Date:	Tue, 23 Jun 2015 01:45:23 +0300
From:	Laurent Pinchart <laurent.pinchart@...asonboard.com>
To:	Magnus Damm <magnus.damm@...il.com>
Cc:	linux-kernel@...r.kernel.org, devicetree@...r.kernel.org,
	laurent.pinchart+renesas@...asonboard.com, geert+renesas@...der.be,
	linux-sh@...r.kernel.org, daniel.lezcano@...aro.org,
	horms+renesas@...ge.net.au, tglx@...utronix.de
Subject: Re: [PATCH 00/08] clocksource: sh_cmt: DT binding rework

Hi Magnus,

Thank you for the patches.

On Sunday 21 June 2015 17:09:03 Magnus Damm wrote:
> clocksource: sh_cmt: DT binding rework
> 
> [PATCH 01/08] devicetree: bindings: Remove sh7372 CMT binding
> [PATCH 02/08] clocksource: sh_cmt: Use 0x3f mask for SH_CMT_48BIT case
> [PATCH 03/08] devicetree: bindings: R-Car Gen2 CMT0 and CMT1 bindings
> [PATCH 04/08] clocksource: sh_cmt: Support separate R-Car Gen2 CMT0/1
> [PATCH 05/08] devicetree: bindings: r8a73a4 and R-Car Gen2 CMT bindings
> [PATCH 06/08] ARM: shmobile: Update CMT compat string users in DTS
> [PATCH 07/08] devicetree: bindings: Deprecate property, update example
> [PATCH 08/08] ARM: shmobile: Remove CMT renesas,channels-mask from DTS
> 
> This series reworks the CMT DT bindings to try to deal with the following:
> - R-Car Gen2 CMT0 and CMT1 hardware instances are not identical
> - The property renesas,channels-mask is not enough to describe the
>   difference
> - DT should describe the hardware, not the software implementation
> - Not all documented DT bindings are actually used
> 
> Without these patches the binding "renesas,cmt-48-gen2" is used for both
> CMT0 and CMT1 on R-Car Gen2 SoCs. CMT0 and CMT1 are currently seen by the
> Linux device driver as compatible hardware, and the device-specific property
> "renesas,channels-mask" is used to point out some of the hardware
> configuration differences. Since the driver is not feature complete only
> some differences are described and when diving into the data sheet we can
> see that:
> 
> 1) CMT0 is not 48-bit at all, instead it only supports 32-bit counters.

How the heck did we come up with these bindings in the first place ?

> 2) Some channels of CMT1 are 48-bit, some 32-bit.
> 3) A couple of CMT1 channels have even more features.
> 
> It turns out that none of the above differences are described in our current
> DT files. And since we use the same compat string for CMT0 and CMT1 the
> driver itself cannot enable features specific only to CMT1 without first
> updating the DTS. So this series is ground work for future feature patches.
> 
> It seems that we have two choices if we want to support CMT1 features:
> A) Keep existing DT bindings, add more properties for CMT1
> B) Rework the compatible strings and keep configuration in the driver
> 
> Judging by above it seems that DT update is inevitable. In my mind it is
> rather painful to update the DT so I'd like to minimize the number of
> updates and let the majority of the changes only happen in the driver.
> And since we should really describe hardware in DT but driver features tend
> to be implemented incrementally then B) seems like a good fit to me.
>
> I wouldn't mind going with A) but to be honest I must say that the existing
> compat string "renesas,cmt-48-gen2" is just too confusing with "48"
> (when CMT0 is 32bit-only) and also "gen2" (used without product line R-Car).
> 
> Because of that I've gone with B) and reworked the bindings to separate
> CMT0 from CMT1 and keep channel configuration in the device driver.

In this specific case B seems to be better given that the CMT features look 
quite random. We would need to encode too many properties in DT.

> While at it I've deprecated "renesas,channels-mask" and the old DT compat
> strings. Per-SoC compat strings have also been reworked, but are not left
> around documented as deprecated since they were never used by the driver.

Shouldn't we remove the renesas,channels-mask property completely from the 
documentation ? The git history could be considered as enough when it comes to 
documenting deprecated properties.

> Comments are very welcome! If all are in favor then it would be good to
> merge patch 1-5 first and patch 6-8 later once 1-5 are in. Old DTBs are
> left working but treated as low-feature CMT0. New DTBs including patch 6-8
> will not work on old kernels.

Apart from Geert's comments on patches 1/8 and 2/8, and my comment on patch 
8/8, the series looks fine to me.

With those problems fixed,

Acked-by: Laurent Pinchart <laurent.pinchart@...asonboard.com>

> Signed-off-by: Magnus Damm <damm+renesas@...nsource.se>
> ---
> 
> Written against renesas-devel-20150619-v4.1-rc8
> 
>  Documentation/devicetree/bindings/timer/renesas,cmt.txt |   55 +++++++-----
>  arch/arm/boot/dts/r8a73a4.dtsi                          |    5 -
>  arch/arm/boot/dts/r8a7740.dtsi                          |    3
>  arch/arm/boot/dts/r8a7790.dtsi                          |   10 --
>  arch/arm/boot/dts/r8a7791.dtsi                          |   10 --
>  arch/arm/boot/dts/r8a7793.dtsi                          |   10 --
>  arch/arm/boot/dts/r8a7794.dtsi                          |   10 --
>  arch/arm/boot/dts/sh73a0.dtsi                           |    3
>  drivers/clocksource/sh_cmt.c                            |   54 ++++++++----
>  9 files changed, 83 insertions(+), 77 deletions(-)

-- 
Regards,

Laurent Pinchart

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists