Re: [PATCH v5 0/4] vfio/dma-buf: add TPH support for peer-to-peer access

["Christian König" <christian.koenig@amd.com>]

On 5/27/26 14:14, Jason Gunthorpe wrote:
> On Wed, May 27, 2026 at 08:55:49AM +0200, Christian König wrote:
>> On 5/26/26 16:43, Zhiping Zhang wrote:
>>> This series adds TLP Processing Hints (TPH) support to the VFIO dma-buf
>>> export path, allowing importing drivers (e.g. mlx5) to use the
>>> exporter's steering tag when performing peer-to-peer DMA into a
>>> VFIO-owned device.
>>
>> I'm not an expert for TPH, but that sounds very strange to me.
>>
>> As far as I know the TLP Processing Hints allow devices to give a
>> steering tag to the root complex together with memory accesses to
>> give fine grained control about cache usage. In other words it is an
>> extension to the classic snoop bit.
> 
> TPHS includes an bit of data on every TLP and the data transits to the
> eventual completer.
> 
> It does not have to be a root port.
>  
>> For P2P that is obviously nonsense because we don't have P2P support
>> for cached accesses.
> 
> For P2P the TPH data on the TLP will transit to the P2P completer
> unchanged.
> 
> It is up to the completer do define what it does with the TPH data.
> 
> Typically root ports in CPUs will use TPH data for cache placement
> instructions. But who knows what a P2P device will use it for.
> 
> In Linux the driver that owns the completing address space gets to
> specify how the TPH data works based on its own device specific
> knowledge.

Yeah that's a good point, I should probably rephrase the question.

I'm aware of how TPH works by adding the extra ST to the TLP.

But my question is how is that useful to a PCIe endpoint? What is the effect of the ST here?

I only know about the caching use case for the root complex and that's clearly not the case here.

Regards,
Christian.


> 
> Jason