drm/nv84/fence: access fences with full virtual address, not offset

Allows most of the code to be shared between nv84/nvc0 implementations,
and paves the way for doing emit/sync on non-VRAM buffers (multi-gpu,
dma-buf).

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

4 files changed, 149 insertions, 191 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.h b/drivers/gpu/drm/nouveau/nouveau_fence.h
index be6166eb966d..2324911fc4c3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.h
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.h
@@ -68,8 +68,27 @@ void nv17_fence_resume(struct nouveau_drm *drm);
 int nv50_fence_create(struct nouveau_drm *);
 int nv84_fence_create(struct nouveau_drm *);
 int nvc0_fence_create(struct nouveau_drm *);
-u64 nvc0_fence_crtc(struct nouveau_channel *, int crtc);
 
 int nouveau_flip_complete(void *chan);
 
+struct nv84_fence_chan {
+        struct nouveau_fence_chan base;
+        struct nouveau_vma vma;
+        struct nouveau_vma dispc_vma[4];
+};
+
+struct nv84_fence_priv {
+        struct nouveau_fence_priv base;
+        struct nouveau_bo *bo;
+        u32 *suspend;
+};
+
+u64  nv84_fence_crtc(struct nouveau_channel *, int);
+u32  nv84_fence_read(struct nouveau_channel *);
+int  nv84_fence_context_new(struct nouveau_channel *);
+void nv84_fence_context_del(struct nouveau_channel *);
+bool nv84_fence_suspend(struct nouveau_drm *);
+void nv84_fence_resume(struct nouveau_drm *);
+void nv84_fence_destroy(struct nouveau_drm *);
+
 #endif
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index c9da4f108924..102a8734d377 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -502,7 +502,7 @@ nv50_display_flip_next(struct drm_crtc *crtc, struct drm_framebuffer *fb,
                 if (ret)
                         return ret;
 
-                if (nv_mclass(chan->object) < NVC0_CHANNEL_IND_CLASS) {
+                if (nv_mclass(chan->object) < NV84_CHANNEL_IND_CLASS) {
                         BEGIN_NV04(chan, 0, NV11_SUBCHAN_DMA_SEMAPHORE, 2);
                         OUT_RING  (chan, NvEvoSema0 + nv_crtc->index);
                         OUT_RING  (chan, sync->sem.offset);
@@ -512,24 +512,36 @@ nv50_display_flip_next(struct drm_crtc *crtc, struct drm_framebuffer *fb,
                         OUT_RING  (chan, sync->sem.offset ^ 0x10);
                         OUT_RING  (chan, 0x74b1e000);
                         BEGIN_NV04(chan, 0, NV11_SUBCHAN_DMA_SEMAPHORE, 1);
-                        if (nv_mclass(chan->object) < NV84_CHANNEL_DMA_CLASS)
-                                OUT_RING  (chan, NvSema);
-                        else
-                                OUT_RING  (chan, chan->vram);
+                        OUT_RING  (chan, NvSema);
+                } else
+                if (nv_mclass(chan->object) < NVC0_CHANNEL_IND_CLASS) {
+                        u64 offset = nv84_fence_crtc(chan, nv_crtc->index);
+                        offset += sync->sem.offset;
+
+                        BEGIN_NV04(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 4);
+                        OUT_RING  (chan, upper_32_bits(offset));
+                        OUT_RING  (chan, lower_32_bits(offset));
+                        OUT_RING  (chan, 0xf00d0000 | sync->sem.value);
+                        OUT_RING  (chan, 0x00000002);
+                        BEGIN_NV04(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 4);
+                        OUT_RING  (chan, upper_32_bits(offset));
+                        OUT_RING  (chan, lower_32_bits(offset ^ 0x10));
+                        OUT_RING  (chan, 0x74b1e000);
+                        OUT_RING  (chan, 0x00000001);
                 } else {
-                        u64 offset = nvc0_fence_crtc(chan, nv_crtc->index);
+                        u64 offset = nv84_fence_crtc(chan, nv_crtc->index);
                         offset += sync->sem.offset;
 
                         BEGIN_NVC0(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 4);
                         OUT_RING  (chan, upper_32_bits(offset));
                         OUT_RING  (chan, lower_32_bits(offset));
                         OUT_RING  (chan, 0xf00d0000 | sync->sem.value);
-                        OUT_RING  (chan, 0x1002);
+                        OUT_RING  (chan, 0x00001002);
                         BEGIN_NVC0(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 4);
                         OUT_RING  (chan, upper_32_bits(offset));
                         OUT_RING  (chan, lower_32_bits(offset ^ 0x10));
                         OUT_RING  (chan, 0x74b1e000);
-                        OUT_RING  (chan, 0x1001);
+                        OUT_RING  (chan, 0x00001001);
                 }
 
