From ca1d36b823944f24b5755311e95883fb5fdb807b Mon Sep 17 00:00:00 2001 From: Qi Zheng Date: Mon, 11 Sep 2023 17:44:41 +0800 Subject: mm: shrinker: make global slab shrink lockless The shrinker_rwsem is a global read-write lock in shrinkers subsystem, which protects most operations such as slab shrink, registration and unregistration of shrinkers, etc. This can easily cause problems in the following cases. 1) When the memory pressure is high and there are many filesystems mounted or unmounted at the same time, slab shrink will be affected (down_read_trylock() failed). Such as the real workload mentioned by Kirill Tkhai: ``` One of the real workloads from my experience is start of an overcommitted node containing many starting containers after node crash (or many resuming containers after reboot for kernel update). In these cases memory pressure is huge, and the node goes round in long reclaim. ``` 2) If a shrinker is blocked (such as the case mentioned in [1]) and a writer comes in (such as mount a fs), then this writer will be blocked and cause all subsequent shrinker-related operations to be blocked. Even if there is no competitor when shrinking slab, there may still be a problem. The down_read_trylock() may become a perf hotspot with frequent calls to shrink_slab(). Because of the poor multicore scalability of atomic operations, this can lead to a significant drop in IPC (instructions per cycle). We used to implement the lockless slab shrink with SRCU [2], but then kernel test robot reported -88.8% regression in stress-ng.ramfs.ops_per_sec test case [3], so we reverted it [4]. This commit uses the refcount+RCU method [5] proposed by Dave Chinner to re-implement the lockless global slab shrink. The memcg slab shrink is handled in the subsequent patch. For now, all shrinker instances are converted to dynamically allocated and will be freed by call_rcu(). So we can use rcu_read_{lock,unlock}() to ensure that the shrinker instance is valid. And the shrinker instance will not be run again after unregistration. So the structure that records the pointer of shrinker instance can be safely freed without waiting for the RCU read-side critical section. In this way, while we implement the lockless slab shrink, we don't need to be blocked in unregister_shrinker(). The following are the test results: stress-ng --timeout 60 --times --verify --metrics-brief --ramfs 9 & 1) Before applying this patchset: setting to a 60 second run per stressor dispatching hogs: 9 ramfs stressor bogo ops real time usr time sys time bogo ops/s bogo ops/s (secs) (secs) (secs) (real time) (usr+sys time) ramfs 473062 60.00 8.00 279.13 7884.12 1647.59 for a 60.01s run time: 1440.34s available CPU time 7.99s user time ( 0.55%) 279.13s system time ( 19.38%) 287.12s total time ( 19.93%) load average: 7.12 2.99 1.15 successful run completed in 60.01s (1 min, 0.01 secs) 2) After applying this patchset: setting to a 60 second run per stressor dispatching hogs: 9 ramfs stressor bogo ops real time usr time sys time bogo ops/s bogo ops/s (secs) (secs) (secs) (real time) (usr+sys time) ramfs 477165 60.00 8.13 281.34 7952.55 1648.40 for a 60.01s run time: 1440.33s available CPU time 8.12s user time ( 0.56%) 281.34s system time ( 19.53%) 289.46s total time ( 20.10%) load average: 6.98 3.03 1.19 successful run completed in 60.01s (1 min, 0.01 secs) We can see that the ops/s has hardly changed. [1]. https://lore.kernel.org/lkml/20191129214541.3110-1-ptikhomirov@virtuozzo.com/ [2]. https://lore.kernel.org/lkml/20230313112819.38938-1-zhengqi.arch@bytedance.com/ [3]. https://lore.kernel.org/lkml/202305230837.db2c233f-yujie.liu@intel.com/ [4]. https://lore.kernel.org/all/20230609081518.3039120-1-qi.zheng@linux.dev/ [5]. https://lore.kernel.org/lkml/ZIJhou1d55d4H1s0@dread.disaster.area/ Link: https://lkml.kernel.org/r/20230911094444.68966-43-zhengqi.arch@bytedance.com Signed-off-by: Qi Zheng Cc: Abhinav Kumar Cc: Alasdair Kergon Cc: Alexander Viro Cc: Alyssa Rosenzweig Cc: Andreas Dilger Cc: Andreas Gruenbacher Cc: Anna Schumaker Cc: Arnd Bergmann Cc: Bob Peterson Cc: Borislav Petkov Cc: Carlos Llamas Cc: Chandan Babu R Cc: Chao Yu Cc: Chris Mason Cc: Christian Brauner Cc: Christian Koenig Cc: Chuck Lever Cc: Coly Li Cc: Dai Ngo Cc: Daniel Vetter Cc: Daniel Vetter Cc: "Darrick J. Wong" Cc: Dave Chinner Cc: Dave Hansen Cc: David Airlie Cc: David Hildenbrand Cc: David Sterba Cc: Dmitry Baryshkov Cc: Gao Xiang Cc: Greg Kroah-Hartman Cc: Huang Rui Cc: Ingo Molnar Cc: Jaegeuk Kim Cc: Jani Nikula Cc: Jan Kara Cc: Jason Wang Cc: Jeff Layton Cc: Jeffle Xu Cc: Joel Fernandes (Google) Cc: Joonas Lahtinen Cc: Josef Bacik Cc: Juergen Gross Cc: Kent Overstreet Cc: Kirill Tkhai Cc: Marijn Suijten Cc: "Michael S. Tsirkin" Cc: Mike Snitzer Cc: Minchan Kim Cc: Muchun Song Cc: Muchun Song Cc: Nadav Amit Cc: Neil Brown Cc: Oleksandr