[PATCH 1/4] drm/ttm: add page order in page pool

[PATCH 1/4] drm/ttm: add page order in page pool

[Roger He]

to indicate page order for each element in the pool

Change-Id: Ic609925ca5d2a5d4ad49d6becf505388ce3624cf
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 ++++++++++++++++++++++++++----------
 1 file changed, 31 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index 72ea037..0a0c653 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -81,6 +81,7 @@ struct ttm_page_pool {
 	char			*name;
 	unsigned long		nfrees;
 	unsigned long		nrefills;
+	unsigned int		order;
 };
 
 /**
@@ -412,6 +413,7 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 	struct ttm_page_pool *pool;
 	int shrink_pages = sc->nr_to_scan;
 	unsigned long freed = 0;
+	unsigned int nr_free_pool;
 
 	if (!mutex_trylock(&lock))
 		return SHRINK_STOP;
@@ -421,10 +423,15 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 		unsigned nr_free = shrink_pages;
 		if (shrink_pages == 0)
 			break;
+
 		pool = &_manager->pools[(i + pool_offset)%NUM_POOLS];
 		/* OK to use static buffer since global mutex is held. */
-		shrink_pages = ttm_page_pool_free(pool, nr_free, true);
-		freed += nr_free - shrink_pages;
+		nr_free_pool = (nr_free >> pool->order);
+		if (nr_free_pool == 0)
+			continue;
+
+		shrink_pages = ttm_page_pool_free(pool, nr_free_pool, true);
+		freed += ((nr_free_pool - shrink_pages) << pool->order);
 	}
 	mutex_unlock(&lock);
 	return freed;
@@ -436,9 +443,12 @@ ttm_pool_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
 {
 	unsigned i;
 	unsigned long count = 0;
+	struct ttm_page_pool *pool;
 
-	for (i = 0; i < NUM_POOLS; ++i)
-		count += _manager->pools[i].npages;
+	for (i = 0; i < NUM_POOLS; ++i) {
+		pool = &_manager->pools[i];
+		count += (pool->npages << pool->order);
+	}
 
 	return count;
 }
@@ -932,7 +942,7 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
 }
 
 static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
-		char *name)
+		char *name, unsigned int order)
 {
 	spin_lock_init(&pool->lock);
 	pool->fill_lock = false;
@@ -940,8 +950,18 @@ static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
 	pool->npages = pool->nfrees = 0;
 	pool->gfp_flags = flags;
 	pool->name = name;
+	pool->order = order;
 }
 
+/**
+ * Actually if TRANSPARENT_HUGEPAGE not enabled, we will not use
+ * wc_pool_huge and uc_pool_huge, so no matter whatever the page
+ * order are for those two pools
+ */
+#ifndef CONFIG_TRANSPARENT_HUGEPAGE
+#define	HPAGE_PMD_ORDER	9
+#endif
+
 int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
 {
 	int ret;
@@ -952,23 +972,23 @@ int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
 
 	_manager = kzalloc(sizeof(*_manager), GFP_KERNEL);
 
-	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc");
+	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc", 0);
 
-	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc");
+	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc", 0);
 
 	ttm_page_pool_init_locked(&_manager->wc_pool_dma32,
-				  GFP_USER | GFP_DMA32, "wc dma");
+				  GFP_USER | GFP_DMA32, "wc dma", 0);
 
 	ttm_page_pool_init_locked(&_manager->uc_pool_dma32,
-				  GFP_USER | GFP_DMA32, "uc dma");
+				  GFP_USER | GFP_DMA32, "uc dma", 0);
 
 	ttm_page_pool_init_locked(&_manager->wc_pool_huge,
 				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP),
-				  "wc huge");
+				  "wc huge", HPAGE_PMD_ORDER);
 
 	ttm_page_pool_init_locked(&_manager->uc_pool_huge,
 				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP)
-				  , "uc huge");
+				  , "uc huge", HPAGE_PMD_ORDER);
 
 	_manager->options.max_size = max_pages;
 	_manager->options.small = SMALL_ALLOCATION;
-- 
2.7.4

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

[PATCH 2/4] drm/ttm: add page order support in ttm_pages_put

[Roger He]

Change-Id: Ia55b206d95812c5afcfd6cec29f580758d1f50f0
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 +++++++++++++++++++++++++++++-------
 1 file changed, 34 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index 0a0c653..de64209 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -285,13 +285,39 @@ static struct ttm_page_pool *ttm_get_pool(int flags, bool huge,
 }
 
 /* set memory back to wb and free the pages. */
-static void ttm_pages_put(struct page *pages[], unsigned npages)
+static void ttm_pages_put(struct page *pages[], unsigned npages,
+		unsigned int order)
 {
-	unsigned i;
-	if (set_pages_array_wb(pages, npages))
-		pr_err("Failed to set %d pages to wb!\n", npages);
-	for (i = 0; i < npages; ++i)
-		__free_page(pages[i]);
+	struct page **pages_to_free = NULL;
+	struct page **pages_array;
+	struct page *p;
+	unsigned int i, j, pages_nr = (1 << order);
+
+	if (order > 0) {
+		pages_to_free = kmalloc_array(pages_nr, sizeof(struct page *),
+					GFP_KERNEL);
+		if (!pages_to_free) {
+			pr_err("Failed to allocate memory for ttm pages put operation\n");
+			return;
+		}
+	}
+
+	for (i = 0; i < npages; ++i) {
+		if (order > 0) {
+			p = pages[i];
+			for (j = 0; j < pages_nr; ++j)
+				pages_to_free[j] = p++;
+
+			pages_array = pages_to_free;
+		} else
+			pages_array = pages;
+
+		if (set_pages_array_wb(pages_array, pages_nr))
+			pr_err("Failed to set %d pages to wb!\n", pages_nr);
+		__free_pages(pages[i], order);
+	}
+
+	kfree(pages_to_free);
 }
 
 static void ttm_pool_update_free_locked(struct ttm_page_pool *pool,
@@ -354,7 +380,7 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
 			 */
 			spin_unlock_irqrestore(&pool->lock, irq_flags);
 
-			ttm_pages_put(pages_to_free, freed_pages);
+			ttm_pages_put(pages_to_free, freed_pages, pool->order);
 			if (likely(nr_free != FREE_ALL_PAGES))
 				nr_free -= freed_pages;
 
@@ -389,7 +415,7 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
 	spin_unlock_irqrestore(&pool->lock, irq_flags);
 
 	if (freed_pages)
-		ttm_pages_put(pages_to_free, freed_pages);
+		ttm_pages_put(pages_to_free, freed_pages, pool->order);
 out:
 	if (pages_to_free != static_buf)
 		kfree(pages_to_free);
-- 
2.7.4

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

Re: [PATCH 2/4] drm/ttm: add page order support in ttm_pages_put

[Christian König]

I would rather completely nuke ttm_pages_put() instead of coming up with 
more workarounds here.

Going to provide a cleanup patch to show what I mean, just give me a minute.

Regards,
Christian.

