Re: [PATCHv3 3/3] clocksource: Add Freescale FlexTimer Module (FTM) timer support

[Daniel Lezcano]

On 05/19/2014 04:26 AM, Li.Xiubo@xxxxxxxxxxxxx wrote:
> > > +#define FTM_CNTIN	0x4C
> > > +
> > > +static void __iomem *clksrc_base;
> > > +static void __iomem *clkevt_base;
> > > +static unsigned long peroidic_cyc;
> > > +static unsigned long ps;
> > > +bool big_endian;
> > > +
> >
> > Usually this is encaspulated in a structure.
> >
> > struct ftm_clock_device {
> > 	void __iomem *clksrc_base;
> > 	...
> > };
> >
> >
> > > +static inline u32 ftm_readl(void __iomem *addr)
> > > +{
> > > +	if (big_endian)
> >
> > I am not a big fan of addressing global variables in the functions, so
> > if you can pass the structure pointer around here and the other
> > functions instead that would be nice.
> >
> > Otherwise the patch sounds ok. Thanks for taking care of encapsulating
> > well the functions and commenting the code.
>
> Yes, I did think so.
>
> But some callbacks like ï
> + static u64 ftm_read_sched_clock(void)
> + {
> +         return ftm_readl(clksrc_base + FTM_CNT);
> + }
>
> Used by :
> + sched_clock_register(ftm_read_sched_clock,....);
>
> If they are encapsulated in a structure, and should the struct instance
> be one global instance too ? I'm doubting whether will this make sense ?

Actually, I plan in a near future to consolidate the code and factor out 
some parts with a common structure across the different drivers. So even 
if you address the base@ with the global instance but pass around the 
structure as parameter, that will be ok because that will be less 
modifications in the future. It is not a strong requirement, just put in 
place some encapsulation to make the life easier for after.

> > > +static int __init ftm_calc_closest_round_cyc(unsigned long freq)
> > > +{
> > > +	ps = 0;
> > > +
> > > +	do {
> > > +		peroidic_cyc = DIV_ROUND_CLOSEST(freq, HZ * (1 << ps++));
> > > +	} while (peroidic_cyc > 0xFFFF);
> > > +
> > > +	if (ps > 7) {
> > > +		pr_err("ftm: the max prescaler is %lu > 7\n", ps);
> > > +		return -EINVAL;
> > > +	}
> > > +
> >
> > Can you explain how this error can happen ?
>
> Yes, the hardware limitation of the 'ps' is 0~7, and the counter register
> Is only using the lower 16 bits.
> If the 'freq' value is too big here, then the periodic_cyc may exceed 0xFFFF.
>
> Or should I add some comment here ?

Yes, a comment will be welcome.

Thanks
  -- Daniel

--
 <http://www.linaro.org/> Linaro.org â Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/