1. Forum
  2. Usenet
  3. LINUX.DEBIAN.KERNEL

  • Bug#1088153: linux-perf: "perf script" is extremely slow

    From Salvatore Bonaccorso@1:229/2 to Ralf Jung on Tue Jun 10 22:20:01 2025
    XPost: linux.debian.bugs.dist
    From: carnil@debian.org

    Contol: tags -1 + moreinfo unreproducible

    Hi Ralf,

    On Sun, Nov 24, 2024 at 10:45:50AM +0100, Ralf Jung wrote:
    Package: linux-perf
    Version: 6.11.9-1
    Severity: normal

    Dear Maintainer,

    I am trying to use cargo-flamegraph to produce a flamegraph from perf data. That
    invokes "perf script" and that has now been running already 15 minutes on a tiny
    perf file containing 3300 samples. Something is extremely slow here, making some
    kinds of perf analysis basically impossible.

    Looking around for perf performance issues, I found https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=911815, which however is supposed to have been fixed years ago. But I still see thousands and thousands
    of addr2line processes being spawned, so somehow <https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=be8ecc57f180415e8a7c1cc5620c5236be2a7e56>
    either did not have the desired effect or is not present in current kernels any
    more?

    When I finally killed "perf script", I saw a lot of this kind of output:

    addr2line /home/r/.debug/.build-id/37/8f73d84f9d89f55dfc18a9ab84bce321cd9b09/elf: could not read first record

    I don't know whether that could cause the slowdown, or what could cause that error.

    Is this still something you can reproduce?

    Regards,
    Salvatore

    --- SoupGate-Win32 v1.05
    * Origin: you cannot sedate... all the things you hate (1:229/2)
  • From Ralf Jung@1:229/2 to All on Wed Jun 11 17:00:01 2025
    XPost: linux.debian.bugs.dist
    From: post@ralfj.de

    Hi,

    On Sun, Nov 24, 2024 at 10:45:50AM +0100, Ralf Jung wrote:
    Package: linux-perf
    Version: 6.11.9-1
    Severity: normal

    Dear Maintainer,

    I am trying to use cargo-flamegraph to produce a flamegraph from perf data. That
    invokes "perf script" and that has now been running already 15 minutes on a tiny
    perf file containing 3300 samples. Something is extremely slow here, making some
    kinds of perf analysis basically impossible.

    Looking around for perf performance issues, I found
    https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=911815, which however is >> supposed to have been fixed years ago. But I still see thousands and thousands
    of addr2line processes being spawned, so somehow
    <https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=be8ecc57f180415e8a7c1cc5620c5236be2a7e56>
    either did not have the desired effect or is not present in current kernels any
    more?

    When I finally killed "perf script", I saw a lot of this kind of output:

    addr2line /home/r/.debug/.build-id/37/8f73d84f9d89f55dfc18a9ab84bce321cd9b09/elf: could not read first record

    I don't know whether that could cause the slowdown, or what could cause that error.

    Is this still something you can reproduce?

    On a quick check, it seems like "perf script" now finishes in a few seconds now.

    Kind regards,
    Ralf

    --- SoupGate-Win32 v1.05
    * Origin: you cannot sedate... all the things you hate (1:229/2)