| Message ID | 20240210-zswap-global-lru-v3-2-200495333595@bytedance.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | mm/zswap: optimize for dynamic zswap_pools | expand |
On Fri, Feb 16, 2024 at 12:55 AM Chengming Zhou <zhouchengming@bytedance.com> wrote: > > All zswap entries will take a reference of zswap_pool when > zswap_store(), and drop it when free. Change it to use the > percpu_ref is better for scalability performance. > > Although percpu_ref use a bit more memory which should be ok > for our use case, since we almost have only one zswap_pool to > be using. The performance gain is for zswap_store/load hotpath. > > Testing kernel build (32 threads) in tmpfs with memory.max=2GB. > (zswap shrinker and writeback enabled with one 50GB swapfile, > on a 128 CPUs x86-64 machine, below is the average of 5 runs) > > mm-unstable zswap-global-lru > real 63.20 63.12 > user 1061.75 1062.95 > sys 268.74 264.44 Idea is straightforward + code looks solid to me FWIW, so: Reviewed-by: Nhat Pham <nphamcs@gmail.com> > > Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> > --- > mm/zswap.c | 36 +++++++++++++++++++++++++++--------- > 1 file changed, 27 insertions(+), 9 deletions(-) > > diff --git a/mm/zswap.c b/mm/zswap.c > index d275eb523fc4..961349162997 100644 > --- a/mm/zswap.c > +++ b/mm/zswap.c > @@ -173,7 +173,7 @@ struct crypto_acomp_ctx { > struct zswap_pool { > struct zpool *zpools[ZSWAP_NR_ZPOOLS]; > struct crypto_acomp_ctx __percpu *acomp_ctx; > - struct kref kref; > + struct percpu_ref ref; > struct list_head list; > struct work_struct release_work; > struct hlist_node node; > @@ -305,6 +305,7 @@ static void zswap_update_total_size(void) > /********************************* > * pool functions > **********************************/ > +static void __zswap_pool_empty(struct percpu_ref *ref); > > static struct zswap_pool *zswap_pool_create(char *type, char *compressor) > { > @@ -358,13 +359,18 @@ static struct zswap_pool *zswap_pool_create(char *type, char *compressor) > /* being the current pool takes 1 ref; this func expects the > * caller to always add the new pool as the current pool > */ > - kref_init(&pool->kref); > + ret = percpu_ref_init(&pool->ref, __zswap_pool_empty, > + PERCPU_REF_ALLOW_REINIT, GFP_KERNEL); > + if (ret) > + goto ref_fail; > INIT_LIST_HEAD(&pool->list); > > zswap_pool_debug("created", pool); > > return pool; > > +ref_fail: > + cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node); > error: > if (pool->acomp_ctx) > free_percpu(pool->acomp_ctx); > @@ -437,8 +443,9 @@ static void __zswap_pool_release(struct work_struct *work) > > synchronize_rcu(); > > - /* nobody should have been able to get a kref... */ > - WARN_ON(kref_get_unless_zero(&pool->kref)); > + /* nobody should have been able to get a ref... */ > + WARN_ON(!percpu_ref_is_zero(&pool->ref)); Ah nice - this is actually even clearer :) For some reason I missed it, my apologies. > + percpu_ref_exit(&pool->ref); > > /* pool is now off zswap_pools list and has no references. */ > zswap_pool_destroy(pool); > @@ -446,11 +453,11 @@ static void __zswap_pool_release(struct work_struct *work) > > static struct zswap_pool *zswap_pool_current(void); > > -static void __zswap_pool_empty(struct kref *kref) > +static void __zswap_pool_empty(struct percpu_ref *ref) > { > struct zswap_pool *pool; > > - pool = container_of(kref, typeof(*pool), kref); > + pool = container_of(ref, typeof(*pool), ref); > > spin_lock(&zswap_pools_lock); > > @@ -469,12 +476,12 @@ static int __must_check zswap_pool_get(struct zswap_pool *pool) > if (!pool) > return 0; > > - return kref_get_unless_zero(&pool->kref); > + return percpu_ref_tryget(&pool->ref); > } > > static void zswap_pool_put(struct zswap_pool *pool) > { > - kref_put(&pool->kref, __zswap_pool_empty); > + percpu_ref_put(&pool->ref); > } > > static struct zswap_pool *__zswap_pool_current(void) > @@ -604,6 +611,17 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp, > > if (!pool) > pool = zswap_pool_create(type, compressor); > + else { > + /* > + * Restore the initial ref dropped by percpu_ref_kill() > + * when the pool was decommissioned and switch it again > + * to percpu mode. > + */ > + percpu_ref_resurrect(&pool->ref); > + > + /* Drop the ref from zswap_pool_find_get(). */ > + zswap_pool_put(pool); > + } > > if (pool) > ret = param_set_charp(s, kp); > @@ -642,7 +660,7 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp, > * or the new pool we failed to add > */ > if (put_pool) > - zswap_pool_put(put_pool); > + percpu_ref_kill(&put_pool->ref); > > return ret; > } > > -- > b4 0.10.1
