[PATCH v2 4/4] vfio/pci: Allow MMIO regions to be exported through dma-buf

[Oded Gabbay]

On Tue, Sep 6, 2022 at 8:59 PM Jason Gunthorpe <jgg at nvidia.com> wrote:
>
> On Tue, Sep 06, 2022 at 03:34:02PM +0300, Oded Gabbay wrote:
>
> > > > > > + /*
> > > > > > +  * Since the memory being mapped is a device memory it could never be in
> > > > > > +  * CPU caches.
> > > > > > +  */
> > > > > DMA_ATTR_SKIP_CPU_SYNC doesn't even apply to dma_map_resource, not sure
> > > > > where this wisdom comes from.
> > >
> > > Habana driver
> > I hate to throw the ball at someone else, but I actually copied the
> > code from the amdgpu driver, from amdgpu_vram_mgr_alloc_sgt() iirc.
> > And if you remember Jason, you asked why we use this specific define
> > in the original review you did and I replied the following (to which
> > you agreed and that's why we added the comment):
>
> Yes, I remember, but Christophs remark is that DMA_ATTR_SKIP_CPU_SYNC
> doesn't even do anything when passed to dma_map_resource().
>
> The only attr that seems to be used is DMA_ATTR_PRIVILEGED from what I
> can see.
>
> Jason
Yes, it appears he is correct.
Probably what happened is that either this was originally copied from
a use of dma_map_page or something similar that does check this
attribute, or maybe dma_map_resource used it in the past and the
underlying code has changed.

Regardless, it seems we can remove it from the calls to
dma_map_resource. I went over the kernel code and it seems only habana
and amdgpu (and amdkfd) are passing this property to dma_map_resource.
All other callers just pass 0.

Oded