Re: [PATCH v10 07/11] tracing/probes: Add $args meta argument for all function args

From: Google
Date: Mon May 15 2023 - 11:11:52 EST


On Mon, 15 May 2023 07:57:01 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Sun, 14 May 2023 23:12:11 +0900
> "Masami Hiramatsu (Google)" <mhiramat@xxxxxxxxxx> wrote:
>
> > From: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> >
> > Add the '$args' meta fetch argument for function-entry probe events. This
> > will be expanded to the all arguments of the function and the tracepoint
> > using BTF function argument information.
> >
> > e.g.
> > # echo 'p vfs_read $args' >> dynamic_events
> > # echo 'f vfs_write $args' >> dynamic_events
> > # echo 't sched_overutilized_tp $args' >> dynamic_events
> > # cat dynamic_events
> > p:kprobes/p_vfs_read_0 vfs_read file=file buf=buf count=count pos=pos
> > f:fprobes/vfs_write__entry vfs_write file=file buf=buf count=count pos=pos
> > t:tracepoints/sched_overutilized_tp sched_overutilized_tp rd=rd overutilized=overutilized
> >
> > NOTE: This is not like other $-vars, you can not use this $args as a
> > part of fetch args, e.g. specifying name "foo=$args" and using it in
> > dereferences "+0($args)" will lead a parse error.
> >
> > Signed-off-by: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> > ---
> > Changes in v10:
> > - Change $$args to $args so that user can use $$ for current task's pid.
>
> I hate coming up with new apis, because you never know if what you pick is
> correct ;-) And then you are stuck with whatever you decided on. :-/
>
> I know I suggested $args, but since it is special, should we call it
> "$arg*" ?
>
> That way it follows bash wildcard semantics?
>
> # echo 'p vfs_read $arg*' >> dynamic_events
>
> I think that is more along the lines of what people would expect.
>
> What do you think?

Good idea!

BTW, user will expect that $arg* will be expanded to available "$argN".
But this $args does different thing at this point, it is expanded
to something equivalent to "<BTF-name>=$argN:<BTF-type>".
So, to give more consistency, I need one more step: when user gives
only "$argN" on BTF supported kernel, it will be translated to
"<BTF-name>=$argN:<BTF-type>". Then, I think it is natural to use '$arg*'.

Thank you,

>
> -- Steve
>
>
> > Changes in v6:
> > - update patch description.
> > ---
> > kernel/trace/trace_fprobe.c | 21 ++++++++-
> > kernel/trace/trace_kprobe.c | 23 ++++++++--
> > kernel/trace/trace_probe.c | 98 +++++++++++++++++++++++++++++++++++++++++++
> > kernel/trace/trace_probe.h | 10 ++++
> > 4 files changed, 144 insertions(+), 8 deletions(-)
> >
> >


--
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>