Skip to contentSkip to conversation

New: kvm/guest crashes when smp > 1 with AMD FX8300; with host kernel oops from abrt as well

Bug 92291 [email protected] 1 participant 11 messages Jan 29, 2015 – Feb 12, 2015 KVM

lore
Share & tools
lore
#0 lore raw

https://bugzilla.kernel.org/show_bug.cgi?id=92291

            Bug ID: 92291
           Summary: kvm/guest crashes when smp > 1 with AMD FX8300; with
                    host kernel oops from abrt as well
           Product: Virtualization
           Version: unspecified
    Kernel Version: 3.18.3-201.fc21.x86_64 [host], 3.13.0-39-generic
                    [ubuntu guest]
          Hardware: x86-64
                OS: Linux
              Tree: Mainline
            Status: NEW
          Severity: high
          Priority: P1
         Component: kvm
          Assignee: [email protected]
          Reporter: [email protected]
        Regression: No

Created attachment 165171
--> https://bugzilla.kernel.org/attachment.cgi?id=165171&action=edit
kvm register dumps

Overview
========
Whenever I launch a kvm guest in qemu-system-x86_64 with -smp > 1 [-cpu host -enable-kvm of course], the guest kernel crashes at some stage; I've seen the kernel crash before completing boot-up, giving the recognizable kernel crash trace, as well as during 'normal' operation when the guest simply 'freezes'. The host kernel gives me an oops as well, although abrt won't let me report it now as I've got Nvidia proprietary drivers 'tainting' the kernel The bug seems non-specific to guest OS, though mainly it's been *buntu guests, with various different kernels including lowlatency, I even think I've seen the bug with a Windows guest. I've tried building qemu from git sources, no particularly noticeable difference;

Steps to reproduce
==================

1) Start linux guest on a multicore system [AMD FX8* series CPU?] with

$ qemu-system-x86_64 -m 2G -cpu host -smp 4 -enable-kvm -hda [image_file] -vga
vmware

2) wait until the guest kernel crashes; virtually every time in less than 10 minutes

Actual results
==============
guest kernel crashes, host kernel gives an oops too, presumably as a result of kvm passthrough

Expected results
================
guest virtual machine should work normally then keep working normally

it would be good to try to pipe dmesg from the guest to the host serial connection, in case that would assist?

Signature · 2 lines
You are receiving this mail because:
You are watching the assignee of the bug.
#2 lore raw

[Bug 92291] kvm/guest crashes when smp > 1 with AMD FX8300; with host kernel oops from abrt as well

In reply to [email protected]

https://bugzilla.kernel.org/show_bug.cgi?id=92291

--- Comment #2 from Mark <[email protected]> --- Created attachment 165191
--> https://bugzilla.kernel.org/attachment.cgi?id=165191&action=edit
serial log when no smp

Signature · 2 lines
You are receiving this mail because:
You are watching the assignee of the bug.
#4 lore raw

[Bug 92291] kvm/guest crashes when smp > 1 with AMD FX8300; with host kernel oops from abrt as well

In reply to [email protected]

https://bugzilla.kernel.org/show_bug.cgi?id=92291

--- Comment #4 from Mark <[email protected]> --- I should just add that that 'code' segment in the kvm dump is identical every time, I'd be happy to try to track what is causing it, providing someone could give me some pointers to debugging that kind of stuff in more detail

Code=00 01 48 c7 c0 8a b0 00 00 31 db 0f b7 0c 01 b8 05 00 00 00 <0f> 01 c1 0f 1f 44 00 00 5b 41 5c 41 5d 41 5e 5d c3 89 f0 31 c9 f0 0f b0 0d fb 26 e6 00 40

Signature · 2 lines
You are receiving this mail because:
You are watching the assignee of the bug.
#5 lore raw

[Bug 92291] kvm/guest crashes when smp > 1 with AMD FX8300; with host kernel oops from abrt as well

In reply to [email protected]

https://bugzilla.kernel.org/show_bug.cgi?id=92291

--- Comment #5 from Mark <[email protected]> --- Created attachment 166021
--> https://bugzilla.kernel.org/attachment.cgi?id=166021&action=edit
result of guest lsmod; identical for mono-cpu / multi-cpu

Signature · 2 lines
You are receiving this mail because:
You are watching the assignee of the bug.
#6 lore raw

[Bug 92291] kvm/guest crashes when smp > 1 with AMD FX8300; with host kernel oops from abrt as well

In reply to [email protected]

https://bugzilla.kernel.org/show_bug.cgi?id=92291

Alan <[email protected]> changed:

What |Removed |Added
----------------------------------------------------------------------------

             Status|NEW                         |NEEDINFO
                 CC|                            |[email protected]

--- Comment #6 from Alan <[email protected]> --- Can you reproduce it without the Nvidia blob loaded on the host ?

Signature · 2 lines
You are receiving this mail because:
You are watching the assignee of the bug.
#7 lore raw

[Bug 92291] kvm/guest crashes when smp > 1 with AMD FX8300; with host kernel oops from abrt as well

In reply to [email protected]

https://bugzilla.kernel.org/show_bug.cgi?id=92291

Paolo Bonzini <[email protected]> changed:

What |Removed |Added
----------------------------------------------------------------------------
CC| |[email protected]

--- Comment #7 from Paolo Bonzini <[email protected]> ---
Your guest is probably missing commit c1118b3602c2329671ad5ec8bdf8e374323d6343.

Signature · 2 lines
You are receiving this mail because:
You are watching the assignee of the bug.
#10 lore raw

[Bug 92291] kvm/guest crashes when smp > 1 with AMD FX8300; with host kernel oops from abrt as well

In reply to [email protected]

https://bugzilla.kernel.org/show_bug.cgi?id=92291

--- Comment #10 from Mark <[email protected]> --- well thanks very much guys, I'll cautiously say that a patched guest kernel seems to resolve it :-)

the bug seemed to appear even when the host is untainted; serial log says

[  375.989736] divide error: 0000 [#1] SMP 

that's a Kubuntu partition that manages without the proprietary drivers while Fedora won't seem to give graphics properly with nouveau;

not entirely satisfactory of course given that I've now had to modify a virtual machine that it was important to try to avoid modifying for testing purposes; although definitely better than crashing :-)

do we keep the bug report open so that the host kvm handling bug gets fixed too / is this the right place to report that?

Signature · 2 lines
You are receiving this mail because:
You are watching the assignee of the bug.