Add ML_ACC parameters agreed in ag/14877105 as a powerhint
Test: pushed powerhint.json to device, sent ML_ACC powerhint, observed a
change to /sys/kernel/vendor_sched/nnapi_uclamp_min
Bug: 191241561
Change-Id: Ic974cd0b3bbcb98532940520cfd2684d156b9330
The property should be based on device max refresh rate.
Move to raviole config files.
Bug: 177493042
Test: build
Change-Id: I34c42c87e92208e9f8140197e00f713aafaa63ba
Bug: 176195960
Test: no Permission denied while accessing the file node
Signed-off-by: Jack Wu <wjack@google.com>
Change-Id: Ia70c110de8bc1b6166c57e9c3df3d4f43f23a428
Kernel modules on the blocklist are not automatically loaded during
second stage init. Modules are often put on the blocklist if we want
them to get loaded only under certain circumstances.
Bug: 190652328
Change-Id: Ib67463555f498f4dc85d5d175e0bf9e4150771dd
The cancelling behavior in MTP has two behaviors.
1. cancel requests one by one
2. batch cancel all requests
This patch will set the property to true to enable
batch cancel all requests.
Bug: 181729410
Test: build pass
Change-Id: I8d92cc676ff7be8fd22a574b32653bafc028b801
Now HWUI app is using ADPF already, for those non-HWUI app (probably
mostly games), let us try to just use FPS based boosting only.
Bug: 177493042
Test: build
Signed-off-by: Wei Wang <wvw@google.com>
Change-Id: I63fe7ff6edcb57819461a8c0ed5f3f43c26e0137
When display and SF falls out of sync, there is a possibility that while
display holds on to a buffer, a composed buffer could be waiting in the
decon thread. Since maxFrameBufferAcquiredBuffers is 2, GPU will have to
wait for display to swap out the buffer it is currently presenting to be
able to compose for the next frame, which leads to SF missing its
timeline for subsequent frames.
Bug: 190842189
Test: perfetto
Signed-off-by: Devika Krishnadas <kdevika@google.com>
Change-Id: I434e8644cff2538589c1c9b2bf731afae0aa8980
Previously, we copied all available kernel modules into the vendor_boot
ramdisk except for approx. 14 modules that we explicitly excluded.
Going forward, the kernel build will distribute the two additional files
vendor_boot.modules.load and vendor_dlkm.modules.load which define,
respectively, the lists of modules that should be loaded from
vendor_boot and vendor_dlkm.
The contents of the two *.modules.load files will be copied almost
verbatim into modules.load files in the respective images except for the
fact that the directory names of the .ko files are stripped. So, for
example, kernel/net/core/pktgen.ko becomes just pktgen.ko.
Additionally, we only copy those .ko files into the vendor_boot image
that are listed in vendor_boot.modules.load. On the other hand, the
vendor_dlkm partition image will contain all .ko files regardless of
whether they are listed in vendor_dlkm.modules.load.
Bug: 190652328
Change-Id: Ibda5b9368547ad089575333e2566697ec0e8465f
android.software.verified_boot.xml should be copied to TARGET_COPY_OUT_VENDOR instead of TARGET_COPY_OUT_PRODUCT
Bug: 190401885
Test: run cts-on-gsi -m CtsSecurityTestCases -t android.security.cts.VerifiedBootTest#testVerifiedBootSupport
Change-Id: Id17176f07e6ae8843f71b10ad9d01b18640fe675
Integrated CHPP- chpp-20210524-39bf8f5.zip
Adjust the noise table for GPS
False L5 SNR removed by adding valid peak check to L5 CN0
Fixed for SyncIn request to time out within 3 seconds
Support multiple measurement requests
Estimated better accuracy uncertainty in GLNS Only fix
Request UTC time from framework even if the request type is BREADCRUMB
Fixed Reset checking timer for default value of CpResetInvalidateSec to properly work
Bug: 188843280
Bug: 189793458
Bug: 188857474
Bug: 188127870
Bug: 188737899
Bug: 185200717
Bug: 184770261
Bug: 185884684
Sanity test: 190960119
Test: build pass
Change-Id: I969b3ce240583d726f28e93da76c74f1c369a505
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
Specify ownership to system and r/w permissions to user/group.
Allows dumpstate to access /dev/trusty-log0.
Bug: 188285071
Test: lshal debug android.hardware.dumpstate@1.1::IDumpstateDevice/default trusty
Change-Id: I935cb9dd11d1e0968ab0e2c9d7349de336502e7c
The cancelling behavior in MTP has two behaviors.
1. cancel requests one by one
2. batch cancel all requests
This patch will set the property to true to enable
batch cancel all requests.
Bug: 181729410
Test: build pass
Change-Id: Iad744e27e6f843ae22fc499efbb803be51e1b820
Bug: 174189876
Test: check bugreport for #19 data.
Signed-off-by: Super Liu <supercjliu@google.com>
Change-Id: I288d8079528f99834a1867ab973535d4e7b383b6
Add ADPF_DISABLE_TA_BOOST hint for masking TA boost when ADPF is active
Test: build, boot to home
Test: adb shell cat /sys/kernel/vendor_sched/ta_uclamp_min
Bug: 177493042
Change-Id: I6bbeebf80fe4ba7b1e957cde5ed0d4202a2da818
This will enable zram and Pixel common config
Test: confirmed first writeback work via dumpsys jobscheduler | grep Zram
check watermark_boost_factor
Bug: 190571517
Bug: 189938926
Bug: 190732106
Signed-off-by: Martin Liu <liumartin@google.com>
Change-Id: I1a475ea5481d21c7f9fa14eb5f01c6d6802def72
Test: Confirmed the initial write working with console log
Bug: 190571517
Signed-off-by: Minchan Kim <minchan@google.com>
Change-Id: Idb9f800455c62251bad333a015385cb5bd5029cc
GCA need intensive processing for camera shot and need to tune the
system performance to speed up the processing time. Adds
GCA_CAMERA_SHOT_{BIGCPU/BIGMIDCPU/ALLCPU} power hints to enable GCA to do
adequate power hint for this situation.
Bug: 187373665
Test: On device with GCA: b/187373665#comment8 with forrest.
Change-Id: Ife75e52a60de4d67369280998b193fe3ca7f8173
This CL configures the first set of experimental parameters for Adaptive
Suspend on targeted devices, including:
* dropping the max backoff from default 60 seconds down to 40
* changing the break-even threshold from 5 seconds (which was just for
collecting data) to 2
* enabling backoff on short suspends
BUG: 171433995
Test: ran on test device and observed suspend behavior
Change-Id: Icebf50aeade713591f1641664df11570d184d352
Boost frequency of CPUBigCluster when running fingerprint algorithm.
Bug: 185893477
Test: Observe from systrace that the CPU frequency is boosted when
running fingerprint algorithm.
Change-Id: I9e8e727a61f9a2679fc2a00d6e2a1dd4a6c9de99
Prefer high cap and prefer idle is to optimize power and
trade off with scheduling latency.
Applying to app side does has no regression on frame drops.
And we can not apply prefer idle to hal process as it impact performance.
Bug: 190127212
Test:
1080p60: 20mW power saving, 59fps no regression
slow-mo 240fps: 239+fps
Change-Id: I548becbd2d6ccac8b576cfe54f9021bbf30b4801
- Modified HEVC decoder error handling which
copy the nearest reference frame to current frame when error is detected.
- Implemented so that SBWC can be off/on during decoding when SBWC is enabled.
- Fix the max I frame size calculation bug when RC mode is CBR_VT. (RC_MODE = 3)
- Modified so that H.264/HEVC decoder can consumed AUD NAL type stream
when AUD is added at the end of frame.
Bug: 176415285
Signed-off-by: Taehwan Kim <t_h.kim@samsung.com>
Change-Id: I6ceb2cdc2544a40dc505772fb50b5bd1ccd720bd
Reason for remove: face enroll fail
Bug: 168654554
Bug: 187759712
Test: check ISP clock freq no change on local device
Change-Id: I7d70c1fdd7e5d067db1941cced69a375b66ebc89
Bug: 190331327
Bug: 190331548
Bug: 189895600
Bug: 190331108
Bug: 182524105
Bug: 183935302
Test: build ROM and check if the modules and sepolicy are still there
Change-Id: Id28ac2094e14051defe0401055d862c6353194f7
