tracing: Fix race where histograms can be called before the event
authorSteven Rostedt (Google) <rostedt@goodmis.org>
Wed, 23 Nov 2022 21:43:23 +0000 (16:43 -0500)
committerSteven Rostedt (Google) <rostedt@goodmis.org>
Thu, 24 Nov 2022 00:05:50 +0000 (19:05 -0500)
commitef38c79a522b660f7f71d45dad2d6244bc741841
tree74676f7bf6dfd660112239b0cf31e77486bd6715
parent022632f6c43a86f2135642dccd5686de318e861d
tracing: Fix race where histograms can be called before the event

commit 94eedf3dded5 ("tracing: Fix race where eprobes can be called before
the event") fixed an issue where if an event is soft disabled, and the
trigger is being added, there's a small window where the event sees that
there's a trigger but does not see that it requires reading the event yet,
and then calls the trigger with the record == NULL.

This could be solved with adding memory barriers in the hot path, or to
make sure that all the triggers requiring a record check for NULL. The
latter was chosen.

Commit 94eedf3dded5 set the eprobe trigger handle to check for NULL, but
the same needs to be done with histograms.

Link: https://lore.kernel.org/linux-trace-kernel/20221118211809.701d40c0f8a757b0df3c025a@kernel.org/
Link: https://lore.kernel.org/linux-trace-kernel/20221123164323.03450c3a@gandalf.local.home
Cc: Tom Zanussi <zanussi@kernel.org>
Cc: stable@vger.kernel.org
Fixes: 7491e2c442781 ("tracing: Add a probe that attaches to trace events")
Reported-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
kernel/trace/trace_events_hist.c