Re: [PATCH 1/2] ARM: SMMU: add devices attached to the SMMU to anIOMMU group

From: Will Deacon
Date: Mon Oct 14 2013 - 13:08:12 EST


On Mon, Oct 14, 2013 at 04:13:15PM +0100, Antonios Motakis wrote:
> On Mon, Oct 14, 2013 at 2:48 PM, Will Deacon <will.deacon@xxxxxxx> wrote:
> > > diff --git a/drivers/iommu/arm-smmu.c b/drivers/iommu/arm-smmu.c
> > > index 0f45a48..8b71332 100644
> > > --- a/drivers/iommu/arm-smmu.c
> > > +++ b/drivers/iommu/arm-smmu.c
> > > @@ -1502,6 +1502,8 @@ static int arm_smmu_add_device(struct device *dev)
> > > {
> > > struct arm_smmu_device *child, *parent, *smmu;
> > > struct arm_smmu_master *master = NULL;
> > > + struct iommu_group *group;
> > > + int ret;
> > >
> > > spin_lock(&arm_smmu_devices_lock);
> > > list_for_each_entry(parent, &arm_smmu_devices, list) {
> > > @@ -1534,13 +1536,27 @@ static int arm_smmu_add_device(struct device *dev)
> > > if (!master)
> > > return -ENODEV;
> > >
> > > + group = iommu_group_get(dev);
> >
> > I'm not especially familiar with IOMMU groups (I understand them as the
> > minimum translation granularity, which would mean single StreamID for the
> > ARM SMMU), but under what circumstances would you expect to receive a
> > non-NULL group here? I can't see any other code adding devices to groups
> > (outside of other drivers)...
> >
>
> You are right, only other IOMMU drivers will add a device to a group.
> There was a discussion about this when I posted a similar patch for
> the Exynos System MMU driver, see
> http://lists.infradead.org/pipermail/linux-arm-kernel/2013-July/185675.html
>
> The idea is to check in the case of add_device() being called multiple
> times, which is not the case most of the time, but still a sane
> safeguard.

Ok, but it feels a bit weird. The current code (arm_smmu_add_device)
basically does a bunch of sanity checking against the DT data in order to
find where the master sits in the device topology. Then it updates
dev->archdata.iommu to point at the relevant SMMU instance.

So, the interesting case is where the device was previously associated with
a *different* IOMMU. In that case, the current code clobbers the iommu field
with the new smmu, whereas the new code could end up getting very confused
with respect to IOMMU groups.

A better way is probably to check that dev->archdata.iommu is NULL before we
assign to it. If not, then spit out a warning and return an error. That
would also mean you could get rid of the group get/put calls.

Will
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/