Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <CAMRc=MdyT7nMJ7dUTf6M3PvRdynaCurb_TftMyA4eCwzpR9fVA@mail.gmail.com>
 <20190212101835.GB20638@dell> <CAMRc=MfYeVbvY6O03xq4eLN9TVpAKWCAXPk0eP5LSBCU7VU9fA@mail.gmail.com>
 <20190212111403.GC20638@dell> <CAMpxmJWXOb1HW4QiEn8gMp3Nh05ahE0CMLXkac+hEKRjtGxTHQ@mail.gmail.com>
 <20190212132016.GA4781@dell> <20190213092553.GE1863@dell> <CAMRc=MeTMq_GWzQB6LBvts1csMftj=On5sB22jsJZzQzONxbNg@mail.gmail.com>
 <20190213095330.GF1863@dell> <CAMpxmJXqEwNtyT1ahmRd05QOjhdi0rT_-5S9Lk8O-PmV8BuA6w@mail.gmail.com>
 <20190213103946.GG1863@dell>
In-Reply-To: <20190213103946.GG1863@dell>
From:   Bartosz Golaszewski <brgl@bgdev.pl>
Date:   Wed, 13 Feb 2019 11:46:30 +0100
Message-ID: <CAMRc=Mctzwh7=gt43h3pra_SPoG8=RyKSVhe2zC3XgbjsaL56Q@mail.gmail.com>
Subject: Re: [PATCH v4 05/10] mfd: max77650: new core mfd driver
To:     Lee Jones <lee.jones@linaro.org>
Cc:     Bartosz Golaszewski <bgolaszewski@baylibre.com>,
        Rob Herring <robh+dt@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Linus Walleij <linus.walleij@linaro.org>,
        Dmitry Torokhov <dmitry.torokhov@gmail.com>,
        Jacek Anaszewski <jacek.anaszewski@gmail.com>,
        Pavel Machek <pavel@ucw.cz>,
        Sebastian Reichel <sre@kernel.org>,
        Liam Girdwood <lgirdwood@gmail.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        "open list:GPIO SUBSYSTEM" <linux-gpio@vger.kernel.org>,
        devicetree <devicetree@vger.kernel.org>,
        Linux Input <linux-input@vger.kernel.org>,
        Linux LED Subsystem <linux-leds@vger.kernel.org>,
        Linux PM list <linux-pm@vger.kernel.org>,
        Mark Brown <broonie@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

