]> pilppa.com Git - linux-2.6-omap-h63xx.git/commit
sched: prevent divide by zero error in cpu_avg_load_per_task, update
authorIngo Molnar <mingo@elte.hu>
Sat, 29 Nov 2008 19:45:15 +0000 (20:45 +0100)
committerIngo Molnar <mingo@elte.hu>
Sat, 29 Nov 2008 19:45:15 +0000 (20:45 +0100)
commitaf6d596fd603219b054c1c90fb16672a9fd441bd
treee3e5a85532079aedc9b0b194a319ec084b8a4449
parent1583715ddb61f822041807a0f18b3b4845e88c76
sched: prevent divide by zero error in cpu_avg_load_per_task, update

Regarding the bug addressed in:

  4cd4262: sched: prevent divide by zero error in cpu_avg_load_per_task

Linus points out that the fix is not complete:

> There's nothing that keeps gcc from deciding not to reload
> rq->nr_running.
>
> Of course, in _practice_, I don't think gcc ever will (if it decides
> that it will spill, gcc is likely going to decide that it will
> literally spill the local variable to the stack rather than decide to
> reload off the pointer), but it's a valid compiler optimization, and
> it even has a name (rematerialization).
>
> So I suspect that your patch does fix the bug, but it still leaves the
> fairly unlikely _potential_ for it to re-appear at some point.
>
> We have ACCESS_ONCE() as a macro to guarantee that the compiler
> doesn't rematerialize a pointer access. That also would clarify
> the fact that we access something unsafe outside a lock.

So make sure our nr_running value is immutable and cannot change
after we check it for nonzero.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched.c