[Xen-devel] Re: mapcache broken?

[Keir Fraser <Keir.Fraser@xxxxxxxxxxxx>]

On 10 Jan 2006, at 09:46, Leendert Van Doorn wrote:

> Your map_domain_page mapcache cleanup from last night breaks VMX 
> 32-bit,
> but funny enough not VMX 64-bit.
>
> The map_io_shared_page determines that there are 0 entries in the e820 
> map.
> The map itself doesn't appear to fail which probably means that a page 
> full
> with 0's is mapped and this would indicate at a problem with the 
> mapcache
> (this is just a conjecture, I haven't actually added to code to test 
> this,
> I only had 10min to look at this).
Yeah, I realised this might be a problem for VMX last night. Problem is 
that certain pagetables get map_domain_page()'ed for the lifetime of 
the guest, and are expected to then be mapped in all address spaces. 
Now the mapcache is per domain, this 'trick' won't work.
A new shadow mode should not do such long-term mappings. I'll see if I 
can come up with a temporary solution for current shadow mode (maybe 
reintroduce ioremap()...).
 -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel