[PATCH 1/1] ARM iMX28: incorrect frac/div calculation

[PATCH 1/1] ARM iMX28: incorrect frac/div calculation

[Matt Burtch]

The frac and div values in _CLK_SET_RATE are calculated incorrectly
for some clock rates ie. CPU clk of 454.73MHz, 392.72MHz, 261.81Mhz,
etc.

Now the frac and div values which result in the closest actual clock
rate to that requested is selected.  Also there are no limitations
on what frac values can be used, as long as they fall in the valid
(18 < frac < 35) range.

Tested on custom iMX28 board.

Signed-off-by: Matt Burtch <matt@grid-net.com>
---
 arch/arm/mach-mxs/clock-mx28.c |    8 ++------
 1 files changed, 2 insertions(+), 6 deletions(-)

diff --git a/arch/arm/mach-mxs/clock-mx28.c b/arch/arm/mach-mxs/clock-mx28.c
index 5d68e41..a058b99 100644
--- a/arch/arm/mach-mxs/clock-mx28.c
+++ b/arch/arm/mach-mxs/clock-mx28.c
@@ -354,19 +354,15 @@ static int name##_set_rate(struct clk *clk, unsigned long rate)		\
 		}							\
 		for (d = 1; d <= div_max; d++) {			\
 			f = parent_rate * 18 / d / rate;		\
-			if ((parent_rate * 18 / d) % rate)		\
-				f++;					\
 			if (f < 18 || f > 35)				\
 				continue;				\
 									\
 			calc_rate = parent_rate * 18 / f / d;		\
-			if (calc_rate > rate)				\
-				continue;				\
 									\
-			if (rate - calc_rate < diff) {			\
+			if (abs(rate - calc_rate) < diff) {		\
 				frac = f;				\
 				div = d;				\
-				diff = rate - calc_rate;		\
+				diff = abs(rate - calc_rate);		\
 			}						\
 									\
 			if (diff == 0)					\
-- 
1.7.5.4

[PATCH 1/1] ARM iMX28: incorrect frac/div calculation

[Shawn Guo]

On Thu, Mar 22, 2012 at 04:05:01PM -0700, Matt Burtch wrote:
> The frac and div values in _CLK_SET_RATE are calculated incorrectly
> for some clock rates ie. CPU clk of 454.73MHz, 392.72MHz, 261.81Mhz,
> etc.
> 
> Now the frac and div values which result in the closest actual clock
> rate to that requested is selected.

My understanding on clk_set_rate is we need to set the clk to a rate
closest to the requested one but never exceeding it.  Does your change
ensure that?

Regards,
Shawn

> Also there are no limitations
> on what frac values can be used, as long as they fall in the valid
> (18 < frac < 35) range.
> 
> Tested on custom iMX28 board.
> 
> Signed-off-by: Matt Burtch <matt@grid-net.com>
> ---
>  arch/arm/mach-mxs/clock-mx28.c |    8 ++------
>  1 files changed, 2 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/arm/mach-mxs/clock-mx28.c b/arch/arm/mach-mxs/clock-mx28.c
> index 5d68e41..a058b99 100644
> --- a/arch/arm/mach-mxs/clock-mx28.c
> +++ b/arch/arm/mach-mxs/clock-mx28.c
> @@ -354,19 +354,15 @@ static int name##_set_rate(struct clk *clk, unsigned long rate)		\
>  		}							\
>  		for (d = 1; d <= div_max; d++) {			\
>  			f = parent_rate * 18 / d / rate;		\
> -			if ((parent_rate * 18 / d) % rate)		\
> -				f++;					\
>  			if (f < 18 || f > 35)				\
>  				continue;				\
>  									\
>  			calc_rate = parent_rate * 18 / f / d;		\
> -			if (calc_rate > rate)				\
> -				continue;				\
>  									\
> -			if (rate - calc_rate < diff) {			\
> +			if (abs(rate - calc_rate) < diff) {		\
>  				frac = f;				\
>  				div = d;				\
> -				diff = rate - calc_rate;		\
> +				diff = abs(rate - calc_rate);		\
>  			}						\
>  									\
>  			if (diff == 0)					\
> -- 
> 1.7.5.4
> 

[PATCH 1/1] ARM iMX28: incorrect frac/div calculation

[Lothar Waßmann]

Hi,

Matt Burtch writes:
> The frac and div values in _CLK_SET_RATE are calculated incorrectly
> for some clock rates ie. CPU clk of 454.73MHz, 392.72MHz, 261.81Mhz,
> etc.
> 
> Now the frac and div values which result in the closest actual clock
> rate to that requested is selected.  Also there are no limitations
> on what frac values can be used, as long as they fall in the valid
> (18 < frac < 35) range.
> 
> Tested on custom iMX28 board.
> 
> Signed-off-by: Matt Burtch <matt@grid-net.com>
> ---
>  arch/arm/mach-mxs/clock-mx28.c |    8 ++------
>  1 files changed, 2 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/arm/mach-mxs/clock-mx28.c b/arch/arm/mach-mxs/clock-mx28.c
> index 5d68e41..a058b99 100644
> --- a/arch/arm/mach-mxs/clock-mx28.c
> +++ b/arch/arm/mach-mxs/clock-mx28.c
> @@ -354,19 +354,15 @@ static int name##_set_rate(struct clk *clk, unsigned long rate)		\
>  		}							\
>  		for (d = 1; d <= div_max; d++) {			\
>  			f = parent_rate * 18 / d / rate;		\
> -			if ((parent_rate * 18 / d) % rate)		\
> -				f++;					\
>  			if (f < 18 || f > 35)				\
>  				continue;				\
>  									\
>  			calc_rate = parent_rate * 18 / f / d;		\
> -			if (calc_rate > rate)				\
> -				continue;				\
>  									\
> -			if (rate - calc_rate < diff) {			\
> +			if (abs(rate - calc_rate) < diff) {		\
Shouldn't this actually be '<=' (also in the original code)?


Lothar Wa?mann
-- 
___________________________________________________________

Ka-Ro electronics GmbH | Pascalstra?e 22 | D - 52076 Aachen
Phone: +49 2408 1402-0 | Fax: +49 2408 1402-10
Gesch?ftsf?hrer: Matthias Kaussen
Handelsregistereintrag: Amtsgericht Aachen, HRB 4996

www.karo-electronics.de | info@karo-electronics.de
___________________________________________________________

[PATCH 1/1] ARM iMX28: incorrect frac/div calculation

[Matt Burtch]

On 03/22/2012 07:52 PM, Shawn Guo wrote:
> On Thu, Mar 22, 2012 at 04:05:01PM -0700, Matt Burtch wrote:
>> The frac and div values in _CLK_SET_RATE are calculated incorrectly
>> for some clock rates ie. CPU clk of 454.73MHz, 392.72MHz, 261.81Mhz,
>> etc.
>>
>> Now the frac and div values which result in the closest actual clock
>> rate to that requested is selected.
> My understanding on clk_set_rate is we need to set the clk to a rate
> closest to the requested one but never exceeding it.  Does your change
> ensure that?
>
> Regards,
> Shawn

It currently doesn't check that the resultant clock is lower than the 
requested.  I can certainly make that change and resubmit.  I wasn't 
aware of this requirement.

So with the patch as it stands, if you request a CPU clock of 391Mhz, 
it'll give you 392.71MHz. Would it be correct for the clock to be set at 
375.65MHz (the next closest valid clock rate without going over)?

Thanks,
Matt

>> Also there are no limitations
>> on what frac values can be used, as long as they fall in the valid
>> (18<  frac<  35) range.
>>
>> Tested on custom iMX28 board.
>>
>> Signed-off-by: Matt Burtch<matt@grid-net.com>
>> ---
>>   arch/arm/mach-mxs/clock-mx28.c |    8 ++------
>>   1 files changed, 2 insertions(+), 6 deletions(-)
>>
>> diff --git a/arch/arm/mach-mxs/clock-mx28.c b/arch/arm/mach-mxs/clock-mx28.c
>> index 5d68e41..a058b99 100644
>> --- a/arch/arm/mach-mxs/clock-mx28.c
>> +++ b/arch/arm/mach-mxs/clock-mx28.c
>> @@ -354,19 +354,15 @@ static int name##_set_rate(struct clk *clk, unsigned long rate)		\
>>   		}							\
>>   		for (d = 1; d<= div_max; d++) {			\
>>   			f = parent_rate * 18 / d / rate;		\
>> -			if ((parent_rate * 18 / d) % rate)		\
>> -				f++;					\
>>   			if (f<  18 || f>  35)				\
>>   				continue;				\
>>   									\
>>   			calc_rate = parent_rate * 18 / f / d;		\
>> -			if (calc_rate>  rate)				\
>> -				continue;				\
>>   									\
>> -			if (rate - calc_rate<  diff) {			\
>> +			if (abs(rate - calc_rate)<  diff) {		\
>>   				frac = f;				\
>>   				div = d;				\
>> -				diff = rate - calc_rate;		\
>> +				diff = abs(rate - calc_rate);		\
>>   			}						\
>>   									\
>>   			if (diff == 0)					\
>> -- 
>> 1.7.5.4
>>

[PATCH 1/1] ARM iMX28: incorrect frac/div calculation

[Matt Burtch]

On 03/23/2012 12:13 AM, Lothar Wa?mann wrote:
> Hi,
>
> Matt Burtch writes:
>> The frac and div values in _CLK_SET_RATE are calculated incorrectly
>> for some clock rates ie. CPU clk of 454.73MHz, 392.72MHz, 261.81Mhz,
>> etc.
>>
>> Now the frac and div values which result in the closest actual clock
>> rate to that requested is selected.  Also there are no limitations
>> on what frac values can be used, as long as they fall in the valid
>> (18<  frac<  35) range.
>>
>> Tested on custom iMX28 board.
>>
>> Signed-off-by: Matt Burtch<matt@grid-net.com>
>> ---
>>   arch/arm/mach-mxs/clock-mx28.c |    8 ++------
>>   1 files changed, 2 insertions(+), 6 deletions(-)
>>
>> diff --git a/arch/arm/mach-mxs/clock-mx28.c b/arch/arm/mach-mxs/clock-mx28.c
>> index 5d68e41..a058b99 100644
>> --- a/arch/arm/mach-mxs/clock-mx28.c
>> +++ b/arch/arm/mach-mxs/clock-mx28.c
>> @@ -354,19 +354,15 @@ static int name##_set_rate(struct clk *clk, unsigned long rate)		\
>>   		}							\
>>   		for (d = 1; d<= div_max; d++) {			\
>>   			f = parent_rate * 18 / d / rate;		\
>> -			if ((parent_rate * 18 / d) % rate)		\
>> -				f++;					\
>>   			if (f<  18 || f>  35)				\
>>   				continue;				\
>>   									\
>>   			calc_rate = parent_rate * 18 / f / d;		\
>> -			if (calc_rate>  rate)				\
>> -				continue;				\
>>   									\
>> -			if (rate - calc_rate<  diff) {			\
>> +			if (abs(rate - calc_rate)<  diff) {		\
> Shouldn't this actually be '<=' (also in the original code)?
Hi Lothar,

It shouldn't make a difference. If it was '<=' it would result in a 
different frac and higher div value but the difference between the 
requested rate and actual would be the same.

Thanks,
Matt Burtch
> Lothar Wa?mann