[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Xen-devel] [PATCH 1 of 4 RFC] xen, p2m: get_entry returns level of entry as well
On Thu, Jun 14, 2012 at 4:01 PM, George Dunlap
> On 14/06/12 11:21, Tim Deegan wrote:
>> At 12:54 +0100 on 08 Jun (1339160091), George Dunlap wrote:
>>> Return the p2m level of the entry which filled this request.
>>> Intended to be used to see if pages returned by the balloon
>>> driver are part of a superpage, and reclaim them if so.
>> This looks broadly correct, but it's a bit invasive. If there's any way
>> to rework patch 2 not to need it that would be helpful. It looks like
>> the main use is for detecting and removing superpage PoD entries; maybe
>> that could be done with a sweep like you have in the non-PoD case?
> You mean the non-balloon case? I forget why I thought that was a bad idea,
> exactly. Let me think about that.
I think I basically didn't want to have to do the full check of each
of the 512 individual entries of the p2m in the case that it wasn't,
in fact, a superpage; but I think in the common case it should either
be a full superpage, or the check at the beginning of
p2m_pod_zero_check_superpage() should fail out relatively early. So
we could probably get away without it.
Why don't I re-write patch 2 of this series to not require patch 1,
and include it in my other series. If there's a performance gain to
be had by avoiding the check, we can discuss that post-4.2.
Xen-devel mailing list