From: Joel Fernandes <joel@joelfernandes.org>
To: Steven Rostedt <rostedt@goodmis.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
Alexei Starovoitov <alexei.starovoitov@gmail.com>,
Peter Zijlstra <peterz@infradead.org>,
linux-kernel <linux-kernel@vger.kernel.org>,
linux-arch <linux-arch@vger.kernel.org>,
Ingo Molnar <mingo@kernel.org>,
Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
"Gustavo A. R. Silva" <gustavo@embeddedor.com>,
Thomas Gleixner <tglx@linutronix.de>,
paulmck <paulmck@kernel.org>,
Josh Triplett <josh@joshtriplett.org>,
Lai Jiangshan <jiangshanlai@gmail.com>,
Andy Lutomirski <luto@kernel.org>,
Tony Luck <tony.luck@intel.com>,
Frederic Weisbecker <frederic@kernel.org>,
dan carpenter <dan.carpenter@oracle.com>,
Masami Hiramatsu <mhiramat@kernel.org>
Subject: Re: [PATCH v4 16/27] tracing: Remove regular RCU context for _rcuidle tracepoints (again)
Date: Fri, 6 Mar 2020 13:45:38 -0500 [thread overview]
Message-ID: <20200306184538.GA92717@google.com> (raw)
In-Reply-To: <20200306125500.6aa75c0d@gandalf.local.home>
On Fri, Mar 06, 2020 at 12:55:00PM -0500, Steven Rostedt wrote:
> On Fri, 6 Mar 2020 11:04:28 -0500 (EST)
> Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
>
> > If we care about not adding those extra branches on the fast-path, there is
> > an alternative way to do things: BPF could provide two distinct probe callbacks,
> > one meant for rcuidle tracepoints (which would have the trace_rcu_enter/exit), and
> > the other for the for 99% of the other callsites which have RCU watching.
> >
> > I would recommend performing benchmarks justifying the choice of one approach over
> > the other though.
>
> I just whipped this up (haven't even tried to compile it), but this should
> satisfy everyone. Those that register a callback that needs RCU protection
> simply registers with one of the _rcu versions, and all will be done. And
> since DO_TRACE is a macro, and rcuidle is a constant, the rcu protection
> code will be compiled out for locations that it is not needed.
>
> With this, perf doesn't even need to do anything extra but register with
> the "_rcu" version.
Looks nice! Some comments below:
> -- Steve
>
> diff --git a/include/linux/tracepoint-defs.h b/include/linux/tracepoint-defs.h
> index b29950a19205..582dece30170 100644
> --- a/include/linux/tracepoint-defs.h
> +++ b/include/linux/tracepoint-defs.h
> @@ -25,6 +25,7 @@ struct tracepoint_func {
> void *func;
> void *data;
> int prio;
> + int requires_rcu;
> };
>
> struct tracepoint {
> diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h
> index 1fb11daa5c53..5f4de82ffa0f 100644
> --- a/include/linux/tracepoint.h
> +++ b/include/linux/tracepoint.h
> @@ -179,25 +179,28 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
> * For rcuidle callers, use srcu since sched-rcu \
> * doesn't work from the idle path. \
> */ \
> - if (rcuidle) { \
> + if (rcuidle) \
> __idx = srcu_read_lock_notrace(&tracepoint_srcu);\
Small addition:
To prevent confusion, we could make more clear that SRCU here is just to
protect the tracepoint function table and not the callbacks themselves.
> - rcu_irq_enter_irqson(); \
> - } \
> \
> it_func_ptr = rcu_dereference_raw((tp)->funcs); \
> \
> if (it_func_ptr) { \
> do { \
> + int rcu_flags; \
> it_func = (it_func_ptr)->func; \
> + if (rcuidle && \
> + (it_func_ptr)->requires_rcu) \
> + rcu_flags = trace_rcu_enter(); \
> __data = (it_func_ptr)->data; \
> ((void(*)(proto))(it_func))(args); \
> + if (rcuidle && \
> + (it_func_ptr)->requires_rcu) \
> + trace_rcu_exit(rcu_flags); \
Nit: If we have incurred the cost of trace_rcu_enter() once, we can call
it only once and then call trace_rcu_exit() after the do-while loop. That way
we pay the price only once.
thanks,
- Joel
> } while ((++it_func_ptr)->func); \
> } \
> \
> - if (rcuidle) { \
> + if (rcuidle) \
> rcu_irq_exit_irqson(); \
> - srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\
> - } \
> \
> preempt_enable_notrace(); \
> } while (0)
> diff --git a/kernel/tracepoint.c b/kernel/tracepoint.c
> index 73956eaff8a9..1797e20fd471 100644
> --- a/kernel/tracepoint.c
> +++ b/kernel/tracepoint.c
> @@ -295,6 +295,7 @@ static int tracepoint_remove_func(struct tracepoint *tp,
> * @probe: probe handler
> * @data: tracepoint data
> * @prio: priority of this function over other registered functions
> + * @rcu: set to non zero if the callback requires RCU protection
> *
> * Returns 0 if ok, error value on error.
