drm/vmwgfx: Abstract placement selection

Problem with explicit placement selection in vmwgfx is that by the time
the buffer object needs to be validated the information about which
placement was supposed to be used is lost. To workaround this the driver
had a bunch of state in various places e.g. as_mob or cpu_blit to
somehow convey the information on which placement was intended.

Fix it properly by allowing the buffer objects to hold their preferred
placement so it can be reused whenever needed. This makes the entire
validation pipeline a lot easier both to understand and maintain.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Martin Krastev <krastevm@vmware.com>
Reviewed-by: Maaz Mombasawala <mombasawalam@vmware.com>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20230131033542.953249-8-zack@kde.org

Author: zackr

drivers/gpu/drm/vmwgfx/vmwgfx_bo.c

@@ -135,11 +135,17 @@ int vmw_bo_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
 		goto out_unreserve;
 	}
 
-	ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, &ctx);
+	vmw_bo_placement_set(buf,
+			     VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
+			     VMW_BO_DOMAIN_GMR);
+	ret = ttm_bo_validate(bo, &buf->placement, &ctx);
 	if (likely(ret == 0) || ret == -ERESTARTSYS)
 		goto out_unreserve;
 
-	ret = ttm_bo_validate(bo, &vmw_vram_placement, &ctx);
+	vmw_bo_placement_set(buf,
+			     VMW_BO_DOMAIN_VRAM,
+			     VMW_BO_DOMAIN_VRAM);
+	ret = ttm_bo_validate(bo, &buf->placement, &ctx);
 
 out_unreserve:
 	if (!ret)
@@ -190,17 +196,8 @@ int vmw_bo_pin_in_start_of_vram(struct vmw_private *dev_priv,
 {
 	struct ttm_operation_ctx ctx = {interruptible, false };
 	struct ttm_buffer_object *bo = &buf->base;
-	struct ttm_placement placement;
-	struct ttm_place place;
 	int ret = 0;
 
-	place = vmw_vram_placement.placement[0];
-	place.lpfn = bo->resource->num_pages;
-	placement.num_placement = 1;
-	placement.placement = &place;
-	placement.num_busy_placement = 1;
-	placement.busy_placement = &place;
-
 	vmw_execbuf_release_pinned_bo(dev_priv);
 	ret = ttm_bo_reserve(bo, interruptible, false, NULL);
 	if (unlikely(ret != 0))
@@ -216,14 +213,21 @@ int vmw_bo_pin_in_start_of_vram(struct vmw_private *dev_priv,
 	    bo->resource->start > 0 &&
 	    buf->base.pin_count == 0) {
 		ctx.interruptible = false;
-		(void) ttm_bo_validate(bo, &vmw_sys_placement, &ctx);
+		vmw_bo_placement_set(buf,
+				     VMW_BO_DOMAIN_SYS,
+				     VMW_BO_DOMAIN_SYS);
+		(void)ttm_bo_validate(bo, &buf->placement, &ctx);
 	}
 
+	vmw_bo_placement_set(buf,
+			     VMW_BO_DOMAIN_VRAM,
+			     VMW_BO_DOMAIN_VRAM);
+	buf->places[0].lpfn = bo->resource->num_pages;
 	if (buf->base.pin_count > 0)
-		ret = ttm_resource_compat(bo->resource, &placement)
+		ret = ttm_resource_compat(bo->resource, &buf->placement)
 			? 0 : -EINVAL;
 	else
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		ret = ttm_bo_validate(bo, &buf->placement, &ctx);
 
 	/* For some reason we didn't end up at the start of vram */
 	WARN_ON(ret == 0 && bo->resource->start != 0);
@@ -431,7 +435,7 @@ int vmw_bo_create_kernel(struct vmw_private *dev_priv, unsigned long size,
 }
 
 int vmw_bo_create(struct vmw_private *vmw,
-		  size_t size, struct ttm_placement *placement,
+		  size_t size, u32 domain, u32 busy_domain,
 		  bool interruptible, bool pin,
 		  struct vmw_bo **p_bo)
 {
@@ -444,7 +448,8 @@ int vmw_bo_create(struct vmw_private *vmw,
 	}
 
 	ret = vmw_bo_init(vmw, *p_bo, size,
-			  placement, interruptible, pin);
+			  domain, busy_domain,
+			  interruptible, pin);
 	if (unlikely(ret != 0))
 		goto out_error;
 
