[lkml]   [2017]   [Jul]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [Xen-devel] [PATCH v2 09/13] xen/pvcalls: implement recvmsg
On 07/26/2017 08:08 PM, Stefano Stabellini wrote:
> On Wed, 26 Jul 2017, Boris Ostrovsky wrote:
>>>> + count++;
>>>> + else
>>>> + wait_event_interruptible(map->active.inflight_conn_req,
>>>> + pvcalls_front_read_todo(map));
>>>> + }
>>> Should we be using PVCALLS_FRONT_MAX_SPIN here? In sendmsg it is
>>> counting non-sleeping iterations but here we are sleeping so
>>> PVCALLS_FRONT_MAX_SPIN (5000) may take a while.
>>> In fact, what shouldn't this waiting be a function of MSG_DONTWAIT
>> err, which it already is. But the question still stands (except for
> The code (admittedly unintuitive) is busy-looping (non-sleeping) for
> 5000 iterations *before* attempting to sleep. So in that regard, recvmsg
> and sendmsg use PVCALLS_FRONT_MAX_SPIN in the same way: only for
> non-sleeping iterations.


Why not go directly into wait_event_interruptible()? I see you write in
the commit message

If not enough data is available on the ring, rather than returning
immediately or sleep-waiting, spin for up to 5000 cycles. This small
optimization turns out to improve performance and latency significantly.

Is this because of scheduling latency? I think this should be mentioned not just in the commit message but also as a comment in the code.

(I also think it's not "not enough data" but rather "no data"?)


 \ /
  Last update: 2017-07-27 16:58    [W:0.057 / U:5.236 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site