Public bug reported:
Backports the upstream 8-patch series **"iommu: Fix
`pci_dev_reset_iommu_prepare/done()`"** (v8) onto `26.04_linux-nvidia-
bos` (NV kernel 7.0). This fixes a failure where a **CXL bus reset**
(`cxl_bus` reset method) aborts with `EBUSY` and emits a WARN from
`pci_dev_reset_iommu_prepare()`.
Root cause:
cxl_reset_bus_function() calls pci_reset_bus_function() internally, and
both call pci_dev_reset_iommu_prepare/done(), producing a nested (re-
entrant) call:
cxl_reset_bus_function():
pci_dev_reset_iommu_prepare(); // outer
pci_reset_bus_function():
pci_dev_reset_iommu_prepare(); // inner (re-entry)
...
pci_dev_reset_iommu_done(); // inner
pci_dev_reset_iommu_done(); // outer
pci_dev_reset_iommu_prepare() did not allow re-entry, so the inner call warns
and returns -EBUSY. The same nesting exists across most of
pci_dev_specific_reset() (which call pcie_flr()). This is the exact issue
described upstream:
Original posting:
https://lore.kernel.org/all/[email protected]/
Merged v8 cover:
https://lore.kernel.org/all/[email protected]/
The reworked series makes the prepare/done bookkeeping per-group_device
and re-entrant, and fixes several related corner cases (nested
callbacks, multi-device groups, PASID attach, ATS invalidation timeouts,
and a UAF/WARN on concurrent detach).
** Affects: linux-nvidia-bos (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2159852
Title:
Backport iommu: Fix pci_dev_reset_iommu_prepare/done()
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-nvidia-bos/+bug/2159852/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
