From a0af9add499cf29a16707b7b148842992077f4cb Mon Sep 17 00:00:00 2001 From: Francisco Jerez Date: Fri, 11 Dec 2009 16:51:09 +0100 Subject: drm/nouveau: Make the MM aware of pre-G80 tiling. This commit has also the following 3 bugfix commits squashed into it from the nouveau git tree: drm/nouveau: Fix up the tiling alignment restrictions for nv1x. drm/nouveau: Fix up the nv2x tiling alignment restrictions. drm/nv50: fix align typo for g9x Signed-off-by: Francisco Jerez --- drivers/gpu/drm/nouveau/nouveau_bo.c | 221 ++++++++++++++++++++++++----------- 1 file changed, 156 insertions(+), 65 deletions(-) (limited to 'drivers/gpu/drm/nouveau/nouveau_bo.c') diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index 0cad6d834eb..1d6036fabd5 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -37,6 +37,7 @@ static void nouveau_bo_del_ttm(struct ttm_buffer_object *bo) { struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); + struct drm_device *dev = dev_priv->dev; struct nouveau_bo *nvbo = nouveau_bo(bo); ttm_bo_kunmap(&nvbo->kmap); @@ -44,12 +45,83 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo) if (unlikely(nvbo->gem)) DRM_ERROR("bo %p still attached to GEM object\n", bo); + if (nvbo->tile) + nv10_mem_expire_tiling(dev, nvbo->tile, NULL); + spin_lock(&dev_priv->ttm.bo_list_lock); list_del(&nvbo->head); spin_unlock(&dev_priv->ttm.bo_list_lock); kfree(nvbo); } +static void +nouveau_bo_fixup_align(struct drm_device *dev, + uint32_t tile_mode, uint32_t tile_flags, + int *align, int *size) +{ + struct drm_nouveau_private *dev_priv = dev->dev_private; + + /* + * Some of the tile_flags have a periodic structure of N*4096 bytes, + * align to to that as well as the page size. Overallocate memory to + * avoid corruption of other buffer objects. + */ + if (dev_priv->card_type == NV_50) { + switch (tile_flags) { + case 0x1800: + case 0x2800: + case 0x4800: + case 0x7a00: + if (dev_priv->chipset >= 0xA0) { + /* This is based on high end cards with 448 bits + * memory bus, could be different elsewhere.*/ + *size += 6 * 28672; + /* 8 * 28672 is the actual alignment requirement + * but we must also align to page size. */ + *align = 2 * 8 * 28672; + } else if (dev_priv->chipset >= 0x90) { + *size += 3 * 16384; + *align = 12 * 16384; + } else { + *size += 3 * 8192; + /* 12 * 8192 is the actual alignment requirement + * but we must also align to page size. */ + *align = 2 * 12 * 8192; + } + break; + default: + break; + } + + } else { + if (tile_mode) { + if (dev_priv->chipset >= 0x40) { + *align = 65536; + *size = roundup(*size, 64 * tile_mode); + + } else if (dev_priv->chipset >= 0x30) { + *align = 32768; + *size = roundup(*size, 64 * tile_mode); + + } else if (dev_priv->chipset >= 0x20) { + *align = 16384; + *size = roundup(*size, 64 * tile_mode); + + } else if (dev_priv->chipset >= 0x10) { + *align = 16384; + *size = roundup(*size, 32 * tile_mode); + } + } + } + + *size = ALIGN(*size, PAGE_SIZE); + + if (dev_priv->card_type == NV_50) { + *size = ALIGN(*size, 65536); + *align = max(65536, *align); + } +} + int nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan, int size, int align, uint32_t flags, uint32_t tile_mode, @@ -70,46 +142,9 @@ nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan, nvbo->tile_mode = tile_mode; nvbo->tile_flags = tile_flags; - /* - * Some of the tile_flags have a periodic structure of N*4096 bytes, - * align to to that as well as the page size. Overallocate memory to - * avoid corruption of other buffer objects. - */ - switch (tile_flags) { - case 0x1800: - case 0x2800: - case 0x4800: - case 0x7a00: - if (dev_priv->chipset >= 0xA0) { - /* This is based on high end cards with 448 bits - * memory bus, could be different elsewhere.*/ - size += 6 * 28672; - /* 8 * 28672 is the actual alignment requirement, - * but we must also align to page size. */ - align = 2 * 8 * 28672; - } else if (dev_priv->chipset >= 0x90) { - size += 3 * 16384; - align = 12 * 16834; - } else { - size += 3 * 8192; - /* 12 * 8192 is the actual alignment requirement, - * but we must also align to page size. */ - align = 2 * 12 * 8192; - } - break; - default: - break; - } - + nouveau_bo_fixup_align(dev, tile_mode, tile_flags, &align, &size); align >>= PAGE_SHIFT; - size = (size + (PAGE_SIZE - 1)) & ~(PAGE_SIZE - 1); - if (dev_priv->card_type == NV_50) { - size = (size + 65535) & ~65535; - if (align < (65536 / PAGE_SIZE)) - align = (65536 / PAGE_SIZE); - } - if (flags & TTM_PL_FLAG_VRAM) nvbo->placements[n++] = TTM_PL_FLAG_VRAM | TTM_PL_MASK_CACHING; if (flags & TTM_PL_FLAG_TT) @@ -421,6 +456,7 @@ nouveau_bo_evict_flags(struct ttm_buffer_object *bo, struct ttm_placement *pl) /* GPU-assisted copy using NV_MEMORY_TO_MEMORY_FORMAT, can access * TTM_PL_{VRAM,TT} directly. */ + static int nouveau_bo_move_accel_cleanup(struct nouveau_channel *chan, struct nouveau_bo *nvbo, bool evict, bool no_wait, @@ -455,11 +491,12 @@ nouveau_bo_mem_ctxdma(struct nouveau_bo *nvbo, struct nouveau_channel *chan, } static int -nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, int no_wait, - struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem) +nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr, + int no_wait, struct ttm_mem_reg *new_mem) { struct nouveau_bo *nvbo = nouveau_bo(bo); struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); + struct ttm_mem_reg *old_mem = &bo->mem; struct nouveau_channel *chan; uint64_t src_offset, dst_offset; uint32_t page_count; @@ -559,7 +596,7 @@ nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr, if (ret) goto out; - ret = nouveau_bo_move_m2mf(bo, true, no_wait, &bo->mem, &tmp_mem); + ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait, &tmp_mem); if (ret) goto out; @@ -597,7 +634,7 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr, if (ret) goto out; - ret = nouveau_bo_move_m2mf(bo, true, no_wait, &bo->mem, new_mem); + ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait, new_mem); if (ret) goto out; @@ -612,52 +649,106 @@ out: } static int -nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr, - bool no_wait, struct ttm_mem_reg *new_mem) +nouveau_bo_vm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem, + struct nouveau_tile_reg **new_tile) { struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); - struct nouveau_bo *nvbo = nouveau_bo(bo); struct drm_device *dev = dev_priv->dev; - struct ttm_mem_reg *old_mem = &bo->mem; + struct nouveau_bo *nvbo = nouveau_bo(bo); + uint64_t offset; int ret; - if (dev_priv->card_type == NV_50 && new_mem->mem_type == TTM_PL_VRAM && - !nvbo->no_vm) { - uint64_t offset = new_mem->mm_node->start << PAGE_SHIFT; + if (nvbo->no_vm || new_mem->mem_type != TTM_PL_VRAM) { + /* Nothing to do. */ + *new_tile = NULL; + return 0; + } + + offset = new_mem->mm_node->start << PAGE_SHIFT; + if (dev_priv->card_type == NV_50) { ret = nv50_mem_vm_bind_linear(dev, offset + dev_priv->vm_vram_base, new_mem->size, nvbo->tile_flags, offset); if (ret) return ret; + + } else if (dev_priv->card_type >= NV_10) { + *new_tile = nv10_mem_set_tiling(dev, offset, new_mem->size, + nvbo->tile_mode); } + return 0; +} + +static void +nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo, + struct nouveau_tile_reg *new_tile, + struct nouveau_tile_reg **old_tile) +{ + struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); + struct drm_device *dev = dev_priv->dev; + + if (dev_priv->card_type >= NV_10 && + dev_priv->card_type < NV_50) { + if (*old_tile) + nv10_mem_expire_tiling(dev, *old_tile, bo->sync_obj); + + *old_tile = new_tile; + } +} + +static int +nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr, + bool no_wait, struct ttm_mem_reg *new_mem) +{ + struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); + struct nouveau_bo *nvbo = nouveau_bo(bo); + struct ttm_mem_reg *old_mem = &bo->mem; + struct nouveau_tile_reg *new_tile = NULL; + int ret = 0; + + ret = nouveau_bo_vm_bind(bo, new_mem, &new_tile); + if (ret) + return ret; + + /* Software copy if the card isn't up and running yet. */ if (dev_priv->init_state != NOUVEAU_CARD_INIT_DONE || - !dev_priv->channel) - return ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); + !dev_priv->channel) { + ret = ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); + goto out; + } + /* Fake bo copy. */ if (old_mem->mem_type == TTM_PL_SYSTEM && !bo->ttm) { BUG_ON(bo->mem.mm_node != NULL); bo->mem = *new_mem; new_mem->mm_node = NULL; - return 0; + goto out; } - if (new_mem->mem_type == TTM_PL_SYSTEM) { - if (old_mem->mem_type == TTM_PL_SYSTEM) - return ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); - if (nouveau_bo_move_flipd(bo, evict, intr, no_wait, new_mem)) - return ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); - } else if (old_mem->mem_type == TTM_PL_SYSTEM) { - if (nouveau_bo_move_flips(bo, evict, intr, no_wait, new_mem)) - return ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); - } else { - if (nouveau_bo_move_m2mf(bo, evict, no_wait, old_mem, new_mem)) - return ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); - } + /* Hardware assisted copy. */ + if (new_mem->mem_type == TTM_PL_SYSTEM) + ret = nouveau_bo_move_flipd(bo, evict, intr, no_wait, new_mem); + else if (old_mem->mem_type == TTM_PL_SYSTEM) + ret = nouveau_bo_move_flips(bo, evict, intr, no_wait, new_mem); + else + ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait, new_mem); - return 0; + if (!ret) + goto out; + + /* Fallback to software copy. */ + ret = ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); + +out: + if (ret) + nouveau_bo_vm_cleanup(bo, NULL, &new_tile); + else + nouveau_bo_vm_cleanup(bo, new_tile, &nvbo->tile); + + return ret; } static int -- cgit v1.2.3-70-g09d2