[PATCH v2 1/4] iommu/arm-smmu-v3: Add feature detection for HTTU

[Jason Gunthorpe]

On Wed, Apr 24, 2024 at 01:20:53PM +0000, Shameerali Kolothum Thodi wrote:

> > On Wed, Apr 24, 2024 at 11:04:43AM +0100, Ryan Roberts wrote:
> > > On 23/04/2024 15:52, Jason Gunthorpe wrote:
> > > > On Tue, Apr 23, 2024 at 03:41:30PM +0100, Ryan Roberts wrote:
> > > >> Hi,
> > > >>
> > > >> I'm aiming to (slowly) get more involved with SMMU activities, although I'm
> > sure
> > > >> it will take a while to get up to speed and provide useful input. It was
> > > >> suggested that this series would be a useful starting point to dip my toe in.
> > > >> Please bear with me while I ask stupid questions...
> > > >
> > > > Nice!
> > > >
> > > > I would like to see this merged as it was part of the original three
> > > > implementations for iommufd dirty tracking. The other two have been
> > > > merged for a long time now..
> > >
> > > My understanding is that this series should pretty much apply on top of
> > > mainline, is that correct? (in practice there are some conflicts, but I think
> > > they are trivial).
> > 
> > It was originally based on my part 2, but I suspect it could apply
> > easially on top of part 2a. I don't think it is worth rebasing to
> > v6.9-rc since I expect Will to take 2a.
> 
> But isn't arm_smmu_domain_alloc_user() gets introduced only in part 3
> of the series? 

Right. "easially" was perhaps a bit too strong. It would have to be
rebased and pull back a little bit of the domain_alloc_user()
implementation.

If it is good otherwise lets consider doing this. Maybe HTTU and part
2b can go together during the v6.11 cycle?

> Do we plan to support DIRTY_TRACKING with domain_alloc( )
> itself by default?

No

Jason