* [PATCH 1/2] mm/vmscan: add sync_shrinkers function @ 2021-04-15 11:56 Christian König 2021-04-15 11:56 ` [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 Christian König 2021-04-15 13:23 ` [PATCH 1/2] mm/vmscan: add sync_shrinkers function Daniel Vetter 0 siblings, 2 replies; 8+ messages in thread From: Christian König @ 2021-04-15 11:56 UTC (permalink / raw) To: linux-mm, linux-kernel, dri-devel; +Cc: akpm, ray.huang, vbabka To be able to switch to a spinlock and reduce lock contention in the TTM shrinker we don't want to hold a mutex while unmapping and freeing pages from the pool. But then we somehow need to prevent a race between (for example) the shrinker trying to free pages and hotplug trying to remove the device which those pages belong to. Taking and releasing the shrinker semaphore on the write side after unmapping and freeing all pages should make sure that no shrinker is running in paralell any more. Signed-off-by: Christian König <christian.koenig@amd.com> --- include/linux/shrinker.h | 1 + mm/vmscan.c | 10 ++++++++++ 2 files changed, 11 insertions(+) diff --git a/include/linux/shrinker.h b/include/linux/shrinker.h index 0f80123650e2..6b75dc372fce 100644 --- a/include/linux/shrinker.h +++ b/include/linux/shrinker.h @@ -92,4 +92,5 @@ extern void register_shrinker_prepared(struct shrinker *shrinker); extern int register_shrinker(struct shrinker *shrinker); extern void unregister_shrinker(struct shrinker *shrinker); extern void free_prealloced_shrinker(struct shrinker *shrinker); +extern void sync_shrinkers(void); #endif diff --git a/mm/vmscan.c b/mm/vmscan.c index 562e87cbd7a1..46cd9c215d73 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -408,6 +408,16 @@ void unregister_shrinker(struct shrinker *shrinker) } EXPORT_SYMBOL(unregister_shrinker); +/** + * sync_shrinker - Wait for all running shrinkers to complete. + */ +void sync_shrinkers(void) +{ + down_write(&shrinker_rwsem); + up_write(&shrinker_rwsem); +} +EXPORT_SYMBOL(sync_shrinkers); + #define SHRINK_BATCH 128 static unsigned long do_shrink_slab(struct shrink_control *shrinkctl, -- 2.25.1 _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply related [flat|nested] 8+ messages in thread
* [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 2021-04-15 11:56 [PATCH 1/2] mm/vmscan: add sync_shrinkers function Christian König @ 2021-04-15 11:56 ` Christian König 2021-04-15 14:33 ` Huang Rui 2021-04-15 20:33 ` Andrew Morton 2021-04-15 13:23 ` [PATCH 1/2] mm/vmscan: add sync_shrinkers function Daniel Vetter 1 sibling, 2 replies; 8+ messages in thread From: Christian König @ 2021-04-15 11:56 UTC (permalink / raw) To: linux-mm, linux-kernel, dri-devel; +Cc: akpm, ray.huang, vbabka Switch back to using a spinlock again by moving the IOMMU unmap outside of the locked region. v2: Add a comment explaining why we need sync_shrinkers(). Signed-off-by: Christian König <christian.koenig@amd.com> --- drivers/gpu/drm/ttm/ttm_pool.c | 44 +++++++++++++++++----------------- 1 file changed, 22 insertions(+), 22 deletions(-) diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c index cb38b1a17b09..955836d569cc 100644 --- a/drivers/gpu/drm/ttm/ttm_pool.c +++ b/drivers/gpu/drm/ttm/ttm_pool.c @@ -70,7 +70,7 @@ static struct ttm_pool_type global_uncached[MAX_ORDER]; static struct ttm_pool_type global_dma32_write_combined[MAX_ORDER]; static struct ttm_pool_type global_dma32_uncached[MAX_ORDER]; -static struct mutex shrinker_lock; +static spinlock_t shrinker_lock; static struct list_head shrinker_list; static struct shrinker mm_shrinker; @@ -263,9 +263,9 @@ static void ttm_pool_type_init(struct ttm_pool_type *pt, struct ttm_pool *pool, spin_lock_init(&pt->lock); INIT_LIST_HEAD(&pt->pages); - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); list_add_tail(&pt->shrinker_list, &shrinker_list); - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); } /* Remove a pool_type from the global shrinker list and free all pages */ @@ -273,9 +273,9 @@ static void ttm_pool_type_fini(struct ttm_pool_type *pt) { struct page *p; - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); list_del(&pt->shrinker_list); - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); while ((p = ttm_pool_type_take(pt))) ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); @@ -313,24 +313,19 @@ static struct ttm_pool_type *ttm_pool_select_type(struct ttm_pool *pool, static unsigned int ttm_pool_shrink(void) { struct ttm_pool_type *pt; - unsigned int num_freed; struct page *p; - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); pt = list_first_entry(&shrinker_list, typeof(*pt), shrinker_list); + list_move_tail(&pt->shrinker_list, &shrinker_list); + spin_unlock(&shrinker_lock); p = ttm_pool_type_take(pt); - if (p) { - ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); - num_freed = 1 << pt->order; - } else { - num_freed = 0; - } - - list_move_tail(&pt->shrinker_list, &shrinker_list); - mutex_unlock(&shrinker_lock); + if (!p) + return 0; - return num_freed; + ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); + return 1 << pt->order; } /* Return the allocation order based for a page */ @@ -530,6 +525,11 @@ void ttm_pool_fini(struct ttm_pool *pool) for (j = 0; j < MAX_ORDER; ++j) ttm_pool_type_fini(&pool->caching[i].orders[j]); } + + /* We removed the pool types from the LRU, but we need to also make sure + * that no shrinker is concurrently freeing pages from the pool. + */ + sync_shrinkers(); } /* As long as pages are available make sure to release at least one */ @@ -604,7 +604,7 @@ static int ttm_pool_debugfs_globals_show(struct seq_file *m, void *data) { ttm_pool_debugfs_header(m); - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); seq_puts(m, "wc\t:"); ttm_pool_debugfs_orders(global_write_combined, m); seq_puts(m, "uc\t:"); @@ -613,7 +613,7 @@ static int ttm_pool_debugfs_globals_show(struct seq_file *m, void *data) ttm_pool_debugfs_orders(global_dma32_write_combined, m); seq_puts(m, "uc 32\t:"); ttm_pool_debugfs_orders(global_dma32_uncached, m); - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); ttm_pool_debugfs_footer(m); @@ -640,7 +640,7 @@ int ttm_pool_debugfs(struct ttm_pool *pool, struct seq_file *m) ttm_pool_debugfs_header(m); - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i) { seq_puts(m, "DMA "); switch (i) { @@ -656,7 +656,7 @@ int ttm_pool_debugfs(struct ttm_pool *pool, struct seq_file *m) } ttm_pool_debugfs_orders(pool->caching[i].orders, m); } - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); ttm_pool_debugfs_footer(m); return 0; @@ -693,7 +693,7 @@ int ttm_pool_mgr_init(unsigned long num_pages) if (!page_pool_size) page_pool_size = num_pages; - mutex_init(&shrinker_lock); + spin_lock_init(&shrinker_lock); INIT_LIST_HEAD(&shrinker_list); for (i = 0; i < MAX_ORDER; ++i) { -- 2.25.1 _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 2021-04-15 11:56 ` [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 Christian König @ 2021-04-15 14:33 ` Huang Rui 2021-04-15 20:33 ` Andrew Morton 1 sibling, 0 replies; 8+ messages in thread From: Huang Rui @ 2021-04-15 14:33 UTC (permalink / raw) To: Christian König; +Cc: linux-kernel, dri-devel, linux-mm, akpm, vbabka On Thu, Apr 15, 2021 at 07:56:24PM +0800, Christian König wrote: > Switch back to using a spinlock again by moving the IOMMU unmap outside > of the locked region. > > v2: Add a comment explaining why we need sync_shrinkers(). > > Signed-off-by: Christian König <christian.koenig@amd.com> Series look good for me as well. Acked-by: Huang Rui <ray.huang@amd.com> > --- > drivers/gpu/drm/ttm/ttm_pool.c | 44 +++++++++++++++++----------------- > 1 file changed, 22 insertions(+), 22 deletions(-) > > diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c > index cb38b1a17b09..955836d569cc 100644 > --- a/drivers/gpu/drm/ttm/ttm_pool.c > +++ b/drivers/gpu/drm/ttm/ttm_pool.c > @@ -70,7 +70,7 @@ static struct ttm_pool_type global_uncached[MAX_ORDER]; > static struct ttm_pool_type global_dma32_write_combined[MAX_ORDER]; > static struct ttm_pool_type global_dma32_uncached[MAX_ORDER]; > > -static struct mutex shrinker_lock; > +static spinlock_t shrinker_lock; > static struct list_head shrinker_list; > static struct shrinker mm_shrinker; > > @@ -263,9 +263,9 @@ static void ttm_pool_type_init(struct ttm_pool_type *pt, struct ttm_pool *pool, > spin_lock_init(&pt->lock); > INIT_LIST_HEAD(&pt->pages); > > - mutex_lock(&shrinker_lock); > + spin_lock(&shrinker_lock); > list_add_tail(&pt->shrinker_list, &shrinker_list); > - mutex_unlock(&shrinker_lock); > + spin_unlock(&shrinker_lock); > } > > /* Remove a pool_type from the global shrinker list and free all pages */ > @@ -273,9 +273,9 @@ static void ttm_pool_type_fini(struct ttm_pool_type *pt) > { > struct page *p; > > - mutex_lock(&shrinker_lock); > + spin_lock(&shrinker_lock); > list_del(&pt->shrinker_list); > - mutex_unlock(&shrinker_lock); > + spin_unlock(&shrinker_lock); > > while ((p = ttm_pool_type_take(pt))) > ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); > @@ -313,24 +313,19 @@ static struct ttm_pool_type *ttm_pool_select_type(struct ttm_pool *pool, > static unsigned int ttm_pool_shrink(void) > { > struct ttm_pool_type *pt; > - unsigned int num_freed; > struct page *p; > > - mutex_lock(&shrinker_lock); > + spin_lock(&shrinker_lock); > pt = list_first_entry(&shrinker_list, typeof(*pt), shrinker_list); > + list_move_tail(&pt->shrinker_list, &shrinker_list); > + spin_unlock(&shrinker_lock); > > p = ttm_pool_type_take(pt); > - if (p) { > - ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); > - num_freed = 1 << pt->order; > - } else { > - num_freed = 0; > - } > - > - list_move_tail(&pt->shrinker_list, &shrinker_list); > - mutex_unlock(&shrinker_lock); > + if (!p) > + return 0; > > - return num_freed; > + ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); > + return 1 << pt->order; > } > > /* Return the allocation order based for a page */ > @@ -530,6 +525,11 @@ void ttm_pool_fini(struct ttm_pool *pool) > for (j = 0; j < MAX_ORDER; ++j) > ttm_pool_type_fini(&pool->caching[i].orders[j]); > } > + > + /* We removed the pool types from the LRU, but we need to also make sure > + * that no shrinker is concurrently freeing pages from the pool. > + */ > + sync_shrinkers(); > } > > /* As long as pages are available make sure to release at least one */ > @@ -604,7 +604,7 @@ static int ttm_pool_debugfs_globals_show(struct seq_file *m, void *data) > { > ttm_pool_debugfs_header(m); > > - mutex_lock(&shrinker_lock); > + spin_lock(&shrinker_lock); > seq_puts(m, "wc\t:"); > ttm_pool_debugfs_orders(global_write_combined, m); > seq_puts(m, "uc\t:"); > @@ -613,7 +613,7 @@ static int ttm_pool_debugfs_globals_show(struct seq_file *m, void *data) > ttm_pool_debugfs_orders(global_dma32_write_combined, m); > seq_puts(m, "uc 32\t:"); > ttm_pool_debugfs_orders(global_dma32_uncached, m); > - mutex_unlock(&shrinker_lock); > + spin_unlock(&shrinker_lock); > > ttm_pool_debugfs_footer(m); > > @@ -640,7 +640,7 @@ int ttm_pool_debugfs(struct ttm_pool *pool, struct seq_file *m) > > ttm_pool_debugfs_header(m); > > - mutex_lock(&shrinker_lock); > + spin_lock(&shrinker_lock); > for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i) { > seq_puts(m, "DMA "); > switch (i) { > @@ -656,7 +656,7 @@ int ttm_pool_debugfs(struct ttm_pool *pool, struct seq_file *m) > } > ttm_pool_debugfs_orders(pool->caching[i].orders, m); > } > - mutex_unlock(&shrinker_lock); > + spin_unlock(&shrinker_lock); > > ttm_pool_debugfs_footer(m); > return 0; > @@ -693,7 +693,7 @@ int ttm_pool_mgr_init(unsigned long num_pages) > if (!page_pool_size) > page_pool_size = num_pages; > > - mutex_init(&shrinker_lock); > + spin_lock_init(&shrinker_lock); > INIT_LIST_HEAD(&shrinker_list); > > for (i = 0; i < MAX_ORDER; ++i) { > -- > 2.25.1 > _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 2021-04-15 11:56 ` [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 Christian König 2021-04-15 14:33 ` Huang Rui @ 2021-04-15 20:33 ` Andrew Morton 2021-04-16 7:08 ` Christian König 1 sibling, 1 reply; 8+ messages in thread From: Andrew Morton @ 2021-04-15 20:33 UTC (permalink / raw) To: Christian König Cc: linux-kernel, dri-devel, linux-mm, ray.huang, vbabka On Thu, 15 Apr 2021 13:56:24 +0200 "Christian König" <ckoenig.leichtzumerken@gmail.com> wrote: > @@ -530,6 +525,11 @@ void ttm_pool_fini(struct ttm_pool *pool) > for (j = 0; j < MAX_ORDER; ++j) > ttm_pool_type_fini(&pool->caching[i].orders[j]); > } > + > + /* We removed the pool types from the LRU, but we need to also make sure > + * that no shrinker is concurrently freeing pages from the pool. > + */ > + sync_shrinkers(); It isn't immediately clear to me how this works. ttm_pool_fini() has already freed all the pages hasn't it? So why would it care if some shrinkers are still playing with the pages? Or is it the case that ttm_pool_fini() is assuming that there will be some further action against these pages, which requires that shrinkers no longer be accessing the pages and which further assumes that future shrinker invocations will not be able to look up these pages? IOW, a bit more explanation about the dynamics here would help! _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 2021-04-15 20:33 ` Andrew Morton @ 2021-04-16 7:08 ` Christian König 2021-04-26 11:15 ` Christian König 0 siblings, 1 reply; 8+ messages in thread From: Christian König @ 2021-04-16 7:08 UTC (permalink / raw) To: Andrew Morton; +Cc: linux-kernel, dri-devel, linux-mm, ray.huang, vbabka Am 15.04.21 um 22:33 schrieb Andrew Morton: > On Thu, 15 Apr 2021 13:56:24 +0200 "Christian König" <ckoenig.leichtzumerken@gmail.com> wrote: > >> @@ -530,6 +525,11 @@ void ttm_pool_fini(struct ttm_pool *pool) >> for (j = 0; j < MAX_ORDER; ++j) >> ttm_pool_type_fini(&pool->caching[i].orders[j]); >> } >> + >> + /* We removed the pool types from the LRU, but we need to also make sure >> + * that no shrinker is concurrently freeing pages from the pool. >> + */ >> + sync_shrinkers(); > It isn't immediately clear to me how this works. ttm_pool_fini() has > already freed all the pages hasn't it? So why would it care if some > shrinkers are still playing with the pages? Yes ttm_pool_fini() has freed up all pages which had been in the pool when the function was called. But the problem is it is possible that a parallel running shrinker has taken a page from the pool and is in the process of freeing it up. When I return here the pool structure and especially the device structure are freed while the parallel running shrinker is still using them. I could go for a design where we have one shrinker per device instead, but that would put a bit to much pressure on the pool in my opinion. > Or is it the case that ttm_pool_fini() is assuming that there will be > some further action against these pages, which requires that shrinkers > no longer be accessing the pages and which further assumes that future > shrinker invocations will not be able to look up these pages? > > IOW, a bit more explanation about the dynamics here would help! Sorry, I'm not a native speaker of English and sometimes still have a hard time explaining things. Regards, Christian. _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 2021-04-16 7:08 ` Christian König @ 2021-04-26 11:15 ` Christian König 0 siblings, 0 replies; 8+ messages in thread From: Christian König @ 2021-04-26 11:15 UTC (permalink / raw) To: Andrew Morton; +Cc: linux-kernel, dri-devel, linux-mm, ray.huang, vbabka Just a gentle ping? Are you ok with this explanation Andrew or should I look for a different approach? Thanks, Christian. Am 16.04.21 um 09:08 schrieb Christian König: > Am 15.04.21 um 22:33 schrieb Andrew Morton: >> On Thu, 15 Apr 2021 13:56:24 +0200 "Christian König" >> <ckoenig.leichtzumerken@gmail.com> wrote: >> >>> @@ -530,6 +525,11 @@ void ttm_pool_fini(struct ttm_pool *pool) >>> for (j = 0; j < MAX_ORDER; ++j) >>> ttm_pool_type_fini(&pool->caching[i].orders[j]); >>> } >>> + >>> + /* We removed the pool types from the LRU, but we need to also >>> make sure >>> + * that no shrinker is concurrently freeing pages from the pool. >>> + */ >>> + sync_shrinkers(); >> It isn't immediately clear to me how this works. ttm_pool_fini() has >> already freed all the pages hasn't it? So why would it care if some >> shrinkers are still playing with the pages? > > Yes ttm_pool_fini() has freed up all pages which had been in the pool > when the function was called. > > But the problem is it is possible that a parallel running shrinker has > taken a page from the pool and is in the process of freeing it up. > > When I return here the pool structure and especially the device > structure are freed while the parallel running shrinker is still using > them. > > I could go for a design where we have one shrinker per device instead, > but that would put a bit to much pressure on the pool in my opinion. > >> Or is it the case that ttm_pool_fini() is assuming that there will be >> some further action against these pages, which requires that shrinkers >> no longer be accessing the pages and which further assumes that future >> shrinker invocations will not be able to look up these pages? >> >> IOW, a bit more explanation about the dynamics here would help! > > Sorry, I'm not a native speaker of English and sometimes still have a > hard time explaining things. > > Regards, > Christian. _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 1/2] mm/vmscan: add sync_shrinkers function 2021-04-15 11:56 [PATCH 1/2] mm/vmscan: add sync_shrinkers function Christian König 2021-04-15 11:56 ` [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 Christian König @ 2021-04-15 13:23 ` Daniel Vetter 1 sibling, 0 replies; 8+ messages in thread From: Daniel Vetter @ 2021-04-15 13:23 UTC (permalink / raw) To: Christian König Cc: linux-kernel, dri-devel, linux-mm, ray.huang, akpm, vbabka On Thu, Apr 15, 2021 at 01:56:23PM +0200, Christian König wrote: > To be able to switch to a spinlock and reduce lock contention in the TTM > shrinker we don't want to hold a mutex while unmapping and freeing pages > from the pool. > > But then we somehow need to prevent a race between (for example) the shrinker > trying to free pages and hotplug trying to remove the device which those pages > belong to. > > Taking and releasing the shrinker semaphore on the write side after > unmapping and freeing all pages should make sure that no shrinker is running in > paralell any more. > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > include/linux/shrinker.h | 1 + > mm/vmscan.c | 10 ++++++++++ > 2 files changed, 11 insertions(+) > > diff --git a/include/linux/shrinker.h b/include/linux/shrinker.h > index 0f80123650e2..6b75dc372fce 100644 > --- a/include/linux/shrinker.h > +++ b/include/linux/shrinker.h > @@ -92,4 +92,5 @@ extern void register_shrinker_prepared(struct shrinker *shrinker); > extern int register_shrinker(struct shrinker *shrinker); > extern void unregister_shrinker(struct shrinker *shrinker); > extern void free_prealloced_shrinker(struct shrinker *shrinker); > +extern void sync_shrinkers(void); > #endif > diff --git a/mm/vmscan.c b/mm/vmscan.c > index 562e87cbd7a1..46cd9c215d73 100644 > --- a/mm/vmscan.c > +++ b/mm/vmscan.c > @@ -408,6 +408,16 @@ void unregister_shrinker(struct shrinker *shrinker) > } > EXPORT_SYMBOL(unregister_shrinker); > > +/** > + * sync_shrinker - Wait for all running shrinkers to complete. Maybe make it clear this is a barrier type thing, it wont stop shrinkers at all, just synchronize with them. Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> But needs an ack from Andrew for merging through drm-misc-next before we push it there. -Daniel > + */ > +void sync_shrinkers(void) > +{ > + down_write(&shrinker_rwsem); > + up_write(&shrinker_rwsem); > +} > +EXPORT_SYMBOL(sync_shrinkers); > + > #define SHRINK_BATCH 128 > > static unsigned long do_shrink_slab(struct shrink_control *shrinkctl, > -- > 2.25.1 > -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 8+ messages in thread
* Adding sync_shrinkers function for TTM pool optimization @ 2021-08-20 12:05 Christian König 2021-08-20 12:05 ` [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 Christian König 0 siblings, 1 reply; 8+ messages in thread From: Christian König @ 2021-08-20 12:05 UTC (permalink / raw) To: akpm; +Cc: dri-devel, daniel Hi Andrew, Daniel suggested that I ping you once more about this. Basically we want to add a barrier function to make sure that our TTM pool shrinker is not freeing up pages from a device while the device is being unplugged. Currently we are having a global mutex to serialize all of this, but this caused contention for unmapping the freed pages in the IOMMU. We just need your Acked-by and I hope my explanation is now more understandable than the last time. Cheers, Christian. ^ permalink raw reply [flat|nested] 8+ messages in thread
* [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 2021-08-20 12:05 Adding sync_shrinkers function for TTM pool optimization Christian König @ 2021-08-20 12:05 ` Christian König 0 siblings, 0 replies; 8+ messages in thread From: Christian König @ 2021-08-20 12:05 UTC (permalink / raw) To: akpm; +Cc: dri-devel, daniel Switch back to using a spinlock again by moving the IOMMU unmap outside of the locked region. This avoids contention especially while freeing pages. v2: Add a comment explaining why we need sync_shrinkers(). Signed-off-by: Christian König <christian.koenig@amd.com> Acked-by: Huang Rui <ray.huang@amd.com> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> --- drivers/gpu/drm/ttm/ttm_pool.c | 40 +++++++++++++++++++--------------- 1 file changed, 22 insertions(+), 18 deletions(-) diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c index cb38b1a17b09..7d4f76d4141d 100644 --- a/drivers/gpu/drm/ttm/ttm_pool.c +++ b/drivers/gpu/drm/ttm/ttm_pool.c @@ -70,7 +70,7 @@ static struct ttm_pool_type global_uncached[MAX_ORDER]; static struct ttm_pool_type global_dma32_write_combined[MAX_ORDER]; static struct ttm_pool_type global_dma32_uncached[MAX_ORDER]; -static struct mutex shrinker_lock; +static spinlock_t shrinker_lock; static struct list_head shrinker_list; static struct shrinker mm_shrinker; @@ -263,9 +263,9 @@ static void ttm_pool_type_init(struct ttm_pool_type *pt, struct ttm_pool *pool, spin_lock_init(&pt->lock); INIT_LIST_HEAD(&pt->pages); - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); list_add_tail(&pt->shrinker_list, &shrinker_list); - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); } /* Remove a pool_type from the global shrinker list and free all pages */ @@ -273,9 +273,9 @@ static void ttm_pool_type_fini(struct ttm_pool_type *pt) { struct page *p; - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); list_del(&pt->shrinker_list); - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); while ((p = ttm_pool_type_take(pt))) ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); @@ -313,24 +313,23 @@ static struct ttm_pool_type *ttm_pool_select_type(struct ttm_pool *pool, static unsigned int ttm_pool_shrink(void) { struct ttm_pool_type *pt; - unsigned int num_freed; + unsigned int num_pages; struct page *p; - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); pt = list_first_entry(&shrinker_list, typeof(*pt), shrinker_list); + list_move_tail(&pt->shrinker_list, &shrinker_list); + spin_unlock(&shrinker_lock); p = ttm_pool_type_take(pt); if (p) { ttm_pool_free_page(pt->pool, pt->caching, pt->order, p); - num_freed = 1 << pt->order; + num_pages = 1 << pt->order; } else { - num_freed = 0; + num_pages = 0; } - list_move_tail(&pt->shrinker_list, &shrinker_list); - mutex_unlock(&shrinker_lock); - - return num_freed; + return num_pages; } /* Return the allocation order based for a page */ @@ -530,6 +529,11 @@ void ttm_pool_fini(struct ttm_pool *pool) for (j = 0; j < MAX_ORDER; ++j) ttm_pool_type_fini(&pool->caching[i].orders[j]); } + + /* We removed the pool types from the LRU, but we need to also make sure + * that no shrinker is concurrently freeing pages from the pool. + */ + sync_shrinkers(); } /* As long as pages are available make sure to release at least one */ @@ -604,7 +608,7 @@ static int ttm_pool_debugfs_globals_show(struct seq_file *m, void *data) { ttm_pool_debugfs_header(m); - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); seq_puts(m, "wc\t:"); ttm_pool_debugfs_orders(global_write_combined, m); seq_puts(m, "uc\t:"); @@ -613,7 +617,7 @@ static int ttm_pool_debugfs_globals_show(struct seq_file *m, void *data) ttm_pool_debugfs_orders(global_dma32_write_combined, m); seq_puts(m, "uc 32\t:"); ttm_pool_debugfs_orders(global_dma32_uncached, m); - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); ttm_pool_debugfs_footer(m); @@ -640,7 +644,7 @@ int ttm_pool_debugfs(struct ttm_pool *pool, struct seq_file *m) ttm_pool_debugfs_header(m); - mutex_lock(&shrinker_lock); + spin_lock(&shrinker_lock); for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i) { seq_puts(m, "DMA "); switch (i) { @@ -656,7 +660,7 @@ int ttm_pool_debugfs(struct ttm_pool *pool, struct seq_file *m) } ttm_pool_debugfs_orders(pool->caching[i].orders, m); } - mutex_unlock(&shrinker_lock); + spin_unlock(&shrinker_lock); ttm_pool_debugfs_footer(m); return 0; @@ -693,7 +697,7 @@ int ttm_pool_mgr_init(unsigned long num_pages) if (!page_pool_size) page_pool_size = num_pages; - mutex_init(&shrinker_lock); + spin_lock_init(&shrinker_lock); INIT_LIST_HEAD(&shrinker_list); for (i = 0; i < MAX_ORDER; ++i) { -- 2.25.1 ^ permalink raw reply related [flat|nested] 8+ messages in thread
end of thread, other threads:[~2021-08-20 12:05 UTC | newest] Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2021-04-15 11:56 [PATCH 1/2] mm/vmscan: add sync_shrinkers function Christian König 2021-04-15 11:56 ` [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 Christian König 2021-04-15 14:33 ` Huang Rui 2021-04-15 20:33 ` Andrew Morton 2021-04-16 7:08 ` Christian König 2021-04-26 11:15 ` Christian König 2021-04-15 13:23 ` [PATCH 1/2] mm/vmscan: add sync_shrinkers function Daniel Vetter 2021-08-20 12:05 Adding sync_shrinkers function for TTM pool optimization Christian König 2021-08-20 12:05 ` [PATCH 2/2] drm/ttm: optimize the pool shrinker a bit v2 Christian König
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).