[PATCH] OMAP/serial: Fix division by zero exception on 3M+ baud rates

[Russell King - ARM Linux]

On Sat, Sep 21, 2013 at 03:43:44AM -0400, Alexey Pelykh wrote:
> diff --git a/drivers/tty/serial/omap-serial.c b/drivers/tty/serial/omap-serial.c
> index 816d1a2..808a880 100644
> --- a/drivers/tty/serial/omap-serial.c
> +++ b/drivers/tty/serial/omap-serial.c
> @@ -240,8 +240,8 @@ serial_omap_baud_is_mode16(struct uart_port *port, unsigned int baud)
>  {
>  	unsigned int n13 = port->uartclk / (13 * baud);
>  	unsigned int n16 = port->uartclk / (16 * baud);
> -	int baudAbsDiff13 = baud - (port->uartclk / (13 * n13));
> -	int baudAbsDiff16 = baud - (port->uartclk / (16 * n16));
> +	int baudAbsDiff13 = n13 ? (baud - (port->uartclk / (13 * n13))) : INT_MAX;
> +	int baudAbsDiff16 = n16 ? (baud - (port->uartclk / (16 * n16))) : INT_MAX;
>  	if(baudAbsDiff13 < 0)
>  		baudAbsDiff13 = -baudAbsDiff13;
>  	if(baudAbsDiff16 < 0)

So, this code is trying to work out whether it's better to use a prescaler
of 13 or 16?  In which case, the above code is rather buggy in many ways.
Let's take an example - what if port->uartclk is 19MHz?

n13 = 19M / 13 * 115200 = 1
n16 = 19M / 16 * 115200 = 1
baudAbsDiff13 = 115200 - (19M / 13 * 1) = 115200 - 146153 = -30953 -> 30953
baudAbsDiff16 = 115200 - (19M / 16 * 1) = 115200 - 118750 = -3550 -> 3350

return (baudAbsDiff13 > baudAbsDiff16); -> 1

That seems like it's right.

Now, what if it's 18MHz?

n13 = 18M / 13 * 115200 = 1
n16 = 18M / 16 * 115200 = 0
baudAbsDiff13 = 115200 - (18M / 13 * 1) = 115200 - 146153 = -23261 -> 23261
baudAbsDiff16 = 115200 - (18M / 16 * 0) = 115200 - 118750 = INT_MAX

return (baudAbsDiff13 > baudAbsDiff16); -> 0

Now, consider the question: with a 18MHz clock, which produces a more
accurate 115200 baud rate - a prescaler of 16 or 13?  Let's go back to
the math.

	115200 * 13 => 1497600
	115200 * 16 => 1843200

So, choosing a prescaler of 16 will give a slower baud rate, but it's
a lot closer to 115200 than using the prescaler of 13.  Yet, the code
will select the latter.

I'd suggest that this code gets rewritten to do "what it says on the tin"
a bit better:

	unsigned n13 = DIV_ROUND_CLOSEST(port->uartclk, 13 * baud);
	unsigned n16 = DIV_ROUND_CLOSEST(port->uartclk, 16 * baud);
	int delta_clk_13 = 13 * baud * n13 - port->uartclk;
	int delta_clk_16 = 16 * baud * n16 - port->uartclk;

	if (delta_clk_13 < 0)
		delta_clk_13 = -delta_clk_13;
	if (delta_clk_16 < 0)
		delta_clk_16 = -delta_clk_16;

	return delta_clk_13 > delta_clk_16;

Note that baud will never be larger than port->uartclk / 13, so n13
will always be greater than 1.  n16 may be zero though, and in this
case, at the point of the test, delta_clk_16 becomes much larger than
delta_clk_13, so the above calculation returns false, meaning we
correctly use a prescaler of 13.

Not only does this avoid the problem with the divider being zero, but
it also selects the prescaler which gives us the closest baud rate to
the one which is being requested.

Finally, serial_omap_get_divisor should also use DIV_ROUND_CLOSEST() -
or for extra points, integrate this into serial_omap_get_divisor(),
and have it also return the prescaler too.