1 files changed, 69 insertions, 98 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 078c068937fe..6b15cad78de9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -88,6 +88,46 @@ struct amdgpu_prt_cb {
 	struct dma_fence_cb cb;
 };
 
+/**
+ * amdgpu_vm_set_pasid - manage pasid and vm ptr mapping
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: amdgpu_vm pointer
+ * @pasid: the pasid the VM is using on this GPU
+ *
+ * Set the pasid this VM is using on this GPU, can also be used to remove the
+ * pasid by passing in zero.
+ *
+ */
+int amdgpu_vm_set_pasid(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			u32 pasid)
+{
+	int r;
+
+	if (vm->pasid == pasid)
+		return 0;
+
+	if (vm->pasid) {
+		r = xa_err(xa_erase_irq(&adev->vm_manager.pasids, vm->pasid));
+		if (r < 0)
+			return r;
+
+		vm->pasid = 0;
+	}
+
+	if (pasid) {
+		r = xa_err(xa_store_irq(&adev->vm_manager.pasids, pasid, vm,
+					GFP_KERNEL));
+		if (r < 0)
+			return r;
+
+		vm->pasid = pasid;
+	}
+
+
+	return 0;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -886,7 +926,7 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
 	bp.size = amdgpu_vm_bo_size(adev, level);
 	bp.byte_align = AMDGPU_GPU_PAGE_SIZE;
 	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
-	bp.domain = amdgpu_bo_get_preferred_pin_domain(adev, bp.domain);
+	bp.domain = amdgpu_bo_get_preferred_domain(adev, bp.domain);
 	bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
 		AMDGPU_GEM_CREATE_CPU_GTT_USWC;
 
@@ -1178,7 +1218,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 		amdgpu_gmc_emit_pasid_mapping(ring, job->vmid, job->pasid);
 
 	if (vm_flush_needed || pasid_mapping_needed) {
-		r = amdgpu_fence_emit(ring, &fence, 0);
+		r = amdgpu_fence_emit(ring, &fence, NULL, 0);
 		if (r)
 			return r;
 	}
@@ -1758,7 +1798,7 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	r = vm->update_funcs->commit(&params, fence);
 
 	if (table_freed)
-		*table_freed = params.table_freed;
+		*table_freed = *table_freed || params.table_freed;
 
 error_unlock:
 	amdgpu_vm_eviction_unlock(vm);
@@ -1816,6 +1856,7 @@ void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem,
  * @adev: amdgpu_device pointer
  * @bo_va: requested BO and VM object
  * @clear: if true clear the entries
+ * @table_freed: return true if page table is freed
  *
  * Fill in the page table entries for @bo_va.
  *
@@ -1823,7 +1864,7 @@ void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem,
  * 0 for success, -EINVAL for failure.
  */
 int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
-			bool clear)
+			bool clear, bool *table_freed)
 {
 	struct amdgpu_bo *bo = bo_va->base.bo;
 	struct amdgpu_vm *vm = bo_va->base.vm;
@@ -1902,7 +1943,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
 						resv, mapping->start,
 						mapping->last, update_flags,
 						mapping->offset, mem,
-						pages_addr, last_update, NULL);
+						pages_addr, last_update, table_freed);
 		if (r)
 			return r;
 	}
@@ -2154,7 +2195,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 
 	list_for_each_entry_safe(bo_va, tmp, &vm->moved, base.vm_status) {
 		/* Per VM BOs never need to bo cleared in the page tables */
-		r = amdgpu_vm_bo_update(adev, bo_va, false);
+		r = amdgpu_vm_bo_update(adev, bo_va, false, NULL);
 		if (r)
 			return r;
 	}
@@ -2173,7 +2214,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 		else
 			clear = true;
 
-		r = amdgpu_vm_bo_update(adev, bo_va, clear);
+		r = amdgpu_vm_bo_update(adev, bo_va, clear, NULL);
 		if (r)
 			return r;
 
@@ -2863,14 +2904,13 @@ long amdgpu_vm_wait_idle(struct amdgpu_vm *vm, long timeout)
  *
  * @adev: amdgpu_device pointer
  * @vm: requested vm
- * @pasid: Process address space identifier
  *
  * Init @vm fields.
  *
  * Returns:
  * 0 for success, error for failure.
  */
-int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, u32 pasid)
+int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
 	struct amdgpu_bo *root_bo;
 	struct amdgpu_bo_vm *root;
@@ -2944,19 +2984,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, u32 pasid)
 
 	amdgpu_bo_unreserve(vm->root.bo);
 
-	if (pasid) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
-		r = idr_alloc(&adev->vm_manager.pasid_idr, vm, pasid, pasid + 1,
-			      GFP_ATOMIC);
-		spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
-		if (r < 0)
-			goto error_free_root;
-
-		vm->pasid = pasid;
-	}
-
 	INIT_KFIFO(vm->faults);
 
 	return 0;
@@ -3012,7 +3039,6 @@ static int amdgpu_vm_check_clean_reserved(struct amdgpu_device *adev,
  *
  * @adev: amdgpu_device pointer
  * @vm: requested vm
- * @pasid: pasid to use
  *
  * This only works on GFX VMs that don't have any BOs added and no
  * page tables allocated yet.
@@ -3020,7 +3046,6 @@ static int amdgpu_vm_check_clean_reserved(struct amdgpu_device *adev,
  * Changes the following VM parameters:
  * - use_cpu_for_update
  * - pte_supports_ats
- * - pasid (old PASID is released, because compute manages its own PASIDs)
  *
  * Reinitializes the page directory to reflect the changed ATS
  * setting.
@@ -3028,8 +3053,7 @@ static int amdgpu_vm_check_clean_reserved(struct amdgpu_device *adev,
  * Returns:
  * 0 for success, -errno for errors.
  */
-int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-			   u32 pasid)
+int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
 	bool pte_support_ats = (adev->asic_type == CHIP_RAVEN);
 	int r;
@@ -3043,19 +3067,6 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	if (r)
 		goto unreserve_bo;
 
-	if (pasid) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
-		r = idr_alloc(&adev->vm_manager.pasid_idr, vm, pasid, pasid + 1,
-			      GFP_ATOMIC);
-		spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
-
-		if (r == -ENOSPC)
-			goto unreserve_bo;
-		r = 0;
-	}
-
 	/* Check if PD needs to be reinitialized and do it before
 	 * changing any other state, in case it fails.
 	 */
@@ -3065,7 +3076,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 				       to_amdgpu_bo_vm(vm->root.bo),
 				       false);
 		if (r)
-			goto free_idr;
+			goto unreserve_bo;
 	}
 
 	/* Update VM state */
@@ -3082,7 +3093,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		r = amdgpu_bo_sync_wait(vm->root.bo,
 					AMDGPU_FENCE_OWNER_UNDEFINED, true);
 		if (r)
-			goto free_idr;
+			goto unreserve_bo;
 
 		vm->update_funcs = &amdgpu_vm_cpu_funcs;
 	} else {
@@ -3092,36 +3103,11 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	vm->last_update = NULL;
 	vm->is_compute_context = true;
 
-	if (vm->pasid) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
-		idr_remove(&adev->vm_manager.pasid_idr, vm->pasid);
-		spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
-
-		/* Free the original amdgpu allocated pasid
-		 * Will be replaced with kfd allocated pasid
-		 */
-		amdgpu_pasid_free(vm->pasid);
-		vm->pasid = 0;
-	}
-
 	/* Free the shadow bo for compute VM */
 	amdgpu_bo_unref(&to_amdgpu_bo_vm(vm->root.bo)->shadow);
 
-	if (pasid)
-		vm->pasid = pasid;
-
 	goto unreserve_bo;
 
-free_idr:
-	if (pasid) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
-		idr_remove(&adev->vm_manager.pasid_idr, pasid);
-		spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
-	}
 unreserve_bo:
 	amdgpu_bo_unreserve(vm->root.bo);
 	return r;
@@ -3137,14 +3123,7 @@ unreserve_bo:
  */
 void amdgpu_vm_release_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
-	if (vm->pasid) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
-		idr_remove(&adev->vm_manager.pasid_idr, vm->pasid);
-		spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
-	}
-	vm->pasid = 0;
+	amdgpu_vm_set_pasid(adev, vm, 0);
 	vm->is_compute_context = false;
 }
 
