[PATCH 1/2] drm/ttm: don't update global memory count for some special cases

[PATCH 1/2] drm/ttm: don't update global memory count for some special cases

[Roger He]

add input parameter for ttm_dma_unpopulate.
when ttm_dma_pool_get_pages or ttm_mem_global_alloc_page fail, don't
call ttm_mem_global_free_page to update global memory count.

Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  2 +-
 drivers/gpu/drm/nouveau/nouveau_bo.c     |  2 +-
 drivers/gpu/drm/radeon/radeon_ttm.c      |  2 +-
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 19 +++++++++++--------
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c   |  2 +-
 include/drm/ttm/ttm_page_alloc.h         |  5 +++--
 6 files changed, 18 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index e4bb435..723ccf1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1046,7 +1046,7 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
 
 #ifdef CONFIG_SWIOTLB
 	if (swiotlb_nr_tbl()) {
-		ttm_dma_unpopulate(&gtt->ttm, adev->dev);
+		ttm_dma_unpopulate(&gtt->ttm, adev->dev, true);
 		return;
 	}
 #endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index ce328ed..3f7c30f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1632,7 +1632,7 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
 
 #if IS_ENABLED(CONFIG_SWIOTLB) && IS_ENABLED(CONFIG_X86)
 	if (swiotlb_nr_tbl()) {
-		ttm_dma_unpopulate((void *)ttm, dev);
+		ttm_dma_unpopulate((void *)ttm, dev, true);
 		return;
 	}
 #endif
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index a0a839b..449cc65 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -789,7 +789,7 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
 
 #ifdef CONFIG_SWIOTLB
 	if (swiotlb_nr_tbl()) {
-		ttm_dma_unpopulate(&gtt->ttm, rdev->dev);
+		ttm_dma_unpopulate(&gtt->ttm, rdev->dev, true);
 		return;
 	}
 #endif
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
index c7f01a4..4cda764 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
@@ -969,7 +969,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
 						pool->size, ctx);
 		if (unlikely(ret != 0)) {
-			ttm_dma_unpopulate(ttm_dma, dev);
+			ttm_dma_unpopulate(ttm_dma, dev, false);
 			return -ENOMEM;
 		}
 
@@ -998,14 +998,14 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 	while (num_pages) {
 		ret = ttm_dma_pool_get_pages(pool, ttm_dma, i);
 		if (ret != 0) {
-			ttm_dma_unpopulate(ttm_dma, dev);
+			ttm_dma_unpopulate(ttm_dma, dev, false);
 			return -ENOMEM;
 		}
 
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
 						pool->size, ctx);
 		if (unlikely(ret != 0)) {
-			ttm_dma_unpopulate(ttm_dma, dev);
+			ttm_dma_unpopulate(ttm_dma, dev, false);
 			return -ENOMEM;
 		}
 
@@ -1016,7 +1016,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 	if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) {
 		ret = ttm_tt_swapin(ttm);
 		if (unlikely(ret != 0)) {
-			ttm_dma_unpopulate(ttm_dma, dev);
+			ttm_dma_unpopulate(ttm_dma, dev, true);
 			return ret;
 		}
 	}
@@ -1027,7 +1027,8 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 EXPORT_SYMBOL_GPL(ttm_dma_populate);
 
 /* Put all pages in pages list to correct pool to wait for reuse */
-void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
+void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
+			bool update_glob_count)
 {
 	struct ttm_tt *ttm = &ttm_dma->ttm;
 	struct dma_pool *pool;
@@ -1049,7 +1050,8 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 				continue;
 
 			count++;
-			ttm_mem_global_free_page(ttm->glob->mem_glob,
+			if (update_glob_count)
+				ttm_mem_global_free_page(ttm->glob->mem_glob,
 						 d_page->p, pool->size);
 			ttm_dma_page_put(pool, d_page);
 		}
@@ -1094,11 +1096,12 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
 
 	if (is_cached) {
 		list_for_each_entry_safe(d_page, next, &ttm_dma->pages_list, page_list) {
-			ttm_mem_global_free_page(ttm->glob->mem_glob,
+			if (update_glob_count)
+				ttm_mem_global_free_page(ttm->glob->mem_glob,
 						 d_page->p, pool->size);
 			ttm_dma_page_put(pool, d_page);
 		}
-	} else {
+	} else if (update_glob_count) {
 		for (i = 0; i < count; i++) {
 			ttm_mem_global_free_page(ttm->glob->mem_glob,
 						 ttm->pages[i], pool->size);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 22231bc..38df05d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -681,7 +681,7 @@ static void vmw_ttm_unpopulate(struct ttm_tt *ttm)
 		size_t size =
 			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
 
