[PATCH] Revert "isolcpus: Affine unbound kernel threads to housekeeping cpus"

[PATCH] Revert "isolcpus: Affine unbound kernel threads to housekeeping cpus"

[Marcelo Tosatti]

commit 9cc5b8656892a72438ee7deb introduced a new housekeeping flag,
HK_FLAG_KTHREAD, that when enabled sets the CPU affinity for the 
kthreadd process (therefore all unbounded kernel threads created
from that point on will use the housekeeping cpumask).

This is not necessary, since its possible to control placement of
kthreadd from userspace:

# taskset -c -p 0 `pgrep kthreadd`
pid 2's current affinity list: 1
pid 2's new affinity list: 0

Unbounded kernel threads started from that point on will inherit
the kthreadd cpumask.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: linux-2.6/include/linux/sched/isolation.h
===================================================================
--- linux-2.6.orig/include/linux/sched/isolation.h
+++ linux-2.6/include/linux/sched/isolation.h
@@ -14,7 +14,6 @@ enum hk_flags {
 	HK_FLAG_DOMAIN		= (1 << 5),
 	HK_FLAG_WQ		= (1 << 6),
 	HK_FLAG_MANAGED_IRQ	= (1 << 7),
-	HK_FLAG_KTHREAD		= (1 << 8),
 };
 
 #ifdef CONFIG_CPU_ISOLATION
Index: linux-2.6/kernel/kthread.c
===================================================================
--- linux-2.6.orig/kernel/kthread.c
+++ linux-2.6/kernel/kthread.c
@@ -27,7 +27,6 @@
 #include <linux/ptrace.h>
 #include <linux/uaccess.h>
 #include <linux/numa.h>
-#include <linux/sched/isolation.h>
 #include <trace/events/sched.h>
 
 
@@ -405,8 +404,7 @@ struct task_struct *__kthread_create_on_
 		 * The kernel thread should not inherit these properties.
 		 */
 		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
-		set_cpus_allowed_ptr(task,
-				     housekeeping_cpumask(HK_FLAG_KTHREAD));
+		set_cpus_allowed_ptr(task, cpu_possible_mask);
 	}
 	kfree(create);
 	return task;
@@ -655,7 +653,7 @@ int kthreadd(void *unused)
 	/* Setup a clean context for our children to inherit. */
 	set_task_comm(tsk, "kthreadd");
 	ignore_signals(tsk);
-	set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
+	set_cpus_allowed_ptr(tsk, cpu_possible_mask);
 	set_mems_allowed(node_states[N_MEMORY]);
 
 	current->flags |= PF_NOFREEZE;
Index: linux-2.6/kernel/sched/isolation.c
===================================================================
--- linux-2.6.orig/kernel/sched/isolation.c
+++ linux-2.6/kernel/sched/isolation.c
@@ -140,8 +140,7 @@ static int __init housekeeping_nohz_full
 {
 	unsigned int flags;
 
-	flags = HK_FLAG_TICK | HK_FLAG_WQ | HK_FLAG_TIMER | HK_FLAG_RCU |
-		HK_FLAG_MISC | HK_FLAG_KTHREAD;
+	flags = HK_FLAG_TICK | HK_FLAG_WQ | HK_FLAG_TIMER | HK_FLAG_RCU | HK_FLAG_MISC;
 
 	return housekeeping_setup(str, flags);
 }

Re: [PATCH] Revert "isolcpus: Affine unbound kernel threads to housekeeping cpus"

[Frederic Weisbecker]

On Fri, May 07, 2021 at 01:57:10PM -0300, Marcelo Tosatti wrote:
> 
> 
> commit 9cc5b8656892a72438ee7deb introduced a new housekeeping flag,
> HK_FLAG_KTHREAD, that when enabled sets the CPU affinity for the 
> kthreadd process (therefore all unbounded kernel threads created
> from that point on will use the housekeeping cpumask).
> 
> This is not necessary, since its possible to control placement of
> kthreadd from userspace:
> 
> # taskset -c -p 0 `pgrep kthreadd`
> pid 2's current affinity list: 1
> pid 2's new affinity list: 0
> 
> Unbounded kernel threads started from that point on will inherit
> the kthreadd cpumask.

Hmm, but look below:

> @@ -405,8 +404,7 @@ struct task_struct *__kthread_create_on_
>  		 * The kernel thread should not inherit these properties.
>  		 */
>  		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
> -		set_cpus_allowed_ptr(task,
> -				     housekeeping_cpumask(HK_FLAG_KTHREAD));
> +		set_cpus_allowed_ptr(task, cpu_possible_mask);

That inheritance is then overriden, right?

Thanks.

Re: [PATCH] Revert "isolcpus: Affine unbound kernel threads to housekeeping cpus"

[Marcelo Tosatti]

On Fri, May 07, 2021 at 10:16:00PM +0200, Frederic Weisbecker wrote:
> On Fri, May 07, 2021 at 01:57:10PM -0300, Marcelo Tosatti wrote:
> > 
> > 
> > commit 9cc5b8656892a72438ee7deb introduced a new housekeeping flag,
> > HK_FLAG_KTHREAD, that when enabled sets the CPU affinity for the 
> > kthreadd process (therefore all unbounded kernel threads created
> > from that point on will use the housekeeping cpumask).
> > 
> > This is not necessary, since its possible to control placement of
> > kthreadd from userspace:
> > 
> > # taskset -c -p 0 `pgrep kthreadd`
> > pid 2's current affinity list: 1
> > pid 2's new affinity list: 0
> > 
> > Unbounded kernel threads started from that point on will inherit
> > the kthreadd cpumask.
> 
> Hmm, but look below:
> 
> > @@ -405,8 +404,7 @@ struct task_struct *__kthread_create_on_
> >  		 * The kernel thread should not inherit these properties.
> >  		 */
> >  		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
> > -		set_cpus_allowed_ptr(task,
> > -				     housekeeping_cpumask(HK_FLAG_KTHREAD));
> > +		set_cpus_allowed_ptr(task, cpu_possible_mask);
> 
> That inheritance is then overriden, right?
> 
> Thanks.

Hi Frederic,

Doh, yes, ignore me.