[PATCH COVER] rockchip: rk3399: fix SPI-NOR flash not found in U-Boot pre-reloc

[Jonas Karlman]

Hi Quentin,

On 2024-06-17 16:03, Quentin Schulz wrote:
> Hi Jonas,
> 
> On 6/17/24 3:44 PM, Jonas Karlman wrote:
>> Hi Quentin,
>>
>> On 2024-06-17 15:29, Quentin Schulz wrote:
>>> Hi Jonas,
>>>
>>> On 6/17/24 3:26 PM, Jonas Karlman wrote:
>>>> Hi Quentin,
>>>>
>>>> On 2024-06-17 15:10, Quentin Schulz wrote:
>>>>> From: Quentin Schulz <quentin.schulz at cherry.de>
>>>>>
>>>>> In commit 100f489f58a6 ("rockchip: rk3399: Fix loading FIT from SD-card
>>>>> when booting from eMMC"), the spi1 bootph properties were mistakenly
>>>>> removed meaning, so re-add them back to fix SPI-NOR flash not being
>>>>> found in U-Boot pre-reloc as required for RK3399 Puma.
>>>>
>>>> Good catch, for TPL/SPL the bootph props is propagated, something that
>>>> is not done for pre-reloc.
>>>>
>>>
>>> Can you tell us a bit more about this? I know that the pinctrl nodes
>>> marked for pre-reloc recursively apply the same to their parent, but
>>> couldn't find anything similar for other subsystems for example. What
>>> did I miss in my 5m search?
>>
>> v2024.04-rc1 added support to propagate bootph props using fdtgrep for
>> the u-boot-tpl/spl.dtb files, however pre-reloc the main u-boot.dtb (or
>> possible the FDT included in FIT) that is not processed by fdtgrep.
>>
>> fdtgrep: Allow propagating properties up to supernodes
>> https://source.denx.de/u-boot/u-boot/-/commit/7a06cc2027c0169c462da63a68fa269c0d59a950
>>
>> Makefile: Use the fdtgrep -u flag
>> https://source.denx.de/u-boot/u-boot/-/commit/aca95282c1b72c41d8e72984b1dceb15f396b2f8
>>
> 
> Thanks for the pointers!
> 
> I'm wondering if we shouldn't do the same for U-Boot proper DTB as well?

I think that would be good, and should also remove the need for the
recursive pinctrl lookup for bootph props during pre-reloc.

> 
> i.e.
> 
> cmd_fdt_rm_props = cat $< | $(objtree)/tools/fdtgrep -r -O dtb - -o $@
> 
> to
> 
> cmd_fdt_rm_props = $(objtree)/tools/fdtgrep $(fdtgrep_props) -u  $< | 
> $(objtree)/tools/fdtgrep -r -O dtb - -o $@
> 
> I assume we need to add a bunch more options though.
> 
> But that would make sense to me, the fact that there's a pre-reloc stage 
> in U-Boot proper is already a bit odd (to me), but if we have something 
> that behaves differently than the pre-reloc stage in earlier stages... 
> I'm not sure this is a good idea?

Fully agree that part of the pre-reloc stage make little sense at least
on Rockchip platform where full memory is available before pre-reloc.

(I am hoping to get an initial series to remove time wasted during
pre-reloc for Rockchip platform out in near future, e.g. ~250ms is
currently wasted looking up e.g. CRU_BASE on rk3399, a constant that is
known at compile time).

> 
> While at it, I'm wondering if we didn't miss the removal of 
> bootph-some-ram in cmd_fdtgrep for the VPL/SPL/TPL?

I think so too, the bootph-some-ram should be safe to remove.

We can/should probably also extend fdtgrep to remove disabled nodes for
the VPL/SPL/TPL.

> 
> Additionally, we should be able to remove bootph-pre-sram and 
> bootph-pre-ram in U-Boot proper DT since those only apply to non-U-Boot 
> proper stages, don't you think?

I do not fully think we can remove them, due to the dual use of 
ofnode_pre_reloc(). During pre-reloc it will return true if node is
required, and after pre-reloc it return true if the node was required
during any prior phase. Unsure if anything depend on this secondry use.

Regards,
Jonas

> 
> Cheers,
> Quentin