Re: Spinlock debugging

[Andrew Bird <ajb@spheresystems.co.uk>]

Alan
	thanks that did the trick. 
One further question, on the later kernels 2.6.17+, I don't have low_latency 
set. Can I still guarantee that after calling tty_flip_buffer_push() I have 
made space in the tty for my buffer? For example, is this legal? 

// in interrupt handler

   if(tty_buffer_request_room(tty, size) < size) {

      spin_unlock(&dc->lock);
      spin_unlock(&port->lock);

      tty_flip_buffer_push(tty);

      spin_lock(&port->lock);
      spin_lock(&dc->lock);

  }
  tty_insert_flip_string(tty, buf, size);
   
Thanks


Andrew

On Monday 11 September 2006 18:24, Alan Cox wrote:
> Ar Llu, 2006-09-11 am 17:38 +0100, ysgrifennodd Andrew Bird:
> > Alan
> > 	Yes, I have low_latency set for kernels lower than 2.6.17. I'm currently
> > testing using 2.6.15. When you mention 'write method for flow control' do
> > you mean for software XON/XOFF etc?
>
> Yes
>
> Basically in low_latency the following is valid
>
>
> 	driver receives bytes
> 		flush_to_ldisc
> 			ldisc calls driver write methods
>
>
> That means if you have a shared lock for read/write you want to drop it
> after you've bashed the hardware and before you flush_to_ldisc. Remember
> the IRQ handler is not re-entrant so another IRQ of the same number
> won't cause further I/O and out of order receives.
>
> Alan