1 files changed, 186 insertions, 155 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 1e998f972c30..35fe2c974699 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -197,7 +197,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 			return -EINVAL;
 
 		vram_size = KFD_XCP_MEMORY_SIZE(adev, xcp_id);
-		if (adev->flags & AMD_IS_APU) {
+		if (adev->apu_prefer_gtt) {
 			system_mem_needed = size;
 			ttm_mem_needed = size;
 		}
@@ -213,19 +213,35 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
 
 	if (kfd_mem_limit.system_mem_used + system_mem_needed >
-	    kfd_mem_limit.max_system_mem_limit)
+	    kfd_mem_limit.max_system_mem_limit) {
 		pr_debug("Set no_system_mem_limit=1 if using shared memory\n");
+		if (!no_system_mem_limit) {
+			ret = -ENOMEM;
+			goto release;
+		}
+	}
 
-	if ((kfd_mem_limit.system_mem_used + system_mem_needed >
-	     kfd_mem_limit.max_system_mem_limit && !no_system_mem_limit) ||
-	    (kfd_mem_limit.ttm_mem_used + ttm_mem_needed >
-	     kfd_mem_limit.max_ttm_mem_limit) ||
-	    (adev && xcp_id >= 0 && adev->kfd.vram_used[xcp_id] + vram_needed >
-	     vram_size - reserved_for_pt - reserved_for_ras - atomic64_read(&adev->vram_pin_size))) {
+	if (kfd_mem_limit.ttm_mem_used + ttm_mem_needed >
+		kfd_mem_limit.max_ttm_mem_limit) {
 		ret = -ENOMEM;
 		goto release;
 	}
 
+	/*if is_app_apu is false and apu_prefer_gtt is true, it is an APU with
+	 * carve out < gtt. In that case, VRAM allocation will go to gtt domain, skip
+	 * VRAM check since ttm_mem_limit check already cover this allocation
+	 */
+
+	if (adev && xcp_id >= 0 && (!adev->apu_prefer_gtt || adev->gmc.is_app_apu)) {
+		uint64_t vram_available =
+			vram_size - reserved_for_pt - reserved_for_ras -
+			atomic64_read(&adev->vram_pin_size);
+		if (adev->kfd.vram_used[xcp_id] + vram_needed > vram_available) {
+			ret = -ENOMEM;
+			goto release;
+		}
+	}
+
 	/* Update memory accounting by decreasing available system
 	 * memory, TTM memory and GPU memory as computed above
 	 */
@@ -234,7 +250,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 	if (adev && xcp_id >= 0) {
 		adev->kfd.vram_used[xcp_id] += vram_needed;
 		adev->kfd.vram_used_aligned[xcp_id] +=
-				(adev->flags & AMD_IS_APU) ?
+				adev->apu_prefer_gtt ?
 				vram_needed :
 				ALIGN(vram_needed, VRAM_AVAILABLITY_ALIGN);
 	}
@@ -262,7 +278,7 @@ void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
 
 		if (adev) {
 			adev->kfd.vram_used[xcp_id] -= size;
-			if (adev->flags & AMD_IS_APU) {
+			if (adev->apu_prefer_gtt) {
 				adev->kfd.vram_used_aligned[xcp_id] -= size;
 				kfd_mem_limit.system_mem_used -= size;
 				kfd_mem_limit.ttm_mem_used -= size;
@@ -370,40 +386,32 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo,
 	return 0;
 }
 
-int amdgpu_amdkfd_remove_fence_on_pt_pd_bos(struct amdgpu_bo *bo)
+/**
+ * amdgpu_amdkfd_remove_all_eviction_fences - Remove all eviction fences
+ * @bo: the BO where to remove the evictions fences from.
+ *
+ * This functions should only be used on release when all references to the BO
+ * are already dropped. We remove the eviction fence from the private copy of
+ * the dma_resv object here since that is what is used during release to
+ * determine of the BO is idle or not.
+ */
+void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo)
 {
-	struct amdgpu_bo *root = bo;
-	struct amdgpu_vm_bo_base *vm_bo;
-	struct amdgpu_vm *vm;
-	struct amdkfd_process_info *info;
-	struct amdgpu_amdkfd_fence *ef;
-	int ret;
-
-	/* we can always get vm_bo from root PD bo.*/
-	while (root->parent)
-		root = root->parent;
-
-	vm_bo = root->vm_bo;
-	if (!vm_bo)
-		return 0;
-
-	vm = vm_bo->vm;
-	if (!vm)
-		return 0;
+	struct dma_resv *resv = &bo->tbo.base._resv;
+	struct dma_fence *fence, *stub;
+	struct dma_resv_iter cursor;
 
