Re: [PATCH v4] iommu/io-pgtable-arm: Add support for contiguous hint bit

[Jason Gunthorpe <jgg@ziepe.ca>]

On Tue, Sep 08, 2026 at 09:36:35AM -0700, Daniel Mentz wrote:
> > > That said, install_leaf() is shared by other callers that reach it through
> > > ops->map_pages() directly, without going through iommu_pgsize(). panthor_vm_map_pages()
> > > (drivers/gpu/drm/panthor/panthor_mmu.c) is one -- it allocates its io_pgtable_ops via
> > > alloc_io_pgtable_ops(ARM_64_LPAE_S1, ...), same as any other LPAE consumer, but does its
> > > own chunking with a local get_pgsize() that only ever returns SZ_4K or SZ_2M, with no
> > > notion of the 64K/32M CONT boundaries. That can hand install_leaf() exactly the mixed
> > > iova=paddr=0x1000, pgcount=34 shape in a single call (panfrost's map loop uses the same
> > > get_pgsize() and hits the same case). So the prefix/aligned-group/suffix handling in
> > > install_leaf() is still needed for that path.
> >
> > Maybe other people can chime in here, but I feel, from an API
> > perspective, we should have only one way to make of use iopgtable. I
> > think we should just mandate that users honor the updated value for
> > pgtbl_cfg.pgsize_bitmap (and pgtbl_cfg.ias) after the
> > alloc_io_pgtable_ops() call and then choose page sizes appropriately.
> > Otherwise, we end up with two pieces of code in the kernel solving the
> > same problem. panthor_mmu won't be able to take advantage of the
> > contiguous bit until it gets updated accordingly, and that's ok.
> 
> On this note, I want to point out that I tried something similar in
> 2025 (see https://lore.kernel.org/linux-iommu/20250430231924.1481493-1-danielmentz@google.com/).
> The feedback I received from Jason at the time was:
> 
> "Wouldn't it be better to use the pgsize_bitmap and rely on the core
> code to indicate that a contiguous page size is being used, so it can
> trivially set the PTE bit without doing all this extra work?"

Still think this is right, Panthor is not using it well if it doesn't
jump through all the hoops. Probably there should be an general entry
point for map/unmap that does the right stuff for the DRM drivers to
call.

Jason