From mboxrd@z Thu Jan  1 00:00:00 1970
From: Dirk Behme <dirk.behme@gmail.com>
Date: Sat, 04 May 2013 12:06:43 +0200
Subject: [U-Boot] [PATCH] spi: mxc_spi: Fix pre and post divider
	calculation
In-Reply-To: <51842265.7010103@boundarydevices.com>
References: <1367492386-20464-1-git-send-email-dirk.behme@de.bosch.com>
	<5182B2AA.9040606@boundarydevices.com>
	<51835215.90602@de.bosch.com>
	<51842265.7010103@boundarydevices.com>
Message-ID: <5184DDB3.90901@gmail.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

Am 03.05.2013 22:47, schrieb Troy Kisky:
> On 5/2/2013 10:58 PM, Dirk Behme wrote:
>> Do you want to say you propose
>>
>> post_div = pre_div / 16;
>> pre_div = 16;
>>
>> ?
> yes, that's what I said
>>
>> If so:
>>
>> First, I agree that we have to use the same dividers in both lines.
>>
>> But, second, this would mean that you use /16 as max pre_div. For
>> the i.MX6 case where clk_src is 60MHz this would result in a
>> pre-divided clock of 3.75Mhz (instead of 4MHz with /15).
>
> That does sound better for i.MX6, what about other processors using
> this file?
>
>>
>> So using /15 or /16 is just a decision of which end clocks most
>> probably are needed.
>>
>> If you want to be able to configure 4MHz, 2MHz, 1MHz, 500kHz etc
>> then /15 is the better choice.
>>
>> If you want to be able to configure 3.75Mhz, 1.875MHz, 937.5kHz,
>> 468.75kHz etc then /16 is the better choice.
>>
>> I vote for /15 as done by my patch.
>
> Thanks for explaining. The downside of using /15 is that you can't get
> the slowest clock possible.

Yes. I was looking for the _highest_ clock possible, though ;) And 
this isn't correctly done by the recent code. This is why I was 
looking into it ...

> How about restructuring the code to improve both. Calculate post_div
> first.
>
> pre_div = DIV_ROUND_UP(clk_src, max_hz);
> /* fls(1) = 1, fls(0x80000000) = 32, fls(16) = 5 */
> post_div = fls(pre_div - 1);
> if (post_div > 4)
>      post_div -= 4;
> else
>      post_div = 0;
>
> if (post_div >= 16) {
>             printf("Error: no divider for the freq: %d\n",
>                                          max_hz);
>             return -1;
> }
> pre_div = (pre_div + (1 << post_div) - 1) >> post_div;

Using my test code gives the correct values using this algorithm. So 
yes, sounds good.

Just a small note: Wouldn't it be better to put the printf and the 
last line with the pre_div calculation into the if(post_div > 4) part? 
I.e.

pre_div = DIV_ROUND_UP(clk_src, max_hz);
/* fls(1) = 1, fls(0x80000000) = 32, fls(16) = 5 */
post_div = fls(pre_div - 1);
if (post_div > 4) {
     post_div -= 4;

     if (post_div >= 16) {
            printf("Error: no divider for the freq: %d\n",
                                         max_hz);
            return -1;
     }
     pre_div = (pre_div + (1 << post_div) - 1) >> post_div;

} else
     post_div = 0;

?

In case we agree on this, I'm thinking about doing 2 patches to make 
clear what we are doing:

1. Re-doing my initial patch with

post_div = pre_div / 16;
pre_div = 16;

This would be the "fix the issues in the existing (non-optimal) 
algorithm but keep that" patch.

2. Replace the existing algorithm with your above version. This would 
be the "improve the algorithm" patch.

What do you think?

Best regards

Dirk