Skip to contentSkip to conversation

sched: Print sched_group::__cpu_power in sched_domain_debug.

PATCH New Gautham R Shenoy 3 participants 13 messages Mar 30, 2009 – Apr 16, 2009

lore
Share & tools
lore
#1 lore raw

PATCH 1/1 sched: Print sched_group::__cpu_power in sched_domain_debug.

In reply to Gautham R Shenoy

If the user changes the value of the sched_mc/smt_power_savings sysfs tunable, it'll trigger a rebuilding of the whole sched_domain tree, with the SD_POWERSAVINGS_BALANCE flag set at certain levels.

As a result, there would be a change in the __cpu_power of sched_groups in the sched_domain hierarchy.

Print the __cpu_power values for each sched_group in sched_domain_debug to help verify this change and correlate it with the change in the load-balancing behavior.

---

 kernel/sched.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
kernel/sched.c +2 -1
index 706517c..fbac83b 100644
@@ -7363,7 +7363,8 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
 		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-		printk(KERN_CONT " %s", str);
+		printk(KERN_CONT " %s (__cpu_power = %d)", str,
+						group->__cpu_power);
 
 		group = group->next;
 	} while (group != sd->groups);
#0 lore raw

PATCH 1/1 sched: Print sched_group::__cpu_power in sched_domain_debug.

A simple patch to print the value of __cpu_power in sched_domain_debug.

---

Gautham R Shenoy (1):
sched: Print sched_group::__cpu_power in sched_domain_debug.

 kernel/sched.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
Signature · 2 lines
Thanks and Regards
gautham.
#2 lore raw

[tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

In reply to Gautham R Shenoy
Commit-ID:  46e0bb9c12f4bab539736f1714cbf16600f681ec
Gitweb:     http://git.kernel.org/tip/46e0bb9c12f4bab539736f1714cbf16600f681ec
Author:     Gautham R Shenoy <[email protected]>
AuthorDate: Mon, 30 Mar 2009 10:25:20 +0530
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 1 Apr 2009 17:58:03 +0200

sched: Print sched_group::__cpu_power in sched_domain_debug

Impact: extend debug info /proc/sched_debug

If the user changes the value of the sched_mc/smt_power_savings sysfs tunable, it'll trigger a rebuilding of the whole sched_domain tree, with the SD_POWERSAVINGS_BALANCE flag set at certain levels.

As a result, there would be a change in the __cpu_power of sched_groups in the sched_domain hierarchy.

Print the __cpu_power values for each sched_group in sched_domain_debug to help verify this change and correlate it with the change in the load-balancing behavior.

LKML-Reference: <[email protected]>

---
 kernel/sched.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
kernel/sched.c +2 -1
index 8d1bdbe..6234d10 100644
@@ -6963,7 +6963,8 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
 		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-		printk(KERN_CONT " %s", str);
+		printk(KERN_CONT " %s (__cpu_power = %d)", str,
+						group->__cpu_power);
 
 		group = group->next;
 	} while (group != sd->groups);
#3 lore raw

Re: [tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

kernel/sched.c static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, · line 6967

Quoted 9 lines
kernel/sched.c +2 -1
@@ -6963,7 +6963,8 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
                cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
                cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-               printk(KERN_CONT " %s", str);
+               printk(KERN_CONT " %s (__cpu_power = %d)", str,
+                                               group->__cpu_power);

This has added a lot of clutter to the console log during boot (especially on large systems).

Here is the start of the diff output comparing old and new console messages on a 16 cpu machine:

77c77
< groups: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
---

kernel/sched.c static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, · line 6968

Quoted 1 line
kernel/sched.c +0 -0
  groups: 0 (__cpu_power = 1024) 1 (__cpu_power = 1024) 2 (__cpu_power = 1024) 3 (__cpu_power = 1024) 4 (__cpu_power = 1024) 5 (__cpu_power = 1024) 6 (__cpu_power = 1024) 7 (__cpu_power = 1024) 8 (

__cpu_power = 1024) 9 (__cpu_power = 1024) 10 (__cpu_power = 1024) 11 (__cpu_power = 1024) 12 (__cpu_power = 1024) 13 (__cpu_power = 1024) 14 (__cpu_power = 1024) 15 (__cpu_power = 1024)
80c80
< groups: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0
---

kernel/sched.c static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, · line 6969

Quoted 1 line
kernel/sched.c +0 -0
  groups: 1 (__cpu_power = 1024) 2 (__cpu_power = 1024) 3 (__cpu_power = 1024) 4 (__cpu_power = 1024) 5 (__cpu_power = 1024) 6 (__cpu_power = 1024) 7 (__cpu_power = 1024) 8 (__cpu_power = 1024) 9 (

__cpu_power = 1024) 10 (__cpu_power = 1024) 11 (__cpu_power = 1024) 12 (__cpu_power = 1024) 13 (__cpu_power = 1024) 14 (__cpu_power = 1024) 15 (__cpu_power = 1024) 0 (__cpu_power = 1024)

continues as each group is reported.

