[7/8] drm/amdgpu: track evicted page tables

Submitted by Christian König on Aug. 10, 2017, 2:50 p.m.

Details

Message ID 1502376647-4519-7-git-send-email-deathsimple@vodafone.de
State New
Headers show
Series "Series without cover letter" ( rev: 1 ) in AMD X.Org drivers

Not browsing as part of any series.

Commit Message

Christian König Aug. 10, 2017, 2:50 p.m.
From: Christian König <christian.koenig@amd.com>

Instead of validating all page tables when one was evicted,
track which one needs a validation.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c     |   7 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c     | 217 ++++++++++++++---------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h     |  14 +-
 4 files changed, 116 insertions(+), 124 deletions(-)

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 3f46b5a..f68ac56 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -632,9 +632,6 @@  static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 
 	amdgpu_cs_report_moved_bytes(p->adev, p->bytes_moved,
 				     p->bytes_moved_vis);
-	fpriv->vm.last_eviction_counter =
-		atomic64_read(&p->adev->num_evictions);
-
 	if (p->bo_list) {
 		struct amdgpu_bo *gds = p->bo_list->gds_obj;
 		struct amdgpu_bo *gws = p->bo_list->gws_obj;
@@ -826,7 +823,7 @@  static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p)
 			if (!bo)
 				continue;
 
-			amdgpu_vm_bo_invalidate(adev, bo);
+			amdgpu_vm_bo_invalidate(adev, bo, false);
 		}
 	}
 
@@ -851,7 +848,7 @@  static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
 	}
 
 	if (p->job->vm) {
-		p->job->vm_pd_addr = amdgpu_bo_gpu_offset(vm->root.bo);
+		p->job->vm_pd_addr = amdgpu_bo_gpu_offset(vm->root.base.bo);
 
 		r = amdgpu_bo_vm_update_pte(p);
 		if (r)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index e7e8991..9a41488 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -929,7 +929,7 @@  void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 		return;
 
 	abo = container_of(bo, struct amdgpu_bo, tbo);
-	amdgpu_vm_bo_invalidate(adev, abo);
+	amdgpu_vm_bo_invalidate(adev, abo, evict);
 
 	amdgpu_bo_kunmap(abo);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index b183965..8e2ce91 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -140,7 +140,7 @@  void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 			 struct list_head *validated,
 			 struct amdgpu_bo_list_entry *entry)
 {
-	entry->robj = vm->root.bo;
+	entry->robj = vm->root.base.bo;
 	entry->priority = 0;
 	entry->tv.bo = &entry->robj->tbo;
 	entry->tv.shared = true;
@@ -149,94 +149,66 @@  void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 }
 
 /**
- * amdgpu_vm_validate_layer - validate a single page table level
+ * amdgpu_vm_validate_pt_bos - validate the page table BOs
  *
- * @parent: parent page table level
+ * @adev: amdgpu device pointer
+ * @vm: vm providing the BOs
  * @validate: callback to do the validation
  * @param: parameter for the validation callback
  *
  * Validate the page table BOs on command submission if neccessary.
  */
-static int amdgpu_vm_validate_level(struct amdgpu_vm_pt *parent,
-				    int (*validate)(void *, struct amdgpu_bo *),
-				    void *param, bool use_cpu_for_update,
-				    struct ttm_bo_global *glob)
+int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			      int (*validate)(void *p, struct amdgpu_bo *bo),
+			      void *param)
 {
-	unsigned i;
+	struct ttm_bo_global *glob = adev->mman.bdev.glob;
 	int r;
 
-	if (parent->bo->shadow) {
-		struct amdgpu_bo *shadow = parent->bo->shadow;
-
-		r = amdgpu_ttm_bind(&shadow->tbo, &shadow->tbo.mem);
-		if (r)
-			return r;
-	}
+	spin_lock(&vm->status_lock);
+	while (!list_empty(&vm->evicted)) {
+		struct amdgpu_vm_bo_base *bo_base;
+		struct amdgpu_bo *bo;
 
