Hi Heiko, Jiaxing, On 9/22/2026 6:06 AM, Heiko Stuebner wrote: > Hi, > > Am Dienstag, 15. September 2026, 12:43:23 Mitteleuropäische Sommerzeit > schrieb Jiaxing Hu: >> The RK3576 NPU domains need a short settle time after the idle request is >> released before the registers behind the domain answer. Without it the QoS >> writes that rockchip_pmu_restore_qos() issues land while the domain is >> still coming up, and the NPU throws an async SError on the first cold >> power-on. > > are you really really sure, this is the fault of the domain itself? > > The regulator also can just need a longer time to bring up the power. > Looking at [0], it seems I actually encountered the same thing back > on the RK3588. > > Only difference is, on RK3588 a fan5355-type regulator normally does > provide the npu supply, while on your board it comes from the main pmic. > > Can you please try the following: > - set your power-domain regulator to always-on in the DT (so it stays > on even when the domain gets turned off) > - try to reproduce your SError issue and see if it still appears > > > And for the other case, look through the pmic datasheet to find the > enable-time values and do something similar for the RK806. > >
Thank you for your attention to this series. Providing some background here: this delay is required by the pmdomain design itself, rather than being dependent on the external RK806 state. Taking the NPU as an example, after its pmdomain is powered on, some internal preparation is needed, which takes approximately 15us. And during this period, the QOS registers *must* not be accessed, otherwise it will cause a hang. -- Best, Chaoyi
