Re: [PATCH 3/4] ARM OMAP2+ GPMC: fix WAITMONITORINGTIME divider bug

[Roger Quadros <rogerq@ti.com>]

On 17/02/15 15:48, Robert Abel wrote:
> Hi Roger,
> 
> On Tue, Feb 17, 2015 at 10:27 AM, Roger Quadros <rogerq@ti.com> wrote:
>>
>> Can you please point out which DT had used this pre-divided workaround? We will have to
>> fix those then.
> 
> I didn't check. A cursory glance reveals all DTS in arch/arm/boot/dts
> to use a value of 0.
> 
>>
>> Can you please rebase your patches on top of v3.19?
>> gpmc.c has been moved to drivers/memory/omap-gpmc.c
> 
> 
> Will do.
> 
>>
>> So all the below occurrences of "unsigned int clk_sel" become "enum gpmc_clksel".
> 
> Yes, I should have opted for a cleaner solution from the start.
> 
>>> +static unsigned int gpmc_ns_to_clk_ticks(unsigned int time_ns, int cs, unsigned int clk_sel)
>>>  {
>>>       unsigned long tick_ps;
>>>
>>>       /* Calculate in picosecs to yield more exact results */
>>> -     tick_ps = gpmc_get_fclk_period();
>>> +     tick_ps = gpmc_get_clk_period(cs, clk_sel);
>>>
>>>       return (time_ns * 1000 + tick_ps - 1) / tick_ps;
>>>  }
>>>
>>> +static unsigned int gpmc_ns_to_ticks(unsigned int time_ns)
>>> +{
>>> +     return gpmc_ns_to_clk_ticks(time_ns, 0, GPMC_FCLK);
>>
>> This function should have been unchanged since we're dealing with GPMC_FCLK
>> and it was using gpmc_get_fclk_period().
> 
> 
> Which function? gpmc_ns_to_ticks? It still uses FCLK. I merely did not
> want to have the picosecond formula in two places. I don't see a good
> reason to do so now either...

Agree with you. I missed to see earlier that it was getting rid of the
picosecond formula.

> 
>>
>> Let's call this GPMC_SET_ONE_CLKSEL()
> 
> 
> Will do.
> 
> 
>>
>> You will also need to correct gpmc_cs_show_timings() to show the
>> correct timing for "wait-monitoring-ns" based on GPMC_CLK.
> 
> 
> I was working off 3.14, checked if it worked for 3.17, so both didn't
> have gpmc_cs_show_timings. I'll take a look.
> 
Thanks.

cheers,
-roger