Skip to contentSkip to conversation

drm/ttm: don't use compound pages for now

PATCH In review Christian König 3 participants 5 messages Oct 10, 2017 2 messages not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/4 drm/ttm: don't use compound pages for now

From: Christian König <[email protected]>

We need to figure out first how to correctly map them into the CPU page tables.

---
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/ttm/ttm_page_alloc_dma.c +1 -0
index e5ef10d..96ad129 100644
@@ -913,6 +913,7 @@ static gfp_t ttm_dma_pool_gfp_flags(struct ttm_dma_tt *ttm_dma, bool huge)
 	if (huge) {
 		gfp_flags |= GFP_TRANSHUGE;
 		gfp_flags &= ~__GFP_MOVABLE;
+		gfp_flags &= ~__GFP_COMP;
 	}
 
 	return gfp_flags;
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 1507625625-8665-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org

#2 lore raw

PATCH 2/4 drm/ttm: add transparent huge page support for cached allocations v2

From: Christian König <[email protected]>

Try to allocate huge pages when it makes sense.

v2: avoid compound pages for now

---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 50 ++++++++++++++++++++++++++++++------
 1 file changed, 42 insertions(+), 8 deletions(-)
drivers/gpu/drm/ttm/ttm_page_alloc.c +42 -8
index 6c852e8..1bc6053 100644
@@ -685,12 +685,24 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 
 	if (pool == NULL) {
 		/* No pool for this memory type so free the pages */
-		for (i = 0; i < npages; i++) {
-			if (pages[i]) {
-				if (page_count(pages[i]) != 1)
-					pr_err("Erroneous page count. Leaking pages.\n");
-				__free_page(pages[i]);
-				pages[i] = NULL;
+		i = 0;
+		while (i < npages) {
+			unsigned order;
+
+			if (!pages[i]) {
+				++i;
+				continue;
+			}
+
+			if (page_count(pages[i]) != 1)
+				pr_err("Erroneous page count. Leaking pages.\n");
+			order = compound_order(pages[i]);
+			__free_pages(pages[i], order);
+
+			order = 1 << order;
+			while (order) {
+				pages[i++] = NULL;
+				--order;
 			}
 		}
 		return;
@@ -740,12 +752,33 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 
 	/* No pool for cached pages */
 	if (pool == NULL) {
+		unsigned i, j;
+
 		if (flags & TTM_PAGE_FLAG_DMA32)
 			gfp_flags |= GFP_DMA32;
 		else
 			gfp_flags |= GFP_HIGHUSER;
 
-		for (r = 0; r < npages; ++r) {
+		i = 0;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+		while (npages >= HPAGE_PMD_NR) {
+			gfp_t huge_flags = gfp_flags;
+
+			huge_flags |= GFP_TRANSHUGE;
+			huge_flags &= ~__GFP_MOVABLE;
+			huge_flags &= ~__GFP_COMP;
+			p = alloc_pages(huge_flags, HPAGE_PMD_ORDER);
+			if (!p)
+				break;
+
+			for (j = 0; j < HPAGE_PMD_NR; ++j)
+				pages[i++] = p++;
+
+			npages -= HPAGE_PMD_NR;
+		}
+#endif
+
+		while (npages) {
 			p = alloc_page(gfp_flags);
 			if (!p) {
 
@@ -753,7 +786,8 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 				return -ENOMEM;
 			}
 
-			pages[r] = p;
+			pages[i++] = p;
+			--npages;
 		}
 		return 0;
 	}
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw

PATCH 3/4 drm/ttm: move more logic into ttm_page_pool_get_pages

In reply to a message not in the archive

From: Christian König <[email protected]>

Make it easier to add huge page pool.

