> +#define PIC32_SDEV_NAME "ttyS"
> +#define PIC32_SDEV_MAJOR TTY_MAJOR
> +#define PIC32_SDEV_MINOR 64
No. Same goes for you as every one of the forty other people a year who
try and claim their console is ttyS. If it's not an 8250 it isn't.
ttyS is the 8250, use dynamic major and minor and a different name.
> +/* serial core request to change current uart setting */
> +static void pic32_uart_set_termios(struct uart_port *port,
> + struct ktermios *new,
> + struct ktermios *old)
> +{
You need to clear any termios features requested but not supported. In
your case that appears to be CMSPAR, as you don't seem to support
mark/space parity.
Similarly if you only support 8N1 or 7E1/7O1 you need to force the CSIZE
bits to match what you ended up setting the UART to do.
> + /* update baud */
> + baud = uart_get_baud_rate(port, new, old, 0, port->uartclk / 16);
> + quot = uart_get_divisor(port, baud) - 1;
> + pic32_uart_write(quot, sport, PIC32_UART_BRG);
> + uart_update_timeout(port, new->c_cflag, baud);
See the 8250 driver for an example: you probably need to write back the
actual rate you got.
> +/* serial core request to release uart iomem */
> +static void pic32_uart_release_port(struct uart_port *port)
> +{
> + struct platform_device *pdev = to_platform_device(port->dev);
> + struct resource *res_mem;
> + unsigned int res_size;
resource_size_t for resources. Or you could just avoid the pointless
variable in the first place 8)
Other oddments - things like kasprintf() returns should be checked
Alan
