Re: [PATCH v3] phy/qcom-qmp-combo: propagate correct return value at phy_power_on()

[Stephen Boyd]

Quoting Abhinav Kumar (2024-04-03 14:28:52)
>
>
> On 4/3/2024 1:04 PM, Stephen Boyd wrote:
> > Quoting Abhinav Kumar (2024-04-03 12:58:50)
> >>
> >>
> >> On 4/3/2024 12:51 PM, Stephen Boyd wrote:
> >>> Quoting Kuogee Hsieh (2024-03-29 12:50:35)
> >>>> Currently qmp_combo_dp_power_on() always return 0 in regardless of
> >>>> return value of cfg->configure_dp_phy(). This patch propagate
> >>>> return value of cfg->configure_dp_phy() all the way back to caller.
> >>>
> >>> Is this found via code inspection or because the phy is failing to power
> >>> on sometimes? I ask because I'm looking at a DP bug on Trogdor with
> >>> chromeos' v6.6 based kernel and wondering if this is related.
> >>>
> >>
> >> No, we actually hit an issue. This issue was originally reported as a
> >> link training issue while bringing up DP on x1e80100.
> >>
> >> While debugging that we noticed that we should not have even proceeded
> >> to link training because the PLL was not getting locked and it was
> >> failing silently since there are no other error prints (and hence the
> >> second part of the patch to improve the error logs), and we do not
> >> return any error code from this driver, we could not catch the PLL
> >> unlocked issue.
> >
> > Did link training succeed in that case and the screen was black? Also,
> > did you figure out why the PLL failed to lock? I sometimes see reports
> > now with an "Unexpected interrupt:" message from the DP driver and the
> > interrupt is the PLL unlocked one (DP_INTR_PLL_UNLOCKED).
> >
>
> No the link training had failed.
>
> Yes, root-cause was that the PLL registers were misconfigured in the
> x1e80100 DP PHY for HBR2. Once we programmed the correct values it
> worked. This was specific to x1e80100.

Ah ok, so that's what the x1e80100 patch is about.

>
> Yes, Doug mentioned this to me on IRC that this issue is still there.
> Surprising because I thought we had pushed
> https://patchwork.freedesktop.org/patch/551847/ long ago and it was
> fixed. It certainly did that time when I had tested this.

I see it on v6.6 and it is also on v5.15.y (stable kernel) so that has
been picked back. Somehow the aux interrupt is still happening though
when the PLL isn't locked. Maybe that interrupt bit should be masked in
most cases and only unmasked when something in the driver is going to
care about it.