---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 98 +++++++++++++++++++-----------------
 1 file changed, 52 insertions(+), 46 deletions(-)
drivers/gpu/drm/ttm/ttm_page_alloc.c +52 -46
index 1bc6053..3974732 100644
@@ -627,19 +627,20 @@ static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool,
 }
 
 /**
- * Cut 'count' number of pages from the pool and put them on the return list.
+ * Allocate pages from the pool and put them on the return list.
  *
- * @return count of pages still required to fulfill the request.
+ * @return zero for success or negative error code.
  */
-static unsigned ttm_page_pool_get_pages(struct ttm_page_pool *pool,
-					struct list_head *pages,
-					int ttm_flags,
-					enum ttm_caching_state cstate,
-					unsigned count)
+static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
+				   struct list_head *pages,
+				   int ttm_flags,
+				   enum ttm_caching_state cstate,
+				   unsigned count)
 {
 	unsigned long irq_flags;
 	struct list_head *p;
 	unsigned i;
+	int r = 0;
 
 	spin_lock_irqsave(&pool->lock, irq_flags);
 	ttm_page_pool_fill_locked(pool, ttm_flags, cstate, count, &irq_flags);
@@ -672,7 +673,35 @@ static unsigned ttm_page_pool_get_pages(struct ttm_page_pool *pool,
 	count = 0;
 out:
 	spin_unlock_irqrestore(&pool->lock, irq_flags);
-	return count;
+
+	/* clear the pages coming from the pool if requested */
+	if (ttm_flags & TTM_PAGE_FLAG_ZERO_ALLOC) {
+		struct page *page;
+
+		list_for_each_entry(page, pages, lru) {
+			if (PageHighMem(page))
+				clear_highpage(page);
+			else
+				clear_page(page_address(page));
+		}
+	}
+
+	/* If pool didn't have enough pages allocate new one. */
+	if (count) {
+		gfp_t gfp_flags = pool->gfp_flags;
+
+		/* set zero flag for page allocation if required */
+		if (ttm_flags & TTM_PAGE_FLAG_ZERO_ALLOC)
+			gfp_flags |= __GFP_ZERO;
+
+		/* ttm_alloc_new_pages doesn't reference pool so we can run
+		 * multiple requests in parallel.
+		 **/
+		r = ttm_alloc_new_pages(pages, gfp_flags, ttm_flags, cstate,
+					count);
+	}
+
+	return r;
 }
 
 /* Put all pages in pages list to correct pool to wait for reuse */
@@ -742,18 +771,18 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 	struct ttm_page_pool *pool = ttm_get_pool(flags, cstate);
 	struct list_head plist;
 	struct page *p = NULL;
-	gfp_t gfp_flags = GFP_USER;
 	unsigned count;
 	int r;
 
-	/* set zero flag for page allocation if required */
-	if (flags & TTM_PAGE_FLAG_ZERO_ALLOC)
-		gfp_flags |= __GFP_ZERO;
-
 	/* No pool for cached pages */
 	if (pool == NULL) {
+		gfp_t gfp_flags = GFP_USER;
 		unsigned i, j;
 
+		/* set zero flag for page allocation if required */
+		if (flags & TTM_PAGE_FLAG_ZERO_ALLOC)
+			gfp_flags |= __GFP_ZERO;
+
 		if (flags & TTM_PAGE_FLAG_DMA32)
 			gfp_flags |= GFP_DMA32;
 		else
@@ -792,44 +821,21 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 		return 0;
 	}
 
-	/* combine zero flag to pool flags */
-	gfp_flags |= pool->gfp_flags;
-
 	/* First we take pages from the pool */
 	INIT_LIST_HEAD(&plist);
-	npages = ttm_page_pool_get_pages(pool, &plist, flags, cstate, npages);
+	r = ttm_page_pool_get_pages(pool, &plist, flags, cstate, npages);
+
 	count = 0;
-	list_for_each_entry(p, &plist, lru) {
+	list_for_each_entry(p, &plist, lru)
 		pages[count++] = p;
-	}
-
-	/* clear the pages coming from the pool if requested */
-	if (flags & TTM_PAGE_FLAG_ZERO_ALLOC) {
-		list_for_each_entry(p, &plist, lru) {
-			if (PageHighMem(p))
-				clear_highpage(p);
-			else
-				clear_page(page_address(p));
-		}
-	}
 
