Re: [PATCH] trace: irqsoff: Fix function tracing in preempt and preemptirqsoff tracers

[Steven Rostedt]

On Sat, 12 Mar 2016 00:13:06 -0800
Joel Fernandes <agnel.joel@xxxxxxxxx> wrote:

> All functions aren't traced in critical sections where preemption is disabled
> and interrupts are enabled in preempt* tracers because current behavior is to
> check if interrupts are disabled and if they are not, then we don't trace these
> functions. So here we do the correct checks depending on tracer type and allow
> these functions to be traced.
> 
> One example is when interrupts are enabled before softirq processing, with the
> patch, these functions are traced as well.
> 
>    <...>-2265    1d.h1 3419us : preempt_count_sub <-irq_exit
>    <...>-2265    1d..1 3419us : __do_softirq <-irq_exit
>    <...>-2265    1d..1 3419us : msecs_to_jiffies <-__do_softirq
>    <...>-2265    1d..1 3420us : irqtime_account_irq <-__do_softirq
>    <...>-2265    1d..1 3420us : __local_bh_disable_ip <-__do_softirq
>    <...>-2265    1..s1 3421us : run_timer_softirq <-__do_softirq
>    <...>-2265    1..s1 3421us : hrtimer_run_pending <-run_timer_softirq
>    <...>-2265    1..s1 3421us : _raw_spin_lock_irq <-run_timer_softirq
>    <...>-2265    1d.s1 3422us : preempt_count_add <-_raw_spin_lock_irq
>    <...>-2265    1d.s2 3422us : _raw_spin_unlock_irq <-run_timer_softirq
>    <...>-2265    1..s2 3422us : preempt_count_sub <-_raw_spin_unlock_irq
>    <...>-2265    1..s1 3423us : rcu_bh_qs <-__do_softirq
>    <...>-2265    1d.s1 3423us : irqtime_account_irq <-__do_softirq
>    <...>-2265    1d.s1 3423us : __local_bh_enable <-__do_softirq
> 
> Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> Signed-off-by: Joel Fernandes <agnel.joel@xxxxxxxxx>
> ---
>  kernel/trace/trace_irqsoff.c | 16 ++++++++++++++--
>  1 file changed, 14 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c
> index e4e5658..ca8f84f 100644
> --- a/kernel/trace/trace_irqsoff.c
> +++ b/kernel/trace/trace_irqsoff.c
> @@ -110,8 +110,20 @@ static int func_prolog_dec(struct trace_array *tr,
>  
>  	local_save_flags(*flags);
>  	/* slight chance to get a false positive on tracing_cpu */
> -	if (!irqs_disabled_flags(*flags))
> -		return 0;
> +	switch (trace_type) {
> +	case (TRACER_IRQS_OFF | TRACER_PREEMPT_OFF):
> +		if (!preempt_trace() && !irqs_disabled_flags(*flags))
> +			return 0;
> +		break;
> +	case TRACER_IRQS_OFF:
> +		if (!irqs_disabled_flags(*flags))
> +			return 0;
> +		break;
> +	case TRACER_PREEMPT_OFF:
> +		if (!preempt_trace())
> +			return 0;
> +		break;
> +	}

Actually this is too complex a fix. If tracing_cpu is set, we want to
trace this cpu. And I'm starting to think we don't even need this
check, because I'm having a hard time seeing how tracing_cpu could be
set without preemption or irqs being disabled, and thus there be no
race.

Anyway, we can just add a check if either irqs are disabled or
interrupts are disabled:

	if (!irqs_disabled_flags(*flags) && !preempt_count())
		return 0;

-- Steve

>  
>  	*data = per_cpu_ptr(tr->trace_buffer.data, cpu);
>  	disabled = atomic_inc_return(&(*data)->disabled);