[PATCH] Add syscall tests when following/detaching from fork

[Keith Seitz]

Hi, Andrew,

Thank you for taking a look at this.

On 12/11/24 7:55 AM, Andrew Burgess wrote:
> Running on my x86-64 machine, with native-extended-gdbserver board (and
> native-gdbserver board too), I'm seeing a KPASS:
> 
>    (gdb) PASS: gdb.base/foll-fork-syscall.exp: follow-fork-mode=parent: detach-on-fork=off: test_catch_syscall: continue to breakpoint after fork
>    continue
>    Continuing.
>    
>    Thread 1.1 "foll-fork-sysca" hit Catchpoint 5 (call to syscall chdir), 0x00007ffff7d73e4b in chdir () from /lib64/libc.so.6
>    (gdb) KPASS: gdb.base/foll-fork-syscall.exp: follow-fork-mode=parent: detach-on-fork=off: test_catch_syscall: continue to chdir syscall (PRMS breakpoints/13457)
>    Remote debugging from host ::1, port 34732
>    Process /tmp/build/gdb/testsuite/outputs/gdb.base/foll-fork-syscall/foll-fork-syscall created; pid = 4189457
>    I am the parent
>    monitor exit
>    (gdb) Killing process(es): 4189457 4189458
> 
> Do you also see this?

Yes, I am also seeing this now. At first I thought this was some sort of
heisenfail, but after re-running this a million times via RACY_ITER and
manually, this is definitely a deterministic result.

The test KFAILs for native (unix) targets but succeeds for gdbserver.
I will account for that.

> In my mind, the ideal solution would be that the new test should
> introduce no new FAILs, just PASS/KFAIL, with all the KFAIL associated
> with a bug number.
> 
> Obviously, there's a limit to which platforms you can test on, but if
> you know that some platforms FAIL, it would be great if we could KFAIL
> rather than FAIL.

I will retest on RHEL/Fedora supported platforms.

Keith