Re: [PATCH 1/4 v10] arm: use devicetree to get smp_twd clock

[Russell King - ARM Linux <linux@arm.linux.org.uk>]

Mark,

Rafael just asked me to look at this patch, though I guess these comments
should be directed to Rob who was the original patch author.

On Fri, Jan 04, 2013 at 10:35:43AM -0600, Mark Langsdorf wrote:
> diff --git a/arch/arm/kernel/smp_twd.c b/arch/arm/kernel/smp_twd.c
> index 49f335d..dad2d81 100644
> --- a/arch/arm/kernel/smp_twd.c
> +++ b/arch/arm/kernel/smp_twd.c
> @@ -239,12 +239,15 @@ static irqreturn_t twd_handler(int irq, void *dev_id)
>  	return IRQ_NONE;
>  }
>  
> -static struct clk *twd_get_clock(void)
> +static struct clk *twd_get_clock(struct device_node *np)
>  {
>  	struct clk *clk;
>  	int err;
>  
> -	clk = clk_get_sys("smp_twd", NULL);
> +	if (np)
> +		clk = of_clk_get(np, 0);
> +	else
> +		clk = clk_get_sys("smp_twd", NULL);
>  	if (IS_ERR(clk)) {
>  		pr_err("smp_twd: clock not found: %d\n", (int)PTR_ERR(clk));
>  		return clk;
> @@ -257,6 +260,7 @@ static struct clk *twd_get_clock(void)
>  		return ERR_PTR(err);
>  	}
>  
> +	twd_timer_rate = clk_get_rate(clk);

Hmm, so this overrides the later clk_get_rate() in twd_timer_setup(), making
the later one redundant.  However...

>  	return clk;
>  }
>  
> @@ -285,7 +289,7 @@ static int __cpuinit twd_timer_setup(struct clock_event_device *clk)
>  	 * during the runtime of the system.
>  	 */
>  	if (!common_setup_called) {
> -		twd_clk = twd_get_clock();
> +		twd_clk = twd_get_clock(NULL);
>  
>  		/*
>  		 * We use IS_ERR_OR_NULL() here, because if the clock stubs
> @@ -373,6 +377,8 @@ int __init twd_local_timer_register(struct twd_local_timer *tlt)
>  	if (!twd_base)
>  		return -ENOMEM;
>  
> +	twd_clk = twd_get_clock(NULL);
> +
>  	return twd_local_timer_common_register();

Ok, so this sets up twd_clk, and also twd_timer_rate, but
twd_local_timer_common_register() just ends up registering the set of
function pointers with the local timer code.  Some point later, the
->setup function is called, and that will happen with common_setup_called
false.  The result will be another call to twd_get_clock().

>  }
>  
> @@ -405,6 +411,8 @@ void __init twd_local_timer_of_register(void)
>  		goto out;
>  	}
>  
> +	twd_clk = twd_get_clock(np);
> +
>  	err = twd_local_timer_common_register();

And a similar thing happens here.  Except... the twd_clk gets overwritten
by the call to twd_get_clock(NULL) from twd_timer_setup().

I wonder if it would be much better to move twd_get_clock() out of
twd_timer_setup() entirely, moving it into twd_local_timer_common_register().
twd_local_timer_common_register() would have to take the dev node.
Also, leave the setting of twd_timer_rate in twd_timer_setup().

An alternative strategy would be to move the initialization of the
timer rate also into twd_local_timer_common_register(), detect the
NULL or error clock, and run the calibration from there (I don't think
we can move the calibration).  If that's chosen, then "common_setup_called"
should probably be renamed to "twd_calibration_done".

What do you think?