The patch titled
cpufreq: ondemand vs suspend deadlock fix
has been added to the -mm tree. Its filename is
cpufreq-ondemand-vs-suspend-deadlock-fix.patch
See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find
out what to do about this
------------------------------------------------------
Subject: cpufreq: ondemand vs suspend deadlock fix
From: Venkatesh Pallipadi <venkatesh.pallipadi@xxxxxxxxx>
Root caused this to a deadlock in cpufreq and ondemand. The deadlock is
due to non-existant ordering between cpu_hotplug lock and dbs_mutex.
Basically a race condition between cpu_down() and do_dbs_timer().
cpu_down() flow:
1 cpu_down() for CPU 1
2 Takes the cpu_hotplug lock
3 Calls pre-down notifiers
4 notifier handler in cpufreq calls cpufreq_driver_target
5 cpufreq_driver_target calls cpu_hotplu lock/unlock
It is OK as cpu_hotplug lock is recusive for same process
6 CPU 1 goes down
7 CPU 0 calls post down notifiers for CPU 1
8 notifier handler in cpufreq calls governor event for stop
9 this ondemand governor routine takes dbs_mutex
Basically cpu_hotplug lock being taken before dbs_mutex in this path.
There is another event that gets triggered periodically in do_dbs_timer().
This runs in the context of ondemand workqueue and it takes dbs_mutex first
and takes cpu_hotplug later, inside __cpufreq_driver_target() call. This
ordering conflicts with mutex ordering in cpu_down and causes a deadlock.
Attached patch fixes the issue for both ondemand and conservative governors.
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@xxxxxxxxx>
Cc: Dave Jones <davej@xxxxxxxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxx>
---
drivers/cpufreq/cpufreq_conservative.c | 12 ++++++++++++
drivers/cpufreq/cpufreq_ondemand.c | 12 ++++++++++++
2 files changed, 24 insertions(+)
diff -puN drivers/cpufreq/cpufreq_conservative.c~cpufreq-ondemand-vs-suspend-deadlock-fix drivers/cpufreq/cpufreq_conservative.c
--- a/drivers/cpufreq/cpufreq_conservative.c~cpufreq-ondemand-vs-suspend-deadlock-fix
+++ a/drivers/cpufreq/cpufreq_conservative.c
@@ -72,6 +72,14 @@ static DEFINE_PER_CPU(struct cpu_dbs_inf
static unsigned int dbs_enable; /* number of CPUs using this policy */
+/*
+ * DEADLOCK ALERT! There is a ordering requirement between cpu_hotplug
+ * lock and dbs_mutex. cpu_hotplug lock should always be held before
+ * dbs_mutex. If any function that can potentially take cpu_hotplug lock
+ * (like __cpufreq_driver_target()) is being called with dbs_mutex taken, then
+ * cpu_hotplug lock should be taken before that. Note that cpu_hotplug lock
+ * is recursive for the same process. -Venki
+ */
static DEFINE_MUTEX (dbs_mutex);
static DECLARE_WORK (dbs_work, do_dbs_timer, NULL);
@@ -414,12 +422,14 @@ static void dbs_check_cpu(int cpu)
static void do_dbs_timer(void *data)
{
int i;
+ lock_cpu_hotplug();
mutex_lock(&dbs_mutex);
for_each_online_cpu(i)
dbs_check_cpu(i);
schedule_delayed_work(&dbs_work,
usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
mutex_unlock(&dbs_mutex);
+ unlock_cpu_hotplug();
}
static inline void dbs_timer_init(void)
@@ -514,6 +524,7 @@ static int cpufreq_governor_dbs(struct c
break;
case CPUFREQ_GOV_LIMITS:
+ lock_cpu_hotplug();
mutex_lock(&dbs_mutex);
if (policy->max < this_dbs_info->cur_policy->cur)
__cpufreq_driver_target(
@@ -524,6 +535,7 @@ static int cpufreq_governor_dbs(struct c
this_dbs_info->cur_policy,
policy->min, CPUFREQ_RELATION_L);
mutex_unlock(&dbs_mutex);
+ unlock_cpu_hotplug();
break;
}
return 0;
diff -puN drivers/cpufreq/cpufreq_ondemand.c~cpufreq-ondemand-vs-suspend-deadlock-fix drivers/cpufreq/cpufreq_ondemand.c
--- a/drivers/cpufreq/cpufreq_ondemand.c~cpufreq-ondemand-vs-suspend-deadlock-fix
+++ a/drivers/cpufreq/cpufreq_ondemand.c
@@ -71,6 +71,14 @@ static DEFINE_PER_CPU(struct cpu_dbs_inf
static unsigned int dbs_enable; /* number of CPUs using this policy */
+/*
+ * DEADLOCK ALERT! There is a ordering requirement between cpu_hotplug
+ * lock and dbs_mutex. cpu_hotplug lock should always be held before
+ * dbs_mutex. If any function that can potentially take cpu_hotplug lock
+ * (like __cpufreq_driver_target()) is being called with dbs_mutex taken, then
+ * cpu_hotplug lock should be taken before that. Note that cpu_hotplug lock
+ * is recursive for the same process. -Venki
+ */
static DEFINE_MUTEX (dbs_mutex);
static DECLARE_WORK (dbs_work, do_dbs_timer, NULL);
@@ -363,12 +371,14 @@ static void dbs_check_cpu(int cpu)
static void do_dbs_timer(void *data)
{
int i;
+ lock_cpu_hotplug();
mutex_lock(&dbs_mutex);
for_each_online_cpu(i)
dbs_check_cpu(i);
queue_delayed_work(dbs_workq, &dbs_work,
usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
mutex_unlock(&dbs_mutex);
+ unlock_cpu_hotplug();
}
static inline void dbs_timer_init(void)
@@ -469,6 +479,7 @@ static int cpufreq_governor_dbs(struct c
break;
case CPUFREQ_GOV_LIMITS:
+ lock_cpu_hotplug();
mutex_lock(&dbs_mutex);
if (policy->max < this_dbs_info->cur_policy->cur)
__cpufreq_driver_target(
@@ -479,6 +490,7 @@ static int cpufreq_governor_dbs(struct c
this_dbs_info->cur_policy,
policy->min, CPUFREQ_RELATION_L);
mutex_unlock(&dbs_mutex);
+ unlock_cpu_hotplug();
break;
}
return 0;
_
Patches currently in -mm which might be from venkatesh.pallipadi@xxxxxxxxx are
git-acpi.patch
cpufreq-ondemand-vs-suspend-deadlock-fix.patch
hpet-rtc-emulation-add-watchdog-timer.patch
-
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
