Re: 2.6.7-mm6

[Nick Piggin]

David S. Miller wrote:
> On Wed, 7 Jul 2004 09:35:10 +0200
> Ingo Molnar <mingo@xxxxxxx> wrote:
>
>
> > the patch below should solve this. Is it safe on sparc to do a
> > fork_by_hand() like this?
>
>
> If the regs are garbage, copy_thread() will explode as it tries
> to interpret the stack pointer in that regs value.
>
> The parent's regs (stored in current_thread_info() at trap time,
> and also needed by copy_thread() processing) will also be garbage
> since we're avoiding the fork syscall trap.
>
> In short, this won't work :)
>
> This is why I use kernel_thread().  Why is that so bad?

We could make CLONE_IDLETASK clones not do the wakeup?

Ingo? I guess an alternative is to have the arch explicitly
make a call to dequeue it.



---

 linux-2.6-npiggin/kernel/fork.c |   12 +++++++-----
 1 files changed, 7 insertions(+), 5 deletions(-)

diff -puN kernel/fork.c~kernelthread-idle-fix kernel/fork.c
--- linux-2.6/kernel/fork.c~kernelthread-idle-fix	2004-07-09 12:42:02.000000000 +1000
+++ linux-2.6-npiggin/kernel/fork.c	2004-07-09 12:43:11.000000000 +1000
@@ -1215,11 +1215,13 @@ long do_fork(unsigned long clone_flags,
 			set_tsk_thread_flag(p, TIF_SIGPENDING);
 		}
 
-		if (!(clone_flags & CLONE_STOPPED))
-			wake_up_new_task(p, clone_flags);
-		else
-			p->state = TASK_STOPPED;
-		++total_forks;
+		if (likely(!(clone_flags & CLONE_IDLETASK))) {
+			if (!(clone_flags & CLONE_STOPPED))
+				wake_up_new_task(p, clone_flags);
+			else
+				p->state = TASK_STOPPED;
+			++total_forks;
+		}
 
 		if (unlikely (trace)) {
 			current->ptrace_message = pid;

_