X-Git-Url: https://git.karo-electronics.de/?a=blobdiff_plain;f=drivers%2Fgpu%2Fdrm%2Fnouveau%2Fnouveau_mem.c;h=3a531d53de4d86c323236cdbdaac718c6656cf02;hb=bfd83aca5a1c4cc07fb18d96ed8660f2c338a8f3;hp=fe4a30dc4b42e929f283d35e1094b26a422f4f2e;hpb=76db8ac45fc738f7d7664fe9b56d15c594a45228;p=mv-sheeva.git diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c index fe4a30dc4b4..3a531d53de4 100644 --- a/drivers/gpu/drm/nouveau/nouveau_mem.c +++ b/drivers/gpu/drm/nouveau/nouveau_mem.c @@ -36,183 +36,112 @@ #include "nouveau_drv.h" #include "nouveau_pm.h" +#include "nouveau_mm.h" +#include "nouveau_vm.h" /* * NV10-NV40 tiling helpers */ static void -nv10_mem_set_region_tiling(struct drm_device *dev, int i, uint32_t addr, - uint32_t size, uint32_t pitch) +nv10_mem_update_tile_region(struct drm_device *dev, + struct nouveau_tile_reg *tile, uint32_t addr, + uint32_t size, uint32_t pitch, uint32_t flags) { struct drm_nouveau_private *dev_priv = dev->dev_private; struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo; struct nouveau_fb_engine *pfb = &dev_priv->engine.fb; struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph; - struct nouveau_tile_reg *tile = &dev_priv->tile[i]; + int i = tile - dev_priv->tile.reg; + unsigned long save; - tile->addr = addr; - tile->size = size; - tile->used = !!pitch; - nouveau_fence_unref((void **)&tile->fence); + nouveau_fence_unref(&tile->fence); + if (tile->pitch) + pfb->free_tile_region(dev, i); + + if (pitch) + pfb->init_tile_region(dev, i, addr, size, pitch, flags); + + spin_lock_irqsave(&dev_priv->context_switch_lock, save); pfifo->reassign(dev, false); pfifo->cache_pull(dev, false); nouveau_wait_for_idle(dev); - pgraph->set_region_tiling(dev, i, addr, size, pitch); - pfb->set_region_tiling(dev, i, addr, size, pitch); + pfb->set_tile_region(dev, i); + pgraph->set_tile_region(dev, i); pfifo->cache_pull(dev, true); pfifo->reassign(dev, true); + spin_unlock_irqrestore(&dev_priv->context_switch_lock, save); } -struct nouveau_tile_reg * -nv10_mem_set_tiling(struct drm_device *dev, uint32_t addr, uint32_t size, - uint32_t pitch) +static struct nouveau_tile_reg * +nv10_mem_get_tile_region(struct drm_device *dev, int i) { struct drm_nouveau_private *dev_priv = dev->dev_private; - struct nouveau_fb_engine *pfb = &dev_priv->engine.fb; - struct nouveau_tile_reg *found = NULL; - unsigned long i, flags; + struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i]; - spin_lock_irqsave(&dev_priv->context_switch_lock, flags); - - for (i = 0; i < pfb->num_tiles; i++) { - struct nouveau_tile_reg *tile = &dev_priv->tile[i]; - - if (tile->used) - /* Tile region in use. */ - continue; - - if (tile->fence && - !nouveau_fence_signalled(tile->fence, NULL)) - /* Pending tile region. */ - continue; - - if (max(tile->addr, addr) < - min(tile->addr + tile->size, addr + size)) - /* Kill an intersecting tile region. */ - nv10_mem_set_region_tiling(dev, i, 0, 0, 0); - - if (pitch && !found) { - /* Free tile region. */ - nv10_mem_set_region_tiling(dev, i, addr, size, pitch); - found = tile; - } - } + spin_lock(&dev_priv->tile.lock); - spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags); + if (!tile->used && + (!tile->fence || nouveau_fence_signalled(tile->fence))) + tile->used = true; + else + tile = NULL; - return found; + spin_unlock(&dev_priv->tile.lock); + return tile; } void -nv10_mem_expire_tiling(struct drm_device *dev, struct nouveau_tile_reg *tile, - struct nouveau_fence *fence) -{ - if (fence) { - /* Mark it as pending. */ - tile->fence = fence; - nouveau_fence_ref(fence); - } - - tile->used = false; -} - -/* - * NV50 VM helpers - */ -int -nv50_mem_vm_bind_linear(struct drm_device *dev, uint64_t virt, uint32_t size, - uint32_t flags, uint64_t phys) +nv10_mem_put_tile_region(struct drm_device *dev, struct nouveau_tile_reg *tile, + struct nouveau_fence *fence) { struct drm_nouveau_private *dev_priv = dev->dev_private; - struct nouveau_gpuobj *pgt; - unsigned block; - int i; - - virt = ((virt - dev_priv->vm_vram_base) >> 16) << 1; - size = (size >> 16) << 1; - - phys |= ((uint64_t)flags << 32); - phys |= 1; - if (dev_priv->vram_sys_base) { - phys += dev_priv->vram_sys_base; - phys |= 0x30; - } - while (size) { - unsigned offset_h = upper_32_bits(phys); - unsigned offset_l = lower_32_bits(phys); - unsigned pte, end; - - for (i = 7; i >= 0; i--) { - block = 1 << (i + 1); - if (size >= block && !(virt & (block - 1))) - break; - } - offset_l |= (i << 7); - - phys += block << 15; - size -= block; - - while (block) { - pgt = dev_priv->vm_vram_pt[virt >> 14]; - pte = virt & 0x3ffe; - - end = pte + block; - if (end > 16384) - end = 16384; - block -= (end - pte); - virt += (end - pte); - - while (pte < end) { - nv_wo32(pgt, (pte * 4) + 0, offset_l); - nv_wo32(pgt, (pte * 4) + 4, offset_h); - pte += 2; - } + if (tile) { + spin_lock(&dev_priv->tile.lock); + if (fence) { + /* Mark it as pending. */ + tile->fence = fence; + nouveau_fence_ref(fence); } - } - dev_priv->engine.instmem.flush(dev); - dev_priv->engine.fifo.tlb_flush(dev); - dev_priv->engine.graph.tlb_flush(dev); - nv50_vm_flush(dev, 6); - return 0; + tile->used = false; + spin_unlock(&dev_priv->tile.lock); + } } -void -nv50_mem_vm_unbind(struct drm_device *dev, uint64_t virt, uint32_t size) +struct nouveau_tile_reg * +nv10_mem_set_tiling(struct drm_device *dev, uint32_t addr, uint32_t size, + uint32_t pitch, uint32_t flags) { struct drm_nouveau_private *dev_priv = dev->dev_private; - struct nouveau_gpuobj *pgt; - unsigned pages, pte, end; - - virt -= dev_priv->vm_vram_base; - pages = (size >> 16) << 1; + struct nouveau_fb_engine *pfb = &dev_priv->engine.fb; + struct nouveau_tile_reg *tile, *found = NULL; + int i; - while (pages) { - pgt = dev_priv->vm_vram_pt[virt >> 29]; - pte = (virt & 0x1ffe0000ULL) >> 15; + for (i = 0; i < pfb->num_tiles; i++) { + tile = nv10_mem_get_tile_region(dev, i); - end = pte + pages; - if (end > 16384) - end = 16384; - pages -= (end - pte); - virt += (end - pte) << 15; + if (pitch && !found) { + found = tile; + continue; - while (pte < end) { - nv_wo32(pgt, (pte * 4), 0); - pte++; + } else if (tile && tile->pitch) { + /* Kill an unused tile region. */ + nv10_mem_update_tile_region(dev, tile, 0, 0, 0, 0); } + + nv10_mem_put_tile_region(dev, tile, NULL); } - dev_priv->engine.instmem.flush(dev); - dev_priv->engine.fifo.tlb_flush(dev); - dev_priv->engine.graph.tlb_flush(dev); - nv50_vm_flush(dev, 6); + if (found) + nv10_mem_update_tile_region(dev, found, addr, size, + pitch, flags); + return found; } /* @@ -312,61 +241,6 @@ nouveau_mem_detect_nforce(struct drm_device *dev) return 0; } -static void -nv50_vram_preinit(struct drm_device *dev) -{ - struct drm_nouveau_private *dev_priv = dev->dev_private; - int i, parts, colbits, rowbitsa, rowbitsb, banks; - u64 rowsize, predicted; - u32 r0, r4, rt, ru; - - r0 = nv_rd32(dev, 0x100200); - r4 = nv_rd32(dev, 0x100204); - rt = nv_rd32(dev, 0x100250); - ru = nv_rd32(dev, 0x001540); - NV_DEBUG(dev, "memcfg 0x%08x 0x%08x 0x%08x 0x%08x\n", r0, r4, rt, ru); - - for (i = 0, parts = 0; i < 8; i++) { - if (ru & (0x00010000 << i)) - parts++; - } - - colbits = (r4 & 0x0000f000) >> 12; - rowbitsa = ((r4 & 0x000f0000) >> 16) + 8; - rowbitsb = ((r4 & 0x00f00000) >> 20) + 8; - banks = ((r4 & 0x01000000) ? 8 : 4); - - rowsize = parts * banks * (1 << colbits) * 8; - predicted = rowsize << rowbitsa; - if (r0 & 0x00000004) - predicted += rowsize << rowbitsb; - - if (predicted != dev_priv->vram_size) { - NV_WARN(dev, "memory controller reports %dMiB VRAM\n", - (u32)(dev_priv->vram_size >> 20)); - NV_WARN(dev, "we calculated %dMiB VRAM\n", - (u32)(predicted >> 20)); - } - - dev_priv->vram_rblock_size = rowsize >> 12; - if (rt & 1) - dev_priv->vram_rblock_size *= 3; - - NV_DEBUG(dev, "rblock %lld bytes\n", - (u64)dev_priv->vram_rblock_size << 12); -} - -static void -nvaa_vram_preinit(struct drm_device *dev) -{ - struct drm_nouveau_private *dev_priv = dev->dev_private; - - /* To our knowledge, there's no large scale reordering of pages - * that occurs on IGP chipsets. - */ - dev_priv->vram_rblock_size = 1; -} - static int nouveau_mem_detect(struct drm_device *dev) { @@ -383,22 +257,8 @@ nouveau_mem_detect(struct drm_device *dev) dev_priv->vram_size &= NV10_PFB_FIFO_DATA_RAM_AMOUNT_MB_MASK; } else if (dev_priv->card_type < NV_C0) { - dev_priv->vram_size = nv_rd32(dev, NV04_PFB_FIFO_DATA); - dev_priv->vram_size |= (dev_priv->vram_size & 0xff) << 32; - dev_priv->vram_size &= 0xffffffff00ll; - - switch (dev_priv->chipset) { - case 0xaa: - case 0xac: - case 0xaf: - dev_priv->vram_sys_base = nv_rd32(dev, 0x100e10); - dev_priv->vram_sys_base <<= 12; - nvaa_vram_preinit(dev); - break; - default: - nv50_vram_preinit(dev); - break; - } + if (nv50_vram_init(dev)) + return -ENOMEM; } else { dev_priv->vram_size = nv_rd32(dev, 0x10f20c) << 20; dev_priv->vram_size *= nv_rd32(dev, 0x121c74); @@ -547,10 +407,6 @@ nouveau_mem_vram_init(struct drm_device *dev) if (ret) return ret; - ret = nouveau_mem_detect(dev); - if (ret) - return ret; - dev_priv->fb_phys = pci_resource_start(dev->pdev, 1); ret = nouveau_ttm_global_init(dev_priv); @@ -566,13 +422,6 @@ nouveau_mem_vram_init(struct drm_device *dev) return ret; } - dev_priv->fb_available_size = dev_priv->vram_size; - dev_priv->fb_mappable_pages = dev_priv->fb_available_size; - if (dev_priv->fb_mappable_pages > pci_resource_len(dev->pdev, 1)) - dev_priv->fb_mappable_pages = - pci_resource_len(dev->pdev, 1); - dev_priv->fb_mappable_pages >>= PAGE_SHIFT; - /* reserve space at end of VRAM for PRAMIN */ if (dev_priv->chipset == 0x40 || dev_priv->chipset == 0x47 || dev_priv->chipset == 0x49 || dev_priv->chipset == 0x4b) @@ -583,6 +432,17 @@ nouveau_mem_vram_init(struct drm_device *dev) else dev_priv->ramin_rsvd_vram = (512 * 1024); + /* initialise