vmstat: explicitly schedule per-cpu work on the CPU we need it to run on

commit 176bed1d upstream.

The vmstat code uses "schedule_delayed_work_on()" to do the initial
startup of the delayed work on the right CPU, but then once it was
started it would use the non-cpu-specific "schedule_delayed_work()" to
re-schedule it on that CPU.

That just happened to schedule it on the same CPU historically (well, in
almost all situations), but the code _requires_ this work to be per-cpu,
and should say so explicitly rather than depend on the non-cpu-specific
scheduling to schedule on the current CPU.

The timer code is being changed to not be as single-minded in always
running things on the calling CPU.

See also commit 874bbfe6 ("workqueue: make sure delayed work run in
local cpu") that for now maintains the local CPU guarantees just in case
there are other broken users that depended on the accidental behavior.

js: 3.12 backport

Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Mike Galbraith <mgalbraith@suse.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
[ kamal: backport to 3.16-stable: use queue_delayed_work_on() ]
Signed-off-by: Kamal Mostafa <kamal@canonical.com>
Signed-off-by: Luis Henriques <luis.henriques@canonical.com>

mm/vmstat.c

@@ -1233,7 +1233,8 @@ int sysctl_stat_interval __read_mostly = HZ;
 static void vmstat_update(struct work_struct *w)
 {
 	refresh_cpu_vm_stats();
-	queue_delayed_work(vmstat_wq, this_cpu_ptr(&vmstat_work),
+	queue_delayed_work_on(smp_processor_id(), vmstat_wq,
+		this_cpu_ptr(&vmstat_work),
 		round_jiffies_relative(sysctl_stat_interval));
 }