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;
Date:   Sat, 17 Mar 2018 10:37:22 -0700
From:   Dmitry Torokhov <dmitry.torokhov@gmail.com>
To:     jacopo mondi <jacopo@jmondi.org>
Cc:     Laurent Pinchart <laurent.pinchart@ideasonboard.com>,
        Lee Jones <lee.jones@linaro.org>,
        Daniel Thompson <daniel.thompson@linaro.org>,
        Jingoo Han <jingoohan1@gmail.com>, linux-fbdev@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-sh@vger.kernel.org,
        Yoshinori Sato <ysato@users.sourceforge.jp>,
        Rich Felker <dalias@libc.org>,
        Linus Walleij <linus.walleij@linaro.org>
Subject: Re: [RFC 2/4] sh: ecovec24: conditionally register backlight device
Message-ID: <20180317173722.GB183240@dtor-ws>
References: <20180315224202.96668-1-dmitry.torokhov@gmail.com>
 <20180315224202.96668-3-dmitry.torokhov@gmail.com>
 <20180316100748.GH16424@w540>
 <20180316233800.GA32310@dtor-ws>
 <20180317092516.g5p5ybka57abyhex@apu3b>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180317092516.g5p5ybka57abyhex@apu3b>
User-Agent: Mutt/1.9.2 (2017-12-15)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Mar 17, 2018 at 10:25:16AM +0100, jacopo mondi wrote:
> Hi Dmitry,
> 
> On Fri, Mar 16, 2018 at 04:38:00PM -0700, Dmitry Torokhov wrote:
> > Hi Jacopo,
> > 
> > On Fri, Mar 16, 2018 at 11:07:48AM +0100, jacopo mondi wrote:
> > > Hello Dmitry
> > > 
> > > FYI I am brushing the ecovec board these days as well
> > > https://www.spinics.net/lists/linux-sh/msg52536.html
> > > 
> > 
> > What is the ecovec board BTW? Is it some devkit or what? It seems quite
> > old to me.
> 
> Yes, it is a SuperH 4 based development board. It is old for sure. I'm
> also working on removing some stuff the ecovec board file is the only
> user of...
> 
> > > And I have a board to test with but without any display panel, I'm
> > > afraid.
> > > 
> > > On Thu, Mar 15, 2018 at 03:42:00PM -0700, Dmitry Torokhov wrote:
> > > > Commit fe79f919f47e ("sh: ecovec24: Use gpio-backlight") removed custom
> > > > backlight support and switched over to generic gpio-backlight driver. The
> > > > comment when we run with DVI states "no backlight", but setting
> > > > gpio_backlight_data.fbdev to NULL actually makes gpio-backlight to react to
> > > > events from any framebuffer device, not ignore them.
> > > >
> > > > We want to get rid of platform data in favor of generic device properties
> > > > in gpio_backlight driver, so we can not have kernel pointers passed around
> > > > to tie the framebuffer device to backlight. Assuming that the intent of the
> > > > above referenced commit was to indeed not export backlight when using DVI,
> > > > let's switch to conditionally registering backlight device so it is not
> > > > present at all in DVI case.
> > > >
> > > > Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
> > > > ---
> > > >  arch/sh/boards/mach-ecovec24/setup.c | 24 +++++++++++++++++-------
> > > >  1 file changed, 17 insertions(+), 7 deletions(-)
> > > >
> > > > diff --git a/arch/sh/boards/mach-ecovec24/setup.c b/arch/sh/boards/mach-ecovec24/setup.c
> > > > index 6f929abe0b50f..67633d2d42390 100644
> > > > --- a/arch/sh/boards/mach-ecovec24/setup.c
> > > > +++ b/arch/sh/boards/mach-ecovec24/setup.c
> > > > @@ -368,7 +368,6 @@ static struct platform_device lcdc_device = {
> > > >  };
> > > >
> > > >  static struct gpio_backlight_platform_data gpio_backlight_data = {
> > > > -	.fbdev = &lcdc_device.dev,
> > > >  	.gpio = GPIO_PTR1,
> > > >  	.def_value = 1,
> > > >  	.name = "backlight",
> > > > @@ -987,7 +986,6 @@ static struct platform_device *ecovec_devices[] __initdata = {
> > > >  	&usb1_common_device,
> > > >  	&usbhs_device,
> > > >  	&lcdc_device,
> > > > -	&gpio_backlight_device,
> > > >  	&ceu0_device,
> > > >  	&ceu1_device,
> > > >  	&keysc_device,
> > > > @@ -1077,6 +1075,8 @@ static int __init arch_setup(void)
> > > >  {
> > > >  	struct clk *clk;
> > > >  	bool cn12_enabled = false;
> > > > +	bool use_backlight = false;
> > > > +	int error;
> > > >
> > > >  	/* register board specific self-refresh code */
> > > >  	sh_mobile_register_self_refresh(SUSP_SH_STANDBY | SUSP_SH_SF |
> > > > @@ -1193,9 +1193,6 @@ static int __init arch_setup(void)
> > > >  		lcdc_info.ch[0].lcd_modes		= ecovec_dvi_modes;
> > > >  		lcdc_info.ch[0].num_modes		= ARRAY_SIZE(ecovec_dvi_modes);
> > > >
> > > > -		/* No backlight */
> > > > -		gpio_backlight_data.fbdev = NULL;
> > > > -
> > > >  		gpio_set_value(GPIO_PTA2, 1);
> > > >  		gpio_set_value(GPIO_PTU1, 1);
> > > >  	} else {
> > > > @@ -1217,6 +1214,8 @@ static int __init arch_setup(void)
> > > >  		/* enable TouchScreen */
> > > >  		i2c_register_board_info(0, &ts_i2c_clients, 1);
> > > >  		irq_set_irq_type(IRQ0, IRQ_TYPE_LEVEL_LOW);
> > > > +
> > > > +		use_backlight = true;
> > > >  	}
> > > >
> > > >  	/* enable CEU0 */
> > > > @@ -1431,8 +1430,19 @@ static int __init arch_setup(void)
> > > >  	gpio_set_value(GPIO_PTG4, 1);
> > > >  #endif
> > > >
> > > > -	return platform_add_devices(ecovec_devices,
> > > > -				    ARRAY_SIZE(ecovec_devices));
> > > > +	error = platform_add_devices(ecovec_devices,
> > > > +				      ARRAY_SIZE(ecovec_devices));
> > > 
> > > I would invert this.
> > > Register the backlight first, then all other devices.
> > 
> > We could do that, but why would that be better?
> > 
> 
> That if backlight device registration fails we do not register all
> other devices. But yes that may be a bit too harsh, isn't it?

Yes, that was my reasoning. With platform_add_devices() that unwinds all
the devices it created successfully we do not really have any option but
return error (which is either ignored or fatal). Here we can simply warn
and still let the device boot so users have easier way to see what went
wrong.

> 
> > > 
> > > 
> > > > +	if (error)
> > > > +		return error;
> > > > +
> > > > +	if (use_backlight) {
> > > > +		error = platform_device_add(&gpio_backlight_device);
> > > > +		if (error)
> > > > +			pr_warn("%s: failed to register backlight: %d\n",
> > > > +				error);
> > > 
> > > Could you use dev_warn here? Also the format is wrong, I assume you
> > 
> > I would rather not, as the backlight device would be in unknown state
> > here, and using dev_warn with device that has not been fully registered
> > does not give any benefits. There is also no ambiguity as there is only
> > one backlight.
> 
> You are very correct, sorry for the fuss.
> 
> > 
> > > are missing a '__func__' as second function argument.
> > 
> > I'll fix this.
> > 
> > > 
> > > Also, you may want to return error.
> > 
> > How would caller handle this error? Should we kill all successfully
> > registered devices on error adding backlight?
> 
> As the function returned an error code for 'platform_add_devices()' I
> thought we may want to return one as well. That's why I proposed to
> invert the registration order :)
> 
> All minor nits btw,  sorry for jumping up, I understand this is an
> RFC and ecovec board file is not the real juice of this series ;)
> 
> Ping me if I can help with testing as I've the board.

Yes, I'll fix up the other mess ups and I'll post a v2.

Thanks!

-- 
Dmitry