Power off stress test cannot recall the system

[Scott Hu]

Steps to reproduce:
1) Execute "stress/poweroff_30"
2) Run testing

Expected Result:
Can pass Poweroff stress test without any problem.

Actual Result:
System can be powered off, but can not power on.

Other info:
The test would be continued after power on manually.
SKU: FAR-MT-DVT2-C1(M4000)
BIOS: 1.2.4
Manifest: A02

[Pierre Equoy (pieq)]

Run manually the single poweroff job using plainbox:

    plainbox run -i .*power-management/poweroff.*

Got the same error (the desktop powers off but never restarts).

At first I thought it could be a RTC issue, but the fwts-wakealarm test looks fine (see following attachment).

[Pierre Equoy (pieq)]

After analysing the logs from above, think the RTC data look correct. For instance, in my latest run we can see:

----
2016-01-26 14:56:36,775 DEBUG Wakeup timestamp: 1453791456 (Tue 26 Jan 2016 02:57:36 PM )
2016-01-26 14:56:36,776 DEBUG RTC data:
- alrm_time: 06:57:36
- periodic IRQ frequency: 1024
- periodic IRQ enabled: no
- periodic_IRQ: no
- update_IRQ: no
- alarm_IRQ: yes
- BCD: yes
- batt_status: okay
- rtc_date: 2016-01-26
- 24hr: yes
- rtc_time: 06:56:36
- DST_enable: no
- alrm_date: 2016-01-26
- max user IRQ frequency: 64
- HPET_emulated: yes
- periodic_freq: 1024
- alrm_pending: no
- update IRQ enabled: no
----

The alarm timestamp has been correctly set 60 seconds after current time (2016-0126 06:56:36 UTC), but nothing happens at that time.

[Jerry Kao (jerry.kao)]

Alex's comment in LP:1161711

RTC's wakeup capability depends on hardware design - from the chipset and from the motherboard. It is possible that chipset supports this function (as most does) but motherboard won't allow it.

See the case study @ http://alexhungdmz.blogspot.tw/2012/04/rtc-fails-to-wake-up-system-from-s4-in.html

If all RTC registers and ACPI registers are configured but a system won't wake up from RTC, it is probably designed as it is except the the case in the above link.

It is, however, necessary to understand that ACPI defines RTC can wakeup a system from a sleeping state, and "the G2/S5 soft off ... are not sleeping states." (more details in 4.8.2.4 Real Time Clock Alarm)

In short, RTC should not wake up a system from S5 at all according to ACPI spec, although it apparently not how Linux is designed.

If this feature is really needed, it is necessary to talk to ODM/OEM to check the hardware/firmware designs.

*We may need to find other way to test S4/S5 stress

[Jerry Kao (jerry.kao)]

We need a reliable method to test power-off stress as it is/will be in whitelist. Raise importance as critical.

[Kent Lin (kent-jclin)]

@ Pierre and Chirs,
Please works with HWE team to find a correct and reliable method to test Power-Off stress if certification team decide to include it into whitelist.

Base on comment #5, RTC should not wake up a system from S5 at all according to ACPI spec. Which means the current method we use to wake up system from S5 is not right method.

[Sylvain Pineau (sylvain-pineau)]

I'm not sure if WoL is widely supported on new systems (an laptops), and even activated by default when supported. But it could be seen as an alternative.

[Jerry Kao (jerry.kao)]

Did not see this issue recently. Set to wishlist. Will raise importance if anyone see this issue again.

[Jerry Kao (jerry.kao)]

Danny confirms this issue still be seen from time to time. On failed platform, issue does not happen on all power off rounds. It fails once but success to boot up in all rest of rounds.

[Nara Huang (narahuang)]

I also encountered this issue.
Version:
checkbox-ng 0.35.0~ppa~ubuntu16.04.1
plainbox-provider-checkbox 0.41.0~ppa~ubuntu16.04.1

[Jerry Kao (jerry.kao)]

I was thinking this issue maybe due to the boot up message sent out before system shutdown completed.

The comment for poweroff stress test is
pm_test --checkbox-respawn-cmd $PLAINBOX_SESSION_SHARE/__respawn_checkbox --r 30 --silent --log-level=notset poweroff

[Jerry Kao (jerry.kao)]

@Scott, please add an argument -w 150 to test comment to extend system wake up time. Then rerun poweroff stress test on fail unit.

pm_test --checkbox-respawn-cmd $PLAINBOX_SESSION_SHARE/__respawn_checkbox --r 30 -w 150 --silent --log-level=notset poweroff

file
/usr/share/plainbox-provider-checkbox/units/stress/jobs.pxu

job
plugin: shell
category_id: com.canonical.plainbox::stress
id: stress/poweroff_30
requires:
 package.name == 'fwts'
 executable.name == 'x-terminal-emulator'
command: pm_test --checkbox-respawn-cmd $PLAINBOX_SESSION_SHARE/__respawn_checkbox --r 30 --silent --log-level=notset poweroff --log-dir=$PLAINBOX_SESSION_SHARE
flags: noreturn autorestart
estimated_duration: 3600
user: root
environ: PLAINBOX_SESSION_SHARE
_description:
 Stress poweroff system (30 cycles)

[Jerry Kao (jerry.kao)]

default wakeup time is 60 seconds

[Scott Hu (huntu207)]

Verified passed on Armani MLK after adding an argument -w 150, please check below for the detail

BIOS: 99.5.0
Manifest: X19
SKU:
AMC-DVT1-C2 (power off 30 times: pass, 300 times: pass)
AMC-DVT1-C3 (power off 30 times: pass, 300 times: pass)