@@ -461,7 +466,8 @@ int vmw_bo_create(struct vmw_private *vmw,
  * @dev_priv: Pointer to the device private struct
  * @vmw_bo: Pointer to the struct vmw_bo to initialize.
  * @size: Buffer object size in bytes.
- * @placement: Initial placement.
+ * @domain: Domain to put the bo in.
+ * @busy_domain: Domain to put the bo if busy.
  * @interruptible: Whether waits should be performed interruptible.
  * @pin: If the BO should be created pinned at a fixed location.
  * Returns: Zero on success, negative error code on error.
@@ -470,7 +476,9 @@ int vmw_bo_create(struct vmw_private *vmw,
  */
 int vmw_bo_init(struct vmw_private *dev_priv,
 		struct vmw_bo *vmw_bo,
-		size_t size, struct ttm_placement *placement,
+		size_t size,
+		u32 domain,
+		u32 busy_domain,
 		bool interruptible, bool pin)
 {
 	struct ttm_operation_ctx ctx = {
@@ -489,9 +497,10 @@ int vmw_bo_init(struct vmw_private *dev_priv,
 	size = ALIGN(size, PAGE_SIZE);
 	drm_gem_private_object_init(vdev, &vmw_bo->base.base, size);
 
+	vmw_bo_placement_set(vmw_bo, domain, busy_domain);
 	ret = ttm_bo_init_reserved(bdev, &vmw_bo->base, size,
 				   ttm_bo_type_device,
-				   placement,
+				   &vmw_bo->placement,
 				   0, &ctx, NULL, NULL, vmw_bo_free);
 	if (unlikely(ret)) {
 		return ret;
@@ -815,3 +824,101 @@ void vmw_bo_move_notify(struct ttm_buffer_object *bo,
 	if (mem->mem_type != VMW_PL_MOB && bo->resource->mem_type == VMW_PL_MOB)
 		vmw_resource_unbind_list(vbo);
 }
+
+static u32
+set_placement_list(struct ttm_place *pl, u32 domain)
+{
+	u32 n = 0;
+
+	/*
+	 * The placements are ordered according to our preferences
+	 */
+	if (domain & VMW_BO_DOMAIN_MOB) {
+		pl[n].mem_type = VMW_PL_MOB;
+		pl[n].flags = 0;
+		pl[n].fpfn = 0;
+		pl[n].lpfn = 0;
+		n++;
+	}
+	if (domain & VMW_BO_DOMAIN_GMR) {
+		pl[n].mem_type = VMW_PL_GMR;
+		pl[n].flags = 0;
+		pl[n].fpfn = 0;
+		pl[n].lpfn = 0;
+		n++;
+	}
+	if (domain & VMW_BO_DOMAIN_VRAM) {
+		pl[n].mem_type = TTM_PL_VRAM;
+		pl[n].flags = 0;
+		pl[n].fpfn = 0;
+		pl[n].lpfn = 0;
+		n++;
+	}
+	WARN_ON((domain & VMW_BO_DOMAIN_WAITABLE_SYS) != 0);
+	if (domain & VMW_BO_DOMAIN_WAITABLE_SYS) {
+		pl[n].mem_type = VMW_PL_SYSTEM;
+		pl[n].flags = 0;
+		pl[n].fpfn = 0;
+		pl[n].lpfn = 0;
+		n++;
+	}
+	if (domain & VMW_BO_DOMAIN_SYS) {
+		pl[n].mem_type = TTM_PL_SYSTEM;
+		pl[n].flags = 0;
+		pl[n].fpfn = 0;
+		pl[n].lpfn = 0;
+		n++;
+	}
+
+	WARN_ON(!n);
+	if (!n) {
+		pl[n].mem_type = TTM_PL_SYSTEM;
+		pl[n].flags = 0;
+		pl[n].fpfn = 0;
+		pl[n].lpfn = 0;
+		n++;
+	}
+	return n;
+}
+
+void vmw_bo_placement_set(struct vmw_bo *bo, u32 domain, u32 busy_domain)
+{
+	struct ttm_device *bdev = bo->base.bdev;
+	struct vmw_private *vmw =
+		container_of(bdev, struct vmw_private, bdev);
+	struct ttm_placement *pl = &bo->placement;
+	bool mem_compatible = false;
+	u32 i;
+
+	pl->placement = bo->places;
+	pl->num_placement = set_placement_list(bo->places, domain);
+
+	if (drm_debug_enabled(DRM_UT_DRIVER) && bo->base.resource) {
+		for (i = 0; i < pl->num_placement; ++i) {
+			if (bo->base.resource->mem_type == TTM_PL_SYSTEM ||
+			    bo->base.resource->mem_type == pl->placement[i].mem_type)
+				mem_compatible = true;
+		}
+		if (!mem_compatible)
+			drm_warn(&vmw->drm,
+				 "%s: Incompatible transition from "
+				 "bo->base.resource->mem_type = %u to domain = %u\n",
+				 __func__, bo->base.resource->mem_type, domain);
+	}
+
+	pl->busy_placement = bo->busy_places;
+	pl->num_busy_placement = set_placement_list(bo->busy_places, busy_domain);
+}
+
+void vmw_bo_placement_set_default_accelerated(struct vmw_bo *bo)
+{
+	struct ttm_device *bdev = bo->base.bdev;
+	struct vmw_private *vmw =
+		container_of(bdev, struct vmw_private, bdev);
+	u32 domain = VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM;
+
+	if (vmw->has_mob)
+		domain = VMW_BO_DOMAIN_MOB;
+
+	vmw_bo_placement_set(bo, domain, domain);
+}

