On Thu, Oct 24, 2019 at 12:13:00PM +0100, Mark Brown wrote:
> On Thu, Oct 24, 2019 at 01:07:57PM +0200, Alvaro Gamez Machado wrote:
> > By leaving this value unset, a default value of 8 was being set later on.
> >
> > If it happens that the SPI master driver doesn't support this value of 8,
> > there will be an initial inconsistency between the SPI master and the device
> > itself. This isn't a problem for most devices because kernel drivers
>
> This will break things, client devices are working on the basis that the
> default transfer width is 8 bits. As I've repeatedly said if we have
> different parts of the system with different ideas about the word size
> we're going to end up with data corruption. Please take this feedback
> on board.
Oh, ok. I didn't understand this cleary from previous mails, now I see what
you mean.
I think then the only way this would be feasible is to check if 8 bits is an
acceptable number for the master and, if it isn't, apply the lowest
available data width. I believe this cannot break anything, as it leaves 8
as the default unless the master can't work with that number, in which case
it really doesn't matter what client device wants because the hardware can't
provide it.
Thanks!
diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c
index 794e20e54237..4e26ac79e133 100644
--- a/drivers/spi/spi.c
+++ b/drivers/spi/spi.c
@@ -3079,8 +3079,12 @@ int spi_setup(struct spi_device *spi)
return -EINVAL;
}
- if (!spi->bits_per_word)
- spi->bits_per_word = 8;
+ if (!spi->bits_per_word) {
+ if (spi->controller->bits_per_word_mask & SPI_BPW_MASK(8))
+ spi->bits_per_word = 8;
+ else
+ spi->bits_per_word = ffs(spi->controller->bits_per_word_mask);
+ }
status = __spi_validate_bits_per_word(spi->controller,
spi->bits_per_word);
--
Alvaro G. M.