-	info = vm->process_info;
-	if (!info || !info->eviction_fence)
-		return 0;
-
-	ef = container_of(dma_fence_get(&info->eviction_fence->base),
-			struct amdgpu_amdkfd_fence, base);
+	dma_resv_assert_held(resv);
 
-	BUG_ON(!dma_resv_trylock(bo->tbo.base.resv));
-	ret = amdgpu_amdkfd_remove_eviction_fence(bo, ef);
-	dma_resv_unlock(bo->tbo.base.resv);
+	stub = dma_fence_get_stub();
+	dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_BOOKKEEP, fence) {
+		if (!to_amdgpu_amdkfd_fence(fence))
+			continue;
 
-	dma_fence_put(&ef->base);
-	return ret;
+		dma_resv_replace_fences(resv, fence->context, stub,
+					DMA_RESV_USAGE_BOOKKEEP);
+	}
+	dma_fence_put(stub);
 }
 
 static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain,
@@ -499,10 +507,11 @@ static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync)
 	if (ret)
 		return ret;
 
-	return amdgpu_sync_fence(sync, vm->last_update);
+	return amdgpu_sync_fence(sync, vm->last_update, GFP_KERNEL);
 }
 
-static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem)
+static uint64_t get_pte_flags(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			      struct kgd_mem *mem)
 {
 	uint32_t mapping_flags = AMDGPU_VM_PAGE_READABLE |
 				 AMDGPU_VM_MTYPE_DEFAULT;
@@ -512,7 +521,7 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem)
 	if (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE)
 		mapping_flags |= AMDGPU_VM_PAGE_EXECUTABLE;
 
-	return amdgpu_gem_va_map_flags(adev, mapping_flags);
+	return mapping_flags;
 }
 
 /**
@@ -531,7 +540,7 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem)
  */
 static struct sg_table *create_sg_table(uint64_t addr, uint32_t size)
 {
-	struct sg_table *sg = kmalloc(sizeof(*sg), GFP_KERNEL);
+	struct sg_table *sg = kmalloc_obj(*sg);
 
 	if (!sg)
 		return NULL;
@@ -564,7 +573,7 @@ kfd_mem_dmamap_userptr(struct kgd_mem *mem,
 	if (WARN_ON(ttm->num_pages != src_ttm->num_pages))
 		return -EINVAL;
 
-	ttm->sg = kmalloc(sizeof(*ttm->sg), GFP_KERNEL);
+	ttm->sg = kmalloc_obj(*ttm->sg);
 	if (unlikely(!ttm->sg))
 		return -ENOMEM;
 
@@ -603,12 +612,6 @@ kfd_mem_dmamap_dmabuf(struct kfd_mem_attachment *attachment)
 {
 	struct ttm_operation_ctx ctx = {.interruptible = true};
 	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
-	int ret;
-
-	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
-	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
-	if (ret)
-		return ret;
 
 	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
 	return ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
@@ -875,6 +878,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
 	struct amdgpu_bo *bo[2] = {NULL, NULL};
 	struct amdgpu_bo_va *bo_va;
 	bool same_hive = false;
+	struct drm_exec exec;
 	int i, ret;
 
 	if (!va) {
@@ -890,7 +894,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
 	 * if peer device has large BAR. In contrast, access over xGMI is
 	 * allowed for both small and large BAR configurations of peer device
 	 */
-	if ((adev != bo_adev && !(adev->flags & AMD_IS_APU)) &&
+	if ((adev != bo_adev && !adev->apu_prefer_gtt) &&
 	    ((mem->domain == AMDGPU_GEM_DOMAIN_VRAM) ||
 	     (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL) ||
 	     (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
@@ -955,19 +959,25 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
 			goto unwind;
 		}
 
