Re: kvm [2087]: load/store instruction decoding not implemented

[Marc Zyngier <marc.zyngier@arm.com>]

On 24/02/15 14:43, Richard W.M. Jones wrote:
> On Tue, Feb 24, 2015 at 02:42:46PM +0000, Marc Zyngier wrote:
>> On 24/02/15 14:36, Richard W.M. Jones wrote:
>>> On Tue, Feb 24, 2015 at 02:10:28PM +0000, Marc Zyngier wrote:
>>>> On 24/02/15 13:45, Richard W.M. Jones wrote:
>>>>> On Tue, Feb 24, 2015 at 01:12:49PM +0000, Marc Zyngier wrote:
>>>>>> Here's my theory: userspace is accessing something it should never
>>>>>> access (outside of RAM, basically), and doing so via a kernel interface.
>>>>>>
>>>>>> Is this process accessing /dev/mem by any chance? dmidecode anyone?
>>>>>
>>>>> Not as far as I know.  The userspace process is inserting modules.
>>>>>
>>>>> Here is the userspace function which is most likely to be running:
>>>>>
>>>>> https://github.com/libguestfs/supermin/blob/master/src/init.c#L292
>>>>
>>>> Hmmm. That seems quite inoffensive indeed...
>>>>
>>>>> Unfortunately because of lack of a full stack trace, I can't be sure
>>>>> exactly what system call is failing, but I'll probably add more debug
>>>>> to the userspace program later.
>>>>>
>>>>> BTW this worked fine in 3.19.  It's started failing in 3.20/4.0.  It
>>>>> also works fine on x86.
>>>>
>>>> Any chance you could find out whether that's a host or guest regression?
>>>
>>> Here is a summary of the test combinations that I have run:
>>>
>>>   guest kernel         host kernel         result
>>> --------------------------------------------------------
>>>   3.19.0-0.rc7         3.19.0-0.rc7        no bug seen
>>>
>>>   3.20.0-0.rc0         3.19.0-0.rc7        bug seen
>>>
>>>   3.19.0-0.rc7         4.0.0-0.rc1         no bug seen
>>>
>>>   4.0.0-0.rc1          4.0.0-0.rc1         bug seen
>>>
>>> So a guest regression, I think?
>>
>> Looks like it. Is your .config stashed somewhere? I'd like to give it a
>> go on my own setup...
> 
> Attached.

Thanks. I can reproduce that behaviour by just doing an "insmod
crc32.ko", having booted the guest using kvmtool.

Which means that meither your userspace or qemu are at fault here, but
that it probably is a genuine kernel bug that KVM happens to uncover.

Off to debug the sucker.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...