We now compute radio active time per application, by distributing
the active time across all applications each time the radio goes
down, weighting it by the number of packets transferred.
Per-app radio power use is now computed using this radio active
time.
This also gives us a new metric "ms per packet", which give an
idea of how effectively an application is using the radio. This
is collected and reported as a new set of stats in the human-
readable checkin. (It can be computed from the raw checkin data).
Also improve sync reporting to include the sync source as used
in wake locks, not just the component name.
Change-Id: I0b0185fadd1e47ae749090ed36728ab78ac24c5e
- Add events for sync.
- Add more descriptive tags for wake events.
- Fix battery reset.
- Fix tracking of wifi data.
Change-Id: Ic07f2a86a5ed33e7da57eb1108c31c777ecd801f
Hide managed profiles from lockscreen user switcher on tablets.
Hide managed profiles from power menu user switcher on phones.
Add flag to enable multi user ui turned off by default.
Change-Id: I4c69a6f7b0f39c249fc85fd940318df1ddab073f
* commit '5f652b9fdfbcc279353955f7ef86b72d2ef9f5fb':
open("/dev/rtc0") failure in AlarmManagerService.setTime() should be non-fatal
Move time setting code from SystemClock to AlarmManagerService
* changes:
open("/dev/rtc0") failure in AlarmManagerService.setTime() should be non-fatal
Move time setting code from SystemClock to AlarmManagerService
On devices using /dev/rtc instead of /dev/alarm, updating the
time-of-day clock and RTC are separate syscalls. Hence the clock and
RTC could be left in inconsistent states if two threads called
SystemClock.setCurrentTimeMillis() simultaneously.
By moving this code into AlarmManagerService, we can put a global lock
around AlarmManagerService.setTime() and prevent the race condition.
Note that access to SystemClock.setCurrentTimeMillis() is now gated by
android.permission.SET_TIME, where before it was gated by filesystem
permissions (i.e., could the process write to /dev/alarm or /dev/rtc).
Change-Id: Ia34899a4cde983656305fd2ef466dfe908ed23c8
Signed-off-by: Greg Hackmann <ghackmann@google.com>
When a doze component has been specified in a config.xml resource
overlay, the power manager will try to start a preconfigured dream
whenever it would have otherwise gone to sleep and turned the
screen off. The dream should render whatever it intends to show
then call startDozing() to tell the power manager to put the display
into a low power "doze" state and allow the application processor
to be suspended. The dream may wake up periodically using the
alarm manager or other features to update the contents of the display.
Added several new config.xml resources related to dreams and dozing.
In particular for dozing there are two new resources that pertain to
decoupling auto-suspend mode and interactive mode from the display
state. This is a requirement to enable the application processor
and other components to be suspended while dozing. Most devices
do not support these features today.
Consolidated the power manager's NAPPING and DREAMING states into one
to simplify the logic. The NAPPING state was mostly superfluous
and simply indicated that the power manager should attempt to start
a new dream. This state is now tracked in the mSandmanSummoned field.
Added a new DOZING state which is analoguous to DREAMING. The normal
state transition is now: AWAKE -> DREAMING -> DOZING -> ASLEEP.
The PowerManager.goToSleep() method now enters the DOZING state instead
of immediately going to sleep.
While in the doze state, the screen remains on. However, we actually
tell the rest of the system that the screen is off. This is somewhat
unfortunate but much of the system makes inappropriate assumptions
about what it means for the screen to be on or off. In particular,
screen on is usually taken to indicate an interactive state where
the user is present but that's not at all true for dozing (and is
only sometimes true while dreaming). We will probably need to add
some more precise externally visible states at some point.
The DozeHardware interface encapsulates a generic microcontroller
interface to allow a doze dream for off-loading rendering or other
functions while dozing. If the device possesses an MCU HAL for dozing
then it is exposed to the DreamService here.
