Re: [PATCH v6 18/18] RDMA/mlx5: Ask P2PDMA whether ATS takes a direct peer-to-peer route

[Jason Gunthorpe <jgg@ziepe.ca>]

On Fri, Sep 18, 2026 at 03:15:00PM +0300, Leon Romanovsky wrote:
> > That is, a flag to tell the topology check that some transactions
> > *will* take the host-bridge path due to IOVA being used, and that the
> > computations including pci_p2pdma_distance() need to check whether that
> > is possible (checking whitelist etc.) and return the corresponding
> > THRU_HOST_BRIDGE mapping type. Translated transactions taking a short-
> > cut using the bus-address would then be hidden from the driver.
> 
> The word *will* puzzles me. As I understand it, if your device has ATS
> enabled all the time, it should always get THRU_HOST_BRIDGE.

The P2P subsystem currently does not understand ATS, it assumes the
device will use untranslated requests and makes the routing
calculation accordingly.

If the device knows it will use ATS it should ask P2P for an ATS path,
and P2P should only return THRU_HOST_BRIDGE or failure.

The purpose of involving the P2P subsystem is to verify that the ACS
flags for the ATS translated path are going to work. There are many ACS
configurations where the fabric will hang :(

For something like mlx5 it is ideal if P2P chooses between ATS or !ATS
based on what provides an optimal transfer. eg there is usually no
reason to use ATS to access system memory for streaming (ie non cache
hitting) transfers.

Jason