-		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
+		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev, true);
 		ttm_mem_global_free(glob, size);
 	} else
 		ttm_pool_unpopulate(ttm);
diff --git a/include/drm/ttm/ttm_page_alloc.h b/include/drm/ttm/ttm_page_alloc.h
index 4d9b019..8518cd5 100644
--- a/include/drm/ttm/ttm_page_alloc.h
+++ b/include/drm/ttm/ttm_page_alloc.h
@@ -92,7 +92,8 @@ int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data);
 
 int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 			struct ttm_operation_ctx *ctx);
-void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev);
+void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
+			bool update_glob_count);
 
 #else
 static inline int ttm_dma_page_alloc_init(struct ttm_mem_global *glob,
@@ -114,7 +115,7 @@ static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma,
 	return -ENOMEM;
 }
 static inline void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
-				      struct device *dev)
+		struct device *dev, bool update_glob_count);
 {
 }
 #endif
-- 
2.7.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 2/2] drm/ttm: only free pages rather than update global memory count together

[Roger He]

if ttm_get_pages or ttm_mem_global_alloc_page fail, should not update
global memory count.

Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index 0eab24e..aebabf1 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -1075,7 +1075,8 @@ int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
 	ret = ttm_get_pages(ttm->pages, ttm->num_pages, ttm->page_flags,
 			    ttm->caching_state);
 	if (unlikely(ret != 0)) {
-		ttm_pool_unpopulate(ttm);
+		ttm_put_pages(ttm->pages, ttm->num_pages, ttm->page_flags,
+			ttm->caching_state);
 		return ret;
 	}
 
@@ -1083,7 +1084,8 @@ int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
 		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
 						PAGE_SIZE, ctx);
 		if (unlikely(ret != 0)) {
-			ttm_pool_unpopulate(ttm);
+			ttm_put_pages(ttm->pages, ttm->num_pages,
+				ttm->page_flags, ttm->caching_state);
 			return -ENOMEM;
 		}
 	}
-- 
2.7.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 1/2] drm/ttm: don't update global memory count for some special cases

[Chunming Zhou]

Can this fix GTT size issue or not?


