Otherwise we are doing redundant work. Especially since the 'unbind' and 'unpopulate' have been merged and nouveau driver ends up calling it quite excessivly. On a GeForce 8600 GT with Gnome Shell (GNOME 3) we end up spending about 54% CPU time in __change_page_attr_set_clr checking the page flags.

The callgraph (annotated) looks as so before this patch:

53.29% gnome-shell [kernel.kallsyms] [k] static_protections | --- static_protections | |--91.80%-- __change_page_attr_set_clr | change_page_attr_set_clr | set_pages_array_wb | | | |--96.55%-- ttm_dma_unpopulate | | nouveau_ttm_tt_unpopulate | | ttm_tt_destroy | | ttm_bo_cleanup_memtype_use | | ttm_bo_release | | kref_put | | ttm_bo_unref | | nouveau_gem_object_del | | drm_gem_object_free | | kref_put | | drm_gem_object_unreference_unlocked | | drm_gem_object_handle_unreference_unlocked.part.1 | | drm_gem_handle_delete | | drm_gem_close_ioctl | | drm_ioctl | | do_vfs_ioctl | | sys_ioctl | | system_call_fastpath | | __GI___ioctl | | | --3.45%-- ttm_dma_pages_put | ttm_dma_page_pool_free | ttm_dma_unpopulate | nouveau_ttm_tt_unpopulate | ttm_tt_destroy | ttm_bo_cleanup_memtype_use | ttm_bo_release | kref_put | ttm_bo_unref | nouveau_gem_object_del | drm_gem_object_free | kref_put | drm_gem_object_unreference_unlocked | drm_gem_object_handle_unreference_unlocked.part.1 | drm_gem_handle_delete | drm_gem_close_ioctl | drm_ioctl | do_vfs_ioctl | sys_ioctl | system_call_fastpath | __GI___ioctl | --8.20%-- change_page_attr_set_clr set_pages_array_wb | |--93.76%-- ttm_dma_unpopulate | nouveau_ttm_tt_unpopulate | ttm_tt_destroy | ttm_bo_cleanup_memtype_use | ttm_bo_release | kref_put | ttm_bo_unref | nouveau_gem_object_del | drm_gem_object_free | kref_put | drm_gem_object_unreference_unlocked | drm_gem_object_handle_unreference_unlocked.part.1 | drm_gem_handle_delete | drm_gem_close_ioctl | drm_ioctl | do_vfs_ioctl | sys_ioctl | system_call_fastpath | __GI___ioctl | --6.24%-- ttm_dma_pages_put ttm_dma_page_pool_free ttm_dma_unpopulate nouveau_ttm_tt_unpopulate ttm_tt_destroy ttm_bo_cleanup_memtype_use ttm_bo_release kref_put ttm_bo_unref nouveau_gem_object_del drm_gem_object_free kref_put drm_gem_object_unreference_unlocked drm_gem_object_handle_unreference_unlocked.part.1 drm_gem_handle_delete drm_gem_close_ioctl drm_ioctl do_vfs_ioctl sys_ioctl system_call_fastpath __GI___ioctl

and after this patch all of that disappears.

Signed-off-by: Konrad Rzeszutek Wilk konrad.wilk@oracle.com --- drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 7 +++++-- 1 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c index 6678abc..6c06d0b 100644 --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c @@ -384,7 +384,9 @@ static void ttm_dma_pages_put(struct dma_pool *pool, struct list_head *d_pages, { struct dma_page *d_page, *tmp;

- if (npages && set_pages_array_wb(pages, npages)) + /* Don't set WB on WB page pool. */ + if (npages && !(pool->type & IS_CACHED) && + set_pages_array_wb(pages, npages)) pr_err(TTM_PFX "%s: Failed to set %d pages to wb!\n", pool->dev_name, npages);

@@ -396,7 +398,8 @@ static void ttm_dma_pages_put(struct dma_pool *pool, struct list_head *d_pages,

static void ttm_dma_page_put(struct dma_pool *pool, struct dma_page *d_page) { - if (set_pages_array_wb(&d_page->p, 1)) + /* Don't set WB on WB page pool. */ + if (!(pool->type & IS_CACHED) && set_pages_array_wb(&d_page->p, 1)) pr_err(TTM_PFX "%s: Failed to set %d pages to wb!\n", pool->dev_name, 1);

The code to figure out how many pages to shrink the pool ends up capping the 'count' at _manager->options.max_size - which is OK. Except that the 'count' is also used when accounting for how many pages are recycled - which we end up with the invalid values. This fixes it by using a different value for the amount of pages to shrink.

On top of that we would free the cached page pool - which is nonsense as they are deleted from the pool - so there are no free pages in that pool..

Also we also missed the opportunity to batch the amount of pages to free (similar to how ttm_page_alloc.c does it). This reintroduces the code that was lost during rebasing.

Reviewed-by: Jerome Glisse jglisse@redhat.com Signed-off-by: Konrad Rzeszutek Wilk konrad.wilk@oracle.com --- drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 15 ++++++++++----- 1 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c index 6c06d0b..156ddcd 100644 --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c @@ -949,7 +949,7 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev) struct dma_page *d_page, *next; enum pool_type type; bool is_cached = false; - unsigned count = 0, i; + unsigned count = 0, i, npages = 0; unsigned long irq_flags;

type = ttm_to_type(ttm->page_flags, ttm->caching_state); @@ -974,8 +974,13 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev) } else { pool->npages_free += count; list_splice(&ttm_dma->pages_list, &pool->free_list); + npages = count; if (pool->npages_free > _manager->options.max_size) { - count = pool->npages_free - _manager->options.max_size; + npages = pool->npages_free - _manager->options.max_size; + /* free at least NUM_PAGES_TO_ALLOC number of pages + * to reduce calls to set_memory_wb */ + if (npages < NUM_PAGES_TO_ALLOC) + npages = NUM_PAGES_TO_ALLOC; } } spin_unlock_irqrestore(&pool->lock, irq_flags); @@ -999,9 +1004,9 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev) ttm_dma->dma_address[i] = 0; }

- /* shrink pool if necessary */ - if (count) - ttm_dma_page_pool_free(pool, count); + /* shrink pool if necessary (only on !is_cached pools)*/ + if (npages) + ttm_dma_page_pool_free(pool, npages); ttm->state = tt_unpopulated; } EXPORT_SYMBOL_GPL(ttm_dma_unpopulate);