-Tony

#4 lore raw

Re: [tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

Quoted 16 lines · Tony Luck

kernel/sched.c static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, · line 6967

Quoted 9 lines
kernel/sched.c +2 -1
@@ -6963,7 +6963,8 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
                cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
                cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-               printk(KERN_CONT " %s", str);
+               printk(KERN_CONT " %s (__cpu_power = %d)", str,
+                                               group->__cpu_power);

This has added a lot of clutter to the console log during boot (especially on large systems).

Here is the start of the diff output comparing old and new console messages on a 16 cpu machine:

Quoted 8 lines

77c77
< groups: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
---

kernel/sched.c static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, · line 6968

Quoted 1 line
kernel/sched.c +0 -0
  groups: 0 (__cpu_power = 1024) 1 (__cpu_power = 1024) 2 (__cpu_power = 1024) 3 (__cpu_power = 1024) 4 (__cpu_power = 1024) 5 (__cpu_power = 1024) 6 (__cpu_power = 1024) 7 (__cpu_power = 1024) 8 (

__cpu_power = 1024) 9 (__cpu_power = 1024) 10 (__cpu_power = 1024) 11 (__cpu_power = 1024) 12 (__cpu_power = 1024) 13 (__cpu_power = 1024) 14 (__cpu_power = 1024) 15 (__cpu_power = 1024)
80c80

indeed ...

I think we should skip the printout in the default (power==1024) case. Gautham?

Ingo

#5 lore raw

Re: [tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

Quoted 28 lines · Ingo Molnar
Quoted 16 lines · Tony Luck

kernel/sched.c static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, · line 6967

Quoted 9 lines
kernel/sched.c +2 -1
@@ -6963,7 +6963,8 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
                cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
                cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-               printk(KERN_CONT " %s", str);
+               printk(KERN_CONT " %s (__cpu_power = %d)", str,
+                                               group->__cpu_power);

This has added a lot of clutter to the console log during boot (especially on large systems).

Here is the start of the diff output comparing old and new console messages on a 16 cpu machine:

Quoted 8 lines

77c77
< groups: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
---

kernel/sched.c static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, · line 6968

Quoted 1 line
kernel/sched.c +0 -0
  groups: 0 (__cpu_power = 1024) 1 (__cpu_power = 1024) 2 (__cpu_power = 1024) 3 (__cpu_power = 1024) 4 (__cpu_power = 1024) 5 (__cpu_power = 1024) 6 (__cpu_power = 1024) 7 (__cpu_power = 1024) 8 (

__cpu_power = 1024) 9 (__cpu_power = 1024) 10 (__cpu_power = 1024) 11 (__cpu_power = 1024) 12 (__cpu_power = 1024) 13 (__cpu_power = 1024) 14 (__cpu_power = 1024) 15 (__cpu_power = 1024)
80c80

Quoted 4 lines

indeed ...

I think we should skip the printout in the default (power==1024) case. Gautham?

Makes sense. Patch appended.

Quoted 2 lines

Ingo

-->

sched: Avoid printing sched_group::__cpu_power for default case.

From: Gautham R Shenoy <[email protected]>

The following commit produces a messy dmesg output while attempting to print the sched_group::__cpu_power for each group in the sched_domain hierarchy.

commit 46e0bb9c12f4bab539736f1714cbf16600f681ec
Author: Gautham R Shenoy <[email protected]>
Date: Mon Mar 30 10:25:20 2009 +0530
sched: Print sched_group::__cpu_power in sched_domain_debug

Fix this by avoid printing the __cpu_power for default cases. (i.e, __cpu_power == SCHED_LOAD_SCALE).

---

 kernel/sched.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
kernel/sched.c +3 -2
index 681d4ae..0584e04 100644
@@ -7467,8 +7467,9 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
 		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-		printk(KERN_CONT " %s (__cpu_power = %d)", str,
-						group->__cpu_power);
+		if (group->__cpu_power != SCHED_LOAD_SCALE)
+			printk(KERN_CONT " %s (__cpu_power = %d)", str,
+							group->__cpu_power);
 
 		group = group->next;
 	} while (group != sd->groups);
Signature · 2 lines
Thanks and Regards
gautham
#6 lore raw

[tip:sched/urgent] sched: Avoid printing sched_group::__cpu_power for default case

Commit-ID:  fa0dacb0eec2f531e457058c05e3fe8b7606ed08
Gitweb:     http://git.kernel.org/tip/fa0dacb0eec2f531e457058c05e3fe8b7606ed08
Author:     Gautham R Shenoy <[email protected]>
AuthorDate: Tue, 14 Apr 2009 09:09:36 +0530
Committer:  Ingo Molnar <[email protected]>
CommitDate: Tue, 14 Apr 2009 14:01:00 +0200

sched: Avoid printing sched_group::__cpu_power for default case

Impact: reduce syslog clutter

Commit 46e0bb9c12f4 ("sched: Print sched_group::__cpu_power in sched_domain_debug") produces a messy dmesg output while attempting to print the sched_group::__cpu_power for each group in the sched_domain hierarchy.

Fix this by avoid printing the __cpu_power for default cases. (i.e, __cpu_power == SCHED_LOAD_SCALE).

LKML-Reference: <[email protected]>

---
 kernel/sched.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
kernel/sched.c +3 -2
index e90e70e..ebd574c 100644
@@ -7367,8 +7367,9 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
 		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-		printk(KERN_CONT " %s (__cpu_power = %d)", str,
-						group->__cpu_power);
+		if (group->__cpu_power != SCHED_LOAD_SCALE)
+			printk(KERN_CONT " %s (__cpu_power = %d)", str,
+							group->__cpu_power);
 
 		group = group->next;
 	} while (group != sd->groups);
#7 lore raw

RE: [tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

In reply to Gautham R Shenoy

- printk(KERN_CONT " %s (__cpu_power = %d)", str,
- group->__cpu_power);
+ if (group->__cpu_power != SCHED_LOAD_SCALE)
+ printk(KERN_CONT " %s (__cpu_power = %d)", str,
+ group->__cpu_power);

Much quieter ... but perhaps a little too quiet?
Is this what you want? Now the console output looks like this:

CPU0 attaching sched-domain:
domain 0: span 0-15 level CPU
groups:
CPU1 attaching sched-domain:
domain 0: span 0-15 level CPU
groups:
CPU2 attaching sched-domain:
domain 0: span 0-15 level CPU
groups:
...

instead of the original:

CPU0 attaching sched-domain:
domain 0: span 0-15 level CPU
groups: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
CPU1 attaching sched-domain:
domain 0: span 0-15 level CPU
groups: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0
CPU2 attaching sched-domain:
domain 0: span 0-15 level CPU
groups: 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0 1
...

Either we don't need the empty "groups:" line, or we should still list the cpus in the group? I'm not really sure what information you are trying to convey here.

-Tony

#8 lore raw

Re: [tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

Quoted 33 lines · Tony
(quoted diff) +3 -2
-               printk(KERN_CONT " %s (__cpu_power = %d)", str,
-                                               group->__cpu_power);
+               if (group->__cpu_power != SCHED_LOAD_SCALE)
+                       printk(KERN_CONT " %s (__cpu_power = %d)", str,
+                                                       group->__cpu_power);

Much quieter ... but perhaps a little too quiet?
Is this what you want? Now the console output looks like this:

CPU0 attaching sched-domain:
domain 0: span 0-15 level CPU
groups:
CPU1 attaching sched-domain:
domain 0: span 0-15 level CPU
groups:
CPU2 attaching sched-domain:
domain 0: span 0-15 level CPU
groups:
...

instead of the original:

CPU0 attaching sched-domain:
domain 0: span 0-15 level CPU
groups: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
CPU1 attaching sched-domain:
domain 0: span 0-15 level CPU
groups: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0
CPU2 attaching sched-domain:
domain 0: span 0-15 level CPU
groups: 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0 1
...

Quoted 3 lines

Either we don't need the empty "groups:" line, or we should still list the cpus in the group? I'm not really sure what information you are trying to convey here.

We should be listing the cpus in the group. We should not be listing the __cpu_power of the group, if the __cpu_power has the default value. In the patch that I sent this morning, I made a mistake by making dependent the printing of both cpus as well as __cpu_power, on the if condition which checks if __cpu_power is default or not.

               if (group->__cpu_power != SCHED_LOAD_SCALE)
                       printk(KERN_CONT " %s (__cpu_power = %d)", str,
                                                       group->__cpu_power);

Unfortunately for me, it's not the first goof up I've been involved in today. Please find the updated patch below.

---->
sched: Avoid printing sched_group::__cpu_power for default case.

From: Gautham R Shenoy <[email protected]>

The following commit produces a messy dmesg output while attempting to print the sched_group::__cpu_power for each group in the sched_domain hierarchy.

	commit 46e0bb9c12f4bab539736f1714cbf16600f681ec
	Author: Gautham R Shenoy <[email protected]>
	Date:   Mon Mar 30 10:25:20 2009 +0530
	sched: Print sched_group::__cpu_power in sched_domain_debug

Fix this by avoid printing the __cpu_power for default cases. (i.e, __cpu_power == SCHED_LOAD_SCALE).

---

 kernel/sched.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
kernel/sched.c +3 -1
index 681d4ae..db2df70 100644
@@ -7467,7 +7467,9 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
 		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-		printk(KERN_CONT " %s (__cpu_power = %d)", str,
+		printk(KERN_CONT " %s", str);
+		if (group->__cpu_power != SCHED_LOAD_SCALE)
+			printk(KERN_CONT " (__cpu_power = %d)",
 						group->__cpu_power);
 
 		group = group->next;
Signature · 2 lines
Thanks and Regards
gautham
Acked-by
#9 lore raw

RE: [tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

-               printk(KERN_CONT " %s (__cpu_power = %d)", str,
+               printk(KERN_CONT " %s", str);
+               if (group->__cpu_power != SCHED_LOAD_SCALE)
+                       printk(KERN_CONT " (__cpu_power = %d)",
                                                group->__cpu_power);

That looks better.

-Tony

#10 lore raw

Re: [tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug

Quoted 9 lines · Luck, Tony
kernel/sched.c +3 -1
-               printk(KERN_CONT " %s (__cpu_power = %d)", str,
+               printk(KERN_CONT " %s", str);
+               if (group->__cpu_power != SCHED_LOAD_SCALE)
+                       printk(KERN_CONT " (__cpu_power = %d)",
                                                group->__cpu_power);

That looks better.

Thanks, picked up this variant!

Ingo

#11 lore raw

[tip:sched/urgent] sched: Avoid printing sched_group::__cpu_power for default case

In reply to Gautham R Shenoy
Commit-ID:  728fc565bdfc98dc7086fdcccb3f1e1edf195b20
Gitweb:     http://git.kernel.org/tip/728fc565bdfc98dc7086fdcccb3f1e1edf195b20
Author:     Gautham R Shenoy <[email protected]>
AuthorDate: Tue, 14 Apr 2009 09:09:36 +0530
Committer:  Ingo Molnar <[email protected]>
CommitDate: Tue, 14 Apr 2009 20:39:40 +0200

sched: Avoid printing sched_group::__cpu_power for default case

Impact: reduce syslog clutter

Commit 46e0bb9c12f4 ("sched: Print sched_group::__cpu_power in sched_domain_debug") produces a messy dmesg output while attempting to print the sched_group::__cpu_power for each group in the sched_domain hierarchy.

Fix this by avoid printing the __cpu_power for default cases. (i.e, __cpu_power == SCHED_LOAD_SCALE).

LKML-Reference: <[email protected]>

---
 kernel/sched.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)
kernel/sched.c +6 -2
index e90e70e..b902e58 100644
@@ -7367,8 +7367,12 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
 		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-		printk(KERN_CONT " %s (__cpu_power = %d)", str,
-						group->__cpu_power);
+
+		printk(KERN_CONT " %s", str);
+		if (group->__cpu_power != SCHED_LOAD_SCALE) {
+			printk(KERN_CONT " (__cpu_power = %d)",
+				group->__cpu_power);
+		}
 
 		group = group->next;
 	} while (group != sd->groups);
#12 lore raw

[tip:sched/urgent] sched: Avoid printing sched_group::__cpu_power for default case

In reply to Gautham R Shenoy
Commit-ID:  381512cf3d27f63f7a45b1bbe7d2d609c2ea3b74
Gitweb:     http://git.kernel.org/tip/381512cf3d27f63f7a45b1bbe7d2d609c2ea3b74
Author:     Gautham R Shenoy <[email protected]>
AuthorDate: Tue, 14 Apr 2009 09:09:36 +0530
Committer:  Ingo Molnar <[email protected]>
CommitDate: Fri, 17 Apr 2009 00:46:05 +0200

sched: Avoid printing sched_group::__cpu_power for default case

Commit 46e0bb9c12f4 ("sched: Print sched_group::__cpu_power in sched_domain_debug") produces a messy dmesg output while attempting to print the sched_group::__cpu_power for each group in the sched_domain hierarchy.

Fix this by avoid printing the __cpu_power for default cases. (i.e, __cpu_power == SCHED_LOAD_SCALE).

[ Impact: reduce syslog clutter ]

LKML-Reference: <[email protected]>

---
 kernel/sched.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)
kernel/sched.c +6 -2
index e90e70e..b902e58 100644
@@ -7367,8 +7367,12 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
 
 		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
-		printk(KERN_CONT " %s (__cpu_power = %d)", str,
-						group->__cpu_power);
+
+		printk(KERN_CONT " %s", str);
+		if (group->__cpu_power != SCHED_LOAD_SCALE) {
+			printk(KERN_CONT " (__cpu_power = %d)",
+				group->__cpu_power);
+		}
 
 		group = group->next;
 	} while (group != sd->groups);