Re: [PATCH 4/5] tty: serial: 8250 core: add runtime pm

[Felipe Balbi <balbi@ti.com>]

[-- Attachment #1: Type: text/plain, Size: 3170 bytes --]

Hi,

On Thu, Jul 17, 2014 at 05:43:00PM +0200, Sebastian Andrzej Siewior wrote:
> * Peter Hurley | 2014-07-17 11:31:59 [-0400]:
> 
> >On 07/16/2014 12:06 PM, Felipe Balbi wrote:
> >>On Wed, Jul 16, 2014 at 05:54:56PM +0200, Sebastian Andrzej Siewior wrote:
> >>>On 07/16/2014 05:16 PM, Felipe Balbi wrote:
> >
> >>>>I wonder if you should get_sync() on start_tx() and only
> >>>>put_autosuspend() at stop_tx(). I guess the outcome would be
> >>>>largely the same, no ?
> >>>
> >>>I just opened minicom on ttyS0 and gave a try. start_tx() was invoked
> >>>each time I pressed a key (sent a character). I haven't seen stop_tx()
> >>>even after after I closed minicom. I guess stop_tx() is invoked if you
> >>>switch half-duplex communication.
> >>
> >>that's bad, I expected stop to be called also after each character.
> >
> >The 8250 core auto-stops tx when the tx ring buffer is empty (except
> >in the case of dma, where stopping tx isn't necessary).
> 
> This is correct. So this is what I have now for the non-dma case:
> 
> diff --git a/drivers/tty/serial/8250/8250_core.c b/drivers/tty/serial/8250/8250_core.c
> index 2e4a93b..480a1c0 100644
> --- a/drivers/tty/serial/8250/8250_core.c
> +++ b/drivers/tty/serial/8250/8250_core.c
> @@ -1283,6 +1283,9 @@ static inline void __stop_tx(struct uart_8250_port *p)
>  	if (p->ier & UART_IER_THRI) {
>  		p->ier &= ~UART_IER_THRI;
>  		serial_out(p, UART_IER, p->ier);
> +
> +		pm_runtime_mark_last_busy(p->port.dev);
> +		pm_runtime_put_autosuspend(p->port.dev);
>  	}
>  }
>  
> @@ -1310,12 +1313,12 @@ static void serial8250_start_tx(struct uart_port *port)
>  	struct uart_8250_port *up =
>  		container_of(port, struct uart_8250_port, port);
>  
> -	pm_runtime_get_sync(port->dev);
>  	if (up->dma && !serial8250_tx_dma(up)) {
>  		goto out;
>  	} else if (!(up->ier & UART_IER_THRI)) {
>  		up->ier |= UART_IER_THRI;
> +		pm_runtime_get_sync(port->dev);
>  		serial_port_out(port, UART_IER, up->ier);
>  
>  		if (up->bugs & UART_BUG_TXEN) {
>  			unsigned char lsr;

this looks better. So we get on start_tx() and put on stop_tx().

> @@ -1500,9 +1503,10 @@ void serial8250_tx_chars(struct uart_8250_port *up)
>  		uart_write_wakeup(port);
>  
>  	DEBUG_INTR("THRE...");
> -
> +#if 0
>  	if (uart_circ_empty(xmit))
>  		__stop_tx(up);
> +#endif
>  }
>  EXPORT_SYMBOL_GPL(serial8250_tx_chars);

is it so that start_tx() gets called one and stop_tx() might be called N
times ? That looks unbalanced to me. If the calls are balanced, then you
shouldn't need to care because pm_runtime will handle reference counting
for you, right?

> and now I need to come up with something that is not if (port != omap)
> for that #if 0 block. The code disables the TX FIFO empty interrupt once
> the transfer is complete. I want to call __stop_tx() once the tx fifo is
> empty.
> Felipe, Would a check for dev->power.use_autosuspend be the right thing
> to do?

probably not, as that's internal to the pm_runtime code. But I wonder if
start/stop tx calls are balanced, if they are then we're good. Unless
I'm missing something else.

-- 
balbi

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]