SMP: bind process to cpu

SMP: bind process to cpu

[Thomas Widmann]

Hi,

I run an 3*XEON 550MHz Primergy with 2GB of RAM.
On this machine, i have compiled kernel 2.4.0SMP.

Is it possible to bind a process to a specific
cpu on this SMP machine (process affinity) ?

I there something like pset ?

Thanks in advance

Regards
Widmann Thomas
Siemens AG - Munich

Re: SMP: bind process to cpu

[Christoph Hellwig]

In article <BGEDIODHBENLENEMBEPAEEDFCAAA.thomas.widmann@icn.siemens.de> you wrote:
> Hi,
>
> I run an 3*XEON 550MHz Primergy with 2GB of RAM.
> On this machine, i have compiled kernel 2.4.0SMP.
>
> Is it possible to bind a process to a specific
> cpu on this SMP machine (process affinity) ?

Linux 2.4 is mostlu ready for process affinity, but it is not (yet)
exported to userspace.  I've attached at patch by Nick Pollitt from SGI
that allows to enable process pinning using prctl().

> I there something like pset ?

I've seen patches for SGI-like psets for 2.2.<something>, but not for 2.4.

	Christoph

-- 
Of course it doesn't work. We've performed a software upgrade.


diff -X /home/npollitt/dontdiff -Nur origlinux/fs/proc/array.c linux/fs/proc/array.c
--- origlinux/fs/proc/array.c	Tue Nov 14 11:22:36 2000
+++ linux/fs/proc/array.c	Thu Jan 25 15:17:35 2001
@@ -347,7 +347,7 @@
 	read_unlock(&tasklist_lock);
 	res = sprintf(buffer,"%d (%s) %c %d %d %d %d %d %lu %lu \
 %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu %lu \
-%lu %lu %lu %lu %lu %lu %lu %lu %d %d\n",
+%lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu\n",
 		task->pid,
 		task->comm,
 		state,
@@ -390,7 +390,8 @@
 		task->nswap,
 		task->cnswap,
 		task->exit_signal,
-		task->processor);
+		task->processor,
+		task->cpus_allowed);
 	if(mm)
 		mmput(mm);
 	return res;
diff -X /home/npollitt/dontdiff -Nur origlinux/include/linux/prctl.h linux/include/linux/prctl.h
--- origlinux/include/linux/prctl.h	Sun Mar 19 11:15:32 2000
+++ linux/include/linux/prctl.h	Thu Jan 25 15:17:35 2001
@@ -20,4 +20,9 @@
 #define PR_GET_KEEPCAPS   7
 #define PR_SET_KEEPCAPS   8
 
+#define PR_GET_RUNON	  9
+#define PR_SET_RUNON	  10
+#define PR_MUSTRUN_PID    11
+#define PR_RUNANY_PID     12
+
 #endif /* _LINUX_PRCTL_H */
diff -X /home/npollitt/dontdiff -Nur origlinux/kernel/sched.c linux/kernel/sched.c
--- origlinux/kernel/sched.c	Thu Jan  4 13:50:38 2001
+++ linux/kernel/sched.c	Thu Jan 25 17:22:23 2001
@@ -108,6 +108,10 @@
 #ifdef CONFIG_SMP
 
 #define idle_task(cpu) (init_tasks[cpu_number_map(cpu)])
+#define can_schedule_goodness(p,cpu) ( (!(p)->has_cpu ||  \
+                                        p->processor == cpu) &&  \
+                                        ((p)->cpus_allowed & (1 << cpu)))
+
 #define can_schedule(p,cpu) ((!(p)->has_cpu) && \
 				((p)->cpus_allowed & (1 << cpu)))
 
