Skip to contentSkip to conversation

kthreads: move sched-realeted initialization from kthreadd context

PATCH In review Oleg Nesterov 2 participants 3 messages Jan 30, 2009 – Feb 1, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/4 kthreads: move sched-realeted initialization from kthreadd context

(on top of kthread-dont-looking-for-a-task-in-create_kthread-2.patch)

kthreadd is the single thread which implements ths "create" request, move sched_setscheduler/etc from create_kthread() to kthread_create() to improve the scalability.

We should be careful with sched_setscheduler(), use _nochek helper.

kernel/kthread.c renamed from kernel/kthread.c~1_SCHED +9 -8
@@ -100,15 +100,7 @@ static void create_kthread(struct kthrea
 	if (pid < 0) {
 		create->result = ERR_PTR(pid);
 	} else {
-		struct sched_param param = { .sched_priority = 0 };
 		wait_for_completion(&create->started);
-		/*
-		 * root may have changed our (kthreadd's) priority or CPU mask.
-		 * The kernel thread should not inherit these properties.
-		 */
-		sched_setscheduler(create->result, SCHED_NORMAL, &param);
-		set_user_nice(create->result, KTHREAD_NICE_LEVEL);
-		set_cpus_allowed_ptr(create->result, CPU_MASK_ALL_PTR);
 	}
 	complete(&create->done);
 }
@@ -152,11 +144,20 @@ struct task_struct *kthread_create(int (
 	wait_for_completion(&create.done);
 
 	if (!IS_ERR(create.result)) {
+		struct sched_param param = { .sched_priority = 0 };
 		va_list args;
+
 		va_start(args, namefmt);
 		vsnprintf(create.result->comm, sizeof(create.result->comm),
 			  namefmt, args);
 		va_end(args);
+		/*
+		 * root may have changed our (kthreadd's) priority or CPU mask.
+		 * The kernel thread should not inherit these properties.
+		 */
+		sched_setscheduler_nocheck(create.result, SCHED_NORMAL, &param);
+		set_user_nice(create.result, KTHREAD_NICE_LEVEL);
+		set_cpus_allowed_ptr(create.result, CPU_MASK_ALL_PTR);
 	}
 	return create.result;
 }
#1 lore raw
Quoted 1 line · On Friday 30 January 2009 23:03:50 Oleg Nesterov wrote:

(on top of kthread-dont-looking-for-a-task-in-create_kthread-2.patch)

Hi Oleg,

Thanks for the cc: Vitaliy, I never saw that patch. I've included it in my queue now.

As to this patch, it seems marginal. I've never been convinced that we should
be trying to rescue root if they choose to set kthreadd's prio anyway, but I'm also wondering why we care about kthread_create scalability!

Still, I'm happy to apply it with one change:

kernel/kthread.c struct task_struct *kthread_create(int ( · line 160

Quoted 7 lines
kernel/kthread.c renamed from kernel/kthread.c~1_SCHED +7 -0
+		/*
+		 * root may have changed our (kthreadd's) priority or CPU mask.
+		 * The kernel thread should not inherit these properties.
+		 */
+		sched_setscheduler_nocheck(create.result, SCHED_NORMAL, &param);
+		set_user_nice(create.result, KTHREAD_NICE_LEVEL);
+		set_cpus_allowed_ptr(create.result, CPU_MASK_ALL_PTR);

cpu_all_mask is the non-deprecated replacement for CPU_MASK_ALL_PTR.

Thanks!
Rusty.

#2 lore raw
Quoted 10 lines · Rusty Russell
Quoted 1 line · On Friday 30 January 2009 23:03:50 Oleg Nesterov wrote:

(on top of kthread-dont-looking-for-a-task-in-create_kthread-2.patch)

Hi Oleg,

Thanks for the cc: Vitaliy, I never saw that patch. I've included it in my queue now.

As to this patch, it seems marginal. I've never been convinced that we should be trying to rescue root if they choose to set kthreadd's prio anyway,

I agree. Personally, I never understood this too. But if we want to remove this code we need the separate discussion, so I just moved it to the caller's context.

Quoted 1 line

but I'm also wondering why we care about kthread_create scalability!

It is always better to speedup things and to decrease the latency ;)

But I also think this and the next patch make the code simpler and more clean. Now the only thing create_kthread() does is a plain fork() and nothing else, this is imho good.

Quoted 11 lines

Still, I'm happy to apply it with one change:

kernel/kthread.c struct task_struct *kthread_create(int ( · line 160

Quoted 7 lines
kernel/kthread.c renamed from kernel/kthread.c~1_SCHED +7 -0
+		/*
+		 * root may have changed our (kthreadd's) priority or CPU mask.
+		 * The kernel thread should not inherit these properties.
+		 */
+		sched_setscheduler_nocheck(create.result, SCHED_NORMAL, &param);
+		set_user_nice(create.result, KTHREAD_NICE_LEVEL);
+		set_cpus_allowed_ptr(create.result, CPU_MASK_ALL_PTR);

cpu_all_mask is the non-deprecated replacement for CPU_MASK_ALL_PTR.

Great, thanks.

(btw, I thought about avoiding CPU_MASK_ALL_PTR too, but I didn't
dare to mix 2 different things in one patch).

Oleg.