64 inst, unsigned long *rflags)
+nvkm_fifo_chan_inst_locked(struct nvkm_fifo *fifo, u64 inst)
 {
         struct nvkm_fifo_chan *chan;
-        unsigned long flags;
-        spin_lock_irqsave(&fifo->lock, flags);
         list_for_each_entry(chan, &fifo->chan, head) {
                 if (chan->inst->addr == inst) {
                         list_del(&chan->head);
                         list_add(&chan->head, &fifo->chan);
-                        *rflags = flags;
                         return chan;
                 }
         }
+        return NULL;
+}
+
+struct nvkm_fifo_chan *
+nvkm_fifo_chan_inst(struct nvkm_fifo *fifo, u64 inst, unsigned long *rflags)
+{
+        struct nvkm_fifo_chan *chan;
+        unsigned long flags;
+        spin_lock_irqsave(&fifo->lock, flags);
+        if ((chan = nvkm_fifo_chan_inst_locked(fifo, inst))) {
+                *rflags = flags;
+                return chan;
+        }
         spin_unlock_irqrestore(&fifo->lock, flags);
         return NULL;
 }
@@ -90,9 +111,34 @@ nvkm_fifo_chan_chid(struct nvkm_fifo *fifo, int chid, unsigned long *rflags)
         return NULL;
 }
 
+void
+nvkm_fifo_kevent(struct nvkm_fifo *fifo, int chid)
+{
+        nvkm_event_send(&fifo->kevent, 1, chid, NULL, 0);
+}
+
 static int
-nvkm_fifo_event_ctor(struct nvkm_object *object, void *data, u32 size,
-                     struct nvkm_notify *notify)
+nvkm_fifo_kevent_ctor(struct nvkm_object *object, void *data, u32 size,
+                      struct nvkm_notify *notify)
+{
+        struct nvkm_fifo_chan *chan = nvkm_fifo_chan(object);
+        if (size == 0) {
+                notify->size  = 0;
+                notify->types = 1;
+                notify->index = chan->chid;
+                return 0;
+        }
+        return -ENOSYS;
+}
+
+static const struct nvkm_event_func
+nvkm_fifo_kevent_func = {
+        .ctor = nvkm_fifo_kevent_ctor,
+};
+
+static int
+nvkm_fifo_cevent_ctor(struct nvkm_object *object, void *data, u32 size,
+                      struct nvkm_notify *notify)
 {
         if (size == 0) {
                 notify->size  = 0;
@@ -104,10 +150,16 @@ nvkm_fifo_event_ctor(struct nvkm_object *object, void *data, u32 size,
 }
 
 static const struct nvkm_event_func
-nvkm_fifo_event_func = {
-        .ctor = nvkm_fifo_event_ctor,
+nvkm_fifo_cevent_func = {
+        .ctor = nvkm_fifo_cevent_ctor,
 };
 
+void
+nvkm_fifo_cevent(struct nvkm_fifo *fifo)
+{
+        nvkm_event_send(&fifo->cevent, 1, 0, NULL, 0);
+}
+
 static void
 nvkm_fifo_uevent_fini(struct nvkm_event *event, int type, int index)
 {
@@ -241,6 +293,7 @@ nvkm_fifo_dtor(struct nvkm_engine *engine)
         void *data = fifo;
         if (fifo->func->dtor)
                 data = fifo->func->dtor(fifo);
+        nvkm_event_fini(&fifo->kevent);
         nvkm_event_fini(&fifo->cevent);
         nvkm_event_fini(&fifo->uevent);
         return data;
@@ -283,5 +336,9 @@ nvkm_fifo_ctor(const struct nvkm_fifo_func *func, struct nvkm_device *device,
                         return ret;
         }
 
-        return nvkm_event_init(&nvkm_fifo_event_func, 1, 1, &fifo->cevent);
+        ret = nvkm_event_init(&nvkm_fifo_cevent_func, 1, 1, &fifo->cevent);
+        if (ret)
+                return ret;
+
+        return nvkm_event_init(&nvkm_fifo_kevent_func, 1, nr, &fifo->kevent);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
index dc6d4678f228..fab760ae922f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
@@ -371,9 +371,9 @@ nvkm_fifo_chan_ctor(const struct nvkm_fifo_chan_func *func,
 
         /* allocate push buffer ctxdma instance */
         if (push) {
-                dmaobj = nvkm_dma_search(device->dma, oclass->client, push);
-                if (!dmaobj)
-                        return -ENOENT;
+                dmaobj = nvkm_dmaobj_search(client, push);
+                if (IS_ERR(dmaobj))
+                        return PTR_ERR(dmaobj);
 
                 ret = nvkm_object_bind(&dmaobj->object, chan->inst, -16,
                                        &chan->push);
@@ -410,6 +410,6 @@ nvkm_fifo_chan_ctor(const struct nvkm_fifo_chan_func *func,
                      base + user * chan->chid;
         chan->size = user;
 
-        nvkm_event_send(&fifo->cevent, 1, 0, NULL, 0);
+        nvkm_fifo_cevent(fifo);
         return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
index 55dc415c5c08..d8019bdacd61 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
@@ -29,5 +29,5 @@ struct nvkm_fifo_chan_oclass {
         struct nvkm_sclass base;
 };
 
-int g84_fifo_chan_ntfy(struct nvkm_fifo_chan *, u32, struct nvkm_event **);
+int gf100_fifo_chan_ntfy(struct nvkm_fifo_chan *, u32, struct nvkm_event **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c
index 15a992b3580a..61797c4dd07a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c
@@ -30,12 +30,12 @@
 
 #include <nvif/cl826e.h>
 
-int
+static int
 g84_fifo_chan_ntfy(struct nvkm_fifo_chan *chan, u32 type,
                    struct nvkm_event **pevent)
 {
         switch (type) {
-        case G82_CHANNEL_DMA_V0_NTFY_UEVENT:
+        case NV826E_V0_NTFY_NON_STALL_INTERRUPT:
                 *pevent = &chan->fifo->uevent;
                 return 0;
         default:
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
index ec68ea9747d5..cd468ab1db12 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
@@ -68,7 +68,14 @@ gf100_fifo_runlist_commit(struct gf100_fifo *fifo)
         }
         nvkm_done(cur);
 
-        target = (nvkm_memory_target(cur) == NVKM_MEM_TARGET_HOST) ? 0x3 : 0x0;
+        switch (nvkm_memory_target(cur)) {
+        case NVKM_MEM_TARGET_VRAM: target = 0; break;
+        case NVKM_MEM_TARGET_NCOH: target = 3; break;
+        default:
+                mutex_unlock(&subdev->mutex);
+                WARN_ON(1);
+                return;
+        }
 
         nvkm_wr32(device, 0x002270, (nvkm_memory_addr(cur) >> 12) |
                                     (target << 28));
@@ -183,6 +190,7 @@ gf100_fifo_recover(struct gf100_fifo *fifo, struct nvkm_engine *engine,
         if (engine != &fifo->base.engine)
                 fifo->recover.mask |= 1ULL << engine->subdev.index;
         schedule_work(&fifo->recover.work);
+        nvkm_fifo_kevent(&fifo->base, chid);
 }
 
 static const struct nvkm_enum
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
index 38c0910722c0..3a24788c3185 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
@@ -27,11 +27,71 @@
 #include <core/client.h>
 #include <core/gpuobj.h>
 #include <subdev/bar.h>
+#include <subdev/timer.h>
 #include <subdev/top.h>
 #include <engine/sw.h>
 
 #include <nvif/class.h>
 
+struct gk104_fifo_engine_status {
+        bool busy;
+        bool faulted;
+        bool chsw;
+        bool save;
+        bool load;
+        struct {
+                bool tsg;
+                u32 id;
+        } prev, next, *chan;
+};
+
+static void
+gk104_fifo_engine_status(struct gk104_fifo *fifo, int engn,
+                         struct gk104_fifo_engine_status *status)
+{
+        struct nvkm_engine *engine = fifo->engine[engn].engine;
+        struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
+        struct nvkm_device *device = subdev->device;
+        u32 stat = nvkm_rd32(device, 0x002640 + (engn * 0x08));
+
+        status->busy     = !!(stat & 0x80000000);
+        status->faulted  = !!(stat & 0x40000000);
+        status->next.tsg = !!(stat & 0x10000000);
+        status->next.id  =   (stat & 0x0fff0000) >> 16;
+        status->chsw     = !!(stat & 0x00008000);
+        status->save     = !!(stat & 0x00004000);
+        status->load     = !!(stat & 0x00002000);
+        status->prev.tsg = !!(stat & 0x00001000);
+        status->prev.id  =   (stat & 0x00000fff);
+        status->chan     = NULL;
+
+        if (status->busy && status->chsw) {
+                if (status->load && status->save) {
+                        if (engine && nvkm_engine_chsw_load(engine))
+                                status->chan = &status->next;
+                        else
+                                status->chan = &status->prev;
+                } else
+                if (status->load) {
+                        status->chan = &status->next;
+                } else {
+                        status->chan = &status->prev;
+                }
+        } else
+        if (status->load) {
+                status->chan = &status->prev;
+        }
+
+        nvkm_debug(subdev, "engine %02d: busy %d faulted %d chsw %d "
+                           "save %d load %d %sid %d%s-> %sid %d%s\n",
+                   engn, status->busy, status->faulted,
+                   status->chsw, status->save, status->load,
+                   status->prev.tsg ? "tsg" : "ch", status->prev.id,
+                   status->chan == &status->prev ? "*" : " ",
+                   status->next.tsg ? "tsg" : "ch", status->next.id,
+                   status->chan == &status->next ? "*" : " ");
+}
+
 static int
 gk104_fifo_class_get(struct nvkm_fifo *base, int index,
                      const struct nvkm_fifo_chan_oclass **psclass)
@@ -83,10 +143,13 @@ gk104_fifo_runlist_commit(struct gk104_fifo *fifo, int runl)
         }
         nvkm_done(mem);
 
-        if (nvkm_memory_target(mem) == NVKM_MEM_TARGET_VRAM)
-                target = 0;
-        else
-                target = 3;
+        switch (nvkm_memory_target(mem)) {
+        case NVKM_MEM_TARGET_VRAM: target = 0; break;
+        case NVKM_MEM_TARGET_NCOH: target = 3; break;
+        default:
+                WARN_ON(1);
+                return;
+        }
 
         nvkm_wr32(device, 0x002270, (nvkm_memory_addr(mem) >> 12) |
                                     (target << 28));
@@ -149,31 +212,137 @@ gk104_fifo_recover_work(struct work_struct *w)
         nvkm_mask(device, 0x002630, runm, 0x00000000);
 }
 
+static void gk104_fifo_recover_engn(struct gk104_fifo *fifo, int engn);
+
 static void
-gk104_fifo_recover(struct gk104_fifo *fifo, struct nvkm_engine *engine,
-                   struct gk104_fifo_chan *chan)
+gk104_fifo_recover_runl(struct gk104_fifo *fifo, int runl)
 {
         struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
         struct nvkm_device *device = subdev->device;
-        u32 chid = chan->base.chid;
-        int engn;
+        const u32 runm = BIT(runl);
 
-        nvkm_error(subdev, "%s engine fault on channel %d, recovering...\n",
-                   nvkm_subdev_name[engine->subdev.index], chid);
         assert_spin_locked(&fifo->base.lock);
+        if (fifo->recover.runm & runm)
+                return;
+        fifo->recover.runm |= runm;
 
-        nvkm_mask(device, 0x800004 + (chid * 0x08), 0x00000800, 0x00000800);
-        list_del_init(&chan->head);
-        chan->killed = true;
+        /* Block runlist to prevent channel assignment(s) from changing. */
+        nvkm_mask(device, 0x002630, runm, runm);
 
-        for (engn = 0; engn < fifo->engine_nr; engn++) {
-                if (fifo->engine[engn].engine == engine) {
-                        fifo->recover.engm |= BIT(engn);
+        /* Schedule recovery. */
+        nvkm_warn(subdev, "runlist %d: scheduled for recovery\n", runl);
+        schedule_work(&fifo->recover.work);
+}
+
+static void
+gk104_fifo_recover_chan(struct nvkm_fifo *base, int chid)
+{
+        struct gk104_fifo *fifo = gk104_fifo(base);
+        struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
+        struct nvkm_device *device = subdev->device;
+        const u32  stat = nvkm_rd32(device, 0x800004 + (chid * 0x08));
+        const u32  runl = (stat & 0x000f0000) >> 16;
+        const bool used = (stat & 0x00000001);
+        unsigned long engn, engm = fifo->runlist[runl].engm;
+        struct gk104_fifo_chan *chan;
+
+        assert_spin_locked(&fifo->base.lock);
+        if (!used)
+                return;
+
+        /* Lookup SW state for channel, and mark it as dead. */
+        list_for_each_entry(chan, &fifo->runlist[runl].chan, head) {
+                if (chan->base.chid == chid) {
+                        list_del_init(&chan->head);
+                        chan->killed = true;
+                        nvkm_fifo_kevent(&fifo->base, chid);
                         break;
                 }
         }
 
-        fifo->recover.runm |= BIT(chan->runl);
+        /* Disable channel. */
+        nvkm_wr32(device, 0x800004 + (chid * 0x08), stat | 0x00000800);
+        nvkm_warn(subdev, "channel %d: killed\n", chid);
+
+        /* Block channel assignments from changing during recovery. */
+        gk104_fifo_recover_runl(fifo, runl);
+
+        /* Schedule recovery for any engines the channel is on. */
+        for_each_set_bit(engn, &engm, fifo->engine_nr) {
+                struct gk104_fifo_engine_status status;
+                gk104_fifo_engine_status(fifo, engn, &status);
+                if (!status.chan || status.chan->id != chid)
+                        continue;
+                gk104_fifo_recover_engn(fifo, engn);
+        }
+}
+
+static void
+gk104_fifo_recover_engn(struct gk104_fifo *fifo, int engn)
+{
+        struct nvkm_engine *engine = fifo->engine[engn].engine;
+        struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
+        struct nvkm_device *device = subdev->device;
+        const u32 runl = fifo->engine[engn].runl;
+        const u32 engm = BIT(engn);
+        struct gk104_fifo_engine_status status;
+        int mmui = -1;
+
+        assert_spin_locked(&fifo->base.lock);
+        if (fifo->recover.engm & engm)
+                return;
+        fifo->recover.engm |= engm;
+
+        /* Block channel assignments from changing during recovery. */
+        gk104_fifo_recover_runl(fifo, runl);
+
+        /* Determine which channel (if any) is currently on the engine. */
+        gk104_fifo_engine_status(fifo, engn, &status);
+        if (status.chan) {
+                /* The channel is not longer viable, kill it. */
+                gk104_fifo_recover_chan(&fifo->base, status.chan->id);
+        }
+
+        /* Determine MMU fault ID for the engine, if we're not being
+         * called from the fault handler already.
+         */
+        if (!status.faulted && engine) {
+                mmui = nvkm_top_fault_id(device, engine->subdev.index);
+                if (mmui < 0) {
+                        const struct nvkm_enum *en = fifo->func->fault.engine;
+                        for (; en && en->name; en++) {
+                                if (en->data2 == engine->subdev.index) {
+                                        mmui = en->value;
+                                        break;
+                                }
+                        }
+                }
+                WARN_ON(mmui < 0);
+        }
+
+        /* Trigger a MMU fault for the engine.
+         *
+         * No good idea why this is needed, but nvgpu does something similar,
+         * and it makes recovery from CTXSW_TIMEOUT a lot more reliable.
+         */
+        if (mmui >= 0) {
+                nvkm_wr32(device, 0x002a30 + (engn * 0x04), 0x00000100 | mmui);
+
+                /* Wait for fault to trigger. */
+                nvkm_msec(device, 2000,
+                        gk104_fifo_engine_status(fifo, engn, &status);
+                        if (status.faulted)
+                                break;
+                );
+
+                /* Release MMU fault trigger, and ACK the fault. */
+                nvkm_wr32(device, 0x002a30 + (engn * 0x04), 0x00000000);
+                nvkm_wr32(device, 0x00259c, BIT(mmui));
+                nvkm_wr32(device, 0x002100, 0x10000000);
+        }
+
+        /* Schedule recovery. */
+        nvkm_warn(subdev, "engine %d: scheduled for recovery\n", engn);
         schedule_work(&fifo->recover.work);
 }
 
@@ -211,34 +380,30 @@ static void
 gk104_fifo_intr_sched_ctxsw(struct gk104_fifo *fifo)
 {
         struct nvkm_device *device = fifo->base.engine.subdev.device;
-        struct gk104_fifo_chan *chan;
-        unsigned long flags;
+        unsigned long flags, engm = 0;
         u32 engn;
 
+        /* We need to ACK the SCHED_ERROR here, and prevent it reasserting,
+         * as MMU_FAULT cannot be triggered while it's pending.
+         */
         spin_lock_irqsave(&fifo->base.lock, flags);
+        nvkm_mask(device, 0x002140, 0x00000100, 0x00000000);
+        nvkm_wr32(device, 0x002100, 0x00000100);
+
         for (engn = 0; engn < fifo->engine_nr; engn++) {
-                struct nvkm_engine *engine = fifo->engine[engn].engine;
-                int runl = fifo->engine[engn].runl;
-                u32 stat = nvkm_rd32(device, 0x002640 + (engn * 0x08));
-                u32 busy = (stat & 0x80000000);
-                u32 next = (stat & 0x0fff0000) >> 16;
-                u32 chsw = (stat & 0x00008000);
-                u32 save = (stat & 0x00004000);
-                u32 load = (stat & 0x00002000);
-                u32 prev = (stat & 0x00000fff);
-                u32 chid = load ? next : prev;
-                (void)save;
-
-                if (!busy || !chsw)
+                struct gk104_fifo_engine_status status;
+
+                gk104_fifo_engine_status(fifo, engn, &status);
+                if (!status.busy || !status.chsw)
                         continue;
 
-                list_for_each_entry(chan, &fifo->runlist[runl].chan, head) {
-                        if (chan->base.chid == chid && engine) {
-                                gk104_fifo_recover(fifo, engine, chan);
-                                break;
-                        }
-                }
+                engm |= BIT(engn);
         }
+
+        for_each_set_bit(engn, &engm, fifo->engine_nr)
+                gk104_fifo_recover_engn(fifo, engn);
+
+        nvkm_mask(device, 0x002140, 0x00000100, 0x00000100);
         spin_unlock_irqrestore(&fifo->base.lock, flags);
 }
 
@@ -301,6 +466,7 @@ gk104_fifo_intr_fault(struct gk104_fifo *fifo, int unit)
         struct nvkm_fifo_chan *chan;
         unsigned long flags;
         char gpcid[8] = "", en[16] = "";
+        int engn;
 
         er = nvkm_enum_find(fifo->func->fault.reason, reason);
         eu = nvkm_enum_find(fifo->func->fault.engine, unit);
@@ -342,7 +508,8 @@ gk104_fifo_intr_fault(struct gk104_fifo *fifo, int unit)
                 snprintf(en, sizeof(en), "%s", eu->name);
         }
 
-        chan = nvkm_fifo_chan_inst(&fifo->base, (u64)inst << 12, &flags);
+        spin_lock_irqsave(&fifo->base.lock, flags);
+        chan = nvkm_fifo_chan_inst_locked(&fifo->base, (u64)inst << 12);
 
         nvkm_error(subdev,
                    "%s fault at %010llx engine %02x [%s] client %02x [%s%s] "
@@ -353,9 +520,23 @@ gk104_fifo_intr_fault(struct gk104_fifo *fifo, int unit)
                    (u64)inst << 12,
                    chan ? chan->object.client->name : "unknown");
 
-        if (engine && chan)
-                gk104_fifo_recover(fifo, engine, (void *)chan);
-        nvkm_fifo_chan_put(&fifo->base, flags, &chan);
+
+        /* Kill the channel that caused the fault. */
+        if (chan)
+                gk104_fifo_recover_chan(&fifo->base, chan->chid);
+
+        /* Channel recovery will probably have already done this for the
+         * correct engine(s), but just in case we can't find the channel
+         * information...
+         */
+        for (engn = 0; engn < fifo->engine_nr && engine; engn++) {
+                if (fifo->engine[engn].engine == engine) {
+                        gk104_fifo_recover_engn(fifo, engn);
+                        break;
+                }
+        }
+
+        spin_unlock_irqrestore(&fifo->base.lock, flags);
 }
 
 static const struct nvkm_bitfield gk104_fifo_pbdma_intr_0[] = {
@@ -716,6 +897,7 @@ gk104_fifo_ = {
         .intr = gk104_fifo_intr,
         .uevent_init = gk104_fifo_uevent_init,
         .uevent_fini = gk104_fifo_uevent_fini,
+        .recover_chan = gk104_fifo_recover_chan,
         .class_get = gk104_fifo_class_get,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
index 12d964260a29..f9e0377d3d24 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
@@ -32,6 +32,23 @@
 #include <nvif/cl906f.h>
 #include <nvif/unpack.h>
 
+int
+gf100_fifo_chan_ntfy(struct nvkm_fifo_chan *chan, u32 type,
+                     struct nvkm_event **pevent)
+{
+        switch (type) {
+        case NV906F_V0_NTFY_NON_STALL_INTERRUPT:
+                *pevent = &chan->fifo->uevent;
+                return 0;
+        case NV906F_V0_NTFY_KILLED:
+                *pevent = &chan->fifo->kevent;
+                return 0;
+        default:
+                break;
+        }
+        return -EINVAL;
+}
+
 static u32
 gf100_fifo_gpfifo_engine_addr(struct nvkm_engine *engine)
 {
@@ -184,7 +201,7 @@ gf100_fifo_gpfifo_func = {
         .dtor = gf100_fifo_gpfifo_dtor,
         .init = gf100_fifo_gpfifo_init,
         .fini = gf100_fifo_gpfifo_fini,
-        .ntfy = g84_fifo_chan_ntfy,
+        .ntfy = gf100_fifo_chan_ntfy,
         .engine_ctor = gf100_fifo_gpfifo_engine_ctor,
         .engine_dtor = gf100_fifo_gpfifo_engine_dtor,
         .engine_init = gf100_fifo_gpfifo_engine_init,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
index a2df4f3e7763..8abf6f8ef445 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
@@ -50,6 +50,7 @@ gk104_fifo_gpfifo_kick(struct gk104_fifo_chan *chan)
         ) < 0) {
                 nvkm_error(subdev, "channel %d [%s] kick timeout\n",
                            chan->base.chid, client->name);
+                nvkm_fifo_recover_chan(&fifo->base, chan->base.chid);
                 ret = -ETIMEDOUT;
         }
         mutex_unlock(&subdev->mutex);
@@ -213,7 +214,7 @@ gk104_fifo_gpfifo_func = {
         .dtor = gk104_fifo_gpfifo_dtor,
         .init = gk104_fifo_gpfifo_init,
         .fini = gk104_fifo_gpfifo_fini,
-        .ntfy = g84_fifo_chan_ntfy,
+        .ntfy = gf100_fifo_chan_ntfy,
         .engine_ctor = gk104_fifo_gpfifo_engine_ctor,
         .engine_dtor = gk104_fifo_gpfifo_engine_dtor,
         .engine_init = gk104_fifo_gpfifo_engine_init,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
index f6dfb37d9429..f889b13b5e41 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
@@ -6,6 +6,12 @@
 int nvkm_fifo_ctor(const struct nvkm_fifo_func *, struct nvkm_device *,
                    int index, int nr, struct nvkm_fifo *);
 void nvkm_fifo_uevent(struct nvkm_fifo *);
+void nvkm_fifo_cevent(struct nvkm_fifo *);
+void nvkm_fifo_kevent(struct nvkm_fifo *, int chid);
+void nvkm_fifo_recover_chan(struct nvkm_fifo *, int chid);
+
+struct nvkm_fifo_chan *
+nvkm_fifo_chan_inst_locked(struct nvkm_fifo *, u64 inst);
 
 struct nvkm_fifo_chan_oclass;
 struct nvkm_fifo_func {
@@ -18,6 +24,7 @@ struct nvkm_fifo_func {
         void (*start)(struct nvkm_fifo *, unsigned long *);
         void (*uevent_init)(struct nvkm_fifo *);
         void (*uevent_fini)(struct nvkm_fifo *);
+        void (*recover_chan)(struct nvkm_fifo *, int chid);
         int (*class_get)(struct nvkm_fifo *, int index,
                          const struct nvkm_fifo_chan_oclass **);
         const struct nvkm_fifo_chan_oclass *chan[];
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c
index 467065d1b4e6..cd8cf6f7024c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c
@@ -25,6 +25,15 @@
 
 #include <engine/fifo.h>
 
+static bool
+nvkm_gr_chsw_load(struct nvkm_engine *engine)
+{
+        struct nvkm_gr *gr = nvkm_gr(engine);
+        if (gr->func->chsw_load)
+                return gr->func->chsw_load(gr);
+        return false;
+}
+
 static void
 nvkm_gr_tile(struct nvkm_engine *engine, int region, struct nvkm_fb_tile *tile)
 {
@@ -106,6 +115,15 @@ nvkm_gr_init(struct nvkm_engine *engine)
         return gr->func->init(gr);
 }
 
+static int
+nvkm_gr_fini(struct nvkm_engine *engine, bool suspend)
+{
+        struct nvkm_gr *gr = nvkm_gr(engine);
+        if (gr->func->fini)
+                return gr->func->fini(gr, suspend);
+        return 0;
+}
+
 static void *
 nvkm_gr_dtor(struct nvkm_engine *engine)
 {
@@ -120,8 +138,10 @@ nvkm_gr = {
         .dtor = nvkm_gr_dtor,
         .oneinit = nvkm_gr_oneinit,
         .init = nvkm_gr_init,
+        .fini = nvkm_gr_fini,
         .intr = nvkm_gr_intr,
         .tile = nvkm_gr_tile,
+        .chsw_load = nvkm_gr_chsw_load,
         .fifo.cclass = nvkm_gr_cclass_new,
         .fifo.sclass = nvkm_gr_oclass_get,
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/g84.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/g84.c
index ce913300539f..da1ba74682b4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/g84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/g84.c
@@ -25,6 +25,8 @@
 
 #include <subdev/timer.h>
 
+#include <nvif/class.h>
+
 static const struct nvkm_bitfield nv50_gr_status[] = {
         { 0x00000001, "BUSY" }, /* set when any bit is set */
         { 0x00000002, "DISPATCH" },
@@ -180,11 +182,11 @@ g84_gr = {
         .tlb_flush = g84_gr_tlb_flush,
         .units = nv50_gr_units,
         .sclass = {
-                { -1, -1, 0x0030, &nv50_gr_object },
-                { -1, -1, 0x502d, &nv50_gr_object },
-                { -1, -1, 0x5039, &nv50_gr_object },
-                { -1, -1, 0x50c0, &nv50_gr_object },
-                { -1, -1, 0x8297, &nv50_gr_object },
+                { -1, -1, NV_NULL_CLASS, &nv50_gr_object },
+                { -1, -1, NV50_TWOD, &nv50_gr_object },
+                { -1, -1, NV50_MEMORY_TO_MEMORY_FORMAT, &nv50_gr_object },
+                { -1, -1, NV50_COMPUTE, &nv50_gr_object },
+                { -1, -1, G82_TESLA, &nv50_gr_object },
                 {}
         }
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
index f65a5b0a1a4d..f9acb8a944d2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
@@ -702,6 +702,22 @@ gf100_gr_pack_mmio[] = {
  * PGRAPH engine/subdev functions
  ******************************************************************************/
 
+static bool
+gf100_gr_chsw_load(struct nvkm_gr *base)
+{
+        struct gf100_gr *gr = gf100_gr(base);
+        if (!gr->firmware) {
+                u32 trace = nvkm_rd32(gr->base.engine.subdev.device, 0x40981c);
+                if (trace & 0x00000040)
+                        return true;
+        } else {
+                u32 mthd = nvkm_rd32(gr->base.engine.subdev.device, 0x409808);
+                if (mthd & 0x00080000)
+                        return true;
+        }
+        return false;
+}
+
 int
 gf100_gr_rops(struct gf100_gr *gr)
 {
@@ -1136,7 +1152,7 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
         if (trap & 0x00000008) {
                 u32 stat = nvkm_rd32(device, 0x408030);
 
-                nvkm_snprintbf(error, sizeof(error), gf100_m2mf_error,
+                nvkm_snprintbf(error, sizeof(error), gf100_ccache_error,
                                stat & 0x3fffffff);
                 nvkm_error(subdev, "CCACHE %08x [%s]\n", stat, error);
                 nvkm_wr32(device, 0x408030, 0xc0000000);
@@ -1391,26 +1407,11 @@ gf100_gr_intr(struct nvkm_gr *base)
 }
 
 static void
-gf100_gr_init_fw(struct gf100_gr *gr, u32 fuc_base,
+gf100_gr_init_fw(struct nvkm_falcon *falcon,
                  struct gf100_gr_fuc *code, struct gf100_gr_fuc *data)
 {
-        struct nvkm_device *device = gr->base.engine.subdev.device;
-        int i;
-
-        nvkm_wr32(device, fuc_base + 0x01c0, 0x01000000);
-        for (i = 0; i < data->size / 4; i++)
-                nvkm_wr32(device, fuc_base + 0x01c4, data->data[i]);
-
-        nvkm_wr32(device, fuc_base + 0x0180, 0x01000000);
-        for (i = 0; i < code->size / 4; i++) {
-                if ((i & 0x3f) == 0)
-                        nvkm_wr32(device, fuc_base + 0x0188, i >> 6);
-                nvkm_wr32(device, fuc_base + 0x0184, code->data[i]);
-        }
-
-        /* code must be padded to 0x40 words */
-        for (; i & 0x3f; i++)
-                nvkm_wr32(device, fuc_base + 0x0184, 0);
+        nvkm_falcon_load_dmem(falcon, data->data, 0x0, data->size, 0);
+        nvkm_falcon_load_imem(falcon, code->data, 0x0, code->size, 0, 0, false);
 }
 
 static void
@@ -1455,162 +1456,149 @@ gf100_gr_init_csdata(struct gf100_gr *gr,
         nvkm_wr32(device, falcon + 0x01c4, star + 4);
 }
 
-int
-gf100_gr_init_ctxctl(struct gf100_gr *gr)
+/* Initialize context from an external (secure or not) firmware */
+static int
+gf100_gr_init_ctxctl_ext(struct gf100_gr *gr)
 {
-        const struct gf100_grctx_func *grctx = gr->func->grctx;
         struct nvkm_subdev *subdev = &gr->base.engine.subdev;
         struct nvkm_device *device = subdev->device;
         struct nvkm_secboot *sb = device->secboot;
-        int i;
         int ret = 0;
 
-        if (gr->firmware) {
-                /* load fuc microcode */
-                nvkm_mc_unk260(device, 0);
-
-                /* securely-managed falcons must be reset using secure boot */
-                if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_FECS))
-                        ret = nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_FECS);
-                else
-                        gf100_gr_init_fw(gr, 0x409000, &gr->fuc409c,
-                                         &gr->fuc409d);
-                if (ret)
-                        return ret;
+        /* load fuc microcode */
+        nvkm_mc_unk260(device, 0);
 
-                if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_GPCCS))
-                        ret = nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_GPCCS);
-                else
-                        gf100_gr_init_fw(gr, 0x41a000, &gr->fuc41ac,
-                                         &gr->fuc41ad);
-                if (ret)
-                        return ret;
+        /* securely-managed falcons must be reset using secure boot */
+        if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_FECS))
+                ret = nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_FECS);
+        else
+                gf100_gr_init_fw(gr->fecs, &gr->fuc409c, &gr->fuc409d);
+        if (ret)
+                return ret;
 
-                nvkm_mc_unk260(device, 1);
-
-                /* start both of them running */
-                nvkm_wr32(device, 0x409840, 0xffffffff);
-                nvkm_wr32(device, 0x41a10c, 0x00000000);
-                nvkm_wr32(device, 0x40910c, 0x00000000);
-
-                if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_GPCCS))
-                        nvkm_secboot_start(sb, NVKM_SECBOOT_FALCON_GPCCS);
-                else
-                        nvkm_wr32(device, 0x41a100, 0x00000002);
-                if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_FECS))
-                        nvkm_secboot_start(sb, NVKM_SECBOOT_FALCON_FECS);
-                else
-                        nvkm_wr32(device, 0x409100, 0x00000002);
-                if (nvkm_msec(device, 2000,
-                        if (nvkm_rd32(device, 0x409800) & 0x00000001)
-                                break;
-                ) < 0)
-                        return -EBUSY;
+        if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_GPCCS))
+                ret = nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_GPCCS);
+        else
+                gf100_gr_init_fw(gr->gpccs, &gr->fuc41ac, &gr->fuc41ad);
+        if (ret)
+                return ret;
+
+        nvkm_mc_unk260(device, 1);
+
+        /* start both of them running */
+        nvkm_wr32(device, 0x409840, 0xffffffff);
+        nvkm_wr32(device, 0x41a10c, 0x00000000);
+        nvkm_wr32(device, 0x40910c, 0x00000000);
+
+        nvkm_falcon_start(gr->gpccs);
+        nvkm_falcon_start(gr->fecs);
 
-                nvkm_wr32(device, 0x409840, 0xffffffff);
-                nvkm_wr32(device, 0x409500, 0x7fffffff);
-                nvkm_wr32(device, 0x409504, 0x00000021);
+        if (nvkm_msec(device, 2000,
+                if (nvkm_rd32(device, 0x409800) & 0x00000001)
+                        break;
+        ) < 0)
+                return -EBUSY;
+
+        nvkm_wr32(device, 0x409840, 0xffffffff);
+        nvkm_wr32(device, 0x409500, 0x7fffffff);
+        nvkm_wr32(device, 0x409504, 0x00000021);
+
+        nvkm_wr32(device, 0x409840, 0xffffffff);
+        nvkm_wr32(device, 0x409500, 0x00000000);
+        nvkm_wr32(device, 0x409504, 0x00000010);
+        if (nvkm_msec(device, 2000,
+                if ((gr->size = nvkm_rd32(device, 0x409800)))
+                        break;
+        ) < 0)
+                return -EBUSY;
+
+        nvkm_wr32(device, 0x409840, 0xffffffff);
+        nvkm_wr32(device, 0x409500, 0x00000000);
+        nvkm_wr32(device, 0x409504, 0x00000016);
+        if (nvkm_msec(device, 2000,
+                if (nvkm_rd32(device, 0x409800))
+                        break;
+        ) < 0)
+                return -EBUSY;
+
+        nvkm_wr32(device, 0x409840, 0xffffffff);
+        nvkm_wr32(device, 0x409500, 0x00000000);
+        nvkm_wr32(device, 0x409504, 0x00000025);
+        if (nvkm_msec(device, 2000,
+                if (nvkm_rd32(device, 0x409800))
+                        break;
+        ) < 0)
+                return -EBUSY;
 
-                nvkm_wr32(device, 0x409840, 0xffffffff);
-                nvkm_wr32(device, 0x409500, 0x00000000);
-                nvkm_wr32(device, 0x409504, 0x00000010);
+        if (device->chipset >= 0xe0) {
+                nvkm_wr32(device, 0x409800, 0x00000000);
+                nvkm_wr32(device, 0x409500, 0x00000001);
+                nvkm_wr32(device, 0x409504, 0x00000030);
                 if (nvkm_msec(device, 2000,
-                        if ((gr->size = nvkm_rd32(device, 0x409800)))
+                        if (nvkm_rd32(device, 0x409800))
                                 break;
                 ) < 0)
                         return -EBUSY;
 
-                nvkm_wr32(device, 0x409840, 0xffffffff);
-                nvkm_wr32(device, 0x409500, 0x00000000);
-                nvkm_wr32(device, 0x409504, 0x00000016);
+                nvkm_wr32(device, 0x409810, 0xb00095c8);
+                nvkm_wr32(device, 0x409800, 0x00000000);
+                nvkm_wr32(device, 0x409500, 0x00000001);
+                nvkm_wr32(device, 0x409504, 0x00000031);
                 if (nvkm_msec(device, 2000,
                         if (nvkm_rd32(device, 0x409800))
                                 break;
                 ) < 0)
                         return -EBUSY;
 
-                nvkm_wr32(device, 0x409840, 0xffffffff);
-                nvkm_wr32(device, 0x409500, 0x00000000);
-                nvkm_wr32(device, 0x409504, 0x00000025);
+                nvkm_wr32(device, 0x409810, 0x00080420);
+                nvkm_wr32(device, 0x409800, 0x00000000);
+                nvkm_wr32(device, 0x409500, 0x00000001);
+                nvkm_wr32(device, 0x409504, 0x00000032);
                 if (nvkm_msec(device, 2000,
                         if (nvkm_rd32(device, 0x409800))
                                 break;
                 ) < 0)
                         return -EBUSY;
 
-                if (device->chipset >= 0xe0) {
-                        nvkm_wr32(device, 0x409800, 0x00000000);
-                        nvkm_wr32(device, 0x409500, 0x00000001);
-                        nvkm_wr32(device, 0x409504, 0x00000030);
-                        if (nvkm_msec(device, 2000,
-                                if (nvkm_rd32(device, 0x409800))
-                                        break;
-                        ) < 0)
-                                return -EBUSY;
-
-                        nvkm_wr32(device, 0x409810, 0xb00095c8);
-                        nvkm_wr32(device, 0x409800, 0x00000000);
-                        nvkm_wr32(device, 0x409500, 0x00000001);
-                        nvkm_wr32(device, 0x409504, 0x00000031);
-                        if (nvkm_msec(device, 2000,
-                                if (nvkm_rd32(device, 0x409800))
-                                        break;
-                        ) < 0)
-                                return -EBUSY;
-
-                        nvkm_wr32(device, 0x409810, 0x00080420);
-                        nvkm_wr32(device, 0x409800, 0x00000000);
-                        nvkm_wr32(device, 0x409500, 0x00000001);
-                        nvkm_wr32(device, 0x409504, 0x00000032);
-                        if (nvkm_msec(device, 2000,
-                                if (nvkm_rd32(device, 0x409800))
-                                        break;
-                        ) < 0)
-                                return -EBUSY;
+                nvkm_wr32(device, 0x409614, 0x00000070);
+                nvkm_wr32(device, 0x409614, 0x00000770);
+                nvkm_wr32(device, 0x40802c, 0x00000001);
+        }
 
-                        nvkm_wr32(device, 0x409614, 0x00000070);
-                        nvkm_wr32(device, 0x409614, 0x00000770);
-                        nvkm_wr32(device, 0x40802c, 0x00000001);
+        if (gr->data == NULL) {
+                int ret = gf100_grctx_generate(gr);
+                if (ret) {
+                        nvkm_error(subdev, "failed to construct context\n");
+                        return ret;
                 }
+        }
 
-                if (gr->data == NULL) {
-                        int ret = gf100_grctx_generate(gr);
-                        if (ret) {
-                                nvkm_error(subdev, "failed to construct context\n");
-                                return ret;
-                        }
-                }
+        return 0;
+}
+
+static int
+gf100_gr_init_ctxctl_int(struct gf100_gr *gr)
+{
+        const struct gf100_grctx_func *grctx = gr->func->grctx;
+        struct nvkm_subdev *subdev = &gr->base.engine.subdev;
+        struct nvkm_device *device = subdev->device;
 
-                return 0;
-        } else
         if (!gr->func->fecs.ucode) {
                 return -ENOSYS;
         }
 
         /* load HUB microcode */
         nvkm_mc_unk260(device, 0);
-        nvkm_wr32(device, 0x4091c0, 0x01000000);
-        for (i = 0; i < gr->func->fecs.ucode->data.size / 4; i++)
-                nvkm_wr32(device, 0x4091c4, gr->func->fecs.ucode->data.data[i]);
-
-        nvkm_wr32(device, 0x409180, 0x01000000);
-        for (i = 0; i < gr->func->fecs.ucode->code.size / 4; i++) {
-                if ((i & 0x3f) == 0)
-                        nvkm_wr32(device, 0x409188, i >> 6);
-                nvkm_wr32(device, 0x409184, gr->func->fecs.ucode->code.data[i]);
-        }
+        nvkm_falcon_load_dmem(gr->fecs, gr->func->fecs.ucode->data.data, 0x0,
+                              gr->func->fecs.ucode->data.size, 0);
+        nvkm_falcon_load_imem(gr->fecs, gr->func->fecs.ucode->code.data, 0x0,
+                              gr->func->fecs.ucode->code.size, 0, 0, false);
 
         /* load GPC microcode */
-        nvkm_wr32(device, 0x41a1c0, 0x01000000);
-        for (i = 0; i < gr->func->gpccs.ucode->data.size / 4; i++)
-                nvkm_wr32(device, 0x41a1c4, gr->func->gpccs.ucode->data.data[i]);
-
-        nvkm_wr32(device, 0x41a180, 0x01000000);
-        for (i = 0; i < gr->func->gpccs.ucode->code.size / 4; i++) {
-                if ((i & 0x3f) == 0)
-                        nvkm_wr32(device, 0x41a188, i >> 6);
-                nvkm_wr32(device, 0x41a184, gr->func->gpccs.ucode->code.data[i]);
-        }
+        nvkm_falcon_load_dmem(gr->gpccs, gr->func->gpccs.ucode->data.data, 0x0,
+                              gr->func->gpccs.ucode->data.size, 0);
+        nvkm_falcon_load_imem(gr->gpccs, gr->func->gpccs.ucode->code.data, 0x0,
+                              gr->func->gpccs.ucode->code.size, 0, 0, false);
         nvkm_mc_unk260(device, 1);
 
         /* load register lists */
@@ -1642,6 +1630,19 @@ gf100_gr_init_ctxctl(struct gf100_gr *gr)
         return 0;
 }
 
+int
+gf100_gr_init_ctxctl(struct gf100_gr *gr)
+{
+        int ret;
+
+        if (gr->firmware)
+                ret = gf100_gr_init_ctxctl_ext(gr);
+        else
+                ret = gf100_gr_init_ctxctl_int(gr);
+
+        return ret;
+}
+
 static int
 gf100_gr_oneinit(struct nvkm_gr *base)
 {
@@ -1711,10 +1712,32 @@ static int
 gf100_gr_init_(struct nvkm_gr *base)
 {
         struct gf100_gr *gr = gf100_gr(base);
+        struct nvkm_subdev *subdev = &base->engine.subdev;
+        u32 ret;
+
         nvkm_pmu_pgob(gr->base.engine.subdev.device->pmu, false);
+
+        ret = nvkm_falcon_get(gr->fecs, subdev);
+        if (ret)
+                return ret;
+
+        ret = nvkm_falcon_get(gr->gpccs, subdev);
+        if (ret)
+                return ret;
+
         return gr->func->init(gr);
 }
 
+static int
+gf100_gr_fini_(struct nvkm_gr *base, bool suspend)
+{
+        struct gf100_gr *gr = gf100_gr(base);
+        struct nvkm_subdev *subdev = &gr->base.engine.subdev;
+        nvkm_falcon_put(gr->gpccs, subdev);
+        nvkm_falcon_put(gr->fecs, subdev);
+        return 0;
+}
+
 void
 gf100_gr_dtor_fw(struct gf100_gr_fuc *fuc)
 {
@@ -1737,6 +1760,9 @@ gf100_gr_dtor(struct nvkm_gr *base)
                 gr->func->dtor(gr);
         kfree(gr->data);
 
+        nvkm_falcon_del(&gr->gpccs);
+        nvkm_falcon_del(&gr->fecs);
+
         gf100_gr_dtor_fw(&gr->fuc409c);
         gf100_gr_dtor_fw(&gr->fuc409d);
         gf100_gr_dtor_fw(&gr->fuc41ac);
@@ -1755,10 +1781,12 @@ gf100_gr_ = {
         .dtor = gf100_gr_dtor,
         .oneinit = gf100_gr_oneinit,
         .init = gf100_gr_init_,
+        .fini = gf100_gr_fini_,
         .intr = gf100_gr_intr,
         .units = gf100_gr_units,
         .chan_new = gf100_gr_chan_new,
         .object_get = gf100_gr_object_get,
+        .chsw_load = gf100_gr_chsw_load,
 };
 
 int
@@ -1828,6 +1856,7 @@ int
 gf100_gr_ctor(const struct gf100_gr_func *func, struct nvkm_device *device,
               int index, struct gf100_gr *gr)
 {
+        struct nvkm_subdev *subdev = &gr->base.engine.subdev;
         int ret;
 
         gr->func = func;
@@ -1840,7 +1869,11 @@ gf100_gr_ctor(const struct gf100_gr_func *func, struct nvkm_device *device,
         if (ret)
                 return ret;
 
-        return 0;
+        ret = nvkm_falcon_v1_new(subdev, "FECS", 0x409000, &gr->fecs);
+        if (ret)
+                return ret;
+
+        return nvkm_falcon_v1_new(subdev, "GPCCS", 0x41a000, &gr->gpccs);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
index 268b8d60ff73..db6ee3b06841 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
@@ -29,6 +29,7 @@
 #include <core/gpuobj.h>
 #include <subdev/ltc.h>
 #include <subdev/mmu.h>
+#include <engine/falcon.h>
 
 #define GPC_MAX 32
 #define TPC_MAX_PER_GPC 8
@@ -75,6 +76,8 @@ struct gf100_gr {
         const struct gf100_gr_func *func;
         struct nvkm_gr base;
 
+        struct nvkm_falcon *fecs;
+        struct nvkm_falcon *gpccs;
         struct gf100_gr_fuc fuc409c;
         struct gf100_gr_fuc fuc409d;
         struct gf100_gr_fuc fuc41ac;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt200.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt200.c
index 2e68919f00b2..c711a55ce392 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt200.c
@@ -23,6 +23,8 @@
  */
 #include "nv50.h"
 
+#include <nvif/class.h>
+
 static const struct nvkm_gr_func
 gt200_gr = {
         .init = nv50_gr_init,
@@ -31,11 +33,11 @@ gt200_gr = {
         .tlb_flush = g84_gr_tlb_flush,
         .units = nv50_gr_units,
         .sclass = {
-                { -1, -1, 0x0030, &nv50_gr_object },
-                { -1, -1, 0x502d, &nv50_gr_object },
-                { -1, -1, 0x5039, &nv50_gr_object },
-                { -1, -1, 0x50c0, &nv50_gr_object },
-                { -1, -1, 0x8397, &nv50_gr_object },
+                { -1, -1, NV_NULL_CLASS, &nv50_gr_object },
+                { -1, -1, NV50_TWOD, &nv50_gr_object },
+                { -1, -1, NV50_MEMORY_TO_MEMORY_FORMAT, &nv50_gr_object },
+                { -1, -1, NV50_COMPUTE, &nv50_gr_object },
+                { -1, -1, GT200_TESLA, &nv50_gr_object },
                 {}
         }
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt215.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt215.c
index 2bf7aac360cc..fa103df32ec7 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt215.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gt215.c
@@ -23,6 +23,8 @@
  */
 #include "nv50.h"
 
+#include <nvif/class.h>
+
 static const struct nvkm_gr_func
 gt215_gr = {
         .init = nv50_gr_init,
@@ -31,12 +33,12 @@ gt215_gr = {
         .tlb_flush = g84_gr_tlb_flush,
         .units = nv50_gr_units,
         .sclass = {
-                { -1, -1, 0x0030, &nv50_gr_object },
-                { -1, -1, 0x502d, &nv50_gr_object },
-                { -1, -1, 0x5039, &nv50_gr_object },
-                { -1, -1, 0x50c0, &nv50_gr_object },
-                { -1, -1, 0x8597, &nv50_gr_object },
-                { -1, -1, 0x85c0, &nv50_gr_object },
+                { -1, -1, NV_NULL_CLASS, &nv50_gr_object },
+                { -1, -1, NV50_TWOD, &nv50_gr_object },
+                { -1, -1, NV50_MEMORY_TO_MEMORY_FORMAT, &nv50_gr_object },
+                { -1, -1, NV50_COMPUTE, &nv50_gr_object },
+                { -1, -1, GT214_TESLA, &nv50_gr_object },
+                { -1, -1, GT214_COMPUTE, &nv50_gr_object },
                 {}
         }
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp79.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp79.c
index 95d5219faf93..eb1a90644752 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp79.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp79.c
@@ -23,6 +23,8 @@
  */
 #include "nv50.h"
 
+#include <nvif/class.h>
+
 static const struct nvkm_gr_func
 mcp79_gr = {
         .init = nv50_gr_init,
@@ -30,11 +32,11 @@ mcp79_gr = {
         .chan_new = nv50_gr_chan_new,
         .units = nv50_gr_units,
         .sclass = {
-                { -1, -1, 0x0030, &nv50_gr_object },
-                { -1, -1, 0x502d, &nv50_gr_object },
-                { -1, -1, 0x5039, &nv50_gr_object },
-                { -1, -1, 0x50c0, &nv50_gr_object },
-                { -1, -1, 0x8397, &nv50_gr_object },
+                { -1, -1, NV_NULL_CLASS, &nv50_gr_object },
+                { -1, -1, NV50_TWOD, &nv50_gr_object },
+                { -1, -1, NV50_MEMORY_TO_MEMORY_FORMAT, &nv50_gr_object },
+                { -1, -1, NV50_COMPUTE, &nv50_gr_object },
+                { -1, -1, GT200_TESLA, &nv50_gr_object },
                 {}
         }
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp89.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp89.c
index 027b58e5976b..c91eb56e9327 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp89.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/mcp89.c
@@ -23,6 +23,8 @@
  */
 #include "nv50.h"
 
+#include <nvif/class.h>
+
 static const struct nvkm_gr_func
 mcp89_gr = {
         .init = nv50_gr_init,
@@ -31,12 +33,12 @@ mcp89_gr = {
         .tlb_flush = g84_gr_tlb_flush,
         .units = nv50_gr_units,
         .sclass = {
-                { -1, -1, 0x0030, &nv50_gr_object },
-                { -1, -1, 0x502d, &nv50_gr_object },
-                { -1, -1, 0x5039, &nv50_gr_object },
-                { -1, -1, 0x50c0, &nv50_gr_object },
-                { -1, -1, 0x85c0, &nv50_gr_object },
-                { -1, -1, 0x8697, &nv50_gr_object },
+                { -1, -1, NV_NULL_CLASS, &nv50_gr_object },
+                { -1, -1, NV50_TWOD, &nv50_gr_object },
+                { -1, -1, NV50_MEMORY_TO_MEMORY_FORMAT, &nv50_gr_object },
+                { -1, -1, NV50_COMPUTE, &nv50_gr_object },
+                { -1, -1, GT214_COMPUTE, &nv50_gr_object },
+                { -1, -1, GT21A_TESLA, &nv50_gr_object },
                 {}
         }
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c
index fca67de43f2b..df16ffda1749 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c
@@ -27,6 +27,8 @@
 #include <core/gpuobj.h>
 #include <engine/fifo.h>
 
+#include <nvif/class.h>
+
 u64
 nv50_gr_units(struct nvkm_gr *gr)
 {
@@ -778,11 +780,11 @@ nv50_gr = {
         .chan_new = nv50_gr_chan_new,
         .units = nv50_gr_units,
         .sclass = {
-                { -1, -1, 0x0030, &nv50_gr_object },
-                { -1, -1, 0x502d, &nv50_gr_object },
-                { -1, -1, 0x5039, &nv50_gr_object },
-                { -1, -1, 0x5097, &nv50_gr_object },
-                { -1, -1, 0x50c0, &nv50_gr_object },
+                { -1, -1, NV_NULL_CLASS, &nv50_gr_object },
+                { -1, -1, NV50_TWOD, &nv50_gr_object },
+                { -1, -1, NV50_MEMORY_TO_MEMORY_FORMAT, &nv50_gr_object },
+                { -1, -1, NV50_TESLA, &nv50_gr_object },
+                { -1, -1, NV50_COMPUTE, &nv50_gr_object },
                 {}
         }
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/gr/priv.h
index d8adcdf6985a..2a52d9f026ec 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/priv.h
@@ -15,6 +15,7 @@ struct nvkm_gr_func {
         void *(*dtor)(struct nvkm_gr *);
         int (*oneinit)(struct nvkm_gr *);
         int (*init)(struct nvkm_gr *);
+        int (*fini)(struct nvkm_gr *, bool);
         void (*intr)(struct nvkm_gr *);
         void (*tile)(struct nvkm_gr *, int region, struct nvkm_fb_tile *);
         int (*tlb_flush)(struct nvkm_gr *);
@@ -24,6 +25,7 @@ struct nvkm_gr_func {
         /* Returns chipset-specific counts of units packed into an u64.
          */
         u64 (*units)(struct nvkm_gr *);
+        bool (*chsw_load)(struct nvkm_gr *);
         struct nvkm_sclass sclass[];
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/falcon/Kbuild b/drivers/gpu/drm/nouveau/nvkm/falcon/Kbuild
new file mode 100644
index 000000000000..584863db9bfc
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/falcon/Kbuild
@@ -0,0 +1,2 @@
+nvkm-y += nvkm/falcon/base.o
+nvkm-y += nvkm/falcon/v1.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/falcon/base.c b/drivers/gpu/drm/nouveau/nvkm/falcon/base.c
new file mode 100644
index 000000000000..4852f313762f
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/falcon/base.c
@@ -0,0 +1,191 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+
+#include <subdev/mc.h>
+
+void
+nvkm_falcon_load_imem(struct nvkm_falcon *falcon, void *data, u32 start,
+                      u32 size, u16 tag, u8 port, bool secure)
+{
+        if (secure && !falcon->secret) {
+                nvkm_warn(falcon->user,
+                          "writing with secure tag on a non-secure falcon!\n");
+                return;
+        }
+
+        falcon->func->load_imem(falcon, data, start, size, tag, port,
+                                secure);
+}
+
+void
+nvkm_falcon_load_dmem(struct nvkm_falcon *falcon, void *data, u32 start,
+                      u32 size, u8 port)
+{
+        falcon->func->load_dmem(falcon, data, start, size, port);
+}
+
+void
+nvkm_falcon_read_dmem(struct nvkm_falcon *falcon, u32 start, u32 size, u8 port,
+                      void *data)
+{
+        falcon->func->read_dmem(falcon, start, size, port, data);
+}
+
+void
+nvkm_falcon_bind_context(struct nvkm_falcon *falcon, struct nvkm_gpuobj *inst)
+{
+        if (!falcon->func->bind_context) {
+                nvkm_error(falcon->user,
+                           "Context binding not supported on this falcon!\n");
+                return;
+        }
+
+        falcon->func->bind_context(falcon, inst);
+}
+
+void
+nvkm_falcon_set_start_addr(struct nvkm_falcon *falcon, u32 start_addr)
+{
+        falcon->func->set_start_addr(falcon, start_addr);
+}
+
+void
+nvkm_falcon_start(struct nvkm_falcon *falcon)
+{
+        falcon->func->start(falcon);
+}
+
+int
+nvkm_falcon_enable(struct nvkm_falcon *falcon)
+{
+        struct nvkm_device *device = falcon->owner->device;
+        enum nvkm_devidx id = falcon->owner->index;
+        int ret;
+
+        nvkm_mc_enable(device, id);
+        ret = falcon->func->enable(falcon);
+        if (ret) {
+                nvkm_mc_disable(device, id);
+                return ret;
+        }
+
+        return 0;
+}
+
+void
+nvkm_falcon_disable(struct nvkm_falcon *falcon)
+{
+        struct nvkm_device *device = falcon->owner->device;
+        enum nvkm_devidx id = falcon->owner->index;
+
+        /* already disabled, return or wait_idle will timeout */
+        if (!nvkm_mc_enabled(device, id))
+                return;
+
+        falcon->func->disable(falcon);
+
+        nvkm_mc_disable(device, id);
+}
+
+int
+nvkm_falcon_reset(struct nvkm_falcon *falcon)
+{
+        nvkm_falcon_disable(falcon);
+        return nvkm_falcon_enable(falcon);
+}
+
+int
+nvkm_falcon_wait_for_halt(struct nvkm_falcon *falcon, u32 ms)
+{
+        return falcon->func->wait_for_halt(falcon, ms);
+}
+
+int
+nvkm_falcon_clear_interrupt(struct nvkm_falcon *falcon, u32 mask)
+{
+        return falcon->func->clear_interrupt(falcon, mask);
+}
+
+void
+nvkm_falcon_put(struct nvkm_falcon *falcon, const struct nvkm_subdev *user)
+{
+        mutex_lock(&falcon->mutex);
+        if (falcon->user == user) {
+                nvkm_debug(falcon->user, "released %s falcon\n", falcon->name);
+                falcon->user = NULL;
+        }
+        mutex_unlock(&falcon->mutex);
+}
+
+int
+nvkm_falcon_get(struct nvkm_falcon *falcon, const struct nvkm_subdev *user)
+{
+        mutex_lock(&falcon->mutex);
+        if (falcon->user) {
+                nvkm_error(user, "%s falcon already acquired by %s!\n",
+                           falcon->name, nvkm_subdev_name[falcon->user->index]);
+                mutex_unlock(&falcon->mutex);
+                return -EBUSY;
+        }
+
+        nvkm_debug(user, "acquired %s falcon\n", falcon->name);
+        falcon->user = user;
+        mutex_unlock(&falcon->mutex);
+        return 0;
+}
+
+void
+nvkm_falcon_ctor(const struct nvkm_falcon_func *func,
+                 struct nvkm_subdev *subdev, const char *name, u32 addr,
+                 struct nvkm_falcon *falcon)
+{
+        u32 reg;
+
+        falcon->func = func;
+        falcon->owner = subdev;
+        falcon->name = name;
+        falcon->addr = addr;
+        mutex_init(&falcon->mutex);
+
+        reg = nvkm_falcon_rd32(falcon, 0x12c);
+        falcon->version = reg & 0xf;
+        falcon->secret = (reg >> 4) & 0x3;
+        falcon->code.ports = (reg >> 8) & 0xf;
+        falcon->data.ports = (reg >> 12) & 0xf;
+
+        reg = nvkm_falcon_rd32(falcon, 0x108);
+        falcon->code.limit = (reg & 0x1ff) << 8;
+        falcon->data.limit = (reg & 0x3fe00) >> 1;
+
+        reg = nvkm_falcon_rd32(falcon, 0xc08);
+        falcon->debug = (reg >> 20) & 0x1;
+}
+
+void
+nvkm_falcon_del(struct nvkm_falcon **pfalcon)
+{
+        if (*pfalcon) {
+                kfree(*pfalcon);
+                *pfalcon = NULL;
+        }
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/falcon/priv.h b/drivers/gpu/drm/nouveau/nvkm/falcon/priv.h
new file mode 100644
index 000000000000..97b56f759d0b
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/falcon/priv.h
@@ -0,0 +1,8 @@
+#ifndef __NVKM_FALCON_PRIV_H__
+#define __NVKM_FALCON_PRIV_H__
+#include <engine/falcon.h>
+
+void
+nvkm_falcon_ctor(const struct nvkm_falcon_func *, struct nvkm_subdev *,
+                 const char *, u32, struct nvkm_falcon *);
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/falcon/v1.c b/drivers/gpu/drm/nouveau/nvkm/falcon/v1.c
new file mode 100644
index 000000000000..b537f111f39c
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/falcon/v1.c
@@ -0,0 +1,266 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+
+#include <core/gpuobj.h>
+#include <core/memory.h>
+#include <subdev/timer.h>
+
+static void
+nvkm_falcon_v1_load_imem(struct nvkm_falcon *falcon, void *data, u32 start,
+                         u32 size, u16 tag, u8 port, bool secure)
+{
+        u8 rem = size % 4;
+        u32 reg;
+        int i;
+
+        size -= rem;
+
+        reg = start | BIT(24) | (secure ? BIT(28) : 0);
+        nvkm_falcon_wr32(falcon, 0x180 + (port * 16), reg);
+        for (i = 0; i < size / 4; i++) {
+                /* write new tag every 256B */
+                if ((i & 0x3f) == 0)
+                        nvkm_falcon_wr32(falcon, 0x188, tag++);
+                nvkm_falcon_wr32(falcon, 0x184, ((u32 *)data)[i]);
+        }
+
+        /*
+         * If size is not a multiple of 4, mask the last work to ensure garbage
+         * does not get written
+         */
+        if (rem) {
+                u32 extra = ((u32 *)data)[i];
+
+                /* write new tag every 256B */
+                if ((i & 0x3f) == 0)
+                        nvkm_falcon_wr32(falcon, 0x188, tag++);
+                nvkm_falcon_wr32(falcon, 0x184, extra & (BIT(rem * 8) - 1));
+                ++i;
+        }
+
+        /* code must be padded to 0x40 words */
+        for (; i & 0x3f; i++)
+                nvkm_falcon_wr32(falcon, 0x184, 0);
+}
+
+static void
+nvkm_falcon_v1_load_dmem(struct nvkm_falcon *falcon, void *data, u32 start,
+                      u32 size, u8 port)
+{
+        u8 rem = size % 4;
+        int i;
+
+        size -= rem;
+
+        nvkm_falcon_wr32(falcon, 0x1c0 + (port * 16), start | (0x1 << 24));
+        for (i = 0; i < size / 4; i++)
+                nvkm_falcon_wr32(falcon, 0x1c4, ((u32 *)data)[i]);
+
+        /*
+         * If size is not a multiple of 4, mask the last work to ensure garbage
+         * does not get read
+         */
+        if (rem) {
+                u32 extra = ((u32 *)data)[i];
+
+                nvkm_falcon_wr32(falcon, 0x1c4, extra & (BIT(rem * 8) - 1));
+        }
+}
+
+static void
+nvkm_falcon_v1_read_dmem(struct nvkm_falcon *falcon, u32 start, u32 size,
+                         u8 port, void *data)
+{
+        u8 rem = size % 4;
+        int i;
+
+        size -= rem;
+
+        nvkm_falcon_wr32(falcon, 0x1c0 + (port * 16), start | (0x1 << 25));
+        for (i = 0; i < size / 4; i++)
+                ((u32 *)data)[i] = nvkm_falcon_rd32(falcon, 0x1c4);
+
+        /*
+         * If size is not a multiple of 4, mask the last work to ensure garbage
+         * does not get read
+         */
+        if (rem) {
+                u32 extra = nvkm_falcon_rd32(falcon, 0x1c4);
+
+                for (i = size; i < size + rem; i++) {
+                        ((u8 *)data)[i] = (u8)(extra & 0xff);
+                        extra >>= 8;
+                }
+        }
+}
+
+static void
+nvkm_falcon_v1_bind_context(struct nvkm_falcon *falcon, struct nvkm_gpuobj *ctx)
+{
+        u32 inst_loc;
+
+        /* disable instance block binding */
+        if (ctx == NULL) {
+                nvkm_falcon_wr32(falcon, 0x10c, 0x0);
+                return;
+        }
+
+        nvkm_falcon_wr32(falcon, 0x10c, 0x1);
+
+        /* setup apertures - virtual */
+        nvkm_falcon_wr32(falcon, 0xe00 + 4 * FALCON_DMAIDX_UCODE, 0x4);
+        nvkm_falcon_wr32(falcon, 0xe00 + 4 * FALCON_DMAIDX_VIRT, 0x0);
+        /* setup apertures - physical */
+        nvkm_falcon_wr32(falcon, 0xe00 + 4 * FALCON_DMAIDX_PHYS_VID, 0x4);
+        nvkm_falcon_wr32(falcon, 0xe00 + 4 * FALCON_DMAIDX_PHYS_SYS_COH, 0x5);
+        nvkm_falcon_wr32(falcon, 0xe00 + 4 * FALCON_DMAIDX_PHYS_SYS_NCOH, 0x6);
+
+        /* Set context */
+        switch (nvkm_memory_target(ctx->memory)) {
+        case NVKM_MEM_TARGET_VRAM: inst_loc = 0; break;
+        case NVKM_MEM_TARGET_NCOH: inst_loc = 3; break;
+        default:
+                WARN_ON(1);
+                return;
+        }
+
+        /* Enable context */
+        nvkm_falcon_mask(falcon, 0x048, 0x1, 0x1);
+        nvkm_falcon_wr32(falcon, 0x480,
+                         ((ctx->addr >> 12) & 0xfffffff) |
+                         (inst_loc << 28) | (1 << 30));
+}
+
+static void
+nvkm_falcon_v1_set_start_addr(struct nvkm_falcon *falcon, u32 start_addr)
+{
+        nvkm_falcon_wr32(falcon, 0x104, start_addr);
+}
+
+static void
+nvkm_falcon_v1_start(struct nvkm_falcon *falcon)
+{
+        u32 reg = nvkm_falcon_rd32(falcon, 0x100);
+
+        if (reg & BIT(6))
+                nvkm_falcon_wr32(falcon, 0x130, 0x2);
+        else
+                nvkm_falcon_wr32(falcon, 0x100, 0x2);
+}
+
+static int
+nvkm_falcon_v1_wait_for_halt(struct nvkm_falcon *falcon, u32 ms)
+{
+        struct nvkm_device *device = falcon->owner->device;
+        int ret;
+
+        ret = nvkm_wait_msec(device, ms, falcon->addr + 0x100, 0x10, 0x10);
+        if (ret < 0)
+                return ret;
+
+        return 0;
+}
+
+static int
+nvkm_falcon_v1_clear_interrupt(struct nvkm_falcon *falcon, u32 mask)
+{
+        struct nvkm_device *device = falcon->owner->device;
+        int ret;
+
+        /* clear interrupt(s) */
+        nvkm_falcon_mask(falcon, 0x004, mask, mask);
+        /* wait until interrupts are cleared */
+        ret = nvkm_wait_msec(device, 10, falcon->addr + 0x008, mask, 0x0);
+        if (ret < 0)
+                return ret;
+
+        return 0;
+}
+
+static int
+falcon_v1_wait_idle(struct nvkm_falcon *falcon)
+{
+        struct nvkm_device *device = falcon->owner->device;
+        int ret;
+
+        ret = nvkm_wait_msec(device, 10, falcon->addr + 0x04c, 0xffff, 0x0);
+        if (ret < 0)
+                return ret;
+
+        return 0;
+}
+
+static int
+nvkm_falcon_v1_enable(struct nvkm_falcon *falcon)
+{
+        struct nvkm_device *device = falcon->owner->device;
+        int ret;
+
+        ret = nvkm_wait_msec(device, 10, falcon->addr + 0x10c, 0x6, 0x0);
+        if (ret < 0) {
+                nvkm_error(falcon->user, "Falcon mem scrubbing timeout\n");
+                return ret;
+        }
+
+        ret = falcon_v1_wait_idle(falcon);
+        if (ret)
+                return ret;
+
+        /* enable IRQs */
+        nvkm_falcon_wr32(falcon, 0x010, 0xff);
+
+        return 0;
+}
+
+static void
+nvkm_falcon_v1_disable(struct nvkm_falcon *falcon)
+{
+        /* disable IRQs and wait for any previous code to complete */
+        nvkm_falcon_wr32(falcon, 0x014, 0xff);
+        falcon_v1_wait_idle(falcon);
+}
+
+static const struct nvkm_falcon_func
+nvkm_falcon_v1 = {
+        .load_imem = nvkm_falcon_v1_load_imem,
+        .load_dmem = nvkm_falcon_v1_load_dmem,
+        .read_dmem = nvkm_falcon_v1_read_dmem,
+        .bind_context = nvkm_falcon_v1_bind_context,
+        .start = nvkm_falcon_v1_start,
+        .wait_for_halt = nvkm_falcon_v1_wait_for_halt,
+        .clear_interrupt = nvkm_falcon_v1_clear_interrupt,
+        .enable = nvkm_falcon_v1_enable,
+        .disable = nvkm_falcon_v1_disable,
+        .set_start_addr = nvkm_falcon_v1_set_start_addr,
+};
+
+int
+nvkm_falcon_v1_new(struct nvkm_subdev *owner, const char *name, u32 addr,
+                   struct nvkm_falcon **pfalcon)
+{
+        struct nvkm_falcon *falcon;
+        if (!(falcon = *pfalcon = kzalloc(sizeof(*falcon), GFP_KERNEL)))
+                return -ENOMEM;
+        nvkm_falcon_ctor(&nvkm_falcon_v1, owner, name, addr, falcon);
+        return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/Kbuild
index be57220a2e01..6b4f1e06a38f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/Kbuild
@@ -19,6 +19,7 @@ nvkm-y += nvkm/subdev/bios/pcir.o
 nvkm-y += nvkm/subdev/bios/perf.o
 nvkm-y += nvkm/subdev/bios/pll.o
 nvkm-y += nvkm/subdev/bios/pmu.o
+nvkm-y += nvkm/subdev/bios/power_budget.o
 nvkm-y += nvkm/subdev/bios/ramcfg.o
 nvkm-y += nvkm/subdev/bios/rammap.o
 nvkm-y += nvkm/subdev/bios/shadow.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/power_budget.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/power_budget.c
new file mode 100644
index 000000000000..617bfffce4ad
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/power_budget.c
@@ -0,0 +1,126 @@
+/*
+ * Copyright 2016 Karol Herbst
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Karol Herbst
+ */
+#include <subdev/bios.h>
+#include <subdev/bios/bit.h>
+#include <subdev/bios/power_budget.h>
+
+static u32
+nvbios_power_budget_table(struct nvkm_bios *bios, u8 *ver, u8 *hdr, u8 *cnt,
+                          u8 *len)
+{
+        struct bit_entry bit_P;
+        u32 power_budget;
+
+        if (bit_entry(bios, 'P', &bit_P) || bit_P.version != 2 ||
+            bit_P.length < 0x2c)
+                return 0;
+
+        power_budget = nvbios_rd32(bios, bit_P.offset + 0x2c);
+        if (!power_budget)
+                return 0;
+
+        *ver = nvbios_rd08(bios, power_budget);
+        switch (*ver) {
+        case 0x20:
+        case 0x30:
+                *hdr = nvbios_rd08(bios, power_budget + 0x1);
+                *len = nvbios_rd08(bios, power_budget + 0x2);
+                *cnt = nvbios_rd08(bios, power_budget + 0x3);
+                return power_budget;
+        default:
+                break;
+        }
+
+        return 0;
+}
+
+int
+nvbios_power_budget_header(struct nvkm_bios *bios,
+                           struct nvbios_power_budget *budget)
+{
+        struct nvkm_subdev *subdev = &bios->subdev;
+        u8 ver, hdr, cnt, len, cap_entry;
+        u32 header;
+
+        if (!bios || !budget)
+                return -EINVAL;
+
+        header = nvbios_power_budget_table(bios, &ver, &hdr, &cnt, &len);
+        if (!header || !cnt)
+                return -ENODEV;
+
+        switch (ver) {
+        case 0x20:
+                cap_entry = nvbios_rd08(bios, header + 0x9);
+                break;
+        case 0x30:
+                cap_entry = nvbios_rd08(bios, header + 0xa);
+                break;
+        default:
+                cap_entry = 0xff;
+        }
+
+        if (cap_entry >= cnt && cap_entry != 0xff) {
+                nvkm_warn(subdev,
+                          "invalid cap_entry in power budget table found\n");
+                budget->cap_entry = 0xff;
+                return -EINVAL;
+        }
+
+        budget->offset = header;
+        budget->ver = ver;
+        budget->hlen = hdr;
+        budget->elen = len;
+        budget->ecount = cnt;
+
+        budget->cap_entry = cap_entry;
+
+        return 0;
+}
+
+int
+nvbios_power_budget_entry(struct nvkm_bios *bios,
+                          struct nvbios_power_budget *budget,
+                          u8 idx, struct nvbios_power_budget_entry *entry)
+{
+        u32 entry_offset;
+
+        if (!bios || !budget || !budget->offset || idx >= budget->ecount
+                || !entry)
+                return -EINVAL;
+
+        entry_offset = budget->offset + budget->hlen + idx * budget->elen;
+
+        if (budget->ver >= 0x20) {
+                entry->min_w = nvbios_rd32(bios, entry_offset + 0x2);
+                entry->avg_w = nvbios_rd32(bios, entry_offset + 0x6);
+                entry->max_w = nvbios_rd32(bios, entry_offset + 0xa);
+        } else {
+                entry->min_w = 0;
+                entry->max_w = nvbios_rd32(bios, entry_offset + 0x2);
+                entry->avg_w = entry->max_w;
+        }
+
+        return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/nv50.c
index 5841f297973c..da1770e47490 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/nv50.c
@@ -112,7 +112,7 @@ read_pll_src(struct nv50_clk *clk, u32 base)
                 M    = (coef & 0x000000ff) >> 0;
                 break;
         default:
-                BUG_ON(1);
+                BUG();
         }
 
