06:58 <qschulz> a3f: /chosen/bootsource should contain the full path to the node, so not sure why /soc/aliases would help with that?

07:15 <qschulz> and if we're talking a mapping from "what the bootrom returns as MMC1" and "what the path in the DT is for MMC1 based on /soc/aliases" we'll have a bad time, since I assume /soc/aliases is like /aliases and may not actually always represent the same controller on the board

07:16 <qschulz> e.g. we could have a board say mmc1 is sdmmc and another saying it's sdhci. This information is irrelevant to the mapping of which MMC controller was used by the BootROM to load the first stage?

07:38 <Kwiboo> mmind00: I am working on pm domain for RK3528 and would like to avoid having to define a power domain as always-on, do you see any issue with allowing to define power-domains for gpio, i2c, saradc devices? see top commits from https://github.com/Kwiboo/linux-rockchip/commits/next-20250509-rk3528-gpu-pmdomain/

07:40 <Kwiboo> there is also one issue that I do not know how to handle properly: there is one clock related to ddrphy in the RK3528_PD_RKVDEC domain, without this domain enabled trying to read scmi_clk_ddr will result in a system reset/reboot :-/

07:41 <Kwiboo> my initial thought is just to skip declaring it in DT and push the issue until we have a rkvdec2 driver ready, or to declare it always-on

07:42 <mmind00> Kwiboo: the first thing to show is the strange association (VPU/VO for gpios and uarts), but that is part of the soc makeup I guess

07:43 <mmind00> Kwiboo: technically I don't have a problem with defining domains for those .... they are connected to the devices afterall

08:00 <Kwiboo> mmind00: agree, the clocks and domains are all over the place on rk3528, I was very supprised when I disabled the rkvenc domain and could not enable/disable regulators using gpio4 without a SError, not what I was expected

08:01 <Kwiboo> vendor kernel just flag all power domains as always-on, and the gpu power-domain is the only one that can be fully powered down, the reset just support putting into idle

08:18 <Kwiboo> I will prepare a proper series for this later

17:08 <CounterPillow> Sounds like a delightful SoC, I was already despairing with the RK3576 non-idempotent PD toggles :D

21:32 <user21> working on getting NPU working for the rock 5b+ board under armbian https://www.armbian.com/rock-5b-plus

21:33 <user21> I need help expressing this for the rock 5b+ https://gitlab.freedesktop.org/tomeu/linux/-/commit/a764c58bd0a1e87efab547bf46315f356adcbbf8

21:34 <user21> basically the right device tree to get the NPU working, this is based on the user space driver here https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29698

21:34 <user21> and the kernel patch here https://lore.kernel.org/linux-rockchip/20250225-6-10-rocket-v2-0-d4dbcfafc141@tomeuvizoso.net/

21:34 <CounterPillow> armbian doesn't even have the tomeu rocket driver in the kernel though, does it?

21:34 <user21> no but there is a patch and I got it applied to the 6.14.5 kernel

21:35 <user21> but that patch has the device tree for quartzpro64

21:35 <user21> I'm still learning the kernel development and don't understand yet this commit https://gitlab.freedesktop.org/tomeu/linux/-/commit/a764c58bd0a1e87efab547bf46315f356adcbbf8

21:35 <CounterPillow> well it's your lucky day https://gitlab.collabora.com/hardware-enablement/rockchip-3588/linux/-/commit/5da44d61b09c345309f76159574d447d071c295d

21:36 <user21> oh wow

21:36 <CounterPillow> should be quite similar for 5bp

21:36 <user21> so there is some effort for 5b already ?

21:36 <CounterPillow> yes

21:37 <CounterPillow> https://gitlab.collabora.com/hardware-enablement/rockchip-3588/linux/-/commits/fratti-tomeu-npu-rk3588?ref_type=heads

21:39 <user21> wow nice, that is literally where I spend last 3 weeks on, got it all merged and compiled user + kernel on 6.14.5

21:40 <user21> the device tree was missing, reached out to Tomeu as well for questions

21:40 <user21> nice

21:43 <user21> was the plan to merge this into mainline any time soon ? I was looking at 6.12 minimum, but the latest kernel the better