Re: [PATCH v3 0/2] Add support for SYSCON reset

[Rob Herring]

On Wed, Jun 15, 2016 at 11:45 AM, Philipp Zabel <p.zabel@xxxxxxxxxxxxxx> wrote:
> Hi Andrew,
>
> Am Mittwoch, den 15.06.2016, 10:48 -0500 schrieb Andrew F. Davis:
>> On 06/01/2016 01:41 PM, Andrew F. Davis wrote:
>> > Some SoCs contain reset controls for modules that are memory-mapped to
>> > areas shared with other module configuration settings. This requires
>> > synchronization across all drivers accessing this memory area. This
>> > series adds a generic SYSCON reset driver to allow resets toggled
>> > by bits in memory-mapped registers through SYSCON.
>> >
>> > Changes from v2:
>> >  - Rebased on v4.7-rc1
>> >  - Removed the need to give reset specifier nodes an index address
>> >
>> > Changes from v1:
>> >  - Reset control information is now described in the reset node, this
>> >    keeps the reset information centralized for easy verification
>> >  - Other small fixups
>> >
>> > Andrew F. Davis (2):
>> >   Documentation: dt: reset: Add syscon reset binding
>> >   reset: add a SYSCON based reset driver
>> >
>> >  .../devicetree/bindings/reset/syscon-reset.txt     | 105 ++++++++
>> >  drivers/reset/Kconfig                              |  10 +
>> >  drivers/reset/Makefile                             |   1 +
>> >  drivers/reset/reset-syscon.c                       | 289 +++++++++++++++++++++
>> >  include/dt-bindings/reset/syscon.h                 |  23 ++
>> >  5 files changed, 428 insertions(+)
>> >  create mode 100644 Documentation/devicetree/bindings/reset/syscon-reset.txt
>> >  create mode 100644 drivers/reset/reset-syscon.c
>> >  create mode 100644 include/dt-bindings/reset/syscon.h
>> >
>>
>> Is there any additional feedback for this driver? I normally try to
>> refrain from pings, but this may begin to block further upstreaming of
>> IPs that uses this reset if it can not be taken this cycle.
>
> There's still Rob's concern that this binding needs a device tree node
> per single register bit in the worst case, which seems a bit overkill.

Yes, that's still my concern with this.

> I'd still prefer to have this information hidden in the drivers, but if
> you absolutely have to put it in the device tree, maybe an approach more
> similar to pinctrl-simple, where you could list all resets, one per
> line, in a single property, would be more acceptable:
>
> pscrst: reset-controller {
>         compatible = "ti,<chip>-pscrst", "syscon-reset";
>
>         /* syscon-reset,bits = <ctrl_reg ctrl_bit stat_reg stat_bit flags>; */

ti,reset-bits

>         syscon-reset,bits = <0xa3c 8 0x83c 8 RESET_ASSERT_CLEAR  /* 0: pcrst-dsp */
>                              0xa40 5 0     0 RESET_TRIGGER_SET>; /* 1: pcrst-example */
> };
>
> dsp0: dsp {
>         resets = <&pscrst 0>;
> };

Otherwise, this seems okay. It is more concise.

>
>> If there is still an objection to calling this a generic reset solution
>> we would not strongly object to relabeling this to something implying
>> more TI exclusivity.
>
> Good. Right now there don't seem to be that many reset controllers in
> the wild for which this binding would be a good fit. If this turns out
> to be useful for others, we can add more compatibles to the driver.

Certainly other users would make a generic solution more compelling.

Rob