The deadline IO scheduler results in lower latency SD, and pulls in
latency outliers (especially when the flash starts getting fragmented).
Switch to deadline post boot.
Test: various workloads, both synthetic (fio) as well as actual
workloads (app launch, ioshark).
Bug: 63630139
Change-Id: I94b90ac09e921fdf01564bd4c6e1a1b809f3f8ec
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
Walleye requres /sys/class/uio/uio1/device/rtel/rtel in userdebug build,
and GSI will change the device to userdebug. So, we need to switch rtel
with the build type in boot time, even the device is in user build.
Bug: 65229896
Test: Boot (aosp_)walleye-(user|userdebug) with/without GSI to home screen
Change-Id: I8fa74c02a0a48feeef87d8e9caa702e4a44aedf7
Bootloader set androidboot.revision on kernel command line and we set
ro.boot.hardware.revision to a numerical string accordingly.
Bug: 64375827
Change-Id: Ib814f765b53155721d86035405fbc6490fb5a2c3
Signed-off-by: Thierry Strudel <tstrudel@google.com>
Tested 1/2x, 1x, 2x, and 4x the current readahead buffer and noop and
cfq I/O schedulers for /dev/block/sda, which contains /data. The most
consistent performance boost was seen across all UFS vendors and storage
sizes using cfq and a readahead buffer of 1024.
Bug: 38000161
Test: Compared 1P app launch times using app-launcher for schedulers
(noop, cfq) and readahead buffer sizes (128,512,1024,2048)
Change-Id: Ifcc9145d6f7a77ba4b3844b7892291eb348d9e1c
With noop scheduler, I see dramatic jank with an irritant i/o thread;
going back to cfq, which doesn't have nearly the same negative impact
on UX.
Bug: 62375787
Test: boot, scroll a lot
Change-Id: I3c1ad17f432b25b5359f813e070fd1266dac1f87
Disable kernel module loading after boot on user builds.
Bug: 38204366
Test: boot walleye, confirm kernel module does not load after boot
Change-Id: I40217ff37c1b1bcf00155e5bec74b92a020481f7