Am 22.11.2017 um 06:36 schrieb Roger He:
> Change-Id: Ia55b206d95812c5afcfd6cec29f580758d1f50f0
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 +++++++++++++++++++++++++++++-------
>   1 file changed, 34 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 0a0c653..de64209 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -285,13 +285,39 @@ static struct ttm_page_pool *ttm_get_pool(int flags, bool huge,
>   }
>   
>   /* set memory back to wb and free the pages. */
> -static void ttm_pages_put(struct page *pages[], unsigned npages)
> +static void ttm_pages_put(struct page *pages[], unsigned npages,
> +		unsigned int order)
>   {
> -	unsigned i;
> -	if (set_pages_array_wb(pages, npages))
> -		pr_err("Failed to set %d pages to wb!\n", npages);
> -	for (i = 0; i < npages; ++i)
> -		__free_page(pages[i]);
> +	struct page **pages_to_free = NULL;
> +	struct page **pages_array;
> +	struct page *p;
> +	unsigned int i, j, pages_nr = (1 << order);
> +
> +	if (order > 0) {
> +		pages_to_free = kmalloc_array(pages_nr, sizeof(struct page *),
> +					GFP_KERNEL);
> +		if (!pages_to_free) {
> +			pr_err("Failed to allocate memory for ttm pages put operation\n");
> +			return;
> +		}
> +	}
> +
> +	for (i = 0; i < npages; ++i) {
> +		if (order > 0) {
> +			p = pages[i];
> +			for (j = 0; j < pages_nr; ++j)
> +				pages_to_free[j] = p++;
> +
> +			pages_array = pages_to_free;
> +		} else
> +			pages_array = pages;
> +
> +		if (set_pages_array_wb(pages_array, pages_nr))
> +			pr_err("Failed to set %d pages to wb!\n", pages_nr);
> +		__free_pages(pages[i], order);
> +	}
> +
> +	kfree(pages_to_free);
>   }
>   
>   static void ttm_pool_update_free_locked(struct ttm_page_pool *pool,
> @@ -354,7 +380,7 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
>   			 */
>   			spin_unlock_irqrestore(&pool->lock, irq_flags);
>   
> -			ttm_pages_put(pages_to_free, freed_pages);
> +			ttm_pages_put(pages_to_free, freed_pages, pool->order);
>   			if (likely(nr_free != FREE_ALL_PAGES))
>   				nr_free -= freed_pages;
>   
> @@ -389,7 +415,7 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
>   	spin_unlock_irqrestore(&pool->lock, irq_flags);
>   
>   	if (freed_pages)
> -		ttm_pages_put(pages_to_free, freed_pages);
> +		ttm_pages_put(pages_to_free, freed_pages, pool->order);
>   out:
>   	if (pages_to_free != static_buf)
>   		kfree(pages_to_free);


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

RE: [PATCH 2/4] drm/ttm: add page order support in ttm_pages_put

[He, Roger]

Hi Christian:

This is old patch, I already send new patches. And it is clean and simple.  Please check that.

Thanks
Roger(Hongbo.He)

-----Original Message-----
From: Christian König [mailto:ckoenig.leichtzumerken@gmail.com] 
Sent: Wednesday, November 22, 2017 5:29 PM
To: He, Roger <Hongbo.He@amd.com>; amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Cc: Koenig, Christian <Christian.Koenig@amd.com>
Subject: Re: [PATCH 2/4] drm/ttm: add page order support in ttm_pages_put

I would rather completely nuke ttm_pages_put() instead of coming up with more workarounds here.

Going to provide a cleanup patch to show what I mean, just give me a minute.

Regards,
Christian.

Am 22.11.2017 um 06:36 schrieb Roger He:
> Change-Id: Ia55b206d95812c5afcfd6cec29f580758d1f50f0
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 +++++++++++++++++++++++++++++-------
>   1 file changed, 34 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c 
> b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 0a0c653..de64209 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -285,13 +285,39 @@ static struct ttm_page_pool *ttm_get_pool(int flags, bool huge,
>   }
>   
>   /* set memory back to wb and free the pages. */ -static void 
> ttm_pages_put(struct page *pages[], unsigned npages)
> +static void ttm_pages_put(struct page *pages[], unsigned npages,
> +		unsigned int order)
>   {
> -	unsigned i;
> -	if (set_pages_array_wb(pages, npages))
> -		pr_err("Failed to set %d pages to wb!\n", npages);
> -	for (i = 0; i < npages; ++i)
> -		__free_page(pages[i]);
> +	struct page **pages_to_free = NULL;
> +	struct page **pages_array;
> +	struct page *p;
> +	unsigned int i, j, pages_nr = (1 << order);
> +
> +	if (order > 0) {
> +		pages_to_free = kmalloc_array(pages_nr, sizeof(struct page *),
> +					GFP_KERNEL);
> +		if (!pages_to_free) {
> +			pr_err("Failed to allocate memory for ttm pages put operation\n");
> +			return;
> +		}
> +	}
> +
> +	for (i = 0; i < npages; ++i) {
> +		if (order > 0) {
> +			p = pages[i];
> +			for (j = 0; j < pages_nr; ++j)
> +				pages_to_free[j] = p++;
> +
> +			pages_array = pages_to_free;
> +		} else
> +			pages_array = pages;
> +
> +		if (set_pages_array_wb(pages_array, pages_nr))
> +			pr_err("Failed to set %d pages to wb!\n", pages_nr);
> +		__free_pages(pages[i], order);
> +	}
> +
> +	kfree(pages_to_free);
>   }
>   
>   static void ttm_pool_update_free_locked(struct ttm_page_pool *pool, 
> @@ -354,7 +380,7 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
>   			 */
>   			spin_unlock_irqrestore(&pool->lock, irq_flags);
>   
> -			ttm_pages_put(pages_to_free, freed_pages);
> +			ttm_pages_put(pages_to_free, freed_pages, pool->order);
>   			if (likely(nr_free != FREE_ALL_PAGES))
>   				nr_free -= freed_pages;
>   
> @@ -389,7 +415,7 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
>   	spin_unlock_irqrestore(&pool->lock, irq_flags);
>   
>   	if (freed_pages)
> -		ttm_pages_put(pages_to_free, freed_pages);
> +		ttm_pages_put(pages_to_free, freed_pages, pool->order);
>   out:
>   	if (pages_to_free != static_buf)
>   		kfree(pages_to_free);


