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);
-
+ nv10_mem_put_tile_region(dev, nvbo->tile, NULL);
kfree(nvbo);
}
nvbo->no_vm = no_vm;
nvbo->tile_mode = tile_mode;
nvbo->tile_flags = tile_flags;
+ nvbo->bo.bdev = &dev_priv->ttm.bdev;
- nouveau_bo_fixup_align(dev, tile_mode, tile_flags, &align, &size);
+ nouveau_bo_fixup_align(dev, tile_mode, nouveau_bo_tile_layout(nvbo),
+ &align, &size);
align >>= PAGE_SHIFT;
nouveau_bo_placement_set(nvbo, flags, 0);
pl[(*n)++] = TTM_PL_FLAG_SYSTEM | flags;
}
+static void
+set_placement_range(struct nouveau_bo *nvbo, uint32_t type)
+{
+ struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev);
+
+ if (dev_priv->card_type == NV_10 &&
+ nvbo->tile_mode && (type & TTM_PL_FLAG_VRAM)) {
+ /*
+ * Make sure that the color and depth buffers are handled
+ * by independent memory controller units. Up to a 9x
+ * speed up when alpha-blending and depth-test are enabled
+ * at the same time.
+ */
+ int vram_pages = dev_priv->vram_size >> PAGE_SHIFT;
+
+ if (nvbo->tile_flags & NOUVEAU_GEM_TILE_ZETA) {
+ nvbo->placement.fpfn = vram_pages / 2;
+ nvbo->placement.lpfn = ~0;
+ } else {
+ nvbo->placement.fpfn = 0;
+ nvbo->placement.lpfn = vram_pages / 2;
+ }
+ }
+}
+
void
nouveau_bo_placement_set(struct nouveau_bo *nvbo, uint32_t type, uint32_t busy)
{
pl->busy_placement = nvbo->busy_placements;
set_placement_list(nvbo->busy_placements, &pl->num_busy_placement,
type | busy, flags);
+
+ set_placement_range(nvbo, type);
}
int
switch (dev_priv->gart_info.type) {
case NOUVEAU_GART_AGP:
man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
- man->available_caching = TTM_PL_FLAG_UNCACHED;
- man->default_caching = TTM_PL_FLAG_UNCACHED;
+ man->available_caching = TTM_PL_FLAG_UNCACHED |
+ TTM_PL_FLAG_WC;
+ man->default_caching = TTM_PL_FLAG_WC;
break;
case NOUVEAU_GART_SGDMA:
man->flags = TTM_MEMTYPE_FLAG_MAPPABLE |
if (ret)
return ret;
- if (nvbo->channel) {
- ret = nouveau_fence_sync(fence, nvbo->channel);
- if (ret)
- goto out;
- }
-
ret = ttm_bo_move_accel_cleanup(&nvbo->bo, fence, NULL, evict,
no_wait_reserve, no_wait_gpu, new_mem);
-out:
- nouveau_fence_unref((void *)&fence);
+ nouveau_fence_unref(&fence);
return ret;
}
stride = 16 * 4;
height = amount / stride;
- if (new_mem->mem_type == TTM_PL_VRAM && nvbo->tile_flags) {
+ if (new_mem->mem_type == TTM_PL_VRAM &&
+ nouveau_bo_tile_layout(nvbo)) {
ret = RING_SPACE(chan, 8);
if (ret)
return ret;
BEGIN_RING(chan, NvSubM2MF, 0x0200, 1);
OUT_RING (chan, 1);
}
- if (old_mem->mem_type == TTM_PL_VRAM && nvbo->tile_flags) {
+ if (old_mem->mem_type == TTM_PL_VRAM &&
+ nouveau_bo_tile_layout(nvbo)) {
ret = RING_SPACE(chan, 8);
if (ret)
return ret;
int ret;
chan = nvbo->channel;
- if (!chan || nvbo->no_vm)
+ if (!chan || nvbo->no_vm) {
chan = dev_priv->channel;
+ mutex_lock_nested(&chan->mutex, NOUVEAU_KCHANNEL_MUTEX);
+ }
if (dev_priv->card_type < NV_50)
ret = nv04_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
else
ret = nv50_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
- if (ret)
- return ret;
+ if (ret == 0) {
+ ret = nouveau_bo_move_accel_cleanup(chan, nvbo, evict,
+ no_wait_reserve,
+ no_wait_gpu, new_mem);
+ }
- return nouveau_bo_move_accel_cleanup(chan, nvbo, evict, no_wait_reserve, no_wait_gpu, new_mem);
+ if (chan == dev_priv->channel)
+ mutex_unlock(&chan->mutex);
+ return ret;
}
static int
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,
+ new_mem->size,
+ nouveau_bo_tile_layout(nvbo),
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);
+ nvbo->tile_mode,
+ nvbo->tile_flags);
}
return 0;
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);
-
+ nv10_mem_put_tile_region(dev, *old_tile, bo->sync_obj);
*old_tile = new_tile;
}
}
* nothing to do here.
*/
if (bo->mem.mem_type != TTM_PL_VRAM) {
- if (dev_priv->card_type < NV_50 || !nvbo->tile_flags)
+ if (dev_priv->card_type < NV_50 ||
+ !nouveau_bo_tile_layout(nvbo))
return 0;
}
return ttm_bo_validate(bo, &nvbo->placement, false, true, false);
}
+void
+nouveau_bo_fence(struct nouveau_bo *nvbo, struct nouveau_fence *fence)
+{
+ struct nouveau_fence *old_fence;
+
+ if (likely(fence))
+ nouveau_fence_ref(fence);
+
+ spin_lock(&nvbo->bo.bdev->fence_lock);
+ old_fence = nvbo->bo.sync_obj;
+ nvbo->bo.sync_obj = fence;
+ spin_unlock(&nvbo->bo.bdev->fence_lock);
+
+ nouveau_fence_unref(&old_fence);
+}
+
struct ttm_bo_driver nouveau_bo_driver = {
.create_ttm_backend_entry = nouveau_bo_create_ttm_backend_entry,
.invalidate_caches = nouveau_bo_invalidate_caches,
.evict_flags = nouveau_bo_evict_flags,
.move = nouveau_bo_move,
.verify_access = nouveau_bo_verify_access,
- .sync_obj_signaled = nouveau_fence_signalled,
- .sync_obj_wait = nouveau_fence_wait,
- .sync_obj_flush = nouveau_fence_flush,
- .sync_obj_unref = nouveau_fence_unref,
- .sync_obj_ref = nouveau_fence_ref,
+ .sync_obj_signaled = __nouveau_fence_signalled,
+ .sync_obj_wait = __nouveau_fence_wait,
+ .sync_obj_flush = __nouveau_fence_flush,
+ .sync_obj_unref = __nouveau_fence_unref,
+ .sync_obj_ref = __nouveau_fence_ref,
.fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
.io_mem_reserve = &nouveau_ttm_io_mem_reserve,
.io_mem_free = &nouveau_ttm_io_mem_free,