diff mbox

cpufreq, store_scaling_governor requires policy->rwsem to be held for duration of changing governors [v2]

Message ID 53E9FBA4.2080609@redhat.com (mailing list archive)
State Not Applicable, archived
Headers show

Commit Message

Prarit Bhargava Aug. 12, 2014, 11:33 a.m. UTC 
On 08/12/2014 05:03 AM, Viresh Kumar wrote:
> On 7 August 2014 15:45, Viresh Kumar <viresh.kumar@linaro.org> wrote:
>> On 7 August 2014 15:42, Prarit Bhargava <prarit@redhat.com> wrote:
>>> That should have done it.  What are your CPUFREQ configs?
>>
>> You can check the same .config I attached last time for that :)
>>
>> CONFIG_CPU_FREQ=y
>> CONFIG_CPU_FREQ_GOV_COMMON=y
>> CONFIG_CPU_FREQ_STAT=y
>> CONFIG_CPU_FREQ_STAT_DETAILS=y
>> # CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set
>> # CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE is not set
>> # CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set
>> CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE=y
>> CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
>> CONFIG_CPU_FREQ_GOV_POWERSAVE=y
>> CONFIG_CPU_FREQ_GOV_USERSPACE=y
>> CONFIG_CPU_FREQ_GOV_ONDEMAND=y
>> CONFIG_CPU_FREQ_GOV_CONSERVATIVE=y

Viresh, sorry for my late reply -- I got distracted by another issue.

>>
>>
>> Anyway, has anybody tried to test what I have been trying now?
>> @Prarit: You can try that on your x86 box as well, which has a
>> single cluster or group of CPUs sharing clock line.
> 

Okay, this is what I have and I can reproduce this *easily* 100% of the time.

I've used your above config options and have enabled LOCKDEP.

In order to restore the locking, I've applied the following patch to the cpufreq
core (sorry for the cut-and-paste):


I do a

cat /sys/devices/system/cpu/cpu2/cpufreq/conservative/*
echo ondemand > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor

and then I immediately see the stack trace.

P.


> Ping!!
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Viresh Kumar Aug. 13, 2014, 7:39 a.m. UTC | #1 
On 12 August 2014 17:03, Prarit Bhargava <prarit@redhat.com> wrote:
> Okay, this is what I have and I can reproduce this *easily* 100% of the time.
>
> I've used your above config options and have enabled LOCKDEP.
>
> In order to restore the locking, I've applied the following patch to the cpufreq
> core (sorry for the cut-and-paste):
>
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index d9fdedd..dfda238 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -2192,9 +2192,7 @@ static int cpufreq_set_policy(struct cpufreq_policy *polic
>         /* end old governor */
>         if (old_gov) {
>                 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
> -               up_write(&policy->rwsem);
>                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
> -               down_write(&policy->rwsem);
>         }
>
>         /* start new governor */
> @@ -2203,9 +2201,7 @@ static int cpufreq_set_policy(struct cpufreq_policy *polic
>                 if (!__cpufreq_governor(policy, CPUFREQ_GOV_START))
>                         goto out;
>
> -               up_write(&policy->rwsem);
>                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
> -               down_write(&policy->rwsem);
>         }
>
>         /* new governor failed, so re-start old one */
>
>
> I've modified the acpi-cpufreq driver to include (sorry for the cut-and-paste)
>
> diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
> index b0c18ed..97653c3 100644
> --- a/drivers/cpufreq/acpi-cpufreq.c
> +++ b/drivers/cpufreq/acpi-cpufreq.c
> @@ -884,6 +884,9 @@ static struct freq_attr *acpi_cpufreq_attr[] = {
>  };
>
>  static struct cpufreq_driver acpi_cpufreq_driver = {
> +       .flags                  = CPUFREQ_STICKY |
> +                                       CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
> +                                       CPUFREQ_NEED_INITIAL_FREQ_CHECK,
>         .verify         = cpufreq_generic_frequency_table_verify,
>         .target_index   = acpi_cpufreq_target,
>         .bios_limit     = acpi_processor_get_bios_limit,
>
> I do a
>
> cat /sys/devices/system/cpu/cpu2/cpufreq/conservative/*
> echo ondemand > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
>
> and then I immediately see the stack trace.