                 FIRE_RING (chan);
diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
index e64e8154a5af..58c2401b18ff 100644
--- a/drivers/gpu/drm/nouveau/nv84_fence.c
+++ b/drivers/gpu/drm/nouveau/nv84_fence.c
@@ -23,6 +23,7 @@
  */
 
 #include <core/object.h>
+#include <core/client.h>
 #include <core/class.h>
 
 #include <engine/fifo.h>
@@ -33,80 +34,96 @@
 
 #include "nv50_display.h"
 
-struct nv84_fence_chan {
-        struct nouveau_fence_chan base;
-};
-
-struct nv84_fence_priv {
-        struct nouveau_fence_priv base;
-        struct nouveau_gpuobj *mem;
-};
+u64
+nv84_fence_crtc(struct nouveau_channel *chan, int crtc)
+{
+        struct nv84_fence_chan *fctx = chan->fence;
+        return fctx->dispc_vma[crtc].offset;
+}
 
 static int
 nv84_fence_emit(struct nouveau_fence *fence)
 {
         struct nouveau_channel *chan = fence->channel;
+        struct nv84_fence_chan *fctx = chan->fence;
         struct nouveau_fifo_chan *fifo = (void *)chan->object;
-        int ret = RING_SPACE(chan, 8);
+        u64 addr = fctx->vma.offset + fifo->chid * 16;
+        int ret;
+
+        ret = RING_SPACE(chan, 8);
         if (ret == 0) {
                 BEGIN_NV04(chan, 0, NV11_SUBCHAN_DMA_SEMAPHORE, 1);
-                OUT_RING  (chan, NvSema);
+                OUT_RING  (chan, chan->vram);
                 BEGIN_NV04(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 5);
-                OUT_RING  (chan, upper_32_bits(fifo->chid * 16));
-                OUT_RING  (chan, lower_32_bits(fifo->chid * 16));
+                OUT_RING  (chan, upper_32_bits(addr));
+                OUT_RING  (chan, lower_32_bits(addr));
                 OUT_RING  (chan, fence->sequence);
                 OUT_RING  (chan, NV84_SUBCHAN_SEMAPHORE_TRIGGER_WRITE_LONG);
                 OUT_RING  (chan, 0x00000000);
                 FIRE_RING (chan);
         }
+
         return ret;
 }
 
-
 static int
 nv84_fence_sync(struct nouveau_fence *fence,
                 struct nouveau_channel *prev, struct nouveau_channel *chan)
 {
+        struct nv84_fence_chan *fctx = chan->fence;
         struct nouveau_fifo_chan *fifo = (void *)prev->object;
-        int ret = RING_SPACE(chan, 7);
+        u64 addr = fctx->vma.offset + fifo->chid * 16;
+        int ret;
+
+        ret = RING_SPACE(chan, 7);
         if (ret == 0) {
                 BEGIN_NV04(chan, 0, NV11_SUBCHAN_DMA_SEMAPHORE, 1);
-                OUT_RING  (chan, NvSema);
+                OUT_RING  (chan, chan->vram);
                 BEGIN_NV04(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 4);
-                OUT_RING  (chan, upper_32_bits(fifo->chid * 16));
-                OUT_RING  (chan, lower_32_bits(fifo->chid * 16));
+                OUT_RING  (chan, upper_32_bits(addr));
+                OUT_RING  (chan, lower_32_bits(addr));
                 OUT_RING  (chan, fence->sequence);
                 OUT_RING  (chan, NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_GEQUAL);
                 FIRE_RING (chan);
         }
+
         return ret;
 }
 
-static u32
+u32
 nv84_fence_read(struct nouveau_channel *chan)
 {
         struct nouveau_fifo_chan *fifo = (void *)chan->object;
         struct nv84_fence_priv *priv = chan->drm->fence;
-        return nv_ro32(priv->mem, fifo->chid * 16);
+        return nouveau_bo_rd32(priv->bo, fifo->chid * 16/4);
 }
 
-static void
+void
 nv84_fence_context_del(struct nouveau_channel *chan)
 {
+        struct drm_device *dev = chan->drm->dev;
+        struct nv84_fence_priv *priv = chan->drm->fence;
         struct nv84_fence_chan *fctx = chan->fence;
+        int i;
+
+        for (i = 0; i < dev->mode_config.num_crtc; i++) {
+                struct nouveau_bo *bo = nv50_display_crtc_sema(dev, i);
+                nouveau_bo_vma_del(bo, &fctx->dispc_vma[i]);
+        }
+
+        nouveau_bo_vma_del(priv->bo, &fctx->vma);
         nouveau_fence_context_del(&fctx->base);
         chan->fence = NULL;
         kfree(fctx);
 }
 
-static int
+int
 nv84_fence_context_new(struct nouveau_channel *chan)
 {
-        struct drm_device *dev = chan->drm->dev;
         struct nouveau_fifo_chan *fifo = (void *)chan->object;
+        struct nouveau_client *client = nouveau_client(fifo);
         struct nv84_fence_priv *priv = chan->drm->fence;
         struct nv84_fence_chan *fctx;
-        struct nouveau_object *object;
         int ret, i;
 
         fctx = chan->fence = kzalloc(sizeof(*fctx), GFP_KERNEL);
@@ -115,43 +132,59 @@ nv84_fence_context_new(struct nouveau_channel *chan)
 
         nouveau_fence_context_new(&fctx->base);
 
-        ret = nouveau_object_new(nv_object(chan->cli), chan->handle,
-                                 NvSema, 0x0002,
-                                 &(struct nv_dma_class) {
-                                        .flags = NV_DMA_TARGET_VRAM |
-                                                 NV_DMA_ACCESS_RDWR,
-                                        .start = priv->mem->addr,
-                                        .limit = priv->mem->addr +
-                                                 priv->mem->size - 1,
-                                 }, sizeof(struct nv_dma_class),
-                                 &object);
-
-        /* dma objects for display sync channel semaphore blocks */
-        for (i = 0; !ret && i < dev->mode_config.num_crtc; i++) {
-                struct nouveau_bo *bo = nv50_display_crtc_sema(dev, i);
+        ret = nouveau_bo_vma_add(priv->bo, client->vm, &fctx->vma);
+        if (ret)
+                nv84_fence_context_del(chan);
 
-                ret = nouveau_object_new(nv_object(chan->cli), chan->handle,
-                                         NvEvoSema0 + i, 0x003d,
-                                         &(struct nv_dma_class) {
-                                                .flags = NV_DMA_TARGET_VRAM |
-                                                         NV_DMA_ACCESS_RDWR,
-                                                .start = bo->bo.offset,
-                                                .limit = bo->bo.offset + 0xfff,
-                                         }, sizeof(struct nv_dma_class),
-                                         &object);
+        /* map display semaphore buffers into channel's vm */
+        for (i = 0; !ret && i < chan->drm->dev->mode_config.num_crtc; i++) {
+                struct nouveau_bo *bo = nv50_display_crtc_sema(chan->drm->dev, i);
+                ret = nouveau_bo_vma_add(bo, client->vm, &fctx->dispc_vma[i]);
         }
 
-        if (ret)
-                nv84_fence_context_del(chan);
-        nv_wo32(priv->mem, fifo->chid * 16, 0x00000000);
+        nouveau_bo_wr32(priv->bo, fifo->chid * 16/4, 0x00000000);
         return ret;
 }
 
-static void
+bool
+nv84_fence_suspend(struct nouveau_drm *drm)
+{
+        struct nouveau_fifo *pfifo = nouveau_fifo(drm->device);
+        struct nv84_fence_priv *priv = drm->fence;
+        int i;
+
+        priv->suspend = vmalloc((pfifo->max + 1) * sizeof(u32));
+        if (priv->suspend) {
+                for (i = 0; i <= pfifo->max; i++)
+                        priv->suspend[i] = nouveau_bo_rd32(priv->bo, i*4);
+        }
+
+        return priv->suspend != NULL;
+}
+
+void
+nv84_fence_resume(struct nouveau_drm *drm)
+{
+        struct nouveau_fifo *pfifo = nouveau_fifo(drm->device);
+        struct nv84_fence_priv *priv = drm->fence;
+        int i;
+
+        if (priv->suspend) {
+                for (i = 0; i <= pfifo->max; i++)
+                        nouveau_bo_wr32(priv->bo, i*4, priv->suspend[i]);
+                vfree(priv->suspend);
+                priv->suspend = NULL;
+        }
+}
+
+void
 nv84_fence_destroy(struct nouveau_drm *drm)
 {
         struct nv84_fence_priv *priv = drm->fence;
-        nouveau_gpuobj_ref(NULL, &priv->mem);
+        nouveau_bo_unmap(priv->bo);
+        if (priv->bo)
+                nouveau_bo_unpin(priv->bo);
+        nouveau_bo_ref(NULL, &priv->bo);
         drm->fence = NULL;
         kfree(priv);
 }
@@ -161,7 +194,6 @@ nv84_fence_create(struct nouveau_drm *drm)
 {
         struct nouveau_fifo *pfifo = nouveau_fifo(drm->device);
         struct nv84_fence_priv *priv;
-        u32 chan = pfifo->max + 1;
         int ret;
 
         priv = drm->fence = kzalloc(sizeof(*priv), GFP_KERNEL);
@@ -169,6 +201,8 @@ nv84_fence_create(struct nouveau_drm *drm)
                 return -ENOMEM;
 
         priv->base.dtor = nv84_fence_destroy;
+        priv->base.suspend = nv84_fence_suspend;
+        priv->base.resume = nv84_fence_resume;
         priv->base.context_new = nv84_fence_context_new;
         priv->base.context_del = nv84_fence_context_del;
         priv->base.emit = nv84_fence_emit;
@@ -178,8 +212,19 @@ nv84_fence_create(struct nouveau_drm *drm)
         init_waitqueue_head(&priv->base.waiting);
         priv->base.uevent = true;
 
-        ret = nouveau_gpuobj_new(drm->device, NULL, chan * 16, 0x1000, 0,
-                                &priv->mem);
+        ret = nouveau_bo_new(drm->dev, 16 * (pfifo->max + 1), 0,
+                             TTM_PL_FLAG_VRAM, 0, 0, NULL, &priv->bo);
+        if (ret == 0) {
+                ret = nouveau_bo_pin(priv->bo, TTM_PL_FLAG_VRAM);
+                if (ret == 0) {
+                        ret = nouveau_bo_map(priv->bo);
+                        if (ret)
+                                nouveau_bo_unpin(priv->bo);
+                }
+                if (ret)
+                        nouveau_bo_ref(NULL, &priv->bo);
+        }
+
         if (ret)
                 nv84_fence_destroy(drm);
         return ret;
diff --git a/drivers/gpu/drm/nouveau/nvc0_fence.c b/drivers/gpu/drm/nouveau/nvc0_fence.c
index d8ed2c5f4fab..e4c4ead24805 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fence.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fence.c
@@ -34,30 +34,11 @@
 
 #include "nv50_display.h"
 
-struct nvc0_fence_priv {
-        struct nouveau_fence_priv base;
-        struct nouveau_bo *bo;
-        u32 *suspend;
-};
-
-struct nvc0_fence_chan {
-        struct nouveau_fence_chan base;
-        struct nouveau_vma vma;
-        struct nouveau_vma dispc_vma[4];
-};
-
-u64
-nvc0_fence_crtc(struct nouveau_channel *chan, int crtc)
-{
-        struct nvc0_fence_chan *fctx = chan->fence;
-        return fctx->dispc_vma[crtc].offset;
-}
-
 static int
 nvc0_fence_emit(struct nouveau_fence *fence)
 {
         struct nouveau_channel *chan = fence->channel;
-        struct nvc0_fence_chan *fctx = chan->fence;
+        struct nv84_fence_chan *fctx = chan->fence;
         struct nouveau_fifo_chan *fifo = (void *)chan->object;
         u64 addr = fctx->vma.offset + fifo->chid * 16;
         int ret;
@@ -80,7 +61,7 @@ static int
 nvc0_fence_sync(struct nouveau_fence *fence,
                 struct nouveau_channel *prev, struct nouveau_channel *chan)
 {
-        struct nvc0_fence_chan *fctx = chan->fence;
+        struct nv84_fence_chan *fctx = chan->fence;
         struct nouveau_fifo_chan *fifo = (void *)prev->object;
         u64 addr = fctx->vma.offset + fifo->chid * 16;
         int ret;
@@ -99,124 +80,25 @@ nvc0_fence_sync(struct nouveau_fence *fence,
         return ret;
 }
 
