Re: [PATCH] VT-d: Tylersburg errata apply to further steppings

[Marek Marczykowski-Górecki <marmarek@xxxxxxxxxxxxxxxxxxxxxx>]

On Tue, Aug 03, 2021 at 03:44:10PM +0200, Jan Beulich wrote:
> On 03.08.2021 15:30, Marek Marczykowski-Górecki wrote:
> > On Tue, Aug 03, 2021 at 03:16:14PM +0200, Jan Beulich wrote:
> >> On 03.08.2021 15:12, Marek Marczykowski-Górecki wrote:
> >>> On Tue, Aug 03, 2021 at 03:06:50PM +0200, Jan Beulich wrote:
> >>>> On 03.08.2021 15:01, Marek Marczykowski-Górecki wrote:
> >>>>> Ok, then, just setting iommu_intremap=false should do the right thing,
> >>>>
> >>>> ... if "iommu=force" is in use (but not otherwise), ...
> >>>
> >>> But that's the purpose of iommu=force, no?
> >>> With "iommu=force": strictly require IOMMU
> >>> Without "iommu=force": use IOMMU on best-effort basis
> >>>
> >>> It makes sense to refuse the boot if intremap is broken in the first
> >>> case. But also, it makes sense to allow using IOMMU without intremp in
> >>> the second case.
> >>
> >> I agree with both statements. What I disagree with is that the latter
> >> happens by default (instead of only upon admin override), including
> >> the case of intremap being unavailable in the first place.
> > 
> > "upon admin override" - do I read the code right, that iommu=no-intremap
> > will actually achieve this effect?
> 
> In the case of this quirk - yes, as the call to the checking function is
> gated by a check of iommu_intremap. But by "admin override" I meant a
> per-guest attribute, not a host-wide one that isn't explicitly meant to
> cover all guests.
> 
> > Will that allow to use IOMMU without
> > interrupt remapping on a hardware where it's broken? In that case, maybe
> > at least add this info to the log message?
> 
> You mean to suggest the use of this option? I'd rather not, to be honest.
> I don't think options like this should be suggested to be used. I'd
> prefer if we had less of such options, i.e. if they went away after some
> initial integration phase.

Indeed, in fact I agree, this should be per-guest configurable option
(and in some cases, it kind of is - toolstack will refuse to boot a 
domain with PCI devices if IOMMU is missing). But, as you noted earlier,
there is no way to require intremap, on per-domain basis (regardless of
what the default behavior would be).

As for optionally requiring IOMMU host-wide, this still makes sense, as
IOMMU could be used not only for PCI passthrough, but also to protect
dom0 from some (possibly hot-plugged) devices - using quarantine
feature. There may be also some desired interactions with Intel TXT
(which AFAIR itself requires working VT-d).

-- 
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Attachment: signature.asc
Description: PGP signature