Re: [PATCH v3 06/20] iommu/mtk: Remove detach_dev callback

[Robin Murphy]

On 2022-11-29 02:07, Baolu Lu wrote:
> On 11/28/22 9:59 PM, Robin Murphy wrote:
> > On 2022-11-28 13:49, Jason Gunthorpe wrote:
> > > On Mon, Nov 28, 2022 at 02:46:34PM +0800, Lu Baolu wrote:
> > > > The IOMMU driver supports default domain, so the detach_dev op will 
> > > > never
> > > > be called. Remove it to avoid dead code.
> > > >
> > > > Signed-off-by: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > > > ---
> > > >   drivers/iommu/mtk_iommu.c | 9 ---------
> > > >   1 file changed, 9 deletions(-)
> > >
> > > I listed this driver as not supporting default domains:
> > >
> > > https://lore.kernel.org/linux-iommu/20220516135741.GV1343366@xxxxxxxxxx/
> > >
> > > ?
> > >
> > > Has something changed? Did I get it wrong?
> >
> > static struct iommu_domain *mtk_iommu_domain_alloc(unsigned type)
> > {
> >      struct mtk_iommu_domain *dom;
> >
> >      if (type != IOMMU_DOMAIN_DMA && type != IOMMU_DOMAIN_UNMANAGED)
> >          return NULL;
> > ...
> >
> >
> > This one runs on arm64, so has always supported default domains for 
> > iommu-dma to work.
>
> This, together with several other ones, only support IOMMU_DOMAIN_DMA
> type of default domain. The iommu core handle this by falling back to
> IOMMU_DOMAIN_DMA if other types fail.
>
>          dom = __iommu_domain_alloc(bus, type);
>          if (!dom && type != IOMMU_DOMAIN_DMA) {
>                  dom = __iommu_domain_alloc(bus, IOMMU_DOMAIN_DMA);
>                  if (dom)
>                          pr_warn("Failed to allocate default IOMMU 
> domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
>                                  type, group->name);
>          }
>
> I have another cleanup series:
>
> https://github.com/LuBaolu/intel-iommu/commits/iommu-use-def_default_type-wip
>
> which adds IOMMU_DOMAIN_DMA default domain type requirement in the
> def_domain_type callback. I planed to bring that to discussion after
> this one.

Per the discussion over on the s390 thread, I think that would be a step 
in the wrong direction. I'd prefer to keep .def_domain_type for 
device-specific requirements and express general driver domain support a 
different way. If the IOMMU_DOMAIN_DMA fallback is worth removing then 
the one for IOMMU_DOMAIN_BLOCKED is as well - no point doing half the job ;)

Thanks,
Robin.