Skip to contentSkip to conversation

cpufreq: Fix wrong time unit conversion

PATCH Applied Frederic Weisbecker 4 participants 9 messages Sep 7, 2013 – Sep 8, 2013 CPU frequency scaling framework

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cpufreq: Fix wrong time unit conversion

The time spent by a CPU under a given frequency is stored in jiffies unit in the cpu var cpufreq_stats_table->time_in_state[i], i being the index of the frequency.

This is what is displayed in the following file on the right column:

     cat /sys/devices/system/cpu/cpuX/cpufreq/stats/time_in_state
     2301000 19835820
     2300000 3172
     [...]

Now cpufreq converts this jiffies unit delta to clock_t before returning it to the user as in the above file. And that conversion is achieved using the API cputime64_to_clock_t().

Although it accidentally works on traditional tick based cputime accounting, where cputime_t maps directly to jiffies, it doesn't work with other types of cputime accounting such as CONFIG_VIRT_CPU_ACCOUNTING_* where cputime_t can map to nsecs or any granularity preffered by the architecture.

For example we get a buggy zero delta on full dyntick configurations:

     cat /sys/devices/system/cpu/cpuX/cpufreq/stats/time_in_state
     2301000 0
     2300000 0
     [...]

Fix this with using the proper jiffies_64_t to clock_t conversion.

Cc Carsten Emde <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Clark Williams <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Paul E. McKenney <[email protected]>
---
 drivers/cpufreq/cpufreq_stats.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq_stats.c +1 -1
index d37568c..10e6138 100644
@@ -81,7 +81,7 @@ static ssize_t show_time_in_state(struct cpufreq_policy *policy, char *buf)
 	for (i = 0; i < stat->state_num; i++) {
 		len += sprintf(buf + len, "%u %llu\n", stat->freq_table[i],
 			(unsigned long long)
-			cputime64_to_clock_t(stat->time_in_state[i]));
+			jiffies_64_to_clock_t(stat->time_in_state[i]));
 	}
 	return len;
 }
Signature · 1 line
1.7.5.4
Acked-by on v1 1/1
#1 lore raw
Quoted 29 lines · Frederic Weisbecker

The time spent by a CPU under a given frequency is stored in jiffies unit in the cpu var cpufreq_stats_table->time_in_state[i], i being the index of the frequency.

This is what is displayed in the following file on the right column:

     cat /sys/devices/system/cpu/cpuX/cpufreq/stats/time_in_state
     2301000 19835820
     2300000 3172
     [...]

Now cpufreq converts this jiffies unit delta to clock_t before returning it to the user as in the above file. And that conversion is achieved using the API cputime64_to_clock_t().

Although it accidentally works on traditional tick based cputime accounting, where cputime_t maps directly to jiffies, it doesn't work with other types of cputime accounting such as CONFIG_VIRT_CPU_ACCOUNTING_* where cputime_t can map to nsecs or any granularity preffered by the architecture.

For example we get a buggy zero delta on full dyntick configurations:

     cat /sys/devices/system/cpu/cpuX/cpufreq/stats/time_in_state
     2301000 0
     2300000 0
     [...]

Fix this with using the proper jiffies_64_t to clock_t conversion.

Quoted 8 lines
Cc Carsten Emde <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Clark Williams <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Paul E. McKenney <[email protected]>

Good catch!

Quoted 20 lines
---
 drivers/cpufreq/cpufreq_stats.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq_stats.c +1 -1
index d37568c..10e6138 100644
@@ -81,7 +81,7 @@ static ssize_t show_time_in_state(struct cpufreq_policy *policy, char *buf)
 	for (i = 0; i < stat->state_num; i++) {
 		len += sprintf(buf + len, "%u %llu\n", stat->freq_table[i],
 			(unsigned long long)
-			cputime64_to_clock_t(stat->time_in_state[i]));
+			jiffies_64_to_clock_t(stat->time_in_state[i]));
 	}
 	return len;
 }
Signature · 1 line
1.7.5.4
#2 lore raw
Quoted 39 lines · Paul E. McKenney
Quoted 29 lines · Frederic Weisbecker

The time spent by a CPU under a given frequency is stored in jiffies unit in the cpu var cpufreq_stats_table->time_in_state[i], i being the index of the frequency.

This is what is displayed in the following file on the right column:

     cat /sys/devices/system/cpu/cpuX/cpufreq/stats/time_in_state
     2301000 19835820
     2300000 3172
     [...]

Now cpufreq converts this jiffies unit delta to clock_t before returning it to the user as in the above file. And that conversion is achieved using the API cputime64_to_clock_t().

Although it accidentally works on traditional tick based cputime accounting, where cputime_t maps directly to jiffies, it doesn't work with other types of cputime accounting such as CONFIG_VIRT_CPU_ACCOUNTING_* where cputime_t can map to nsecs or any granularity preffered by the architecture.

For example we get a buggy zero delta on full dyntick configurations:

     cat /sys/devices/system/cpu/cpuX/cpufreq/stats/time_in_state
     2301000 0
     2300000 0
     [...]

Fix this with using the proper jiffies_64_t to clock_t conversion.

Quoted 8 lines
Cc Carsten Emde <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Clark Williams <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Paul E. McKenney <[email protected]>
Quoted 3 lines

Good catch!

Applied, will push for 3.12-rc1.

Thanks Frederic!

Quoted 19 lines
Quoted 18 lines
---
 drivers/cpufreq/cpufreq_stats.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq_stats.c +1 -1
index d37568c..10e6138 100644
@@ -81,7 +81,7 @@ static ssize_t show_time_in_state(struct cpufreq_policy *policy, char *buf)
 	for (i = 0; i < stat->state_num; i++) {
 		len += sprintf(buf + len, "%u %llu\n", stat->freq_table[i],
 			(unsigned long long)
-			cputime64_to_clock_t(stat->time_in_state[i]));
+			jiffies_64_to_clock_t(stat->time_in_state[i]));
 	}
 	return len;
 }
Signature · 2 lines
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
#5 lore raw
In reply to Andreas Schwab
Quoted 3 lines · Andreas Schwab

See also <http://permalink.gmane.org/gmane.linux.kernel.cpufreq/11672>.

Andreas.

Ah I missed it.

Raphael, if it is not too late, may be you could still set Andreas as the original author?

Thanks.

Quoted 5 lines
Signature · 4 lines
-- 
Andreas Schwab, [email protected]
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#7 lore raw
Quoted 11 lines · Rafael J. Wysocki
Quoted 8 lines · Frederic Weisbecker
Quoted 3 lines · Andreas Schwab

See also <http://permalink.gmane.org/gmane.linux.kernel.cpufreq/11672>.

Andreas.

Ah I missed it.

Raphael, if it is not too late, may be you could still set Andreas as the original author?

Done.

BTW, CCing power management patches to [email protected] helps a bit, because I can see them in patchwork then and that makes overlooking them or forgetting about them much less likely.

Thanks,
Rafael

#8 lore raw
Quoted 9 lines · Rafael J. Wysocki
Quoted 8 lines · Rafael J. Wysocki
Quoted 7 lines · Frederic Weisbecker
Quoted 3 lines · Andreas Schwab

See also <http://permalink.gmane.org/gmane.linux.kernel.cpufreq/11672>.

Andreas.

Ah I missed it.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Raphael, if it is not too late, may be you could still set Andreas as the original author?

Done.

BTW, CCing power management patches to [email protected] helps a bit, because I can see them in patchwork then and that makes overlooking them or forgetting about them much less likely.

Ok, I'll do that next time.

Thanks.