sched: Print sched_group::__cpu_power in sched_domain_debug.
PATCH 1/1 sched: Print sched_group::__cpu_power in sched_domain_debug.
In reply to Gautham R ShenoyIf 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
@@ -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);
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.
[tip:sched/urgent] sched: Print sched_group::__cpu_power in sched_domain_debug
In reply to Gautham R ShenoyCommit-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
@@ -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);
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
-0groups: 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
-0groups: 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
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
-0groups: 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
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
-0groups: 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
@@ -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
[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
@@ -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);
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
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
@@ -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
Luck, Tony
Acked-byRE: [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
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.
Acked-by Tony Luck <[email protected]>
Thanks, picked up this variant!
Ingo
[tip:sched/urgent] sched: Avoid printing sched_group::__cpu_power for default case
In reply to Gautham R ShenoyCommit-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
@@ -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);
[tip:sched/urgent] sched: Avoid printing sched_group::__cpu_power for default case
In reply to Gautham R ShenoyCommit-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
@@ -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);