Hi Maxime,

On 03/24/2015 01:21 PM, Maxime Coquelin wrote:

Hi Peter,

2015-03-19 18:35 GMT+01:00 Maxime Coquelin [off-list ref]:

quoted

2015-03-19 15:58 GMT+01:00 Peter Hurley [off-list ref]:

quoted

On 03/19/2015 09:55 AM, Maxime Coquelin wrote:

quoted

quoted

quoted

quoted

quoted

+static void stm32_set_termios(struct uart_port *port, struct ktermios *termios,
+                           struct ktermios *old)

[...]

quoted

quoted

quoted

quoted

quoted

+       usardiv = (port->uartclk * 25) / (baud * 4);
+       mantissa = (usardiv / 100) << USART_BRR_DIV_M_SHIFT;
+       fraction = DIV_ROUND_CLOSEST((usardiv % 100) * 16, 100);
+       if (fraction & ~USART_BRR_DIV_F_MASK) {
+               fraction = 0;
+               mantissa += (1 << USART_BRR_DIV_M_SHIFT);
+       }

[...]

quoted

Really, I would prefer keeping this fractional divider as it is
implemented today.

You have to admit that's basically an unintelligible mess;
how would anyone ever be able to refactor and replace that with a
common divider implementation?

At the very least, please comment on the formula and format.

Ok, I will refactor the implementation, and comment it.

The implementation was indeed a mess.
I found some time to refactor the code, and also added support for 8
times oversampling (16 by default).
It will allow to achieve higher speeds, with the side effect of being
less tolerant to clock deviations.

What do you think about the code below?


    usartdiv = DIV_ROUND_CLOSEST(port->uartclk, baud);

    /*
     * The USART supports 16 or 8 times oversampling.
     * By default we prefer 16 times oversampling, so that the receiver
     * has a better tolerance to clock deviations.
     * 8 times oversampling is only used to achieve higher speeds.
     */
    if (usartdiv < 16) {
        oversampling = 8;
        stm32_set_bits(port, USART_CR1, USART_CR1_OVER8);
    } else {
        oversampling = 16;
        stm32_clr_bits(port, USART_CR1, USART_CR1_OVER8);
    }

    mantissa = (usartdiv / oversampling) << USART_BRR_DIV_M_SHIFT;
    fraction = usartdiv % oversampling;
    writel_relaxed(mantissa | fraction, port->membase + USART_BRR)

Thanks! Way better :)
Much more obvious this is a fixed-point divisor.

Regards,
Peter Hurley