From mboxrd@z Thu Jan  1 00:00:00 1970
From: Jerry Van Baren <gerald.vanbaren@ge.com>
Date: Tue, 08 Jul 2008 13:50:47 -0400
Subject: [U-Boot-Users] [PATCH] Round the serial port clock divisor
 value returned by calc_divisor()
In-Reply-To: <1215538353-22785-1-git-send-email-hugo.villeneuve@lyrtech.com>
References: <1215538353-22785-1-git-send-email-hugo.villeneuve@lyrtech.com>
Message-ID: <4873A8F7.5080305@ge.com>
List-Id: <u-boot.lists.denx.de>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: u-boot@lists.denx.de

Hugo Villeneuve wrote:
> Round the serial port clock divisor value returned by
> calc_divisor().
> 
> Signed-off-by: Hugo Villeneuve <hugo.villeneuve@lyrtech.com>
> 
> ---
> 
> Rounding is important, especially when using high baud rates
> values like 115200bps. When using the non-rounded value, some
> boards will work and some won't.
> 
>  drivers/serial/serial.c |    7 ++++++-
>  1 files changed, 6 insertions(+), 1 deletions(-)
> 
> diff --git a/drivers/serial/serial.c b/drivers/serial/serial.c
> index 76425d8..7e315ad 100644
> --- a/drivers/serial/serial.c
> +++ b/drivers/serial/serial.c
> @@ -124,6 +124,8 @@ static NS16550_t serial_ports[4] = {
>  
>  static int calc_divisor (NS16550_t port)
>  {
> +	u32 divisor_x10;
> +
>  #ifdef CONFIG_OMAP1510
>  	/* If can't cleanly clock 115200 set div to 1 */
>  	if ((CFG_NS16550_CLK == 12000000) && (gd->baudrate == 115200)) {
> @@ -144,8 +146,11 @@ static int calc_divisor (NS16550_t port)
>  #else
>  #define MODE_X_DIV 16
>  #endif
> -	return (CFG_NS16550_CLK / MODE_X_DIV / gd->baudrate);
>  
> +	/* Compute divisor value with rounding by adding 0.5 */
> +	divisor_x10 = (10 * CFG_NS16550_CLK) / MODE_X_DIV / gd->baudrate;
> +
> +	return (divisor_x10 + 5) / 10;
>  }
>  
>  #if !defined(CONFIG_SERIAL_MULTI)

Hi Hugo,

Will a real rounding work?  Work better?  If I got my mental math and 
parenthesis right and the resulting math doesn't overflow your 
registers, the following will add 1/2 the baud rate scaled by the 
MODE_X_DIV and then perform the divide which will do full rounding.

	return (((CFG_NS16550_CLK + ((gd->baudrate / 2)* MODE_X_DIV))
		/ MODE_X_DIV) / gd->baudrate);

Alternately, I prefer to scale up by 16 and then divide by 8 since 
processors can do that very efficiently ( << 4 followed by >> 3).

Best regards,
gvb