Re: [crash] NULL pointer dereference at IP: [<ffffffff812e9ccb>] uart_close+0x2a/0x1e4

[Ingo Molnar <mingo@elte.hu>]

* Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:

> > Unfortunately 46d57a449aa13d9c6adcc9d1dbc7b9a0ecfb69d8 is not easy to 
> > revert, at all. Would be glad to try any patch - fixes or some manual 
> > revert as well (if that's feasible).
> 
> The patch doesn't do anything but create local variables of 
> state->port. I'm now glad I did it that way as it makes it easy to 
> spot the mistake.
>
> the tty_port lock is ->lock, the uart lock is ->lock both are 
> different..
> 
> Spot the mistake:
> 
>  	if (state->flags & UIF_INITIALIZED) {
>  		unsigned long flags;
>  		spin_lock_irqsave(&port->lock, flags);
> -		port->ops->stop_rx(port);
> +		uport->ops->stop_rx(uport);
>  		spin_unlock_irqrestore(&port->lock, flags);
>  		/*
>  		 * Before we drop DTR, make sure the UART transmitter
>  		 * has completely drained; this is especially
>  		 * important if there is a transmit FIFO!
>  		 */
> -		uart_wait_until_sent(tty, port->timeout);
> +		uart_wait_until_sent(tty, uport->timeout);
>  	}
> 
> 
> so uart_close takes the wrong lock. I've checked the rest of the patch 
> for the same error and I don't see any other screwups.

Cool! This very much looks like something that could fix both problems. 
I've started testing your fix.

And yeah, i too never ever regretted having done 'too finegrained' 
patches. (only ever the other way around) They are a pleasure to bisect 
back to.

> [...] The race in question is very hard to hit but fortunately Ingo 
> hit it twice..

a hundred thousand bootups a year go a long way in finding narrow races 
;-)

	Ingo