Den 01.09.2015 19:52, skrev Brüns, Stefan:
On Tuesday, September 01, 2015 16:57:14 Noralf Trønnes wrote:
IMHO, this is a bad idea. A controller supporting more than 32 bpw can
not set the bpw mask, as the mask is limited to 32 bits. Thus it has to
check the bpw in its own setup function, and may or may not reject 9 bpw.
Why should one penalize any controllers having an unset bpw mask, which
is a completely valid configuration?
BTW, the SPI core patch needed for this to work is in the 4.3 pull
request.
I have acked a patch from Stefan Wahren adressing this issue:
http://driverdev.linuxdriverproject.org/pipermail/driverdev-devel/2015-Augus
t/075423.html
Two issues:
1st: Stefan Wahren posted a patch large part identical to mine, 18 hours
later. Coincidence?
For my part I admit to being sloppy here, I should have pulled you into
the thread.
According to the commit message, Stefan Wahren seem to have detected
this problem
on the ARM MXS platform and you refere to xilinx, so I guess it's a
coincidence.
2nd: Modifying bpw and not calling spi_setup() is IMHO wrong. The controller
setup function could e.g. modify max_speed_hz.
We are only changing bits_per_word, and spi_setup() has already been called.
Callchain for DT registered spi devices:
spi_register_master->of_register_spi_devices->of_register_spi_device->
spi_add_device->spi_setup
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
