Re: [PATCH v7] memory: tegra210: Support interconnect framework

[Nicolas Chauvet]

Le lun. 6 avr. 2026 à 08:36, Aaron Kling via B4 Relay
<devnull+webgeek1234.gmail.com@xxxxxxxxxx> a écrit :
>
> From: Aaron Kling <webgeek1234@xxxxxxxxx>
>
> This makes mc and emc interconnect providers and allows for dynamic
> memory clock scaling.
>
> Signed-off-by: Aaron Kling <webgeek1234@xxxxxxxxx>
> ---
> Changes in v7:

I've tried this patch (on top of the memory-controller-drv-7.1 tag
which contains the base-commit mentioned).

I have the following error at boot:
---
OF: /actmon@6000c800: could not get #interconnect-cells for
/memory-controller@70019000
of: _read_bw: Mismatch between opp-peak-kBps and paths (1 0)
tegra-devfreq 6000c800.actmon: _opp_add_static_v2: opp key field not found
tegra-devfreq 6000c800.actmon: _of_add_opp_table_v2: Failed to add OPP, -22
tegra-devfreq 6000c800.actmon: Failed to add OPP table: -22
tegra-devfreq 6000c800.actmon: probe with driver tegra-devfreq failed
with error -22
---
But then, I can still have a usable display IIRC. Also worth noting
that I add explicitly loaded tegra30-devfreq.

If I add the next-patch that enables only interconnect-cells for MC
and EMC (without the interconnect lines on dc0 and dc1),
then in this case I'm experiencing some freeze at boot and the display
does not work. (display stays black with some flickering while I
should have a text console prompt).
Nothing obvious from logs.

Using the original patch that enabled the interconnect in (and
emc-tables passed from bootloader).
https://lore.kernel.org/linux-tegra/20251021-t210-actmon-p4-v5-2-4a4dbc49fbc8@xxxxxxxxx/
This patch sometimes works, but I have some random behavior that
sometimes prevents booting to userspace
I'm suspecting some module load ordering issue, or maybe a state that
isn't properly re-initialized on cold boot...

When the system boots, the emc scaling seems to provide good results
under gnome (no more flickering display with mesa-25.1.9).
glmark2-wayland provided good results (except for one scene that
breaks, as previously).
So this looks promising...

I will test again with 7.1-rc1+ and with a lower patch list. hopefully
by the end of May.