-	/* If pool didn't have enough pages allocate new one. */
-	if (npages > 0) {
-		/* ttm_alloc_new_pages doesn't reference pool so we can run
-		 * multiple requests in parallel.
-		 **/
-		INIT_LIST_HEAD(&plist);
-		r = ttm_alloc_new_pages(&plist, gfp_flags, flags, cstate, npages);
-		list_for_each_entry(p, &plist, lru) {
-			pages[count++] = p;
-		}
-		if (r) {
-			/* If there is any pages in the list put them back to
-			 * the pool. */
-			pr_err("Failed to allocate extra pages for large request\n");
-			ttm_put_pages(pages, count, flags, cstate);
-			return r;
-		}
+	if (r) {
+		/* If there is any pages in the list put them back to
+		 * the pool.
+		 */
+		pr_err("Failed to allocate extra pages for large request\n");
+		ttm_put_pages(pages, count, flags, cstate);
+		return r;
 	}
 
 	return 0;
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#4 lore raw

PATCH 4/4 drm/ttm: add transparent huge page support for wc or uc allocations v2

In reply to Christian König

From: Christian König <[email protected]>

Add a new huge page pool and try to allocate from it when it makes sense.

v2: avoid compound pages for now

---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 136 ++++++++++++++++++++++++++++-------
 1 file changed, 109 insertions(+), 27 deletions(-)
drivers/gpu/drm/ttm/ttm_page_alloc.c +109 -27
index 3974732..b6f16e7ff 100644
@@ -95,7 +95,7 @@ struct ttm_pool_opts {
 	unsigned	small;
 };
 
-#define NUM_POOLS 4
+#define NUM_POOLS 6
 
 /**
  * struct ttm_pool_manager - Holds memory pools for fst allocation
@@ -122,6 +122,8 @@ struct ttm_pool_manager {
 			struct ttm_page_pool	uc_pool;
 			struct ttm_page_pool	wc_pool_dma32;
 			struct ttm_page_pool	uc_pool_dma32;
+			struct ttm_page_pool	wc_pool_huge;
+			struct ttm_page_pool	uc_pool_huge;
 		} ;
 	};
 };
@@ -256,8 +258,8 @@ static int set_pages_array_uc(struct page **pages, int addrinarray)
 
 /**
  * Select the right pool or requested caching state and ttm flags. */
-static struct ttm_page_pool *ttm_get_pool(int flags,
-		enum ttm_caching_state cstate)
+static struct ttm_page_pool *ttm_get_pool(int flags, bool huge,
+					  enum ttm_caching_state cstate)
 {
 	int pool_index;
 
@@ -269,9 +271,15 @@ static struct ttm_page_pool *ttm_get_pool(int flags,
 	else
 		pool_index = 0x1;
 
-	if (flags & TTM_PAGE_FLAG_DMA32)
+	if (flags & TTM_PAGE_FLAG_DMA32) {
+		if (huge)
+			return NULL;
 		pool_index |= 0x2;
 
+	} else if (huge) {
+		pool_index |= 0x4;
+	}
+
 	return &_manager->pools[pool_index];
 }
 
@@ -494,12 +502,14 @@ static void ttm_handle_caching_state_failure(struct list_head *pages,
  * pages returned in pages array.
  */
 static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
-		int ttm_flags, enum ttm_caching_state cstate, unsigned count)
+			       int ttm_flags, enum ttm_caching_state cstate,
+			       unsigned count, unsigned order)
 {
 	struct page **caching_array;
 	struct page *p;
 	int r = 0;
-	unsigned i, cpages;
+	unsigned i, j, cpages;
+	unsigned npages = 1 << order;
 	unsigned max_cpages = min(count,
 			(unsigned)(PAGE_SIZE/sizeof(struct page *)));
 
@@ -512,7 +522,7 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
 	}
 
 	for (i = 0, cpages = 0; i < count; ++i) {
-		p = alloc_page(gfp_flags);
+		p = alloc_pages(gfp_flags, order);
 
 		if (!p) {
 			pr_err("Unable to get page %u\n", i);
@@ -531,14 +541,18 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
 			goto out;
 		}
 
+		list_add(&p->lru, 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;
+		for (j = 0; j < npages; ++j) {
+			caching_array[cpages++] = p++;
 			if (cpages == max_cpages) {
 
 				r = ttm_set_pages_caching(caching_array,
@@ -552,8 +566,6 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
 				cpages = 0;
 			}
 		}
-
-		list_add(&p->lru, pages);
 	}
 
 	if (cpages) {
@@ -573,9 +585,9 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
  * Fill the given pool if there aren't enough pages and the requested number of
  * pages is small.
  */
-static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool,
-		int ttm_flags, enum ttm_caching_state cstate, unsigned count,
-		unsigned long *irq_flags)
+static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool, int ttm_flags,
+				      enum ttm_caching_state cstate,
+				      unsigned count, unsigned long *irq_flags)
 {
 	struct page *p;
 	int r;
@@ -605,7 +617,7 @@ static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool,
 
