Re: [PATCH 0 of 3] Update paging/sharing/access interfaces v2

["Andres Lagar-Cavilla" <andres@lagarcavilla.org>]

> At 01:08 -0500 on 09 Feb (1328749705), Andres Lagar-Cavilla wrote:
>> i(Was switch from domctl to memops)
>> Changes from v1 posted Feb 2nd 2012
>>
>> - Patches 1 & 2 Acked-by Tim Deegan on the hypervisor side
>> - Added patch 3 to clean up the enable domctl interface, based on
>>   discussion with Ian Campbell
>>
>> Description from original post follows:
>>
>> Per page operations in the paging, sharing, and access tracking
>> subsystems are
>> all implemented with domctls (e.g. a domctl to evict one page, or to
>> share one
>> page).
>>
>> Under heavy load, the domctl path reveals a lack of scalability. The
>> domctl
>> lock serializes dom0's vcpus in the hypervisor. When performing
>> thousands of
>> per-page operations on dozens of domains, these vcpus will spin in the
>> hypervisor. Beyond the aggressive locking, an added inefficiency of
>> blocking
>> vcpus in the domctl lock is that dom0 is prevented from re-scheduling
>> any of
>> its other work-starved processes.
>>
>> We retain the domctl interface for setting up and tearing down
>> paging/sharing/mem access for a domain. But we migrate all the per page
>> operations to use the memory_op hypercalls (e.g XENMEM_*).
>>
>> This is a backwards-incompatible ABI change. It's been floating on the
>> list for
>> a couple weeks now, with no nacks thus far.
>>
>> Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla>
>> Signed-off-by: Adin Scannell <adin@scannell.ca>
>
> Applied 1 and 2; thanks.
>
> I'll leave patch 3 for others to comment -- I know there are out-of-tree
> users of the mem-access interface, and changing the hypercalls is less
> disruptive than changing the libxc interface.

Makes a lot of sense. Thanks.

I don't view this change as sine qua-non, yet, "it would be nice if"...

Is there a timeout mechanism if out-of-tree consumers are not on the ball?

Actually, this hiatus allows me to float a perhaps cleaner way to map the
ring: the known problem is that the pager may die abruptly, and Xen is
still posting events to a page now belonging to some other dom0 process.
This is dealt with in the qemu-dm case by stuffing the ring in an unused
pfn (presumably somewhere in the mmio hole?)

Would that work? Is there a policy for parceling out these "magic pfn's"?

Andres

>
> Tim.
>