aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2013-11-22 13:56:11 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2013-11-22 13:56:11 -0500
commitaecde27c4fc4939f7c16ae13645f896438190567 (patch)
tree6a867d321f71eb1ed9602ae2158b21ac15da6770
parente3414786ffb40a1fd524624369d350ed70a90a08 (diff)
parent760c960bd6880cf22a57c0af9ff60c96250aad39 (diff)
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Pull DRM fixes from Dave Airlie: "I was going to leave this until post -rc1 but sysfs fixes broke hotplug in userspace, so I had to fix it harder, otherwise a set of pulls from intel, radeon and vmware, The vmware/ttm changes are bit larger but since its early and they are unlikely to break anything else I put them in, it lets vmware work with dri3" * 'drm-fixes' of git://people.freedesktop.org/~airlied/linux: (36 commits) drm/sysfs: fix hotplug regression since lifetime changes drm/exynos: g2d: fix memory leak to userptr drm/i915: Fix gen3 self-refresh watermarks drm/ttm: Remove set_need_resched from the ttm fault handler drm/ttm: Don't move non-existing data drm/radeon: hook up backlight functions for CI and KV family. drm/i915: Replicate BIOS eDP bpp clamping hack for hsw drm/i915: Do not enable package C8 on unsupported hardware drm/i915: Hold pc8 lock around toggling pc8.gpu_idle drm/i915: encoder->get_config is no longer optional drm/i915/tv: add ->get_config callback drm/radeon/cik: Add macrotile mode array query drm/radeon/cik: Return backend map information to userspace drm/vmwgfx: Make vmwgfx dma buffers prime aware drm/vmwgfx: Make surfaces prime-aware drm/vmwgfx: Hook up the prime ioctls drm/ttm: Add a minimal prime implementation for ttm base objects drm/vmwgfx: Fix false lockdep warning drm/ttm: Allow execbuf util reserves without ticket drm/i915: restore the early forcewake cleanup ...
Diffstat
-rw-r--r--arch/x86/kernel/early-quirks.c4
-rw-r--r--drivers/gpu/drm/drm_sysfs.c40
-rw-r--r--drivers/gpu/drm/exynos/exynos_drm_g2d.c2
-rw-r--r--drivers/gpu/drm/i915/i915_drv.h1
-rw-r--r--drivers/gpu/drm/i915/intel_bios.c7
-rw-r--r--drivers/gpu/drm/i915/intel_ddi.c20
-rw-r--r--drivers/gpu/drm/i915/intel_display.c33
-rw-r--r--drivers/gpu/drm/i915/intel_dp.c2
-rw-r--r--drivers/gpu/drm/i915/intel_pm.c4
-rw-r--r--drivers/gpu/drm/i915/intel_tv.c8
-rw-r--r--drivers/gpu/drm/i915/intel_uncore.c26
-rw-r--r--drivers/gpu/drm/radeon/atombios_i2c.c6
-rw-r--r--drivers/gpu/drm/radeon/cik.c57
-rw-r--r--drivers/gpu/drm/radeon/cik_sdma.c13
-rw-r--r--drivers/gpu/drm/radeon/cypress_dpm.c2
-rw-r--r--drivers/gpu/drm/radeon/evergreen_dma.c9
-rw-r--r--drivers/gpu/drm/radeon/ni_dpm.c2
-rw-r--r--drivers/gpu/drm/radeon/r100.c3
-rw-r--r--drivers/gpu/drm/radeon/r600.c13
-rw-r--r--drivers/gpu/drm/radeon/r600_dma.c13
-rw-r--r--drivers/gpu/drm/radeon/radeon.h38
-rw-r--r--drivers/gpu/drm/radeon/radeon_asic.c4
-rw-r--r--drivers/gpu/drm/radeon/radeon_asic.h18
-rw-r--r--drivers/gpu/drm/radeon/radeon_cs.c9
-rw-r--r--drivers/gpu/drm/radeon/radeon_device.c47
-rw-r--r--drivers/gpu/drm/radeon/radeon_drv.c3
-rw-r--r--drivers/gpu/drm/radeon/radeon_fence.c30
-rw-r--r--drivers/gpu/drm/radeon/radeon_gart.c6
-rw-r--r--drivers/gpu/drm/radeon/radeon_kms.c11
-rw-r--r--drivers/gpu/drm/radeon/radeon_legacy_crtc.c28
-rw-r--r--drivers/gpu/drm/radeon/radeon_pm.c2
-rw-r--r--drivers/gpu/drm/radeon/radeon_ring.c46
-rw-r--r--drivers/gpu/drm/radeon/radeon_semaphore.c129
-rw-r--r--drivers/gpu/drm/radeon/radeon_trace.h36
-rw-r--r--drivers/gpu/drm/radeon/rv770_dma.c9
-rw-r--r--drivers/gpu/drm/radeon/si_dma.c9
-rw-r--r--drivers/gpu/drm/radeon/trinity_dpm.c6
-rw-r--r--drivers/gpu/drm/radeon/uvd_v1_0.c4
-rw-r--r--drivers/gpu/drm/radeon/uvd_v3_1.c4
-rw-r--r--drivers/gpu/drm/ttm/ttm_bo.c35
-rw-r--r--drivers/gpu/drm/ttm/ttm_bo_util.c7
-rw-r--r--drivers/gpu/drm/ttm/ttm_bo_vm.c26
-rw-r--r--drivers/gpu/drm/ttm/ttm_execbuf_util.c32
-rw-r--r--drivers/gpu/drm/ttm/ttm_object.c254
-rw-r--r--drivers/gpu/drm/vmwgfx/Makefile2
-rw-r--r--drivers/gpu/drm/vmwgfx/vmwgfx_drv.c7
-rw-r--r--drivers/gpu/drm/vmwgfx/vmwgfx_drv.h14
-rw-r--r--drivers/gpu/drm/vmwgfx/vmwgfx_prime.c137
-rw-r--r--drivers/gpu/drm/vmwgfx/vmwgfx_resource.c63
-rw-r--r--drivers/gpu/drm/vmwgfx/vmwgfx_surface.c30
-rw-r--r--include/drm/ttm/ttm_bo_api.h4
-rw-r--r--include/drm/ttm/ttm_execbuf_util.h3
-rw-r--r--include/drm/ttm/ttm_object.h61
-rw-r--r--include/uapi/drm/radeon_drm.h2
54 files changed, 1057 insertions, 324 deletions
diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c
index 96f958d8cd45..bc4a088f9023 100644
--- a/arch/x86/kernel/early-quirks.c
+++ b/arch/x86/kernel/early-quirks.c
@@ -330,8 +330,8 @@ static struct pci_device_id intel_stolen_ids[] __initdata = {
330 INTEL_I915GM_IDS(gen3_stolen_size), 330 INTEL_I915GM_IDS(gen3_stolen_size),
331 INTEL_I945G_IDS(gen3_stolen_size), 331 INTEL_I945G_IDS(gen3_stolen_size),
332 INTEL_I945GM_IDS(gen3_stolen_size), 332 INTEL_I945GM_IDS(gen3_stolen_size),
333 INTEL_VLV_M_IDS(gen3_stolen_size), 333 INTEL_VLV_M_IDS(gen6_stolen_size),
334 INTEL_VLV_D_IDS(gen3_stolen_size), 334 INTEL_VLV_D_IDS(gen6_stolen_size),
335 INTEL_PINEVIEW_IDS(gen3_stolen_size), 335 INTEL_PINEVIEW_IDS(gen3_stolen_size),
336 INTEL_I965G_IDS(gen3_stolen_size), 336 INTEL_I965G_IDS(gen3_stolen_size),
337 INTEL_G33_IDS(gen3_stolen_size), 337 INTEL_G33_IDS(gen3_stolen_size),
diff --git a/drivers/gpu/drm/drm_sysfs.c b/drivers/gpu/drm/drm_sysfs.c
index 1a35ea53106b..bd2bca395792 100644
--- a/drivers/gpu/drm/drm_sysfs.c
+++ b/drivers/gpu/drm/drm_sysfs.c
@@ -489,6 +489,11 @@ void drm_sysfs_hotplug_event(struct drm_device *dev)
489} 489}
490EXPORT_SYMBOL(drm_sysfs_hotplug_event); 490EXPORT_SYMBOL(drm_sysfs_hotplug_event);
491 491
492static void drm_sysfs_release(struct device *dev)
493{
494 kfree(dev);
495}
496
492/** 497/**
493 * drm_sysfs_device_add - adds a class device to sysfs for a character driver 498 * drm_sysfs_device_add - adds a class device to sysfs for a character driver
494 * @dev: DRM device to be added 499 * @dev: DRM device to be added
@@ -501,6 +506,7 @@ EXPORT_SYMBOL(drm_sysfs_hotplug_event);
501int drm_sysfs_device_add(struct drm_minor *minor) 506int drm_sysfs_device_add(struct drm_minor *minor)
502{ 507{
503 char *minor_str; 508 char *minor_str;
509 int r;
504 510
505 if (minor->type == DRM_MINOR_CONTROL) 511 if (minor->type == DRM_MINOR_CONTROL)
506 minor_str = "controlD%d"; 512 minor_str = "controlD%d";
@@ -509,14 +515,34 @@ int drm_sysfs_device_add(struct drm_minor *minor)
509 else 515 else
510 minor_str = "card%d"; 516 minor_str = "card%d";
511 517
512 minor->kdev = device_create(drm_class, minor->dev->dev, 518 minor->kdev = kzalloc(sizeof(*minor->kdev), GFP_KERNEL);
513 MKDEV(DRM_MAJOR, minor->index), 519 if (!minor->dev) {
514 minor, minor_str, minor->index); 520 r = -ENOMEM;
515 if (IS_ERR(minor->kdev)) { 521 goto error;
516 DRM_ERROR("device create failed %ld\n", PTR_ERR(minor->kdev));
517 return PTR_ERR(minor->kdev);
518 } 522 }
523
524 device_initialize(minor->kdev);
525 minor->kdev->devt = MKDEV(DRM_MAJOR, minor->index);
526 minor->kdev->class = drm_class;
527 minor->kdev->type = &drm_sysfs_device_minor;
528 minor->kdev->parent = minor->dev->dev;
529 minor->kdev->release = drm_sysfs_release;
530 dev_set_drvdata(minor->kdev, minor);
531
532 r = dev_set_name(minor->kdev, minor_str, minor->index);
533 if (r < 0)
534 goto error;
535
536 r = device_add(minor->kdev);
537 if (r < 0)
538 goto error;
539
519 return 0; 540 return 0;
541
542error:
543 DRM_ERROR("device create failed %d\n", r);
544 put_device(minor->kdev);
545 return r;
520} 546}
521 547
522/** 548/**
@@ -529,7 +555,7 @@ int drm_sysfs_device_add(struct drm_minor *minor)
529void drm_sysfs_device_remove(struct drm_minor *minor) 555void drm_sysfs_device_remove(struct drm_minor *minor)
530{ 556{
531 if (minor->kdev) 557 if (minor->kdev)
532 device_destroy(drm_class, MKDEV(DRM_MAJOR, minor->index)); 558 device_unregister(minor->kdev);
533 minor->kdev = NULL; 559 minor->kdev = NULL;
534} 560}
535 561
diff --git a/drivers/gpu/drm/exynos/exynos_drm_g2d.c b/drivers/gpu/drm/exynos/exynos_drm_g2d.c
index 3271fd4b1724..7bccedca487a 100644
--- a/drivers/gpu/drm/exynos/exynos_drm_g2d.c
+++ b/drivers/gpu/drm/exynos/exynos_drm_g2d.c
@@ -383,6 +383,8 @@ out:
383 g2d_userptr->npages, 383 g2d_userptr->npages,
384 g2d_userptr->vma); 384 g2d_userptr->vma);
385 385
386 exynos_gem_put_vma(g2d_userptr->vma);
387
386 if (!g2d_userptr->out_of_list) 388 if (!g2d_userptr->out_of_list)
387 list_del_init(&g2d_userptr->list); 389 list_del_init(&g2d_userptr->list);
388 390
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 8600c315b4c4..ccdbecca070d 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1816,6 +1816,7 @@ struct drm_i915_file_private {
1816#define HAS_POWER_WELL(dev) (IS_HASWELL(dev) || IS_BROADWELL(dev)) 1816#define HAS_POWER_WELL(dev) (IS_HASWELL(dev) || IS_BROADWELL(dev))
1817#define HAS_FPGA_DBG_UNCLAIMED(dev) (INTEL_INFO(dev)->has_fpga_dbg) 1817#define HAS_FPGA_DBG_UNCLAIMED(dev) (INTEL_INFO(dev)->has_fpga_dbg)
1818#define HAS_PSR(dev) (IS_HASWELL(dev) || IS_BROADWELL(dev)) 1818#define HAS_PSR(dev) (IS_HASWELL(dev) || IS_BROADWELL(dev))
1819#define HAS_PC8(dev) (IS_HASWELL(dev)) /* XXX HSW:ULX */
1819 1820
1820#define INTEL_PCH_DEVICE_ID_MASK 0xff00 1821#define INTEL_PCH_DEVICE_ID_MASK 0xff00
1821#define INTEL_PCH_IBX_DEVICE_ID_TYPE 0x3b00 1822#define INTEL_PCH_IBX_DEVICE_ID_TYPE 0x3b00
diff --git a/drivers/gpu/drm/i915/intel_bios.c b/drivers/gpu/drm/i915/intel_bios.c
index 6dd622d733b9..e4fba39631a5 100644
--- a/drivers/gpu/drm/i915/intel_bios.c
+++ b/drivers/gpu/drm/i915/intel_bios.c
@@ -790,7 +790,12 @@ init_vbt_defaults(struct drm_i915_private *dev_priv)
790 790
791 /* Default to using SSC */ 791 /* Default to using SSC */
792 dev_priv->vbt.lvds_use_ssc = 1; 792 dev_priv->vbt.lvds_use_ssc = 1;
793 dev_priv->vbt.lvds_ssc_freq = intel_bios_ssc_frequency(dev, 1); 793 /*
794 * Core/SandyBridge/IvyBridge use alternative (120MHz) reference
795 * clock for LVDS.
796 */
797 dev_priv->vbt.lvds_ssc_freq = intel_bios_ssc_frequency(dev,
798 !HAS_PCH_SPLIT(dev));
794 DRM_DEBUG_KMS("Set default to SSC at %dMHz\n", dev_priv->vbt.lvds_ssc_freq); 799 DRM_DEBUG_KMS("Set default to SSC at %dMHz\n", dev_priv->vbt.lvds_ssc_freq);
795 800
796 for (port = PORT_A; port < I915_MAX_PORTS; port++) { 801 for (port = PORT_A; port < I915_MAX_PORTS; port++) {
diff --git a/drivers/gpu/drm/i915/intel_ddi.c b/drivers/gpu/drm/i915/intel_ddi.c
index 1591576a6101..330077bcd0bd 100644
--- a/drivers/gpu/drm/i915/intel_ddi.c
+++ b/drivers/gpu/drm/i915/intel_ddi.c
@@ -1406,6 +1406,26 @@ void intel_ddi_get_config(struct intel_encoder *encoder,
1406 default: 1406 default:
1407 break; 1407 break;
1408 } 1408 }
1409
1410 if (encoder->type == INTEL_OUTPUT_EDP && dev_priv->vbt.edp_bpp &&
1411 pipe_config->pipe_bpp > dev_priv->vbt.edp_bpp) {
1412 /*
1413 * This is a big fat ugly hack.
1414 *
1415 * Some machines in UEFI boot mode provide us a VBT that has 18
1416 * bpp and 1.62 GHz link bandwidth for eDP, which for reasons
1417 * unknown we fail to light up. Yet the same BIOS boots up with
1418 * 24 bpp and 2.7 GHz link. Use the same bpp as the BIOS uses as
1419 * max, not what it tells us to use.
1420 *
1421 * Note: This will still be broken if the eDP panel is not lit
1422 * up by the BIOS, and thus we can't get the mode at module
1423 * load.
1424 */
1425 DRM_DEBUG_KMS("pipe has %d bpp for eDP panel, overriding BIOS-provided max %d bpp\n",
1426 pipe_config->pipe_bpp, dev_priv->vbt.edp_bpp);
1427 dev_priv->vbt.edp_bpp = pipe_config->pipe_bpp;
1428 }
1409} 1429}
1410 1430
1411static void intel_ddi_destroy(struct drm_encoder *encoder) 1431static void intel_ddi_destroy(struct drm_encoder *encoder)
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 3cddd508d110..7ec8b488bb1d 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -6518,6 +6518,9 @@ static void __hsw_disable_package_c8(struct drm_i915_private *dev_priv)
6518 6518
6519void hsw_enable_package_c8(struct drm_i915_private *dev_priv) 6519void hsw_enable_package_c8(struct drm_i915_private *dev_priv)
6520{ 6520{
6521 if (!HAS_PC8(dev_priv->dev))
6522 return;
6523
6521 mutex_lock(&dev_priv->pc8.lock); 6524 mutex_lock(&dev_priv->pc8.lock);
6522 __hsw_enable_package_c8(dev_priv); 6525 __hsw_enable_package_c8(dev_priv);
6523 mutex_unlock(&dev_priv->pc8.lock); 6526 mutex_unlock(&dev_priv->pc8.lock);
@@ -6525,6 +6528,9 @@ void hsw_enable_package_c8(struct drm_i915_private *dev_priv)
6525 6528
6526void hsw_disable_package_c8(struct drm_i915_private *dev_priv) 6529void hsw_disable_package_c8(struct drm_i915_private *dev_priv)
6527{ 6530{
6531 if (!HAS_PC8(dev_priv->dev))
6532 return;
6533
6528 mutex_lock(&dev_priv->pc8.lock); 6534 mutex_lock(&dev_priv->pc8.lock);
6529 __hsw_disable_package_c8(dev_priv); 6535 __hsw_disable_package_c8(dev_priv);
6530 mutex_unlock(&dev_priv->pc8.lock); 6536 mutex_unlock(&dev_priv->pc8.lock);
@@ -6562,6 +6568,9 @@ static void hsw_update_package_c8(struct drm_device *dev)
6562 struct drm_i915_private *dev_priv = dev->dev_private; 6568 struct drm_i915_private *dev_priv = dev->dev_private;
6563 bool allow; 6569 bool allow;
6564 6570
6571 if (!HAS_PC8(dev_priv->dev))
6572 return;
6573
6565 if (!i915_enable_pc8) 6574 if (!i915_enable_pc8)
6566 return; 6575 return;
6567 6576
@@ -6585,18 +6594,28 @@ done:
6585 6594
6586static void hsw_package_c8_gpu_idle(struct drm_i915_private *dev_priv) 6595static void hsw_package_c8_gpu_idle(struct drm_i915_private *dev_priv)
6587{ 6596{
6597 if (!HAS_PC8(dev_priv->dev))
6598 return;
6599
6600 mutex_lock(&dev_priv->pc8.lock);
6588 if (!dev_priv->pc8.gpu_idle) { 6601 if (!dev_priv->pc8.gpu_idle) {
6589 dev_priv->pc8.gpu_idle = true; 6602 dev_priv->pc8.gpu_idle = true;
6590 hsw_enable_package_c8(dev_priv); 6603 __hsw_enable_package_c8(dev_priv);
6591 } 6604 }
6605 mutex_unlock(&dev_priv->pc8.lock);
6592} 6606}
6593 6607
6594static void hsw_package_c8_gpu_busy(struct drm_i915_private *dev_priv) 6608static void hsw_package_c8_gpu_busy(struct drm_i915_private *dev_priv)
6595{ 6609{
6610 if (!HAS_PC8(dev_priv->dev))
6611 return;
6612
6613 mutex_lock(&dev_priv->pc8.lock);
6596 if (dev_priv->pc8.gpu_idle) { 6614 if (dev_priv->pc8.gpu_idle) {
6597 dev_priv->pc8.gpu_idle = false; 6615 dev_priv->pc8.gpu_idle = false;
6598 hsw_disable_package_c8(dev_priv); 6616 __hsw_disable_package_c8(dev_priv);
6599 } 6617 }
6618 mutex_unlock(&dev_priv->pc8.lock);
6600} 6619}
6601 6620
6602#define for_each_power_domain(domain, mask) \ 6621#define for_each_power_domain(domain, mask) \
@@ -7184,7 +7203,9 @@ static void i9xx_update_cursor(struct drm_crtc *crtc, u32 base)
7184 intel_crtc->cursor_visible = visible; 7203 intel_crtc->cursor_visible = visible;
7185 } 7204 }
7186 /* and commit changes on next vblank */ 7205 /* and commit changes on next vblank */
7206 POSTING_READ(CURCNTR(pipe));
7187 I915_WRITE(CURBASE(pipe), base); 7207 I915_WRITE(CURBASE(pipe), base);
7208 POSTING_READ(CURBASE(pipe));
7188} 7209}
7189 7210
7190static void ivb_update_cursor(struct drm_crtc *crtc, u32 base) 7211static void ivb_update_cursor(struct drm_crtc *crtc, u32 base)
@@ -7213,7 +7234,9 @@ static void ivb_update_cursor(struct drm_crtc *crtc, u32 base)
7213 intel_crtc->cursor_visible = visible; 7234 intel_crtc->cursor_visible = visible;
7214 } 7235 }
7215 /* and commit changes on next vblank */ 7236 /* and commit changes on next vblank */
7237 POSTING_READ(CURCNTR_IVB(pipe));
7216 I915_WRITE(CURBASE_IVB(pipe), base); 7238 I915_WRITE(CURBASE_IVB(pipe), base);
7239 POSTING_READ(CURBASE_IVB(pipe));
7217} 7240}
7218 7241
7219/* If no-part of the cursor is visible on the framebuffer, then the GPU may hang... */ 7242/* If no-part of the cursor is visible on the framebuffer, then the GPU may hang... */
@@ -9248,8 +9271,7 @@ check_crtc_state(struct drm_device *dev)
9248 enum pipe pipe; 9271 enum pipe pipe;
9249 if (encoder->base.crtc != &crtc->base) 9272 if (encoder->base.crtc != &crtc->base)
9250 continue; 9273 continue;
9251 if (encoder->get_config && 9274 if (encoder->get_hw_state(encoder, &pipe))
9252 encoder->get_hw_state(encoder, &pipe))
9253 encoder->get_config(encoder, &pipe_config); 9275 encoder->get_config(encoder, &pipe_config);
9254 } 9276 }
9255 9277
@@ -10909,8 +10931,7 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev)
10909 if (encoder->get_hw_state(encoder, &pipe)) { 10931 if (encoder->get_hw_state(encoder, &pipe)) {
10910 crtc = to_intel_crtc(dev_priv->pipe_to_crtc_mapping[pipe]); 10932 crtc = to_intel_crtc(dev_priv->pipe_to_crtc_mapping[pipe]);
10911 encoder->base.crtc = &crtc->base; 10933 encoder->base.crtc = &crtc->base;
10912 if (encoder->get_config) 10934 encoder->get_config(encoder, &crtc->config);
10913 encoder->get_config(encoder, &crtc->config);
10914 } else { 10935 } else {
10915 encoder->base.crtc = NULL; 10936 encoder->base.crtc = NULL;
10916 } 10937 }
diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c
index eb8139da9763..0b2e842fef01 100644
--- a/drivers/gpu/drm/i915/intel_dp.c
+++ b/drivers/gpu/drm/i915/intel_dp.c
@@ -1774,7 +1774,7 @@ static void intel_disable_dp(struct intel_encoder *encoder)
1774 * ensure that we have vdd while we switch off the panel. */ 1774 * ensure that we have vdd while we switch off the panel. */
1775 ironlake_edp_panel_vdd_on(intel_dp); 1775 ironlake_edp_panel_vdd_on(intel_dp);
1776 ironlake_edp_backlight_off(intel_dp); 1776 ironlake_edp_backlight_off(intel_dp);
1777 intel_dp_sink_dpms(intel_dp, DRM_MODE_DPMS_ON); 1777 intel_dp_sink_dpms(intel_dp, DRM_MODE_DPMS_OFF);
1778 ironlake_edp_panel_off(intel_dp); 1778 ironlake_edp_panel_off(intel_dp);
1779 1779
1780 /* cpu edp my only be disable _after_ the cpu pipe/plane is disabled. */ 1780 /* cpu edp my only be disable _after_ the cpu pipe/plane is disabled. */
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index 0a07d7c9cafc..caf2ee4e5441 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -1625,7 +1625,7 @@ static void i9xx_update_wm(struct drm_crtc *unused_crtc)
1625 &to_intel_crtc(enabled)->config.adjusted_mode; 1625 &to_intel_crtc(enabled)->config.adjusted_mode;
1626 int clock = adjusted_mode->crtc_clock; 1626 int clock = adjusted_mode->crtc_clock;
1627 int htotal = adjusted_mode->htotal; 1627 int htotal = adjusted_mode->htotal;
1628 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w; 1628 int hdisplay = to_intel_crtc(enabled)->config.pipe_src_w;
1629 int pixel_size = enabled->fb->bits_per_pixel / 8; 1629 int pixel_size = enabled->fb->bits_per_pixel / 8;
1630 unsigned long line_time_us; 1630 unsigned long line_time_us;
1631 int entries; 1631 int entries;
@@ -3888,7 +3888,7 @@ static void gen6_enable_rps(struct drm_device *dev)
3888 3888
3889 I915_WRITE(GEN6_RC_SLEEP, 0); 3889 I915_WRITE(GEN6_RC_SLEEP, 0);
3890 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); 3890 I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
3891 if (INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) 3891 if (IS_IVYBRIDGE(dev))
3892 I915_WRITE(GEN6_RC6_THRESHOLD, 125000); 3892 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
3893 else 3893 else
3894 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); 3894 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
diff --git a/drivers/gpu/drm/i915/intel_tv.c b/drivers/gpu/drm/i915/intel_tv.c
index 18c406246a2d..22cf0f4ba248 100644
--- a/drivers/gpu/drm/i915/intel_tv.c
+++ b/drivers/gpu/drm/i915/intel_tv.c
@@ -902,6 +902,13 @@ intel_tv_mode_valid(struct drm_connector *connector,
902} 902}
903 903
904 904
905static void
906intel_tv_get_config(struct intel_encoder *encoder,
907 struct intel_crtc_config *pipe_config)
908{
909 pipe_config->adjusted_mode.crtc_clock = pipe_config->port_clock;
910}
911
905static bool 912static bool
906intel_tv_compute_config(struct intel_encoder *encoder, 913intel_tv_compute_config(struct intel_encoder *encoder,
907 struct intel_crtc_config *pipe_config) 914 struct intel_crtc_config *pipe_config)
@@ -1621,6 +1628,7 @@ intel_tv_init(struct drm_device *dev)
1621 DRM_MODE_ENCODER_TVDAC); 1628 DRM_MODE_ENCODER_TVDAC);
1622 1629
1623 intel_encoder->compute_config = intel_tv_compute_config; 1630 intel_encoder->compute_config = intel_tv_compute_config;
1631 intel_encoder->get_config = intel_tv_get_config;
1624 intel_encoder->mode_set = intel_tv_mode_set; 1632 intel_encoder->mode_set = intel_tv_mode_set;
1625 intel_encoder->enable = intel_enable_tv; 1633 intel_encoder->enable = intel_enable_tv;
1626 intel_encoder->disable = intel_disable_tv; 1634 intel_encoder->disable = intel_disable_tv;
diff --git a/drivers/gpu/drm/i915/intel_uncore.c b/drivers/gpu/drm/i915/intel_uncore.c
index f9883ceff946..0b02078a0b84 100644
--- a/drivers/gpu/drm/i915/intel_uncore.c
+++ b/drivers/gpu/drm/i915/intel_uncore.c
@@ -217,6 +217,19 @@ static void gen6_force_wake_work(struct work_struct *work)
217 spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags); 217 spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
218} 218}
219 219
220static void intel_uncore_forcewake_reset(struct drm_device *dev)
221{
222 struct drm_i915_private *dev_priv = dev->dev_private;
223
224 if (IS_VALLEYVIEW(dev)) {
225 vlv_force_wake_reset(dev_priv);
226 } else if (INTEL_INFO(dev)->gen >= 6) {
227 __gen6_gt_force_wake_reset(dev_priv);
228 if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
229 __gen6_gt_force_wake_mt_reset(dev_priv);
230 }
231}
232
220void intel_uncore_early_sanitize(struct drm_device *dev) 233void intel_uncore_early_sanitize(struct drm_device *dev)
221{ 234{
222 struct drm_i915_private *dev_priv = dev->dev_private; 235 struct drm_i915_private *dev_priv = dev->dev_private;
@@ -234,19 +247,8 @@ void intel_uncore_early_sanitize(struct drm_device *dev)
234 dev_priv->ellc_size = 128; 247 dev_priv->ellc_size = 128;
235 DRM_INFO("Found %zuMB of eLLC\n", dev_priv->ellc_size); 248 DRM_INFO("Found %zuMB of eLLC\n", dev_priv->ellc_size);
236 } 249 }
237}
238 250
239static void intel_uncore_forcewake_reset(struct drm_device *dev) 251 intel_uncore_forcewake_reset(dev);
240{
241 struct drm_i915_private *dev_priv = dev->dev_private;
242
243 if (IS_VALLEYVIEW(dev)) {
244 vlv_force_wake_reset(dev_priv);
245 } else if (INTEL_INFO(dev)->gen >= 6) {
246 __gen6_gt_force_wake_reset(dev_priv);
247 if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
248 __gen6_gt_force_wake_mt_reset(dev_priv);
249 }
250} 252}
251 253
252void intel_uncore_sanitize(struct drm_device *dev) 254void intel_uncore_sanitize(struct drm_device *dev)
diff --git a/drivers/gpu/drm/radeon/atombios_i2c.c b/drivers/gpu/drm/radeon/atombios_i2c.c
index deaf98cdca3a..0652ee0a2098 100644
--- a/drivers/gpu/drm/radeon/atombios_i2c.c
+++ b/drivers/gpu/drm/radeon/atombios_i2c.c
@@ -56,8 +56,10 @@ static int radeon_process_i2c_ch(struct radeon_i2c_chan *chan,
56 return -EINVAL; 56 return -EINVAL;
57 } 57 }
58 args.ucRegIndex = buf[0]; 58 args.ucRegIndex = buf[0];
59 if (num > 1) 59 if (num > 1) {
60 memcpy(&out, &buf[1], num - 1); 60 num--;
61 memcpy(&out, &buf[1], num);
62 }
61 args.lpI2CDataOut = cpu_to_le16(out); 63 args.lpI2CDataOut = cpu_to_le16(out);
62 } else { 64 } else {
63 if (num > ATOM_MAX_HW_I2C_READ) { 65 if (num > ATOM_MAX_HW_I2C_READ) {
diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c
index ae92aa041c6a..b43a3a3c9067 100644
--- a/drivers/gpu/drm/radeon/cik.c
+++ b/drivers/gpu/drm/radeon/cik.c
@@ -1560,17 +1560,17 @@ u32 cik_get_xclk(struct radeon_device *rdev)
1560 * cik_mm_rdoorbell - read a doorbell dword 1560 * cik_mm_rdoorbell - read a doorbell dword
1561 * 1561 *
1562 * @rdev: radeon_device pointer 1562 * @rdev: radeon_device pointer
1563 * @offset: byte offset into the aperture 1563 * @index: doorbell index
1564 * 1564 *
1565 * Returns the value in the doorbell aperture at the 1565 * Returns the value in the doorbell aperture at the
1566 * requested offset (CIK). 1566 * requested doorbell index (CIK).
1567 */ 1567 */
1568u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset) 1568u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 index)
1569{ 1569{
1570 if (offset < rdev->doorbell.size) { 1570 if (index < rdev->doorbell.num_doorbells) {
1571 return readl(((void __iomem *)rdev->doorbell.ptr) + offset); 1571 return readl(rdev->doorbell.ptr + index);
1572 } else { 1572 } else {
1573 DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", offset); 1573 DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
1574 return 0; 1574 return 0;
1575 } 1575 }
1576} 1576}
@@ -1579,18 +1579,18 @@ u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset)
1579 * cik_mm_wdoorbell - write a doorbell dword 1579 * cik_mm_wdoorbell - write a doorbell dword
1580 * 1580 *
1581 * @rdev: radeon_device pointer 1581 * @rdev: radeon_device pointer
1582 * @offset: byte offset into the aperture 1582 * @index: doorbell index
1583 * @v: value to write 1583 * @v: value to write
1584 * 1584 *
1585 * Writes @v to the doorbell aperture at the 1585 * Writes @v to the doorbell aperture at the
1586 * requested offset (CIK). 1586 * requested doorbell index (CIK).
1587 */ 1587 */
1588void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v) 1588void cik_mm_wdoorbell(struct radeon_device *rdev, u32 index, u32 v)
1589{ 1589{
1590 if (offset < rdev->doorbell.size) { 1590 if (index < rdev->doorbell.num_doorbells) {
1591 writel(v, ((void __iomem *)rdev->doorbell.ptr) + offset); 1591 writel(v, rdev->doorbell.ptr + index);
1592 } else { 1592 } else {
1593 DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", offset); 1593 DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
1594 } 1594 }
1595} 1595}
1596 1596
@@ -2427,6 +2427,7 @@ static void cik_tiling_mode_table_init(struct radeon_device *rdev)
2427 gb_tile_moden = 0; 2427 gb_tile_moden = 0;
2428 break; 2428 break;
2429 } 2429 }
2430 rdev->config.cik.macrotile_mode_array[reg_offset] = gb_tile_moden;
2430 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden); 2431 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2431 } 2432 }
2432 } else if (num_pipe_configs == 4) { 2433 } else if (num_pipe_configs == 4) {
@@ -2773,6 +2774,7 @@ static void cik_tiling_mode_table_init(struct radeon_device *rdev)
2773 gb_tile_moden = 0; 2774 gb_tile_moden = 0;
2774 break; 2775 break;
2775 } 2776 }
2777 rdev->config.cik.macrotile_mode_array[reg_offset] = gb_tile_moden;
2776 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden); 2778 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2777 } 2779 }
2778 } else if (num_pipe_configs == 2) { 2780 } else if (num_pipe_configs == 2) {
@@ -2990,6 +2992,7 @@ static void cik_tiling_mode_table_init(struct radeon_device *rdev)
2990 gb_tile_moden = 0; 2992 gb_tile_moden = 0;
2991 break; 2993 break;
2992 } 2994 }
2995 rdev->config.cik.macrotile_mode_array[reg_offset] = gb_tile_moden;
2993 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden); 2996 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2994 } 2997 }
2995 } else 2998 } else
@@ -3556,17 +3559,24 @@ void cik_fence_compute_ring_emit(struct radeon_device *rdev,
3556 radeon_ring_write(ring, 0); 3559 radeon_ring_write(ring, 0);
3557} 3560}
3558 3561
3559void cik_semaphore_ring_emit(struct radeon_device *rdev, 3562bool cik_semaphore_ring_emit(struct radeon_device *rdev,
3560 struct radeon_ring *ring, 3563 struct radeon_ring *ring,
3561 struct radeon_semaphore *semaphore, 3564 struct radeon_semaphore *semaphore,
3562 bool emit_wait) 3565 bool emit_wait)
3563{ 3566{
3567/* TODO: figure out why semaphore cause lockups */
3568#if 0
3564 uint64_t addr = semaphore->gpu_addr; 3569 uint64_t addr = semaphore->gpu_addr;
3565 unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL; 3570 unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL;
3566 3571
3567 radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1)); 3572 radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1));
3568 radeon_ring_write(ring, addr & 0xffffffff); 3573 radeon_ring_write(ring, addr & 0xffffffff);
3569 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel); 3574 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel);
3575
3576 return true;
3577#else
3578 return false;
3579#endif
3570} 3580}
3571 3581
3572/** 3582/**
@@ -3609,13 +3619,8 @@ int cik_copy_cpdma(struct radeon_device *rdev,
3609 return r; 3619 return r;
3610 } 3620 }
3611 3621
3612 if (radeon_fence_need_sync(*fence, ring->idx)) { 3622 radeon_semaphore_sync_to(sem, *fence);
3613 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 3623 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
3614 ring->idx);
3615 radeon_fence_note_sync(*fence, ring->idx);
3616 } else {
3617 radeon_semaphore_free(rdev, &sem, NULL);
3618 }
3619 3624
3620 for (i = 0; i < num_loops; i++) { 3625 for (i = 0; i < num_loops; i++) {
3621 cur_size_in_bytes = size_in_bytes; 3626 cur_size_in_bytes = size_in_bytes;
@@ -4052,7 +4057,7 @@ void cik_compute_ring_set_wptr(struct radeon_device *rdev,
4052 struct radeon_ring *ring) 4057 struct radeon_ring *ring)
4053{ 4058{
4054 rdev->wb.wb[ring->wptr_offs/4] = cpu_to_le32(ring->wptr); 4059 rdev->wb.wb[ring->wptr_offs/4] = cpu_to_le32(ring->wptr);
4055 WDOORBELL32(ring->doorbell_offset, ring->wptr); 4060 WDOORBELL32(ring->doorbell_index, ring->wptr);
4056} 4061}
4057 4062
4058/** 4063/**
@@ -4393,10 +4398,6 @@ static int cik_cp_compute_resume(struct radeon_device *rdev)
4393 return r; 4398 return r;
4394 } 4399 }
4395 4400
4396 /* doorbell offset */
4397 rdev->ring[idx].doorbell_offset =
4398 (rdev->ring[idx].doorbell_page_num * PAGE_SIZE) + 0;
4399
4400 /* init the mqd struct */ 4401 /* init the mqd struct */
4401 memset(buf, 0, sizeof(struct bonaire_mqd)); 4402 memset(buf, 0, sizeof(struct bonaire_mqd));
4402 4403
@@ -4508,7 +4509,7 @@ static int cik_cp_compute_resume(struct radeon_device *rdev)
4508 RREG32(CP_HQD_PQ_DOORBELL_CONTROL); 4509 RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
4509 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_OFFSET_MASK; 4510 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_OFFSET_MASK;
4510 mqd->queue_state.cp_hqd_pq_doorbell_control |= 4511 mqd->queue_state.cp_hqd_pq_doorbell_control |=
4511 DOORBELL_OFFSET(rdev->ring[idx].doorbell_offset / 4); 4512 DOORBELL_OFFSET(rdev->ring[idx].doorbell_index);
4512 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN; 4513 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
4513 mqd->queue_state.cp_hqd_pq_doorbell_control &= 4514 mqd->queue_state.cp_hqd_pq_doorbell_control &=
4514 ~(DOORBELL_SOURCE | DOORBELL_HIT); 4515 ~(DOORBELL_SOURCE | DOORBELL_HIT);
@@ -7839,14 +7840,14 @@ int cik_init(struct radeon_device *rdev)
7839 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]; 7840 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7840 ring->ring_obj = NULL; 7841 ring->ring_obj = NULL;
7841 r600_ring_init(rdev, ring, 1024 * 1024); 7842 r600_ring_init(rdev, ring, 1024 * 1024);
7842 r = radeon_doorbell_get(rdev, &ring->doorbell_page_num); 7843 r = radeon_doorbell_get(rdev, &ring->doorbell_index);
7843 if (r) 7844 if (r)
7844 return r; 7845 return r;
7845 7846
7846 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]; 7847 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7847 ring->ring_obj = NULL; 7848 ring->ring_obj = NULL;
7848 r600_ring_init(rdev, ring, 1024 * 1024); 7849 r600_ring_init(rdev, ring, 1024 * 1024);
7849 r = radeon_doorbell_get(rdev, &ring->doorbell_page_num); 7850 r = radeon_doorbell_get(rdev, &ring->doorbell_index);
7850 if (r) 7851 if (r)
7851 return r; 7852 return r;
7852 7853
diff --git a/drivers/gpu/drm/radeon/cik_sdma.c b/drivers/gpu/drm/radeon/cik_sdma.c
index 9c9529de20ee..0300727a4f70 100644
--- a/drivers/gpu/drm/radeon/cik_sdma.c
+++ b/drivers/gpu/drm/radeon/cik_sdma.c
@@ -130,7 +130,7 @@ void cik_sdma_fence_ring_emit(struct radeon_device *rdev,
130 * Add a DMA semaphore packet to the ring wait on or signal 130 * Add a DMA semaphore packet to the ring wait on or signal
131 * other rings (CIK). 131 * other rings (CIK).
132 */ 132 */
133void cik_sdma_semaphore_ring_emit(struct radeon_device *rdev, 133bool cik_sdma_semaphore_ring_emit(struct radeon_device *rdev,
134 struct radeon_ring *ring, 134 struct radeon_ring *ring,
135 struct radeon_semaphore *semaphore, 135 struct radeon_semaphore *semaphore,
136 bool emit_wait) 136 bool emit_wait)
@@ -141,6 +141,8 @@ void cik_sdma_semaphore_ring_emit(struct radeon_device *rdev,
141 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits)); 141 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits));
142 radeon_ring_write(ring, addr & 0xfffffff8); 142 radeon_ring_write(ring, addr & 0xfffffff8);
143 radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff); 143 radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
144
145 return true;
144} 146}
145 147
146/** 148/**
@@ -443,13 +445,8 @@ int cik_copy_dma(struct radeon_device *rdev,
443 return r; 445 return r;
444 } 446 }
445 447
446 if (radeon_fence_need_sync(*fence, ring->idx)) { 448 radeon_semaphore_sync_to(sem, *fence);
447 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 449 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
448 ring->idx);
449 radeon_fence_note_sync(*fence, ring->idx);
450 } else {
451 radeon_semaphore_free(rdev, &sem, NULL);
452 }
453 450
454 for (i = 0; i < num_loops; i++) { 451 for (i = 0; i < num_loops; i++) {
455 cur_size_in_bytes = size_in_bytes; 452 cur_size_in_bytes = size_in_bytes;
diff --git a/drivers/gpu/drm/radeon/cypress_dpm.c b/drivers/gpu/drm/radeon/cypress_dpm.c
index 91bb470de0a3..920e1e4a52c5 100644
--- a/drivers/gpu/drm/radeon/cypress_dpm.c
+++ b/drivers/gpu/drm/radeon/cypress_dpm.c
@@ -299,7 +299,9 @@ void cypress_program_response_times(struct radeon_device *rdev)
299static int cypress_pcie_performance_request(struct radeon_device *rdev, 299static int cypress_pcie_performance_request(struct radeon_device *rdev,
300 u8 perf_req, bool advertise) 300 u8 perf_req, bool advertise)
301{ 301{
302#if defined(CONFIG_ACPI)
302 struct evergreen_power_info *eg_pi = evergreen_get_pi(rdev); 303 struct evergreen_power_info *eg_pi = evergreen_get_pi(rdev);
304#endif
303 u32 tmp; 305 u32 tmp;
304 306
305 udelay(10); 307 udelay(10);
diff --git a/drivers/gpu/drm/radeon/evergreen_dma.c b/drivers/gpu/drm/radeon/evergreen_dma.c
index 6a0656d00ed0..a37b54436382 100644
--- a/drivers/gpu/drm/radeon/evergreen_dma.c
+++ b/drivers/gpu/drm/radeon/evergreen_dma.c
@@ -131,13 +131,8 @@ int evergreen_copy_dma(struct radeon_device *rdev,
131 return r; 131 return r;
132 } 132 }
133 133
134 if (radeon_fence_need_sync(*fence, ring->idx)) { 134 radeon_semaphore_sync_to(sem, *fence);
135 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 135 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
136 ring->idx);
137 radeon_fence_note_sync(*fence, ring->idx);
138 } else {
139 radeon_semaphore_free(rdev, &sem, NULL);
140 }
141 136
142 for (i = 0; i < num_loops; i++) { 137 for (i = 0; i < num_loops; i++) {
143 cur_size_in_dw = size_in_dw; 138 cur_size_in_dw = size_in_dw;
diff --git a/drivers/gpu/drm/radeon/ni_dpm.c b/drivers/gpu/drm/radeon/ni_dpm.c
index f26339028154..cdc003085a76 100644
--- a/drivers/gpu/drm/radeon/ni_dpm.c
+++ b/drivers/gpu/drm/radeon/ni_dpm.c
@@ -3445,9 +3445,9 @@ static int ni_enable_smc_cac(struct radeon_device *rdev,
3445static int ni_pcie_performance_request(struct radeon_device *rdev, 3445static int ni_pcie_performance_request(struct radeon_device *rdev,
3446 u8 perf_req, bool advertise) 3446 u8 perf_req, bool advertise)
3447{ 3447{
3448#if defined(CONFIG_ACPI)
3448 struct evergreen_power_info *eg_pi = evergreen_get_pi(rdev); 3449 struct evergreen_power_info *eg_pi = evergreen_get_pi(rdev);
3449 3450
3450#if defined(CONFIG_ACPI)
3451 if ((perf_req == PCIE_PERF_REQ_PECI_GEN1) || 3451 if ((perf_req == PCIE_PERF_REQ_PECI_GEN1) ||
3452 (perf_req == PCIE_PERF_REQ_PECI_GEN2)) { 3452 (perf_req == PCIE_PERF_REQ_PECI_GEN2)) {
3453 if (eg_pi->pcie_performance_request_registered == false) 3453 if (eg_pi->pcie_performance_request_registered == false)
diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index 784983d78158..10abc4d5a6cc 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -869,13 +869,14 @@ void r100_fence_ring_emit(struct radeon_device *rdev,
869 radeon_ring_write(ring, RADEON_SW_INT_FIRE); 869 radeon_ring_write(ring, RADEON_SW_INT_FIRE);
870} 870}
871 871
872void r100_semaphore_ring_emit(struct radeon_device *rdev, 872bool r100_semaphore_ring_emit(struct radeon_device *rdev,
873 struct radeon_ring *ring, 873 struct radeon_ring *ring,
874 struct radeon_semaphore *semaphore, 874 struct radeon_semaphore *semaphore,
875 bool emit_wait) 875 bool emit_wait)
876{ 876{
877 /* Unused on older asics, since we don't have semaphores or multiple rings */ 877 /* Unused on older asics, since we don't have semaphores or multiple rings */
878 BUG(); 878 BUG();
879 return false;
879} 880}
880 881
881int r100_copy_blit(struct radeon_device *rdev, 882int r100_copy_blit(struct radeon_device *rdev,
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index 4e609e8a8d2b..9ad06732a78b 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -2650,7 +2650,7 @@ void r600_fence_ring_emit(struct radeon_device *rdev,
2650 } 2650 }
2651} 2651}
2652 2652
2653void r600_semaphore_ring_emit(struct radeon_device *rdev, 2653bool r600_semaphore_ring_emit(struct radeon_device *rdev,
2654 struct radeon_ring *ring, 2654 struct radeon_ring *ring,
2655 struct radeon_semaphore *semaphore, 2655 struct radeon_semaphore *semaphore,
2656 bool emit_wait) 2656 bool emit_wait)
@@ -2664,6 +2664,8 @@ void r600_semaphore_ring_emit(struct radeon_device *rdev,
2664 radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1)); 2664 radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1));
2665 radeon_ring_write(ring, addr & 0xffffffff); 2665 radeon_ring_write(ring, addr & 0xffffffff);
2666 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | sel); 2666 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | sel);
2667
2668 return true;
2667} 2669}
2668 2670
2669/** 2671/**
@@ -2706,13 +2708,8 @@ int r600_copy_cpdma(struct radeon_device *rdev,
2706 return r; 2708 return r;
2707 } 2709 }
2708 2710
2709 if (radeon_fence_need_sync(*fence, ring->idx)) { 2711 radeon_semaphore_sync_to(sem, *fence);
2710 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 2712 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
2711 ring->idx);
2712 radeon_fence_note_sync(*fence, ring->idx);
2713 } else {
2714 radeon_semaphore_free(rdev, &sem, NULL);
2715 }
2716 2713
2717 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 2714 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2718 radeon_ring_write(ring, (WAIT_UNTIL - PACKET3_SET_CONFIG_REG_OFFSET) >> 2); 2715 radeon_ring_write(ring, (WAIT_UNTIL - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
diff --git a/drivers/gpu/drm/radeon/r600_dma.c b/drivers/gpu/drm/radeon/r600_dma.c
index 3b317456512a..7844d15c139f 100644
--- a/drivers/gpu/drm/radeon/r600_dma.c
+++ b/drivers/gpu/drm/radeon/r600_dma.c
@@ -311,7 +311,7 @@ void r600_dma_fence_ring_emit(struct radeon_device *rdev,
311 * Add a DMA semaphore packet to the ring wait on or signal 311 * Add a DMA semaphore packet to the ring wait on or signal
312 * other rings (r6xx-SI). 312 * other rings (r6xx-SI).
313 */ 313 */
314void r600_dma_semaphore_ring_emit(struct radeon_device *rdev, 314bool r600_dma_semaphore_ring_emit(struct radeon_device *rdev,
315 struct radeon_ring *ring, 315 struct radeon_ring *ring,
316 struct radeon_semaphore *semaphore, 316 struct radeon_semaphore *semaphore,
317 bool emit_wait) 317 bool emit_wait)
@@ -322,6 +322,8 @@ void r600_dma_semaphore_ring_emit(struct radeon_device *rdev,
322 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SEMAPHORE, 0, s, 0)); 322 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SEMAPHORE, 0, s, 0));
323 radeon_ring_write(ring, addr & 0xfffffffc); 323 radeon_ring_write(ring, addr & 0xfffffffc);
324 radeon_ring_write(ring, upper_32_bits(addr) & 0xff); 324 radeon_ring_write(ring, upper_32_bits(addr) & 0xff);
325
326 return true;
325} 327}
326 328
327/** 329/**
@@ -462,13 +464,8 @@ int r600_copy_dma(struct radeon_device *rdev,
462 return r; 464 return r;
463 } 465 }
464 466
465 if (radeon_fence_need_sync(*fence, ring->idx)) { 467 radeon_semaphore_sync_to(sem, *fence);
466 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 468 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
467 ring->idx);
468 radeon_fence_note_sync(*fence, ring->idx);
469 } else {
470 radeon_semaphore_free(rdev, &sem, NULL);
471 }
472 469
473 for (i = 0; i < num_loops; i++) { 470 for (i = 0; i < num_loops; i++) {
474 cur_size_in_dw = size_in_dw; 471 cur_size_in_dw = size_in_dw;
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index b9ee99258602..ecf2a3960c07 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -348,6 +348,7 @@ int radeon_fence_emit(struct radeon_device *rdev, struct radeon_fence **fence, i
348void radeon_fence_process(struct radeon_device *rdev, int ring); 348void radeon_fence_process(struct radeon_device *rdev, int ring);
349bool radeon_fence_signaled(struct radeon_fence *fence); 349bool radeon_fence_signaled(struct radeon_fence *fence);
350int radeon_fence_wait(struct radeon_fence *fence, bool interruptible); 350int radeon_fence_wait(struct radeon_fence *fence, bool interruptible);
351int radeon_fence_wait_locked(struct radeon_fence *fence);
351int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring); 352int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring);
352int radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring); 353int radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring);
353int radeon_fence_wait_any(struct radeon_device *rdev, 354int radeon_fence_wait_any(struct radeon_device *rdev,
@@ -548,17 +549,20 @@ struct radeon_semaphore {
548 struct radeon_sa_bo *sa_bo; 549 struct radeon_sa_bo *sa_bo;
549 signed waiters; 550 signed waiters;
550 uint64_t gpu_addr; 551 uint64_t gpu_addr;
552 struct radeon_fence *sync_to[RADEON_NUM_RINGS];
551}; 553};
552 554
553int radeon_semaphore_create(struct radeon_device *rdev, 555int radeon_semaphore_create(struct radeon_device *rdev,
554 struct radeon_semaphore **semaphore); 556 struct radeon_semaphore **semaphore);
555void radeon_semaphore_emit_signal(struct radeon_device *rdev, int ring, 557bool radeon_semaphore_emit_signal(struct radeon_device *rdev, int ring,
556 struct radeon_semaphore *semaphore); 558 struct radeon_semaphore *semaphore);
557void radeon_semaphore_emit_wait(struct radeon_device *rdev, int ring, 559bool radeon_semaphore_emit_wait(struct radeon_device *rdev, int ring,
558 struct radeon_semaphore *semaphore); 560 struct radeon_semaphore *semaphore);
561void radeon_semaphore_sync_to(struct radeon_semaphore *semaphore,
562 struct radeon_fence *fence);
559int radeon_semaphore_sync_rings(struct radeon_device *rdev, 563int radeon_semaphore_sync_rings(struct radeon_device *rdev,
560 struct radeon_semaphore *semaphore, 564 struct radeon_semaphore *semaphore,
561 int signaler, int waiter); 565 int waiting_ring);
562void radeon_semaphore_free(struct radeon_device *rdev, 566void radeon_semaphore_free(struct radeon_device *rdev,
563 struct radeon_semaphore **semaphore, 567 struct radeon_semaphore **semaphore,
564 struct radeon_fence *fence); 568 struct radeon_fence *fence);
@@ -645,13 +649,15 @@ void radeon_scratch_free(struct radeon_device *rdev, uint32_t reg);
645/* 649/*
646 * GPU doorbell structures, functions & helpers 650 * GPU doorbell structures, functions & helpers
647 */ 651 */
652#define RADEON_MAX_DOORBELLS 1024 /* Reserve at most 1024 doorbell slots for radeon-owned rings. */
653
648struct radeon_doorbell { 654struct radeon_doorbell {
649 u32 num_pages;
650 bool free[1024];
651 /* doorbell mmio */ 655 /* doorbell mmio */
652 resource_size_t base; 656 resource_size_t base;
653 resource_size_t size; 657 resource_size_t size;
654 void __iomem *ptr; 658 u32 __iomem *ptr;
659 u32 num_doorbells; /* Number of doorbells actually reserved for radeon. */
660 unsigned long used[DIV_ROUND_UP(RADEON_MAX_DOORBELLS, BITS_PER_LONG)];
655}; 661};
656 662
657int radeon_doorbell_get(struct radeon_device *rdev, u32 *page); 663int radeon_doorbell_get(struct radeon_device *rdev, u32 *page);
@@ -765,7 +771,6 @@ struct radeon_ib {
765 struct radeon_fence *fence; 771 struct radeon_fence *fence;
766 struct radeon_vm *vm; 772 struct radeon_vm *vm;
767 bool is_const_ib; 773 bool is_const_ib;
768 struct radeon_fence *sync_to[RADEON_NUM_RINGS];
769 struct radeon_semaphore *semaphore; 774 struct radeon_semaphore *semaphore;
770}; 775};
771 776
@@ -799,8 +804,7 @@ struct radeon_ring {
799 u32 pipe; 804 u32 pipe;
800 u32 queue; 805 u32 queue;
801 struct radeon_bo *mqd_obj; 806 struct radeon_bo *mqd_obj;
802 u32 doorbell_page_num; 807 u32 doorbell_index;
803 u32 doorbell_offset;
804 unsigned wptr_offs; 808 unsigned wptr_offs;
805}; 809};
806 810
@@ -921,7 +925,6 @@ int radeon_ib_get(struct radeon_device *rdev, int ring,
921 struct radeon_ib *ib, struct radeon_vm *vm, 925 struct radeon_ib *ib, struct radeon_vm *vm,
922 unsigned size); 926 unsigned size);
923void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib); 927void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib);
924void radeon_ib_sync_to(struct radeon_ib *ib, struct radeon_fence *fence);
925int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib, 928int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
926 struct radeon_ib *const_ib); 929 struct radeon_ib *const_ib);
927int radeon_ib_pool_init(struct radeon_device *rdev); 930int radeon_ib_pool_init(struct radeon_device *rdev);
@@ -1638,7 +1641,7 @@ struct radeon_asic_ring {
1638 /* command emmit functions */ 1641 /* command emmit functions */
1639 void (*ib_execute)(struct radeon_device *rdev, struct radeon_ib *ib); 1642 void (*ib_execute)(struct radeon_device *rdev, struct radeon_ib *ib);
1640 void (*emit_fence)(struct radeon_device *rdev, struct radeon_fence *fence); 1643 void (*emit_fence)(struct radeon_device *rdev, struct radeon_fence *fence);
1641 void (*emit_semaphore)(struct radeon_device *rdev, struct radeon_ring *cp, 1644 bool (*emit_semaphore)(struct radeon_device *rdev, struct radeon_ring *cp,
1642 struct radeon_semaphore *semaphore, bool emit_wait); 1645 struct radeon_semaphore *semaphore, bool emit_wait);
1643 void (*vm_flush)(struct radeon_device *rdev, int ridx, struct radeon_vm *vm); 1646 void (*vm_flush)(struct radeon_device *rdev, int ridx, struct radeon_vm *vm);
1644 1647
@@ -1979,6 +1982,7 @@ struct cik_asic {
1979 1982
1980 unsigned tile_config; 1983 unsigned tile_config;
1981 uint32_t tile_mode_array[32]; 1984 uint32_t tile_mode_array[32];
1985 uint32_t macrotile_mode_array[16];
1982}; 1986};
1983 1987
1984union radeon_asic_config { 1988union radeon_asic_config {
@@ -2239,8 +2243,8 @@ void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v,
2239u32 r100_io_rreg(struct radeon_device *rdev, u32 reg); 2243u32 r100_io_rreg(struct radeon_device *rdev, u32 reg);
2240void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v); 2244void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v);
2241 2245
2242u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset); 2246u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 index);
2243void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v); 2247void cik_mm_wdoorbell(struct radeon_device *rdev, u32 index, u32 v);
2244 2248
2245/* 2249/*
2246 * Cast helper 2250 * Cast helper
@@ -2303,8 +2307,8 @@ void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v);
2303#define RREG32_IO(reg) r100_io_rreg(rdev, (reg)) 2307#define RREG32_IO(reg) r100_io_rreg(rdev, (reg))
2304#define WREG32_IO(reg, v) r100_io_wreg(rdev, (reg), (v)) 2308#define WREG32_IO(reg, v) r100_io_wreg(rdev, (reg), (v))
2305 2309
2306#define RDOORBELL32(offset) cik_mm_rdoorbell(rdev, (offset)) 2310#define RDOORBELL32(index) cik_mm_rdoorbell(rdev, (index))
2307#define WDOORBELL32(offset, v) cik_mm_wdoorbell(rdev, (offset), (v)) 2311#define WDOORBELL32(index, v) cik_mm_wdoorbell(rdev, (index), (v))
2308 2312
2309/* 2313/*
2310 * Indirect registers accessor 2314 * Indirect registers accessor
diff --git a/drivers/gpu/drm/radeon/radeon_asic.c b/drivers/gpu/drm/radeon/radeon_asic.c
index 50853c0cb49d..e354ce94cdd1 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.c
+++ b/drivers/gpu/drm/radeon/radeon_asic.c
@@ -2015,6 +2015,8 @@ static struct radeon_asic ci_asic = {
2015 .bandwidth_update = &dce8_bandwidth_update, 2015 .bandwidth_update = &dce8_bandwidth_update,
2016 .get_vblank_counter = &evergreen_get_vblank_counter, 2016 .get_vblank_counter = &evergreen_get_vblank_counter,
2017 .wait_for_vblank = &dce4_wait_for_vblank, 2017 .wait_for_vblank = &dce4_wait_for_vblank,
2018 .set_backlight_level = &atombios_set_backlight_level,
2019 .get_backlight_level = &atombios_get_backlight_level,
2018 .hdmi_enable = &evergreen_hdmi_enable, 2020 .hdmi_enable = &evergreen_hdmi_enable,
2019 .hdmi_setmode = &evergreen_hdmi_setmode, 2021 .hdmi_setmode = &evergreen_hdmi_setmode,
2020 }, 2022 },
@@ -2114,6 +2116,8 @@ static struct radeon_asic kv_asic = {
2114 .bandwidth_update = &dce8_bandwidth_update, 2116 .bandwidth_update = &dce8_bandwidth_update,
2115 .get_vblank_counter = &evergreen_get_vblank_counter, 2117 .get_vblank_counter = &evergreen_get_vblank_counter,
