linux-kernel - Re: [PATCH v4 1/5] perf stat: Add AGGR_PCIE

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 for Android: free password hash cracker in your pocket

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

Date:   Mon, 8 Feb 2021 14:30:21 +0300
From:   Alexander Antonov <alexander.antonov@...ux.intel.com>
To:     Namhyung Kim <namhyung@...nel.org>
Cc:     Arnaldo Carvalho de Melo <acme@...nel.org>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        Jiri Olsa <jolsa@...hat.com>, Andi Kleen <ak@...ux.intel.com>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Mark Rutland <mark.rutland@....com>,
        Ian Rogers <irogers@...gle.com>,
        Ingo Molnar <mingo@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>
Subject: Re: [PATCH v4 1/5] perf stat: Add AGGR_PCIE_PORT mode

On 2/4/2021 3:07 PM, Namhyung Kim wrote:
> Hello,
>
> On Wed, Feb 3, 2021 at 10:58 PM Alexander Antonov
> <alexander.antonov@...ux.intel.com> wrote:
>> Adding AGGR_PCIE_PORT mode to be able to distinguish aggr_mode
>> for root ports in following patches.
> I'm not sure adding the AGGR_PCIE_PORT is the right way.
> In my understanding, the aggr mode is to specify how we aggregate
> counter values of a single event from different cpus.  But this seems
> to aggregate counter values from different events.  Also the new
> mode is basically the same as AGGR_GLOBAL.
>
> As you will add stat_config.iostat_run to distinguish the iostat
> command, probably we just want to use the global aggr mode
> (and it's the default!) and get rid of the AGGR_PCIE_PORT.
>
> Thoughts?
>
> Thanks,
> Namhyung
Hello Namhyung,

Actually, you are right. We aggregate counter values from different 
events of a
single IIO stack (PCIe root port) to calculate metrics for this IO stack.
But the reason is to prevent using of '-e' and '-M' options in 'iostat' mode
because it can be a reason for the mess in the output that can confuse 
users.

There is an idea to use your suggestion for this part:

status = iostat_prepare(...);
if (status < 0)
         goto out;
if (status == IOSTAT_LIST)
         iostat_list(...);
else
         ...

So, we can check if evlist is empty inside iostat_prepare(). If not, print
a warning, for example, "The -e and -M options are not supported. All chosen
events/metrics will be dropped". Then we can free of evlist by using
evlist__delete(), create new one by using evlist__new() and fill the evlist.

In this case the body of iostat_prepare() function would be:

iostat_prepare()
{
     If (!is_evlist_empty) {
         pr_warning();
         evlist__delete();
         evlist__new()
     }

     iostat_event_group();
}

It will allow to get rid of the AGGR_PCIE_PORT.
What do you think?

Thank you,
Alexander