-	if (use_cpu_for_update) {
-		r = amdgpu_bo_kmap(parent->bo, NULL);
-		if (r)
-			return r;
-	}
+		bo_base = list_first_entry(&vm->evicted,
+					   struct amdgpu_vm_bo_base,
+					   vm_status);
+		spin_unlock(&vm->status_lock);
 
-	if (!parent->entries)
-		return 0;
+		bo = bo_base->bo;
+		BUG_ON(!bo);
+		if (bo->parent) {
+			r = validate(param, bo);
+			if (r)
+				return r;
 
-	for (i = 0; i <= parent->last_entry_used; ++i) {
-		struct amdgpu_vm_pt *entry = &parent->entries[i];
+			spin_lock(&glob->lru_lock);
+			ttm_bo_move_to_lru_tail(&bo->tbo);
+			if (bo->shadow)
+				ttm_bo_move_to_lru_tail(&bo->shadow->tbo);
+			spin_unlock(&glob->lru_lock);
+		}
 
-		if (!entry->bo)
-			continue;
+		if (bo->shadow) {
+			struct amdgpu_bo *shadow = bo->shadow;
 
-		r = validate(param, entry->bo);
-		if (r)
-			return r;
+			r = amdgpu_ttm_bind(&shadow->tbo, &shadow->tbo.mem);
+			if (r)
+				return r;
+		}
 
-		spin_lock(&glob->lru_lock);
-		ttm_bo_move_to_lru_tail(&entry->bo->tbo);
-		if (entry->bo->shadow)
-			ttm_bo_move_to_lru_tail(&entry->bo->shadow->tbo);
-		spin_unlock(&glob->lru_lock);
+		if (vm->use_cpu_for_update) {
+			r = amdgpu_bo_kmap(bo, NULL);
+			if (r)
+				return r;
+		}
 
-		/*
-		 * Recurse into the sub directory. This is harmless because we
-		 * have only a maximum of 5 layers.
-		 */
-		r = amdgpu_vm_validate_level(entry, validate, param,
-					     use_cpu_for_update, glob);
-		if (r)
-			return r;
+		spin_lock(&vm->status_lock);
+		list_del_init(&bo_base->vm_status);
 	}
+	spin_unlock(&vm->status_lock);
 
-	return r;
-}
-
-/**
- * amdgpu_vm_validate_pt_bos - validate the page table BOs
- *
- * @adev: amdgpu device pointer
- * @vm: vm providing the BOs
- * @validate: callback to do the validation
- * @param: parameter for the validation callback
- *
- * Validate the page table BOs on command submission if neccessary.
- */
-int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-			      int (*validate)(void *p, struct amdgpu_bo *bo),
-			      void *param)
-{
-	uint64_t num_evictions;
-
-	/* We only need to validate the page tables
-	 * if they aren't already valid.
-	 */
-	num_evictions = atomic64_read(&adev->num_evictions);
-	if (num_evictions == vm->last_eviction_counter)
-		return 0;
-
-	return amdgpu_vm_validate_level(&vm->root, validate, param,
-					vm->use_cpu_for_update,
-					adev->mman.bdev.glob);
+	return 0;
 }
 
 /**
@@ -301,11 +273,11 @@  static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
 
 	/* walk over the address space and allocate the page tables */
 	for (pt_idx = from; pt_idx <= to; ++pt_idx) {
-		struct reservation_object *resv = vm->root.bo->tbo.resv;
+		struct reservation_object *resv = vm->root.base.bo->tbo.resv;
 		struct amdgpu_vm_pt *entry = &parent->entries[pt_idx];
 		struct amdgpu_bo *pt;
 
-		if (!entry->bo) {
+		if (!entry->base.bo) {
 			r = amdgpu_bo_create(adev,
 					     amdgpu_vm_bo_size(adev, level),
 					     AMDGPU_GPU_PAGE_SIZE, true,
@@ -326,9 +298,12 @@  static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
 			/* Keep a reference to the root directory to avoid
 			* freeing them up in the wrong order.
 			*/
-			pt->parent = amdgpu_bo_ref(vm->root.bo);
+			pt->parent = amdgpu_bo_ref(vm->root.base.bo);
 
