1 files changed, 399 insertions, 186 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 4299ce386322..3911c78f8282 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -36,6 +36,7 @@
 #include <drm/ttm/ttm_tt.h>
 #include <drm/drm_exec.h>
 #include "amdgpu.h"
+#include "amdgpu_vm.h"
 #include "amdgpu_trace.h"
 #include "amdgpu_amdkfd.h"
 #include "amdgpu_gmc.h"
@@ -311,6 +312,111 @@ static void amdgpu_vm_bo_reset_state_machine(struct amdgpu_vm *vm)
 }
 
 /**
+ * amdgpu_vm_update_shared - helper to update shared memory stat
+ * @base: base structure for tracking BO usage in a VM
+ *
+ * Takes the vm status_lock and updates the shared memory stat. If the basic
+ * stat changed (e.g. buffer was moved) amdgpu_vm_update_stats need to be called
+ * as well.
+ */
+static void amdgpu_vm_update_shared(struct amdgpu_vm_bo_base *base)
+{
+	struct amdgpu_vm *vm = base->vm;
+	struct amdgpu_bo *bo = base->bo;
+	uint64_t size = amdgpu_bo_size(bo);
+	uint32_t bo_memtype = amdgpu_bo_mem_stats_placement(bo);
+	bool shared;
+
+	spin_lock(&vm->status_lock);
+	shared = drm_gem_object_is_shared_for_memory_stats(&bo->tbo.base);
+	if (base->shared != shared) {
+		base->shared = shared;
+		if (shared) {
+			vm->stats[bo_memtype].drm.shared += size;
+			vm->stats[bo_memtype].drm.private -= size;
+		} else {
+			vm->stats[bo_memtype].drm.shared -= size;
+			vm->stats[bo_memtype].drm.private += size;
+		}
+	}
+	spin_unlock(&vm->status_lock);
+}
+
+/**
+ * amdgpu_vm_bo_update_shared - callback when bo gets shared/unshared
+ * @bo: amdgpu buffer object
+ *
+ * Update the per VM stats for all the vm if needed from private to shared or
+ * vice versa.
+ */
+void amdgpu_vm_bo_update_shared(struct amdgpu_bo *bo)
+{
+	struct amdgpu_vm_bo_base *base;
+
+	for (base = bo->vm_bo; base; base = base->next)
+		amdgpu_vm_update_shared(base);
+}
+
+/**
+ * amdgpu_vm_update_stats_locked - helper to update normal memory stat
+ * @base: base structure for tracking BO usage in a VM
+ * @res:  the ttm_resource to use for the purpose of accounting, may or may not
+ *        be bo->tbo.resource
+ * @sign: if we should add (+1) or subtract (-1) from the stat
+ *
+ * Caller need to have the vm status_lock held. Useful for when multiple update
+ * need to happen at the same time.
+ */
+static void amdgpu_vm_update_stats_locked(struct amdgpu_vm_bo_base *base,
+			    struct ttm_resource *res, int sign)
+{
+	struct amdgpu_vm *vm = base->vm;
+	struct amdgpu_bo *bo = base->bo;
+	int64_t size = sign * amdgpu_bo_size(bo);
+	uint32_t bo_memtype = amdgpu_bo_mem_stats_placement(bo);
+
+	/* For drm-total- and drm-shared-, BO are accounted by their preferred
+	 * placement, see also amdgpu_bo_mem_stats_placement.
+	 */
+	if (base->shared)
+		vm->stats[bo_memtype].drm.shared += size;
+	else
+		vm->stats[bo_memtype].drm.private += size;
+
+	if (res && res->mem_type < __AMDGPU_PL_NUM) {
+		uint32_t res_memtype = res->mem_type;
+
+		vm->stats[res_memtype].drm.resident += size;
+		/* BO only count as purgeable if it is resident,
+		 * since otherwise there's nothing to purge.
+		 */
+		if (bo->flags & AMDGPU_GEM_CREATE_DISCARDABLE)
+			vm->stats[res_memtype].drm.purgeable += size;
+		if (!(bo->preferred_domains & amdgpu_mem_type_to_domain(res_memtype)))
+			vm->stats[bo_memtype].evicted += size;
+	}
+}
+
+/**
+ * amdgpu_vm_update_stats - helper to update normal memory stat
+ * @base: base structure for tracking BO usage in a VM
+ * @res:  the ttm_resource to use for the purpose of accounting, may or may not
+ *        be bo->tbo.resource
+ * @sign: if we should add (+1) or subtract (-1) from the stat
+ *
+ * Updates the basic memory stat when bo is added/deleted/moved.
+ */
+void amdgpu_vm_update_stats(struct amdgpu_vm_bo_base *base,
+			    struct ttm_resource *res, int sign)
+{
+	struct amdgpu_vm *vm = base->vm;
+
+	spin_lock(&vm->status_lock);
+	amdgpu_vm_update_stats_locked(base, res, sign);
+	spin_unlock(&vm->status_lock);
+}
+
+/**
  * amdgpu_vm_bo_base_init - Adds bo to the list of bos associated with the vm
  *
  * @base: base structure for tracking BO usage in a VM
@@ -333,7 +439,12 @@ void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
 	base->next = bo->vm_bo;
 	bo->vm_bo = base;
 
-	if (bo->tbo.base.resv != vm->root.bo->tbo.base.resv)
+	spin_lock(&vm->status_lock);
+	base->shared = drm_gem_object_is_shared_for_memory_stats(&bo->tbo.base);
+	amdgpu_vm_update_stats_locked(base, bo->tbo.resource, +1);
+	spin_unlock(&vm->status_lock);
+
+	if (!amdgpu_vm_is_bo_always_valid(vm, bo))
 		return;
 
 	dma_resv_assert_held(vm->root.bo->tbo.base.resv);
@@ -434,7 +545,7 @@ uint64_t amdgpu_vm_generation(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	if (!vm)
 		return result;
 
-	result += vm->generation;
+	result += lower_32_bits(vm->generation);
 	/* Add one if the page tables will be re-generated on next CS */
 	if (drm_sched_entity_error(&vm->delayed))
 		++result;
@@ -463,13 +574,13 @@ int amdgpu_vm_validate(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		       int (*validate)(void *p, struct amdgpu_bo *bo),
 		       void *param)
 {
+	uint64_t new_vm_generation = amdgpu_vm_generation(adev, vm);
 	struct amdgpu_vm_bo_base *bo_base;
-	struct amdgpu_bo *shadow;
 	struct amdgpu_bo *bo;
 	int r;
 
-	if (drm_sched_entity_error(&vm->delayed)) {
-		++vm->generation;
+	if (vm->generation != new_vm_generation) {
+		vm->generation = new_vm_generation;
 		amdgpu_vm_bo_reset_state_machine(vm);
 		amdgpu_vm_fini_entities(vm);
 		r = amdgpu_vm_init_entities(adev, vm);
@@ -485,16 +596,10 @@ int amdgpu_vm_validate(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		spin_unlock(&vm->status_lock);
 
 		bo = bo_base->bo;
-		shadow = amdgpu_bo_shadowed(bo);
 
 		r = validate(param, bo);
 		if (r)
 			return r;
-		if (shadow) {
-			r = validate(param, shadow);
-			if (r)
-				return r;
-		}
 
 		if (bo->tbo.type != ttm_bo_type_kernel) {
 			amdgpu_vm_bo_moved(bo_base);
@@ -649,6 +754,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 		    bool need_pipe_sync)
 {
 	struct amdgpu_device *adev = ring->adev;
+	struct amdgpu_isolation *isolation = &adev->isolation[ring->xcp_id];
 	unsigned vmhub = ring->vm_hub;
 	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
 	struct amdgpu_vmid *id = &id_mgr->ids[job->vmid];
@@ -656,8 +762,9 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 	bool gds_switch_needed = ring->funcs->emit_gds_switch &&
 		job->gds_switch_needed;
 	bool vm_flush_needed = job->vm_needs_flush;
-	struct dma_fence *fence = NULL;
+	bool cleaner_shader_needed = false;
 	bool pasid_mapping_needed = false;
+	struct dma_fence *fence = NULL;
 	unsigned int patch;
 	int r;
 
@@ -680,7 +787,13 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 	pasid_mapping_needed &= adev->gmc.gmc_funcs->emit_pasid_mapping &&
 		ring->funcs->emit_wreg;
 
-	if (!vm_flush_needed && !gds_switch_needed && !need_pipe_sync)
+	cleaner_shader_needed = job->run_cleaner_shader &&
+		adev->gfx.enable_cleaner_shader &&
+		ring->funcs->emit_cleaner_shader && job->base.s_fence &&
+		&job->base.s_fence->scheduled == isolation->spearhead;
+
+	if (!vm_flush_needed && !gds_switch_needed && !need_pipe_sync &&
+	    !cleaner_shader_needed)
 		return 0;
 
 	amdgpu_ring_ib_begin(ring);
@@ -691,6 +804,9 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 	if (need_pipe_sync)
 		amdgpu_ring_emit_pipeline_sync(ring);
 
+	if (cleaner_shader_needed)
+		ring->funcs->emit_cleaner_shader(ring);
+
 	if (vm_flush_needed) {
 		trace_amdgpu_vm_flush(ring, job->vmid, job->vm_pd_addr);
 		amdgpu_ring_emit_vm_flush(ring, job->vmid, job->vm_pd_addr);
@@ -702,7 +818,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 	if (spm_update_needed && adev->gfx.rlc.funcs->update_spm_vmid)
 		adev->gfx.rlc.funcs->update_spm_vmid(adev, ring, job->vmid);
 
-	if (!ring->is_mes_queue && ring->funcs->emit_gds_switch &&
+	if (ring->funcs->emit_gds_switch &&
 	    gds_switch_needed) {
 		amdgpu_ring_emit_gds_switch(ring, job->vmid, job->gds_base,
 					    job->gds_size, job->gws_base,
@@ -710,7 +826,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 					    job->oa_size);
 	}
 
-	if (vm_flush_needed || pasid_mapping_needed) {
+	if (vm_flush_needed || pasid_mapping_needed || cleaner_shader_needed) {
 		r = amdgpu_fence_emit(ring, &fence, NULL, 0);
 		if (r)
 			return r;
@@ -732,6 +848,18 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 		id->pasid_mapping = dma_fence_get(fence);
 		mutex_unlock(&id_mgr->lock);
 	}
+
+	/*
+	 * Make sure that all other submissions wait for the cleaner shader to
+	 * finish before we push them to the HW.
+	 */
+	if (cleaner_shader_needed) {
+		trace_amdgpu_cleaner_shader(ring, fence);
+		mutex_lock(&adev->enforce_isolation_mutex);
+		dma_fence_put(isolation->spearhead);
+		isolation->spearhead = dma_fence_get(fence);
+		mutex_unlock(&adev->enforce_isolation_mutex);
+	}
 	dma_fence_put(fence);
 
 	amdgpu_ring_patch_cond_exec(ring, patch);
@@ -741,6 +869,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 		amdgpu_ring_emit_switch_buffer(ring);
 		amdgpu_ring_emit_switch_buffer(ring);
 	}
+
 	amdgpu_ring_ib_end(ring);
 	return 0;
 }
@@ -837,7 +966,7 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev,
 	params.vm = vm;
 	params.immediate = immediate;
 
-	r = vm->update_funcs->prepare(&params, NULL, AMDGPU_SYNC_EXPLICIT);
+	r = vm->update_funcs->prepare(&params, NULL);
 	if (r)
 		goto error;
 
@@ -886,6 +1015,46 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
 }
 
 /**
+ * amdgpu_vm_tlb_flush - prepare TLB flush
+ *
+ * @params: parameters for update
+ * @fence: input fence to sync TLB flush with
+ * @tlb_cb: the callback structure
+ *
+ * Increments the tlb sequence to make sure that future CS execute a VM flush.
+ */
+static void
+amdgpu_vm_tlb_flush(struct amdgpu_vm_update_params *params,
+		    struct dma_fence **fence,
+		    struct amdgpu_vm_tlb_seq_struct *tlb_cb)
+{
+	struct amdgpu_vm *vm = params->vm;
+
+	tlb_cb->vm = vm;
+	if (!fence || !*fence) {
+		amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb);
+		return;
+	}
+
+	if (!dma_fence_add_callback(*fence, &tlb_cb->cb,
+				    amdgpu_vm_tlb_seq_cb)) {
+		dma_fence_put(vm->last_tlb_flush);
+		vm->last_tlb_flush = dma_fence_get(*fence);
+	} else {
+		amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb);
+	}
+
+	/* Prepare a TLB flush fence to be attached to PTs */
+	if (!params->unlocked && vm->is_compute_context) {
+		amdgpu_vm_tlb_fence_create(params->adev, vm, fence);
+
+		/* Makes sure no PD/PT is freed before the flush */
+		dma_resv_add_fence(vm->root.bo->tbo.base.resv, *fence,
+				   DMA_RESV_USAGE_BOOKKEEP);
+	}
+}
+
+/**
  * amdgpu_vm_update_range - update a range in the vm page table
  *
  * @adev: amdgpu_device pointer to use for commands
@@ -894,7 +1063,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
  * @unlocked: unlocked invalidation during MM callback
  * @flush_tlb: trigger tlb invalidation after update completed
  * @allow_override: change MTYPE for local NUMA nodes
- * @resv: fences we need to sync to
+ * @sync: fences we need to sync to
  * @start: start of mapped range
  * @last: last mapped entry
  * @flags: flags for the entries
@@ -910,16 +1079,16 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
  * 0 for success, negative erro code for failure.
  */
 int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-			   bool immediate, bool unlocked, bool flush_tlb, bool allow_override,
-			   struct dma_resv *resv, uint64_t start, uint64_t last,
-			   uint64_t flags, uint64_t offset, uint64_t vram_base,
+			   bool immediate, bool unlocked, bool flush_tlb,
+			   bool allow_override, struct amdgpu_sync *sync,
+			   uint64_t start, uint64_t last, uint64_t flags,
+			   uint64_t offset, uint64_t vram_base,
 			   struct ttm_resource *res, dma_addr_t *pages_addr,
 			   struct dma_fence **fence)
 {
-	struct amdgpu_vm_update_params params;
 	struct amdgpu_vm_tlb_seq_struct *tlb_cb;
+	struct amdgpu_vm_update_params params;
 	struct amdgpu_res_cursor cursor;
-	enum amdgpu_sync_mode sync_mode;
 	int r, idx;
 
 	if (!drm_dev_enter(adev_to_drm(adev), &idx))
@@ -927,8 +1096,8 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 
 	tlb_cb = kmalloc(sizeof(*tlb_cb), GFP_KERNEL);
 	if (!tlb_cb) {
-		r = -ENOMEM;
-		goto error_unlock;
+		drm_dev_exit(idx);
+		return -ENOMEM;
 	}
 
 	/* Vega20+XGMI where PTEs get inadvertently cached in L2 texture cache,
@@ -948,15 +1117,9 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	params.immediate = immediate;
 	params.pages_addr = pages_addr;
 	params.unlocked = unlocked;
+	params.needs_flush = flush_tlb;
 	params.allow_override = allow_override;
-
-	/* Implicitly sync to command submissions in the same VM before
-	 * unmapping. Sync to moving fences before mapping.
-	 */
-	if (!(flags & AMDGPU_PTE_VALID))
-		sync_mode = AMDGPU_SYNC_EQ_OWNER;
-	else
-		sync_mode = AMDGPU_SYNC_EXPLICIT;
+	INIT_LIST_HEAD(&params.tlb_flush_waitlist);
 
 	amdgpu_vm_eviction_lock(vm);
 	if (vm->evicting) {
@@ -972,7 +1135,7 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		dma_fence_put(tmp);
 	}
 
-	r = vm->update_funcs->prepare(&params, resv, sync_mode);
+	r = vm->update_funcs->prepare(&params, sync);
 	if (r)
 		goto error_free;
 
@@ -1015,7 +1178,7 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 				params.pages_addr = NULL;
 			}
 
-		} else if (flags & (AMDGPU_PTE_VALID | AMDGPU_PTE_PRT)) {
+		} else if (flags & (AMDGPU_PTE_VALID | AMDGPU_PTE_PRT_FLAG(adev))) {
 			addr = vram_base + cursor.start;
 		} else {
 			addr = 0;
@@ -1031,74 +1194,28 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	}
 
 	r = vm->update_funcs->commit(&params, fence);
+	if (r)
+		goto error_free;
 
