Skip to contentSkip to conversation

drm/amdgpu: Verify root PD is mapped into kernel address space.

PATCH Superseded Andrey Grodzovsky 3 participants 3 messages Jul 4, 2018 – Jul 5, 2018 2 messages not in the archive

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/amdgpu: Verify root PD is mapped into kernel address space.

Problem: When PD/PT update made by CPU root PD was not yet mapped causing page fault.

Fix: Move amdgpu_bo_kmap into amdgpu_vm_bo_base_init to cover all cases and avoid code duplication with amdgpu_vm_alloc_levels.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 47 ++++++++++++++++++++++------------
 1 file changed, 31 insertions(+), 16 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +31 -16
index 845f73a..f546afa 100644
@@ -143,25 +143,36 @@ struct amdgpu_prt_cb {
  * Initialize a bo_va_base structure and add it to the appropriate lists
  *
  */
-static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
+static int amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
 				   struct amdgpu_vm *vm,
 				   struct amdgpu_bo *bo)
 {
+	int r = 0;
 	base->vm = vm;
 	base->bo = bo;
 	INIT_LIST_HEAD(&base->bo_list);
 	INIT_LIST_HEAD(&base->vm_status);
 
 	if (!bo)
-		return;
+		return r;
+
 	list_add_tail(&base->bo_list, &bo->va);
 
+	if (vm->use_cpu_for_update && bo->tbo.type == ttm_bo_type_kernel) {
+		r = amdgpu_bo_kmap(bo, NULL);
+		if (r) {
+			amdgpu_bo_unref(&bo->shadow);
+			amdgpu_bo_unref(&bo);
+			return r;
+		}
+	}
+
 	if (bo->tbo.resv != vm->root.base.bo->tbo.resv)
-		return;
+		return r;
 
 	if (bo->preferred_domains &
 	    amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type))
-		return;
+		return r;
 
 	/*
 	 * we checked all the prerequisites, but it looks like this per vm bo
@@ -169,6 +180,8 @@ static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
 	 * is validated on next vm use to avoid fault.
 	 * */
 	list_move_tail(&base->vm_status, &vm->evicted);
+
+	return r;
 }
 
 /**
@@ -525,21 +538,15 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
 				return r;
 			}
 
-			if (vm->use_cpu_for_update) {
-				r = amdgpu_bo_kmap(pt, NULL);
-				if (r) {
-					amdgpu_bo_unref(&pt->shadow);
-					amdgpu_bo_unref(&pt);
-					return r;
-				}
-			}
-
 			/* Keep a reference to the root directory to avoid
 			* freeing them up in the wrong order.
 			*/
 			pt->parent = amdgpu_bo_ref(parent->base.bo);
 
-			amdgpu_vm_bo_base_init(&entry->base, vm, pt);
+			r = amdgpu_vm_bo_base_init(&entry->base, vm, pt);
+			if (r)
+				return r;
+
 			list_move(&entry->base.vm_status, &vm->relocated);
 		}
 
@@ -1992,12 +1999,17 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
 				      struct amdgpu_bo *bo)
 {
 	struct amdgpu_bo_va *bo_va;
+	int r;
 
 	bo_va = kzalloc(sizeof(struct amdgpu_bo_va), GFP_KERNEL);
 	if (bo_va == NULL) {
 		return NULL;
 	}
-	amdgpu_vm_bo_base_init(&bo_va->base, vm, bo);
+
+	if ((r = amdgpu_vm_bo_base_init(&bo_va->base, vm, bo))) {
+		WARN_ONCE(1,"r = %d\n", r);
+		return NULL;
+	}
 
 	bo_va->ref_count = 1;
 	INIT_LIST_HEAD(&bo_va->valids);
@@ -2613,7 +2625,10 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	if (r)
 		goto error_unreserve;
 
-	amdgpu_vm_bo_base_init(&vm->root.base, vm, root);
+	r = amdgpu_vm_bo_base_init(&vm->root.base, vm, root);
+	if (r)
+			goto error_unreserve;
+
 	amdgpu_bo_unreserve(vm->root.base.bo);
 
 	if (pasid) {
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1530733765-25535-1-git-send-email-andrey.grodzovsky-5C7GfCeVMHo@public.gmane.org

#2 lore raw
Quoted 32 lines · Andrey Grodzovsky

Problem: When PD/PT update made by CPU root PD was not yet mapped causing page fault.

Fix: Move amdgpu_bo_kmap into amdgpu_vm_bo_base_init to cover all cases and avoid code duplication with amdgpu_vm_alloc_levels.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 47 ++++++++++++++++++++++------------
  1 file changed, 31 insertions(+), 16 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2 -1
index 845f73a..f546afa 100644
@@ -143,25 +143,36 @@ struct amdgpu_prt_cb {
   * Initialize a bo_va_base structure and add it to the appropriate lists
   *
   */
-static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
+static int amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
  				   struct amdgpu_vm *vm,
  				   struct amdgpu_bo *bo)
  {
+	int r = 0;
  	base->vm = vm;
  	base->bo = bo;
  	INIT_LIST_HEAD(&base->bo_list);
  	INIT_LIST_HEAD(&base->vm_status);
 
  	if (!bo)

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_prt_cb { · line 165

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +7 -1
-		return;
+		return r;
+
  	list_add_tail(&base->bo_list, &bo->va);
 
+	if (vm->use_cpu_for_update && bo->tbo.type == ttm_bo_type_kernel) {
+		r = amdgpu_bo_kmap(bo, NULL);
+		if (r) {
+			amdgpu_bo_unref(&bo->shadow);
+			amdgpu_bo_unref(&bo);

I feel these two lines should move out of helper function, ref/unref should appear in where used with pair.

Regards,
David Zhou

Quoted 71 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +18 -13
+			return r;
+		}
+	}
+
  	if (bo->tbo.resv != vm->root.base.bo->tbo.resv)
-		return;
+		return r;
 
  	if (bo->preferred_domains &
  	    amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type))
-		return;
+		return r;
 
  	/*
  	 * we checked all the prerequisites, but it looks like this per vm bo
@@ -169,6 +180,8 @@ static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
  	 * is validated on next vm use to avoid fault.
  	 * */
  	list_move_tail(&base->vm_status, &vm->evicted);
+
+	return r;
  }
 
  /**
@@ -525,21 +538,15 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
  				return r;
  			}
 
-			if (vm->use_cpu_for_update) {
-				r = amdgpu_bo_kmap(pt, NULL);
-				if (r) {
-					amdgpu_bo_unref(&pt->shadow);
-					amdgpu_bo_unref(&pt);
-					return r;
-				}
-			}
-
  			/* Keep a reference to the root directory to avoid
  			* freeing them up in the wrong order.
  			*/
  			pt->parent = amdgpu_bo_ref(parent->base.bo);
 
-			amdgpu_vm_bo_base_init(&entry->base, vm, pt);
+			r = amdgpu_vm_bo_base_init(&entry->base, vm, pt);
+			if (r)
+				return r;
+
  			list_move(&entry->base.vm_status, &vm->relocated);
  		}
 
@@ -1992,12 +1999,17 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
  				      struct amdgpu_bo *bo)
  {
  	struct amdgpu_bo_va *bo_va;
+	int r;
 
  	bo_va = kzalloc(sizeof(struct amdgpu_bo_va), GFP_KERNEL);
  	if (bo_va == NULL) {
  		return NULL;
  	}
-	amdgpu_vm_bo_base_init(&bo_va->base, vm, bo);
+
+	if ((r = amdgpu_vm_bo_base_init(&bo_va->base, vm, bo))) {
+		WARN_ONCE(1,"r = %d\n", r);
+		return NULL;
+	}
 
  	bo_va->ref_count = 1;
  	INIT_LIST_HEAD(&bo_va->valids);
@@ -2613,7 +2625,10 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
  	if (r)

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, · line 2634

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -1
  		goto error_unreserve;
 
-	amdgpu_vm_bo_base_init(&vm->root.base, vm, root);
+	r = amdgpu_vm_bo_base_init(&vm->root.base, vm, root);
+	if (r)
+			goto error_unreserve;
+
  	amdgpu_bo_unreserve(vm->root.base.bo);
 
  	if (pasid) {

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 25f954f3-f726-5c93-c5c6-ca267e0db7df-5C7GfCeVMHo@public.gmane.org

#4 lore raw

Am 05.07.2018 um 04:09 schrieb zhoucm1:

Quoted 37 lines · Andrey Grodzovsky
Quoted 34 lines

Problem: When PD/PT update made by CPU root PD was not yet mapped causing page fault.

