Merge tag 'drm-next-2021-08-31-1' of git://anongit.freedesktop.org/drm/drm

Pull drm updates from Dave Airlie:
 "Highlights:

   - i915 has seen a lot of refactoring and uAPI cleanups due to a
     change in the upstream direction going forward

     This has all been audited with known userspace, but there may be
     some pitfalls that were missed.

   - i915 now uses common TTM to enable discrete memory on DG1/2 GPUs

   - i915 enables Jasper and Elkhart Lake by default and has preliminary
     XeHP/DG2 support

   - amdgpu adds support for Cyan Skillfish

   - lots of implicit fencing rules documented and fixed up in drivers

   - msm now uses the core scheduler

   - the irq midlayer has been removed for non-legacy drivers

   - the sysfb code now works on more than x86.

  Otherwise the usual smattering of stuff everywhere, panels, bridges,
  refactorings.

  Detailed summary:

  core:
   - extract i915 eDP backlight into core
   - DP aux bus support
   - drm_device.irq_enabled removed
   - port drivers to native irq interfaces
   - export gem shadow plane handling for vgem
   - print proper driver name in framebuffer registration
   - driver fixes for implicit fencing rules
   - ARM fixed rate compression modifier added
   - updated fb damage handling
   - rmfb ioctl logging/docs
   - drop drm_gem_object_put_locked
   - define DRM_FORMAT_MAX_PLANES
   - add gem fb vmap/vunmap helpers
   - add lockdep_assert(once) helpers
   - mark drm irq midlayer as legacy
   - use offset adjusted bo mapping conversion

  vgaarb:
   - cleanups

  fbdev:
   - extend efifb handling to all arches
   - div by 0 fixes for multiple drivers

  udmabuf:
   - add hugepage mapping support

  dma-buf:
   - non-dynamic exporter fixups
   - document implicit fencing rules

  amdgpu:
   - Initial Cyan Skillfish support
   - switch virtual DCE over to vkms based atomic
   - VCN/JPEG power down fixes
   - NAVI PCIE link handling fixes
   - AMD HDMI freesync fixes
   - Yellow Carp + Beige Goby fixes
   - Clockgating/S0ix/SMU/EEPROM fixes
   - embed hw fence in job
   - rework dma-resv handling
   - ensure eviction to system ram

  amdkfd:
   - uapi: SVM address range query added
   - sysfs leak fix
   - GPUVM TLB optimizations
   - vmfault/migration counters

  i915:
   - Enable JSL and EHL by default
   - preliminary XeHP/DG2 support
   - remove all CNL support (never shipped)
   - move to TTM for discrete memory support
   - allow mixed object mmap handling
   - GEM uAPI spring cleaning
       - add I915_MMAP_OBJECT_FIXED
       - reinstate ADL-P mmap ioctls
       - drop a bunch of unused by userspace features
       - disable and remove GPU relocations
   - revert some i915 misfeatures
   - major refactoring of GuC for Gen11+
   - execbuffer object locking separate step
   - reject caching/set-domain on discrete
   - Enable pipe DMC loading on XE-LPD and ADL-P
   - add PSF GV point support
   - Refactor and fix DDI buffer translations
   - Clean up FBC CFB allocation code
   - Finish INTEL_GEN() and friends macro conversions

  nouveau:
   - add eDP backlight support
   - implicit fence fix

  msm:
   - a680/7c3 support
   - drm/scheduler conversion

  panfrost:
   - rework GPU reset

  virtio:
   - fix fencing for planes

  ast:
   - add detect support

  bochs:
   - move to tiny GPU driver

  vc4:
   - use hotplug irqs
   - HDMI codec support

  vmwgfx:
   - use internal vmware device headers

  ingenic:
   - demidlayering irq

  rcar-du:
   - shutdown fixes
   - convert to bridge connector helpers

  zynqmp-dsub:
   - misc fixes

  mgag200:
   - convert PLL handling to atomic

  mediatek:
   - MT8133 AAL support
   - gem mmap object support
   - MT8167 support

  etnaviv:
   - NXP Layerscape LS1028A SoC support
   - GEM mmap cleanups

  tegra:
   - new user API

  exynos:
   - missing unlock fix
   - build warning fix
   - use refcount_t"

* tag 'drm-next-2021-08-31-1' of git://anongit.freedesktop.org/drm/drm: (1318 commits)
  drm/amd/display: Move AllowDRAMSelfRefreshOrDRAMClockChangeInVblank to bounding box
  drm/amd/display: Remove duplicate dml init
  drm/amd/display: Update bounding box states (v2)
  drm/amd/display: Update number of DCN3 clock states
  drm/amdgpu: disable GFX CGCG in aldebaran
  drm/amdgpu: Clear RAS interrupt status on aldebaran
  drm/amdgpu: Add support for RAS XGMI err query
  drm/amdkfd: Account for SH/SE count when setting up cu masks.
  drm/amdgpu: rename amdgpu_bo_get_preferred_pin_domain
  drm/amdgpu: drop redundant cancel_delayed_work_sync call
  drm/amdgpu: add missing cleanups for more ASICs on UVD/VCE suspend
  drm/amdgpu: add missing cleanups for Polaris12 UVD/VCE on suspend
  drm/amdkfd: map SVM range with correct access permission
  drm/amdkfd: check access permisson to restore retry fault
  drm/amdgpu: Update RAS XGMI Error Query
  drm/amdgpu: Add driver infrastructure for MCA RAS
  drm/amd/display: Add Logging for HDMI color depth information
  drm/amd/amdgpu: consolidate PSP TA init shared buf functions
  drm/amd/amdgpu: add name field back to ras_common_if
  drm/amdgpu: Fix build with missing pm_suspend_target_state module export
  ...

