drm/vmwgfx: Abstract placement selection
authorZack Rusin <zackr@vmware.com>
Tue, 31 Jan 2023 03:35:41 +0000 (22:35 -0500)
committerZack Rusin <zackr@vmware.com>
Tue, 14 Feb 2023 03:37:55 +0000 (22:37 -0500)
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
22 files changed:
drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
drivers/gpu/drm/vmwgfx/vmwgfx_bo.h
drivers/gpu/drm/vmwgfx/vmwgfx_context.c
drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
drivers/gpu/drm/vmwgfx/vmwgfx_gem.c
drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
drivers/gpu/drm/vmwgfx/vmwgfx_resource_priv.h
drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
drivers/gpu/drm/vmwgfx/vmwgfx_so.c
drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
drivers/gpu/drm/vmwgfx/vmwgfx_streamoutput.c
drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
drivers/gpu/drm/vmwgfx/vmwgfx_va.c
drivers/gpu/drm/vmwgfx/vmwgfx_validation.c
drivers/gpu/drm/vmwgfx/vmwgfx_validation.h

index 4beaaf4..dbee64c 100644 (file)
@@ -124,11 +124,17 @@ int vmw_bo_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
        if (unlikely(ret != 0))
                goto err;
 
-       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)
@@ -179,17 +185,8 @@ 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 = PFN_UP(bo->resource->size);
-       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))
@@ -205,10 +202,17 @@ 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);
        }
 
-       ret = ttm_bo_validate(bo, &placement, &ctx);
+       vmw_bo_placement_set(buf,
+                            VMW_BO_DOMAIN_VRAM,
+                            VMW_BO_DOMAIN_VRAM);
+       buf->places[0].lpfn = PFN_UP(bo->resource->size);
+       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);
@@ -416,7 +420,7 @@ error_free:
 }
 
 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)
 {
@@ -429,7 +433,8 @@ 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;
 
@@ -446,7 +451,8 @@ out_error:
  * @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.
@@ -455,7 +461,9 @@ out_error:
  */
 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 = {
@@ -474,8 +482,9 @@ 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, ttm_bo_type_device,
-                                  placement, 0, &ctx, NULL, NULL, vmw_bo_free);
+                                  &vmw_bo->placement, 0, &ctx, NULL, NULL, vmw_bo_free);
        if (unlikely(ret)) {
                return ret;
        }
@@ -810,3 +819,101 @@ 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);
+}
index b874888..f146e8e 100644 (file)
@@ -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
@@ -53,6 +62,11 @@ 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;
 
        atomic_t cpu_writers;
@@ -64,17 +78,24 @@ 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);
index cc02be6..84ef5f0 100644 (file)
@@ -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,
index 9193faa..d49db81 100644 (file)
@@ -136,7 +136,8 @@ 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,
@@ -424,7 +425,8 @@ 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");
@@ -465,7 +467,10 @@ 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;
index d6d90fb..d6300aa 100644 (file)
@@ -398,7 +398,8 @@ 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;
 
index 7eb3787..f903f78 100644 (file)
@@ -932,9 +932,7 @@ 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);
index f269a66..d135db9 100644 (file)
@@ -478,9 +478,13 @@ static int vmw_resource_context_res_add(struct vmw_private *dev_priv,
                struct vmw_bo *dx_query_mob;
 
                dx_query_mob = vmw_context_get_dx_query_mob(ctx);
-               if (dx_query_mob)
+               if (dx_query_mob) {
+                       vmw_bo_placement_set(dx_query_mob,
+                                            VMW_BO_DOMAIN_MOB,
+                                            VMW_BO_DOMAIN_MOB);
                        ret = vmw_validation_add_bo(sw_context->ctx,
-                                                   dx_query_mob, true, false);
+                                                   dx_query_mob);
+               }
        }
 
        mutex_unlock(&dev_priv->binding_mutex);
@@ -1036,17 +1040,17 @@ static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv,
 
                if (unlikely(sw_context->cur_query_bo != NULL)) {
                        sw_context->needs_post_query_barrier = true;
+                       vmw_bo_placement_set_default_accelerated(sw_context->cur_query_bo);
                        ret = vmw_validation_add_bo(sw_context->ctx,
-                                                   sw_context->cur_query_bo,
-                                                   dev_priv->has_mob, false);
+                                                   sw_context->cur_query_bo);
                        if (unlikely(ret != 0))
                                return ret;
                }
                sw_context->cur_query_bo = new_query_bo;
 
+               vmw_bo_placement_set_default_accelerated(dev_priv->dummy_query_bo);
                ret = vmw_validation_add_bo(sw_context->ctx,
-                                           dev_priv->dummy_query_bo,
-                                           dev_priv->has_mob, false);
+                                           dev_priv->dummy_query_bo);
                if (unlikely(ret != 0))
                        return ret;
        }
@@ -1158,7 +1162,8 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
                drm_dbg(&dev_priv->drm, "Could not find or use MOB buffer.\n");
                return PTR_ERR(vmw_bo);
        }
-       ret = vmw_validation_add_bo(sw_context->ctx, vmw_bo, true, false);
+       vmw_bo_placement_set(vmw_bo, VMW_BO_DOMAIN_MOB, VMW_BO_DOMAIN_MOB);
+       ret = vmw_validation_add_bo(sw_context->ctx, vmw_bo);
        ttm_bo_put(&vmw_bo->base);
        if (unlikely(ret != 0))
                return ret;
@@ -1212,7 +1217,9 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
                drm_dbg(&dev_priv->drm, "Could not find or use GMR region.\n");
                return PTR_ERR(vmw_bo);
        }
-       ret = vmw_validation_add_bo(sw_context->ctx, vmw_bo, false, false);
+       vmw_bo_placement_set(vmw_bo, VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
+                            VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM);
+       ret = vmw_validation_add_bo(sw_context->ctx, vmw_bo);
        ttm_bo_put(&vmw_bo->base);
        if (unlikely(ret != 0))
                return ret;
@@ -4362,13 +4369,17 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
        if (dev_priv->pinned_bo == NULL)
                goto out_unlock;
 
-       ret = vmw_validation_add_bo(&val_ctx, dev_priv->pinned_bo, false,
-                                   false);
+       vmw_bo_placement_set(dev_priv->pinned_bo,
+                            VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
+                            VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM);
+       ret = vmw_validation_add_bo(&val_ctx, dev_priv->pinned_bo);
        if (ret)
                goto out_no_reserve;
 
-       ret = vmw_validation_add_bo(&val_ctx, dev_priv->dummy_query_bo, false,
-                                   false);
+       vmw_bo_placement_set(dev_priv->dummy_query_bo,
+                            VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
+                            VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM);
+       ret = vmw_validation_add_bo(&val_ctx, dev_priv->dummy_query_bo);
        if (ret)
                goto out_no_reserve;
 
index c7ebcd4..5f38357 100644 (file)
@@ -121,9 +121,8 @@ int vmw_gem_object_create_with_handle(struct vmw_private *dev_priv,
        int ret;
 
        ret = vmw_bo_create(dev_priv, size,
-                           (dev_priv->has_mob) ?
-                                   &vmw_sys_placement :
-                                   &vmw_vram_sys_placement,
+                           (dev_priv->has_mob) ? VMW_BO_DOMAIN_SYS : VMW_BO_DOMAIN_VRAM,
+                           VMW_BO_DOMAIN_SYS,
                            true, false, p_vbo);
 
        (*p_vbo)->base.base.funcs = &vmw_gem_object_funcs;
index e83286e..d975c0a 100644 (file)
@@ -1270,9 +1270,9 @@ int vmw_kms_readback(struct vmw_private *dev_priv,
                                            user_fence_rep, vclips, num_clips,
                                            NULL);
        case vmw_du_screen_target:
-               return vmw_kms_stdu_dma(dev_priv, file_priv, vfb,
-                                       user_fence_rep, NULL, vclips, num_clips,
-                                       1, false, true, NULL);
+               return vmw_kms_stdu_readback(dev_priv, file_priv, vfb,
+                                            user_fence_rep, NULL, vclips, num_clips,
+                                            1, NULL);
        default:
                WARN_ONCE(true,
                          "Readback called with invalid display system.\n");
@@ -2999,8 +2999,20 @@ int vmw_du_helper_plane_update(struct vmw_du_update_plane *update)
                struct vmw_framebuffer_bo *vfbbo =
                        container_of(update->vfb, typeof(*vfbbo), base);
 
-               ret = vmw_validation_add_bo(&val_ctx, vfbbo->buffer, false,
-                                           update->cpu_blit);
+               /*
+                * For screen targets we want a mappable bo, for everything else we want
+                * accelerated i.e. host backed (vram or gmr) bo. If the display unit
+                * is not screen target then mob's shouldn't be available.
+                */
+               if (update->dev_priv->active_display_unit == vmw_du_screen_target) {
+                       vmw_bo_placement_set(vfbbo->buffer,
+                                            VMW_BO_DOMAIN_SYS | VMW_BO_DOMAIN_MOB | VMW_BO_DOMAIN_GMR,
+                                            VMW_BO_DOMAIN_SYS | VMW_BO_DOMAIN_MOB | VMW_BO_DOMAIN_GMR);
+               } else {
+                       WARN_ON(update->dev_priv->has_mob);
+                       vmw_bo_placement_set_default_accelerated(vfbbo->buffer);
+               }
+               ret = vmw_validation_add_bo(&val_ctx, vfbbo->buffer);
        } else {
                struct vmw_framebuffer_surface *vfbs =
                        container_of(update->vfb, typeof(*vfbs), base);
index 7a97e53..1bdf601 100644 (file)
@@ -126,7 +126,6 @@ struct vmw_du_update_plane {
        struct vmw_framebuffer *vfb;
        struct vmw_fence_obj **out_fence;
        struct mutex *mutex;
-       bool cpu_blit;
        bool intr;
 };
 
@@ -564,17 +563,15 @@ int vmw_kms_stdu_surface_dirty(struct vmw_private *dev_priv,
                               unsigned num_clips, int inc,
                               struct vmw_fence_obj **out_fence,
                               struct drm_crtc *crtc);
-int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
-                    struct drm_file *file_priv,
-                    struct vmw_framebuffer *vfb,
-                    struct drm_vmw_fence_rep __user *user_fence_rep,
-                    struct drm_clip_rect *clips,
-                    struct drm_vmw_rect *vclips,
-                    uint32_t num_clips,
-                    int increment,
-                    bool to_surface,
-                    bool interruptible,
-                    struct drm_crtc *crtc);
+int vmw_kms_stdu_readback(struct vmw_private *dev_priv,
+                         struct drm_file *file_priv,
+                         struct vmw_framebuffer *vfb,
+                         struct drm_vmw_fence_rep __user *user_fence_rep,
+                         struct drm_clip_rect *clips,
+                         struct drm_vmw_rect *vclips,
+                         uint32_t num_clips,
+                         int increment,
+                         struct drm_crtc *crtc);
 
 int vmw_du_helper_plane_update(struct vmw_du_update_plane *update);
 
index 54e412f..6780a36 100644 (file)
@@ -332,7 +332,7 @@ static int vmw_resource_buf_alloc(struct vmw_resource *res,
        }
 
        ret = vmw_bo_create(res->dev_priv, res->backup_size,
-                           res->func->backup_placement,
+                           res->func->domain, res->func->busy_domain,
                            interruptible, false, &backup);
        if (unlikely(ret != 0))
                goto out_no_bo;
@@ -529,8 +529,10 @@ vmw_resource_check_buffer(struct ww_acquire_ctx *ticket,
                return 0;
 
        backup_dirty = res->backup_dirty;
+       vmw_bo_placement_set(res->backup, res->func->domain,
+                            res->func->busy_domain);
        ret = ttm_bo_validate(&res->backup->base,
-                             res->func->backup_placement,
+                             &res->backup->placement,
                              &ctx);
 
        if (unlikely(ret != 0))
@@ -968,9 +970,12 @@ int vmw_resource_pin(struct vmw_resource *res, bool interruptible)
                        if (ret)
                                goto out_no_validate;
                        if (!vbo->base.pin_count) {
+                               vmw_bo_placement_set(vbo,
+                                                    res->func->domain,
+                                                    res->func->busy_domain);
                                ret = ttm_bo_validate
                                        (&vbo->base,
-                                        res->func->backup_placement,
+                                        &vbo->placement,
                                         &ctx);
                                if (ret) {
                                        ttm_bo_unreserve(&vbo->base);
index 3b7438b..2c24e09 100644 (file)
@@ -83,7 +83,8 @@ struct vmw_res_func {
        enum vmw_res_type res_type;
        bool needs_backup;
        const char *type_name;
-       struct ttm_placement *backup_placement;
+       u32 domain;
+       u32 busy_domain;
        bool may_evict;
        u32 prio;
        u32 dirty_prio;
index a04897f..e9d03ef 100644 (file)
@@ -445,7 +445,8 @@ vmw_sou_primary_plane_prepare_fb(struct drm_plane *plane,
         */
        vmw_overlay_pause_all(dev_priv);
        ret = vmw_bo_create(dev_priv, size,
-                           &vmw_vram_placement,
+                           VMW_BO_DOMAIN_VRAM,
+                           VMW_BO_DOMAIN_VRAM,
                            false, true, &vps->bo);
        vmw_overlay_resume_all(dev_priv);
        if (ret) {
@@ -547,7 +548,6 @@ static int vmw_sou_plane_update_bo(struct vmw_private *dev_priv,
        bo_update.base.vfb = vfb;
        bo_update.base.out_fence = out_fence;
        bo_update.base.mutex = NULL;
-       bo_update.base.cpu_blit = false;
        bo_update.base.intr = true;
 
        bo_update.base.calc_fifo_size = vmw_sou_bo_fifo_size;
@@ -708,7 +708,6 @@ static int vmw_sou_plane_update_surface(struct vmw_private *dev_priv,
        srf_update.base.vfb = vfb;
        srf_update.base.out_fence = out_fence;
        srf_update.base.mutex = &dev_priv->cmdbuf_mutex;
-       srf_update.base.cpu_blit = false;
        srf_update.base.intr = true;
 
        srf_update.base.calc_fifo_size = vmw_sou_surface_fifo_size;
@@ -1224,7 +1223,9 @@ int vmw_kms_sou_do_bo_dirty(struct vmw_private *dev_priv,
        DECLARE_VAL_CONTEXT(val_ctx, NULL, 0);
        int ret;
 
-       ret = vmw_validation_add_bo(&val_ctx, buf, false, false);
+       vmw_bo_placement_set(buf, VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
+                            VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM);
+       ret = vmw_validation_add_bo(&val_ctx, buf);
        if (ret)
                return ret;
 
@@ -1330,7 +1331,9 @@ int vmw_kms_sou_readback(struct vmw_private *dev_priv,
        DECLARE_VAL_CONTEXT(val_ctx, NULL, 0);
        int ret;
 
-       ret = vmw_validation_add_bo(&val_ctx, buf, false, false);
+       vmw_bo_placement_set(buf, VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
+                            VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM);
+       ret = vmw_validation_add_bo(&val_ctx, buf);
        if (ret)
                return ret;
 
index b186d09..9920c10 100644 (file)
@@ -94,7 +94,8 @@ static const struct vmw_res_func vmw_gb_shader_func = {
        .prio = 3,
        .dirty_prio = 3,
        .type_name = "guest backed shaders",
-       .backup_placement = &vmw_mob_placement,
+       .domain = VMW_BO_DOMAIN_MOB,
+       .busy_domain = VMW_BO_DOMAIN_MOB,
        .create = vmw_gb_shader_create,
        .destroy = vmw_gb_shader_destroy,
        .bind = vmw_gb_shader_bind,
@@ -108,7 +109,8 @@ static const struct vmw_res_func vmw_dx_shader_func = {
        .prio = 3,
        .dirty_prio = 3,
        .type_name = "dx shaders",
-       .backup_placement = &vmw_mob_placement,
+       .domain = VMW_BO_DOMAIN_MOB,
+       .busy_domain = VMW_BO_DOMAIN_MOB,
        .create = vmw_dx_shader_create,
        /*
         * The destroy callback is only called with a committed resource on
@@ -893,7 +895,9 @@ int vmw_compat_shader_add(struct vmw_private *dev_priv,
        if (!vmw_shader_id_ok(user_key, shader_type))
                return -EINVAL;
 
-       ret = vmw_bo_create(dev_priv, size, &vmw_sys_placement,
+       ret = vmw_bo_create(dev_priv, size,
+                           VMW_BO_DOMAIN_SYS,
+                           VMW_BO_DOMAIN_SYS,
                            true, true, &buf);
        if (unlikely(ret != 0))
                goto out;
@@ -913,7 +917,10 @@ int vmw_compat_shader_add(struct vmw_private *dev_priv,
        WARN_ON(is_iomem);
 
        ttm_bo_kunmap(&map);
-       ret = ttm_bo_validate(&buf->base, &vmw_sys_placement, &ctx);
+       vmw_bo_placement_set(buf,
+                            VMW_BO_DOMAIN_SYS,
+                            VMW_BO_DOMAIN_SYS);
+       ret = ttm_bo_validate(&buf->base, &buf->placement, &ctx);
        WARN_ON(ret != 0);
        ttm_bo_unreserve(&buf->base);
 
index 4ea32b0..603175b 100644 (file)
@@ -24,6 +24,7 @@
  *
  **************************************************************************/
 
+#include "vmwgfx_bo.h"
 #include "vmwgfx_drv.h"
 #include "vmwgfx_resource_priv.h"
 #include "vmwgfx_so.h"
@@ -84,7 +85,8 @@ static const struct vmw_res_func vmw_view_func = {
        .needs_backup = false,
        .may_evict = false,
        .type_name = "DX view",
-       .backup_placement = NULL,
+       .domain = VMW_BO_DOMAIN_SYS,
+       .busy_domain = VMW_BO_DOMAIN_SYS,
        .create = vmw_view_create,
        .commit_notify = vmw_view_commit_notify,
 };
index 4745537..0df8640 100644 (file)
@@ -66,7 +66,6 @@ enum stdu_content_type {
  */
 struct vmw_stdu_dirty {
        struct vmw_kms_dirty base;
-       SVGA3dTransferType  transfer;
        s32 left, right, top, bottom;
        s32 fb_left, fb_top;
        u32 pitch;
@@ -137,12 +136,6 @@ static void vmw_stdu_destroy(struct vmw_screen_target_display_unit *stdu);
  * Screen Target Display Unit CRTC Functions
  *****************************************************************************/
 