> * Note: if @tp is within a module, the caller is responsible for
> @@ -302,8 +303,8 @@ static int tracepoint_remove_func(struct tracepoint *tp,
> * performed either with a tracepoint module going notifier, or from
> * within module exit functions.
> */
> -int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
> - void *data, int prio)
> +int tracepoint_probe_register_prio_rcu(struct tracepoint *tp, void *probe,
> + void *data, int prio, int rcu)
> {
> struct tracepoint_func tp_func;
> int ret;
> @@ -312,12 +313,52 @@ int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
> tp_func.func = probe;
> tp_func.data = data;
> tp_func.prio = prio;
> + tp_func.requires_rcu = rcu;
> ret = tracepoint_add_func(tp, &tp_func, prio);
> mutex_unlock(&tracepoints_mutex);
> return ret;
> }
> +EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_rcu);
> +
> +/**
> + * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
> + * @tp: tracepoint
> + * @probe: probe handler
> + * @data: tracepoint data
> + * @prio: priority of this function over other registered functions
> + *
> + * Returns 0 if ok, error value on error.
> + * Note: if @tp is within a module, the caller is responsible for
> + * unregistering the probe before the module is gone. This can be
> + * performed either with a tracepoint module going notifier, or from
> + * within module exit functions.
> + */
> +int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
> + void *data, int prio)
> +{
> + return tracepoint_probe_register_prio_rcu(tp, probe, data, prio, 0);
> +}
> EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
>
> +/**
> + * tracepoint_probe_register_rcu - Connect a probe to a tracepoint
> + * @tp: tracepoint
> + * @probe: probe handler
> + * @data: tracepoint data
> + *
> + * Returns 0 if ok, error value on error.
> + * Note: if @tp is within a module, the caller is responsible for
> + * unregistering the probe before the module is gone. This can be
> + * performed either with a tracepoint module going notifier, or from
> + * within module exit functions.
> + */
> +int tracepoint_probe_register_rcu(struct tracepoint *tp, void *probe, void *data)
> +{
> + return tracepoint_probe_register_prio_rcu(tp, probe, data,
> + TRACEPOINT_DEFAULT_PRIO, 1);
> +}
> +EXPORT_SYMBOL_GPL(tracepoint_probe_register_rcu);
> +
> /**
> * tracepoint_probe_register - Connect a probe to a tracepoint
> * @tp: tracepoint
> @@ -332,7 +373,8 @@ EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
> */
> int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
> {
> - return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
> + return tracepoint_probe_register_prio_rcu(tp, probe, data,
> + TRACEPOINT_DEFAULT_PRIO, 0);
> }
> EXPORT_SYMBOL_GPL(tracepoint_probe_register);
>
next prev parent reply other threads:[~2020-03-06 18:45 UTC|newest]
Thread overview: 84+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-02-21 13:34 [PATCH v4 00/27] tracing vs world Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 01/27] lockdep: Teach lockdep about "USED" <- "IN-NMI" inversions Peter Zijlstra
2020-02-21 15:08 ` Steven Rostedt
2020-02-21 20:25 ` Peter Zijlstra
2020-02-21 20:28 ` Steven Rostedt
2020-02-21 22:01 ` Frederic Weisbecker
2020-02-22 3:08 ` Joel Fernandes
2020-02-24 10:10 ` Peter Zijlstra
2020-02-25 2:12 ` Joel Fernandes
2020-03-20 12:58 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 02/27] hardirq/nmi: Allow nested nmi_enter() Peter Zijlstra
2020-02-21 22:21 ` Frederic Weisbecker
2020-02-24 12:13 ` Petr Mladek
2020-02-25 1:30 ` Frederic Weisbecker
2020-02-24 16:13 ` Peter Zijlstra
2020-02-25 3:09 ` Frederic Weisbecker
2020-02-25 15:41 ` Peter Zijlstra
2020-02-25 16:21 ` Frederic Weisbecker
2020-02-25 22:10 ` Frederic Weisbecker
2020-02-27 9:10 ` Peter Zijlstra
2020-02-27 13:34 ` Frederic Weisbecker
2020-02-21 13:34 ` [PATCH v4 03/27] x86/entry: Flip _TIF_SIGPENDING and _TIF_NOTIFY_RESUME handling Peter Zijlstra
2020-02-21 16:14 ` Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 04/27] x86/mce: Delete ist_begin_non_atomic() Peter Zijlstra
2020-02-21 19:07 ` Andy Lutomirski
2020-02-21 23:40 ` Frederic Weisbecker
2020-02-21 13:34 ` [PATCH v4 05/27] x86: Replace ist_enter() with nmi_enter() Peter Zijlstra
2020-02-21 19:05 ` Andy Lutomirski
