Re: [PULL 11/24] tcg/optimize: Use tcg_constant_internal with constant folding

[David Hildenbrand]

On 04.02.21 10:03, Richard W.M. Jones wrote:
> On Thu, Feb 04, 2021 at 09:38:45AM +0100, David Hildenbrand wrote:
> > On 04.02.21 08:55, David Hildenbrand wrote:
> > > On 04.02.21 07:41, David Hildenbrand wrote:
> > > > > Am 04.02.2021 um 03:22 schrieb Richard Henderson <richard.henderson@linaro.org>:
> > > > >
> > > > > On 2/1/21 10:45 AM, Richard W.M. Jones wrote:
> > > > > > This commit breaks running certain s390x binaries, at least
> > > > > > the "mount" command (or a library it uses) breaks.
> > > > > >
> > > > > > More details in this BZ:
> > > > > >
> > > > > > https://bugzilla.redhat.com/show_bug.cgi?id=1922248
> > > > > >
> > > > > > Could we revert this change since it seems to have caused other
> > > > > > problems as well?
> > > > >
> > > > > Well, the other problems have been fixed (which were in fact latent, and could
> > > > > have been produced by other means).  I would not like to sideline this patch
> > > > > set indefinitely.
> > > > >
> > > > > Could you give me some help extracting the relevant binaries?  "Begin with an
> > > > > s390x host" is a non-starter.
> > > >
> > > > Hi,
> > > >
> > > > I‘m planning on reproducing it today or tomorrow. Especially, finding a 
> > > > reproducer and trying reproducing on x86-64 host.
> > >
> > > FWIW, on an x86-64 host, I can boot F32, Fedora rawhide, and RHEL8.X
> > > just fine from qcow2 (so "mount" seems to work in that environment as
> > > expected). Maybe it's really s390x-host specific? I'll give it a try.
> >
> > F33 qcow2 [1] fails booting on an s390x/TCG host.
>
> What did the failure look like?

It starts booting just fine until

[   10.869011] Core dump to |/bin/false pipe failed
[   10.915968] systemd[1]: Finished Create list of static device nodes for the 
current kernel.
[   10.946424] systemd[1]: systemd-journald.service: Main process exited, 
code=killed, status=31/SYS
[   10.966677] systemd[1]: systemd-journald.service: Failed with result 
'signal'.
[   11.017545] systemd[1]: Failed to start Journal Service.
[FAILED] Failed to start Journal Service.
See 'systemctl status systemd-journald.service' for details.

which repeats a couple of times. Then things go nuts

[   32.488899] systemd[1]: Failed to start Rule-based Manager for Device Events 
and Files.
[FAILED] Failed to start Rule-based…r for Device Events and Files.
See 'systemctl status systemd-udevd.service' for details.
[   32.501449] systemd[1]: systemd-udevd.service: Scheduled restart job, 
restart counter is at 1.
[   32.502134] systemd[1]: Stopped Rule-based Manager for Device Events and 
Files.


Looks also related to /dev / udev.

> > I tried "-cpu qemu" and "-qemu qemu=vx=off". The same image boots on
> > x86-64/TCG host just fine.
> >
> >
> > With
> >
> > commit 8f17a975e60b773d7c366a81c0d9bbe304f30859
> > Author: Richard Henderson <richard.henderson@linaro.org>
> > Date:   Mon Mar 30 19:52:02 2020 -0700
> >
> >      tcg/optimize: Adjust TempOptInfo allocation
> >
> > The image boots just fine on s390x/TCG as well.
>
> Let me try this in a minute on my original test machine.

That's the commit exactly before the problematic one (didn't want to mess with 
reverts for now).

--
Thanks,

David / dhildenb