[PATCH u-boot-mvebu 00/31] kwboot / kwbimage improvements

[Stefan Roese]

Hi Marek,

On 09.07.21 16:08, Marek Behún wrote:
> On Fri, 9 Jul 2021 14:35:15 +0200
> Stefan Roese <sr at denx.de> wrote:
> 
>> Thanks. And could you please also do one (or more) boot-time
>> comparisons, old vs. new version? This way we can see, if and how the
>> boot-time is affected (perhaps improved, which would be great) by this
>> patch series.
> 
> For Turris Omnia, the size of the u-boot-spl.kwb binary decreases by
> ~37 KiB, mainly due to not needing padding anymore. The size of the SPL
> binary itself (u-boot-spl.bin) decreases by 7.5 KiB (due to code
> reduction - no SPI NOR code).
> 
> Unfortunately the boot time increases by 815 ms, from 2341 ms (on
> average) to 3156 ms (on average).

Ups. This boot-time increase is quite big. And there might be some
Armada / MVEBU U-Boot targets that are tuned for fast booting. I can
speak for the Armada XP theadorable board, where boot-time is quite
critical. I remember that we tuned and optimized here and an increase
of perhaps 100ms would be acceptable. But an increase of nearly 1
second is definitely not acceptable here, I'm afraid.

Please see below.

> This is probably because BootROM read the memory differently (lower
> frequency or some other reason).
> 
> I think this 815 ms increase in boot time into U-Boot prompt is worth
> the 37 KiB saved space for U-Boot code (mainly because on Omnia we have
> 960 KiB for U-Boot and currently we are at 860 KiB, so enabling some
> other features in the future may push us to the limit), but others may
> not agree.

Yes, I disagree here. Boot-time is critical for some platforms and this
increase is just too high. I might be able to do some tests on
theadorable to get actual numbers for this platform.

Could you perhaps add this "SPL returns to BootROM" support as an
optional feature that can be selected on a per-board basis?

Thanks,
Stefan

> I am going to put this information also into v2 cover letter.
> 
> Marek
>