Re: Keyboard lost after exit s2idle on ASUS UX433FN

[Chris Chiu]

On Thu, Aug 16, 2018 at 4:54 PM, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> On Thu, Aug 16, 2018 at 10:18 AM Chris Chiu <chiu@xxxxxxxxxxxx> wrote:
>>
>> On Thu, Aug 16, 2018 at 3:26 PM, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
>> > On Thu, Aug 16, 2018 at 4:45 AM Chris Chiu <chiu@xxxxxxxxxxxx> wrote:
>> >>
>> >> Hi,
>> >>     We recently hit a weird problem on the ASUS laptop UX433FN with
>> >> latest Intel Core i7-8565U CPU on kernel 4.18. The keyboard stops
>> >> functioning after exit s2idle. It stops firing interrupts after resume
>> >> on any keypress. We thought it should be something wrong with i8042
>> >> driver or even atkbd driver, so we tried to skip the suspend/resume
>> >> path of i8042 and input devices but no luck.
>> >>
>> >>     Then we tried to hack the s2idle code to fail right before it goes
>> >> into the idle state to find out which code really cause the keyboard
>> >> broken. It comes with an interesting finding that if it aborts s2idle
>> >> before cpuidle_resume() in s2idle_enter(), the keyboard is fine. If it
>> >> aborts after cpuidle_resume, then the keyboard down. At least it
>> >> proves that even with dpm_noirq_begin() and
>> >> dpm_noirq_suspend_devices() executed, the keyboard is still alive.
>> >> There should be something wrong with the cpuidle.
>> >>
>> >>     Going deeper into intel_idle_s2idle() which is invoked by
>> >> cpuidle_enter_s2idle(), we found that the keyboard interrupt will no
>> >> longer function after mwait_idle_with_hints() which just simply
>> >> executing intel monitor and mwait instructions. So I don't know what
>> >> should be the next step I can take. Can anyone give some pieces of
>> >> advice?
>> >
>> > It may indicate that the deepest C-states used by s2idle simply don't
>> > work correctly on the affected system.
>> >
>> > To verify this, you can try to disable the deepest C-states via sysfs
>> > using the "disable" attribute under
>> > /sys/devices/system/cpu/cpuX/cpuidle/stateX/
>> >
>> Thanks for the great help. I can have the keyboard work after s2idle with
>> the following command
>>
>> echo 1 > /sys/devices/system/cpu/cpuX/cpuidle/state8/disable
>>
>> cpuX, X can be 0 to 7 on this CPU.
>> And only state8 disable would work, it fails on state 0~7 disabled.
>
> Well, if you disable one of states 0-7 alone, it will still attempt to
> use state8. :-)
>
> It generally uses the deepest one enabled.
>
>> There're 9 (0-8) states on the cpuidle of this machine. However, I have another
>> laptop with the same CPU Intel Core i7-8565U which only have 4(0-3) states
>> under the same '/sys/devices/system/cpu/cpuX/cpuidle/' entry. Sorry that I
>> don't have enough background knowledge about cpuidle. I thought the #C
>> state should be depend on CPU, but why the same i7-8565 has different
>> number of C-states?
>
> What is there in /sys/devices/system/cpu/cpuidle/current_driver on
> each of these systems?
>
>>  And for this case, should I just disable C-8 state or there
>> should be a generic fix with it?
>
> No, it just doesn't work specifically on your system, unfortunately,
> so you need to disable state8 manually, at least for the time being.
> If all UX433FNs turn out to be affected, we may consider adding a
> quirk for them, but for now we just don't know.
>
> Do the "name" and "desc" sysfs files of state8 on the affected system
> contain "C10" and "MWAIT 0x60", respectively?
>
> To make a more persistent record of this issue, you can file a bug at
> bugzilla.kernel.org against cpuidle and put all of the relevant
> information into it.  It is possible that some special magic is needed
> to make the deepest idle state safe on this system and it may be hard
> to find out, but at least it is good to know that this happens.
>

Thanks. I've reported on https://bugzilla.kernel.org/show_bug.cgi?id=200829.
Also, include information you request. Please let me know if anything missing.

Chris

>> > Is s2idle the default suspend method on that system?  If so, have you
>> > checked whether or not suspend-to-RAM works too?
>> >
>>
>> Yes, it's default s2idle. If I do 'echo mem > /sys/power/state', the keyboard
>> would still fail.
>
> OK
>
> What's there in /sys/power/mem_sleep ?
>
> Thanks,
> Rafael