Skip to contentSkip to conversation

add allowed CPUs check into find_idlest_group()

PATCH In review M.Baris Demiray 2 participants 5 messages Jun 5, 2005 – Jun 6, 2005

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 add allowed CPUs check into find_idlest_group()

Hello,
following patch adds check for allowed CPUs into sched.c:find_idlest_group() -as told in comment line that had removed-. But, I have several questions about that comment.

Firstly, I've understood it as "Check whether process p is allowed to run on each CPU of to-be-found idlest group"; is that right?

If so, isn't it more appropriate to do check in find_idlest_cpu()? Because, we're only interested in CPUs that are in idlest group but doing a check in find_idlest_group() also checks for CPUs that are not in idlest group (since we're traversing all the groups in given domain). Checking this after finding the idlest group (in find_idlest_cpu() with ordinary call order as in sched_balance_self()) will save us from extra overhead.

Although I've questions in my mind, I'm sending a patch following that comment. Any explanation and comment on patch will be appreciated.

Regards.

kernel/sched.c renamed from kernel/sched.c.orig +6 -1
@@ -1040,7 +1040,12 @@
 		int i;
 
 		local_group = cpu_isset(this_cpu, group->cpumask);
-		/* XXX: put a cpus allowed check */
+
+		/* Check whether all CPUs in the group is allowed to run on */
+		for_each_cpu_mask(i, group->cpumask) {
+			if (!cpu_isset(i, p->cpus_allowed))
+				continue;
+		}
 
 		/* Tally up the load of all CPUs in the group */
 		avg_load = 0;
Signature · 4 lines
"You have to understand, most of these people are not ready to be
unplugged. And many of them are no inert, so hopelessly dependent
on the system, that they will fight to protect it."
                                                         Morpheus

begin:vcard
fn:M.Baris Demiray
n:Demiray;M.Baris
org:Labris Teknoloji
adr:;;Teknokent Silikon Bina No:24 ODTU;Ankara;;06531;Turkey email;internet:[email protected] title:Yazilim Gelistirme Uzmani tel;work:+903122101490
tel;fax:+903122101492
x-mozilla-html:FALSE url:http://www.labristeknoloji.com
version:2.1
end:vcard

#1 lore raw
Quoted 9 lines · M.Baris Demiray

Hello,
following patch adds check for allowed CPUs into sched.c:find_idlest_group() -as told in comment line that had removed-. But, I have several questions about that comment.

Firstly, I've understood it as "Check whether process p is allowed to run on each CPU of to-be-found idlest group"; is that right?

Close.

Probably it would be better to take the intersection of (group->cpumask, p->cpus_allowed), and skip the group if the intersection is empty.

In addition to that, do a patch for find_idlest_cpu that skips cpus that aren't allowed. You needn't do the cpumask check each time round the loop, again just take the intersection of the group->cpumask and p->cpus_allowed, and loop over that.

Wanna do a patch for that?

Quoted 8 lines

If so, isn't it more appropriate to do check in find_idlest_cpu()? Because, we're only interested in CPUs that are in idlest group but doing a check in find_idlest_group() also checks for CPUs that are not in idlest group (since we're traversing all the groups in given domain). Checking this after finding the idlest group (in find_idlest_cpu() with ordinary call order as in sched_balance_self()) will save us from extra overhead.

Quoted 4 lines

Although I've questions in my mind, I'm sending a patch following that comment. Any explanation and comment on patch will be appreciated.

I don't think it does anything ;)

Quoted 24 lines

Regards.

--- linux-2.6.12-rc5-mm2/kernel/sched.c.orig 2005-06-05
12:31:04.000000000 +0000
+++ linux-2.6.12-rc5-mm2/kernel/sched.c 2005-06-05 16:49:49.000000000
+0000

kernel/sched.c renamed from kernel/sched.c.orig +6 -1
@@ -1040,7 +1040,12 @@
         int i;
 
         local_group = cpu_isset(this_cpu, group->cpumask);
-        /* XXX: put a cpus allowed check */
+
+        /* Check whether all CPUs in the group is allowed to run on */
+        for_each_cpu_mask(i, group->cpumask) {
+            if (!cpu_isset(i, p->cpus_allowed))
+                continue;
+        }
 
         /* Tally up the load of all CPUs in the group */
         avg_load = 0;
Signature · 3 lines
SUSE Labs, Novell Inc.

Send instant messages to your online friends http://au.messenger.yahoo.com 
#2 lore raw

Hello Nick,

Quoted 8 lines · Nick Piggin

M.Baris Demiray wrote:

[...]
Close.

Probably it would be better to take the intersection of (group->cpumask, p->cpus_allowed), and skip the group if the intersection is empty.

But, isn't it required for us to be allowed to run on _every_ CPU in that group. If we take intersection and continue if that's not empty, then there could be CPUs in group that are not allowed. Since any CPU then can be the idlest in that group we can be assigned to a CPU that is not allowed. Missing something?

Quoted 5 lines

In addition to that, do a patch for find_idlest_cpu that skips cpus that aren't allowed. You needn't do the cpumask check each time round the loop, again just take the intersection of the group->cpumask and p->cpus_allowed, and loop over that.

Will do it.

Quoted 1 line

Wanna do a patch for that?

Sure. But I'll wait EOT and do read something more to fill the gaps.

Quoted 3 lines

[...]

I don't think it does anything ;)

LOL. Hope next one'll do.

Meanwhile, what is the problem with that patch? Not traversing the CPUs correctly or continue;ing is wrong?

	for_each_cpu_mask(i, group->cpumask) {
		if (!cpu_isset(i, p->cpus_allowed))
			continue;
	}

Thanks for comments.

Quoted 1 line
Signature · 4 lines
"You have to understand, most of these people are not ready to be
unplugged. And many of them are no inert, so hopelessly dependent
on the system, that they will fight to protect it."
                                                        Morpheus

begin:vcard
fn:M.Baris Demiray
n:Demiray;M.Baris
org:Labris Teknoloji
adr:;;Teknokent Silikon Bina No:24 ODTU;Ankara;;06531;Turkey email;internet:[email protected] title:Yazilim Gelistirme Uzmani tel;work:+903122101490
tel;fax:+903122101492
x-mozilla-html:FALSE url:http://www.labristeknoloji.com
version:2.1
end:vcard

#3 lore raw
Quoted 15 lines · M.Baris Demiray

Hello Nick,

Quoted 8 lines · Nick Piggin

M.Baris Demiray wrote:

[...]
Close.

Probably it would be better to take the intersection of (group->cpumask, p->cpus_allowed), and skip the group if the intersection is empty.

Quoted 7 lines

But, isn't it required for us to be allowed to run on _every_ CPU in that group. If we take intersection and continue if that's not empty, then there could be CPUs in group that are not allowed. Since any CPU then can be the idlest in that group we can be assigned to a CPU that is not allowed. Missing something?

That should be OK. We basically aren't too interested in exactly which CPU it should go to, but just that it should go to that group.

If the group is determined to be the idlest, and there is a CPU that can run the task, then that's all we need.

Quoted 8 lines
Quoted 4 lines

[...]

I don't think it does anything ;)

LOL. Hope next one'll do.

Quoted 8 lines

Meanwhile, what is the problem with that patch? Not traversing the CPUs correctly or continue;ing is wrong?

    for_each_cpu_mask(i, group->cpumask) {
        if (!cpu_isset(i, p->cpus_allowed))
            continue;
    }

In Linux, the for_* macros actually *are* for loops. So that is that loop that your continue continues, and seeing as it is at the end of that for loop, it does nothing.

Thanks,
Nick

Signature · 3 lines
SUSE Labs, Novell Inc.

Send instant messages to your online friends http://au.messenger.yahoo.com 
#4 lore raw
Quoted 11 lines · Nick Piggin

M.Baris Demiray wrote:
[...]

Quoted 8 lines

But, isn't it required for us to be allowed to run on _every_ CPU in that group. If we take intersection and continue if that's not empty, then there could be CPUs in group that are not allowed. Since any CPU then can be the idlest in that group we can be assigned to a CPU that is not allowed. Missing something?

Quoted 6 lines

That should be OK. We basically aren't too interested in exactly which CPU it should go to, but just that it should go to that group.

If the group is determined to be the idlest, and there is a CPU that can run the task, then that's all we need.

OK. And also I missed the point that you requested a second check in find_idlest_cpu() which'll prevent an assignment to an unallowed CPU. That will solve the problem.

Quoted 11 lines

[...]

Quoted 9 lines

Meanwhile, what is the problem with that patch? Not traversing the CPUs correctly or continue;ing is wrong?

    for_each_cpu_mask(i, group->cpumask) {
        if (!cpu_isset(i, p->cpus_allowed))
            continue;
    }
Quoted 3 lines

In Linux, the for_* macros actually *are* for loops. So that is that loop that your continue continues, and seeing as it is at the end of that for loop, it does nothing.

Argh. I'll look at these.

Thanks Nick.

Quoted 3 lines

Thanks,
Nick

Signature · 4 lines
"You have to understand, most of these people are not ready to be
unplugged. And many of them are no inert, so hopelessly dependent
on the system, that they will fight to protect it."
                                                        Morpheus

begin:vcard
fn:M.Baris Demiray
n:Demiray;M.Baris
org:Labris Teknoloji
adr:;;Teknokent Silikon Bina No:24 ODTU;Ankara;;06531;Turkey email;internet:[email protected] title:Yazilim Gelistirme Uzmani tel;work:+903122101490
tel;fax:+903122101492
x-mozilla-html:FALSE url:http://www.labristeknoloji.com
version:2.1
end:vcard