Re: [PATCH] serial: 8250_omap: Fix gpio check for auto RTS and CTS

[Yegor Yefremov <yegorslists@googlemail.com>]

On Sun, Oct 6, 2019 at 5:49 PM Adam Ford <aford173@gmail.com> wrote:
>
> On Sun, Oct 6, 2019 at 7:34 AM Yegor Yefremov
> <yegorslists@googlemail.com> wrote:
> >
> > On Sun, Oct 6, 2019 at 1:38 PM Adam Ford <aford173@gmail.com> wrote:
> > >
> > > On Sun, Oct 6, 2019 at 6:21 AM Yegor Yefremov
> > > <yegorslists@googlemail.com> wrote:
> > > >
> > > > Hi Adam,
> > > >
> > > > On Sun, Oct 6, 2019 at 4:33 AM Adam Ford <aford173@gmail.com> wrote:
> > > > >
> > > > > There are two checks to see if the manual gpio is configured, but
> > > > > these the check is seeing if the structure is NULL instead it
> > > > > should check to see if there are CTS and/or RTS pins defined.
> > > > >
> > > > > This patch uses checks for those individual pins instead of
> > > > > checking for the structure itself.
> > > diff --git a/drivers/tty/serial/8250/8250_core.c b/drivers/tty/serial/8250/8250_core.c
> > index e682390..d5fdb71 100644
> > --- a/drivers/tty/serial/8250/8250_core.c
> > +++ b/drivers/tty/serial/8250/8250_core.c
> > @@ -1031,6 +1031,8 @@ int serial8250_register_8250_port(struct
> > uart_8250_port *up)
> >                         } else {
> >                                 uart->gpios = gpios;
> >                         }
> > +               } else {
> > +                       uart->gpios = NULL;
> >                 }
> >
> >                 serial8250_set_defaults(uart);
> > > >
> > > > > Signed-off-by: Adam Ford <aford173@gmail.com>
> > > > >
> > > > > diff --git a/drivers/tty/serial/8250/8250_omap.c b/drivers/tty/serial/8250/8250_omap.c
> > > > > index c68e2b3a1634..836e736ae188 100644
> > > > > --- a/drivers/tty/serial/8250/8250_omap.c
> > > > > +++ b/drivers/tty/serial/8250/8250_omap.c
> > > > > @@ -141,7 +141,7 @@ static void omap8250_set_mctrl(struct uart_port *port, unsigned int mctrl)
> > > > >
> > > > >         serial8250_do_set_mctrl(port, mctrl);
> > > > >
> > > > > -       if (!up->gpios) {
> > > > > +       if (!mctrl_gpio_to_gpiod(up->gpios, UART_GPIO_RTS)) {
> > > > >                 /*
> > > > >                  * Turn off autoRTS if RTS is lowered and restore autoRTS
> > > > >                  * setting if RTS is raised
> > > > > @@ -456,7 +456,8 @@ static void omap_8250_set_termios(struct uart_port *port,
> > > > >         up->port.status &= ~(UPSTAT_AUTOCTS | UPSTAT_AUTORTS | UPSTAT_AUTOXOFF);
> > > > >
> > > > >         if (termios->c_cflag & CRTSCTS && up->port.flags & UPF_HARD_FLOW &&
> > > > > -           !up->gpios) {
> > > > > +           !mctrl_gpio_to_gpiod(up->gpios, UART_GPIO_RTS) &&
> > > > > +           !mctrl_gpio_to_gpiod(up->gpios, UART_GPIO_CTS)) {
> > > > >                 /* Enable AUTOCTS (autoRTS is enabled when RTS is raised) */
> > > > >                 up->port.status |= UPSTAT_AUTOCTS | UPSTAT_AUTORTS;
> > > > >                 priv->efr |= UART_EFR_CTS;
> > > >
> > > > Looks good to me but !up->gpios must remain as otherwise, we will get
> > > > NULL pointer dereference. What do you think?
>
> What we if add a check to make sure we deference it?  I was thinking
> something like:
>
> diff --git a/drivers/tty/serial/serial_mctrl_gpio.c
> b/drivers/tty/serial/serial_mctrl_gpio.c
> index d9074303c88e..fb4781292d40 100644
> --- a/drivers/tty/serial/serial_mctrl_gpio.c
> +++ b/drivers/tty/serial/serial_mctrl_gpio.c
> @@ -66,6 +66,9 @@ EXPORT_SYMBOL_GPL(mctrl_gpio_set);
>  struct gpio_desc *mctrl_gpio_to_gpiod(struct mctrl_gpios *gpios,
>                                       enum mctrl_gpio_idx gidx)
>  {
> +       if (gpios == NULL)
> +               return NULL;
> +
>         return gpios->gpio[gidx];
>  }
>  EXPORT_SYMBOL_GPL(mctrl_gpio_to_gpiod);
>
> This should make my previous patch tolerate situations where gpios is
> NULL.  It also explictly looks for conditions where the gpios are
> manually configured for RTS and/or CTS.

I like this idea. This would solve my concern.

Yegor

> > >
> > > I was not seeing up->gpios ever NULL so the contents inside the check
> > > never was executed.  When I removed the check, the performance came
> > > back.  I looked at examples on how other devices checked for RTS and
> > > CTS, and I noticed that the Atmel serial driver did something like the
> > > above.
> > >
> > > >
> > > > Also adding some more people who can be interested in testing this approach.
> > >
> > > I am open for ideas.  If something is better, but something is either
> > > incorrectly setting up->gpios to non-NULL or the check for non-NULL is
> > > wrong.
> >
> > I wonder whether we forgot to add this assignment for the ACPI systems:
> >
>
> I unwound my patch and applied your patch instead, but it is back to
> having corrupt frames again.
>
> > diff --git a/drivers/tty/serial/8250/8250_core.c
> > b/drivers/tty/serial/8250/8250_core.c
> > index e682390..d5fdb71 100644
> > --- a/drivers/tty/serial/8250/8250_core.c
> > +++ b/drivers/tty/serial/8250/8250_core.c
> > @@ -1031,6 +1031,8 @@ int serial8250_register_8250_port(struct
> > uart_8250_port *up)
> >                         } else {
> >                                 uart->gpios = gpios;
> >                         }
> > +               } else {
> > +                       uart->gpios = NULL;
> >                 }
> >
> I did not check the value of uart->gpios after this patch, but since
> the flow control code isn't executing, I am guessing it's still not
> NULL.
> >                 serial8250_set_defaults(uart);
>
> adam