Skip to contentSkip to conversation

Add P state driver for Intel Core Processors

PATCH In review [email protected] 2 participants 25 messages Feb 5, 2013 – Feb 6, 2013

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

cover Add P state driver for Intel Core Processors

From: Dirk Brandewie <[email protected]>

This driver implements a scaling driver with an internal governor for Intel Core processors. The driver follows the same model as the Transmeta scaling driver (longrun.c) and implements the setpolicy() instead of target(). Scaling drivers that implement setpolicy() are assmuned to implement internal governors by the cpufreq core. All the logic for selecting the current P state is contained within the driver no external governor is used by the cpufreq core.

At the moment only Intel SandyBridge processors are supported. As testing on SandyBridge+ processors is completed support will be added to the driver.

New sysfs files for controlling P state selection have been added to /sys/devices/system/cpu/intel_pstate/
max_perf_pct: limits the maximum P state that will be requested by the driver stated as a percentage of the avail performance.

min_perf_pct: limits the minimum P state that will be requested by the driver stated as a percentage of the avail performance.

no_turbo: limits the driver to selecting P states below the turbo frequency range.

The units for these for these files are purposely abstract and stated in terms of available performance and not frequency. In idea that frequency can be set to a single frequency is a fiction for Intel Core processors. Even if the scaling driver selects a single P state the actual frequency the processor will run at is selected by the processor

This patch set is based on Rafael's bleeding edge branch commit df9d5b7c Patch 1:
Viresh's patch for the sake of completeness

patch 2-5:
Fix issues related to scaling drivers that implement the setpolicy() interface instead of target()

patch 6:
Fix cpufreq_stats to handle the scaling driver not exposing a frequency table while processing CPU_DOWN_PREPARE notification
Patch 7:
The driver itself

Dirk Brandewie (6):
cpufreq: Retrieve current frequency from scaling drivers with
internal governors
cpufreq: Only query drivers that implement cpufreq_driver.target()
cpufreq: Do not track governor name for scaling drivers with internal
governors.
cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using
setpolicy
cpufreq_stats: do not remove sysfs files if frequency table is not
present
cpufreq/x86: Add P-state driver for sandy bridge.

Viresh Kumar (1):
cpufreq: Don't remove sysfs link for policy->cpu

 drivers/cpufreq/Kconfig.x86     |   18 +
 drivers/cpufreq/Makefile        |    1 +
 drivers/cpufreq/cpufreq.c       |   21 +-
 drivers/cpufreq/cpufreq_stats.c |    5 +
 drivers/cpufreq/intel_pstate.c  |  829 +++++++++++++++++++++++++++++++++++++++
 5 files changed, 870 insertions(+), 4 deletions(-)
 create mode 100644 drivers/cpufreq/intel_pstate.c
Signature · 1 line
1.7.7.6
#1 lore raw

PATCH 1/7 cpufreq: Don't remove sysfs link for policy->cpu

From: Viresh Kumar <[email protected]>

"cpufreq" directory in policy->cpu is never created using sysfs_create_link(), but using kobject_init_and_add(). And so we shouldn't call sysfs_remove_link() for policy->cpu(). sysfs stuff for policy->cpu is automatically removed when we call kobject_put() for dying policy.

---
 drivers/cpufreq/cpufreq.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
drivers/cpufreq/cpufreq.c +3 -2
index 9656420..2817c3c 100644
@@ -1058,7 +1058,9 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
 	cpus = cpumask_weight(data->cpus);
 	cpumask_clear_cpu(cpu, data->cpus);
 
-	if (unlikely((cpu == data->cpu) && (cpus > 1))) {
+	if (cpu != data->cpu) {
+		sysfs_remove_link(&dev->kobj, "cpufreq");
+	} else if (cpus > 1) {
 		/* first sibling now owns the new sysfs dir */
 		cpu_dev = get_cpu_device(cpumask_first(data->cpus));
 		sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
@@ -1083,7 +1085,6 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
 	pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
 	cpufreq_cpu_put(data);
 	unlock_policy_rwsem_write(cpu);
-	sysfs_remove_link(&dev->kobj, "cpufreq");
 
 	/* If cpu is last user of policy, free policy */
 	if (cpus == 1) {
Signature · 1 line
1.7.7.6
maintainer Tested-by Dirk Brandewie on v1 1/7
#2 lore raw

Re: [PATCH 1/7] cpufreq: Don't remove sysfs link for policy->cpu

Quoted 9 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Viresh Kumar <[email protected]>

"cpufreq" directory in policy->cpu is never created using sysfs_create_link(), but using kobject_init_and_add(). And so we shouldn't call sysfs_remove_link() for policy->cpu(). sysfs stuff for policy->cpu is automatically removed when we call kobject_put() for dying policy.

Its already included by Rafael, don't include it in your next version:

commit 73bf0fc2b03d1f4fdada0ec430dc20bfb089cfd5

Author: Viresh Kumar <[email protected]>
Date:   Tue Feb 5 22:21:14 2013 +0100

cpufreq: Don't remove sysfs link for policy->cpu

"cpufreq" directory in policy->cpu is never created using sysfs_create_link(), but using kobject_init_and_add(). And so we shouldn't call sysfs_remove_link() for policy->cpu(). sysfs stuff for policy->cpu is automatically removed when we call kobject_put() for dying policy.

#3 lore raw

PATCH 2/7 cpufreq: Retrieve current frequency from scaling drivers with internal governors

In reply to [email protected]

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement the cpufreq_driver.setpolicy() versus the cpufreq_driver.target() interface do not set policy->cur.

Normally policy->cur is set during the call to cpufreq_driver.target() when the frequnecy request is made by the governor.

If the scaling driver implements cpufreq_driver.setpolicy() and cpufreq_driver.get() interfaces use cpufreq_driver.get() to retrieve the current frequency.

From: Dirk Brandewie <[email protected]>
Date: 2013-02-04 18:44:40 -0800

cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

There is an additional reference added to the driver in cpufreq_add_dev() that is removed in__cpufreq_governor() if the driver implements target(). Remove the last reference when the driver implements setpolicy()

---
 drivers/cpufreq/cpufreq.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq.c +7 -0
index 2817c3c..96bc302 100644
@@ -1181,6 +1181,13 @@ unsigned int cpufreq_quick_get(unsigned int cpu)
 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 	unsigned int ret_freq = 0;
 
+	if (cpufreq_driver && cpufreq_driver->setpolicy &&
+		cpufreq_driver->get) {
+		ret_freq = cpufreq_driver->get(cpu);
+		cpufreq_cpu_put(policy);
+		return ret_freq;
+	}
+
 	if (policy) {
 		ret_freq = policy->cur;
 		cpufreq_cpu_put(policy);
Signature · 1 line
1.7.7.6
maintainer
#4 lore raw

Re: [PATCH 2/7] cpufreq: Retrieve current frequency from scaling drivers with internal governors

Quoted 5 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement the cpufreq_driver.setpolicy() versus the cpufreq_driver.target() interface do not set policy->cur.

Quoted 8 lines

From: Dirk Brandewie <[email protected]>
Date: 2013-02-04 18:44:40 -0800

cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

Wow!! Something happened here. :)

maintainer
#5 lore raw

Re: [PATCH 2/7] cpufreq: Retrieve current frequency from scaling drivers with internal governors

In reply to [email protected]
Quoted 5 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:
drivers/cpufreq/cpufreq.c +0 -0
index 2817c3c..96bc302 100644

drivers/cpufreq/cpufreq.c unsigned int cpufreq_quick_get(unsigned int cpu) · line 1190

Quoted 10 lines
drivers/cpufreq/cpufreq.c +7 -0
@@ -1181,6 +1181,13 @@ unsigned int cpufreq_quick_get(unsigned int cpu)
        struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
        unsigned int ret_freq = 0;
 
+       if (cpufreq_driver && cpufreq_driver->setpolicy &&
+               cpufreq_driver->get) {
+               ret_freq = cpufreq_driver->get(cpu);
+               cpufreq_cpu_put(policy);
+               return ret_freq;
+       }
+

This was my comment on the last version:

"You are required to do cpufreq_cpu_put() in this case too... Better do cpufreq_cpu_get() after your check."

It still applies.

#6 lore raw

Re: [PATCH 2/7] cpufreq: Retrieve current frequency from scaling drivers with internal governors

Quoted 9 lines · Viresh Kumar
Quoted 8 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:
drivers/cpufreq/cpufreq.c +0 -0
index 2817c3c..96bc302 100644
@@ -1181,6 +1181,13 @@ unsigned int cpufreq_quick_get(unsigned int cpu)
         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
         unsigned int ret_freq = 0;
 
Quoted 8 lines

drivers/cpufreq/cpufreq.c unsigned int cpufreq_quick_get(unsigned int cpu) · line 1190

Quoted 7 lines
drivers/cpufreq/cpufreq.c +7 -0
+       if (cpufreq_driver && cpufreq_driver->setpolicy &&
+               cpufreq_driver->get) {
+               ret_freq = cpufreq_driver->get(cpu);
+               cpufreq_cpu_put(policy);
+               return ret_freq;
+       }
+

How about this?

drivers/cpufreq/cpufreq.c +6 -1
index 2817c3c..9c0eac4 100644
@@ -1182,7 +1182,12 @@ unsigned int cpufreq_quick_get(unsigned int cpu)
 	unsigned int ret_freq = 0;
 
 	if (policy) {
-		ret_freq = policy->cur;
+		if (cpufreq_driver && cpufreq_driver->setpolicy &&
+			cpufreq_driver->get) {
+			ret_freq = cpufreq_driver->get(cpu);
+		} else {
+			ret_freq = policy->cur;
+		}
 		cpufreq_cpu_put(policy);
 	}
 
maintainer
#7 lore raw

Re: [PATCH 2/7] cpufreq: Retrieve current frequency from scaling drivers with internal governors

Quoted 10 lines · Dirk Brandewie

How about this?

drivers/cpufreq/cpufreq.c +0 -0
index 2817c3c..9c0eac4 100644
@@ -1182,7 +1182,12 @@ unsigned int cpufreq_quick_get(unsigned int cpu)
 
        unsigned int ret_freq = 0;
 

drivers/cpufreq/cpufreq.c unsigned int cpufreq_quick_get(unsigned int cpu) · line 1193

Quoted 10 lines
drivers/cpufreq/cpufreq.c +6 -1
        if (policy) {
-               ret_freq = policy->cur;
 
+               if (cpufreq_driver && cpufreq_driver->setpolicy &&
+                       cpufreq_driver->get) {
+                       ret_freq = cpufreq_driver->get(cpu);
+               } else {
+                       ret_freq = policy->cur;
+               }
                cpufreq_cpu_put(policy);

The problem is: You don't need to do get/put in your case at all.

#8 lore raw

Re: [PATCH 2/7] cpufreq: Retrieve current frequency from scaling drivers with internal governors

Quoted 14 lines · Viresh Kumar
Quoted 13 lines · Dirk Brandewie

How about this?

drivers/cpufreq/cpufreq.c +0 -1
index 2817c3c..9c0eac4 100644
@@ -1182,7 +1182,12 @@ unsigned int cpufreq_quick_get(unsigned int cpu)
 
         unsigned int ret_freq = 0;
 
         if (policy) {
-               ret_freq = policy->cur;
 
Quoted 7 lines

drivers/cpufreq/cpufreq.c unsigned int cpufreq_quick_get(unsigned int cpu) · line 1193

Quoted 7 lines
drivers/cpufreq/cpufreq.c +6 -0
+               if (cpufreq_driver && cpufreq_driver->setpolicy &&
+                       cpufreq_driver->get) {
+                       ret_freq = cpufreq_driver->get(cpu);
+               } else {
+                       ret_freq = policy->cur;
+               }
                 cpufreq_cpu_put(policy);
drivers/cpufreq/cpufreq.c +6 -1
index 2817c3c..7516b7d 100644
@@ -1178,9 +1178,14 @@ static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
  */
 unsigned int cpufreq_quick_get(unsigned int cpu)
 {
-	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+	struct cpufreq_policy *policy;
 	unsigned int ret_freq = 0;
 
+	if (cpufreq_driver && cpufreq_driver->setpolicy &&
+			cpufreq_driver->get)
+			return cpufreq_driver->get(cpu);
+
+	policy = cpufreq_cpu_get(cpu);
 	if (policy) {
 		ret_freq = policy->cur;
 		cpufreq_cpu_put(policy);
maintainer Acked-by
#9 lore raw

Re: [PATCH 2/7] cpufreq: Retrieve current frequency from scaling drivers with internal governors

Quoted 20 lines · Dirk Brandewie
drivers/cpufreq/cpufreq.c +6 -1
index 2817c3c..7516b7d 100644
@@ -1178,9 +1178,14 @@ static void cpufreq_out_of_sync(unsigned int cpu,
 unsigned int old_freq,
  */
 
 unsigned int cpufreq_quick_get(unsigned int cpu)
 {
-       struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+       struct cpufreq_policy *policy;
 
        unsigned int ret_freq = 0;
 
+       if (cpufreq_driver && cpufreq_driver->setpolicy &&
+                       cpufreq_driver->get)
+                       return cpufreq_driver->get(cpu);
+
+       policy = cpufreq_cpu_get(cpu);
Quoted 3 lines
        if (policy) {
                ret_freq = policy->cur;
                cpufreq_cpu_put(policy);
#10 lore raw

PATCH 3/7 cpufreq: Only query drivers that implement cpufreq_driver.target()

In reply to [email protected]

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement cpufreq_driver.setpolicy() have internal governors and may/will change the current operating frequency very frequently this will cause cpufreq_out_of_sync() to be called every time. Only call cpufreq_driver.get() for drivers that implement cpufreq_driver.target()

---
 drivers/cpufreq/cpufreq.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq.c +1 -1
index 96bc302..d8daa4b 100644
@@ -1787,7 +1787,7 @@ int cpufreq_update_policy(unsigned int cpu)
 
 	/* BIOS might change freq behind our back
 	  -> ask driver for current freq and notify governors about a change */
-	if (cpufreq_driver->get) {
+	if (cpufreq_driver->get && cpufreq_driver->target) {
 		policy.cur = cpufreq_driver->get(cpu);
 		if (!data->cur) {
 			pr_debug("Driver did not initialize current freq");
Signature · 1 line
1.7.7.6
maintainer
#11 lore raw

Re: [PATCH 3/7] cpufreq: Only query drivers that implement cpufreq_driver.target()

Quoted 18 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement cpufreq_driver.setpolicy() have internal governors and may/will change the current operating frequency very frequently this will cause cpufreq_out_of_sync() to be called every time. Only call cpufreq_driver.get() for drivers that implement cpufreq_driver.target()

---
 drivers/cpufreq/cpufreq.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index 96bc302..d8daa4b 100644

drivers/cpufreq/cpufreq.c int cpufreq_update_policy(unsigned int cpu) · line 1793

Quoted 8 lines
drivers/cpufreq/cpufreq.c +1 -1
@@ -1787,7 +1787,7 @@ int cpufreq_update_policy(unsigned int cpu)
 
        /* BIOS might change freq behind our back
          -> ask driver for current freq and notify governors about a change */
-       if (cpufreq_driver->get) {
+       if (cpufreq_driver->get && cpufreq_driver->target) {
                policy.cur = cpufreq_driver->get(cpu);
                if (!data->cur) {
                        pr_debug("Driver did not initialize current freq");

I am really not liking copy-pasting my older comments here :(

"This would mean policy->cur has a garbage value. I don't really know how would other routine behave on this. Would it make sense to make policy->cur zero atleast?
"

#12 lore raw

Re: [PATCH 3/7] cpufreq: Only query drivers that implement cpufreq_driver.target()

Quoted 30 lines · Viresh Kumar
Quoted 18 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement cpufreq_driver.setpolicy() have internal governors and may/will change the current operating frequency very frequently this will cause cpufreq_out_of_sync() to be called every time. Only call cpufreq_driver.get() for drivers that implement cpufreq_driver.target()

---
  drivers/cpufreq/cpufreq.c |    2 +-
  1 files changed, 1 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index 96bc302..d8daa4b 100644

drivers/cpufreq/cpufreq.c int cpufreq_update_policy(unsigned int cpu) · line 1793

Quoted 8 lines
drivers/cpufreq/cpufreq.c +1 -1
@@ -1787,7 +1787,7 @@ int cpufreq_update_policy(unsigned int cpu)
 
         /* BIOS might change freq behind our back
           -> ask driver for current freq and notify governors about a change */
-       if (cpufreq_driver->get) {
+       if (cpufreq_driver->get && cpufreq_driver->target) {
                 policy.cur = cpufreq_driver->get(cpu);
                 if (!data->cur) {
                         pr_debug("Driver did not initialize current freq");

I am really not liking copy-pasting my older comments here :(

Quoted 5 lines

"This would mean policy->cur has a garbage value. I don't really know how would other routine behave on this. Would it make sense to make policy->cur zero atleast?
"

The driver implements get() and will return a valid value but the other components that track the current frequency will not have been notified about any change so there is nothing to be out of sync with. There is no reason to call cpufreq_out_of_sync() where the driver being used implements an internal governor.

