linux-kernel - Re: [PATCH] sched/fair: Rearrange select_task_rq

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 24 Apr 2018 15:34:37 -0700
From:   Rohit Jain <rohit.k.jain@...cle.com>
To:     Joel Fernandes <joel.opensrc@...il.com>,
        Peter Zijlstra <peterz@...radead.org>
Cc:     Valentin Schneider <valentin.schneider@....com>,
        Viresh Kumar <viresh.kumar@...aro.org>,
        Ingo Molnar <mingo@...hat.com>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Daniel Lezcano <daniel.lezcano@...aro.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Quentin Perret <quentin.perret@....com>,
        Joel Fernandes <joelaf@...gle.com>
Subject: Re: [PATCH] sched/fair: Rearrange select_task_rq_fair() to optimize
 it



On 04/24/2018 08:47 AM, Joel Fernandes wrote:
> On Tue, Apr 24, 2018 at 8:46 AM, Joel Fernandes <joel.opensrc@...il.com> wrote:
>> On Tue, Apr 24, 2018 at 5:35 AM, Peter Zijlstra <peterz@...radead.org> wrote:
>>> On Tue, Apr 24, 2018 at 12:19:07PM +0100, Valentin Schneider wrote:
>>>> On 24/04/18 11:43, Peter Zijlstra wrote:
>>>>> On Tue, Apr 24, 2018 at 11:02:26AM +0100, Valentin Schneider wrote:
>>>>>> I'd argue making things easier to read is a non-negligible part as well.
>>>>> Right, so I don't object to either of these (I think); but it would be
>>>>> good to see this in combination with that proposed EAS change.
>>>>>
>>>> True, I would've said the call to find_energy_efficient_cpu() ([1]) could
>>>> simply be added to the if (sd) {} case, but...
>>> I think the proposal was to put it before the for_each_domain() loop
>>> entirely, however...
>>>
>>>>> I think you (valentin) wanted to side-step the entire domain loop in
>>>>> that case or something.
>>>>>
>>>> ...this would change more things. Admittedly I've been sort of out of the loop
>>>> (no pun intended) lately, but this doesn't ring a bell. That might have been
>>>> the other frenchie (Quentin) :)
>>> It does indeed appear I confused the two of you, it was Quentin playing
>>> with that.
>>>
>>> In any case, if there not going to be conflicts here, this all looks
>>> good.
>> Both Viresh's and Valentin's patch looks lovely to me too. I couldn't
>> spot anything wrong with them either. One suggestion I was thinking
>> off is can we add better comments to this code (atleast label fast
>> path vs slow path) ?
>>
>> Also, annotate the conditions for the fast/slow path with
>> likely/unlikely since fast path is the common case? so like:
>>
>> if (unlikely(sd)) {
>>    /* Fast path, common case */
>>    ...
>> } else if (...) {
>>    /* Slow path */
>> }
> Aargh, I messed that up, I meant:
>
> if (unlikely(sd)) {
>     /* Slow path */
>     ...
> } else if (...) {
>     /* Fast path */
> }

Including the "unlikely" suggestion and the original patch, as expected
with a quick hackbench test on a 44 core 2 socket x86 machine causes no
change in performance.

Thanks,
Rohit

<snip>