All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Michel Dänzer" <michel@daenzer.net>
To: "Christian König" <christian.koenig@amd.com>
Cc: dri-devel@lists.freedesktop.org, amd-gfx@lists.freedesktop.org,
	linux-kernel@vger.kernel.org
Subject: [PATCH v2 1/2] drm/ttm: Only allocate huge pages with new flag TTM_PAGE_FLAG_TRANSHUGE
Date: Fri, 27 Apr 2018 15:08:11 +0200	[thread overview]
Message-ID: <20180427130811.7642-1-michel@daenzer.net> (raw)
In-Reply-To: <20180426150618.13470-1-michel@daenzer.net>

From: Michel Dänzer <michel.daenzer@amd.com>

Previously, TTM would always (with CONFIG_TRANSPARENT_HUGEPAGE enabled)
try to allocate huge pages. However, not all drivers can take advantage
of huge pages, but they would incur the overhead for allocating and
freeing them anyway.

Now, drivers which can take advantage of huge pages need to set the new
flag TTM_PAGE_FLAG_TRANSHUGE to get them. Drivers not setting this flag
no longer incur any overhead for allocating or freeing huge pages.

v2:
* Also guard swapping of consecutive pages in ttm_get_pages
* Reword commit log, hopefully clearer now

Cc: stable@vger.kernel.org
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  2 +-
 drivers/gpu/drm/ttm/ttm_page_alloc.c     | 35 +++++++++++++++++-------
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 ++++--
 include/drm/ttm/ttm_tt.h                 |  1 +
 4 files changed, 32 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index dfd22db13fb1..e03e9e361e2a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -988,7 +988,7 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
 		return NULL;
 	}
 	gtt->ttm.ttm.func = &amdgpu_backend_func;
-	if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags)) {
+	if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags | TTM_PAGE_FLAG_TRANSHUGE)) {
 		kfree(gtt);
 		return NULL;
 	}
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index f0481b7b60c5..476d668e1cbd 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -760,7 +760,7 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 {
 	struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
+	struct ttm_page_pool *huge = NULL;
 #endif
 	unsigned long irq_flags;
 	unsigned i;
@@ -780,7 +780,8 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 			}
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-			if (!(flags & TTM_PAGE_FLAG_DMA32)) {
+			if ((flags & (TTM_PAGE_FLAG_DMA32 | TTM_PAGE_FLAG_TRANSHUGE)) ==
+			    TTM_PAGE_FLAG_TRANSHUGE) {
 				for (j = 0; j < HPAGE_PMD_NR; ++j)
 					if (p++ != pages[i + j])
 					    break;
@@ -805,6 +806,8 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 
 	i = 0;
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if (flags & TTM_PAGE_FLAG_TRANSHUGE)
+		huge = ttm_get_pool(flags, true, cstate);
 	if (huge) {
 		unsigned max_size, n2free;
 
@@ -877,7 +880,7 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 {
 	struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
+	struct ttm_page_pool *huge = NULL;
 #endif
 	struct list_head plist;
 	struct page *p = NULL;
@@ -906,7 +909,8 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 
 		i = 0;
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-		if (!(gfp_flags & GFP_DMA32)) {
+		if ((flags & (TTM_PAGE_FLAG_DMA32 | TTM_PAGE_FLAG_TRANSHUGE)) ==
+		    TTM_PAGE_FLAG_TRANSHUGE) {
 			while (npages >= HPAGE_PMD_NR) {
 				gfp_t huge_flags = gfp_flags;
 
@@ -933,9 +937,13 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 				return -ENOMEM;
 			}
 
-			/* Swap the pages if we detect consecutive order */
-			if (i > first && pages[i - 1] == p - 1)
-				swap(p, pages[i - 1]);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+			if (flags & TTM_PAGE_FLAG_TRANSHUGE) {
+				/* Swap the pages if we detect consecutive order */
+				if (i > first && pages[i - 1] == p - 1)
+					swap(p, pages[i - 1]);
+			}
+#endif
 
 			pages[i++] = p;
 			--npages;
@@ -946,6 +954,8 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 	count = 0;
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if (flags & TTM_PAGE_FLAG_TRANSHUGE)
+		huge = ttm_get_pool(flags, true, cstate);
 	if (huge && npages >= HPAGE_PMD_NR) {
 		INIT_LIST_HEAD(&plist);
 		ttm_page_pool_get_pages(huge, &plist, flags, cstate,
@@ -969,9 +979,14 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 	list_for_each_entry(p, &plist, lru) {
 		struct page *tmp = p;
 
