[PATCH 03/74] ST SPEAr: Formalized timer support

Shiraz Hashim shiraz.hashim at st.com
Mon Sep 13 00:04:44 EDT 2010


Dear Jean,

On 9/13/2010 9:07 AM, Jean-Christophe PLAGNIOL-VILLARD wrote:
> On 08:52 Mon 13 Sep     , Shiraz Hashim wrote:
>> Dear Jean,
>>
>> On 9/7/2010 4:25 AM, Jean-Christophe PLAGNIOL-VILLARD wrote:
>>>>  
>>>> -#endif
>>>> +static void __init spear3xx_timer_init(void)
>>>> +{
>>>> +	spear_setup_timer();
>>> why not call it directly?
>>
>> A single level of abstraction was provided so that if archs (spear3xx/6xx/13xx)
>> required to do something extra.
>
> it does not seems to be needed at all so for now call it directly is enough
>

OK. So, I think what could be better is to set the parent clock of the gpt in
the arch specific function like spear3xx_timer_init. This would also remove this
part from the timer code.

>>>> +/* following defines the parent clock to be used */
>>>> +#ifdef CONFIG_ARCH_SPEAR13XX
>>>> +static char pclk_name[] = "osc1_24m_clk";
>>>> +#else
>>>> +static char pclk_name[] = "pll3_48m_clk";
>>>> +#endif
>>> why 2 name?
>>
>> spear13xx has different parent clock for gpt than spear3xx/6xx.
> so use a generic name which will be an alias specify on the soc clock
> so here no need to do the ifdef specially if you want to support both of them
> in the same kernel

As described above would move the parent clock selection to the respective timer
init function for archs. Would that be fine ?

>>
>>> it will be better to use clkdev and create an alias
>>
>> This is specifically to chose parent clock, which can be different in different
>> architectures. clkdev corresponding to parent clock is already defined. We use
>> name here to get the clock.
>>
>>> btw how about move the timer to drivers/clocksource?
>>
>> I don't know whether this is the right place. Every body else has placed timer
>> code in his respective mach/plat directory.
>>
>>>> +
>>>>  static void clockevent_set_mode(enum clock_event_mode mode,
>>>>  				struct clock_event_device *clk_event_dev);
>>>>  static int clockevent_next_event(unsigned long evt,
>>>> @@ -215,7 +222,8 @@ static void __init spear_clockevent_init(void)
>>>>  
>>>>  void __init spear_setup_timer(void)
>>>>  {
>>>> -	struct clk *pll3_clk;
>>>> +	struct clk *clk;
>>>> +	int ret;
>>>>  
>>>>  	if (!request_mem_region(SPEAR_GPT0_BASE, SZ_1K, "gpt0")) {
>>>>  		pr_err("%s:cannot get IO addr\n", __func__);
>>>> @@ -234,26 +242,32 @@ void __init spear_setup_timer(void)
>>>>  		goto err_iomap;
>>>>  	}
>>>>  
>>>> -	pll3_clk = clk_get(NULL, "pll3_48m_clk");
>>>> -	if (!pll3_clk) {
>>>> -		pr_err("%s:couldn't get PLL3 as parent for gpt\n", __func__);
>>>> +	/* get the parent clock */
>>>> +	clk = clk_get(NULL, pclk_name);
>>>> +
>>>> +	if (!clk) {
>>>> +		pr_err("%s:couldn't get %s as parent for gpt\n",
>>>> +				__func__, pclk_name);
>>>>  		goto err_iomap;
>>>>  	}
>>>>  
>>>> -	clk_set_parent(gpt_clk, pll3_clk);
>>>> +	clk_set_parent(gpt_clk, clk);
>>> ??
>>> why not do this in the clock file?

Better to be done in timer init function of respective archs like spear3xx_timer_init

>>
>> The functional clock of the timer very much depends on the parent clock it
>> selects. This is a part of the timer itself, hence it is kept here.
>
> it's seems to be more soc depends than timer
>

yes, OK, would above solution be fine.

regards
Shiraz



More information about the linux-arm-kernel mailing list