linux-kernel - Re: [RESEND PATCH 2/2] perf/x86: improve the event scheduling to avoid unnecessary pmu

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: <YiIyrFn7upPEouVt@hirez.programming.kicks-ass.net>
Date:   Fri, 4 Mar 2022 16:39:24 +0100
From:   Peter Zijlstra <peterz@...radead.org>
To:     Wen Yang <simon.wy@...baba-inc.com>
Cc:     Ingo Molnar <mingo@...hat.com>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Stephane Eranian <eranian@...gle.com>,
        mark rutland <mark.rutland@....com>,
        jiri olsa <jolsa@...hat.com>,
        namhyung kim <namhyung@...nel.org>,
        borislav petkov <bp@...en8.de>, x86@...nel.org,
        Wen Yang <wenyang@...ux.alibaba.com>,
        "h. peter anvin" <hpa@...or.com>, linux-perf-users@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [RESEND PATCH 2/2] perf/x86: improve the event scheduling to
 avoid unnecessary pmu_stop/start

On Fri, Mar 04, 2022 at 07:03:51PM +0800, Wen Yang wrote:
> this issue has been there for a long time, we could reproduce it as follows:

What issue? You've not described an issue. So you cannot reference one.

This is still completely unreadable gibberish.

> 1, run a script that periodically collects perf data, eg:
> while true
> do
>     perf stat -e cache-misses,cache-misses,cache-misses -c 1 sleep 2
>     perf stat -e cache-misses -c 1 sleep 2
>     sleep 1
> done
> 
> 2, run another one to capture the ipc, eg:
> perf stat -e cycles:d,instructions:d  -c 1 -i 1000

<snip line noise>

> the reason is that the nmi watchdog permanently consumes one fp
> (*cycles*). therefore, when the above shell script obtains *cycles*
> again, only one gp can be used, and its weight is 5.
> but other events (like *cache-misses*) have a weight of 4,
> so the counter used by *cycles* will often be taken away, as in
> the raw data above:
> [1]
>   n_events = 3
>   assign = {33, 1, 32, ...}
> -->
>   n_events = 6
>   assign = {33, 3, 32, 0, 1, 2, ...}

Again unreadable... what do any of those numbers mean?

> 
> so it will cause unnecessary pmu_stop/start and also cause abnormal cpi.

How?!?

> Cloud servers usually continuously monitor the cpi data of some important
> services. This issue affects performance and misleads monitoring.
> 
> The current event scheduling algorithm is more than 10 years old:
> commit 1da53e023029 ("perf_events, x86: Improve x86 event scheduling")

irrelevant

> we wish it could be optimized a bit.

I wish for a unicorn ...

> The fields msk_counters and msk_events are added to indicate currently
> used counters and events so that the used ones can be skipped
> in __perf_sched_find_counter and perf_sched_next_event functions to avoid
> unnecessary pmu_stop/start.

Still not sure what your actual problem is, nor what the actual proposal
is.

Why should I attempt to reverse engineer your code without basic
understanding of what you're actually trying to achieve?