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

[PATCH 3/4] drm/ttm: add static buffer for ttm_pages_put to support pool shrink

[Roger He]

Change-Id: Ic20c016eb3043d7cfedc2e3648790a017168da6c
Signed-off-by: Roger He <Hongbo.He@amd.com>
---
 drivers/gpu/drm/ttm/ttm_page_alloc.c | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
index de64209..37c2f2f 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -285,20 +285,27 @@ static struct ttm_page_pool *ttm_get_pool(int flags, bool huge,
 }
 
 /* set memory back to wb and free the pages. */
+#define STATIC_BUF_PUT_SIZE 512
 static void ttm_pages_put(struct page *pages[], unsigned npages,
-		unsigned int order)
+		unsigned int order, bool use_static)
 {
+	static struct page *static_buf_put[STATIC_BUF_PUT_SIZE];
 	struct page **pages_to_free = NULL;
 	struct page **pages_array;
 	struct page *p;
 	unsigned int i, j, pages_nr = (1 << order);
 
+	BUG_ON(use_static && pages_nr > STATIC_BUF_PUT_SIZE);
 	if (order > 0) {
-		pages_to_free = kmalloc_array(pages_nr, sizeof(struct page *),
-					GFP_KERNEL);
-		if (!pages_to_free) {
-			pr_err("Failed to allocate memory for ttm pages put operation\n");
-			return;
+		if (use_static)
+			pages_to_free = static_buf_put;
+		else {
+			pages_to_free = kmalloc_array(pages_nr,
+				sizeof(struct page *), GFP_KERNEL);
+			if (!pages_to_free) {
+				pr_err("Failed to allocate memory for ttm pages put operation\n");
+				return;
+			}
 		}
 	}
 
@@ -317,7 +324,8 @@ static void ttm_pages_put(struct page *pages[], unsigned npages,
 		__free_pages(pages[i], order);
 	}
 
-	kfree(pages_to_free);
+	if (!use_static)
+		kfree(pages_to_free);
 }
 
 static void ttm_pool_update_free_locked(struct ttm_page_pool *pool,
@@ -380,7 +388,8 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
 			 */
 			spin_unlock_irqrestore(&pool->lock, irq_flags);
 
-			ttm_pages_put(pages_to_free, freed_pages, pool->order);
+			ttm_pages_put(pages_to_free, freed_pages, pool->order,
+					use_static);
 			if (likely(nr_free != FREE_ALL_PAGES))
 				nr_free -= freed_pages;
 
@@ -415,7 +424,8 @@ static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free,
 	spin_unlock_irqrestore(&pool->lock, irq_flags);
 
 	if (freed_pages)
-		ttm_pages_put(pages_to_free, freed_pages, pool->order);
+		ttm_pages_put(pages_to_free, freed_pages, pool->order,
+			use_static);
 out:
 	if (pages_to_free != static_buf)
 		kfree(pages_to_free);
-- 
2.7.4

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

[PATCH 4/4] drm/ttm: free one in huge pool even shrink request less than one element

[Roger He]

Change-Id: Id8bd4d1ecff9f3ab14355e2dbd1c59b9fe824e01
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 37c2f2f..f80fc5b 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -463,11 +463,13 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 		pool = &_manager->pools[(i + pool_offset)%NUM_POOLS];
 		/* OK to use static buffer since global mutex is held. */
 		nr_free_pool = (nr_free >> pool->order);
-		if (nr_free_pool == 0)
-			continue;
+		if (!nr_free_pool && pool->order)
+			nr_free_pool = 1;
 
 		shrink_pages = ttm_page_pool_free(pool, nr_free_pool, true);
 		freed += ((nr_free_pool - shrink_pages) << pool->order);
+		if (freed >= sc->nr_to_scan)
+			break;
 	}
 	mutex_unlock(&lock);
 	return freed;
-- 
2.7.4

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

Re: [PATCH 1/4] drm/ttm: add page order in page pool

[Chunming Zhou]

On 2017年11月22日 13:36, Roger He wrote:
> to indicate page order for each element in the pool
>
> Change-Id: Ic609925ca5d2a5d4ad49d6becf505388ce3624cf
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 ++++++++++++++++++++++++++----------
>   1 file changed, 31 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 72ea037..0a0c653 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -81,6 +81,7 @@ struct ttm_page_pool {
>   	char			*name;
>   	unsigned long		nfrees;
>   	unsigned long		nrefills;
> +	unsigned int		order;
>   };
>   
>   /**
> @@ -412,6 +413,7 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>   	struct ttm_page_pool *pool;
>   	int shrink_pages = sc->nr_to_scan;
>   	unsigned long freed = 0;
> +	unsigned int nr_free_pool;
>   
>   	if (!mutex_trylock(&lock))
>   		return SHRINK_STOP;
> @@ -421,10 +423,15 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>   		unsigned nr_free = shrink_pages;
>   		if (shrink_pages == 0)
>   			break;
> +
>   		pool = &_manager->pools[(i + pool_offset)%NUM_POOLS];
>   		/* OK to use static buffer since global mutex is held. */
> -		shrink_pages = ttm_page_pool_free(pool, nr_free, true);
> -		freed += nr_free - shrink_pages;
> +		nr_free_pool = (nr_free >> pool->order);
How about nr_free_pool = roundup(nr_free >> pool->order, 1)?
  this way, your patch#4 is not need.

Regards,
David Zhou
> +		if (nr_free_pool == 0)
> +			continue;
> +
> +		shrink_pages = ttm_page_pool_free(pool, nr_free_pool, true);
> +		freed += ((nr_free_pool - shrink_pages) << pool->order);
>   	}
>   	mutex_unlock(&lock);
>   	return freed;
> @@ -436,9 +443,12 @@ ttm_pool_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
>   {
>   	unsigned i;
>   	unsigned long count = 0;
> +	struct ttm_page_pool *pool;
>   
> -	for (i = 0; i < NUM_POOLS; ++i)
> -		count += _manager->pools[i].npages;
> +	for (i = 0; i < NUM_POOLS; ++i) {
> +		pool = &_manager->pools[i];
> +		count += (pool->npages << pool->order);
> +	}
>   
>   	return count;
>   }
> @@ -932,7 +942,7 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
>   }
>   
>   static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
> -		char *name)
> +		char *name, unsigned int order)
>   {
>   	spin_lock_init(&pool->lock);
>   	pool->fill_lock = false;
> @@ -940,8 +950,18 @@ static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
>   	pool->npages = pool->nfrees = 0;
>   	pool->gfp_flags = flags;
>   	pool->name = name;
> +	pool->order = order;
>   }
>   
> +/**
> + * Actually if TRANSPARENT_HUGEPAGE not enabled, we will not use
> + * wc_pool_huge and uc_pool_huge, so no matter whatever the page
> + * order are for those two pools
> + */
> +#ifndef CONFIG_TRANSPARENT_HUGEPAGE
> +#define	HPAGE_PMD_ORDER	9
> +#endif
> +
>   int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
>   {
>   	int ret;
> @@ -952,23 +972,23 @@ int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
>   
>   	_manager = kzalloc(sizeof(*_manager), GFP_KERNEL);
>   
> -	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc");
> +	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc", 0);
>   
> -	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc");
> +	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->wc_pool_dma32,
> -				  GFP_USER | GFP_DMA32, "wc dma");
> +				  GFP_USER | GFP_DMA32, "wc dma", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->uc_pool_dma32,
> -				  GFP_USER | GFP_DMA32, "uc dma");
> +				  GFP_USER | GFP_DMA32, "uc dma", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->wc_pool_huge,
>   				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP),
> -				  "wc huge");
> +				  "wc huge", HPAGE_PMD_ORDER);
>   
>   	ttm_page_pool_init_locked(&_manager->uc_pool_huge,
>   				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP)
> -				  , "uc huge");
> +				  , "uc huge", HPAGE_PMD_ORDER);
>   
>   	_manager->options.max_size = max_pages;
>   	_manager->options.small = SMALL_ALLOCATION;

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

Re: [PATCH 1/4] drm/ttm: add page order in page pool

[Christian König]

Am 22.11.2017 um 06:36 schrieb Roger He:
> to indicate page order for each element in the pool
>
> Change-Id: Ic609925ca5d2a5d4ad49d6becf505388ce3624cf
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 ++++++++++++++++++++++++++----------
>   1 file changed, 31 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 72ea037..0a0c653 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -81,6 +81,7 @@ struct ttm_page_pool {
>   	char			*name;
>   	unsigned long		nfrees;
>   	unsigned long		nrefills;
> +	unsigned int		order;
>   };
>   
>   /**
> @@ -412,6 +413,7 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>   	struct ttm_page_pool *pool;
>   	int shrink_pages = sc->nr_to_scan;
>   	unsigned long freed = 0;
> +	unsigned int nr_free_pool;
>   
>   	if (!mutex_trylock(&lock))
>   		return SHRINK_STOP;
> @@ -421,10 +423,15 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>   		unsigned nr_free = shrink_pages;
>   		if (shrink_pages == 0)
>   			break;
> +
>   		pool = &_manager->pools[(i + pool_offset)%NUM_POOLS];
>   		/* OK to use static buffer since global mutex is held. */
> -		shrink_pages = ttm_page_pool_free(pool, nr_free, true);
> -		freed += nr_free - shrink_pages;
> +		nr_free_pool = (nr_free >> pool->order);
> +		if (nr_free_pool == 0)
> +			continue;
> +
> +		shrink_pages = ttm_page_pool_free(pool, nr_free_pool, true);
> +		freed += ((nr_free_pool - shrink_pages) << pool->order);
>   	}
>   	mutex_unlock(&lock);
>   	return freed;
> @@ -436,9 +443,12 @@ ttm_pool_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
>   {
>   	unsigned i;
>   	unsigned long count = 0;
> +	struct ttm_page_pool *pool;
>   
> -	for (i = 0; i < NUM_POOLS; ++i)
> -		count += _manager->pools[i].npages;
> +	for (i = 0; i < NUM_POOLS; ++i) {
> +		pool = &_manager->pools[i];
> +		count += (pool->npages << pool->order);
> +	}
>   
>   	return count;
>   }
> @@ -932,7 +942,7 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
>   }
>   
>   static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
> -		char *name)
> +		char *name, unsigned int order)
>   {
>   	spin_lock_init(&pool->lock);
>   	pool->fill_lock = false;
> @@ -940,8 +950,18 @@ static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
>   	pool->npages = pool->nfrees = 0;
>   	pool->gfp_flags = flags;
>   	pool->name = name;
> +	pool->order = order;
>   }
>   
> +/**
> + * Actually if TRANSPARENT_HUGEPAGE not enabled, we will not use
> + * wc_pool_huge and uc_pool_huge, so no matter whatever the page
> + * order are for those two pools
> + */
> +#ifndef CONFIG_TRANSPARENT_HUGEPAGE
> +#define	HPAGE_PMD_ORDER	9
> +#endif
> +

That still won't work and sorry I wasn't 100% clear in the last mail.

When CONFIG_TRANSPARENT_HUGEPAGE isn't set HPAGE_PMD_ORDER is defined as 
BUILD_BUG().

So you will still run into problems when that config option isn't set.

>   int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
>   {
>   	int ret;

I suggest to just handle it here like this

#ifdef CONFIG_TRANSPARENT_HUGEPAGE
     unsigned order = HPAGE_PMD_ORDER;
#else
     unsigned order = 0;
#endif

Apart from that the patch looks good to me,
Christian.

> @@ -952,23 +972,23 @@ int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
>   
>   	_manager = kzalloc(sizeof(*_manager), GFP_KERNEL);
>   
> -	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc");
> +	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc", 0);
>   
> -	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc");
> +	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->wc_pool_dma32,
> -				  GFP_USER | GFP_DMA32, "wc dma");
> +				  GFP_USER | GFP_DMA32, "wc dma", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->uc_pool_dma32,
> -				  GFP_USER | GFP_DMA32, "uc dma");
> +				  GFP_USER | GFP_DMA32, "uc dma", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->wc_pool_huge,
>   				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP),
> -				  "wc huge");
> +				  "wc huge", HPAGE_PMD_ORDER);
>   
>   	ttm_page_pool_init_locked(&_manager->uc_pool_huge,
>   				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP)
> -				  , "uc huge");
> +				  , "uc huge", HPAGE_PMD_ORDER);
>   
>   	_manager->options.max_size = max_pages;
>   	_manager->options.small = SMALL_ALLOCATION;


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

RE: [PATCH 1/4] drm/ttm: add page order in page pool

[He, Roger]

-----Original Message-----
From: Koenig, Christian 
Sent: Wednesday, November 22, 2017 3:48 PM
To: He, Roger <Hongbo.He@amd.com>; amd-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org
Subject: Re: [PATCH 1/4] drm/ttm: add page order in page pool

Am 22.11.2017 um 06:36 schrieb Roger He:
> to indicate page order for each element in the pool
>
> Change-Id: Ic609925ca5d2a5d4ad49d6becf505388ce3624cf
> Signed-off-by: Roger He <Hongbo.He@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 ++++++++++++++++++++++++++----------
>   1 file changed, 31 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c 
> b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 72ea037..0a0c653 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -81,6 +81,7 @@ struct ttm_page_pool {
>   	char			*name;
>   	unsigned long		nfrees;
>   	unsigned long		nrefills;
> +	unsigned int		order;
>   };
>   
>   /**
> @@ -412,6 +413,7 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>   	struct ttm_page_pool *pool;
>   	int shrink_pages = sc->nr_to_scan;
>   	unsigned long freed = 0;
> +	unsigned int nr_free_pool;
>   
>   	if (!mutex_trylock(&lock))
>   		return SHRINK_STOP;
> @@ -421,10 +423,15 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>   		unsigned nr_free = shrink_pages;
>   		if (shrink_pages == 0)
>   			break;
> +
>   		pool = &_manager->pools[(i + pool_offset)%NUM_POOLS];
>   		/* OK to use static buffer since global mutex is held. */
> -		shrink_pages = ttm_page_pool_free(pool, nr_free, true);
> -		freed += nr_free - shrink_pages;
> +		nr_free_pool = (nr_free >> pool->order);
> +		if (nr_free_pool == 0)
> +			continue;
> +
> +		shrink_pages = ttm_page_pool_free(pool, nr_free_pool, true);
> +		freed += ((nr_free_pool - shrink_pages) << pool->order);
>   	}
>   	mutex_unlock(&lock);
>   	return freed;
> @@ -436,9 +443,12 @@ ttm_pool_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
>   {
>   	unsigned i;
>   	unsigned long count = 0;
> +	struct ttm_page_pool *pool;
>   
> -	for (i = 0; i < NUM_POOLS; ++i)
> -		count += _manager->pools[i].npages;
> +	for (i = 0; i < NUM_POOLS; ++i) {
> +		pool = &_manager->pools[i];
> +		count += (pool->npages << pool->order);
> +	}
>   
>   	return count;
>   }
> @@ -932,7 +942,7 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
>   }
>   
>   static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
> -		char *name)
> +		char *name, unsigned int order)
>   {
>   	spin_lock_init(&pool->lock);
>   	pool->fill_lock = false;
> @@ -940,8 +950,18 @@ static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
>   	pool->npages = pool->nfrees = 0;
>   	pool->gfp_flags = flags;
>   	pool->name = name;
> +	pool->order = order;
>   }
>   
> +/**
> + * Actually if TRANSPARENT_HUGEPAGE not enabled, we will not use
> + * wc_pool_huge and uc_pool_huge, so no matter whatever the page
> + * order are for those two pools
> + */
> +#ifndef CONFIG_TRANSPARENT_HUGEPAGE
> +#define	HPAGE_PMD_ORDER	9
> +#endif
> +

That still won't work and sorry I wasn't 100% clear in the last mail.

When CONFIG_TRANSPARENT_HUGEPAGE isn't set HPAGE_PMD_ORDER is defined as BUILD_BUG().

So you will still run into problems when that config option isn't set.

>   int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
>   {
>   	int ret;

I suggest to just handle it here like this

#ifdef CONFIG_TRANSPARENT_HUGEPAGE
     unsigned order = HPAGE_PMD_ORDER;
#else
     unsigned order = 0;
#endif

Apart from that the patch looks good to me, Christian.


Ok, going to modify it. Thanks!

Thanks
Roger(Hongbo.He)

> @@ -952,23 +972,23 @@ int ttm_page_alloc_init(struct ttm_mem_global 
> *glob, unsigned max_pages)
>   
>   	_manager = kzalloc(sizeof(*_manager), GFP_KERNEL);
>   
> -	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc");
> +	ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc", 
> +0);
>   
> -	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc");
> +	ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc", 
> +0);
>   
>   	ttm_page_pool_init_locked(&_manager->wc_pool_dma32,
> -				  GFP_USER | GFP_DMA32, "wc dma");
> +				  GFP_USER | GFP_DMA32, "wc dma", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->uc_pool_dma32,
> -				  GFP_USER | GFP_DMA32, "uc dma");
> +				  GFP_USER | GFP_DMA32, "uc dma", 0);
>   
>   	ttm_page_pool_init_locked(&_manager->wc_pool_huge,
>   				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP),
> -				  "wc huge");
> +				  "wc huge", HPAGE_PMD_ORDER);
>   
>   	ttm_page_pool_init_locked(&_manager->uc_pool_huge,
>   				  GFP_TRANSHUGE	& ~(__GFP_MOVABLE | __GFP_COMP)
> -				  , "uc huge");
> +				  , "uc huge", HPAGE_PMD_ORDER);
>   
>   	_manager->options.max_size = max_pages;
>   	_manager->options.small = SMALL_ALLOCATION;


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

[PATCH 0/4] *** fix memory leak for HUGE PAGE ***

[Roger He]

1. add page order into pool to support gtt huge page
2. add page order support when ttm pages put
3. update memory shrinker with order handling


Roger He (4):
  drm/ttm: add page order in page pool
  drm/ttm: use NUM_PAGES_TO_ALLOC always
  drm/ttm: add page order support in ttm_pages_put
  drm/ttm: free one in huge pool even shrink request less than one
    element

 drivers/gpu/drm/ttm/ttm_page_alloc.c | 80 ++++++++++++++++++++++++++----------
 1 file changed, 59 insertions(+), 21 deletions(-)

-- 
2.7.4

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

[PATCH 4/4] drm/ttm: free one in huge pool even shrink request less than one element

[Roger He]

Change-Id: Id8bd4d1ecff9f3ab14355e2dbd1c59b9fe824e01
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 90546fd..c194a51 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
@@ -453,11 +453,13 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 		pool = &_manager->pools[(i + pool_offset)%NUM_POOLS];
 		/* OK to use static buffer since global mutex is held. */
 		nr_free_pool = (nr_free >> pool->order);
-		if (nr_free_pool == 0)
-			continue;
+		if (!nr_free_pool && pool->order)
+			nr_free_pool = 1;
 
 		shrink_pages = ttm_page_pool_free(pool, nr_free_pool, true);
 		freed += ((nr_free_pool - shrink_pages) << pool->order);
+		if (freed > sc->nr_to_scan)
+			break;
 	}
 	mutex_unlock(&lock);
 	return freed;
-- 
2.7.4

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