On 2/14/26 10:29 AM, Denys Dmytriyenko wrote:
On Thu, Feb 12, 2026 at 04:03:34PM -0600, Andrew Davis via
lists.yoctoproject.org wrote:
On 2/12/26 5:42 AM, Vitor Soares wrote:
On Wed, 2026-02-11 at 18:52 -0600, Andrew Davis wrote:
On 2/11/26 11:15 AM, Vitor Soares via lists.yoctoproject.org wrote:
Hi,
While working on our Toradex Aquila AM69 SoM support, we noticed that k3.inc
defines:
SERIAL_CONSOLES = "115200;ttyS0 115200;ttyS2"
This caused systemd to enable [email protected] on our board, where
ttyS0 maps to main_uart4 - a user-facing UART, not a debug/console UART. The
getty continuously outputs a login prompt, interfering with the interface.
This made us question whether SERIAL_CONSOLES belongs in SoC-level include
files
at all. Our understanding is that the console getty is already handled
automatically by systemd-getty-generator from the console= kernel command
line argument, which makes SERIAL_CONSOLES potentially redundant for the
console
UART. Any additional UARTs beyond the console are board-specific and
arguably
belong in each machine config rather than a shared SoC include.
We are not 100% certain about this reasoning and would love to hear the
rationale behind the current approach. Are we missing something? Would it
make
sense to either remove SERIAL_CONSOLES from the SoC includes or at minimum
use
a weak assignment (?=) to allow boards to override cleanly?
We simply noticed almost all our EVM boards used the same console UARTs and so
refactored the definition into a more common place. But yes you are correct
that a file like k3.inc seems like it should contain only items common to all
users of the K3 SoC family, and not simply common to TI EVM boards. Changing
the assignment in k3.inc to a weak one seem reasonable (although I think you
can still just override regular assignments after you include k3.conf in
your board conf)
You're right we can override it in our board configs, but I wanted to share why
we think this is worth reconsidering.
The issue is that SERIAL_CONSOLES in k3.inc mixes SoC capabilities with board-
specific UART usage.
The K3 SoC provides the UART peripherals, but doesn't determine which ones are
used for console, user interfaces, or other purposes - that's decided at the
board level.
Different boards use UARTs differently. On TI EVMs, ttyS2 or ttyS0 might be the
debug console. On our Aquila AM69, ttyS2 is the console but ttyS0 is a user-
facing UART. Same SoC, different UART assignments.
SERIAL_CONSOLES creates getty login prompts on listed UARTs, which is a
board/system integration concern.
I think the same could be said about `IMAGE_FSTYPES` which is also in `k3.inc`
but really should depend on the type of media on a board.
If TI's EVM boards share the same UART configuration, would it make sense to
have a ti-evm-common.inc instead of putting it in the SoC include?
Happy to help with refactoring if you think this direction makes sense.
This directions makes sense to me, I'd have no issue with a ti-evm-common.inc.
This seems like an overkill for a single variable that can be easily resolved
with a weak assignment. Moreover, this can quickly get out of hand with SKs
and Beagles and how those sometimes have convoluted mixed ownership...
Or even since it is just 1 or 2 lines that would need to be moved out, might
be easier to just move those lines back out to the different board configs.
If you want to send the patch then up to you.
Isn't it how it used to be originally?
Yes, I'm basically saying I'm okay with reverting my patch[0] that consolidated
these[0].
The idea of a generic setting getting weakly assigned with a sane default
is specifically to make it easy for downstream layers to change, while keep
things maintainable upstream.
We've been slowly trying to unify, cleanup and consolidate such settings in
the past. To that extent, I keep reminding Ryan and others to avoid using
:append override, which is hard to undo downstream, as well as use weak +=
instead of = for assigning SoC and machine settings.
The patch that moved these to k3.inc used =, it should have used +=, so
changing that to += is also an option if we don't want to revert [0].
Andrew
[0]
https://git.yoctoproject.org/meta-ti/commit/?id=bc894bf52eaa4359eaa64a64e950accd099a81cb
We've seen instances where a sane generic default works for most of the cases,
but for a couple of cases need to be changed in more specific location. That's
why we have hierarchical include structure from most generic to most specific:
ti-soc.inc -> k3.inc -> j784s4.inc -> j784s4-evm.conf
On the other hand, specifying U-boot defconfig (UBOOT_MACHINE/UBOOT_CONFIG) is
almost always done in corresponding specific machine configs, simply becasue
there's no generic default...
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19517):
https://lists.yoctoproject.org/g/meta-ti/message/19517
Mute This Topic: https://lists.yoctoproject.org/mt/117760181/21656
Group Owner: [email protected]
Unsubscribe: https://lists.yoctoproject.org/g/meta-ti/unsub
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
