Skip to contentSkip to conversation

OMAP3: decrease cpufreq transition latency

PATCH In review Mike Turquette 4 participants 6 messages Oct 29, 2009 – Nov 10, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 OMAP3: decrease cpufreq transition latency

Adjust OMAP3 frequency transition latency from 10,000,000uS to a more reasonable 300,000uS. This causes ondemand and conservative governors to sample CPU load more often resulting in more responsive behavior.

Tested on Android 2.6.29; using this value and conservative governor, CORE power consumption on Zoom2 was comparable to the old and unresponsive 10,000,000uS value while UI responsiveness was greatly improved.

This patch applies against Tony's omap-fixes branch.

---
 arch/arm/plat-omap/cpu-omap.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/plat-omap/cpu-omap.c +1 -1
index 1868c0d..341235c 100644
@@ -127,7 +127,7 @@ static int __init omap_cpu_init(struct cpufreq_policy *policy)
 	}
 
 	/* FIXME: what's the actual transition time? */
-	policy->cpuinfo.transition_latency = 10 * 1000 * 1000;
+	policy->cpuinfo.transition_latency = 300 * 1000;
 
 	return 0;
 }
Signature · 1 line
1.6.3.2
#1 lore raw
Quoted 20 lines · Mike Turquette

Adjust OMAP3 frequency transition latency from 10,000,000uS to a more reasonable 300,000uS. This causes ondemand and conservative governors to sample CPU load more often resulting in more responsive behavior.

Tested on Android 2.6.29; using this value and conservative governor, CORE power consumption on Zoom2 was comparable to the old and unresponsive 10,000,000uS value while UI responsiveness was greatly improved.

This patch applies against Tony's omap-fixes branch.

---
 arch/arm/plat-omap/cpu-omap.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/plat-omap/cpu-omap.c +0 -0
index 1868c0d..341235c 100644

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 129

Quoted 3 lines
arch/arm/plat-omap/cpu-omap.c +0 -0
@@ -127,7 +127,7 @@ static int __init omap_cpu_init(struct cpufreq_policy *policy)
 	}
 
 	/* FIXME: what's the actual transition time? */

Can probably drop this 'FIXME' comment now.

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 130

Quoted 2 lines
arch/arm/plat-omap/cpu-omap.c +1 -1
-	policy->cpuinfo.transition_latency = 10 * 1000 * 1000;
+	policy->cpuinfo.transition_latency = 300 * 1000;

otherwise,

Tony, can this still be queued for .32-rc?

Kevin

#2 lore raw
Quoted 24 lines · Kevin Hilman
Quoted 22 lines · Mike Turquette

Adjust OMAP3 frequency transition latency from 10,000,000uS to a more reasonable 300,000uS. This causes ondemand and conservative governors to sample CPU load more often resulting in more responsive behavior.

Tested on Android 2.6.29; using this value and conservative governor, CORE power consumption on Zoom2 was comparable to the old and unresponsive 10,000,000uS value while UI responsiveness was greatly improved.

This patch applies against Tony's omap-fixes branch.

---
 arch/arm/plat-omap/cpu-omap.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/plat-omap/cpu-omap.c +0 -0
index 1868c0d..341235c 100644
@@ -127,7 +127,7 @@ static int __init omap_cpu_init(struct cpufreq_policy *policy)
 	}
 
Quoted 3 lines

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 129

Quoted 1 line
arch/arm/plat-omap/cpu-omap.c +0 -0
 	/* FIXME: what's the actual transition time? */

Can probably drop this 'FIXME' comment now.

I chose not to drop the FIXME since the worst-case transition time is still unknown. At some point I'd like to revisit this number with something better supported by data from the hardware guys. This number was deemed "safe", but may not be optimal.

Mike

Quoted 11 lines

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 130

Quoted 2 lines
arch/arm/plat-omap/cpu-omap.c +1 -1
-	policy->cpuinfo.transition_latency = 10 * 1000 * 1000;
+	policy->cpuinfo.transition_latency = 300 * 1000;

otherwise,

Tony, can this still be queued for .32-rc?

Kevin

#3 lore raw
Quoted 28 lines · Mike Turquette
Quoted 17 lines · Kevin Hilman
Quoted 15 lines · Mike Turquette

Adjust OMAP3 frequency transition latency from 10,000,000uS to a more reasonable 300,000uS. This causes ondemand and conservative governors to sample CPU load more often resulting in more responsive behavior.

Tested on Android 2.6.29; using this value and conservative governor, CORE power consumption on Zoom2 was comparable to the old and unresponsive 10,000,000uS value while UI responsiveness was greatly improved.

This patch applies against Tony's omap-fixes branch.

---
 arch/arm/plat-omap/cpu-omap.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
Quoted 9 lines

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 128

Quoted 7 lines
arch/arm/plat-omap/cpu-omap.c +0 -0
index 1868c0d..341235c 100644
@@ -127,7 +127,7 @@ static int __init omap_cpu_init(struct cpufreq_policy *policy)
 	}
  	/* FIXME: what's the actual transition time? */

Can probably drop this 'FIXME' comment now.

Quoted 4 lines

I chose not to drop the FIXME since the worst-case transition time is still unknown. At some point I'd like to revisit this number with something better supported by data from the hardware guys. This number was deemed "safe", but may not be optimal.

OK by me.

Kevin

Quoted 12 lines
Quoted 12 lines

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 130

Quoted 3 lines
arch/arm/plat-omap/cpu-omap.c +1 -1
-	policy->cpuinfo.transition_latency = 10 * 1000 * 1000;
+	policy->cpuinfo.transition_latency = 300 * 1000;

otherwise,

Tony, can this still be queued for .32-rc?

Kevin

#4 lore raw
In reply to Kevin Hilman
Quoted 28 lines · Kevin Hilman
Quoted 20 lines · Mike Turquette

Adjust OMAP3 frequency transition latency from 10,000,000uS to a more reasonable 300,000uS. This causes ondemand and conservative governors to sample CPU load more often resulting in more responsive behavior.

Tested on Android 2.6.29; using this value and conservative governor, CORE power consumption on Zoom2 was comparable to the old and unresponsive 10,000,000uS value while UI responsiveness was greatly improved.

This patch applies against Tony's omap-fixes branch.

---
 arch/arm/plat-omap/cpu-omap.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/arm/plat-omap/cpu-omap.c +0 -0
index 1868c0d..341235c 100644

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 129

Quoted 3 lines
arch/arm/plat-omap/cpu-omap.c +0 -0
@@ -127,7 +127,7 @@ static int __init omap_cpu_init(struct cpufreq_policy *policy)
 	}
 
 	/* FIXME: what's the actual transition time? */

Can probably drop this 'FIXME' comment now.

Quoted 8 lines

arch/arm/plat-omap/cpu-omap.c static int __init omap_cpu_init(struct cpufreq_policy *policy) · line 130

Quoted 2 lines
arch/arm/plat-omap/cpu-omap.c +1 -1
-	policy->cpuinfo.transition_latency = 10 * 1000 * 1000;
+	policy->cpuinfo.transition_latency = 300 * 1000;

otherwise,

Tony, can this still be queued for .32-rc?

Sure, I'll put together one more omap-fixes series this week.

Regards,

Tony

#5 lore raw

PATCH 1/1 OMAP3: decrease cpufreq transition latency

In reply to Mike Turquette

This patch has been applied to the linux-omap
by youw fwiendly patch wobot.

Branch in linux-omap: omap-fixes

Initial commit ID (Likely to change): 2a817bb46c1cd125e2091cf2d7209ec7995adf84

PatchWorks
http://patchwork.kernel.org/patch/56520/

Git (Likely to change, and takes a while to get mirrored)
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=commit;h=2a817bb46c1cd125e2091cf2d7209ec7995adf84