Re: [PATCH] tty reference count fix

[Paul Fulghum <paulkf@microgate.com>]

Jason Baron wrote:
> patch looks good to me. Or you could even drop the tty_sem completely from 
> the release_dev path (patch below) since lock_kernel() is held in both 
> tty_open() and the release_dev paths()

The historical posts on locking in release_dev()
I looked at suggest that a move from BKL to more
fine grained synchronization (tty_sem) was a goal.

The code looks like it originally relied on the BKL.
The later addition of tty_sem was done in a way
that allowed sleeping between setting tty_closing
and TTY_CLOSING flag:

down(&tty_sem)
if (tty->count == 1)
	tty_closing = 1;
up(&tty_sem)
...
down(&tty_sem) <<<< can sleep, open can increase count
tty->count--
up(&tty_sem)
...
if (tty_closing)
	set_bit(TTY_CLOSING)

 > (and there is no sleeping b/w setting the tty_closing
 > flag and setting the TTY_CLOSING bit).

Your patch leaves the schedule() call at the bottom of
the while loop between setting tty_closing and
setting TTY_CLOSING flag.

Thanks,
Paul

--
Paul Fulghum
Microgate Systems, Ltd