Messages in this thread Patch in this message | | | Date | Tue, 11 Nov 2008 14:46:38 +0100 | From | "Vegard Nossum" <> | Subject | Re: [Bug #11989] Suspend failure on NForce4-based boards due to chanes in stop_machine |
| |
On Tue, Nov 11, 2008 at 2:36 PM, Vegard Nossum <vegard.nossum@gmail.com> wrote: > On Tue, Nov 11, 2008 at 11:52 AM, Ingo Molnar <mingo@elte.hu> wrote: >> [ Cc:-ed workqueue/locking/suspend-race-condition experts. ] > > Heh. I am not expert, but I looked at the code. The obvious suspicious > thing to see is the use of unpaired barriers? Maybe like this:
...
> 55 /* Last one to ack a state moves to the next state. */ > 56 static void ack_state(void) > 57 { > 58 if (atomic_dec_and_test(&thread_ack)) > > Maybe > + /* force ordering between thread_ack/state */ > + smp_rmb(); > here?
Oops, I am wrong (after a small investigation).
"1490 Any atomic operation that modifies some state in memory and returns information 1491 about the state (old or new) implies an SMP-conditional general memory barrier 1492 (smp_mb()) on each side of the actual operation (with the exception of 1493 explicit lock operations, described later). These include: 1494 ... 1503 atomic_dec_and_test();"
Won't fix the problem at hand, but maybe something like this would be nice for future generations :-)
diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c index 0e688c6..6796bb1 100644 --- a/kernel/stop_machine.c +++ b/kernel/stop_machine.c @@ -55,6 +55,7 @@ static void set_state(enum stopmachine_state newstate) /* Last one to ack a state moves to the next state. */ static void ack_state(void) { + /* Implicit memory barrier; no smp_rmb() needed */ if (atomic_dec_and_test(&thread_ack)) set_state(state + 1); }
Vegard
-- "The animistic metaphor of the bug that maliciously sneaked in while the programmer was not looking is intellectually dishonest as it disguises that the error is the programmer's own creation." -- E. W. Dijkstra, EWD1036
| |