Message-ID: <CAP-5=fX-Z6cKX30TRNYQLnYHzDSaY-yw56BZwk8db0_R3_mcNQ@mail.gmail.com>
Date: Wed, 5 Nov 2025 09:07:23 -0800
From: Ian Rogers <irogers@...gle.com>
To: Guilherme Amadio <amadio@...too.org>
Cc: Arnaldo Carvalho de Melo <acme@...nel.org>, namhyung@...nel.org, linux-kernel@...r.kernel.org, 
	linux-perf-users@...r.kernel.org
Subject: Re: Crash when running perf top with perf-6.17.7

On Wed, Nov 5, 2025 at 8:36 AM Guilherme Amadio <amadio@...too.org> wrote:
>
> Hi Arnaldo,
>
> After updating to perf-6.17.7, perf top crashes for me (seems to happen
> since perf-6.17, but doesn't happen with perf-6.16):
>
> gentoo perf $ perf top
> perf: Segmentation fault
> -------- backtrace --------
> perf: Segmentation fault
> -------- backtrace --------
> Segmentation fault         (core dumped) perf top
>
> The stack trace from the crash is below. If this is not enough information,
> please let me know what additional information you'd like to have.

This is a BUILD_NONDISTRO=1 build that is default off and
non-distributable as libbfd is incompatible with perf's GPLv2 license.
I think the perf libbfd code should just be deleted so we can focus
energy on distributable code.

Most recently the libbfd code was factored into its own file:
https://lore.kernel.org/lkml/20250929190805.201446-5-irogers@google.com/
we still don't have the dlopen support:
https://lore.kernel.org/lkml/20251007163835.3152881-1-irogers@google.com/
even though shipping with libLLVM/capstone linked (ie not dlopen)
isn't possible in distributions like SuSE:
https://lore.kernel.org/linux-perf-users/aPkivW2WoySU1Zkc@suse.de/
For addr2line the fallback options are fairly clear (LLVM first, then
libbfd then the forked command):
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/srcline.c?h=perf-tools-next#n117
there should probably be a similar clean up for read build id. That
said, the symbol-minimal.c code:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/symbol-minimal.c?h=perf-tools-next#n33
has no library or forked tool dependencies, so perhaps we can just use
this for all our build ID needs.

I'll see if there's a cleanup for read build id similar to the
addr2line cleanup that can be done and try to fix the libbfd
implementation in the process. However, as I said at the top, this
would be less work if we just delete the code. Do people really care
if a build ID, line number of objdump was provided to them by libbfd?
I suspect I've introduced this regression by making build ID mmaps the
default, anyway...

