On 11/19/25 11:09 AM, Chaoyi Chen wrote:
Hi Quentin,
On 11/19/2025 5:52 PM, Quentin Schulz wrote:
Hi Chaoyi Chen,
On 11/19/25 2:55 AM, Chaoyi Chen wrote:
From: Chaoyi Chen <[email protected]>
Some pmdomains rely on RK806 PMIC for power supply. If we don't
enable them in U-Boot, random panic may occur during the Kernel
boot process.
The kernel shouldn't rely on U-Boot doing something to work, so please
fix this with upstream Linux kernel.
Can you tell us what's the issue?
The issue with this patch here is that we probably would need **all**
RK3588 boards to have the PMIC enabled, and all the appropriate
regulators brought up before booting into the kernel. It'd be *much*
better this is fixed in Linux (and depending on what we're talking
about, it may need another fix in U-Boot so that whatever depends on
the PM domain works as well).
Also, why the PMIC in SPL but not the regulators?
It make sense, but I haven't thought of a specific solution for Kernel yet.
To be specific, I'm trying to get RKVENC to work. It uses an independent
power domain. If RK806 is not enabled in U-Boot, the first power-up
attempt for the pmdomain corresponding to RKVENC will fail in the Kernel.
I assume you tried to add
domain-supply = <&vdd_vdenc_s0>;
to power-domain@RK3588_PD_VCODEC node to model what the HW is expecting?
Doing something similar to the NPU essentially, which has a similar
issue as far as I remember (I am not too involved in kernel development
nowadays so I only get snippets of info here and there).
And essentially, the operations of pmdomains in the Kernel are dependent
on regulators. But it seems that pmdomains driver haven't dealt with
this. I have seen some cases, but it seems that the dependency issue has
not been fully resolved [0].
The problem with working around the issue by applying bandage fixes in
U-Boot is that nobody will actually work on fixing the actual issue.
We've seen that with SoCs with IO domains which are still not properly
supported by the kernel but since IO domains are now initialized by
U-Boot, the issue doesn't show anymore, c.f.
https://lore.kernel.org/linux-rockchip/[email protected]/.
Cheers,
Quentin
