[PATCH 3/3] drm/ttm: add transparent huge page support for DMA allocations
Alex Deucher
alexdeucher at gmail.com
Fri Jul 14 16:24:49 UTC 2017
On Thu, Jul 13, 2017 at 9:56 AM, Christian König
<deathsimple at vodafone.de> wrote:
> From: Christian König <christian.koenig at amd.com>
>
> Try to allocate huge pages when it makes sense.
>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
> drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 217 ++++++++++++++++++++++++-------
> 1 file changed, 169 insertions(+), 48 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> index 2081e20..e51d3fd 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> @@ -57,21 +57,25 @@
> #define NUM_PAGES_TO_ALLOC (PAGE_SIZE/sizeof(struct page *))
> #define SMALL_ALLOCATION 4
> #define FREE_ALL_PAGES (~0U)
> +#define VADDR_FLAG_HUGE_POOL 1UL
>
> enum pool_type {
> IS_UNDEFINED = 0,
> IS_WC = 1 << 1,
> IS_UC = 1 << 2,
> IS_CACHED = 1 << 3,
> - IS_DMA32 = 1 << 4
> + IS_DMA32 = 1 << 4,
> + IS_HUGE = 1 << 5
> };
>
> /*
> - * The pool structure. There are usually six pools:
> + * The pool structure. There are up to seven pools:
> * - generic (not restricted to DMA32):
> * - write combined, uncached, cached.
> * - dma32 (up to 2^32 - so up 4GB):
> * - write combined, uncached, cached.
> + * - huge (not restricted to DMA32):
> + * - cached.
Any need for uncached or write combined huge pages?
> * for each 'struct device'. The 'cached' is for pages that are actively used.
> * The other ones can be shrunk by the shrinker API if neccessary.
> * @pools: The 'struct device->dma_pools' link.
> @@ -111,13 +115,14 @@ struct dma_pool {
> * The accounting page keeping track of the allocated page along with
> * the DMA address.
> * @page_list: The link to the 'page_list' in 'struct dma_pool'.
> - * @vaddr: The virtual address of the page
> + * @vaddr: The virtual address of the page and a flag if the page belongs to a
> + * huge pool
> * @dma: The bus address of the page. If the page is not allocated
> * via the DMA API, it will be -1.
> */
> struct dma_page {
> struct list_head page_list;
> - void *vaddr;
> + unsigned long vaddr;
> struct page *p;
> dma_addr_t dma;
> };
> @@ -316,7 +321,8 @@ static int ttm_set_pages_caching(struct dma_pool *pool,
> static void __ttm_dma_free_page(struct dma_pool *pool, struct dma_page *d_page)
> {
> dma_addr_t dma = d_page->dma;
> - dma_free_coherent(pool->dev, pool->size, d_page->vaddr, dma);
> + d_page->vaddr &= ~VADDR_FLAG_HUGE_POOL;
> + dma_free_coherent(pool->dev, pool->size, (void *)d_page->vaddr, dma);
>
> kfree(d_page);
> d_page = NULL;
> @@ -324,19 +330,22 @@ static void __ttm_dma_free_page(struct dma_pool *pool, struct dma_page *d_page)
> static struct dma_page *__ttm_dma_alloc_page(struct dma_pool *pool)
> {
> struct dma_page *d_page;
> + void *vaddr;
>
> d_page = kmalloc(sizeof(struct dma_page), GFP_KERNEL);
> if (!d_page)
> return NULL;
>
> - d_page->vaddr = dma_alloc_coherent(pool->dev, pool->size,
> - &d_page->dma,
> - pool->gfp_flags);
