kpatch load/unload patch transition has stalled

[oceansue]

Description

I have encountered an issue with kpatch where it experiences stalling or timeouts when performing load/unload operations on modules,。

Upon reviewing the kpatch code, I noticed that the method used to check the system interface at /sys/kernel/livepatch/livepatch_diff/transition involved a sleep 1 interval, which seemed to be inefficient or unsuitable for my current working environment. This method appeared to contribute to the stalling or timeouts observed.

To address this, I modified the method in kpatch that checks this interface. My changes have resulted in a significant improvement in the efficiency of module load/unload operations. I believe this modification could be beneficial for others facing similar issues, especially on new architectures.

Environment

• OS: [kylin os]
• Kernel：4.19 and 5.10
• kpatch version: 6115a1a

Additional Information

• I have created a patch that addresses this issue and submitted it to the repository.
• I propose that the patch I submitted be reviewed and, if found suitable, integrated into the main branch to resolve the stalling/timeout issue.
• The patch file is WIP:kpatch/kpatch: Optimize patch transition timing with a while loop #1408

Attachments

Before Applying the Patch

• load module

[root@localhost kpatch]# ./kpatch load ../kpatch-build/livepatch-diff.ko
loading patch module: ../kpatch-build/livepatch-diff.ko
waiting (up to 15 seconds) for patch transition to complete...
patch transition has stalled!
signaling stalled process(es):
waiting (up to 60 seconds) for patch transition to complete...

Stalled processes:
module livepatch_diff did not complete its transition, unloading...
disabling patch module: livepatch_diff
waiting (up to 15 seconds) for patch transition to complete...
transition complete (2 seconds)
unloading patch module: livepatch_diff
kpatch: error: failed to load module livepatch_diff (transition stalled)
[root@localhost kpatch]# 

• unload module

[root@localhost kpatch]# ./kpatch unload ../kpatch-build/livepatch-diff.ko
disabling patch module: livepatch_diff
waiting (up to 15 seconds) for patch transition to complete...
patch transition has stalled!
signaling stalled process(es):
waiting (up to 60 seconds) for patch transition to complete...
kpatch: transition stalled for livepatch_diff
[root@localhost kpatch]# 

By observing the dmesg logs, it can be seen that the module transition time is over 1500 seconds, of course, this time is not fixed, and sometimes it might complete the load/unload in just a few seconds with good luck.

After Applying the Patch

• load module

[root@localhost kpatch]# ./kpatch load ../kpatch-build/livepatch-diff.ko
loading patch module: ../kpatch-build/livepatch-diff.ko
waiting (up to 15 seconds) for patch transition to complete...
transition complete (14 seconds)

• unload module

[root@localhost kpatch]# ./kpatch unload ../kpatch-build/livepatch-diff.ko
disabling patch module: livepatch_diff
waiting (up to 15 seconds) for patch transition to complete...
transition complete (1 seconds)
unloading patch module: livepatch_diff
[root@localhost kpatch]# 

other

If you need me to provide any additional information, please contact me. I would also appreciate it if you could offer me some better solutions.
/cc @oceansue

[github-actions]

This issue has been open for 30 days with no activity and no assignee. It will be closed in 7 days unless a comment is added.