2008-03-12 12:34:19

by Oleg Nesterov

Subject: [PATCH 2/2] signals: document CLD_CONTINUED notification mechanics

A couple of small comments about how CLD_CONTINUED notification works.

Signed-off-by: Oleg Nesterov <[email protected]>

--- 25/kernel/signal.c~2_DOC_NOTIFY 2008-03-12 13:29:42.000000000 +0300
+++ 25/kernel/signal.c 2008-03-12 15:16:52.000000000 +0300
@@ -630,6 +630,11 @@ static int prepare_signal(int sig, struc
why |= SIGNAL_CLD_STOPPED;

if (why) {
+ /*
+ * The first thread which returns from finish_stop()
+ * will take ->siglock, notice SIGNAL_CLD_MASK, and
+ * notify its parent. See get_signal_to_deliver().
+ */
signal->flags = why | SIGNAL_STOP_CONTINUED;
signal->group_stop_count = 0;
signal->group_exit_code = 0;
@@ -1652,7 +1657,7 @@ relock:
try_to_freeze();

spin_lock_irq(&sighand->siglock);
-
+ /* see prepare_signal(SIGCONT) */
if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
int why = (signal->flags & SIGNAL_STOP_CONTINUED)
? CLD_CONTINUED : CLD_STOPPED;

2008-03-12 22:03:19

by Roland McGrath

Subject: Re: [PATCH 2/2] signals: document CLD_CONTINUED notification mechanics

> A couple of small comments about how CLD_CONTINUED notification works.

A worthy addition. I would make the get_signal_to_deliver comment say a
little more. In particular, it's kind of nonobvious that though this
happens at the beginning of the function, the importance of its placement
is really that it will always be run (via the relock: loop) just after any
wake-up from having been in TASK_STOPPED.


Thanks,
Roland