Subject: Re: CVS commit: src/sys/arch/xen/xen
To: YAMAMOTO Takashi <firstname.lastname@example.org>
From: Manuel Bouyer <email@example.com>
Date: 10/08/2005 22:28:07
On Thu, Oct 06, 2005 at 10:02:54PM +0200, Manuel Bouyer wrote:
> > > events sent to the frontend.
> > txring->event isn't enough?
> In SMP systems I think there can be races between the domains on this,
> leading to a number of interrupt several times larger than what it should be.
> But with the other changes I've made, maybe this isn't an issue any more,
> and releasing the ring slot sooner may lead to less pages map/unmap.
> I'll do more tests on this.
OK, because xni_page can be shared between different xennetback instances,
more work has to be done inside the loop, so there's no benefit in delaying
the response. Doing it ASAP in xennetback_tx_free() makes the code much
simpler, and doesn't need the packet list. There is even a small performances
increase with a domU that deals properly with txring->event (that is,
NetBSD-current or linux 2.6, but not NetBSD-3)
Thanks for the comments !
Manuel Bouyer <firstname.lastname@example.org>
NetBSD: 26 ans d'experience feront toujours la difference