Skip to contentSkip to conversation

kvm: repeatable kernel crash with Athlon II cpu

DISCUSSION Stas Sergeev 2 participants 9 messages Mar 29, 2016 – Mar 31, 2016 KVM

lore
Share & tools
lore
#0 lore raw

Hello.

I've been running dosemu and found out that it hangs or reboots one of my PCs. This happens with any fedora-23 kernels and hand-compiled kernels. The latest I tried, were:
4.4.6-300.fc23.x86_64
4.5.0-rc6

I tried to put the debug printfs into dosemu, and it seems the crash happens on KVM_RUN ioctl, but of course the one can't say for sure when an instant reboot happens. Yet if I disable the KVM use, or try on a different CPU, there is no crash too, so I am pretty sure it is the KVM bug.

The CPU on which the problem happens:
AMD Athlon(tm) II X2 245 Processor

The CPUs on which the problem doesn't happen:
AMD FX 8350
Intel(R) Core(TM) i7-5500U CPU @ 2.40GHz

I've set up the test-case here:
https://github.com/stsp/dosemu2/tree/bug

After checking out this branch, you need to get the file c_drive/c.tar.gz
and unpack it into ~/.dosemu/drives so that ~/.dosemu/drives/c exists.
After that you need to build and run dosemu.
Make sure to install slang-devel first, so that it can work on a text-console (kms or non-kms).
Then run dosemu. Dos Navigator should load. Hit Alt-x to exit, then Enter to confirm. At this point the host machine reboots, or, if there is a different CPU - dosemu just exits. The test must be repeated 2-3 times to make sure, but in most cases it works from the first time.

Hope someone can find such an old CPU an reproduce the problem. Please let me know what info should I provide or what debugging to do to help getting that fixed.

maintainer
#1 lore raw
Quoted 5 lines · Stas Sergeev

I've been running dosemu and found out that it hangs or reboots one of my PCs. This happens with any fedora-23 kernels and hand-compiled kernels. The latest I tried, were:
4.4.6-300.fc23.x86_64
4.5.0-rc6

Can you check if 4.3 works?

Quoted 5 lines

I tried to put the debug printfs into dosemu, and it seems the crash happens on KVM_RUN ioctl, but of course the one can't say for sure when an instant reboot happens. Yet if I disable the KVM use, or try on a different CPU, there is no crash too, so I am pretty sure it is the KVM bug.

Yes, that's likely.

Quoted 3 lines

Hope someone can find such an old CPU an reproduce the problem. Please let me know what info should I provide or what debugging to do to help getting that fixed.

Do you have experience bisecting the kernel? If so perhaps you can try doing that.

Paolo

#2 lore raw

29.03.2016 18:37, Paolo Bonzini пишет:

Quoted 10 lines · Stas Sergeev
Quoted 5 lines

I've been running dosemu and found out that it hangs or reboots one of my PCs. This happens with any fedora-23 kernels and hand-compiled kernels. The latest I tried, were:
4.4.6-300.fc23.x86_64
4.5.0-rc6

Can you check if 4.3 works?

Same crash on kernel-4.2.3-300.fc23.x86_64 - the oldest available for f23.

Quoted 6 lines
Quoted 3 lines

Hope someone can find such an old CPU an reproduce the problem. Please let me know what info should I provide or what debugging to do to help getting that fixed.

Do you have experience bisecting the kernel? If so perhaps you can try doing that.

But it looks like a genuine bug, not a regression. Of course I can check some 3.x versions...

maintainer
#3 lore raw
Quoted 11 lines · Stas Sergeev
Quoted 7 lines
Quoted 5 lines

I've been running dosemu and found out that it hangs or reboots one of my PCs. This happens with any fedora-23 kernels and hand-compiled kernels. The latest I tried, were:
4.4.6-300.fc23.x86_64
4.5.0-rc6

Can you check if 4.3 works?

Same crash on kernel-4.2.3-300.fc23.x86_64 - the oldest available for f23.

Quoted 7 lines
Quoted 6 lines
Quoted 6 lines
Quoted 3 lines
Quoted 3 lines

Hope someone can find such an old CPU an reproduce the problem. Please let me know what info should I provide or what debugging to do to help getting that fixed.

