Skip to contentSkip to conversation

ACPI: cap off P-state transition latency from buggy BIOSes

PATCH Applied Pallipadi, Venkatesh 5 participants 6 messages Mar 19, 2009 – Mar 28, 2009

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

PATCH 1/1 ACPI: cap off P-state transition latency from buggy BIOSes

Some BIOSes report very high frequency transition latency which are plainly wrong on CPus that can change frequency using native MSR interface.

One such system is IBM T42 (2327-8ZU) as reported by Owen Taylor and Rik van Riel.

cpufreq_ondemand driver uses this transition latency to come up with a reasonable sampling interval to sample CPU usage and with such high latency value, ondemand sampling interval ends up being very high (0.5 sec, in this particular case), resulting in performance impact due to slow response to increasing frequency.

Fix it by capping-off the transition latency to 20uS for native MSR based frequency transitions.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |   12 ++++++++++++
 1 file changed, 12 insertions(+)
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +12 -0
Index: linux-2.6/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
===================================================================
@@ -659,6 +659,18 @@ static int acpi_cpufreq_cpu_init(struct 
 			    perf->states[i].transition_latency * 1000;
 	}
 
+	/* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
+	if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
+	    policy->cpuinfo.transition_latency > 20 * 1000) {
+		static int print_once;
+		policy->cpuinfo.transition_latency = 20 * 1000;
+		if (!print_once) {
+			print_once = 1;
+			printk(KERN_INFO "Capping off P-state tranision latency"
+				" at 20 uS\n");
+		}
+	}
+
 	data->max_freq = perf->states[0].core_frequency * 1000;
 	/* table init */
 	for (i=0; i<perf->state_count; i++) {
#2 lore raw
Quoted 9 lines · Matthew Garrett
Quoted 4 lines · Venkatesh

Fix it by capping-off the transition latency to 20uS for native MSR based frequency transitions.

We've confirmed that this also helps on the X31.

Any chance of sending this to -stable after it hits mainline? It is the sort of stuff a huge number of thinkpad users will want in their kernels ASAP...

Signature · 4 lines
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh
#3 lore raw
Quoted 7 lines · Henrique de Moraes Holschuh
Quoted 6 lines · Matthew Garrett
Quoted 4 lines · Venkatesh

Fix it by capping-off the transition latency to 20uS for native MSR based frequency transitions.

Quoted 7 lines
Quoted 3 lines

We've confirmed that this also helps on the X31.

Any chance of sending this to -stable after it hits mainline? It is the sort of stuff a huge number of thinkpad users will want in their kernels ASAP...

I don't see why not.

Signature · 1 line
Matthew Garrett | [email protected]
#4 lore raw
In reply to Pallipadi, Venkatesh
Quoted 30 lines · Pallipadi, Venkatesh

Some BIOSes report very high frequency transition latency which are plainly wrong on CPus that can change frequency using native MSR interface.

One such system is IBM T42 (2327-8ZU) as reported by Owen Taylor and Rik van Riel.

cpufreq_ondemand driver uses this transition latency to come up with a reasonable sampling interval to sample CPU usage and with such high latency value, ondemand sampling interval ends up being very high (0.5 sec, in this particular case), resulting in performance impact due to slow response to increasing frequency.

Fix it by capping-off the transition latency to 20uS for native MSR based frequency transitions.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |   12 ++++++++++++
 1 file changed, 12 insertions(+)

Index: linux-2.6/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
===================================================================

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +0 -0
@@ -659,6 +659,18 @@ static int acpi_cpufreq_cpu_init(struct 
 			    perf->states[i].transition_latency * 1000;
 	}
 

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int acpi_cpufreq_cpu_init(struct · line 671

Quoted 10 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +10 -0
+	/* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
+	if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
+	    policy->cpuinfo.transition_latency > 20 * 1000) {
+		static int print_once;
+		policy->cpuinfo.transition_latency = 20 * 1000;
+		if (!print_once) {
+			print_once = 1;
+			printk(KERN_INFO "Capping off P-state tranision latency"
+				" at 20 uS\n");
+		}

btw.., in the next merge window we'll have printk_once():

f036be9: printk: introduce printk_once()

so then the above can be cleaned up to:

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int acpi_cpufreq_cpu_init(struct

Quoted 8 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +8 -0
+	/* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
+	if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
+	    policy->cpuinfo.transition_latency > 20 * 1000) {
+
+		policy->cpuinfo.transition_latency = 20 * 1000;
+		printk_once(KERN_INFO
+		    "Capping off P-state tranision latency at 20 uS\n");
+	}

Ingo

reviewer Applied
#5 lore raw
In reply to Pallipadi, Venkatesh
Quoted 36 lines · Venkatesh

Some BIOSes report very high frequency transition latency which are plainly wrong on CPus that can change frequency using native MSR interface.

One such system is IBM T42 (2327-8ZU) as reported by Owen Taylor and Rik van Riel.

cpufreq_ondemand driver uses this transition latency to come up with a reasonable sampling interval to sample CPU usage and with such high latency value, ondemand sampling interval ends up being very high (0.5 sec, in this particular case), resulting in performance impact due to slow response to increasing frequency.

Fix it by capping-off the transition latency to 20uS for native MSR based frequency transitions.

---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |   12 ++++++++++++
 1 file changed, 12 insertions(+)

Index: linux-2.6/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
===================================================================

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +6 -0
@@ -659,6 +659,18 @@ static int acpi_cpufreq_cpu_init(struct 
 			    perf->states[i].transition_latency * 1000;
 	}
 
+	/* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
+	if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
+	    policy->cpuinfo.transition_latency > 20 * 1000) {
+		static int print_once;
+		policy->cpuinfo.transition_latency = 20 * 1000;
+		if (!print_once) {

arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c static int acpi_cpufreq_cpu_init(struct

Quoted 10 lines
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c +6 -1
+			print_once = 1;
+			printk(KERN_INFO "Capping off P-state tranision latency"
+				" at 20 uS\n");
+		}
+	}
+
 	data->max_freq = perf->states[0].core_frequency * 1000;
 	/* table init */
 	for (i=0; i<perf->state_count; i++) {
--

Sort of makes you wonder if Windows is using this bogus info, or ignoring it...

applied.

thanks,
Len Brown, Intel Open Source Technology Center