> - if (d_page->vaddr) {
> - if (is_vmalloc_addr(d_page->vaddr))
> - d_page->p = vmalloc_to_page(d_page->vaddr);
> + vaddr = dma_alloc_coherent(pool->dev, pool->size, &d_page->dma,
> + pool->gfp_flags);
> + if (vaddr) {
> + if (is_vmalloc_addr(vaddr))
> + d_page->p = vmalloc_to_page(vaddr);
> else
> - d_page->p = virt_to_page(d_page->vaddr);
> + d_page->p = virt_to_page(vaddr);
> + d_page->vaddr = (unsigned long)vaddr;
> + if (pool->type & IS_HUGE)
> + d_page->vaddr |= VADDR_FLAG_HUGE_POOL;
> } else {
> kfree(d_page);
> d_page = NULL;
> @@ -368,11 +377,40 @@ static void ttm_pool_update_free_locked(struct dma_pool *pool,
> }
>
> /* set memory back to wb and free the pages. */
> +static void ttm_dma_page_put(struct dma_pool *pool, struct dma_page *d_page)
> +{
> + struct page *page = d_page->p;
> + unsigned i, num_pages;
> + int ret;
> +
> + /* Don't set WB on WB page pool. */
> + if (!(pool->type & IS_CACHED)) {
> + num_pages = pool->size / PAGE_SIZE;
> + for (i = 0; i < num_pages; ++i, ++page) {
> + ret = set_pages_array_wb(&page, 1);
> + if (ret) {
> + pr_err("%s: Failed to set %d pages to wb!\n",
> + pool->dev_name, 1);
> + }
> + }
> + }
> +
> + list_del(&d_page->page_list);
> + __ttm_dma_free_page(pool, d_page);
> +}
> +
> static void ttm_dma_pages_put(struct dma_pool *pool, struct list_head *d_pages,
> struct page *pages[], unsigned npages)
> {
> struct dma_page *d_page, *tmp;
>
> + if (pool->type & IS_HUGE) {
> + list_for_each_entry_safe(d_page, tmp, d_pages, page_list)
> + ttm_dma_page_put(pool, d_page);
> +
> + return;
> + }
> +
> /* Don't set WB on WB page pool. */
> if (npages && !(pool->type & IS_CACHED) &&
> set_pages_array_wb(pages, npages))
> @@ -385,17 +423,6 @@ 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)
> -{
> - /* Don't set WB on WB page pool. */
> - if (!(pool->type & IS_CACHED) && set_pages_array_wb(&d_page->p, 1))
> - pr_err("%s: Failed to set %d pages to wb!\n",
> - pool->dev_name, 1);
> -
> - list_del(&d_page->page_list);
> - __ttm_dma_free_page(pool, d_page);
> -}
> -
> /*
> * Free pages from pool.
> *
> @@ -564,8 +591,8 @@ static int ttm_dma_pool_match(struct device *dev, void *res, void *match_data)
> static struct dma_pool *ttm_dma_pool_init(struct device *dev, gfp_t flags,
> enum pool_type type)
> {
> - char *n[] = {"wc", "uc", "cached", " dma32", "unknown",};
> - enum pool_type t[] = {IS_WC, IS_UC, IS_CACHED, IS_DMA32, IS_UNDEFINED};
> + const char *n[] = {"wc", "uc", "cached", " dma32", "huge"};
> + enum pool_type t[] = {IS_WC, IS_UC, IS_CACHED, IS_DMA32, IS_HUGE};
> struct device_pools *sec_pool = NULL;
> struct dma_pool *pool = NULL, **ptr;
> unsigned i;
> @@ -602,11 +629,18 @@ static struct dma_pool *ttm_dma_pool_init(struct device *dev, gfp_t flags,
> pool->npages_free = pool->npages_in_use = 0;
> pool->nfrees = 0;
> pool->gfp_flags = flags;
> - pool->size = PAGE_SIZE;
> + if (type & IS_HUGE)
> +#if CONFIG_TRANSPARENT_HUGEPAGE
I think this should be #ifdef for consistency with the HUGEPAGE
support in the rest of the kernel.
> + pool->size = HPAGE_PMD_SIZE;
> +#else
> + BUG();
> +#endif
> + else
> + pool->size = PAGE_SIZE;
> pool->type = type;
> pool->nrefills = 0;
> p = pool->name;
> - for (i = 0; i < 5; i++) {
> + for (i = 0; i < ARRAY_SIZE(t); i++) {
> if (type & t[i]) {
> p += snprintf(p, sizeof(pool->name) - (p - pool->name),
> "%s", n[i]);
> @@ -710,7 +744,7 @@ static int ttm_dma_pool_alloc_new_pages(struct dma_pool *pool,
> struct dma_page *dma_p;
> struct page *p;
> int r = 0;
> - unsigned i, cpages;
> + unsigned i, j, npages, cpages;
> unsigned max_cpages = min(count,
> (unsigned)(PAGE_SIZE/sizeof(struct page *)));
>
> @@ -748,28 +782,32 @@ static int ttm_dma_pool_alloc_new_pages(struct dma_pool *pool,
> goto out;
> }
> p = dma_p->p;
> + list_add(&dma_p->page_list, d_pages);
> +
> #ifdef CONFIG_HIGHMEM
> /* gfp flags of highmem page should never be dma32 so we
> * we should be fine in such case
> */
> - if (!PageHighMem(p))
> + if (PageHighMem(p))
> + continue;
> #endif
> - {
> - caching_array[cpages++] = p;
> +
> + npages = pool->size / PAGE_SIZE;
> + for (j = 0; j < npages; ++j) {
> + caching_array[cpages++] = p + j;
> if (cpages == max_cpages) {
> /* Note: Cannot hold the spinlock */
> r = ttm_set_pages_caching(pool, caching_array,
> - cpages);
> + cpages);
> if (r) {
> ttm_dma_handle_caching_state_failure(
> - pool, d_pages, caching_array,
> - cpages);
> + pool, d_pages, caching_array,
> + cpages);
> goto out;
> }
> cpages = 0;
> }
> }
> - list_add(&dma_p->page_list, d_pages);
> }
>
> if (cpages) {
> @@ -857,6 +895,26 @@ static int ttm_dma_pool_get_pages(struct dma_pool *pool,
> return r;
> }
>
> +static gfp_t ttm_dma_pool_gfp_flags(struct ttm_dma_tt *ttm_dma, bool huge)
> +{
> + struct ttm_tt *ttm = &ttm_dma->ttm;
> + gfp_t gfp_flags;
> +
> + if (ttm->page_flags & TTM_PAGE_FLAG_DMA32)
> + gfp_flags = GFP_USER | GFP_DMA32;
> + else
> + gfp_flags = GFP_HIGHUSER;
> + if (ttm->page_flags & TTM_PAGE_FLAG_ZERO_ALLOC)
> + gfp_flags |= __GFP_ZERO;
> +
> + if (huge) {
> + gfp_flags |= GFP_TRANSHUGE;
> + gfp_flags &= ~__GFP_MOVABLE;
> + }
> +
> + return gfp_flags;
> +}
> +
> /*
> * On success pages list will hold count number of correctly
> * cached pages. On failure will hold the negative return value (-ENOMEM, etc).
> @@ -865,6 +923,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
> {
> struct ttm_tt *ttm = &ttm_dma->ttm;
> struct ttm_mem_global *mem_glob = ttm->glob->mem_glob;
> + unsigned long num_pages = ttm->num_pages;
> struct dma_pool *pool;
> enum pool_type type;
> unsigned i;
> @@ -873,26 +932,61 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
> if (ttm->state != tt_unpopulated)
> return 0;
>
> + INIT_LIST_HEAD(&ttm_dma->pages_list);
> + i = 0;
> +
> type = ttm_to_type(ttm->page_flags, ttm->caching_state);
> - pool = ttm_dma_find_pool(dev, type);
> +
> +#if CONFIG_TRANSPARENT_HUGEPAGE
Same here.
> + if (ttm->page_flags & TTM_PAGE_FLAG_DMA32)
> + goto skip_huge;
> +
> + pool = ttm_dma_find_pool(dev, type | IS_HUGE);
> if (!pool) {
> - gfp_t gfp_flags;
> + gfp_t gfp_flags = ttm_dma_pool_gfp_flags(ttm_dma, true);
>
> - if (ttm->page_flags & TTM_PAGE_FLAG_DMA32)
> - gfp_flags = GFP_USER | GFP_DMA32;
> - else
> - gfp_flags = GFP_HIGHUSER;
> - if (ttm->page_flags & TTM_PAGE_FLAG_ZERO_ALLOC)
> - gfp_flags |= __GFP_ZERO;
> + pool = ttm_dma_pool_init(dev, gfp_flags, type | IS_HUGE);
> + if (IS_ERR_OR_NULL(pool))
> + goto skip_huge;
> + }
>
> - pool = ttm_dma_pool_init(dev, gfp_flags, type);
> - if (IS_ERR_OR_NULL(pool)) {
> + while (num_pages >= HPAGE_PMD_NR) {
> + unsigned j;
> +
> + ret = ttm_dma_pool_get_pages(pool, ttm_dma, i);
> + if (ret != 0)
> + break;
> +
> + ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
> + pool->size);
> + if (unlikely(ret != 0)) {
> + ttm_dma_unpopulate(ttm_dma, dev);
> return -ENOMEM;
> }
> +
> + for (j = i + 1; j < (i + HPAGE_PMD_NR); ++j) {
> + ttm->pages[j] = ttm->pages[j - 1] + 1;
> + ttm_dma->dma_address[j] = ttm_dma->dma_address[j - 1] +
> + PAGE_SIZE;
> + }
> +
> + i += HPAGE_PMD_NR;
> + num_pages -= HPAGE_PMD_NR;
> }
>
> - INIT_LIST_HEAD(&ttm_dma->pages_list);
> - for (i = 0; i < ttm->num_pages; ++i) {
> +skip_huge:
> +#endif
> +
> + pool = ttm_dma_find_pool(dev, type);
> + if (!pool) {
> + gfp_t gfp_flags = ttm_dma_pool_gfp_flags(ttm_dma, false);
> +
> + pool = ttm_dma_pool_init(dev, gfp_flags, type);
> + if (IS_ERR_OR_NULL(pool))
> + return -ENOMEM;
> + }
> +
> + while (num_pages) {
> ret = ttm_dma_pool_get_pages(pool, ttm_dma, i);
> if (ret != 0) {
> ttm_dma_unpopulate(ttm_dma, dev);
> @@ -905,6 +999,9 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev)
> ttm_dma_unpopulate(ttm_dma, dev);
> return -ENOMEM;
> }
> +
> + ++i;
> + --num_pages;
> }
>
> if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) {
> @@ -928,10 +1025,33 @@ 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, npages = 0;
> + unsigned count, i, npages = 0;
> unsigned long irq_flags;
>
> type = ttm_to_type(ttm->page_flags, ttm->caching_state);
> +
> +#if CONFIG_TRANSPARENT_HUGEPAGE
Same here.
> + pool = ttm_dma_find_pool(dev, type | IS_HUGE);
> + if (pool) {
> + count = 0;
> + list_for_each_entry_safe(d_page, next, &ttm_dma->pages_list,
> + page_list) {
> + if (!(d_page->vaddr & VADDR_FLAG_HUGE_POOL))
> + continue;
> +
> + count++;
> + ttm_mem_global_free_page(ttm->glob->mem_glob,
> + d_page->p, pool->size);
> + ttm_dma_page_put(pool, d_page);
> + }
> +
> + spin_lock_irqsave(&pool->lock, irq_flags);
> + pool->npages_in_use -= count;
> + pool->nfrees += count;
> + spin_unlock_irqrestore(&pool->lock, irq_flags);
> + }
> +#endif
> +
> pool = ttm_dma_find_pool(dev, type);
> if (!pool)
> return;
> @@ -940,6 +1060,7 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
> ttm_to_type(ttm->page_flags, tt_cached)) == pool);
>
> /* make sure pages array match list and count number of pages */
> + count = 0;
> list_for_each_entry(d_page, &ttm_dma->pages_list, page_list) {
> ttm->pages[count] = d_page->p;
> count++;
> --
> 2.7.4
>
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
More information about the dri-devel
mailing list