On 11/09/2019 01:27:10+0100, Linus Walleij wrote:

On Thu, Sep 5, 2019 at 3:13 PM Alexandre Belloni
[off-list ref] wrote:

quoted

Implement .get_multiple and .set_multiple to allow reading or setting
multiple pins simultaneously. Pins in the same bank will all be switched at
the same time, improving synchronization and performances.

Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>

Good initiative!

quoted

+       for (bank = 0; bank < atmel_pioctrl->nbanks; bank++) {> +               unsigned int word = bank;
+               unsigned int offset = 0;
+               unsigned int reg;
+
+#if ATMEL_PIO_NPINS_PER_BANK != BITS_PER_LONG

Should it not be > rather than != ?

Realistically, the only case that could happen would be
ATMEL_PIO_NPINS_PER_BANK == 32 and BITS_PER_LONG ==64. so I would go for
ATMEL_PIO_NPINS_PER_BANK < BITS_PER_LONG

quoted

+               word = BIT_WORD(bank * ATMEL_PIO_NPINS_PER_BANK);
+               offset = bank * ATMEL_PIO_NPINS_PER_BANK % BITS_PER_LONG;
+#endif

This doesn't look good for multiplatform kernels.

I don't think we have multiplatform kernels that run both in 32 and 64
bits. I don't believe ATMEL_PIO_NPINS_PER_BANK will ever change, it has
been 32 on all the atmel SoCs since 2001.

We need to get rid of any compiletime constants like this.

Not your fault I suppose it is already there, but this really need
to be fixed. Any ideas?

I can go for a variable instead of a constant but the fact is that there
is currently no 64bit SoC with that IP. I added the compile time check
just in case a 64 bit SoC appears with that IP one day.


-- 
Alexandre Belloni, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel