diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2023-09-02 00:44:45 +0200 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2023-09-02 00:44:45 +0200 |
commit | e987af4546ac5de50e514182c1d0ca33843fa665 (patch) | |
tree | ed2fa9e146d4cd58ce35130e6dd08a00b2cb7a38 /lib | |
parent | Merge tag 'v6.6-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/... (diff) | |
parent | kernel/fork: group allocation/free of per-cpu counters for mm struct (diff) | |
download | linux-e987af4546ac5de50e514182c1d0ca33843fa665.tar.xz linux-e987af4546ac5de50e514182c1d0ca33843fa665.zip |
Merge tag 'percpu-for-6.6' of git://git.kernel.org/pub/scm/linux/kernel/git/dennis/percpu
Pull percpu updates from Dennis Zhou:
"One bigger change to percpu_counter's api allowing for init and
destroy of multiple counters via percpu_counter_init_many() and
percpu_counter_destroy_many(). This is used to help begin remediating
a performance regression with percpu rss stats.
Additionally, it seems larger core count machines are feeling the
burden of the single threaded allocation of percpu. Mateusz is
thinking about it and I will spend some time on it too.
percpu:
- A couple cleanups by Baoquan He and Bibo Mao. The only behavior
change is to start printing messages if we're under the warn limit
for failed atomic allocations.
percpu_counter:
- Shakeel introduced percpu counters into mm_struct which caused
percpu allocations be on the hot path [1]. Originally I spent some
time trying to improve the percpu allocator, but instead preferred
what Mateusz Guzik proposed grouping at the allocation site,
percpu_counter_init_many(). This allows a single percpu allocation
to be shared by the counters. I like this approach because it
creates a shared lifetime by the allocations. Additionally, I
believe many inits have higher level synchronization requirements,
like percpu_counter does against HOTPLUG_CPU. Therefore we can
group these optimizations together"
Link: https://lore.kernel.org/linux-mm/20221024052841.3291983-1-shakeelb@google.com/ [1]
* tag 'percpu-for-6.6' of git://git.kernel.org/pub/scm/linux/kernel/git/dennis/percpu:
kernel/fork: group allocation/free of per-cpu counters for mm struct
pcpcntr: add group allocation/free
mm/percpu.c: print error message too if atomic alloc failed
mm/percpu.c: optimize the code in pcpu_setup_first_chunk() a little bit
mm/percpu.c: remove redundant check
mm/percpu: Remove some local variables in pcpu_populate_pte
Diffstat (limited to 'lib')
-rw-r--r-- | lib/percpu_counter.c | 62 |
1 files changed, 43 insertions, 19 deletions
diff --git a/lib/percpu_counter.c b/lib/percpu_counter.c index 5004463c4f9f..9073430dc865 100644 --- a/lib/percpu_counter.c +++ b/lib/percpu_counter.c @@ -151,48 +151,72 @@ s64 __percpu_counter_sum(struct percpu_counter *fbc) } EXPORT_SYMBOL(__percpu_counter_sum); -int __percpu_counter_init(struct percpu_counter *fbc, s64 amount, gfp_t gfp, - struct lock_class_key *key) +int __percpu_counter_init_many(struct percpu_counter *fbc, s64 amount, + gfp_t gfp, u32 nr_counters, + struct lock_class_key *key) { unsigned long flags __maybe_unused; - - raw_spin_lock_init(&fbc->lock); - lockdep_set_class(&fbc->lock, key); - fbc->count = amount; - fbc->counters = alloc_percpu_gfp(s32, gfp); - if (!fbc->counters) + size_t counter_size; + s32 __percpu *counters; + u32 i; + + counter_size = ALIGN(sizeof(*counters), __alignof__(*counters)); + counters = __alloc_percpu_gfp(nr_counters * counter_size, + __alignof__(*counters), gfp); + if (!counters) { + fbc[0].counters = NULL; return -ENOMEM; + } - debug_percpu_counter_activate(fbc); + for (i = 0; i < nr_counters; i++) { + raw_spin_lock_init(&fbc[i].lock); + lockdep_set_class(&fbc[i].lock, key); +#ifdef CONFIG_HOTPLUG_CPU + INIT_LIST_HEAD(&fbc[i].list); +#endif + fbc[i].count = amount; + fbc[i].counters = (void *)counters + (i * counter_size); + + debug_percpu_counter_activate(&fbc[i]); + } #ifdef CONFIG_HOTPLUG_CPU - INIT_LIST_HEAD(&fbc->list); spin_lock_irqsave(&percpu_counters_lock, flags); - list_add(&fbc->list, &percpu_counters); + for (i = 0; i < nr_counters; i++) + list_add(&fbc[i].list, &percpu_counters); spin_unlock_irqrestore(&percpu_counters_lock, flags); #endif return 0; } -EXPORT_SYMBOL(__percpu_counter_init); +EXPORT_SYMBOL(__percpu_counter_init_many); -void percpu_counter_destroy(struct percpu_counter *fbc) +void percpu_counter_destroy_many(struct percpu_counter *fbc, u32 nr_counters) { unsigned long flags __maybe_unused; + u32 i; + + if (WARN_ON_ONCE(!fbc)) + return; - if (!fbc->counters) + if (!fbc[0].counters) return; - debug_percpu_counter_deactivate(fbc); + for (i = 0; i < nr_counters; i++) + debug_percpu_counter_deactivate(&fbc[i]); #ifdef CONFIG_HOTPLUG_CPU spin_lock_irqsave(&percpu_counters_lock, flags); - list_del(&fbc->list); + for (i = 0; i < nr_counters; i++) + list_del(&fbc[i].list); spin_unlock_irqrestore(&percpu_counters_lock, flags); #endif - free_percpu(fbc->counters); - fbc->counters = NULL; + + free_percpu(fbc[0].counters); + + for (i = 0; i < nr_counters; i++) + fbc[i].counters = NULL; } -EXPORT_SYMBOL(percpu_counter_destroy); +EXPORT_SYMBOL(percpu_counter_destroy_many); int percpu_counter_batch __read_mostly = 32; EXPORT_SYMBOL(percpu_counter_batch); |