Date:   Thu, 9 Feb 2023 23:23:05 +0900
From:   Masami Hiramatsu (Google) <mhiramat@kernel.org>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     Steven Rostedt <rostedt@goodmis.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Linux Trace Kernel <linux-trace-kernel@vger.kernel.org>,
        Masami Hiramatsu <mhiramat@kernel.org>,
        Shuah Khan <shuah@kernel.org>,
        Shuah Khan <skhan@linuxfoundation.org>
Subject: Re: [PATCH] tracing/selftests: Ignore __pfx_ symbols in kprobe test
Message-Id: <20230209232305.672dd0249e0db41552b28763@kernel.org>
In-Reply-To: <Y+PkVD1+myADYns+@hirez.programming.kicks-ass.net>
References: <20230207135147.5ce618d6@gandalf.local.home>
        <20230207135402.38f73bb6@gandalf.local.home>
        <Y+PkVD1+myADYns+@hirez.programming.kicks-ass.net>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Wed, 8 Feb 2023 19:05:08 +0100
Peter Zijlstra <peterz@infradead.org> wrote:

> On Tue, Feb 07, 2023 at 01:54:02PM -0500, Steven Rostedt wrote:
> > On Tue, 7 Feb 2023 13:51:47 -0500
> > Steven Rostedt <rostedt@goodmis.org> wrote:
> > 
> > > From: "Steven Rostedt (Google)" <rostedt@goodmis.org>
> > > 
> > > The kprobe probepoint.tc test started failing because of the added __pfx_
> > > symbols that were added because of -fpatchable-function-entry=X,Y causing
> > > unwinders to see them as part of the previous functions. But kprobes can
> > > not be added on top of them. The selftest looks for tracefs_create_dir and
> > > picks it and the previous and following functions to add at their address.
> > > This caused it to include __pfx_tracefs_create_dir which is invalid to
> > > attach a kprobe to and caused the test to fail.
> > > 
> > > Fixes: 9f2899fe36a62 ("objtool: Add option to generate prefix symbols")
> > > Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
> > 
> > This is assuming that kprobes can not be added on top of these. But another
> > solution could be to have kprobes just pick the function the __pfx_ is for.
> > Would that be a better solution?
> 
> Simply refusing them is simplest. I don't see a compelling reason to
> make this complicated.

Yeah, and __pfx_ symbols has some "range", that means it is hard to translate
the probe address if user specify __pfx_*+offset.

BTW, currently kprobe event rejects this __pfx_ symbols because it is notrace
symbols, thus we can trace it if CONFIG_KPROBE_EVENTS_ON_NOTRACE=y.
But I guess it should not probe that place always because it should never
executed right?

Thank you,


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>