Hi Niklas,
On Mon, Oct 13, 2025 at 11:53:55AM +0200, Niklas Schnelle wrote:
On Sun, 2025-10-12 at 17:04 -0700, Nicolin Chen wrote:
quoted
Add a new test_dev domain op for driver to test the compatibility between
a domain and a device at the driver level, before calling into the actual
attachment/replacement of a domain. Support pasid for set_dev_pasid call.
Move existing core-level compatibility tests to a helper function. Invoke
it prior to:
* __iommu_attach_device() or its wrapper __iommu_device_set_domain()
* __iommu_set_group_pasid()
Should this list also include iommu_deferred_attach()? The code does
include it.
iommu_deferred_attach() invokes __iommu_attach_device(), so it is
already included in the list :)
quoted
/**
* struct iommu_domain_ops - domain specific operations
- * @attach_dev: attach an iommu domain to a device
+ * @test_dev: Test compatibility prior to an @attach_dev or @set_dev_pasid call.
+ * A driver-level callback of this op should do a thorough sanity, to
You're missing the word "check" above.
Ack.
quoted
+ * make sure a device is compatible with the domain. So the following
+ * @attach_dev and @set_dev_pasid functions would likely succeed with
+ * only one exception due to a temporary failure like out of memory.
Nit: "… only one exception …" / "… like out of memory …" this sounds a
bit odd to me because on the one hand it's one exception but then also
a group (temporary failures).
Maybe better:
"… would likely succeed with only the exception of temporary failures
like out of memory."?
Sure. I can do that. Fixing both parts, it would be:
* @test_dev: Test compatibility prior to an @attach_dev or @set_dev_pasid call.
* A driver callback of this op should do a thorough sanity check, to
* make sure a device is compatible with the domain, so the following
* @attach_dev and @set_dev_pasid functions would likely succeed with
* only the exception of temporary failures like out of memory.
quoted hunk ↗ jump to hunk
quoted
@@ -3615,6 +3657,11 @@ int iommu_replace_device_pasid(struct iommu_domain *domain,
ret = 0;
if (curr_domain != domain) {
+ ret = __iommu_domain_test_device(domain, dev, pasid,
+ curr_domain);
+ if (ret)
+ goto out_unlock;
+
ret = __iommu_set_group_pasid(domain, group,
pasid, curr_domain);
if (ret)
Apart from the comment and commit description nits mentioned above this
looks good to me.
Reviewed-by: Niklas Schnelle <schnelle@linux.ibm.com>
Thanks for the review!
Nicolin