-static bool vmw_stdu_use_cpu_blit(const struct vmw_private *vmw)
-{
-       return !(vmw->capabilities & SVGA_CAP_3D) || vmw->vram_size < (32 * 1024 * 1024);
-}
-
-
 /**
  * vmw_stdu_crtc_destroy - cleans up the STDU
  *
@@ -452,93 +445,6 @@ static void vmw_stdu_crtc_atomic_disable(struct drm_crtc *crtc,
 }
 
 /**
- * vmw_stdu_bo_clip - Callback to encode a suface DMA command cliprect
- *
- * @dirty: The closure structure.
- *
- * Encodes a surface DMA command cliprect and updates the bounding box
- * for the DMA.
- */
-static void vmw_stdu_bo_clip(struct vmw_kms_dirty *dirty)
-{
-       struct vmw_stdu_dirty *ddirty =
-               container_of(dirty, struct vmw_stdu_dirty, base);
-       struct vmw_stdu_dma *cmd = dirty->cmd;
-       struct SVGA3dCopyBox *blit = (struct SVGA3dCopyBox *) &cmd[1];
-
-       blit += dirty->num_hits;
-       blit->srcx = dirty->fb_x;
-       blit->srcy = dirty->fb_y;
-       blit->x = dirty->unit_x1;
-       blit->y = dirty->unit_y1;
-       blit->d = 1;
-       blit->w = dirty->unit_x2 - dirty->unit_x1;
-       blit->h = dirty->unit_y2 - dirty->unit_y1;
-       dirty->num_hits++;
-
-       if (ddirty->transfer != SVGA3D_WRITE_HOST_VRAM)
-               return;
-
-       /* Destination bounding box */
-       ddirty->left = min_t(s32, ddirty->left, dirty->unit_x1);
-       ddirty->top = min_t(s32, ddirty->top, dirty->unit_y1);
-       ddirty->right = max_t(s32, ddirty->right, dirty->unit_x2);
-       ddirty->bottom = max_t(s32, ddirty->bottom, dirty->unit_y2);
-}
-
-/**
- * vmw_stdu_bo_fifo_commit - Callback to fill in and submit a DMA command.
- *
- * @dirty: The closure structure.
- *
- * Fills in the missing fields in a DMA command, and optionally encodes
- * a screen target update command, depending on transfer direction.
- */
-static void vmw_stdu_bo_fifo_commit(struct vmw_kms_dirty *dirty)
-{
-       struct vmw_stdu_dirty *ddirty =
-               container_of(dirty, struct vmw_stdu_dirty, base);
-       struct vmw_screen_target_display_unit *stdu =
-               container_of(dirty->unit, typeof(*stdu), base);
-       struct vmw_stdu_dma *cmd = dirty->cmd;
-       struct SVGA3dCopyBox *blit = (struct SVGA3dCopyBox *) &cmd[1];
-       SVGA3dCmdSurfaceDMASuffix *suffix =
-               (SVGA3dCmdSurfaceDMASuffix *) &blit[dirty->num_hits];
-       size_t blit_size = sizeof(*blit) * dirty->num_hits + sizeof(*suffix);
-
-       if (!dirty->num_hits) {
-               vmw_cmd_commit(dirty->dev_priv, 0);
-               return;
-       }
-
-       cmd->header.id = SVGA_3D_CMD_SURFACE_DMA;
-       cmd->header.size = sizeof(cmd->body) + blit_size;
-       vmw_bo_get_guest_ptr(&ddirty->buf->base, &cmd->body.guest.ptr);
-       cmd->body.guest.pitch = ddirty->pitch;
-       cmd->body.host.sid = stdu->display_srf->res.id;
-       cmd->body.host.face = 0;
-       cmd->body.host.mipmap = 0;
-       cmd->body.transfer = ddirty->transfer;
-       suffix->suffixSize = sizeof(*suffix);
-       suffix->maximumOffset = ddirty->buf->base.base.size;
-
-       if (ddirty->transfer == SVGA3D_WRITE_HOST_VRAM) {
-               blit_size += sizeof(struct vmw_stdu_update);
-
-               vmw_stdu_populate_update(&suffix[1], stdu->base.unit,
-                                        ddirty->left, ddirty->right,
-                                        ddirty->top, ddirty->bottom);
-       }
-
-       vmw_cmd_commit(dirty->dev_priv, sizeof(*cmd) + blit_size);
-
-       stdu->display_srf->res.res_dirty = true;
-       ddirty->left = ddirty->top = S32_MAX;
-       ddirty->right = ddirty->bottom = S32_MIN;
-}
-
-
-/**
  * vmw_stdu_bo_cpu_clip - Callback to encode a CPU blit
  *
  * @dirty: The closure structure.
@@ -598,62 +504,21 @@ static void vmw_stdu_bo_cpu_commit(struct vmw_kms_dirty *dirty)
                return;
 
        /* Assume we are blitting from Guest (bo) to Host (display_srf) */
-       dst_pitch = stdu->display_srf->metadata.base_size.width * stdu->cpp;
-       dst_bo = &stdu->display_srf->res.backup->base;
-       dst_offset = ddirty->top * dst_pitch + ddirty->left * stdu->cpp;
-
-       src_pitch = ddirty->pitch;
-       src_bo = &ddirty->buf->base;
-       src_offset = ddirty->fb_top * src_pitch + ddirty->fb_left * stdu->cpp;
+       src_pitch = stdu->display_srf->metadata.base_size.width * stdu->cpp;
+       src_bo = &stdu->display_srf->res.backup->base;
+       src_offset = ddirty->top * dst_pitch + ddirty->left * stdu->cpp;
 
-       /* Swap src and dst if the assumption was wrong. */
-       if (ddirty->transfer != SVGA3D_WRITE_HOST_VRAM) {
-               swap(dst_pitch, src_pitch);
-               swap(dst_bo, src_bo);
-               swap(src_offset, dst_offset);
-       }
+       dst_pitch = ddirty->pitch;
+       dst_bo = &ddirty->buf->base;
+       dst_offset = ddirty->fb_top * src_pitch + ddirty->fb_left * stdu->cpp;
 
        (void) vmw_bo_cpu_blit(dst_bo, dst_offset, dst_pitch,
                               src_bo, src_offset, src_pitch,
                               width * stdu->cpp, height, &diff);
-
-       if (ddirty->transfer == SVGA3D_WRITE_HOST_VRAM &&
-           drm_rect_visible(&diff.rect)) {
-               struct vmw_private *dev_priv;
-               struct vmw_stdu_update *cmd;
-               struct drm_clip_rect region;
-               int ret;
-
-               /* We are updating the actual surface, not a proxy */
-               region.x1 = diff.rect.x1;
-               region.x2 = diff.rect.x2;
-               region.y1 = diff.rect.y1;
-               region.y2 = diff.rect.y2;
-               ret = vmw_kms_update_proxy(&stdu->display_srf->res, &region,
-                                          1, 1);
-               if (ret)
-                       goto out_cleanup;
-
-
-               dev_priv = vmw_priv(stdu->base.crtc.dev);
-               cmd = VMW_CMD_RESERVE(dev_priv, sizeof(*cmd));
-               if (!cmd)
-                       goto out_cleanup;
-
-               vmw_stdu_populate_update(cmd, stdu->base.unit,
-                                        region.x1, region.x2,
-                                        region.y1, region.y2);
-
-               vmw_cmd_commit(dev_priv, sizeof(*cmd));
-       }
-
-out_cleanup:
-       ddirty->left = ddirty->top = ddirty->fb_left = ddirty->fb_top = S32_MAX;
-       ddirty->right = ddirty->bottom = S32_MIN;
 }
 
 /**
- * vmw_kms_stdu_dma - Perform a DMA transfer between a buffer-object backed
+ * vmw_kms_stdu_readback - Perform a readback from a buffer-object backed
  * framebuffer and the screen target system.
  *
  * @dev_priv: Pointer to the device private structure.
@@ -666,9 +531,6 @@ out_cleanup:
  * be NULL.
  * @num_clips: Number of clip rects in @clips or @vclips.
  * @increment: Increment to use when looping over @clips or @vclips.
- * @to_surface: Whether to DMA to the screen target system as opposed to
- * from the screen target system.
- * @interruptible: Whether to perform waits interruptible if possible.
  * @crtc: If crtc is passed, perform stdu dma on that crtc only.
  *
  * If DMA-ing till the screen target system, the function will also notify
@@ -677,59 +539,49 @@ out_cleanup:
  * Returns 0 on success, negative error code on failure. -ERESTARTSYS if
  * interrupted.
  */
-int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
-                    struct drm_file *file_priv,
-                    struct vmw_framebuffer *vfb,
-                    struct drm_vmw_fence_rep __user *user_fence_rep,
-                    struct drm_clip_rect *clips,
-                    struct drm_vmw_rect *vclips,
-                    uint32_t num_clips,
-                    int increment,
-                    bool to_surface,
-                    bool interruptible,
-                    struct drm_crtc *crtc)
+int vmw_kms_stdu_readback(struct vmw_private *dev_priv,
+                         struct drm_file *file_priv,
+                         struct vmw_framebuffer *vfb,
+                         struct drm_vmw_fence_rep __user *user_fence_rep,
+                         struct drm_clip_rect *clips,
+                         struct drm_vmw_rect *vclips,
+                         uint32_t num_clips,
+                         int increment,
+                         struct drm_crtc *crtc)
 {
        struct vmw_bo *buf =
                container_of(vfb, struct vmw_framebuffer_bo, base)->buffer;
        struct vmw_stdu_dirty ddirty;
        int ret;
-       bool cpu_blit = vmw_stdu_use_cpu_blit(dev_priv);
        DECLARE_VAL_CONTEXT(val_ctx, NULL, 0);
 
        /*
-        * VMs without 3D support don't have the surface DMA command and
-        * we'll be using a CPU blit, and the framebuffer should be moved out
-        * of VRAM.
+        * The GMR domain might seem confusing because it might seem like it should
+        * never happen with screen targets but e.g. the xorg vmware driver issues
+        * CMD_SURFACE_DMA for various pixmap updates which might transition our bo to
+        * a GMR. Instead of forcing another transition we can optimize the readback
+        * by reading directly from the GMR.
         */
-       ret = vmw_validation_add_bo(&val_ctx, buf, false, cpu_blit);
+       vmw_bo_placement_set(buf,
+                            VMW_BO_DOMAIN_MOB | VMW_BO_DOMAIN_SYS | VMW_BO_DOMAIN_GMR,
+                            VMW_BO_DOMAIN_MOB | VMW_BO_DOMAIN_SYS | VMW_BO_DOMAIN_GMR);
+       ret = vmw_validation_add_bo(&val_ctx, buf);
        if (ret)
                return ret;
 
-       ret = vmw_validation_prepare(&val_ctx, NULL, interruptible);
+       ret = vmw_validation_prepare(&val_ctx, NULL, true);
        if (ret)
                goto out_unref;
 
-       ddirty.transfer = (to_surface) ? SVGA3D_WRITE_HOST_VRAM :
-               SVGA3D_READ_HOST_VRAM;
        ddirty.left = ddirty.top = S32_MAX;
        ddirty.right = ddirty.bottom = S32_MIN;
        ddirty.fb_left = ddirty.fb_top = S32_MAX;
        ddirty.pitch = vfb->base.pitches[0];
        ddirty.buf = buf;
-       ddirty.base.fifo_commit = vmw_stdu_bo_fifo_commit;
-       ddirty.base.clip = vmw_stdu_bo_clip;
-       ddirty.base.fifo_reserve_size = sizeof(struct vmw_stdu_dma) +
-               num_clips * sizeof(SVGA3dCopyBox) +
-               sizeof(SVGA3dCmdSurfaceDMASuffix);
-       if (to_surface)
-               ddirty.base.fifo_reserve_size += sizeof(struct vmw_stdu_update);
-
-
-       if (cpu_blit) {
-               ddirty.base.fifo_commit = vmw_stdu_bo_cpu_commit;
-               ddirty.base.clip = vmw_stdu_bo_cpu_clip;
-               ddirty.base.fifo_reserve_size = 0;
-       }
+
+       ddirty.base.fifo_commit = vmw_stdu_bo_cpu_commit;
+       ddirty.base.clip = vmw_stdu_bo_cpu_clip;
+       ddirty.base.fifo_reserve_size = 0;
 
        ddirty.base.crtc = crtc;
 