-	if (flush_tlb || params.table_freed) {
-		tlb_cb->vm = vm;
-		if (fence && *fence &&
-		    !dma_fence_add_callback(*fence, &tlb_cb->cb,
-					   amdgpu_vm_tlb_seq_cb)) {
-			dma_fence_put(vm->last_tlb_flush);
-			vm->last_tlb_flush = dma_fence_get(*fence);
-		} else {
-			amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb);
-		}
+	if (params.needs_flush) {
+		amdgpu_vm_tlb_flush(&params, fence, tlb_cb);
 		tlb_cb = NULL;
 	}
 
+	amdgpu_vm_pt_free_list(adev, &params);
+
 error_free:
 	kfree(tlb_cb);
-
-error_unlock:
 	amdgpu_vm_eviction_unlock(vm);
 	drm_dev_exit(idx);
 	return r;
 }
 
-static void amdgpu_vm_bo_get_memory(struct amdgpu_bo_va *bo_va,
-				    struct amdgpu_mem_stats *stats)
-{
-	struct amdgpu_vm *vm = bo_va->base.vm;
-	struct amdgpu_bo *bo = bo_va->base.bo;
-
-	if (!bo)
-		return;
-
-	/*
-	 * For now ignore BOs which are currently locked and potentially
-	 * changing their location.
-	 */
-	if (bo->tbo.base.resv != vm->root.bo->tbo.base.resv &&
-	    !dma_resv_trylock(bo->tbo.base.resv))
-		return;
-
-	amdgpu_bo_get_memory(bo, stats);
-	if (bo->tbo.base.resv != vm->root.bo->tbo.base.resv)
-	    dma_resv_unlock(bo->tbo.base.resv);
-}
-
 void amdgpu_vm_get_memory(struct amdgpu_vm *vm,
-			  struct amdgpu_mem_stats *stats)
+			  struct amdgpu_mem_stats stats[__AMDGPU_PL_NUM])
 {
-	struct amdgpu_bo_va *bo_va, *tmp;
-
 	spin_lock(&vm->status_lock);
-	list_for_each_entry_safe(bo_va, tmp, &vm->idle, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->evicted, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->relocated, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->moved, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->invalidated, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats);
-
-	list_for_each_entry_safe(bo_va, tmp, &vm->done, base.vm_status)
-		amdgpu_vm_bo_get_memory(bo_va, stats);
+	memcpy(stats, vm->stats, sizeof(*stats) * __AMDGPU_PL_NUM);
 	spin_unlock(&vm->status_lock);
 }
 
@@ -1120,23 +1237,40 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
 	struct amdgpu_bo *bo = bo_va->base.bo;
 	struct amdgpu_vm *vm = bo_va->base.vm;
 	struct amdgpu_bo_va_mapping *mapping;
+	struct dma_fence **last_update;
 	dma_addr_t *pages_addr = NULL;
 	struct ttm_resource *mem;
-	struct dma_fence **last_update;
+	struct amdgpu_sync sync;
 	bool flush_tlb = clear;
-	bool uncached;
-	struct dma_resv *resv;
 	uint64_t vram_base;
 	uint64_t flags;
+	bool uncached;
 	int r;
 
-	if (clear || !bo) {
+	amdgpu_sync_create(&sync);
+	if (clear) {
 		mem = NULL;
-		resv = vm->root.bo->tbo.base.resv;
+
+		/* Implicitly sync to command submissions in the same VM before
+		 * unmapping.
+		 */
+		r = amdgpu_sync_resv(adev, &sync, vm->root.bo->tbo.base.resv,
+				     AMDGPU_SYNC_EQ_OWNER, vm);
+		if (r)
+			goto error_free;
+		if (bo) {
+			r = amdgpu_sync_kfd(&sync, bo->tbo.base.resv);
+			if (r)
+				goto error_free;
+		}
+	} else if (!bo) {
+		mem = NULL;
+
+		/* PRT map operations don't need to sync to anything. */
+
 	} else {
 		struct drm_gem_object *obj = &bo->tbo.base;
 
-		resv = bo->tbo.base.resv;
 		if (obj->import_attach && bo_va->is_xgmi) {
 			struct dma_buf *dma_buf = obj->import_attach->dmabuf;
 			struct drm_gem_object *gobj = dma_buf->priv;
@@ -1150,6 +1284,12 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
 		if (mem && (mem->mem_type == TTM_PL_TT ||
 			    mem->mem_type == AMDGPU_PL_PREEMPT))
 			pages_addr = bo->tbo.ttm->dma_address;
+
+		/* Implicitly sync to moving fences before mapping anything */
+		r = amdgpu_sync_resv(adev, &sync, bo->tbo.base.resv,
+				     AMDGPU_SYNC_EXPLICIT, vm);
+		if (r)
+			goto error_free;
 	}
 
 	if (bo) {
@@ -1169,8 +1309,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
 		uncached = false;
 	}
 
-	if (clear || (bo && bo->tbo.base.resv ==
-		      vm->root.bo->tbo.base.resv))
+	if (clear || amdgpu_vm_is_bo_always_valid(vm, bo))
 		last_update = &vm->last_update;
 	else
 		last_update = &bo_va->last_pt_update;
@@ -1200,23 +1339,22 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
 		trace_amdgpu_vm_bo_update(mapping);
 
 		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb,
-					   !uncached, resv, mapping->start, mapping->last,
-					   update_flags, mapping->offset,
-					   vram_base, mem, pages_addr,
-					   last_update);
+					   !uncached, &sync, mapping->start,
+					   mapping->last, update_flags,
+					   mapping->offset, vram_base, mem,
+					   pages_addr, last_update);
 		if (r)
-			return r;
+			goto error_free;
 	}
 
 	/* If the BO is not in its preferred location add it back to
 	 * the evicted list so that it gets validated again on the
 	 * next command submission.
 	 */
-	if (bo && bo->tbo.base.resv == vm->root.bo->tbo.base.resv) {
-		uint32_t mem_type = bo->tbo.resource->mem_type;
-
-		if (!(bo->preferred_domains &
-		      amdgpu_mem_type_to_domain(mem_type)))
+	if (amdgpu_vm_is_bo_always_valid(vm, bo)) {
+		if (bo->tbo.resource &&
+		    !(bo->preferred_domains &
+		      amdgpu_mem_type_to_domain(bo->tbo.resource->mem_type)))
 			amdgpu_vm_bo_evicted(&bo_va->base);
 		else
 			amdgpu_vm_bo_idle(&bo_va->base);
@@ -1233,7 +1371,9 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
 			trace_amdgpu_vm_bo_mapping(mapping);
 	}
 
-	return 0;
+error_free:
+	amdgpu_sync_free(&sync);
+	return r;
 }
 
 /**
@@ -1335,7 +1475,7 @@ static void amdgpu_vm_free_mapping(struct amdgpu_device *adev,
 				   struct amdgpu_bo_va_mapping *mapping,
 				   struct dma_fence *fence)
 {
-	if (mapping->flags & AMDGPU_PTE_PRT)
+	if (mapping->flags & AMDGPU_PTE_PRT_FLAG(adev))
 		amdgpu_vm_add_prt_cb(adev, fence);
 	kfree(mapping);
 }
@@ -1380,25 +1520,34 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
 			  struct amdgpu_vm *vm,
 			  struct dma_fence **fence)
 {
-	struct dma_resv *resv = vm->root.bo->tbo.base.resv;
 	struct amdgpu_bo_va_mapping *mapping;
-	uint64_t init_pte_value = 0;
 	struct dma_fence *f = NULL;
+	struct amdgpu_sync sync;
 	int r;
 
+
+	/*
+	 * Implicitly sync to command submissions in the same VM before
+	 * unmapping.
+	 */
+	amdgpu_sync_create(&sync);
+	r = amdgpu_sync_resv(adev, &sync, vm->root.bo->tbo.base.resv,
+			     AMDGPU_SYNC_EQ_OWNER, vm);
+	if (r)
+		goto error_free;
+
 	while (!list_empty(&vm->freed)) {
 		mapping = list_first_entry(&vm->freed,
 			struct amdgpu_bo_va_mapping, list);
 		list_del(&mapping->list);
 
 		r = amdgpu_vm_update_range(adev, vm, false, false, true, false,
-					   resv, mapping->start, mapping->last,
-					   init_pte_value, 0, 0, NULL, NULL,
-					   &f);
+					   &sync, mapping->start, mapping->last,
+					   0, 0, 0, NULL, NULL, &f);
 		amdgpu_vm_free_mapping(adev, vm, mapping, f);
 		if (r) {
 			dma_fence_put(f);
-			return r;
+			goto error_free;
 		}
 	}
 
@@ -1409,7 +1558,9 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
 		dma_fence_put(f);
 	}
 
-	return 0;
+error_free:
+	amdgpu_sync_free(&sync);
+	return r;
 
 }
 
@@ -1603,16 +1754,46 @@ static void amdgpu_vm_bo_insert_map(struct amdgpu_device *adev,
 	list_add(&mapping->list, &bo_va->invalids);
 	amdgpu_vm_it_insert(mapping, &vm->va);
 
-	if (mapping->flags & AMDGPU_PTE_PRT)
+	if (mapping->flags & AMDGPU_PTE_PRT_FLAG(adev))
 		amdgpu_vm_prt_get(adev);
 
-	if (bo && bo->tbo.base.resv == vm->root.bo->tbo.base.resv &&
-	    !bo_va->base.moved) {
+	if (amdgpu_vm_is_bo_always_valid(vm, bo) && !bo_va->base.moved)
 		amdgpu_vm_bo_moved(&bo_va->base);
-	}
+
 	trace_amdgpu_vm_bo_map(bo_va, mapping);
 }
 
+/* Validate operation parameters to prevent potential abuse */
+static int amdgpu_vm_verify_parameters(struct amdgpu_device *adev,
+					  struct amdgpu_bo *bo,
+					  uint64_t saddr,
+					  uint64_t offset,
+					  uint64_t size)
+{
+	uint64_t tmp, lpfn;
+
+	if (saddr & AMDGPU_GPU_PAGE_MASK
+	    || offset & AMDGPU_GPU_PAGE_MASK
+	    || size & AMDGPU_GPU_PAGE_MASK)
+		return -EINVAL;
+
+	if (check_add_overflow(saddr, size, &tmp)
+	    || check_add_overflow(offset, size, &tmp)
+	    || size == 0 /* which also leads to end < begin */)
+		return -EINVAL;
+
+	/* make sure object fit at this offset */
+	if (bo && offset + size > amdgpu_bo_size(bo))
+		return -EINVAL;
+
+	/* Ensure last pfn not exceed max_pfn */
+	lpfn = (saddr + size - 1) >> AMDGPU_GPU_PAGE_SHIFT;
+	if (lpfn >= adev->vm_manager.max_pfn)
+		return -EINVAL;
+
+	return 0;
+}
+
 /**
  * amdgpu_vm_bo_map - map bo inside a vm
  *
@@ -1639,21 +1820,14 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	struct amdgpu_bo *bo = bo_va->base.bo;
 	struct amdgpu_vm *vm = bo_va->base.vm;
 	uint64_t eaddr;
+	int r;
 
