On Sun, 2013-06-23 at 16:58 -0500, Anthony Liguori wrote: > On Sun, Jun 23, 2013 at 4:39 PM, Benjamin Herrenschmidt > <b...@kernel.crashing.org> wrote: > > On Sun, 2013-06-23 at 10:02 -0500, Anthony Liguori wrote: > >> It doesn't make a lot of sense to have Power do MSI decoding in > >> userspace while x86 does it in the kernel. I assume the kernel > >> doesn't have enough information to do the translation on Power? > > > > Well, it would need the host bridge which it doesn't have and it > > would need a concept of mapping that to interrupt numbers which > > we don't want in the kernel. > > > > We can keep the kernel side a LOT simpler (completely avoiding > > the whole route bloatware) by sticking to our "simple" direct map kernel > > stuff and doing the mapping from address/data to interrupts in qemu, so > > we'll stick to that. > > Yeah, but none of this is Power specific... so we can do the same for x86, > no? > > I'm still trying to wrap my head around why we need MSI knowledge at > all in the kernel for x86. I presume it's to fast-path irqfd when > doing vhost?
Or device assignment. Any paths where we want to inject an MSI interrupt without going through userspace. Thanks, Alex
