Re: [GIT PULL RFC] pvclock cleanups and pvclock vsyscall support

[Avi Kivity <avi@redhat.com>]

On 10/18/2009 05:18 PM, Jeremy Fitzhardinge wrote:
> On 10/18/09 15:43, Avi Kivity wrote:
>    
>> On 10/16/2009 04:46 AM, Jeremy Fitzhardinge wrote:
>>      
>>> Care to cook up a patch to implement the kvm bits to make sure it all
>>> works OK for you?
>>>
>>>        
>> I've started to do that, but it occurs to me that we're missing out on
>> NUMA placement by forcing all clocks to be on the same page. OTOH, if
>> the clocks are heavily used, they'll stay in cache, and if not, who
>> cares.
>>      
> Yes, I'd say so.  I'd expect the data to be very close to read-only, so
> the lines should be shared pretty efficiently.
>
>    

There wouldn't be any sharing since each clock is on its own cache 
line.  But the point is valid regardless.

> On the other hand, there's nothing to stop us from moving to multiple
> pages in future (either to support NUMA placement, or just more than 64
> cpus).
>    

I'm already allocating multiple pages, so we'd just need to adjust the 
fixmap.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.