Re: [PATCH 4/6] kvm-userland.git mmu notifier compat

[Avi Kivity <avi@qumranet.com>]

Andrea Arcangeli wrote:
> On Tue, Jul 29, 2008 at 03:49:08PM +0300, Avi Kivity wrote:
>   
>> Applied; thanks.  I think page pinning can be worked around by pinning all 
>> pages on memslot registration (and unpinning on memslot removal); this will 
>> slow down virtual machine startup, but is at least simple.
>>     
>
> This has the benefit that will also fix the tlb issues, but if we go
> this way if a sles/rhel user takes the next kvm release, all ram will
> be pinned and it won't swap anything for him, like with my patch that
> added VM_LOCKED. Said that swap wasn't a reliable feature before so
> nobody should have depended on that.
>   

A 64-bit guest would pin all pages anyway as soon as it touched them the 
first time, through the direct map.

> If we really go this memslot-wide-pinning way I could submit to rhel
> and sles a mmu notifier backport so the rhel/sles users won't have to
> wait to get their kernel upgraded to 2.6.27 to get full kvm-paging.
>   

I really doubt they would take it; Red Hat will want to point people at 
ovirt and Novel at Xen.

It's not a simple patch.

-- 
error compiling committee.c: too many arguments to function