This convert radeon to use new TTM validation API, it doesn't really take advantage of it beside in the eviction case.
Signed-off-by: Jerome Glisse <jgli...@redhat.com> --- drivers/gpu/drm/radeon/radeon.h | 3 ++ drivers/gpu/drm/radeon/radeon_object.c | 56 +++++++++++++++++++++---------- drivers/gpu/drm/radeon/radeon_ttm.c | 57 ++++++++++++++----------------- 3 files changed, 67 insertions(+), 49 deletions(-) diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h index 57416d2..99767d6 100644 --- a/drivers/gpu/drm/radeon/radeon.h +++ b/drivers/gpu/drm/radeon/radeon.h @@ -208,6 +208,7 @@ struct radeon_bo { /* Protected by gem.mutex */ struct list_head list; /* Protected by tbo.reserved */ + struct ttm_placement placement; struct ttm_buffer_object tbo; struct ttm_bo_kmap_obj kmap; unsigned pin_count; @@ -1000,6 +1001,8 @@ extern void radeon_surface_init(struct radeon_device *rdev); extern int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data); extern void radeon_legacy_set_clock_gating(struct radeon_device *rdev, int enable); extern void radeon_atom_set_clock_gating(struct radeon_device *rdev, int enable); +extern void radeon_ttm_placement_from_domain(u32 domain, + struct ttm_placement *placement); /* r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280 */ struct r100_mc_save { diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c index bec4943..814f4c3 100644 --- a/drivers/gpu/drm/radeon/radeon_object.c +++ b/drivers/gpu/drm/radeon/radeon_object.c @@ -75,6 +75,34 @@ static inline u32 radeon_ttm_flags_from_domain(u32 domain) return flags; } +void radeon_ttm_placement_from_domain(u32 domain, + struct ttm_placement *placement) +{ + u32 c = 0; + + placement->fpfn = 0; + placement->lpfn = 0; + placement->placements = 0; + placement->flags = 0; + if (domain & RADEON_GEM_DOMAIN_VRAM) { + placement->flags |= TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED; + placement->placements |= TTM_PL_VRAM << (4 * (++c)); + } + if (domain & RADEON_GEM_DOMAIN_GTT) { + placement->flags |= TTM_PL_MASK_CACHING; + placement->placements |= TTM_PL_TT << (4 * (++c)); + } + if (domain & RADEON_GEM_DOMAIN_CPU) { + placement->flags |= TTM_PL_MASK_CACHING; + placement->placements |= TTM_PL_SYSTEM << (4 * (++c)); + } + placement->placements |= c; + if (!placement->placements) { + placement->flags = TTM_PL_MASK_CACHING; + placement->placements = 1; + } +} + int radeon_bo_create(struct radeon_device *rdev, struct drm_gem_object *gobj, unsigned long size, bool kernel, u32 domain, struct radeon_bo **bo_ptr) @@ -169,24 +197,18 @@ void radeon_bo_unref(struct radeon_bo **bo) int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr) { - u32 flags; - u32 tmp; int r; - flags = radeon_ttm_flags_from_domain(domain); + radeon_ttm_placement_from_domain(domain, &bo->placement); if (bo->pin_count) { bo->pin_count++; if (gpu_addr) *gpu_addr = radeon_bo_gpu_offset(bo); return 0; } - tmp = bo->tbo.mem.placement; - ttm_flag_masked(&tmp, flags, TTM_PL_MASK_MEM); - bo->tbo.proposed_placement = tmp | TTM_PL_FLAG_NO_EVICT | - TTM_PL_MASK_CACHING; + bo->placement.flags |= TTM_PL_FLAG_NO_EVICT; retry: - r = ttm_buffer_object_validate(&bo->tbo, bo->tbo.proposed_placement, - true, false); + r = ttm_buffer_object_validate(&bo->tbo, &bo->placement, true, false); if (likely(r == 0)) { bo->pin_count = 1; if (gpu_addr != NULL) @@ -211,11 +233,9 @@ int radeon_bo_unpin(struct radeon_bo *bo) bo->pin_count--; if (bo->pin_count) return 0; - bo->tbo.proposed_placement = bo->tbo.mem.placement & - ~TTM_PL_FLAG_NO_EVICT; + bo->placement.flags &= ~TTM_PL_FLAG_NO_EVICT; retry: - r = ttm_buffer_object_validate(&bo->tbo, bo->tbo.proposed_placement, - true, false); + r = ttm_buffer_object_validate(&bo->tbo, &bo->placement, true, false); if (unlikely(r != 0)) { if (r == -ERESTART) goto retry; @@ -326,15 +346,15 @@ int radeon_bo_list_validate(struct list_head *head, void *fence) bo = lobj->bo; if (!bo->pin_count) { if (lobj->wdomain) { - bo->tbo.proposed_placement = - radeon_ttm_flags_from_domain(lobj->wdomain); + radeon_ttm_placement_from_domain(lobj->wdomain, + &bo->placement); } else { - bo->tbo.proposed_placement = - radeon_ttm_flags_from_domain(lobj->rdomain); + radeon_ttm_placement_from_domain(lobj->rdomain, + &bo->placement); } retry: r = ttm_buffer_object_validate(&bo->tbo, - bo->tbo.proposed_placement, + &bo->placement, true, false); if (unlikely(r)) { if (r == -ERESTART) diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c index bdb46c8..152c1d0 100644 --- a/drivers/gpu/drm/radeon/radeon_ttm.c +++ b/drivers/gpu/drm/radeon/radeon_ttm.c @@ -197,15 +197,21 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, return 0; } -static uint32_t radeon_evict_flags(struct ttm_buffer_object *bo) +static void radeon_evict_flags(struct ttm_buffer_object *bo, + struct ttm_placement *placement) { - uint32_t cur_placement = bo->mem.placement & ~TTM_PL_MASK_MEMTYPE; - switch (bo->mem.mem_type) { + case TTM_PL_VRAM: + radeon_ttm_placement_from_domain(RADEON_GEM_DOMAIN_GTT, + placement); + break; + case TTM_PL_TT: + radeon_ttm_placement_from_domain(RADEON_GEM_DOMAIN_CPU, + placement); + break; default: - return (cur_placement & ~TTM_PL_MASK_CACHING) | - TTM_PL_FLAG_SYSTEM | - TTM_PL_FLAG_CACHED; + radeon_ttm_placement_from_domain(RADEON_GEM_DOMAIN_CPU, + placement); } } @@ -283,14 +289,14 @@ static int radeon_move_vram_ram(struct ttm_buffer_object *bo, struct radeon_device *rdev; struct ttm_mem_reg *old_mem = &bo->mem; struct ttm_mem_reg tmp_mem; - uint32_t proposed_placement; + struct ttm_placement placement; int r; rdev = radeon_get_rdev(bo->bdev); tmp_mem = *new_mem; tmp_mem.mm_node = NULL; - proposed_placement = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING; - r = ttm_bo_mem_space(bo, proposed_placement, &tmp_mem, + radeon_ttm_placement_from_domain(RADEON_GEM_DOMAIN_GTT, &placement); + r = ttm_bo_mem_space(bo, &placement, &tmp_mem, interruptible, no_wait); if (unlikely(r)) { return r; @@ -329,15 +335,14 @@ static int radeon_move_ram_vram(struct ttm_buffer_object *bo, struct radeon_device *rdev; struct ttm_mem_reg *old_mem = &bo->mem; struct ttm_mem_reg tmp_mem; - uint32_t proposed_flags; + struct ttm_placement placement; int r; rdev = radeon_get_rdev(bo->bdev); tmp_mem = *new_mem; tmp_mem.mm_node = NULL; - proposed_flags = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING; - r = ttm_bo_mem_space(bo, proposed_flags, &tmp_mem, - interruptible, no_wait); + radeon_ttm_placement_from_domain(RADEON_GEM_DOMAIN_GTT, &placement); + r = ttm_bo_mem_space(bo, &placement, &tmp_mem, interruptible, no_wait); if (unlikely(r)) { return r; } @@ -407,18 +412,6 @@ memcpy: return r; } -const uint32_t radeon_mem_prios[] = { - TTM_PL_VRAM, - TTM_PL_TT, - TTM_PL_SYSTEM, -}; - -const uint32_t radeon_busy_prios[] = { - TTM_PL_TT, - TTM_PL_VRAM, - TTM_PL_SYSTEM, -}; - static int radeon_sync_obj_wait(void *sync_obj, void *sync_arg, bool lazy, bool interruptible) { @@ -446,10 +439,6 @@ static bool radeon_sync_obj_signaled(void *sync_obj, void *sync_arg) } static struct ttm_bo_driver radeon_bo_driver = { - .mem_type_prio = radeon_mem_prios, - .mem_busy_prio = radeon_busy_prios, - .num_mem_type_prio = ARRAY_SIZE(radeon_mem_prios), - .num_mem_busy_prio = ARRAY_SIZE(radeon_busy_prios), .create_ttm_backend_entry = &radeon_create_ttm_backend_entry, .invalidate_caches = &radeon_invalidate_caches, .init_mem_type = &radeon_init_mem_type, @@ -483,7 +472,7 @@ int radeon_ttm_init(struct radeon_device *rdev) return r; } r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM, - 0, rdev->mc.real_vram_size >> PAGE_SHIFT); + rdev->mc.real_vram_size >> PAGE_SHIFT); if (r) { DRM_ERROR("Failed initializing VRAM heap.\n"); return r; @@ -506,7 +495,7 @@ int radeon_ttm_init(struct radeon_device *rdev) DRM_INFO("radeon: %uM of VRAM memory ready\n", (unsigned)rdev->mc.real_vram_size / (1024 * 1024)); r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_TT, - 0, rdev->mc.gtt_size >> PAGE_SHIFT); + rdev->mc.gtt_size >> PAGE_SHIFT); if (r) { DRM_ERROR("Failed initializing GTT heap.\n"); return r; @@ -537,11 +526,17 @@ void radeon_ttm_fini(struct radeon_device *rdev) } radeon_bo_unref(&rdev->stollen_vga_memory); } +printk(KERN_INFO "%s %d\n", __func__, __LINE__); ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_VRAM); +printk(KERN_INFO "%s %d\n", __func__, __LINE__); ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_TT); +printk(KERN_INFO "%s %d\n", __func__, __LINE__); ttm_bo_device_release(&rdev->mman.bdev); +printk(KERN_INFO "%s %d\n", __func__, __LINE__); radeon_gart_fini(rdev); +printk(KERN_INFO "%s %d\n", __func__, __LINE__); radeon_ttm_global_fini(rdev); +printk(KERN_INFO "%s %d\n", __func__, __LINE__); DRM_INFO("radeon: ttm finalized\n"); } -- 1.6.5.2 ------------------------------------------------------------------------------ Join us December 9, 2009 for the Red Hat Virtual Experience, a free event focused on virtualization and cloud computing. Attend in-depth sessions from your desk. Your couch. Anywhere. http://p.sf.net/sfu/redhat-sfdev2dev -- _______________________________________________ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel