Re: [PATCH 3.19 016/175] ksoftirqd: Enable IRQs and call cond_resched() before poking RCU

[Michal Kubecek]

On Sat, May 02, 2015 at 05:14:50AM +0200, Mike Galbraith wrote:
> On Fri, 2015-05-01 at 22:52 +0200, Greg Kroah-Hartman wrote:
> > On Fri, May 01, 2015 at 03:00:00PM -0500, Josh Hunt wrote:
> > > On Wed, Mar 4, 2015 at 12:13 AM, Greg Kroah-Hartman
> > > <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > > 3.19-stable review patch.  If anyone has any objections, please let me know.
> > > >
> > > > ------------------
> > > >
> > > > From: Calvin Owens <calvinowens@xxxxxx>
> > > >
> > > > commit 28423ad283d5348793b0c45cc9b1af058e776fd6 upstream.
> > > >
> > > > While debugging an issue with excessive softirq usage, I encountered the
> > > > following note in commit 3e339b5dae24a706 ("softirq: Use hotplug thread
> > > > infrastructure"):
> > > >
> > > >     [ paulmck: Call rcu_note_context_switch() with interrupts enabled. ]
> > > >
> > > > ...but despite this note, the patch still calls RCU with IRQs disabled.
> > > >
> > > > This seemingly innocuous change caused a significant regression in softirq
> > > > CPU usage on the sending side of a large TCP transfer (~1 GB/s): when
> > > > introducing 0.01% packet loss, the softirq usage would jump to around 25%,
> > > > spiking as high as 50%. Before the change, the usage would never exceed 5%.
> > > >
> > > > Moving the call to rcu_note_context_switch() after the cond_sched() call,
> > > > as it was originally before the hotplug patch, completely eliminated this
> > > > problem.
> > > >
> > > > Signed-off-by: Calvin Owens <calvinowens@xxxxxx>
> > > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> > > > Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> > > >
> > > > ---
> > > >  kernel/softirq.c |    6 +++++-
> > > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > > >
> > > > --- a/kernel/softirq.c
> > > > +++ b/kernel/softirq.c
> > > > @@ -656,9 +656,13 @@ static void run_ksoftirqd(unsigned int c
> > > >                  * in the task stack here.
> > > >                  */
> > > >                 __do_softirq();
> > > > -               rcu_note_context_switch();
> > > >                 local_irq_enable();
> > > >                 cond_resched();
> > > > +
> > > > +               preempt_disable();
> > > > +               rcu_note_context_switch();
> > > > +               preempt_enable();
> > > > +
> > > >                 return;
> > > >         }
> > > >         local_irq_enable();
> > > >
> > > >
> > > > --
> > > > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > > > the body of a message to majordomo@xxxxxxxxxxxxxxx
> > > > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > > > Please read the FAQ at  http://www.tux.org/lkml/
> > > 
> > > Sorry for the delay in noticing this, but should this be applied to
> > > 3.14-stable as well?
> > 
> > Why should it?
> 
> The regression inducing change arrived in 3.7-rc1.

I guess stable-3.12.y should have it too, then (with a trivial refresh
for the comment added in 3.13).

Michal Kubecek

> >   And odds are, if I didn't apply it there, it was either
> > because it didn't apply, or it broke the build.
> 
> a. [x]  rcu_note_context_switch(cpu) -> rcu_note_context_switch()
> 
> From 28423ad283d5348793b0c45cc9b1af058e776fd6 Mon Sep 17 00:00:00 2001
> From: Calvin Owens <calvinowens@xxxxxx>
> Date: Tue, 13 Jan 2015 13:16:18 -0800
> Subject: ksoftirqd: Enable IRQs and call cond_resched() before poking RCU
> 
> While debugging an issue with excessive softirq usage, I encountered the
> following note in commit 3e339b5dae24a706 ("softirq: Use hotplug thread
> infrastructure"):
> 
>     [ paulmck: Call rcu_note_context_switch() with interrupts enabled. ]
> 
> ...but despite this note, the patch still calls RCU with IRQs disabled.
> 
> This seemingly innocuous change caused a significant regression in softirq
> CPU usage on the sending side of a large TCP transfer (~1 GB/s): when
> introducing 0.01% packet loss, the softirq usage would jump to around 25%,
> spiking as high as 50%. Before the change, the usage would never exceed 5%.
> 
> Moving the call to rcu_note_context_switch() after the cond_sched() call,
> as it was originally before the hotplug patch, completely eliminated this
> problem.
> 
> Signed-off-by: Calvin Owens <calvinowens@xxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> Signed-off-by: Mike Galbraith <mgalbraith@xxxxxxx>
> ---
>  kernel/softirq.c |    6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> --- a/kernel/softirq.c
> +++ b/kernel/softirq.c
> @@ -657,9 +657,13 @@ static void run_ksoftirqd(unsigned int c
>  		 * in the task stack here.
>  		 */
>  		__do_softirq();
> -		rcu_note_context_switch(cpu);
>  		local_irq_enable();
>  		cond_resched();
> +
> +		preempt_disable();
> +		rcu_note_context_switch(cpu);
> +		preempt_enable();
> +
>  		return;
>  	}
>  	local_irq_enable();
> 
> 
> --
> To unsubscribe from this list: send the line "unsubscribe stable" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/