-	/* validate the parameters */
-	if (saddr & ~PAGE_MASK || offset & ~PAGE_MASK || size & ~PAGE_MASK)
-		return -EINVAL;
-	if (saddr + size <= saddr || offset + size <= offset)
-		return -EINVAL;
-
-	/* make sure object fit at this offset */
-	eaddr = saddr + size - 1;
-	if ((bo && offset + size > amdgpu_bo_size(bo)) ||
-	    (eaddr >= adev->vm_manager.max_pfn << AMDGPU_GPU_PAGE_SHIFT))
-		return -EINVAL;
+	r = amdgpu_vm_verify_parameters(adev, bo, saddr, offset, size);
+	if (r)
+		return r;
 
 	saddr /= AMDGPU_GPU_PAGE_SIZE;
-	eaddr /= AMDGPU_GPU_PAGE_SIZE;
+	eaddr = saddr + (size - 1) / AMDGPU_GPU_PAGE_SIZE;
 
 	tmp = amdgpu_vm_it_iter_first(&vm->va, saddr, eaddr);
 	if (tmp) {
@@ -1706,17 +1880,9 @@ int amdgpu_vm_bo_replace_map(struct amdgpu_device *adev,
 	uint64_t eaddr;
 	int r;
 
-	/* validate the parameters */
-	if (saddr & ~PAGE_MASK || offset & ~PAGE_MASK || size & ~PAGE_MASK)
-		return -EINVAL;
-	if (saddr + size <= saddr || offset + size <= offset)
-		return -EINVAL;
-
-	/* make sure object fit at this offset */
-	eaddr = saddr + size - 1;
-	if ((bo && offset + size > amdgpu_bo_size(bo)) ||
-	    (eaddr >= adev->vm_manager.max_pfn << AMDGPU_GPU_PAGE_SHIFT))
-		return -EINVAL;
+	r = amdgpu_vm_verify_parameters(adev, bo, saddr, offset, size);
+	if (r)
+		return r;
 
 	/* Allocate all the needed memory */
 	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
@@ -1730,7 +1896,7 @@ int amdgpu_vm_bo_replace_map(struct amdgpu_device *adev,
 	}
 
 	saddr /= AMDGPU_GPU_PAGE_SIZE;
-	eaddr /= AMDGPU_GPU_PAGE_SIZE;
+	eaddr = saddr + (size - 1) / AMDGPU_GPU_PAGE_SIZE;
 
 	mapping->start = saddr;
 	mapping->last = eaddr;
@@ -1817,10 +1983,14 @@ int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev,
 	struct amdgpu_bo_va_mapping *before, *after, *tmp, *next;
 	LIST_HEAD(removed);
 	uint64_t eaddr;
+	int r;
+
+	r = amdgpu_vm_verify_parameters(adev, NULL, saddr, 0, size);
+	if (r)
+		return r;
 
-	eaddr = saddr + size - 1;
 	saddr /= AMDGPU_GPU_PAGE_SIZE;
-	eaddr /= AMDGPU_GPU_PAGE_SIZE;
+	eaddr = saddr + (size - 1) / AMDGPU_GPU_PAGE_SIZE;
 
 	/* Allocate all the needed memory */
 	before = kzalloc(sizeof(*before), GFP_KERNEL);
@@ -1885,10 +2055,10 @@ int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev,
 		struct amdgpu_bo *bo = before->bo_va->base.bo;
 
 		amdgpu_vm_it_insert(before, &vm->va);
-		if (before->flags & AMDGPU_PTE_PRT)
+		if (before->flags & AMDGPU_PTE_PRT_FLAG(adev))
 			amdgpu_vm_prt_get(adev);
 
-		if (bo && bo->tbo.base.resv == vm->root.bo->tbo.base.resv &&
+		if (amdgpu_vm_is_bo_always_valid(vm, bo) &&
 		    !before->bo_va->base.moved)
 			amdgpu_vm_bo_moved(&before->bo_va->base);
 	} else {
@@ -1900,10 +2070,10 @@ int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev,
 		struct amdgpu_bo *bo = after->bo_va->base.bo;
 
 		amdgpu_vm_it_insert(after, &vm->va);
-		if (after->flags & AMDGPU_PTE_PRT)
+		if (after->flags & AMDGPU_PTE_PRT_FLAG(adev))
 			amdgpu_vm_prt_get(adev);
 
-		if (bo && bo->tbo.base.resv == vm->root.bo->tbo.base.resv &&
+		if (amdgpu_vm_is_bo_always_valid(vm, bo) &&
 		    !after->bo_va->base.moved)
 			amdgpu_vm_bo_moved(&after->bo_va->base);
 	} else {
@@ -1983,7 +2153,7 @@ void amdgpu_vm_bo_del(struct amdgpu_device *adev,
 
 	if (bo) {
 		dma_resv_assert_held(bo->tbo.base.resv);
-		if (bo->tbo.base.resv == vm->root.bo->tbo.base.resv)
+		if (amdgpu_vm_is_bo_always_valid(vm, bo))
 			ttm_bo_set_bulk_move(&bo->tbo, NULL);
 
 		for (base = &bo_va->base.bo->vm_bo; *base;
@@ -1991,6 +2161,7 @@ void amdgpu_vm_bo_del(struct amdgpu_device *adev,
 			if (*base != &bo_va->base)
 				continue;
 
