Re: [GIT PULL] scheduler changes for v5.3

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Jul 09, 2019 at 10:48:49PM -0700, John Stultz wrote:
> On Mon, Jul 8, 2019 at 9:33 AM Ingo Molnar <mingo@kernel.org> wrote:
> > Please pull the latest sched-core-for-linus git tree from:
> >
> >    git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched-core-for-linus
> ....
> > Peter Zijlstra (1):
> >       sched/core: Optimize try_to_wake_up() for local wakeups
> 
> Hey Peter, Ingo,
>    Since this change landed in Linus' tree, I've been seeing a lot of
> the following dmesg noise when running AOSP on the HiKey960 board.
> 
> [  173.162712] CPU: 2 PID: 731 Comm: ndroid.systemui Tainted: G S
>           5.2.0-rc5-00110-g6751c43d94d6 #447
> [  173.162721] Hardware name: HiKey960 (DT)
> [  173.171194] caller is try_to_wake_up+0x3e4/0x788
> [  173.179605] Call trace:
> [  173.179617]  dump_backtrace+0x0/0x140
> [  173.179626]  show_stack+0x14/0x20
> [  173.179638]  dump_stack+0x9c/0xc4
> [  173.179649]  debug_smp_processor_id+0x148/0x150
> [  173.179659]  try_to_wake_up+0x3e4/0x788
> [  173.179669]  wake_up_q+0x5c/0x98
> [  173.179681]  futex_wake+0x170/0x1a8
> [  173.179696]  do_futex+0x560/0xf30
> [  173.284541]  __arm64_sys_futex+0xfc/0x148
> [  173.288570]  el0_svc_common.constprop.0+0x64/0x188
> [  173.293371]  el0_svc_handler+0x28/0x78
> [  173.297131]  el0_svc+0x8/0xc
> [  173.300045] CPU: 0 PID: 1258 Comm: Binder:363_5 Tainted: G S
>         5.2.0-rc5-00110-g6751c43d94d6 #447
> [  173.301130] BUG: using smp_processor_id() in preemptible [00000000]
> code: ndroid.systemui/731
> [  173.310074] Hardware name: HiKey960 (DT)
> [  173.310084] Call trace:
> [  173.310112]  dump_backtrace+0x0/0x140
> [  173.310131]  show_stack+0x14/0x20
> [  173.318685] caller is try_to_wake_up+0x3e4/0x788
> [  173.322583]  dump_stack+0x9c/0xc4
> [  173.322595]  debug_smp_processor_id+0x148/0x150
> [  173.322605]  try_to_wake_up+0x3e4/0x788
> [  173.322615]  wake_up_q+0x5c/0x98
> [  173.322628]  futex_wake+0x170/0x1a8
> [  173.322641]  do_futex+0x560/0xf30
> [  173.358367]  __arm64_sys_futex+0xfc/0x148
> [  173.362397]  el0_svc_common.constprop.0+0x64/0x188
> [  173.367199]  el0_svc_handler+0x28/0x78
> [  173.370956]  el0_svc+0x8/0xc
> 

Urgh.. however didn't we find that before :/ stupid stats.

Something like the below ought to fix, but let me see if I can come up
with something saner...

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 108449526f11..0b22e55cebe8 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2399,6 +2399,7 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 	unsigned long flags;
 	int cpu, success = 0;
 
+	preempt_disable();
 	if (p == current) {
 		/*
 		 * We're waking current, this means 'p->on_rq' and 'task_cpu(p)
@@ -2412,7 +2413,7 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 		 *    it disabling IRQs (this allows not taking ->pi_lock).
 		 */
 		if (!(p->state & state))
-			return false;
+			goto out;
 
 		success = 1;
 		cpu = task_cpu(p);
@@ -2526,6 +2527,7 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 out:
 	if (success)
 		ttwu_stat(p, cpu, wake_flags);
+	preempt_enable();
 
 	return success;
 }