Re: [PATCH 1/2] PF_DEAD: cleanup usage

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@osdl.org> wrote:

> Well, in all fairness, it's entirely possible (nay, likely) that 
> PF_DEAD just isn't relevant the way it used to be.

yes, that's the case. I do believe we should go with Oleg's original 
patches.

> Looking at the history, the immediate reason for PF_DEAD was (I think) 
> that we had a race where we would first mark the task as a ZOMBIE, and 
> then if it was self-reaping, we'd mark it dead in release_task(). But 
> we dropped the tasklist_lock in between, which meant that the real 
> parent could come in and reap it just before it reaped itself, and all 
> hell would break lose.

yeah. Another problem was if we accidentally scheduled somewhere after 
the task was marked TASK_DEAD. (That used to be caught by another 
PF_DEAD check before the final schedule()).

all this stuff got changed as part of the PREEMPT_VOLUNTARY changes, and 
now we exit in a much more robust way.

> HOWEVER. I just noticed something strange. EXIT_DEAD should be in 
> "task->exit_state", not in "task->state". So there's something strange 
> going on in that neck of the woods _anyway_. That whole
> 
> 	...
>         if (unlikely(prev->flags & PF_DEAD))
>                 prev->state = EXIT_DEAD;
> 	...
> 
> in kernel/sched.c seems totally bogus.

no, it's not bogus. PF_DEAD is basically just a fancy, persistent flag 
for "mark the task state as EXIT_DEAD atomically before the final 
schedule". The TASK_DEAD flag is still necessary so that we dont stay on 
the runqueue. [we could use TASK_UNINTERRUPTIBLE - it should just not be 
TASK_RUNNING or TASK_INTERRUPTIBLE.]

in any case, i agree with Oleg's patches: with the current code it's 
unnecessary to signal towards schedule() to turn the PF_DEAD flag into 
EXIT_DEAD - we can do it right where we set PF_DEAD: immediately before 
calling the final schedule(). PF_DEAD is a relic from the days when the 
exit path was structured differently.

	Ingo