[PATCH v2 0/9] spl: atf: add support for LOAD_IMAGE_V2

[Michael Walle]

Am 2020-11-20 12:15, schrieb Michal Simek:
> On 20. 11. 20 11:48, Michael Walle wrote:
>> Am 2020-11-20 11:14, schrieb Michal Simek:
>>> Hi,
>>> 
>>> On 18. 11. 20 17:45, Michael Walle wrote:
>>>> Newer TF-A versions provide a new image loading protocol. This is
>>>> used on
>>>> (newer?) NXP's SoCs. Normally, the bootflow is bl1 -> bl2 -> bl31 ->
>>>> u-boot. With this series it is possible that U-Boot SPL loads the 
>>>> bl31
>>>> directly and thus replacing bl1 and bl2 from the TF-A.
>>>> 
>>>> This was tested on the Kontron sl28 board using NXPs bl31 and the
>>>> upstream
>>>> version of the OP-TEE Trusted OS.
>>> 
>>> I still have some questions about this.
>>> 
>>> As I see from TFA previous image format has been removed in 2018 by
>>> 
>>> commit ed51b51f7a9163a7fc48289c5ed97a3fe4fe504f
>>> Author:     Roberto Vargas <roberto.vargas at arm.com>
>>> AuthorDate: Mon Sep 24 17:20:48 2018 +0100
>>> Commit:     Antonio Nino Diaz <antonio.ninodiaz at arm.com>
>>> CommitDate: Fri Sep 28 15:31:52 2018 +0100
>>> 
>>>     Remove build option LOAD_IMAGE_V2
>>> 
>>>     The code of LOAD_IMAGE_V2=0 has been removed.
>>> 
>>>     Change-Id: Iea03e5bebb90c66889bdb23f85c07d0c9717fffe
>>>     Co-authored-by: Antonio Nino Diaz <antonio.ninodiaz at arm.com>
>>>     Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz at arm.com>
>>> 
>>> 
>> 
>> DOH! Lol, I'm using just one non-upstream part for the whole board
>> and of course it is doing something miserable. I wasn't aware of this.
>> 
>>> On Xilinx ZynqMP I use SPL->bl31 loading but not using that TFA
>>> structure because xilinx is using own format.
>>> 
>>> But I am curious if V2 was removed in 2018 who is really using 
>>> previous
>>> one and also if current implemenation is origin or also not full v2.
>> 
>> NXP,
>> https://source.codeaurora.org/external/qoriq/qoriq-components/atf/tree/plat/nxp/common/common.mk#n55
>> 
>> 
>> The last non-nxp commit there was from around March 2018..
>> 
>>> And these patches are not breaking boot on zynqmp that's why not big
>>> deal for me.
>> 
>> I was looking at porting TFA to upstream for this board but there is
>> such a huge gap. Therefore, it seemed to be easier to just use the
>> vendor version for now.
> 
> Please get this reviewed by people who are using current blX code.

What do you mean by blX code? Nobody is using this flow for now, i.e.
   spl -> bl31 -> u-boot
For the LOAD_IMAGE_V2 case.

NXP is using bl1 -> bl2 -> bl3 -> u-boot.

> As I said this series is not breaking our flow on xilinx zynqmp soc but
> maybe your code is more or less duplication of what's there now.
> Or maybe if there is any NXP private way it should be handled
> differently as I do for zynqmp.

The code is split into the generic v2 handling and layerscape LS1028A
specific stuff, hence the _default() function which is used in the
board files and then arch specific parameters are applied.
Also I don't know wether unifying the
   bl2_plat_get_bl31_params_default() and
   bl2_plat_get_bl31_params_v2_default()
is worth it. If that is what you mean by "more or less a duplication".

-michael