Re: [PATCH] Revert "mfd: syscon: Remove repetition of the regmap_get_val_endian()"

[Andy Shevchenko]

On Sat, Oct 08, 2022 at 09:45:16AM -0700, Linus Torvalds wrote:
> On Sat, Oct 8, 2022 at 8:47 AM Jason A. Donenfeld <Jason@xxxxxxxxx> wrote:
> >
> > This reverts commit 72a95859728a7866522e6633818bebc1c2519b17. It broke
> > reboots on big-endian MIPS and MIPS64 malta QEMU instances, which use
> > the syscon driver. Little-endian is not effected, which means likely
> > it's important to handle regmap_get_val_endian() in this function after
> > all.
> 
> Hmm. The revert may indeed be the right thing to do, but we're still
> early in the release process, so let's go through the channels.
> 
> I do note that commit 72a95859728a points to commit 0dbdb76c0ca8
> ("regmap: mmio: Parse endianness definitions from DT") as the reason
> why it's not necessary any more, but that commit
> 
>  (a) doesn't seem to set config->val_format_endian (which somebody may
> care about). It does set the operation pointers etc, but doesn't set
> that field.

It should.

of_syscon_register() calls to regmap_init_mmio() with syscon_config data
structure as a parameter.

Before 72a95859728a the of_syscon_register() fills the val_format_endian with
something it parses from DT. After that commit the default value (0) is
REGMAP_ENDIAN_DEFAULT. Now when __regmap_init_mmio_clk() is called it
creates a context base on DT since the field is 0.

>  (b) it uses regmap_get_val_endian(), which doesn't actually even look
> at the OF properties unless config->val_format_endian is
> REGMAP_ENDIAN_DEFAULT

Which is 0!

> so the code that commit 72a95859728a removed was actually quite a bit
> different from the code in commit 0dbdb76c0ca8.
> 
> Maybe the problem is related to those semantic differences, and is
> easy to fix for somebody who knows what the heck that stuff is doing.

But while looking into this, I think I know what is going on,
of_syscon_register() calls regmap API with dev == NULL, hence
fwnode == NULL, hence nothing to read from DT.

But default (via regmap bus configuration) is LE and LE works fine.

> And if not, please just send me the revert through the normal channels. Ok?

Yeah, revert is a good move here.

For real deduplication we need to either add a regmap_get_val_endian() kind
that takes fwnode as a parameter and call it explicitly or propagate fwnode
to __regmap_init_mmio_clk() somehow.

-- 
With Best Regards,
Andy Shevchenko