From 8f7286f8e4e80f7b868ba3d117ae900f0d207cbe Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Mon, 14 Feb 2011 09:57:35 +1000 Subject: [PATCH] drm/nv50: support for compression Signed-off-by: Ben Skeggs --- drivers/gpu/drm/nouveau/nouveau_drv.h | 1 + drivers/gpu/drm/nouveau/nouveau_mem.c | 2 +- drivers/gpu/drm/nouveau/nouveau_vm.c | 5 +++- drivers/gpu/drm/nouveau/nouveau_vm.h | 7 ++--- drivers/gpu/drm/nouveau/nv50_fb.c | 51 ++++++++++++++++++++++++----------- drivers/gpu/drm/nouveau/nv50_vm.c | 8 +++++- drivers/gpu/drm/nouveau/nv50_vram.c | 27 ++++++++++++++++--- drivers/gpu/drm/nouveau/nvc0_vm.c | 2 +- drivers/gpu/drm/nouveau/nvc0_vram.c | 2 +- include/drm/nouveau_drm.h | 1 + 10 files changed, 79 insertions(+), 27 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h index 4bdc726..00aff22 100644 --- a/drivers/gpu/drm/nouveau/nouveau_drv.h +++ b/drivers/gpu/drm/nouveau/nouveau_drv.h @@ -72,6 +72,7 @@ struct nouveau_mem { struct nouveau_vma tmp_vma; u8 page_shift; + struct drm_mm_node *tag; struct list_head regions; dma_addr_t *pages; u32 memtype; diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c index 73f37bd..63b9040 100644 --- a/drivers/gpu/drm/nouveau/nouveau_mem.c +++ b/drivers/gpu/drm/nouveau/nouveau_mem.c @@ -740,7 +740,7 @@ nouveau_vram_manager_new(struct ttm_mem_type_manager *man, ret = vram->get(dev, mem->num_pages << PAGE_SHIFT, mem->page_alignment << PAGE_SHIFT, size_nc, - (nvbo->tile_flags >> 8) & 0xff, &node); + (nvbo->tile_flags >> 8) & 0x3ff, &node); if (ret) return ret; diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.c b/drivers/gpu/drm/nouveau/nouveau_vm.c index 3d9c159..62824c8 100644 --- a/drivers/gpu/drm/nouveau/nouveau_vm.c +++ b/drivers/gpu/drm/nouveau/nouveau_vm.c @@ -40,6 +40,7 @@ nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node) u32 max = 1 << (vm->pgt_bits - bits); u32 end, len; + delta = 0; list_for_each_entry(r, &node->regions, rl_entry) { u64 phys = (u64)r->offset << 12; u32 num = r->length >> bits; @@ -52,7 +53,7 @@ nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node) end = max; len = end - pte; - vm->map(vma, pgt, node, pte, len, phys); + vm->map(vma, pgt, node, pte, len, phys, delta); num -= len; pte += len; @@ -60,6 +61,8 @@ nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node) pde++; pte = 0; } + + delta += (u64)len << vma->node->type; } } diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.h b/drivers/gpu/drm/nouveau/nouveau_vm.h index 0e00264..2e06b55 100644 --- a/drivers/gpu/drm/nouveau/nouveau_vm.h +++ b/drivers/gpu/drm/nouveau/nouveau_vm.h @@ -67,7 +67,8 @@ struct nouveau_vm { void (*map_pgt)(struct nouveau_gpuobj *pgd, u32 pde, struct nouveau_gpuobj *pgt[2]); void (*map)(struct nouveau_vma *, struct nouveau_gpuobj *, - struct nouveau_mem *, u32 pte, u32 cnt, u64 phys); + struct nouveau_mem *, u32 pte, u32 cnt, + u64 phys, u64 delta); void (*map_sg)(struct nouveau_vma *, struct nouveau_gpuobj *, struct nouveau_mem *, u32 pte, u32 cnt, dma_addr_t *); void (*unmap)(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt); @@ -93,7 +94,7 @@ void nouveau_vm_map_sg(struct nouveau_vma *, u64 offset, u64 length, void nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde, struct nouveau_gpuobj *pgt[2]); void nv50_vm_map(struct nouveau_vma *, struct nouveau_gpuobj *, - struct nouveau_mem *, u32 pte, u32 cnt, u64 phys); + struct nouveau_mem *, u32 pte, u32 cnt, u64 phys, u64 delta); void