iommu/arm-smmu-v3: Fail aliasing StreamIDs more gracefully

We've never supported StreamID aliasing between devices, and as such
they will never have had functioning DMA, but this is not fatal to the
SMMU itself. Although aliasing between hard-wired platform device
StreamIDs would tend to raise questions about the whole system, in
practice it's far more likely to occur relatively innocently due to
legacy PCI bridges, where the underlying StreamID mappings are still
perfectly reasonable.

As such, return a more benign -ENODEV when failing probe for such an
unsupported device (and log a more obvious error message), so that it
doesn't break the entire SMMU probe now that bus_iommu_probe() runs in
the right order and can propagate that error back. The end result is
still that the device doesn't get an IOMMU group and probably won't
work, same as before.

Signed-off-by: Robin Murphy <[email protected]>
Link: https://lore.kernel.org/r/39d54e49c8476efc4653e352150d44b185d6d50f.1744380554.git.robin.murphy@arm.com
Signed-off-by: Will Deacon <[email protected]>

Author: rmurphy-arm

drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c

@@ -3411,9 +3411,9 @@ static int arm_smmu_insert_master(struct arm_smmu_device *smmu,
 				continue;
 
 			dev_warn(master->dev,
-				 "stream %u already in tree from dev %s\n", sid,
-				 dev_name(existing_master->dev));
-			ret = -EINVAL;
+				 "Aliasing StreamID 0x%x (from %s) unsupported, expect DMA to be broken\n",
+				 sid, dev_name(existing_master->dev));
+			ret = -ENODEV;
 			break;
 		}
 	}