Re: [PATCH 3/6] PCI: Indicate that we want to force strict DMA for untrusted devices

[Doug Anderson]

Hi,

On Thu, Jun 24, 2021 at 6:38 AM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Jun 21, 2021 at 04:52:45PM -0700, Douglas Anderson wrote:
> > At the moment the generic IOMMU framework reaches into the PCIe device
> > to check the "untrusted" state and uses this information to figure out
> > if it should be running the IOMMU in strict or non-strict mode. Let's
> > instead set the new boolean in "struct device" to indicate when we
> > want forced strictness.
> >
> > NOTE: we still continue to set the "untrusted" bit in PCIe since that
> > apparently is used for more than just IOMMU strictness. It probably
> > makes sense for a later patchset to clarify all of the other needs we
> > have for "untrusted" PCIe devices (perhaps add more booleans into the
> > "struct device") so we can fully eliminate the need for the IOMMU
> > framework to reach into a PCIe device.
>
> It feels like the iommu code should not be messing with pci devices at
> all, please don't do this.

I think it's generally agreed that having the IOMMU code reach into
the PCIe code is pretty non-ideal, but that's not something that my
patch series added. The IOMMU code already has special cases to reach
into PCIe devices to decide strictness. I was actually trying to
reduce the amount of it.

> Why does this matter?  Why wouldn't a pci device use "strict" iommu at
> all times?  What happens if it does not?  Why are PCI devices special?

This is something pre-existing in Linux. In my patch series I was
trying to make PCI devices less special and take some of the concepts
from there and expand them, but in a cleaner way. It sounds like in my
v2 I should steer away from this and leave the existing PCI hacks
alone.

-Doug