1 files changed, 80 insertions, 64 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 72d9b92b1754..14499f0de32d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -129,30 +129,50 @@ static u32 amdgpu_fence_read(struct amdgpu_ring *ring)
  *
  * @ring: ring the fence is associated with
  * @f: resulting fence object
+ * @job: job the fence is embedded in
  * @flags: flags to pass into the subordinate .emit_fence() call
  *
  * Emits a fence command on the requested ring (all asics).
  * Returns 0 on success, -ENOMEM on failure.
  */
-int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f,
+int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amdgpu_job *job,
 		      unsigned flags)
 {
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_fence *fence;
+	struct dma_fence *fence;
+	struct amdgpu_fence *am_fence;
 	struct dma_fence __rcu **ptr;
 	uint32_t seq;
 	int r;
 
-	fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL);
-	if (fence == NULL)
-		return -ENOMEM;
+	if (job == NULL) {
+		/* create a sperate hw fence */
+		am_fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_ATOMIC);
+		if (am_fence == NULL)
+			return -ENOMEM;
+		fence = &am_fence->base;
+		am_fence->ring = ring;
+	} else {
+		/* take use of job-embedded fence */
+		fence = &job->hw_fence;
+	}
 
 	seq = ++ring->fence_drv.sync_seq;
-	fence->ring = ring;
-	dma_fence_init(&fence->base, &amdgpu_fence_ops,
-		       &ring->fence_drv.lock,
-		       adev->fence_context + ring->idx,
-		       seq);
+	if (job != NULL && job->job_run_counter) {
+		/* reinit seq for resubmitted jobs */
+		fence->seqno = seq;
+	} else {
+		dma_fence_init(fence, &amdgpu_fence_ops,
+				&ring->fence_drv.lock,
+				adev->fence_context + ring->idx,
+				seq);
+	}
+
+	if (job != NULL) {
+		/* mark this fence has a parent job */
+		set_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &fence->flags);
+	}
+
 	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
 			       seq, flags | AMDGPU_FENCE_FLAG_INT);
 	pm_runtime_get_noresume(adev_to_drm(adev)->dev);
@@ -175,9 +195,9 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f,
 	/* This function can't be called concurrently anyway, otherwise
 	 * emitting the fence would mess up the hardware ring buffer.
 	 */
-	rcu_assign_pointer(*ptr, dma_fence_get(&fence->base));
+	rcu_assign_pointer(*ptr, dma_fence_get(fence));
 
-	*f = &fence->base;
+	*f = fence;
 
 	return 0;
 }
@@ -417,9 +437,6 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 	}
 	amdgpu_fence_write(ring, atomic_read(&ring->fence_drv.last_seq));
 
-	if (irq_src)
-		amdgpu_irq_get(adev, irq_src, irq_type);
-
 	ring->fence_drv.irq_src = irq_src;
 	ring->fence_drv.irq_type = irq_type;
 	ring->fence_drv.initialized = true;
@@ -490,7 +507,7 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 
 	r = drm_sched_init(&ring->sched, &amdgpu_sched_ops,
 			   num_hw_submission, amdgpu_job_hang_limit,
-			   timeout, sched_score, ring->name);
+			   timeout, NULL, sched_score, ring->name);
 	if (r) {
 		DRM_ERROR("Failed to create scheduler on ring %s.\n",
 			  ring->name);
@@ -501,7 +518,7 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 }
 
 /**
- * amdgpu_fence_driver_init - init the fence driver
+ * amdgpu_fence_driver_sw_init - init the fence driver
  * for all possible rings.
  *
  * @adev: amdgpu device pointer
@@ -512,20 +529,20 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
  * amdgpu_fence_driver_start_ring().
  * Returns 0 for success.
  */
-int amdgpu_fence_driver_init(struct amdgpu_device *adev)
+int amdgpu_fence_driver_sw_init(struct amdgpu_device *adev)
 {
 	return 0;
 }
 
 /**
- * amdgpu_fence_driver_fini - tear down the fence driver
+ * amdgpu_fence_driver_hw_fini - tear down the fence driver
  * for all possible rings.
  *
  * @adev: amdgpu device pointer
  *
  * Tear down the fence driver for all possible rings (all asics).
  */
-void amdgpu_fence_driver_fini_hw(struct amdgpu_device *adev)
+void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
 {
 	int i, r;
 
@@ -534,8 +551,7 @@ void amdgpu_fence_driver_fini_hw(struct amdgpu_device *adev)
 
 		if (!ring || !ring->fence_drv.initialized)
 			continue;
-		if (!ring->no_scheduler)
-			drm_sched_fini(&ring->sched);
+
 		/* You can't wait for HW to signal if it's gone */
 		if (!drm_dev_is_unplugged(&adev->ddev))
 			r = amdgpu_fence_wait_empty(ring);
@@ -553,7 +569,7 @@ void amdgpu_fence_driver_fini_hw(struct amdgpu_device *adev)
 	}
 }
 
-void amdgpu_fence_driver_fini_sw(struct amdgpu_device *adev)
+void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev)
 {
 	unsigned int i, j;
 
@@ -563,6 +579,9 @@ void amdgpu_fence_driver_fini_sw(struct amdgpu_device *adev)
 		if (!ring || !ring->fence_drv.initialized)
 			continue;
 
+		if (!ring->no_scheduler)
+			drm_sched_fini(&ring->sched);
+
 		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
 			dma_fence_put(ring->fence_drv.fences[j]);
 		kfree(ring->fence_drv.fences);
@@ -572,49 +591,18 @@ void amdgpu_fence_driver_fini_sw(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_fence_driver_suspend - suspend the fence driver
- * for all possible rings.
- *
- * @adev: amdgpu device pointer
- *
- * Suspend the fence driver for all possible rings (all asics).
- */
-void amdgpu_fence_driver_suspend(struct amdgpu_device *adev)
-{
-	int i, r;
-
-	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		if (!ring || !ring->fence_drv.initialized)
-			continue;
-
-		/* wait for gpu to finish processing current batch */
-		r = amdgpu_fence_wait_empty(ring);
-		if (r) {
-			/* delay GPU reset to resume */
-			amdgpu_fence_driver_force_completion(ring);
-		}
-
-		/* disable the interrupt */
-		if (ring->fence_drv.irq_src)
-			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
-				       ring->fence_drv.irq_type);
-	}
-}
-
-/**
- * amdgpu_fence_driver_resume - resume the fence driver
+ * amdgpu_fence_driver_hw_init - enable the fence driver
  * for all possible rings.
  *
  * @adev: amdgpu device pointer
  *
- * Resume the fence driver for all possible rings (all asics).
+ * Enable the fence driver for all possible rings (all asics).
  * Not all asics have all rings, so each asic will only
  * start the fence driver on the rings it has using
  * amdgpu_fence_driver_start_ring().
  * Returns 0 for success.
  */
-void amdgpu_fence_driver_resume(struct amdgpu_device *adev)
+void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev)
 {
 	int i;
 
@@ -653,8 +641,16 @@ static const char *amdgpu_fence_get_driver_name(struct dma_fence *fence)
 
 static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
 {
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	return (const char *)fence->ring->name;
+	struct amdgpu_ring *ring;
+
+	if (test_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &f->flags)) {
+		struct amdgpu_job *job = container_of(f, struct amdgpu_job, hw_fence);
+
+		ring = to_amdgpu_ring(job->base.sched);
+	} else {
+		ring = to_amdgpu_fence(f)->ring;
+	}
+	return (const char *)ring->name;
 }
 
 /**
@@ -667,13 +663,20 @@ static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
  */
 static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
 {
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	struct amdgpu_ring *ring = fence->ring;
+	struct amdgpu_ring *ring;
+
+	if (test_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &f->flags)) {
+		struct amdgpu_job *job = container_of(f, struct amdgpu_job, hw_fence);
+
+		ring = to_amdgpu_ring(job->base.sched);
+	} else {
+		ring = to_amdgpu_fence(f)->ring;
+	}
 
 	if (!timer_pending(&ring->fence_drv.fallback_timer))
 		amdgpu_fence_schedule_fallback(ring);
 
-	DMA_FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx);
+	DMA_FENCE_TRACE(f, "armed on ring %i!\n", ring->idx);
 
 	return true;
 }
@@ -688,8 +691,20 @@ static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
 static void amdgpu_fence_free(struct rcu_head *rcu)
 {
 	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	kmem_cache_free(amdgpu_fence_slab, fence);
+
+	if (test_bit(AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT, &f->flags)) {
+	/* free job if fence has a parent job */
+		struct amdgpu_job *job;
+
+		job = container_of(f, struct amdgpu_job, hw_fence);
+		kfree(job);
+	} else {
+	/* free fence_slab if it's separated fence*/
+		struct amdgpu_fence *fence;
+
+		fence = to_amdgpu_fence(f);
+		kmem_cache_free(amdgpu_fence_slab, fence);
+	}
 }
 
 /**
@@ -712,6 +727,7 @@ static const struct dma_fence_ops amdgpu_fence_ops = {
 	.release = amdgpu_fence_release,
 };
 
+
 /*
  * Fence debugfs
  */