* Gary Bisson [off-list ref] [170314 07:57]:

Hi Linus,

On Tue, Mar 14, 2017 at 03:47:58PM +0100, Linus Walleij wrote:

quoted

On Tue, Feb 28, 2017 at 7:00 AM, Mika Penttilä
[off-list ref] wrote:

quoted

Recent pulls for mainline pre 4.11 introduced pinctrl setup changes and moved pinctrl-imx to
use generic helpers.

Net effect was that hog group could not be resolved. I made it work for myself
with a two stage setup with create and start separated, and dt probe in between them.


Signed-off-by: Mika Penttilä <redacted>

Sorry for including the whole mail body, some people may have missed the
mail. Notably the i.MX maintainers.

Your patch reminds me of the pinctrl_register() vs pinctrl_register_and_init()
introduced by Tony Lindgren, can you look into this in commit
950b0d91dc108f54bccca5a2f75bb46f2df63d29
"pinctrl: core: Fix regression caused by delayed work for hogs"?

Does switching pinctrl_register_and_init() back to pinctrl_register()
solve your problem?

Gary: have you seen any problem like this?

Yes, Mika brought it up as soon as the first 4.11-rc appeared and indeed
this issue can be reproduced on any i.MX6 platform.

Then Tony offered a patch but you haven't replied to him yet:
https://lkml.org/lkml/2017/2/28/140

I think as a regression fix for the -rc series, we should use Mika's
patch with the following tweaks:

1. Instead of adding yet another devm_pinctrl_register_and_init(),
   let's just make devm_pinctrl_register_and_init() not call start
   automatically. If people prefer to use Mika's naming for it
   with nostart, I have no problem with that. But let's then rename
   the current devm_pinctrl_register_and_init() so we are not stuck
   with another interface with issues.

2. Update the four drivers using devm_pinctrl_register_and_init()
   to also call start separately.

Mika, care to update your patch for that?

Regards,

Tony