lkml.org 
[lkml]   [2022]   [Nov]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v3 13/14] serial: liteuart: add IRQ support for the TX path
On Tue, Nov 15, 2022 at 06:14:50PM +0200, Ilpo Järvinen wrote:
> On Sat, 12 Nov 2022, Gabriel Somlo wrote:
>
> > Modify the TX path to operate in an IRQ-compatible way, while
> > maintaining support for polling mode via the poll timer.
> >
> > Signed-off-by: Gabriel Somlo <gsomlo@gmail.com>
> > ---
> > drivers/tty/serial/liteuart.c | 67 ++++++++++++++++++++++++-----------
> > 1 file changed, 47 insertions(+), 20 deletions(-)
> >
> > diff --git a/drivers/tty/serial/liteuart.c b/drivers/tty/serial/liteuart.c
> > index e30adb30277f..307c27398e30 100644
> > --- a/drivers/tty/serial/liteuart.c
> > +++ b/drivers/tty/serial/liteuart.c
> > @@ -46,6 +46,7 @@ struct liteuart_port {
> > struct uart_port port;
> > struct timer_list timer;
> > u32 id;
> > + bool poll_tx_started;
> > };
> >
> > #define to_liteuart_port(port) container_of(port, struct liteuart_port, port)
> > @@ -78,29 +79,24 @@ static void liteuart_putchar(struct uart_port *port, unsigned char ch)
> >
> > static void liteuart_stop_tx(struct uart_port *port)
> > {
> > - /* not used in LiteUART, but called unconditionally from serial_core */
>
> Drop this comment from the earlier patch too given you remove it here. It
> just adds useless churn in diff for no useful reason.

Right -- I already had this lined up for v4 :)

> > + if (port->irq) {
> > + u8 irq_mask = litex_read8(port->membase + OFF_EV_ENABLE);
> > + litex_write8(port->membase + OFF_EV_ENABLE, irq_mask & ~EV_TX);
>
> If you put irq_mask into liteuart_port you wouldn't need to read it
> back here?

So, instead of `bool poll_tx_started` I should just keep a copy of the
irq_mask there, and take `&port->lock` whenever I need to *both* update
the mask *and* write it out to the actual device register?

> > + } else {
> > + struct liteuart_port *uart = to_liteuart_port(port);
> > + uart->poll_tx_started = false;
> > + }
> > }
> >
> > static void liteuart_start_tx(struct uart_port *port)
> > {
> > - struct circ_buf *xmit = &port->state->xmit;
> > - unsigned char ch;
> > -
> > - if (unlikely(port->x_char)) {
> > - litex_write8(port->membase + OFF_RXTX, port->x_char);
> > - port->icount.tx++;
> > - port->x_char = 0;
> > - } else if (!uart_circ_empty(xmit)) {
> > - while (xmit->head != xmit->tail) {
> > - ch = xmit->buf[xmit->tail];
> > - xmit->tail = (xmit->tail + 1) & (UART_XMIT_SIZE - 1);
> > - port->icount.tx++;
>
> This is not based on tty-next tree. Please rebase on top of it (you
> might have noted it already, IIRC, somebody noted uart_xmit_advance
> conflict in some patch, perhaps it was you :-)).

Yeah, I did notice that right after I sent out v3. I've already
rebased it on top of your patch using uart_xmit_advance.

> > - liteuart_putchar(port, ch);
> > - }
> > + if (port->irq) {
> > + u8 irq_mask = litex_read8(port->membase + OFF_EV_ENABLE);
> > + litex_write8(port->membase + OFF_EV_ENABLE, irq_mask | EV_TX);
>
> ->irq_mask?

I'll switch to s/bool poll_tx_started/u8 irq_mask/g in v4, hopefully
it should make it all look cleaner.

> > + } else {
> > + struct liteuart_port *uart = to_liteuart_port(port);
> > + uart->poll_tx_started = true;
> > }
> > -
> > - if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
> > - uart_write_wakeup(port);
> > }
> >
> > static void liteuart_stop_rx(struct uart_port *port)
> > @@ -131,18 +127,47 @@ static void liteuart_rx_chars(struct uart_port *port)
> > tty_flip_buffer_push(&port->state->port);
> > }
> >
> > +static void liteuart_tx_chars(struct uart_port *port)
> > +{
> > + struct circ_buf *xmit = &port->state->xmit;
> > +
> > + if (unlikely(port->x_char)) {
> > + litex_write8(port->membase + OFF_RXTX, port->x_char);
> > + port->x_char = 0;
> > + port->icount.tx++;
> > + return;
> > + }
> > +
> > + while (!litex_read8(port->membase + OFF_TXFULL)) {
> > + if (xmit->head == xmit->tail)
>
> There exists a helper for this condition.

Is that in the released linus tree, or still only in tty-next?

> > + break;
> > + litex_write8(port->membase + OFF_RXTX, xmit->buf[xmit->tail]);
> > + xmit->tail = (xmit->tail + 1) & (UART_XMIT_SIZE - 1);
> > + port->icount.tx++;
>
> uart_xmit_advance()

Already lined up for v4.

>
> > + }
> > +
> > + if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
> > + uart_write_wakeup(port);
> > +
> > + if (uart_circ_empty(xmit))
> > + liteuart_stop_tx(port);
> > +}
>
> You might want to check if you can generate this whole function with
> Jiri's tx helpers (IIRC, they're only in tty-next tree currently).

Looks like I should switch to tty-next for this whole series, which
makes sense, since it's a tty I'm working on :)

I'll rebase on top of that before I send out v4, hopefully later this
afternoon.

> > +
> > static irqreturn_t liteuart_interrupt(int irq, void *data)
> > {
> > struct liteuart_port *uart = data;
> > struct uart_port *port = &uart->port;
> > u8 isr = litex_read8(port->membase + OFF_EV_PENDING);
> >
> > - /* for now, only rx path triggers interrupts */
> > - isr &= EV_RX;
> > + if (!(port->irq || uart->poll_tx_started))
> > + isr &= ~EV_TX; /* polling mode with tx stopped */
> >
> > spin_lock(&port->lock);
> > if (isr & EV_RX)
> > liteuart_rx_chars(port);
> > + if (isr & EV_TX) {
> > + liteuart_tx_chars(port);
> > + }
>
> Extra braces.

Got it, thanks!

> > spin_unlock(&port->lock);
> >
> > return IRQ_RETVAL(isr);
> > @@ -196,6 +221,7 @@ static int liteuart_startup(struct uart_port *port)
> > }
> >
> > if (!port->irq) {
> > + uart->poll_tx_started = false;
>
> Can poll_tx_started ever be true here?

Proably not, but it shouldn't matter if I switch to using `u8 irq_mask`,
instead, which should be initialized to 0 during probe().

Thanks again for the feedback!

Best,
--Gabriel

> > timer_setup(&uart->timer, liteuart_timer, 0);
> > mod_timer(&uart->timer, jiffies + uart_poll_timeout(port));
> > }
> > @@ -210,6 +236,7 @@ static void liteuart_shutdown(struct uart_port *port)
> > struct liteuart_port *uart = to_liteuart_port(port);
> >
> > litex_write8(port->membase + OFF_EV_ENABLE, 0);
> > + uart->poll_tx_started = false;
> >
> > if (port->irq)
> > free_irq(port->irq, port);
> >
>
> --
> i.
>

\
 
 \ /
  Last update: 2022-11-15 18:14    [W:0.289 / U:0.956 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site