-		/* Swap the pages if we detect consecutive order */
-		if (count > first && pages[count - 1] == tmp - 1)
-			swap(tmp, pages[count - 1]);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+		if (flags & TTM_PAGE_FLAG_TRANSHUGE) {
+			/* Swap the pages if we detect consecutive order */
+			if (count > first && pages[count - 1] == tmp - 1)
+				swap(tmp, pages[count - 1]);
+		}
+#endif
+
 		pages[count++] = tmp;
 	}
 
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
index 8a25d1974385..291b04213ec5 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
@@ -949,7 +949,8 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 	type = ttm_to_type(ttm->page_flags, ttm->caching_state);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	if (ttm->page_flags & TTM_PAGE_FLAG_DMA32)
+	if ((ttm->page_flags & (TTM_PAGE_FLAG_DMA32 | TTM_PAGE_FLAG_TRANSHUGE))
+	    != TTM_PAGE_FLAG_TRANSHUGE)
 		goto skip_huge;
 
 	pool = ttm_dma_find_pool(dev, type | IS_HUGE);
@@ -1035,7 +1036,7 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 {
 	struct ttm_tt *ttm = &ttm_dma->ttm;
 	struct ttm_mem_global *mem_glob = ttm->bdev->glob->mem_glob;
-	struct dma_pool *pool;
+	struct dma_pool *pool = NULL;
 	struct dma_page *d_page, *next;
 	enum pool_type type;
 	bool is_cached = false;
@@ -1045,7 +1046,8 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 	type = ttm_to_type(ttm->page_flags, ttm->caching_state);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	pool = ttm_dma_find_pool(dev, type | IS_HUGE);
+	if (ttm->page_flags & TTM_PAGE_FLAG_TRANSHUGE)
+		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,
diff --git a/include/drm/ttm/ttm_tt.h b/include/drm/ttm/ttm_tt.h
index c0e928abf592..c7d2120f0362 100644
--- a/include/drm/ttm/ttm_tt.h
+++ b/include/drm/ttm/ttm_tt.h
@@ -41,6 +41,7 @@ struct ttm_operation_ctx;
 #define TTM_PAGE_FLAG_DMA32           (1 << 7)
 #define TTM_PAGE_FLAG_SG              (1 << 8)
 #define TTM_PAGE_FLAG_NO_RETRY	      (1 << 9)
+#define TTM_PAGE_FLAG_TRANSHUGE       (1 << 10)
 
 enum ttm_caching_state {
 	tt_uncached,
-- 
2.17.0

WARNING: multiple messages have this Message-ID (diff)
From: "Michel Dänzer" <michel-otUistvHUpPR7s880joybQ@public.gmane.org>
To: "Christian König" <christian.koenig-5C7GfCeVMHo@public.gmane.org>
Cc: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org,
	dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org,
	linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Subject: [PATCH v2 1/2] drm/ttm: Only allocate huge pages with new flag TTM_PAGE_FLAG_TRANSHUGE
Date: Fri, 27 Apr 2018 15:08:11 +0200	[thread overview]
Message-ID: <20180427130811.7642-1-michel@daenzer.net> (raw)
In-Reply-To: <20180426150618.13470-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org>

From: Michel Dänzer <michel.daenzer@amd.com>

Previously, TTM would always (with CONFIG_TRANSPARENT_HUGEPAGE enabled)
try to allocate huge pages. However, not all drivers can take advantage
of huge pages, but they would incur the overhead for allocating and
freeing them anyway.

Now, drivers which can take advantage of huge pages need to set the new
flag TTM_PAGE_FLAG_TRANSHUGE to get them. Drivers not setting this flag
no longer incur any overhead for allocating or freeing huge pages.

v2:
* Also guard swapping of consecutive pages in ttm_get_pages
* Reword commit log, hopefully clearer now

Cc: stable@vger.kernel.org
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  2 +-
 drivers/gpu/drm/ttm/ttm_page_alloc.c     | 35 +++++++++++++++++-------
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 ++++--
 include/drm/ttm/ttm_tt.h                 |  1 +
 4 files changed, 32 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index dfd22db13fb1..e03e9e361e2a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -988,7 +988,7 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
 		return NULL;
 	}
 	gtt->ttm.ttm.func = &amdgpu_backend_func;
-	if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags)) {
+	if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags | TTM_PAGE_FLAG_TRANSHUGE)) {
 		kfree(gtt);
 		return NULL;
 	}
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index f0481b7b60c5..476d668e1cbd 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -760,7 +760,7 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 {
 	struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
+	struct ttm_page_pool *huge = NULL;
 #endif
 	unsigned long irq_flags;
 	unsigned i;
@@ -780,7 +780,8 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 			}
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-			if (!(flags & TTM_PAGE_FLAG_DMA32)) {
+			if ((flags & (TTM_PAGE_FLAG_DMA32 | TTM_PAGE_FLAG_TRANSHUGE)) ==
+			    TTM_PAGE_FLAG_TRANSHUGE) {
 				for (j = 0; j < HPAGE_PMD_NR; ++j)
 					if (p++ != pages[i + j])
 					    break;
@@ -805,6 +806,8 @@ static void ttm_put_pages(struct page **pages, unsigned npages, int flags,
 
 	i = 0;
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if (flags & TTM_PAGE_FLAG_TRANSHUGE)
+		huge = ttm_get_pool(flags, true, cstate);
 	if (huge) {
 		unsigned max_size, n2free;
 
@@ -877,7 +880,7 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 {
 	struct ttm_page_pool *pool = ttm_get_pool(flags, false, cstate);
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	struct ttm_page_pool *huge = ttm_get_pool(flags, true, cstate);
+	struct ttm_page_pool *huge = NULL;
 #endif
 	struct list_head plist;
 	struct page *p = NULL;
@@ -906,7 +909,8 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 
 		i = 0;
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-		if (!(gfp_flags & GFP_DMA32)) {
+		if ((flags & (TTM_PAGE_FLAG_DMA32 | TTM_PAGE_FLAG_TRANSHUGE)) ==
+		    TTM_PAGE_FLAG_TRANSHUGE) {
 			while (npages >= HPAGE_PMD_NR) {
 				gfp_t huge_flags = gfp_flags;
 
@@ -933,9 +937,13 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 				return -ENOMEM;
 			}
 
-			/* Swap the pages if we detect consecutive order */
-			if (i > first && pages[i - 1] == p - 1)
-				swap(p, pages[i - 1]);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+			if (flags & TTM_PAGE_FLAG_TRANSHUGE) {
+				/* Swap the pages if we detect consecutive order */
+				if (i > first && pages[i - 1] == p - 1)
+					swap(p, pages[i - 1]);
+			}
+#endif
 
 			pages[i++] = p;
 			--npages;
@@ -946,6 +954,8 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 	count = 0;
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	if (flags & TTM_PAGE_FLAG_TRANSHUGE)
+		huge = ttm_get_pool(flags, true, cstate);
 	if (huge && npages >= HPAGE_PMD_NR) {
 		INIT_LIST_HEAD(&plist);
 		ttm_page_pool_get_pages(huge, &plist, flags, cstate,
@@ -969,9 +979,14 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 	list_for_each_entry(p, &plist, lru) {
 		struct page *tmp = p;
 
-		/* Swap the pages if we detect consecutive order */
-		if (count > first && pages[count - 1] == tmp - 1)
-			swap(tmp, pages[count - 1]);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+		if (flags & TTM_PAGE_FLAG_TRANSHUGE) {
+			/* Swap the pages if we detect consecutive order */
+			if (count > first && pages[count - 1] == tmp - 1)
+				swap(tmp, pages[count - 1]);
+		}
+#endif
+
 		pages[count++] = tmp;
 	}
 