Do you have experience bisecting the kernel? If so perhaps you can try doing that.

But it looks like a genuine bug, not a regression.

Yeah, I was hoping it was introduced recently.

Quoted 1 line

Of course I can check some 3.x versions...

That would help a lot, thanks.

Paolo

#4 lore raw

29.03.2016 19:27, Paolo Bonzini пишет:

Quoted 18 lines · Stas Sergeev
Quoted 17 lines
Quoted 7 lines
Quoted 5 lines

I've been running dosemu and found out that it hangs or reboots one of my PCs. This happens with any fedora-23 kernels and hand-compiled kernels. The latest I tried, were:
4.4.6-300.fc23.x86_64
4.5.0-rc6

Can you check if 4.3 works?

Same crash on kernel-4.2.3-300.fc23.x86_64 - the oldest available for f23.

Quoted 6 lines
Quoted 6 lines
Quoted 3 lines
Quoted 3 lines

Hope someone can find such an old CPU an reproduce the problem. Please let me know what info should I provide or what debugging to do to help getting that fixed.

Do you have experience bisecting the kernel? If so perhaps you can try doing that.

Quoted 7 lines
Quoted 1 line

But it looks like a genuine bug, not a regression.

Yeah, I was hoping it was introduced recently.

Quoted 1 line

Of course I can check some 3.x versions...

That would help a lot, thanks.

Same on 3.18, and that's the oldest kernel that can be built on f23 (gcc5).
So that's a genuine bug after all, not a regression.
So does anyone have the AMD AthlonII X2 245 to try reproducing this?

maintainer
#5 lore raw
Quoted 2 lines · Stas Sergeev

Same on 3.18, and that's the oldest kernel that can be built on f23 (gcc5).

Can you try getting even older kernel directly from Koji? Worst case it doesn't boot. :)

Paolo

Quoted 2 lines

So that's a genuine bug after all, not a regression.
So does anyone have the AMD AthlonII X2 245 to try reproducing this?

#6 lore raw

29.03.2016 20:13, Paolo Bonzini пишет:

Quoted 8 lines · Stas Sergeev
Quoted 2 lines

Same on 3.18, and that's the oldest kernel that can be built on f23 (gcc5).

Can you try getting even older kernel directly from Koji? Worst case it doesn't boot. :)

Could you please send me the URLs?
This koji server is too slow to successfully browse it.

maintainer
#7 lore raw
Quoted 12 lines · Stas Sergeev

29.03.2016 20:13, Paolo Bonzini пишет:

Quoted 8 lines · Stas Sergeev
Quoted 2 lines

Same on 3.18, and that's the oldest kernel that can be built on f23 (gcc5).

Can you try getting even older kernel directly from Koji? Worst case it doesn't boot. :)

Could you please send me the URLs?
This koji server is too slow to successfully browse it.

Here they are:

kernel-2.6.35.6-45.fc14
http://koji.fedoraproject.org/koji/buildinfo?buildID=201002

kernel-2.6.38.6-26.rc1.fc15
http://koji.fedoraproject.org/koji/buildinfo?buildID=243161

kernel-3.1.0-7.fc16
http://koji.fedoraproject.org/koji/buildinfo?buildID=271832

kernel-3.3.4-5.fc17
http://koji.fedoraproject.org/koji/buildinfo?buildID=317848

kernel-3.6.10-4.fc18
http://koji.fedoraproject.org/koji/buildinfo?buildID=372260

kernel-3.9.5-301.fc19
http://koji.fedoraproject.org/koji/buildinfo?buildID=426020

kernel-3.11.10-301.fc20
http://koji.fedoraproject.org/koji/buildinfo?buildID=483008

kernel-3.17.4-301.fc21
http://koji.fedoraproject.org/koji/buildinfo?buildID=596012

Paolo

#8 lore raw
In reply to Paolo Bonzini

29.03.2016 20:13, Paolo Bonzini пишет:

Quoted 6 lines · Stas Sergeev
Quoted 2 lines

Same on 3.18, and that's the oldest kernel that can be built on f23 (gcc5).

Can you try getting even older kernel directly from Koji? Worst case it doesn't boot. :)

kernel-3.9.5-301.fc19.x86_64: crash happens
any older: no boot.