@@ -1161,11 +1013,8 @@ vmw_stdu_primary_plane_prepare_fb(struct drm_plane *plane,
        /*
         * This should only happen if the buffer object is too large to create a
         * proxy surface for.
-        * If we are a 2D VM with a buffer object then we have to use CPU blit
-        * so cache these mappings
         */
-       if (vps->content_fb_type == SEPARATE_BO &&
-           vmw_stdu_use_cpu_blit(dev_priv))
+       if (vps->content_fb_type == SEPARATE_BO)
                vps->cpp = new_fb->pitches[0] / new_fb->width;
 
        return 0;
@@ -1175,14 +1024,6 @@ out_srf_unref:
        return ret;
 }
 
-static uint32_t vmw_stdu_bo_fifo_size(struct vmw_du_update_plane *update,
-                                     uint32_t num_hits)
-{
-       return sizeof(struct vmw_stdu_dma) + sizeof(SVGA3dCopyBox) * num_hits +
-               sizeof(SVGA3dCmdSurfaceDMASuffix) +
-               sizeof(struct vmw_stdu_update);
-}
-
 static uint32_t vmw_stdu_bo_fifo_size_cpu(struct vmw_du_update_plane *update,
                                          uint32_t num_hits)
 {
@@ -1190,68 +1031,6 @@ static uint32_t vmw_stdu_bo_fifo_size_cpu(struct vmw_du_update_plane *update,
                sizeof(struct vmw_stdu_update);
 }
 
-static uint32_t vmw_stdu_bo_populate_dma(struct vmw_du_update_plane  *update,
-                                        void *cmd, uint32_t num_hits)
-{
-       struct vmw_screen_target_display_unit *stdu;
-       struct vmw_framebuffer_bo *vfbbo;
-       struct vmw_stdu_dma *cmd_dma = cmd;
-
-       stdu = container_of(update->du, typeof(*stdu), base);
-       vfbbo = container_of(update->vfb, typeof(*vfbbo), base);
-
-       cmd_dma->header.id = SVGA_3D_CMD_SURFACE_DMA;
-       cmd_dma->header.size = sizeof(cmd_dma->body) +
-               sizeof(struct SVGA3dCopyBox) * num_hits +
-               sizeof(SVGA3dCmdSurfaceDMASuffix);
-       vmw_bo_get_guest_ptr(&vfbbo->buffer->base, &cmd_dma->body.guest.ptr);
-       cmd_dma->body.guest.pitch = update->vfb->base.pitches[0];
-       cmd_dma->body.host.sid = stdu->display_srf->res.id;
-       cmd_dma->body.host.face = 0;
-       cmd_dma->body.host.mipmap = 0;
-       cmd_dma->body.transfer = SVGA3D_WRITE_HOST_VRAM;
-
-       return sizeof(*cmd_dma);
-}
-
-static uint32_t vmw_stdu_bo_populate_clip(struct vmw_du_update_plane  *update,
-                                         void *cmd, struct drm_rect *clip,
-                                         uint32_t fb_x, uint32_t fb_y)
-{
-       struct SVGA3dCopyBox *box = cmd;
-
-       box->srcx = fb_x;
-       box->srcy = fb_y;
-       box->srcz = 0;
-       box->x = clip->x1;
-       box->y = clip->y1;
-       box->z = 0;
-       box->w = drm_rect_width(clip);
-       box->h = drm_rect_height(clip);
-       box->d = 1;
-
-       return sizeof(*box);
-}
-
-static uint32_t vmw_stdu_bo_populate_update(struct vmw_du_update_plane  *update,
-                                           void *cmd, struct drm_rect *bb)
-{
-       struct vmw_screen_target_display_unit *stdu;
-       struct vmw_framebuffer_bo *vfbbo;
-       SVGA3dCmdSurfaceDMASuffix *suffix = cmd;
-
-       stdu = container_of(update->du, typeof(*stdu), base);
-       vfbbo = container_of(update->vfb, typeof(*vfbbo), base);
-
-       suffix->suffixSize = sizeof(*suffix);
-       suffix->maximumOffset = vfbbo->buffer->base.base.size;
-
-       vmw_stdu_populate_update(&suffix[1], stdu->base.unit, bb->x1, bb->x2,
-                                bb->y1, bb->y2);
-
-       return sizeof(*suffix) + sizeof(struct vmw_stdu_update);
-}
-
 static uint32_t vmw_stdu_bo_pre_clip_cpu(struct vmw_du_update_plane  *update,
                                         void *cmd, uint32_t num_hits)
 {
@@ -1369,24 +1148,12 @@ static int vmw_stdu_plane_update_bo(struct vmw_private *dev_priv,
        bo_update.base.vfb = vfb;
        bo_update.base.out_fence = out_fence;
        bo_update.base.mutex = NULL;
-       bo_update.base.cpu_blit = vmw_stdu_use_cpu_blit(dev_priv);
        bo_update.base.intr = false;
 
-       /*
-        * VM without 3D support don't have surface DMA command and framebuffer
-        * should be moved out of VRAM.
-        */
-       if (bo_update.base.cpu_blit) {
-               bo_update.base.calc_fifo_size = vmw_stdu_bo_fifo_size_cpu;
-               bo_update.base.pre_clip = vmw_stdu_bo_pre_clip_cpu;
-               bo_update.base.clip = vmw_stdu_bo_clip_cpu;
-               bo_update.base.post_clip = vmw_stdu_bo_populate_update_cpu;
-       } else {
-               bo_update.base.calc_fifo_size = vmw_stdu_bo_fifo_size;
-               bo_update.base.pre_clip = vmw_stdu_bo_populate_dma;
-               bo_update.base.clip = vmw_stdu_bo_populate_clip;
-               bo_update.base.post_clip = vmw_stdu_bo_populate_update;
-       }
+       bo_update.base.calc_fifo_size = vmw_stdu_bo_fifo_size_cpu;
+       bo_update.base.pre_clip = vmw_stdu_bo_pre_clip_cpu;
+       bo_update.base.clip = vmw_stdu_bo_clip_cpu;
+       bo_update.base.post_clip = vmw_stdu_bo_populate_update_cpu;
 
        return vmw_du_helper_plane_update(&bo_update.base);
 }
@@ -1549,7 +1316,6 @@ static int vmw_stdu_plane_update_surface(struct vmw_private *dev_priv,
        srf_update.vfb = vfb;
        srf_update.out_fence = out_fence;
        srf_update.mutex = &dev_priv->cmdbuf_mutex;
-       srf_update.cpu_blit = false;
        srf_update.intr = true;
 
        if (vfbs->is_bo_proxy)
index 71ce891..b35ac19 100644 (file)
@@ -66,7 +66,8 @@ static const struct vmw_res_func vmw_dx_streamoutput_func = {
        .needs_backup = true,
        .may_evict = false,
        .type_name = "DX streamoutput",
-       .backup_placement = &vmw_mob_placement,
+       .domain = VMW_BO_DOMAIN_MOB,
+       .busy_domain = VMW_BO_DOMAIN_MOB,
        .create = vmw_dx_streamoutput_create,
        .destroy = NULL, /* Command buffer managed resource. */
        .bind = vmw_dx_streamoutput_bind,
index 296d903..9c6a691 100644 (file)
@@ -131,7 +131,8 @@ static const struct vmw_res_func vmw_legacy_surface_func = {
        .prio = 1,
        .dirty_prio = 1,
        .type_name = "legacy surfaces",
-       .backup_placement = &vmw_srf_placement,
+       .domain = VMW_BO_DOMAIN_GMR,
+       .busy_domain = VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
        .create = &vmw_legacy_srf_create,
        .destroy = &vmw_legacy_srf_destroy,
        .bind = &vmw_legacy_srf_bind,
@@ -145,7 +146,8 @@ static const struct vmw_res_func vmw_gb_surface_func = {
        .prio = 1,
        .dirty_prio = 2,
        .type_name = "guest backed surfaces",
-       .backup_placement = &vmw_mob_placement,
+       .domain = VMW_BO_DOMAIN_MOB,
+       .busy_domain = VMW_BO_DOMAIN_MOB,
        .create = vmw_gb_surface_create,
        .destroy = vmw_gb_surface_destroy,
        .bind = vmw_gb_surface_bind,
index 0f86bc8..1a61ebf 100644 (file)
@@ -78,20 +78,6 @@ static const struct ttm_place vram_gmr_placement_flags[] = {
        }
 };
 
-static const struct ttm_place gmr_vram_placement_flags[] = {
-       {
-               .fpfn = 0,
-               .lpfn = 0,
-               .mem_type = VMW_PL_GMR,
-               .flags = 0
-       }, {
-               .fpfn = 0,
-               .lpfn = 0,
-               .mem_type = TTM_PL_VRAM,
-               .flags = 0
-       }
-};
-
 static const struct ttm_place vmw_sys_placement_flags = {
        .fpfn = 0,
        .lpfn = 0,
@@ -127,32 +113,6 @@ struct ttm_placement vmw_pt_sys_placement = {
        .busy_placement = &vmw_sys_placement_flags
 };
 
-static const struct ttm_place nonfixed_placement_flags[] = {
-       {
-               .fpfn = 0,
-               .lpfn = 0,
-               .mem_type = TTM_PL_SYSTEM,
-               .flags = 0
-       }, {
-               .fpfn = 0,
-               .lpfn = 0,
-               .mem_type = VMW_PL_GMR,
-               .flags = 0
-       }, {
-               .fpfn = 0,
-               .lpfn = 0,
-               .mem_type = VMW_PL_MOB,
-               .flags = 0
-       }
-};
-
-struct ttm_placement vmw_srf_placement = {
-       .num_placement = 1,
-       .num_busy_placement = 2,
-       .placement = &gmr_placement_flags,
-       .busy_placement = gmr_vram_placement_flags
-};
-
 struct ttm_placement vmw_mob_placement = {
        .num_placement = 1,
        .num_busy_placement = 1,
@@ -160,13 +120,6 @@ struct ttm_placement vmw_mob_placement = {
        .busy_placement = &mob_placement_flags
 };
 
-struct ttm_placement vmw_nonfixed_placement = {
-       .num_placement = 3,
-       .placement = nonfixed_placement_flags,
-       .num_busy_placement = 1,
-       .busy_placement = &sys_placement_flags
-};
-
 const size_t vmw_tt_size = sizeof(struct vmw_ttm_tt);
 
 /**
index 6ad744a..c968180 100644 (file)
@@ -25,6 +25,7 @@
  *
  **************************************************************************/
 
+#include "vmwgfx_bo.h"
 #include "vmwgfx_drv.h"
 #include "vmwgfx_resource_priv.h"
 
@@ -83,7 +84,8 @@ static const struct vmw_simple_resource_func va_stream_func = {
                .needs_backup = false,
                .may_evict = false,
                .type_name = "overlay stream",
-               .backup_placement = NULL,
+               .domain = VMW_BO_DOMAIN_SYS,
+               .busy_domain = VMW_BO_DOMAIN_SYS,
                .create = NULL,
                .destroy = NULL,
                .bind = NULL,
index 770b1b5..05f0909 100644 (file)
@@ -27,6 +27,7 @@
  **************************************************************************/
 #include "vmwgfx_bo.h"
 #include "vmwgfx_drv.h"
+#include "vmwgfx_resource_priv.h"
 #include "vmwgfx_validation.h"
 
