Re: [PATCH] fs/epoll: Enable non-blocking busypoll with epoll timeout of 0

[Eric Dumazet <eric.dumazet@gmail.com>]

On 6/24/20 9:32 AM, Samudrala, Sridhar wrote:
> Adding Dave, Eric for review and see if we can get this in via net-next
> as this is mainly useful for networking workloads doing busypoll.
> 
> Thanks
> Sridhar
> 
> On 6/19/2020 11:13 AM, Sridhar Samudrala wrote:
>> This patch triggers non-blocking busy poll when busy_poll is enabled and
>> epoll is called with a timeout of 0 and is associated with a napi_id.
>> This enables an app thread to go through napi poll routine once by calling
>> epoll with a 0 timeout.
>>
>> poll/select with a 0 timeout behave in a similar manner.
>>
>> Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
>> ---
>>   fs/eventpoll.c | 13 +++++++++++++
>>   1 file changed, 13 insertions(+)
>>
>> diff --git a/fs/eventpoll.c b/fs/eventpoll.c
>> index 12eebcdea9c8..5f55078d6381 100644
>> --- a/fs/eventpoll.c
>> +++ b/fs/eventpoll.c
>> @@ -1847,6 +1847,19 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
>>           eavail = ep_events_available(ep);
>>           write_unlock_irq(&ep->lock);
>>   +        /*
>> +         * Trigger non-blocking busy poll if timeout is 0 and there are
>> +         * no events available. Passing timed_out(1) to ep_busy_loop
>> +         * will make sure that busy polling is triggered only once and
>> +         * only if sysctl.net.core.busy_poll is set to non-zero value.
>> +         */
>> +        if (!eavail) {

Maybe avoid all this stuff for the typical case of busy poll being not used ?

            if (!evail && net_busy_loop_on)) {

>> +            ep_busy_loop(ep, timed_out);


>> +            write_lock_irq(&ep->lock);
>> +            eavail = ep_events_available(ep);
>> +            write_unlock_irq(&ep->lock);
>> +        }
>> +
>>           goto send_events;
>>       }
>>