-		/* Add BO to VM internal data structures */
-		ret = amdgpu_bo_reserve(bo[i], false);
-		if (ret) {
-			pr_debug("Unable to reserve BO during memory attach");
-			goto unwind;
+		drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0);
+		drm_exec_until_all_locked(&exec) {
+			ret = amdgpu_vm_lock_pd(vm, &exec, 0);
+			drm_exec_retry_on_contention(&exec);
+			if (unlikely(ret))
+				goto unwind;
+			ret = drm_exec_lock_obj(&exec, &bo[i]->tbo.base);
+			drm_exec_retry_on_contention(&exec);
+			if (unlikely(ret))
+				goto unwind;
 		}
+
 		bo_va = amdgpu_vm_bo_find(vm, bo[i]);
 		if (!bo_va)
 			bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
 		else
 			++bo_va->ref_count;
 		attachment[i]->bo_va = bo_va;
-		amdgpu_bo_unreserve(bo[i]);
+		drm_exec_fini(&exec);
 		if (unlikely(!attachment[i]->bo_va)) {
 			ret = -ENOMEM;
 			pr_err("Failed to add BO object to VM. ret == %d\n",
@@ -975,7 +985,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
 			goto unwind;
 		}
 		attachment[i]->va = va;
-		attachment[i]->pte_flags = get_pte_flags(adev, mem);
+		attachment[i]->pte_flags = get_pte_flags(adev, vm, mem);
 		attachment[i]->adev = adev;
 		list_add(&attachment[i]->list, &mem->attachments);
 
@@ -1054,7 +1064,7 @@ static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr,
 	struct amdkfd_process_info *process_info = mem->process_info;
 	struct amdgpu_bo *bo = mem->bo;
 	struct ttm_operation_ctx ctx = { true, false };
-	struct hmm_range *range;
+	struct amdgpu_hmm_range *range;
 	int ret = 0;
 
 	mutex_lock(&process_info->lock);
@@ -1086,8 +1096,15 @@ static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr,
 		return 0;
 	}
 
-	ret = amdgpu_ttm_tt_get_user_pages(bo, bo->tbo.ttm->pages, &range);
+	range = amdgpu_hmm_range_alloc(NULL);
+	if (unlikely(!range)) {
+		ret = -ENOMEM;
+		goto unregister_out;
+	}
+
+	ret = amdgpu_ttm_tt_get_user_pages(bo, range);
 	if (ret) {
+		amdgpu_hmm_range_free(range);
 		if (ret == -EAGAIN)
 			pr_debug("Failed to get user pages, try again\n");
 		else
@@ -1100,6 +1117,9 @@ static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr,
 		pr_err("%s: Failed to reserve BO\n", __func__);
 		goto release_out;
 	}
+
+	amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, range);
+
 	amdgpu_bo_placement_from_domain(bo, mem->domain);
 	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
 	if (ret)
@@ -1107,7 +1127,7 @@ static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr,
 	amdgpu_bo_unreserve(bo);
 
 release_out:
-	amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm, range);
+	amdgpu_hmm_range_free(range);
 unregister_out:
 	if (ret)
 		amdgpu_hmm_unregister(bo);
@@ -1261,9 +1281,13 @@ static int unmap_bo_from_gpuvm(struct kgd_mem *mem,
 
 	(void)amdgpu_vm_bo_unmap(adev, bo_va, entry->va);
 
+	/* VM entity stopped if process killed, don't clear freed pt bo */
+	if (!amdgpu_vm_ready(vm))
+		return 0;
+
 	(void)amdgpu_vm_clear_freed(adev, vm, &bo_va->last_pt_update);
 
-	(void)amdgpu_sync_fence(sync, bo_va->last_pt_update);
+	(void)amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL);
 
 	return 0;
 }
@@ -1287,7 +1311,7 @@ static int update_gpuvm_pte(struct kgd_mem *mem,
 		return ret;
 	}
 
-	return amdgpu_sync_fence(sync, bo_va->last_pt_update);
+	return amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL);
 }
 
 static int map_bo_to_gpuvm(struct kgd_mem *mem,
@@ -1380,10 +1404,12 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
 		       struct dma_fence **ef)
 {
 	struct amdkfd_process_info *info = NULL;
+	struct kfd_process *process = NULL;
 	int ret;
 
+	process = container_of(process_info, struct kfd_process, kgd_process_info);
 	if (!*process_info) {
-		info = kzalloc(sizeof(*info), GFP_KERNEL);
+		info = kzalloc_obj(*info);
 		if (!info)
 			return -ENOMEM;
 
@@ -1397,21 +1423,26 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
 		info->eviction_fence =
 			amdgpu_amdkfd_fence_create(dma_fence_context_alloc(1),
 						   current->mm,
-						   NULL);
+						   NULL, process->context_id);
 		if (!info->eviction_fence) {
 			pr_err("Failed to create eviction fence\n");
 			ret = -ENOMEM;
 			goto create_evict_fence_fail;
 		}
 
-		info->pid = get_task_pid(current->group_leader, PIDTYPE_PID);
+		info->pid = get_task_pid(current, PIDTYPE_TGID);
 		INIT_DELAYED_WORK(&info->restore_userptr_work,
 				  amdgpu_amdkfd_restore_userptr_worker);
 
+		info->context_id = process->context_id;
+
 		*process_info = info;
 	}
 
-	vm->process_info = *process_info;
+	if (cmpxchg(&vm->process_info, NULL, *process_info) != NULL) {
+		ret = -EINVAL;
+		goto already_acquired;
+	}
 
 	/* Validate page directory and attach eviction fence */
 	ret = amdgpu_bo_reserve(vm->root.bo, true);
@@ -1451,6 +1482,7 @@ validate_pd_fail:
 	amdgpu_bo_unreserve(vm->root.bo);
 reserve_pd_fail:
 	vm->process_info = NULL;
+already_acquired:
 	if (info) {
 		dma_fence_put(&info->eviction_fence->base);
 		*process_info = NULL;
@@ -1529,27 +1561,6 @@ static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo)
 	amdgpu_bo_unreserve(bo);
 }
 
-int amdgpu_amdkfd_gpuvm_set_vm_pasid(struct amdgpu_device *adev,
-				     struct amdgpu_vm *avm, u32 pasid)
-
-{
-	int ret;
-
-	/* Free the original amdgpu allocated pasid,
-	 * will be replaced with kfd allocated pasid.
-	 */
-	if (avm->pasid) {
-		amdgpu_pasid_free(avm->pasid);
-		amdgpu_vm_set_pasid(adev, avm, 0);
-	}
-
-	ret = amdgpu_vm_set_pasid(adev, avm, pasid);
-	if (ret)
-		return ret;
-
-	return 0;
-}
-
 int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,
 					   struct amdgpu_vm *avm,
 					   void **process_info,
@@ -1607,27 +1618,6 @@ void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
 	}
 }
 
-void amdgpu_amdkfd_gpuvm_release_process_vm(struct amdgpu_device *adev,
-					    void *drm_priv)
-{
-	struct amdgpu_vm *avm;
-
-	if (WARN_ON(!adev || !drm_priv))
-		return;
-
-	avm = drm_priv_to_vm(drm_priv);
-
-	pr_debug("Releasing process vm %p\n", avm);
-
-	/* The original pasid of amdgpu vm has already been
-	 * released during making a amdgpu vm to a compute vm
-	 * The current pasid is managed by kfd and will be
-	 * released on kfd process destroy. Set amdgpu pasid
-	 * to 0 to avoid duplicate release.
-	 */
-	amdgpu_vm_release_compute(adev, avm);
-}
-
 uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv)
 {
 	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
@@ -1682,13 +1672,17 @@ size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev,
 	uint64_t vram_available, system_mem_available, ttm_mem_available;
 
