Atmel AT91 SD/MMC clock too fast

Atmel AT91 SD/MMC clock too fast

[George Pontis]

Due to integer truncation, the calculated divisor for the SD/MMC clock may
be too low, resulting in a clock that is too fast. For example, the
at91sam9g45 is typically run with a master clock at 133 MHz. There is no
divisor possible that will result in a 25 or 50 MHz clock as is typically
used for a SD card. The existing code will set the clock at 33 MHz or 66 MHz
when 25 or 50 MHz is requested. A Kingston class 10 8G SDHC was found to not
work properly at 33 MHz, but did after patching. The patch, in this case,
generated a 22.1 MHz clock ( 133/6 ). This is the same frequency that this
card sees after booting Linux. If a perfect integer division is possible,
the patch will provide an exact clock rate. If not, the frequency will be
the closest possible without exceeding the target.

The following was based on barebox 2012-07:

--- a/drivers/mci/atmel_mci.c   2012-07-02 01:31:52.000000000 -0700
+++ b/drivers/mci/atmel_mci.c   2012-07-15 23:14:02.203872596 -0700
@@ -76,7 +76,8 @@
        unsigned int clk_in = clk_get_rate(host->clk);

        if (clk_ios > 0) {
-               divider = (clk_in / clk_ios) / 2;
+               /* don't allow integer truncation to result in too small a
divider */
+               divider = (clk_in + 2 * clk_ios - 1) / (2 * clk_ios);
                if (divider > 0)
                        divider -= 1;
        }



_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

Re: Atmel AT91 SD/MMC clock too fast

[Sascha Hauer]

Hi George,

On Mon, Jul 16, 2012 at 12:22:13AM -0700, George Pontis wrote:
> Due to integer truncation, the calculated divisor for the SD/MMC clock may
> be too low, resulting in a clock that is too fast. For example, the
> at91sam9g45 is typically run with a master clock at 133 MHz. There is no
> divisor possible that will result in a 25 or 50 MHz clock as is typically
> used for a SD card. The existing code will set the clock at 33 MHz or 66 MHz
> when 25 or 50 MHz is requested. A Kingston class 10 8G SDHC was found to not
> work properly at 33 MHz, but did after patching. The patch, in this case,
> generated a 22.1 MHz clock ( 133/6 ). This is the same frequency that this
> card sees after booting Linux. If a perfect integer division is possible,
> the patch will provide an exact clock rate. If not, the frequency will be
> the closest possible without exceeding the target.
> 
> The following was based on barebox 2012-07:
> 
> --- a/drivers/mci/atmel_mci.c   2012-07-02 01:31:52.000000000 -0700
> +++ b/drivers/mci/atmel_mci.c   2012-07-15 23:14:02.203872596 -0700
> @@ -76,7 +76,8 @@
>         unsigned int clk_in = clk_get_rate(host->clk);
> 
>         if (clk_ios > 0) {
> -               divider = (clk_in / clk_ios) / 2;
> +               /* don't allow integer truncation to result in too small a
> divider */
> +               divider = (clk_in + 2 * clk_ios - 1) / (2 * clk_ios);

DIV_ROUND_UP(clk_in, clk_ios / 2) instead?

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

RE: Atmel AT91 SD/MMC clock too fast

[George Pontis]

> -----Original Message-----
> From: Sascha Hauer [mailto:s.hauer@pengutronix.de]
> Sent: Monday, July 16, 2012 1:33 AM
> To: George Pontis
> Cc: barebox@lists.infradead.org
> Subject: Re: Atmel AT91 SD/MMC clock too fast
> 
> Hi George,
> 
> On Mon, Jul 16, 2012 at 12:22:13AM -0700, George Pontis wrote:
> > Due to integer truncation, the calculated divisor for the SD/MMC clock
may
> > be too low, resulting in a clock that is too fast. For example, the
> > at91sam9g45 is typically run with a master clock at 133 MHz. There is no
> > divisor possible that will result in a 25 or 50 MHz clock as is
typically
> > used for a SD card. The existing code will set the clock at 33 MHz or 66
> MHz
> > when 25 or 50 MHz is requested. A Kingston class 10 8G SDHC was found to
> not
> > work properly at 33 MHz, but did after patching. The patch, in this
case,
> > generated a 22.1 MHz clock ( 133/6 ). This is the same frequency that
this
> > card sees after booting Linux. If a perfect integer division is
possible,
> > the patch will provide an exact clock rate. If not, the frequency will
be
> > the closest possible without exceeding the target.
> >
> > The following was based on barebox 2012-07:
> >
> > --- a/drivers/mci/atmel_mci.c   2012-07-02 01:31:52.000000000 -0700
> > +++ b/drivers/mci/atmel_mci.c   2012-07-15 23:14:02.203872596 -0700
> > @@ -76,7 +76,8 @@
> >         unsigned int clk_in = clk_get_rate(host->clk);
> >
> >         if (clk_ios > 0) {
> > -               divider = (clk_in / clk_ios) / 2;
> > +               /* don't allow integer truncation to result in too small
a
> > divider */
> > +               divider = (clk_in + 2 * clk_ios - 1) / (2 * clk_ios);
> 
> DIV_ROUND_UP(clk_in, clk_ios / 2) instead?
> 
> Sascha
> 

Looks good, but it needs a divisor of 2*clk_ios. Using that macro:


--- a/drivers/mci/atmel_mci.c   2012-07-02 01:31:52.000000000 -0700
+++ b/drivers/mci/atmel_mci.c   2012-07-15 23:14:02.203872596 -0700
@@ -76,7 +76,8 @@
        unsigned int clk_in = clk_get_rate(host->clk);

        if (clk_ios > 0) {
-               divider = (clk_in / clk_ios) / 2;
+               /* don't allow integer truncation to result in too small a
divider */
+               divider = DIV_ROUND_UP( clk_in, (2 * clk_ios));



_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox