Assertion failure when booting fbsd 8.2-release w/ sandy bridge

[jasonbking]

When booting the freebsd 8.2-release ISO under kvm, the system panics with:
assertion failed: ent->fields.trig_mode == IOAPIC_LEVEL_TRIG, file: kvm_ioapic.c, line: 249

[rmustacc]

Can we get the QEMU command line options you used? Also, is this kvm and kvm-cmd compiled from source, using the OI package, SmartOS?

[jasonbking]

pfexec qemu-system-x86_64 -enable-kvm
-drive file=FreeBSD-8.2-RELEASE-amd64-dvd1.iso,media=cdrom,if=ide
-boot order=dc
-smp 2
-m 1024
-vnc 0.0.0.0:0
-usbdevice tablet
-vga std
-drive file=/dev/zvol/rdsk/rpool/vm/fbsd1,if=ide,index=1

and compiled from source on oi151 w/ the joyent ACPI updates

fyi (I can put this as a separate bug) centos 6.0 also seems to be unhappy with the illumos-kvm apic support as well (except it just panics the vm, not the illumos host, and using 'noapic' seems to make it happy).

[rmustacc]

Does this fail if you add in -no-acpi and -no-hpet?

Can you confirm that you built kvm-cmd via ./build.sh?

I'll try to get some cycles to try out the 8.2 release. Though we've had no issues with using 8.1 in our testing. As for centos, please file a separate bug and include the arguments used to launch it.

[jasonbking]

It doesn't trigger the assertion, however it gets to a certain point in the boot process, then the VM hangs. The host is fine, I can kill the VM without any apparent issues. I seem to hit a panic with 8.1 ISO as well (I haven't looked at the dump yet to see if it's the same assertion failure). This is on Sandy Bridge. I will try to do a bit more diagnostics this week to see if I can find a way to make fbsd boot & update with the results, as well as try it out w/ SmartOS.

I tried (with the same results) the latest (as of the evening of 9/12) kvm module. However, the Illumos kernel is a few weeks old (8/26)-- most recent change set is 13441 - '1402 fix for #534 causes dire performance regression' (with the acpi fixes applied on top of that). However, looking at subsequent change sets after that one, I don't see anything that looks like it could help (though I'm happy to try with the absolute latest illumos tree + acpi if that'd help).

And I did use the build instructions -- the only change was altering the illumos-kvm-cmd makefile to use bash instead of sh for a few of the targets that were invoking sh (since they were in fact using bash-isms) since it was built on OpenIndiana (and I don't put /usr/gnu/bin in $PATH). I can post/submit a diff if interested.

[rmustacc]

I just tried booting centos 6 with -no-acpi. Without -no-acpi it would fail (note acpi and not apic). With -no-acpi it works. We'll be investigating this acpi issue. I suspect the same is true with your FreeBSD case. Can you confirm?

[jasonbking]

With '-no-acpi' (QEMU option I'm assuming), it begins to boot, but does not make any forward progress (I can provide a console screenshot if you like) -- it stops at 'md0: Preloaded image </boot/mfsroot> 4194304 bytes at 0xffffffff80d9fe20'
And of course no Illumos(host) panics.

[rmustacc]

The FreeBSD issue should be fixed with 3d2e6c9. Similarly all the CentOS issues should be fixed and no longer require extra QEMU flags.