drivers/gpu/drm/vmwgfx/vmwgfx_bo.h

@@ -31,6 +31,7 @@
 #include "device_include/svga_reg.h"
 
 #include <drm/ttm/ttm_bo.h>
+#include <drm/ttm/ttm_placement.h>
 
 #include <linux/rbtree_types.h>
 #include <linux/types.h>
@@ -40,6 +41,14 @@ struct vmw_fence_obj;
 struct vmw_private;
 struct vmw_resource;
 
+enum vmw_bo_domain {
+	VMW_BO_DOMAIN_SYS           = BIT(0),
+	VMW_BO_DOMAIN_WAITABLE_SYS  = BIT(1),
+	VMW_BO_DOMAIN_VRAM          = BIT(2),
+	VMW_BO_DOMAIN_GMR           = BIT(3),
+	VMW_BO_DOMAIN_MOB           = BIT(4),
+};
+
 /**
  * struct vmw_bo - TTM buffer object with vmwgfx additions
  * @base: The TTM buffer object
@@ -54,6 +63,11 @@ struct vmw_resource;
  */
 struct vmw_bo {
 	struct ttm_buffer_object base;
+
+	struct ttm_placement placement;
+	struct ttm_place places[5];
+	struct ttm_place busy_places[5];
+
 	struct rb_root res_tree;
 	/* For KMS atomic helpers: ttm bo mapping count */
 	atomic_t base_mapped_count;
@@ -67,17 +81,24 @@ struct vmw_bo {
 	struct vmw_bo_dirty *dirty;
 };
 
+void vmw_bo_placement_set(struct vmw_bo *bo, u32 domain, u32 busy_domain);
+void vmw_bo_placement_set_default_accelerated(struct vmw_bo *bo);
+
 int vmw_bo_create_kernel(struct vmw_private *dev_priv,
 			 unsigned long size,
 			 struct ttm_placement *placement,
 			 struct ttm_buffer_object **p_bo);
 int vmw_bo_create(struct vmw_private *dev_priv,
-		  size_t size, struct ttm_placement *placement,
+		  size_t size,
+		  u32 domain,
+		  u32 busy_domain,
 		  bool interruptible, bool pin,
 		  struct vmw_bo **p_bo);
 int vmw_bo_init(struct vmw_private *dev_priv,
 		struct vmw_bo *vmw_bo,
-		size_t size, struct ttm_placement *placement,
+		size_t size,
+		u32 domain,
+		u32 busy_domain,
 		bool interruptible, bool pin);
 int vmw_bo_unref_ioctl(struct drm_device *dev, void *data,
 		       struct drm_file *file_priv);