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
-	vram_available = KFD_XCP_MEMORY_SIZE(adev, xcp_id)
-		- adev->kfd.vram_used_aligned[xcp_id]
-		- atomic64_read(&adev->vram_pin_size)
-		- reserved_for_pt
-		- reserved_for_ras;
+	if (adev->apu_prefer_gtt && !adev->gmc.is_app_apu)
+		vram_available = KFD_XCP_MEMORY_SIZE(adev, xcp_id)
+			- adev->kfd.vram_used_aligned[xcp_id];
+	else
+		vram_available = KFD_XCP_MEMORY_SIZE(adev, xcp_id)
+			- adev->kfd.vram_used_aligned[xcp_id]
+			- atomic64_read(&adev->vram_pin_size)
+			- reserved_for_pt
+			- reserved_for_ras;
 
-	if (adev->flags & AMD_IS_APU) {
+	if (adev->apu_prefer_gtt) {
 		system_mem_available = no_system_mem_limit ?
 					kfd_mem_limit.max_system_mem_limit :
 					kfd_mem_limit.max_system_mem_limit -
@@ -1736,12 +1730,13 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
 		domain = alloc_domain = AMDGPU_GEM_DOMAIN_VRAM;
 
-		if (adev->flags & AMD_IS_APU) {
+		if (adev->apu_prefer_gtt) {
 			domain = AMDGPU_GEM_DOMAIN_GTT;
 			alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
 			alloc_flags = 0;
 		} else {
-			alloc_flags = AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE;
+			alloc_flags = AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE |
+				AMDGPU_GEM_CREATE_VRAM_CLEARED;
 			alloc_flags |= (flags & KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC) ?
 			AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED : 0;
 
@@ -1783,7 +1778,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED)
 		alloc_flags |= AMDGPU_GEM_CREATE_UNCACHED;
 
-	*mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	*mem = kzalloc_obj(struct kgd_mem);
 	if (!*mem) {
 		ret = -ENOMEM;
 		goto err;
@@ -1919,13 +1914,6 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 
 	mutex_lock(&mem->lock);
 
-	/* Unpin MMIO/DOORBELL BO's that were pinned during allocation */
-	if (mem->alloc_flags &
-	    (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
-	     KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
-		amdgpu_amdkfd_gpuvm_unpin_bo(mem->bo);
-	}
-
 	mapped_to_gpu_memory = mem->mapped_to_gpu_memory;
 	is_imported = mem->is_imported;
 	mutex_unlock(&mem->lock);
@@ -1939,23 +1927,32 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 		return -EBUSY;
 	}
 
+	/* At this point the BO is guaranteed to be freed, so unpin the
+	 * MMIO/DOORBELL BOs that were pinned during allocation.
+	 */
+	if (mem->alloc_flags &
+	    (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+	     KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
+		amdgpu_amdkfd_gpuvm_unpin_bo(mem->bo);
+	}
+
 	/* Make sure restore workers don't access the BO any more */
 	mutex_lock(&process_info->lock);
-	list_del(&mem->validate_list);
+	if (!list_empty(&mem->validate_list))
+		list_del_init(&mem->validate_list);
 	mutex_unlock(&process_info->lock);
 
+	ret = reserve_bo_and_cond_vms(mem, NULL, BO_VM_ALL, &ctx);
+	if (unlikely(ret))
+		return ret;
+
 	/* Cleanup user pages and MMU notifiers */
 	if (amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm)) {
 		amdgpu_hmm_unregister(mem->bo);
-		mutex_lock(&process_info->notifier_lock);
-		amdgpu_ttm_tt_discard_user_pages(mem->bo->tbo.ttm, mem->range);
-		mutex_unlock(&process_info->notifier_lock);
+		amdgpu_hmm_range_free(mem->range);
+		mem->range = NULL;
 	}
 
-	ret = reserve_bo_and_cond_vms(mem, NULL, BO_VM_ALL, &ctx);
-	if (unlikely(ret))
-		return ret;
-
 	amdgpu_amdkfd_remove_eviction_fence(mem->bo,
 					process_info->eviction_fence);
 	pr_debug("Release VA 0x%llx - 0x%llx\n", mem->va,
@@ -1986,9 +1983,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 	 */
 	if (size) {
 		if (!is_imported &&
-		   (mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_VRAM ||
-		   ((adev->flags & AMD_IS_APU) &&
-		    mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT)))
+		   mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM)
 			*size = bo_size;
 		else
 			*size = 0;
@@ -2010,7 +2005,8 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 	drm_gem_object_put(&mem->bo->tbo.base);
 
 	/*
-	 * For kgd_mem allocated in amdgpu_amdkfd_gpuvm_import_dmabuf(),
+	 * For kgd_mem allocated in import_obj_create() via
+	 * amdgpu_amdkfd_gpuvm_import_dmabuf_fd(),
 	 * explicitly free it here.
 	 */
 	if (!use_release_notifier)
@@ -2233,7 +2229,7 @@ int amdgpu_amdkfd_gpuvm_sync_memory(
  * @bo_gart: Return bo reference
  *
  * Before return, bo reference count is incremented. To release the reference and unpin/
- * unmap the BO, call amdgpu_amdkfd_free_gtt_mem.
+ * unmap the BO, call amdgpu_amdkfd_free_kernel_mem.
  */
 int amdgpu_amdkfd_map_gtt_bo_to_gart(struct amdgpu_bo *bo, struct amdgpu_bo **bo_gart)
 {
@@ -2361,10 +2357,9 @@ void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_mem *mem)
 int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
 					  struct kfd_vm_fault_info *mem)
 {
-	if (atomic_read(&adev->gmc.vm_fault_info_updated) == 1) {
+	if (atomic_read_acquire(&adev->gmc.vm_fault_info_updated) == 1) {
 		*mem = *adev->gmc.vm_fault_info;
-		mb(); /* make sure read happened */
-		atomic_set(&adev->gmc.vm_fault_info_updated, 0);
+		atomic_set_release(&adev->gmc.vm_fault_info_updated, 0);
 	}
 	return 0;
 }
@@ -2386,7 +2381,7 @@ static int import_obj_create(struct amdgpu_device *adev,
 		/* Only VRAM and GTT BOs are supported */
 		return -EINVAL;
 
-	*mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	*mem = kzalloc_obj(struct kgd_mem);
 	if (!*mem)
 		return -ENOMEM;
 
@@ -2414,7 +2409,7 @@ static int import_obj_create(struct amdgpu_device *adev,
 	(*mem)->bo = bo;
 	(*mem)->va = va;
 	(*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) &&
-			 !(adev->flags & AMD_IS_APU) ?
+			 !adev->apu_prefer_gtt ?
 			 AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT;
 
 	(*mem)->mapped_to_gpu_memory = 0;
@@ -2575,7 +2570,7 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
 
 		bo = mem->bo;
 
-		amdgpu_ttm_tt_discard_user_pages(bo->tbo.ttm, mem->range);
+		amdgpu_hmm_range_free(mem->range);
 		mem->range = NULL;
 
 		/* BO reservations and getting user pages (hmm_range_fault)
@@ -2599,10 +2594,14 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
 			}
 		}
 
+		mem->range = amdgpu_hmm_range_alloc(NULL);
+		if (unlikely(!mem->range))
+			return -ENOMEM;
 		/* Get updated user pages */
-		ret = amdgpu_ttm_tt_get_user_pages(bo, bo->tbo.ttm->pages,
-						   &mem->range);
+		ret = amdgpu_ttm_tt_get_user_pages(bo, mem->range);
 		if (ret) {
+			amdgpu_hmm_range_free(mem->range);
+			mem->range = NULL;
 			pr_debug("Failed %d to get user pages\n", ret);
 
 			/* Return -EFAULT bad address error as success. It will
@@ -2615,9 +2614,28 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
 			if (ret != -EFAULT)
 				return ret;
 
+			/* If applications unmap memory before destroying the userptr
+			 * from the KFD, trigger a segmentation fault in VM debug mode.
+			 */
+			if (amdgpu_ttm_adev(bo->tbo.bdev)->debug_vm_userptr) {
+				struct kfd_process *p;
+
+				pr_err("Pid %d unmapped memory before destroying userptr at GPU addr 0x%llx\n",
+								pid_nr(process_info->pid), mem->va);
+
+				// Send GPU VM fault to user space
+				p = kfd_lookup_process_by_pid(process_info->pid);
+				if (p) {
+					kfd_signal_vm_fault_event_with_userptr(p, mem->va);
+					kfd_unref_process(p);
+				}
+			}
+
 			ret = 0;
 		}
 
+		amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, mem->range);
+
 		mutex_lock(&process_info->notifier_lock);
 
 		/* Mark the BO as valid unless it was invalidated
@@ -2756,8 +2774,8 @@ static int confirm_valid_user_pages_locked(struct amdkfd_process_info *process_i
 			continue;
 
 		/* Only check mem with hmm range associated */
-		valid = amdgpu_ttm_tt_get_user_pages_done(
-					mem->bo->tbo.ttm, mem->range);
+		valid = amdgpu_hmm_range_valid(mem->range);
+		amdgpu_hmm_range_free(mem->range);
 
 		mem->range = NULL;
 		if (!valid) {
@@ -2969,7 +2987,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu *
 		}
 		dma_resv_for_each_fence(&cursor, bo->tbo.base.resv,
 					DMA_RESV_USAGE_KERNEL, fence) {
-			ret = amdgpu_sync_fence(&sync_obj, fence);
+			ret = amdgpu_sync_fence(&sync_obj, fence, GFP_KERNEL);
 			if (ret) {
 				pr_debug("Memory eviction: Sync BO fence failed. Try again\n");
 				goto validate_map_fail;
@@ -2983,7 +3001,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu *
 	/* Validate PDs, PTs and evicted DMABuf imports last. Otherwise BO
 	 * validations above would invalidate DMABuf imports again.
 	 */
-	ret = process_validate_vms(process_info, &exec.ticket);
+	ret = process_validate_vms(process_info, drm_exec_ticket(&exec));
 	if (ret) {
 		pr_debug("Validating VMs failed, ret: %d\n", ret);
 		goto validate_map_fail;
@@ -3013,9 +3031,22 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu *
 		struct amdgpu_device *adev = amdgpu_ttm_adev(
 			peer_vm->root.bo->tbo.bdev);
 
-		ret = amdgpu_vm_handle_moved(adev, peer_vm, &exec.ticket);
+		struct amdgpu_fpriv *fpriv =
+			container_of(peer_vm, struct amdgpu_fpriv, vm);
+
+		ret = amdgpu_vm_bo_update(adev, fpriv->prt_va, false);
+		if (ret) {
+			dev_dbg(adev->dev,
+				"Memory eviction: handle PRT moved failed, pid %8d. Try again.\n",
+				pid_nr(process_info->pid));
+			goto validate_map_fail;
+		}
+
+		ret = amdgpu_vm_handle_moved(adev, peer_vm, drm_exec_ticket(&exec));
 		if (ret) {
-			pr_debug("Memory eviction: handle moved failed. Try again\n");
+			dev_dbg(adev->dev,
+				"Memory eviction: handle moved failed, pid %8d. Try again.\n",
+				pid_nr(process_info->pid));
 			goto validate_map_fail;
 		}
 	}
@@ -3054,7 +3085,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu *
 			amdgpu_amdkfd_fence_create(
 				process_info->eviction_fence->base.context,
 				process_info->eviction_fence->mm,
-				NULL);
+				NULL, process_info->context_id);
 
 		if (!new_fence) {
 			pr_err("Failed to create eviction fence\n");
@@ -3105,7 +3136,7 @@ int amdgpu_amdkfd_add_gws_to_process(void *info, void *gws, struct kgd_mem **mem
 	if (!info || !gws)
 		return -EINVAL;
 
-	*mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	*mem = kzalloc_obj(struct kgd_mem);
 	if (!*mem)
 		return -ENOMEM;