Message-ID: <20151109115727.GE963@e106497-lin.cambridge.arm.com>
Date:	Mon, 9 Nov 2015 11:57:27 +0000
From:	Liviu Dudau <Liviu.Dudau@....com>
To:	Russell King - ARM Linux <linux@....linux.org.uk>
Cc:	Mark yao <mark.yao@...k-chips.com>,
	David Airlie <airlied@...ux.ie>,
	Daniel Vetter <daniel.vetter@...ll.ch>,
	Philipp Zabel <p.zabel@...gutronix.de>,
	Heiko Stuebner <heiko@...ech.de>,
	dri-devel <dri-devel@...ts.freedesktop.org>,
	linux-rockchip <linux-rockchip@...ts.infradead.org>,
	LAKML <linux-arm-kernel@...ts.infradead.org>,
	LKML <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH v3 1/4] drm: Introduce generic probe function for
 component based masters.

On Mon, Nov 09, 2015 at 11:43:00AM +0000, Russell King - ARM Linux wrote:
> On Mon, Nov 09, 2015 at 05:39:25PM +0800, Mark yao wrote:
> > Hi Liviu
> >       Rockchip drm can't work with drm_of_component_probe function now,
> > 
> >       At drm_of_component_probe:
> >             component_match_add(dev, &match, compare_of, port);
> >       And original rockchip drm use:
> >             component_match_add(dev, &match, compare_of, port->parent);
> > 
> >      That different "port" and "port->parent" cause crtc device node always
> > mis-match.
> > 
> >      I'm confused that rockchip use same dts node map as imx drm driver, but
> > it works
> > for imx drm, not work on rockchip drm.
> 
> iMX is rather confusing because the whole device is rather complex.  It's
> a complete image processor unit, which has multiple functions within a
> single device.  It's a less than perfect example of how to deal with
> these issues (each time I look at it, I find more stuff it shouldn't be
> doing... like what I've found today.)
> 
> Basically, the device is declared in DT as:
> 
>                 ipu1: ipu@...00000 {
>                         compatible = "fsl,imx6q-ipu";
>                         ipu1_csi0: port@0 {
> ...
>                         };
>                         ipu1_csi1: port@1 {
> ...
>                         };
>                         ipu1_di0: port@2 {
> ...
>                         };
>                         ipu1_di1: port@3 {
> ...
>                         };
>                 };
> 
> ipu1 is the platform device, which is the _entire_ IPU device, containing
> multiple sub-devices - eg, two camera interfaces (csi) and two "CRTCs"
> (di).
> 
> The ipuv3 code creates platform devices for these, calling the CSI
> devices "imx-ipuv3-camera" and the DI devices "imx-ipuv3-crtc".
> Initially, these have no of_node attached (yuck!) but later have an
> of_node attached when the device is probed (yuck yuck yuck!) by
> ipu_drm_probe() - the of_node being the ipu1_di0 or ipu1_di1 node.
> 
> The display-subsystem property references these ipu1_di0/ipu1_di1 nodes:
> 
>         display-subsystem {
>                 compatible = "fsl,imx-display-subsystem";
>                 ports = <&ipu1_di0>, <&ipu1_di1>, <&ipu2_di0>, <&ipu2_di1>;
>         };
> 
> and so finds the "imx-ipuv3-crtc" platform devices rather than the
> parent ipu1 device.
> 
> It's not nice - I'd like to see this:
> 
>         if (!dev->of_node) {
>                 /* Associate crtc device with the corresponding DI port node */
>                 dev->of_node = ipu_drm_get_port_by_id(dev->parent->of_node,
>                                                       pdata->di + 2);
>                 if (!dev->of_node) {
>                         dev_err(dev, "missing port@%d node in %s\n",
>                                 pdata->di + 2, dev->parent->of_node->full_name);                        return -ENODEV;
>                 }
>         }
> 
> moved out of drivers/gpu/drm/imx/ipuv3-crtc.c and into
> drivers/gpu/ipu-v3/ipu-common.c where the platform devices are created
> so that we're only setting device of_node pointers at device creation
> time and not randomly during the probe sequence, even though the above
> is "safe" as far as the component helper's concerned.  There's no reason
> why it can't be done at the device creation time.

Hi Russell,

Thanks for your analysis, I keep delaying looking into the imx code more
seriously even if I have a SabreLite board that I could play with (not on my
day-to-day list of tasks).

> 
> Now, as to how to handle the differences here, I think a solution would
> be to pass in two compare_of function pointers: one for CRTCs and one
> for the encoders, since the two will need to be handled separately
> depending on the implementation.  Where you have one "parent" device of
> the CRTC node containing exactly one CRTC, then the "rockchip" method
> makes sense - though comparing the parent of the port node in
> CRTC compare_of would be where I'd put it.  If you have multiple
> CRTCs within one parent device, then something more complex like the
> iMX solution would be required.
> 
> In any case, passing port->parent is a data loss in the generic case:
> you lose the information in the compare_of function about exactly which
> port is required, so that must go into the CRTC compare_of.
> 
> So...
> 
> rockchip's CRTC compare_of() should be:
> 
> static int crtc_compare_of(struct device *dev, void *data)
> {
>         struct device_node *np = data;
> 
>         return dev->of_node == np->parent;
> }
> 
> and it should have an encoder_compare_of() which is its existing
> compare_of() renamed as such.
> 
> Then, we need drm_of_component_probe() to take _two_ comparison
> functions, one for the CRTCs and one for the encoders.

This hits very close to my experience. After sending this patchset I started
converting my HDLCD driver to use it and hit exactly the same issue, that the
same compare function is used for selecting CRTCs and encoders. My hardware
is even simpler, with one CRTC connected to only one encoder. My board has
two of each but the only link between CRTCs is the fact that they are fed most
of the time from the same clock source. In the end I've backed out of using
the drm_of_component_probe() code as it meant I've had to modify the device
tree in a way that did no longer describe the hardware accurately. I will
have another look this week if I can extend the function and make it work
for all cases.

Meanwhile, what is your suggestion regarding the patchset. I've seen David has
sent Linus a pull request for 4.4-rc1 that includes it. Should we send a
revert for rockchip commit and then patch later the function?

Best regards,
Liviu


> 
> -- 
> FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
> according to speedtest.net.
> 

-- 
====================
| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists