Re: Multi driver SPI chip select count unenforced.

[Lukas Wunner <lukas@xxxxxxxxx>]

On Fri, Apr 16, 2021 at 11:55:45AM +0100, Joe Burmeister wrote:
> However, after I found the same broken pattern in a few places.
> You can find suspects by grep'ing in the SPI folder for "->chip_select]".

You need to cc the maintainers of the affected drivers.
You can retrieve them with "scripts/get_maintainer.pl drivers/spi/spi-...".

Please don't attach patches but rather submit each one separately.
Pass a commit range (abcdefg..0123456) to "git format-patch" to
get numbered patches, then submit them with msmtp or "git send-email".


> Subject: [PATCH] Handle SPI device setup callback failure.

Prepend the subsystem to the subject and drop the period, e.g.:

spi: Handle SPI device setup callback failure

Patch otherwise LGTM.


> Subject: [PATCH] Remove BCM2835 SPI chipselect limit.

Prepend "spi: bcm2835: " in this case and drop the period.
Use "git log --oneline <filename>" to see what the prefix should look
like for a particular file.


> The limit of 4 chipselects for the BCM2835 was not required and also was
> not inforced. Without inforcement it was possible to make a device tree
> over this limit which would trample memory.
> 
> The chipselect count is now obtained from the device tree and expanded
> if more devices are added.

I'd prefer it if you could just raise BCM2835_SPI_NUM_CS to 6
(or 8 or whatever you need).  Use commit 603e92ff10a8 as a template.

Then submit a separate patch to error out of bcm2835_spi_setup()
if bcm2835_spi_setup >= BCM2835_SPI_NUM_CS.

Honestly I think the additional code complexity isn't worth it to allow
for dynamic resizing.  Raising BCM2835_SPI_NUM_CS results in just a
few additional bytes, that's probably smaller than the increase of the
text segment due to the additional resizing code.

Nit: I think the correct spelling is "enforce".  (Not an English native
speaker but my dictionary says so.)

Thanks!

Lukas