Re: Re: 2.6.26-rc8 pv_ops causes Unhandled invalid opcode fault/trap

[Jeremy Fitzhardinge <jeremy@goop.org>]

Guillaume Rousse wrote:
> Jeremy Fitzhardinge a écrit :
>> Guillaume Rousse wrote:
>>> Jeremy Fitzhardinge a écrit :
>>>> Guillaume Rousse wrote:
>>>>> Jeremy Fitzhardinge a écrit :
>>>>>>> FWIW, I just tried 2.6.26-rc1 and it booted fine.  I can narrow 
>>>>>>> it down further if that'd be useful to you.
>>>>>>>   
>>>>>>
>>>>>> Yes, please.
>>>>> Was there any final conclusion there ? Because I'm currently 
>>>>> trying a 2.6.27.5 with the same kind of issue:
>>>>
>>>> I don't know of any outstanding bugs which cause this kind of 
>>>> symptom.  What's happening on the Linux side when the domain 
>>>> crashes this way?
>>> Immediate crash, without further information.
>>>
>>> I've opened a but report on mandriva bugzilla:
>>> https://qa.mandriva.com/show_bug.cgi?id=46470
>>
>> Does earlyprintk=xen reveal anything more?  What does the rip 
>> (ffffffff8020b3ed) map to in the kernel?
> I guess earlyprintk=xen is a dom0 option, because passing it to domU 
> didn't change anything.

It will be most helpful if you're running a debug=y build of Xen, so the 
domU's early console output appears on the Xen console.

> The given adress map to set_page_prot, according to the System.map file.

That's interesting.

> BTW, it only happens with ISCSI-stored hosts, whereas it works OK with 
> LVM-stored hosts (tough on slightly different hardware). 

Uh, that's very mysterious.  The underlying storage shouldn't have any 
effect on whether the kernel can boot - it's either loaded or not.
> On other hosts, the guest kernel boots OK, and then fails to mount 
> underlying FS (but that's a different issue). I updated upstream 
> report with details.


I'm on vacation at the moment, but I'll give this a closer look when I 
return.

    J