On 2024-05-28 10:17, Heiko Stübner wrote: > Am Freitag, 17. Mai 2024, 08:58:32 CEST schrieb Luca Ceresoli: >> Hello Dmitry, >> >> On Thu, 16 May 2024 17:06:46 +0500 >> Dmitry Yashin <dmt.yashin@xxxxxxxxx> wrote: >> >>> Hi Luca, >>> >>> On 15.05.24 21:29, Luca Ceresoli wrote: >>>> I'm skeptical about this being bound to a new DT compatible. As far as I >>>> know the RK3308 and RK3308B are mostly equivalent, so it looks as the >>>> pinctrl implementation could be detected at runtime. This would let >>>> products to be built with either chip version and work on any without >>>> any DT change. >>> >>> >>> Thanks for your feedback. >>> >>> Indeed, these SoC's have a lot in common, but as I can see the rk3308b >>> has more blocks, like extra PWM's (rk3308 datasheet 1.5 [0] shows only >>> 1x PWM 4ch, when rk3308b and rk3308b-s have 3x PWM 4ch), 1-wire and >>> CAN controller (mentioned in the TRM, but dropped from rk3308b >>> datasheet for some reason). >>> >>> So, in my view, it really makes sense to add rk3308b.dtsi, where extra >>> PWM's, pinctrl compatible and its pin functions can be moved. And if >>> its not worth it, then I will try to adapt the entire series to runtime >>> config based on cpuid like you suggested. >> >> Having a rk3308b.dtsi would probably make sense, yes, as there are >> several differences as you described. However for the pinctrl it seems >> probably not necessary. >> >> I've seen actual products being manufactured with two different RK3308 >> variants in different lots of production, but with the same DT that has >> rockchip,rk3308-pinctrl in it. Those would need a _selective_ DT >> upgrade in order to benefit from your changes. >> >> And even if a product had always used the B variant, it would need DT >> upgrade when upgrading to a kernel with your changes. Otherwise with >> patch 1/3 of this series the pictrl driver would lose many routes after >> upgrading the kernel (but not the DT): can this lead to >> previously-working devices to stop working? I think this is a >> fundamental question to reply. > > If things can be runtime-detectable, they should be detected at runtime. > So yes, while we need to know that it is a rk3308-something before > via the dt, if we can distinguish between the rk3308 variants at runtime > we should definitly do so. The GRF_CHIP_ID reg (0xFF000800) can be used to identify what model is used at runtime: RK3308: 0xCEA (errata: chip id value is 32'h0cea (32'd3306)) RK3308B: 0x3308 RK3308BS: 0x3308C Vendor U-Boot make use of this reg to identify what model is running: https://github.com/rockchip-linux/u-boot/blob/next-dev/arch/arm/include/asm/arch-rockchip/cpu.h#L68-L82 I can only validate on real hw that the reg value is 0x3308 for RK3308B. Regards, Jonas > > Heiko >
