Re: 2.6.29-rc8 pv_ops dom0 BUG / unable to handle kernel paging request

[Jeremy Fitzhardinge <jeremy@goop.org>]

Pasi Kärkkäinen wrote:
> On Sun, Mar 22, 2009 at 07:04:23PM +0200, Pasi Kärkkäinen wrote:
>   
>> On Sun, Mar 22, 2009 at 01:51:51PM +0200, Pasi Kärkkäinen wrote:
>>     
>>> On Sat, Mar 21, 2009 at 09:28:55PM -0700, Jeremy Fitzhardinge wrote:
>>>       
>>>> Pasi Kärkkäinen wrote:
>>>>         
>>>>> On Sun, Mar 22, 2009 at 12:50:31AM +0200, Pasi Kärkkäinen wrote:
>>>>>  
>>>>>           
>>>>>> On Sat, Mar 21, 2009 at 10:16:52PM +0200, Pasi Kärkkäinen wrote:
>>>>>>    
>>>>>>             
>>>>>>>> Also, do you see this problem before you've started any other domains?  
>>>>>>>> Or does it only happen once you've run a domU (or only while a domU is 
>>>>>>>> running)?
>>>>>>>>
>>>>>>>>        
>>>>>>>>                 
>>>>>>> I'm not running any other domains.. Only dom0 is running.
>>>>>>>
>>>>>>> Steps to reproduce this BUG on my pv_ops dom0 testbox:
>>>>>>>
>>>>>>> 1) Reboot the box to pv_ops dom0 kernel
>>>>>>> 2) Login to dom0 via ssh
>>>>>>> 3) Start kernel compilation on dom0 (make bzImage && make modules)
>>>>>>> 4) Wait some minutes and pv_ops dom0 kernel BUGs
>>>>>>>
>>>>>>> So no other domains has been or is running when this happens..
>>>>>>>
>>>>>>> I'll try disabling CONFIG_HIGHPTE now, and see if that makes any 
>>>>>>> difference.
>>>>>>>
>>>>>>>      
>>>>>>>               
>>>>>> CONFIG_HIGHPTE=y and pv_ops dom0 survives up for maybe 30 mins, and then
>>>>>> BUGs (during kernel compilation):
>>>>>> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-22-xen331-linux-2.6.29-rc8-bug-with-highpte.txt
>>>>>>
>>>>>>
>>>>>> CONFIG_HIGHPTE=n and I get BUG during system startup when udev is started:
>>>>>> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-23-xen331-linux-2.6.29-rc8-bug-no-highpte.txt
>>>>>>
>>>>>> Starting udev: BUG: unable to handle kernel paging request at 70007823
>>>>>> IP: [<e30ce245>] pdc_common_ops+0x171/0xfffffcfe [sata_promise]
>>>>>> *pdpt = 000000005f781001 
>>>>>> Oops: 0002 [#1] SMP 
>>>>>>
>>>>>> So yeah..  with CONFIG_HIGHPTE=n it seems to happen when sata_promise is 
>>>>>> loaded.. What should I try next? 
>>>>>>    
>>>>>>             
>>>>> Actually it's not only sata_promise. I tried 2 more times with the
>>>>> CONFIG_HIGHPTE=n pv_ops dom0 kernel:
>>>>>
>>>>> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-23-xen331-linux-2.6.29-rc8-bug-no-highpte.txt
>>>>> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-23-xen331-linux-2.6.29-rc8-bug-no-highpte-2.txt
>>>>> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-23-xen331-linux-2.6.29-rc8-bug-no-highpte-3.txt
>>>>>
>>>>> BUG: unable to handle kernel paging request at a536462c
>>>>> IP: [<e30f4278>] classes+0x688/0xfffffa30 [parport]
>>>>> *pdpt = 000000005f759001 
>>>>> Oops: 0002 [#1] SMP 
>>>>>  
>>>>>           
>>>> Hm, OK.  Something is clearly drastically amiss.  I'll try to repro.
>>>>
>>>>         
>>> Actually it seems CONFIG_HIGHPTE=n kernel fails also on baremetal:
>>> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-24-baremetal-2.6.29-rc8-bug-no-highpte.txt
>>>
>>> Starting udev: invalid opcode: 0000 [#1] SMP 
>>>
>>> Summary:
>>> CONFIG_HIGHPTE=n: both dom0 and baremetal fail during system startup when udev is started
>>> CONFIG_HIGHPTE=y: baremetal works OK, dom0 fails with BUG after around 30 mins of kernel compilation
>>>
>>>       
>> Please ignore this summary, there was something wrong with my kernel builds or
>> something. 
>>
>> I'll post new summary soon when I'm finished with testing. 
>>
>>     
>
> Ok, I did new fresh kernel+modules builds and re-tested everything.
>
> New summary:
> CONFIG_HIGHPTE=n: both dom0 and baremetal work OK, both survive kernel compilation.
> CONFIG_HIGHPTE=y: baremetal works OK and survives kernel compilation, but dom0 fails with BUG after around 20-30 mins of kernel compilation
>   

Thanks for getting a consistent test result; the other reports looked, 
frankly, scary and I wouldn't want to be on that wild goose chase.

These ones look much more tractable, though I don't really have a theory 
for them.  I'll have a look next week sometime.

    J