Re: [PATCH v13 09/14] pmdomain: rockchip: add optional per-domain power-on settle delay

[Jiaxing Hu <gahing@gahingwoo.com>]

Hi Heiko,

Chaoyi has given the mechanism, and I ran your experiment on v13 as
posted, since the SError behind this patch was first seen in June on a
board that held vdd_npu_s0 up with a local regulator-always-on and had
no domain-supply.

Same kernel in every arm (this series, an unrelated dw-hdmi-qp patch,
and a test-only override of the table delay), v13's rock-4d dtb, and for
the last arm that dtb with only regulator-always-on added to vdd_npu_s0:

  delay 15 us, rail follows the domain   536 cold power-ons, clean
  delay 0,     rail follows the domain   SError at the first power-on
  delay 0,     regulator-always-on       SError at the first power-on

Both failures are the same:

  Kernel panic - not syncing: Asynchronous SError Interrupt
   regmap_write+0x58/0x78
   rockchip_pd_power+0x4b4/0x63c
   rockchip_pd_power_on+0x7c/0xd4

In the always-on arm the rail has been enabled since the regulator
registered, so nothing is ramping at that first power-on (1.4 s,
deferred probe), and it still faults. The fault needs only the domain,
as Chaoyi says, and the delay removes it. The enable time is already
described: rk3576-rock-4d.dts gives vdd_npu_s0
regulator-enable-ramp-delay = <400>.

Limits: one boot per failing arm, nothing tried between 0 and 15 us, and
neither failing arm got past its first power-on.

Regards,
Jiaxing