@@ -3168,15 +3147,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 
 	root = amdgpu_bo_ref(vm->root.bo);
 	amdgpu_bo_reserve(root, true);
-	if (vm->pasid) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
-		idr_remove(&adev->vm_manager.pasid_idr, vm->pasid);
-		spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
-		vm->pasid = 0;
-	}
-
+	amdgpu_vm_set_pasid(adev, vm, 0);
 	dma_fence_wait(vm->last_unlocked, false);
 	dma_fence_put(vm->last_unlocked);
 
@@ -3258,8 +3229,7 @@ void amdgpu_vm_manager_init(struct amdgpu_device *adev)
 	adev->vm_manager.vm_update_mode = 0;
 #endif
 
-	idr_init(&adev->vm_manager.pasid_idr);
-	spin_lock_init(&adev->vm_manager.pasid_lock);
+	xa_init_flags(&adev->vm_manager.pasids, XA_FLAGS_LOCK_IRQ);
 }
 
 /**
@@ -3271,8 +3241,8 @@ void amdgpu_vm_manager_init(struct amdgpu_device *adev)
  */
 void amdgpu_vm_manager_fini(struct amdgpu_device *adev)
 {
-	WARN_ON(!idr_is_empty(&adev->vm_manager.pasid_idr));
-	idr_destroy(&adev->vm_manager.pasid_idr);
+	WARN_ON(!xa_empty(&adev->vm_manager.pasids));
+	xa_destroy(&adev->vm_manager.pasids);
 
 	amdgpu_vmid_mgr_fini(adev);
 }
@@ -3341,13 +3311,13 @@ void amdgpu_vm_get_task_info(struct amdgpu_device *adev, u32 pasid,
 	struct amdgpu_vm *vm;
 	unsigned long flags;
 
-	spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
+	xa_lock_irqsave(&adev->vm_manager.pasids, flags);
 
-	vm = idr_find(&adev->vm_manager.pasid_idr, pasid);
+	vm = xa_load(&adev->vm_manager.pasids, pasid);
 	if (vm)
 		*task_info = vm->task_info;
 
-	spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
+	xa_unlock_irqrestore(&adev->vm_manager.pasids, flags);
 }
 
 /**
@@ -3375,12 +3345,13 @@ void amdgpu_vm_set_task_info(struct amdgpu_vm *vm)
  * @adev: amdgpu device pointer
  * @pasid: PASID of the VM
  * @addr: Address of the fault
+ * @write_fault: true is write fault, false is read fault
  *
  * Try to gracefully handle a VM fault. Return true if the fault was handled and
  * shouldn't be reported any more.
  */
 bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
-			    uint64_t addr)
+			    uint64_t addr, bool write_fault)
 {
 	bool is_compute_context = false;
 	struct amdgpu_bo *root;
@@ -3389,15 +3360,15 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
 	struct amdgpu_vm *vm;
 	int r;
 
-	spin_lock_irqsave(&adev->vm_manager.pasid_lock, irqflags);
-	vm = idr_find(&adev->vm_manager.pasid_idr, pasid);
+	xa_lock_irqsave(&adev->vm_manager.pasids, irqflags);
+	vm = xa_load(&adev->vm_manager.pasids, pasid);
 	if (vm) {
 		root = amdgpu_bo_ref(vm->root.bo);
 		is_compute_context = vm->is_compute_context;
 	} else {
 		root = NULL;
 	}
-	spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, irqflags);
+	xa_unlock_irqrestore(&adev->vm_manager.pasids, irqflags);
 
 	if (!root)
 		return false;
@@ -3405,7 +3376,7 @@ 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, addr)) {
+	    !svm_range_restore_pages(adev, pasid, addr, write_fault)) {
 		amdgpu_bo_unref(&root);
 		return true;
 	}
@@ -3415,11 +3386,11 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
 		goto error_unref;
 
 	/* Double check that the VM still exists */
-	spin_lock_irqsave(&adev->vm_manager.pasid_lock, irqflags);
-	vm = idr_find(&adev->vm_manager.pasid_idr, pasid);
+	xa_lock_irqsave(&adev->vm_manager.pasids, irqflags);
+	vm = xa_load(&adev->vm_manager.pasids, pasid);
 	if (vm && vm->root.bo != root)
 		vm = NULL;
-	spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, irqflags);
+	xa_unlock_irqrestore(&adev->vm_manager.pasids, irqflags);
 	if (!vm)
 		goto error_unlock;