Public bug reported:

Description

This PR backports three upstream fixes for tegra241-cmdqv teardown. They move 
SMMU teardown to devm, stop VINTF deinit from issuing a CMD_SYNC on a freed 
command queue during probe-failure unwind and vIOMMU destroy, and quiesce the 
VINTFs on shutdown and kexec.
Commits

    7d8c870b84bb iommu/arm-smmu-v3: Manage teardown with devm (upstream 
2bd22a0d4050)
    3f303b85dd4d iommu/tegra241-cmdqv: Fix CMD_SYNC use-after-free on teardown 
(upstream 9ff145a25c5c)
    a19853cb9e2f iommu/arm-smmu-v3: Disable the impl before disabling the SMMU 
on shutdown (upstream ab0dbf411ace)

Testing

Tested on a Vera host with CMDQV enabled

    Boot and DMA: the SMMUs and CMDQVs probe cleanly. NVMe fio wrote and 
verified 32 GiB, and NIC transfers arrived intact, in both passthrough and 
translated (iommu.passthrough=0) modes.
    Host reboot and kexec: kexec plus 3 reboots, including a shutdown with 
translated domains active. No oops, CMD_SYNC timeouts or CMDQV errors on 
serial, and a clean probe every time.
    vIOMMU destroy: QEMU with iommufd, arm-smmuv3,accel=on,cmdqv=on and GPU 
passthrough. After both a clean poweroff and a kill -9, the host showed no 
errors and the VM restarted cleanly.
    Guest reboot and kexec: the same kernel running in the vCMDQ VM survived 3 
guest reboots and a guest kexec into translated mode. The guest console showed 
no CMD_SYNC timeouts or CMDQV errors, and the guest's CMDQV and VCMDQs probed 
cleanly every time.
    Probe-failure unwind: forcing the VCMDQ enable to fail with a debug patch 
(not included) gives probe ... failed with error -110 and no oops.

** Affects: linux-nvidia (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/2173553

Title:
  NVIDIA: Backport iommu/smmuv3 fixes for cmdqv teardown

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-nvidia/+bug/2173553/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to