Re: [PATCH v5 13/13] xen/arm: mmu: enable SMMU subsystem only in MMU

[Julien Grall <julien@xxxxxxx>]

On 22/08/2023 09:48, Henry Wang wrote:
> Hi Julien,
>
> > On Aug 22, 2023, at 16:18, Julien Grall <julien@xxxxxxx> wrote:
> >
> > On 22/08/2023 03:11, Henry Wang wrote:
> > > Hi Julien,
> > Hi,
> >
> > > > On Aug 22, 2023, at 05:34, Julien Grall <julien@xxxxxxx> wrote:
> > > >
> > > > Hi,
> > > >
> > > > On 14/08/2023 05:25, Henry Wang wrote:
> > > > > From: Penny Zheng <Penny.Zheng@xxxxxxx>
> > > > > SMMU subsystem is only supported in MMU system, so we make it dependent
> > > > > on CONFIG_HAS_MMU.
> > > > "only supported" as in it doesn't work with Xen or the HW is not supporting it?
> > > I think currently there are no hardware combination of MPU + SMMU, but
> > > theoretically I think this is a valid combination since SMMU supports the linear
> > > mapping. So would below reword looks good to you:
> > > “Currently the hardware use case of connecting SMMU to MPU system is rarely
> > > seen, so we make CONFIG_ARM_SMMU and CONFIG_ARM_SMMU_V3
> > > dependent on CONFIG_MMU."
> > I read this as there might be MPU system with SMMU in development. What you 
> > want to explain is why we can't let the developper to select the SMMU driver on 
> > an MPU system.
> >
> >  From my understanding this is because the drivers are expecting to use the 
> > page-tables and the concept doesn't exist in the MPU system. So the drivers are 
> > not ready for the MPU.
> I agree.
>
> > > > Also, I am not entirely convinced that anything in passthrough would properly 
> > > > work with MPU. At least none of the IOMMU drivers are. So I would consider to 
> > > > completely disable HAS_PASSTHROUGH.
> > > I agree, do you think adding below addition diff to this patch makes sense to 
> > > you?
> > I think it should be a replacement because none of the IOMMU drivers works for the MPU. 
> > So I would rather prefer if we avoid adding "depends on" on all of them (even 
> > if there are only 3) for now.
> I am a bit confused, I read your above explanation to the commit message as you
> agree with the idea that making SMMU driver not selectable by MPU system. 
No. I would rather prefer if HAS_PASSTHROUGH is completely disabled 
because I doubt the IOMMU infrastructure will work without any change 
for the MPU.
So it sounds incorrect to enable HAS_PASSTHROUGH until one of you 
confirm it works.
Cheers,

--
Julien Grall