[Intel-gfx] [PATCH v1 1/5] iommufd: Create access in vfio_iommufd_emulated_bind()

[Liu, Yi L]

> From: Tian, Kevin <kevin.tian at intel.com>
> Sent: Thursday, March 16, 2023 1:49 PM
> 
> > From: Nicolin Chen
> > Sent: Thursday, March 16, 2023 1:44 PM
> >
> > On Thu, Mar 16, 2023 at 05:38:41AM +0000, Tian, Kevin wrote:
> > > External email: Use caution opening links or attachments
> > >
> > >
> > > > From: Nicolin Chen <nicolinc at nvidia.com>
> > > > Sent: Thursday, March 16, 2023 1:33 PM
> > > >
> > > > Hi Kevin,
> > > >
> > > > I've fixed the other two commits. Here is the one that I am
> > > > not sure about:
> > > >
> > > > On Thu, Mar 16, 2023 at 02:53:50AM +0000, Tian, Kevin wrote:
> > > >
> > > > > > [2] This adds iommufd_access_detach() in the cdev series:
> > > > > >     "iommufd/device: Add iommufd_access_detach() API"
> > > > > >
> > > > > >
> > > >
> >
> https://github.com/nicolinc/iommufd/commit/4110522146ca1fc0d5321c04a
> > > > > > 097e2c9d9e26af4
> > > > >
> > > > > also add a check if old_ioas exists it must equal to the new_ioas in
> > attach.
> > > >
> > > > This is the commit adding detach(). And there's a check in it:
> > > >       if (WARN_ON(!access->ioas))
> > > >
> > > > Do you mean having an "if (access->ioas) return -EBUSY;" line
> > > > in the commit adding attach()?
> > >
> > > if (access->ioas && access->ioas != new_ioas)
> > >         return -EBUSY;
> > >
> > > yes this is for attach.
> >
> > OK. For attach(), the access->ioas shouldn't be !NULL, I think.
> > At the point of adding attach(), the uAPI doesn't support the
> > replacement use case yet. And later we have a separate API for
> > that.
> 
> what about user calling attach twice in cdev?

In the cdev series, VFIO only one attach is allowed so far. If
vfio_device->iommufd_attached is set, would return -EBUSY to user.
But this does not prevent the iommufd side to allow attach twice.
Nic's replace series would relax it if user means to replace existing
attached hwpt/ioas.

Regards,
Yi Liu