On Wed, Jan 18, 2017 at 6:42 AM, Andy Shevchenko
[off-list ref] wrote:

On Mon, 2017-01-16 at 16:54 -0600, Rob Herring wrote:

quoted

Add a serdev controller driver for tty ports.

The controller is registered with serdev when tty ports are registered
with the TTY core. As the TTY core is built-in only, this has the side
effect of making serdev built-in as well.

quoted

+if SERIAL_DEV_BUS
+
+config SERIAL_DEV_CTRL_TTYPORT
+     bool "Serial device TTY port controller"
+     depends on TTY

quoted

+     depends on SERIAL_DEV_BUS != m

Since you have this line the
 if SERIAL_DEV_BUS
is redundant for it.

It is not. It is the standard pattern of

menuconfig BLAH

if BLAH
...
endif
<EOF>

If I remove the "if", then SERIAL_DEV_CTRL_TTYPORT can be enabled when
SERIAL_DEV_BUS=n which breaks the build

So, leave either one or another (as an example you can look at
DMADEVICES).

quoted

+
+#define SERPORT_BUSY 1
+#define SERPORT_ACTIVE       2
+#define SERPORT_DEAD 3
+
+struct serport {
+     struct tty_port *port;
+     struct tty_struct *tty;

quoted

+     struct tty_driver *tty_drv;
+     int tty_idx;

Do you need tty_ prefix for them?

It's just to be clear it's the tty driver and index rather than this
driver's driver or index.

quoted

+static int ttyport_open(struct serdev_controller *ctrl)
+{
+     struct serport *serport =
serdev_controller_get_drvdata(ctrl);
+     struct tty_struct *tty;
+     struct ktermios ktermios;
+
+     tty = tty_init_dev(serport->tty_drv, serport->tty_idx);
+     serport->tty = tty;
+
+     serport->port->client_ops = &client_ops;
+     serport->port->client_data = ctrl;
+

quoted

+     tty->receive_room = 65536;

Magic?

Probably. It's just what every ldisc uses. I suppose we could need
clients to set this, but we can add that as needed.

quoted

+     if (tty->ops->open)
+             tty->ops->open(serport->tty, NULL);
+     else
+             tty_port_open(serport->port, tty, NULL);
+
+     /* Bring the UART into a known 8 bits no parity hw fc state
*/
+     ktermios = tty->termios;
+     ktermios.c_iflag &= ~(IGNBRK | BRKINT | PARMRK | ISTRIP |
+                           INLCR | IGNCR | ICRNL | IXON);
+     ktermios.c_oflag &= ~OPOST;
+     ktermios.c_lflag &= ~(ECHO | ECHONL | ICANON | ISIG |
IEXTEN);
+     ktermios.c_cflag &= ~(CSIZE | PARENB);
+     ktermios.c_cflag |= CS8;
+     ktermios.c_cflag |= CRTSCTS;
+     tty_set_termios(tty, &ktermios);
+
+     set_bit(TTY_DO_WRITE_WAKEUP, &tty->flags);
+

quoted

+     mutex_lock(&serport->lock);
+     set_bit(SERPORT_ACTIVE, &serport->flags);
+     mutex_unlock(&serport->lock);

So, some clarification would be good to have to understand why you need
mutex _and_ atomic operation together.

What does mutex protect?

Paranoia. Actually, looking at this closer, we can get rid of the
mutex altogether.

quoted

+void serdev_tty_port_unregister(struct tty_port *port)
+{
+     struct serdev_controller *ctrl = port->client_data;
+     struct serport *serport =
serdev_controller_get_drvdata(ctrl);
+

quoted

+     if (!serport)
+             return;

What this check prevents from?

Didn't you ask this last time? See patch #9. tty_port_destructor()
calls this unconditionally as it doesn't know whether there's a serdev
or not. ctrl may be NULL, and then serport may be NULL.

Rob