Removed a number of catch blocks in DreamService that caught Throwable
and attempted to cause the dream to finish itself. We actually just
want to let the process crash. Cleanup will happen automatically if
needed. Catching these exceptions results in mysterious undefined
behavior and broken dreams.
Bug: 12494706
Change-Id: Ie78336b37dde7250d1ce65b3d367879e3bfb2b8b
Add battery stats tracking of wifi state.
Also update when we retrieve the current time to use a more
consistent value across stats tracking.
Change-Id: I6a7c3efd58ff2c8ea86dac141c8f848e7996d63f
We now always turn on network state tracking for mobile,
and push this information down to battery stats.
In battery stats we use this to both log the changes in
the history and keep track of the total time the mobile
radio was active.
Power computation is switched over to using this information
to help determine power use, which will hopefully make it
more accurate (not counting inaccuracies in knowing when it
actually goes down).
Note yet done is aggregating this data per-uid, to better
emphasize which apps are causing the radio to be up. Right
now we just spread the total time across all uids weighted
by the total number of packets they have sent and received.
Also put in the battery stats infrastructure for bluetooth to
address issue #12973036: Improve power_profile.xml
Change-Id: I39d11b7ff6ae4f336f253d1cba308d8569de7e0d
Java plumbing to expose methods to get/set encryption type in
IMountService, and hooking up of those methods to the Settings app
so the type is set correctly.
Needs matching vold changes from
https://googleplex-android-review.googlesource.com/#/c/412649/
Bug: 8769627
Change-Id: I70c0ed72d11f5ab6f0958a7f9c101b6822b13baa
Treat arrays other than actual Object arrays (i.e. those whose component type
is actually Object) and the primitive arrays already handled specially by
Parcel as Serializable's.
Issue: 64583
Change-Id: I3ff797f3262e77e4de27f35709bceee5410d1ed4
Related users are a group that will share things like notifications.
Managed profiles are a profile of another user.
Change-Id: I2d0532f1abf939810f0fa3fc7c77ad13fa567833
Add new history events for top application package and
foreground application packages.
Doing this involved a fair amount of improvement to history
events. The event code is now separated out to have "start"
and "finish" identifies, and we use that to now keep track
of which events are active. With that, when resetting the
stats, we can spit out all of the currently active events at
the front of the new history.
Also fixed some problems when I re-arranged the history delta
int bits that were conflicting with the packing of the battery
status bits. These packing structures are changed to work
together correctly.
Change-Id: Ic8b815060dd8a50ff4a0a209efc2e1044215cd88
This allows DownloadManager to use FDs, paving the way for downloading
directly to content:// Uris.
Also return flag indicating if deleteOlderFiles() actually deleted
anything. Update tests to verify.
Bug: 5287571
Change-Id: I2579e5e2113f31b2860d7b021bd61c91b6310963
We now try to have a better label for the first wake lock
that is acquired in the log. This is done in two ways:
- The alarm manager now sorts the alarms it is going to
execute so that wakeup alarms are first, which are more
important w.r.t. which one should be logged.
- There is a new power manager facility to make a wake lock
as "unimportant for logging," which just means in battery
stats that a wake lock acquired after that can be considered
the actual one to log. This is only used by the alarm manager
to mark its TIME_TICK alarms as unimportant for logging.
Also reworked the battery history code to be cleaner and a bit
smaller. There is no longer a separate EVENT command, instead
the event code and tag are just another thing that can be included
in an UPDATE command.
The bits used in the first history int are also re-arrange, so
that only the ones that really change a fair amount in the state
bits are up at the top and there is no longer space used for
the command code (since now it is always just UPDATE). This
allows us to have more room for the time delta at the bottom,
to better avoid situations where we need to write an int delta.
Change-Id: I1bb860ae5b558a248800b090b03a84fbf7acd68a
- Better batching of history items. Fixed problems where empty
entries would be created because state toggles got lost.
- The string pool is now a HistoryTag pool, containing both a string
and uid; now an entry only requires 16 bits in the history data.
- Acquiring the first wake lock also now includes a HistoryTag
identifying who did the aquisition.
