Re: [PATCH 1/5] ARM: dts: ti: Enable overlays for all DTB files

[Andreas Kemnade <andreas@kemnade.info>]

On Fri, 13 Feb 2026 06:33:46 +0100
"H. Nikolaus Schaller" <hns@goldelico.com> wrote:

> Hi,
> 
> > Am 12.02.2026 um 23:19 schrieb Andreas Kemnade <andreas@kemnade.info>:
> > 
> > On Thu, 12 Feb 2026 17:55:43 +0100
> > "H. Nikolaus Schaller" <hns@goldelico.com> wrote:
> >   
> >>> Am 12.02.2026 um 17:47 schrieb Andreas Kemnade <andreas@kemnade.info>:
> >>> 
> >>> On Thu, 12 Feb 2026 16:49:43 +0100
> >>> "H. Nikolaus Schaller" <hns@goldelico.com> wrote:
> >>>   
> >>>>> Am 12.02.2026 um 16:26 schrieb Kory Maincent (TI) <kory.maincent@bootlin.com>:
> >>>>> 
> >>>>> Allow overlays to be applied to any DTB. This adds around ~40% to the
> >>>>> total size of the DTB files on average.      
> >>>> 
> >>>> Is this unconditionally enabled or can it be turned off by some CONFIG? We have
> >>>> our own defconfig so I would not worry if if is enabled in omap2plus_defconfig
> >>>> and disabled in ours.
> >>>> 
> >>>> We have several devices where the boot loader can't handle overlays (never touch
> >>>> a working boot-loader :) So this seems to only contribute to build and load time
> >>>> without benefit.
> >>>>   
> >>> As long as you do not add overlays, the bootloader does not care. I would
> >>> like to simply carry around the 1-bit mmc overlay for one broken board.
> >>> That would help me. So I think there is a benefit but nobody forces
> >>> you to use it.    
> >> 
> >> Well, it does not force to use the really good feature, but it forces to add
> >> ~40% more file size and some more compile time, if I understand it correctly.
> >>   
> > Compile time, hardly measurable even if you just do make dtbs.
> > 
> > Size on disk:
> > a) if it lives around in a /boot partitions with kernels and initrams in it,
> >   then we are around 1% more space needed.  
> 
> Ah, I see. I was too focussed on the "adds around ~40% to the total size of the DTB files".
> 
> For the Letux arm distro all DTBs are around 8.1 MB at the moment so it will grow not that much
> (there are non-TI devices included).
> 
> So you are right, it is ~1% of the total if the kernel image is counted.
> 
> Therefore, space should not be something we should be too concerned about (although I remember
> discussions for driver code where every single byte did count).
> 
> On the other hand this increases load time from (sometimes slow) µSD for a specific DTB by 40%.
> That should at least be discussed.
> 
so also 1% of total loading time. And symbols can be stripped away during
e.g. deployment.
fdtput -r some_board.dtb /__symbols__

I am not strongly against it. It is just we are doing something unusual here.
Are there reasons why it is unusual I do not understand?
I think it can be useful for every board in some situations. For the
ones with can carry a head, they are the most useful.

What was the reasoning to have it enabled for k3 for every board? K3 and Broadcom-64bit
seem to be the only arch doing that. Broadcom mostly many RPis, some routers. 

Regards,
Andreas