Re: [PATCH 0/1] kthreads: simplify !kthreadd_task logic, kill kthreadd_task_init_done

[Wu Fei <at.wufei@gmail.com>]

On Fri, Sep 04, 2009 at 09:37:49AM +0200, Ingo Molnar wrote:
> 
> * Oleg Nesterov <oleg@redhat.com> wrote:
> 
> > On 09/01, Eric W. Biederman wrote:
> > >
> > > Oleg Nesterov <oleg@redhat.com> writes:
> > >
> > > > On 09/01, Oleg Nesterov wrote:
> > > >>
> > > >> On 09/01, Ingo Molnar wrote:
> > > >> >
> > > >> > * Oleg Nesterov <oleg@redhat.com> wrote:
> > > >> >
> > > >> > > On 09/01, Ingo Molnar wrote:
> > > >> > > >
> > > >> > > > * Oleg Nesterov <oleg@redhat.com> wrote:
> > > >> > > >
> > > >> > > > > Yes, this should work. But I _think_ we can make the better fix...
> > > >> > > > >
> > > >> > > > > I'll try to make the patch soon. Afaics we don't need
> > > >> > > > > kthreadd_task_init_done.
> > > >> > > >
> > > >> > > > ok.
> > > >> > >
> > > >> > > Just in case, the patch is ready. [...]
> > > >> >
> > > >> > yes - that's roughly the cleanup i referred to in the commit log.
> > > >> >
> > > >> > way too late for -rc8 though - the minimal fix i did _might_ be
> > > >> > eligible.
> > > >> >
> > > >> > agreed?
> > > >>
> > > >> Agreed. Then I will sent the patch on top of this change.
> > > >
> > > > OK, I am sending the patch on top of your fix. Not sure how to really
> > > > test it, but at least the kernel works when I apply the debugging patch
> > > > below on top.
> > >
> > > Stupid question.  How is it that we wind up trying to start kernel threads
> > > before it is safe to do so?
> > >
> > > Races should be impossible because the scheduler isn't running until a few
> > > lines later.
> > 
> > Yes, I am confused too.
> > 
> > At first I thought I understand the race, now I don't. Please see 
> > the whole thread: http://marc.info/?t=125180592500005
> 
> I dont understand it either - and the .config being !SMP excludes 
> any sort of SMP race as well.
> 
> Lets delay this until i can debug it more fully.
> 
CONFIG_PREEMPT_VOLUNTARY looks like the key of this problem,
might_resched becomes _cond_resched, and since d86ee480 changes 
it not to check SYSTEM_RUNNING, this function may call schedule().
And might_resched may be called even from do_fork(), so this is a
scenario:

kernel_thread(kernel_init, ...)
pid = kernel_thread(kthreadd, ...)
	-- switch to thread kernel_init, which refers to  kthreadd_task
           and NULL reference happens.  Note, because of
	   !CONFIG_LOCK_KERNEL,  lock_kernel() is nop.

What about just creating kthreadd before kernel_init as the following,
it works for me?


diff --git a/init/main.c b/init/main.c
index b34fd8e..ae86699 100644
--- a/init/main.c
+++ b/init/main.c
@@ -452,10 +452,10 @@ static noinline void __init_refok rest_init(void)
 	int pid;
 
 	rcu_scheduler_starting();
-	kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
-	numa_default_policy();
 	pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
 	kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
+	kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
+	numa_default_policy();
 	unlock_kernel();
 
 	/*