         if (M)
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/nv50.c
index c714b097719c..59362f8dee22 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/nv50.c
@@ -50,7 +50,7 @@ nv50_devinit_pll_set(struct nvkm_devinit *init, u32 type, u32 freq)
         ret = nv04_pll_calc(subdev, &info, freq, &N1, &M1, &N2, &M2, &P);
         if (!ret) {
                 nvkm_error(subdev, "failed pll calculation\n");
-                return ret;
+                return -EINVAL;
         }
 
         switch (info.type) {
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgf100.c
index 093223d1df4f..6758da93a3a1 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgf100.c
@@ -445,7 +445,7 @@ gf100_ram_get(struct nvkm_ram *ram, u64 size, u32 align, u32 ncmin,
 {
         struct nvkm_ltc *ltc = ram->fb->subdev.device->ltc;
         struct nvkm_mm *mm = &ram->vram;
-        struct nvkm_mm_node *r;
+        struct nvkm_mm_node **node, *r;
         struct nvkm_mem *mem;
         int type = (memtype & 0x0ff);
         int back = (memtype & 0x800);
@@ -462,7 +462,6 @@ gf100_ram_get(struct nvkm_ram *ram, u64 size, u32 align, u32 ncmin,
         if (!mem)
                 return -ENOMEM;
 
-        INIT_LIST_HEAD(&mem->regions);
         mem->size = size;
 
         mutex_lock(&ram->fb->subdev.mutex);
@@ -478,6 +477,7 @@ gf100_ram_get(struct nvkm_ram *ram, u64 size, u32 align, u32 ncmin,
         }
         mem->memtype = type;
 
+        node = &mem->mem;
         do {
                 if (back)
                         ret = nvkm_mm_tail(mm, 0, 1, size, ncmin, align, &r);
@@ -489,13 +489,13 @@ gf100_ram_get(struct nvkm_ram *ram, u64 size, u32 align, u32 ncmin,
                         return ret;
                 }
 
-                list_add_tail(&r->rl_entry, &mem->regions);
+                *node = r;
+                node = &r->next;
                 size -= r->length;
         } while (size);
         mutex_unlock(&ram->fb->subdev.mutex);
 
-        r = list_first_entry(&mem->regions, struct nvkm_mm_node, rl_entry);
-        mem->offset = (u64)r->offset << NVKM_RAM_MM_SHIFT;
+        mem->offset = (u64)mem->mem->offset << NVKM_RAM_MM_SHIFT;
         *pmem = mem;
         return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgk104.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgk104.c
index 7904fa41acef..fb8a1239743d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgk104.c
@@ -989,7 +989,7 @@ gk104_pll_calc_hiclk(int target_khz, int crystal,
                 int *N1, int *fN1, int *M1, int *P1,
                 int *N2, int *M2, int *P2)
 {
-        int best_clk = 0, best_err = target_khz, p_ref, n_ref;
+        int best_err = target_khz, p_ref, n_ref;
         bool upper = false;
 
         *M1 = 1;
@@ -1010,7 +1010,6 @@ gk104_pll_calc_hiclk(int target_khz, int crystal,
                         /* we found a better combination */
                         if (cur_err < best_err) {
                                 best_err = cur_err;
-                                best_clk = cur_clk;
                                 *N2 = cur_N;
                                 *N1 = n_ref;
                                 *P1 = p_ref;
@@ -1022,7 +1021,6 @@ gk104_pll_calc_hiclk(int target_khz, int crystal,
                                 - target_khz;
                         if (cur_err < best_err) {
                                 best_err = cur_err;
-                                best_clk = cur_clk;
                                 *N2 = cur_N;
                                 *N1 = n_ref;
                                 *P1 = p_ref;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/rammcp77.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/rammcp77.c
index 0a0e44b75577..017a91de74a0 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/rammcp77.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/rammcp77.c
@@ -39,7 +39,7 @@ mcp77_ram_init(struct nvkm_ram *base)
         u32 flush  = ((ram->base.size - (ram->poller_base + 0x40)) >> 5) - 1;
 
         /* Enable NISO poller for various clients and set their associated
-         * read address, only for MCP77/78 and MCP79/7A. (fd#25701)
+         * read address, only for MCP77/78 and MCP79/7A. (fd#27501)
          */
         nvkm_wr32(device, 0x100c18, dniso);
         nvkm_mask(device, 0x100c14, 0x00000000, 0x00000001);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c
index 87bde8ff2d6b..6549b0588309 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c
@@ -496,15 +496,12 @@ nv50_ram_tidy(struct nvkm_ram *base)
 void
 __nv50_ram_put(struct nvkm_ram *ram, struct nvkm_mem *mem)
 {
-        struct nvkm_mm_node *this;
-
-        while (!list_empty(&mem->regions)) {
-                this = list_first_entry(&mem->regions, typeof(*this), rl_entry);
-
-                list_del(&this->rl_entry);
-                nvkm_mm_free(&ram->vram, &this);
+        struct nvkm_mm_node *next = mem->mem;
+        struct nvkm_mm_node *node;
+        while ((node = next)) {
+                next = node->next;
+                nvkm_mm_free(&ram->vram, &node);
         }
-
         nvkm_mm_free(&ram->tags, &mem->tag);
 }
 
@@ -530,7 +527,7 @@ nv50_ram_get(struct nvkm_ram *ram, u64 size, u32 align, u32 ncmin,
 {
         struct nvkm_mm *heap = &ram->vram;
         struct nvkm_mm *tags = &ram->tags;
-        struct nvkm_mm_node *r;
+        struct nvkm_mm_node **node, *r;
         struct nvkm_mem *mem;
         int comp = (memtype & 0x300) >> 8;
         int type = (memtype & 0x07f);
@@ -559,11 +556,11 @@ nv50_ram_get(struct nvkm_ram *ram, u64 size, u32 align, u32 ncmin,
                         comp = 0;
         }
 
-        INIT_LIST_HEAD(&mem->regions);
         mem->memtype = (comp << 7) | type;
         mem->size = max;
 
         type = nv50_fb_memtype[type];
+        node = &mem->mem;
         do {
                 if (back)
                         ret = nvkm_mm_tail(heap, 0, type, max, min, align, &r);
@@ -575,13 +572,13 @@ nv50_ram_get(struct nvkm_ram *ram, u64 size, u32 align, u32 ncmin,
                         return ret;
                 }
 
-                list_add_tail(&r->rl_entry, &mem->regions);
+                *node = r;
+                node = &r->next;
                 max -= r->length;
         } while (max);
         mutex_unlock(&ram->fb->subdev.mutex);
 
-        r = list_first_entry(&mem->regions, struct nvkm_mm_node, rl_entry);
-        mem->offset = (u64)r->offset << NVKM_RAM_MM_SHIFT;
+        mem->offset = (u64)mem->mem->offset << NVKM_RAM_MM_SHIFT;
         *pmem = mem;
         return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c
index f0af2a381eea..fecfa6afcf54 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c
@@ -26,6 +26,7 @@
 #include <subdev/bios.h>
 #include <subdev/bios/extdev.h>
 #include <subdev/bios/iccsense.h>
+#include <subdev/bios/power_budget.h>
 #include <subdev/i2c.h>
 
 static bool
@@ -216,10 +217,25 @@ nvkm_iccsense_oneinit(struct nvkm_subdev *subdev)
 {
         struct nvkm_iccsense *iccsense = nvkm_iccsense(subdev);
         struct nvkm_bios *bios = subdev->device->bios;
+        struct nvbios_power_budget budget;
         struct nvbios_iccsense stbl;
-        int i;
+        int i, ret;
 
-        if (!bios || nvbios_iccsense_parse(bios, &stbl) || !stbl.nr_entry)
+        if (!bios)
+                return 0;
+
+        ret = nvbios_power_budget_header(bios, &budget);
+        if (!ret && budget.cap_entry != 0xff) {
+                struct nvbios_power_budget_entry entry;
+                ret = nvbios_power_budget_entry(bios, &budget,
+                                                budget.cap_entry, &entry);
+                if (!ret) {
+                        iccsense->power_w_max  = entry.avg_w;
+                        iccsense->power_w_crit = entry.max_w;
+                }
+        }
+
+        if (nvbios_iccsense_parse(bios, &stbl) || !stbl.nr_entry)
                 return 0;
 
         iccsense->data_valid = true;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c
index a6a7fa0d7679..9dec58ec3d9f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c
@@ -116,7 +116,7 @@ struct gk20a_instmem {
 static enum nvkm_memory_target
 gk20a_instobj_target(struct nvkm_memory *memory)
 {
-        return NVKM_MEM_TARGET_HOST;
+        return NVKM_MEM_TARGET_NCOH;
 }
 
 static u64
@@ -305,11 +305,11 @@ gk20a_instobj_dtor_iommu(struct nvkm_memory *memory)
         struct gk20a_instobj_iommu *node = gk20a_instobj_iommu(memory);
         struct gk20a_instmem *imem = node->base.imem;
         struct device *dev = imem->base.subdev.device->dev;
-        struct nvkm_mm_node *r;
+        struct nvkm_mm_node *r = node->base.mem.mem;
         unsigned long flags;
         int i;
 
-        if (unlikely(list_empty(&node->base.mem.regions)))
+        if (unlikely(!r))
                 goto out;
 
         spin_lock_irqsave(&imem->lock, flags);
@@ -320,9 +320,6 @@ gk20a_instobj_dtor_iommu(struct nvkm_memory *memory)
 
         spin_unlock_irqrestore(&imem->lock, flags);
 
-        r = list_first_entry(&node->base.mem.regions, struct nvkm_mm_node,
-                             rl_entry);
-
         /* clear IOMMU bit to unmap pages */
         r->offset &= ~BIT(imem->iommu_bit - imem->iommu_pgshift);
 
@@ -404,10 +401,7 @@ gk20a_instobj_ctor_dma(struct gk20a_instmem *imem, u32 npages, u32 align,
         node->r.length = (npages << PAGE_SHIFT) >> 12;
 
         node->base.mem.offset = node->handle;
-
-        INIT_LIST_HEAD(&node->base.mem.regions);
-        list_add_tail(&node->r.rl_entry, &node->base.mem.regions);
-
+        node->base.mem.mem = &node->r;
         return 0;
 }
 
@@ -484,10 +478,7 @@ gk20a_instobj_ctor_iommu(struct gk20a_instmem *imem, u32 npages, u32 align,
         r->offset |= BIT(imem->iommu_bit - imem->iommu_pgshift);
 
         node->base.mem.offset = ((u64)r->offset) << imem->iommu_pgshift;
-
-        INIT_LIST_HEAD(&node->base.mem.regions);
-        list_add_tail(&r->rl_entry, &node->base.mem.regions);
-
+        node->base.mem.mem = r;
         return 0;
 
 release_area:
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c
index 6b25e25f9eba..09f669ac6630 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c
@@ -161,6 +161,16 @@ nvkm_mc_enable(struct nvkm_device *device, enum nvkm_devidx devidx)
         }
 }
 
+bool
+nvkm_mc_enabled(struct nvkm_device *device, enum nvkm_devidx devidx)
+{
+        u64 pmc_enable = nvkm_mc_reset_mask(device, false, devidx);
+
+        return (pmc_enable != 0) &&
+               ((nvkm_rd32(device, 0x000200) & pmc_enable) == pmc_enable);
+}
+
+
 static int
 nvkm_mc_fini(struct nvkm_subdev *subdev, bool suspend)
 {
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
index 5df9669ea39c..d06ad2c372bf 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
@@ -31,7 +31,7 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
 {
         struct nvkm_vm *vm = vma->vm;
         struct nvkm_mmu *mmu = vm->mmu;
-        struct nvkm_mm_node *r;
+        struct nvkm_mm_node *r = node->mem;
         int big = vma->node->type != mmu->func->spg_shift;
         u32 offset = vma->node->offset + (delta >> 12);
         u32 bits = vma->node->type - 12;
@@ -41,7 +41,7 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
         u32 end, len;
 
         delta = 0;
-        list_for_each_entry(r, &node->regions, rl_entry) {
+        while (r) {
                 u64 phys = (u64)r->offset << 12;
                 u32 num  = r->length >> bits;
 
@@ -65,7 +65,8 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
 
                         delta += (u64)len << vma->node->type;
                 }
-        }
+                r = r->next;
+        };
 
         mmu->func->flush(vm);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild
index 2a31b7d66a6d..87bf41cef0c6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild
@@ -6,6 +6,7 @@ nvkm-y += nvkm/subdev/pci/nv40.o
 nvkm-y += nvkm/subdev/pci/nv46.o
 nvkm-y += nvkm/subdev/pci/nv4c.o
 nvkm-y += nvkm/subdev/pci/g84.o
+nvkm-y += nvkm/subdev/pci/g92.o
 nvkm-y += nvkm/subdev/pci/g94.o
 nvkm-y += nvkm/subdev/pci/gf100.o
 nvkm-y += nvkm/subdev/pci/gf106.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/g92.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/g92.c
new file mode 100644
index 000000000000..48874359d5f6
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/g92.c
@@ -0,0 +1,57 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "priv.h"
+
+int
+g92_pcie_version_supported(struct nvkm_pci *pci)
+{
+        if ((nvkm_pci_rd32(pci, 0x460) & 0x200) == 0x200)
+                return 2;
+        return 1;
+}
+
+static const struct nvkm_pci_func
+g92_pci_func = {
+        .init = g84_pci_init,
+        .rd32 = nv40_pci_rd32,
+        .wr08 = nv40_pci_wr08,
+        .wr32 = nv40_pci_wr32,
+        .msi_rearm = nv46_pci_msi_rearm,
+
+        .pcie.init = g84_pcie_init,
+        .pcie.set_link = g84_pcie_set_link,
+
+        .pcie.max_speed = g84_pcie_max_speed,
+        .pcie.cur_speed = g84_pcie_cur_speed,
+
+        .pcie.set_version = g84_pcie_set_version,
+        .pcie.version = g84_pcie_version,
+        .pcie.version_supported = g92_pcie_version_supported,
+};
+
+int
+g92_pci_new(struct nvkm_device *device, int index, struct nvkm_pci **ppci)
+{
+        return nvkm_pci_new_(&g92_pci_func, device, index, ppci);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/g94.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/g94.c
index 43444123bc04..09adb37a5664 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/g94.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/g94.c
@@ -23,14 +23,6 @@
  */
 #include "priv.h"
 
-int
-g94_pcie_version_supported(struct nvkm_pci *pci)
-{
-        if ((nvkm_pci_rd32(pci, 0x460) & 0x200) == 0x200)
-                return 2;
-        return 1;
-}
-
 static const struct nvkm_pci_func
 g94_pci_func = {
         .init = g84_pci_init,
@@ -47,7 +39,7 @@ g94_pci_func = {
 
         .pcie.set_version = g84_pcie_set_version,
         .pcie.version = g84_pcie_version,
-        .pcie.version_supported = g94_pcie_version_supported,
+        .pcie.version_supported = g92_pcie_version_supported,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf100.c
index e30ea676baf6..00a5e7d3ee9d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf100.c
@@ -92,7 +92,7 @@ gf100_pci_func = {
 
         .pcie.set_version = gf100_pcie_set_version,
         .pcie.version = gf100_pcie_version,
-        .pcie.version_supported = g94_pcie_version_supported,
+        .pcie.version_supported = g92_pcie_version_supported,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf106.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf106.c
index c3b798c5c6dd..11bf419afe3f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf106.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gf106.c
@@ -39,7 +39,7 @@ gf106_pci_func = {
 
         .pcie.set_version = gf100_pcie_set_version,
         .pcie.version = gf100_pcie_version,
-        .pcie.version_supported = g94_pcie_version_supported,
+        .pcie.version_supported = g92_pcie_version_supported,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/priv.h
index 23de3180aae5..86921ec962d6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/priv.h
@@ -44,7 +44,7 @@ enum nvkm_pcie_speed g84_pcie_max_speed(struct nvkm_pci *);
 int g84_pcie_init(struct nvkm_pci *);
 int g84_pcie_set_link(struct nvkm_pci *, enum nvkm_pcie_speed, u8);
 
-int g94_pcie_version_supported(struct nvkm_pci *);
+int g92_pcie_version_supported(struct nvkm_pci *);
 
 void gf100_pcie_set_version(struct nvkm_pci *, u8);
 int gf100_pcie_version(struct nvkm_pci *);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild
index 51fb4bf94a44..ca57c1e491b0 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild
@@ -8,5 +8,6 @@ nvkm-y += nvkm/subdev/pmu/gk110.o
 nvkm-y += nvkm/subdev/pmu/gk208.o
 nvkm-y += nvkm/subdev/pmu/gk20a.o
 nvkm-y += nvkm/subdev/pmu/gm107.o
+nvkm-y += nvkm/subdev/pmu/gm20b.o
 nvkm-y += nvkm/subdev/pmu/gp100.o
 nvkm-y += nvkm/subdev/pmu/gp102.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
index e611ce80f8ef..a73f690eb4b5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
@@ -116,6 +116,8 @@ nvkm_pmu_init(struct nvkm_subdev *subdev)
 static void *
 nvkm_pmu_dtor(struct nvkm_subdev *subdev)
 {
+        struct nvkm_pmu *pmu = nvkm_pmu(subdev);
+        nvkm_falcon_del(&pmu->falcon);
         return nvkm_pmu(subdev);
 }
 
@@ -129,15 +131,22 @@ nvkm_pmu = {
 };
 
 int
+nvkm_pmu_ctor(const struct nvkm_pmu_func *func, struct nvkm_device *device,
+              int index, struct nvkm_pmu *pmu)
+{
+        nvkm_subdev_ctor(&nvkm_pmu, device, index, &pmu->subdev);
+        pmu->func = func;
+        INIT_WORK(&pmu->recv.work, nvkm_pmu_recv);
+        init_waitqueue_head(&pmu->recv.wait);
+        return nvkm_falcon_v1_new(&pmu->subdev, "PMU", 0x10a000, &pmu->falcon);
+}
+
+int
 nvkm_pmu_new_(const struct nvkm_pmu_func *func, struct nvkm_device *device,
               int index, struct nvkm_pmu **ppmu)
 {
         struct nvkm_pmu *pmu;
         if (!(pmu = *ppmu = kzalloc(sizeof(*pmu), GFP_KERNEL)))
                 return -ENOMEM;
-        nvkm_subdev_ctor(&nvkm_pmu, device, index, &pmu->subdev);
-        pmu->func = func;
-        INIT_WORK(&pmu->recv.work, nvkm_pmu_recv);
-        init_waitqueue_head(&pmu->recv.wait);
-        return 0;
+        return nvkm_pmu_ctor(func, device, index, *ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c
index f996d90c9f0d..9ca0db796cbe 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c
@@ -19,7 +19,7 @@
  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  * DEALINGS IN THE SOFTWARE.
  */
-#define gk20a_pmu(p) container_of((p), struct gk20a_pmu, base.subdev)
+#define gk20a_pmu(p) container_of((p), struct gk20a_pmu, base)
 #include "priv.h"
 
 #include <subdev/clk.h>
@@ -43,9 +43,8 @@ struct gk20a_pmu {
 };
 
 struct gk20a_pmu_dvfs_dev_status {
-        unsigned long total;
-        unsigned long busy;
-        int cur_state;
+        u32 total;
+        u32 busy;
 };
 
 static int
@@ -56,13 +55,12 @@ gk20a_pmu_dvfs_target(struct gk20a_pmu *pmu, int *state)
         return nvkm_clk_astate(clk, *state, 0, false);
 }
 
-static int
+static void
 gk20a_pmu_dvfs_get_cur_state(struct gk20a_pmu *pmu, int *state)
 {
         struct nvkm_clk *clk = pmu->base.subdev.device->clk;
 
         *state = clk->pstate;
-        return 0;
 }
 
 static int
@@ -90,28 +88,26 @@ gk20a_pmu_dvfs_get_target_state(struct gk20a_pmu *pmu,
 
         *state = level;
 
-        if (level == cur_level)
-                return 0;
-        else
-                return 1;
+        return (level != cur_level);
 }
 
-static int
+static void
 gk20a_pmu_dvfs_get_dev_status(struct gk20a_pmu *pmu,
                               struct gk20a_pmu_dvfs_dev_status *status)
 {
-        struct nvkm_device *device = pmu->base.subdev.device;
-        status->busy = nvkm_rd32(device, 0x10a508 + (BUSY_SLOT * 0x10));
-        status->total= nvkm_rd32(device, 0x10a508 + (CLK_SLOT * 0x10));
-        return 0;
+        struct nvkm_falcon *falcon = pmu->base.falcon;
+
+        status->busy = nvkm_falcon_rd32(falcon, 0x508 + (BUSY_SLOT * 0x10));
+        status->total= nvkm_falcon_rd32(falcon, 0x508 + (CLK_SLOT * 0x10));
 }
 
 static void
 gk20a_pmu_dvfs_reset_dev_status(struct gk20a_pmu *pmu)
 {
-        struct nvkm_device *device = pmu->base.subdev.device;
-        nvkm_wr32(device, 0x10a508 + (BUSY_SLOT * 0x10), 0x80000000);
-        nvkm_wr32(device, 0x10a508 + (CLK_SLOT * 0x10), 0x80000000);
+        struct nvkm_falcon *falcon = pmu->base.falcon;
+
+        nvkm_falcon_wr32(falcon, 0x508 + (BUSY_SLOT * 0x10), 0x80000000);
+        nvkm_falcon_wr32(falcon, 0x508 + (CLK_SLOT * 0x10), 0x80000000);
 }
 
 static void
@@ -127,7 +123,7 @@ gk20a_pmu_dvfs_work(struct nvkm_alarm *alarm)
         struct nvkm_timer *tmr = device->timer;
         struct nvkm_volt *volt = device->volt;
         u32 utilization = 0;
-        int state, ret;
+        int state;
 
         /*
          * The PMU is initialized before CLK and VOLT, so we have to make sure the
@@ -136,11 +132,7 @@ gk20a_pmu_dvfs_work(struct nvkm_alarm *alarm)
         if (!clk || !volt)
                 goto resched;
 
-        ret = gk20a_pmu_dvfs_get_dev_status(pmu, &status);
-        if (ret) {
-                nvkm_warn(subdev, "failed to get device status\n");
-                goto resched;
-        }
+        gk20a_pmu_dvfs_get_dev_status(pmu, &status);
 
         if (status.total)
                 utilization = div_u64((u64)status.busy * 100, status.total);
@@ -150,11 +142,7 @@ gk20a_pmu_dvfs_work(struct nvkm_alarm *alarm)
         nvkm_trace(subdev, "utilization = %d %%, avg_load = %d %%\n",
                    utilization, data->avg_load);
 
-        ret = gk20a_pmu_dvfs_get_cur_state(pmu, &state);
-        if (ret) {
-                nvkm_warn(subdev, "failed to get current state\n");
-                goto resched;
-        }
+        gk20a_pmu_dvfs_get_cur_state(pmu, &state);
 
         if (gk20a_pmu_dvfs_get_target_state(pmu, &state, data->avg_load)) {
                 nvkm_trace(subdev, "set new state to %d\n", state);
@@ -166,32 +154,36 @@ resched:
         nvkm_timer_alarm(tmr, 100000000, alarm);
 }
 
-static int
-gk20a_pmu_fini(struct nvkm_subdev *subdev, bool suspend)
+static void
+gk20a_pmu_fini(struct nvkm_pmu *pmu)
 {
-        struct gk20a_pmu *pmu = gk20a_pmu(subdev);
-        nvkm_timer_alarm_cancel(subdev->device->timer, &pmu->alarm);
-        return 0;
-}
+        struct gk20a_pmu *gpmu = gk20a_pmu(pmu);
+        nvkm_timer_alarm_cancel(pmu->subdev.device->timer, &gpmu->alarm);
 
-static void *
-gk20a_pmu_dtor(struct nvkm_subdev *subdev)
-{
-        return gk20a_pmu(subdev);
+        nvkm_falcon_put(pmu->falcon, &pmu->subdev);
 }
 
 static int
-gk20a_pmu_init(struct nvkm_subdev *subdev)
+gk20a_pmu_init(struct nvkm_pmu *pmu)
 {
-        struct gk20a_pmu *pmu = gk20a_pmu(subdev);
-        struct nvkm_device *device = pmu->base.subdev.device;
+        struct gk20a_pmu *gpmu = gk20a_pmu(pmu);
+        struct nvkm_subdev *subdev = &pmu->subdev;
+        struct nvkm_device *device = pmu->subdev.device;
+        struct nvkm_falcon *falcon = pmu->falcon;
+        int ret;
+
+        ret = nvkm_falcon_get(falcon, subdev);
+        if (ret) {
+                nvkm_error(subdev, "cannot acquire %s falcon!\n", falcon->name);
+                return ret;
+        }
 
         /* init pwr perf counter */
-        nvkm_wr32(device, 0x10a504 + (BUSY_SLOT * 0x10), 0x00200001);
-        nvkm_wr32(device, 0x10a50c + (BUSY_SLOT * 0x10), 0x00000002);
-        nvkm_wr32(device, 0x10a50c + (CLK_SLOT * 0x10), 0x00000003);
+        nvkm_falcon_wr32(falcon, 0x504 + (BUSY_SLOT * 0x10), 0x00200001);
+        nvkm_falcon_wr32(falcon, 0x50c + (BUSY_SLOT * 0x10), 0x00000002);
+        nvkm_falcon_wr32(falcon, 0x50c + (CLK_SLOT * 0x10), 0x00000003);
 
-        nvkm_timer_alarm(device->timer, 2000000000, &pmu->alarm);
+        nvkm_timer_alarm(device->timer, 2000000000, &gpmu->alarm);
         return 0;
 }
 
@@ -202,26 +194,26 @@ gk20a_dvfs_data= {
         .p_smooth = 1,
 };
 
-static const struct nvkm_subdev_func
+static const struct nvkm_pmu_func
 gk20a_pmu = {
         .init = gk20a_pmu_init,
         .fini = gk20a_pmu_fini,
-        .dtor = gk20a_pmu_dtor,
+        .reset = gt215_pmu_reset,
 };
 
 int
 gk20a_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-        static const struct nvkm_pmu_func func = {};
         struct gk20a_pmu *pmu;
 
         if (!(pmu = kzalloc(sizeof(*pmu), GFP_KERNEL)))
                 return -ENOMEM;
-        pmu->base.func = &func;
         *ppmu = &pmu->base;
 
-        nvkm_subdev_ctor(&gk20a_pmu, device, index, &pmu->base.subdev);
+        nvkm_pmu_ctor(&gk20a_pmu, device, index, &pmu->base);
+
         pmu->data = &gk20a_dvfs_data;
         nvkm_alarm_init(&pmu->alarm, gk20a_pmu_dvfs_work);
+
         return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
new file mode 100644
index 000000000000..0b8a1cc4a0ee
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include "priv.h"
+
+static const struct nvkm_pmu_func
+gm20b_pmu = {
+        .reset = gt215_pmu_reset,
+};
+
+int
+gm20b_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
+{
+        return nvkm_pmu_new_(&gm20b_pmu, device, index, ppmu);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
index 2e2179a4ad17..096cba069f72 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
@@ -4,6 +4,8 @@
 #include <subdev/pmu.h>
 #include <subdev/pmu/fuc/os.h>
 