Fix: Move amdgpu_bo_kmap into amdgpu_vm_bo_base_init to cover all cases and avoid code duplication with amdgpu_vm_alloc_levels.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 47
++++++++++++++++++++++------------

  1 file changed, 31 insertions(+), 16 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3 -2
index 845f73a..f546afa 100644
@@ -143,25 +143,36 @@ struct amdgpu_prt_cb {
    * Initialize a bo_va_base structure and add it to the appropriate lists
    *
    */
-static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
+static int amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
                      struct amdgpu_vm *vm,
                      struct amdgpu_bo *bo)
   {
+    int r = 0;
       base->vm = vm;
       base->bo = bo;
       INIT_LIST_HEAD(&base->bo_list);
       INIT_LIST_HEAD(&base->vm_status);
         if (!bo)
-        return;
+        return r;
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_prt_cb { · line 165

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +5 -0
+
       list_add_tail(&base->bo_list, &bo->va);
   +    if (vm->use_cpu_for_update && bo->tbo.type == ttm_bo_type_kernel) {
+        r = amdgpu_bo_kmap(bo, NULL);
+        if (r) {
+            amdgpu_bo_unref(&bo->shadow);
+            amdgpu_bo_unref(&bo);

I feel these two lines should move out of helper function, ref/unref should appear in where used with pair.

Yeah agree, but there is an approach which is even cleaner I think.

Quoted 15 lines

Regards,
David Zhou

Quoted 12 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +5 -1
+            return r;
+        }
+    }
+
       if (bo->tbo.resv != vm->root.base.bo->tbo.resv)
-        return;
+        return r;

        if (bo->preferred_domains & amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type)) -        return;
+        return r;
        /*

Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c static void amdgpu_vm_bo_base_init(struct · line 185

Quoted 8 lines

       * we checked all the prerequisites, but it looks like this per vm bo

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2 -0
@@ -169,6 +180,8 @@ static void amdgpu_vm_bo_base_init(struct 
 amdgpu_vm_bo_base *base,
        * is validated on next vm use to avoid fault.
        * */
       list_move_tail(&base->vm_status, &vm->evicted);
+
+    return r;

Here we move it into the evicted status when the placement isn't ok.

Now for PDs/PTs the caller moves it immediately into the relocated status to make sure that the parent entries are updated:

Quoted 2 lines

amdgpu_vm_bo_base_init(&entry->base, vm, pt); list_move(&entry->base.vm_status, &vm->relocated);

And amdgpu_vm_update_directories() maps all relocated BOs anyway before it begins:

Quoted 5 lines

                list_for_each_entry(bo_base, &vm->relocated, vm_status) { r = amdgpu_bo_kmap(bo_base->bo, NULL); if (unlikely(r))
                                return r;
                }

So all we need to do is to make sure that we add the root PD to the relocated list as well.

Cleanest approach would probably be to move the "list_move(&entry->base.vm_status, &vm->relocated);" into amdgpu_vm_bo_base_init().

Regards,
Christian.

Quoted 46 lines
Quoted 46 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -8
   }
     /**
@@ -525,21 +538,15 @@ static int amdgpu_vm_alloc_levels(struct 
 amdgpu_device *adev,
                   return r;
               }
   -            if (vm->use_cpu_for_update) {
-                r = amdgpu_bo_kmap(pt, NULL);
-                if (r) {
-                    amdgpu_bo_unref(&pt->shadow);
-                    amdgpu_bo_unref(&pt);
-                    return r;
-                }
-            }
-
               /* Keep a reference to the root directory to avoid
               * freeing them up in the wrong order.
               */
               pt->parent = amdgpu_bo_ref(parent->base.bo);
   -            amdgpu_vm_bo_base_init(&entry->base, vm, pt);
+            r = amdgpu_vm_bo_base_init(&entry->base, vm, pt);
+            if (r)
+                return r;
+
               list_move(&entry->base.vm_status, &vm->relocated);
           }

  @@ -1992,12 +1999,17 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, struct amdgpu_bo *bo)
  {
      struct amdgpu_bo_va *bo_va;
+    int r;
        bo_va = kzalloc(sizeof(struct amdgpu_bo_va), GFP_KERNEL);
      if (bo_va == NULL) {
          return NULL;
      }
-    amdgpu_vm_bo_base_init(&bo_va->base, vm, bo);
+
+    if ((r = amdgpu_vm_bo_base_init(&bo_va->base, vm, bo))) {
+        WARN_ONCE(1,"r = %d\n", r);
+        return NULL;
+    }
        bo_va->ref_count = 1;
      INIT_LIST_HEAD(&bo_va->valids);

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0 -0
@@ -2613,7 +2625,10 @@ int amdgpu_vm_init(struct amdgpu_device *adev, 
 struct amdgpu_vm *vm,
       if (r)
Quoted 9 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, · line 2634

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -0
           goto error_unreserve;
   -    amdgpu_vm_bo_base_init(&vm->root.base, vm, root);
+    r = amdgpu_vm_bo_base_init(&vm->root.base, vm, root);
+    if (r)
+            goto error_unreserve;
+
       amdgpu_bo_unreserve(vm->root.base.bo);
         if (pasid) {

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx