> From: "Ted Unangst" <t...@tedunangst.com>
> Date: Mon, 01 Apr 2019 05:55:34 -0400
>
> Mark Kettenis wrote:
> > > Date: Thu, 28 Mar 2019 09:00:35 -0700
> > > From: Chris Cappuccio <ch...@nmedia.net>
> > >
> > > I think the current MSI-X implementation is a minimal skeleton,
> > > enough for some devices under virtualization. I don't know if it's
> > > enough for NVMe on real hardware.
> >
> > The main problem is that the MSI-X implementation has
> > machine-depenedent bits that are not implemented on all platforms.
>
> We do have macros everywhere though.
>
> #define pci_intr_map_msix(pa, vec, ihp) (-1)
>
> I think that should be good enough to allow the code to continue building on
> all platforms and fallback, no?
Hmm, maybe, yes.
> > > > - if (pci_intr_map_msi(pa, &ih) != 0) {
> > > > + if ((pci_intr_map_msix(pa, 0, &ih) != 0) && (pci_intr_map_msi(pa,
> > > > &ih) !=
> > > > 0)) {
> > > > if (pci_intr_map(pa, &ih) != 0) {
> > > > printf(": unable to map interrupt\n");
> > > > goto unmap;
> > >
> > >
> >
>
