Hello,
On Tue, 20 Jul 2021, luigi burdo wrote:
i was able to build on my quad qemu, a ram bank was die and make me the issue
with gcc..
this is what happening if i run pegasos 2 with --enable-kvm.
OK. Can you also show the full command so we know what options you used?
via_superio_cfg: unimplemented register 0xf2
via_superio_cfg: unimplemented register 0xf4
via_superio_cfg: unimplemented register 0xf6
via_superio_cfg: unimplemented register 0xf7
via_superio_cfg: unimplemented register 0xf4
via_superio_cfg: unimplemented register 0xf2
PegasosII Boot Strap (c) 2002-2003 bplan GmbH
Running on CPU PVR:000C0209
Enable L1 ICache... Done.
mv64361_write: Unimplemented register write 0x108 = 0
Reading W83194 : FAILED.
Setting Front Side Bus to 133MHz... FAILED.
Invalid access at addr 0xFE000E43, size 1, region '(null)', reason: rejected
So this shows that what I've seen on emulated KVM (running QEMU in a PPC
Linux guest running on qemu-system-ppc64 -M mac99) does not match what
real hardware does so that could be a bug in emulated KVM. As shown at the
end of this message:
https://lists.nongnu.org/archive/html/qemu-ppc/2021-06/msg00146.html
I did not get the Invalid access warning but instead got endless kvm exits
with the NIP not incrementing past the instruction doing this invalid
access so probably there's a problem with handling invalid access with
emulated KVM PR but I don't know where to look for that problem or how to
fix it. It could also be a bug in guest kernel or QEMU, I'm not sure. Hope
somebody with more knowledge about PPC KVM could give some hints.
Aparently this is not a problem on real machine where it works as expected
(the Invalid address is because we don't emulate this device but it's not
needed and it boots without it and we get the same warnings with TCG).
Invalid access at addr 0xFE000E44, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE000E41, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE000E42, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE000E40, size 1, region '(null)', reason: rejected
Configuring DDR...mv64361_write: Unimplemented register write 0x1494 = 291
[...]
Releasing IDE reset ... Done.
Configuring Legacy Devices
Initializing KBD...Invalid access at addr 0xFE0003F0, size 1, region '(null)',
reason: rejected
Done.
via_superio_cfg: unimplemented register 0xf6
via_superio_cfg: unimplemented register 0xf7
via_superio_cfg: unimplemented register 0xf2
Invalid access at addr 0xFE000084, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE000085, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE000086, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE000088, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE00008C, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE00008D, size 1, region '(null)', reason: rejected
Invalid access at addr 0xFE00008E, size 1, region '(null)', reason: rejected
Testing 10000000 Bytes, Pass: 00000000 Failed: 00000000
So does it stop here or do you get to the firmware ok prompt?
In dmesg i had many of this.
[ 3634.418495] kvmppc_exit_pr_progint: emulation at 700 failed (00000700)
[ 3634.418525] Couldn't emulate instruction 0x00000700 (op 0 xop 896)
[ 3634.418551] Couldn't emulate instruction 0x00000700 (op 0 xop 896)
[ 3634.418577] Couldn't emulate instruction 0x00000700 (op 0 xop 896)
[ 3634.418603] Couldn't emulate instr...
This looks like we're getting Program Exceptions that's raised for invalid
instructions so either it's going off track and tries to execute something
it should not or could be that since we're running G4 code on a G5 there
may be some differences so an instruction valid on G4 could be invalid on
G5? KVM should handle these but not sure it does for all such cases or it
works correctly as this was probably not tested for a while. We would need
to know where it got the exception but the above logs don't show that. Was
there anything before these? Maybe there's a kvm trace you can enable or
adding -d int would show the exception but that would generate a lot of
logs so you may need to filter out some unneeded ones to find the relevant
logs. We'd need the address where it got the exception that make it go to
0x700 and the opcode at that address. The logs above only show that it got
to 0x700 and got endless exceptions where as a handler for Program
Exception should be at 0x700 but probably the firmware did not install one
as it should not get such exceptions when working normally on a G4. Which
firmware version are you using? The 1.1 from up050404 should work but the
1.2 from MorphOS storage is known to have a cmpd instruction that's
invalid on G4 yet it works but may raise an exception or do something else
on G5. With TCG that's fixed by this patch:
https://lists.nongnu.org/archive/html/qemu-ppc/2021-07/msg00207.html
that's not in master yet but if using that firmware version this may also
cause problem in KVM if it does not do the same on G5 than on G4. I don't
know if that's the case but it's a possibility. Once we find out what
opcode is causing the program exception we can look up in the G5 and G4
manuals to see why it gets the exception and what can we do about it.
i will build latest linux kernel and do new test.
Apart from that you could also try what happens with the sc 1 calls that
is used but VOF when you use -kernel boot.img instead of -bios
pegasos2.rom. I got the host QEMU crash with that when trying on emulated
KVM but if it works on real hardware that could also be fixed in emulation
so I can at least reproduce the problems you find on real hardware.
Regards,
BALATON Zoltan
