On Thu, 26 May 2022 12:48:41 -0700 Yonghong Song <yhs@xxxxxx> wrote: > So I suspect there is a race condition between kretprobe_dispatcher() > (or higher level kretprobe_trampoline_handler()) and > unregister_kretprobes(). I looked at kernel/trace code and had not > found an obvious race yet. I will continue to check. > But at the same time, I would like to seek some expert advice to see > whether you are aware of any potential issues in 5.12 or not and where > are possible places I should focus on to add debug codes for experiments. First thing I'll ask is if you can reproduce this on a vanilla 5.12 kernel. If not, then we can't help you. If you can, I would ask if it is reproducible on the latest mainline kernel. If it is, then we'll help you look into it. If not, we'll ask if you can try to bisect it to at least find what version it was fixed in. -- Steve
