Re: [PATCH V2 2/2] cpufreq: dt: Refactor initialization to handle probe deferral properly

[Geert Uytterhoeven]

Hi Viresh,

On Wed, Oct 28, 2020 at 6:48 AM Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> On 27-10-20, 17:29, Geert Uytterhoeven wrote:
> > On plain v5.9, with #define DEBUG and a few extra debug prints
> > added, I get:
> >
> >     cpufreq_dt: cpufreq_init:164: policy->cpu = 0
> >     cpufreq_dt: cpufreq_init:165: policy->cpus = 0
> >     cpufreq_dt: cpufreq_init:166: policy->related_cpus =
> >     cpufreq_dt: cpufreq_init:167: policy->real_cpus =
> >     cpu cpu0: dev_pm_opp_of_get_sharing_cpus: Couldn't find opp node.
> >     of: dev_pm_opp_of_cpumask_add_table:1049
> >     of: dev_pm_opp_of_cpumask_add_table:1054: cpu 0
> >     cpu cpu0: dev_pm_opp_of_add_table:954
> >     cpu cpu0: dev_pm_opp_of_add_table:956:
> > dev_pm_opp_get_opp_table_indexed() returned (ptrval)
> >     cpu cpu0: _of_add_opp_table_v1:891
> >     cpu cpu0: _of_add_opp_table_v1:893: _find_opp_table() returned (ptrval)
> >     cpu cpu0: _of_add_opp_table_v1:909: 6 entries
> >     cpu cpu0: dev_pm_opp_get_opp_count:331
> >     cpu cpu0: dev_pm_opp_get_opp_count:333: _find_opp_table() returned (ptrval)
> >     cpu cpu0: dev_pm_opp_get_opp_count:342: _get_opp_count() returned 6
> >     cpu cpu0: dev_pm_opp_get_opp_count:331
> >     cpu cpu0: dev_pm_opp_get_opp_count:333: _find_opp_table() returned (ptrval)
> >     cpu cpu0: dev_pm_opp_get_opp_count:342: _get_opp_count() returned 6
> >     cpu cpu0: dev_pm_opp_get_opp_count:331
> >     cpu cpu0: dev_pm_opp_get_opp_count:333: _find_opp_table() returned (ptrval)
> >     cpu cpu0: dev_pm_opp_get_opp_count:342: _get_opp_count() returned 6
> >     cpu cpu0: Couldn't find proper 'dynamic-power-coefficient' in DT
> >     cpu cpu0: Couldn't register Energy Model -22
> >
> > This happens quite late in the boot sequence, long after cpu1 has been
> > brought online.
> > So it finds the v1 opp table for cpu0, which has 6 entries.
> > The last two messages should be harmless, right?
>
> Yes.
>
> > So you say cpufreq is not working? How can I verify that?
>
> I said it because your earlier logs showed that we defered probed
> again or the count was 0 and we failed. Something like that.
>
> Give output of this to verify if cpufreq is working or not:
>
> grep . /sys/devices/system/cpu/cpufreq/policy*/*
>
> This will be empty if there is no cpufreq.

/sys/devices/system/cpu/cpufreq/policy0/affected_cpus:0 1
/sys/devices/system/cpu/cpufreq/policy0/cpuinfo_cur_freq:375000
/sys/devices/system/cpu/cpufreq/policy0/cpuinfo_max_freq:1500000
/sys/devices/system/cpu/cpufreq/policy0/cpuinfo_min_freq:375000
/sys/devices/system/cpu/cpufreq/policy0/cpuinfo_transition_latency:300000
/sys/devices/system/cpu/cpufreq/policy0/related_cpus:0 1
/sys/devices/system/cpu/cpufreq/policy0/scaling_available_frequencies:375000
750000 937500 1125000 1312500 1500000
/sys/devices/system/cpu/cpufreq/policy0/scaling_available_governors:conservative
ondemand userspace powersave performance schedutil
/sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq:375000
/sys/devices/system/cpu/cpufreq/policy0/scaling_driver:cpufreq-dt
/sys/devices/system/cpu/cpufreq/policy0/scaling_governor:schedutil
/sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq:1500000
/sys/devices/system/cpu/cpufreq/policy0/scaling_min_freq:375000
/sys/devices/system/cpu/cpufreq/policy0/scaling_setspeed:<unsupported>

So it works in v5.9, but not in v5.10-rc1.

Bisection says it was broken by commit 90d46d71cce279d8 ("opp: Handle
multiple calls for same OPP table in _of_add_opp_table_v1()").

> >     cpu cpu0: dev_pm_opp_get_opp_count:331
> >     cpu cpu0: dev_pm_opp_get_opp_count:333: _find_opp_table() returned (ptrval)
> >     cpu cpu0: dev_pm_opp_get_opp_count:342: _get_opp_count() returned 0
> >     cpu cpu0: OPP table can't be empty
> >
> > Wait, _get_opp_count() returns 0?
>
> Does this fix it for you as well ?
>
> https://lore.kernel.org/lkml/2c73ab54717ef358b118ea0cfb727b1427e7730a.1602648719.git.viresh.kumar@xxxxxxxxxx/

Thanks, it does. I had arrived at the same conclusion after bisection.

> I didn't point you to this earlier as your logs said something else.

All my logs said _get_opp_count() returns 0.

> > During s2ram, v5.10-rc1, it redoes most of the above, incl. touching the
> > PMIC, which it shouldn't due in this phase of system resume:
> >
> >     Disabling non-boot CPUs ...
> >     Enabling non-boot CPUs ...
> >     cpufreq_dt: cpufreq_init:112: policy->cpu = 1
> >     cpufreq_dt: cpufreq_init:113: policy->cpus = 1
> >     cpufreq_dt: cpufreq_init:114: policy->related_cpus =
> >     cpufreq_dt: cpufreq_init:115: policy->real_cpus =
> >     of: dev_pm_opp_of_cpumask_add_table:1075
> >     of: dev_pm_opp_of_cpumask_add_table:1080: cpu 0
> >     cpu cpu0: dev_pm_opp_of_add_table:980
> >     cpu cpu0: dev_pm_opp_of_add_table:982:
> > dev_pm_opp_get_opp_table_indexed() returned f680980b
> >     cpu cpu0: _of_add_opp_table_v1:914
> >     cpu cpu0: _of_add_opp_table_v1:916: _find_opp_table() returned a4afd426
> >     cpu cpu0: _of_add_opp_table_v1:937: 6 entries
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >
> > The i2c controller is suspended, this could go boom...
> >
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     of: dev_pm_opp_of_cpumask_add_table:1080: cpu 1
> >     cpu cpu1: dev_pm_opp_of_add_table:980
> >     cpu cpu1: dev_pm_opp_of_add_table:982:
> > dev_pm_opp_get_opp_table_indexed() returned f680980b
> >     cpu cpu1: _of_add_opp_table_v1:914
> >     cpu cpu1: _of_add_opp_table_v1:916: _find_opp_table() returned 9087c76d
> >     cpu cpu1: _of_add_opp_table_v1:937: 6 entries
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     i2c-sh_mobile e60b0000.i2c: Transfer request timed out
> >     cpu cpu0: dev_pm_opp_get_opp_count:331
> >     cpu cpu0: dev_pm_opp_get_opp_count:333: _find_opp_table() returned f680980b
> >     cpu cpu0: dev_pm_opp_get_opp_count:342: _get_opp_count() returned 0
> >     cpu cpu0: OPP table can't be empty
> >     CPU1 is up
>
> Lets make the normal boot work first and see about this later.

This is also fixed by your patch: the PMIC is no longer accessed while
suspended.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds