Re: [PATCH] regulator: qcom-rpmh: Revert "regulator: qcom-rpmh: Use PROBE_FORCE_SYNCHRONOUS"
From: Doug Anderson
Date: Mon May 15 2023 - 11:03:46 EST
On Mon, May 15, 2023 at 7:42 AM Amit Pundir <amit.pundir@xxxxxxxxxx> wrote:
> On Sun, 14 May 2023 at 18:11, Caleb Connolly <caleb.connolly@xxxxxxxxxx> wrote:
> > On 13/05/2023 18:08, Amit Pundir wrote:
> > > On Fri, 24 Mar 2023 at 19:05, Douglas Anderson <dianders@xxxxxxxxxxxx> wrote:
> > >>
> > >> This reverts commit 58973046c1bf ("regulator: qcom-rpmh: Use
> > >> PROBE_FORCE_SYNCHRONOUS"). Further digging into the problems that
> > >> prompted the us to switch to synchronous probe showed that the root
> > >> cause was a missing "rootwait" in the kernel command line
> > >> arguments. Let's reinstate asynchronous probe.
> > >
> > > Hi, the asynchronous probe is broken on Dragonboard 845c (SDM845)
> > > running AOSP (Android Open Source Project) with v6.4-rc1
> > > https://bugs.linaro.org/show_bug.cgi?id=5975.
> > > Can we please go back to synchronous probe.
> > >
> > > AOSP do not make use of rootwait, IIRC, but it is added by the
> > > bootloader anyway. And the device fails to boot AOSP regardless of
> > > "rootwait" bootarg being present or not.
> > Could you try applying this diff to enable some log spam and let me know
> > what you get? I'm keen to try and figure this one out. My mail client
> > might crunch this a bit so I have pasted it here too
> > https://p.calebs.dev/ab74b7@raw
> These prints add just enough delay for the UFS probe to succeed that I
> can't reproduce the failure anymore.
I'd prefer doing at least a little debugging before jumping to a
revert. From looking at your dmesg , it looks as if the async probe
is allowing RPMH to probe at the same time as "qcom-vadc-common".
That's something that talks on the SPMI bus and is (potentially)
talking to the same PMICs that RPMH-regulator is, right? I'm by no
means an expert on how Qualcomm's PMICs work, but it seems plausible
that the "qcom-vadc-common" is somehow causing problems and screwing
up RPMH. Does that seem plausible to you?
If so, one interesting way to track it down would be to move around
delays. Put ~500ms sleep at the _end_ of vadc_probe(). Presumably that
_won't_ fix the problem. Now put a ~500ms sleep at the beginning of
vadc_probe(). Maybe that will fix the problem? If so, you can move the
delay around to narrow down the conflict. My wild guess would be that
vadc_reset() could be throwing things for a loop?
If the above doesn't work, maybe we could add more tracing / printouts
to see what is probing at the same time as RPMH?