On Fri, Aug 23, 2024 at 12:47:50PM +0200, Heiko Stübner wrote: > Am Donnerstag, 22. August 2024, 10:41:10 CEST schrieb Conor Dooley: > > On Thu, Aug 22, 2024 at 09:01:34AM +0200, Heiko Stübner wrote: > > > @Conor: just for me, did some shift happen in our understanding of dt- > > > best-practices in terms of syscon via phandle vs. syscon via compatible? > > > > > > Because Rockchip boards are referencing their GRFs via phandes forever > > > but similar to the soc vs non-soc node thing, I'd like to stay on top of > > > best-practices ;-) > > > > If IP blocks, and thus drivers, are going to be reused between devices, > > using the phandles makes sense given that it is unlikely that syscon > > nodes can make use of fallback compatibles due to bits within that "glue" > > changing between devices. It also makes sense when there are multiple > > instances of an IP on the device, which need to use different syscons. > > My goal is to ask people why they are using these type of syscons > > phandle properties, cos often they are not required at all - for example > > with clocks where you effectively need a whole new driver for every > > single soc and having a phandle property buys you nothing. > > I guess I'm of two minds here. > > For me at least it makes sense to spell out the dependency to the > syscon in the devicetree and not just have that hidden away inside the > driver. > > But on the other hand, we already have the per-soc configuration [0] > defining which grf bits needs to be accessed, so adding a > > .lanecfg1_grf_compat = "rockchip,rk3568-vo" > > would not create overhad, as the grf regs and bits and rearranged > all the time anyway. Right, that's the other side of it. Raw phandles aren't that great if the bits are gonna move around the register and you have to use the match data to figure out where they are. phandle + offset is the other option for that kind of scenario, particularly in cases where there are multiple copies of a block on an SoC and each uses either a different syscon or a different set of registers within one. > > > [0] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/rockchip/dw-mipi-dsi-rockchip.c#n1652 > taking DSI as an example, where this is even more obvious > >
Attachment:
signature.asc
Description: PGP signature