+int nvkm_pmu_ctor(const struct nvkm_pmu_func *, struct nvkm_device *,
+                  int index, struct nvkm_pmu *);
 int nvkm_pmu_new_(const struct nvkm_pmu_func *, struct nvkm_device *,
                   int index, struct nvkm_pmu **);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/Kbuild
index b02b868a6589..5076d1500f47 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/Kbuild
@@ -1,3 +1,7 @@
 nvkm-y += nvkm/subdev/secboot/base.o
+nvkm-y += nvkm/subdev/secboot/ls_ucode_gr.o
+nvkm-y += nvkm/subdev/secboot/acr.o
+nvkm-y += nvkm/subdev/secboot/acr_r352.o
+nvkm-y += nvkm/subdev/secboot/acr_r361.o
 nvkm-y += nvkm/subdev/secboot/gm200.o
 nvkm-y += nvkm/subdev/secboot/gm20b.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr.c
new file mode 100644
index 000000000000..75dc06557877
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr.c
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include "acr.h"
+
+#include <core/firmware.h>
+
+/**
+ * Convenience function to duplicate a firmware file in memory and check that
+ * it has the required minimum size.
+ */
+void *
+nvkm_acr_load_firmware(const struct nvkm_subdev *subdev, const char *name,
+                       size_t min_size)
+{
+        const struct firmware *fw;
+        void *blob;
+        int ret;
+
+        ret = nvkm_firmware_get(subdev->device, name, &fw);
+        if (ret)
+                return ERR_PTR(ret);
+        if (fw->size < min_size) {
+                nvkm_error(subdev, "%s is smaller than expected size %zu\n",
+                           name, min_size);
+                nvkm_firmware_put(fw);
+                return ERR_PTR(-EINVAL);
+        }
+        blob = kmemdup(fw->data, fw->size, GFP_KERNEL);
+        nvkm_firmware_put(fw);
+        if (!blob)
+                return ERR_PTR(-ENOMEM);
+
+        return blob;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr.h b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr.h
new file mode 100644
index 000000000000..97795b342b6f
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr.h
@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+#ifndef __NVKM_SECBOOT_ACR_H__
+#define __NVKM_SECBOOT_ACR_H__
+
+#include "priv.h"
+
+struct nvkm_acr;
+
+/**
+ * struct nvkm_acr_func - properties and functions specific to an ACR
+ *
+ * @load: make the ACR ready to run on the given secboot device
+ * @reset: reset the specified falcon
+ * @start: start the specified falcon (assumed to have been reset)
+ */
+struct nvkm_acr_func {
+        void (*dtor)(struct nvkm_acr *);
+        int (*oneinit)(struct nvkm_acr *, struct nvkm_secboot *);
+        int (*fini)(struct nvkm_acr *, struct nvkm_secboot *, bool);
+        int (*load)(struct nvkm_acr *, struct nvkm_secboot *,
+                    struct nvkm_gpuobj *, u64);
+        int (*reset)(struct nvkm_acr *, struct nvkm_secboot *,
+                     enum nvkm_secboot_falcon);
+        int (*start)(struct nvkm_acr *, struct nvkm_secboot *,
+                     enum nvkm_secboot_falcon);
+};
+
+/**
+ * struct nvkm_acr - instance of an ACR
+ *
+ * @boot_falcon: ID of the falcon that will perform secure boot
+ * @managed_falcons: bitfield of falcons managed by this ACR
+ * @start_address: virtual start address of the HS bootloader
+ */
+struct nvkm_acr {
+        const struct nvkm_acr_func *func;
+        const struct nvkm_subdev *subdev;
+
+        enum nvkm_secboot_falcon boot_falcon;
+        unsigned long managed_falcons;
+        u32 start_address;
+};
+
+void *nvkm_acr_load_firmware(const struct nvkm_subdev *, const char *, size_t);
+
+struct nvkm_acr *acr_r352_new(unsigned long);
+struct nvkm_acr *acr_r361_new(unsigned long);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r352.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r352.c
new file mode 100644
index 000000000000..1aa37ea18580
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r352.c
@@ -0,0 +1,936 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include "acr_r352.h"
+
+#include <core/gpuobj.h>
+#include <core/firmware.h>
+#include <engine/falcon.h>
+
+/**
+ * struct hsf_fw_header - HS firmware descriptor
+ * @sig_dbg_offset:     offset of the debug signature
+ * @sig_dbg_size:       size of the debug signature
+ * @sig_prod_offset:    offset of the production signature
+ * @sig_prod_size:      size of the production signature
+ * @patch_loc:          offset of the offset (sic) of where the signature is
+ * @patch_sig:          offset of the offset (sic) to add to sig_*_offset
+ * @hdr_offset:         offset of the load header (see struct hs_load_header)
+ * @hdr_size:           size of above header
+ *
+ * This structure is embedded in the HS firmware image at
+ * hs_bin_hdr.header_offset.
+ */
+struct hsf_fw_header {
+        u32 sig_dbg_offset;
+        u32 sig_dbg_size;
+        u32 sig_prod_offset;
+        u32 sig_prod_size;
+        u32 patch_loc;
+        u32 patch_sig;
+        u32 hdr_offset;
+        u32 hdr_size;
+};
+
+/**
+ * struct acr_r352_flcn_bl_desc - DMEM bootloader descriptor
+ * @signature:          16B signature for secure code. 0s if no secure code
+ * @ctx_dma:            DMA context to be used by BL while loading code/data
+ * @code_dma_base:      256B-aligned Physical FB Address where code is located
+ *                      (falcon's $xcbase register)
+ * @non_sec_code_off:   offset from code_dma_base where the non-secure code is
+ *                      located. The offset must be multiple of 256 to help perf
+ * @non_sec_code_size:  the size of the nonSecure code part.
+ * @sec_code_off:       offset from code_dma_base where the secure code is
+ *                      located. The offset must be multiple of 256 to help perf
+ * @sec_code_size:      offset from code_dma_base where the secure code is
+ *                      located. The offset must be multiple of 256 to help perf
+ * @code_entry_point:   code entry point which will be invoked by BL after
+ *                      code is loaded.
+ * @data_dma_base:      256B aligned Physical FB Address where data is located.
+ *                      (falcon's $xdbase register)
+ * @data_size:          size of data block. Should be multiple of 256B
+ *
+ * Structure used by the bootloader to load the rest of the code. This has
+ * to be filled by host and copied into DMEM at offset provided in the
+ * hsflcn_bl_desc.bl_desc_dmem_load_off.
+ */
+struct acr_r352_flcn_bl_desc {
+        u32 reserved[4];
+        u32 signature[4];
+        u32 ctx_dma;
+        u32 code_dma_base;
+        u32 non_sec_code_off;
+        u32 non_sec_code_size;
+        u32 sec_code_off;
+        u32 sec_code_size;
+        u32 code_entry_point;
+        u32 data_dma_base;
+        u32 data_size;
+        u32 code_dma_base1;
+        u32 data_dma_base1;
+};
+
+/**
+ * acr_r352_generate_flcn_bl_desc - generate generic BL descriptor for LS image
+ */
+static void
+acr_r352_generate_flcn_bl_desc(const struct nvkm_acr *acr,
+                               const struct ls_ucode_img *_img, u64 wpr_addr,
+                               void *_desc)
+{
+        struct ls_ucode_img_r352 *img = ls_ucode_img_r352(_img);
+        struct acr_r352_flcn_bl_desc *desc = _desc;
+        const struct ls_ucode_img_desc *pdesc = &_img->ucode_desc;
+        u64 base, addr_code, addr_data;
+
+        base = wpr_addr + img->lsb_header.ucode_off + pdesc->app_start_offset;
+        addr_code = (base + pdesc->app_resident_code_offset) >> 8;
+        addr_data = (base + pdesc->app_resident_data_offset) >> 8;
+
+        desc->ctx_dma = FALCON_DMAIDX_UCODE;
+        desc->code_dma_base = lower_32_bits(addr_code);
+        desc->code_dma_base1 = upper_32_bits(addr_code);
+        desc->non_sec_code_off = pdesc->app_resident_code_offset;
+        desc->non_sec_code_size = pdesc->app_resident_code_size;
+        desc->code_entry_point = pdesc->app_imem_entry;
+        desc->data_dma_base = lower_32_bits(addr_data);
+        desc->data_dma_base1 = upper_32_bits(addr_data);
+        desc->data_size = pdesc->app_resident_data_size;
+}
+
+
+/**
+ * struct hsflcn_acr_desc - data section of the HS firmware
+ *
+ * This header is to be copied at the beginning of DMEM by the HS bootloader.
+ *
+ * @signature:          signature of ACR ucode
+ * @wpr_region_id:      region ID holding the WPR header and its details
+ * @wpr_offset:         offset from the WPR region holding the wpr header
+ * @regions:            region descriptors
+ * @nonwpr_ucode_blob_size:     size of LS blob
+ * @nonwpr_ucode_blob_start:    FB location of LS blob is
+ */
+struct hsflcn_acr_desc {
+        union {
+                u8 reserved_dmem[0x200];
+                u32 signatures[4];
+        } ucode_reserved_space;
+        u32 wpr_region_id;
+        u32 wpr_offset;
+        u32 mmu_mem_range;
+#define FLCN_ACR_MAX_REGIONS 2
+        struct {
+                u32 no_regions;
+                struct {
+                        u32 start_addr;
+                        u32 end_addr;
+                        u32 region_id;
+                        u32 read_mask;
+                        u32 write_mask;
+                        u32 client_mask;
+                } region_props[FLCN_ACR_MAX_REGIONS];
+        } regions;
+        u32 ucode_blob_size;
+        u64 ucode_blob_base __aligned(8);
+        struct {
+                u32 vpr_enabled;
+                u32 vpr_start;
+                u32 vpr_end;
+                u32 hdcp_policies;
+        } vpr_desc;
+};
+
+
+/*
+ * Low-secure blob creation
+ */
+
+/**
+ * ls_ucode_img_load() - create a lsf_ucode_img and load it
+ */
+struct ls_ucode_img *
+acr_r352_ls_ucode_img_load(const struct acr_r352 *acr,
+                           enum nvkm_secboot_falcon falcon_id)
+{
+        const struct nvkm_subdev *subdev = acr->base.subdev;
+        struct ls_ucode_img_r352 *img;
+        int ret;
+
+        img = kzalloc(sizeof(*img), GFP_KERNEL);
+        if (!img)
+                return ERR_PTR(-ENOMEM);
+
+        img->base.falcon_id = falcon_id;
+
+        ret = acr->func->ls_func[falcon_id]->load(subdev, &img->base);
+
+        if (ret) {
+                kfree(img->base.ucode_data);
+                kfree(img->base.sig);
+                kfree(img);
+                return ERR_PTR(ret);
+        }
+
+        /* Check that the signature size matches our expectations... */
+        if (img->base.sig_size != sizeof(img->lsb_header.signature)) {
+                nvkm_error(subdev, "invalid signature size for %s falcon!\n",
+                           nvkm_secboot_falcon_name[falcon_id]);
+                return ERR_PTR(-EINVAL);
+        }
+
+        /* Copy signature to the right place */
+        memcpy(&img->lsb_header.signature, img->base.sig, img->base.sig_size);
+
+        /* not needed? the signature should already have the right value */
+        img->lsb_header.signature.falcon_id = falcon_id;
+
+        return &img->base;
+}
+
+#define LSF_LSB_HEADER_ALIGN 256
+#define LSF_BL_DATA_ALIGN 256
+#define LSF_BL_DATA_SIZE_ALIGN 256
+#define LSF_BL_CODE_SIZE_ALIGN 256
+#define LSF_UCODE_DATA_ALIGN 4096
+
+/**
+ * acr_r352_ls_img_fill_headers - fill the WPR and LSB headers of an image
+ * @acr:        ACR to use
+ * @img:        image to generate for
+ * @offset:     offset in the WPR region where this image starts
+ *
+ * Allocate space in the WPR area from offset and write the WPR and LSB headers
+ * accordingly.
+ *
+ * Return: offset at the end of this image.
+ */
+static u32
+acr_r352_ls_img_fill_headers(struct acr_r352 *acr,
+                             struct ls_ucode_img_r352 *img, u32 offset)
+{
+        struct ls_ucode_img *_img = &img->base;
+        struct acr_r352_lsf_wpr_header *whdr = &img->wpr_header;
+        struct acr_r352_lsf_lsb_header *lhdr = &img->lsb_header;
+        struct ls_ucode_img_desc *desc = &_img->ucode_desc;
+        const struct acr_r352_ls_func *func =
+                                            acr->func->ls_func[_img->falcon_id];
+
+        /* Fill WPR header */
+        whdr->falcon_id = _img->falcon_id;
+        whdr->bootstrap_owner = acr->base.boot_falcon;
+        whdr->status = LSF_IMAGE_STATUS_COPY;
+
+        /* Skip bootstrapping falcons started by someone else than ACR */
+        if (acr->lazy_bootstrap & BIT(_img->falcon_id))
+                whdr->lazy_bootstrap = 1;
+
+        /* Align, save off, and include an LSB header size */
+        offset = ALIGN(offset, LSF_LSB_HEADER_ALIGN);
+        whdr->lsb_offset = offset;
+        offset += sizeof(*lhdr);
+
+        /*
+         * Align, save off, and include the original (static) ucode
+         * image size
+         */
+        offset = ALIGN(offset, LSF_UCODE_DATA_ALIGN);
+        lhdr->ucode_off = offset;
+        offset += _img->ucode_size;
+
+        /*
+         * For falcons that use a boot loader (BL), we append a loader
+         * desc structure on the end of the ucode image and consider
+         * this the boot loader data. The host will then copy the loader
+         * desc args to this space within the WPR region (before locking
+         * down) and the HS bin will then copy them to DMEM 0 for the
+         * loader.
+         */
+        lhdr->bl_code_size = ALIGN(desc->bootloader_size,
+                                   LSF_BL_CODE_SIZE_ALIGN);
+        lhdr->ucode_size = ALIGN(desc->app_resident_data_offset,
+                                 LSF_BL_CODE_SIZE_ALIGN) + lhdr->bl_code_size;
+        lhdr->data_size = ALIGN(desc->app_size, LSF_BL_CODE_SIZE_ALIGN) +
+                                lhdr->bl_code_size - lhdr->ucode_size;
+        /*
+         * Though the BL is located at 0th offset of the image, the VA
+         * is different to make sure that it doesn't collide the actual
+         * OS VA range
+         */
+        lhdr->bl_imem_off = desc->bootloader_imem_offset;
+        lhdr->app_code_off = desc->app_start_offset +
+                             desc->app_resident_code_offset;
+        lhdr->app_code_size = desc->app_resident_code_size;
+        lhdr->app_data_off = desc->app_start_offset +
+                             desc->app_resident_data_offset;
+        lhdr->app_data_size = desc->app_resident_data_size;
+
+        lhdr->flags = func->lhdr_flags;
+        if (_img->falcon_id == acr->base.boot_falcon)
+                lhdr->flags |= LSF_FLAG_DMACTL_REQ_CTX;
+
+        /* Align and save off BL descriptor size */
+        lhdr->bl_data_size = ALIGN(func->bl_desc_size, LSF_BL_DATA_SIZE_ALIGN);
+
+        /*
+         * Align, save off, and include the additional BL data
+         */
+        offset = ALIGN(offset, LSF_BL_DATA_ALIGN);
+        lhdr->bl_data_off = offset;
+        offset += lhdr->bl_data_size;
+
+        return offset;
+}
+
+/**
+ * acr_r352_ls_fill_headers - fill WPR and LSB headers of all managed images
+ */
+int
+acr_r352_ls_fill_headers(struct acr_r352 *acr, struct list_head *imgs)
+{
+        struct ls_ucode_img_r352 *img;
+        struct list_head *l;
+        u32 count = 0;
+        u32 offset;
+
+        /* Count the number of images to manage */
+        list_for_each(l, imgs)
+                count++;
+
+        /*
+         * Start with an array of WPR headers at the base of the WPR.
+         * The expectation here is that the secure falcon will do a single DMA
+         * read of this array and cache it internally so it's ok to pack these.
+         * Also, we add 1 to the falcon count to indicate the end of the array.
+         */
+        offset = sizeof(img->wpr_header) * (count + 1);
+
+        /*
+         * Walk the managed falcons, accounting for the LSB structs
+         * as well as the ucode images.
+         */
+        list_for_each_entry(img, imgs, base.node) {
+                offset = acr_r352_ls_img_fill_headers(acr, img, offset);
+        }
+
+        return offset;
+}
+
+/**
+ * acr_r352_ls_write_wpr - write the WPR blob contents
+ */
+int
+acr_r352_ls_write_wpr(struct acr_r352 *acr, struct list_head *imgs,
+                      struct nvkm_gpuobj *wpr_blob, u32 wpr_addr)
+{
+        struct ls_ucode_img *_img;
+        u32 pos = 0;
+
+        nvkm_kmap(wpr_blob);
+
+        list_for_each_entry(_img, imgs, node) {
+                struct ls_ucode_img_r352 *img = ls_ucode_img_r352(_img);
+                const struct acr_r352_ls_func *ls_func =
+                                            acr->func->ls_func[_img->falcon_id];
+                u8 gdesc[ls_func->bl_desc_size];
+
+                nvkm_gpuobj_memcpy_to(wpr_blob, pos, &img->wpr_header,
+                                      sizeof(img->wpr_header));
+
+                nvkm_gpuobj_memcpy_to(wpr_blob, img->wpr_header.lsb_offset,
+                                     &img->lsb_header, sizeof(img->lsb_header));
+
+                /* Generate and write BL descriptor */
+                memset(gdesc, 0, ls_func->bl_desc_size);
+                ls_func->generate_bl_desc(&acr->base, _img, wpr_addr, gdesc);
+
+                nvkm_gpuobj_memcpy_to(wpr_blob, img->lsb_header.bl_data_off,
+                                      gdesc, ls_func->bl_desc_size);
+
+                /* Copy ucode */
+                nvkm_gpuobj_memcpy_to(wpr_blob, img->lsb_header.ucode_off,
+                                      _img->ucode_data, _img->ucode_size);
+
+                pos += sizeof(img->wpr_header);
+        }
+
+        nvkm_wo32(wpr_blob, pos, NVKM_SECBOOT_FALCON_INVALID);
+
+        nvkm_done(wpr_blob);
+
+        return 0;
+}
+
+/* Both size and address of WPR need to be 128K-aligned */
+#define WPR_ALIGNMENT   0x20000
+/**
+ * acr_r352_prepare_ls_blob() - prepare the LS blob
+ *
+ * For each securely managed falcon, load the FW, signatures and bootloaders and
+ * prepare a ucode blob. Then, compute the offsets in the WPR region for each
+ * blob, and finally write the headers and ucode blobs into a GPU object that
+ * will be copied into the WPR region by the HS firmware.
+ */
+static int
+acr_r352_prepare_ls_blob(struct acr_r352 *acr, u64 wpr_addr, u32 wpr_size)
+{
+        const struct nvkm_subdev *subdev = acr->base.subdev;
+        struct list_head imgs;
+        struct ls_ucode_img *img, *t;
+        unsigned long managed_falcons = acr->base.managed_falcons;
+        int managed_count = 0;
+        u32 image_wpr_size;
+        int falcon_id;
+        int ret;
+
+        INIT_LIST_HEAD(&imgs);
+
+        /* Load all LS blobs */
+        for_each_set_bit(falcon_id, &managed_falcons, NVKM_SECBOOT_FALCON_END) {
+                struct ls_ucode_img *img;
+
+                img = acr->func->ls_ucode_img_load(acr, falcon_id);
+                if (IS_ERR(img)) {
+                        ret = PTR_ERR(img);
+                        goto cleanup;
+                }
+
+                list_add_tail(&img->node, &imgs);
+                managed_count++;
+        }
+
+        /*
+         * Fill the WPR and LSF headers with the right offsets and compute
+         * required WPR size
+         */
+        image_wpr_size = acr->func->ls_fill_headers(acr, &imgs);
+        image_wpr_size = ALIGN(image_wpr_size, WPR_ALIGNMENT);
+
+        /* Allocate GPU object that will contain the WPR region */
+        ret = nvkm_gpuobj_new(subdev->device, image_wpr_size, WPR_ALIGNMENT,
+                              false, NULL, &acr->ls_blob);
+        if (ret)
+                goto cleanup;
+
+        nvkm_debug(subdev, "%d managed LS falcons, WPR size is %d bytes\n",
+                    managed_count, image_wpr_size);
+
+        /* If WPR address and size are not fixed, set them to fit the LS blob */
+        if (wpr_size == 0) {
+                wpr_addr = acr->ls_blob->addr;
+                wpr_size = image_wpr_size;
+        /*
+         * But if the WPR region is set by the bootloader, it is illegal for
+         * the HS blob to be larger than this region.
+         */
+        } else if (image_wpr_size > wpr_size) {
+                nvkm_error(subdev, "WPR region too small for FW blob!\n");
+                nvkm_error(subdev, "required: %dB\n", image_wpr_size);
+                nvkm_error(subdev, "available: %dB\n", wpr_size);
+                ret = -ENOSPC;
+                goto cleanup;
+        }
+
+        /* Write LS blob */
+        ret = acr->func->ls_write_wpr(acr, &imgs, acr->ls_blob, wpr_addr);
+        if (ret)
+                nvkm_gpuobj_del(&acr->ls_blob);
+
+cleanup:
+        list_for_each_entry_safe(img, t, &imgs, node) {
+                kfree(img->ucode_data);
+                kfree(img->sig);
+                kfree(img);
+        }
+
+        return ret;
+}
+
+
+
+
+/**
+ * acr_r352_hsf_patch_signature() - patch HS blob with correct signature
+ */
+static void
+acr_r352_hsf_patch_signature(struct nvkm_secboot *sb, void *acr_image)
+{
+        struct fw_bin_header *hsbin_hdr = acr_image;
+        struct hsf_fw_header *fw_hdr = acr_image + hsbin_hdr->header_offset;
+        void *hs_data = acr_image + hsbin_hdr->data_offset;
+        void *sig;
+        u32 sig_size;
+
+        /* Falcon in debug or production mode? */
+        if (sb->boot_falcon->debug) {
+                sig = acr_image + fw_hdr->sig_dbg_offset;
+                sig_size = fw_hdr->sig_dbg_size;
+        } else {
+                sig = acr_image + fw_hdr->sig_prod_offset;
+                sig_size = fw_hdr->sig_prod_size;
+        }
+
+        /* Patch signature */
+        memcpy(hs_data + fw_hdr->patch_loc, sig + fw_hdr->patch_sig, sig_size);
+}
+
+static void
+acr_r352_fixup_hs_desc(struct acr_r352 *acr, struct nvkm_secboot *sb,
+                       struct hsflcn_acr_desc *desc)
+{
+        struct nvkm_gpuobj *ls_blob = acr->ls_blob;
+
+        /* WPR region information if WPR is not fixed */
+        if (sb->wpr_size == 0) {
+                u32 wpr_start = ls_blob->addr;
+                u32 wpr_end = wpr_start + ls_blob->size;
+
+                desc->wpr_region_id = 1;
+                desc->regions.no_regions = 2;
+                desc->regions.region_props[0].start_addr = wpr_start >> 8;
+                desc->regions.region_props[0].end_addr = wpr_end >> 8;
+                desc->regions.region_props[0].region_id = 1;
+                desc->regions.region_props[0].read_mask = 0xf;
+                desc->regions.region_props[0].write_mask = 0xc;
+                desc->regions.region_props[0].client_mask = 0x2;
+        } else {
+                desc->ucode_blob_base = ls_blob->addr;
+                desc->ucode_blob_size = ls_blob->size;
+        }
+}
+
+static void
+acr_r352_generate_hs_bl_desc(const struct hsf_load_header *hdr, void *_bl_desc,
+                             u64 offset)
+{
+        struct acr_r352_flcn_bl_desc *bl_desc = _bl_desc;
+        u64 addr_code, addr_data;
+
+        addr_code = offset >> 8;
+        addr_data = (offset + hdr->data_dma_base) >> 8;
+
+        bl_desc->ctx_dma = FALCON_DMAIDX_VIRT;
+        bl_desc->code_dma_base = lower_32_bits(addr_code);
+        bl_desc->non_sec_code_off = hdr->non_sec_code_off;
+        bl_desc->non_sec_code_size = hdr->non_sec_code_size;
+        bl_desc->sec_code_off = hdr->app[0].sec_code_off;
+        bl_desc->sec_code_size = hdr->app[0].sec_code_size;
+        bl_desc->code_entry_point = 0;
+        bl_desc->data_dma_base = lower_32_bits(addr_data);
+        bl_desc->data_size = hdr->data_size;
+}
+
+/**
+ * acr_r352_prepare_hs_blob - load and prepare a HS blob and BL descriptor
+ *
+ * @sb secure boot instance to prepare for
+ * @fw name of the HS firmware to load
+ * @blob pointer to gpuobj that will be allocated to receive the HS FW payload
+ * @bl_desc pointer to the BL descriptor to write for this firmware
+ * @patch whether we should patch the HS descriptor (only for HS loaders)
+ */
+static int
+acr_r352_prepare_hs_blob(struct acr_r352 *acr, struct nvkm_secboot *sb,
+                         const char *fw, struct nvkm_gpuobj **blob,
+                         struct hsf_load_header *load_header, bool patch)
+{
+        struct nvkm_subdev *subdev = &sb->subdev;
+        void *acr_image;
+        struct fw_bin_header *hsbin_hdr;
+        struct hsf_fw_header *fw_hdr;
+        struct hsf_load_header *load_hdr;
+        void *acr_data;
+        int ret;
+
+        acr_image = nvkm_acr_load_firmware(subdev, fw, 0);
+        if (IS_ERR(acr_image))
+                return PTR_ERR(acr_image);
+
+        hsbin_hdr = acr_image;
+        fw_hdr = acr_image + hsbin_hdr->header_offset;
+        load_hdr = acr_image + fw_hdr->hdr_offset;
+        acr_data = acr_image + hsbin_hdr->data_offset;
+
+        /* Patch signature */
+        acr_r352_hsf_patch_signature(sb, acr_image);
+
+        /* Patch descriptor with WPR information? */
+        if (patch) {
+                struct hsflcn_acr_desc *desc;
+
+                desc = acr_data + load_hdr->data_dma_base;
+                acr_r352_fixup_hs_desc(acr, sb, desc);
+        }
+
+        if (load_hdr->num_apps > ACR_R352_MAX_APPS) {
+                nvkm_error(subdev, "more apps (%d) than supported (%d)!",
+                           load_hdr->num_apps, ACR_R352_MAX_APPS);
+                ret = -EINVAL;
+                goto cleanup;
+        }
+        memcpy(load_header, load_hdr, sizeof(*load_header) +
+                               (sizeof(load_hdr->app[0]) * load_hdr->num_apps));
+
+        /* Create ACR blob and copy HS data to it */
+        ret = nvkm_gpuobj_new(subdev->device, ALIGN(hsbin_hdr->data_size, 256),
+                              0x1000, false, NULL, blob);
+        if (ret)
+                goto cleanup;
+
+        nvkm_kmap(*blob);
+        nvkm_gpuobj_memcpy_to(*blob, 0, acr_data, hsbin_hdr->data_size);
+        nvkm_done(*blob);
+
+cleanup:
+        kfree(acr_image);
+
+        return ret;
+}
+
+static int
+acr_r352_prepare_hsbl_blob(struct acr_r352 *acr)
+{
+        const struct nvkm_subdev *subdev = acr->base.subdev;
+        struct fw_bin_header *hdr;
+        struct fw_bl_desc *hsbl_desc;
+
+        acr->hsbl_blob = nvkm_acr_load_firmware(subdev, "acr/bl", 0);
+        if (IS_ERR(acr->hsbl_blob)) {
+                int ret = PTR_ERR(acr->hsbl_blob);
+
+                acr->hsbl_blob = NULL;
+                return ret;
+        }
+
+        hdr = acr->hsbl_blob;
+        hsbl_desc = acr->hsbl_blob + hdr->header_offset;
+
+        /* virtual start address for boot vector */
+        acr->base.start_address = hsbl_desc->start_tag << 8;
+
+        return 0;
+}
+
+/**
+ * acr_r352_load_blobs - load blobs common to all ACR V1 versions.
+ *
+ * This includes the LS blob, HS ucode loading blob, and HS bootloader.
+ *
+ * The HS ucode unload blob is only used on dGPU if the WPR region is variable.
+ */
+int
+acr_r352_load_blobs(struct acr_r352 *acr, struct nvkm_secboot *sb)
+{
+        int ret;
+
+        /* Firmware already loaded? */
+        if (acr->firmware_ok)
+                return 0;
+
+        /* Load and prepare the managed falcon's firmwares */
+        ret = acr_r352_prepare_ls_blob(acr, sb->wpr_addr, sb->wpr_size);
+        if (ret)
+                return ret;
+
+        /* Load the HS firmware that will load the LS firmwares */
+        if (!acr->load_blob) {
+                ret = acr_r352_prepare_hs_blob(acr, sb, "acr/ucode_load",
+                                               &acr->load_blob,
+                                               &acr->load_bl_header, true);
+                if (ret)
+                        return ret;
+        }
+
+        /* If the ACR region is dynamically programmed, we need an unload FW */
+        if (sb->wpr_size == 0) {
+                ret = acr_r352_prepare_hs_blob(acr, sb, "acr/ucode_unload",
+                                               &acr->unload_blob,
+                                               &acr->unload_bl_header, false);
+                if (ret)
+                        return ret;
+        }
+
+        /* Load the HS firmware bootloader */
+        if (!acr->hsbl_blob) {
+                ret = acr_r352_prepare_hsbl_blob(acr);
+                if (ret)
+                        return ret;
+        }
+
+        acr->firmware_ok = true;
+        nvkm_debug(&sb->subdev, "LS blob successfully created\n");
+
+        return 0;
+}
+
+/**
+ * acr_r352_load() - prepare HS falcon to run the specified blob, mapped
+ * at GPU address offset.
+ */
+static int
+acr_r352_load(struct nvkm_acr *_acr, struct nvkm_secboot *sb,
+              struct nvkm_gpuobj *blob, u64 offset)
+{
+        struct acr_r352 *acr = acr_r352(_acr);
+        struct nvkm_falcon *falcon = sb->boot_falcon;
+        struct fw_bin_header *hdr = acr->hsbl_blob;
+        struct fw_bl_desc *hsbl_desc = acr->hsbl_blob + hdr->header_offset;
+        void *blob_data = acr->hsbl_blob + hdr->data_offset;
+        void *hsbl_code = blob_data + hsbl_desc->code_off;
+        void *hsbl_data = blob_data + hsbl_desc->data_off;
+        u32 code_size = ALIGN(hsbl_desc->code_size, 256);
+        const struct hsf_load_header *load_hdr;
+        const u32 bl_desc_size = acr->func->hs_bl_desc_size;
+        u8 bl_desc[bl_desc_size];
+
+        /* Find the bootloader descriptor for our blob and copy it */
+        if (blob == acr->load_blob) {
+                load_hdr = &acr->load_bl_header;
+        } else if (blob == acr->unload_blob) {
+                load_hdr = &acr->unload_bl_header;
+        } else {
+                nvkm_error(_acr->subdev, "invalid secure boot blob!\n");
+                return -EINVAL;
+        }
+
+        /*
+         * Copy HS bootloader data
+         */
+        nvkm_falcon_load_dmem(falcon, hsbl_data, 0x0, hsbl_desc->data_size, 0);
+
+        /* Copy HS bootloader code to end of IMEM */
+        nvkm_falcon_load_imem(falcon, hsbl_code, falcon->code.limit - code_size,
+                              code_size, hsbl_desc->start_tag, 0, false);
+
+        /* Generate the BL header */
+        memset(bl_desc, 0, bl_desc_size);
+        acr->func->generate_hs_bl_desc(load_hdr, bl_desc, offset);
+
+        /*
+         * Copy HS BL header where the HS descriptor expects it to be
+         */
+        nvkm_falcon_load_dmem(falcon, bl_desc, hsbl_desc->dmem_load_off,
+                              bl_desc_size, 0);
+
+        return 0;
+}
+
+static int
+acr_r352_shutdown(struct acr_r352 *acr, struct nvkm_secboot *sb)
+{
+        int i;
+
+        /* Run the unload blob to unprotect the WPR region */
+        if (acr->unload_blob && sb->wpr_set) {
+                int ret;
+
+                nvkm_debug(&sb->subdev, "running HS unload blob\n");
+                ret = sb->func->run_blob(sb, acr->unload_blob);
+                if (ret)
+                        return ret;
+                nvkm_debug(&sb->subdev, "HS unload blob completed\n");
+        }
+
+        for (i = 0; i < NVKM_SECBOOT_FALCON_END; i++)
+                acr->falcon_state[i] = NON_SECURE;
+
+        sb->wpr_set = false;
+
+        return 0;
+}
+
+static int
+acr_r352_bootstrap(struct acr_r352 *acr, struct nvkm_secboot *sb)
+{
+        int ret;
+
+        if (sb->wpr_set)
+                return 0;
+
+        /* Make sure all blobs are ready */
+        ret = acr_r352_load_blobs(acr, sb);
+        if (ret)
+                return ret;
+
+        nvkm_debug(&sb->subdev, "running HS load blob\n");
+        ret = sb->func->run_blob(sb, acr->load_blob);
+        /* clear halt interrupt */
+        nvkm_falcon_clear_interrupt(sb->boot_falcon, 0x10);
+        if (ret)
+                return ret;
+        nvkm_debug(&sb->subdev, "HS load blob completed\n");
+
+        sb->wpr_set = true;
+
+        return 0;
+}
+
+/*
+ * acr_r352_reset() - execute secure boot from the prepared state
+ *
+ * Load the HS bootloader and ask the falcon to run it. This will in turn
+ * load the HS firmware and run it, so once the falcon stops all the managed
+ * falcons should have their LS firmware loaded and be ready to run.
+ */
+static int
+acr_r352_reset(struct nvkm_acr *_acr, struct nvkm_secboot *sb,
+               enum nvkm_secboot_falcon falcon)
+{
+        struct acr_r352 *acr = acr_r352(_acr);
+        int ret;
+
+        /*
+         * Dummy GM200 implementation: perform secure boot each time we are
+         * called on FECS. Since only FECS and GPCCS are managed and started
+         * together, this ought to be safe.
+         *
+         * Once we have proper PMU firmware and support, this will be changed
+         * to a proper call to the PMU method.
+         */
+        if (falcon != NVKM_SECBOOT_FALCON_FECS)
+                goto end;
+
+        ret = acr_r352_shutdown(acr, sb);
+        if (ret)
+                return ret;
+
+        acr_r352_bootstrap(acr, sb);
+        if (ret)
+                return ret;
+
+end:
+        acr->falcon_state[falcon] = RESET;
+        return 0;
+}
+
+static int
+acr_r352_start(struct nvkm_acr *_acr, struct nvkm_secboot *sb,
+                    enum nvkm_secboot_falcon falcon)
+{
+        struct acr_r352 *acr = acr_r352(_acr);
+        const struct nvkm_subdev *subdev = &sb->subdev;
+        int base;
+
+        switch (falcon) {
+        case NVKM_SECBOOT_FALCON_FECS:
+                base = 0x409000;
+                break;
+        case NVKM_SECBOOT_FALCON_GPCCS:
+                base = 0x41a000;
+                break;
+        default:
+                nvkm_error(subdev, "cannot start unhandled falcon!\n");
+                return -EINVAL;
+        }
+
+        nvkm_wr32(subdev->device, base + 0x130, 0x00000002);
+        acr->falcon_state[falcon] = RUNNING;
+
+        return 0;
+}
+
+static int
+acr_r352_fini(struct nvkm_acr *_acr, struct nvkm_secboot *sb, bool suspend)
+{
+        struct acr_r352 *acr = acr_r352(_acr);
+
+        return acr_r352_shutdown(acr, sb);
+}
+
+static void
+acr_r352_dtor(struct nvkm_acr *_acr)
+{
+        struct acr_r352 *acr = acr_r352(_acr);
+
+        nvkm_gpuobj_del(&acr->unload_blob);
+
+        kfree(acr->hsbl_blob);
+        nvkm_gpuobj_del(&acr->load_blob);
+        nvkm_gpuobj_del(&acr->ls_blob);
+
+        kfree(acr);
+}
+
+const struct acr_r352_ls_func
+acr_r352_ls_fecs_func = {
+        .load = acr_ls_ucode_load_fecs,
+        .generate_bl_desc = acr_r352_generate_flcn_bl_desc,
+        .bl_desc_size = sizeof(struct acr_r352_flcn_bl_desc),
+};
+
+const struct acr_r352_ls_func
+acr_r352_ls_gpccs_func = {
+        .load = acr_ls_ucode_load_gpccs,
+        .generate_bl_desc = acr_r352_generate_flcn_bl_desc,
+        .bl_desc_size = sizeof(struct acr_r352_flcn_bl_desc),
+        /* GPCCS will be loaded using PRI */
+        .lhdr_flags = LSF_FLAG_FORCE_PRIV_LOAD,
+};
+
+const struct acr_r352_func
+acr_r352_func = {
+        .generate_hs_bl_desc = acr_r352_generate_hs_bl_desc,
+        .hs_bl_desc_size = sizeof(struct acr_r352_flcn_bl_desc),
+        .ls_ucode_img_load = acr_r352_ls_ucode_img_load,
+        .ls_fill_headers = acr_r352_ls_fill_headers,
+        .ls_write_wpr = acr_r352_ls_write_wpr,
+        .ls_func = {
+                [NVKM_SECBOOT_FALCON_FECS] = &acr_r352_ls_fecs_func,
+                [NVKM_SECBOOT_FALCON_GPCCS] = &acr_r352_ls_gpccs_func,
+        },
+};
+
+static const struct nvkm_acr_func
+acr_r352_base_func = {
+        .dtor = acr_r352_dtor,
+        .fini = acr_r352_fini,
+        .load = acr_r352_load,
+        .reset = acr_r352_reset,
+        .start = acr_r352_start,
+};
+
+struct nvkm_acr *
+acr_r352_new_(const struct acr_r352_func *func,
+              enum nvkm_secboot_falcon boot_falcon,
+              unsigned long managed_falcons)
+{
+        struct acr_r352 *acr;
+
+        acr = kzalloc(sizeof(*acr), GFP_KERNEL);
+        if (!acr)
+                return ERR_PTR(-ENOMEM);
+
+        acr->base.boot_falcon = boot_falcon;
+        acr->base.managed_falcons = managed_falcons;
+        acr->base.func = &acr_r352_base_func;
+        acr->func = func;
+
+        return &acr->base;
+}
+
+struct nvkm_acr *
+acr_r352_new(unsigned long managed_falcons)
+{
+        return acr_r352_new_(&acr_r352_func, NVKM_SECBOOT_FALCON_PMU,
+                             managed_falcons);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r352.h b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r352.h
new file mode 100644
index 000000000000..ad5923b0fd3c
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r352.h
@@ -0,0 +1,250 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+#ifndef __NVKM_SECBOOT_ACR_R352_H__
+#define __NVKM_SECBOOT_ACR_R352_H__
+
+#include "acr.h"
+#include "ls_ucode.h"
+
+struct ls_ucode_img;
+
+#define ACR_R352_MAX_APPS 8
+
+/*
+ *
+ * LS blob structures
+ *
+ */
+
+/**
+ * struct acr_r352_lsf_lsb_header - LS firmware header
+ * @signature:          signature to verify the firmware against
+ * @ucode_off:          offset of the ucode blob in the WPR region. The ucode
+ *                      blob contains the bootloader, code and data of the
+ *                      LS falcon
+ * @ucode_size:         size of the ucode blob, including bootloader
+ * @data_size:          size of the ucode blob data
+ * @bl_code_size:       size of the bootloader code
+ * @bl_imem_off:        offset in imem of the bootloader
+ * @bl_data_off:        offset of the bootloader data in WPR region
+ * @bl_data_size:       size of the bootloader data
+ * @app_code_off:       offset of the app code relative to ucode_off
+ * @app_code_size:      size of the app code
+ * @app_data_off:       offset of the app data relative to ucode_off
+ * @app_data_size:      size of the app data
+ * @flags:              flags for the secure bootloader
+ *
+ * This structure is written into the WPR region for each managed falcon. Each
+ * instance is referenced by the lsb_offset member of the corresponding
+ * lsf_wpr_header.
+ */
+struct acr_r352_lsf_lsb_header {
+        /**
+         * LS falcon signatures
+         * @prd_keys:           signature to use in production mode
+         * @dgb_keys:           signature to use in debug mode
+         * @b_prd_present:      whether the production key is present
+         * @b_dgb_present:      whether the debug key is present
+         * @falcon_id:          ID of the falcon the ucode applies to
+         */
+        struct {
+                u8 prd_keys[2][16];
+                u8 dbg_keys[2][16];
+                u32 b_prd_present;
+                u32 b_dbg_present;
+                u32 falcon_id;
+        } signature;
+        u32 ucode_off;
+        u32 ucode_size;
+        u32 data_size;
+        u32 bl_code_size;
+        u32 bl_imem_off;
+        u32 bl_data_off;
+        u32 bl_data_size;
+        u32 app_code_off;
+        u32 app_code_size;
+        u32 app_data_off;
+        u32 app_data_size;
+        u32 flags;
+#define LSF_FLAG_LOAD_CODE_AT_0         1
+#define LSF_FLAG_DMACTL_REQ_CTX         4
+#define LSF_FLAG_FORCE_PRIV_LOAD        8
+};
+
+/**
+ * struct acr_r352_lsf_wpr_header - LS blob WPR Header
+ * @falcon_id:          LS falcon ID
+ * @lsb_offset:         offset of the lsb_lsf_header in the WPR region
+ * @bootstrap_owner:    secure falcon reponsible for bootstrapping the LS falcon
+ * @lazy_bootstrap:     skip bootstrapping by ACR
+ * @status:             bootstrapping status
+ *
+ * An array of these is written at the beginning of the WPR region, one for
+ * each managed falcon. The array is terminated by an instance which falcon_id
+ * is LSF_FALCON_ID_INVALID.
+ */
+struct acr_r352_lsf_wpr_header {
+        u32 falcon_id;
+        u32 lsb_offset;
+        u32 bootstrap_owner;
+        u32 lazy_bootstrap;
+        u32 status;
+#define LSF_IMAGE_STATUS_NONE                           0
+#define LSF_IMAGE_STATUS_COPY                           1
+#define LSF_IMAGE_STATUS_VALIDATION_CODE_FAILED         2
+#define LSF_IMAGE_STATUS_VALIDATION_DATA_FAILED         3
+#define LSF_IMAGE_STATUS_VALIDATION_DONE                4
+#define LSF_IMAGE_STATUS_VALIDATION_SKIPPED             5
+#define LSF_IMAGE_STATUS_BOOTSTRAP_READY                6
+};
+
+/**
+ * struct ls_ucode_img_r352 - ucode image augmented with r352 headers
+ */
+struct ls_ucode_img_r352 {
+        struct ls_ucode_img base;
+
+        struct acr_r352_lsf_wpr_header wpr_header;
+        struct acr_r352_lsf_lsb_header lsb_header;
+};
+#define ls_ucode_img_r352(i) container_of(i, struct ls_ucode_img_r352, base)
+
+
+/*
+ * HS blob structures
+ */
+
+struct hsf_load_header_app {
+        u32 sec_code_off;
+        u32 sec_code_size;
+};
+
+/**
+ * struct hsf_load_header - HS firmware load header
+ */
+struct hsf_load_header {
+        u32 non_sec_code_off;
+        u32 non_sec_code_size;
+        u32 data_dma_base;
+        u32 data_size;
+        u32 num_apps;
+        struct hsf_load_header_app app[0];
+};
+
+/**
+ * struct acr_r352_ls_func - manages a single LS firmware
+ *
+ * @load: load the external firmware into a ls_ucode_img
+ * @generate_bl_desc: function called on a block of bl_desc_size to generate the
+ *                    proper bootloader descriptor for this LS firmware
+ * @bl_desc_size: size of the bootloader descriptor
+ * @lhdr_flags: LS flags
+ */
+struct acr_r352_ls_func {
+        int (*load)(const struct nvkm_subdev *, struct ls_ucode_img *);
+        void (*generate_bl_desc)(const struct nvkm_acr *,
+                                 const struct ls_ucode_img *, u64, void *);
+        u32 bl_desc_size;
+        u32 lhdr_flags;
+};
+
+struct acr_r352;
+
+/**
+ * struct acr_r352_func - manages nuances between ACR versions
+ *
+ * @generate_hs_bl_desc: function called on a block of bl_desc_size to generate
+ *                       the proper HS bootloader descriptor
+ * @hs_bl_desc_size: size of the HS bootloader descriptor
+ */
+struct acr_r352_func {
+        void (*generate_hs_bl_desc)(const struct hsf_load_header *, void *,
+                                    u64);
+        u32 hs_bl_desc_size;
+
+        struct ls_ucode_img *(*ls_ucode_img_load)(const struct acr_r352 *,
+                                                  enum nvkm_secboot_falcon);
+        int (*ls_fill_headers)(struct acr_r352 *, struct list_head *);
+        int (*ls_write_wpr)(struct acr_r352 *, struct list_head *,
+                            struct nvkm_gpuobj *, u32);
+
+        const struct acr_r352_ls_func *ls_func[NVKM_SECBOOT_FALCON_END];
+};
+
+/**
+ * struct acr_r352 - ACR data for driver release 352 (and beyond)
+ */
+struct acr_r352 {
+        struct nvkm_acr base;
+        const struct acr_r352_func *func;
+
+        /*
+         * HS FW - lock WPR region (dGPU only) and load LS FWs
+         * on Tegra the HS FW copies the LS blob into the fixed WPR instead
+         */
+        struct nvkm_gpuobj *load_blob;
+        struct {
+                struct hsf_load_header load_bl_header;
+                struct hsf_load_header_app __load_apps[ACR_R352_MAX_APPS];
+        };
+
+        /* HS FW - unlock WPR region (dGPU only) */
+        struct nvkm_gpuobj *unload_blob;
+        struct {
+                struct hsf_load_header unload_bl_header;
+                struct hsf_load_header_app __unload_apps[ACR_R352_MAX_APPS];
+        };
+
+        /* HS bootloader */
+        void *hsbl_blob;
+
+        /* LS FWs, to be loaded by the HS ACR */
+        struct nvkm_gpuobj *ls_blob;
+
+        /* Firmware already loaded? */
+        bool firmware_ok;
+
+        /* Falcons to lazy-bootstrap */
+        u32 lazy_bootstrap;
+
+        /* To keep track of the state of all managed falcons */
+        enum {
+                /* In non-secure state, no firmware loaded, no privileges*/
+                NON_SECURE = 0,
+                /* In low-secure mode and ready to be started */
+                RESET,
+                /* In low-secure mode and running */
+                RUNNING,
+        } falcon_state[NVKM_SECBOOT_FALCON_END];
+};
+#define acr_r352(acr) container_of(acr, struct acr_r352, base)
+
+struct nvkm_acr *acr_r352_new_(const struct acr_r352_func *,
+                               enum nvkm_secboot_falcon, unsigned long);
+
+struct ls_ucode_img *acr_r352_ls_ucode_img_load(const struct acr_r352 *,
+                                                enum nvkm_secboot_falcon);
+int acr_r352_ls_fill_headers(struct acr_r352 *, struct list_head *);
+int acr_r352_ls_write_wpr(struct acr_r352 *, struct list_head *,
+                          struct nvkm_gpuobj *, u32);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r361.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r361.c
new file mode 100644
index 000000000000..f0aff1d98474
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/acr_r361.c
@@ -0,0 +1,138 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include "acr_r352.h"
+
+#include <engine/falcon.h>
+
+/**
+ * struct acr_r361_flcn_bl_desc - DMEM bootloader descriptor
+ * @signature:          16B signature for secure code. 0s if no secure code
+ * @ctx_dma:            DMA context to be used by BL while loading code/data
+ * @code_dma_base:      256B-aligned Physical FB Address where code is located
+ *                      (falcon's $xcbase register)
+ * @non_sec_code_off:   offset from code_dma_base where the non-secure code is
+ *                      located. The offset must be multiple of 256 to help perf
+ * @non_sec_code_size:  the size of the nonSecure code part.
+ * @sec_code_off:       offset from code_dma_base where the secure code is
+ *                      located. The offset must be multiple of 256 to help perf
+ * @sec_code_size:      offset from code_dma_base where the secure code is
+ *                      located. The offset must be multiple of 256 to help perf
+ * @code_entry_point:   code entry point which will be invoked by BL after
+ *                      code is loaded.
+ * @data_dma_base:      256B aligned Physical FB Address where data is located.
+ *                      (falcon's $xdbase register)
+ * @data_size:          size of data block. Should be multiple of 256B
+ *
+ * Structure used by the bootloader to load the rest of the code. This has
+ * to be filled by host and copied into DMEM at offset provided in the
+ * hsflcn_bl_desc.bl_desc_dmem_load_off.
+ */
+struct acr_r361_flcn_bl_desc {
+        u32 reserved[4];
+        u32 signature[4];
+        u32 ctx_dma;
+        struct flcn_u64 code_dma_base;
+        u32 non_sec_code_off;
+        u32 non_sec_code_size;
+        u32 sec_code_off;
+        u32 sec_code_size;
+        u32 code_entry_point;
+        struct flcn_u64 data_dma_base;
+        u32 data_size;
+};
+
+static void
+acr_r361_generate_flcn_bl_desc(const struct nvkm_acr *acr,
+                               const struct ls_ucode_img *_img, u64 wpr_addr,
+                               void *_desc)
+{
+        struct ls_ucode_img_r352 *img = ls_ucode_img_r352(_img);
+        struct acr_r361_flcn_bl_desc *desc = _desc;
+        const struct ls_ucode_img_desc *pdesc = &img->base.ucode_desc;
+        u64 base, addr_code, addr_data;
+
+        base = wpr_addr + img->lsb_header.ucode_off + pdesc->app_start_offset;
+        addr_code = base + pdesc->app_resident_code_offset;
+        addr_data = base + pdesc->app_resident_data_offset;
+
+        desc->ctx_dma = FALCON_DMAIDX_UCODE;
+        desc->code_dma_base = u64_to_flcn64(addr_code);
+        desc->non_sec_code_off = pdesc->app_resident_code_offset;
+        desc->non_sec_code_size = pdesc->app_resident_code_size;
+        desc->code_entry_point = pdesc->app_imem_entry;
+        desc->data_dma_base = u64_to_flcn64(addr_data);
+        desc->data_size = pdesc->app_resident_data_size;
+}
+
+static void
+acr_r361_generate_hs_bl_desc(const struct hsf_load_header *hdr, void *_bl_desc,
+                            u64 offset)
+{
+        struct acr_r361_flcn_bl_desc *bl_desc = _bl_desc;
+
+        bl_desc->ctx_dma = FALCON_DMAIDX_VIRT;
+        bl_desc->code_dma_base = u64_to_flcn64(offset);
+        bl_desc->non_sec_code_off = hdr->non_sec_code_off;
+        bl_desc->non_sec_code_size = hdr->non_sec_code_size;
+        bl_desc->sec_code_off = hdr->app[0].sec_code_off;
+        bl_desc->sec_code_size = hdr->app[0].sec_code_size;
+        bl_desc->code_entry_point = 0;
+        bl_desc->data_dma_base = u64_to_flcn64(offset + hdr->data_dma_base);
+        bl_desc->data_size = hdr->data_size;
+}
+
+const struct acr_r352_ls_func
+acr_r361_ls_fecs_func = {
+        .load = acr_ls_ucode_load_fecs,
+        .generate_bl_desc = acr_r361_generate_flcn_bl_desc,
+        .bl_desc_size = sizeof(struct acr_r361_flcn_bl_desc),
+};
+
+const struct acr_r352_ls_func
+acr_r361_ls_gpccs_func = {
+        .load = acr_ls_ucode_load_gpccs,
+        .generate_bl_desc = acr_r361_generate_flcn_bl_desc,
+        .bl_desc_size = sizeof(struct acr_r361_flcn_bl_desc),
+        /* GPCCS will be loaded using PRI */
+        .lhdr_flags = LSF_FLAG_FORCE_PRIV_LOAD,
+};
+
+const struct acr_r352_func
+acr_r361_func = {
+        .generate_hs_bl_desc = acr_r361_generate_hs_bl_desc,
+        .hs_bl_desc_size = sizeof(struct acr_r361_flcn_bl_desc),
+        .ls_ucode_img_load = acr_r352_ls_ucode_img_load,
+        .ls_fill_headers = acr_r352_ls_fill_headers,
+        .ls_write_wpr = acr_r352_ls_write_wpr,
+        .ls_func = {
+                [NVKM_SECBOOT_FALCON_FECS] = &acr_r361_ls_fecs_func,
+                [NVKM_SECBOOT_FALCON_GPCCS] = &acr_r361_ls_gpccs_func,
+        },
+};
+
+struct nvkm_acr *
+acr_r361_new(unsigned long managed_falcons)
+{
+        return acr_r352_new_(&acr_r361_func, NVKM_SECBOOT_FALCON_PMU,
+                             managed_falcons);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c
index 314be2192b7d..27c9dfffb9a6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c
@@ -19,184 +19,108 @@
  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  * DEALINGS IN THE SOFTWARE.
  */
+
+/*
+ * Secure boot is the process by which NVIDIA-signed firmware is loaded into
+ * some of the falcons of a GPU. For production devices this is the only way
+ * for the firmware to access useful (but sensitive) registers.
+ *
+ * A Falcon microprocessor supporting advanced security modes can run in one of
+ * three modes:
+ *
+ * - Non-secure (NS). In this mode, functionality is similar to Falcon
+ *   architectures before security modes were introduced (pre-Maxwell), but
+ *   capability is restricted. In particular, certain registers may be
+ *   inaccessible for reads and/or writes, and physical memory access may be
+ *   disabled (on certain Falcon instances). This is the only possible mode that
+ *   can be used if you don't have microcode cryptographically signed by NVIDIA.
+ *
+ * - Heavy Secure (HS). In this mode, the microprocessor is a black box - it's
+ *   not possible to read or write any Falcon internal state or Falcon registers
+ *   from outside the Falcon (for example, from the host system). The only way
+ *   to enable this mode is by loading microcode that has been signed by NVIDIA.
+ *   (The loading process involves tagging the IMEM block as secure, writing the
+ *   signature into a Falcon register, and starting execution. The hardware will
+ *   validate the signature, and if valid, grant HS privileges.)
+ *
+ * - Light Secure (LS). In this mode, the microprocessor has more privileges
+ *   than NS but fewer than HS. Some of the microprocessor state is visible to
+ *   host software to ease debugging. The only way to enable this mode is by HS
+ *   microcode enabling LS mode. Some privileges available to HS mode are not
+ *   available here. LS mode is introduced in GM20x.
+ *
+ * Secure boot consists in temporarily switching a HS-capable falcon (typically
+ * PMU) into HS mode in order to validate the LS firmwares of managed falcons,
+ * load them, and switch managed falcons into LS mode. Once secure boot
+ * completes, no falcon remains in HS mode.
+ *
+ * Secure boot requires a write-protected memory region (WPR) which can only be
+ * written by the secure falcon. On dGPU, the driver sets up the WPR region in
+ * video memory. On Tegra, it is set up by the bootloader and its location and
+ * size written into memory controller registers.
+ *
+ * The secure boot process takes place as follows:
+ *
+ * 1) A LS blob is constructed that contains all the LS firmwares we want to
+ *    load, along with their signatures and bootloaders.
+ *
+ * 2) A HS blob (also called ACR) is created that contains the signed HS
+ *    firmware in charge of loading the LS firmwares into their respective
+ *    falcons.
+ *
+ * 3) The HS blob is loaded (via its own bootloader) and executed on the
+ *    HS-capable falcon. It authenticates itself, switches the secure falcon to
+ *    HS mode and setup the WPR region around the LS blob (dGPU) or copies the
+ *    LS blob into the WPR region (Tegra).
+ *
+ * 4) The LS blob is now secure from all external tampering. The HS falcon
+ *    checks the signatures of the LS firmwares and, if valid, switches the
+ *    managed falcons to LS mode and makes them ready to run the LS firmware.
+ *
+ * 5) The managed falcons remain in LS mode and can be started.
+ *
+ */
+
 #include "priv.h"
+#include "acr.h"
 
 #include <subdev/mc.h>
 #include <subdev/timer.h>
+#include <subdev/pmu.h>
 
-static const char *
-managed_falcons_names[] = {
+const char *
+nvkm_secboot_falcon_name[] = {
         [NVKM_SECBOOT_FALCON_PMU] = "PMU",
         [NVKM_SECBOOT_FALCON_RESERVED] = "<reserved>",
         [NVKM_SECBOOT_FALCON_FECS] = "FECS",
         [NVKM_SECBOOT_FALCON_GPCCS] = "GPCCS",
         [NVKM_SECBOOT_FALCON_END] = "<invalid>",
 };
-
-/*
- * Helper falcon functions
- */
-
-static int
-falcon_clear_halt_interrupt(struct nvkm_device *device, u32 base)
-{
-        int ret;
-
-        /* clear halt interrupt */
-        nvkm_mask(device, base + 0x004, 0x10, 0x10);
-        /* wait until halt interrupt is cleared */
-        ret = nvkm_wait_msec(device, 10, base + 0x008, 0x10, 0x0);
-        if (ret < 0)
-                return ret;
-
-        return 0;
-}
-
-static int
-falcon_wait_idle(struct nvkm_device *device, u32 base)
-{
-        int ret;
-
-        ret = nvkm_wait_msec(device, 10, base + 0x04c, 0xffff, 0x0);
-        if (ret < 0)
-                return ret;
-
-        return 0;
-}
-
-static int
-nvkm_secboot_falcon_enable(struct nvkm_secboot *sb)
-{
-        struct nvkm_device *device = sb->subdev.device;
-        int ret;
-
-        /* enable engine */
-        nvkm_mc_enable(device, sb->devidx);
-        ret = nvkm_wait_msec(device, 10, sb->base + 0x10c, 0x6, 0x0);
-        if (ret < 0) {
-                nvkm_error(&sb->subdev, "Falcon mem scrubbing timeout\n");
-                nvkm_mc_disable(device, sb->devidx);
-                return ret;
-        }
-
-        ret = falcon_wait_idle(device, sb->base);
-        if (ret)
-                return ret;
-
-        /* enable IRQs */
-        nvkm_wr32(device, sb->base + 0x010, 0xff);
-        nvkm_mc_intr_mask(device, sb->devidx, true);
-
-        return 0;
-}
-
-static int
-nvkm_secboot_falcon_disable(struct nvkm_secboot *sb)
-{
-        struct nvkm_device *device = sb->subdev.device;
-
-        /* disable IRQs and wait for any previous code to complete */
-        nvkm_mc_intr_mask(device, sb->devidx, false);
-        nvkm_wr32(device, sb->base + 0x014, 0xff);
-
-        falcon_wait_idle(device, sb->base);
-
-        /* disable engine */
-        nvkm_mc_disable(device, sb->devidx);
-
-        return 0;
-}
-
-int
-nvkm_secboot_falcon_reset(struct nvkm_secboot *sb)
-{
-        int ret;
-
-        ret = nvkm_secboot_falcon_disable(sb);
-        if (ret)
-                return ret;
-
-        ret = nvkm_secboot_falcon_enable(sb);
-        if (ret)
-                return ret;
-
-        return 0;
-}
-
-/**
- * nvkm_secboot_falcon_run - run the falcon that will perform secure boot
- *
- * This function is to be called after all chip-specific preparations have
- * been completed. It will start the falcon to perform secure boot, wait for
- * it to halt, and report if an error occurred.
- */
-int
-nvkm_secboot_falcon_run(struct nvkm_secboot *sb)
-{
-        struct nvkm_device *device = sb->subdev.device;
-        int ret;
-
-        /* Start falcon */
-        nvkm_wr32(device, sb->base + 0x100, 0x2);
-
-        /* Wait for falcon halt */
-        ret = nvkm_wait_msec(device, 100, sb->base + 0x100, 0x10, 0x10);
-        if (ret < 0)
-                return ret;
-
-        /* If mailbox register contains an error code, then ACR has failed */
-        ret = nvkm_rd32(device, sb->base + 0x040);
-        if (ret) {
-                nvkm_error(&sb->subdev, "ACR boot failed, ret 0x%08x", ret);
-                falcon_clear_halt_interrupt(device, sb->base);
-                return -EINVAL;
-        }
-
-        return 0;
-}
-
-
 /**
  * nvkm_secboot_reset() - reset specified falcon
  */
 int
-nvkm_secboot_reset(struct nvkm_secboot *sb, u32 falcon)
+nvkm_secboot_reset(struct nvkm_secboot *sb, enum nvkm_secboot_falcon falcon)
 {
         /* Unmanaged falcon? */
-        if (!(BIT(falcon) & sb->func->managed_falcons)) {
+        if (!(BIT(falcon) & sb->acr->managed_falcons)) {
                 nvkm_error(&sb->subdev, "cannot reset unmanaged falcon!\n");
                 return -EINVAL;
         }
 
-        return sb->func->reset(sb, falcon);
-}
-
-/**
- * nvkm_secboot_start() - start specified falcon
- */
-int
-nvkm_secboot_start(struct nvkm_secboot *sb, u32 falcon)
-{
-        /* Unmanaged falcon? */
-        if (!(BIT(falcon) & sb->func->managed_falcons)) {
-                nvkm_error(&sb->subdev, "cannot start unmanaged falcon!\n");
-                return -EINVAL;
-        }
-
-        return sb->func->start(sb, falcon);
+        return sb->acr->func->reset(sb->acr, sb, falcon);
 }
 
 /**
  * nvkm_secboot_is_managed() - check whether a given falcon is securely-managed
  */
 bool
-nvkm_secboot_is_managed(struct nvkm_secboot *secboot,
-                        enum nvkm_secboot_falcon fid)
+nvkm_secboot_is_managed(struct nvkm_secboot *sb, enum nvkm_secboot_falcon fid)
 {
-        if (!secboot)
+        if (!sb)
                 return false;
 
-        return secboot->func->managed_falcons & BIT(fid);
+        return sb->acr->managed_falcons & BIT(fid);
 }
 
 static int
@@ -205,9 +129,19 @@ nvkm_secboot_oneinit(struct nvkm_subdev *subdev)
         struct nvkm_secboot *sb = nvkm_secboot(subdev);
         int ret = 0;
 
+        switch (sb->acr->boot_falcon) {
+        case NVKM_SECBOOT_FALCON_PMU:
+                sb->boot_falcon = subdev->device->pmu->falcon;
+                break;
+        default:
+                nvkm_error(subdev, "Unmanaged boot falcon %s!\n",
+                                        nvkm_secboot_falcon_name[sb->acr->boot_falcon]);
+                return -EINVAL;
+        }
+
         /* Call chip-specific init function */
-        if (sb->func->init)
-                ret = sb->func->init(sb);
+        if (sb->func->oneinit)
+                ret = sb->func->oneinit(sb);
         if (ret) {
                 nvkm_error(subdev, "Secure Boot initialization failed: %d\n",
                            ret);
@@ -249,7 +183,7 @@ nvkm_secboot = {
 };
 
 int
-nvkm_secboot_ctor(const struct nvkm_secboot_func *func,
+nvkm_secboot_ctor(const struct nvkm_secboot_func *func, struct nvkm_acr *acr,
                   struct nvkm_device *device, int index,
                   struct nvkm_secboot *sb)
 {
@@ -257,22 +191,14 @@ nvkm_secboot_ctor(const struct nvkm_secboot_func *func,
 
         nvkm_subdev_ctor(&nvkm_secboot, device, index, &sb->subdev);
         sb->func = func;
-
-        /* setup the performing falcon's base address and masks */
-        switch (func->boot_falcon) {
-        case NVKM_SECBOOT_FALCON_PMU:
-                sb->devidx = NVKM_SUBDEV_PMU;
-                sb->base = 0x10a000;
-                break;
-        default:
-                nvkm_error(&sb->subdev, "invalid secure boot falcon\n");
-                return -EINVAL;
-        };
+        sb->acr = acr;
+        acr->subdev = &sb->subdev;
 
         nvkm_debug(&sb->subdev, "securely managed falcons:\n");
-        for_each_set_bit(fid, &sb->func->managed_falcons,
+        for_each_set_bit(fid, &sb->acr->managed_falcons,
                          NVKM_SECBOOT_FALCON_END)
-                nvkm_debug(&sb->subdev, "- %s\n", managed_falcons_names[fid]);
+                nvkm_debug(&sb->subdev, "- %s\n",
+                           nvkm_secboot_falcon_name[fid]);
 
         return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c
index ec48e4ace37a..813c4eb0b25f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c
@@ -20,1313 +20,84 @@
  * DEALINGS IN THE SOFTWARE.
  */
 
-/*
- * Secure boot is the process by which NVIDIA-signed firmware is loaded into
- * some of the falcons of a GPU. For production devices this is the only way
- * for the firmware to access useful (but sensitive) registers.
- *
- * A Falcon microprocessor supporting advanced security modes can run in one of
- * three modes:
- *
- * - Non-secure (NS). In this mode, functionality is similar to Falcon
- *   architectures before security modes were introduced (pre-Maxwell), but
- *   capability is restricted. In particular, certain registers may be
- *   inaccessible for reads and/or writes, and physical memory access may be
- *   disabled (on certain Falcon instances). This is the only possible mode that
- *   can be used if you don't have microcode cryptographically signed by NVIDIA.
- *
- * - Heavy Secure (HS). In this mode, the microprocessor is a black box - it's
- *   not possible to read or write any Falcon internal state or Falcon registers
- *   from outside the Falcon (for example, from the host system). The only way
- *   to enable this mode is by loading microcode that has been signed by NVIDIA.
- *   (The loading process involves tagging the IMEM block as secure, writing the
- *   signature into a Falcon register, and starting execution. The hardware will
- *   validate the signature, and if valid, grant HS privileges.)
- *
- * - Light Secure (LS). In this mode, the microprocessor has more privileges
- *   than NS but fewer than HS. Some of the microprocessor state is visible to
- *   host software to ease debugging. The only way to enable this mode is by HS
- *   microcode enabling LS mode. Some privileges available to HS mode are not
- *   available here. LS mode is introduced in GM20x.
- *
- * Secure boot consists in temporarily switching a HS-capable falcon (typically
- * PMU) into HS mode in order to validate the LS firmwares of managed falcons,
- * load them, and switch managed falcons into LS mode. Once secure boot
- * completes, no falcon remains in HS mode.
- *
- * Secure boot requires a write-protected memory region (WPR) which can only be
- * written by the secure falcon. On dGPU, the driver sets up the WPR region in
- * video memory. On Tegra, it is set up by the bootloader and its location and
- * size written into memory controller registers.
- *
- * The secure boot process takes place as follows:
- *
- * 1) A LS blob is constructed that contains all the LS firmwares we want to
- *    load, along with their signatures and bootloaders.
- *
- * 2) A HS blob (also called ACR) is created that contains the signed HS
- *    firmware in charge of loading the LS firmwares into their respective
- *    falcons.
- *
- * 3) The HS blob is loaded (via its own bootloader) and executed on the
- *    HS-capable falcon. It authenticates itself, switches the secure falcon to
- *    HS mode and setup the WPR region around the LS blob (dGPU) or copies the
- *    LS blob into the WPR region (Tegra).
- *
- * 4) The LS blob is now secure from all external tampering. The HS falcon
- *    checks the signatures of the LS firmwares and, if valid, switches the
- *    managed falcons to LS mode and makes them ready to run the LS firmware.
- *
- * 5) The managed falcons remain in LS mode and can be started.
- *
- */
 
-#include "priv.h"
+#include "acr.h"
+#include "gm200.h"
 
 #include <core/gpuobj.h>
-#include <core/firmware.h>
 #include <subdev/fb.h>
-
-enum {
-        FALCON_DMAIDX_UCODE             = 0,
-        FALCON_DMAIDX_VIRT              = 1,
-        FALCON_DMAIDX_PHYS_VID          = 2,
-        FALCON_DMAIDX_PHYS_SYS_COH      = 3,
-        FALCON_DMAIDX_PHYS_SYS_NCOH     = 4,
-};
-
-/**
- * struct fw_bin_header - header of firmware files
- * @bin_magic:          always 0x3b1d14f0
- * @bin_ver:            version of the bin format
- * @bin_size:           entire image size including this header
- * @header_offset:      offset of the firmware/bootloader header in the file
- * @data_offset:        offset of the firmware/bootloader payload in the file
- * @data_size:          size of the payload
- *
- * This header is located at the beginning of the HS firmware and HS bootloader
- * files, to describe where the headers and data can be found.
- */
-struct fw_bin_header {
-        u32 bin_magic;
-        u32 bin_ver;
-        u32 bin_size;
-        u32 header_offset;
-        u32 data_offset;
-        u32 data_size;
-};
-
-/**
- * struct fw_bl_desc - firmware bootloader descriptor
- * @start_tag:          starting tag of bootloader
- * @desc_dmem_load_off: DMEM offset of flcn_bl_dmem_desc
- * @code_off:           offset of code section
- * @code_size:          size of code section
- * @data_off:           offset of data section
- * @data_size:          size of data section
- *
- * This structure is embedded in bootloader firmware files at to describe the
- * IMEM and DMEM layout expected by the bootloader.
- */
-struct fw_bl_desc {
-        u32 start_tag;
-        u32 dmem_load_off;
-        u32 code_off;
-        u32 code_size;
-        u32 data_off;
-        u32 data_size;
-};
-
-
-/*
- *
- * LS blob structures
- *
- */
-
-/**
- * struct lsf_ucode_desc - LS falcon signatures
- * @prd_keys:           signature to use when the GPU is in production mode
- * @dgb_keys:           signature to use when the GPU is in debug mode
- * @b_prd_present:      whether the production key is present
- * @b_dgb_present:      whether the debug key is present
- * @falcon_id:          ID of the falcon the ucode applies to
- *
- * Directly loaded from a signature file.
- */
-struct lsf_ucode_desc {
-        u8  prd_keys[2][16];
-        u8  dbg_keys[2][16];
-        u32 b_prd_present;
-        u32 b_dbg_present;
-        u32 falcon_id;
-};
-
-/**
- * struct lsf_lsb_header - LS firmware header
- * @signature:          signature to verify the firmware against
- * @ucode_off:          offset of the ucode blob in the WPR region. The ucode
- *                      blob contains the bootloader, code and data of the
- *                      LS falcon
- * @ucode_size:         size of the ucode blob, including bootloader
- * @data_size:          size of the ucode blob data
- * @bl_code_size:       size of the bootloader code
- * @bl_imem_off:        offset in imem of the bootloader
- * @bl_data_off:        offset of the bootloader data in WPR region
- * @bl_data_size:       size of the bootloader data
- * @app_code_off:       offset of the app code relative to ucode_off
- * @app_code_size:      size of the app code
- * @app_data_off:       offset of the app data relative to ucode_off
- * @app_data_size:      size of the app data
- * @flags:              flags for the secure bootloader
- *
- * This structure is written into the WPR region for each managed falcon. Each
- * instance is referenced by the lsb_offset member of the corresponding
- * lsf_wpr_header.
- */
-struct lsf_lsb_header {
-        struct lsf_ucode_desc signature;
-        u32 ucode_off;
-        u32 ucode_size;
-        u32 data_size;
-        u32 bl_code_size;
-        u32 bl_imem_off;
-        u32 bl_data_off;
-        u32 bl_data_size;
-        u32 app_code_off;
-        u32 app_code_size;
-        u32 app_data_off;
-        u32 app_data_size;
-        u32 flags;
-#define LSF_FLAG_LOAD_CODE_AT_0         1
-#define LSF_FLAG_DMACTL_REQ_CTX         4
-#define LSF_FLAG_FORCE_PRIV_LOAD        8
-};
-
-/**
- * struct lsf_wpr_header - LS blob WPR Header
- * @falcon_id:          LS falcon ID
- * @lsb_offset:         offset of the lsb_lsf_header in the WPR region
- * @bootstrap_owner:    secure falcon reponsible for bootstrapping the LS falcon
- * @lazy_bootstrap:     skip bootstrapping by ACR
- * @status:             bootstrapping status
- *
- * An array of these is written at the beginning of the WPR region, one for
- * each managed falcon. The array is terminated by an instance which falcon_id
- * is LSF_FALCON_ID_INVALID.
- */
-struct lsf_wpr_header {
-        u32  falcon_id;
-        u32  lsb_offset;
-        u32  bootstrap_owner;
-        u32  lazy_bootstrap;
-        u32  status;
-#define LSF_IMAGE_STATUS_NONE                           0
-#define LSF_IMAGE_STATUS_COPY                           1
-#define LSF_IMAGE_STATUS_VALIDATION_CODE_FAILED         2
-#define LSF_IMAGE_STATUS_VALIDATION_DATA_FAILED         3
-#define LSF_IMAGE_STATUS_VALIDATION_DONE                4
-#define LSF_IMAGE_STATUS_VALIDATION_SKIPPED             5
-#define LSF_IMAGE_STATUS_BOOTSTRAP_READY                6
-};
-
-
-/**
- * struct ls_ucode_img_desc - descriptor of firmware image
- * @descriptor_size:            size of this descriptor
- * @image_size:                 size of the whole image
- * @bootloader_start_offset:    start offset of the bootloader in ucode image
- * @bootloader_size:            size of the bootloader
- * @bootloader_imem_offset:     start off set of the bootloader in IMEM
- * @bootloader_entry_point:     entry point of the bootloader in IMEM
- * @app_start_offset:           start offset of the LS firmware
- * @app_size:                   size of the LS firmware's code and data
- * @app_imem_offset:            offset of the app in IMEM
- * @app_imem_entry:             entry point of the app in IMEM
- * @app_dmem_offset:            offset of the data in DMEM
- * @app_resident_code_offset:   offset of app code from app_start_offset
- * @app_resident_code_size:     size of the code
- * @app_resident_data_offset:   offset of data from app_start_offset
- * @app_resident_data_size:     size of data
- *
- * A firmware image contains the code, data, and bootloader of a given LS
- * falcon in a single blob. This structure describes where everything is.
- *
- * This can be generated from a (bootloader, code, data) set if they have
- * been loaded separately, or come directly from a file.
- */
-struct ls_ucode_img_desc {
-        u32 descriptor_size;
-        u32 image_size;
-        u32 tools_version;
-        u32 app_version;
-        char date[64];
-        u32 bootloader_start_offset;
-        u32 bootloader_size;
-        u32 bootloader_imem_offset;
-        u32 bootloader_entry_point;
-        u32 app_start_offset;
-        u32 app_size;
-        u32 app_imem_offset;
-        u32 app_imem_entry;
-        u32 app_dmem_offset;
-        u32 app_resident_code_offset;
-        u32 app_resident_code_size;
-        u32 app_resident_data_offset;
-        u32 app_resident_data_size;
-        u32 nb_overlays;
-        struct {u32 start; u32 size; } load_ovl[64];
-        u32 compressed;
-};
-
-/**
- * struct ls_ucode_img - temporary storage for loaded LS firmwares
- * @node:               to link within lsf_ucode_mgr
- * @falcon_id:          ID of the falcon this LS firmware is for
- * @ucode_desc:         loaded or generated map of ucode_data
- * @ucode_header:       header of the firmware
- * @ucode_data:         firmware payload (code and data)
- * @ucode_size:         size in bytes of data in ucode_data
- * @wpr_header:         WPR header to be written to the LS blob
- * @lsb_header:         LSB header to be written to the LS blob
- *
- * Preparing the WPR LS blob requires information about all the LS firmwares
- * (size, etc) to be known. This structure contains all the data of one LS
- * firmware.
- */
-struct ls_ucode_img {
-        struct list_head node;
-        enum nvkm_secboot_falcon falcon_id;
-
-        struct ls_ucode_img_desc ucode_desc;
-        u32 *ucode_header;
-        u8 *ucode_data;
-        u32 ucode_size;
-
-        struct lsf_wpr_header wpr_header;
-        struct lsf_lsb_header lsb_header;
-};
-
-/**
- * struct ls_ucode_mgr - manager for all LS falcon firmwares
- * @count:      number of managed LS falcons
- * @wpr_size:   size of the required WPR region in bytes
- * @img_list:   linked list of lsf_ucode_img
- */
-struct ls_ucode_mgr {
-        u16 count;
-        u32 wpr_size;
-        struct list_head img_list;
-};
-
-
-/*
- *
- * HS blob structures
- *
- */
-
-/**
- * struct hsf_fw_header - HS firmware descriptor
- * @sig_dbg_offset:     offset of the debug signature
- * @sig_dbg_size:       size of the debug signature
- * @sig_prod_offset:    offset of the production signature
- * @sig_prod_size:      size of the production signature
- * @patch_loc:          offset of the offset (sic) of where the signature is
- * @patch_sig:          offset of the offset (sic) to add to sig_*_offset
- * @hdr_offset:         offset of the load header (see struct hs_load_header)
- * @hdr_size:           size of above header
- *
- * This structure is embedded in the HS firmware image at
- * hs_bin_hdr.header_offset.
- */
-struct hsf_fw_header {
-        u32 sig_dbg_offset;
-        u32 sig_dbg_size;
-        u32 sig_prod_offset;
-        u32 sig_prod_size;
-        u32 patch_loc;
-        u32 patch_sig;
-        u32 hdr_offset;
-        u32 hdr_size;
-};
-
-/**
- * struct hsf_load_header - HS firmware load header
- */
-struct hsf_load_header {
-        u32 non_sec_code_off;
-        u32 non_sec_code_size;
-        u32 data_dma_base;
-        u32 data_size;
-        u32 num_apps;
-        struct {
-                u32 sec_code_off;
-                u32 sec_code_size;
-        } app[0];
-};
-
-/**
- * Convenience function to duplicate a firmware file in memory and check that
- * it has the required minimum size.
- */
-static void *
-gm200_secboot_load_firmware(struct nvkm_subdev *subdev, const char *name,
-                    size_t min_size)
-{
-        const struct firmware *fw;
-        void *blob;
-        int ret;
-
-        ret = nvkm_firmware_get(subdev->device, name, &fw);
-        if (ret)
-                return ERR_PTR(ret);
-        if (fw->size < min_size) {
-                nvkm_error(subdev, "%s is smaller than expected size %zu\n",
-                           name, min_size);
-                nvkm_firmware_put(fw);
-                return ERR_PTR(-EINVAL);
-        }
-        blob = kmemdup(fw->data, fw->size, GFP_KERNEL);
-        nvkm_firmware_put(fw);
-        if (!blob)
-                return ERR_PTR(-ENOMEM);
-
-        return blob;
-}
-
-
-/*
- * Low-secure blob creation
- */
-
-#define BL_DESC_BLK_SIZE 256
-/**
- * Build a ucode image and descriptor from provided bootloader, code and data.
- *
- * @bl:         bootloader image, including 16-bytes descriptor
- * @code:       LS firmware code segment
- * @data:       LS firmware data segment
- * @desc:       ucode descriptor to be written
- *
- * Return: allocated ucode image with corresponding descriptor information. desc
- *         is also updated to contain the right offsets within returned image.
- */
-static void *
-ls_ucode_img_build(const struct firmware *bl, const struct firmware *code,
-                   const struct firmware *data, struct ls_ucode_img_desc *desc)
-{
-        struct fw_bin_header *bin_hdr = (void *)bl->data;
-        struct fw_bl_desc *bl_desc = (void *)bl->data + bin_hdr->header_offset;
-        void *bl_data = (void *)bl->data + bin_hdr->data_offset;
-        u32 pos = 0;
-        void *image;
-
-        desc->bootloader_start_offset = pos;
-        desc->bootloader_size = ALIGN(bl_desc->code_size, sizeof(u32));
-        desc->bootloader_imem_offset = bl_desc->start_tag * 256;
-        desc->bootloader_entry_point = bl_desc->start_tag * 256;
-
-        pos = ALIGN(pos + desc->bootloader_size, BL_DESC_BLK_SIZE);
-        desc->app_start_offset = pos;
-        desc->app_size = ALIGN(code->size, BL_DESC_BLK_SIZE) +
-                         ALIGN(data->size, BL_DESC_BLK_SIZE);
-        desc->app_imem_offset = 0;
-        desc->app_imem_entry = 0;
-        desc->app_dmem_offset = 0;
-        desc->app_resident_code_offset = 0;
-        desc->app_resident_code_size = ALIGN(code->size, BL_DESC_BLK_SIZE);
-
-        pos = ALIGN(pos + desc->app_resident_code_size, BL_DESC_BLK_SIZE);
-        desc->app_resident_data_offset = pos - desc->app_start_offset;
-        desc->app_resident_data_size = ALIGN(data->size, BL_DESC_BLK_SIZE);
-
-        desc->image_size = ALIGN(bl_desc->code_size, BL_DESC_BLK_SIZE) +
-                           desc->app_size;
-
-        image = kzalloc(desc->image_size, GFP_KERNEL);
-        if (!image)
-                return ERR_PTR(-ENOMEM);
-
-        memcpy(image + desc->bootloader_start_offset, bl_data,
-               bl_desc->code_size);
-        memcpy(image + desc->app_start_offset, code->data, code->size);
-        memcpy(image + desc->app_start_offset + desc->app_resident_data_offset,
-               data->data, data->size);
-
-        return image;
-}
-
-/**
- * ls_ucode_img_load_generic() - load and prepare a LS ucode image
- *
- * Load the LS microcode, bootloader and signature and pack them into a single
- * blob. Also generate the corresponding ucode descriptor.
- */
-static int
-ls_ucode_img_load_generic(struct nvkm_subdev *subdev,
-                          struct ls_ucode_img *img, const char *falcon_name,
-                          const u32 falcon_id)
-{
-        const struct firmware *bl, *code, *data;
-        struct lsf_ucode_desc *lsf_desc;
-        char f[64];
-        int ret;
-
-        img->ucode_header = NULL;
-
-        snprintf(f, sizeof(f), "gr/%s_bl", falcon_name);
-        ret = nvkm_firmware_get(subdev->device, f, &bl);
-        if (ret)
-                goto error;
-
-        snprintf(f, sizeof(f), "gr/%s_inst", falcon_name);
-        ret = nvkm_firmware_get(subdev->device, f, &code);
-        if (ret)
-                goto free_bl;
-
-        snprintf(f, sizeof(f), "gr/%s_data", falcon_name);
-        ret = nvkm_firmware_get(subdev->device, f, &data);
-        if (ret)
-                goto free_inst;
-
-        img->ucode_data = ls_ucode_img_build(bl, code, data,
-                                             &img->ucode_desc);
-        if (IS_ERR(img->ucode_data)) {
-                ret = PTR_ERR(img->ucode_data);
-                goto free_data;
-        }
-        img->ucode_size = img->ucode_desc.image_size;
-
-        snprintf(f, sizeof(f), "gr/%s_sig", falcon_name);
-        lsf_desc = gm200_secboot_load_firmware(subdev, f, sizeof(*lsf_desc));
-        if (IS_ERR(lsf_desc)) {
-                ret = PTR_ERR(lsf_desc);
-                goto free_image;
-        }
-        /* not needed? the signature should already have the right value */
-        lsf_desc->falcon_id = falcon_id;
-        memcpy(&img->lsb_header.signature, lsf_desc, sizeof(*lsf_desc));
-        img->falcon_id = lsf_desc->falcon_id;
-        kfree(lsf_desc);
-
-        /* success path - only free requested firmware files */
-        goto free_data;
-
-free_image:
-        kfree(img->ucode_data);
-free_data:
-        nvkm_firmware_put(data);
-free_inst:
-        nvkm_firmware_put(code);
-free_bl:
-        nvkm_firmware_put(bl);
-error:
-        return ret;
-}
-
-typedef int (*lsf_load_func)(struct nvkm_subdev *, struct ls_ucode_img *);
-
-static int
-ls_ucode_img_load_fecs(struct nvkm_subdev *subdev, struct ls_ucode_img *img)
-{
-        return ls_ucode_img_load_generic(subdev, img, "fecs",
-                                         NVKM_SECBOOT_FALCON_FECS);
-}
-
-static int
-ls_ucode_img_load_gpccs(struct nvkm_subdev *subdev, struct ls_ucode_img *img)
-{
-        return ls_ucode_img_load_generic(subdev, img, "gpccs",
-                                         NVKM_SECBOOT_FALCON_GPCCS);
-}
-
-/**
- * ls_ucode_img_load() - create a lsf_ucode_img and load it
- */
-static struct ls_ucode_img *
-ls_ucode_img_load(struct nvkm_subdev *subdev, lsf_load_func load_func)
-{
-        struct ls_ucode_img *img;
-        int ret;
-
-        img = kzalloc(sizeof(*img), GFP_KERNEL);
-        if (!img)
-                return ERR_PTR(-ENOMEM);
-
-        ret = load_func(subdev, img);
-        if (ret) {
-                kfree(img);
-                return ERR_PTR(ret);
-        }
-
-        return img;
-}
-
-static const lsf_load_func lsf_load_funcs[] = {
-        [NVKM_SECBOOT_FALCON_END] = NULL, /* reserve enough space */
-        [NVKM_SECBOOT_FALCON_FECS] = ls_ucode_img_load_fecs,
-        [NVKM_SECBOOT_FALCON_GPCCS] = ls_ucode_img_load_gpccs,
-};
-
-/**
- * ls_ucode_img_populate_bl_desc() - populate a DMEM BL descriptor for LS image
- * @img:        ucode image to generate against
- * @desc:       descriptor to populate
- * @sb:         secure boot state to use for base addresses
- *
- * Populate the DMEM BL descriptor with the information contained in a
- * ls_ucode_desc.
- *
- */
-static void
-ls_ucode_img_populate_bl_desc(struct ls_ucode_img *img, u64 wpr_addr,
-                              struct gm200_flcn_bl_desc *desc)
-{
-        struct ls_ucode_img_desc *pdesc = &img->ucode_desc;
-        u64 addr_base;
-
-        addr_base = wpr_addr + img->lsb_header.ucode_off +
-                    pdesc->app_start_offset;
-
-        memset(desc, 0, sizeof(*desc));
-        desc->ctx_dma = FALCON_DMAIDX_UCODE;
-        desc->code_dma_base.lo = lower_32_bits(
-                (addr_base + pdesc->app_resident_code_offset));
-        desc->code_dma_base.hi = upper_32_bits(
-                (addr_base + pdesc->app_resident_code_offset));
-        desc->non_sec_code_size = pdesc->app_resident_code_size;
-        desc->data_dma_base.lo = lower_32_bits(
-                (addr_base + pdesc->app_resident_data_offset));
-        desc->data_dma_base.hi = upper_32_bits(
-                (addr_base + pdesc->app_resident_data_offset));
-        desc->data_size = pdesc->app_resident_data_size;
-        desc->code_entry_point = pdesc->app_imem_entry;
-}
-
-#define LSF_LSB_HEADER_ALIGN 256
-#define LSF_BL_DATA_ALIGN 256
-#define LSF_BL_DATA_SIZE_ALIGN 256
-#define LSF_BL_CODE_SIZE_ALIGN 256
-#define LSF_UCODE_DATA_ALIGN 4096
-
-/**
- * ls_ucode_img_fill_headers - fill the WPR and LSB headers of an image
- * @gsb:        secure boot device used
- * @img:        image to generate for
- * @offset:     offset in the WPR region where this image starts
- *
- * Allocate space in the WPR area from offset and write the WPR and LSB headers
- * accordingly.
- *
- * Return: offset at the end of this image.
- */
-static u32
-ls_ucode_img_fill_headers(struct gm200_secboot *gsb, struct ls_ucode_img *img,
-                          u32 offset)
-{
-        struct lsf_wpr_header *whdr = &img->wpr_header;
-        struct lsf_lsb_header *lhdr = &img->lsb_header;
-        struct ls_ucode_img_desc *desc = &img->ucode_desc;
-
-        if (img->ucode_header) {
-                nvkm_fatal(&gsb->base.subdev,
-                            "images withough loader are not supported yet!\n");
-                return offset;
-        }
-
-        /* Fill WPR header */
-        whdr->falcon_id = img->falcon_id;
-        whdr->bootstrap_owner = gsb->base.func->boot_falcon;
-        whdr->status = LSF_IMAGE_STATUS_COPY;
-
-        /* Align, save off, and include an LSB header size */
-        offset = ALIGN(offset, LSF_LSB_HEADER_ALIGN);
-        whdr->lsb_offset = offset;
-        offset += sizeof(struct lsf_lsb_header);
-
-        /*
-         * Align, save off, and include the original (static) ucode
-         * image size
-         */
-        offset = ALIGN(offset, LSF_UCODE_DATA_ALIGN);
-        lhdr->ucode_off = offset;
-        offset += img->ucode_size;
-
-        /*
-         * For falcons that use a boot loader (BL), we append a loader
-         * desc structure on the end of the ucode image and consider
-         * this the boot loader data. The host will then copy the loader
-         * desc args to this space within the WPR region (before locking
-         * down) and the HS bin will then copy them to DMEM 0 for the
-         * loader.
-         */
-        lhdr->bl_code_size = ALIGN(desc->bootloader_size,
-                                   LSF_BL_CODE_SIZE_ALIGN);
-        lhdr->ucode_size = ALIGN(desc->app_resident_data_offset,
-                                 LSF_BL_CODE_SIZE_ALIGN) + lhdr->bl_code_size;
-        lhdr->data_size = ALIGN(desc->app_size, LSF_BL_CODE_SIZE_ALIGN) +
-                                lhdr->bl_code_size - lhdr->ucode_size;
-        /*
-         * Though the BL is located at 0th offset of the image, the VA
-         * is different to make sure that it doesn't collide the actual
-         * OS VA range
-         */
-        lhdr->bl_imem_off = desc->bootloader_imem_offset;
-        lhdr->app_code_off = desc->app_start_offset +
-                             desc->app_resident_code_offset;
-        lhdr->app_code_size = desc->app_resident_code_size;
-        lhdr->app_data_off = desc->app_start_offset +
-                             desc->app_resident_data_offset;
-        lhdr->app_data_size = desc->app_resident_data_size;
-
-        lhdr->flags = 0;
-        if (img->falcon_id == gsb->base.func->boot_falcon)
-                lhdr->flags = LSF_FLAG_DMACTL_REQ_CTX;
-
-        /* GPCCS will be loaded using PRI */
-        if (img->falcon_id == NVKM_SECBOOT_FALCON_GPCCS)
-                lhdr->flags |= LSF_FLAG_FORCE_PRIV_LOAD;
-
-        /* Align (size bloat) and save off BL descriptor size */
-        lhdr->bl_data_size = ALIGN(sizeof(struct gm200_flcn_bl_desc),
-                                   LSF_BL_DATA_SIZE_ALIGN);
-        /*
-         * Align, save off, and include the additional BL data
-         */
-        offset = ALIGN(offset, LSF_BL_DATA_ALIGN);
-        lhdr->bl_data_off = offset;
-        offset += lhdr->bl_data_size;
-
-        return offset;
-}
-
-static void
-ls_ucode_mgr_init(struct ls_ucode_mgr *mgr)
-{
-        memset(mgr, 0, sizeof(*mgr));
-        INIT_LIST_HEAD(&mgr->img_list);
-}
-
-static void
-ls_ucode_mgr_cleanup(struct ls_ucode_mgr *mgr)
-{
-        struct ls_ucode_img *img, *t;
-
-        list_for_each_entry_safe(img, t, &mgr->img_list, node) {
-                kfree(img->ucode_data);
-                kfree(img->ucode_header);
-                kfree(img);
-        }
-}
-
-static void
-ls_ucode_mgr_add_img(struct ls_ucode_mgr *mgr, struct ls_ucode_img *img)
-{
-        mgr->count++;
-        list_add_tail(&img->node, &mgr->img_list);
-}
-
-/**
- * ls_ucode_mgr_fill_headers - fill WPR and LSB headers of all managed images
- */
-static void
-ls_ucode_mgr_fill_headers(struct gm200_secboot *gsb, struct ls_ucode_mgr *mgr)
-{
-        struct ls_ucode_img *img;
-        u32 offset;
-
-        /*
-         * Start with an array of WPR headers at the base of the WPR.
-         * The expectation here is that the secure falcon will do a single DMA
-         * read of this array and cache it internally so it's ok to pack these.
-         * Also, we add 1 to the falcon count to indicate the end of the array.
-         */
-        offset = sizeof(struct lsf_wpr_header) * (mgr->count + 1);
-
-        /*
-         * Walk the managed falcons, accounting for the LSB structs
-         * as well as the ucode images.
-         */
-        list_for_each_entry(img, &mgr->img_list, node) {
-                offset = ls_ucode_img_fill_headers(gsb, img, offset);
-        }
-
-        mgr->wpr_size = offset;
-}
-
-/**
- * ls_ucode_mgr_write_wpr - write the WPR blob contents
- */
-static int
-ls_ucode_mgr_write_wpr(struct gm200_secboot *gsb, struct ls_ucode_mgr *mgr,
-                       struct nvkm_gpuobj *wpr_blob)
-{
-        struct ls_ucode_img *img;
-        u32 pos = 0;
-
-        nvkm_kmap(wpr_blob);
-
-        list_for_each_entry(img, &mgr->img_list, node) {
-                nvkm_gpuobj_memcpy_to(wpr_blob, pos, &img->wpr_header,
-                                      sizeof(img->wpr_header));
-
-                nvkm_gpuobj_memcpy_to(wpr_blob, img->wpr_header.lsb_offset,
-                                     &img->lsb_header, sizeof(img->lsb_header));
-
-                /* Generate and write BL descriptor */
-                if (!img->ucode_header) {
-                        u8 desc[gsb->func->bl_desc_size];
-                        struct gm200_flcn_bl_desc gdesc;
-
-                        ls_ucode_img_populate_bl_desc(img, gsb->wpr_addr,
-                                                      &gdesc);
-                        gsb->func->fixup_bl_desc(&gdesc, &desc);
-                        nvkm_gpuobj_memcpy_to(wpr_blob,
-                                              img->lsb_header.bl_data_off,
-                                              &desc, gsb->func->bl_desc_size);
-                }
-
-                /* Copy ucode */
-                nvkm_gpuobj_memcpy_to(wpr_blob, img->lsb_header.ucode_off,
-                                      img->ucode_data, img->ucode_size);
-
-                pos += sizeof(img->wpr_header);
-        }
-
-        nvkm_wo32(wpr_blob, pos, NVKM_SECBOOT_FALCON_INVALID);
-
-        nvkm_done(wpr_blob);
-
-        return 0;
-}
-
-/* Both size and address of WPR need to be 128K-aligned */
-#define WPR_ALIGNMENT   0x20000
-/**
- * gm200_secboot_prepare_ls_blob() - prepare the LS blob
- *
- * For each securely managed falcon, load the FW, signatures and bootloaders and
- * prepare a ucode blob. Then, compute the offsets in the WPR region for each
- * blob, and finally write the headers and ucode blobs into a GPU object that
- * will be copied into the WPR region by the HS firmware.
- */
-static int
-gm200_secboot_prepare_ls_blob(struct gm200_secboot *gsb)
-{
-        struct nvkm_secboot *sb = &gsb->base;
-        struct nvkm_device *device = sb->subdev.device;
-        struct ls_ucode_mgr mgr;
-        int falcon_id;
-        int ret;
-
-        ls_ucode_mgr_init(&mgr);
-
-        /* Load all LS blobs */
-        for_each_set_bit(falcon_id, &gsb->base.func->managed_falcons,
-                         NVKM_SECBOOT_FALCON_END) {
-                struct ls_ucode_img *img;
-
-                img = ls_ucode_img_load(&sb->subdev, lsf_load_funcs[falcon_id]);
-
-                if (IS_ERR(img)) {
-                        ret = PTR_ERR(img);
-                        goto cleanup;
-                }
-                ls_ucode_mgr_add_img(&mgr, img);
-        }
-
-        /*
-         * Fill the WPR and LSF headers with the right offsets and compute
-         * required WPR size
-         */
-        ls_ucode_mgr_fill_headers(gsb, &mgr);
-        mgr.wpr_size = ALIGN(mgr.wpr_size, WPR_ALIGNMENT);
-
-        /* Allocate GPU object that will contain the WPR region */
-        ret = nvkm_gpuobj_new(device, mgr.wpr_size, WPR_ALIGNMENT, false, NULL,
-                              &gsb->ls_blob);
-        if (ret)
-                goto cleanup;
-
-        nvkm_debug(&sb->subdev, "%d managed LS falcons, WPR size is %d bytes\n",
-                    mgr.count, mgr.wpr_size);
-
-        /* If WPR address and size are not fixed, set them to fit the LS blob */
-        if (!gsb->wpr_size) {
-                gsb->wpr_addr = gsb->ls_blob->addr;
-                gsb->wpr_size = gsb->ls_blob->size;
-        }
-
-        /* Write LS blob */
-        ret = ls_ucode_mgr_write_wpr(gsb, &mgr, gsb->ls_blob);
-        if (ret)
-                nvkm_gpuobj_del(&gsb->ls_blob);
-
-cleanup:
-        ls_ucode_mgr_cleanup(&mgr);
-
-        return ret;
-}
-
-/*
- * High-secure blob creation
- */
-
-/**
- * gm200_secboot_hsf_patch_signature() - patch HS blob with correct signature
- */
-static void
-gm200_secboot_hsf_patch_signature(struct gm200_secboot *gsb, void *acr_image)
-{
-        struct nvkm_secboot *sb = &gsb->base;
-        struct fw_bin_header *hsbin_hdr = acr_image;
-        struct hsf_fw_header *fw_hdr = acr_image + hsbin_hdr->header_offset;
-        void *hs_data = acr_image + hsbin_hdr->data_offset;
-        void *sig;
-        u32 sig_size;
-
-        /* Falcon in debug or production mode? */
-        if ((nvkm_rd32(sb->subdev.device, sb->base + 0xc08) >> 20) & 0x1) {
-                sig = acr_image + fw_hdr->sig_dbg_offset;
-                sig_size = fw_hdr->sig_dbg_size;
-        } else {
-                sig = acr_image + fw_hdr->sig_prod_offset;
-                sig_size = fw_hdr->sig_prod_size;
-        }
-
-        /* Patch signature */
-        memcpy(hs_data + fw_hdr->patch_loc, sig + fw_hdr->patch_sig, sig_size);
-}
-
-/**
- * gm200_secboot_populate_hsf_bl_desc() - populate BL descriptor for HS image
- */
-static void
-gm200_secboot_populate_hsf_bl_desc(void *acr_image,
-                                   struct gm200_flcn_bl_desc *bl_desc)
-{
-        struct fw_bin_header *hsbin_hdr = acr_image;
-        struct hsf_fw_header *fw_hdr = acr_image + hsbin_hdr->header_offset;
-        struct hsf_load_header *load_hdr = acr_image + fw_hdr->hdr_offset;
-
-        /*
-         * Descriptor for the bootloader that will load the ACR image into
-         * IMEM/DMEM memory.
-         */
-        fw_hdr = acr_image + hsbin_hdr->header_offset;
-        load_hdr = acr_image + fw_hdr->hdr_offset;
-        memset(bl_desc, 0, sizeof(*bl_desc));
-        bl_desc->ctx_dma = FALCON_DMAIDX_VIRT;
-        bl_desc->non_sec_code_off = load_hdr->non_sec_code_off;
-        bl_desc->non_sec_code_size = load_hdr->non_sec_code_size;
-        bl_desc->sec_code_off = load_hdr->app[0].sec_code_off;
-        bl_desc->sec_code_size = load_hdr->app[0].sec_code_size;
-        bl_desc->code_entry_point = 0;
-        /*
-         * We need to set code_dma_base to the virtual address of the acr_blob,
-         * and add this address to data_dma_base before writing it into DMEM
-         */
-        bl_desc->code_dma_base.lo = 0;
-        bl_desc->data_dma_base.lo = load_hdr->data_dma_base;
-        bl_desc->data_size = load_hdr->data_size;
-}
-
-/**
- * gm200_secboot_prepare_hs_blob - load and prepare a HS blob and BL descriptor
- *
- * @gsb secure boot instance to prepare for
- * @fw name of the HS firmware to load
- * @blob pointer to gpuobj that will be allocated to receive the HS FW payload
- * @bl_desc pointer to the BL descriptor to write for this firmware
- * @patch whether we should patch the HS descriptor (only for HS loaders)
- */
-static int
-gm200_secboot_prepare_hs_blob(struct gm200_secboot *gsb, const char *fw,
-                              struct nvkm_gpuobj **blob,
-                              struct gm200_flcn_bl_desc *bl_desc, bool patch)
-{
-        struct nvkm_subdev *subdev = &gsb->base.subdev;
-        void *acr_image;
-        struct fw_bin_header *hsbin_hdr;
-        struct hsf_fw_header *fw_hdr;
-        void *acr_data;
-        struct hsf_load_header *load_hdr;
-        struct hsflcn_acr_desc *desc;
-        int ret;
-
-        acr_image = gm200_secboot_load_firmware(subdev, fw, 0);
-        if (IS_ERR(acr_image))
-                return PTR_ERR(acr_image);
-        hsbin_hdr = acr_image;
-
-        /* Patch signature */
-        gm200_secboot_hsf_patch_signature(gsb, acr_image);
-
-        acr_data = acr_image + hsbin_hdr->data_offset;
-
-        /* Patch descriptor? */
-        if (patch) {
-                fw_hdr = acr_image + hsbin_hdr->header_offset;
-                load_hdr = acr_image + fw_hdr->hdr_offset;
-                desc = acr_data + load_hdr->data_dma_base;
-                gsb->func->fixup_hs_desc(gsb, desc);
-        }
-
-        /* Generate HS BL descriptor */
-        gm200_secboot_populate_hsf_bl_desc(acr_image, bl_desc);
-
-        /* Create ACR blob and copy HS data to it */
-        ret = nvkm_gpuobj_new(subdev->device, ALIGN(hsbin_hdr->data_size, 256),
-                              0x1000, false, NULL, blob);
-        if (ret)
-                goto cleanup;
-
-        nvkm_kmap(*blob);
-        nvkm_gpuobj_memcpy_to(*blob, 0, acr_data, hsbin_hdr->data_size);
-        nvkm_done(*blob);
-
-cleanup:
-        kfree(acr_image);
-
-        return ret;
-}
-
-/*
- * High-secure bootloader blob creation
- */
-
-static int
-gm200_secboot_prepare_hsbl_blob(struct gm200_secboot *gsb)
-{
-        struct nvkm_subdev *subdev = &gsb->base.subdev;
-
-        gsb->hsbl_blob = gm200_secboot_load_firmware(subdev, "acr/bl", 0);
-        if (IS_ERR(gsb->hsbl_blob)) {
-                int ret = PTR_ERR(gsb->hsbl_blob);
-
-                gsb->hsbl_blob = NULL;
-                return ret;
-        }
-
-        return 0;
-}
+#include <engine/falcon.h>
+#include <subdev/mc.h>
 