gpu-specific vram backend */ + ret = nouveau_mem_detect(dev); + if (ret) + return ret; + + dev_priv->fb_available_size = dev_priv->vram_size; + dev_priv->fb_mappable_pages = dev_priv->fb_available_size; + if (dev_priv->fb_mappable_pages > pci_resource_len(dev->pdev, 1)) + dev_priv->fb_mappable_pages = pci_resource_len(dev->pdev, 1); + dev_priv->fb_mappable_pages >>= PAGE_SHIFT; + dev_priv->fb_available_size -= dev_priv->ramin_rsvd_vram; dev_priv->fb_aper_free = dev_priv->fb_available_size; @@ -799,3 +659,109 @@ nouveau_mem_timing_fini(struct drm_device *dev) kfree(mem->timing); } + +static int +nouveau_vram_manager_init(struct ttm_mem_type_manager *man, unsigned long p_size) +{ + struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev); + struct nouveau_mm *mm; + u32 b_size; + int ret; + + p_size = (p_size << PAGE_SHIFT) >> 12; + b_size = dev_priv->vram_rblock_size >> 12; + + ret = nouveau_mm_init(&mm, 0, p_size, b_size); + if (ret) + return ret; + + man->priv = mm; + return 0; +} + +static int +nouveau_vram_manager_fini(struct ttm_mem_type_manager *man) +{ + struct nouveau_mm *mm = man->priv; + int ret; + + ret = nouveau_mm_fini(&mm); + if (ret) + return ret; + + man->priv = NULL; + return 0; +} + +static void +nouveau_vram_manager_del(struct ttm_mem_type_manager *man, + struct ttm_mem_reg *mem) +{ + struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev); + struct drm_device *dev = dev_priv->dev; + + nv50_vram_del(dev, (struct nouveau_vram **)&mem->mm_node); +} + +static int +nouveau_vram_manager_new(struct ttm_mem_type_manager *man, + struct ttm_buffer_object *bo, + struct ttm_placement *placement, + struct ttm_mem_reg *mem) +{ + struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev); + struct drm_device *dev = dev_priv->dev; + struct nouveau_bo *nvbo = nouveau_bo(bo); + struct nouveau_vram *vram; + int ret; + + ret = nv50_vram_new(dev, mem->num_pages << PAGE_SHIFT, + mem->page_alignment << PAGE_SHIFT, 0, + (nvbo->tile_flags >> 8) & 0x7f, &vram); + if (ret) + return ret; + + mem->mm_node = vram; + mem->start = vram->offset >> PAGE_SHIFT; + return 0; +} + +void +nouveau_vram_manager_debug(struct ttm_mem_type_manager *man, const char *prefix) +{ + struct ttm_bo_global *glob = man->bdev->glob; + struct nouveau_mm *mm = man->priv; + struct nouveau_mm_node *r; + u64 total = 0, ttotal[3] = {}, tused[3] = {}, tfree[3] = {}; + int i; + + mutex_lock(&mm->mutex); + list_for_each_entry(r, &mm->nodes, nl_entry) { + printk(KERN_DEBUG "%s %s-%d: 0x%010llx 0x%010llx\n", + prefix, r->free ? "free" : "used", r->type, + ((u64)r->offset << 12), + (((u64)r->offset + r->length) << 12)); + total += r->length; + ttotal[r->type] += r->length; + if (r->free) + tfree[r->type] += r->length; + else + tused[r->type] += r->length; + } + mutex_unlock(&mm->mutex); + + printk(KERN_DEBUG "%s total: 0x%010llx\n", prefix, total << 12); + for (i = 0; i < 3; i++) { + printk(KERN_DEBUG "%s type %d: 0x%010llx, " + "used 0x%010llx, free 0x%010llx\n", prefix, + i, ttotal[i] << 12, tused[i] << 12, tfree[i] << 12); + } +} + +const struct ttm_mem_type_manager_func nouveau_vram_manager = { + nouveau_vram_manager_init, + nouveau_vram_manager_fini, + nouveau_vram_manager_new, + nouveau_vram_manager_del, + nouveau_vram_manager_debug +};