-			entry->bo = pt;
+			entry->base.vm = vm;
+			entry->base.bo = pt;
+			list_add_tail(&entry->base.bo_list, &pt->va);
+			INIT_LIST_HEAD(&entry->base.vm_status);
 			entry->addr = 0;
 		}
 
@@ -995,7 +970,7 @@  static int amdgpu_vm_wait_pd(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	int r;
 
 	amdgpu_sync_create(&sync);
-	amdgpu_sync_resv(adev, &sync, vm->root.bo->tbo.resv, owner);
+	amdgpu_sync_resv(adev, &sync, vm->root.base.bo->tbo.resv, owner);
 	r = amdgpu_sync_wait(&sync, true);
 	amdgpu_sync_free(&sync);
 
@@ -1034,10 +1009,10 @@  static int amdgpu_vm_update_level(struct amdgpu_device *adev,
 
 	memset(&params, 0, sizeof(params));
 	params.adev = adev;
-	shadow = parent->bo->shadow;
+	shadow = parent->base.bo->shadow;
 
 	if (vm->use_cpu_for_update) {
-		pd_addr = (unsigned long)amdgpu_bo_kptr(parent->bo);
+		pd_addr = (unsigned long)amdgpu_bo_kptr(parent->base.bo);
 		r = amdgpu_vm_wait_pd(adev, vm, AMDGPU_FENCE_OWNER_VM);
 		if (unlikely(r))
 			return r;
@@ -1053,7 +1028,7 @@  static int amdgpu_vm_update_level(struct amdgpu_device *adev,
 		/* assume the worst case */
 		ndw += parent->last_entry_used * 6;
 
-		pd_addr = amdgpu_bo_gpu_offset(parent->bo);
+		pd_addr = amdgpu_bo_gpu_offset(parent->base.bo);
 
 		if (shadow) {
 			shadow_addr = amdgpu_bo_gpu_offset(shadow);
@@ -1073,7 +1048,7 @@  static int amdgpu_vm_update_level(struct amdgpu_device *adev,
 
 	/* walk over the address space and update the directory */
 	for (pt_idx = 0; pt_idx <= parent->last_entry_used; ++pt_idx) {
-		struct amdgpu_bo *bo = parent->entries[pt_idx].bo;
+		struct amdgpu_bo *bo = parent->entries[pt_idx].base.bo;
 		uint64_t pde, pt;
 
 		if (bo == NULL)
@@ -1115,7 +1090,7 @@  static int amdgpu_vm_update_level(struct amdgpu_device *adev,
 	}
 
 	if (count) {
-		if (vm->root.bo->shadow)
+		if (vm->root.base.bo->shadow)
 			params.func(&params, last_shadow, last_pt,
 				    count, incr, AMDGPU_PTE_VALID);
 
@@ -1128,7 +1103,8 @@  static int amdgpu_vm_update_level(struct amdgpu_device *adev,
 			amdgpu_job_free(job);
 		} else {
 			amdgpu_ring_pad_ib(ring, params.ib);
-			amdgpu_sync_resv(adev, &job->sync, parent->bo->tbo.resv,
+			amdgpu_sync_resv(adev, &job->sync,
+					 parent->base.bo->tbo.resv,
 					 AMDGPU_FENCE_OWNER_VM);
 			if (shadow)
 				amdgpu_sync_resv(adev, &job->sync,
@@ -1141,7 +1117,7 @@  static int amdgpu_vm_update_level(struct amdgpu_device *adev,
 			if (r)
 				goto error_free;
 
-			amdgpu_bo_fence(parent->bo, fence, true);
+			amdgpu_bo_fence(parent->base.bo, fence, true);
 			dma_fence_put(vm->last_dir_update);
 			vm->last_dir_update = dma_fence_get(fence);
 			dma_fence_put(fence);
@@ -1154,7 +1130,7 @@  static int amdgpu_vm_update_level(struct amdgpu_device *adev,
 	for (pt_idx = 0; pt_idx <= parent->last_entry_used; ++pt_idx) {
 		struct amdgpu_vm_pt *entry = &parent->entries[pt_idx];
 
