linux-kernel - Re: [PATCH] perf/rapl: restart perf rapl counter after resume

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] [day] [month] [year] [list]

Message-ID: <d663757d-8395-bab8-cde3-e6b1ecab0cda@linux.intel.com>
Date:   Thu, 20 Jun 2019 10:33:48 -0400
From:   "Liang, Kan" <kan.liang@...ux.intel.com>
To:     Peter Zijlstra <peterz@...radead.org>,
        Zhang Rui <rui.zhang@...el.com>
Cc:     linux-x86 <x86@...nel.org>, LKML <linux-kernel@...r.kernel.org>,
        mingo@...hat.com, acme@...nel.org,
        alexander.shishkin@...ux.intel.com, jolsa@...hat.com,
        namhyung@...nel.org, tglx@...utronix.de,
        "Liang, Kan" <kan.liang@...el.com>
Subject: Re: [PATCH] perf/rapl: restart perf rapl counter after resume



On 6/20/2019 8:50 AM, Peter Zijlstra wrote:
> On Mon, Jun 17, 2019 at 09:41:37PM +0800, Zhang Rui wrote:
> 
>> After S3 suspend/resume, "perf stat -I 1000 -e power/energy-pkg/ -a"
>> reports an insane value for the very first sampling period after resume
>> as shown below.
>>
>>      19.278989977               2.16 Joules power/energy-pkg/
>>      20.279373569               1.96 Joules power/energy-pkg/
>>      21.279765481               2.09 Joules power/energy-pkg/
>>      22.280305420               2.10 Joules power/energy-pkg/
>>      25.504782277   4,294,966,686.01 Joules power/energy-pkg/
>>      26.505114993               3.58 Joules power/energy-pkg/
>>      27.505471758               1.66 Joules power/energy-pkg/
>>
>> Fix this by resetting the counter right after resume.
> 
> Cute...
> 
> 
>> +#ifdef CONFIG_PM
>> +
>> +static int perf_rapl_suspend(void)
>> +{
>> +	int i;
>> +
>> +	get_online_cpus();
>> +	for (i = 0; i < rapl_pmus->maxpkg; i++)
>> +		rapl_pmu_update_all(rapl_pmus->pmus[i]);
>> +	put_online_cpus();
>> +	return 0;
>> +}
>> +
>> +static void perf_rapl_resume(void)
>> +{
>> +	int i;
>> +
>> +	get_online_cpus();
>> +	for (i = 0; i < rapl_pmus->maxpkg; i++)
>> +		rapl_pmu_restart_all(rapl_pmus->pmus[i]);
>> +	put_online_cpus();
>> +}
> 
> What's the reason for that get/put_online_cpus() here ?
>

It looks like syscore_* functions are executed with one CPU on-line.
If so, they may not be the right place for the rapl callback.

Rapl is per socket. The driver manipulates the registers on the first 
CPU of each socket. I think we need to update/restart the counters on 
all sockets. That's the reason I add get/put_online_cpus() in the 
original patch.

Besides, I think we also need to call rapl_pmu_restart/update_all() on 
the target cpu.


Thanks,
Kan