Re: [PATCH v6 3/8] clk: sunxi-ng: a733: Add PRCM CCU

[Enzo Adriano]

Hi Junhui,

> I'd prefer not to mark these clocks as critical just for the vendor
> firmware. You can see my reply in v5:

I agree that native PSCI and a future OR100 Crust port are separate from
the vendor SCP on E902, and that future E902 remoteproc/mailbox drivers
should own these clocks. Vinicius's native-PSCI reboot report used locally
modified TF-A on A7Z [1]. Upstream TF-A still has no A733 target; its port
is under review [2]. The immediate problem is that the new CCU's
unused-clock cleanup breaks reboot with the firmware these boards ship.

On one Cubie A7S, RAM-booting v7.4-rc1 + CCU v6 + DT v3 and its dependencies
with vendor boot0/BL31/SCP/U-Boot and no clk_ignore_unused, the run without
critical flags stalled at reboot -f. With only bus-r-riscv and bus-r-msgbox
critical, two warm reboots returned to vendor Linux while r-timer0,
bus-r-timer and bus-msgbox0 were 0/0 (enable/prepare); five flags also passed
once.

> Since you've already submitted a separate patch adding these critical
> flags, I'll leave that to you.

I will narrow my separate patch [3] to bus-r-riscv and bus-r-msgbox.
Chen-Yu, Jernej, Samuel: would you take these two flags as a separate patch
until E902/mailbox drivers can own the clocks? There is precedent in A523
bus-r-cpucfg (1dba74abf3e2) and Meson AXG fclk_div3, kept critical for SCPI
firmware until the driver claims its clocks and a handoff exists
(drivers/clk/meson/axg.c).

I have not tested power-off, suspend, either flag alone on A7S, or other
boards or firmware; Vinicius found either flag alone insufficient on A7Z
with vendor BL31 [4].

[1] https://lore.kernel.org/r/20261011200153.575096-1-vinicius.eduardo.pedrosa@xxxxxxxxx
[2] https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/55200
[3] https://lore.kernel.org/r/20261031020640-263f7b05-a7s-clock@xxxxxxxxx
[4] https://lore.kernel.org/r/20261005183038.417798-1-vinicius.eduardo.pedrosa@xxxxxxxxx

Thanks,
Enzo