2013-07-29 17:19:30

by Abbas Raza

Subject: [PATCH] sched: don't wait in kthread_bind for presmp initcalls

From: Abbas Raza <[email protected]>

wait_task_inactive shouldn't be called in kthread_bind for presmp
initcalls the same way it is done in !SMP case.

more info here:
http://permalink.gmane.org/gmane.linux.kernel.embedded/4046

This patch improves boot time for pre-smp initcalls as given below

initcall spawn_ksoftirqd+0x0/0x54 returned 0 after 0 usecs
initcall init_workqueues+0x0/0x358 returned 0 after 0 usecs
initcall cpu_stop_init+0x0/0xcc returned 0 after 0 usecs

boot time without this patch:

initcall spawn_ksoftirqd+0x0/0x54 returned 0 after 9765 usecs
initcall init_workqueues+0x0/0x358 returned 0 after 9765 usecs
initcall cpu_stop_init+0x0/0xcc returned 0 after 19531 usecs

Signed-off-by: Abbas Raza <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Ingo Molnar <[email protected]>

---
kernel/kthread.c | 13 +++++++++----
1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/kernel/kthread.c b/kernel/kthread.c
index b579af5..80f19b5 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -221,10 +221,15 @@ EXPORT_SYMBOL(kthread_create_on_node);
*/
void kthread_bind(struct task_struct *p, unsigned int cpu)
{
- /* Must have done schedule() in kthread() before we set_task_cpu */
- if (!wait_task_inactive(p, TASK_UNINTERRUPTIBLE)) {
- WARN_ON(1);
- return;
+ if (!((num_online_cpus() == 1) && !cpu)) {
+ /*
+ * Must have done schedule() in kthread() before
+ * we set_task_cpu
+ */
+ if (!wait_task_inactive(p, TASK_UNINTERRUPTIBLE)) {
+ WARN_ON(1);
+ return;
+ }
}

/* It's safe because the task is inactive. */
--
1.8.3.2

2013-07-30 03:27:50

by Xie XiuQi

Subject: Re: [PATCH] sched: don't wait in kthread_bind for presmp initcalls

On 2013/7/30 1:19, Abbas Raza wrote:
> From: Abbas Raza <[email protected]>
>
> wait_task_inactive shouldn't be called in kthread_bind for presmp
> initcalls the same way it is done in !SMP case.
>
> more info here:
> http://permalink.gmane.org/gmane.linux.kernel.embedded/4046
>
> This patch improves boot time for pre-smp initcalls as given below
>
> initcall spawn_ksoftirqd+0x0/0x54 returned 0 after 0 usecs
> initcall init_workqueues+0x0/0x358 returned 0 after 0 usecs
> initcall cpu_stop_init+0x0/0xcc returned 0 after 0 usecs
>
> boot time without this patch:
>
> initcall spawn_ksoftirqd+0x0/0x54 returned 0 after 9765 usecs
> initcall init_workqueues+0x0/0x358 returned 0 after 9765 usecs
> initcall cpu_stop_init+0x0/0xcc returned 0 after 19531 usecs
>
> Signed-off-by: Abbas Raza <[email protected]>
> Cc: Peter Zijlstra <[email protected]>
> Cc: Ingo Molnar <[email protected]>
>
> ---
> kernel/kthread.c | 13 +++++++++----
> 1 file changed, 9 insertions(+), 4 deletions(-)
>
> diff --git a/kernel/kthread.c b/kernel/kthread.c
> index b579af5..80f19b5 100644
> --- a/kernel/kthread.c
> +++ b/kernel/kthread.c
> @@ -221,10 +221,15 @@ EXPORT_SYMBOL(kthread_create_on_node);
> */
> void kthread_bind(struct task_struct *p, unsigned int cpu)
> {
> - /* Must have done schedule() in kthread() before we set_task_cpu */
> - if (!wait_task_inactive(p, TASK_UNINTERRUPTIBLE)) {
> - WARN_ON(1);
> - return;

Hi Abbas, this patch isn't based the current mainline, and kthread_bind()
has been changed since v3.9.

> + if (!((num_online_cpus() == 1) && !cpu)) {
> + /*
> + * Must have done schedule() in kthread() before
> + * we set_task_cpu
> + */
> + if (!wait_task_inactive(p, TASK_UNINTERRUPTIBLE)) {
> + WARN_ON(1);
> + return;
> + }
> }
>
> /* It's safe because the task is inactive. */
>