Thanks,
Ian


> Best regards,
> -Guilherme
>
> Core was generated by `perf top'.
> Program terminated with signal SIGSEGV, Segmentation fault.
> #0  close_one () at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c:197
>
> warning: 197    /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c: No such file or directory
> [Current thread is 1 (Thread 0x7f8769c466c0 (LWP 144140))]
> (gdb) bt
> #0  close_one () at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c:197
> #1  0x00007f87885286dd in _bfd_cache_init_unlocked (abfd=abfd@...ry=0x5567f738eb40) at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c:561
> #2  0x00007f8788529021 in bfd_cache_init (abfd=abfd@...ry=0x5567f738eb40) at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c:587
> #3  0x00007f878853618a in bfd_fopen (filename=filename@...ry=0x7f8769c3fa98 "/usr/bin/perf", target=target@...ry=0x0, mode=0x7f878860c4b9 "r", fd=84)
>     at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/opncls.c:291
> #4  0x00007f87885363c9 in bfd_fdopenr (filename=filename@...ry=0x7f8769c3fa98 "/usr/bin/perf", target=target@...ry=0x0, fd=<optimized out>)
>     at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/opncls.c:407
> #5  0x00005567c5cf4436 in read_build_id (filename=filename@...ry=0x7f8769c3fa98 "/usr/bin/perf", bid=bid@...ry=0x7f8769c3d8f0, block=block@...ry=false)
>     at util/symbol-elf.c:886
> #6  0x00005567c5cf7240 in filename__read_build_id (filename=filename@...ry=0x7f8769c3fa98 "/usr/bin/perf", bid=bid@...ry=0x7f8769c3d8f0, block=block@...ry=false)
>     at util/symbol-elf.c:968
> #7  0x00005567c5caaba4 in perf_record_mmap2__read_build_id (event=event@...ry=0x7f8769c3fa50, machine=machine@...ry=0x5567f7392a00, is_kernel=is_kernel@...ry=false)
>     at util/synthetic-events.c:404
> #8  0x00005567c5cabff2 in perf_event__synthesize_mmap_events (tool=tool@...ry=0x0, event=event@...ry=0x7f8769c3fa50, pid=pid@...ry=144076, tgid=tgid@...ry=144076,
>     process=process@...ry=0x5567c5c5d491 <mmap_handler>, machine=machine@...ry=0x5567f7392a00, mmap_data=true) at util/synthetic-events.c:536
> #9  0x00005567c5c598fb in machine__init_live (machine=machine@...ry=0x5567f7392a00, pid=pid@...ry=144076) at util/machine.c:167
> #10 0x00005567c5c5b987 in machine__new_live (host_env=host_env@...ry=0x7f8769c40b70, kernel_maps=kernel_maps@...ry=false, pid=pid@...ry=144076) at util/machine.c:178
> #11 0x00005567c5c56b13 in __dump_stack (file=0x7f87887f25c0 <_IO_2_1_stdout_>, stackdump=stackdump@...ry=0x7f8769c40d90, stackdump_size=15) at util/debug.c:318
> #12 0x00005567c5bfb299 in ui__signal_backtrace (sig=<optimized out>) at ui/tui/setup.c:111
> #13 <signal handler called>
> #14 close_one () at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c:197
> #15 0x00007f87885286dd in _bfd_cache_init_unlocked (abfd=abfd@...ry=0x5567f738e8c0) at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c:561
> #16 0x00007f8788529021 in bfd_cache_init (abfd=abfd@...ry=0x5567f738e8c0) at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/cache.c:587
> #17 0x00007f878853618a in bfd_fopen (filename=filename@...ry=0x5567f73ca048 "/usr/lib64/libfribidi.so.0.4.0", target=target@...ry=0x0, mode=0x7f878860c4b9 "r", fd=63)
>     at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/opncls.c:291
> #18 0x00007f87885363c9 in bfd_fdopenr (filename=filename@...ry=0x5567f73ca048 "/usr/lib64/libfribidi.so.0.4.0", target=target@...ry=0x0, fd=<optimized out>)
>     at /tmp/portage/sys-libs/binutils-libs-2.45-r1/work/binutils-2.45/bfd/opncls.c:407
> #19 0x00005567c5cf4436 in read_build_id (filename=filename@...ry=0x5567f73ca048 "/usr/lib64/libfribidi.so.0.4.0", bid=bid@...ry=0x7f8769c425d0, block=block@...ry=false)
>     at util/symbol-elf.c:886
> #20 0x00005567c5cf7240 in filename__read_build_id (filename=filename@...ry=0x5567f73ca048 "/usr/lib64/libfribidi.so.0.4.0", bid=bid@...ry=0x7f8769c425d0,
>     block=block@...ry=false) at util/symbol-elf.c:968
> #21 0x00005567c5caaba4 in perf_record_mmap2__read_build_id (event=event@...ry=0x5567f73ca000, machine=machine@...ry=0x5567f7280b18, is_kernel=is_kernel@...ry=false)
>     at util/synthetic-events.c:404
> #22 0x00005567c5cabff2 in perf_event__synthesize_mmap_events (tool=tool@...ry=0x0, event=event@...ry=0x5567f73ca000, pid=pid@...ry=5227, tgid=5227,
>     process=process@...ry=0x5567c5c19264 <perf_event__process>, machine=machine@...ry=0x5567f7280b18, mmap_data=false) at util/synthetic-events.c:536
> #23 0x00005567c5cac2a5 in __event__synthesize_thread (comm_event=comm_event@...ry=0x5567f73c8060, mmap_event=mmap_event@...ry=0x5567f73ca000,
>     fork_event=fork_event@...ry=0x5567f73c8080, namespaces_event=namespaces_event@...ry=0x5567f7246870, pid=5227, full=full@...ry=1,
>     process=0x5567c5c19264 <perf_event__process>, tool=0x0, machine=0x5567f7280b18, needs_mmap=true, mmap_data=false) at util/synthetic-events.c:851
> #24 0x00005567c5cac50d in __perf_event__synthesize_threads (tool=0x0, process=0x5567c5c19264 <perf_event__process>, machine=0x5567f7280b18, needs_mmap=true,
>     mmap_data=false, dirent=0x5567f7377400, start=475, num=19) at util/synthetic-events.c:986
> #25 0x00005567c5cac5c4 in synthesize_threads_worker (arg=<optimized out>) at util/synthetic-events.c:1018
> #26 0x00007f87886c0379 in start_thread (arg=<optimized out>) at pthread_create.c:448
> #27 0x00007f8788728fac in __GI___clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78
>
>

Powered by blists - more mailing lists