2020-02-21 20:22 ` Peter Zijlstra
2020-02-24 10:43 ` Peter Zijlstra
2020-02-24 16:27 ` Steven Rostedt
2020-02-24 16:34 ` Peter Zijlstra
2020-02-24 16:47 ` Steven Rostedt
2020-02-24 21:31 ` Peter Zijlstra
2020-02-24 22:02 ` Steven Rostedt
2020-02-26 10:25 ` Peter Zijlstra
2020-02-26 13:16 ` Peter Zijlstra
2020-02-26 10:27 ` Peter Zijlstra
2020-02-26 15:20 ` Steven Rostedt
2020-03-07 1:53 ` Masami Hiramatsu
2020-02-21 13:34 ` [PATCH v4 06/27] x86/doublefault: Remove memmove() call Peter Zijlstra
2020-02-21 19:10 ` Andy Lutomirski
2020-02-21 13:34 ` [PATCH v4 07/27] rcu: Make RCU IRQ enter/exit functions rely on in_nmi() Peter Zijlstra
2020-02-26 0:23 ` Frederic Weisbecker
2020-02-21 13:34 ` [PATCH v4 08/27] rcu/kprobes: Comment why rcu_nmi_enter() is marked NOKPROBE Peter Zijlstra
2020-02-26 0:27 ` Frederic Weisbecker
2020-02-21 13:34 ` [PATCH v4 09/27] rcu: Rename rcu_irq_{enter,exit}_irqson() Peter Zijlstra
2020-02-21 20:21 ` Steven Rostedt
2020-02-24 10:24 ` Peter Zijlstra
2020-02-26 0:35 ` Frederic Weisbecker
2020-02-21 13:34 ` [PATCH v4 10/27] rcu: Mark rcu_dynticks_curr_cpu_in_eqs() inline Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 11/27] rcu,tracing: Create trace_rcu_{enter,exit}() Peter Zijlstra
2020-03-06 11:50 ` Peter Zijlstra
2020-03-06 12:40 ` Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 12/27] sched,rcu,tracing: Avoid tracing before in_nmi() is correct Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 13/27] x86,tracing: Add comments to do_nmi() Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 14/27] perf,tracing: Prepare the perf-trace interface for RCU changes Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 15/27] tracing: Employ trace_rcu_{enter,exit}() Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 16/27] tracing: Remove regular RCU context for _rcuidle tracepoints (again) Peter Zijlstra
2020-03-06 10:43 ` Peter Zijlstra
2020-03-06 11:31 ` Peter Zijlstra
2020-03-06 15:51 ` Alexei Starovoitov
2020-03-06 16:04 ` Mathieu Desnoyers
2020-03-06 17:55 ` Steven Rostedt
2020-03-06 18:45 ` Joel Fernandes [this message]
2020-03-06 18:59 ` Steven Rostedt
2020-03-06 19:14 ` Joel Fernandes
2020-03-06 20:22 ` Mathieu Desnoyers
2020-03-06 20:45 ` Steven Rostedt
2020-03-06 20:55 ` Mathieu Desnoyers
2020-03-06 21:06 ` Steven Rostedt
2020-03-06 23:10 ` Alexei Starovoitov
2020-03-06 17:21 ` Joel Fernandes
2020-02-21 13:34 ` [PATCH v4 17/27] perf,tracing: Allow function tracing when !RCU Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 18/27] x86/int3: Ensure that poke_int3_handler() is not traced Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 19/27] locking/atomics, kcsan: Add KCSAN instrumentation Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 20/27] asm-generic/atomic: Use __always_inline for pure wrappers Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 21/27] asm-generic/atomic: Use __always_inline for fallback wrappers Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 22/27] compiler: Simple READ/WRITE_ONCE() implementations Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 23/27] locking/atomics: Flip fallbacks and instrumentation Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 24/27] x86/int3: Avoid atomic instrumentation Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 25/27] lib/bsearch: Provide __always_inline variant Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 26/27] x86/int3: Inline bsearch() Peter Zijlstra
2020-02-21 13:34 ` [PATCH v4 27/27] x86/int3: Ensure that poke_int3_handler() is not sanitized Peter Zijlstra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20200306184538.GA92717@google.com \
--to=joel@joelfernandes.org \
--cc=alexei.starovoitov@gmail.com \
--cc=dan.carpenter@oracle.com \
--cc=frederic@kernel.org \
--cc=gregkh@linuxfoundation.org \
--cc=gustavo@embeddedor.com \
--cc=jiangshanlai@gmail.com \
--cc=josh@joshtriplett.org \
--cc=linux-arch@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=luto@kernel.org \
--cc=mathieu.desnoyers@efficios.com \
--cc=mhiramat@kernel.org \
--cc=mingo@kernel.org \
--cc=paulmck@kernel.org \
--cc=peterz@infradead.org \
--cc=rostedt@goodmis.org \
--cc=tglx@linutronix.de \
--cc=tony.luck@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).