linux-kernel - Re: [PATCH 1/4] cputime,cpuacct: Include guest time in user time in cpuacct.stat

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]

Message-ID: <dee3d594-65bb-ffa2-f009-7900a4cf3f89@yandex-team.com>
Date:   Fri, 20 Aug 2021 12:37:28 +0300
From:   Andrey Ryabinin <arbn@...dex-team.com>
To:     Daniel Jordan <daniel.m.jordan@...cle.com>,
        Ingo Molnar <mingo@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>,
        Juri Lelli <juri.lelli@...hat.com>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Tejun Heo <tj@...nel.org>, Zefan Li <lizefan.x@...edance.com>,
        Johannes Weiner <hannes@...xchg.org>,
        Dietmar Eggemann <dietmar.eggemann@....com>,
        Steven Rostedt <rostedt@...dmis.org>,
        Ben Segall <bsegall@...gle.com>, Mel Gorman <mgorman@...e.de>,
        Daniel Bristot de Oliveira <bristot@...hat.com>
Cc:     Boris Burkov <boris@....io>,
        Bharata B Rao <bharata@...ux.vnet.ibm.com>,
        cgroups@...r.kernel.org, linux-kernel@...r.kernel.org,
        stable@...r.kernel.org
Subject: Re: [PATCH 1/4] cputime,cpuacct: Include guest time in user time in
 cpuacct.stat


Sorry for abandoning this, got distracted by lots of other stuff.


On 3/18/21 1:09 AM, Daniel Jordan wrote:
> Andrey Ryabinin <arbn@...dex-team.com> writes:
> 
>> cpuacct.stat in no-root cgroups shows user time without guest time
>> included int it. This doesn't match with user time shown in root
>> cpuacct.stat and /proc/<pid>/stat.
> 
> Yeah, that's inconsistent.
> 
>> Make account_guest_time() to add user time to cgroup's cpustat to
>> fix this.
> 
> Yep.
> 
> cgroup2's cpu.stat is broken the same way for child cgroups, and this
> happily fixes it.  Probably deserves a mention in the changelog.
> 

Sure.

> The problem with cgroup2 was, if the workload was mostly guest time,
> cpu.stat's user and system together reflected it, but it was split
> unevenly across the two.  I think guest time wasn't actually included in
> either bucket, it was just that the little user and system time there
> was got scaled up in cgroup_base_stat_cputime_show -> cputime_adjust to
> match sum_exec_runtime, which did have it.
> 
> The stats look ok now for both cgroup1 and 2.  Just slightly unsure
> whether we want to change the way both interfaces expose the accounting
> in case something out there depends on it.  Seems like we should, but
> it'd be good to hear more opinions.
> 
>> @@ -148,11 +146,11 @@ void account_guest_time(struct task_struct *p, u64 cputime)
>>  
>>  	/* Add guest time to cpustat. */
>>  	if (task_nice(p) > 0) {
>> -		cpustat[CPUTIME_NICE] += cputime;
>> -		cpustat[CPUTIME_GUEST_NICE] += cputime;
>> +		task_group_account_field(p, CPUTIME_NICE, cputime);
>> +		task_group_account_field(p, CPUTIME_GUEST_NICE, cputime);
>>  	} else {
>> -		cpustat[CPUTIME_USER] += cputime;
>> -		cpustat[CPUTIME_GUEST] += cputime;
>> +		task_group_account_field(p, CPUTIME_USER, cputime);
>> +		task_group_account_field(p, CPUTIME_GUEST, cputime);
>>  	}
> 
> Makes sense for _USER and _NICE, but it doesn't seem cgroup1 or 2
> actually use _GUEST and _GUEST_NICE.
> 
> Could go either way.  Consistency is nice, but I probably wouldn't
> change the GUEST ones so people aren't confused about why they're
> accounted.  It's also extra cycles for nothing, even though most of the
> data is probably in the cache.
> 

Agreed, will live the _GUEST* as is.