Re: [Xen-devel] Implementing working set tracker in xen

[Tim Deegan <tim@xxxxxxx>]

At 13:27 +0200 on 03 Apr (1428067639), HANNAS YAYA Issa wrote:
> Thank you a lot Tim.Your Method which consist of using shadow page 
> table help me.
> However when I use the method shadow_blow_tables() I don't think it 
> makes all page unavailable.

I'm pretty sure it does -- e.g. live migration of PV guetst relies on
it at the moment. :)  If you can find a reproducible counterexample
please let us know.

> because the the number of page fault (I get it from sh_page_fault) does 
> not seem to increase.
> 
> when shadow_blow_tables() is called this code is executed:
> 
> for_each_vcpu(d, v)
>          for ( i = 0 ; i < 4 ; i++ )
>              if ( !pagetable_is_null(v->arch.shadow_table[i]) )
>                  shadow_unhook_mappings(v,
>                                 
> pagetable_get_mfn(v->arch.shadow_table[i]), 0);
> 
> I wonder about the '4' in the for loop. does it means that there is 
> only 4 shadow page table per-vcpu?

No - it means that the top level of the shadow pagetable that's
currently in use by this vcpu may be up to four pages long.
Read the big comment in mm/shadow/common.c:1173 for an explanation.

Cheers,

Tim.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel