From 4c7886791264f03428d5424befb1b96f08fc90f4 Mon Sep 17 00:00:00 2001 From: Jerome Glisse Date: Fri, 20 Nov 2009 14:29:23 +0100 Subject: drm/radeon/kms: Rework radeon object handling The locking & protection of radeon object was somewhat messy. This patch completely rework it to now use ttm reserve as a protection for the radeon object structure member. It also shrink down the various radeon object structure by removing field which were redondant with the ttm information. Last it converts few simple functions to inline which should with performances. airlied: rebase on top of r600 and other changes. Signed-off-by: Jerome Glisse Signed-off-by: Dave Airlie --- drivers/gpu/drm/radeon/radeon_ttm.c | 36 +++++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 13 deletions(-) (limited to 'drivers/gpu/drm/radeon/radeon_ttm.c') diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c index 1381e06d6af3..bdb46c8cadd1 100644 --- a/drivers/gpu/drm/radeon/radeon_ttm.c +++ b/drivers/gpu/drm/radeon/radeon_ttm.c @@ -150,7 +150,7 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, man->default_caching = TTM_PL_FLAG_CACHED; break; case TTM_PL_TT: - man->gpu_offset = 0; + man->gpu_offset = rdev->mc.gtt_location; man->available_caching = TTM_PL_MASK_CACHING; man->default_caching = TTM_PL_FLAG_CACHED; man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | TTM_MEMTYPE_FLAG_CMA; @@ -180,7 +180,7 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, break; case TTM_PL_VRAM: /* "On-card" video ram */ - man->gpu_offset = 0; + man->gpu_offset = rdev->mc.vram_location; man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_NEEDS_IOREMAP | TTM_MEMTYPE_FLAG_MAPPABLE; @@ -482,27 +482,31 @@ int radeon_ttm_init(struct radeon_device *rdev) DRM_ERROR("failed initializing buffer object driver(%d).\n", r); return r; } - r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM, 0, - ((rdev->mc.real_vram_size) >> PAGE_SHIFT)); + r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM, + 0, rdev->mc.real_vram_size >> PAGE_SHIFT); if (r) { DRM_ERROR("Failed initializing VRAM heap.\n"); return r; } - r = radeon_object_create(rdev, NULL, 256 * 1024, true, - RADEON_GEM_DOMAIN_VRAM, false, - &rdev->stollen_vga_memory); + r = radeon_bo_create(rdev, NULL, 256 * 1024, true, + RADEON_GEM_DOMAIN_VRAM, + &rdev->stollen_vga_memory); if (r) { return r; } - r = radeon_object_pin(rdev->stollen_vga_memory, RADEON_GEM_DOMAIN_VRAM, NULL); + r = radeon_bo_reserve(rdev->stollen_vga_memory, false); + if (r) + return r; + r = radeon_bo_pin(rdev->stollen_vga_memory, RADEON_GEM_DOMAIN_VRAM, NULL); + radeon_bo_unreserve(rdev->stollen_vga_memory); if (r) { - radeon_object_unref(&rdev->stollen_vga_memory); + radeon_bo_unref(&rdev->stollen_vga_memory); return r; } 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)); + r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_TT, + 0, rdev->mc.gtt_size >> PAGE_SHIFT); if (r) { DRM_ERROR("Failed initializing GTT heap.\n"); return r; @@ -523,9 +527,15 @@ int radeon_ttm_init(struct radeon_device *rdev) void radeon_ttm_fini(struct radeon_device *rdev) { + int r; + if (rdev->stollen_vga_memory) { - radeon_object_unpin(rdev->stollen_vga_memory); - radeon_object_unref(&rdev->stollen_vga_memory); + r = radeon_bo_reserve(rdev->stollen_vga_memory, false); + if (r == 0) { + radeon_bo_unpin(rdev->stollen_vga_memory); + radeon_bo_unreserve(rdev->stollen_vga_memory); + } + radeon_bo_unref(&rdev->stollen_vga_memory); } ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_VRAM); ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_TT); -- cgit v1.2.3 From 312ea8da049a1830aa50c6e00002e50e30df476e Mon Sep 17 00:00:00 2001 From: Jerome Glisse Date: Mon, 7 Dec 2009 15:52:58 +0100 Subject: drm/radeon/kms: Convert radeon to new TTM validation API (V2) 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 Signed-off-by: Dave Airlie --- drivers/gpu/drm/radeon/radeon.h | 3 ++ drivers/gpu/drm/radeon/radeon_object.c | 54 +++++++++++++++++++----------- drivers/gpu/drm/radeon/radeon_ttm.c | 61 +++++++++++++++++----------------- 3 files changed, 67 insertions(+), 51 deletions(-) (limited to 'drivers/gpu/drm/radeon/radeon_ttm.c') diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h index a15cf9ceb9a7..5941e7d2d7ff 100644 --- a/drivers/gpu/drm/radeon/radeon.h +++ b/drivers/gpu/drm/radeon/radeon.h @@ -208,6 +208,8 @@ struct radeon_bo { /* Protected by gem.mutex */ struct list_head list; /* Protected by tbo.reserved */ + u32 placements[3]; + struct ttm_placement placement; struct ttm_buffer_object tbo; struct ttm_bo_kmap_obj kmap; unsigned pin_count; @@ -1012,6 +1014,7 @@ 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(struct radeon_bo *rbo, u32 domain); /* 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 