Re: [PATCH 0/2] PM / core: genpd fix and pm_runtime_force_suspend|resume() rework

[Geert Uytterhoeven]

Hi Rafael,

On Mon, Jan 15, 2018 at 1:04 AM, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> On Sun, Jan 14, 2018 at 10:48 AM, Geert Uytterhoeven
> <geert@xxxxxxxxxxxxxx> wrote:
>> On Sat, Jan 13, 2018 at 1:38 AM, Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote:
>>> On Friday, January 12, 2018 3:31:09 PM CET Geert Uytterhoeven wrote:
>>>> On Fri, Jan 12, 2018 at 2:00 PM, Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote:
>>>> > This comes from the recent discussion/testing effort that ensued after my
>>>> > pm_runtime_force_suspend|resume() changes proposal:
>>>> >
>>>> > https://marc.info/?t=151497772000004&r=1&w=2
>>>> >
>>>> > Patch [1/2] basically is https://patchwork.kernel.org/patch/10152873/ rebased
>>>> > on top of the current linux-next branch of the linux-pm.git tree (the relevant
>>>> > part should be there in the linux-next tree proper ATM).  It applies on top
>>>> > of https://patchwork.kernel.org/patch/10156077/ which should apply to the Linus'
>>>> > tree cleanly.
>>>> >
>>>> > Patch [2/2] is a resend of https://patchwork.kernel.org/patch/10142047/ with
>>>> > a very minor changelog modification and the R-b tag from Ulf.
>>>> >
>>>> > Geert, if possible, please test this on the Renesas systems that had the
>>>> > problem with https://patchwork.kernel.org/patch/10142047/ previously and let
>>>> > me know if you still see issues.
>>>>
>>>> I've tested this on two very similar systems: Salvator-XS with R-Car H3 ES2.0,
>>>> and Salvator-X with R-Car M3-W ES1.0.
>>>>
>>>> On the M3-based system, everything seems to work fine.
>>>
>>> Good.
>>>
>>>> On the H3-based system, the serial console (the /dev/ttySC0 device, not kernel
>>>> serial output) is dead after resume from s2ram, with and without
>>>> no_console_suspend.
>>>>
>>>> With no_console_suspend, I see:
>>>>
>>>>     ttySC ttySC0: 1 input overrun(s)
>>>>
>>>> after typing on the serial console, so it looks like an interrupt problem.
>>>>
>>>> This issue seems to be caused by patch [1/2]. But I have no idea what's
>>>> really happening, and why the two systems behave differently.
>>
>> Could be a firmware issue, too.
>> While the kernel images are identical, the ARM trusted firmware configs aren't
>> (same version, though).
>>
>> I'll do some more investigation...
>
> OK, thanks!
>
> It also would be good to know the topology of the device hierarchy and
> how that maps to the domains on the failing system (and which UART
> clocks are operated by genpd).

The topology is the same on both systems.
The UART's module clock is operated by genpd, on both systems.

>>> Well, that's not dramatic.
>>>
>>> Let's make a deal that we'll fix this on top of [1/2].
>>
>> ;-)
>>
>>> Which driver is this BTW?  sh-sci?  That one doesn't even support runtime
>>> PM, confusingly enough.
>>
>> Yes, sh-sci. It does make pm_runtime_*() calls.
>
> Hmm.  I overlooked that part.
>
> This is sort of unusual, because the driver doesn't provide any
> runtime PM callbacks, but still it does provided system suspend ones.
> It looks like the idea is to never put it into runtime suspend if any
> ports are enabled and always put it into runtime suspend otherwise.
>
> Which one is the case in your testing?  Is the port disabled or
> enabled during system-wide suspend?

It's enabled on both systems, as a getty is running.

>> And of course there's uart_ops.pm, which is driven from serial_core...
>
> What does this point to for that particular device?

sci_pm(), on both systems.

See, there's no difference in topology on both systems, so I'll have to look
a bit deeper first...

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