Date:   Wed, 19 Jan 2022 19:12:23 +0300
From:   Sergey Shtylyov <s.shtylyov@....ru>
To:     Uwe Kleine-König <u.kleine-koenig@...gutronix.de>,
        "Geert Uytterhoeven" <geert@...ux-m68k.org>
CC:     Andrew Lunn <andrew@...n.ch>, Ulf Hansson <ulf.hansson@...aro.org>,
        Vignesh Raghavendra <vigneshr@...com>,
        KVM list <kvm@...r.kernel.org>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        <linux-iio@...r.kernel.org>,
        "Linus Walleij" <linus.walleij@...aro.org>,
        Amit Kucheria <amitk@...nel.org>,
        "ALSA Development Mailing List" <alsa-devel@...a-project.org>,
        Jaroslav Kysela <perex@...ex.cz>,
        Guenter Roeck <groeck@...omium.org>,
        Thierry Reding <thierry.reding@...il.com>,
        MTD Maling List <linux-mtd@...ts.infradead.org>,
        Linux I2C <linux-i2c@...r.kernel.org>,
        Miquel Raynal <miquel.raynal@...tlin.com>,
        <linux-phy@...ts.infradead.org>,
        linux-spi <linux-spi@...r.kernel.org>,
        Jiri Slaby <jirislaby@...nel.org>,
        "David S. Miller" <davem@...emloft.net>,
        Khuong Dinh <khuong@...amperecomputing.com>,
        Florian Fainelli <f.fainelli@...il.com>,
        Matthias Schiffer <matthias.schiffer@...tq-group.com>,
        Kamal Dasu <kdasu.kdev@...il.com>,
        "Greg Kroah-Hartman" <gregkh@...uxfoundation.org>,
        Lee Jones <lee.jones@...aro.org>,
        Bartosz Golaszewski <brgl@...ev.pl>,
        Daniel Lezcano <daniel.lezcano@...aro.org>,
        Kishon Vijay Abraham I <kishon@...com>,
        bcm-kernel-feedback-list <bcm-kernel-feedback-list@...adcom.com>,
        "open list:SERIAL DRIVERS" <linux-serial@...r.kernel.org>,
        Jakub Kicinski <kuba@...nel.org>,
        Zhang Rui <rui.zhang@...el.com>,
        <platform-driver-x86@...r.kernel.org>,
        Linux PWM List <linux-pwm@...r.kernel.org>,
        Robert Richter <rric@...nel.org>,
        "Saravanan Sekar" <sravanhome@...il.com>,
        Corey Minyard <minyard@....org>,
        Linux PM list <linux-pm@...r.kernel.org>,
        Liam Girdwood <lgirdwood@...il.com>,
        "Mauro Carvalho Chehab" <mchehab@...nel.org>,
        John Garry <john.garry@...wei.com>,
        Peter Korsgaard <peter@...sgaard.com>,
        William Breathitt Gray <vilhelm.gray@...il.com>,
        Mark Gross <markgross@...nel.org>,
        Hans de Goede <hdegoede@...hat.com>,
        Alex Williamson <alex.williamson@...hat.com>,
        "Mark Brown" <broonie@...nel.org>, Borislav Petkov <bp@...en8.de>,
        Takashi Iwai <tiwai@...e.com>,
        Matthias Brugger <matthias.bgg@...il.com>,
        <openipmi-developer@...ts.sourceforge.net>,
        Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
        Benson Leung <bleung@...omium.org>,
        Pengutronix Kernel Team <kernel@...gutronix.de>,
        Linux ARM <linux-arm-kernel@...ts.infradead.org>,
        <linux-edac@...r.kernel.org>, "Tony Luck" <tony.luck@...el.com>,
        Richard Weinberger <richard@....at>,
        Mun Yew Tham <mun.yew.tham@...el.com>,
        Eric Auger <eric.auger@...hat.com>,
        netdev <netdev@...r.kernel.org>,
        "open list:GPIO SUBSYSTEM" <linux-gpio@...r.kernel.org>,
        Cornelia Huck <cohuck@...hat.com>,
        "Linux MMC List" <linux-mmc@...r.kernel.org>,
        Joakim Zhang <qiangqing.zhang@....com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Linux-Renesas <linux-renesas-soc@...r.kernel.org>,
        Vinod Koul <vkoul@...nel.org>,
        "James Morse" <james.morse@....com>,
        Zha Qipeng <qipeng.zha@...el.com>,
        "Sebastian Reichel" <sre@...nel.org>,
        Niklas Söderlund <niklas.soderlund@...natech.se>,
        <linux-mediatek@...ts.infradead.org>,
        "Brian Norris" <computersforpeace@...il.com>,
        Yoshihiro Shimoda <yoshihiro.shimoda.uh@...esas.com>
Subject: Re: [PATCH 1/2] platform: make platform_get_irq_optional() optional

