Re: [PATCH 1/3] tracepoints: Do not use call_rcu_sched() before early_initcall()

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

----- Original Message -----
> From: "Steven Rostedt" <rostedt@goodmis.org>
> To: linux-kernel@vger.kernel.org
> Cc: "Linus Torvalds" <torvalds@linux-foundation.org>, "Ingo Molnar" <mingo@kernel.org>, "Andrew Morton"
> <akpm@linux-foundation.org>, "Thomas Gleixner" <tglx@linutronix.de>, "Mathieu Desnoyers"
> <mathieu.desnoyers@efficios.com>, "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> Sent: Sunday, December 14, 2014 11:41:05 AM
> Subject: [PATCH 1/3] tracepoints: Do not use call_rcu_sched() before early_initcall()
> 
> From: "Steven Rostedt (Red Hat)" <rostedt@goodmis.org>
> 
> In order to move enabling of trace events to just after mm_init(), the
> tracepoint enable code can not use call_rcu_sched() because rcu isn't
> even initialized yet.  Since this can only happen before SMP is set up
> (and even before interrupts are set up), there's no reason to use
> call_rcu_sched() at this point.
> 
> Instead, create a variable called tracepoint_rcu_safe that gets enabled
> via early_initcall() and if that is not set, free the code directly
> instead of using call_rcu_sched().
> 
> This allows us to enable tracepoints early without issues.
> 
> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
> ---
>  kernel/tracepoint.c | 27 ++++++++++++++++++++++++++-
>  1 file changed, 26 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/tracepoint.c b/kernel/tracepoint.c
> index 3490407dc7b7..9b90ef0ac731 100644
> --- a/kernel/tracepoint.c
> +++ b/kernel/tracepoint.c
> @@ -33,6 +33,15 @@ extern struct tracepoint * const
> __stop___tracepoints_ptrs[];
>  /* Set to 1 to enable tracepoint debug output */
>  static const int tracepoint_debug;
>  
> +/*
> + * traceoint_rcu_is_safe is set to true at early_initcall().
> + * Tracepoints can be enabled before rcu is initialized.
> + * When that happens, there's no reason to free via call_rcu()
> + * because the system isn't even in SMP mode yet, and rcu isn't
> + * initialized. Just directly free the old tracepoints instead.
> + */
> +static bool tracepoint_rcu_is_safe;
> +
>  #ifdef CONFIG_MODULES
>  /*
>   * Tracepoint module list mutex protects the local module list.
> @@ -76,7 +85,16 @@ static inline void release_probes(struct tracepoint_func
> *old)
>  	if (old) {
>  		struct tp_probes *tp_probes = container_of(old,
>  			struct tp_probes, probes[0]);
> -		call_rcu_sched(&tp_probes->rcu, rcu_free_old_probes);
> +		/*
> +		 * Tracepoints can be enabled before RCU is initialized
> +		 * at boot up. If that is the case, do not bother using
> +		 * call_rcu() (because that will fail), but instead just
> +		 * free directly.
> +		 */
> +		if (likely(tracepoint_rcu_is_safe))
> +			call_rcu_sched(&tp_probes->rcu, rcu_free_old_probes);
> +		else
> +			rcu_free_old_probes(&tp_probes->rcu);

Would it makes sense to have call_rcu() and call_rcu_sched()
provide this "direct call" fallback at early boot instead
of having this in the caller ?

Thanks,

Mathieu

>  	}
>  }
>  
> @@ -518,3 +536,10 @@ void syscall_unregfunc(void)
>  	}
>  }
>  #endif
> +
> +static __init int init_tracepoint_rcu(void)
> +{
> +	tracepoint_rcu_is_safe = true;
> +	return 0;
> +}
> +early_initcall(init_tracepoint_rcu);
> --
> 2.1.3
> 
> 
> 

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com