Re: [RFC PATCH 0/2] Expose available KVM free memory slot count to help avoid aborts

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Jan 26, 2011 at 11:54:21AM +0200, Avi Kivity wrote:
> On 01/26/2011 11:39 AM, Michael S. Tsirkin wrote:
> >On Wed, Jan 26, 2011 at 11:23:21AM +0200, Avi Kivity wrote:
> >>  On 01/26/2011 11:20 AM, Michael S. Tsirkin wrote:
> >>  >On Wed, Jan 26, 2011 at 11:17:11AM +0200, Avi Kivity wrote:
> >>  >>   On 01/25/2011 07:58 PM, Michael S. Tsirkin wrote:
> >>  >>   >On Tue, Jan 25, 2011 at 07:33:40PM +0200, Avi Kivity wrote:
> >>  >>   >>    On 01/25/2011 04:59 PM, Michael S. Tsirkin wrote:
> >>  >>   >>    >On Tue, Jan 25, 2011 at 04:53:44PM +0200, Avi Kivity wrote:
> >>  >>   >>    >>     For the other lookups, which we
> >>  >>   >>    >>     believe will succeed, we can assume the probablity of a match is
> >>  >>   >>    >>     related to the slot size, and sort the slots by page count.
> >>  >>   >>    >
> >>  >>   >>    >Unlikely to be true for assigned device BARs.
> >>  >>   >>
> >>  >>   >>    We'll have a large slot at 4G+ - EOM, a medium slot at 1M-3G, and
> >>  >>   >>    lots of small slots for BARs and such.
> >>  >>   >>
> >>  >>   >>    The vast majority of faults will either touch one of the two largest
> >>  >>   >>    slots, or will miss all slots.  Relatively few will hit the small
> >>  >>   >>    slots.
> >>  >>   >
> >>  >>   >Not if you are using one of the assigned devices and don't
> >>  >>   >do any faults on memory :)
> >>  >>
> >>  >>   It's impossible not to fault on memory.
> >>  >
> >>  >No I mean the RAM.
> >>
> >>  No idea what you mean.  It's impossible not to fault on RAM, either
> >>  (unless you don't use it at all).
> >
> >I just mean that once you fault you map sptes and then you can use them
> >without exits.  mmio will cause exits each time. Right?
> 
> The swapper scanning sptes, ksmd, khugepaged, and swapping can all
> cause a page to be unmapped.  Though it should certainly happen with
> a much lower frequency than mmio.

Right. That's why I say that sorting by size might not be optimal.
Maybe a cache ...

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