Re: [PATCH] lockdep: call trace_hardirqs_off after clearing hardirqs_enabled

[Peter Zijlstra <peterz@infradead.org>]

Thanks for Cc'ing me.

On Mon, Mar 13, 2017 at 06:08:48PM +0800, Qi Hou wrote:
> The current order of calls within trace_hardirqs_off() would provoke an
> "unannotated irqs-off" WARNING.
> 
> This warning was reported by check_flags() when it found that the hardirqs has
> been disabled but the irq-flags state, hardirqs_enabled, has not been cleared.
> 
> Check_flags() is called via trace_hardirqs_off(), and trace_hardirqs_off() will
> be called by functions kind of local_irq_disable() if CONFIG_TRACE_IRQFLAGS has
> been enabled.
> 
> The calltrace from functions kind of local_irq_diasble() to check_flags()
> simplied as below:
> 
> local_irq_disable()
>  |- trace_hardirqs_off()
>      |- trace_hardirqs_off_caller()
>          |- preemptirqsoff_hist()
>             |- __DO_TRACE()
>                |- rcu_dereference_sched()
>                    |- rcu_read_lock_sched_held()
>                        |- lock_is_held()
>                            |- check_flags()
> 
> The order of calls within trace_hardirqs_off() can be simplified as below:
> 
> local_irq_disable()
>  |- raw_local_irq_disable()
>  |- trace_hardirqs_off()
>      |- trace_hardirqs_off() --> check_flags()
>      |- hardirqs_enabled = 0
> 
> We could find that the call to check_flags() has been made between disabling
> hardirqs and clearing hardirqs_enabled. That will definitely cause an
> "unannotated irqs-off" warning.
> 
> To fix this "unannotated irqs-off" warning, we must call trace_hardirqs_off()
> after that we clear hardirqs_enabled.

Here you say: "trace_hardirqs_off()"

> Signed-off-by: Qi Hou <qi.hou@windriver.com>
> ---
>  kernel/locking/lockdep.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
> index 577f026..e8b35e4 100644
> --- a/kernel/locking/lockdep.c
> +++ b/kernel/locking/lockdep.c
> @@ -2627,8 +2627,6 @@ __visible void trace_hardirqs_off_caller(unsigned long ip)
>  {
>  	struct task_struct *curr = current;
>  
> -	time_hardirqs_off(CALLER_ADDR0, ip);

This is: "TIME_hardirqs_off()", notice the difference?

Care to try to write a changelog that makes sense?