 /**
- * gm20x_secboot_prepare_blobs - load blobs common to all GM20X GPUs.
+ * gm200_secboot_run_blob() - run the given high-secure blob
  *
- * This includes the LS blob, HS ucode loading blob, and HS bootloader.
- *
- * The HS ucode unload blob is only used on dGPU.
  */
 int
-gm20x_secboot_prepare_blobs(struct gm200_secboot *gsb)
-{
-        int ret;
-
-        /* Load and prepare the managed falcon's firmwares */
-        if (!gsb->ls_blob) {
-                ret = gm200_secboot_prepare_ls_blob(gsb);
-                if (ret)
-                        return ret;
-        }
-
-        /* Load the HS firmware that will load the LS firmwares */
-        if (!gsb->acr_load_blob) {
-                ret = gm200_secboot_prepare_hs_blob(gsb, "acr/ucode_load",
-                                                &gsb->acr_load_blob,
-                                                &gsb->acr_load_bl_desc, true);
-                if (ret)
-                        return ret;
-        }
-
-        /* Load the HS firmware bootloader */
-        if (!gsb->hsbl_blob) {
-                ret = gm200_secboot_prepare_hsbl_blob(gsb);
-                if (ret)
-                        return ret;
-        }
-
-        return 0;
-}
-
-static int
-gm200_secboot_prepare_blobs(struct gm200_secboot *gsb)
-{
-        int ret;
-
-        ret = gm20x_secboot_prepare_blobs(gsb);
-        if (ret)
-                return ret;
-
-        /* dGPU only: load the HS firmware that unprotects the WPR region */
-        if (!gsb->acr_unload_blob) {
-                ret = gm200_secboot_prepare_hs_blob(gsb, "acr/ucode_unload",
-                                               &gsb->acr_unload_blob,
-                                               &gsb->acr_unload_bl_desc, false);
-                if (ret)
-                        return ret;
-        }
-
-        return 0;
-}
-
-static int
-gm200_secboot_blobs_ready(struct gm200_secboot *gsb)
+gm200_secboot_run_blob(struct nvkm_secboot *sb, struct nvkm_gpuobj *blob)
 {
+        struct gm200_secboot *gsb = gm200_secboot(sb);
         struct nvkm_subdev *subdev = &gsb->base.subdev;
+        struct nvkm_falcon *falcon = gsb->base.boot_falcon;
+        struct nvkm_vma vma;
         int ret;
 
-        /* firmware already loaded, nothing to do... */
-        if (gsb->firmware_ok)
-                return 0;
-
-        ret = gsb->func->prepare_blobs(gsb);
-        if (ret) {
-                nvkm_error(subdev, "failed to load secure firmware\n");
-                return ret;
-        }
-
-        gsb->firmware_ok = true;
-
-        return 0;
-}
-
-
-/*
- * Secure Boot Execution
- */
-
-/**
- * gm200_secboot_load_hs_bl() - load HS bootloader into DMEM and IMEM
- */
-static void
-gm200_secboot_load_hs_bl(struct gm200_secboot *gsb, void *data, u32 data_size)
-{
-        struct nvkm_device *device = gsb->base.subdev.device;
-        struct fw_bin_header *hdr = gsb->hsbl_blob;
-        struct fw_bl_desc *hsbl_desc = gsb->hsbl_blob + hdr->header_offset;
-        void *blob_data = gsb->hsbl_blob + hdr->data_offset;
-        void *hsbl_code = blob_data + hsbl_desc->code_off;
-        void *hsbl_data = blob_data + hsbl_desc->data_off;
-        u32 code_size = ALIGN(hsbl_desc->code_size, 256);
-        const u32 base = gsb->base.base;
-        u32 blk;
-        u32 tag;
-        int i;
-
-        /*
-         * Copy HS bootloader data
-         */
-        nvkm_wr32(device, base + 0x1c0, (0x00000000 | (0x1 << 24)));
-        for (i = 0; i < hsbl_desc->data_size / 4; i++)
-                nvkm_wr32(device, base + 0x1c4, ((u32 *)hsbl_data)[i]);
-
-        /*
-         * Copy HS bootloader interface structure where the HS descriptor
-         * expects it to be
-         */
-        nvkm_wr32(device, base + 0x1c0,
-                  (hsbl_desc->dmem_load_off | (0x1 << 24)));
-        for (i = 0; i < data_size / 4; i++)
-                nvkm_wr32(device, base + 0x1c4, ((u32 *)data)[i]);
-
-        /* Copy HS bootloader code to end of IMEM */
-        blk = (nvkm_rd32(device, base + 0x108) & 0x1ff) - (code_size >> 8);
-        tag = hsbl_desc->start_tag;
-        nvkm_wr32(device, base + 0x180, ((blk & 0xff) << 8) | (0x1 << 24));
-        for (i = 0; i < code_size / 4; i++) {
-                /* write new tag every 256B */
-                if ((i & 0x3f) == 0) {
-                        nvkm_wr32(device, base + 0x188, tag & 0xffff);
-                        tag++;
-                }
-                nvkm_wr32(device, base + 0x184, ((u32 *)hsbl_code)[i]);
-        }
-        nvkm_wr32(device, base + 0x188, 0);
-}
-
-/**
- * gm200_secboot_setup_falcon() - set up the secure falcon for secure boot
- */
-static int
-gm200_secboot_setup_falcon(struct gm200_secboot *gsb)
-{
-        struct nvkm_device *device = gsb->base.subdev.device;
-        struct fw_bin_header *hdr = gsb->hsbl_blob;
-        struct fw_bl_desc *hsbl_desc = gsb->hsbl_blob + hdr->header_offset;
-        /* virtual start address for boot vector */
-        u32 virt_addr = hsbl_desc->start_tag << 8;
-        const u32 base = gsb->base.base;
-        const u32 reg_base = base + 0xe00;
-        u32 inst_loc;
-        int ret;
-
-        ret = nvkm_secboot_falcon_reset(&gsb->base);
+        ret = nvkm_falcon_get(falcon, subdev);
         if (ret)
                 return ret;
 
-        /* setup apertures - virtual */
-        nvkm_wr32(device, reg_base + 4 * (FALCON_DMAIDX_UCODE), 0x4);
-        nvkm_wr32(device, reg_base + 4 * (FALCON_DMAIDX_VIRT), 0x0);
-        /* setup apertures - physical */
-        nvkm_wr32(device, reg_base + 4 * (FALCON_DMAIDX_PHYS_VID), 0x4);
-        nvkm_wr32(device, reg_base + 4 * (FALCON_DMAIDX_PHYS_SYS_COH),
-                  0x4 | 0x1);
-        nvkm_wr32(device, reg_base + 4 * (FALCON_DMAIDX_PHYS_SYS_NCOH),
-                  0x4 | 0x2);
-
-        /* Set context */
-        if (nvkm_memory_target(gsb->inst->memory) == NVKM_MEM_TARGET_VRAM)
-                inst_loc = 0x0; /* FB */
-        else
-                inst_loc = 0x3; /* Non-coherent sysmem */
-
-        nvkm_mask(device, base + 0x048, 0x1, 0x1);
-        nvkm_wr32(device, base + 0x480,
-                  ((gsb->inst->addr >> 12) & 0xfffffff) |
-                  (inst_loc << 28) | (1 << 30));
-
-        /* Set boot vector to code's starting virtual address */
-        nvkm_wr32(device, base + 0x104, virt_addr);
-
-        return 0;
-}
-
-/**
- * gm200_secboot_run_hs_blob() - run the given high-secure blob
- */
-static int
-gm200_secboot_run_hs_blob(struct gm200_secboot *gsb, struct nvkm_gpuobj *blob,
-                          struct gm200_flcn_bl_desc *desc)
-{
-        struct nvkm_vma vma;
-        u64 vma_addr;
-        const u32 bl_desc_size = gsb->func->bl_desc_size;
-        u8 bl_desc[bl_desc_size];
-        int ret;
-
         /* Map the HS firmware so the HS bootloader can see it */
         ret = nvkm_gpuobj_map(blob, gsb->vm, NV_MEM_ACCESS_RW, &vma);
-        if (ret)
+        if (ret) {
+                nvkm_falcon_put(falcon, subdev);
                 return ret;
+        }
 
-        /* Add the mapping address to the DMA bases */
-        vma_addr = flcn64_to_u64(desc->code_dma_base) + vma.offset;
-        desc->code_dma_base.lo = lower_32_bits(vma_addr);
-        desc->code_dma_base.hi = upper_32_bits(vma_addr);
-        vma_addr = flcn64_to_u64(desc->data_dma_base) + vma.offset;
-        desc->data_dma_base.lo = lower_32_bits(vma_addr);
-        desc->data_dma_base.hi = upper_32_bits(vma_addr);
-
-        /* Fixup the BL header */
-        gsb->func->fixup_bl_desc(desc, &bl_desc);
-
-        /* Reset the falcon and make it ready to run the HS bootloader */
-        ret = gm200_secboot_setup_falcon(gsb);
+        /* Reset and set the falcon up */
+        ret = nvkm_falcon_reset(falcon);
         if (ret)
-                goto done;
+                goto end;
+        nvkm_falcon_bind_context(falcon, gsb->inst);
 
         /* Load the HS bootloader into the falcon's IMEM/DMEM */
-        gm200_secboot_load_hs_bl(gsb, &bl_desc, bl_desc_size);
-
-        /* Start the HS bootloader */
-        ret = nvkm_secboot_falcon_run(&gsb->base);
+        ret = sb->acr->func->load(sb->acr, &gsb->base, blob, vma.offset);
         if (ret)
-                goto done;
-
-done:
-        /* Restore the original DMA addresses */
-        vma_addr = flcn64_to_u64(desc->code_dma_base) - vma.offset;
-        desc->code_dma_base.lo = lower_32_bits(vma_addr);
-        desc->code_dma_base.hi = upper_32_bits(vma_addr);
-        vma_addr = flcn64_to_u64(desc->data_dma_base) - vma.offset;
-        desc->data_dma_base.lo = lower_32_bits(vma_addr);
-        desc->data_dma_base.hi = upper_32_bits(vma_addr);
-
-        /* We don't need the ACR firmware anymore */
-        nvkm_gpuobj_unmap(&vma);
+                goto end;
 
-        return ret;
-}
+        /* Disable interrupts as we will poll for the HALT bit */
+        nvkm_mc_intr_mask(sb->subdev.device, falcon->owner->index, false);
 
-/*
- * gm200_secboot_reset() - execute secure boot from the prepared state
- *
- * Load the HS bootloader and ask the falcon to run it. This will in turn
- * load the HS firmware and run it, so once the falcon stops all the managed
- * falcons should have their LS firmware loaded and be ready to run.
- */
-int
-gm200_secboot_reset(struct nvkm_secboot *sb, enum nvkm_secboot_falcon falcon)
-{
-        struct gm200_secboot *gsb = gm200_secboot(sb);
-        int ret;
+        /* Set default error value in mailbox register */
+        nvkm_falcon_wr32(falcon, 0x040, 0xdeada5a5);
 
-        /* Make sure all blobs are ready */
-        ret = gm200_secboot_blobs_ready(gsb);
+        /* Start the HS bootloader */
+        nvkm_falcon_set_start_addr(falcon, sb->acr->start_address);
+        nvkm_falcon_start(falcon);
+        ret = nvkm_falcon_wait_for_halt(falcon, 100);
         if (ret)
-                return ret;
-
-        /*
-         * Dummy GM200 implementation: perform secure boot each time we are
-         * called on FECS. Since only FECS and GPCCS are managed and started
-         * together, this ought to be safe.
-         *
-         * Once we have proper PMU firmware and support, this will be changed
-         * to a proper call to the PMU method.
-         */
-        if (falcon != NVKM_SECBOOT_FALCON_FECS)
                 goto end;
 
-        /* If WPR is set and we have an unload blob, run it to unlock WPR */
-        if (gsb->acr_unload_blob &&
-            gsb->falcon_state[NVKM_SECBOOT_FALCON_FECS] != NON_SECURE) {
-                ret = gm200_secboot_run_hs_blob(gsb, gsb->acr_unload_blob,
-                                                &gsb->acr_unload_bl_desc);
-                if (ret)
-                        return ret;
+        /* If mailbox register contains an error code, then ACR has failed */
+        ret = nvkm_falcon_rd32(falcon, 0x040);
+        if (ret) {
+                nvkm_error(subdev, "ACR boot failed, ret 0x%08x", ret);
+                ret = -EINVAL;
+                goto end;
         }
 
-        /* Reload all managed falcons */
-        ret = gm200_secboot_run_hs_blob(gsb, gsb->acr_load_blob,
-                                        &gsb->acr_load_bl_desc);
-        if (ret)
-                return ret;
-
 end:
-        gsb->falcon_state[falcon] = RESET;
-        return 0;
-}
+        /* Reenable interrupts */
+        nvkm_mc_intr_mask(sb->subdev.device, falcon->owner->index, true);
 
-int
-gm200_secboot_start(struct nvkm_secboot *sb, enum nvkm_secboot_falcon falcon)
-{
-        struct gm200_secboot *gsb = gm200_secboot(sb);
-        int base;
-
-        switch (falcon) {
-        case NVKM_SECBOOT_FALCON_FECS:
-                base = 0x409000;
-                break;
-        case NVKM_SECBOOT_FALCON_GPCCS:
-                base = 0x41a000;
-                break;
-        default:
-                nvkm_error(&sb->subdev, "cannot start unhandled falcon!\n");
-                return -EINVAL;
-        }
-
-        nvkm_wr32(sb->subdev.device, base + 0x130, 0x00000002);
-        gsb->falcon_state[falcon] = RUNNING;
+        /* We don't need the ACR firmware anymore */
+        nvkm_gpuobj_unmap(&vma);
+        nvkm_falcon_put(falcon, subdev);
 
-        return 0;
+        return ret;
 }
 
-
-
 int
-gm200_secboot_init(struct nvkm_secboot *sb)
+gm200_secboot_oneinit(struct nvkm_secboot *sb)
 {
         struct gm200_secboot *gsb = gm200_secboot(sb);
         struct nvkm_device *device = sb->subdev.device;
@@ -1361,24 +132,22 @@ gm200_secboot_init(struct nvkm_secboot *sb)
         nvkm_wo32(gsb->inst, 0x20c, upper_32_bits(vm_area_len - 1));
         nvkm_done(gsb->inst);
 
+        if (sb->acr->func->oneinit) {
+                ret = sb->acr->func->oneinit(sb->acr, sb);
+                if (ret)
+                        return ret;
+        }
+
         return 0;
 }
 
-static int
+int
 gm200_secboot_fini(struct nvkm_secboot *sb, bool suspend)
 {
-        struct gm200_secboot *gsb = gm200_secboot(sb);
         int ret = 0;
-        int i;
 
-        /* Run the unload blob to unprotect the WPR region */
-        if (gsb->acr_unload_blob &&
-            gsb->falcon_state[NVKM_SECBOOT_FALCON_FECS] != NON_SECURE)
-                ret = gm200_secboot_run_hs_blob(gsb, gsb->acr_unload_blob,
-                                                &gsb->acr_unload_bl_desc);
-
-        for (i = 0; i < NVKM_SECBOOT_FALCON_END; i++)
-                gsb->falcon_state[i] = NON_SECURE;
+        if (sb->acr->func->fini)
+                ret = sb->acr->func->fini(sb->acr, sb, suspend);
 
         return ret;
 }
@@ -1388,11 +157,7 @@ gm200_secboot_dtor(struct nvkm_secboot *sb)
 {
         struct gm200_secboot *gsb = gm200_secboot(sb);
 
-        nvkm_gpuobj_del(&gsb->acr_unload_blob);
-
-        kfree(gsb->hsbl_blob);
-        nvkm_gpuobj_del(&gsb->acr_load_blob);
-        nvkm_gpuobj_del(&gsb->ls_blob);
+        sb->acr->func->dtor(sb->acr);
 
         nvkm_vm_ref(NULL, &gsb->vm, gsb->pgd);
         nvkm_gpuobj_del(&gsb->pgd);
@@ -1405,50 +170,9 @@ gm200_secboot_dtor(struct nvkm_secboot *sb)
 static const struct nvkm_secboot_func
 gm200_secboot = {
         .dtor = gm200_secboot_dtor,
-        .init = gm200_secboot_init,
+        .oneinit = gm200_secboot_oneinit,
         .fini = gm200_secboot_fini,
-        .reset = gm200_secboot_reset,
-        .start = gm200_secboot_start,
-        .managed_falcons = BIT(NVKM_SECBOOT_FALCON_FECS) |
-                           BIT(NVKM_SECBOOT_FALCON_GPCCS),
-        .boot_falcon = NVKM_SECBOOT_FALCON_PMU,
-};
-
-/**
- * gm200_fixup_bl_desc - just copy the BL descriptor
- *
- * Use the GM200 descriptor format by default.
- */
-static void
-gm200_secboot_fixup_bl_desc(const struct gm200_flcn_bl_desc *desc, void *ret)
-{
-        memcpy(ret, desc, sizeof(*desc));
-}
-
-static void
-gm200_secboot_fixup_hs_desc(struct gm200_secboot *gsb,
-                            struct hsflcn_acr_desc *desc)
-{
-        desc->ucode_blob_base = gsb->ls_blob->addr;
-        desc->ucode_blob_size = gsb->ls_blob->size;
-
-        desc->wpr_offset = 0;
-
-        /* WPR region information for the HS binary to set up */
-        desc->wpr_region_id = 1;
-        desc->regions.no_regions = 1;
-        desc->regions.region_props[0].region_id = 1;
-        desc->regions.region_props[0].start_addr = gsb->wpr_addr >> 8;
-        desc->regions.region_props[0].end_addr =
-                (gsb->wpr_addr + gsb->wpr_size) >> 8;
-}
-
-static const struct gm200_secboot_func
-gm200_secboot_func = {
-        .bl_desc_size = sizeof(struct gm200_flcn_bl_desc),
-        .fixup_bl_desc = gm200_secboot_fixup_bl_desc,
-        .fixup_hs_desc = gm200_secboot_fixup_hs_desc,
-        .prepare_blobs = gm200_secboot_prepare_blobs,
+        .run_blob = gm200_secboot_run_blob,
 };
 
 int
@@ -1457,6 +181,12 @@ gm200_secboot_new(struct nvkm_device *device, int index,
 {
         int ret;
         struct gm200_secboot *gsb;
+        struct nvkm_acr *acr;
+
+        acr = acr_r361_new(BIT(NVKM_SECBOOT_FALCON_FECS) |
+                           BIT(NVKM_SECBOOT_FALCON_GPCCS));
+        if (IS_ERR(acr))
+                return PTR_ERR(acr);
 
         gsb = kzalloc(sizeof(*gsb), GFP_KERNEL);
         if (!gsb) {
@@ -1465,15 +195,14 @@ gm200_secboot_new(struct nvkm_device *device, int index,
         }
         *psb = &gsb->base;
 
-        ret = nvkm_secboot_ctor(&gm200_secboot, device, index, &gsb->base);
+        ret = nvkm_secboot_ctor(&gm200_secboot, acr, device, index, &gsb->base);
         if (ret)
                 return ret;
 
-        gsb->func = &gm200_secboot_func;
-
         return 0;
 }
 
+
 MODULE_FIRMWARE("nvidia/gm200/acr/bl.bin");
 MODULE_FIRMWARE("nvidia/gm200/acr/ucode_load.bin");
 MODULE_FIRMWARE("nvidia/gm200/acr/ucode_unload.bin");
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.h b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.h
new file mode 100644
index 000000000000..45adf1a3bc20
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef __NVKM_SECBOOT_GM200_H__
+#define __NVKM_SECBOOT_GM200_H__
+
+#include "priv.h"
+
+struct gm200_secboot {
+        struct nvkm_secboot base;
+
+        /* Instance block & address space used for HS FW execution */
+        struct nvkm_gpuobj *inst;
+        struct nvkm_gpuobj *pgd;
+        struct nvkm_vm *vm;
+};
+#define gm200_secboot(sb) container_of(sb, struct gm200_secboot, base)
+
+int gm200_secboot_oneinit(struct nvkm_secboot *);
+int gm200_secboot_fini(struct nvkm_secboot *, bool);
+void *gm200_secboot_dtor(struct nvkm_secboot *);
+int gm200_secboot_run_blob(struct nvkm_secboot *, struct nvkm_gpuobj *);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c
index d5395ebfe8d3..6707b8edc086 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c
@@ -20,103 +20,8 @@
  * DEALINGS IN THE SOFTWARE.
  */
 
-#include "priv.h"
-
-#include <core/gpuobj.h>
-
-/*
- * The BL header format used by GM20B's firmware is slightly different
- * from the one of GM200. Fix the differences here.
- */
-struct gm20b_flcn_bl_desc {
-        u32 reserved[4];
-        u32 signature[4];
-        u32 ctx_dma;
-        u32 code_dma_base;
-        u32 non_sec_code_off;
-        u32 non_sec_code_size;
-        u32 sec_code_off;
-        u32 sec_code_size;
-        u32 code_entry_point;
-        u32 data_dma_base;
-        u32 data_size;
-};
-
-static int
-gm20b_secboot_prepare_blobs(struct gm200_secboot *gsb)
-{
-        struct nvkm_subdev *subdev = &gsb->base.subdev;
-        int acr_size;
-        int ret;
-
-        ret = gm20x_secboot_prepare_blobs(gsb);
-        if (ret)
-                return ret;
-
-        acr_size = gsb->acr_load_blob->size;
-        /*
-         * On Tegra the WPR region is set by the bootloader. It is illegal for
-         * the HS blob to be larger than this region.
-         */
-        if (acr_size > gsb->wpr_size) {
-                nvkm_error(subdev, "WPR region too small for FW blob!\n");
-                nvkm_error(subdev, "required: %dB\n", acr_size);
-                nvkm_error(subdev, "WPR size: %dB\n", gsb->wpr_size);
-                return -ENOSPC;
-        }
-
-        return 0;
-}
-
-/**
- * gm20b_secboot_fixup_bl_desc - adapt BL descriptor to format used by GM20B FW
- *
- * There is only a slight format difference (DMA addresses being 32-bits and
- * 256B-aligned) to address.
- */
-static void
-gm20b_secboot_fixup_bl_desc(const struct gm200_flcn_bl_desc *desc, void *ret)
-{
-        struct gm20b_flcn_bl_desc *gdesc = ret;
-        u64 addr;
-
-        memcpy(gdesc->reserved, desc->reserved, sizeof(gdesc->reserved));
-        memcpy(gdesc->signature, desc->signature, sizeof(gdesc->signature));
-        gdesc->ctx_dma = desc->ctx_dma;
-        addr = desc->code_dma_base.hi;
-        addr <<= 32;
-        addr |= desc->code_dma_base.lo;
-        gdesc->code_dma_base = lower_32_bits(addr >> 8);
-        gdesc->non_sec_code_off = desc->non_sec_code_off;
-        gdesc->non_sec_code_size = desc->non_sec_code_size;
-        gdesc->sec_code_off = desc->sec_code_off;
-        gdesc->sec_code_size = desc->sec_code_size;
-        gdesc->code_entry_point = desc->code_entry_point;
-        addr = desc->data_dma_base.hi;
-        addr <<= 32;
-        addr |= desc->data_dma_base.lo;
-        gdesc->data_dma_base = lower_32_bits(addr >> 8);
-        gdesc->data_size = desc->data_size;
-}
-
-static void
-gm20b_secboot_fixup_hs_desc(struct gm200_secboot *gsb,
-                            struct hsflcn_acr_desc *desc)
-{
-        desc->ucode_blob_base = gsb->ls_blob->addr;
-        desc->ucode_blob_size = gsb->ls_blob->size;
-
-        desc->wpr_offset = 0;
-}
-
-static const struct gm200_secboot_func
-gm20b_secboot_func = {
-        .bl_desc_size = sizeof(struct gm20b_flcn_bl_desc),
-        .fixup_bl_desc = gm20b_secboot_fixup_bl_desc,
-        .fixup_hs_desc = gm20b_secboot_fixup_hs_desc,
-        .prepare_blobs = gm20b_secboot_prepare_blobs,
-};
-
+#include "acr.h"
+#include "gm200.h"
 
 #ifdef CONFIG_ARCH_TEGRA
 #define TEGRA_MC_BASE                           0x70019000
@@ -144,15 +49,15 @@ gm20b_tegra_read_wpr(struct gm200_secboot *gsb)
                 nvkm_error(&sb->subdev, "Cannot map Tegra MC registers\n");
                 return PTR_ERR(mc);
         }
-        gsb->wpr_addr = ioread32_native(mc + MC_SECURITY_CARVEOUT2_BOM_0) |
+        sb->wpr_addr = ioread32_native(mc + MC_SECURITY_CARVEOUT2_BOM_0) |
               ((u64)ioread32_native(mc + MC_SECURITY_CARVEOUT2_BOM_HI_0) << 32);
-        gsb->wpr_size = ioread32_native(mc + MC_SECURITY_CARVEOUT2_SIZE_128K)
+        sb->wpr_size = ioread32_native(mc + MC_SECURITY_CARVEOUT2_SIZE_128K)
                 << 17;
         cfg = ioread32_native(mc + MC_SECURITY_CARVEOUT2_CFG0);
         iounmap(mc);
 
         /* Check that WPR settings are valid */
-        if (gsb->wpr_size == 0) {
+        if (sb->wpr_size == 0) {
                 nvkm_error(&sb->subdev, "WPR region is empty\n");
                 return -EINVAL;
         }
@@ -174,7 +79,7 @@ gm20b_tegra_read_wpr(struct gm200_secboot *gsb)
 #endif
 
 static int
-gm20b_secboot_init(struct nvkm_secboot *sb)
+gm20b_secboot_oneinit(struct nvkm_secboot *sb)
 {
         struct gm200_secboot *gsb = gm200_secboot(sb);
         int ret;
@@ -183,17 +88,15 @@ gm20b_secboot_init(struct nvkm_secboot *sb)
         if (ret)
                 return ret;
 
-        return gm200_secboot_init(sb);
+        return gm200_secboot_oneinit(sb);
 }
 
 static const struct nvkm_secboot_func
 gm20b_secboot = {
         .dtor = gm200_secboot_dtor,
-        .init = gm20b_secboot_init,
-        .reset = gm200_secboot_reset,
-        .start = gm200_secboot_start,
-        .managed_falcons = BIT(NVKM_SECBOOT_FALCON_FECS),
-        .boot_falcon = NVKM_SECBOOT_FALCON_PMU,
+        .oneinit = gm20b_secboot_oneinit,
+        .fini = gm200_secboot_fini,
+        .run_blob = gm200_secboot_run_blob,
 };
 
 int
@@ -202,6 +105,11 @@ gm20b_secboot_new(struct nvkm_device *device, int index,
 {
         int ret;
         struct gm200_secboot *gsb;
+        struct nvkm_acr *acr;
+
+        acr = acr_r352_new(BIT(NVKM_SECBOOT_FALCON_FECS));
+        if (IS_ERR(acr))
+                return PTR_ERR(acr);
 
         gsb = kzalloc(sizeof(*gsb), GFP_KERNEL);
         if (!gsb) {
@@ -210,12 +118,10 @@ gm20b_secboot_new(struct nvkm_device *device, int index,
         }
         *psb = &gsb->base;
 
-        ret = nvkm_secboot_ctor(&gm20b_secboot, device, index, &gsb->base);
+        ret = nvkm_secboot_ctor(&gm20b_secboot, acr, device, index, &gsb->base);
         if (ret)
                 return ret;
 
-        gsb->func = &gm20b_secboot_func;
-
         return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/ls_ucode.h b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/ls_ucode.h
new file mode 100644
index 000000000000..00886cee57eb
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/ls_ucode.h
@@ -0,0 +1,151 @@
+/*
+ * Copyright (c) 2014, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef __NVKM_SECBOOT_LS_UCODE_H__
+#define __NVKM_SECBOOT_LS_UCODE_H__
+
+#include <core/os.h>
+#include <core/subdev.h>
+#include <subdev/secboot.h>
+
+
+/**
+ * struct ls_ucode_img_desc - descriptor of firmware image
+ * @descriptor_size:            size of this descriptor
+ * @image_size:                 size of the whole image
+ * @bootloader_start_offset:    start offset of the bootloader in ucode image
+ * @bootloader_size:            size of the bootloader
+ * @bootloader_imem_offset:     start off set of the bootloader in IMEM
+ * @bootloader_entry_point:     entry point of the bootloader in IMEM
+ * @app_start_offset:           start offset of the LS firmware
+ * @app_size:                   size of the LS firmware's code and data
+ * @app_imem_offset:            offset of the app in IMEM
+ * @app_imem_entry:             entry point of the app in IMEM
+ * @app_dmem_offset:            offset of the data in DMEM
+ * @app_resident_code_offset:   offset of app code from app_start_offset
+ * @app_resident_code_size:     size of the code
+ * @app_resident_data_offset:   offset of data from app_start_offset
+ * @app_resident_data_size:     size of data
+ *
+ * A firmware image contains the code, data, and bootloader of a given LS
+ * falcon in a single blob. This structure describes where everything is.
+ *
+ * This can be generated from a (bootloader, code, data) set if they have
+ * been loaded separately, or come directly from a file.
+ */
+struct ls_ucode_img_desc {
+        u32 descriptor_size;
+        u32 image_size;
+        u32 tools_version;
+        u32 app_version;
+        char date[64];
+        u32 bootloader_start_offset;
+        u32 bootloader_size;
+        u32 bootloader_imem_offset;
+        u32 bootloader_entry_point;
+        u32 app_start_offset;
+        u32 app_size;
+        u32 app_imem_offset;
+        u32 app_imem_entry;
+        u32 app_dmem_offset;
+        u32 app_resident_code_offset;
+        u32 app_resident_code_size;
+        u32 app_resident_data_offset;
+        u32 app_resident_data_size;
+        u32 nb_overlays;
+        struct {u32 start; u32 size; } load_ovl[64];
+        u32 compressed;
+};
+
+/**
+ * struct ls_ucode_img - temporary storage for loaded LS firmwares
+ * @node:               to link within lsf_ucode_mgr
+ * @falcon_id:          ID of the falcon this LS firmware is for
+ * @ucode_desc:         loaded or generated map of ucode_data
+ * @ucode_data:         firmware payload (code and data)
+ * @ucode_size:         size in bytes of data in ucode_data
+ * @sig:                signature for this firmware
+ * @sig:size:           size of the signature in bytes
+ *
+ * Preparing the WPR LS blob requires information about all the LS firmwares
+ * (size, etc) to be known. This structure contains all the data of one LS
+ * firmware.
+ */
+struct ls_ucode_img {
+        struct list_head node;
+        enum nvkm_secboot_falcon falcon_id;
+
+        struct ls_ucode_img_desc ucode_desc;
+        u8 *ucode_data;
+        u32 ucode_size;
+
+        u8 *sig;
+        u32 sig_size;
+};
+
+/**
+ * struct fw_bin_header - header of firmware files
+ * @bin_magic:          always 0x3b1d14f0
+ * @bin_ver:            version of the bin format
+ * @bin_size:           entire image size including this header
+ * @header_offset:      offset of the firmware/bootloader header in the file
+ * @data_offset:        offset of the firmware/bootloader payload in the file
+ * @data_size:          size of the payload
+ *
+ * This header is located at the beginning of the HS firmware and HS bootloader
+ * files, to describe where the headers and data can be found.
+ */
+struct fw_bin_header {
+        u32 bin_magic;
+        u32 bin_ver;
+        u32 bin_size;
+        u32 header_offset;
+        u32 data_offset;
+        u32 data_size;
+};
+
+/**
+ * struct fw_bl_desc - firmware bootloader descriptor
+ * @start_tag:          starting tag of bootloader
+ * @desc_dmem_load_off: DMEM offset of flcn_bl_dmem_desc
+ * @code_off:           offset of code section
+ * @code_size:          size of code section
+ * @data_off:           offset of data section
+ * @data_size:          size of data section
+ *
+ * This structure is embedded in bootloader firmware files at to describe the
+ * IMEM and DMEM layout expected by the bootloader.
+ */
+struct fw_bl_desc {
+        u32 start_tag;
+        u32 dmem_load_off;
+        u32 code_off;
+        u32 code_size;
+        u32 data_off;
+        u32 data_size;
+};
+
+int acr_ls_ucode_load_fecs(const struct nvkm_subdev *, struct ls_ucode_img *);
+int acr_ls_ucode_load_gpccs(const struct nvkm_subdev *, struct ls_ucode_img *);
+
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/ls_ucode_gr.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/ls_ucode_gr.c
new file mode 100644
index 000000000000..40a6df77bb8a
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/ls_ucode_gr.c
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+
+#include "ls_ucode.h"
+#include "acr.h"
+
+#include <core/firmware.h>
+
+#define BL_DESC_BLK_SIZE 256
+/**
+ * Build a ucode image and descriptor from provided bootloader, code and data.
+ *
+ * @bl:         bootloader image, including 16-bytes descriptor
+ * @code:       LS firmware code segment
+ * @data:       LS firmware data segment
+ * @desc:       ucode descriptor to be written
+ *
+ * Return: allocated ucode image with corresponding descriptor information. desc
+ *         is also updated to contain the right offsets within returned image.
+ */
+static void *
+ls_ucode_img_build(const struct firmware *bl, const struct firmware *code,
+                   const struct firmware *data, struct ls_ucode_img_desc *desc)
+{
+        struct fw_bin_header *bin_hdr = (void *)bl->data;
+        struct fw_bl_desc *bl_desc = (void *)bl->data + bin_hdr->header_offset;
+        void *bl_data = (void *)bl->data + bin_hdr->data_offset;
+        u32 pos = 0;
+        void *image;
+
+        desc->bootloader_start_offset = pos;
+        desc->bootloader_size = ALIGN(bl_desc->code_size, sizeof(u32));
+        desc->bootloader_imem_offset = bl_desc->start_tag * 256;
+        desc->bootloader_entry_point = bl_desc->start_tag * 256;
+
+        pos = ALIGN(pos + desc->bootloader_size, BL_DESC_BLK_SIZE);
+        desc->app_start_offset = pos;
+        desc->app_size = ALIGN(code->size, BL_DESC_BLK_SIZE) +
+                         ALIGN(data->size, BL_DESC_BLK_SIZE);
+        desc->app_imem_offset = 0;
+        desc->app_imem_entry = 0;
+        desc->app_dmem_offset = 0;
+        desc->app_resident_code_offset = 0;
+        desc->app_resident_code_size = ALIGN(code->size, BL_DESC_BLK_SIZE);
+
+        pos = ALIGN(pos + desc->app_resident_code_size, BL_DESC_BLK_SIZE);
+        desc->app_resident_data_offset = pos - desc->app_start_offset;
+        desc->app_resident_data_size = ALIGN(data->size, BL_DESC_BLK_SIZE);
+
+        desc->image_size = ALIGN(bl_desc->code_size, BL_DESC_BLK_SIZE) +
+                           desc->app_size;
+
+        image = kzalloc(desc->image_size, GFP_KERNEL);
+        if (!image)
+                return ERR_PTR(-ENOMEM);
+
+        memcpy(image + desc->bootloader_start_offset, bl_data,
+               bl_desc->code_size);
+        memcpy(image + desc->app_start_offset, code->data, code->size);
+        memcpy(image + desc->app_start_offset + desc->app_resident_data_offset,
+               data->data, data->size);
+
+        return image;
+}
+
+/**
+ * ls_ucode_img_load_gr() - load and prepare a LS GR ucode image
+ *
+ * Load the LS microcode, bootloader and signature and pack them into a single
+ * blob. Also generate the corresponding ucode descriptor.
+ */
+static int
+ls_ucode_img_load_gr(const struct nvkm_subdev *subdev, struct ls_ucode_img *img,
+                     const char *falcon_name)
+{
+        const struct firmware *bl, *code, *data, *sig;
+        char f[64];
+        int ret;
+
+        snprintf(f, sizeof(f), "gr/%s_bl", falcon_name);
+        ret = nvkm_firmware_get(subdev->device, f, &bl);
+        if (ret)
+                goto error;
+
+        snprintf(f, sizeof(f), "gr/%s_inst", falcon_name);
+        ret = nvkm_firmware_get(subdev->device, f, &code);
+        if (ret)
+                goto free_bl;
+
+        snprintf(f, sizeof(f), "gr/%s_data", falcon_name);
+        ret = nvkm_firmware_get(subdev->device, f, &data);
+        if (ret)
+                goto free_inst;
+
+        snprintf(f, sizeof(f), "gr/%s_sig", falcon_name);
+        ret = nvkm_firmware_get(subdev->device, f, &sig);
+        if (ret)
+                goto free_data;
+        img->sig = kmemdup(sig->data, sig->size, GFP_KERNEL);
+        if (!img->sig) {
+                ret = -ENOMEM;
+                goto free_sig;
+        }
+        img->sig_size = sig->size;
+
+        img->ucode_data = ls_ucode_img_build(bl, code, data,
+                                             &img->ucode_desc);
+        if (IS_ERR(img->ucode_data)) {
+                ret = PTR_ERR(img->ucode_data);
+                goto free_data;
+        }
+        img->ucode_size = img->ucode_desc.image_size;
+
+free_sig:
+        nvkm_firmware_put(sig);
+free_data:
+        nvkm_firmware_put(data);
+free_inst:
+        nvkm_firmware_put(code);
+free_bl:
+        nvkm_firmware_put(bl);
+error:
+        return ret;
+}
+
+int
+acr_ls_ucode_load_fecs(const struct nvkm_subdev *subdev,
+                       struct ls_ucode_img *img)
+{
+        return ls_ucode_img_load_gr(subdev, img, "fecs");
+}
+
+int
+acr_ls_ucode_load_gpccs(const struct nvkm_subdev *subdev,
+                        struct ls_ucode_img *img)
+{
+        return ls_ucode_img_load_gr(subdev, img, "gpccs");
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h
index a9a8a0e1017e..936a65f5658c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h
@@ -27,20 +27,16 @@
 #include <subdev/mmu.h>
 
 struct nvkm_secboot_func {
-        int (*init)(struct nvkm_secboot *);
+        int (*oneinit)(struct nvkm_secboot *);
         int (*fini)(struct nvkm_secboot *, bool suspend);
         void *(*dtor)(struct nvkm_secboot *);
-        int (*reset)(struct nvkm_secboot *, enum nvkm_secboot_falcon);
-        int (*start)(struct nvkm_secboot *, enum nvkm_secboot_falcon);
-
-        /* ID of the falcon that will perform secure boot */
-        enum nvkm_secboot_falcon boot_falcon;
-        /* Bit-mask of IDs of managed falcons */
-        unsigned long managed_falcons;
+        int (*run_blob)(struct nvkm_secboot *, struct nvkm_gpuobj *);
 };
 
-int nvkm_secboot_ctor(const struct nvkm_secboot_func *, struct nvkm_device *,
-                      int index, struct nvkm_secboot *);
+extern const char *nvkm_secboot_falcon_name[];
+
+int nvkm_secboot_ctor(const struct nvkm_secboot_func *, struct nvkm_acr *,
+                      struct nvkm_device *, int, struct nvkm_secboot *);
 int nvkm_secboot_falcon_reset(struct nvkm_secboot *);
 int nvkm_secboot_falcon_run(struct nvkm_secboot *);
 
@@ -48,187 +44,20 @@ struct flcn_u64 {
         u32 lo;
         u32 hi;
 };
+
 static inline u64 flcn64_to_u64(const struct flcn_u64 f)
 {
         return ((u64)f.hi) << 32 | f.lo;
 }
 
-/**
- * struct gm200_flcn_bl_desc - DMEM bootloader descriptor
- * @signature:          16B signature for secure code. 0s if no secure code
- * @ctx_dma:            DMA context to be used by BL while loading code/data
- * @code_dma_base:      256B-aligned Physical FB Address where code is located
- *                      (falcon's $xcbase register)
- * @non_sec_code_off:   offset from code_dma_base where the non-secure code is
- *                      located. The offset must be multiple of 256 to help perf
- * @non_sec_code_size:  the size of the nonSecure code part.
- * @sec_code_off:       offset from code_dma_base where the secure code is
- *                      located. The offset must be multiple of 256 to help perf
- * @sec_code_size:      offset from code_dma_base where the secure code is
- *                      located. The offset must be multiple of 256 to help perf
- * @code_entry_point:   code entry point which will be invoked by BL after
- *                      code is loaded.
- * @data_dma_base:      256B aligned Physical FB Address where data is located.
- *                      (falcon's $xdbase register)
- * @data_size:          size of data block. Should be multiple of 256B
- *
- * Structure used by the bootloader to load the rest of the code. This has
- * to be filled by host and copied into DMEM at offset provided in the
- * hsflcn_bl_desc.bl_desc_dmem_load_off.
- */
-struct gm200_flcn_bl_desc {
-        u32 reserved[4];
-        u32 signature[4];
-        u32 ctx_dma;
-        struct flcn_u64 code_dma_base;
-        u32 non_sec_code_off;
-        u32 non_sec_code_size;
-        u32 sec_code_off;
-        u32 sec_code_size;
-        u32 code_entry_point;
-        struct flcn_u64 data_dma_base;
-        u32 data_size;
-};
-
-/**
- * struct hsflcn_acr_desc - data section of the HS firmware
- *
- * This header is to be copied at the beginning of DMEM by the HS bootloader.
- *
- * @signature:          signature of ACR ucode
- * @wpr_region_id:      region ID holding the WPR header and its details
- * @wpr_offset:         offset from the WPR region holding the wpr header
- * @regions:            region descriptors
- * @nonwpr_ucode_blob_size:     size of LS blob
- * @nonwpr_ucode_blob_start:    FB location of LS blob is
- */
-struct hsflcn_acr_desc {
-        union {
-                u8 reserved_dmem[0x200];
-                u32 signatures[4];
-        } ucode_reserved_space;
-        u32 wpr_region_id;
-        u32 wpr_offset;
-        u32 mmu_mem_range;
-#define FLCN_ACR_MAX_REGIONS 2
-        struct {
-                u32 no_regions;
-                struct {
-                        u32 start_addr;
-                        u32 end_addr;
-                        u32 region_id;
-                        u32 read_mask;
-                        u32 write_mask;
-                        u32 client_mask;
-                } region_props[FLCN_ACR_MAX_REGIONS];
-        } regions;
-        u32 ucode_blob_size;
-        u64 ucode_blob_base __aligned(8);
-        struct {
-                u32 vpr_enabled;
-                u32 vpr_start;
-                u32 vpr_end;
-                u32 hdcp_policies;
-        } vpr_desc;
-};
-
-/**
- * Contains the whole secure boot state, allowing it to be performed as needed
- * @wpr_addr:           physical address of the WPR region
- * @wpr_size:           size in bytes of the WPR region
- * @ls_blob:            LS blob of all the LS firmwares, signatures, bootloaders
- * @ls_blob_size:       size of the LS blob
- * @ls_blob_nb_regions: number of LS firmwares that will be loaded
- * @acr_blob:           HS blob
- * @acr_blob_vma:       mapping of the HS blob into the secure falcon's VM
- * @acr_bl_desc:        bootloader descriptor of the HS blob
- * @hsbl_blob:          HS blob bootloader
- * @inst:               instance block for HS falcon
- * @pgd:                page directory for the HS falcon
- * @vm:                 address space used by the HS falcon
- * @falcon_state:       current state of the managed falcons
- * @firmware_ok:        whether the firmware blobs have been created
- */
-struct gm200_secboot {
-        struct nvkm_secboot base;
-        const struct gm200_secboot_func *func;
-
-        /*
-         * Address and size of the WPR region. On dGPU this will be the
-         * address of the LS blob. On Tegra this is a fixed region set by the
-         * bootloader
-         */
-        u64 wpr_addr;
-        u32 wpr_size;
-
-        /*
-         * HS FW - lock WPR region (dGPU only) and load LS FWs
-         * on Tegra the HS FW copies the LS blob into the fixed WPR instead
-         */
-        struct nvkm_gpuobj *acr_load_blob;
-        struct gm200_flcn_bl_desc acr_load_bl_desc;
-
-        /* HS FW - unlock WPR region (dGPU only) */
-        struct nvkm_gpuobj *acr_unload_blob;
-        struct gm200_flcn_bl_desc acr_unload_bl_desc;
-
-        /* HS bootloader */
-        void *hsbl_blob;
-
-        /* LS FWs, to be loaded by the HS ACR */
-        struct nvkm_gpuobj *ls_blob;
-
-        /* Instance block & address space used for HS FW execution */
-        struct nvkm_gpuobj *inst;
-        struct nvkm_gpuobj *pgd;
-        struct nvkm_vm *vm;
-
-        /* To keep track of the state of all managed falcons */
-        enum {
-                /* In non-secure state, no firmware loaded, no privileges*/
-                NON_SECURE = 0,
-                /* In low-secure mode and ready to be started */
-                RESET,
-                /* In low-secure mode and running */
-                RUNNING,
-        } falcon_state[NVKM_SECBOOT_FALCON_END];
-
-        bool firmware_ok;
-};
-#define gm200_secboot(sb) container_of(sb, struct gm200_secboot, base)
-
-/**
- * Contains functions we wish to abstract between GM200-like implementations
- * @bl_desc_size:       size of the BL descriptor used by this chip.
- * @fixup_bl_desc:      hook that generates the proper BL descriptor format from
- *                      the generic GM200 format into a data array of size
- *                      bl_desc_size
- * @fixup_hs_desc:      hook that twiddles the HS descriptor before it is used
- * @prepare_blobs:      prepares the various blobs needed for secure booting
- */
-struct gm200_secboot_func {
-        /*
-         * Size of the bootloader descriptor for this chip. A block of this
-         * size is allocated before booting a falcon and the fixup_bl_desc
-         * callback is called on it
-         */
-        u32 bl_desc_size;
-        void (*fixup_bl_desc)(const struct gm200_flcn_bl_desc *, void *);
-
-        /*
-         * Chip-specific modifications of the HS descriptor can be done here.
-         * On dGPU this is used to fill the information about the WPR region
-         * we want the HS FW to set up.
-         */
-        void (*fixup_hs_desc)(struct gm200_secboot *, struct hsflcn_acr_desc *);
-        int (*prepare_blobs)(struct gm200_secboot *);
-};
+static inline struct flcn_u64 u64_to_flcn64(u64 u)
+{
+        struct flcn_u64 ret;
 
-int gm200_secboot_init(struct nvkm_secboot *);
-void *gm200_secboot_dtor(struct nvkm_secboot *);
-int gm200_secboot_reset(struct nvkm_secboot *, u32);
-int gm200_secboot_start(struct nvkm_secboot *, u32);
+        ret.hi = upper_32_bits(u);
+        ret.lo = lower_32_bits(u);
 
-int gm20x_secboot_prepare_blobs(struct gm200_secboot *);
+        return ret;
+}
 
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c
index 8894fee30cbc..df949fa7d05d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c
@@ -64,10 +64,9 @@ nvkm_therm_update_trip(struct nvkm_therm *therm)
 }
 
 static int
-nvkm_therm_update_linear(struct nvkm_therm *therm)
+nvkm_therm_compute_linear_duty(struct nvkm_therm *therm, u8 linear_min_temp,
+                               u8 linear_max_temp)
 {
-        u8  linear_min_temp = therm->fan->bios.linear_min_temp;
-        u8  linear_max_temp = therm->fan->bios.linear_max_temp;
         u8  temp = therm->func->temp_get(therm);
         u16 duty;
 
@@ -85,6 +84,21 @@ nvkm_therm_update_linear(struct nvkm_therm *therm)
         return duty;
 }
 
+static int
+nvkm_therm_update_linear(struct nvkm_therm *therm)
+{
+        u8  min = therm->fan->bios.linear_min_temp;
+        u8  max = therm->fan->bios.linear_max_temp;
+        return nvkm_therm_compute_linear_duty(therm, min, max);
+}
+
+static int
+nvkm_therm_update_linear_fallback(struct nvkm_therm *therm)
+{
+        u8 max = therm->bios_sensor.thrs_fan_boost.temp;
+        return nvkm_therm_compute_linear_duty(therm, 30, max);
+}
+
 static void
 nvkm_therm_update(struct nvkm_therm *therm, int mode)
 {
@@ -119,6 +133,8 @@ nvkm_therm_update(struct nvkm_therm *therm, int mode)
                 case NVBIOS_THERM_FAN_OTHER:
                         if (therm->cstate)
                                 duty = therm->cstate;
+                        else
+                                duty = nvkm_therm_update_linear_fallback(therm);
                         poll = false;
                         break;
                 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c
index fe063d5728e2..67ada1d9a28c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c
@@ -95,6 +95,20 @@ nvkm_top_intr(struct nvkm_device *device, u32 intr, u64 *psubdevs)
         return intr & ~handled;
 }
 
+int
+nvkm_top_fault_id(struct nvkm_device *device, enum nvkm_devidx devidx)
+{
+        struct nvkm_top *top = device->top;
+        struct nvkm_top_device *info;
+
+        list_for_each_entry(info, &top->device, head) {
+                if (info->index == devidx && info->fault >= 0)
+                        return info->fault;
+        }
+
+        return -ENOENT;
+}
+
 enum nvkm_devidx
 nvkm_top_fault(struct nvkm_device *device, int fault)
 {