[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20250623135718.Zc4hKa-u@linutronix.de>
Date: Mon, 23 Jun 2025 15:57:18 +0200
From: Nam Cao <namcao@...utronix.de>
To: Junxuan Liao <ljx@...wisc.edu>
Cc: linux-trace-kernel@...r.kernel.org, linux-kernel@...r.kernel.org,
x86@...nel.org, Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
Dave Hansen <dave.hansen@...ux.intel.com>,
"H. Peter Anvin" <hpa@...or.com>,
Steven Rostedt <rostedt@...dmis.org>,
Masami Hiramatsu <mhiramat@...nel.org>,
Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
"Paul E. McKenney" <paulmck@...nel.org>
Subject: Re: [PATCH v3 RESEND] x86/tracing: introduce enter/exit tracepoint
pairs for page faults
On Sun, Jun 15, 2025 at 07:32:00PM -0500, Junxuan Liao wrote:
> Merge page_fault_{user,kernel}, rename it page_fault_enter, and add
> page_fault_exit. This pair is useful for measuring page fault handling
> latencies.
>
> Add a new field to the merged tracepoints to indicate whether the page
> fault happened in userspace. We no longer need the static key associated,
> since it was used just to avoid checking user_mode when the tracepoints
> were disabled.
(I think) this breaks lttng which uses the old tracepoint names [1]. But
I'm not sure if we should worry about that.
Do you plan to add this to other architectures?
Perhaps it is better to implement the other approach suggested by Frederic
[2], using trace_user_exit/trace_user_enter? So that more architectures are
covered, and avoid breaking user tools?
(Btw, you would need to rebase this patch. The tracepoint has been moved,
and the static key has been deleted.)
Best regards,
Nam
[1] https://github.com/lttng/lttng-modules/blob/master/include/instrumentation/events/arch/x86/exceptions.h#L88C48-L88C63
[2] https://lore.kernel.org/lkml/Z_mO6_m0bV-Q8NEa@pavilion.home/#t
Powered by blists - more mailing lists