bec494384825..d9b239bce12a 100644 --- a/drivers/gpu/drm/radeon/radeon_object.c +++ b/drivers/gpu/drm/radeon/radeon_object.c @@ -75,6 +75,25 @@ static inline u32 radeon_ttm_flags_from_domain(u32 domain) return flags; } +void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain) +{ + u32 c = 0; + + rbo->placement.fpfn = 0; + rbo->placement.lpfn = 0; + rbo->placement.placement = rbo->placements; + rbo->placement.busy_placement = rbo->placements; + if (domain & RADEON_GEM_DOMAIN_VRAM) + rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | + TTM_PL_FLAG_VRAM; + if (domain & RADEON_GEM_DOMAIN_GTT) + rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT; + if (domain & RADEON_GEM_DOMAIN_CPU) + rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM; + rbo->placement.num_placement = c; + rbo->placement.num_busy_placement = c; +} + 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 +188,20 @@ 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; + int r, i; - flags = radeon_ttm_flags_from_domain(domain); + radeon_ttm_placement_from_domain(bo, domain); 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; + radeon_ttm_placement_from_domain(bo, domain); + for (i = 0; i < bo->placement.num_placement; i++) + bo->placements[i] |= 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) @@ -202,7 +217,7 @@ retry: int radeon_bo_unpin(struct radeon_bo *bo) { - int r; + int r, i; if (!bo->pin_count) { dev_warn(bo->rdev->dev, "%p unpin not necessary\n", bo); @@ -211,11 +226,10 @@ 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; + for (i = 0; i < bo->placement.num_placement; i++) + bo->placements[i] &= ~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 +340,15 @@ 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(bo, + lobj->wdomain); } else { - bo->tbo.proposed_placement = - radeon_ttm_flags_from_domain(lobj->rdomain); + radeon_ttm_placement_from_domain(bo, + lobj->rdomain); } 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 bdb46c8cadd1..4ca7dfc44310 100644 --- a/drivers/gpu/drm/radeon/radeon_ttm.c +++ b/drivers/gpu/drm/radeon/radeon_ttm.c @@ -197,15 +197,17 @@ 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; - + struct radeon_bo *rbo = container_of(bo, struct radeon_bo, tbo); switch (bo->mem.mem_type) { + case TTM_PL_VRAM: + radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_GTT); + break; + case TTM_PL_TT: default: - return (cur_placement & ~TTM_PL_MASK_CACHING) | - TTM_PL_FLAG_SYSTEM | - TTM_PL_FLAG_CACHED; + radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_CPU); } } @@ -283,14 +285,21 @@ 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; + u32 placements; + 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, + placement.fpfn = 0; + placement.lpfn = 0; + placement.num_placement = 1; + placement.placement = &placements; + placement.num_busy_placement = 1; + placement.busy_placement = &placements; + placements = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT; + r = ttm_bo_mem_space(bo, &placement, &tmp_mem, interruptible, no_wait); if (unlikely(r)) { return r; @@ -329,15 +338,21 @@ 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; + u32 placements; 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); + placement.fpfn = 0; + placement.lpfn = 0; + placement.num_placement = 1; + placement.placement = &placements; + placement.num_busy_placement = 1; + placement.busy_placement = &placements; + placements = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT; + r = ttm_bo_mem_space(bo, &placement, &tmp_mem, interruptible, no_wait); if (unlikely(r)) { return r; } @@ -407,18 +422,6 @@ 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 +449,6 @@ 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 +482,7 @@ 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 +505,7 @@ 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; -- cgit v1.2.3 From eaa5fd1a66fefd7cc918d80250d66fa48b10b81f Mon Sep 17 00:00:00 2001 From: Jerome Glisse Date: Wed, 9 Dec 2009 21:57:37 +0100 Subject: drm/radeon/kms: actualy set the eviction placements we choose Stupid bug, somehow copying the eviction placements into the result structure was missing. Signed-off-by: Jerome Glisse Signed-off-by: Dave Airlie --- drivers/gpu/drm/radeon/radeon_ttm.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/gpu/drm/radeon/radeon_ttm.c') diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c index 4ca7dfc44310..d2ed896cca01 100644 --- a/drivers/gpu/drm/radeon/radeon_ttm.c +++ b/drivers/gpu/drm/radeon/radeon_ttm.c @@ -209,6 +209,7 @@ static void radeon_evict_flags(struct ttm_buffer_object *bo, default: radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_CPU); } + *placement = rbo->placement; } static int radeon_verify_access(struct ttm_buffer_object *bo, struct file *filp) -- cgit v1.2.3