 #include <linux/slab.h>
@@ -40,8 +41,6 @@
  * @hash: A hash entry used for the duplicate detection hash table.
  * @coherent_count: If switching backup buffers, number of new coherent
  * resources that will have this buffer as a backup buffer.
- * @as_mob: Validate as mob.
- * @cpu_blit: Validate for cpu blit access.
  *
  * Bit fields are used since these structures are allocated and freed in
  * large numbers and space conservation is desired.
@@ -50,8 +49,6 @@ struct vmw_validation_bo_node {
        struct ttm_validate_buffer base;
        struct vmwgfx_hash_item hash;
        unsigned int coherent_count;
-       u32 as_mob : 1;
-       u32 cpu_blit : 1;
 };
 /**
  * struct vmw_validation_res_node - Resource validation metadata.
@@ -260,26 +257,16 @@ out:
  * vmw_validation_add_bo - Add a buffer object to the validation context.
  * @ctx: The validation context.
  * @vbo: The buffer object.
- * @as_mob: Validate as mob, otherwise suitable for GMR operations.
- * @cpu_blit: Validate in a page-mappable location.
  *
  * Return: Zero on success, negative error code otherwise.
  */
 int vmw_validation_add_bo(struct vmw_validation_context *ctx,
-                         struct vmw_bo *vbo,
-                         bool as_mob,
-                         bool cpu_blit)
+                         struct vmw_bo *vbo)
 {
        struct vmw_validation_bo_node *bo_node;
 
        bo_node = vmw_validation_find_bo_dup(ctx, vbo);
-       if (bo_node) {
-               if (bo_node->as_mob != as_mob ||
-                   bo_node->cpu_blit != cpu_blit) {
-                       DRM_ERROR("Inconsistent buffer usage.\n");
-                       return -EINVAL;
-               }
-       } else {
+       if (!bo_node) {
                struct ttm_validate_buffer *val_buf;
 
                bo_node = vmw_validation_mem_alloc(ctx, sizeof(*bo_node));
@@ -297,8 +284,6 @@ int vmw_validation_add_bo(struct vmw_validation_context *ctx,
                        return -ESRCH;
                val_buf->num_shared = 0;
                list_add_tail(&val_buf->head, &ctx->bo_list);
-               bo_node->as_mob = as_mob;
-               bo_node->cpu_blit = cpu_blit;
        }
 
        return 0;
@@ -455,9 +440,10 @@ int vmw_validation_res_reserve(struct vmw_validation_context *ctx,
                if (res->backup) {
                        struct vmw_bo *vbo = res->backup;
 
-                       ret = vmw_validation_add_bo
-                               (ctx, vbo, vmw_resource_needs_backup(res),
-                                false);
+                       vmw_bo_placement_set(vbo,
+                                            res->func->domain,
+                                            res->func->busy_domain);
+                       ret = vmw_validation_add_bo(ctx, vbo);
                        if (ret)
                                goto out_unreserve;
                }
@@ -519,14 +505,12 @@ void vmw_validation_res_unreserve(struct vmw_validation_context *ctx,
  * vmw_validation_bo_validate_single - Validate a single buffer object.
  * @bo: The TTM buffer object base.
  * @interruptible: Whether to perform waits interruptible if possible.
- * @validate_as_mob: Whether to validate in MOB memory.
  *
  * Return: Zero on success, -ERESTARTSYS if interrupted. Negative error
  * code on failure.
  */
-int vmw_validation_bo_validate_single(struct ttm_buffer_object *bo,
-                                     bool interruptible,
-                                     bool validate_as_mob)
+static int vmw_validation_bo_validate_single(struct ttm_buffer_object *bo,
+                                            bool interruptible)
 {
        struct vmw_bo *vbo =
                container_of(bo, struct vmw_bo, base);
@@ -542,27 +526,17 @@ int vmw_validation_bo_validate_single(struct ttm_buffer_object *bo,
        if (vbo->base.pin_count > 0)
                return 0;
 
-       if (validate_as_mob)
-               return ttm_bo_validate(bo, &vmw_mob_placement, &ctx);
-
-       /**
-        * Put BO in VRAM if there is space, otherwise as a GMR.
-        * If there is no space in VRAM and GMR ids are all used up,
-        * start evicting GMRs to make room. If the DMA buffer can't be
-        * used as a GMR, this will return -ENOMEM.
-        */
-
-       ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, &ctx);
+       ret = ttm_bo_validate(bo, &vbo->placement, &ctx);
        if (ret == 0 || ret == -ERESTARTSYS)
                return ret;
 
-       /**
-        * If that failed, try VRAM again, this time evicting
+       /*
+        * If that failed, try again, this time evicting
         * previous contents.
         */
+       ctx.allow_res_evict = true;
 
-       ret = ttm_bo_validate(bo, &vmw_vram_placement, &ctx);
-       return ret;
+       return ttm_bo_validate(bo, &vbo->placement, &ctx);
 }
 
 /**
@@ -583,18 +557,8 @@ int vmw_validation_bo_validate(struct vmw_validation_context *ctx, bool intr)
                struct vmw_bo *vbo =
                        container_of(entry->base.bo, typeof(*vbo), base);
 
-               if (entry->cpu_blit) {
-                       struct ttm_operation_ctx ttm_ctx = {
-                               .interruptible = intr,
-                               .no_wait_gpu = false
-                       };
-
-                       ret = ttm_bo_validate(entry->base.bo,
-                                             &vmw_nonfixed_placement, &ttm_ctx);
-               } else {
-                       ret = vmw_validation_bo_validate_single
-                       (entry->base.bo, intr, entry->as_mob);
-               }
+               ret = vmw_validation_bo_validate_single(entry->base.bo, intr);
+
                if (ret)
                        return ret;
 
@@ -655,9 +619,9 @@ int vmw_validation_res_validate(struct vmw_validation_context *ctx, bool intr)
                if (backup && res->backup && (backup != res->backup)) {
                        struct vmw_bo *vbo = res->backup;
 
-                       ret = vmw_validation_add_bo
-                               (ctx, vbo, vmw_resource_needs_backup(res),
-                                false);
+                       vmw_bo_placement_set(vbo, res->func->domain,
+                                            res->func->busy_domain);
+                       ret = vmw_validation_add_bo(ctx, vbo);
                        if (ret)
                                return ret;
                }
index 4aa4f70..240ee0c 100644 (file)
@@ -159,11 +159,7 @@ static inline unsigned int vmw_validation_align(unsigned int val)
 }
 
 int vmw_validation_add_bo(struct vmw_validation_context *ctx,
-                         struct vmw_bo *vbo,
-                         bool as_mob, bool cpu_blit);
-int vmw_validation_bo_validate_single(struct ttm_buffer_object *bo,
-                                     bool interruptible,
-                                     bool validate_as_mob);
+                         struct vmw_bo *vbo);
 int vmw_validation_bo_validate(struct vmw_validation_context *ctx, bool intr);
 void vmw_validation_unref_lists(struct vmw_validation_context *ctx);
 int vmw_validation_add_resource(struct vmw_validation_context *ctx,