Re: [PATCH v1 1/2] fbtft: Rectify GPIO handling

[Andy Shevchenko]

On Fri, Apr 16, 2021 at 04:37:48PM +0300, Andy Shevchenko wrote:
> On Fri, Apr 16, 2021 at 02:51:59PM +0200, Greg Kroah-Hartman wrote:
> > On Fri, Apr 16, 2021 at 03:31:16PM +0300, Andy Shevchenko wrote:
> > > The infamous commit c440eee1a7a1 ("Staging: fbtft: Switch to
> > > the GPIO descriptor interface") broke GPIO handling completely.
> > > It has already four commits to rectify and it seems not enough.
> > > In order to fix the mess here we:
> > > 
> > >   1) Set default to "inactive" for all requested pins
> > > 
> > >   2) Fix CS, RD, and WR pins polarity since it's active low and
> > >      GPIO descriptor interface takes it into consideration from
> > >      the Device Tree or ACPI
> > > 
> > >   3) Fix RESET pin polarity in the places missed by the commit
> > >      b918d1c27066 ("Staging: fbtft: Fix reset assertion when using gpio descriptor")
> > > 
> > >   4) Consolidate chip activation (CS assert) under default
> > >      ->reset() callback
> > > 
> > > To summarize the expectations about polarity for GPIOs:
> > > 
> > >    #RD			Low
> > >    #WR			Low
> > >    #CS			Low
> > >    #RESET		Low
> > >    DC or RS		High
> > >    RW			High
> > >    Data	0..15		High
> > > 
> > > See also Adafruit learning course [1] for the example of the schematics.
> > > 
> > > While at it, drop unneeded NULL checks, since GPIO API is tolerant to that.
> > > At the end, update TODO to mark this job eventually done.
> > > 
> > > [1]: https://learn.adafruit.com/adafruit-2-8-and-3-2-color-tft-touchscreen-breakout-v2/downloads
> > 
> > Shouldn't this be broken up into "one patch per thing" from your list
> > above?  Feels like you did a lot of different things all in the same
> > patch :(
> 
> I am aware, but breaking to the things here will bring it to the state where
> the functionality is still broken in between. Another point is that the drop
> of unneeded checks will bring the modification of the same line in the code
> twice. Or if you look at 5), for instance, due to CS management breakage,
> fixing it w/o 5) will bring it to the weird case that previously handled CS
> due to lucky defaults from firmware or bootloader, suddenly won't work and
> nothing can help it. The split in this case would look like adding the CS
> handling to all drivers followed by removal all of them. I think it's ugly.
> And so on. Believe me, I really tried hard to split this, but it always
> becomes to undesired result.
> 
> Any ideas how to split that we fix stuff in one commit?

Okay, item 3) can be split to a separate patch.

-- 
With Best Regards,
Andy Shevchenko