--Dirk

maintainer
#13 lore raw

Re: [PATCH 3/7] cpufreq: Only query drivers that implement cpufreq_driver.target()

Quoted 41 lines · Dirk Brandewie
Quoted 35 lines · Viresh Kumar
Quoted 28 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement cpufreq_driver.setpolicy() have internal governors and may/will change the current operating frequency very frequently this will cause cpufreq_out_of_sync() to be called every time. Only call cpufreq_driver.get() for drivers that implement cpufreq_driver.target()

---
  drivers/cpufreq/cpufreq.c |    2 +-
  1 files changed, 1 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index 96bc302..d8daa4b 100644
@@ -1787,7 +1787,7 @@ int cpufreq_update_policy(unsigned int cpu)
 
         /* BIOS might change freq behind our back
           -> ask driver for current freq and notify governors about a

change */

drivers/cpufreq/cpufreq.c +1 -1
-       if (cpufreq_driver->get) {
+       if (cpufreq_driver->get && cpufreq_driver->target) {
                 policy.cur = cpufreq_driver->get(cpu);
                 if (!data->cur) {
                         pr_debug("Driver did not initialize current
Quoted 1 line

freq");

I am really not liking copy-pasting my older comments here :(

Quoted 5 lines

"This would mean policy->cur has a garbage value. I don't really know how would other routine behave on this. Would it make sense to make policy->cur zero atleast?
"

Quoted 6 lines

The driver implements get() and will return a valid value but the other components that track the current frequency will not have been notified about any change so there is nothing to be out of sync with. There is no reason to call cpufreq_out_of_sync() where the driver being used implements an internal
governor.

Not sure if we are discussing the same issue. What i am saying is, with your patch we aren't calling following line:

Quoted 1 line
Quoted 1 line
Quoted 1 line
                 policy.cur = cpufreq_driver->get(cpu);

and policy was a local variable. Hence policy->cur is having a garbage value. policy->cur is used at multiple places in cpufreq.c . Please check everywhere if this garbage value doesn't break set_policy() type systems.

#14 lore raw

PATCH 4/7 cpufreq: Do not track governor name for scaling drivers with internal governors.

In reply to [email protected]

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement internal governors do not have governor structures assocaited with them. Only track the name of the governor associated with the CPU if the driver does not implement cpufreq_driver.setpolicy()

---
 drivers/cpufreq/cpufreq.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq.c +3 -1
index d8daa4b..622e282 100644
@@ -1050,7 +1050,9 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
 		__cpufreq_governor(data, CPUFREQ_GOV_STOP);
 
 #ifdef CONFIG_HOTPLUG_CPU
-	strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name,
+	if (!cpufreq_driver->setpolicy)
+		strncpy(per_cpu(cpufreq_cpu_governor, cpu),
+			data->governor->name,
 			CPUFREQ_NAME_LEN);
 #endif
 
Signature · 1 line
1.7.7.6
maintainer Acked-by on v1 4/7
#15 lore raw

Re: [PATCH 4/7] cpufreq: Do not track governor name for scaling drivers with internal governors.

Quoted 17 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

Scaling drivers that implement internal governors do not have governor structures assocaited with them. Only track the name of the governor associated with the CPU if the driver does not implement cpufreq_driver.setpolicy()

---
 drivers/cpufreq/cpufreq.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index d8daa4b..622e282 100644

drivers/cpufreq/cpufreq.c static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif · line 1056

Quoted 8 lines
drivers/cpufreq/cpufreq.c +3 -1
@@ -1050,7 +1050,9 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
                __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 
 #ifdef CONFIG_HOTPLUG_CPU
-       strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name,
+       if (!cpufreq_driver->setpolicy)
+               strncpy(per_cpu(cpufreq_cpu_governor, cpu),
+                       data->governor->name,
                        CPUFREQ_NAME_LEN);
#16 lore raw

PATCH 5/7 cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

In reply to [email protected]

From: Dirk Brandewie <[email protected]>

There is an additional reference added to the driver in cpufreq_add_dev() that is removed in__cpufreq_governor() if the driver implements target(). Remove the last reference when the driver implements setpolicy()

---
 drivers/cpufreq/cpufreq.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq.c +3 -0
index 622e282..d17477b 100644
@@ -1049,6 +1049,9 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
 	if (cpufreq_driver->target)
 		__cpufreq_governor(data, CPUFREQ_GOV_STOP);
 
+	if (cpufreq_driver->setpolicy)
+		cpufreq_cpu_put(data);
+
 #ifdef CONFIG_HOTPLUG_CPU
 	if (!cpufreq_driver->setpolicy)
 		strncpy(per_cpu(cpufreq_cpu_governor, cpu),
Signature · 1 line
1.7.7.6
maintainer
#17 lore raw

Re: [PATCH 5/7] cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

Quoted 17 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

There is an additional reference added to the driver in cpufreq_add_dev() that is removed in__cpufreq_governor() if the driver implements target(). Remove the last reference when the driver implements setpolicy()

---
 drivers/cpufreq/cpufreq.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index 622e282..d17477b 100644

drivers/cpufreq/cpufreq.c static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif · line 1053

Quoted 5 lines
drivers/cpufreq/cpufreq.c +2 -0
@@ -1049,6 +1049,9 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
        if (cpufreq_driver->target)
                __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 
+       if (cpufreq_driver->setpolicy)
+               cpufreq_cpu_put(data);

I don't understand this patch at all.. I grepped both cpufreq_cpu_get() & put() in bleeding-edge and found everything to be correct.

Can you please point me to the exact line numbers ?

#18 lore raw

Re: [PATCH 5/7] cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

Quoted 21 lines · Viresh Kumar
Quoted 20 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

There is an additional reference added to the driver in cpufreq_add_dev() that is removed in__cpufreq_governor() if the driver implements target(). Remove the last reference when the driver implements setpolicy()

---
  drivers/cpufreq/cpufreq.c |    3 +++
  1 files changed, 3 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index 622e282..d17477b 100644
@@ -1049,6 +1049,9 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
         if (cpufreq_driver->target)
                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 
Quoted 8 lines

drivers/cpufreq/cpufreq.c static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif · line 1053

Quoted 2 lines
drivers/cpufreq/cpufreq.c +2 -0
+       if (cpufreq_driver->setpolicy)
+               cpufreq_cpu_put(data);

I don't understand this patch at all.. I grepped both cpufreq_cpu_get() & put() in bleeding-edge and found everything to be correct.

Can you please point me to the exact line numbers ?

Line 878 in cpufreq_add_dev()

--Dirk

maintainer
#19 lore raw

Re: [PATCH 5/7] cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

Quoted 36 lines · Dirk Brandewie
Quoted 35 lines · Viresh Kumar
Quoted 25 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

There is an additional reference added to the driver in cpufreq_add_dev() that is removed in__cpufreq_governor() if the

driver implements target(). Remove the last reference when the driver implements setpolicy()

---
  drivers/cpufreq/cpufreq.c |    3 +++
  1 files changed, 3 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index 622e282..d17477b 100644
@@ -1049,6 +1049,9 @@ static int __cpufreq_remove_dev(struct device *dev,

struct subsys_interface *sif

         if (cpufreq_driver->target)
                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);

drivers/cpufreq/cpufreq.c static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif · line 1053

Quoted 2 lines
drivers/cpufreq/cpufreq.c +2 -0
+       if (cpufreq_driver->setpolicy)
+               cpufreq_cpu_put(data);

I don't understand this patch at all.. I grepped both cpufreq_cpu_get() & put()
in bleeding-edge and found everything to be correct.

Quoted 4 lines
Quoted 2 lines

Can you please point me to the exact line numbers ?

Line 878 in cpufreq_add_dev()

Following is line 878:

	for_each_online_cpu(sibling) {
		struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
		if (cp && cpumask_test_cpu(cpu, cp->related_cpus))
			return cpufreq_add_policy_cpu(cpu, sibling, dev);
	}

How is this related to your patch?

