On Fri, Oct 8, 2021 at 4:37 PM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote: > > On Fri, 8 Oct 2021 15:42:31 -0700 > Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote: > > > orgot to actually cc Steven, oops. Steven, I've run into the problem > > when running a few selftests that do uprobe/kprobe attachment. At some > > point, they started complaining that files like > > /sys/kernel/debug/tracing/events/syscalls/sys_enter_nanosleep/id don't > > exist. And this condition persisted. When I checked > > /sys/kernel/debug/tracing in QEMU, it was empty. Is this a known > > problem? > > The tracefs directory should automatically be mounted in the debugfs > "tracing" directory when debugfs is mounted. Yeah, we do that for our testing environment and it works reliably. This time we started running a bunch of tests completely in parallel, and in one case I ran into this situation that caused "permanent damage". Unfortunately I haven't checked if debugfs is still mounted or not. Yucong mentioned that one of the selftests (tc_redirect.c) is manipulating /sys mount, so I wonder if this could have caused some consequences like this... Jussi, any idea? > > Does /sys/kernel/tracing exist? Yeah, it was there, but was empty. > > -- Steve
