[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[Linus Walleij]

On Wed, Jun 18, 2014 at 3:37 PM, Houcheng Lin <houcheng@gmail.com> wrote:

> The Problem
> -----------
> The reset signal on a hardware board is send either:
>     - during machine initialization
>     - during bus master's initialization

I just thought about this a bit, since there isn't already a generic GPIO
reset driver, just call this drivers/reset/reset-gpio.c and make the
ability to deferral just a configuration detail of the GPIO reset driver.

Yours,
Linus Walleij

[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[Maxime Ripard]

On Tue, Jul 08, 2014 at 09:52:03AM +0200, Linus Walleij wrote:
> On Wed, Jun 18, 2014 at 3:37 PM, Houcheng Lin <houcheng@gmail.com> wrote:
> 
> > The Problem
> > -----------
> > The reset signal on a hardware board is send either:
> >     - during machine initialization
> >     - during bus master's initialization
> 
> I just thought about this a bit, since there isn't already a generic GPIO
> reset driver, just call this drivers/reset/reset-gpio.c and make the
> ability to deferral just a configuration detail of the GPIO reset driver.

Philipp has been working on one for quite some time. See
http://www.spinics.net/lists/arm-kernel/msg321927.html

However, it seems to progress slowly, and we don't seem to be able to
reach a consensus here.

If you ask me, having to set a few extra properties like this just
advocates for a regular reset driver and DT node for the reset GPIO,
but I'm pretty sure Philipp will feel otherwise :)

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20140708/0509ca19/attachment.sig>

[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[Linus Walleij]

On Tue, Jul 8, 2014 at 10:05 AM, Maxime Ripard
<maxime.ripard@free-electrons.com> wrote:
> On Tue, Jul 08, 2014 at 09:52:03AM +0200, Linus Walleij wrote:
>> On Wed, Jun 18, 2014 at 3:37 PM, Houcheng Lin <houcheng@gmail.com> wrote:
>>
>> > The Problem
>> > -----------
>> > The reset signal on a hardware board is send either:
>> >     - during machine initialization
>> >     - during bus master's initialization
>>
>> I just thought about this a bit, since there isn't already a generic GPIO
>> reset driver, just call this drivers/reset/reset-gpio.c and make the
>> ability to deferral just a configuration detail of the GPIO reset driver.
>
> Philipp has been working on one for quite some time. See
> http://www.spinics.net/lists/arm-kernel/msg321927.html
>
> However, it seems to progress slowly, and we don't seem to be able to
> reach a consensus here.
>
> If you ask me, having to set a few extra properties like this just
> advocates for a regular reset driver and DT node for the reset GPIO,
> but I'm pretty sure Philipp will feel otherwise :)

Hm haha yeah let's fight it out :-)

This is not my subsystem so I'm getting some popcorn.

Yours,
Linus Walleij

[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[Philipp Zabel]

Am Dienstag, den 08.07.2014, 11:38 +0200 schrieb Linus Walleij:
> On Tue, Jul 8, 2014 at 10:05 AM, Maxime Ripard
> <maxime.ripard@free-electrons.com> wrote:
> > On Tue, Jul 08, 2014 at 09:52:03AM +0200, Linus Walleij wrote:
> >> On Wed, Jun 18, 2014 at 3:37 PM, Houcheng Lin <houcheng@gmail.com> wrote:
> >>
> >> > The Problem
> >> > -----------
> >> > The reset signal on a hardware board is send either:
> >> >     - during machine initialization
> >> >     - during bus master's initialization
> >>
> >> I just thought about this a bit, since there isn't already a generic GPIO
> >> reset driver, just call this drivers/reset/reset-gpio.c and make the
> >> ability to deferral just a configuration detail of the GPIO reset driver.
> >
> > Philipp has been working on one for quite some time. See
> > http://www.spinics.net/lists/arm-kernel/msg321927.html
> >
> > However, it seems to progress slowly, and we don't seem to be able to
> > reach a consensus here.

Mostly because Maxime and I seem to have a completely different opinion
and nobody else argued one way or the other.

> > If you ask me, having to set a few extra properties like this just
> > advocates for a regular reset driver and DT node for the reset GPIO,
> > but I'm pretty sure Philipp will feel otherwise :)
> 
> Hm haha yeah let's fight it out :-)
> 
> This is not my subsystem so I'm getting some popcorn.

Sorry about missing this earlier, I hope the popcorn is not stale.

I think a reset that needs to be released before a fixed device appears
on the bus, should be handled by the bus driver. The reset framework
could be made to help with that, but I don't think a separate entity
that scans the whole device tree itself is the right solution.

regards
Philipp

[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[Maxime Ripard]

On Fri, Aug 08, 2014 at 04:23:09PM +0200, Philipp Zabel wrote:
> Am Dienstag, den 08.07.2014, 11:38 +0200 schrieb Linus Walleij:
> > On Tue, Jul 8, 2014 at 10:05 AM, Maxime Ripard
> > <maxime.ripard@free-electrons.com> wrote:
> > > On Tue, Jul 08, 2014 at 09:52:03AM +0200, Linus Walleij wrote:
> > >> On Wed, Jun 18, 2014 at 3:37 PM, Houcheng Lin <houcheng@gmail.com> wrote:
> > >>
> > >> > The Problem
> > >> > -----------
> > >> > The reset signal on a hardware board is send either:
> > >> >     - during machine initialization
> > >> >     - during bus master's initialization
> > >>
> > >> I just thought about this a bit, since there isn't already a generic GPIO
> > >> reset driver, just call this drivers/reset/reset-gpio.c and make the
> > >> ability to deferral just a configuration detail of the GPIO reset driver.
> > >
> > > Philipp has been working on one for quite some time. See
> > > http://www.spinics.net/lists/arm-kernel/msg321927.html
> > >
> > > However, it seems to progress slowly, and we don't seem to be able to
> > > reach a consensus here.
> 
> Mostly because Maxime and I seem to have a completely different opinion
> and nobody else argued one way or the other.

Yep, mostly because I don't see how a generic approach can work.

The existing reset-gpios property only provide the gpio to use, but
some informations are encoded in the driver, such as the reset
duration, or a reset sequence if any.

How do you plan on giving that information to your generic driver?

The only solution I can think of would be to add an extra property
that your code would parse. But then, you break the existing DT
bindings.

And if we're going to break those bindings, at least do it in a way
consistent with reset bindings.

Plus, your approach doesn't cover the weird corner cases such as:
  - reset-gpio
  - wlf,reset-gpios
  - phy-reset-gpios
  - snps,reset-gpio
  - the drivers that need several gpio and expect the reset one as a
    positional argument.
  - etc.

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20140811/8b38b659/attachment.sig>

[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[Philipp Zabel]

Hi Maxime,

Am Montag, den 11.08.2014, 19:33 +0200 schrieb Maxime Ripard:
> > Mostly because Maxime and I seem to have a completely different opinion
> > and nobody else argued one way or the other.
> 
> Yep, mostly because I don't see how a generic approach can work.
> 
> The existing reset-gpios property only provide the gpio to use, but
> some informations are encoded in the driver, such as the reset
> duration, or a reset sequence if any.

The driver should provide the duration. I'd really like to see an
example where sequencing is necessary.
I agree that as soon as things get significantly more complicated than
pulsing a single GPIO, the reset-gpios binding is too limited.
Still, I'm not happy to mandate a separate gpio reset device for each
reset line if most devices are simple enough for it to work without.
What about using reset-gpios for the majority of simple cases and have a
separate gpio-reset-sequencer driver when multiple GPIO resets have to
be timed?

> How do you plan on giving that information to your generic driver?
> 
> The only solution I can think of would be to add an extra property
> that your code would parse. But then, you break the existing DT
> bindings.
>
> And if we're going to break those bindings, at least do it in a way
> consistent with reset bindings.

For the backwards compatibility case, the driver already has to provide
the duration. I don't want to break the existing bindings at all.

> Plus, your approach doesn't cover the weird corner cases such as:
>   - reset-gpio
>   - wlf,reset-gpios
>   - phy-reset-gpios
>   - snps,reset-gpio
>   - the drivers that need several gpio and expect the reset one as a
>     positional argument.
>   - etc.

Those are just an issue of the implementation I posted earlier because
gpiod_get doesn't support custom names other than %s-gpios. This could
be extended and handled just as well if deemed necessary.

regards
Philipp

[RFC PATCH v3] reset: Add a defer reset object to send board specific reset

[Maxime Ripard]

On Thu, Aug 14, 2014 at 11:36:38AM +0200, Philipp Zabel wrote:
> Hi Maxime,
> 
> Am Montag, den 11.08.2014, 19:33 +0200 schrieb Maxime Ripard:
> > > Mostly because Maxime and I seem to have a completely different opinion
> > > and nobody else argued one way or the other.
> > 
> > Yep, mostly because I don't see how a generic approach can work.
> > 
> > The existing reset-gpios property only provide the gpio to use, but
> > some informations are encoded in the driver, such as the reset
> > duration, or a reset sequence if any.
> 
> The driver should provide the duration.

How? This used to be in the code, and reset_control_reset doesn't take
such argument.

> I'd really like to see an example where sequencing is necessary.

Well, if you have several reset lines, the sequencing between each
might be important.

> I agree that as soon as things get significantly more complicated than
> pulsing a single GPIO, the reset-gpios binding is too limited.

While the generic reset bindings are perfect for that.

> Still, I'm not happy to mandate a separate gpio reset device for each
> reset line if most devices are simple enough for it to work without.

Well, it's pretty much already the case for other subsystems, such as
regulator.

I guess we can treat this as a legacy option, but allow the reset-gpio
code to be a full driver of its own, if we need more advanced use
cases?

> What about using reset-gpios for the majority of simple cases and have a
> separate gpio-reset-sequencer driver when multiple GPIO resets have to
> be timed?

I don't know. I feel like it should be in the driver itself, rather
than in a generic layer.

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20140814/36248234/attachment.sig>