#20 lore raw

Re: [PATCH 5/7] cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

Quoted 44 lines · Viresh Kumar
Quoted 36 lines · Dirk Brandewie
Quoted 35 lines · Viresh Kumar
Quoted 25 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

There is an additional reference added to the driver in cpufreq_add_dev() that is removed in__cpufreq_governor() if the

driver implements target(). Remove the last reference when the driver implements setpolicy()

---
   drivers/cpufreq/cpufreq.c |    3 +++
   1 files changed, 3 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq.c +0 -0
index 622e282..d17477b 100644
@@ -1049,6 +1049,9 @@ static int __cpufreq_remove_dev(struct device *dev,

struct subsys_interface *sif

          if (cpufreq_driver->target)
                  __cpufreq_governor(data, CPUFREQ_GOV_STOP);

drivers/cpufreq/cpufreq.c static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif · line 1053

Quoted 2 lines
drivers/cpufreq/cpufreq.c +2 -0
+       if (cpufreq_driver->setpolicy)
+               cpufreq_cpu_put(data);

I don't understand this patch at all.. I grepped both cpufreq_cpu_get() & put()
in bleeding-edge and found everything to be correct.

Quoted 4 lines
Quoted 2 lines

Can you please point me to the exact line numbers ?

Line 878 in cpufreq_add_dev()

Following is line 878:

Quoted 8 lines
	for_each_online_cpu(sibling) {
		struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
		if (cp && cpumask_test_cpu(cpu, cp->related_cpus))
			return cpufreq_add_policy_cpu(cpu, sibling, dev);
	}

How is this related to your patch?

our files are clearly out of sync :-) The code in cpufreq_add_dev() is

#ifdef CONFIG_SMP
	/* check whether a different CPU already registered this
	 * CPU because it is in the same boat. */
	policy = cpufreq_cpu_get(cpu);
	if (unlikely(policy)) {
		cpufreq_cpu_put(policy);
		return 0;
	}

The reference added by this cpufreq_cpu_get() is finally dropped in __cpufreq_remove_dev() with the call to __cpufreq_governor()

	if (driver->target)
		__cpufreq_governor(data, CPUFREQ_GOV_STOP);

Without this change I hang at:

		pr_debug("waiting for dropping of refcount\n");
		wait_for_completion(cmp);

--Dirk

maintainer
#21 lore raw

Re: [PATCH 5/7] cpufreq: balance out cpufreq_cpu_{get,put} for scaling drivers using setpolicy

Quoted 1 line · Dirk Brandewie

our files are clearly out of sync :-) The code in cpufreq_add_dev() is

I was at latest bleeding-edge then, but i still have below code in my HEAD.

Quoted 11 lines
#ifdef CONFIG_SMP
        /* check whether a different CPU already registered this
         * CPU because it is in the same boat. */
        policy = cpufreq_cpu_get(cpu);
        if (unlikely(policy)) {
                cpufreq_cpu_put(policy);
                return 0;
        }

The reference added by this cpufreq_cpu_get() is finally dropped in __cpufreq_remove_dev() with the call to __cpufreq_governor()

__cpufreq_governor() is not calling cpufreq_cpu_put() at all and is dropping reference for the governor only, not for policy.

Quoted 6 lines
        if (driver->target)
                __cpufreq_governor(data, CPUFREQ_GOV_STOP);

Without this change I hang at:

                pr_debug("waiting for dropping of refcount\n");
                wait_for_completion(cmp);

Below should have fixed it for you i believe.

	pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
	cpufreq_cpu_put(data);
	unlock_policy_rwsem_write(cpu);

--
viresh

#22 lore raw

PATCH 6/7 cpufreq_stats: do not remove sysfs files if frequency table is not present

In reply to [email protected]

From: Dirk Brandewie <[email protected]>

The sysfs files for cpufreq_stats are created in cpufreq_stats_create_table() called from cpufreq_stat_notifier_policy() when a policy is added to the cpu. cpufreq_stats_create_table() will not be called if the scaling driver does not export a frequency table to cpufreq. Use the same fence on tear down.

---
 drivers/cpufreq/cpufreq_stats.c |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq_stats.c +5 -0
index 572124c..fb001e2 100644
@@ -185,6 +185,11 @@ static void cpufreq_stats_free_table(unsigned int cpu)
 static void cpufreq_stats_free_sysfs(unsigned int cpu)
 {
 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+	struct cpufreq_frequency_table *table;
+
+	table = cpufreq_frequency_get_table(cpu);
+	if (!table)
+		return;
 	if (policy && !policy_is_shared(policy)) {
 		pr_debug("%s: Free sysfs stat\n", __func__);
 		sysfs_remove_group(&policy->kobj, &stats_attr_group);
Signature · 1 line
1.7.7.6
maintainer
#23 lore raw

Re: [PATCH 6/7] cpufreq_stats: do not remove sysfs files if frequency table is not present

Quoted 18 lines · On Tue, Feb 5, 2013 at 11:54 PM, <[email protected]> wrote:

From: Dirk Brandewie <[email protected]>

The sysfs files for cpufreq_stats are created in cpufreq_stats_create_table() called from cpufreq_stat_notifier_policy() when a policy is added to the cpu. cpufreq_stats_create_table() will not be called if the scaling driver does not export a frequency table to cpufreq. Use the same fence on tear down.

