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

[Lu Baolu]

Hi Bjorn,

On 11/16/21 4:38 AM, Bjorn Helgaas wrote:
> On Mon, Nov 15, 2021 at 10:05:42AM +0800, Lu Baolu wrote:
> >  From the perspective of who is initiating the device to do DMA, device
> > DMA could be divided into the following types:
> >
> >          DMA_OWNER_KERNEL: kernel device driver intiates the DMA
> >          DMA_OWNER_USER: userspace device driver intiates the DMA
>
> s/intiates/initiates/ (twice)

Yes.

> As your first sentence suggests, the driver doesn't actually
> *initiate* the DMA in either case.  One of the drivers programs the
> device, and the *device* initiates the DMA.

You are right. I could rephrase it as:

"From the perspective of who is controlling the device to do DMA ..."

> > DMA_OWNER_KERNEL and DMA_OWNER_USER are exclusive for all devices in
> > same iommu group as an iommu group is the smallest granularity of device
> > isolation and protection that the IOMMU subsystem can guarantee.
>
> I think this basically says DMA_OWNER_KERNEL and DMA_OWNER_USER are
> attributes of the iommu_group (not an individual device), and it
> applies to all devices in the iommu_group.  Below, you allude to the
> fact that the interfaces are per-device.  It's not clear to me why you
> made a per-device interface instead of a per-group interface.

Yes, the attributes are of the iommu_group. We have both per-device and
per-iommu_group interfaces. The former is for device drivers and the
latter is only for vfio who has an iommu_group based iommu abstract.

> > This
> > extends the iommu core to enforce this exclusion when devices are
> > assigned to userspace.
> >
> > Basically two new interfaces are provided:
> >
> >          int iommu_device_set_dma_owner(struct device *dev,
> >                  enum iommu_dma_owner mode, struct file *user_file);
> >          void iommu_device_release_dma_owner(struct device *dev,
> >                  enum iommu_dma_owner mode);
> >
> > Although above interfaces are per-device, DMA owner is tracked per group
> > under the hood. An iommu group cannot have both DMA_OWNER_KERNEL
> > and DMA_OWNER_USER set at the same time. Violation of this assumption
> > fails iommu_device_set_dma_owner().
> >
> > Kernel driver which does DMA have DMA_OWNER_KENREL automatically
> > set/released in the driver binding process (see next patch).
>
> s/DMA_OWNER_KENREL/DMA_OWNER_KERNEL/

Yes. Sorry for the typo.

> > Kernel driver which doesn't do DMA should not set the owner type (via a
> > new suppress flag in next patch). Device bound to such driver is considered
> > same as a driver-less device which is compatible to all owner types.
> >
> > Userspace driver framework (e.g. vfio) should set DMA_OWNER_USER for
> > a device before the userspace is allowed to access it, plus a fd pointer to
> > mark the user identity so a single group cannot be operated by multiple
> > users simultaneously. Vice versa, the owner type should be released after
> > the user access permission is withdrawn.

Best regards,
baolu