Comment 12 for bug 1421455

Looking at the strace output one can see that powerd gets the power button press notification, processes it, and then nothing happens for half a second:

[pid 949] 1430318963.039809 send(3, "<31>Apr 29 16:49:23 powerd[949]: exiting suspend", 48, MSG_NOSIGNAL) = 48
[pid 949] 1430318963.040868 gettimeofday({1430318963, 40921}, NULL) = 0
[pid 949] 1430318963.041289 send(3, "<31>Apr 29 16:49:23 powerd[949]: libsuspend: exit_suspend.", 58, MSG_NOSIGNAL) = 58
[pid 949] 1430318963.042623 open("/sys/power/state", O_WRONLY) = 14
[pid 949] 1430318963.043483 write(14</sys/power/state>, "on", 2 <unfinished ...>

<Note: 508 milliseconds delay>

[pid 8256] 1430318963.515663 <... futex resumed> ) = -1 ETIMEDOUT (Connection timed out)
[pid 8256] 1430318963.515862 clock_gettime(CLOCK_MONOTONIC, {27280, 493365943}) = 0
[pid 8256] 1430318963.517005 clock_gettime(CLOCK_MONOTONIC, {27280, 493965482}) = 0
[pid 8256] 1430318963.518207 futex(0xaa430c, FUTEX_WAIT, 30, {14, 999399518} <unfinished ...>
[pid 971] 1430318963.587966 <... read resumed> "a", 1) = 1
[pid 971] 1430318963.588263 close(7</sys/power/wait_for_fb_wake>) = 0
[pid 971] 1430318963.589699 open("/sys/power/wait_for_fb_sleep", O_RDONLY) = 7
[pid 971] 1430318963.590953 read(7</sys/power/wait_for_fb_sleep>, <unfinished ...>
[pid 949] 1430318963.592034 <... write resumed> ) = 2

If you look close enough you can see that pid 949 does a blocking write of the string "on" to /sys/power/state, and this syscall comes back half a second later. Half a second. Exactly the missing time we are looking for.

So I pressed the power button to turn off the screen, waited a couple of seconds to allow the system go to sleep, and did the following:

root@ubuntu-phablet:~# time echo "on" > /sys/power/state

real 0m0.545s
user 0m0.000s
sys 0m0.000s

Something in the kernel seems to take roughly half a second to wake up.

I looked at the Documentation/power/states.txt file of the Linux kernel and was am a bit confused: There should be only four valid values one can write to /sys/power/state, namely "mem", "standby", "freeze" and "disk". On the bq phone the file contains "mem disk", so only those two states are available. I found out that "on" seems to be an accepted value on Android kernels only.

So the root of the problem my lie in kernel land at the end.