---
 drivers/cpufreq/cpufreq_stats.c |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)
drivers/cpufreq/cpufreq_stats.c +0 -0
index 572124c..fb001e2 100644

drivers/cpufreq/cpufreq_stats.c static void cpufreq_stats_free_table(unsigned int cpu) · line 192

Quoted 8 lines
drivers/cpufreq/cpufreq_stats.c +5 -0
@@ -185,6 +185,11 @@ static void cpufreq_stats_free_table(unsigned int cpu)
 static void cpufreq_stats_free_sysfs(unsigned int cpu)
 {
        struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+       struct cpufreq_frequency_table *table;
+
+       table = cpufreq_frequency_get_table(cpu);
+       if (!table)
+               return;

Because we aren't using table later on, i would write it as:

Quoted 2 lines
drivers/cpufreq/cpufreq_stats.c +2 -0
+       if (!cpufreq_frequency_get_table(cpu))
+               return;
#24 lore raw

PATCH 7/7 cpufreq/x86: Add P-state driver for sandy bridge.

In reply to [email protected]

From: Dirk Brandewie <[email protected]>

Add a P-state driver for the Intel Sandy bridge processor. In cpufreq terminology this driver implements a scaling driver with an internal governor.

When built into the the kernel this driver will be the preferred scaling driver for Sandy bridge processors.

In addition to the interfaces provided by the cpufreq subsystem for controlling scaling drivers. The user may control the behavior of the driver via three sysfs files located in "/sys/devices/system/cpu/intel_pstate".

max_perf_pct: limits the maximum P state that will be requested by the driver stated as a percentage of the avail performance.

min_perf_pct: limits the minimum P state that will be requested by the driver stated as a percentage of the avail performance.

no_turbo: limits the driver to selecting P states below the turbo frequency range.

---
 drivers/cpufreq/Kconfig.x86    |   18 +
 drivers/cpufreq/Makefile       |    1 +
 drivers/cpufreq/intel_pstate.c |  829 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 848 insertions(+), 0 deletions(-)
 create mode 100644 drivers/cpufreq/intel_pstate.c
drivers/cpufreq/Kconfig.x86 +18 -0
index 7227cd7..6aa7053 100644
@@ -2,6 +2,24 @@
 # x86 CPU Frequency scaling drivers
 #
 
+config X86_INTEL_PSTATE
+       tristate "Intel P state control"
+       depends on X86
+       help
+          This driver provides a P state for Intel core processors.
+	  The driver implements an internal governor and will become
+          the scaling driver and governor for Sandy bridge processors.
+
+	  When this driver is enabled it will become the perferred
+          scaling driver for Sandy bridge processors.
+
+	  Note: This driver should be built with the same settings as
+	  the other scaling drivers configured into the system
+	  (module/built-in) in order for the driver to register itself
+	  as the scaling driver on the system.
+
+	  If in doubt, say N.
+
 config X86_PCC_CPUFREQ
 	tristate "Processor Clocking Control interface driver"
 	depends on ACPI && ACPI_PROCESSOR
drivers/cpufreq/Makefile +1 -0
index f7aed54..39ad489 100644
@@ -40,6 +40,7 @@ obj-$(CONFIG_X86_SPEEDSTEP_SMI)		+= speedstep-smi.o
 obj-$(CONFIG_X86_SPEEDSTEP_CENTRINO)	+= speedstep-centrino.o
 obj-$(CONFIG_X86_P4_CLOCKMOD)		+= p4-clockmod.o
 obj-$(CONFIG_X86_CPUFREQ_NFORCE2)	+= cpufreq-nforce2.o
+obj-$(CONFIG_X86_INTEL_PSTATE)		+= intel_pstate.o
 
 ##################################################################################
 # ARM SoC drivers
drivers/cpufreq/intel_pstate.c new file +829 -0

829 lines not on this page · raw message

Signature · 1 line
1.7.7.6