Re: [PATCHv8 RFC] pwm: Add Freescale FTM PWM driver support

[Bill Pringlemeir <bpringlemeir@nbsps.com>]

On  3 Jan 2014, Li.Xiubo@freescale.com wrote:

> The FTM PWM device can be found on Vybrid VF610 Tower and
> Layerscape LS-1 SoCs.
>
> Signed-off-by: Xiubo Li <Li.Xiubo@freescale.com>
> Signed-off-by: Alison Wang <b18965@freescale.com>
> Signed-off-by: Jingchang Lu <b35083@freescale.com>
> Reviewed-by: Sascha Hauer <s.hauer@pengutronix.de>
> ---
>
> Hi Thierry, Bill

[snip]

> +static unsigned long fsl_pwm_calculate_period_cycles(struct fsl_pwm_chip *fpc,
> +						     unsigned long period_ns,
> +						     enum fsl_pwm_clk index)
> +{
> +	bool bg = fpc->big_endian;
> +	int ret;
> +
> +	fpc->counter_clk_select = FTM_SC_CLK(bg, index);

Yes, this is the spirit of what I was suggesting.  The code is much less
efficient/bigger on the Vybrid with this run-time detection; but this is
more efficient/smaller than previous versions.  I think that 'bg' can be
a compiler '#define' base on the configured SOC-systems.  Ie, if the
kernel config only has 'Vybrid' or only 'LayerScape', then 'bg' can be a
hard coded value.  The compiler will produce much better code in these
cases.

Also, maybe 'distro' people may want to make a 'hand-held' (Debian) or a
'router' (OpenWRT) distribution and they would only pick either 'Vybrid'
or 'LayerScape'.  However, if someone wants an 'every ARM under the
sun', then the code still works.  So, I think that the code is better
setup for a subsequent patch set like this (or at least just a good).

Especially, the stuff on the I/O swapping in the 'readl()' and
'writel()' is no longer needed; I think you can use the same function
for both SOCs.

> +#define __FTM_SWAP32(v) ((u32)(\
> +	(((u32)(v) & (u32)0x000000ffUL) << 24) |\
> +	(((u32)(v) & (u32)0x0000ff00UL) <<  8) |\
> +	(((u32)(v) & (u32)0x00ff0000UL) >>  8) |\
> +	(((u32)(v) & (u32)0xff000000UL) >> 24)))
> +#define FTM_SWAP32(b, v)	(b ? __FTM_SWAP32(v) : v)

I think that there are macros that you could use here.  For instance,
'#include <linux/swab.h>' (powerpc and arm) has some assembler macros
that are quite fast for swapping.  If the kernel config has ARCH >= 6
for ARM, then the very fast 'rev' instruction is used.  If not, then a
generic version is used as you have coded.  The PowerPC (another
possible future ARCH for QorIQ/Layerscape SOC?) always has inline
assembler macros.

So,

+ #include <linux/swab.h>
...
+ #define FTM_SWAP32(b, v)	(b ? __swab32(v) : v)

might be better.

Suggested-by: Bill Pringlemeir <bpringlemeir@nbsps.com>

Thanks,
Bill Pringlemeir.