Re: [PATCH v3 02/11] iommu/arm-smmu: Introduce iommu_group notifier block

[Will Deacon <will.deacon@arm.com>]

On Wed, Jan 22, 2014 at 01:14:13PM +0000, Varun Sethi wrote:
> > On Tue, Jan 21, 2014 at 05:48:02PM +0000, Varun Sethi wrote:
> > > > +static int arm_smmu_group_notifier(struct notifier_block *nb,
> > > > +				unsigned long action, void *data) {
> > > > +	struct device *dev = data;
> > > > +	struct dma_iommu_mapping *mapping;
> > > > +	struct arm_smmu_device *smmu;
> > > > +	int ret;
> > > > +
> > > > +	switch (action) {
> > > > +	case IOMMU_GROUP_NOTIFY_BIND_DRIVER:
> > > > +
> > > > +		smmu = dev->archdata.iommu;
> > > > +		if (!smmu || !(smmu->options & ARM_SMMU_OPT_ISOLATE_DEVICES))
> > > > +			break;
> > > [Sethi Varun-B16395] Should this check be really done here? The
> > > "Isolate devices" property would allow us to set up iommu groups. My
> > > understanding is that if we specify the isolate devices property, then
> > > each device would have a separate iommu group otherwise all devices
> > > connected to the SMMU would share the iommu group.
> > 
> > That's not what currently happens (at least, in the patch I have queued
> > for groups). The code queued adds each device to its own group in
> > arm_smmu_add_device, which I think is the right thing to do.
> > 
> > > With that logic, we should link the mapping to the iommu group.
> > 
> > Ok, so are you suggesting that we perform the isolation mapping in
> > arm_smmu_add_device and drop the notifier altogether?
> I think that should be fine, until we want to delay mapping creation till
> driver bind time.

Is there a hard dependency on that?

> But the "isolate device" property should dictate iommu group creation.

The reason we added automatic group creation (per-device) is for VFIO, which
expects all devices to be in a group regardless of the device isolation
configuration.

Will