Bug 216321

Summary:	cpuset: can cpuset_hotplug_workfn() be skipped during suspend/resume
Product:	Process Management	Reporter:	Rick Yiu (rickyiu)
Component:	Other	Assignee:	process_other
Status:	NEW ---
Severity:	normal	CC:	rickyiu, tj, wvw
Priority:	P1
Hardware:	All
OS:	Linux
Kernel Version:	5.10.134	Subsystem:
Regression:	No	Bisected commit-id:

Description Rick Yiu 2022-08-03 09:06:13 UTC

Hi,

We are checking suspend/resume performance on our platform. We found that when device resumes, it took 10+ ms in the function cpuset_hotplug_workfn(), and to be more specific, the time is spending in the function rebuild_root_domains() mostly.

From the comment of cpuset_hotplug_workfn(), it seems like there is no necessary to do rebuild_root_domains() or even cpuset_hotplug_workfn() itself in suspend/resume path. I tried to bypass the functions for suspend/resume, and I didn't see problem so far.

Because I am not an expert of this part, I would like to know if there is any risk to skip either function when suspend/resume. Thanks.



/*
...
 * Note that CPU offlining during suspend is ignored.  We don't modify
 * cpusets across suspend/resume cycles at all.
 */
void cpuset_hotplug_workfn(struct work_struct *work)