- Cleaned up printing of signal strengths and cell radio types.
- There was code that tried to allow you to add new history entries
while iterating the history... but these should never happen
together, so turned that into a failure... and fixed an issue
where we could leave the battery stats in a state where it
thinks it is continually iterating.
Change-Id: I1afa57ee2d66b186932c502dbdd633cdd4aed353
The method handleMessage(Message msg) from mHandler variable was
not checking if the timer was cancelled, so
sendMessageDelayed(obtainMessage(MSG), delay) was keeping the
timer alive. The patch simply adds a boolean and checks if the
CountDownTimer was cancelled before calling
sendMessageDelayed(obtainMessage(MSG), delay)
bug: https://code.google.com/p/android/issues/detail?id=58668
Change-Id: Ic6bbb9d33a3616f8503db222513cc14ad2270cb8
Signed-off-by: jl1990 <jlcarrasco1990@gmail.com>
Changes include
[x] Long is used to store native pointers
[x] Added new method obtain(long obj) to Parcel. Binder
uses this method instead of obtain(int obj).
[x] obtain(int) has been changed to throw unsupported
operation exception.
Change-Id: I408e0f2a24deb28c9277d86670653a51eb314266
Signed-off-by: Ashok Bhat <ashok.bhat@arm.com>
Signed-off-by: Craig Barber <craig.barber@arm.com>
Signed-off-by: Kévin PETIT <kevin.petit@arm.com>
This will break on 64 bit because 64 bit pointers are
wider than java int.
Also, this method technically isn't a public API because
it's a protected method on a final class. It shows up in
our api.txt files because of a quirk in the API generation
system.
Change-Id: I2b574943008ed34696b048195c0cac6af1aa0a3b
- Adjust total power use when there is unaccounted power so that our
percentages don't end up > 100%.
- Fix accounting of isolated uids to be against the owning real app
uids.
- Rework how we put cpu use into the battery stats to no longer need
this uid name cache that can confuse the uid it is associated with.
- Finish implementing events in the history, adding a string pool and
reading/writing/dumping them.
- Add first two events: processes starting and finishing.
- Fix alarm manager reporting of wakeup alarms to be adjusted by the
WorkSource associated with the alarm, so they are blamed on the
correct app.
- New "--history" dump option allows you to perform a checkin of
only the history data.
- Fixed BitDescription bug that would cause incorrect printing of
changes in some states.
Change-Id: Ifbdd0740132ed178033851c58f165adc0d50f716
Transform DisplayManagerService into a SystemService and start cleaning
up other local services that it uses from window manager and input manager.
Clean up service thread initialization.
Remove unnecessary static variables from ActivityManagerService.
It's starting to become clear that we really need a better way to manage
service dependencies. Boot phases don't quite cut it.
Change-Id: If319dbd7cbfbd4812fe55ece969e818d4b20755b
Change PowerManagerService to start the pre-recovery service rather
than rebooting directly, when requested to reboot into recovery. Add
a new RECOVERY permission which a caller needs (in addition to REBOOT)
in order to go to recovery.
Bug: 12188746
Change-Id: I39121b701c4724558fe751adfbad79f8567faa43
Otherwise we sometimes end up counting objects strongly referenced
by the finalizer queue, and falsely detecting an instance leak.
Bug: 12533002
Change-Id: I144c941a3dfb0cbb837b98e80d2f49ffc9ca13c7
Refactored SystemServer to get rid of a bunch of legacy cruft related
to how the ServerThread used to be started up.
Create system context first when system server starts. This removes
the tangled initialization order dependency that forced us to start
the activity manager service before most anything else.
Moved factory test related constants into the FactoryTest class.
Partially migrated Installer, ActivityManagerService, and
PowerManagerService to the new SystemService pattern. There's more
work to be done here, particularly around the lifecycle of the
power manager.
Bug: 12172368
Change-Id: Ia527dd56e3b3fd90f9eeb41289dbe044921230d4
