Le 24/01/2024 à 15:59, Mark Brown a écrit :
> On Wed, Jan 24, 2024 at 02:41:03PM +0100, Jonas Gorski wrote:
> 
>> For some reason we don't store neither the actual number of supported
>> parallel chipselects in the controller, nor the amount of chipselects
>> used by the spi device, so all loops always  need to iterate
>> SPI_CS_CNT_MAX times and check for the chipselect numbers not being
>> 0xff instead of limiting by the (possible to know) actual number of
>> chip selects in use.
> 
> Yes, we really can do a lot better here if we keep a bit more data
> around.

When I see all those loops over SPI_CS_CNT_MAX I have the feeling it 
could have been done a lot easier, for instance by using bitmaps.

Should we revert that commit 4d8ff6b0991d ("spi: Add multi-cs memories 
support in SPI core") and implement something simpler ?

Also have the impression that the commit is doing several things at once 
and should have been split in several commits, for instance that 'if 
((of_property_read_bool(nc, "parallel-memories")) ' stuff seems 
unrelated to the implementation of the generic support of multi-chipselects.

Christophe