On 2023/12/19 14:49, Ethan Zhao wrote:
diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
index 74e8e4c17e81..182eb5df244d 100644
--- a/drivers/iommu/intel/pasid.c
+++ b/drivers/iommu/intel/pasid.c
@@ -476,6 +476,23 @@ devtlb_invalidation_with_pasid(struct
intel_iommu *iommu,
{
struct device_domain_info *info;
u16 sid, qdep, pfsid;
+ struct pci_dev *pdev;
+
+ pdev = to_pci_dev(dev);
+ if (!pdev)
+ return;
+
+ /*
+ * If endpoint device's link was brough down by user's pci
configuration
+ * access to it's hotplug capable slot link control register, as
sequence
+ * response for DLLSC, pciehp_ist() will set the device
error_state to
+ * pci_channel_io_perm_failure. Checking device's state here to
avoid
+ * issuing meaningless devTLB flush request to it, that might
cause lockup
+ * warning or deadlock because too long time waiting in
interrupt context.
+ */
+
+ if (pci_dev_is_disconnected(pdev))
+ return;
info = dev_iommu_priv_get(dev);
if (!info || !info->ats_enabled)
It's likely better to check the device status after verifying
ats_enabled. How about below change?
diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
index 74e8e4c17e81..fa19c6cdfd8b 100644
--- a/drivers/iommu/intel/pasid.c
+++ b/drivers/iommu/intel/pasid.c
@@ -481,6 +481,9 @@ devtlb_invalidation_with_pasid(struct intel_iommu
*iommu,
if (!info || !info->ats_enabled)
return;
+ if (pci_dev_is_disconnected(to_pci_dev(dev)))
I like this kind of simplicity, but rationalist always brings me back to
the no-error(ugly)
style. 🙂
The rational is that Intel IOMMU driver only supports PCI ATS. So if
device is marked as ATS supported, then it must be a PCI device.
Therefore, it's safe to convert it to pci_device with to_pci_dev().
If you move this up before ATS checking, then you need to check
dev_is_pci().
Best regards,
baolu
