Re: WARNING: at /usr/src/modules/kvm/mmu.c:390 account_shadowed()

[Avi Kivity <avi@qumranet.com>]

Marcelo Tosatti wrote:
> On Wed, Apr 23, 2008 at 09:30:06AM +0300, Avi Kivity wrote:
>   
>>> as I got no reply, I guess it is a bad setup on my part. If that might
>>> help, this happenned while I was doing a "make -j" on webkit svn tree
>>> (ie. heavy c++ compilation workload) .
>>>
>>>   
>>>       
>> No this is not bad setup.  No amount of bad setup should give this warning.
>>
>> You didn't get a reply because no one knows what to make of it, and 
>> because it's much more fun to debate endianess or contemplete guests 
>> with eighty thousand disks than to fix those impossible bugs.  If you 
>> can give clear instructions on how to reproduce this, we will try it 
>> out.  Please be sure to state OS name and versions for the guest as well 
>> as the host.
>>     
>
> It is valid to have more than PAGES_PER_HPAGE in the largepage's
> shadowed count. If the gpte read races with a pte-update-from-guest (and
> the pte update results in a different sp->role), it might account twice
> for a single gfn.
>
> Such "zombie" shadow pages should eventually be removed through
> recycling, allowing for instantiation of a large page, unless references
> can be leaked. Can't spot such leakage problem though.
>
>   

That strikes me as unlikely (though a valid scenario).

An alternative explanation is that we're seeing a nonpae guest, so each 
page can be shadowed in two different roles (two quadrants for a pte 
page) or even four (for a pgd page).

Thomas, are you running a 32-bit nonpae guest?

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone