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

[Luis Chamberlain]

On Tue, Mar 25, 2025 at 04:41:58PM +0200, Leon Romanovsky wrote:
> On Tue, Mar 25, 2025 at 09:36:37AM -0300, Jason Gunthorpe wrote:
> > On Fri, Mar 21, 2025 at 04:05:22PM +0000, Robin Murphy wrote:
> 
> <...>
> 
> > > 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?
> 
> Yes, SWIOTLB works

We will double check too.

> and Christoph said it more than once that he tested
> NVMe conversion patches and they worked.

We've taken this entire series and the NVMe patches and have built on
top of them. The nvme-pci driver does not have scatter list chaining
support, and we don't want to support that because it is backwards.
Instead, the two step DMA API lets us actually remove all that scatter
list cruft and provide a single solution for direct IO and io-uring
command passthrough to support large IOs [0] [1] and logical block sizes
up to 2 MiB.

We continue to plan to work on this and are happy to test this further.

Clearly, we don't want any regressions on NVMe.

[0] https://lore.kernel.org/all/20250320111328.2841690-1-mcgrof@kernel.org/
[1] https://lore.kernel.org/all/Z9v-1xjl7dD7Tr-H@bombadil.infradead.org/

  Luis