On Tue, 2012-06-05 at 12:26 +0200, Jan Kiszka wrote: > On 2012-06-05 12:20, Michael S. Tsirkin wrote: > > On Mon, Jun 04, 2012 at 04:56:01PM +0200, Jan Kiszka wrote: > >> No user in sight. > >> > >> Signed-off-by: Jan Kiszka <[email protected]> > > > > Just to note, currently (since the memory API changes) all users assume > > a single 4K page for MSI-X, we will need some API to remove that > > assumption. > > Still the patch is fine. > > Yep, I have a patch for this here (I think I posted an earlier version > before). It's required to do device assignment using the MSI layer.
Would you mind sending a pointer or reposting, I can't find it. Trying to switch vfio to not require existing capabilities checks I was reminded of the horror of msix_init(). We need a pretty good overhaul here to be able to allow the flexibility of defining where both the vector table a pba live to make it more compatible with real devices. I honestly don't see the benefit of allowing msix_init() the ability to resize a BAR to give itself room. There are only two other users currently, ivshmem and virtio, both of which allocate a BAR exclusively for MSI-X. Thanks, Alex