What's your system configuration? How many clusters/cpus/etc..
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Prarit Bhargava Aug. 13, 2014, 9:58 a.m. UTC | #2 
On 08/13/2014 03:39 AM, Viresh Kumar wrote:
> On 12 August 2014 17:03, Prarit Bhargava <prarit@redhat.com> wrote:
>> Okay, this is what I have and I can reproduce this *easily* 100% of the time.
>>
>> I've used your above config options and have enabled LOCKDEP.
>>
>> In order to restore the locking, I've applied the following patch to the cpufreq
>> core (sorry for the cut-and-paste):
>>
>> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
>> index d9fdedd..dfda238 100644
>> --- a/drivers/cpufreq/cpufreq.c
>> +++ b/drivers/cpufreq/cpufreq.c
>> @@ -2192,9 +2192,7 @@ static int cpufreq_set_policy(struct cpufreq_policy *polic
>>         /* end old governor */
>>         if (old_gov) {
>>                 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
>> -               up_write(&policy->rwsem);
>>                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
>> -               down_write(&policy->rwsem);
>>         }
>>
>>         /* start new governor */
>> @@ -2203,9 +2201,7 @@ static int cpufreq_set_policy(struct cpufreq_policy *polic
>>                 if (!__cpufreq_governor(policy, CPUFREQ_GOV_START))
>>                         goto out;
>>
>> -               up_write(&policy->rwsem);
>>                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
>> -               down_write(&policy->rwsem);
>>         }
>>
>>         /* new governor failed, so re-start old one */
>>
>>
>> I've modified the acpi-cpufreq driver to include (sorry for the cut-and-paste)
>>
>> diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
>> index b0c18ed..97653c3 100644
>> --- a/drivers/cpufreq/acpi-cpufreq.c
>> +++ b/drivers/cpufreq/acpi-cpufreq.c
>> @@ -884,6 +884,9 @@ static struct freq_attr *acpi_cpufreq_attr[] = {
>>  };
>>
>>  static struct cpufreq_driver acpi_cpufreq_driver = {
>> +       .flags                  = CPUFREQ_STICKY |
>> +                                       CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
>> +                                       CPUFREQ_NEED_INITIAL_FREQ_CHECK,
>>         .verify         = cpufreq_generic_frequency_table_verify,
>>         .target_index   = acpi_cpufreq_target,
>>         .bios_limit     = acpi_processor_get_bios_limit,
>>
>> I do a
>>
>> cat /sys/devices/system/cpu/cpu2/cpufreq/conservative/*
>> echo ondemand > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
>>
>> and then I immediately see the stack trace.
> 
> What's your system configuration? How many clusters/cpus/etc..

Anywhere from 2-4 sockets, 8 - 240 cpus (depending on # of sockets), x86 arch.

P.
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Viresh Kumar Aug. 14, 2014, 4:19 a.m. UTC | #3 
On 13 August 2014 15:28, Prarit Bhargava <prarit@redhat.com> wrote:
> Anywhere from 2-4 sockets, 8 - 240 cpus (depending on # of sockets), x86 arch.

That's what. We know that it does happen on multi cluster systems
and I was reproducing it on a single cluster one. i.e. all CPUs share
clock line.
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index d9fdedd..dfda238 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -2192,9 +2192,7 @@  static int cpufreq_set_policy(struct cpufreq_policy *polic
        /* end old governor */
        if (old_gov) {
                __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
-               up_write(&policy->rwsem);
                __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
-               down_write(&policy->rwsem);
        }

        /* start new governor */
@@ -2203,9 +2201,7 @@  static int cpufreq_set_policy(struct cpufreq_policy *polic
                if (!__cpufreq_governor(policy, CPUFREQ_GOV_START))
                        goto out;

-               up_write(&policy->rwsem);
                __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
-               down_write(&policy->rwsem);
        }

        /* new governor failed, so re-start old one */


I've modified the acpi-cpufreq driver to include (sorry for the cut-and-paste)

diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
index b0c18ed..97653c3 100644
--- a/drivers/cpufreq/acpi-cpufreq.c
+++ b/drivers/cpufreq/acpi-cpufreq.c
@@ -884,6 +884,9 @@  static struct freq_attr *acpi_cpufreq_attr[] = {
 };

 static struct cpufreq_driver acpi_cpufreq_driver = {
+       .flags                  = CPUFREQ_STICKY |
+                                       CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
+                                       CPUFREQ_NEED_INITIAL_FREQ_CHECK,
        .verify         = cpufreq_generic_frequency_table_verify,
        .target_index   = acpi_cpufreq_target,
        .bios_limit     = acpi_processor_get_bios_limit,