diff --git a/mm/zswap.c b/mm/zswap.c index d275eb523fc4..961349162997 100644 --- a/mm/zswap.c +++ b/mm/zswap.c @@ -173,7 +173,7 @@ struct crypto_acomp_ctx { struct zswap_pool { struct zpool *zpools[ZSWAP_NR_ZPOOLS]; struct crypto_acomp_ctx __percpu *acomp_ctx; - struct kref kref; + struct percpu_ref ref; struct list_head list; struct work_struct release_work; struct hlist_node node; @@ -305,6 +305,7 @@ static void zswap_update_total_size(void) /********************************* * pool functions **********************************/ +static void __zswap_pool_empty(struct percpu_ref *ref); static struct zswap_pool *zswap_pool_create(char *type, char *compressor) { @@ -358,13 +359,18 @@ static struct zswap_pool *zswap_pool_create(char *type, char *compressor) /* being the current pool takes 1 ref; this func expects the * caller to always add the new pool as the current pool */ - kref_init(&pool->kref); + ret = percpu_ref_init(&pool->ref, __zswap_pool_empty, + PERCPU_REF_ALLOW_REINIT, GFP_KERNEL); + if (ret) + goto ref_fail; INIT_LIST_HEAD(&pool->list); zswap_pool_debug("created", pool); return pool; +ref_fail: + cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node); error: if (pool->acomp_ctx) free_percpu(pool->acomp_ctx); @@ -437,8 +443,9 @@ static void __zswap_pool_release(struct work_struct *work) synchronize_rcu(); - /* nobody should have been able to get a kref... */ - WARN_ON(kref_get_unless_zero(&pool->kref)); + /* nobody should have been able to get a ref... */ + WARN_ON(!percpu_ref_is_zero(&pool->ref)); + percpu_ref_exit(&pool->ref); /* pool is now off zswap_pools list and has no references. */ zswap_pool_destroy(pool); @@ -446,11 +453,11 @@ static void __zswap_pool_release(struct work_struct *work) static struct zswap_pool *zswap_pool_current(void); -static void __zswap_pool_empty(struct kref *kref) +static void __zswap_pool_empty(struct percpu_ref *ref) { struct zswap_pool *pool; - pool = container_of(kref, typeof(*pool), kref); + pool = container_of(ref, typeof(*pool), ref); spin_lock(&zswap_pools_lock); @@ -469,12 +476,12 @@ static int __must_check zswap_pool_get(struct zswap_pool *pool) if (!pool) return 0; - return kref_get_unless_zero(&pool->kref); + return percpu_ref_tryget(&pool->ref); } static void zswap_pool_put(struct zswap_pool *pool) { - kref_put(&pool->kref, __zswap_pool_empty); + percpu_ref_put(&pool->ref); } static struct zswap_pool *__zswap_pool_current(void) @@ -604,6 +611,17 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp, if (!pool) pool = zswap_pool_create(type, compressor); + else { + /* + * Restore the initial ref dropped by percpu_ref_kill() + * when the pool was decommissioned and switch it again + * to percpu mode. + */ + percpu_ref_resurrect(&pool->ref); + + /* Drop the ref from zswap_pool_find_get(). */ + zswap_pool_put(pool); + } if (pool) ret = param_set_charp(s, kp); @@ -642,7 +660,7 @@ static int __zswap_param_set(const char *val, const struct kernel_param *kp, * or the new pool we failed to add */ if (put_pool) - zswap_pool_put(put_pool); + percpu_ref_kill(&put_pool->ref); return ret; }
All zswap entries will take a reference of zswap_pool when zswap_store(), and drop it when free. Change it to use the percpu_ref is better for scalability performance. Although percpu_ref use a bit more memory which should be ok for our use case, since we almost have only one zswap_pool to be using. The performance gain is for zswap_store/load hotpath. Testing kernel build (32 threads) in tmpfs with memory.max=2GB. (zswap shrinker and writeback enabled with one 50GB swapfile, on a 128 CPUs x86-64 machine, below is the average of 5 runs) mm-unstable zswap-global-lru real 63.20 63.12 user 1061.75 1062.95 sys 268.74 264.44 Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> --- mm/zswap.c | 36 +++++++++++++++++++++++++++--------- 1 file changed, 27 insertions(+), 9 deletions(-)