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
index 8a25d1974385..291b04213ec5 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
@@ -949,7 +949,8 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 	type = ttm_to_type(ttm->page_flags, ttm->caching_state);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	if (ttm->page_flags & TTM_PAGE_FLAG_DMA32)
+	if ((ttm->page_flags & (TTM_PAGE_FLAG_DMA32 | TTM_PAGE_FLAG_TRANSHUGE))
+	    != TTM_PAGE_FLAG_TRANSHUGE)
 		goto skip_huge;
 
 	pool = ttm_dma_find_pool(dev, type | IS_HUGE);
@@ -1035,7 +1036,7 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 {
 	struct ttm_tt *ttm = &ttm_dma->ttm;
 	struct ttm_mem_global *mem_glob = ttm->bdev->glob->mem_glob;
-	struct dma_pool *pool;
+	struct dma_pool *pool = NULL;
 	struct dma_page *d_page, *next;
 	enum pool_type type;
 	bool is_cached = false;
@@ -1045,7 +1046,8 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 	type = ttm_to_type(ttm->page_flags, ttm->caching_state);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-	pool = ttm_dma_find_pool(dev, type | IS_HUGE);
+	if (ttm->page_flags & TTM_PAGE_FLAG_TRANSHUGE)
+		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,
diff --git a/include/drm/ttm/ttm_tt.h b/include/drm/ttm/ttm_tt.h
index c0e928abf592..c7d2120f0362 100644
--- a/include/drm/ttm/ttm_tt.h
+++ b/include/drm/ttm/ttm_tt.h
@@ -41,6 +41,7 @@ struct ttm_operation_ctx;
 #define TTM_PAGE_FLAG_DMA32           (1 << 7)
 #define TTM_PAGE_FLAG_SG              (1 << 8)
 #define TTM_PAGE_FLAG_NO_RETRY	      (1 << 9)
+#define TTM_PAGE_FLAG_TRANSHUGE       (1 << 10)
 
 enum ttm_caching_state {
 	tt_uncached,
-- 
2.17.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

  parent reply	other threads:[~2018-04-27 13:08 UTC|newest]

Thread overview: 29+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-04-26 15:06 [PATCH 1/2] drm/ttm: Add TTM_PAGE_FLAG_TRANSHUGE Michel Dänzer
2018-04-26 15:06 ` [PATCH 2/2] drm/ttm: Use GFP_TRANSHUGE_LIGHT for allocating huge pages Michel Dänzer
2018-04-26 15:06   ` Michel Dänzer
2018-04-29  7:04   ` Christian König
2018-04-29  7:04     ` Christian König
2018-04-27  2:51 ` [PATCH 1/2] drm/ttm: Add TTM_PAGE_FLAG_TRANSHUGE zhoucm1
2018-04-27  2:51   ` zhoucm1
2018-04-27  8:41   ` Michel Dänzer
2018-04-27  8:41     ` Michel Dänzer
2018-04-27 13:08 ` Michel Dänzer [this message]
2018-04-27 13:08   ` [PATCH v2 1/2] drm/ttm: Only allocate huge pages with new flag TTM_PAGE_FLAG_TRANSHUGE Michel Dänzer
2018-04-28 16:30   ` Ilia Mirkin
2018-04-28 16:30     ` Ilia Mirkin
2018-04-28 23:02     ` Michel Dänzer
2018-04-28 23:02       ` Michel Dänzer
2018-04-28 23:56       ` Ilia Mirkin
2018-04-28 23:56         ` Ilia Mirkin
2018-05-02  8:08         ` Michel Dänzer
2018-05-02  8:08           ` Michel Dänzer
2018-04-29  7:02       ` Christian König
2018-04-29  7:02         ` Christian König
2018-04-30 16:33         ` Michel Dänzer
2018-04-30 16:33           ` Michel Dänzer
2018-04-30 18:22           ` Christian König
2018-04-30 18:22             ` Christian König
2018-04-30 23:15             ` Dave Airlie
2018-04-30 23:15               ` Dave Airlie
2018-05-01 13:59               ` Michel Dänzer
2018-05-01 13:59                 ` Michel Dänzer

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20180427130811.7642-1-michel@daenzer.net \
    --to=michel@daenzer.net \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=christian.koenig@amd.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=linux-kernel@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.