diff mbox

[1/2] drm/ttm: Add TTM_PAGE_FLAG_TRANSHUGE

Message ID 20180426150618.13470-1-michel@daenzer.net (mailing list archive)
State New, archived
Headers show

Commit Message

Michel Dänzer April 26, 2018, 3:06 p.m. UTC 
From: Michel Dänzer <michel.daenzer@amd.com>

When it's set, TTM tries to allocate huge pages if possible. Drivers
which can take advantage of huge pages should set it.

Drivers not setting this flag no longer incur any overhead related to
allocating or freeing huge pages.

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     | 14 ++++++++++----
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 +++++---
 include/drm/ttm/ttm_tt.h                 |  1 +
 4 files changed, 17 insertions(+), 8 deletions(-)

Comments

Chunming Zhou April 27, 2018, 2:51 a.m. UTC | #1 
On 2018年04月26日 23:06, Michel Dänzer wrote:
> From: Michel Dänzer <michel.daenzer@amd.com>
>
> When it's set, TTM tries to allocate huge pages if possible.
Do you mean original driver doesn't do this?
 From the code, driver always try huge pages if 
CONFIG_TRANSPARENT_HUGEPAGE is enabled.

Regards,
David Zhou
>   Drivers
> which can take advantage of huge pages should set it.
>
> Drivers not setting this flag no longer incur any overhead related to
> allocating or freeing huge pages.
>
> 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     | 14 ++++++++++----
>   drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  8 +++++---
>   include/drm/ttm/ttm_tt.h                 |  1 +
>   4 files changed, 17 insertions(+), 8 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..2ce91272b111 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;
>   
> @@ -946,6 +950,8 @@ 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,
> 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,
Michel Dänzer April 27, 2018, 8:41 a.m. UTC | #2 
[ Dropping Roger He, his e-mail address seems to bounce ]

On 2018-04-27 04:51 AM, zhoucm1 wrote:
> On 2018年04月26日 23:06, Michel Dänzer wrote:
>> From: Michel Dänzer <michel.daenzer@amd.com>
>>
>> When it's set, TTM tries to allocate huge pages if possible.
> Do you mean original driver doesn't do this?
> From the code, driver always try huge pages if
> CONFIG_TRANSPARENT_HUGEPAGE is enabled.

Right, before this change, TTM would do this unconditionally for all
drivers. The point of this change is not to incur any huge page related
overhead for drivers which can't take advantage of huge pages anyway.
I'll try changing the commit log to make this clearer in v2.
diff mbox

Patch

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..2ce91272b111 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;
 
@@ -946,6 +950,8 @@  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,
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,