 		INIT_LIST_HEAD(&new_pages);
 		r = ttm_alloc_new_pages(&new_pages, pool->gfp_flags, ttm_flags,
-				cstate,	alloc_size);
+					cstate, alloc_size, 0);
 		spin_lock_irqsave(&pool->lock, *irq_flags);
 
 		if (!r) {
@@ -635,7 +647,7 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
 				   struct list_head *pages,
 				   int ttm_flags,
 				   enum ttm_caching_state cstate,
-				   unsigned count)
+				   unsigned count, unsigned order)
 {
 	unsigned long irq_flags;
 	struct list_head *p;
@@ -643,7 +655,9 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
 	int r = 0;
 
 	spin_lock_irqsave(&pool->lock, irq_flags);
-	ttm_page_pool_fill_locked(pool, ttm_flags, cstate, count, &irq_flags);
+	if (!order)
+		ttm_page_pool_fill_locked(pool, ttm_flags, cstate, count,
+					  &irq_flags);
 
 	if (count >= pool->npages) {
 		/* take all pages from the pool */
@@ -698,7 +712,7 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
 		 * multiple requests in parallel.
 		 **/
 		r = ttm_alloc_new_pages(pages, gfp_flags, ttm_flags, cstate,
-					count);
+					count, order);
 	}
 
 	return r;
@@ -708,8 +722,9 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
 static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 			  enum ttm_caching_state cstate)
 {
+	struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
+	struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
 	unsigned long irq_flags;
-	struct ttm_page_pool *pool = ttm_get_pool(flags, cstate);
 	unsigned i;
 
 	if (pool == NULL) {
@@ -737,8 +752,48 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 		return;
 	}
 
+	i = 0;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if (huge) {
+		unsigned max_size, n2free;
+
+		spin_lock_irqsave(&huge->lock, irq_flags);
+		while (i < npages) {
+			struct page *p = pages[i];
+			unsigned j;
+
+			if (!p)
+				break;
+
+			for (j = 0; j < HPAGE_PMD_NR; ++j)
+				if (p++ != pages[i + j])
+				    break;
+
+			if (j != HPAGE_PMD_NR)
+				break;
+
+			list_add_tail(&pages[i]->lru, &huge->list);
+
+			for (j = 0; j < HPAGE_PMD_NR; ++j)
+				pages[i++] = NULL;
+			huge->npages++;
+		}
+
+		/* Check that we don't go over the pool limit */
+		max_size = _manager->options.max_size;
+		max_size /= HPAGE_PMD_NR;
+		if (huge->npages > max_size)
+			n2free = huge->npages - max_size;
+		else
+			n2free = 0;
+		spin_unlock_irqrestore(&huge->lock, irq_flags);
+		if (n2free)
+			ttm_page_pool_free(huge, n2free, false);
+	}
+#endif
+
 	spin_lock_irqsave(&pool->lock, irq_flags);
-	for (i = 0; i < npages; i++) {
+	while (i < npages) {
 		if (pages[i]) {
 			if (page_count(pages[i]) != 1)
 				pr_err("Erroneous page count. Leaking pages.\n");
@@ -746,6 +801,7 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 			pages[i] = NULL;
 			pool->npages++;
 		}
+		++i;
 	}
 	/* Check that we don't go over the pool limit */
 	npages = 0;
@@ -768,7 +824,8 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 			 enum ttm_caching_state cstate)
 {
-	struct ttm_page_pool *pool = ttm_get_pool(flags, cstate);
+	struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
+	struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
 	struct list_head plist;
 	struct page *p = NULL;
 	unsigned count;
@@ -821,11 +878,28 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 		return 0;
 	}
 
-	/* First we take pages from the pool */
+	count = 0;
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if (huge && npages >= HPAGE_PMD_NR) {
+		INIT_LIST_HEAD(&plist);
+		ttm_page_pool_get_pages(huge, &plist, flags, cstate,
+					npages / HPAGE_PMD_NR,
+					HPAGE_PMD_ORDER);
+
+		list_for_each_entry(p, &plist, lru) {
+			unsigned j;
+
+			for (j = 0; j < HPAGE_PMD_NR; ++j)
+				pages[count++] = &p[j];
+		}
+	}
+#endif
+
 	INIT_LIST_HEAD(&plist);
-	r = ttm_page_pool_get_pages(pool, &plist, flags, cstate, npages);
+	r = ttm_page_pool_get_pages(pool, &plist, flags, cstate,
+				    npages - count, 0);
 
-	count = 0;
 	list_for_each_entry(p, &plist, lru)
 		pages[count++] = p;
 
@@ -872,6 +946,14 @@ int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
 	ttm_page_pool_init_locked(&_manager->uc_pool_dma32,
 				  GFP_USER | GFP_DMA32, "uc dma");
 
+	ttm_page_pool_init_locked(&_manager->wc_pool_huge,
+				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP),
+				  "wc huge");
+
+	ttm_page_pool_init_locked(&_manager->uc_pool_huge,
+				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP)
+				  , "uc huge");
+
 	_manager->options.max_size = max_pages;
 	_manager->options.small = SMALL_ALLOCATION;
 	_manager->options.alloc_size = NUM_PAGES_TO_ALLOC;
@@ -1041,12 +1123,12 @@ int ttm_page_alloc_debugfs(struct seq_file *m, void *data)
 		seq_printf(m, "No pool allocator running.\n");
 		return 0;
 	}
-	seq_printf(m, "%6s %12s %13s %8s\n",
+	seq_printf(m, "%7s %12s %13s %8s\n",
 			h[0], h[1], h[2], h[3]);
 	for (i = 0; i < NUM_POOLS; ++i) {
 		p = &_manager->pools[i];
 
-		seq_printf(m, "%6s %12ld %13ld %8d\n",
+		seq_printf(m, "%7s %12ld %13ld %8d\n",
 				p->name, p->nrefills,
 				p->nfrees, p->npages);
 	}
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 1507625625-8665-4-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org

Acked-by
#6 lore raw

Re: [PATCH 4/4] drm/ttm: add transparent huge page support for wc or uc allocations v2

Quoted 8 lines · Christian König

From: Christian König <[email protected]>

Add a new huge page pool and try to allocate from it when it makes sense.

v2: avoid compound pages for now

Series is:

Quoted 306 lines
---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 136 ++++++++++++++++++++++++++++-------
 1 file changed, 109 insertions(+), 27 deletions(-)
drivers/gpu/drm/ttm/ttm_page_alloc.c +109 -28
index 3974732..b6f16e7ff 100644
@@ -95,7 +95,7 @@ struct ttm_pool_opts {
        unsigned        small;
 };
 
-#define NUM_POOLS 4
+#define NUM_POOLS 6
 
 /**
  * struct ttm_pool_manager - Holds memory pools for fst allocation
@@ -122,6 +122,8 @@ struct ttm_pool_manager {
                        struct ttm_page_pool    uc_pool;
                        struct ttm_page_pool    wc_pool_dma32;
                        struct ttm_page_pool    uc_pool_dma32;
+                       struct ttm_page_pool    wc_pool_huge;
+                       struct ttm_page_pool    uc_pool_huge;
                } ;
        };
 };
@@ -256,8 +258,8 @@ static int set_pages_array_uc(struct page **pages, int addrinarray)
 
 /**
  * Select the right pool or requested caching state and ttm flags. */
-static struct ttm_page_pool *ttm_get_pool(int flags,
-               enum ttm_caching_state cstate)
+static struct ttm_page_pool *ttm_get_pool(int flags, bool huge,
+                                         enum ttm_caching_state cstate)
 {
        int pool_index;
 
@@ -269,9 +271,15 @@ static struct ttm_page_pool *ttm_get_pool(int flags,
        else
                pool_index = 0x1;
 
-       if (flags & TTM_PAGE_FLAG_DMA32)
+       if (flags & TTM_PAGE_FLAG_DMA32) {
+               if (huge)
+                       return NULL;
                pool_index |= 0x2;
 