=C5=9Br., 13 lut 2019 o 11:39 Lee Jones <lee.jones@linaro.org> napisa=C5=82=
(a):
>
> On Wed, 13 Feb 2019, Bartosz Golaszewski wrote:
>
> > =C5=9Br., 13 lut 2019 o 10:53 Lee Jones <lee.jones@linaro.org> napisa=
=C5=82(a):
> > >
> > > On Wed, 13 Feb 2019, Bartosz Golaszewski wrote:
> > >
> > > > =C5=9Br., 13 lut 2019 o 10:25 Lee Jones <lee.jones@linaro.org> napi=
sa=C5=82(a):
> > > > >
> > > > > On Tue, 12 Feb 2019, Lee Jones wrote:
> > > > >
> > > > > > On Tue, 12 Feb 2019, Bartosz Golaszewski wrote:
> > > > > >
> > > > > > > wt., 12 lut 2019 o 12:14 Lee Jones <lee.jones@linaro.org> nap=
isa=C5=82(a):
> > > > > > > >
> > > > > > > > On Tue, 12 Feb 2019, Bartosz Golaszewski wrote:
> > > > > > > >
> > > > > > > > > wt., 12 lut 2019 o 11:18 Lee Jones <lee.jones@linaro.org>=
 napisa=C5=82(a):
> > > > > > > > > >
> > > > > > > > > > On Tue, 12 Feb 2019, Bartosz Golaszewski wrote:
> > > > > > > > > >
> > > > > > > > > > > wt., 12 lut 2019 o 10:55 Lee Jones <lee.jones@linaro.=
org> napisa=C5=82(a):
> > > > > > > > > > > >
> > > > > > > > > > > >  * The declaration of a superfluous struct
> > > > > > > > > > > >  * 100 lines of additional/avoidable code
> > > > > > > > > > > >  * Hacky hoop jumping trying to fudge VIRQs into re=
sources
> > > > > > > > > > > >  * Resources were designed for HWIRQs (unless a dom=
ain is present)
> > > > > > > > > > > >  * Loads of additional/avoidable CPU cycles setting=
 all this up
> > > > > > > > > > >
> > > > > > > > > > > While the above may be right, this one is negligible =
and you know it. :)
> > > > > > > > > >
> > > > > > > > > > You have nested for() loops.  You *are* wasting lots of=
 cycles.
> > > > > > > > > >
> > > > > > > > > > > > Need I go on? :)
> > > > > > > > > > > >
> > > > > > > > > > > > Surely the fact that you are using both sides of an=
 API
> > > > > > > > > > > > (devm_regmap_init_i2c and regmap_irq_get_*) in the =
same driver, must
> > > > > > > > > > > > set some alarm bells ringing?
> > > > > > > > > > > >
> > > > > > > > > > > > This whole HWIRQ setting, VIRQ getting, resource ha=
cking is a mess.
> > > > > > > > > > > >
> > > > > > > > > > > > And for what?  To avoid passing IRQ data to a child=
 driver?
> > > > > > > > > > >
> > > > > > > > > > > What do you propose? Should I go back to the approach=
 in v1 and pass
> > > > > > > > > > > the regmap_irq_chip_data to child drivers?
> > > > > > > > > >
> > > > > > > > > > I'm saying you should remove all of this hackery and pa=
ss IRQs as they
> > > > > > > > > > are supposed to be passed (like everyone else does).
> > > > > > > > >
> > > > > > > > > I'm not sure what you mean by "like everyone else does" -=
 different
> > > > > > > > > mfd drivers seem to be doing different things. Is a simpl=
e struct
> > > > > > > > > containing virtual irq numbers passed to sub-drivers fine=
?
> > > > > > > >
> > > > > > > > How do you plan on deriving the VIRQs to place into the str=
uct?
> > > > > > >
> > > > > > > Exampe:
> > > > > > >
> > > > > > > struct max77650_gpio_pdata {
> > > > > > >     int gpi_irq;
> > > > > > > };
> > > > > > >
> > > > > > > In MFD driver:
> > > > > > >
> > > > > > > struct max77650_gpio_pdata *gpio_data =3D devm_kmalloc(dev, s=
izeof(*gpio_data));
> > > > > > >
> > > > > > > gpio_data->gpi_irq =3D regmap_irq_get_virq(irqchip_data, GPI_=
NUM);
> > > > > > >
> > > > > > > gpio_cell.platform_data =3D gpio_data;
> > > > > > >
> > > > > > > In GPIO driver:
> > > > > > >
> > > > > > > struct max77650_gpio_pdata *gpio_data =3D pdev->dev.platform_=
data;
> > > > > > >
> > > > > > > int irq =3D gpio_data->gpi_irq;
> > > > > >
> > > > > > Definitely not.  What you're trying to do is a hack.
> > > > > >
> > > > > > If you're using Regmap to handle your IRQs, then you should use=
 Regmap
> > > > > > in the client to pull them out.  Setting them via Regmap, then =
pulling
> > > > > > them out again in the *same driver*, only to store them in plat=
form
> > > > > > data to be passed to a child device is bonkers.
> > > > > >
> > > > > > *Either* use the MFD provided platform-data helpers *or* pass a=
nd
> > > > > > handle them via the Regmap APIs, *not* both.
> > > > >
> > > > > Right, a plan has been formed.
> > > > >
> > > > > Hopefully this works and you can avoid all this dancing around.
> > > > >
> > > > > Firstly, you need to make a small change to:
> > > > >
> > > > >   drivers/base/regmap/regmap-irq.c
> > > > >
> > > > > Add the following function:
> > > > >
> > > > >   struct irq_domain *regmap_irq_get_domain(struct regmap *map)
> > > >
> > > > We already do have such function and a lot of mfd drivers actually =
use it.
> > >
> > > Even better.
> > >
> > > > > As you can see, it will return the IRQ Domain for the chip.
> > > > >
> > > > > You can then pass this IRQ domain to mfd_add_devices() and it wil=
l do
> > > > > the HWIRQ =3D> VIRQ mapping for you on the fly.  Meaning that you=
 can
> > > > > remove all the nastiness in max77650_setup_irqs() and have the In=
put
> > > > > device use the standard (e.g. platform_get_irq()) APIs.
> > > > >
> > > > > How does that Sound?
> > > >
> > > > This does sound better! Why didn't you lead with that in the first =
place?
> > >
> > > I'm not even going to dignify that stupid question with a response.
> >
> > It's not a stupid question and you're being unnecessarily rude. As an
> > expert in the subsystem you maintain you could have answered simply
> > with a "this is wrong, retrieve the irq domain from the regmap
> > irq_chip and pass it over to mfd_add_devices, the mfd core will create
> > appropriate mappings".
>
> Could be culture clash, but I found the question offensive which is
> why I chose not to answer it.  The reason is actually explained below:
>

It wasn't meant to be offensive. I guess when dealing with non-native
English speakers over e-mail it's best to assume good faith.

>  "It's only the craziness in this patch which forced me to look into how
>   Regmap handles IRQs and come up with a subsequent solution which fits
>   your use-case."
>
> Thus the fact that a) Regmap used IRQ domains and b) the IRQ domain
> could be fetched and reused here didn't enter my thought process until
> I delved into the inner workings of Regmap.
>
> Yes, I know MFD pretty well, but I only tend to deep-dive into other
> subsystems, particularly ones as complicated as Regmap, when it's
> necessary to do so.  Now I know a little more about it, I can provide
> the feedback you suggest going forward.
>
> > > > It's a pity it's not documented, I had to look at the code to find =
out
> > > > irq resources would get translated in mfd_add_devices() if a domain=
 is
> > > > present.
> > >
> > > Where is it likely to be documented?  MFD is pretty simple and seldom
> > > needs explanation.  A 3 second look at the API you're trying to use
> > > (instead of blind copy & paste) would have told you that it's possibl=
e
> > > to take an IRQ domain as an argument.
> > >
> > > It's only the craziness in this patch which forced me to look into ho=
w
> > > Regmap handles IRQs and come up with a subsequent solution which fits
> > > your use-case.
> > >
> > > > In that case - I really don't see a reason to create a superfluous
> > > > structure to only hold the main regmap - we can very well get it fr=
om
> > > > the parent device in sub-drivers as I do now.
> > >
> > > The whole point of this solution is that you don't need to pass
> > > anything non-standard (i.e. not provided by the current APIs) to the
> > > child device.
> >
> > I don't understand what you're saying here.
>
> I'm saying that the structure you speak of is no longer required.
>

Thanks for clarifying.

Bartosz