@@ -568,7 +572,7 @@
 still_running_back:
 	list_for_each(tmp, &runqueue_head) {
 		p = list_entry(tmp, struct task_struct, run_list);
-		if (can_schedule(p, this_cpu)) {
+		if (can_schedule_goodness(p, this_cpu)) {
 			int weight = goodness(p, this_cpu, prev->active_mm);
 			if (weight > c)
 				c = weight, next = p;
diff -X /home/npollitt/dontdiff -Nur origlinux/kernel/sys.c linux/kernel/sys.c
--- origlinux/kernel/sys.c	Mon Oct 16 12:58:51 2000
+++ linux/kernel/sys.c	Thu Jan 25 15:17:35 2001
@@ -1203,12 +1203,95 @@
 			}
 			current->keep_capabilities = arg2;
 			break;
+		case PR_GET_RUNON:
+			error = put_user(current->cpus_allowed, (long *)arg2);
+			break;
+		case PR_SET_RUNON:
+			if (arg2 == 0)
+				arg2 = 1 << smp_processor_id();
+			arg2 &= cpu_online_map;
+			if (!arg2)
+				error = -EINVAL;
+			else {
+				current->cpus_allowed = arg2;
+				if (!(arg2 & (1 << smp_processor_id())))
+					current->need_resched = 1;
+			}
+			break;
+		case PR_MUSTRUN_PID:
+			/* arg2 is cpu, arg3 is pid */
+			if (arg2 == 0)
+				arg2 = 1 << smp_processor_id();
+			arg2 &= cpu_online_map;
+			if (!arg2)
+				error = -EINVAL;
+			error = mp_mustrun_pid(arg2, arg3);
+			break;
+		case PR_RUNANY_PID:
+			/* arg2 is pid */
+			if (!arg2)
+				error = -EINVAL;
+			error = mp_runany_pid(arg2);
+			break;
 		default:
 			error = -EINVAL;
 			break;
 	}
 	return error;
 }
+
+static int mp_mustrun_pid(int cpu, int pid) 
+{
+	struct task_struct *p;
+	int ret;
+
+	ret = -EPERM;
+	/* Not allowed to change 1 */
+	if (pid == 1) 
+		goto out;
+
+	read_lock(&tasklist_lock);
+	p = find_task_by_pid(pid);
+	if (p)
+		get_task_struct(p);
+	read_unlock(&tasklist_lock);
+	if (!p)
+		ret = -ESRCH;
+
+	p->cpus_allowed = cpu;
+	p->need_resched = 1;
+	free_task_struct(p);
+	ret = 0;
+out:
+	return ret;
+}
+
+static int mp_runany_pid(int pid) 
+{
+	struct task_struct *p;
+	int ret;
+
+	ret = -EPERM;
+	/* Not allowed to change 1 */
+	if (pid == 1) 
+		goto out;
+
+	read_lock(&tasklist_lock);
+	p = find_task_by_pid(pid);
+	if (p)
+		get_task_struct(p);
+	read_unlock(&tasklist_lock);
+	if (!p)
+		ret = -ESRCH;
+
+	p->cpus_allowed = 0xFFFFFFFF;
+	p->need_resched = 0;
+	free_task_struct(p);
+	ret = 0;
+out:
+	return ret;
+}
+
 
 EXPORT_SYMBOL(notifier_chain_register);
 EXPORT_SYMBOL(notifier_chain_unregister);

Re: SMP: bind process to cpu

[Manfred Spraul]

Christoph Hellwig wrote:
> 
> In article <BGEDIODHBENLENEMBEPAEEDFCAAA.thomas.widmann@icn.siemens.de> you wrote:
> > Hi,
> >
> > I run an 3*XEON 550MHz Primergy with 2GB of RAM.
> > On this machine, i have compiled kernel 2.4.0SMP.
> >
> > Is it possible to bind a process to a specific
> > cpu on this SMP machine (process affinity) ?
> 
> Linux 2.4 is mostlu ready for process affinity, but it is not (yet)
> exported to userspace.  I've attached at patch by Nick Pollitt from SGI
> that allows to enable process pinning using prctl().
> 
> > I there something like pset ?
> 
> I've seen patches for SGI-like psets for 2.2.<something>, but not for 2.4.
>
>         Christoph
>
You must also update wake_process_synchroneous(), otherwise you can get
lost wakeups with pipes.

Something like

>         if (!(p->cpus_allowed & (1 << smp_processor_id()))
>                 reschedule_idle(p);

must be added after add_to_runqueue().

Ingo Molnar did some testing with tux2, and under high load wakeups were
lost without such a patch.

--
	Manfred

Re: SMP: bind process to cpu

[Christoph Hellwig]

	[Nick, I've added you to the Cc list so you can look at
	 it for future versions of your patch]


On Sat, Feb 17, 2001 at 03:13:45PM +0100, Manfred Spraul wrote:
> You must also update wake_process_synchroneous(), otherwise you can get
> lost wakeups with pipes.
> 
> Something like
> 
> >         if (!(p->cpus_allowed & (1 << smp_processor_id()))
> >                 reschedule_idle(p);
> 
> must be added after add_to_runqueue().

Ok.

> Ingo Molnar did some testing with tux2, and under high load wakeups were
> lost without such a patch.

(s/tux2/tux/ I suppose)

Yepp - but tux is again not userspace...

	Christoph

-- 
Of course it doesn't work. We've performed a software upgrade.

Re: SMP: bind process to cpu

[Andrew Morton]

Thomas Widmann wrote:
> 
> Hi,
> 
> I run an 3*XEON 550MHz Primergy with 2GB of RAM.
> On this machine, i have compiled kernel 2.4.0SMP.
> 
> Is it possible to bind a process to a specific
> cpu on this SMP machine (process affinity) ?
> 
> I there something like pset ?

A patch which creates /proc/<pid>/cpus_allowed is at

	http://www.uow.edu.au/~andrewm/linux/#cpus_allowed

You just write a bitmask into it.

Re: SMP: bind process to cpu

[Thomas Widmann]

Hi,

* Andrew Morton wrote:

> > Hi,
> > 
> > I run an 3*XEON 550MHz Primergy with 2GB of RAM.
> > On this machine, i have compiled kernel 2.4.0SMP.
> > 
> > Is it possible to bind a process to a specific
> > cpu on this SMP machine (process affinity) ?
> > 
> > I there something like pset ?
> 
> A patch which creates /proc/<pid>/cpus_allowed is at
> 
> 	http://www.uow.edu.au/~andrewm/linux/#cpus_allowed
> 
> You just write a bitmask into it.

Thanks for this information. I patched my the kernel with it.
After rebooting with the new kernel i can see the bitmask
for every process running on my server.

#cat /proc/1310/cpus_allowed
ffffffff

Now, if i want to run this process on only one cpu, i which way
do i have to set the bitmask ?
Let's say, i want to run it on cpu0. how look's the bitmask ?

Thanks 

Regards
Thomas

Re: SMP: bind process to cpu

[Francis Galiegue]

On Sat, 17 Feb 2001, Thomas Widmann wrote:

> 
> #cat /proc/1310/cpus_allowed
> ffffffff
> 
> Now, if i want to run this process on only one cpu, i which way
> do i have to set the bitmask ?
> Let's say, i want to run it on cpu0. how look's the bitmask ?
> 

Wild guess: as this is a bitmask, you must "or" the bitmask with (1 <<
cpu_number). 1 for CPU 0 only, 5 for CPU 0 and 2, etc, etc.

-- 
Francis Galiegue, fg@mandrakesoft.com - Normand et fier de l'être
"Programming is a race between programmers, who try and make more and more
idiot-proof software, and universe, which produces more and more remarkable
idiots. Until now, universe leads the race"  -- R. Cook

Re: SMP: bind process to cpu

[Andrew Morton]

Thomas Widmann wrote:
> ...
> * Andrew Morton wrote:
> 
> >       http://www.uow.edu.au/~andrewm/linux/#cpus_allowed
> >
> > You just write a bitmask into it.
> 
> Thanks for this information. I patched my the kernel with it.
> After rebooting with the new kernel i can see the bitmask
> for every process running on my server.
> 
> #cat /proc/1310/cpus_allowed
> ffffffff
> 
> Now, if i want to run this process on only one cpu, i which way
> do i have to set the bitmask ?
> Let's say, i want to run it on cpu0. how look's the bitmask ?

Each bit corresponds to a logical CPU on which the task
is permitted to run.  So 1->CPU0, 2->CPU1, 5->CPU0+CPU2, etc.

But it does seem there are problems with the scheduler
which occur when cpus_allowed it not all-ones.  I didn't
observe any problems in the 1-2 hours testing which I
needed that patch for, so it should be OK for experimentation
purposes..

-

Re: SMP: bind process to cpu

[Tim Hockin]

> Is it possible to bind a process to a specific
> cpu on this SMP machine (process affinity) ?
> 
> I there something like pset ?

http://isunix.it.ilstu.edu/~thockin/pset  - pset for linux-2.2 (not ported
to 2.4 yet)