-		if (!entry->bo)
+		if (!entry->base.bo)
 			continue;
 
 		r = amdgpu_vm_update_level(adev, vm, entry, level + 1);
@@ -1187,7 +1163,7 @@  static void amdgpu_vm_invalidate_level(struct amdgpu_vm_pt *parent)
 	for (pt_idx = 0; pt_idx <= parent->last_entry_used; ++pt_idx) {
 		struct amdgpu_vm_pt *entry = &parent->entries[pt_idx];
 
-		if (!entry->bo)
+		if (!entry->base.bo)
 			continue;
 
 		entry->addr = ~0ULL;
@@ -1242,7 +1218,7 @@  void amdgpu_vm_get_entry(struct amdgpu_pte_update_params *p, uint64_t addr,
 	*entry = &p->vm->root;
 	while ((*entry)->entries) {
 		idx = addr >> (p->adev->vm_manager.block_size * level--);
-		idx %= amdgpu_bo_size((*entry)->bo) / 8;
+		idx %= amdgpu_bo_size((*entry)->base.bo) / 8;
 		*parent = *entry;
 		*entry = &(*entry)->entries[idx];
 	}
@@ -1278,7 +1254,7 @@  static void amdgpu_vm_handle_huge_pages(struct amdgpu_pte_update_params *p,
 	    p->func == amdgpu_vm_do_copy_ptes ||
 	    !(flags & AMDGPU_PTE_VALID)) {
 
-		dst = amdgpu_bo_gpu_offset(entry->bo);
+		dst = amdgpu_bo_gpu_offset(entry->base.bo);
 		dst = amdgpu_gart_get_vm_pde(p->adev, dst);
 		flags = AMDGPU_PTE_VALID;
 	} else {
@@ -1291,16 +1267,16 @@  static void amdgpu_vm_handle_huge_pages(struct amdgpu_pte_update_params *p,
 	entry->addr = (dst | flags);
 
 	if (use_cpu_update) {
-		pd_addr = (unsigned long)amdgpu_bo_kptr(parent->bo);
+		pd_addr = (unsigned long)amdgpu_bo_kptr(parent->base.bo);
 		pde = pd_addr + (entry - parent->entries) * 8;
 		amdgpu_vm_cpu_set_ptes(p, pde, dst, 1, 0, flags);
 	} else {
-		if (parent->bo->shadow) {
-			pd_addr = amdgpu_bo_gpu_offset(parent->bo->shadow);
+		if (parent->base.bo->shadow) {
+			pd_addr = amdgpu_bo_gpu_offset(parent->base.bo->shadow);
 			pde = pd_addr + (entry - parent->entries) * 8;
 			amdgpu_vm_do_set_ptes(p, pde, dst, 1, 0, flags);
 		}
-		pd_addr = amdgpu_bo_gpu_offset(parent->bo);
+		pd_addr = amdgpu_bo_gpu_offset(parent->base.bo);
 		pde = pd_addr + (entry - parent->entries) * 8;
 		amdgpu_vm_do_set_ptes(p, pde, dst, 1, 0, flags);
 	}
@@ -1350,7 +1326,7 @@  static int amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params,
 		if (entry->addr & AMDGPU_PDE_PTE)
 			continue;
 
-		pt = entry->bo;
+		pt = entry->base.bo;
 		if (use_cpu_update) {
 			pe_start = (unsigned long)amdgpu_bo_kptr(pt);
 		} else {
@@ -1572,12 +1548,12 @@  static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	if (r)
 		goto error_free;
 
-	r = amdgpu_sync_resv(adev, &job->sync, vm->root.bo->tbo.resv,
+	r = amdgpu_sync_resv(adev, &job->sync, vm->root.base.bo->tbo.resv,
 			     owner);
 	if (r)
 		goto error_free;
 
-	r = reservation_object_reserve_shared(vm->root.bo->tbo.resv);
+	r = reservation_object_reserve_shared(vm->root.base.bo->tbo.resv);
 	if (r)
 		goto error_free;
 
@@ -1592,7 +1568,7 @@  static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	if (r)
 		goto error_free;
 
-	amdgpu_bo_fence(vm->root.bo, f, true);
+	amdgpu_bo_fence(vm->root.base.bo, f, true);
 	dma_fence_put(*fence);
 	*fence = f;
 	return 0;
@@ -1899,7 +1875,7 @@  static void amdgpu_vm_free_mapping(struct amdgpu_device *adev,
  */
 static void amdgpu_vm_prt_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
-	struct reservation_object *resv = vm->root.bo->tbo.resv;
+	struct reservation_object *resv = vm->root.base.bo->tbo.resv;
 	struct dma_fence *excl, **shared;
 	unsigned i, shared_count;
 	int r;
@@ -2384,11 +2360,24 @@  void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
  * Mark @bo as invalid.
  */
 void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
-			     struct amdgpu_bo *bo)
+			     struct amdgpu_bo *bo, bool evicted)
 {
 	struct amdgpu_vm_bo_base *bo_base;
 
 	list_for_each_entry(bo_base, &bo->va, bo_list) {
+		struct amdgpu_vm *vm = bo_base->vm;
+
+		if (evicted && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
+			spin_lock(&bo_base->vm->status_lock);
+			list_move(&bo_base->vm_status, &vm->evicted);
+			spin_unlock(&bo_base->vm->status_lock);
+			continue;
+		}
+
+		/* Don't add page tables to the moved state */
+		if (bo->tbo.type == ttm_bo_type_kernel)
+			continue;
+
 		spin_lock(&bo_base->vm->status_lock);
 		if (list_empty(&bo_base->vm_status))
 			list_add(&bo_base->vm_status,
@@ -2461,6 +2450,7 @@  int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	for (i = 0; i < AMDGPU_MAX_VMHUBS; i++)
 		vm->reserved_vmid[i] = NULL;
 	spin_lock_init(&vm->status_lock);
+	INIT_LIST_HEAD(&vm->evicted);
 	INIT_LIST_HEAD(&vm->moved);
 	INIT_LIST_HEAD(&vm->cleared);
 	INIT_LIST_HEAD(&vm->freed);
@@ -2506,30 +2496,31 @@  int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	r = amdgpu_bo_create(adev, amdgpu_vm_bo_size(adev, 0), align, true,
 			     AMDGPU_GEM_DOMAIN_VRAM,
 			     flags,
-			     NULL, NULL, init_pde_value, &vm->root.bo);
+			     NULL, NULL, init_pde_value, &vm->root.base.bo);
 	if (r)
 		goto error_free_sched_entity;
 
-	r = amdgpu_bo_reserve(vm->root.bo, false);
-	if (r)
-		goto error_free_root;
-
-	vm->last_eviction_counter = atomic64_read(&adev->num_evictions);
+	vm->root.base.vm = vm;
+	list_add_tail(&vm->root.base.bo_list, &vm->root.base.bo->va);
+	INIT_LIST_HEAD(&vm->root.base.vm_status);
 
 	if (vm->use_cpu_for_update) {
-		r = amdgpu_bo_kmap(vm->root.bo, NULL);
+		r = amdgpu_bo_reserve(vm->root.base.bo, false);
 		if (r)
 			goto error_free_root;
-	}
 
-	amdgpu_bo_unreserve(vm->root.bo);
+		r = amdgpu_bo_kmap(vm->root.base.bo, NULL);
+		if (r)
+			goto error_free_root;
+		amdgpu_bo_unreserve(vm->root.base.bo);
+	}
 
 	return 0;
 
 error_free_root:
-	amdgpu_bo_unref(&vm->root.bo->shadow);
-	amdgpu_bo_unref(&vm->root.bo);
-	vm->root.bo = NULL;
+	amdgpu_bo_unref(&vm->root.base.bo->shadow);
+	amdgpu_bo_unref(&vm->root.base.bo);
+	vm->root.base.bo = NULL;
 
 error_free_sched_entity:
 	amd_sched_entity_fini(&ring->sched, &vm->entity);
@@ -2548,9 +2539,11 @@  static void amdgpu_vm_free_levels(struct amdgpu_vm_pt *level)
 {
 	unsigned i;
 
-	if (level->bo) {
-		amdgpu_bo_unref(&level->bo->shadow);
-		amdgpu_bo_unref(&level->bo);
+	if (level->base.bo) {
+		list_del(&level->base.bo_list);
+		list_del(&level->base.vm_status);
+		amdgpu_bo_unref(&level->base.bo->shadow);
+		amdgpu_bo_unref(&level->base.bo);
 	}
 
 	if (level->entries)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index a740b57..a649950 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -113,12 +113,12 @@  struct amdgpu_vm_bo_base {
 };
 
 struct amdgpu_vm_pt {
-	struct amdgpu_bo	*bo;
-	uint64_t		addr;
+	struct amdgpu_vm_bo_base	base;
+	uint64_t			addr;
 
 	/* array of page tables, one for each directory entry */
-	struct amdgpu_vm_pt	*entries;
-	unsigned		last_entry_used;
+	struct amdgpu_vm_pt		*entries;
+	unsigned			last_entry_used;
 };
 
 struct amdgpu_vm {
@@ -128,6 +128,9 @@  struct amdgpu_vm {
 	/* protecting invalidated */
 	spinlock_t		status_lock;
 
+	/* BOs who needs a validation */
+	struct list_head	evicted;
+
 	/* BOs moved, but not yet updated in the PT */
 	struct list_head	moved;
 
@@ -140,7 +143,6 @@  struct amdgpu_vm {
 	/* contains the page directory */
 	struct amdgpu_vm_pt     root;
 	struct dma_fence	*last_dir_update;
-	uint64_t		last_eviction_counter;
 
 	/* protecting freed */
 	spinlock_t		freed_lock;
@@ -253,7 +255,7 @@  int amdgpu_vm_bo_update(struct amdgpu_device *adev,
 			struct amdgpu_bo_va *bo_va,
 			bool clear);
 void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
-			     struct amdgpu_bo *bo);
+			     struct amdgpu_bo *bo, bool evicted);
 struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm,
 				       struct amdgpu_bo *bo);
 struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,

Comments

On 2017-08-10 10:50 AM, Christian König wrote:
> @@ -2384,11 +2360,24 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
>   * Mark @bo as invalid.
>   */
>  void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
> -			     struct amdgpu_bo *bo)
> +			     struct amdgpu_bo *bo, bool evicted)
>  {
>  	struct amdgpu_vm_bo_base *bo_base;
>  
>  	list_for_each_entry(bo_base, &bo->va, bo_list) {
> +		struct amdgpu_vm *vm = bo_base->vm;
> +
> +		if (evicted && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
> +			spin_lock(&bo_base->vm->status_lock);
> +			list_move(&bo_base->vm_status, &vm->evicted);
> +			spin_unlock(&bo_base->vm->status_lock);
> +			continue;
> +		}
> +
> +		/* Don't add page tables to the moved state */
> +		if (bo->tbo.type == ttm_bo_type_kernel)
> +			continue;
> +

It looks like you're using two different conditions to detect page
tables. When evicted is true, you use (bo->tbo.resv ==
vm->root.base.bo->tbo.resv). Otherwise you use (bo->tbo.type ==
ttm_bo_type_kernel). Are these really the same or is there a subtle
difference I'm missing? I'd find the following logic easier to read, but
I'm not sure if that's really your intention:

	if (bo->tbo.resv == vm->root.base.bo->tbo.resv) {
		spin_lock(&bo_base->vm->status_lock);

		if (evicted)
			list_move(&bo_base->vm_status, &vm->evicted);
		else if (list_empty(&bo_base->vm_status))
			/* this is from the next commit */
			list_add(&bo_base->vm_status, &vm->relocated);

		spin_unlock(&bo_base->vm->status_lock);
		continue;
	}

Regards,
  Felix

>  		spin_lock(&bo_base->vm->status_lock);
>  		if (list_empty(&bo_base->vm_status))
>  			list_add(&bo_base->vm_status,