drivers/gpu/drm/vmwgfx/vmwgfx_context.c

@@ -76,7 +76,8 @@ static const struct vmw_res_func vmw_legacy_context_func = {
 	.needs_backup = false,
 	.may_evict = false,
 	.type_name = "legacy contexts",
-	.backup_placement = NULL,
+	.domain = VMW_BO_DOMAIN_SYS,
+	.busy_domain = VMW_BO_DOMAIN_SYS,
 	.create = NULL,
 	.destroy = NULL,
 	.bind = NULL,
@@ -90,7 +91,8 @@ static const struct vmw_res_func vmw_gb_context_func = {
 	.prio = 3,
 	.dirty_prio = 3,
 	.type_name = "guest backed contexts",
-	.backup_placement = &vmw_mob_placement,
+	.domain = VMW_BO_DOMAIN_MOB,
+	.busy_domain = VMW_BO_DOMAIN_MOB,
 	.create = vmw_gb_context_create,
 	.destroy = vmw_gb_context_destroy,
 	.bind = vmw_gb_context_bind,
@@ -104,7 +106,8 @@ static const struct vmw_res_func vmw_dx_context_func = {
 	.prio = 3,
 	.dirty_prio = 3,
 	.type_name = "dx contexts",
-	.backup_placement = &vmw_mob_placement,
+	.domain = VMW_BO_DOMAIN_MOB,
+	.busy_domain = VMW_BO_DOMAIN_MOB,
 	.create = vmw_dx_context_create,
 	.destroy = vmw_dx_context_destroy,
 	.bind = vmw_dx_context_bind,

drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c

@@ -123,7 +123,8 @@ static const struct vmw_res_func vmw_cotable_func = {
 	.prio = 3,
 	.dirty_prio = 3,
 	.type_name = "context guest backed object tables",
-	.backup_placement = &vmw_mob_placement,
+	.domain = VMW_BO_DOMAIN_MOB,
+	.busy_domain = VMW_BO_DOMAIN_MOB,
 	.create = vmw_cotable_create,
 	.destroy = vmw_cotable_destroy,
 	.bind = vmw_cotable_bind,
@@ -408,7 +409,8 @@ static int vmw_cotable_resize(struct vmw_resource *res, size_t new_size)
 	 * for the new COTable. Initially pin the buffer object to make sure
 	 * we can use tryreserve without failure.
 	 */
-	ret = vmw_bo_create(dev_priv, new_size, &vmw_mob_placement,
+	ret = vmw_bo_create(dev_priv, new_size,
+			    VMW_BO_DOMAIN_MOB, VMW_BO_DOMAIN_MOB,
 			    true, true, &buf);
 	if (ret) {
 		DRM_ERROR("Failed initializing new cotable MOB.\n");
@@ -449,7 +451,10 @@ static int vmw_cotable_resize(struct vmw_resource *res, size_t new_size)
 	}
 
 	/* Unpin new buffer, and switch backup buffers. */
-	ret = ttm_bo_validate(bo, &vmw_mob_placement, &ctx);
+	vmw_bo_placement_set(buf,
+			     VMW_BO_DOMAIN_MOB,
+			     VMW_BO_DOMAIN_MOB);
+	ret = ttm_bo_validate(bo, &buf->placement, &ctx);
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("Failed validating new COTable backup buffer.\n");
 		goto out_wait;

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c

@@ -401,7 +401,8 @@ static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 	 * user of the bo currently.
 	 */
 	ret = vmw_bo_create(dev_priv, PAGE_SIZE,
-			    &vmw_sys_placement, false, true, &vbo);
+			    VMW_BO_DOMAIN_SYS, VMW_BO_DOMAIN_SYS,
+			    false, true, &vbo);
 	if (unlikely(ret != 0))
 		return ret;
 

drivers/gpu/drm/vmwgfx/vmwgfx_drv.h

@@ -927,9 +927,7 @@ extern struct ttm_placement vmw_vram_placement;
 extern struct ttm_placement vmw_vram_sys_placement;
 extern struct ttm_placement vmw_vram_gmr_placement;
 extern struct ttm_placement vmw_sys_placement;
-extern struct ttm_placement vmw_srf_placement;
 extern struct ttm_placement vmw_mob_placement;
-extern struct ttm_placement vmw_nonfixed_placement;
 extern struct ttm_device_funcs vmw_bo_driver;
 extern const struct vmw_sg_table *
 vmw_bo_sg_table(struct ttm_buffer_object *bo);