[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-rt-users
Subject:    Re: [RFC v1] sched/completion: convert completions to use simple wait queues
From:       Daniel Wagner <wagi () monom ! org>
Date:       2016-03-30 15:29:45
Message-ID: 56FBF0E9.40203 () monom ! org
[Download RAW message or body]

On 03/30/2016 05:21 PM, Peter Zijlstra wrote:
> On Wed, Mar 30, 2016 at 05:17:29PM +0200, Sebastian Andrzej Siewior wrote:
>> On 03/30/2016 05:07 PM, Peter Zijlstra wrote:
>>> On Wed, Mar 30, 2016 at 04:53:05PM +0200, Daniel Wagner wrote:
>>>> From: Daniel Wagner <daniel.wagner@bmw-carit.de>
>>>>
>>>> Completions have no long lasting callbacks and therefore do not need
>>>> the complex waitqueue variant.  Use simple waitqueues which reduces
>>>> the contention on the waitqueue lock.
>>>
>>> Changelog really should have talk about the determinism thing. The last
>>> time you posted this the point was raised that we should wake the
>>> highest prio waiter in the defer case, you did not address this.
>>
>> So we really want to go this road? 
> 
> Dunno, but at least mention why it wouldn't matter.

It seems I put to much effort into the cover letter. I should have spent
that time in the changelog. Anyway, I am going through the users of
complete_all() and it looks like most of them are either some setup code
paths and the other bunch of calls are just making sure the single
waiter really wakes up.

>> I didn't find any numbers what the
>> highest count of queued sleepers was in Daniel's complete_all() testing.
>>
>> As for the latest -RT I received only one report from Clark Williams
>> with something like 3 to 9 sleepers waked up during one complete_all()
>> and this happens in the resume code.
>> Based on this, deferring wake-ups from IRQ-context and a RB-tree (or
>> something like that for priority sorting) looks like a lot of complexity
>> and it does not look like we gain much.
> 
> Sure, but that equally puts the whole defer thing into question, if we
> can put a hard cap on the max number (and WARN when exceeded) we're also
> good.
> 
>>> Also, you make no mention of the reduction of UINT_MAX to USHORT_MAX and
>>> the implications of that.
>>
>> Wasn't this
>> |To avoid a size increase of struct completion, I spitted the done
>> |field into two half.
>>
>> later he mentions that we can't have 2M sleepers anymore.
> 
> That wasn't in this changelog, therefore it wasn't read ;-)

Got it, next version has all info in the changelog and not in the cover
letter.
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
[prev in list] [next in list] [prev in thread] [next in thread]