Re: [PATCH v3 0/3] Generic SMMU Bindings

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Tue, 2 Feb 2021, Julien Grall wrote:
> On 02/02/2021 17:44, Stefano Stabellini wrote:
> > On Tue, 2 Feb 2021, Rahul Singh wrote:
> > > Hello Stefano,
> > > 
> > > > On 26 Jan 2021, at 10:58 pm, Stefano Stabellini <sstabellini@xxxxxxxxxx>
> > > > wrote:
> > > > 
> > > > Hi all,
> > > > 
> > > > This series introduces support for the generic SMMU bindings to
> > > > xen/drivers/passthrough/arm/smmu.c.
> > > > 
> > > > The last version of the series was
> > > > https://marc.info/?l=xen-devel&m=159539053406643
> > > > 
> > > > I realize that it is late for 4.15 -- I think it is OK if this series
> > > > goes in afterwards.
> > > 
> > > I tested the series on the Juno board it is woking fine.
> > > I found one issue in SMMU driver while testing this series that is not
> > > related to this series but already existing issue in SMMU driver.
> > > 
> > > If there are more than one device behind SMMU and they share the same
> > > Stream-Id, SMMU driver is creating the new SMR entry without checking the
> > > already configured SMR entry if SMR entry correspond to stream-id is
> > > already configured.  Because of this I observed the stream match conflicts
> > > on Juno board.
> > > 
> > > (XEN) smmu: /iommu@7fb30000: Unexpected global fault, this could be
> > > serious
> > > (XEN) smmu: /iommu@7fb30000:      GFSR 0x00000004, GFSYNR0 0x00000006,
> > > GFSYNR1 0x00000000, GFSYNR2 0x00000000
> > > 
> > > 
> > > Below two patches is required to be ported to Xen to fix the issue from
> > > Linux driver.
> > > 
> > > https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/iommu/arm-smmu.c?h=linux-5.8.y&id=1f3d5ca43019bff1105838712d55be087d93c0da
> > > https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/iommu/arm-smmu.c?h=linux-5.8.y&id=21174240e4f4439bb8ed6c116cdbdc03eba2126e
> > 
> > 
> > Good catch and thanks for the pointers! Do you have any interest in
> > backporting these two patches or should I put them on my TODO list?
> > 
> > Unrelated to who does the job, we should discuss if it makes sense to
> > try to fix the bug for 4.15. The patches don't seem trivial so I am
> > tempted to say that it might be best to leave the bug unfixed for 4.15
> > and fix it later.
> 
> SMMU support on Juno is not that interesting because IIRC the stream-ID is the
> same for all the devices. So it is all or nothing passthrough.
> 
> For other HW, this may be a useful feature. Yet we would need a way to group
> the devices for passthrough.
> 
> In this context, I would consider it more a feature than a bug because the
> SMMU driver never remotely work on such HW.

I see. To be honest I wasn't thinking of Juno (I wasn't aware of its
limitations) but of potential genuine situations where stream-ids are
the same for 2 devices. I know it can happen with PCI devices for
instance, although I am aware we don't have PCI passthrough yet. I don't
know if it is possible for it to happen with non-PCI devices but I
wouldn't be surprised if it can.