On 1/18/22 5:29 PM, Uwe Kleine-König wrote:

>> nst the magic not-found value (so no implementation detail magic
>>>>> leaks into the caller code) and just pass it to the next API function=
>> .
>>>>> (And my expectation would be that if you chose to represent not-found=
>>  by
>>>>> (void *)66 instead of NULL, you won't have to adapt any user, just th=
>> e
>>>>> framework internal checks. This is a good thing!)
>>>>
>>>> Ah, there is the wrong assumption: drivers sometimes do need to know
>>>> if the resource was found, and thus do need to know about (void *)66,
>>>> -ENODEV, or -ENXIO.  I already gave examples for IRQ and clk before.
>>>> I can imagine these exist for gpiod and regulator, too, as soon as
>>>> you go beyond the trivial "enable" and "disable" use-cases.
>>>
>>> My premise is that every user who has to check for "not found"
>>> explicitly should not use (clk|gpiod)_get_optional() but
>>> (clk|gpiod)_get() and do proper (and explicit) error handling for
>>> -ENODEV. (clk|gpiod)_get_optional() is only for these trivial use-cases.
>>>
>>>> And 0/NULL vs. > 0 is the natural check here: missing, but not
>>>> an error.
>>>
>>> For me it it 100% irrelevant if "not found" is an error for the query
>>> function or not. I just have to be able to check for "not found" and
>>> react accordingly.
>>>
>>> And adding a function
>>>
>>>         def platform_get_irq_opional():
>>>                 ret =3D platform_get_irq()
>>>                 if ret =3D=3D -ENXIO:
>>>                         return 0
>>>                 return ret
>>>
>>> it's not a useful addition to the API if I cannot use 0 as a dummy
>>> because it doesn't simplify the caller enough to justify the additional
>>> function.
>>>
>>> The only thing I need to be able is to distinguish the cases "there is
>>> an irq", "there is no irq" and anything else is "there is a problem I
>>> cannot handle and so forward it to my caller". The semantic of
>>> platform_get_irq() is able to satisfy this requirement[1], so why introdu=
>> ce
>>> platform_get_irq_opional() for the small advantage that I can check for
>>> not-found using
>>>
>>>         if (!irq)
>>>
>>> instead of
>>>
>>>         if (irq !=3D -ENXIO)
>>>
>>> ? The semantic of platform_get_irq() is easier ("Either a usable
>>> non-negative irq number or a negative error number") compared to
>>> platform_get_irq_optional() ("Either a usable positive irq number or a
>>> negative error number or 0 meaning not found"). Usage of
>>> platform_get_irq() isn't harder or more expensive (neither for a human
>>> reader nor for a maching running the resulting compiled code).
>>> For a human reader
>>>
>>>         if (irq !=3D -ENXIO)
>>>
>>> is even easier to understand because for
>>>
>>>         if (!irq)
>>>
>>> they have to check where the value comes from, see it's
>>> platform_get_irq_optional() and understand that 0 means not-found.
>>
>> "vIRQ zero does not exist."
> 
> With that statement in mind I would expect that a function that gives me
> an (v)irq number never returns 0.
> 
>>> This function just adds overhead because as a irq framework user I have
>>> to understand another function. For me the added benefit is too small to
>>> justify the additional function. And you break out-of-tree drivers.
>>> These are all no major counter arguments, but as the advantage isn't
>>> major either, they still matter.
>>>
>>> Best regards
>>> Uwe
>>>
>>> [1] the only annoying thing is the error message.
>>
>> So there's still a need for two functions.
> 
> Or a single function not emitting an error message together with the
> callers being responsible for calling dev_err().
> 
> So the options in my preference order (first is best) are:
> 
>  - Remove the printk from platform_get_irq() and remove
>    platform_get_irq_optional();

   Strong NAK here:
- dev_err() in our function saves a lot of (repeatable!) comments;
- we've already discussed that it's more optimal to check againt 0 than
  against -ENXIO in the callers.

>  - Rename platform_get_irq_optional() to platform_get_irq_silently()

   NAK as well. We'd better off complaining about irq < 0 in this function.

>  - Keep platform_get_irq_optional() as is

   NAK, it's suboptimal in the call sites.

>  - Collect underpants
> 
>  - ?

   You're on your own here. :-)

>  - Change semantic of platform_get_irq_optional()

   Yes, we should change the semantics if it serves our goals better. 

> Best regards
> Uwe

MBR, Sergey

Powered by blists - more mailing lists