diff mbox series

[RFC,5/6] drm/xe: Use DMA_RESV_USAGE_PREEMPT for preempt fences

Message ID 20241109172942.482630-6-matthew.brost@intel.com (mailing list archive)
State New, archived
Headers show
Series Common preempt fences and semantics | expand

Commit Message

Matthew Brost Nov. 9, 2024, 5:29 p.m. UTC
Use the new DMA_RESV_USAGE_PREEMPT dma-resv slots in Xe for preemptive
fences, and call them in dma-resv/scheduler rather than open-coding the
enabling of signaling before waiting.

Cc: Dave Airlie <airlied@redhat.com>
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Christian Koenig <christian.koenig@amd.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/xe/xe_bo.c      | 22 +++++-----------------
 drivers/gpu/drm/xe/xe_migrate.c |  4 ++--
 drivers/gpu/drm/xe/xe_pt.c      | 12 ++----------
 drivers/gpu/drm/xe/xe_vm.c      | 22 ++++------------------
 4 files changed, 13 insertions(+), 47 deletions(-)
diff mbox series

Patch

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 549866da5cd1..7ab7d27b66c6 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -484,23 +484,11 @@  static int xe_ttm_io_mem_reserve(struct ttm_device *bdev,
 static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
 				const struct ttm_operation_ctx *ctx)
 {
-	struct dma_resv_iter cursor;
-	struct dma_fence *fence;
 	struct drm_gem_object *obj = &bo->ttm.base;
 	struct drm_gpuvm_bo *vm_bo;
 	bool idle = false;
 	int ret = 0;
 
-	dma_resv_assert_held(bo->ttm.base.resv);
-
-	if (!list_empty(&bo->ttm.base.gpuva.list)) {
-		dma_resv_iter_begin(&cursor, bo->ttm.base.resv,
-				    DMA_RESV_USAGE_BOOKKEEP);
-		dma_resv_for_each_fence_unlocked(&cursor, fence)
-			dma_fence_enable_sw_signaling(fence);
-		dma_resv_iter_end(&cursor);
-	}
-
 	drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
 		struct xe_vm *vm = gpuvm_to_vm(vm_bo->vm);
 		struct drm_gpuva *gpuva;
@@ -515,11 +503,11 @@  static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
 
 			if (ctx->no_wait_gpu &&
 			    !dma_resv_test_signaled(bo->ttm.base.resv,
-						    DMA_RESV_USAGE_BOOKKEEP))
+						    DMA_RESV_USAGE_PREEMPT))
 				return -EBUSY;
 
 			timeout = dma_resv_wait_timeout(bo->ttm.base.resv,
-							DMA_RESV_USAGE_BOOKKEEP,
+							DMA_RESV_USAGE_PREEMPT,
 							ctx->interruptible,
 							MAX_SCHEDULE_TIMEOUT);
 			if (!timeout)
@@ -723,7 +711,7 @@  static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
 	if (old_mem_type == XE_PL_TT &&
 	    new_mem->mem_type == XE_PL_SYSTEM) {
 		long timeout = dma_resv_wait_timeout(ttm_bo->base.resv,
-						     DMA_RESV_USAGE_BOOKKEEP,
+						     DMA_RESV_USAGE_PREEMPT,
 						     true,
 						     MAX_SCHEDULE_TIMEOUT);
 		if (timeout < 0) {
@@ -1056,7 +1044,7 @@  static void xe_ttm_bo_release_notify(struct ttm_buffer_object *ttm_bo)
 	 * unbind.
 	 */
 	dma_resv_for_each_fence(&cursor, ttm_bo->base.resv,
-				DMA_RESV_USAGE_BOOKKEEP, fence) {
+				DMA_RESV_USAGE_PREEMPT, fence) {
 		if (xe_fence_is_xe_preempt(fence) &&
 		    !dma_fence_is_signaled(fence)) {
 			if (!replacement)
@@ -1065,7 +1053,7 @@  static void xe_ttm_bo_release_notify(struct ttm_buffer_object *ttm_bo)
 			dma_resv_replace_fences(ttm_bo->base.resv,
 						fence->context,
 						replacement,
-						DMA_RESV_USAGE_BOOKKEEP);
+						DMA_RESV_USAGE_PREEMPT);
 		}
 	}
 	dma_fence_put(replacement);
diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
index cfd31ae49cc1..75067c584581 100644
--- a/drivers/gpu/drm/xe/xe_migrate.c
+++ b/drivers/gpu/drm/xe/xe_migrate.c
@@ -895,10 +895,10 @@  struct dma_fence *xe_migrate_copy(struct xe_migrate *m,
 		xe_sched_job_add_migrate_flush(job, flush_flags);
 		if (!fence) {
 			err = xe_sched_job_add_deps(job, src_bo->ttm.base.resv,
-						    DMA_RESV_USAGE_BOOKKEEP);
+						    DMA_RESV_USAGE_PREEMPT);
 			if (!err && src_bo != dst_bo)
 				err = xe_sched_job_add_deps(job, dst_bo->ttm.base.resv,
-							    DMA_RESV_USAGE_BOOKKEEP);
+							    DMA_RESV_USAGE_PREEMPT);
 			if (err)
 				goto err_job;
 		}
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index f27f579f4d85..00358e748d0d 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -1115,7 +1115,7 @@  static int xe_pt_vm_dependencies(struct xe_sched_job *job,
 	if (pt_update_ops->wait_vm_bookkeep || pt_update_ops->wait_vm_kernel) {
 		err = job_test_add_deps(job, xe_vm_resv(vm),
 					pt_update_ops->wait_vm_bookkeep ?
-					DMA_RESV_USAGE_BOOKKEEP :
+					DMA_RESV_USAGE_PREEMPT :
 					DMA_RESV_USAGE_KERNEL);
 		if (err)
 			return err;
@@ -1231,18 +1231,10 @@  static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,
 		spin_unlock(&vm->userptr.invalidated_lock);
 
 		if (xe_vm_in_preempt_fence_mode(vm)) {
-			struct dma_resv_iter cursor;
-			struct dma_fence *fence;
 			long err;
 
-			dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
-					    DMA_RESV_USAGE_BOOKKEEP);
-			dma_resv_for_each_fence_unlocked(&cursor, fence)
-				dma_fence_enable_sw_signaling(fence);
-			dma_resv_iter_end(&cursor);
-
 			err = dma_resv_wait_timeout(xe_vm_resv(vm),
-						    DMA_RESV_USAGE_BOOKKEEP,
+						    DMA_RESV_USAGE_PREEMPT,
 						    false, MAX_SCHEDULE_TIMEOUT);
 			XE_WARN_ON(err <= 0);
 		}
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 624133fae5f5..568395530c49 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -195,7 +195,7 @@  static int add_preempt_fences(struct xe_vm *vm, struct xe_bo *bo)
 		if (q->lr.pfence) {
 			dma_resv_add_fence(bo->ttm.base.resv,
 					   q->lr.pfence,
-					   DMA_RESV_USAGE_BOOKKEEP);
+					   DMA_RESV_USAGE_PREEMPT);
 		}
 
 	return 0;
@@ -213,7 +213,7 @@  static void resume_and_reinstall_preempt_fences(struct xe_vm *vm,
 		q->ops->resume(q);
 
 		drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, q->lr.pfence,
-					 DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP);
+					 DMA_RESV_USAGE_PREEMPT, DMA_RESV_USAGE_PREEMPT);
 	}
 }
 
@@ -250,7 +250,7 @@  int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
 	down_read(&vm->userptr.notifier_lock);
 
 	drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, pfence,
-				 DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP);
+				 DMA_RESV_USAGE_PREEMPT, DMA_RESV_USAGE_PREEMPT);
 
 	/*
 	 * Check to see if a preemption on VM is in flight or userptr
@@ -588,8 +588,6 @@  static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
 	struct xe_userptr_vma *uvma = container_of(userptr, typeof(*uvma), userptr);
 	struct xe_vma *vma = &uvma->vma;
 	struct xe_vm *vm = xe_vma_vm(vma);
-	struct dma_resv_iter cursor;
-	struct dma_fence *fence;
 	long err;
 
 	xe_assert(vm->xe, xe_vma_is_userptr(vma));
@@ -625,20 +623,8 @@  static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
 
 	up_write(&vm->userptr.notifier_lock);
 
-	/*
-	 * Preempt fences turn into schedule disables, pipeline these.
-	 * Note that even in fault mode, we need to wait for binds and
-	 * unbinds to complete, and those are attached as BOOKMARK fences
-	 * to the vm.
-	 */
-	dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
-			    DMA_RESV_USAGE_BOOKKEEP);
-	dma_resv_for_each_fence_unlocked(&cursor, fence)
-		dma_fence_enable_sw_signaling(fence);
-	dma_resv_iter_end(&cursor);
-
 	err = dma_resv_wait_timeout(xe_vm_resv(vm),
-				    DMA_RESV_USAGE_BOOKKEEP,
+				    DMA_RESV_USAGE_PREEMPT,
 				    false, MAX_SCHEDULE_TIMEOUT);
 	XE_WARN_ON(err <= 0);