+       } else if (huge) {
+               pool_index |= 0x4;
+       }
+
        return &_manager->pools[pool_index];
 }
 
@@ -494,12 +502,14 @@ static void ttm_handle_caching_state_failure(struct list_head *pages,
  * pages returned in pages array.
  */
 static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
-               int ttm_flags, enum ttm_caching_state cstate, unsigned count)
+                              int ttm_flags, enum ttm_caching_state cstate,
+                              unsigned count, unsigned order)
 {
        struct page **caching_array;
        struct page *p;
        int r = 0;
-       unsigned i, cpages;
+       unsigned i, j, cpages;
+       unsigned npages = 1 << order;
        unsigned max_cpages = min(count,
                        (unsigned)(PAGE_SIZE/sizeof(struct page *)));
 
@@ -512,7 +522,7 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
        }
 
        for (i = 0, cpages = 0; i < count; ++i) {
-               p = alloc_page(gfp_flags);
+               p = alloc_pages(gfp_flags, order);
 
                if (!p) {
                        pr_err("Unable to get page %u\n", i);
@@ -531,14 +541,18 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
                        goto out;
                }
 
+               list_add(&p->lru, 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;
+               for (j = 0; j < npages; ++j) {
+                       caching_array[cpages++] = p++;
                        if (cpages == max_cpages) {
 
                                r = ttm_set_pages_caching(caching_array,
@@ -552,8 +566,6 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
                                cpages = 0;
                        }
                }
-
-               list_add(&p->lru, pages);
        }
 
        if (cpages) {
@@ -573,9 +585,9 @@ static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags,
  * Fill the given pool if there aren't enough pages and the requested number of
  * pages is small.
  */
-static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool,
-               int ttm_flags, enum ttm_caching_state cstate, unsigned count,
-               unsigned long *irq_flags)
+static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool, int ttm_flags,
+                                     enum ttm_caching_state cstate,
+                                     unsigned count, unsigned long *irq_flags)
 {
        struct page *p;
        int r;
@@ -605,7 +617,7 @@ static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool,
 
                INIT_LIST_HEAD(&new_pages);
                r = ttm_alloc_new_pages(&new_pages, pool->gfp_flags, ttm_flags,
-                               cstate, alloc_size);
+                                       cstate, alloc_size, 0);
                spin_lock_irqsave(&pool->lock, *irq_flags);
 
                if (!r) {
@@ -635,7 +647,7 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
                                   struct list_head *pages,
                                   int ttm_flags,
                                   enum ttm_caching_state cstate,
-                                  unsigned count)
+                                  unsigned count, unsigned order)
 {
        unsigned long irq_flags;
        struct list_head *p;
@@ -643,7 +655,9 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
        int r = 0;
 
        spin_lock_irqsave(&pool->lock, irq_flags);
-       ttm_page_pool_fill_locked(pool, ttm_flags, cstate, count, &irq_flags);
+       if (!order)
+               ttm_page_pool_fill_locked(pool, ttm_flags, cstate, count,
+                                         &irq_flags);
 
        if (count >= pool->npages) {
                /* take all pages from the pool */
@@ -698,7 +712,7 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
                 * multiple requests in parallel.
                 **/
                r = ttm_alloc_new_pages(pages, gfp_flags, ttm_flags, cstate,
-                                       count);
+                                       count, order);
        }
 
        return r;
@@ -708,8 +722,9 @@ static int ttm_page_pool_get_pages(struct ttm_page_pool *pool,
 static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
                          enum ttm_caching_state cstate)
 {
+       struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
+       struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
        unsigned long irq_flags;
-       struct ttm_page_pool *pool = ttm_get_pool(flags, cstate);
        unsigned i;
 
        if (pool == NULL) {
@@ -737,8 +752,48 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
                return;
        }
 
