Re: generic.MP stability on SMP QEMU?

пн, 8 дек. 2025 г., 22:44 Andrew Randrianasulu <[email protected]>:

> On Mon, Dec 8, 2025 at 11:22 PM Andrew Randrianasulu
> <[email protected]> wrote:
> >
> > I might experience something described in this thread:
> >
> > https://mail-index.netbsd.org/port-sparc/2024/12/03/msg002823.html
> >
> > But on qemu-emulated SS-20 with latest 10.1_stable kernel
> >
> https://nycdn.netbsd.org/pub/NetBSD-daily/netbsd-10/20251129160226Z/sparc/binary/kernel/
> >
> > and 10.1 userland.
> >
> > I thought it was lib/57628 but it was pulled up to 10 branch before 10.1
> was cut?
> >
> > https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=57628
> >
> > Note, I use self-compiled 32bit i586 qemu on Slackware
> >
> >
> > qemu-system-sparc -hda
> /mnt/zip/home/guest/QEMU/sparc-netbsd-15g-compressed-r1.qcow2 -m 1536 -g
> 1024x768x24 -smp1 -acceltcg,thread=multi,tb-size=256 -M SS-20
> >
> > if I set -smp to 2, 3 or 4 compiling bigger C++ things like re2c (needed
> by ninja, needed by meson, needed by gobject-introspection, needed by gtk2)
> results in hung build process (make install from /usr/pkgsrc)
> >
> > I see some messages _lwp_ctl (?) can't allocate memory, but powerdown
> usually works (on 10.1 kernel it resulted in some kernel panic in ffs, so
> fsck was forced on new boot).
> >
> > I hope I'll try to install tests.tgz and see if atf catch anything ...
>
>
> https://mastodon.online/@Andrew_R/115686358603684472
>
> seems to hang on futext_wake_op_cmp test .. (not repeated yet)
>


on qemu with -smp 1 tests still running past this point, like 584 out of
960.

I looked up bug database and found exactly one entry from 2017 with query
port sparc + SMP + all prio levels selected:

https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=52155

will try to repeat on qemu with -nographic and create new bugreport if my
hang at specific test is reproducible.

some ATF tests take minute and more, but I can press "Enter" on keyboard
and it will make new line on terminal. In hang case I can't do that, or
ctrl-c out to shell. So far I was running tests as root as suggested. I
think I disabled core generation for saving space, and some tests fail due
to this because they assume core must be generated.



> >
> >
>