time: Move clock_was_set_seq update before updating shadow-timekeeper
authorJohn Stultz <[email protected]>
Thu, 11 Jun 2015 22:54:53 +0000 (15:54 -0700)
committerThomas Gleixner <[email protected]>
Fri, 12 Jun 2015 08:56:20 +0000 (10:56 +0200)
It was reported that 868a3e915f7f5eba (hrtimer: Make offset
update smarter) was causing timer problems after suspend/resume.

The problem with that change is the modification to
clock_was_set_seq in timekeeping_update is done prior to
mirroring the time state to the shadow-timekeeper. Thus the
next time we do update_wall_time() the updated sequence is
overwritten by whats in the shadow copy.

This patch moves the shadow-timekeeper mirroring to the end
of the function, after all updates have been made, so all data
is kept in sync.

(This patch also affects the update_fast_timekeeper calls which
were also problematically done prior to the mirroring).

Reported-and-tested-by: Jeremiah Mahler <[email protected]>
Signed-off-by: John Stultz <[email protected]>
Cc: Preeti U Murthy <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Viresh Kumar <[email protected]>
Cc: Marcelo Tosatti <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
kernel/time/timekeeping.c

index 90ed5db67c1dc3cab6405af2ccab664d062a9eb2..849b93265904f8a5fb5e9d894a9b38f085c52890 100644 (file)
@@ -585,15 +585,19 @@ static void timekeeping_update(struct timekeeper *tk, unsigned int action)
        update_vsyscall(tk);
        update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
 
-       if (action & TK_MIRROR)
-               memcpy(&shadow_timekeeper, &tk_core.timekeeper,
-                      sizeof(tk_core.timekeeper));
-
        update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono);
        update_fast_timekeeper(&tk->tkr_raw,  &tk_fast_raw);
 
        if (action & TK_CLOCK_WAS_SET)
                tk->clock_was_set_seq++;
+       /*
+        * The mirroring of the data to the shadow-timekeeper needs
+        * to happen last here to ensure we don't over-write the
+        * timekeeper structure on the next update with stale data
+        */
+       if (action & TK_MIRROR)
+               memcpy(&shadow_timekeeper, &tk_core.timekeeper,
+                      sizeof(tk_core.timekeeper));
 }
 
 /**