Comment 20 for bug 1626436

My problem is probably the same that Doug Smythies described in bug 1626564 . I used the "echo workqueue:workqueue_queue_work > /sys/kernel/debug/tracing/set_event" method and had about 1000 work items enqueued by "memcg_kmem_cache_create_func" all at pretty much the same time, which correlates nicely with the 1000 kworker processes I see after boot.

If Martin has the same issue with lots of kworkers (I'm still not quite sure?), this might be the same issue.

So I suspect the bad commit might be:
81ae6d03952c1bfb96e1a716809bd65e7cd14360
"mm/slub.c: replace kick_all_cpus_sync() with synchronize_sched() in kmem_cache_shrink()"
according to https://bugzilla.kernel.org/show_bug.cgi?id=172991

This commit was released in Linux 4.7.

This patch might fix the issue, but I haven't tested it:
https://patchwork.kernel.org/patch/9359271/