drm/amdgpu: IOCTL interface for PRT support v4

Till GFX8 we can only enable PRT support globally, but with the next hardware
generation we can do this on a per page basis.

Keep the interface consistent by adding PRT mappings and enable
support globally on current hardware when the first mapping is made.

v2: disable PRT support delayed and on all error paths
v3: PRT and other permissions are mutal exclusive,
    PRT mappings don't need a BO.
v4: update PRT mappings durign CS as well, make va_flags 64bit
Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -701,6 +701,7 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
 
 struct amdgpu_fpriv {
 	struct amdgpu_vm	vm;
+	struct amdgpu_bo_va	*prt_va;
 	struct mutex		bo_list_lock;
 	struct idr		bo_list_handles;
 	struct amdgpu_ctx_mgr	ctx_mgr;

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -759,10 +759,11 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 	amdgpu_bo_unref(&parser->uf_entry.robj);
 }
 
-static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
-				   struct amdgpu_vm *vm)
+static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p)
 {
 	struct amdgpu_device *adev = p->adev;
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_va *bo_va;
 	struct amdgpu_bo *bo;
 	int i, r;
@@ -779,6 +780,15 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 	if (r)
 		return r;
 
+	r = amdgpu_vm_bo_update(adev, fpriv->prt_va, false);
+	if (r)
+		return r;
+
+	r = amdgpu_sync_fence(adev, &p->job->sync,
+			      fpriv->prt_va->last_pt_update);
+	if (r)
+		return r;
+
 	if (amdgpu_sriov_vf(adev)) {
 		struct dma_fence *f;
 		bo_va = vm->csa_bo_va;
@@ -855,7 +865,7 @@ 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->page_directory);
 
-		r = amdgpu_bo_vm_update_pte(p, vm);
+		r = amdgpu_bo_vm_update_pte(p);
 		if (r)
 			return r;
 	}

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c

@@ -553,6 +553,12 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 			  struct drm_file *filp)
 {
+	const uint32_t valid_flags = AMDGPU_VM_DELAY_UPDATE |
+		AMDGPU_VM_PAGE_READABLE | AMDGPU_VM_PAGE_WRITEABLE |
+		AMDGPU_VM_PAGE_EXECUTABLE;
+	const uint32_t prt_flags = AMDGPU_VM_DELAY_UPDATE |
+		AMDGPU_VM_PAGE_PRT;
+
 	struct drm_amdgpu_gem_va *args = data;
 	struct drm_gem_object *gobj;
 	struct amdgpu_device *adev = dev->dev_private;
@@ -563,7 +569,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	struct ttm_validate_buffer tv;
 	struct ww_acquire_ctx ticket;
 	struct list_head list;
-	uint32_t invalid_flags, va_flags = 0;
+	uint64_t va_flags = 0;
 	int r = 0;
 
 	if (!adev->vm_manager.enabled)
@@ -577,11 +583,9 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 		return -EINVAL;
 	}
 
-	invalid_flags = ~(AMDGPU_VM_DELAY_UPDATE | AMDGPU_VM_PAGE_READABLE |
-			AMDGPU_VM_PAGE_WRITEABLE | AMDGPU_VM_PAGE_EXECUTABLE);
-	if ((args->flags & invalid_flags)) {
-		dev_err(&dev->pdev->dev, "invalid flags 0x%08X vs 0x%08X\n",
-			args->flags, invalid_flags);
+	if ((args->flags & ~valid_flags) && (args->flags & ~prt_flags)) {
+		dev_err(&dev->pdev->dev, "invalid flags combination 0x%08X\n",
+			args->flags);
 		return -EINVAL;
 	}
 
@@ -595,28 +599,34 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 		return -EINVAL;
 	}
 
+	INIT_LIST_HEAD(&list);
+	if (!(args->flags & AMDGPU_VM_PAGE_PRT)) {
 		gobj = drm_gem_object_lookup(filp, args->handle);
 		if (gobj == NULL)
 			return -ENOENT;
 		abo = gem_to_amdgpu_bo(gobj);
-	INIT_LIST_HEAD(&list);
 		tv.bo = &abo->tbo;
 		tv.shared = false;
 		list_add(&tv.head, &list);
+	} else {
+		gobj = NULL;
+		abo = NULL;
+	}
 
 	amdgpu_vm_get_pd_bo(&fpriv->vm, &list, &vm_pd);
 
 	r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL);
-	if (r) {
-		drm_gem_object_unreference_unlocked(gobj);
-		return r;
-	}
+	if (r)
+		goto error_unref;
 
+	if (abo) {
 		bo_va = amdgpu_vm_bo_find(&fpriv->vm, abo);
 		if (!bo_va) {
-		ttm_eu_backoff_reservation(&ticket, &list);
-		drm_gem_object_unreference_unlocked(gobj);
-		return -ENOENT;
+			r = -ENOENT;
+			goto error_backoff;
+		}
+	} else {
+		bo_va = fpriv->prt_va;
 	}
 
 	switch (args->operation) {
@@ -627,6 +637,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 			va_flags |= AMDGPU_PTE_WRITEABLE;
 		if (args->flags & AMDGPU_VM_PAGE_EXECUTABLE)
 			va_flags |= AMDGPU_PTE_EXECUTABLE;
+		if (args->flags & AMDGPU_VM_PAGE_PRT)
+			va_flags |= AMDGPU_PTE_PRT;
 		r = amdgpu_vm_bo_map(adev, bo_va, args->va_address,
 				     args->offset_in_bo, args->map_size,
 				     va_flags);
@@ -637,11 +649,13 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	default:
 		break;
 	}
-	if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE) &&
-	    !amdgpu_vm_debug)
+	if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE) && !amdgpu_vm_debug)
 		amdgpu_gem_va_update_vm(adev, bo_va, &list, args->operation);
+
+error_backoff:
 	ttm_eu_backoff_reservation(&ticket, &list);
 
+error_unref:
 	drm_gem_object_unreference_unlocked(gobj);
 	return r;
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c

@@ -655,6 +655,14 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 		goto out_suspend;
 	}
 
+	fpriv->prt_va = amdgpu_vm_bo_add(adev, &fpriv->vm, NULL);
+	if (!fpriv->prt_va) {
+		r = -ENOMEM;
+		amdgpu_vm_fini(adev, &fpriv->vm);
+		kfree(fpriv);
+		goto out_suspend;
+	}
+
 	if (amdgpu_sriov_vf(adev)) {
 		r = amdgpu_map_static_csa(adev, &fpriv->vm);
 		if (r)
@@ -699,6 +707,8 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
 	amdgpu_uvd_free_handles(adev, file_priv);
 	amdgpu_vce_free_handles(adev, file_priv);
 
+	amdgpu_vm_bo_rmv(adev, fpriv->prt_va);
+
 	if (amdgpu_sriov_vf(adev)) {
 		/* TODO: how to handle reserve failure */
 		BUG_ON(amdgpu_bo_reserve(adev->virt.csa_obj, false));

include/uapi/drm/amdgpu_drm.h

@@ -361,6 +361,8 @@ struct drm_amdgpu_gem_op {
 #define AMDGPU_VM_PAGE_WRITEABLE	(1 << 2)
 /* executable mapping, new for VI */
 #define AMDGPU_VM_PAGE_EXECUTABLE	(1 << 3)
+/* partially resident texture */
+#define AMDGPU_VM_PAGE_PRT		(1 << 4)
 
 struct drm_amdgpu_gem_va {
 	/** GEM object handle */