On Mon, Aug 25, 2025 at 06:46:50AM -0700, Christoph Hellwig wrote:

On Thu, Aug 21, 2025 at 01:44:19PM -0700, Keith Busch wrote:

Also use the chance to document why all this is PAGE_SIZE based and
not based on either the iommu granule size or the virt boundary.

This is a good opportunity to double check my assumptions:

PAGE_SIZEs, iommu granules, and virt boundaries are all power-of-two
values, and PAGE_SIZE is always the largest (or tied for largest) of
these.

If that's accurate, storing the lowest page offset is sufficient to
cover all the boundary masks.

If that's not accurate, then this kind of falls apart. I didn't find
anything enforcing this assumption, but I can't imagine it would make
sense for a device to require the virtual boundary be larger than the
page size. It'd be difficult to do IO to that. I also know the iommu
granule may be differant than PAGE_SIZE too, but it's always the smaller
of the two if they are not the same.