Hi Praan,

On 10/06/2026 16:27, Pranjal Shrivastava wrote:
> On Fri, May 29, 2026 at 08:05:53PM -0300, Jason Gunthorpe wrote:
>> On Wed, May 27, 2026 at 03:09:46PM -0600, Alex Williamson wrote:
>>> This deserves a /* CONFIG_PCI_P2PDMA */ reference like the previous
>>> long span, though potentially a better solution would be to split the
>>> core code out to a separate file with shared header.  I asked opus to
>>> see what this would look like and it generated the patch below
>>> (unverified, compile tested only).  Bigger diff, harder to follow
>>> blame, but the shape of the core is much more apparent.  You're welcome
>>> to take or leave it.
>>
>> I think it is a good idea, it will make it easier to add more stuff to
>> these different aspects.
>>
>> Really they are now two different things with one layered on top of
>> the other.
>>
> 
> A little late to this party. I agree with this split per status quo.
> Although, one observation from me, I've posted a concurrent RFC [1] that
> implements ZONE_DEVICE P2P registration for VFIO. That work relies on
> pci_p2pdma_add_resource(), which requires ZONE_DEVICE and the full
> CONFIG_PCI_P2PDMA stack rather than just the CORE part.
> 
> I guess I can be careful to guard the ZONE_DEVICE P2P registration 
> feature in VFIO with the full PCI_P2PDMA config. Apart from that, this
> split provides a clean foundation for both series to coexist.

Thanks for the heads-up/Cc.

>From your description, I agree, just depend on/select CONFIG_PCI_P2PDMA
as usual.  The only new thing in this split is the
CONFIG_PCI_P2PDMA_CORE allows the subset, but the default case (as used
by a couple of other drivers) stays the same.

I just posted a v3 including this split so you can see it in practice.
(I'd done part of it manually anyway so just finished it off, not
character-identical to Alex's patch but does the same thing.)

Thanks,


Matt


> 
> Thanks,
> Praan
> 
> [1] https://lore.kernel.org/all/[email protected]/

Reply via email to