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: <Z0ZvAB0vcL-D-a6f@x1>
Date: Tue, 26 Nov 2024 21:59:44 -0300
From: Arnaldo Carvalho de Melo <acme@...nel.org>
To: Namhyung Kim <namhyung@...nel.org>
Cc: Linus Torvalds <torvalds@...ux-foundation.org>,
	Ian Rogers <irogers@...gle.com>,
	Kan Liang <kan.liang@...ux.intel.com>,
	Peter Zijlstra <peterz@...radead.org>,
	Ingo Molnar <mingo@...nel.org>, LKML <linux-kernel@...r.kernel.org>,
	linux-perf-users@...r.kernel.org
Subject: Re: [GIT PULL] perf-tools changes for v6.13

On Tue, Nov 26, 2024 at 04:00:12PM -0800, Namhyung Kim wrote:
> Hello,
> 
> On Tue, Nov 26, 2024 at 03:33:10PM -0800, Linus Torvalds wrote:
> > On Sun, 24 Nov 2024 at 23:17, Namhyung Kim <namhyung@...nel.org> wrote:
> > >
> > > Please consider pulling the following changes in perf tools for v6.13.
> > 
> > Hmm. Has the default search path for the vmlinux image changed?
> 
> I don't think so.
 
> > Doing profiling with this, I get
> > 
> >     openat(AT_FDCWD, "vmlinux", O_RDONLY)   = 5
> > 
> > which is very very wrong for my kernel build - it basically picks up
> > the vmlinux image from the build directory, which has absolutely
> > *nothign* to do with the image that is being run at the moment.
> 
> Sure, at least it should match the build-id before use.

I tried it here and at first I reproduced Linus results, which I found
very strange, then I tried to bisect it, and all the previous versions
were producing this:

root@...ber:/home/acme/git/perf-tools# perf -v ; perf report -vv |& grep -A5 "^Looking at the vmlinux_path"
perf version 6.12.gb50ecc5aca4d
Looking at the vmlinux_path (8 entries long)
symsrc__init: build id mismatch for vmlinux.
symsrc__init: cannot get elf header.
overlapping maps:
 ffffffffc034b75c-ffffffffc034b89b 0 bpf_prog_40ddf486530245f5_sd_devices
 ffffffffc0000000-ffffffffff000000 7fffc0004000 vmlinux
root@...ber:/home/acme/git/perf-tools#

Which is what I got from the perf-tools branch you asked him to pull.

The algorithm in perf is similar to the one in pahole (that stole it
from perf):

root@...ber:/home/acme/git/perf-tools# strace -e openat pahole --running_kernel_vmlinux |& tail -12
openat(AT_FDCWD, "/sys/kernel/notes", O_RDONLY) = 3
openat(AT_FDCWD, "vmlinux", O_RDONLY)   = 3
openat(AT_FDCWD, "/boot/vmlinux", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/boot/vmlinux-6.11.8-200.fc40.x86_64", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib/debug/boot/vmlinux-6.11.8-200.fc40.x86_64", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/lib/modules/6.11.8-200.fc40.x86_64/build/vmlinux", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib/debug/lib/modules/6.11.8-200.fc40.x86_64/vmlinux", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib/debug/boot/vmlinux-6.11.8-200.fc40.x86_64.debug", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/root/.cache/debuginfod_client/55152ddca5da77ca62deb3dd0db105e82b3711e0/debuginfo", O_RDONLY) = -1 ENOENT (No such file or directory)
pahole: couldn't find a vmlinux that matches the running kernel
HINT: Maybe you're inside a container or missing a debuginfo package?
+++ exited with 1 +++
root@...ber:/home/acme/git/perf-tools#

The one in perf does very much the same:

root@...ber:/home/acme/git/perf-tools# perf -v ; strace -e openat -o /tmp/strace.output perf report -vv |& grep -A5 "^Looking at the vmlinux_path" ; grep '"vmlinux"' -B1 -A10 /tmp/strace.output 
perf version 6.12.gb50ecc5aca4d
Looking at the vmlinux_path (8 entries long)
symsrc__init: build id mismatch for vmlinux.
symsrc__init: cannot get elf header.
overlapping maps:
 ffffffffc034b75c-ffffffffc034b89b 0 bpf_prog_40ddf486530245f5_sd_devices
 ffffffffc0000000-ffffffffff000000 7fffc0004000 vmlinux
openat(AT_FDCWD, "/root/.debug/.build-id/55/152ddca5da77ca62deb3dd0db105e82b3711e0/elf", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "vmlinux", O_RDONLY)   = 142
openat(AT_FDCWD, "/boot/vmlinux", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/boot/vmlinux-6.11.8-200.fc40.x86_64", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib/debug/boot/vmlinux-6.11.8-200.fc40.x86_64", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/lib/modules/6.11.8-200.fc40.x86_64/build/vmlinux", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib/debug/lib/modules/6.11.8-200.fc40.x86_64/vmlinux", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/usr/lib/debug/boot/vmlinux-6.11.8-200.fc40.x86_64.debug", O_RDONLY) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/root/.debug/.build-id/55/152ddca5da77ca62deb3dd0db105e82b3711e0/kallsyms", O_RDONLY) = 142
openat(AT_FDCWD, "/sys/kernel/notes", O_RDONLY) = 142
openat(AT_FDCWD, "/proc/kcore", O_RDONLY) = 142
openat(AT_FDCWD, "/proc/kallsyms", O_RDONLY) = 142
root@...ber:/home/acme/git/perf-tools# 
root@...ber:/home/acme/git/perf-tools# perf buildid-list -h -k

 Usage: perf buildid-list [<options>]

    -k, --kernel          Show current kernel build id

root@...ber:/home/acme/git/perf-tools# perf buildid-list -k
55152ddca5da77ca62deb3dd0db105e82b3711e0
root@...ber:/home/acme/git/perf-tools# file vmlinux 
vmlinux: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), statically linked, BuildID[sha1]=e86cad482822bd4b4caabb01e7cd00161fd51a38, with debug_info, not stripped
root@...ber:/home/acme/git/perf-tools#

And it ends up using /proc/kallsyms

Today I did some bisects, etc, but even with at first reproducing what
Linus described, i.e. it used the non-matching vmlinux in the current
directory to resolve symbols, I couldn't reproduce it after trying to
bisect it :-\
 
> > So now I need to point "perf report" at the actual image, which I
> > didn't need to do before.
> 
> Ok, I'll try to reproduce it and take a look.

Hope you root cause this, late in this part of the world, going AFK now.

- Arnaldo

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