Re: [PATCH 01/11] iommu: Add device dma ownership set/release interfaces

[Lu Baolu]

On 11/17/21 9:35 PM, Jason Gunthorpe wrote:
> On Wed, Nov 17, 2021 at 01:22:19PM +0800, Lu Baolu wrote:
> > Hi Jason,
> >
> > On 11/16/21 9:46 PM, Jason Gunthorpe wrote:
> > > On Tue, Nov 16, 2021 at 09:57:30AM +0800, Lu Baolu wrote:
> > > > Hi Christoph,
> > > >
> > > > On 11/15/21 9:14 PM, Christoph Hellwig wrote:
> > > > > On Mon, Nov 15, 2021 at 10:05:42AM +0800, Lu Baolu wrote:
> > > > > > +enum iommu_dma_owner {
> > > > > > +	DMA_OWNER_NONE,
> > > > > > +	DMA_OWNER_KERNEL,
> > > > > > +	DMA_OWNER_USER,
> > > > > > +};
> > > > > > +
> > > > >
> > > > > > +	enum iommu_dma_owner dma_owner;
> > > > > > +	refcount_t owner_cnt;
> > > > > > +	struct file *owner_user_file;
> > > > >
> > > > > I'd just overload the ownership into owner_user_file,
> > > > >
> > > > >     NULL			-> no owner
> > > > >     (struct file *)1UL)	-> kernel
> > > > >     real pointer		-> user
> > > > >
> > > > > Which could simplify a lot of the code dealing with the owner.
> > > >
> > > > Yeah! Sounds reasonable. I will make this in the next version.
> > >
> > > It would be good to figure out how to make iommu_attach_device()
> > > enforce no other driver binding as a kernel user without a file *, as
> > > Robin pointed to, before optimizing this.
> > >
> > > This fixes an existing bug where iommu_attach_device() only checks the
> > > group size and is vunerable to a hot plug increasing the group size
> > > after it returns. That check should be replaced by this series's logic
> > > instead.
> >
> > As my my understanding, the essence of this problem is that only the
> > user owner of the iommu_group could attach an UNMANAGED domain to it.
> > If I understand it right, how about introducing a new interface to
> > allocate a user managed domain and storing the user file pointer in it.
>
> For iommu_attach_device() the semantic is simple non-sharing, so there
> is no need for the file * at all, it can just be NULL.

The file * being NULL means the device is only owned by the kernel
driver. Perhaps we can check this pointer in iommu_attach_device() to
avoid using it for user domain attachment.

> > Does above help here?
>
> No, iommu_attach_device() is kernel only and should not interact with
> userspace.

The existing iommu_attach_device() allows only for singleton group. As
we have added group ownership attribute, we can enforce this interface
only for kernel domain usage.

> I'm also going to see if I can learn what Tegra is doing with
> iommu_attach_group()

Okay! Thank you!

> Jason

Best regards,
baolu