[U-Boot] [PATCH 4/8] ARM: socfpga: arria10: Add generic handoff devicetree include

[Simon Goldschmidt]

There's something wrong with your mailer: indentation of replies doesn't seem
to work. It gets kind of hard to read who wrote what...

On Mon, Oct 7, 2019 at 4:34 PM Dalon L Westergreen
<dalon.westergreen at linux.intel.com> wrote:
>
> On Sun, 2019-10-06 at 20:05 +0200, Simon Goldschmidt wrote:
>
> Am 06.10.2019 um 19:44 schrieb Dalon L Westergreen:
>
> On Sun, 2019-10-06 at 15:44 +0200, Marek Vasut wrote:
>
> On 10/6/19 1:19 AM, Dalon L Westergreen wrote:
>
> On Sat, 2019-10-05 at 01:51 +0200, Marek Vasut wrote:
>
> On 10/5/19 12:30 AM, Dalon Westergreen wrote:
>
> From: Dalon Westergreen <
>
> dalon.westergreen at intel.com
>
>
>  <mailto:
>
> dalon.westergreen at intel.com
>
> >
>
>
> Generic handoff devicetree include uses a header generated bythe qts-filter-
>
> a10.sh script in mach-socfpga.  The scriptcreates the header based on design
>
> specific implementationsfor clock and pinmux configurations.
>
>
> [...]
>
> diff --git a/arch/arm/dts/socfpga_arria10_handoff_u-boot.dtsi
>
> b/arch/arm/dts/socfpga_arria10_handoff_u-boot.dtsi
>
>
> [...]
>
> - clock_manager at 0xffd04000 {+ clkmgr at 0xffd04000 {+ compatible =
>
> "altr,socfpga-a10-clk-init";+ reg = <0xffd04000 0x00000200>;+
>
> reg-names = "soc_clock_manager_OCP_SLV"; u-boot,dm-pre-
>
> reloc;   mainpll {+ vco0-psrc =
>
> <MAINPLLGRP_VCO0_PSRC>;+ vco1-denom =
>
> <MAINPLLGRP_VCO1_DENOM>;+ vco1-numer =
>
> <MAINPLLGRP_VCO1_NUMER>;
>
>
> But these bits are board-specific , they shouldn't be in common DT.
>
>
> This common dtsi requires that the top level u-boot.dtsi include the board
>
> specific header.  The format
>
> and #define names are in fact common.
>
>
> OK, I now see what you're doing here. Can you explain that in a bit more
>
> detail in the commit message ?
>
>
> Basically socfpga_board.h is included socfpga_board.dts , and then the
>
> preprocessor correctly expands the values from socfpga_board.h in the
>
> socfpga_board.dts , so this works for multiple boards too ?
>
>
> Exactly. Will add more detail in the commit message, and slim down the
>
> included clocks in
>
> the u-boot.dtsi
>
>
> I'm (still) working on a series to bring gen5 completely to devicetree,
>
> so that the 'qts' directories can be removed. I chose a different
>
> approach however, in that I generated everything into a '-handoff.dtsi'
>
> file (*). I'd be happy if we could find a common mechanism for all
>
> socfpga sub-archs. How does S10/Agilex handle this?
>
>
> (*): Gen5 has the downside that we're low on memory in SPL (regarding
>
> DM), and as we require large binary arrays there, I chose to encode the
>
> binary blob arrays in host byte order so that they could be used
>
> in-place instead of copying them from BE (dtb) to LE (stack/heap). Maybe
>
> that doesn't fit A10/S10/Agilex anyway?
>
>
> Can you share your patch set with me, privately or otherwise, just so we can take a similar
> approach in the devicetree?

Give me a week max. and I'll send the series as RFC.

>
> Also, have you looked at the bsp-editor python source that generates the "generated" files
> from the bsp-editor?
>
> 19.1std/647/linux64/ip/altera/preloader/scripts

No, I mainly focused on the U-Boot part for now. I just wrote a C program
that did the conversion from 'official generated' files to DTS...

>
> I am hoping to replace these with a script included in the u-boot source. iocsr.py does
> a bunch of binary parsing.

Right, a direct conversion from compilation output to U-Boot would be better.
However, that would require Intel to keep that output consistent!

Regards,
Simon

>
> --dalon
>
>
> Regards,
>
> Simon