Re: [PATCH] tracing: Add trace_trigger kernel command line option

[Steven Rostedt]

On Thu, 20 Oct 2022 23:33:18 +0900
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx> wrote:

> On Wed, 19 Oct 2022 20:01:37 -0400
> Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> 
> > From: "Steven Rostedt (Google)" <rostedt@xxxxxxxxxxx>
> > 
> > Allow triggers to be enabled at kernel boot up. For example:
> > 
> >   trace_trigger="sched_switch.stacktrace if prev_state == 2"
> > 
> > The above will enable the stacktrace trigger on top of the sched_switch
> > event and only trigger if its prev_state is 2 (TASK_UNINTERRUPTIBLE). Then
> > at boot up, a stacktrace will trigger and be recorded in the tracing ring
> > buffer every time the sched_switch happens where the previous state is
> > TASK_INTERRUPTIBLE.
> > 
> > As this calls into tracepoint logic during very early boot (before
> > interrupts are enabled), a check has to be done to see if early boot
> > interrupts are still disabled, and if so, avoid any call to RCU
> > synchronization, as that will enable interrupts and cause boot up issues.  
> 
> Just out of curiousity, can you do it by boot-time tracer?
> (Is it too late for your issue?)

Yeah, I'm looking at adding triggers very early. I could even add
"traceoff" trigger too here. I'm guessing that bootconf is done later in
boot up? Especially since I needed to fix RCU calls to get this working.

> 
> $ cat >> stacktrace.bconf
> ftrace.event.sched.sched_switch.actions = "stacktrace if prev_state == 2"
> ^D
> $ bootconfig -a stacktrace.bconf initrd.img
> 

Not to mention, I'm doing this for Chromebooks where it's easy to update
the command line (on dev devices), but not as easy to modify the kernel.

-- Steve