new qemu triggers illegal instruction on boot on ppc64el

[Christian Ehrhardt ]

Hi,
with pre-builds of upcoming qemu 4.2 and lbivirt 6.0 that can be found on [1] I've seen issues on ppc64.

Directly at boot I run into this crash:
ubuntu@dradis:~$ virsh start focal-t1 --console
Domain focal-t1 started
Connected to domain focal-t1
Escape character is ^]
Populating /vdevice methods
Populating /vdevice/vty@30000000
Populating /vdevice/nvram@71000000
Populating /pci@800000020000000

( 700 ) Program Exception [ 0 ]

    R0 .. R7 R8 .. R15 R16 .. R23 R24 .. R31
000000000dbf0b14 000000000dc63030 0000000000000000 0000000000008000
000000000e67eff0 000000000e47b010 000000000e7451bc 000000000000f003
000000000dc25e00 000000000dc63028 0000000000000000 0000000000000006
000000000e7592e8 000000000fbd00c8 000000000e771373 000000000dc1bc00
0000000000000000 0000000000000000 000000000dc63040 000000000dc20778
0000000000000000 0000000000000000 000000000dbf4750 0000000000000003
0000000000000000 0000000000000000 000000000dc20bb8 000000000000f001
0000000000000000 0000000000000000 000000000dc20948 ffffffffffffffff

    CR / XER LR / CTR SRR0 / SRR1 DAR / DSISR
        80000402 000000000dbf0b14 0000000000000000 0000000000000000
0000000020040000 0000000000000000 8000000000081000 00000000

Past issues with a similar signature [2][3] make me expect a illegal instruction issue.
But I can't see where exactly things break and what to do.

I'd ask to mirror that to IBM to get their help on this rather sooner than later (Focal feature freeze is coming).

Testing is as easy as:
 $ qemu-system-ppc64 -nographic

This leaves me in an interactive shell of the FW it seems, if there is anything I should poke there let me know:
5 >

Versions:
qemu-system-ppc 1:4.2-1ubuntu1~ppa4
qemu-slof 20180702+dfsg-1

Maybe it is "just" an update to slof that is needed.
But still the new hipervisor shouldn't Illegal-Op on "old FW".
I might understand vice versa, but not this way around.

[1]: https://launchpad.net/~ci-train-ppa-service/+archive/ubuntu/3883/+packages
[2]: https://bugzilla.redhat.com/show_bug.cgi?id=1400476
[3]: https://bugs.launchpad.net/ubuntu/+source/grub2/+bug/1459706

[Christian Ehrhardt  (paelzer)]

Also found in Debian as https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=950005

[Christian Ehrhardt  (paelzer)]

Breaks the same way with and without --enable-kvm
Breaks the same way with old machine types e.g. -M pseries-3.1
Breaks the same way on older -cpu settings power9 power8 power8e power7+

[Christian Ehrhardt  (paelzer)]

The SLOF tagged with qemu 4.2 is 20191209
Lets try those or a later version just to be sure.
Those are still not built from qemu source even though I'm working on that with MJT to slowly convert all of them.
[1]: is the latest currently available upstream.
I'm unsure if [2] could be a problem, the referenced qemu change isn't upstream
Lets go with the tagged version for now. [3]

Throwing that onto the test seems to fix the issue!
I'll open an MP for Debian to go in sync - if there is no response I can have Ubuntu go ahead.

[1]: https://github.com/aik/SLOF/releases/tag/qemu-slof-20191217
[2]: https://github.com/aik/SLOF/commit/7477a174abc03d3e6f508412f69ea0bbd63b93ac
[3]: https://github.com/aik/SLOF/releases/tag/qemu-slof-20191209

[Christian Ehrhardt  (paelzer)]

Fix submitted to Debian as https://salsa.debian.org/qemu-team/slof/merge_requests/1

[Christian Ehrhardt  (paelzer)]

My changes are in Debian now and synced to Focal