+       i = 0;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+       if (huge) {
+               unsigned max_size, n2free;
+
+               spin_lock_irqsave(&huge->lock, irq_flags);
+               while (i < npages) {
+                       struct page *p = pages[i];
+                       unsigned j;
+
+                       if (!p)
+                               break;
+
+                       for (j = 0; j < HPAGE_PMD_NR; ++j)
+                               if (p++ != pages[i + j])
+                                   break;
+
+                       if (j != HPAGE_PMD_NR)
+                               break;
+
+                       list_add_tail(&pages[i]->lru, &huge->list);
+
+                       for (j = 0; j < HPAGE_PMD_NR; ++j)
+                               pages[i++] = NULL;
+                       huge->npages++;
+               }
+
+               /* Check that we don't go over the pool limit */
+               max_size = _manager->options.max_size;
+               max_size /= HPAGE_PMD_NR;
+               if (huge->npages > max_size)
+                       n2free = huge->npages - max_size;
+               else
+                       n2free = 0;
+               spin_unlock_irqrestore(&huge->lock, irq_flags);
+               if (n2free)
+                       ttm_page_pool_free(huge, n2free, false);
+       }
+#endif
+
        spin_lock_irqsave(&pool->lock, irq_flags);
-       for (i = 0; i < npages; i++) {
+       while (i < npages) {
                if (pages[i]) {
                        if (page_count(pages[i]) != 1)
                                pr_err("Erroneous page count. Leaking pages.\n");
@@ -746,6 +801,7 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
                        pages[i] = NULL;
                        pool->npages++;
                }
+               ++i;
        }
        /* Check that we don't go over the pool limit */
        npages = 0;
@@ -768,7 +824,8 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
                         enum ttm_caching_state cstate)
 {
-       struct ttm_page_pool *pool = ttm_get_pool(flags, cstate);
+       struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
+       struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
        struct list_head plist;
        struct page *p = NULL;
        unsigned count;
@@ -821,11 +878,28 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
                return 0;
        }
 
-       /* First we take pages from the pool */
+       count = 0;
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+       if (huge && npages >= HPAGE_PMD_NR) {
+               INIT_LIST_HEAD(&plist);
+               ttm_page_pool_get_pages(huge, &plist, flags, cstate,
+                                       npages / HPAGE_PMD_NR,
+                                       HPAGE_PMD_ORDER);
+
+               list_for_each_entry(p, &plist, lru) {
+                       unsigned j;
+
+                       for (j = 0; j < HPAGE_PMD_NR; ++j)
+                               pages[count++] = &p[j];
+               }
+       }
+#endif
+
        INIT_LIST_HEAD(&plist);
-       r = ttm_page_pool_get_pages(pool, &plist, flags, cstate, npages);
+       r = ttm_page_pool_get_pages(pool, &plist, flags, cstate,
+                                   npages - count, 0);
 
-       count = 0;
        list_for_each_entry(p, &plist, lru)
                pages[count++] = p;
 
@@ -872,6 +946,14 @@ int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
        ttm_page_pool_init_locked(&_manager->uc_pool_dma32,
                                  GFP_USER | GFP_DMA32, "uc dma");
 
+       ttm_page_pool_init_locked(&_manager->wc_pool_huge,
+                                 GFP_TRANSHUGE & ~(__GFP_MOVABLE | __GFP_COMP),
+                                 "wc huge");
+
+       ttm_page_pool_init_locked(&_manager->uc_pool_huge,
+                                 GFP_TRANSHUGE & ~(__GFP_MOVABLE | __GFP_COMP)
+                                 , "uc huge");
+
        _manager->options.max_size = max_pages;
        _manager->options.small = SMALL_ALLOCATION;
        _manager->options.alloc_size = NUM_PAGES_TO_ALLOC;
@@ -1041,12 +1123,12 @@ int ttm_page_alloc_debugfs(struct seq_file *m, void *data)
                seq_printf(m, "No pool allocator running.\n");
                return 0;
        }
-       seq_printf(m, "%6s %12s %13s %8s\n",
+       seq_printf(m, "%7s %12s %13s %8s\n",
                        h[0], h[1], h[2], h[3]);
        for (i = 0; i < NUM_POOLS; ++i) {
                p = &_manager->pools[i];
 
-               seq_printf(m, "%6s %12ld %13ld %8d\n",
+               seq_printf(m, "%7s %12ld %13ld %8d\n",
                                p->name, p->nrefills,
                                p->nfrees, p->npages);
        }
--

2.7.4

Quoted 4 lines

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx