Re: [PATCH v3 1/9] PCI/P2PDMA: Add CONFIG_PCI_P2PDMA_CORE

[Matt Evans]

Hi Bjorn,

On 11/06/2026 17:07, Bjorn Helgaas wrote:
> On Wed, Jun 10, 2026 at 04:43:15PM +0100, Matt Evans wrote:
>> The P2PDMA code currently provides two features under the same
>> CONFIG_PCI_P2PDMA option:
>>
>>  1.  Locate providers via pcim_p2pdma_provider()
>>  2.  Manage actual P2P DMA
>>
>> Some drivers (such as vfio-pci) depend on 1, without having a hard
>> dependency on 2.
>>
>> A future commit expands the use of DMABUF in vfio-pci for non-P2P
>> scenarios, relying on pcim_p2pdma_provider() always being present.  If
>> that depended on CONFIG_PCI_P2PDMA, it would make vfio-pci only
>> available if CONFIG_ZONE_DEVICE is present (e.g. 64-bit systems), even
>> when P2P is not needed.
>>
>> To resolve this, introduce CONFIG_PCI_P2PDMA_CORE and refactor the
>> basic provider functionality into a new p2pdma_core.c file.  This is
>> available even if the CONFIG_PCI_P2PDMA feature is disabled (or
>> unavailable due to !CONFIG_ZONE_DEVICE).  Then, drivers can enable any
>> additional P2P features with the original CONFIG_PCI_P2PDMA (available
>> when CONFIG_ZONE_DEVICE is set).
>>
>> Signed-off-by: Matt Evans <matt@xxxxxxxxxx>
> 
> I thought this was going to be just a code move and new Kconfig
> option, but it involves a little more than that, e.g., adding
> pci_p2pdma_release_pool() and tweaking the RCU synchronization.
> 
> If possible, it would be nice to do that refactoring in a smaller
> preliminary patch so it's easier to review/bisect/etc and make this
> one a pure code move.

pci_p2pdma_release_pool() is really part of the split, not a functional
change, i.e. factoring the (P2P-only) pool handling out from the rest of
the pci_p2pdma_release() actions still needed for the _CORE case.

I guess .._release() could be split in place in a small patch, and then
moved alongside the rest in a second.

> I guess CONFIG_PCI_P2PDMA_CORE selects just part 1 ("Locate providers
> via pcim_p2pdma_provider()"), right?

Yes, I can reword that last paragraph of the commit message to make this
clearer.

>> +++ b/drivers/pci/p2pdma.h
>> @@ -0,0 +1,29 @@
>> +/* SPDX-License-Identifier: GPL-2.0 */
>> +/*
>> + * PCI Peer 2 Peer DMA support.
>> + */
>> +
>> +#ifndef _PCI_P2PDMA_H
>> +#define _PCI_P2PDMA_H
>> +
>> +#include <linux/genalloc.h>
>> +#include <linux/pci-p2pdma.h>
>> +#include <linux/xarray.h>
>> +
>> +struct pci_p2pdma {
>> +	struct gen_pool *pool;
>> +	bool p2pmem_published;
>> +	struct xarray map_types;
>> +	struct p2pdma_provider mem[PCI_STD_NUM_BARS];
>> +};
>> +
>> +#ifdef CONFIG_PCI_P2PDMA
>> +void pci_p2pdma_release_pool(struct pci_dev *pdev, struct pci_p2pdma *p2pdma);
>> +#else
>> +static inline void pci_p2pdma_release_pool(struct pci_dev *pdev, struct pci_p2pdma *p2pdma)
> 
> Wrap to fit in 80 columns like the rest of drivers/pci/

Fixed.

>> +{
>> +}
>> +#endif
>> +
>> +#endif
>> +
> 
> Spurious blank line at end.

Doh, fixed.

>> +++ b/drivers/pci/p2pdma_core.c
>> @@ -0,0 +1,118 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * PCI Peer 2 Peer DMA support core, providing a bare-bones
> 
> In this English text, I think I would spell out "Peer to Peer" instead
> of relying on the "2" homophone.  Same in p2pdma.h.

I would've preferred that too, it isn't correct; I was seeking to keep
the blurb here (and in p2pdma.h) as close to the original p2pdma.c as
possible.  I'll correct it in the new files (hyphenated "peer-to-peer"
in PCI-SIG language).


Thanks,


Matt