Tyshchenko Cc: Olga Kornievskaia Cc: Paul E. McKenney Cc: Richard Weinberger Cc: Rob Clark Cc: Rob Herring Cc: Rodrigo Vivi Cc: Roman Gushchin Cc: Sean Paul Cc: Sergey Senozhatsky Cc: Song Liu Cc: Stefano Stabellini Cc: Steven Price Cc: "Theodore Ts'o" Cc: Thomas Gleixner Cc: Tomeu Vizoso Cc: Tom Talpey Cc: Trond Myklebust Cc: Tvrtko Ursulin Cc: Vlastimil Babka Cc: Xuan Zhuo Cc: Yue Hu Signed-off-by: Andrew Morton --- mm/shrinker.c | 89 +++++++++++++++++++++++++++++++++++++++++++++-------------- 1 file changed, 68 insertions(+), 21 deletions(-) (limited to 'mm/shrinker.c') diff --git a/mm/shrinker.c b/mm/shrinker.c index e2e832b15803..9bb2e61092b3 100644 --- a/mm/shrinker.c +++ b/mm/shrinker.c @@ -2,6 +2,7 @@ #include #include #include +#include #include #include "internal.h" @@ -577,33 +578,50 @@ unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg, if (!mem_cgroup_disabled() && !mem_cgroup_is_root(memcg)) return shrink_slab_memcg(gfp_mask, nid, memcg, priority); - if (!down_read_trylock(&shrinker_rwsem)) - goto out; - - list_for_each_entry(shrinker, &shrinker_list, list) { + /* + * lockless algorithm of global shrink. + * + * In the unregistration setp, the shrinker will be freed asynchronously + * via RCU after its refcount reaches 0. So both rcu_read_lock() and + * shrinker_try_get() can be used to ensure the existence of the shrinker. + * + * So in the global shrink: + * step 1: use rcu_read_lock() to guarantee existence of the shrinker + * and the validity of the shrinker_list walk. + * step 2: use shrinker_try_get() to try get the refcount, if successful, + * then the existence of the shrinker can also be guaranteed, + * so we can release the RCU lock to do do_shrink_slab() that + * may sleep. + * step 3: *MUST* to reacquire the RCU lock before calling shrinker_put(), + * which ensures that neither this shrinker nor the next shrinker + * will be freed in the next traversal operation. + * step 4: do shrinker_put() paired with step 2 to put the refcount, + * if the refcount reaches 0, then wake up the waiter in + * shrinker_free() by calling complete(). + */ + rcu_read_lock(); + list_for_each_entry_rcu(shrinker, &shrinker_list, list) { struct shrink_control sc = { .gfp_mask = gfp_mask, .nid = nid, .memcg = memcg, }; + if (!shrinker_try_get(shrinker)) + continue; + + rcu_read_unlock(); + ret = do_shrink_slab(&sc, shrinker, priority); if (ret == SHRINK_EMPTY) ret = 0; freed += ret; - /* - * Bail out if someone want to register a new shrinker to - * prevent the registration from being stalled for long periods - * by parallel ongoing shrinking. - */ - if (rwsem_is_contended(&shrinker_rwsem)) { - freed = freed ? : 1; - break; - } + + rcu_read_lock(); + shrinker_put(shrinker); } - up_read(&shrinker_rwsem); -out: + rcu_read_unlock(); cond_resched(); return freed; } @@ -676,13 +694,29 @@ void shrinker_register(struct shrinker *shrinker) } down_write(&shrinker_rwsem); - list_add_tail(&shrinker->list, &shrinker_list); + list_add_tail_rcu(&shrinker->list, &shrinker_list); shrinker->flags |= SHRINKER_REGISTERED; shrinker_debugfs_add(shrinker); up_write(&shrinker_rwsem); + + init_completion(&shrinker->done); + /* + * Now the shrinker is fully set up, take the first reference to it to + * indicate that lookup operations are now allowed to use it via + * shrinker_try_get(). + */ + refcount_set(&shrinker->refcount, 1); } EXPORT_SYMBOL_GPL(shrinker_register); +static void shrinker_free_rcu_cb(struct rcu_head *head) +{ + struct shrinker *shrinker = container_of(head, struct shrinker, rcu); + + kfree(shrinker->nr_deferred); + kfree(shrinker); +} + void shrinker_free(struct shrinker *shrinker) { struct dentry *debugfs_entry = NULL; @@ -691,9 +725,25 @@ void shrinker_free(struct shrinker *shrinker) if (!shrinker) return; + if (shrinker->flags & SHRINKER_REGISTERED) { + /* drop the initial refcount */ + shrinker_put(shrinker); + /* + * Wait for all lookups of the shrinker to complete, after that, + * no shrinker is running or will run again, then we can safely + * free it asynchronously via RCU and safely free the structure + * where the shrinker is located, such as super_block etc. + */ + wait_for_completion(&shrinker->done); + } + down_write(&shrinker_rwsem); if (shrinker->flags & SHRINKER_REGISTERED) { - list_del(&shrinker->list); + /* + * Now we can safely remove it from the shrinker_list and then + * free it. + */ + list_del_rcu(&shrinker->list); debugfs_entry = shrinker_debugfs_detach(shrinker, &debugfs_id); shrinker->flags &= ~SHRINKER_REGISTERED; } @@ -707,9 +757,6 @@ void shrinker_free(struct shrinker *shrinker) if (debugfs_entry) shrinker_debugfs_remove(debugfs_entry, debugfs_id); - kfree(shrinker->nr_deferred); - shrinker->nr_deferred = NULL; - - kfree(shrinker); + call_rcu(&shrinker->rcu, shrinker_free_rcu_cb); } EXPORT_SYMBOL_GPL(shrinker_free); -- cgit v1.2.3