2116 .wait_for_vblank = &dce4_wait_for_vblank, 2118 .wait_for_vblank = &dce4_wait_for_vblank,
2119 .set_backlight_level = &atombios_set_backlight_level,
2120 .get_backlight_level = &atombios_get_backlight_level,
2117 .hdmi_enable = &evergreen_hdmi_enable, 2121 .hdmi_enable = &evergreen_hdmi_enable,
2118 .hdmi_setmode = &evergreen_hdmi_setmode, 2122 .hdmi_setmode = &evergreen_hdmi_setmode,
2119 }, 2123 },
diff --git a/drivers/gpu/drm/radeon/radeon_asic.h b/drivers/gpu/drm/radeon/radeon_asic.h
index f2833ee3a613..c9fd97b58076 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.h
+++ b/drivers/gpu/drm/radeon/radeon_asic.h
@@ -80,7 +80,7 @@ int r100_irq_set(struct radeon_device *rdev);
80int r100_irq_process(struct radeon_device *rdev); 80int r100_irq_process(struct radeon_device *rdev);
81void r100_fence_ring_emit(struct radeon_device *rdev, 81void r100_fence_ring_emit(struct radeon_device *rdev,
82 struct radeon_fence *fence); 82 struct radeon_fence *fence);
83void r100_semaphore_ring_emit(struct radeon_device *rdev, 83bool r100_semaphore_ring_emit(struct radeon_device *rdev,
84 struct radeon_ring *cp, 84 struct radeon_ring *cp,
85 struct radeon_semaphore *semaphore, 85 struct radeon_semaphore *semaphore,
86 bool emit_wait); 86 bool emit_wait);
@@ -313,13 +313,13 @@ int r600_cs_parse(struct radeon_cs_parser *p);
313int r600_dma_cs_parse(struct radeon_cs_parser *p); 313int r600_dma_cs_parse(struct radeon_cs_parser *p);
314void r600_fence_ring_emit(struct radeon_device *rdev, 314void r600_fence_ring_emit(struct radeon_device *rdev,
315 struct radeon_fence *fence); 315 struct radeon_fence *fence);
316void r600_semaphore_ring_emit(struct radeon_device *rdev, 316bool r600_semaphore_ring_emit(struct radeon_device *rdev,
317 struct radeon_ring *cp, 317 struct radeon_ring *cp,
318 struct radeon_semaphore *semaphore, 318 struct radeon_semaphore *semaphore,
319 bool emit_wait); 319 bool emit_wait);
320void r600_dma_fence_ring_emit(struct radeon_device *rdev, 320void r600_dma_fence_ring_emit(struct radeon_device *rdev,
321 struct radeon_fence *fence); 321 struct radeon_fence *fence);
322void r600_dma_semaphore_ring_emit(struct radeon_device *rdev, 322bool r600_dma_semaphore_ring_emit(struct radeon_device *rdev,
323 struct radeon_ring *ring, 323 struct radeon_ring *ring,
324 struct radeon_semaphore *semaphore, 324 struct radeon_semaphore *semaphore,
325 bool emit_wait); 325 bool emit_wait);
@@ -566,10 +566,6 @@ int sumo_dpm_force_performance_level(struct radeon_device *rdev,
566 */ 566 */
567void cayman_fence_ring_emit(struct radeon_device *rdev, 567void cayman_fence_ring_emit(struct radeon_device *rdev,
568 struct radeon_fence *fence); 568 struct radeon_fence *fence);
569void cayman_uvd_semaphore_emit(struct radeon_device *rdev,
570 struct radeon_ring *ring,
571 struct radeon_semaphore *semaphore,
572 bool emit_wait);
573void cayman_pcie_gart_tlb_flush(struct radeon_device *rdev); 569void cayman_pcie_gart_tlb_flush(struct radeon_device *rdev);
574int cayman_init(struct radeon_device *rdev); 570int cayman_init(struct radeon_device *rdev);
575void cayman_fini(struct radeon_device *rdev); 571void cayman_fini(struct radeon_device *rdev);
@@ -697,7 +693,7 @@ void cik_pciep_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v);
697int cik_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk); 693int cik_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk);
698void cik_sdma_fence_ring_emit(struct radeon_device *rdev, 694void cik_sdma_fence_ring_emit(struct radeon_device *rdev,
699 struct radeon_fence *fence); 695 struct radeon_fence *fence);
700void cik_sdma_semaphore_ring_emit(struct radeon_device *rdev, 696bool cik_sdma_semaphore_ring_emit(struct radeon_device *rdev,
701 struct radeon_ring *ring, 697 struct radeon_ring *ring,
702 struct radeon_semaphore *semaphore, 698 struct radeon_semaphore *semaphore,
703 bool emit_wait); 699 bool emit_wait);
@@ -717,7 +713,7 @@ void cik_fence_gfx_ring_emit(struct radeon_device *rdev,
717 struct radeon_fence *fence); 713 struct radeon_fence *fence);
718void cik_fence_compute_ring_emit(struct radeon_device *rdev, 714void cik_fence_compute_ring_emit(struct radeon_device *rdev,
719 struct radeon_fence *fence); 715 struct radeon_fence *fence);
720void cik_semaphore_ring_emit(struct radeon_device *rdev, 716bool cik_semaphore_ring_emit(struct radeon_device *rdev,
721 struct radeon_ring *cp, 717 struct radeon_ring *cp,
722 struct radeon_semaphore *semaphore, 718 struct radeon_semaphore *semaphore,
723 bool emit_wait); 719 bool emit_wait);
@@ -807,7 +803,7 @@ void uvd_v1_0_stop(struct radeon_device *rdev);
807 803
808int uvd_v1_0_ring_test(struct radeon_device *rdev, struct radeon_ring *ring); 804int uvd_v1_0_ring_test(struct radeon_device *rdev, struct radeon_ring *ring);
809int uvd_v1_0_ib_test(struct radeon_device *rdev, struct radeon_ring *ring); 805int uvd_v1_0_ib_test(struct radeon_device *rdev, struct radeon_ring *ring);
810void uvd_v1_0_semaphore_emit(struct radeon_device *rdev, 806bool uvd_v1_0_semaphore_emit(struct radeon_device *rdev,
811 struct radeon_ring *ring, 807 struct radeon_ring *ring,
812 struct radeon_semaphore *semaphore, 808 struct radeon_semaphore *semaphore,
813 bool emit_wait); 809 bool emit_wait);
@@ -819,7 +815,7 @@ void uvd_v2_2_fence_emit(struct radeon_device *rdev,
819 struct radeon_fence *fence); 815 struct radeon_fence *fence);
820 816
821/* uvd v3.1 */ 817/* uvd v3.1 */
822void uvd_v3_1_semaphore_emit(struct radeon_device *rdev, 818bool uvd_v3_1_semaphore_emit(struct radeon_device *rdev,
823 struct radeon_ring *ring, 819 struct radeon_ring *ring,
824 struct radeon_semaphore *semaphore, 820 struct radeon_semaphore *semaphore,
825 bool emit_wait); 821 bool emit_wait);
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 26ca223d12d6..f41594b2eeac 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -159,7 +159,8 @@ static void radeon_cs_sync_rings(struct radeon_cs_parser *p)
159 if (!p->relocs[i].robj) 159 if (!p->relocs[i].robj)
160 continue; 160 continue;
161 161
162 radeon_ib_sync_to(&p->ib, p->relocs[i].robj->tbo.sync_obj); 162 radeon_semaphore_sync_to(p->ib.semaphore,
163 p->relocs[i].robj->tbo.sync_obj);
163 } 164 }
164} 165}
165 166
@@ -411,9 +412,9 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
411 goto out; 412 goto out;
412 } 413 }
413 radeon_cs_sync_rings(parser); 414 radeon_cs_sync_rings(parser);
414 radeon_ib_sync_to(&parser->ib, vm->fence); 415 radeon_semaphore_sync_to(parser->ib.semaphore, vm->fence);
415 radeon_ib_sync_to(&parser->ib, radeon_vm_grab_id( 416 radeon_semaphore_sync_to(parser->ib.semaphore,
416 rdev, vm, parser->ring)); 417 radeon_vm_grab_id(rdev, vm, parser->ring));
417 418
418 if ((rdev->family >= CHIP_TAHITI) && 419 if ((rdev->family >= CHIP_TAHITI) &&
419 (parser->chunk_const_ib_idx != -1)) { 420 (parser->chunk_const_ib_idx != -1)) {
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index b9234c43f43d..39b033b441d2 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -251,28 +251,23 @@ void radeon_scratch_free(struct radeon_device *rdev, uint32_t reg)
251 */ 251 */
252int radeon_doorbell_init(struct radeon_device *rdev) 252int radeon_doorbell_init(struct radeon_device *rdev)
253{ 253{
254 int i;
255
256 /* doorbell bar mapping */ 254 /* doorbell bar mapping */
257 rdev->doorbell.base = pci_resource_start(rdev->pdev, 2); 255 rdev->doorbell.base = pci_resource_start(rdev->pdev, 2);
258 rdev->doorbell.size = pci_resource_len(rdev->pdev, 2); 256 rdev->doorbell.size = pci_resource_len(rdev->pdev, 2);
259 257
260 /* limit to 4 MB for now */ 258 rdev->doorbell.num_doorbells = min_t(u32, rdev->doorbell.size / sizeof(u32), RADEON_MAX_DOORBELLS);
261 if (rdev->doorbell.size > (4 * 1024 * 1024)) 259 if (rdev->doorbell.num_doorbells == 0)
262 rdev->doorbell.size = 4 * 1024 * 1024; 260 return -EINVAL;
263 261
264 rdev->doorbell.ptr = ioremap(rdev->doorbell.base, rdev->doorbell.size); 262 rdev->doorbell.ptr = ioremap(rdev->doorbell.base, rdev->doorbell.num_doorbells * sizeof(u32));
265 if (rdev->doorbell.ptr == NULL) { 263 if (rdev->doorbell.ptr == NULL) {
266 return -ENOMEM; 264 return -ENOMEM;
267 } 265 }
268 DRM_INFO("doorbell mmio base: 0x%08X\n", (uint32_t)rdev->doorbell.base); 266 DRM_INFO("doorbell mmio base: 0x%08X\n", (uint32_t)rdev->doorbell.base);
269 DRM_INFO("doorbell mmio size: %u\n", (unsigned)rdev->doorbell.size); 267 DRM_INFO("doorbell mmio size: %u\n", (unsigned)rdev->doorbell.size);
270 268
271 rdev->doorbell.num_pages = rdev->doorbell.size / PAGE_SIZE; 269 memset(&rdev->doorbell.used, 0, sizeof(rdev->doorbell.used));
272 270
273 for (i = 0; i < rdev->doorbell.num_pages; i++) {
274 rdev->doorbell.free[i] = true;
275 }
276 return 0; 271 return 0;
277} 272}
278 273
@@ -290,40 +285,38 @@ void radeon_doorbell_fini(struct radeon_device *rdev)
290} 285}
291 286
292/** 287/**
293 * radeon_doorbell_get - Allocate a doorbell page 288 * radeon_doorbell_get - Allocate a doorbell entry
294 * 289 *
295 * @rdev: radeon_device pointer 290 * @rdev: radeon_device pointer
296 * @doorbell: doorbell page number 291 * @doorbell: doorbell index
297 * 292 *
298 * Allocate a doorbell page for use by the driver (all asics). 293 * Allocate a doorbell for use by the driver (all asics).
299 * Returns 0 on success or -EINVAL on failure. 294 * Returns 0 on success or -EINVAL on failure.
300 */ 295 */
301int radeon_doorbell_get(struct radeon_device *rdev, u32 *doorbell) 296int radeon_doorbell_get(struct radeon_device *rdev, u32 *doorbell)
302{ 297{
303 int i; 298 unsigned long offset = find_first_zero_bit(rdev->doorbell.used, rdev->doorbell.num_doorbells);
304 299 if (offset < rdev->doorbell.num_doorbells) {
305 for (i = 0; i < rdev->doorbell.num_pages; i++) { 300 __set_bit(offset, rdev->doorbell.used);
306 if (rdev->doorbell.free[i]) { 301 *doorbell = offset;
307 rdev->doorbell.free[i] = false; 302 return 0;
308 *doorbell = i; 303 } else {
309 return 0; 304 return -EINVAL;
310 }
311 } 305 }
312 return -EINVAL;
313} 306}
314 307
315/** 308/**
316 * radeon_doorbell_free - Free a doorbell page 309 * radeon_doorbell_free - Free a doorbell entry
317 * 310 *
318 * @rdev: radeon_device pointer 311 * @rdev: radeon_device pointer
319 * @doorbell: doorbell page number 312 * @doorbell: doorbell index
320 * 313 *
321 * Free a doorbell page allocated for use by the driver (all asics) 314 * Free a doorbell allocated for use by the driver (all asics)
322 */ 315 */
323void radeon_doorbell_free(struct radeon_device *rdev, u32 doorbell) 316void radeon_doorbell_free(struct radeon_device *rdev, u32 doorbell)
324{ 317{
325 if (doorbell < rdev->doorbell.num_pages) 318 if (doorbell < rdev->doorbell.num_doorbells)
326 rdev->doorbell.free[doorbell] = true; 319 __clear_bit(doorbell, rdev->doorbell.used);
327} 320}
328 321
329/* 322/*
diff --git a/drivers/gpu/drm/radeon/radeon_drv.c b/drivers/gpu/drm/radeon/radeon_drv.c
index 1aee32213f66..9f5ff28864f6 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.c
+++ b/drivers/gpu/drm/radeon/radeon_drv.c
@@ -76,9 +76,10 @@
76 * 2.32.0 - new info request for rings working 76 * 2.32.0 - new info request for rings working
77 * 2.33.0 - Add SI tiling mode array query 77 * 2.33.0 - Add SI tiling mode array query
78 * 2.34.0 - Add CIK tiling mode array query 78 * 2.34.0 - Add CIK tiling mode array query
79 * 2.35.0 - Add CIK macrotile mode array query
79 */ 80 */
80#define KMS_DRIVER_MAJOR 2 81#define KMS_DRIVER_MAJOR 2
81#define KMS_DRIVER_MINOR 34 82#define KMS_DRIVER_MINOR 35
82#define KMS_DRIVER_PATCHLEVEL 0 83#define KMS_DRIVER_PATCHLEVEL 0
83int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags); 84int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags);
84int radeon_driver_unload_kms(struct drm_device *dev); 85int radeon_driver_unload_kms(struct drm_device *dev);
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
index 281d14c22a47..d3a86e43c012 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -472,6 +472,36 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
472} 472}
473 473
474/** 474/**
475 * radeon_fence_wait_locked - wait for a fence to signal
476 *
477 * @fence: radeon fence object
478 *
479 * Wait for the requested fence to signal (all asics).
480 * Returns 0 if the fence has passed, error for all other cases.
481 */
482int radeon_fence_wait_locked(struct radeon_fence *fence)
483{
484 uint64_t seq[RADEON_NUM_RINGS] = {};
485 int r;
486
487 if (fence == NULL) {
488 WARN(1, "Querying an invalid fence : %p !\n", fence);
489 return -EINVAL;
490 }
491
492 seq[fence->ring] = fence->seq;
493 if (seq[fence->ring] == RADEON_FENCE_SIGNALED_SEQ)
494 return 0;
495
496 r = radeon_fence_wait_seq(fence->rdev, seq, false, false);
497 if (r)
498 return r;
499
500 fence->seq = RADEON_FENCE_SIGNALED_SEQ;
501 return 0;
502}
503
504/**
475 * radeon_fence_wait_next_locked - wait for the next fence to signal 505 * radeon_fence_wait_next_locked - wait for the next fence to signal
476 * 506 *
477 * @rdev: radeon device pointer 507 * @rdev: radeon device pointer
diff --git a/drivers/gpu/drm/radeon/radeon_gart.c b/drivers/gpu/drm/radeon/radeon_gart.c
index 8a83b89d4709..3044e504f4ec 100644
--- a/drivers/gpu/drm/radeon/radeon_gart.c
+++ b/drivers/gpu/drm/radeon/radeon_gart.c
@@ -651,7 +651,7 @@ retry:
651 radeon_asic_vm_set_page(rdev, &ib, vm->pd_gpu_addr, 651 radeon_asic_vm_set_page(rdev, &ib, vm->pd_gpu_addr,
652 0, pd_entries, 0, 0); 652 0, pd_entries, 0, 0);
653 653
654 radeon_ib_sync_to(&ib, vm->fence); 654 radeon_semaphore_sync_to(ib.semaphore, vm->fence);
655 r = radeon_ib_schedule(rdev, &ib, NULL); 655 r = radeon_ib_schedule(rdev, &ib, NULL);
656 if (r) { 656 if (r) {
657 radeon_ib_free(rdev, &ib); 657 radeon_ib_free(rdev, &ib);
@@ -1209,6 +1209,8 @@ int radeon_vm_bo_update_pte(struct radeon_device *rdev,
1209 return -ENOMEM; 1209 return -ENOMEM;
1210 1210
1211 r = radeon_ib_get(rdev, R600_RING_TYPE_DMA_INDEX, &ib, NULL, ndw * 4); 1211 r = radeon_ib_get(rdev, R600_RING_TYPE_DMA_INDEX, &ib, NULL, ndw * 4);
1212 if (r)
1213 return r;
1212 ib.length_dw = 0; 1214 ib.length_dw = 0;
1213 1215
1214 r = radeon_vm_update_pdes(rdev, vm, &ib, bo_va->soffset, bo_va->eoffset); 1216 r = radeon_vm_update_pdes(rdev, vm, &ib, bo_va->soffset, bo_va->eoffset);
@@ -1220,7 +1222,7 @@ int radeon_vm_bo_update_pte(struct radeon_device *rdev,
1220 radeon_vm_update_ptes(rdev, vm, &ib, bo_va->soffset, bo_va->eoffset, 1222 radeon_vm_update_ptes(rdev, vm, &ib, bo_va->soffset, bo_va->eoffset,
1221 addr, radeon_vm_page_flags(bo_va->flags)); 1223 addr, radeon_vm_page_flags(bo_va->flags));
1222 1224
1223 radeon_ib_sync_to(&ib, vm->fence); 1225 radeon_semaphore_sync_to(ib.semaphore, vm->fence);
1224 r = radeon_ib_schedule(rdev, &ib, NULL); 1226 r = radeon_ib_schedule(rdev, &ib, NULL);
1225 if (r) { 1227 if (r) {
1226 radeon_ib_free(rdev, &ib); 1228 radeon_ib_free(rdev, &ib);
diff --git a/drivers/gpu/drm/radeon/radeon_kms.c b/drivers/gpu/drm/radeon/radeon_kms.c
index bb8710531a1b..55d0b474bd37 100644
--- a/drivers/gpu/drm/radeon/radeon_kms.c
+++ b/drivers/gpu/drm/radeon/radeon_kms.c
@@ -340,7 +340,7 @@ int radeon_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
340 break; 340 break;
341 case RADEON_INFO_BACKEND_MAP: 341 case RADEON_INFO_BACKEND_MAP:
342 if (rdev->family >= CHIP_BONAIRE) 342 if (rdev->family >= CHIP_BONAIRE)
343 return -EINVAL; 343 *value = rdev->config.cik.backend_map;
344 else if (rdev->family >= CHIP_TAHITI) 344 else if (rdev->family >= CHIP_TAHITI)
345 *value = rdev->config.si.backend_map; 345 *value = rdev->config.si.backend_map;
346 else if (rdev->family >= CHIP_CAYMAN) 346 else if (rdev->family >= CHIP_CAYMAN)
@@ -449,6 +449,15 @@ int radeon_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
449 return -EINVAL; 449 return -EINVAL;
450 } 450 }
451 break; 451 break;
452 case RADEON_INFO_CIK_MACROTILE_MODE_ARRAY:
453 if (rdev->family >= CHIP_BONAIRE) {
454 value = rdev->config.cik.macrotile_mode_array;
455 value_size = sizeof(uint32_t)*16;
456 } else {
457 DRM_DEBUG_KMS("macrotile mode array is cik+ only!\n");
458 return -EINVAL;
459 }
460 break;
452 case RADEON_INFO_SI_CP_DMA_COMPUTE: 461 case RADEON_INFO_SI_CP_DMA_COMPUTE:
453 *value = 1; 462 *value = 1;
454 break; 463 break;
diff --git a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
index 0c7b8c66301b..0b158f98d287 100644
--- a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
+++ b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
@@ -422,6 +422,7 @@ int radeon_crtc_do_set_base(struct drm_crtc *crtc,
422 /* Pin framebuffer & get tilling informations */ 422 /* Pin framebuffer & get tilling informations */
423 obj = radeon_fb->obj; 423 obj = radeon_fb->obj;
424 rbo = gem_to_radeon_bo(obj); 424 rbo = gem_to_radeon_bo(obj);
425retry:
425 r = radeon_bo_reserve(rbo, false); 426 r = radeon_bo_reserve(rbo, false);
426 if (unlikely(r != 0)) 427 if (unlikely(r != 0))
427 return r; 428 return r;
@@ -430,6 +431,33 @@ int radeon_crtc_do_set_base(struct drm_crtc *crtc,
430 &base); 431 &base);
431 if (unlikely(r != 0)) { 432 if (unlikely(r != 0)) {
432 radeon_bo_unreserve(rbo); 433 radeon_bo_unreserve(rbo);
434
435 /* On old GPU like RN50 with little vram pining can fails because
436 * current fb is taking all space needed. So instead of unpining
437 * the old buffer after pining the new one, first unpin old one
438 * and then retry pining new one.
