Re: [PULL 26/34] migration/multifd: Join the TLS thread

[Fabiano Rosas]

Michael Tokarev <mjt@tls.msk.ru> writes:

> 14.02.2024 16:27, Fabiano Rosas :
>> Michael Tokarev <mjt@tls.msk.ru> writes:
> ..>>> This change, which is suggested for -stable, while simple by its own, 
> seems
>>>> to depend on the previous changes in this series, which are not for 
>>>> -stable.
>>>> In particular, whole "Finally recycle all the threads" loop in 
>>>> multifd_send_terminate_threads()
>>>> (to which the join is being added by this change) is moved from elsewhere 
>>>> by
>>>> 12808db3b8 "migration/multifd: Cleanup multifd_save_cleanup()" (patch 24 in
>>>> this same series).
>>>>
>>> We can probably add the missing join right into the previous location of 
>>> this
>>> loop (before 12808db3b8).  I did this in the attached variant for 8.2, is
>>> this correct?
>
> I forgot to attach the patch.  It just moves the join from 
> multifd_send_terminate_threads()
> back to multifd_save_cleanup.  Attached now.
>
>> It should work. This was originally developed without the rest of the
>> changes on this PR.
>> 
>>> And this does not pass even the basic tests, so it's not that simple :)
>> 
>> Do you have a log of what failed?
>
> Re-running it again...  I haven't even tried to push it somewhere for CI to 
> run,
> I run local `ninja test', which painted some migration tests in red.  Here:
>
> 202/844 qemu:qtest+qtest-aarch64 / qtest-aarch64/migration-test   ERROR   
> 70.26s   killed by signal 6 SIGABRT
> 330/844 qemu:qtest+qtest-i386 / qtest-i386/migration-test         ERROR   
> 85.33s   killed by signal 6 SIGABRT
> 454/844 qemu:qtest+qtest-x86_64 / qtest-x86_64/migration-test     ERROR  
> 101.02s   killed by signal 6 SIGABRT
>
> Unfortunately I don't see anything interesting in the log:
>
> # starting QEMU: exec ./qemu-system-x86_64 -qtest unix:/tmp/qtest-463614.sock 
> -qtest-log /dev/null -chardev socket,path=/tmp/qtest-463614.qmp,id=char0 
> -mon chardev=char0,mode=control -display none -audio none -accel kvm -accel 
> tcg -machine pc-q35-8.2, -name target,debug-threads=on -m 150M -serial 
> file:/tmp/migration-test-SPJTI2/dest_serial -incoming defer -drive 
> if=none,id=d0,file=/tmp/migration-test-SPJTI2/bootsect,format=raw -device 
> ide-hd,drive=d0,secs=1,cyls=1,heads=1    2>/dev/null -accel qtest
> ----------------------------------- stderr -----------------------------------
> ../../build/qemu/8.2/tests/qtest/libqtest.c:204: kill_qemu() detected QEMU 
> death from signal 6 (Aborted)
> (test program exited with status code -6)
>
> Without the attached patch it works.

Ah ok, this is hitting the bug fixed by patch 31. Let's leave patches
26, 27 and 31 out of stable, it would be too risky to backport.