Re: [PATCH v2 RESEND] PREEMPT_RT: USB serial gadget: Fix locking issue

[Felipe Balbi]

On Thu, May 22, 2014 at 04:57:31PM +0200, Jean-Jacques Hiblot wrote:
> With CONFIG_PREEMPT_RT_FULL the tty_flip_buffer_push(..) actions are executed
> immeditely (same behaviour as if low_latency flag was set). We thus have to
> release port_lock before callng tty_flip_buffer_push().
> 
> This issue resultst of commits:
> 44a0c019: USB: g_serial: don't set low_latency flag
> dc0cbd1b: serial: 8250: Call flush_to_ldisc when the irq is threaded
> 
> Signed-off-by: Jean-Jacques Hiblot <jjhiblot@xxxxxxxxxxxxxxx>
> ---
> changes since v1:
>  - fix compilation problem: use port.tty instead of port->port_tty
> 
>  drivers/usb/gadget/u_serial.c | 23 ++++++++++++++++++-----
>  1 file changed, 18 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/usb/gadget/u_serial.c b/drivers/usb/gadget/u_serial.c
> index b369292..3d4246d 100644
> --- a/drivers/usb/gadget/u_serial.c
> +++ b/drivers/usb/gadget/u_serial.c
> @@ -549,12 +549,25 @@ static void gs_rx_push(unsigned long _port)
>  		port->read_started--;
>  	}
> 
> -	/* Push from tty to ldisc; without low_latency set this is handled by
> -	 * a workqueue, so we won't get callbacks and can hold port_lock
> -	 */
> -	if (do_push)
> +	/* Push from tty to ldisc */
> +	if (do_push) {
> +#ifndef CONFIG_PREEMPT_RT_FULL

can we drop the ifdef somehow ? Are there no better options here ?

> +		/* without low_latency set this is handled by a workqueue, so
> +		 * we won't get callbacks and can hold port_lock
> +		*/
>  		tty_flip_buffer_push(&port->port);
> -
> +#else
> +		/* CONFIG_PREEMPT_RT_FULL makes this a synchronous action. and
> +		 * it may trigger synchronous callbacks to this driver. We need
> +		 * to release the lock.
> +		*/
> +		spin_unlock_irq(&port->port_lock);
> +		tty_flip_buffer_push(&port->port);
> +		spin_lock_irq(&port->port_lock);
> +		/* tty may have been closed */
> +		tty = port->port.tty;
> +#endif
> +	}
> 
>  	/* We want our data queue to become empty ASAP, keeping data
>  	 * in the tty and ldisc (not here).  If we couldn't push any
> --
> 1.9.2
> 

-- 
balbi

Attachment: signature.asc
Description: Digital signature