Hi Will,
On Wed, May 20, 2015 at 06:09:26PM +0100, Will Deacon wrote:
> On Tue, May 19, 2015 at 04:24:35PM +0100, Joerg Roedel wrote:
> > > + /* Page sizes */
> > > + if (reg & IDR5_GRAN64K)
> > > + pgsize_bitmap |= SZ_64K | SZ_512M;
> > > + if (reg & IDR5_GRAN16K)
> > > + pgsize_bitmap |= SZ_16K | SZ_32M;
> > > + if (reg & IDR5_GRAN4K)
> > > + pgsize_bitmap |= SZ_4K | SZ_2M | SZ_1G;
> > > +
> > > + arm_smmu_ops.pgsize_bitmap &= pgsize_bitmap;
> >
> > So this could effictivly lead to a zero pgsize_bitmap when there are
> > SMMUs in the system with support for different page sizes, no?
>
> Indeed, if there is no common page size then we end up not being able to
> support any. I tried to resolve this by moving the bitmap out of the
> iommu_ops and into the iommu_domain, but you weren't fond of that idea ;)
Well, what you could do (and what I think the core should do at some
point) is to build the resulting page-size bitmap by taking the biggest
minimum page-size from all iommus and OR the page-size bigger than that
together. For a system with 3 SMMUs supporting all of the above
pgsize_bitmaps the resulting bitmap would look like this:
SZ_64K | SZ_32M | SZ_2M | SZ_512M | SZ_1G;
With the biggest minimum page-size all of the bigger page-sizes can be
emulated. But that is not necessary for this patch-set, just a
suggestion for future work.
Joerg
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu
