[PATCH] x86_64: Move ondemand timer into own work queue
authorAndi Kleen <ak@suse.de>
Mon, 8 May 2006 13:17:31 +0000 (15:17 +0200)
committerLinus Torvalds <torvalds@g5.osdl.org>
Mon, 8 May 2006 16:34:56 +0000 (09:34 -0700)
Taking the cpu hotplug semaphore in a normal events workqueue
is unsafe because other tasks can wait for any workqueues with
it hold. This results in a deadlock.

Move the DBS timer into its own work queue which is not
affected by other work queue flushes to avoid this.

Has been acked by Venkatesh.

Cc: venkatesh.pallipadi@intel.com
Cc: cpufreq@lists.linux.org.uk
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
drivers/cpufreq/cpufreq_ondemand.c

index 956d121cb16151c2c3f564c4ccabde21250afbac..3e6ffcaa5af4c2a10941c2ad6d30478aead32897 100644 (file)
@@ -74,6 +74,8 @@ static unsigned int dbs_enable;       /* number of CPUs using this policy */
 static DEFINE_MUTEX (dbs_mutex);
 static DECLARE_WORK    (dbs_work, do_dbs_timer, NULL);
 
+static struct workqueue_struct *dbs_workq;
+
 struct dbs_tuners {
        unsigned int sampling_rate;
        unsigned int sampling_down_factor;
@@ -364,23 +366,29 @@ static void do_dbs_timer(void *data)
        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));
+       queue_delayed_work(dbs_workq, &dbs_work,
+                          usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
        mutex_unlock(&dbs_mutex);
 }
 
 static inline void dbs_timer_init(void)
 {
        INIT_WORK(&dbs_work, do_dbs_timer, NULL);
-       schedule_delayed_work(&dbs_work,
-                       usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
+       if (!dbs_workq)
+               dbs_workq = create_singlethread_workqueue("ondemand");
+       if (!dbs_workq) {
+               printk(KERN_ERR "ondemand: Cannot initialize kernel thread\n");
+               return;
+       }
+       queue_delayed_work(dbs_workq, &dbs_work,
+                          usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
        return;
 }
 
 static inline void dbs_timer_exit(void)
 {
-       cancel_delayed_work(&dbs_work);
-       return;
+       if (dbs_workq)
+               cancel_rearming_delayed_workqueue(dbs_workq, &dbs_work);
 }
 
 static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
@@ -489,8 +497,12 @@ static int __init cpufreq_gov_dbs_init(void)
 
 static void __exit cpufreq_gov_dbs_exit(void)
 {
-       /* Make sure that the scheduled work is indeed not running */
-       flush_scheduled_work();
+       /* Make sure that the scheduled work is indeed not running.
+          Assumes the timer has been cancelled first. */
+       if (dbs_workq) {
+               flush_workqueue(dbs_workq);
+               destroy_workqueue(dbs_workq);
+       }
 
        cpufreq_unregister_governor(&cpufreq_gov_dbs);
 }