+			amdgpu_vm_update_stats(*base, bo->tbo.resource, -1);
 			*base = bo_va->base.next;
 			break;
 		}
@@ -2059,25 +2230,19 @@ bool amdgpu_vm_evictable(struct amdgpu_bo *bo)
 /**
  * amdgpu_vm_bo_invalidate - mark the bo as invalid
  *
- * @adev: amdgpu_device pointer
  * @bo: amdgpu buffer object
  * @evicted: is the BO evicted
  *
  * Mark @bo as invalid.
  */
-void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
-			     struct amdgpu_bo *bo, bool evicted)
+void amdgpu_vm_bo_invalidate(struct amdgpu_bo *bo, bool evicted)
 {
 	struct amdgpu_vm_bo_base *bo_base;
 
-	/* shadow bo doesn't have bo base, its validation needs its parent */
-	if (bo->parent && (amdgpu_bo_shadowed(bo->parent) == bo))
-		bo = bo->parent;
-
 	for (bo_base = bo->vm_bo; bo_base; bo_base = bo_base->next) {
 		struct amdgpu_vm *vm = bo_base->vm;
 
-		if (evicted && bo->tbo.base.resv == vm->root.bo->tbo.base.resv) {
+		if (evicted && amdgpu_vm_is_bo_always_valid(vm, bo)) {
 			amdgpu_vm_bo_evicted(bo_base);
 			continue;
 		}
@@ -2088,7 +2253,7 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
 
 		if (bo->tbo.type == ttm_bo_type_kernel)
 			amdgpu_vm_bo_relocated(bo_base);
-		else if (bo->tbo.base.resv == vm->root.bo->tbo.base.resv)
+		else if (amdgpu_vm_is_bo_always_valid(vm, bo))
 			amdgpu_vm_bo_moved(bo_base);
 		else
 			amdgpu_vm_bo_invalidated(bo_base);
@@ -2096,6 +2261,32 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
 }
 
 /**
+ * amdgpu_vm_bo_move - handle BO move
+ *
+ * @bo: amdgpu buffer object
+ * @new_mem: the new placement of the BO move
+ * @evicted: is the BO evicted
+ *
+ * Update the memory stats for the new placement and mark @bo as invalid.
+ */
+void amdgpu_vm_bo_move(struct amdgpu_bo *bo, struct ttm_resource *new_mem,
+		       bool evicted)
+{
+	struct amdgpu_vm_bo_base *bo_base;
+
+	for (bo_base = bo->vm_bo; bo_base; bo_base = bo_base->next) {
+		struct amdgpu_vm *vm = bo_base->vm;
+
+		spin_lock(&vm->status_lock);
+		amdgpu_vm_update_stats_locked(bo_base, bo->tbo.resource, -1);
+		amdgpu_vm_update_stats_locked(bo_base, new_mem, +1);
+		spin_unlock(&vm->status_lock);
+	}
+
+	amdgpu_vm_bo_invalidate(bo, evicted);
+}
+
+/**
  * amdgpu_vm_get_block_size - calculate VM page table size as power of two
  *
  * @vm_size: VM size
@@ -2165,7 +2356,7 @@ void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint32_t min_vm_size,
 		phys_ram_gb = ((uint64_t)si.totalram * si.mem_unit +
 			       (1 << 30) - 1) >> 30;
 		vm_size = roundup_pow_of_two(
-			min(max(phys_ram_gb * 3, min_vm_size), max_size));
+			clamp(phys_ram_gb * 3, min_vm_size, max_size));
 	}
 
 	adev->vm_manager.max_pfn = (uint64_t)vm_size << 18;
@@ -2360,14 +2551,14 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->status_lock);
 	INIT_LIST_HEAD(&vm->freed);
 	INIT_LIST_HEAD(&vm->done);
-	INIT_LIST_HEAD(&vm->pt_freed);
-	INIT_WORK(&vm->pt_free_work, amdgpu_vm_pt_free_work);
 	INIT_KFIFO(vm->faults);
 
 	r = amdgpu_vm_init_entities(adev, vm);
 	if (r)
 		return r;
 
+	ttm_lru_bulk_move_init(&vm->lru_bulk_move);
+
 	vm->is_compute_context = false;
 
 	vm->use_cpu_for_update = !!(adev->vm_manager.vm_update_mode &
@@ -2387,10 +2578,11 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	vm->last_update = dma_fence_get_stub();
 	vm->last_unlocked = dma_fence_get_stub();
 	vm->last_tlb_flush = dma_fence_get_stub();
-	vm->generation = 0;
+	vm->generation = amdgpu_vm_generation(adev, NULL);
 
 	mutex_init(&vm->eviction_lock);
 	vm->evicting = false;
+	vm->tlb_fence_context = dma_fence_context_alloc(1);
 
 	r = amdgpu_vm_pt_create(adev, vm, adev->vm_manager.root_level,
 				false, &root, xcp_id);
@@ -2400,7 +2592,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	root_bo = amdgpu_bo_ref(&root->bo);
 	r = amdgpu_bo_reserve(root_bo, true);
 	if (r) {
-		amdgpu_bo_unref(&root->shadow);
 		amdgpu_bo_unref(&root_bo);
 		goto error_free_delayed;
 	}
@@ -2431,6 +2622,7 @@ error_free_root:
 error_free_delayed:
 	dma_fence_put(vm->last_tlb_flush);
 	dma_fence_put(vm->last_unlocked);
+	ttm_lru_bulk_move_fini(&adev->mman.bdev, &vm->lru_bulk_move);
 	amdgpu_vm_fini_entities(vm);
 
 	return r;
@@ -2492,28 +2684,19 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	vm->last_update = dma_fence_get_stub();
 	vm->is_compute_context = true;
 
