[PATCH 3/4] arm: dts: k3-*-binman.dtsi: Clean up and templatize boot binaries

[Neha Malcom Francis]

Hi Michael

Sorry for the late response.

On 28/03/24 17:26, Michael Walle wrote:
> Hi,
> 
> On Thu Mar 28, 2024 at 12:18 PM CET, Neha Malcom Francis wrote:
>> On 27-Mar-24 8:03 PM, Michael Walle wrote:
>>> On Wed Mar 27, 2024 at 8:01 AM CET, Neha Malcom Francis wrote:
>>>> On 26/03/24 19:18, Michael Walle wrote:
>>>>> On Fri Mar 22, 2024 at 2:10 PM CET, Neha Malcom Francis wrote:
>>>>>> Clean up templatized boot binaries for all K3 boards. This includes
>>>>>> modifying the k3-binman.dtsi to use SPL_BOARD_DTB, BOARD_DESCRIPTION and
>>>>>> UBOOT_BOARD_DESCRIPTION from the files that include it to further reuse
>>>>>> code.
>>>>>>
>>>>>> All k3-<soc>-binman.dtsi will contain only templates. Only required boot
>>>>>> binaries can be built from the templates in the boards' respective
>>>>>> -u-boot.dtsi file (or k3-<board>-binman.dtsi if it exists). This allows
>>>>>> clear distinction between the SoC common stuff vs. what is additionally
>>>>>> needed to boot up a specific board.
>>>>>
>>>>> I appreciate the cleanup. But as far as I can see, a board might
>>>>> only have one device tree. How would that work if the uboot proper
>>>>> must support multiple device trees?
>>>>>
>>>>
>>>>    From the discussions that took place in the mailing list [1] the consensus
>>>> seems to be to not focus on multiple devicetree support as it leads to confusion
>>>> for downstream users.
>>>
>>> What are users in this regard? I don't think you'd confuse
>>> developers.
>>>
>>> Anyway, I'm planning on upstreaming a TI board which will have
>>> different memory configurations and different variants of the board.
>>
>> I am assuming you are reusing an existing TI SoC?
> 
> Not really yet. It's the j722s.
> 
> 
>>> And on top of that, it will just be a base board and there will
>>> likely be some carrier device trees (overlay? I'm not sure yet).
>>>
>>> As far as I can tell, you've put the memory configuration into the
>>> device tree, so I'll probably need to switch between them somehow.
>>
>> The "k3-<soc>-ddr.dtsi" file will be present in your k3-<board>r5.dts
>> which makes sense, the memory configuration depends on the board.
> 

k3-<board>-ddr.dtsi* (e.g J721E EVM vs. SK boards consume different memory 
configurations.

> And one board might have multiple configuration depending on the
> variant of the board. Typically, one board is available with
> different memory options. i.e. 1GiB, 4GiB and so on. The actual RAM
> chips can come from different manufacturers. So all all, I presume
> there will be different RAM settings, i.e. different
> k3-<soc>-ddr.dtsi. But I have to switch between the setting during
> runtime because there will be only one boot image for that board.

This is a runtime dynamic DDR configuration support you are describing correct? 
This means you would be including all the supported memory option DTSIs in your 
k3-<board>-r5.dts correct and probably do some board magic code in the SPL DDR 
driver to choose the DTB. How is this affecting the packing of the final 
bootloader which will anyways pack the whole R5 DTB?
> 
>>> Also, regarding the board variants, I'll probably need to choose
>>> between multiple device trees. That is invisible to the user,
>>> because u-boot will choose the correct DTB according a board
>>> strapping, which btw. works really fine, see for example
>>> (boards/kontron/sl28/spl.c:board_fit_config_name_match).
>>
>> Again, this is assuming that there is some HW blown register available
>> for the board to use (or in our earlier K3 case, the EEPROM) but that is
>> not necessarily true every time.
> 
> No, that is of course board dependent. It is just an example that
> there are boards with more than one DTB.
> 
> Let's step back a bit. Right now, there is
>    k3-<soc>-<board>-binman.dtsi
> which is fine. But it seems, that TI is heading towards a common
>    k3-<soc>-binman.dtsi
> which is intended to be used by all the boards that are using that
> particular SoC, correct me if I'm wrong here. Now the problem with
> that is that you hardcode the FIT configuations which are really
> board dependent and assume that there will be exactly one DTB per
> board, i.e. your "#define SPL_BOARD_DTB" etc.
> 

Correct, but as I mentioned in the earlier message, if your board supports more 
than 1 FIT configuration, you can easily extend the image and add more 
configurations.

> Thus, what I was trying to say is that you should split all the
> board independent configuration (dt fragments) from the board
> specific configuration.
> 
> And again, of course I could just ignore the k3-<soc>-binman.dtsi
> and just use a suitable copy "k3-<soc>-<myboard>-binman.dtsi" for my
> board. But as I said, I'm not sure, this is the way to go and I have
> a slight feeling I will be asked to reuse the "k3-<soc>-binman.dtsi"
> when I submit my board support.
> 
>>>
>>> I don't think it makes much sense to hardcode your generic
>>> *-binman.dtsi to just one FIT configuration. I'd rather see a split
>>> between generic things which are shared across all boards and board
>>> specifics, like the FIT configuration. I mean I could just copy all
>>
>> Correct me if I'm wrong, but my understanding is that you would want to
>> add more FDT blobs in the *-binman.dtsi correct? That is still possible,
>> adding another "fdt-1" and "conf-1" in the
>>
>> Something like this in your <board>-u-boot.dtsi,
>>
>> tispl {
>> 	insert-template = <&ti_spl>;
>> 	fit {
>> 		images {
>> 			fdt-1 {
>> 				...
>> 			};
>> 		};
>> 		configurations {
>> 			conf-1 {
>> 				...
>> 			};
>> 		};
>> 	};
>> };
> 
> Then you have the information at two places. One being the "#define
> SPL_BOARD_DTB" stuff and the other one being in this additional DT
> fragment. That is really confusing.
> 

Hm... maybe. I personally don't see it as confusing. Even when picking between 
multiple DTBs, you will have a default DTB in any case, marking that as a macro 
wouldn't be confusing right? We'll need to get a third opinion on here then, I 
had seen your ping on IRC [1], putting it here for the others as well.

>> provided you have the support to handle this multi-dtb FIT. But as far
>> as reusing the k3-binman.dtsi and k3-<soc>-binman.dtsi goes; you should
>> be able to do it.
>>
>>> the binman and tiboot3.bin and tispl.bin magic and put it into my
>>> own "-u-boot.dtsi". But I'm not sure that will make things any
>>> better.
>>>
>>> -michael
> 

[1] https://libera.irclog.whitequark.org/u-boot/2024-03-28

-- 
Thanking You
Neha Malcom Francis