Re: [PATCH] ACPI / LPSS: Don't skip late system PM ops for hibernate on BYT/CHT

[Hans de Goede]

Hi Robert,

On 09-05-19 20:09, Robert R. Howell wrote:
> Hi Hans
>
> On 5/9/19 2:50 AM, Hans de Goede wrote:
>
> > Hi,
> >
> > On 09-05-19 06:24, Robert R. Howell wrote:
> > > On 4/30/19 8:39 AM, Hans de Goede wrote:
> > > >
>
> > > > I've just tried to reproduce the "Error i2c_dw_xfer call while suspended" error
> > > > on suspend/resume on my own T100TA and I could not reproduce this.
> > > >
> > > > Can you try without the BT keyboard paired and waking up from suspend using the
> > > > tablet part's power-button ?
> > > >
> > > > Also do you still have the scripts to rmmod some modules before suspend ?
> > >
> > > The T100TA keyboard is actually a hardwired connection rather than Bluetooth but I
> > > did physically disconnect the keyboard, and also unpaired all the actual Bluetooth
> > > devices (such as the mouse) and then powered down the T100TA bluetooth adapter.
> > > When I suspend, then resume using the tablet power button, I still get the
> > > i2c_dw_xfererror error during the resume. But whatever causes this error isn't fatal,
> > > in the sense that after resume the sound and other i2c functions do still work OK.
> > >
> > > While I always get this i2c_dw_xfer error on resume from suspend or hibernation on the T100TA,
> > > I also have a T100TAM and curiously, it NEVER shows that error -- although all the
> > > other suspend and hibernate behavior seems similar. I'm not sure if the following could
> > > be the difference, but the T100TA uses an i2c connected ATML1000 touchscreen controller
> > > while the T100TAM uses an i2c connected SIS0817 touchscreen controller. Other than that
> > > the hardware seems almost identical.
> >
> > I've been testing on an actual T100TA, with the ATML1000 touchscreen controller.
> >
> > Maybe it is a difference in BIOS version, my T100TA is running the latest BIOS, what
> > is the output of:
> >
> > cat /sys/class/dmi/id/bios_version /sys/class/dmi/id/bios_date
> On the T100TA which shows the i2c_dw_xfer error the above cat reports:
> T100TA.307
> 05/09/2014
>
> while the T100TA which does NOT show the i2c_dw_xfer error reports:
> T100TAM.205
> 07/25/2014
> > Also do you perhaps have a microsd card inserted? (I'm trying to figure out the
> > different between our setups so that I can hopefully reproduce the issue myself).
> I do have a microsd card inserted in both the T100TA and the T100TAM.

Ah, ok I already suspected that and I think that is the difference between our
2 setups. I will try to reproduce the suspend/resume problem again with a microsd
card inserted and mounted.

> > > Regarding scripts, while I do still need a systemd hibernate script which removes the
> > > brcmfmac and the hci_uart (bluetooth related) drivers, I've found that I no longer need
> > > any script for suspend.
> >
> > Ok, so you are not doing any rmmod-s on suspend, right?
> Correct -- I am NOT using a script and am not doing any explicit rmmod's on suspend, just on hibernate.
> > Regards,
> >
> > Hans
>
> All my previous tests were done using a 5.1.0-rc5, 5.1.0-rc3, or earlier kernel.
> But I just compiled the released 5.1.0 kernel and the behavior has changed for the T100TA,
> resulting in a different i2c error and a call trace.  (I still continue to NOT see any
> suspend/resume errors on the T100TAM.)

This is expected we changed / improved the code generating the:
"i2c_designware 80860F41:00: Transfer while suspended"

Warning to include a trace, so that we know which code initiated the transfer,
which, as more or less expected, is the ACPI subsystem, like some power_on (_PS0)
or off (_PS3) method.

> Note that for all the tests described in this message I'm applying your patch
> regarding .poweroff_noirq and .restore_noirq, and I'm applying my own patch removing the
> DPM_FLAG_SMART_SUSPEND flag.  I haven't yet tried to explore varying those patches
> for the 5.1.0 release as I did for the earlier rc's, as described in previous messages.

Hmm, for future testing please leave out the patch removing the DPM_FLAG_SMART_SUSPEND
flag. Usually when asking you to test something we assume you are using a pristine kernel.
What does that patch attempt to fix and what happens during suspend/resume without it ?

Regards,

Hans