Replaced with sched qos profiles.
Bug: 355612114
Flag: NONE vendor task_profiles.json does not support flag
Test: build pass
Change-Id: Ib3ade136be980afeed5c63fbc88bac8cdcc2a4c5
instead of using ADPF, let's use task_profile to boost input tasks'
priroity
Bug: 378761104
Flag: EXEMPT new task_profile behavior
Change-Id: Ice311b1abcf3d629917e16f1490d4cb63fd83796
Signed-off-by: Chungkai Mei <chungkai@google.com>
Add task profiles and set permissions.
Bug: 355612114
Flag: NONE init rc file does not support flag
Test: build pass
Change-Id: I44ab2097bbf919c1487d40ee68e74100c11b8891
Bug: 347122505
Test: check the priority and latency of input thread in Perffeto
Flag: NONE early device setting updates
Change-Id: I0c0a39e3fc03d8df4e9e87cf61fb8e1dafcfc996
Vendor sched procfs path hierarchy gets updated in kernel.
Update the paths in init rc file and task profiles to match the change.
Bug: 289151587
Test: UIbench ab test
Change-Id: I6f8aa26587578860483d6ededf88e9c04a926644
With vendor group throttling, we could relax the ota cpuset.
Bug: 201260585
Test: cpuset changed as expected
Change-Id: I1c5ce6f96fca9f17db9aea9271d89f99ecd30fce
Add ProcFilePath for vendor group change of a process.
Bug: 218684257
Test: function works
Change-Id: I2d87a1a5d27dee699b37de6fac8d8b52334ba54e
(cherry picked from commit 72b0f768ee)
Previous setting is based on cpu shares, which is disabled first, so
we need to change the setting.
Bug: 218420168
Test: build pass
Change-Id: I5528fdccb77f82fc6b4b463476b85aff6b72dbc3
Use a new vendor group, ota, for OtaProfiles, and set its uclamp.max
and group_throttle to 40%, which benefits both performance and power
of OTA process.
Bug: 199006579
Test: value set as expected
Change-Id: I63375f3a286f27ab6c54413d6488b175305e6607
Bg groups are disabled on sc-v2 first, re-enable them on master for
more evaluation.
Bug: 171740453
Test: check bg groups are active
Change-Id: I6bfbbd79fa898cde99d3ba31ff5766de0e4b4c7a
Given the current suspects that the bg throttling may cause extra longer
runnable in P21, remove its usage.
Bug: 208895940
Test: book and check cfs group
Signed-off-by: Wei Wang <wvw@google.com>
Change-Id: If7f20646e944eab8d42e6698e08069b608a0ea7a
high-group is running on CPU 6-7.
mid-group is running on CPU 4-5.
mid-high-gropu is running on CPU 4-7.
Move camera tasks to CPU 4-7 for 1 seconds when AP launch.
Bug: 194271751
Test: GCA
Change-Id: I8e13e9ad0057b8f42f8ed3958bf0c6d58a5a02bc
Add ResetUclampGrp and NoResetUclampGrp taskprofiles to manage vendor
uclamp_fork_set flag.
Bug: 191973176
Test: build and manual test
Change-Id: Ia9c5677c7d5855bb874a714678246ee06c9471bb
We have seen cases SF main thread running longer occasionally which may
run in danger of missing frames. This CL lifts cpufreq for SF when touch
boost is active to compensate the sudden jump of its cpu demand.
Bug: 192099706
Test: Build
Signed-off-by: Wei Wang <wvw@google.com>
Change-Id: I49b2d589249f60a38eab907afbadfbcfdd233d1c
SurfaceFlinger IPC improved with mid and big cores, relaxing cpuset will
help improve its efficiency. For render engine, we set uclamp.min to
skip scheduling on little cores.
FWIW, we let the RT scheduler figure out what is good for perf/power.
Bug: 170507315
Bug: 191054286
Test: Build and boot
Signed-off-by: Wei Wang <wvw@google.com>
Change-Id: I880cf549e33a3a2f629e8260366c4ce8c34c3ae6
NormalPerformance is for system, and ServicePerformance is for
system-background.
Bug: 182509410
Test: build pass
Change-Id: Ifeec11de42fa880cd00b34d504bf6d8ff90dd15c
This reverts commit 637fc9e8d727121276cf8dea2976ac52f8da8d15.
Reason for revert: Camera EBUF overflow observed after launching GCA.
Bug: 187760882
Change-Id: I2357d943f4b4d96ffa7a156671ac7c790640a5ed