linux-kernel - Re: [PATCH -tip] x86/locking/atomic: Use asm

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: <gh8qpil9d3.fsf@gouders.net>
Date: Thu, 06 Mar 2025 11:52:08 +0100
From: Dirk Gouders <dirk@...ders.net>
To: Ingo Molnar <mingo@...nel.org>
Cc: Uros Bizjak <ubizjak@...il.com>, Borislav Petkov <bp@...en8.de>,
        Dave
 Hansen <dave.hansen@...el.com>, x86@...nel.org,
        linux-kernel@...r.kernel.org, Peter Zijlstra <peterz@...radead.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Dave Hansen
 <dave.hansen@...ux.intel.com>,
        "H. Peter Anvin" <hpa@...or.com>,
        Linus
 Torvalds <torvalds@...uxfoundation.org>
Subject: Re: [PATCH -tip] x86/locking/atomic: Use asm_inline for atomic
 locking insns

Hi Ingo,

my interest comes, because I just started to try to better understand
PCL and am reading the perf manual pages.  Perhaps I should therefore
keep my RO-bit permanent for some more months, but:

> And if the benchmark is context-switching heavy, you'll want to use 
> 'perf stat -a' option to not have PMU context switching costs, and the 

I'm sure you know what you are talking about so I don't doubt the above
is correct but perhaps, the manual page should also clarify -a:

-a::
--all-cpus::
        system-wide collection from all CPUs (default if no target is specified)

In the last example -a is combined with -C 2 which is even more irritating when
you just started with the manual pages.


But the main reason why I thought it might be OK to once toggle my
RO-bit is that I tried your examples and with the first one I have way
higher numbers than yours and I thought that must be, because you just
own the faster machine (as I would have expected):

>  starship:~> perf bench sched pipe
>  # Running 'sched/pipe' benchmark:
>  # Executed 1000000 pipe operations between two processes
>
>      Total time: 6.939 [sec]
>
>        6.939128 usecs/op
>          144110 ops/sec

lena:~> perf bench sched pipe
# Running 'sched/pipe' benchmark:
# Executed 1000000 pipe operations between two processes

     Total time: 11.129 [sec]

      11.129952 usecs/op
          89847 ops/sec

And I expected this to continue throughout the examples.

But -- to keep this short -- with the last example, my numbers are
suddenly significantly lower than yours:

>  starship:~> taskset 0x4 perf stat -a -C 2 -e instructions --repeat 5 perf bench sched pipe
>        5.808068 usecs/op
>        5.843716 usecs/op
>        5.826543 usecs/op
>        5.801616 usecs/op
>        5.793129 usecs/op
>
>  Performance counter stats for 'system wide' (5 runs):
>
>     32,244,691,275      instructions                                                            ( +-  0.21% )
>
>            5.81624 +- 0.00912 seconds time elapsed  ( +-  0.16% )

lena:~> taskset 0x4 perf stat -a -C 2 -e instructions --repeat 5 perf bench sched pipe
       4.204444 usecs/op
       4.169279 usecs/op
       4.186812 usecs/op
       4.217039 usecs/op
       4.208538 usecs/op

 Performance counter stats for 'system wide' (5 runs):

    14,196,762,588      instructions                                                            ( +-  0.04% )

           4.20203 +- 0.00854 seconds time elapsed  ( +-  0.20% )



Of course, I don't want to waste anyones time if this is a so obvious
thing that only newbies don't understand.  So, feel free to just ignore this.

Regards

Dirk