hrtimer: Remove HRTIMER_STATE_MIGRATE
authorOleg Nesterov <oleg@redhat.com>
Thu, 11 Jun 2015 12:46:44 +0000 (14:46 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Thu, 18 Jun 2015 22:09:56 +0000 (00:09 +0200)
I do not understand HRTIMER_STATE_MIGRATE. Unless I am totally
confused it looks buggy and simply unneeded.

migrate_hrtimer_list() sets it to keep hrtimer_active() == T, but this
is not enough: this can fool, say, hrtimer_is_queued() in
dequeue_signal().

Can't migrate_hrtimer_list() simply use HRTIMER_STATE_ENQUEUED?
This fixes the race and we can kill STATE_MIGRATE.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: ktkhai@parallels.com
Cc: rostedt@goodmis.org
Cc: juri.lelli@gmail.com
Cc: pang.xunlei@linaro.org
Cc: wanpeng.li@linux.intel.com
Cc: umgwanakikbuti@gmail.com
Link: http://lkml.kernel.org/r/20150611124743.072387650@infradead.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
include/linux/hrtimer.h
kernel/time/hrtimer.c

index 3f82a7edc03d198409686c5587c1927e58c38dec..2f9e57d3d126c71869f97b036f26cfba7ba9d59b 100644 (file)
@@ -70,17 +70,13 @@ enum hrtimer_restart {
  * the handling of the timer.
  *
  * The HRTIMER_STATE_ENQUEUED bit is always or'ed to the current state
- * to preserve the HRTIMER_STATE_CALLBACK in the above scenario. This
- * also affects HRTIMER_STATE_MIGRATE where the preservation is not
- * necessary. HRTIMER_STATE_MIGRATE is cleared after the timer is
- * enqueued on the new cpu.
+ * to preserve the HRTIMER_STATE_CALLBACK in the above scenario.
  *
  * All state transitions are protected by cpu_base->lock.
  */
 #define HRTIMER_STATE_INACTIVE 0x00
 #define HRTIMER_STATE_ENQUEUED 0x01
 #define HRTIMER_STATE_CALLBACK 0x02
-#define HRTIMER_STATE_MIGRATE  0x04
 
 /**
  * struct hrtimer - the basic hrtimer structure
index 278d4b36fd9436f48279296535b386b67c54109b..b1b795e5e0b107ffec290e950a5e482e03e8167f 100644 (file)
@@ -1508,11 +1508,11 @@ static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
                debug_deactivate(timer);
 
                /*
-                * Mark it as STATE_MIGRATE not INACTIVE otherwise the
+                * Mark it as ENQUEUED not INACTIVE otherwise the
                 * timer could be seen as !active and just vanish away
                 * under us on another CPU
                 */
-               __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
+               __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
                timer->base = new_base;
                /*
                 * Enqueue the timers on the new cpu. This does not
@@ -1523,9 +1523,6 @@ static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
                 * event device.
                 */
                enqueue_hrtimer(timer, new_base);
-
-               /* Clear the migration state bit */
-               timer->state &= ~HRTIMER_STATE_MIGRATE;
        }
 }