Re: [PATCH v4 07/10] vfio/pci: Add dma-buf export config for MMIO regions

[Leon Romanovsky <leon@kernel.org>]

On Tue, Sep 30, 2025 at 10:07:58AM -0600, Alex Williamson wrote:
> On Tue, 30 Sep 2025 10:57:48 +0300
> Leon Romanovsky <leon@kernel.org> wrote:
> 
> > On Mon, Sep 29, 2025 at 03:17:40PM -0600, Alex Williamson wrote:
> > > On Sun, 28 Sep 2025 17:50:17 +0300
> > > Leon Romanovsky <leon@kernel.org> wrote:
> > >   
> > > > From: Leon Romanovsky <leonro@nvidia.com>
> > > > 
> > > > Add new kernel config which indicates support for dma-buf export
> > > > of MMIO regions, which implementation is provided in next patches.
> > > > 
> > > > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > > > ---
> > > >  drivers/vfio/pci/Kconfig | 20 ++++++++++++++++++++
> > > >  1 file changed, 20 insertions(+)
> > > > 
> > > > diff --git a/drivers/vfio/pci/Kconfig b/drivers/vfio/pci/Kconfig
> > > > index 2b0172f54665..55ae888bf26a 100644
> > > > --- a/drivers/vfio/pci/Kconfig
> > > > +++ b/drivers/vfio/pci/Kconfig
> > > > @@ -55,6 +55,26 @@ config VFIO_PCI_ZDEV_KVM
> > > >  
> > > >  	  To enable s390x KVM vfio-pci extensions, say Y.
> > > >  
> > > > +config VFIO_PCI_DMABUF
> > > > +	bool "VFIO PCI extensions for DMA-BUF"
> > > > +	depends on VFIO_PCI_CORE
> > > > +	depends on PCI_P2PDMA && DMA_SHARED_BUFFER
> > > > +	default y
> > > > +	help
> > > > +	  Enable support for VFIO PCI extensions that allow exporting
> > > > +	  device MMIO regions as DMA-BUFs for peer devices to access via
> > > > +	  peer-to-peer (P2P) DMA.
> > > > +
> > > > +	  This feature enables a VFIO-managed PCI device to export a portion
> > > > +	  of its MMIO BAR as a DMA-BUF file descriptor, which can be passed
> > > > +	  to other userspace drivers or kernel subsystems capable of
> > > > +	  initiating DMA to that region.
> > > > +
> > > > +	  Say Y here if you want to enable VFIO DMABUF-based MMIO export
> > > > +	  support for peer-to-peer DMA use cases.
> > > > +
> > > > +	  If unsure, say N.
> > > > +
> > > >  source "drivers/vfio/pci/mlx5/Kconfig"
> > > >  
> > > >  source "drivers/vfio/pci/hisilicon/Kconfig"  
> > > 
> > > This is only necessary if we think there's a need to build a kernel with
> > > P2PDMA and VFIO_PCI, but not VFIO_PCI_DMABUF.  Does that need really
> > > exist?  
> > 
> > It is used to filter build of vfio_pci_dmabuf.c - drivers/vfio/pci/Makefile:
> > vfio-pci-core-$(CONFIG_VFIO_PCI_DMABUF) += vfio_pci_dmabuf.o
> 
> Maybe my question of whether it needs to exist at all is too broad.
> Does it need to be a user visible Kconfig option?  Where do we see the
> need to preclude this feature from vfio-pci if the dependencies are
> enabled?

The dependencies are for the platform and not for the devices. For
example, hisilicon device mentioned in other email doesn't support
p2p, but the platform most likely support.

I don't have strong feelings about this config and at least for our use
case will always be enabled. I can hide it from the users.

> 
> > > I also find it unusual to create the Kconfig before adding the
> > > supporting code.  Maybe this could be popped to the end or rolled into
> > > the last patch if we decided to keep it.  Thanks,  
> > 
> > It is leftover from previous version, I can squash it, but first we need
> > to decide what to do with pcim_p2pdma_init() call, if it needs to be
> > guarded or not.
> 
> As in the other thread, I think it would be cleaner in an IS_ENABLED
> branch.  I'm tempted to suggest we filter out EOPNOTSUPP to allow it to
> be unconditional, but I understand your point with the list_head
> initialization.  Thanks,

We can add dmabuf list to struct unconditionally, as memory overhead is
negligible. It will allow us to drop IS_ENABLED() too.

Thanks

> 
> Alex
> 
>