[PATCH] fdtdec: drop needlessly convoluted CONFIG_PHANDLE_CHECK_SEQ

[Eugen.Hristev at microchip.com]

On 5/19/22 3:16 PM, Rasmus Villemoes wrote:
> On 19/05/2022 13.50, Aswath Govindraju wrote:
> 
>> Understood, thanks for the explanation. I am good with this patch.
>>
>> Acked-by: Aswath Govindraju <a-govindraju at ti.com>
>>
> 
> Thanks.
> 
> For completeness, to expand on this:
> 
>>> it's somewhat fragile to rely on (at least one of) the
>>> nodes in question to even have a phandle.
> 
> One way in which to ensure all nodes (with a label) do get a phandle is
> to build with the -@ flag to dtc, which is implied by setting
> CONFIG_OF_LIBFDT_OVERLAY=y. Three of the four _defconfigs that set
> PHANDLE_CHECK_SEQ also set that option, so for them it was guaranteed to
> work, but mostly by chance - if I randomly discovered
> CONFIG_PHANDLE_CHECK_SEQ and found I needed to enable it for my board, I
> wouldn't know to also enable CONFIG_OF_LIBFDT_OVERLAY. And if I didn't
> need overlay support as such, it would also bloat the .dtb (and U-Boot
> itself) needlessly; the difference for sama7g5ek_mmc_defconfig is 34784
> bytes for the current .dtb and 26920 bytes if one disables
> OF_LIBFDT_OVERLAY.
> 
> The last _defconfig with PHANDLE_CHECK_SEQ=y didn't actually seem to
> need it; I've built it and looked at u-boot.dtb, and there are no
> collisions in basenames in the aliases. Commit ddd778ae doesn't say
> anything about why it was added.
> 
> Rasmus
> 


Hi Rasmus,

For sama7g5 boards, we have the following situation:

some node {
	i2c1: i2c at 600: {
		some props;
	};
};

some other node {
	i2c2: i2c at 600: {
		other props;
	};
};

and then we want to be able to reference
aliases {
	i2c0 = &i2c1;
	i2c1 = &i2c2;
};


Without CONFIG_PHANDLE_CHECK_SEQ , both i2c0 and i2c1 aliases pointed to 
the same 'i2c at 600' node, because the code was searching through the DT 
for 'i2c at 600' and it was hitting the first in both cases.

With your patch, it can happen that we hit the same problem again ? Or 
we should be safe ?

Thanks,
Eugen