439 *
440 * As only master can set mode only master can pin and it is
441 * unlikely the master client will race with itself especialy
442 * on those old gpu with single crtc.
443 *
444 * We don't shutdown the display controller because new buffer
445 * will end up in same spot.
446 */
447 if (!atomic && fb && fb != crtc->fb) {
448 struct radeon_bo *old_rbo;
449 unsigned long nsize, osize;
450
451 old_rbo = gem_to_radeon_bo(to_radeon_framebuffer(fb)->obj);
452 osize = radeon_bo_size(old_rbo);
453 nsize = radeon_bo_size(rbo);
454 if (nsize <= osize && !radeon_bo_reserve(old_rbo, false)) {
455 radeon_bo_unpin(old_rbo);
456 radeon_bo_unreserve(old_rbo);
457 fb = NULL;
458 goto retry;
459 }
460 }
433 return -EINVAL; 461 return -EINVAL;
434 } 462 }
435 radeon_bo_get_tiling_flags(rbo, &tiling_flags, NULL); 463 radeon_bo_get_tiling_flags(rbo, &tiling_flags, NULL);
diff --git a/drivers/gpu/drm/radeon/radeon_pm.c b/drivers/gpu/drm/radeon/radeon_pm.c
index 866ace070b91..d1385ccc672c 100644
--- a/drivers/gpu/drm/radeon/radeon_pm.c
+++ b/drivers/gpu/drm/radeon/radeon_pm.c
@@ -1252,7 +1252,6 @@ int radeon_pm_init(struct radeon_device *rdev)
1252 case CHIP_RS780: 1252 case CHIP_RS780:
1253 case CHIP_RS880: 1253 case CHIP_RS880:
1254 case CHIP_CAYMAN: 1254 case CHIP_CAYMAN:
1255 case CHIP_ARUBA:
1256 case CHIP_BONAIRE: 1255 case CHIP_BONAIRE:
1257 case CHIP_KABINI: 1256 case CHIP_KABINI:
1258 case CHIP_KAVERI: 1257 case CHIP_KAVERI:
@@ -1284,6 +1283,7 @@ int radeon_pm_init(struct radeon_device *rdev)
1284 case CHIP_BARTS: 1283 case CHIP_BARTS:
1285 case CHIP_TURKS: 1284 case CHIP_TURKS:
1286 case CHIP_CAICOS: 1285 case CHIP_CAICOS:
1286 case CHIP_ARUBA:
1287 case CHIP_TAHITI: 1287 case CHIP_TAHITI:
1288 case CHIP_PITCAIRN: 1288 case CHIP_PITCAIRN:
1289 case CHIP_VERDE: 1289 case CHIP_VERDE:
diff --git a/drivers/gpu/drm/radeon/radeon_ring.c b/drivers/gpu/drm/radeon/radeon_ring.c
index 18254e1c3e71..9214403ae173 100644
--- a/drivers/gpu/drm/radeon/radeon_ring.c
+++ b/drivers/gpu/drm/radeon/radeon_ring.c
@@ -61,7 +61,7 @@ int radeon_ib_get(struct radeon_device *rdev, int ring,
61 struct radeon_ib *ib, struct radeon_vm *vm, 61 struct radeon_ib *ib, struct radeon_vm *vm,
62 unsigned size) 62 unsigned size)
63{ 63{
64 int i, r; 64 int r;
65 65
66 r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256, true); 66 r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256, true);
67 if (r) { 67 if (r) {
@@ -87,8 +87,6 @@ int radeon_ib_get(struct radeon_device *rdev, int ring,
87 ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo); 87 ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo);
88 } 88 }
89 ib->is_const_ib = false; 89 ib->is_const_ib = false;
90 for (i = 0; i < RADEON_NUM_RINGS; ++i)
91 ib->sync_to[i] = NULL;
92 90
93 return 0; 91 return 0;
94} 92}
@@ -109,25 +107,6 @@ void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib)
109} 107}
110 108
111/** 109/**
112 * radeon_ib_sync_to - sync to fence before executing the IB
113 *
114 * @ib: IB object to add fence to
115 * @fence: fence to sync to
116 *
117 * Sync to the fence before executing the IB
118 */
119void radeon_ib_sync_to(struct radeon_ib *ib, struct radeon_fence *fence)
120{
121 struct radeon_fence *other;
122
123 if (!fence)
124 return;
125
126 other = ib->sync_to[fence->ring];
127 ib->sync_to[fence->ring] = radeon_fence_later(fence, other);
128}
129
130/**
131 * radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring 110 * radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring
132 * 111 *
133 * @rdev: radeon_device pointer 112 * @rdev: radeon_device pointer
@@ -151,8 +130,7 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
151 struct radeon_ib *const_ib) 130 struct radeon_ib *const_ib)
152{ 131{
153 struct radeon_ring *ring = &rdev->ring[ib->ring]; 132 struct radeon_ring *ring = &rdev->ring[ib->ring];
154 bool need_sync = false; 133 int r = 0;
155 int i, r = 0;
156 134
157 if (!ib->length_dw || !ring->ready) { 135 if (!ib->length_dw || !ring->ready) {
158 /* TODO: Nothings in the ib we should report. */ 136 /* TODO: Nothings in the ib we should report. */
@@ -166,19 +144,15 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
166 dev_err(rdev->dev, "scheduling IB failed (%d).\n", r); 144 dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
167 return r; 145 return r;
168 } 146 }
169 for (i = 0; i < RADEON_NUM_RINGS; ++i) { 147
170 struct radeon_fence *fence = ib->sync_to[i]; 148 /* sync with other rings */
171 if (radeon_fence_need_sync(fence, ib->ring)) { 149 r = radeon_semaphore_sync_rings(rdev, ib->semaphore, ib->ring);
172 need_sync = true; 150 if (r) {
173 radeon_semaphore_sync_rings(rdev, ib->semaphore, 151 dev_err(rdev->dev, "failed to sync rings (%d)\n", r);
174 fence->ring, ib->ring); 152 radeon_ring_unlock_undo(rdev, ring);
175 radeon_fence_note_sync(fence, ib->ring); 153 return r;
176 }
177 }
178 /* immediately free semaphore when we don't need to sync */
179 if (!need_sync) {
180 radeon_semaphore_free(rdev, &ib->semaphore, NULL);
181 } 154 }
155
182 /* if we can't remember our last VM flush then flush now! */ 156 /* if we can't remember our last VM flush then flush now! */
183 /* XXX figure out why we have to flush for every IB */ 157 /* XXX figure out why we have to flush for every IB */
184 if (ib->vm /*&& !ib->vm->last_flush*/) { 158 if (ib->vm /*&& !ib->vm->last_flush*/) {
diff --git a/drivers/gpu/drm/radeon/radeon_semaphore.c b/drivers/gpu/drm/radeon/radeon_semaphore.c
index 8dcc20f53d73..2b42aa1914f2 100644
--- a/drivers/gpu/drm/radeon/radeon_semaphore.c
+++ b/drivers/gpu/drm/radeon/radeon_semaphore.c
@@ -29,12 +29,12 @@
29 */ 29 */
30#include <drm/drmP.h> 30#include <drm/drmP.h>
31#include "radeon.h" 31#include "radeon.h"
32 32#include "radeon_trace.h"
33 33
34int radeon_semaphore_create(struct radeon_device *rdev, 34int radeon_semaphore_create(struct radeon_device *rdev,
35 struct radeon_semaphore **semaphore) 35 struct radeon_semaphore **semaphore)
36{ 36{
37 int r; 37 int i, r;
38 38
39 *semaphore = kmalloc(sizeof(struct radeon_semaphore), GFP_KERNEL); 39 *semaphore = kmalloc(sizeof(struct radeon_semaphore), GFP_KERNEL);
40 if (*semaphore == NULL) { 40 if (*semaphore == NULL) {
@@ -50,54 +50,121 @@ int radeon_semaphore_create(struct radeon_device *rdev,
50 (*semaphore)->waiters = 0; 50 (*semaphore)->waiters = 0;
51 (*semaphore)->gpu_addr = radeon_sa_bo_gpu_addr((*semaphore)->sa_bo); 51 (*semaphore)->gpu_addr = radeon_sa_bo_gpu_addr((*semaphore)->sa_bo);
52 *((uint64_t*)radeon_sa_bo_cpu_addr((*semaphore)->sa_bo)) = 0; 52 *((uint64_t*)radeon_sa_bo_cpu_addr((*semaphore)->sa_bo)) = 0;
53
54 for (i = 0; i < RADEON_NUM_RINGS; ++i)
55 (*semaphore)->sync_to[i] = NULL;
56
53 return 0; 57 return 0;
54} 58}
55 59
56void radeon_semaphore_emit_signal(struct radeon_device *rdev, int ring, 60bool radeon_semaphore_emit_signal(struct radeon_device *rdev, int ridx,
57 struct radeon_semaphore *semaphore) 61 struct radeon_semaphore *semaphore)
58{ 62{
59 --semaphore->waiters; 63 struct radeon_ring *ring = &rdev->ring[ridx];
60 radeon_semaphore_ring_emit(rdev, ring, &rdev->ring[ring], semaphore, false); 64
65 trace_radeon_semaphore_signale(ridx, semaphore);
66
67 if (radeon_semaphore_ring_emit(rdev, ridx, ring, semaphore, false)) {
68 --semaphore->waiters;
69
70 /* for debugging lockup only, used by sysfs debug files */
71 ring->last_semaphore_signal_addr = semaphore->gpu_addr;
72 return true;
73 }
74 return false;
61} 75}
62 76
63void radeon_semaphore_emit_wait(struct radeon_device *rdev, int ring, 77bool radeon_semaphore_emit_wait(struct radeon_device *rdev, int ridx,
64 struct radeon_semaphore *semaphore) 78 struct radeon_semaphore *semaphore)
65{ 79{
66 ++semaphore->waiters; 80 struct radeon_ring *ring = &rdev->ring[ridx];
67 radeon_semaphore_ring_emit(rdev, ring, &rdev->ring[ring], semaphore, true); 81
82 trace_radeon_semaphore_wait(ridx, semaphore);
83
84 if (radeon_semaphore_ring_emit(rdev, ridx, ring, semaphore, true)) {
85 ++semaphore->waiters;
86
87 /* for debugging lockup only, used by sysfs debug files */
88 ring->last_semaphore_wait_addr = semaphore->gpu_addr;
89 return true;
90 }
91 return false;
92}
93
94/**
95 * radeon_semaphore_sync_to - use the semaphore to sync to a fence
96 *
97 * @semaphore: semaphore object to add fence to
98 * @fence: fence to sync to
99 *
100 * Sync to the fence using this semaphore object
101 */
102void radeon_semaphore_sync_to(struct radeon_semaphore *semaphore,
103 struct radeon_fence *fence)
104{
105 struct radeon_fence *other;
106
107 if (!fence)
108 return;
109
110 other = semaphore->sync_to[fence->ring];
111 semaphore->sync_to[fence->ring] = radeon_fence_later(fence, other);
68} 112}
69 113
70/* caller must hold ring lock */ 114/**
115 * radeon_semaphore_sync_rings - sync ring to all registered fences
116 *
117 * @rdev: radeon_device pointer
118 * @semaphore: semaphore object to use for sync
119 * @ring: ring that needs sync
120 *
121 * Ensure that all registered fences are signaled before letting
122 * the ring continue. The caller must hold the ring lock.
123 */
71int radeon_semaphore_sync_rings(struct radeon_device *rdev, 124int radeon_semaphore_sync_rings(struct radeon_device *rdev,
72 struct radeon_semaphore *semaphore, 125 struct radeon_semaphore *semaphore,
73 int signaler, int waiter) 126 int ring)
74{ 127{
75 int r; 128 int i, r;
76 129
77 /* no need to signal and wait on the same ring */ 130 for (i = 0; i < RADEON_NUM_RINGS; ++i) {
78 if (signaler == waiter) { 131 struct radeon_fence *fence = semaphore->sync_to[i];
79 return 0;
80 }
81 132
82 /* prevent GPU deadlocks */ 133 /* check if we really need to sync */
83 if (!rdev->ring[signaler].ready) { 134 if (!radeon_fence_need_sync(fence, ring))
84 dev_err(rdev->dev, "Trying to sync to a disabled ring!"); 135 continue;
85 return -EINVAL;
86 }
87 136
88 r = radeon_ring_alloc(rdev, &rdev->ring[signaler], 8); 137 /* prevent GPU deadlocks */
89 if (r) { 138 if (!rdev->ring[i].ready) {
90 return r; 139 dev_err(rdev->dev, "Syncing to a disabled ring!");
91 } 140 return -EINVAL;
92 radeon_semaphore_emit_signal(rdev, signaler, semaphore); 141 }
93 radeon_ring_commit(rdev, &rdev->ring[signaler]);
94 142
95 /* we assume caller has already allocated space on waiters ring */ 143 /* allocate enough space for sync command */
96 radeon_semaphore_emit_wait(rdev, waiter, semaphore); 144 r = radeon_ring_alloc(rdev, &rdev->ring[i], 16);
145 if (r) {
146 return r;
147 }
97 148
98 /* for debugging lockup only, used by sysfs debug files */ 149 /* emit the signal semaphore */
99 rdev->ring[signaler].last_semaphore_signal_addr = semaphore->gpu_addr; 150 if (!radeon_semaphore_emit_signal(rdev, i, semaphore)) {
100 rdev->ring[waiter].last_semaphore_wait_addr = semaphore->gpu_addr; 151 /* signaling wasn't successful wait manually */
152 radeon_ring_undo(&rdev->ring[i]);
153 radeon_fence_wait_locked(fence);
154 continue;
155 }
156
157 /* we assume caller has already allocated space on waiters ring */
158 if (!radeon_semaphore_emit_wait(rdev, ring, semaphore)) {
159 /* waiting wasn't successful wait manually */
160 radeon_ring_undo(&rdev->ring[i]);
161 radeon_fence_wait_locked(fence);
162 continue;
163 }
164
165 radeon_ring_commit(rdev, &rdev->ring[i]);
166 radeon_fence_note_sync(fence, ring);
167 }
101 168
102 return 0; 169 return 0;
103} 170}
diff --git a/drivers/gpu/drm/radeon/radeon_trace.h b/drivers/gpu/drm/radeon/radeon_trace.h
index 811bca691b36..9f0e18172b6e 100644
--- a/drivers/gpu/drm/radeon/radeon_trace.h
+++ b/drivers/gpu/drm/radeon/radeon_trace.h
@@ -111,6 +111,42 @@ DEFINE_EVENT(radeon_fence_request, radeon_fence_wait_end,
111 TP_ARGS(dev, seqno) 111 TP_ARGS(dev, seqno)
112); 112);
113 113
114DECLARE_EVENT_CLASS(radeon_semaphore_request,
115
116 TP_PROTO(int ring, struct radeon_semaphore *sem),
117
118 TP_ARGS(ring, sem),
119
120 TP_STRUCT__entry(
121 __field(int, ring)
122 __field(signed, waiters)
123 __field(uint64_t, gpu_addr)
124 ),
125
126 TP_fast_assign(
127 __entry->ring = ring;
128 __entry->waiters = sem->waiters;
129 __entry->gpu_addr = sem->gpu_addr;
130 ),
131
132 TP_printk("ring=%u, waiters=%d, addr=%010Lx", __entry->ring,
133 __entry->waiters, __entry->gpu_addr)
134);
135
136DEFINE_EVENT(radeon_semaphore_request, radeon_semaphore_signale,
137
138 TP_PROTO(int ring, struct radeon_semaphore *sem),
139
140 TP_ARGS(ring, sem)
141);
142
143DEFINE_EVENT(radeon_semaphore_request, radeon_semaphore_wait,
144
145 TP_PROTO(int ring, struct radeon_semaphore *sem),
146
147 TP_ARGS(ring, sem)
148);
149
114#endif 150#endif
115 151
116/* This part must be outside protection */ 152/* This part must be outside protection */
diff --git a/drivers/gpu/drm/radeon/rv770_dma.c b/drivers/gpu/drm/radeon/rv770_dma.c
index f9b02e3d6830..aca8cbe8a335 100644
--- a/drivers/gpu/drm/radeon/rv770_dma.c
+++ b/drivers/gpu/drm/radeon/rv770_dma.c
@@ -66,13 +66,8 @@ int rv770_copy_dma(struct radeon_device *rdev,
66 return r; 66 return r;
67 } 67 }
68 68
69 if (radeon_fence_need_sync(*fence, ring->idx)) { 69 radeon_semaphore_sync_to(sem, *fence);
70 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 70 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
71 ring->idx);
72 radeon_fence_note_sync(*fence, ring->idx);
73 } else {
74 radeon_semaphore_free(rdev, &sem, NULL);
75 }
76 71
77 for (i = 0; i < num_loops; i++) { 72 for (i = 0; i < num_loops; i++) {
78 cur_size_in_dw = size_in_dw; 73 cur_size_in_dw = size_in_dw;
diff --git a/drivers/gpu/drm/radeon/si_dma.c b/drivers/gpu/drm/radeon/si_dma.c
index 8e8f46133532..59be2cfcbb47 100644
--- a/drivers/gpu/drm/radeon/si_dma.c
+++ b/drivers/gpu/drm/radeon/si_dma.c
@@ -195,13 +195,8 @@ int si_copy_dma(struct radeon_device *rdev,
195 return r; 195 return r;
196 } 196 }
197 197
198 if (radeon_fence_need_sync(*fence, ring->idx)) { 198 radeon_semaphore_sync_to(sem, *fence);
199 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 199 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
200 ring->idx);
201 radeon_fence_note_sync(*fence, ring->idx);
202 } else {
203 radeon_semaphore_free(rdev, &sem, NULL);
204 }
205 200
206 for (i = 0; i < num_loops; i++) { 201 for (i = 0; i < num_loops; i++) {
207 cur_size_in_bytes = size_in_bytes; 202 cur_size_in_bytes = size_in_bytes;
diff --git a/drivers/gpu/drm/radeon/trinity_dpm.c b/drivers/gpu/drm/radeon/trinity_dpm.c
index 9364129ba292..d700698a1f22 100644
--- a/drivers/gpu/drm/radeon/trinity_dpm.c
+++ b/drivers/gpu/drm/radeon/trinity_dpm.c
@@ -1873,9 +1873,9 @@ int trinity_dpm_init(struct radeon_device *rdev)
1873 pi->enable_sclk_ds = true; 1873 pi->enable_sclk_ds = true;
1874 pi->enable_gfx_power_gating = true; 1874 pi->enable_gfx_power_gating = true;
1875 pi->enable_gfx_clock_gating = true; 1875 pi->enable_gfx_clock_gating = true;
1876 pi->enable_mg_clock_gating = true; 1876 pi->enable_mg_clock_gating = false;
1877 pi->enable_gfx_dynamic_mgpg = true; /* ??? */ 1877 pi->enable_gfx_dynamic_mgpg = false;
1878 pi->override_dynamic_mgpg = true; 1878 pi->override_dynamic_mgpg = false;
1879 pi->enable_auto_thermal_throttling = true; 1879 pi->enable_auto_thermal_throttling = true;
1880 pi->voltage_drop_in_dce = false; /* need to restructure dpm/modeset interaction */ 1880 pi->voltage_drop_in_dce = false; /* need to restructure dpm/modeset interaction */
1881 pi->uvd_dpm = true; /* ??? */ 1881 pi->uvd_dpm = true; /* ??? */
diff --git a/drivers/gpu/drm/radeon/uvd_v1_0.c b/drivers/gpu/drm/radeon/uvd_v1_0.c
index 7266805d9786..d4a68af1a279 100644
--- a/drivers/gpu/drm/radeon/uvd_v1_0.c
+++ b/drivers/gpu/drm/radeon/uvd_v1_0.c
@@ -357,7 +357,7 @@ int uvd_v1_0_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
357 * 357 *
358 * Emit a semaphore command (either wait or signal) to the UVD ring. 358 * Emit a semaphore command (either wait or signal) to the UVD ring.
359 */ 359 */
360void uvd_v1_0_semaphore_emit(struct radeon_device *rdev, 360bool uvd_v1_0_semaphore_emit(struct radeon_device *rdev,
361 struct radeon_ring *ring, 361 struct radeon_ring *ring,
362 struct radeon_semaphore *semaphore, 362 struct radeon_semaphore *semaphore,
363 bool emit_wait) 363 bool emit_wait)
@@ -372,6 +372,8 @@ void uvd_v1_0_semaphore_emit(struct radeon_device *rdev,
372 372
373 radeon_ring_write(ring, PACKET0(UVD_SEMA_CMD, 0)); 373 radeon_ring_write(ring, PACKET0(UVD_SEMA_CMD, 0));
374 radeon_ring_write(ring, emit_wait ? 1 : 0); 374 radeon_ring_write(ring, emit_wait ? 1 : 0);
375
376 return true;
375} 377}
376 378
377/** 379/**
diff --git a/drivers/gpu/drm/radeon/uvd_v3_1.c b/drivers/gpu/drm/radeon/uvd_v3_1.c
index 5b6fa1f62d4e..d722db2cf340 100644
--- a/drivers/gpu/drm/radeon/uvd_v3_1.c
+++ b/drivers/gpu/drm/radeon/uvd_v3_1.c
@@ -37,7 +37,7 @@
37 * 37 *
38 * Emit a semaphore command (either wait or signal) to the UVD ring. 38 * Emit a semaphore command (either wait or signal) to the UVD ring.
39 */ 39 */
40void uvd_v3_1_semaphore_emit(struct radeon_device *rdev, 40bool uvd_v3_1_semaphore_emit(struct radeon_device *rdev,
41 struct radeon_ring *ring, 41 struct radeon_ring *ring,
42 struct radeon_semaphore *semaphore, 42 struct radeon_semaphore *semaphore,
43 bool emit_wait) 43 bool emit_wait)
@@ -52,4 +52,6 @@ void uvd_v3_1_semaphore_emit(struct radeon_device *rdev,
52 52
53 radeon_ring_write(ring, PACKET0(UVD_SEMA_CMD, 0)); 53 radeon_ring_write(ring, PACKET0(UVD_SEMA_CMD, 0));
54 radeon_ring_write(ring, 0x80 | (emit_wait ? 1 : 0)); 54 radeon_ring_write(ring, 0x80 | (emit_wait ? 1 : 0));
55
56 return true;
55} 57}
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 8d5a646ebe6a..07e02c4bf5a8 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -151,7 +151,7 @@ static void ttm_bo_release_list(struct kref *list_kref)
151 atomic_dec(&bo->glob->bo_count); 151 atomic_dec(&bo->glob->bo_count);
152 if (bo->resv == &bo->ttm_resv) 152 if (bo->resv == &bo->ttm_resv)
153 reservation_object_fini(&bo->ttm_resv); 153 reservation_object_fini(&bo->ttm_resv);
154 154 mutex_destroy(&bo->wu_mutex);
155 if (bo->destroy) 155 if (bo->destroy)
156 bo->destroy(bo); 156 bo->destroy(bo);
157 else { 157 else {
@@ -1123,6 +1123,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
1123 INIT_LIST_HEAD(&bo->ddestroy); 1123 INIT_LIST_HEAD(&bo->ddestroy);
1124 INIT_LIST_HEAD(&bo->swap); 1124 INIT_LIST_HEAD(&bo->swap);
1125 INIT_LIST_HEAD(&bo->io_reserve_lru); 1125 INIT_LIST_HEAD(&bo->io_reserve_lru);
1126 mutex_init(&bo->wu_mutex);
1126 bo->bdev = bdev; 1127 bo->bdev = bdev;
1127 bo->glob = bdev->glob; 1128 bo->glob = bdev->glob;
1128 bo->type = type; 1129 bo->type = type;
@@ -1704,3 +1705,35 @@ void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
1704 ; 1705 ;
1705} 1706}
1706EXPORT_SYMBOL(ttm_bo_swapout_all); 1707EXPORT_SYMBOL(ttm_bo_swapout_all);
1708
1709/**
1710 * ttm_bo_wait_unreserved - interruptible wait for a buffer object to become
1711 * unreserved
1712 *
1713 * @bo: Pointer to buffer
1714 */
1715int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo)
1716{
1717 int ret;
1718
1719 /*
1720 * In the absense of a wait_unlocked API,
1721 * Use the bo::wu_mutex to avoid triggering livelocks due to
1722 * concurrent use of this function. Note that this use of
1723 * bo::wu_mutex can go away if we change locking order to
1724 * mmap_sem -> bo::reserve.
1725 */
1726 ret = mutex_lock_interruptible(&bo->wu_mutex);
1727 if (unlikely(ret != 0))
1728 return -ERESTARTSYS;
1729 if (!ww_mutex_is_locked(&bo->resv->lock))
1730 goto out_unlock;
1731 ret = ttm_bo_reserve_nolru(bo, true, false, false, NULL);
1732 if (unlikely(ret != 0))
1733 goto out_unlock;
1734 ww_mutex_unlock(&bo->resv->lock);
1735
1736out_unlock:
1737 mutex_unlock(&bo->wu_mutex);
1738 return ret;
1739}
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 4834c463c38b..15b86a94949d 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -350,10 +350,13 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
350 goto out2; 350 goto out2;
351 351
352 /* 352 /*
353 * Move nonexistent data. NOP. 353 * Don't move nonexistent data. Clear destination instead.
354 */ 354 */
355 if (old_iomap == NULL && ttm == NULL) 355 if (old_iomap == NULL &&
356 (ttm == NULL || ttm->state == tt_unpopulated)) {
357 memset_io(new_iomap, 0, new_mem->num_pages*PAGE_SIZE);
356 goto out2; 358 goto out2;
359 }
357 360
358 /* 361 /*
359 * TTM might be null for moves within the same region. 362 * TTM might be null for moves within the same region.
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index ac617f3ecd0c..b249ab9b1eb2 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -107,13 +107,28 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
107 /* 107 /*
108 * Work around locking order reversal in fault / nopfn 108 * Work around locking order reversal in fault / nopfn
109 * between mmap_sem and bo_reserve: Perform a trylock operation 109 * between mmap_sem and bo_reserve: Perform a trylock operation
110 * for reserve, and if it fails, retry the fault after scheduling. 110 * for reserve, and if it fails, retry the fault after waiting
111 * for the buffer to become unreserved.
111 */ 112 */
112 113 ret = ttm_bo_reserve(bo, true, true, false, NULL);
113 ret = ttm_bo_reserve(bo, true, true, false, 0);
114 if (unlikely(ret != 0)) { 114 if (unlikely(ret != 0)) {
115 if (ret == -EBUSY) 115 if (ret != -EBUSY)
116 set_need_resched(); 116 return VM_FAULT_NOPAGE;
117
118 if (vmf->flags & FAULT_FLAG_ALLOW_RETRY) {
119 if (!(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
120 up_read(&vma->vm_mm->mmap_sem);
121 (void) ttm_bo_wait_unreserved(bo);
122 }
123
124 return VM_FAULT_RETRY;
125 }
126
127 /*
128 * If we'd want to change locking order to
129 * mmap_sem -> bo::reserve, we'd use a blocking reserve here
130 * instead of retrying the fault...
131 */
117 return VM_FAULT_NOPAGE; 132 return VM_FAULT_NOPAGE;
118 } 133 }
119 134
@@ -123,7 +138,6 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
123 case 0: 138 case 0:
124 break; 139 break;
125 case -EBUSY: 140 case -EBUSY:
126 set_need_resched();
127 case -ERESTARTSYS: 141 case -ERESTARTSYS:
128 retval = VM_FAULT_NOPAGE; 142 retval = VM_FAULT_NOPAGE;
129 goto out_unlock; 143 goto out_unlock;
diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
index 6c911789ae5c..479e9418e3d7 100644
--- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c
+++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
@@ -32,8 +32,7 @@
32#include <linux/sched.h> 32#include <linux/sched.h>
33#include <linux/module.h> 33#include <linux/module.h>
34 34
35static void ttm_eu_backoff_reservation_locked(struct list_head *list, 35static void ttm_eu_backoff_reservation_locked(struct list_head *list)
36 struct ww_acquire_ctx *ticket)
37{ 36{
38 struct ttm_validate_buffer *entry; 37 struct ttm_validate_buffer *entry;
39 38
@@ -93,8 +92,9 @@ void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket,
93 entry = list_first_entry(list, struct ttm_validate_buffer, head); 92 entry = list_first_entry(list, struct ttm_validate_buffer, head);
94 glob = entry->bo->glob; 93 glob = entry->bo->glob;
95 spin_lock(&glob->lru_lock); 94 spin_lock(&glob->lru_lock);
96 ttm_eu_backoff_reservation_locked(list, ticket); 95 ttm_eu_backoff_reservation_locked(list);
97 ww_acquire_fini(ticket); 96 if (ticket)
97 ww_acquire_fini(ticket);
98 spin_unlock(&glob->lru_lock); 98 spin_unlock(&glob->lru_lock);
99} 99}
100EXPORT_SYMBOL(ttm_eu_backoff_reservation); 100EXPORT_SYMBOL(ttm_eu_backoff_reservation);
@@ -130,7 +130,8 @@ int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket,
130 entry = list_first_entry(list, struct ttm_validate_buffer, head); 130 entry = list_first_entry(list, struct ttm_validate_buffer, head);
131 glob = entry->bo->glob; 131 glob = entry->bo->glob;
132 132
133 ww_acquire_init(ticket, &reservation_ww_class); 133 if (ticket)
134 ww_acquire_init(ticket, &reservation_ww_class);
134retry: 135retry:
135 list_for_each_entry(entry, list, head) { 136 list_for_each_entry(entry, list, head) {
136 struct ttm_buffer_object *bo = entry->bo; 137 struct ttm_buffer_object *bo = entry->bo;
@@ -139,16 +140,17 @@ retry:
139 if (entry->reserved) 140 if (entry->reserved)
140 continue; 141 continue;
141 142
142 143 ret = ttm_bo_reserve_nolru(bo, true, (ticket == NULL), true,
143 ret = ttm_bo_reserve_nolru(bo, true, false, true, ticket); 144 ticket);
144 145
145 if (ret == -EDEADLK) { 146 if (ret == -EDEADLK) {
146 /* uh oh, we lost out, drop every reservation and try 147 /* uh oh, we lost out, drop every reservation and try
147 * to only reserve this buffer, then start over if 148 * to only reserve this buffer, then start over if
148 * this succeeds. 149 * this succeeds.
149 */ 150 */
151 BUG_ON(ticket == NULL);
150 spin_lock(&glob->lru_lock); 152 spin_lock(&glob->lru_lock);
151 ttm_eu_backoff_reservation_locked(list, ticket); 153 ttm_eu_backoff_reservation_locked(list);
152 spin_unlock(&glob->lru_lock); 154 spin_unlock(&glob->lru_lock);
153 ttm_eu_list_ref_sub(list); 155 ttm_eu_list_ref_sub(list);
154 ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock, 156 ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock,
@@ -175,7 +177,8 @@ retry:
175 } 177 }
176 } 178 }
177 179
178 ww_acquire_done(ticket); 180 if (ticket)
181 ww_acquire_done(ticket);
179 spin_lock(&glob->lru_lock); 182 spin_lock(&glob->lru_lock);
180 ttm_eu_del_from_lru_locked(list); 183 ttm_eu_del_from_lru_locked(list);
181 spin_unlock(&glob->lru_lock); 184 spin_unlock(&glob->lru_lock);
@@ -184,12 +187,14 @@ retry:
184 187
185err: 188err:
186 spin_lock(&glob->lru_lock); 189 spin_lock(&glob->lru_lock);
187 ttm_eu_backoff_reservation_locked(list, ticket); 190 ttm_eu_backoff_reservation_locked(list);
188 spin_unlock(&glob->lru_lock); 191 spin_unlock(&glob->lru_lock);
189 ttm_eu_list_ref_sub(list); 192 ttm_eu_list_ref_sub(list);
190err_fini: 193err_fini:
191 ww_acquire_done(ticket); 194 if (ticket) {
192 ww_acquire_fini(ticket); 195 ww_acquire_done(ticket);
196 ww_acquire_fini(ticket);
197 }
193 return ret; 198 return ret;
194} 199}
195EXPORT_SYMBOL(ttm_eu_reserve_buffers); 200EXPORT_SYMBOL(ttm_eu_reserve_buffers);
@@ -224,7 +229,8 @@ void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket,
224 } 229 }
225 spin_unlock(&bdev->fence_lock); 230 spin_unlock(&bdev->fence_lock);
226 spin_unlock(&glob->lru_lock); 231 spin_unlock(&glob->lru_lock);
227 ww_acquire_fini(ticket); 232 if (ticket)
233 ww_acquire_fini(ticket);
228 234
229 list_for_each_entry(entry, list, head) { 235 list_for_each_entry(entry, list, head) {
230 if (entry->old_sync_obj) 236 if (entry->old_sync_obj)
diff --git a/drivers/gpu/drm/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c
index a868176c258a..6fe7b92a82d1 100644
--- a/drivers/gpu/drm/ttm/ttm_object.c
+++ b/drivers/gpu/drm/ttm/ttm_object.c
@@ -1,6 +1,6 @@
1/************************************************************************** 1/**************************************************************************
2 * 2 *
3 * Copyright (c) 2009 VMware, Inc., Palo Alto, CA., USA 3 * Copyright (c) 2009-2013 VMware, Inc., Palo Alto, CA., USA
4 * All Rights Reserved. 4 * All Rights Reserved.
5 * 5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a 6 * Permission is hereby granted, free of charge, to any person obtaining a
@@ -26,6 +26,12 @@
26 **************************************************************************/ 26 **************************************************************************/
27/* 27/*
28 * Authors: Thomas Hellstrom <thellstrom-at-vmware-dot-com> 28 * Authors: Thomas Hellstrom <thellstrom-at-vmware-dot-com>
29 *
30 * While no substantial code is shared, the prime code is inspired by
31 * drm_prime.c, with
32 * Authors:
33 * Dave Airlie <airlied@redhat.com>
34 * Rob Clark <rob.clark@linaro.org>
29 */ 35 */
30/** @file ttm_ref_object.c 36/** @file ttm_ref_object.c
31 * 37 *
@@ -34,6 +40,7 @@
34 * and release on file close. 40 * and release on file close.
35 */ 41 */
36 42
43
37/** 44/**
38 * struct ttm_object_file 45 * struct ttm_object_file
39 * 46 *
@@ -84,6 +91,9 @@ struct ttm_object_device {
84 struct drm_open_hash object_hash; 91 struct drm_open_hash object_hash;
85 atomic_t object_count; 92 atomic_t object_count;
86 struct ttm_mem_global *mem_glob; 93 struct ttm_mem_global *mem_glob;
94 struct dma_buf_ops ops;
95 void (*dmabuf_release)(struct dma_buf *dma_buf);
96 size_t dma_buf_size;
87}; 97};
88 98
89/** 99/**
@@ -116,6 +126,8 @@ struct ttm_ref_object {
116 struct ttm_object_file *tfile; 126 struct ttm_object_file *tfile;
117}; 127};
118 128
129static void ttm_prime_dmabuf_release(struct dma_buf *dma_buf);
130
119static inline struct ttm_object_file * 131static inline struct ttm_object_file *
120ttm_object_file_ref(struct ttm_object_file *tfile) 132ttm_object_file_ref(struct ttm_object_file *tfile)
121{ 133{
@@ -416,9 +428,10 @@ out_err:
416} 428}
417EXPORT_SYMBOL(ttm_object_file_init); 429EXPORT_SYMBOL(ttm_object_file_init);
418 430
419struct ttm_object_device *ttm_object_device_init(struct ttm_mem_global 431struct ttm_object_device *
420 *mem_glob, 432ttm_object_device_init(struct ttm_mem_global *mem_glob,
421 unsigned int hash_order) 433 unsigned int hash_order,
434 const struct dma_buf_ops *ops)
422{ 435{
423 struct ttm_object_device *tdev = kmalloc(sizeof(*tdev), GFP_KERNEL); 436 struct ttm_object_device *tdev = kmalloc(sizeof(*tdev), GFP_KERNEL);
424 int ret; 437 int ret;
@@ -430,10 +443,17 @@ struct ttm_object_device *ttm_object_device_init(struct ttm_mem_global
430 spin_lock_init(&tdev->object_lock); 443 spin_lock_init(&tdev->object_lock);
431 atomic_set(&tdev->object_count, 0); 444 atomic_set(&tdev->object_count, 0);
432 ret = drm_ht_create(&tdev->object_hash, hash_order); 445 ret = drm_ht_create(&tdev->object_hash, hash_order);
446 if (ret != 0)
447 goto out_no_object_hash;
433 448
434 if (likely(ret == 0)) 449 tdev->ops = *ops;
435 return tdev; 450 tdev->dmabuf_release = tdev->ops.release;
451 tdev->ops.release = ttm_prime_dmabuf_release;
452 tdev->dma_buf_size = ttm_round_pot(sizeof(struct dma_buf)) +
453 ttm_round_pot(sizeof(struct file));
454 return tdev;
436 455
456out_no_object_hash:
437 kfree(tdev); 457 kfree(tdev);
438 return NULL; 458 return NULL;
439} 459}
@@ -452,3 +472,225 @@ void ttm_object_device_release(struct ttm_object_device **p_tdev)
452 kfree(tdev); 472 kfree(tdev);
453} 473}
454EXPORT_SYMBOL(ttm_object_device_release); 474EXPORT_SYMBOL(ttm_object_device_release);
475
476/**
477 * get_dma_buf_unless_doomed - get a dma_buf reference if possible.
478 *
479 * @dma_buf: Non-refcounted pointer to a struct dma-buf.
480 *
481 * Obtain a file reference from a lookup structure that doesn't refcount
482 * the file, but synchronizes with its release method to make sure it has
483 * not been freed yet. See for example kref_get_unless_zero documentation.
484 * Returns true if refcounting succeeds, false otherwise.
485 *
486 * Nobody really wants this as a public API yet, so let it mature here
487 * for some time...
488 */
489static bool __must_check get_dma_buf_unless_doomed(struct dma_buf *dmabuf)
490{
491 return atomic_long_inc_not_zero(&dmabuf->file->f_count) != 0L;
492}
493
494/**
495 * ttm_prime_refcount_release - refcount release method for a prime object.
496 *
497 * @p_base: Pointer to ttm_base_object pointer.
498 *
499 * This is a wrapper that calls the refcount_release founction of the
500 * underlying object. At the same time it cleans up the prime object.
501 * This function is called when all references to the base object we
502 * derive from are gone.
503 */
504static void ttm_prime_refcount_release(struct ttm_base_object **p_base)
505{
506 struct ttm_base_object *base = *p_base;
507 struct ttm_prime_object *prime;
508
509 *p_base = NULL;
510 prime = container_of(base, struct ttm_prime_object, base);
511 BUG_ON(prime->dma_buf != NULL);
512 mutex_destroy(&prime->mutex);
513 if (prime->refcount_release)
514 prime->refcount_release(&base);
515}
516
517/**
518 * ttm_prime_dmabuf_release - Release method for the dma-bufs we export
519 *
520 * @dma_buf:
521 *
522 * This function first calls the dma_buf release method the driver
523 * provides. Then it cleans up our dma_buf pointer used for lookup,
524 * and finally releases the reference the dma_buf has on our base
525 * object.
526 */
527static void ttm_prime_dmabuf_release(struct dma_buf *dma_buf)
528{
529 struct ttm_prime_object *prime =
530 (struct ttm_prime_object *) dma_buf->priv;
531 struct ttm_base_object *base = &prime->base;
532 struct ttm_object_device *tdev = base->tfile->tdev;
533
534 if (tdev->dmabuf_release)
535 tdev->dmabuf_release(dma_buf);
536 mutex_lock(&prime->mutex);
537 if (prime->dma_buf == dma_buf)
538 prime->dma_buf = NULL;
539 mutex_unlock(&prime->mutex);
540 ttm_mem_global_free(tdev->mem_glob, tdev->dma_buf_size);
541 ttm_base_object_unref(&base);
542}
543
544/**
545 * ttm_prime_fd_to_handle - Get a base object handle from a prime fd
546 *
547 * @tfile: A struct ttm_object_file identifying the caller.
548 * @fd: The prime / dmabuf fd.
549 * @handle: The returned handle.
550 *
551 * This function returns a handle to an object that previously exported
552 * a dma-buf. Note that we don't handle imports yet, because we simply
553 * have no consumers of that implementation.
554 */
555int ttm_prime_fd_to_handle(struct ttm_object_file *tfile,
556 int fd, u32 *handle)
557{
558 struct ttm_object_device *tdev = tfile->tdev;
559 struct dma_buf *dma_buf;
560 struct ttm_prime_object *prime;
561 struct ttm_base_object *base;
562 int ret;
563
564 dma_buf = dma_buf_get(fd);
565 if (IS_ERR(dma_buf))
566 return PTR_ERR(dma_buf);
567
568 if (dma_buf->ops != &tdev->ops)
569 return -ENOSYS;
570
571 prime = (struct ttm_prime_object *) dma_buf->priv;
572 base = &prime->base;
573 *handle = base->hash.key;
574 ret = ttm_ref_object_add(tfile, base, TTM_REF_USAGE, NULL);
575
576 dma_buf_put(dma_buf);
577
578 return ret;
579}
580EXPORT_SYMBOL_GPL(ttm_prime_fd_to_handle);
581
582/**
583 * ttm_prime_handle_to_fd - Return a dma_buf fd from a ttm prime object
584 *
585 * @tfile: Struct ttm_object_file identifying the caller.
586 * @handle: Handle to the object we're exporting from.
587 * @flags: flags for dma-buf creation. We just pass them on.
588 * @prime_fd: The returned file descriptor.
589 *
590 */
591int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
592 uint32_t handle, uint32_t flags,
593 int *prime_fd)
594{
595 struct ttm_object_device *tdev = tfile->tdev;
596 struct ttm_base_object *base;
597 struct dma_buf *dma_buf;
598 struct ttm_prime_object *prime;
599 int ret;
600
601 base = ttm_base_object_lookup(tfile, handle);
602 if (unlikely(base == NULL ||
603 base->object_type != ttm_prime_type)) {
604 ret = -ENOENT;
605 goto out_unref;
606 }
607
608 prime = container_of(base, struct ttm_prime_object, base);
609 if (unlikely(!base->shareable)) {
610 ret = -EPERM;
611 goto out_unref;
612 }
613
614 ret = mutex_lock_interruptible(&prime->mutex);
615 if (unlikely(ret != 0)) {
616 ret = -ERESTARTSYS;
617 goto out_unref;
618 }
619
620 dma_buf = prime->dma_buf;
621 if (!dma_buf || !get_dma_buf_unless_doomed(dma_buf)) {
622
623 /*
624 * Need to create a new dma_buf, with memory accounting.
625 */
626 ret = ttm_mem_global_alloc(tdev->mem_glob, tdev->dma_buf_size,
627 false, true);
628 if (unlikely(ret != 0)) {
629 mutex_unlock(&prime->mutex);
630 goto out_unref;
631 }
632
633 dma_buf = dma_buf_export(prime, &tdev->ops,
634 prime->size, flags);
635 if (IS_ERR(dma_buf)) {
636 ret = PTR_ERR(dma_buf);
637 ttm_mem_global_free(tdev->mem_glob,
638 tdev->dma_buf_size);
639 mutex_unlock(&prime->mutex);
640 goto out_unref;
641 }
642
643 /*
644 * dma_buf has taken the base object reference
645 */
646 base = NULL;
647 prime->dma_buf = dma_buf;
648 }
649 mutex_unlock(&prime->mutex);
650
651 ret = dma_buf_fd(dma_buf, flags);
652 if (ret >= 0) {
653 *prime_fd = ret;
654 ret = 0;
655 } else
656 dma_buf_put(dma_buf);
657
658out_unref:
659 if (base)
660 ttm_base_object_unref(&base);
661 return ret;
662}
663EXPORT_SYMBOL_GPL(ttm_prime_handle_to_fd);
664
665/**
666 * ttm_prime_object_init - Initialize a ttm_prime_object
667 *
668 * @tfile: struct ttm_object_file identifying the caller
669 * @size: The size of the dma_bufs we export.
670 * @prime: The object to be initialized.
671 * @shareable: See ttm_base_object_init
672 * @type: See ttm_base_object_init
673 * @refcount_release: See ttm_base_object_init
674 * @ref_obj_release: See ttm_base_object_init
675 *
676 * Initializes an object which is compatible with the drm_prime model
677 * for data sharing between processes and devices.
678 */
679int ttm_prime_object_init(struct ttm_object_file *tfile, size_t size,
680 struct ttm_prime_object *prime, bool shareable,
681 enum ttm_object_type type,
682 void (*refcount_release) (struct ttm_base_object **),
683 void (*ref_obj_release) (struct ttm_base_object *,
684 enum ttm_ref_type ref_type))
685{
686 mutex_init(&prime->mutex);
687 prime->size = PAGE_ALIGN(size);
688 prime->real_type = type;
689 prime->dma_buf = NULL;
690 prime->refcount_release = refcount_release;
691 return ttm_base_object_init(tfile, &prime->base, shareable,
692 ttm_prime_type,
693 ttm_prime_refcount_release,
694 ref_obj_release);
695}
696EXPORT_SYMBOL(ttm_prime_object_init);
diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile
index 2cc6cd91ac11..9f8b690bcf52 100644
--- a/drivers/gpu/drm/vmwgfx/Makefile
+++ b/drivers/gpu/drm/vmwgfx/Makefile
@@ -6,6 +6,6 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
6 vmwgfx_fifo.o vmwgfx_irq.o vmwgfx_ldu.o vmwgfx_ttm_glue.o \ 6 vmwgfx_fifo.o vmwgfx_irq.o vmwgfx_ldu.o vmwgfx_ttm_glue.o \
7 vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \ 7 vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \
8 vmwgfx_fence.o vmwgfx_dmabuf.o vmwgfx_scrn.o vmwgfx_context.o \ 8 vmwgfx_fence.o vmwgfx_dmabuf.o vmwgfx_scrn.o vmwgfx_context.o \
9 vmwgfx_surface.o 9 vmwgfx_surface.o vmwgfx_prime.o
10 10
11obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o 11obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 20d5485eaf98..c7a549694e59 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -677,7 +677,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
677 } 677 }
678 678
679 dev_priv->tdev = ttm_object_device_init 679 dev_priv->tdev = ttm_object_device_init
680 (dev_priv->mem_global_ref.object, 12); 680 (dev_priv->mem_global_ref.object, 12, &vmw_prime_dmabuf_ops);
681 681
682 if (unlikely(dev_priv->tdev == NULL)) { 682 if (unlikely(dev_priv->tdev == NULL)) {
683 DRM_ERROR("Unable to initialize TTM object management.\n"); 683 DRM_ERROR("Unable to initialize TTM object management.\n");
@@ -1210,7 +1210,7 @@ static const struct file_operations vmwgfx_driver_fops = {
1210 1210
1211static struct drm_driver driver = { 1211static struct drm_driver driver = {
1212 .driver_features = DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | 1212 .driver_features = DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED |
1213 DRIVER_MODESET, 1213 DRIVER_MODESET | DRIVER_PRIME,
1214 .load = vmw_driver_load, 1214 .load = vmw_driver_load,
1215 .unload = vmw_driver_unload, 1215 .unload = vmw_driver_unload,
1216 .lastclose = vmw_lastclose, 1216 .lastclose = vmw_lastclose,
@@ -1235,6 +1235,9 @@ static struct drm_driver driver = {
1235 .dumb_map_offset = vmw_dumb_map_offset, 1235 .dumb_map_offset = vmw_dumb_map_offset,
1236 .dumb_destroy = vmw_dumb_destroy, 1236 .dumb_destroy = vmw_dumb_destroy,
1237 1237
1238 .prime_fd_to_handle = vmw_prime_fd_to_handle,
1239 .prime_handle_to_fd = vmw_prime_handle_to_fd,
1240
1238 .fops = &vmwgfx_driver_fops, 1241 .fops = &vmwgfx_driver_fops,
1239 .name = VMWGFX_DRIVER_NAME, 1242 .name = VMWGFX_DRIVER_NAME,
1240 .desc = VMWGFX_DRIVER_DESC, 1243 .desc = VMWGFX_DRIVER_DESC,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index e401d5dbcb96..db85985c7086 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -819,6 +819,20 @@ int vmw_overlay_num_free_overlays(struct vmw_private *dev_priv);
819extern const struct ttm_mem_type_manager_func vmw_gmrid_manager_func; 819extern const struct ttm_mem_type_manager_func vmw_gmrid_manager_func;
820 820
821/** 821/**
822 * Prime - vmwgfx_prime.c
823 */
824
825extern const struct dma_buf_ops vmw_prime_dmabuf_ops;
826extern int vmw_prime_fd_to_handle(struct drm_device *dev,
827 struct drm_file *file_priv,
828 int fd, u32 *handle);
829extern int vmw_prime_handle_to_fd(struct drm_device *dev,
830 struct drm_file *file_priv,
831 uint32_t handle, uint32_t flags,
832 int *prime_fd);
833
834
835/**
822 * Inline helper functions 836 * Inline helper functions
823 */ 837 */
824 838
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_prime.c b/drivers/gpu/drm/vmwgfx/vmwgfx_prime.c
new file mode 100644
index 000000000000..31fe32d8d65a
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_prime.c
@@ -0,0 +1,137 @@
1/**************************************************************************
2 *
3 * Copyright © 2013 VMware, Inc., Palo Alto, CA., USA
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 * USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27/*
28 * Authors:
29 * Thomas Hellstrom <thellstrom@vmware.com>
30 *
31 */
32
33#include "vmwgfx_drv.h"
34#include <linux/dma-buf.h>
35#include <drm/ttm/ttm_object.h>
36
37/*
38 * DMA-BUF attach- and mapping methods. No need to implement
39 * these until we have other virtual devices use them.
40 */
41
42static int vmw_prime_map_attach(struct dma_buf *dma_buf,
43 struct device *target_dev,
44 struct dma_buf_attachment *attach)
45{
46 return -ENOSYS;
47}
48
49static void vmw_prime_map_detach(struct dma_buf *dma_buf,
50 struct dma_buf_attachment *attach)
51{
52}
53
54static struct sg_table *vmw_prime_map_dma_buf(struct dma_buf_attachment *attach,
55 enum dma_data_direction dir)
56{
57 return ERR_PTR(-ENOSYS);
58}
59
60static void vmw_prime_unmap_dma_buf(struct dma_buf_attachment *attach,
61 struct sg_table *sgb,
62 enum dma_data_direction dir)
63{
64}
65
66static void *vmw_prime_dmabuf_vmap(struct dma_buf *dma_buf)
67{
68 return NULL;
69}
70
71static void vmw_prime_dmabuf_vunmap(struct dma_buf *dma_buf, void *vaddr)
72{
73}
74
75static void *vmw_prime_dmabuf_kmap_atomic(struct dma_buf *dma_buf,
76 unsigned long page_num)
77{
78 return NULL;
79}
80
81static void vmw_prime_dmabuf_kunmap_atomic(struct dma_buf *dma_buf,
82 unsigned long page_num, void *addr)
83{
84
85}
86static void *vmw_prime_dmabuf_kmap(struct dma_buf *dma_buf,
87 unsigned long page_num)
88{
89 return NULL;
90}
91
92static void vmw_prime_dmabuf_kunmap(struct dma_buf *dma_buf,
93 unsigned long page_num, void *addr)
94{
95
96}
97
98static int vmw_prime_dmabuf_mmap(struct dma_buf *dma_buf,
99 struct vm_area_struct *vma)
100{
101 WARN_ONCE(true, "Attempted use of dmabuf mmap. Bad.\n");
102 return -ENOSYS;
103}
104
105const struct dma_buf_ops vmw_prime_dmabuf_ops = {
106 .attach = vmw_prime_map_attach,
107 .detach = vmw_prime_map_detach,
108 .map_dma_buf = vmw_prime_map_dma_buf,
109 .unmap_dma_buf = vmw_prime_unmap_dma_buf,
110 .release = NULL,
111 .kmap = vmw_prime_dmabuf_kmap,
112 .kmap_atomic = vmw_prime_dmabuf_kmap_atomic,
113 .kunmap = vmw_prime_dmabuf_kunmap,
114 .kunmap_atomic = vmw_prime_dmabuf_kunmap_atomic,
115 .mmap = vmw_prime_dmabuf_mmap,
116 .vmap = vmw_prime_dmabuf_vmap,
117 .vunmap = vmw_prime_dmabuf_vunmap,
118};
119
120int vmw_prime_fd_to_handle(struct drm_device *dev,
121 struct drm_file *file_priv,
122 int fd, u32 *handle)
123{
124 struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
125
126 return ttm_prime_fd_to_handle(tfile, fd, handle);
127}
128
129int vmw_prime_handle_to_fd(struct drm_device *dev,
130 struct drm_file *file_priv,
131 uint32_t handle, uint32_t flags,
132 int *prime_fd)
133{
134 struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
135
136 return ttm_prime_handle_to_fd(tfile, handle, flags, prime_fd);
137}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 252501a54def..efe2b74c5eb1 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -35,7 +35,7 @@
35#define VMW_RES_EVICT_ERR_COUNT 10 35#define VMW_RES_EVICT_ERR_COUNT 10
36 36
37struct vmw_user_dma_buffer { 37struct vmw_user_dma_buffer {
38 struct ttm_base_object base; 38 struct ttm_prime_object prime;
39 struct vmw_dma_buffer dma; 39 struct vmw_dma_buffer dma;
40}; 40};
41 41
@@ -297,7 +297,7 @@ int vmw_user_resource_lookup_handle(struct vmw_private *dev_priv,
297 if (unlikely(base == NULL)) 297 if (unlikely(base == NULL))
298 return -EINVAL; 298 return -EINVAL;
299 299
300 if (unlikely(base->object_type != converter->object_type)) 300 if (unlikely(ttm_base_object_type(base) != converter->object_type))
301 goto out_bad_resource; 301 goto out_bad_resource;
302 302
303 res = converter->base_obj_to_res(base); 303 res = converter->base_obj_to_res(base);
@@ -387,7 +387,7 @@ static void vmw_user_dmabuf_destroy(struct ttm_buffer_object *bo)
387{ 387{
388 struct vmw_user_dma_buffer *vmw_user_bo = vmw_user_dma_buffer(bo); 388 struct vmw_user_dma_buffer *vmw_user_bo = vmw_user_dma_buffer(bo);
389 389
390 ttm_base_object_kfree(vmw_user_bo, base); 390 ttm_prime_object_kfree(vmw_user_bo, prime);
391} 391}
392 392
393static void vmw_user_dmabuf_release(struct ttm_base_object **p_base) 393static void vmw_user_dmabuf_release(struct ttm_base_object **p_base)
@@ -401,7 +401,8 @@ static void vmw_user_dmabuf_release(struct ttm_base_object **p_base)
401 if (unlikely(base == NULL)) 401 if (unlikely(base == NULL))
402 return; 402 return;
403 403
404 vmw_user_bo = container_of(base, struct vmw_user_dma_buffer, base); 404 vmw_user_bo = container_of(base, struct vmw_user_dma_buffer,
405 prime.base);
405 bo = &vmw_user_bo->dma.base; 406 bo = &vmw_user_bo->dma.base;
406 ttm_bo_unref(&bo); 407 ttm_bo_unref(&bo);
407} 408}
@@ -442,18 +443,19 @@ int vmw_user_dmabuf_alloc(struct vmw_private *dev_priv,
442 return ret; 443 return ret;
443 444
444 tmp = ttm_bo_reference(&user_bo->dma.base); 445 tmp = ttm_bo_reference(&user_bo->dma.base);
445 ret = ttm_base_object_init(tfile, 446 ret = ttm_prime_object_init(tfile,
446 &user_bo->base, 447 size,
447 shareable, 448 &user_bo->prime,
448 ttm_buffer_type, 449 shareable,
449 &vmw_user_dmabuf_release, NULL); 450 ttm_buffer_type,
451 &vmw_user_dmabuf_release, NULL);
450 if (unlikely(ret != 0)) { 452 if (unlikely(ret != 0)) {
451 ttm_bo_unref(&tmp); 453 ttm_bo_unref(&tmp);
452 goto out_no_base_object; 454 goto out_no_base_object;
453 } 455 }
454 456
455 *p_dma_buf = &user_bo->dma; 457 *p_dma_buf = &user_bo->dma;
456 *handle = user_bo->base.hash.key; 458 *handle = user_bo->prime.base.hash.key;
457 459
458out_no_base_object: 460out_no_base_object:
459 return ret; 461 return ret;
@@ -475,8 +477,8 @@ int vmw_user_dmabuf_verify_access(struct ttm_buffer_object *bo,
475 return -EPERM; 477 return -EPERM;
476 478
477 vmw_user_bo = vmw_user_dma_buffer(bo); 479 vmw_user_bo = vmw_user_dma_buffer(bo);
478 return (vmw_user_bo->base.tfile == tfile || 480 return (vmw_user_bo->prime.base.tfile == tfile ||
479 vmw_user_bo->base.shareable) ? 0 : -EPERM; 481 vmw_user_bo->prime.base.shareable) ? 0 : -EPERM;
480} 482}
481 483
482int vmw_dmabuf_alloc_ioctl(struct drm_device *dev, void *data, 484int vmw_dmabuf_alloc_ioctl(struct drm_device *dev, void *data,
@@ -538,14 +540,15 @@ int vmw_user_dmabuf_lookup(struct ttm_object_file *tfile,
538 return -ESRCH; 540 return -ESRCH;
539 } 541 }
540 542
541 if (unlikely(base->object_type != ttm_buffer_type)) { 543 if (unlikely(ttm_base_object_type(base) != ttm_buffer_type)) {
542 ttm_base_object_unref(&base); 544 ttm_base_object_unref(&base);
543 printk(KERN_ERR "Invalid buffer object handle 0x%08lx.\n", 545 printk(KERN_ERR "Invalid buffer object handle 0x%08lx.\n",
544 (unsigned long)handle); 546 (unsigned long)handle);
545 return -EINVAL; 547 return -EINVAL;
546 } 548 }
547 549
548 vmw_user_bo = container_of(base, struct vmw_user_dma_buffer, base); 550 vmw_user_bo = container_of(base, struct vmw_user_dma_buffer,
551 prime.base);
549 (void)ttm_bo_reference(&vmw_user_bo->dma.base); 552 (void)ttm_bo_reference(&vmw_user_bo->dma.base);
550 ttm_base_object_unref(&base); 553 ttm_base_object_unref(&base);
551 *out = &vmw_user_bo->dma; 554 *out = &vmw_user_bo->dma;
@@ -562,7 +565,8 @@ int vmw_user_dmabuf_reference(struct ttm_object_file *tfile,
562 return -EINVAL; 565 return -EINVAL;
563 566
564 user_bo = container_of(dma_buf, struct vmw_user_dma_buffer, dma); 567 user_bo = container_of(dma_buf, struct vmw_user_dma_buffer, dma);
565 return ttm_ref_object_add(tfile, &user_bo->base, TTM_REF_USAGE, NULL); 568 return ttm_ref_object_add(tfile, &user_bo->prime.base,
569 TTM_REF_USAGE, NULL);
566} 570}
567 571
568/* 572/*
@@ -807,15 +811,16 @@ int vmw_dumb_create(struct drm_file *file_priv,
807 goto out_no_dmabuf; 811 goto out_no_dmabuf;
808 812
809 tmp = ttm_bo_reference(&vmw_user_bo->dma.base); 813 tmp = ttm_bo_reference(&vmw_user_bo->dma.base);
810 ret = ttm_base_object_init(vmw_fpriv(file_priv)->tfile, 814 ret = ttm_prime_object_init(vmw_fpriv(file_priv)->tfile,
811 &vmw_user_bo->base, 815 args->size,
812 false, 816 &vmw_user_bo->prime,
813 ttm_buffer_type, 817 false,
814 &vmw_user_dmabuf_release, NULL); 818 ttm_buffer_type,
819 &vmw_user_dmabuf_release, NULL);
815 if (unlikely(ret != 0)) 820 if (unlikely(ret != 0))
816 goto out_no_base_object; 821 goto out_no_base_object;
817 822
818 args->handle = vmw_user_bo->base.hash.key; 823 args->handle = vmw_user_bo->prime.base.hash.key;
819 824
820out_no_base_object: 825out_no_base_object:
821 ttm_bo_unref(&tmp); 826 ttm_bo_unref(&tmp);
@@ -994,7 +999,6 @@ void vmw_resource_unreserve(struct vmw_resource *res,
994 */ 999 */
995static int 1000static int
996vmw_resource_check_buffer(struct vmw_resource *res, 1001vmw_resource_check_buffer(struct vmw_resource *res,
997 struct ww_acquire_ctx *ticket,
998 bool interruptible, 1002 bool interruptible,
999 struct ttm_validate_buffer *val_buf) 1003 struct ttm_validate_buffer *val_buf)
1000{ 1004{
@@ -1011,7 +1015,7 @@ vmw_resource_check_buffer(struct vmw_resource *res,
1011 INIT_LIST_HEAD(&val_list); 1015 INIT_LIST_HEAD(&val_list);
1012 val_buf->bo = ttm_bo_reference(&res->backup->base); 1016 val_buf->bo = ttm_bo_reference(&res->backup->base);
1013 list_add_tail(&val_buf->head, &val_list); 1017 list_add_tail(&val_buf->head, &val_list);
1014 ret = ttm_eu_reserve_buffers(ticket, &val_list); 1018 ret = ttm_eu_reserve_buffers(NULL, &val_list);
1015 if (unlikely(ret != 0)) 1019 if (unlikely(ret != 0))
1016 goto out_no_reserve; 1020 goto out_no_reserve;
1017 1021
@@ -1029,7 +1033,7 @@ vmw_resource_check_buffer(struct vmw_resource *res,
1029 return 0; 1033 return 0;
1030 1034
1031out_no_validate: 1035out_no_validate:
1032 ttm_eu_backoff_reservation(ticket, &val_list); 1036 ttm_eu_backoff_reservation(NULL, &val_list);
1033out_no_reserve: 1037out_no_reserve:
1034 ttm_bo_unref(&val_buf->bo); 1038 ttm_bo_unref(&val_buf->bo);
1035 if (backup_dirty) 1039 if (backup_dirty)
@@ -1074,8 +1078,7 @@ int vmw_resource_reserve(struct vmw_resource *res, bool no_backup)
1074 * @val_buf: Backup buffer information. 1078 * @val_buf: Backup buffer information.
1075 */ 1079 */
1076static void 1080static void
1077vmw_resource_backoff_reservation(struct ww_acquire_ctx *ticket, 1081vmw_resource_backoff_reservation(struct ttm_validate_buffer *val_buf)
1078 struct ttm_validate_buffer *val_buf)
1079{ 1082{
1080 struct list_head val_list; 1083 struct list_head val_list;
1081 1084
@@ -1084,7 +1087,7 @@ vmw_resource_backoff_reservation(struct ww_acquire_ctx *ticket,
1084 1087
1085 INIT_LIST_HEAD(&val_list); 1088 INIT_LIST_HEAD(&val_list);
1086 list_add_tail(&val_buf->head, &val_list); 1089 list_add_tail(&val_buf->head, &val_list);
1087 ttm_eu_backoff_reservation(ticket, &val_list); 1090 ttm_eu_backoff_reservation(NULL, &val_list);
1088 ttm_bo_unref(&val_buf->bo); 1091 ttm_bo_unref(&val_buf->bo);
1089} 1092}
1090 1093
@@ -1099,14 +1102,12 @@ int vmw_resource_do_evict(struct vmw_resource *res, bool interruptible)
1099{ 1102{
1100 struct ttm_validate_buffer val_buf; 1103 struct ttm_validate_buffer val_buf;
1101 const struct vmw_res_func *func = res->func; 1104 const struct vmw_res_func *func = res->func;
1102 struct ww_acquire_ctx ticket;
1103 int ret; 1105 int ret;
1104 1106
1105 BUG_ON(!func->may_evict); 1107 BUG_ON(!func->may_evict);
1106 1108
1107 val_buf.bo = NULL; 1109 val_buf.bo = NULL;
1108 ret = vmw_resource_check_buffer(res, &ticket, interruptible, 1110 ret = vmw_resource_check_buffer(res, interruptible, &val_buf);
1109 &val_buf);
1110 if (unlikely(ret != 0)) 1111 if (unlikely(ret != 0))
1111 return ret; 1112 return ret;
1112 1113
@@ -1121,7 +1122,7 @@ int vmw_resource_do_evict(struct vmw_resource *res, bool interruptible)
1121 res->backup_dirty = true; 1122 res->backup_dirty = true;
1122 res->res_dirty = false; 1123 res->res_dirty = false;
1123out_no_unbind: 1124out_no_unbind:
1124 vmw_resource_backoff_reservation(&ticket, &val_buf); 1125 vmw_resource_backoff_reservation(&val_buf);
1125 1126
1126 return ret; 1127 return ret;
1127} 1128}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
index 582814339748..7de2ea8bd553 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
@@ -38,7 +38,7 @@
38 * @size: TTM accounting size for the surface. 38 * @size: TTM accounting size for the surface.
39 */ 39 */
40struct vmw_user_surface { 40struct vmw_user_surface {
41 struct ttm_base_object base; 41 struct ttm_prime_object prime;
42 struct vmw_surface srf; 42 struct vmw_surface srf;
43 uint32_t size; 43 uint32_t size;
44 uint32_t backup_handle; 44 uint32_t backup_handle;
@@ -580,7 +580,8 @@ static int vmw_surface_init(struct vmw_private *dev_priv,
580static struct vmw_resource * 580static struct vmw_resource *
581vmw_user_surface_base_to_res(struct ttm_base_object *base) 581vmw_user_surface_base_to_res(struct ttm_base_object *base)
582{ 582{
583 return &(container_of(base, struct vmw_user_surface, base)->srf.res); 583 return &(container_of(base, struct vmw_user_surface,
584 prime.base)->srf.res);
584} 585}
585 586
586/** 587/**
@@ -599,7 +600,7 @@ static void vmw_user_surface_free(struct vmw_resource *res)
599 kfree(srf->offsets); 600 kfree(srf->offsets);
600 kfree(srf->sizes); 601 kfree(srf->sizes);
601 kfree(srf->snooper.image); 602 kfree(srf->snooper.image);
602 ttm_base_object_kfree(user_srf, base); 603 ttm_prime_object_kfree(user_srf, prime);
603 ttm_mem_global_free(vmw_mem_glob(dev_priv), size); 604 ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
604} 605}
605 606
@@ -616,7 +617,7 @@ static void vmw_user_surface_base_release(struct ttm_base_object **p_base)
616{ 617{
617 struct ttm_base_object *base = *p_base; 618 struct ttm_base_object *base = *p_base;
618 struct vmw_user_surface *user_srf = 619 struct vmw_user_surface *user_srf =
619 container_of(base, struct vmw_user_surface, base); 620 container_of(base, struct vmw_user_surface, prime.base);
620 struct vmw_resource *res = &user_srf->srf.res; 621 struct vmw_resource *res = &user_srf->srf.res;
621 622
622 *p_base = NULL; 623 *p_base = NULL;
@@ -790,8 +791,8 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
790 } 791 }
791 srf->snooper.crtc = NULL; 792 srf->snooper.crtc = NULL;
792 793
793 user_srf->base.shareable = false; 794 user_srf->prime.base.shareable = false;
794 user_srf->base.tfile = NULL; 795 user_srf->prime.base.tfile = NULL;
795 796
796 /** 797 /**
797 * From this point, the generic resource management functions 798 * From this point, the generic resource management functions
@@ -803,9 +804,9 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
803 goto out_unlock; 804 goto out_unlock;
804 805
805 tmp = vmw_resource_reference(&srf->res); 806 tmp = vmw_resource_reference(&srf->res);
806 ret = ttm_base_object_init(tfile, &user_srf->base, 807 ret = ttm_prime_object_init(tfile, res->backup_size, &user_srf->prime,
807 req->shareable, VMW_RES_SURFACE, 808 req->shareable, VMW_RES_SURFACE,
808 &vmw_user_surface_base_release, NULL); 809 &vmw_user_surface_base_release, NULL);
809 810
810 if (unlikely(ret != 0)) { 811 if (unlikely(ret != 0)) {
811 vmw_resource_unreference(&tmp); 812 vmw_resource_unreference(&tmp);
@@ -813,7 +814,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
813 goto out_unlock; 814 goto out_unlock;
814 } 815 }
815 816
816 rep->sid = user_srf->base.hash.key; 817 rep->sid = user_srf->prime.base.hash.key;
817 vmw_resource_unreference(&res); 818 vmw_resource_unreference(&res);
818 819
819 ttm_read_unlock(&vmaster->lock); 820 ttm_read_unlock(&vmaster->lock);
@@ -823,7 +824,7 @@ out_no_copy:
823out_no_offsets: 824out_no_offsets:
824 kfree(srf->sizes); 825 kfree(srf->sizes);
825out_no_sizes: 826out_no_sizes:
826 ttm_base_object_kfree(user_srf, base); 827 ttm_prime_object_kfree(user_srf, prime);
827out_no_user_srf: 828out_no_user_srf:
828 ttm_mem_global_free(vmw_mem_glob(dev_priv), size); 829 ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
829out_unlock: 830out_unlock:
@@ -859,13 +860,14 @@ int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
859 return -EINVAL; 860 return -EINVAL;
860 } 861 }
861 862
862 if (unlikely(base->object_type != VMW_RES_SURFACE)) 863 if (unlikely(ttm_base_object_type(base) != VMW_RES_SURFACE))
863 goto out_bad_resource; 864 goto out_bad_resource;
864 865
865 user_srf = container_of(base, struct vmw_user_surface, base); 866 user_srf = container_of(base, struct vmw_user_surface, prime.base);
866 srf = &user_srf->srf; 867 srf = &user_srf->srf;
867 868
868 ret = ttm_ref_object_add(tfile, &user_srf->base, TTM_REF_USAGE, NULL); 869 ret = ttm_ref_object_add(tfile, &user_srf->prime.base,
870 TTM_REF_USAGE, NULL);
869 if (unlikely(ret != 0)) { 871 if (unlikely(ret != 0)) {
870 DRM_ERROR("Could not add a reference to a surface.\n"); 872 DRM_ERROR("Could not add a reference to a surface.\n");
871 goto out_no_reference; 873 goto out_no_reference;
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 751eaffbf0d5..ee127ec33c60 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -169,6 +169,7 @@ struct ttm_tt;
169 * @offset: The current GPU offset, which can have different meanings 169 * @offset: The current GPU offset, which can have different meanings
170 * depending on the memory type. For SYSTEM type memory, it should be 0. 170 * depending on the memory type. For SYSTEM type memory, it should be 0.
171 * @cur_placement: Hint of current placement. 171 * @cur_placement: Hint of current placement.
172 * @wu_mutex: Wait unreserved mutex.
172 * 173 *
173 * Base class for TTM buffer object, that deals with data placement and CPU 174 * Base class for TTM buffer object, that deals with data placement and CPU
174 * mappings. GPU mappings are really up to the driver, but for simpler GPUs 175 * mappings. GPU mappings are really up to the driver, but for simpler GPUs
@@ -250,6 +251,7 @@ struct ttm_buffer_object {
250 251
251 struct reservation_object *resv; 252 struct reservation_object *resv;
252 struct reservation_object ttm_resv; 253 struct reservation_object ttm_resv;
254 struct mutex wu_mutex;
253}; 255};
254 256
255/** 257/**
@@ -702,5 +704,5 @@ extern ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
702 size_t count, loff_t *f_pos, bool write); 704 size_t count, loff_t *f_pos, bool write);
703 705
704extern void ttm_bo_swapout_all(struct ttm_bo_device *bdev); 706extern void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
705 707extern int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
706#endif 708#endif
diff --git a/include/drm/ttm/ttm_execbuf_util.h b/include/drm/ttm/ttm_execbuf_util.h
index ec8a1d306510..16db7d01a336 100644
--- a/include/drm/ttm/ttm_execbuf_util.h
+++ b/include/drm/ttm/ttm_execbuf_util.h
@@ -70,7 +70,8 @@ extern void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket,
70/** 70/**
71 * function ttm_eu_reserve_buffers 71 * function ttm_eu_reserve_buffers
72 * 72 *
73 * @ticket: [out] ww_acquire_ctx returned by call. 73 * @ticket: [out] ww_acquire_ctx filled in by call, or NULL if only
74 * non-blocking reserves should be tried.
74 * @list: thread private list of ttm_validate_buffer structs. 75 * @list: thread private list of ttm_validate_buffer structs.
75 * 76 *
76 * Tries to reserve bos pointed to by the list entries for validation. 77 * Tries to reserve bos pointed to by the list entries for validation.
diff --git a/include/drm/ttm/ttm_object.h b/include/drm/ttm/ttm_object.h
index fc0cf0649901..58b029894eb3 100644
--- a/include/drm/ttm/ttm_object.h
+++ b/include/drm/ttm/ttm_object.h
@@ -41,6 +41,7 @@
41#include <drm/drm_hashtab.h> 41#include <drm/drm_hashtab.h>
42#include <linux/kref.h> 42#include <linux/kref.h>
43#include <linux/rcupdate.h> 43#include <linux/rcupdate.h>
44#include <linux/dma-buf.h>
44#include <ttm/ttm_memory.h> 45#include <ttm/ttm_memory.h>
45 46
46/** 47/**
@@ -77,6 +78,7 @@ enum ttm_object_type {
77 ttm_fence_type, 78 ttm_fence_type,
78 ttm_buffer_type, 79 ttm_buffer_type,
79 ttm_lock_type, 80 ttm_lock_type,
81 ttm_prime_type,
80 ttm_driver_type0 = 256, 82 ttm_driver_type0 = 256,
81 ttm_driver_type1, 83 ttm_driver_type1,
82 ttm_driver_type2, 84 ttm_driver_type2,
@@ -132,6 +134,30 @@ struct ttm_base_object {
132 enum ttm_ref_type ref_type); 134 enum ttm_ref_type ref_type);
133}; 135};
134 136
137
138/**
139 * struct ttm_prime_object - Modified base object that is prime-aware
140 *
141 * @base: struct ttm_base_object that we derive from
142 * @mutex: Mutex protecting the @dma_buf member.
143 * @size: Size of the dma_buf associated with this object
144 * @real_type: Type of the underlying object. Needed since we're setting
145 * the value of @base::object_type to ttm_prime_type
146 * @dma_buf: Non ref-coutned pointer to a struct dma_buf created from this
147 * object.
148 * @refcount_release: The underlying object's release method. Needed since
149 * we set @base::refcount_release to our own release method.
150 */
151
152struct ttm_prime_object {
153 struct ttm_base_object base;
154 struct mutex mutex;
155 size_t size;
156 enum ttm_object_type real_type;
157 struct dma_buf *dma_buf;
158 void (*refcount_release) (struct ttm_base_object **);
159};
160
135/** 161/**
136 * ttm_base_object_init 162 * ttm_base_object_init
137 * 163 *
@@ -248,14 +274,18 @@ extern void ttm_object_file_release(struct ttm_object_file **p_tfile);
248/** 274/**
249 * ttm_object device init - initialize a struct ttm_object_device 275 * ttm_object device init - initialize a struct ttm_object_device
250 * 276 *
277 * @mem_glob: struct ttm_mem_global for memory accounting.
251 * @hash_order: Order of hash table used to hash the base objects. 278 * @hash_order: Order of hash table used to hash the base objects.
279 * @ops: DMA buf ops for prime objects of this device.
252 * 280 *
253 * This function is typically called on device initialization to prepare 281 * This function is typically called on device initialization to prepare
254 * data structures needed for ttm base and ref objects. 282 * data structures needed for ttm base and ref objects.
255 */ 283 */
256 284
257extern struct ttm_object_device *ttm_object_device_init 285extern struct ttm_object_device *
258 (struct ttm_mem_global *mem_glob, unsigned int hash_order); 286ttm_object_device_init(struct ttm_mem_global *mem_glob,
287 unsigned int hash_order,
288 const struct dma_buf_ops *ops);
259 289
260/** 290/**
261 * ttm_object_device_release - release data held by a ttm_object_device 291 * ttm_object_device_release - release data held by a ttm_object_device
@@ -272,4 +302,31 @@ extern void ttm_object_device_release(struct ttm_object_device **p_tdev);
272 302
273#define ttm_base_object_kfree(__object, __base)\ 303#define ttm_base_object_kfree(__object, __base)\
274 kfree_rcu(__object, __base.rhead) 304 kfree_rcu(__object, __base.rhead)
305
306extern int ttm_prime_object_init(struct ttm_object_file *tfile,
307 size_t size,
308 struct ttm_prime_object *prime,
309 bool shareable,
310 enum ttm_object_type type,
311 void (*refcount_release)
312 (struct ttm_base_object **),
313 void (*ref_obj_release)
314 (struct ttm_base_object *,
315 enum ttm_ref_type ref_type));
316
317static inline enum ttm_object_type
318ttm_base_object_type(struct ttm_base_object *base)
319{
320 return (base->object_type == ttm_prime_type) ?
321 container_of(base, struct ttm_prime_object, base)->real_type :
322 base->object_type;
323}
324extern int ttm_prime_fd_to_handle(struct ttm_object_file *tfile,
325 int fd, u32 *handle);
326extern int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
327 uint32_t handle, uint32_t flags,
328 int *prime_fd);
329
330#define ttm_prime_object_kfree(__obj, __prime) \
331 kfree_rcu(__obj, __prime.base.rhead)
275#endif 332#endif
diff --git a/include/uapi/drm/radeon_drm.h b/include/uapi/drm/radeon_drm.h
index 46d41e8b0dcc..2f3f7ea8c77b 100644
--- a/include/uapi/drm/radeon_drm.h
+++ b/include/uapi/drm/radeon_drm.h
@@ -981,6 +981,8 @@ struct drm_radeon_cs {
981#define RADEON_INFO_SI_TILE_MODE_ARRAY 0x16 981#define RADEON_INFO_SI_TILE_MODE_ARRAY 0x16
982/* query if CP DMA is supported on the compute ring */ 982/* query if CP DMA is supported on the compute ring */
983#define RADEON_INFO_SI_CP_DMA_COMPUTE 0x17 983#define RADEON_INFO_SI_CP_DMA_COMPUTE 0x17
984/* CIK macrotile mode array */
985#define RADEON_INFO_CIK_MACROTILE_MODE_ARRAY 0x18
984 986
985 987
986struct drm_radeon_info { 988struct drm_radeon_info {
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-27 01:04:22 -0500