nv50_vm_map_sg(struct nouveau_vma *, struct nouveau_gpuobj *, struct nouveau_mem *, u32 pte, u32 cnt, dma_addr_t *); void nv50_vm_unmap(struct nouveau_gpuobj *, u32 pte, u32 cnt); @@ -104,7 +105,7 @@ void nv50_vm_flush_engine(struct drm_device *, int engine); void nvc0_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde, struct nouveau_gpuobj *pgt[2]); void nvc0_vm_map(struct nouveau_vma *, struct nouveau_gpuobj *, - struct nouveau_mem *, u32 pte, u32 cnt, u64 phys); + struct nouveau_mem *, u32 pte, u32 cnt, u64 phys, u64 delta); void nvc0_vm_map_sg(struct nouveau_vma *, struct nouveau_gpuobj *, struct nouveau_mem *, u32 pte, u32 cnt, dma_addr_t *); void nvc0_vm_unmap(struct nouveau_gpuobj *, u32 pte, u32 cnt); diff --git a/drivers/gpu/drm/nouveau/nv50_fb.c b/drivers/gpu/drm/nouveau/nv50_fb.c index 50290de..ed411d8 100644 --- a/drivers/gpu/drm/nouveau/nv50_fb.c +++ b/drivers/gpu/drm/nouveau/nv50_fb.c @@ -8,31 +8,61 @@ struct nv50_fb_priv { dma_addr_t r100c08; }; +static void +nv50_fb_destroy(struct drm_device *dev) +{ + struct drm_nouveau_private *dev_priv = dev->dev_private; + struct nouveau_fb_engine *pfb = &dev_priv->engine.fb; + struct nv50_fb_priv *priv = pfb->priv; + + if (drm_mm_initialized(&pfb->tag_heap)) + drm_mm_takedown(&pfb->tag_heap); + + if (priv->r100c08_page) { + pci_unmap_page(dev->pdev, priv->r100c08, PAGE_SIZE, + PCI_DMA_BIDIRECTIONAL); + __free_page(priv->r100c08_page); + } + + kfree(priv); + pfb->priv = NULL; +} + static int nv50_fb_create(struct drm_device *dev) { struct drm_nouveau_private *dev_priv = dev->dev_private; + struct nouveau_fb_engine *pfb = &dev_priv->engine.fb; struct nv50_fb_priv *priv; + u32 tagmem; + int ret; priv = kzalloc(sizeof(*priv), GFP_KERNEL); if (!priv) return -ENOMEM; + pfb->priv = priv; priv->r100c08_page = alloc_page(GFP_KERNEL | __GFP_ZERO); if (!priv->r100c08_page) { - kfree(priv); + nv50_fb_destroy(dev); return -ENOMEM; } priv->r100c08 = pci_map_page(dev->pdev, priv->r100c08_page, 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); if (pci_dma_mapping_error(dev->pdev, priv->r100c08)) { - __free_page(priv->r100c08_page); - kfree(priv); + nv50_fb_destroy(dev); return -EFAULT; } - dev_priv->engine.fb.priv = priv; + tagmem = nv_rd32(dev, 0x100320); + NV_DEBUG(dev, "%d tags available\n", tagmem); + ret = drm_mm_init(&pfb->tag_heap, 0, tagmem); + if (ret) { + nv50_fb_destroy(dev); + return ret; + } + return 0; } @@ -81,18 +111,7 @@ nv50_fb_init(struct drm_device *dev) void nv50_fb_takedown(struct drm_device *dev) { - struct drm_nouveau_private *dev_priv = dev->dev_private; - struct nv50_fb_priv *priv; - - priv = dev_priv->engine.fb.priv; - if (!priv) - return; - dev_priv->engine.fb.priv = NULL; - - pci_unmap_page(dev->pdev, priv->r100c08, PAGE_SIZE, - PCI_DMA_BIDIRECTIONAL); - __free_page(priv->r100c08_page); - kfree(priv); + nv50_fb_destroy(dev); } void diff --git a/drivers/gpu/drm/nouveau/nv50_vm.c b/drivers/gpu/drm/nouveau/nv50_vm.c index 37f941b..b23794c 100644 --- a/drivers/gpu/drm/nouveau/nv50_vm.c +++ b/drivers/gpu/drm/nouveau/nv50_vm.c @@ -83,8 +83,9 @@ nv50_vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target) void nv50_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt, - struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys) + struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta) { + u32 comp = (mem->memtype & 0x180) >> 7; u32 block; int i; @@ -105,6 +106,11 @@ nv50_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt, phys += block << (vma->node->type - 3); cnt -= block; + if (comp) { + u32 tag = mem->tag->start + ((delta >> 16) * comp); + offset_h |= (tag << 17); + delta += block << (vma->node->type - 3); + } while (block) { nv_wo32(pgt, pte + 0, offset_l); diff --git a/drivers/gpu/drm/nouveau/nv50_vram.c b/drivers/gpu/drm/nouveau/nv50_vram.c index ff6cbae..ffbc3d8 100644 --- a/drivers/gpu/drm/nouveau/nv50_vram.c +++ b/drivers/gpu/drm/nouveau/nv50_vram.c @@ -69,6 +69,11 @@ nv50_vram_del(struct drm_device *dev, struct nouveau_mem **pmem) list_del(&this->rl_entry); nouveau_mm_put(mm, this); } + + if (mem->tag) { + drm_mm_put_block(mem->tag); + mem->tag = NULL; + } mutex_unlock(&mm->mutex); kfree(mem); @@ -76,7 +81,7 @@ nv50_vram_del(struct drm_device *dev, struct nouveau_mem **pmem) int nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc, - u32 type, struct nouveau_mem **pmem) + u32 memtype, struct nouveau_mem **pmem) { struct drm_nouveau_private *dev_priv = dev->dev_private; struct ttm_bo_device *bdev = &dev_priv->ttm.bdev; @@ -84,6 +89,8 @@ nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc, struct nouveau_mm *mm = man->priv; struct nouveau_mm_node *r; struct nouveau_mem *mem; + int comp = (memtype & 0x300) >> 8; + int type = (memtype & 0x07f); int ret; if (!types[type]) @@ -96,12 +103,26 @@ nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc, if (!mem) return -ENOMEM; + mutex_lock(&mm->mutex); + if (comp) { + if (align == 16) { + struct nouveau_fb_engine *pfb = &dev_priv->engine.fb; + int n = (size >> 4) * comp; + + mem->tag = drm_mm_search_free(&pfb->tag_heap, n, 0, 0); + if (mem->tag) + mem->tag = drm_mm_get_block(mem->tag, n, 0); + } + + if (unlikely(!mem->tag)) + comp = 0; + } + INIT_LIST_HEAD(&mem->regions); mem->dev = dev_priv->dev; - mem->memtype = type; + mem->memtype = (comp << 7) | type; mem->size = size; - mutex_lock(&mm->mutex); do { ret = nouveau_mm_get(mm, types[type], size, size_nc, align, &r); if (ret) { diff --git a/drivers/gpu/drm/nouveau/nvc0_vm.c b/drivers/gpu/drm/nouveau/nvc0_vm.c index 2b984b2..69af0ba 100644 --- a/drivers/gpu/drm/nouveau/nvc0_vm.c +++ b/drivers/gpu/drm/nouveau/nvc0_vm.c @@ -59,7 +59,7 @@ nvc0_vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target) void nvc0_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt, - struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys) + struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta) { u32 next = 1 << (vma->node->type - 8); diff --git a/drivers/gpu/drm/nouveau/nvc0_vram.c b/drivers/gpu/drm/nouveau/nvc0_vram.c index 6d777a2..67c6ec6 100644 --- a/drivers/gpu/drm/nouveau/nvc0_vram.c +++ b/drivers/gpu/drm/nouveau/nvc0_vram.c @@ -78,7 +78,7 @@ nvc0_vram_new(struct drm_device *dev, u64 size, u32 align, u32 ncmin, INIT_LIST_HEAD(&mem->regions); mem->dev = dev_priv->dev; - mem->memtype = type; + mem->memtype = (type & 0xff); mem->size = size; mutex_lock(&mm->mutex); diff --git a/include/drm/nouveau_drm.h b/include/drm/nouveau_drm.h index e2cfe80..5edd3a7 100644 --- a/include/drm/nouveau_drm.h +++ b/include/drm/nouveau_drm.h @@ -94,6 +94,7 @@ struct drm_nouveau_setparam { #define NOUVEAU_GEM_DOMAIN_GART (1 << 2) #define NOUVEAU_GEM_DOMAIN_MAPPABLE (1 << 3) +#define NOUVEAU_GEM_TILE_COMP 0x00030000 /* nv50-only */ #define NOUVEAU_GEM_TILE_LAYOUT_MASK 0x0000ff00 #define NOUVEAU_GEM_TILE_16BPP 0x00000001 #define NOUVEAU_GEM_TILE_32BPP 0x00000002 -- 2.7.4