2002-08-19 01:54:55

by Paul Mackerras

Subject: [PATCH] fix bug in yield()

This patch makes yield() actually call schedule.

Without this patch, 2.4.20-pre3 will hang on boot for me, looping in
spawn_ksoftirqd waiting for ksoftirqd to start and calling yield().
ksoftirqd never gets to run, however, because yield never actually
calls schedule. (Note that sys_sched_yield as a syscall is OK because
the syscall exit path will check current->need_resched and call
schedule).

Paul.

diff -urN linux-2.4/kernel/sched.c pmac/kernel/sched.c
--- linux-2.4/kernel/sched.c Wed Aug 7 18:10:01 2002
+++ pmac/kernel/sched.c Mon Aug 19 10:39:39 2002
@@ -1081,6 +1081,7 @@
{
set_current_state(TASK_RUNNING);
sys_sched_yield();
+ schedule();
}

void __cond_resched(void)

2002-08-19 11:40:04

by Hiroshi Takekawa

Subject: Re: [PATCH] fix bug in yield()

Hi.

> This patch makes yield() actually call schedule.

The trouble hanging after "Initializing RT netlink socket" which
several report, just gone. And seems to work fine.

But I wonder whether doing schedule() here is good or not. In 2.5,
schedule() is done in sys_sched_yield(). I don't know much about
kernel things. So I just compared...

--
Hiroshi Takekawa <[email protected]>


> Without this patch, 2.4.20-pre3 will hang on boot for me, looping in
> spawn_ksoftirqd waiting for ksoftirqd to start and calling yield().
> ksoftirqd never gets to run, however, because yield never actually
> calls schedule. (Note that sys_sched_yield as a syscall is OK because
> the syscall exit path will check current->need_resched and call
> schedule).

> Paul.

> diff -urN linux-2.4/kernel/sched.c pmac/kernel/sched.c
> --- linux-2.4/kernel/sched.c Wed Aug 7 18:10:01 2002
> +++ pmac/kernel/sched.c Mon Aug 19 10:39:39 2002
> @@ -1081,6 +1081,7 @@
> {
> set_current_state(TASK_RUNNING);
> sys_sched_yield();
> + schedule();
> }

> void __cond_resched(void)