Re: [PATCH 12/12] xen/arm: call iomem_permit_access for passthrough devices

[Julien Grall <julien@xxxxxxx>]

Hi Stefano,

On 26/05/2020 17:46, Stefano Stabellini wrote:
> On Sun, 24 May 2020, Julien Grall wrote:
> > On 30/04/2020 14:01, Julien Grall wrote:
> > > On 29/04/2020 21:47, Stefano Stabellini wrote:
> > > > On Wed, 15 Apr 2020, Julien Grall wrote: But doesn't it make sense to give
> > > > domU permission if it is going to get
> > > > the memory mapped? But admittedly I can't think of something that would
> > > > break because of the lack of the iomem_permit_access call in this code
> > > > path.
> > > On Arm, the permissions are only useful if you plan you DomU to delegate the
> > > regions to another domain. As your domain is not even aware it is running on
> > > Xen (we don't expose 'xen' node in the DT), it makes little sense to add the
> > > permission.
> > I actually found one use when helping a user last week. You can dump the list
> > of MMIO regions assigned to a guest from Xen Console.
> >
> > This will use d->iomem_caps that is modified via iomem_permit_access().
> > Without it, there is no easy way to confirm the list of MMIO regions assigned
> > to a guest. Although...
> >
> > > Even today, you can map IOMEM to a DomU and then revert the permission right
> > > after. They IOMEM will still be mapped in the guest and it will act normaly.
> > ... this would not help the case where permissions are reverted. But I am
> > assuming this shouldn't happen for Dom0less.
> Thank you for looking into this
>
>
> > Stefano, I am not sure what's your plan for the series itself for Xen 4.14. I
> > think this patch could go in now. Any thoughts?
> For the series: I have addresses all comments in my working tree except
> for the ones on memory allocation (the thread "xen: introduce
> reserve_heap_pages"). It looks like that part requires a complete
> rewrite, and it seems that the new code is not trivial to write. So I am
> thinking of not targeting 4.14. What do you think? Do you think the new
> code should be "easy" enough that I could target 4.14?
It may be a stretch with the code freeze on Friday. I would suggest to 
send it when it is ready and we can either include in Xen 4.14 or as 
soon as the tree re-open.
> For this patch: it is fine to go in now, doesn't have to wait for the
> series.
Feel free to add my ack on the patch.

Cheers,

--
Julien Grall