On 24/11/2020 14:22, John Garry wrote:
> On 22/11/2020 15:35, Pavel Begunkov wrote:
>> map->swap_lock protects map->cleared from concurrent modification,
>> however sbitmap_deferred_clear() is already atomically drains it, so
>> it's guaranteed to not loose bits on concurrent
>> sbitmap_deferred_clear().
>>
>> A one threaded tag heavy test on top of nullbk showed ~1.5% t-put
>> increase, and 3% -> 1% cycle reduction of sbitmap_get() according to perf.
>>
>> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
>> ---
>> include/linux/sbitmap.h | 5 -----
>> lib/sbitmap.c | 14 +++-----------
>> 2 files changed, 3 insertions(+), 16 deletions(-)
>>
>> diff --git a/include/linux/sbitmap.h b/include/linux/sbitmap.h
>> index e40d019c3d9d..74cc6384715e 100644
>> --- a/include/linux/sbitmap.h
>> +++ b/include/linux/sbitmap.h
>> @@ -32,11 +32,6 @@ struct sbitmap_word {
>> * @cleared: word holding cleared bits
>> */
>> unsigned long cleared ____cacheline_aligned_in_smp;
>> -
>> - /**
>> - * @swap_lock: Held while swapping word <-> cleared
>> - */
>> - spinlock_t swap_lock;
>> } ____cacheline_aligned_in_smp;
>> /**
>> diff --git a/lib/sbitmap.c b/lib/sbitmap.c
>> index c1c8a4e69325..4fd877048ba8 100644
>> --- a/lib/sbitmap.c
>> +++ b/lib/sbitmap.c
>> @@ -15,13 +15,9 @@
>> static inline bool sbitmap_deferred_clear(struct sbitmap_word *map)
>> {
>> unsigned long mask, val;
>> - bool ret = false;
>> - unsigned long flags;
>> - spin_lock_irqsave(&map->swap_lock, flags);
>> -
>> - if (!map->cleared)
>> - goto out_unlock;
>> + if (!READ_ONCE(map->cleared))
>> + return false;
>
> So if we race with another cpu, won't the 2nd cpu see that the mask is 0 returned from the xchg (not shown)? If so, it's odd to continue to do the CAS - or atomic not, from later patch - on a mask of 0.
Yeah, but this part is legit and I don't expect it to be so
contended to need an additional check, especially with atomic
and from [3/4].
I'm more concerned about sbitmap_resize*() callers to do right
synchronisation (e.g. quiesce) and not rely on that critical
section I remove. Would be great if anyone can confirm that.
>
> Thanks,
> John
>
>> /*
>> * First get a stable cleared mask, setting the old mask to 0.
>> @@ -35,10 +31,7 @@ static inline bool sbitmap_deferred_clear(struct sbitmap_word *map)
>> val = map->word;
>> } while (cmpxchg(&map->word, val, val & ~mask) != val);
>> - ret = true;
>> -out_unlock:
>> - spin_unlock_irqrestore(&map->swap_lock, flags);
>> - return ret;
>> + return true;
>> }
>> int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
>> @@ -80,7 +73,6 @@ int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
>> for (i = 0; i < sb->map_nr; i++) {
>> sb->map[i].depth = min(depth, bits_per_word);
>> depth -= sb->map[i].depth;
>> - spin_lock_init(&sb->map[i].swap_lock);
>> }
>> return 0;
>> }
>>
>
--
Pavel Begunkov
