[PATCH v7 00/17] Provide a new two step DMA mapping API

[Jason Gunthorpe]

On Fri, Mar 21, 2025 at 04:05:22PM +0000, Robin Murphy wrote:

> What everyone seems to have missed is that while it is technically true that
> the streaming DMA API doesn't need a literal struct page, it still very much
> depends on something which having a struct page makes it sufficiently safe
> to assume: that what it's being given is valid kernel memory that it can do
> things like phys_to_virt() or kmap_atomic() on.

No one has missed this, we are not yet at the point of implementing a
non-struct page PFN only path. That is going to be a followup series,
and yes there are going to need to be some cases where DMA will get
EOPNOTSUPP. You can't swiotlb something without a kmap, or MMIO for
instance.

> efficiently. And pushing the complexity into every caller to encourage and
> normalise drivers calling virt_to_phys() all over (_so_ many bugs there...)

That is unlikely to be how things end up.

> and pass magic flags to influence internal behaviour of the API
> implementation clearly isn't scalable. Don't think I haven't seen the other
> thread where Christian had the same concern that this "sounds like an
> absolutely horrible design."

Christian's perspective is thinking about DMABUF exporters using CPU
PFNs to mmap them to VMAs. Which is a uniquely DRM API abuse.

I think everyone who has really dug into this stuff understands that
the driver that is going to perform the DMA should be the one to do
the DMA mapping. It makes little sense for the driver providing the
memory to do the DMA mapping on behalf of the driver programming the
HW for DMA.

Regardless it doesn't really change this series as the same DMA API
interface to the driver is required to do the work. It doesn't matter
if the DMABUF API puts the calls on the exporter or importer side of
it's API.

> So what is it now, a layering violation in a hat with still no clear path to
> support SWIOTLB?

I was under the impression Leon had been testing SWIOTLB?

What does "no clear path to support SWIOTLB" mean?

Jason