core: msg_receive should yield

[OlegHahm]

If a thread sends blocking, but the target thread is not currently in
receive mode, the sender gets queued. If it has a higher priority it
should run again as soon as the target goes into receiving mode.

[Kijewski]

The logic in this PR is incorrect.

Use this instead:

diff --git a/core/msg.c b/core/msg.c
index ec16e94..6a0a83a 100644
--- a/core/msg.c
+++ b/core/msg.c
@@ -310,12 +310,15 @@ static int _msg_receive(msg_t *m, int block)
         *m = *sender_msg;

         /* remove sender from queue */
+        int16_t other_prio = -1;
         if (sender->status != STATUS_REPLY_BLOCKED) {
             sender->wait_data = NULL;
             sched_set_status(sender, STATUS_PENDING);
+            other_prio = sender->priority;
         }

         eINT();
+        if (other_prio >= 0) {
+            sched_switch(other_prio);
+        }
         return 1;
     }

The thread will only yield if it's appropriate.

[OlegHahm]

I wouldn't call it incorrect, but less efficient, yes. I will switch to your proposal if there's a consensus that this change does not have any side effects and is wanted behavior.

[kaspar030]

Why is other_prio >= 0 sufficient as switching condition? Shouldn't there be other_prio < my_prio?

[Kijewski]

That's handled in sched_switch.

[OlegHahm]

With #1836 you're patch would be mostly superfluous if I replace thread_yield() with the new yield function, right?

[Kijewski]

Yes, my PR handles this function.

[OlegHahm]

I think even with your PR it's less expensive to check the sender priority before hand. If I just call thread_yield_higher() instead I save some conditional jumps and assignments, but to the cost of a software interrupt.

[Kijewski]

ACK

[OlegHahm]

@kaspar030, do you agree?

[OlegHahm]

@kaspar030, @LudwigOrtmann, @authmillenon, can I get some of you for a second ACK?

[kaspar030]

yep! ACK after squash.

[OlegHahm]

Squashed.