On Tue 22 Jun 15:23 CDT 2021, Stephen Boyd wrote: > Quoting Bjorn Andersson (2021-06-18 14:41:50) > > On Fri 18 Jun 15:49 CDT 2021, Stephen Boyd wrote: > > > > > Quoting khsieh@xxxxxxxxxxxxxx (2021-06-10 09:54:05) > > > > On 2021-06-08 16:10, Bjorn Andersson wrote: > > > > > On Tue 08 Jun 17:44 CDT 2021, Stephen Boyd wrote: > > > > > > > > > >> Honestly I suspect the DP PHY is _not_ in the CX domain as CX is for > > > > >> digital logic. Probably the PLL is the hardware that has some minimum > > > > >> CX > > > > >> requirement, and that flows down into the various display clks like > > > > >> the > > > > >> link clk that actually clock the DP controller hardware. The mdss_gdsc > > > > >> probably gates CX for the display subsystem (mdss) so if we had proper > > > > >> corner aggregation logic we could indicate that mdss_gdsc is a child > > > > >> of > > > > >> the CX domain and then make requests from the DP driver for particular > > > > >> link frequencies on the mdss_gdsc and then have that bubble up to CX > > > > >> appropriately. I don't think any of that sort of code is in place > > > > >> though, right? > > > > > > > > > > I haven't checked sc7180, but I'm guessing that it's following the > > > > > other > > > > > modern platforms, where all the MDSS related pieces (including e.g. > > > > > dispcc) lives in the MMCX domain, which is separate from CX. > > > > > > > > > > So the parent of MDSS_GDSC should be MMCX, while Kuogee's answer (and > > > > > the dp-opp-table) tells us that the PLL lives in the CX domain. > > > > > > Isn't MMCX a "child" of CX? At least my understanding is that MMCX is > > > basically a GDSC that clamps all of multimedia hardware block power > > > logic so that the leakage is minimized when multimedia isn't in use, > > > i.e. the device is suspended. In terms of bumping up the voltage we have > > > to pin that on CX though as far as I know because that's the only power > > > domain that can actually change voltage, while MMCX merely gates that > > > voltage for multimedia. > > > > > > > No, MMCX is a separate rail from CX, which powers the display blocks and > > is parent of MDSS_GDSC. But I see in rpmhpd that sc7180 is not one of > > these platforms, so I presume this means that the displayport controller > > thereby sits in MDSS_GDSC parented by CX. > > > > But in line with what you're saying, the naming of the supplies to the > > QMP indicates that the power for the PLLs is static. As such the only > > moving things would be the clock rates in the DP controller and as such > > that's what needs to scale the voltage. > > > > So if the resources we're scaling is the clocks in the DP controller > > then the gist of the patch is correct. The only details I see is that > > the DP controller actually sits in MDSS_GDSC - while it should control > > the level of its parent (CX). Not sure if we can describe that in a > > simple way. > > Right. I'm not sure things could be described any better right now. If > we need to change this to be MDSS_GDSC power domain and control the > level of the parent then I suppose we'll have to make some sort of DT > change and pair that with a driver change. Maybe if that happens we can > just pick a new compatible and leave the old code in place. > I would prefer that we stay away from making up a new compatible for that, but let's see when we get there. > Are you happy enough with this current patch? > Yes, I think this looks good. > > > > > > PS. Why does the node name of the opp-table have to be globally unique? > > Presumably the opp table node name can be 'opp-table' as long as it > lives under the node that's using it. If the opp table is at / or /soc > then it will need to be unique. I'd prefer just 'opp-table' if possible. I asked the same question (if it has to be globally unique) in the patch adding sdhci nodes for sc7280 and I didn't get a sufficient answer... So now I do want to know why "opp-table" wouldn't be sufficient name for these device-internal nodes. Regards, Bjorn