Re: [PATCH v2 1/2] gpiolib: Fix a mess with the GPIO_* flags

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Fri, Apr 12, 2024 at 10:20:24AM +0200, Linus Walleij wrote:
> On Tue, Apr 9, 2024 at 1:17 AM Andy Shevchenko
> <andriy.shevchenko@linux.intel.com> wrote:
> 
> > The GPIO_* flag definitions are *almost* duplicated in two files
> > (with unmatches OPEN_SOURCE / OPEN_DRAIN). Moreover, some code relies
> > on one set of definitions while the rest is on the other. Clean up
> > this mess by providing only one source of the definitions to all.
> >
> > Fixes: b424808115cb ("brcm80211: brcmsmac: Move LEDs to GPIO descriptors")
> > Fixes: 5923ea6c2ce6 ("gpio: pass lookup and descriptor flags to request_own")
> > Fixes: fed7026adc7c ("gpiolib: Make use of enum gpio_lookup_flags consistent")
> > Fixes: 4c0facddb7d8 ("gpio: core: Decouple open drain/source flag with active low/high")
> > Fixes: 69d301fdd196 ("gpio: add DT bindings for existing consumer flags")
> > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> 
> The way the line lookup flags ("lflags") were conceived was through
> support for non-DT systems using descriptor tables, and that is how
> enum gpio_lookup_flags came to be.
> 
> When OF support was added it was bolted on on the side, in essence
> assuming that the DT/OF ABI was completely separate (and they/we
> sure like to think about it that way...) and thus needed translation from
> OF flags to kernel-internal enum gpio_lookup_flags.
> 
> The way *I* thought about this when writing it was certainly that the
> DT bindings was a separate thing (<dt-bindings/*.h> didn't even exist
> at the time I think) and that translation from OF to kernel-internal
> lflags would happen in *one* place.
> 
> The main reasoning still holds: the OF define is an ABI, so it can
> *never* be changed, but the enum gpio_lookup_flags is subject to
> Documentation/process/stable-api-nonsense.rst and that means
> that if we want to swap around the order of the definitions we can.
> 
> But admittedly this is a bit over-belief in process and separation of
> concerns and practical matters may be something else...

Got it. But we have a name clash and the mess added to the users.
I can redo this to separate these entities.

Note, that there is code in the kernel that *does* use
#include <dt-bindings/*.h>
for Linux internals.

$ git grep -lw '^#include <dt-bindings/.*\.h>' -- drivers/ | xargs dirname | cut -f 1,2 -d '/' | sort -u
drivers/bus
drivers/clk
drivers/clocksource
drivers/cpufreq
drivers/dma
drivers/firmware
drivers/gpio
drivers/gpu
drivers/hwtracing
drivers/i2c
drivers/iio
drivers/input
drivers/interconnect
drivers/iommu
drivers/irqchip
drivers/leds
drivers/mailbox
drivers/media
drivers/memory
drivers/mfd
drivers/net
drivers/phy
drivers/pinctrl
drivers/platform
drivers/pmdomain
drivers/power
drivers/pwm
drivers/regulator
drivers/remoteproc
drivers/reset
drivers/rtc
drivers/soc
drivers/spmi
drivers/thermal
drivers/tty
drivers/video
drivers/watchdog

P.S>
One of the patch this tries to fix is yours IIRC :-)


-- 
With Best Regards,
Andy Shevchenko