-static u32
-nvc0_fence_read(struct nouveau_channel *chan)
-{
-        struct nouveau_fifo_chan *fifo = (void *)chan->object;
-        struct nvc0_fence_priv *priv = chan->drm->fence;
-        return nouveau_bo_rd32(priv->bo, fifo->chid * 16/4);
-}
-
-static void
-nvc0_fence_context_del(struct nouveau_channel *chan)
-{
-        struct drm_device *dev = chan->drm->dev;
-        struct nvc0_fence_priv *priv = chan->drm->fence;
-        struct nvc0_fence_chan *fctx = chan->fence;
-        int i;
-
-        for (i = 0; i < dev->mode_config.num_crtc; i++) {
-                struct nouveau_bo *bo = nv50_display_crtc_sema(dev, i);
-                nouveau_bo_vma_del(bo, &fctx->dispc_vma[i]);
-        }
-
-        nouveau_bo_vma_del(priv->bo, &fctx->vma);
-        nouveau_fence_context_del(&fctx->base);
-        chan->fence = NULL;
-        kfree(fctx);
-}
-
-static int
-nvc0_fence_context_new(struct nouveau_channel *chan)
-{
-        struct nouveau_fifo_chan *fifo = (void *)chan->object;
-        struct nouveau_client *client = nouveau_client(fifo);
-        struct nvc0_fence_priv *priv = chan->drm->fence;
-        struct nvc0_fence_chan *fctx;
-        int ret, i;
-
-        fctx = chan->fence = kzalloc(sizeof(*fctx), GFP_KERNEL);
-        if (!fctx)
-                return -ENOMEM;
-
-        nouveau_fence_context_new(&fctx->base);
-
-        ret = nouveau_bo_vma_add(priv->bo, client->vm, &fctx->vma);
-        if (ret)
-                nvc0_fence_context_del(chan);
-
-        /* map display semaphore buffers into channel's vm */
-        for (i = 0; !ret && i < chan->drm->dev->mode_config.num_crtc; i++) {
-                struct nouveau_bo *bo = nv50_display_crtc_sema(chan->drm->dev, i);
-                ret = nouveau_bo_vma_add(bo, client->vm, &fctx->dispc_vma[i]);
-        }
-
-        nouveau_bo_wr32(priv->bo, fifo->chid * 16/4, 0x00000000);
-        return ret;
-}
-
-static bool
-nvc0_fence_suspend(struct nouveau_drm *drm)
-{
-        struct nouveau_fifo *pfifo = nouveau_fifo(drm->device);
-        struct nvc0_fence_priv *priv = drm->fence;
-        int i;
-
-        priv->suspend = vmalloc((pfifo->max + 1) * sizeof(u32));
-        if (priv->suspend) {
-                for (i = 0; i <= pfifo->max; i++)
-                        priv->suspend[i] = nouveau_bo_rd32(priv->bo, i);
-        }
-
-        return priv->suspend != NULL;
-}
-
-static void
-nvc0_fence_resume(struct nouveau_drm *drm)
-{
-        struct nouveau_fifo *pfifo = nouveau_fifo(drm->device);
-        struct nvc0_fence_priv *priv = drm->fence;
-        int i;
-
-        if (priv->suspend) {
-                for (i = 0; i <= pfifo->max; i++)
-                        nouveau_bo_wr32(priv->bo, i, priv->suspend[i]);
-                vfree(priv->suspend);
-                priv->suspend = NULL;
-        }
-}
-
-static void
-nvc0_fence_destroy(struct nouveau_drm *drm)
-{
-        struct nvc0_fence_priv *priv = drm->fence;
-        nouveau_bo_unmap(priv->bo);
-        if (priv->bo)
-                nouveau_bo_unpin(priv->bo);
-        nouveau_bo_ref(NULL, &priv->bo);
-        drm->fence = NULL;
-        kfree(priv);
-}
-
 int
 nvc0_fence_create(struct nouveau_drm *drm)
 {
         struct nouveau_fifo *pfifo = nouveau_fifo(drm->device);
-        struct nvc0_fence_priv *priv;
+        struct nv84_fence_priv *priv;
         int ret;
 
         priv = drm->fence = kzalloc(sizeof(*priv), GFP_KERNEL);
         if (!priv)
                 return -ENOMEM;
 
-        priv->base.dtor = nvc0_fence_destroy;
-        priv->base.suspend = nvc0_fence_suspend;
-        priv->base.resume = nvc0_fence_resume;
-        priv->base.context_new = nvc0_fence_context_new;
-        priv->base.context_del = nvc0_fence_context_del;
+        priv->base.dtor = nv84_fence_destroy;
+        priv->base.suspend = nv84_fence_suspend;
+        priv->base.resume = nv84_fence_resume;
+        priv->base.context_new = nv84_fence_context_new;
+        priv->base.context_del = nv84_fence_context_del;
         priv->base.emit = nvc0_fence_emit;
         priv->base.sync = nvc0_fence_sync;
-        priv->base.read = nvc0_fence_read;
+        priv->base.read = nv84_fence_read;
 
         init_waitqueue_head(&priv->base.waiting);
         priv->base.uevent = true;
@@ -235,6 +117,6 @@ nvc0_fence_create(struct nouveau_drm *drm)
         }
 
         if (ret)
-                nvc0_fence_destroy(drm);
+                nv84_fence_destroy(drm);
         return ret;
 }