> selftests/bpf: Test faulting build IDs from a sleepable raw tracepoint

TL;DR:
- Test issue: raise() may call getpid again, so the BPF programs may overwrite
  the stack the test inspects; results may depend on libc
- Nit: usage message omits the new uprobe-paged-out-stop mode

> diff --git a/tools/testing/selftests/bpf/uprobe_multi.c 
> b/tools/testing/selftests/bpf/uprobe_multi.c
> --- a/tools/testing/selftests/bpf/uprobe_multi.c
> +++ b/tools/testing/selftests/bpf/uprobe_multi.c

[ ... ]

> +int __attribute__((weak)) trigger_stop(void)
> +{
> +     int err = prep_build_id_page(false /* page-out build ID */);
> +
> +     if (err)
> +             return err;
> +
> +     /*
> +      * Make one syscall that the BPF program under test filters on, then
> +      * stop this process so that the recorded stack cannot be overwritten
> +      * by another syscall while the caller inspects it.
> +      */
> +     (void)syscall(SYS_getpid);
> +     raise(SIGSTOP);

With glibc, raise() goes through pthread_kill() which calls __getpid() to
get the tgid for tgkill().  So rawtp_nofault and rawtp_sleepable can match
a second time on id == SYS_getpid, from inside raise(), before SIGSTOP
takes effect.

That second match overwrites stack_rawtp_* and res_rawtp_*, so the stack
that subtest_raw_tp() inspects comes from the getpid stub called by
__pthread_kill_implementation(), not from the explicit syscall(SYS_getpid).
In the sleepable case, that stack is also captured after the first match
has already faulted the build ID page back in.

Whether a uprobe_multi frame shows up then depends on how libc compiles
raise()/pthread_kill().

Could the BPF programs keep only the first match, for example by returning
early once res_rawtp_* is non-zero?  Or could the comment and commit
message mention that raise() issues a matching getpid() as well?

[ ... ]

> +     if (!strcmp("uprobe-paged-out-stop", argv[1]))
> +             return trigger_stop();
>  
>  error:
>       fprintf(stderr,

This isn't a bug, but main() now accepts "uprobe-paged-out-stop" while the
usage message that follows still lists only
<bench|usdt|uprobe-paged-out|uprobe-paged-in>.  Should the new mode be
added there?


---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/41802507259

Reply via email to