Re: [kvm-devel] [PATCH 10/13] KVM: Wire up hypercall handlers toa central arch-independent location

From: Anthony Liguori
Date: Fri Feb 23 2007 - 09:56:13 EST


Avi Kivity wrote:
Arjan van de Ven wrote:
Can't it be done by a helper function using a timer and a signal (or whatever mechanism we use to wake up vcpus)?
one could do that but it's not optimal; if the process DOES get waken up
earlier, it should take the interrupt then immediately, so that it
doesn't have to wake up again when the timer fires.

(in fact it would be nice if the guest could somehow poll at the irq
mask at waking from idle, so that it wouldn't need a vmexit/entry for
every such interrupt, but just do the right thing for all pending work)

In addition, yes it'll be a helper function, but since all drivers will
want the functionality its probably best off in the generic code


That's what Xen does (I'm not sure about the timer, but they do avoid unnecessary interrupts). As you can have memory shared between the guest and host, it's quite simple:

0. host puts data in ring
1. host sets interrupt timer
2. guest wakes up for unrelated reasons
3. guest sees data in ring, consumes it, and bumps the consumer pointer
4. timer fires, host sees consumer == producer, no need to issue interrupt

It's not quite a timer. It's an asynchronous notification mechanism. Guest sets a bit in the appropriate domain's shared_info and when the domain gets scheduled again, it checks it's bits.

An easy way to simulate this in KVM would be to have the guest set a bit in the paravirt page and then on the next trip down to userspace, we check the bits and dispatch appropriately. We know we'll get down to userspace at least for the next EINTR.

They have that code abstracted out and reusable by multiple drivers.

Why don't we see how many drivers we get first. I have my doubts that a paravirt disk driver will significantly out perform scsi emulation.

Regards,

Anthony Liguori

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/