-	/* Free the shadow bo for compute VM */
-	amdgpu_bo_unref(&to_amdgpu_bo_vm(vm->root.bo)->shadow);
-
-	goto unreserve_bo;
-
 unreserve_bo:
 	amdgpu_bo_unreserve(vm->root.bo);
 	return r;
 }
 
-/**
- * amdgpu_vm_release_compute - release a compute vm
- * @adev: amdgpu_device pointer
- * @vm: a vm turned into compute vm by calling amdgpu_vm_make_compute
- *
- * This is a correspondant of amdgpu_vm_make_compute. It decouples compute
- * pasid from vm. Compute should stop use of vm after this call.
- */
-void amdgpu_vm_release_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
+static int amdgpu_vm_stats_is_zero(struct amdgpu_vm *vm)
 {
-	amdgpu_vm_set_pasid(adev, vm, 0);
-	vm->is_compute_context = false;
+	for (int i = 0; i < __AMDGPU_PL_NUM; ++i) {
+		if (!(drm_memory_stats_is_zero(&vm->stats[i].drm) &&
+		      vm->stats[i].evicted == 0))
+			return false;
+	}
+	return true;
 }
 
 /**
@@ -2535,11 +2718,8 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 
 	amdgpu_amdkfd_gpuvm_destroy_cb(adev, vm);
 
-	flush_work(&vm->pt_free_work);
-
 	root = amdgpu_bo_ref(vm->root.bo);
 	amdgpu_bo_reserve(root, true);
-	amdgpu_vm_put_task_info(vm->task_info);
 	amdgpu_vm_set_pasid(adev, vm, 0);
 	dma_fence_wait(vm->last_unlocked, false);
 	dma_fence_put(vm->last_unlocked);
@@ -2550,7 +2730,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	dma_fence_put(vm->last_tlb_flush);
 
 	list_for_each_entry_safe(mapping, tmp, &vm->freed, list) {
-		if (mapping->flags & AMDGPU_PTE_PRT && prt_fini_needed) {
+		if (mapping->flags & AMDGPU_PTE_PRT_FLAG(adev) && prt_fini_needed) {
 			amdgpu_vm_prt_fini(adev, vm);
 			prt_fini_needed = false;
 		}
@@ -2587,6 +2767,17 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 		}
 	}
 
+	ttm_lru_bulk_move_fini(&adev->mman.bdev, &vm->lru_bulk_move);
+
+	if (!amdgpu_vm_stats_is_zero(vm)) {
+		struct amdgpu_task_info *ti = vm->task_info;
+
+		dev_warn(adev->dev,
+			 "VM memory stats for proc %s(%d) task %s(%d) is non-zero when fini\n",
+			 ti->process_name, ti->pid, ti->task_name, ti->tgid);
+	}
+
+	amdgpu_vm_put_task_info(vm->task_info);
 }
 
 /**
@@ -2700,6 +2891,7 @@ int amdgpu_vm_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
  * amdgpu_vm_handle_fault - graceful handling of VM faults.
  * @adev: amdgpu device pointer
  * @pasid: PASID of the VM
+ * @ts: Timestamp of the fault
  * @vmid: VMID, only used for GFX 9.4.3.
  * @node_id: Node_id received in IH cookie. Only applicable for
  *           GFX 9.4.3.
@@ -2710,7 +2902,7 @@ int amdgpu_vm_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
  * shouldn't be reported any more.
  */
 bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
-			    u32 vmid, u32 node_id, uint64_t addr,
+			    u32 vmid, u32 node_id, uint64_t addr, uint64_t ts,
 			    bool write_fault)
 {
 	bool is_compute_context = false;
@@ -2736,7 +2928,7 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
 	addr /= AMDGPU_GPU_PAGE_SIZE;
 
 	if (is_compute_context && !svm_range_restore_pages(adev, pasid, vmid,
-	    node_id, addr, write_fault)) {
+	    node_id, addr, ts, write_fault)) {
 		amdgpu_bo_unref(&root);
 		return true;
 	}
@@ -2924,6 +3116,14 @@ void amdgpu_vm_update_fault_cache(struct amdgpu_device *adev,
 	if (vm && status) {
 		vm->fault_info.addr = addr;
 		vm->fault_info.status = status;
+		/*
+		 * Update the fault information globally for later usage
+		 * when vm could be stale or freed.
+		 */
+		adev->vm_manager.fault_info.addr = addr;
+		adev->vm_manager.fault_info.vmhub = vmhub;
+		adev->vm_manager.fault_info.status = status;
+
 		if (AMDGPU_IS_GFXHUB(vmhub)) {
 			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_GFX;
 			vm->fault_info.vmhub |=
@@ -2943,3 +3143,16 @@ void amdgpu_vm_update_fault_cache(struct amdgpu_device *adev,
 	xa_unlock_irqrestore(&adev->vm_manager.pasids, flags);
 }
 
+/**
+ * amdgpu_vm_is_bo_always_valid - check if the BO is VM always valid
+ *
+ * @vm: VM to test against.
+ * @bo: BO to be tested.
+ *
+ * Returns true if the BO shares the dma_resv object with the root PD and is
+ * always guaranteed to be valid inside the VM.
+ */
+bool amdgpu_vm_is_bo_always_valid(struct amdgpu_vm *vm, struct amdgpu_bo *bo)
+{
+	return bo && bo->tbo.base.resv == vm->root.bo->tbo.base.resv;
+}