Re: [PATCH v2 1/4] reset: Add support for the Amlogic Meson SoC Reset Controller

[Kevin Hilman]

Philipp Zabel <p.zabel@xxxxxxxxxxxxxx> writes:

> Am Mittwoch, den 25.05.2016, 19:42 -0700 schrieb Kevin Hilman:
>> Neil Armstrong <narmstrong@xxxxxxxxxxxx> writes:
>> 
>> > This patch adds the platform driver for the Amlogic Meson SoC Reset
>> > Controller.
>> >
>> > The Meson8b and GXBB SoCs are supported.
>> >
>> > Signed-off-by: Neil Armstrong <narmstrong@xxxxxxxxxxxx>
>> 
>> Maybe a question for Philipp, but when testog this version with the
>> stmmac ethernet driver on the Amlogic boards, I noticed that ->reset was
>> never getting called.
>> 
>> Turns out, the stmmac driver only uses reset_control_assert() and
>> reset_control_deassert(), both of which return -ENOTSUPP since this
>> driver doesn't provide ->assert or ->deassert, so the driver's ->reset()
>> never gets called.
>> 
>> I haven't looked into the reset framework in detail, but if there's only
>> a ->reset hook, I kind of expected that reset_control_deassert() would
>> use that instead of returning -ENOTSUPP.
>> 
>> If not, what's the proper way of handling hardware that only supports a
>> write-only reset pulse?  Should users of reset_control_* be adapted to
>> check if ->deassert returns -ENOTSUPP and call ->reset?
>
> I initially came up with this for i.MX6, which has a reset controller
> that just like the Meson ones doesn't support manual assertion and
> deassertion of the reset line.
> My assumption then was that reset() is the default for all devices that
> just need their internal state to be reset at some point. I didn't
> expect the clock-like usage of manual deassert()/assert() for power
> management to become so common.
> So assert() and deassert() return -ENOTSUPP if the reset controller
> doesn't support manually asserting or deasserting the reset line.
>
> Of course you can argue that after a reset() pulse the reset line is
> deasserted, and if you instead interpret the API in terms of expected
> outcome, that would be similar to the state after call to deassert()
> (iff the line was asserted before).
> That would blur the line between reset() and deassert() somewhat, and in
> my opinion having a call to deassert() first doing the exact opposite
> isn't good nomenclature.
> Personally, I'd prefer if the driver could switch to only using
> 	reset_control_reset(rstc);

Then what would happen if that same driver is used on platforms that
have ->assert and ->deassert but no -reset?

The bind I'm in is that I'm using an exsting network driver
(stmicro/stmmac) which is used on a bunch of platforms and I dont' know
what all those reset controllers are actually doing, so going and
changing the driver seems problematic.

> or, if the reset line needs to stay asserted during powerdown where the
> reset controller supports it, but doesn't matter whert not, use:
> 	ret = reset_control_deassert(rstc);
> 	if (ret == -ENOTSUPP)
> 		ret = reset_control_reset(rstc);
>
> We could add a helper reset_control_deassert_or_reset() for that.

IMO, that would be a useful helper function, but I still think that
should be the default behavior of _deassert, otherwise it will still
require changing all the drivers.

Kevin