Re: [PATCH v3] ARM: shmobile: lager: Add DT reference

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Magnus,

On Friday 05 July 2013 01:25:18 Magnus Damm wrote:
> On Fri, Jul 5, 2013 at 12:02 AM, Laurent Pinchart wrote:
> > On Wednesday 03 July 2013 16:57:31 Simon Horman wrote:
> >> This is sufficient to allow boot of the Lager board with a console
> >> without boards-lager.c compiled into the kernel. This is an example of a
> >> minimal but still useful initialisation of the board using DT as much as
> >> possible.
> >> 
> >> As such it is the same as the boot of Lager that can be achieved without
> >> a board file. The intention of adding this file is to facilitate further
> >> work to allow board specific devices to be initialised via DT.
> > 
> > At some (hopefully not too distant) point we should stop adding both
> > traditional and DT board files. DT is supposed to lower the number of
> > board files, not multiply it by two.
> 
> Well, mr-multiply-by-two: While I dislike code duplication I can't recall
> when the number of files actually decreased with DT.
>
> I realize that when we can use DT fully to describe a board then a single
> DTS file is enough to describe board specific hardware together with per-SoC
> C code. I can't however see how that would reduce the number of files, but
> whatever. I always thought the goal was to reduce the memory footprint of
> the kernel and that the source would not come with that many limitations,
> but it seems that I got that all wrong...

I meant board files, sorry. The total number of files hasn't decreased much, 
if at all.

> Anyway, the current split on mach-shmobile is intentional:
> 
> 1) The C version of board support (board-xxx.c)
> 
> This is where platform devices are used (with power domains in some cases)
> and where out-of-tree people can experiment with things. This code is using
> legacy SH clock code and is not targeted for multiplatform. These should go
> away, the sooner the better.
> 
> 2) The DT reference support (board-xxx-reference.c)
> 
> This file contains whatever bits of C code needed to get the DT support
> going. All the board specific devices should be described via DT. The DT
> reference board support is where we build incremental support to catch up
> with 1) board-xxx.c. Ideally we wouldn't need the board-xxx-reference.c file
> and instead only go with .dts, but in some cases we need some extra glue and
> until pinctrl DT is done we also need those written in C. The DT reference
> support will use common clocks as soon as they are available.
> 
> 3) SoC DT support (setup-xxx.c)
> 
> This is what will be used when 2) is caught up with 1). Common clocks and
> multiplatform, one SoC at a time.

Thanks for the clarification.

> > Work is still needed in order to achieve this, as we lack DT bindings for
> > core (and non core) devices. That's something we need to work on. As a
> > first step, do we have a roadmap with a list of devices that lack DT
> > support ? At the very least I see more work needed on
> > 
> > - Common clock framework
> 
> Yes, I agree. Please see below for more about that.
> 
> > - SCI
> 
> Indeed, someone needs to continue earlier efforts for the SCIF driver.

It shouldn't be too difficult, Bastian has already layed out a plan, but the 
dependency on DT clock bindings has prevented the patches from going upstream.

> Fortunately, with EMEV2 our UART driver already supports DT, so we can
> implement DT-only multiplatform support on EMEV2 regardless of the state of
> SCIF.
> 
> > - Timers
> 
> We can use DT-only with Arch timer, but unfortunately another clock evenit
> is needed unconditionally for high resolution timer support, search for
> clock event and the C3STOP flag if you want to see me moan about that.
> 
> Regardless I believe someone took a couple of first steps towards DT support
> for our timers some time ago. But as we discussed face to face about a month
> ago, it makes sense to describe the hardware via DT but in case of timers
> and clock event / clock sources it does not make sense to associate 87 timer
> channels for operating system time keeping. So we need some user space
> configuration interface to associate different kind of timer applications
> (operating system timer, PWM, clock generation, etc) to different channels.

I agree that we need an API for this. It might make sense to restrict the API 
to in-kernel usage to start with, but we need an API nonetheless. Has this 
topic been discussed with clocks/timers people ? It might be worth bringing it 
up for LPC or KS.

> Since you mention it: =)
> - PINCTRL DT and GPIO DT
>
> I'd like to start picking up the remaining PINCTRL and GPIO bits for
> v3.12 merge. Can you please rebase and resend on top of the latest
> renesas-next tag?

Patches for v3.12 have been sent yesterday :-)

> > We shouldn't delay CCF much longer, v3.12 seems an achievable target to
> > me. Who will work on that ?
> 
> So I am currently preparing EMEV2 for multiplatform, will most likely have
> patches posted for that next week. And yes, I intend to send a new DT
> reference board support file.
> 
> In parallel Kuribayashi-san is taking a stab at moving to common clocks for
> EMEV2 as a first step.

That's great news. I'm looking forward to reviewing the patches :-)

> I expect the above to be in order for v3.12, but not much more that that.

That sounds like a reasonable time frame to me.

-- 
Cheers,

Laurent Pinchart