On Thu, Aug 28, 2025 at 08:45:42PM -0300, Jason Gunthorpe wrote:

On Thu, Aug 28, 2025 at 02:54:35PM -0600, Keith Busch wrote:

quoted

In truth though, I hadn't tried p2p metadata before today, and it looks
like bio_integrity_map_user() is missing the P2P extraction flags to
make that work. Just added this patch below, now I can set p2p or host
memory independently for data and integrity payloads:

I think it is a bit more than that, you have to make sure all the meta
data is the same, either all p2p or all cpu and then record this
somehow so the DMA mapping knows what kind it is.

Sure, I can get all that added in for the real patch.
 

Once that is all done then the above should still be OK, the dma unmap
of the data can follow Leon's new flag and the dma unmap of the
integrity can follow however integrity kept track (in the
bio_integrity_payload perhaps?) ??

We have available bits in the bio_integrity_payload bip_flags, so that
sounds doable. I think we'll need to rearrange some things so we can
reuse the important code for data and metadata mapping/unmapping, but
doesn't look too bad. I'll get started on that.