Regards,
David Zhou
On 2018年01月12日 14:14, Roger He wrote:
> add input parameter for ttm_dma_unpopulate.
> when ttm_dma_pool_get_pages or ttm_mem_global_alloc_page fail, don't
> call ttm_mem_global_free_page to update global memory count.
>
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  2 +-
>   drivers/gpu/drm/nouveau/nouveau_bo.c     |  2 +-
>   drivers/gpu/drm/radeon/radeon_ttm.c      |  2 +-
>   drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 19 +++++++++++--------
>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c   |  2 +-
>   include/drm/ttm/ttm_page_alloc.h         |  5 +++--
>   6 files changed, 18 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index e4bb435..723ccf1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -1046,7 +1046,7 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate(&gtt->ttm, adev->dev);
> +		ttm_dma_unpopulate(&gtt->ttm, adev->dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index ce328ed..3f7c30f 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1632,7 +1632,7 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
>   
>   #if IS_ENABLED(CONFIG_SWIOTLB) && IS_ENABLED(CONFIG_X86)
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate((void *)ttm, dev);
> +		ttm_dma_unpopulate((void *)ttm, dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
> index a0a839b..449cc65 100644
> --- a/drivers/gpu/drm/radeon/radeon_ttm.c
> +++ b/drivers/gpu/drm/radeon/radeon_ttm.c
> @@ -789,7 +789,7 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate(&gtt->ttm, rdev->dev);
> +		ttm_dma_unpopulate(&gtt->ttm, rdev->dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> index c7f01a4..4cda764 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> @@ -969,7 +969,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
> @@ -998,14 +998,14 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   	while (num_pages) {
>   		ret = ttm_dma_pool_get_pages(pool, ttm_dma, i);
>   		if (ret != 0) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
> @@ -1016,7 +1016,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   	if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) {
>   		ret = ttm_tt_swapin(ttm);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, true);
>   			return ret;
>   		}
>   	}
> @@ -1027,7 +1027,8 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   EXPORT_SYMBOL_GPL(ttm_dma_populate);
>   
>   /* Put all pages in pages list to correct pool to wait for reuse */
> -void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
> +void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			bool update_glob_count)
>   {
>   	struct ttm_tt *ttm = &ttm_dma->ttm;
>   	struct dma_pool *pool;
> @@ -1049,7 +1050,8 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   				continue;
>   
>   			count++;
> -			ttm_mem_global_free_page(ttm->glob->mem_glob,
> +			if (update_glob_count)
> +				ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 d_page->p, pool->size);
>   			ttm_dma_page_put(pool, d_page);
>   		}
> @@ -1094,11 +1096,12 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   
>   	if (is_cached) {
>   		list_for_each_entry_safe(d_page, next, &ttm_dma->pages_list, page_list) {
> -			ttm_mem_global_free_page(ttm->glob->mem_glob,
> +			if (update_glob_count)
> +				ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 d_page->p, pool->size);
>   			ttm_dma_page_put(pool, d_page);
>   		}
> -	} else {
> +	} else if (update_glob_count) {
>   		for (i = 0; i < count; i++) {
>   			ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 ttm->pages[i], pool->size);
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index 22231bc..38df05d 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -681,7 +681,7 @@ static void vmw_ttm_unpopulate(struct ttm_tt *ttm)
>   		size_t size =
>   			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
>   
> -		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
> +		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev, true);
>   		ttm_mem_global_free(glob, size);
>   	} else
>   		ttm_pool_unpopulate(ttm);
> diff --git a/include/drm/ttm/ttm_page_alloc.h b/include/drm/ttm/ttm_page_alloc.h
> index 4d9b019..8518cd5 100644
> --- a/include/drm/ttm/ttm_page_alloc.h
> +++ b/include/drm/ttm/ttm_page_alloc.h
> @@ -92,7 +92,8 @@ int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data);
>   
>   int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   			struct ttm_operation_ctx *ctx);
> -void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev);
> +void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			bool update_glob_count);
>   
>   #else
>   static inline int ttm_dma_page_alloc_init(struct ttm_mem_global *glob,
> @@ -114,7 +115,7 @@ static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma,
>   	return -ENOMEM;
>   }
>   static inline void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
> -				      struct device *dev)
> +		struct device *dev, bool update_glob_count);
>   {
>   }
>   #endif

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

RE: [PATCH 1/2] drm/ttm: don't update global memory count for some special cases

[He, Roger]

It is the first step for the GTT size issue.
Need other patches I am working on it.

Thanks
Roger(Hongbo.He)
-----Original Message-----
From: Zhou, David(ChunMing) 
Sent: Friday, January 12, 2018 2:46 PM
To: He, Roger <Hongbo.He@amd.com>; dri-devel@lists.freedesktop.org
Cc: Koenig, Christian <Christian.Koenig@amd.com>
Subject: Re: [PATCH 1/2] drm/ttm: don't update global memory count for some special cases

Can this fix GTT size issue or not?


Regards,
David Zhou
On 2018年01月12日 14:14, Roger He wrote:
> add input parameter for ttm_dma_unpopulate.
> when ttm_dma_pool_get_pages or ttm_mem_global_alloc_page fail, don't 
> call ttm_mem_global_free_page to update global memory count.
>
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  2 +-
>   drivers/gpu/drm/nouveau/nouveau_bo.c     |  2 +-
>   drivers/gpu/drm/radeon/radeon_ttm.c      |  2 +-
>   drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 19 +++++++++++--------
>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c   |  2 +-
>   include/drm/ttm/ttm_page_alloc.h         |  5 +++--
>   6 files changed, 18 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index e4bb435..723ccf1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -1046,7 +1046,7 @@ static void amdgpu_ttm_tt_unpopulate(struct 
> ttm_tt *ttm)
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate(&gtt->ttm, adev->dev);
> +		ttm_dma_unpopulate(&gtt->ttm, adev->dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c 
> b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index ce328ed..3f7c30f 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1632,7 +1632,7 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
>   
>   #if IS_ENABLED(CONFIG_SWIOTLB) && IS_ENABLED(CONFIG_X86)
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate((void *)ttm, dev);
> +		ttm_dma_unpopulate((void *)ttm, dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c 
> b/drivers/gpu/drm/radeon/radeon_ttm.c
> index a0a839b..449cc65 100644
> --- a/drivers/gpu/drm/radeon/radeon_ttm.c
> +++ b/drivers/gpu/drm/radeon/radeon_ttm.c
> @@ -789,7 +789,7 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt 
> *ttm)
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate(&gtt->ttm, rdev->dev);
> +		ttm_dma_unpopulate(&gtt->ttm, rdev->dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c 
> b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> index c7f01a4..4cda764 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> @@ -969,7 +969,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
> @@ -998,14 +998,14 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   	while (num_pages) {
>   		ret = ttm_dma_pool_get_pages(pool, ttm_dma, i);
>   		if (ret != 0) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
> @@ -1016,7 +1016,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   	if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) {
>   		ret = ttm_tt_swapin(ttm);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, true);
>   			return ret;
>   		}
>   	}
> @@ -1027,7 +1027,8 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   EXPORT_SYMBOL_GPL(ttm_dma_populate);
>   
>   /* Put all pages in pages list to correct pool to wait for reuse */ 
> -void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device 
> *dev)
> +void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			bool update_glob_count)
>   {
>   	struct ttm_tt *ttm = &ttm_dma->ttm;
>   	struct dma_pool *pool;
> @@ -1049,7 +1050,8 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   				continue;
>   
>   			count++;
> -			ttm_mem_global_free_page(ttm->glob->mem_glob,
> +			if (update_glob_count)
> +				ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 d_page->p, pool->size);
>   			ttm_dma_page_put(pool, d_page);
>   		}
> @@ -1094,11 +1096,12 @@ void ttm_dma_unpopulate(struct ttm_dma_tt 
> *ttm_dma, struct device *dev)
>   
>   	if (is_cached) {
>   		list_for_each_entry_safe(d_page, next, &ttm_dma->pages_list, page_list) {
> -			ttm_mem_global_free_page(ttm->glob->mem_glob,
> +			if (update_glob_count)
> +				ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 d_page->p, pool->size);
>   			ttm_dma_page_put(pool, d_page);
>   		}
> -	} else {
> +	} else if (update_glob_count) {
>   		for (i = 0; i < count; i++) {
>   			ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 ttm->pages[i], pool->size);
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c 
> b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index 22231bc..38df05d 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -681,7 +681,7 @@ static void vmw_ttm_unpopulate(struct ttm_tt *ttm)
>   		size_t size =
>   			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
>   
> -		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
> +		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev, true);
>   		ttm_mem_global_free(glob, size);
>   	} else
>   		ttm_pool_unpopulate(ttm);
> diff --git a/include/drm/ttm/ttm_page_alloc.h 
> b/include/drm/ttm/ttm_page_alloc.h
> index 4d9b019..8518cd5 100644
> --- a/include/drm/ttm/ttm_page_alloc.h
> +++ b/include/drm/ttm/ttm_page_alloc.h
> @@ -92,7 +92,8 @@ int ttm_dma_page_alloc_debugfs(struct seq_file *m, 
> void *data);
>   
>   int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   			struct ttm_operation_ctx *ctx);
> -void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device 
> *dev);
> +void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			bool update_glob_count);
>   
>   #else
>   static inline int ttm_dma_page_alloc_init(struct ttm_mem_global 
> *glob, @@ -114,7 +115,7 @@ static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma,
>   	return -ENOMEM;
>   }
>   static inline void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
> -				      struct device *dev)
> +		struct device *dev, bool update_glob_count);
>   {
>   }
>   #endif

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 2/2] drm/ttm: only free pages rather than update global memory count together

[Andrey Grodzovsky]

Both patches

Tested-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>


Andrey


On 01/12/2018 01:14 AM, Roger He wrote:
> if ttm_get_pages or ttm_mem_global_alloc_page fail, should not update
> global memory count.
>
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_page_alloc.c | 6 ++++--
>   1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 0eab24e..aebabf1 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -1075,7 +1075,8 @@ int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   	ret = ttm_get_pages(ttm->pages, ttm->num_pages, ttm->page_flags,
>   			    ttm->caching_state);
>   	if (unlikely(ret != 0)) {
> -		ttm_pool_unpopulate(ttm);
> +		ttm_put_pages(ttm->pages, ttm->num_pages, ttm->page_flags,
> +			ttm->caching_state);
>   		return ret;
>   	}
>   
> @@ -1083,7 +1084,8 @@ int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						PAGE_SIZE, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_pool_unpopulate(ttm);
> +			ttm_put_pages(ttm->pages, ttm->num_pages,
> +				ttm->page_flags, ttm->caching_state);
>   			return -ENOMEM;
>   		}
>   	}

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 1/2] drm/ttm: don't update global memory count for some special cases

[Christian König]

Am 12.01.2018 um 07:14 schrieb Roger He:
> add input parameter for ttm_dma_unpopulate.
> when ttm_dma_pool_get_pages or ttm_mem_global_alloc_page fail, don't
> call ttm_mem_global_free_page to update global memory count.
>
> Signed-off-by: Roger He <Hongbo.He@amd.com>

Good catch, but that doesn't looks correct to me.

See the handling in ttm_dma_populate():
>                 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, ctx);
>                 if (unlikely(ret != 0)) {
>                         ttm_dma_unpopulate(ttm_dma, dev);
>                         return -ENOMEM;
>                 }
>

When ttm_mem_global_alloc_page() fails we only missed to account the 
last allocated page, not all of them.

Additional to that I really don't like adding the new parameter to the 
drivers.

Regards,
Christian.

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  2 +-
>   drivers/gpu/drm/nouveau/nouveau_bo.c     |  2 +-
>   drivers/gpu/drm/radeon/radeon_ttm.c      |  2 +-
>   drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 19 +++++++++++--------
>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c   |  2 +-
>   include/drm/ttm/ttm_page_alloc.h         |  5 +++--
>   6 files changed, 18 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index e4bb435..723ccf1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -1046,7 +1046,7 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate(&gtt->ttm, adev->dev);
> +		ttm_dma_unpopulate(&gtt->ttm, adev->dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index ce328ed..3f7c30f 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1632,7 +1632,7 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
>   
>   #if IS_ENABLED(CONFIG_SWIOTLB) && IS_ENABLED(CONFIG_X86)
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate((void *)ttm, dev);
> +		ttm_dma_unpopulate((void *)ttm, dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
> index a0a839b..449cc65 100644
> --- a/drivers/gpu/drm/radeon/radeon_ttm.c
> +++ b/drivers/gpu/drm/radeon/radeon_ttm.c
> @@ -789,7 +789,7 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
>   
>   #ifdef CONFIG_SWIOTLB
>   	if (swiotlb_nr_tbl()) {
> -		ttm_dma_unpopulate(&gtt->ttm, rdev->dev);
> +		ttm_dma_unpopulate(&gtt->ttm, rdev->dev, true);
>   		return;
>   	}
>   #endif
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> index c7f01a4..4cda764 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> @@ -969,7 +969,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
> @@ -998,14 +998,14 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   	while (num_pages) {
>   		ret = ttm_dma_pool_get_pages(pool, ttm_dma, i);
>   		if (ret != 0) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						pool->size, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, false);
>   			return -ENOMEM;
>   		}
>   
> @@ -1016,7 +1016,7 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   	if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) {
>   		ret = ttm_tt_swapin(ttm);
>   		if (unlikely(ret != 0)) {
> -			ttm_dma_unpopulate(ttm_dma, dev);
> +			ttm_dma_unpopulate(ttm_dma, dev, true);
>   			return ret;
>   		}
>   	}
> @@ -1027,7 +1027,8 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   EXPORT_SYMBOL_GPL(ttm_dma_populate);
>   
>   /* Put all pages in pages list to correct pool to wait for reuse */
> -void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
> +void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			bool update_glob_count)
>   {
>   	struct ttm_tt *ttm = &ttm_dma->ttm;
>   	struct dma_pool *pool;
> @@ -1049,7 +1050,8 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   				continue;
>   
>   			count++;
> -			ttm_mem_global_free_page(ttm->glob->mem_glob,
> +			if (update_glob_count)
> +				ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 d_page->p, pool->size);
>   			ttm_dma_page_put(pool, d_page);
>   		}
> @@ -1094,11 +1096,12 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev)
>   
>   	if (is_cached) {
>   		list_for_each_entry_safe(d_page, next, &ttm_dma->pages_list, page_list) {
> -			ttm_mem_global_free_page(ttm->glob->mem_glob,
> +			if (update_glob_count)
> +				ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 d_page->p, pool->size);
>   			ttm_dma_page_put(pool, d_page);
>   		}
> -	} else {
> +	} else if (update_glob_count) {
>   		for (i = 0; i < count; i++) {
>   			ttm_mem_global_free_page(ttm->glob->mem_glob,
>   						 ttm->pages[i], pool->size);
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index 22231bc..38df05d 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -681,7 +681,7 @@ static void vmw_ttm_unpopulate(struct ttm_tt *ttm)
>   		size_t size =
>   			ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
>   
> -		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
> +		ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev, true);
>   		ttm_mem_global_free(glob, size);
>   	} else
>   		ttm_pool_unpopulate(ttm);
> diff --git a/include/drm/ttm/ttm_page_alloc.h b/include/drm/ttm/ttm_page_alloc.h
> index 4d9b019..8518cd5 100644
> --- a/include/drm/ttm/ttm_page_alloc.h
> +++ b/include/drm/ttm/ttm_page_alloc.h
> @@ -92,7 +92,8 @@ int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data);
>   
>   int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
>   			struct ttm_operation_ctx *ctx);
> -void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev);
> +void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev,
> +			bool update_glob_count);
>   
>   #else
>   static inline int ttm_dma_page_alloc_init(struct ttm_mem_global *glob,
> @@ -114,7 +115,7 @@ static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma,
>   	return -ENOMEM;
>   }
>   static inline void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
> -				      struct device *dev)
> +		struct device *dev, bool update_glob_count);
>   {
>   }
>   #endif

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 2/2] drm/ttm: only free pages rather than update global memory count together

[Christian König]

Am 12.01.2018 um 07:14 schrieb Roger He:
> if ttm_get_pages or ttm_mem_global_alloc_page fail, should not update
> global memory count.
>
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_page_alloc.c | 6 ++++--
>   1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 0eab24e..aebabf1 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -1075,7 +1075,8 @@ int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   	ret = ttm_get_pages(ttm->pages, ttm->num_pages, ttm->page_flags,
>   			    ttm->caching_state);
>   	if (unlikely(ret != 0)) {
> -		ttm_pool_unpopulate(ttm);
> +		ttm_put_pages(ttm->pages, ttm->num_pages, ttm->page_flags,
> +			ttm->caching_state);

That is not correctly indented. The "ttm->caching_state" should be one 
the same column as the "ttm->pages" above. It looks like your editor 
only inserts tabs, not spaces.

Apart from that the patch is Reviewed-by: Christian König 
<christian.koenig@amd.com>.

Regards,
Christian.

>   		return ret;
>   	}
>   
> @@ -1083,7 +1084,8 @@ int ttm_pool_populate(struct ttm_tt *ttm, struct ttm_operation_ctx *ctx)
>   		ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i],
>   						PAGE_SIZE, ctx);
>   		if (unlikely(ret != 0)) {
> -			ttm_pool_unpopulate(ttm);
> +			ttm_put_pages(ttm->pages, ttm->num_pages,
> +				ttm->page_flags, ttm->caching_state);
>   			return -ENOMEM;
>   		}
>   	}

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel