On 25/05/16 01:07, Emilio G. Cota wrote:
> On Mon, May 23, 2016 at 23:28:27 +0300, Sergey Fedorov wrote:
>> What if we turn qht::lock into a mutex and change the function as follows:
>>
>> static inline
>> struct qht_bucket *qht_bucket_lock__no_stale(struct qht *ht,
>> uint32_t hash,
>> struct qht_map
>> **pmap)
>> {
>> struct qht_bucket *b;
>> struct qht_map *map;
>>
>> map = atomic_rcu_read(&ht->map);
>> b = qht_map_to_bucket(map, hash);
>>
>> qemu_spin_lock(&b->lock);
>> /* 'ht->map' access is serialized by 'b->lock' here */
>> if (likely(map == ht->map)) {
>> /* no resize in progress; we're done */
>> *pmap = map;
>> return b;
>> }
>> qemu_spin_unlock(&b->lock);
>>
>> /* resize in progress; retry grabbing 'ht->lock' */
>> qemu_mutex_lock(&ht->lock);
>> b = qht_map_to_bucket(ht->map, hash);
>> *pmap = ht->map;
>> qemu_spin_lock(&b->lock);
>> qemu_mutex_unlock(&ht->lock);
>>
>> return b;
>> }
>>
>>
>> With this implementation we could:
>> (1) get rid of qht_map::stale
>> (2) don't waste cycles waiting for resize to complete
> I'll include this in v6.
How is it by perf?
Regards,
Sergey
