* Since we have libperfmgr to boost devfreq, there is no need to use it.
* Followed wahoo's set
Change-Id: Ic9a63aac98f315c83335d36f6fdfc1bffc6bf7da
Signed-off-by: Ratoriku <a1063021545@gmail.com>
Google has mentioned that 512 KB performs worse under heavy memory pressure.
Therefore, hardcode 128 KB read-ahead for all which is optimal.
Also, remove the unused block paths.
Signed-off-by: Tashfin Shakeer Rhythm <tashfinshakeerrhythm@gmail.com>
Signed-off-by: pix106 <sbordenave@gmail.com>
CAF's CPU boost has been disabled in kernel. So, this is useless.
Signed-off-by: Tashfin Shakeer Rhythm <tashfinshakeerrhythm@gmail.com>
Signed-off-by: pix106 <sbordenave@gmail.com>
rtg boost and pl have been removed from schedutil in kernel. So, this is useless.
Signed-off-by: Tashfin Shakeer Rhythm <tashfinshakeerrhythm@gmail.com>
Signed-off-by: pix106 <sbordenave@gmail.com>
Set the default IRQ affinity to the primary cluster for SDM660,
SDM630 and SDM845. When a CPU is isolated/hotplugged, the IRQ
affinity is adjusted to one of the CPUs from the default IRQ
affinity mask.
Change-Id: I1f33fdf2a06e1cb086904dabaaa0414aafcf07c5
Signed-off-by: pix106 <sbordenave@gmail.com>
If we set 2048KB ra value in the beginning of the
boot process, services initialized at this time would
keep 2048KB ra value even though we adjust the ra value
to 128KB after boot completion.
This may cause memory pressure.
Bug: 174700194
Signed-off-by: Randall Huang <huangrandall@google.com>
Change-Id: I97ee0ee9511eaada011f2634afd68d68cdca4570
Signed-off-by: pix106 <sbordenave@gmail.com>
This prevents to change the frequency in a very short
manner which seems to fast for the hardware and
causes performance drops.
Change-Id: I678d6dc3dc81d756ba1b2031cd369d95d9298ac0
Signed-off-by: pix106 <sbordenave@gmail.com>
- In my testing, i faced some CPU Stalls with the old setup. (caused some spikes in jitter and non-consistent hackbench).
- 1000 was perfect in my testing and did not hit over power usage.
Signed-off-by: clarencelol <clarencekuiek@icloud.com>
* In favor of Low Memory Killer Daemon (lmkd).
* Adaptive LMK was also dropped quite sometime ago
making it useless to keep this script around.
Change-Id: Ie221d6f669d261bb871a77fed712b0df3f5ba248
* No matter how hard QC tried to improve these features, they still suffer
from stability issues, mainly due to the unpredictable nature of vmpressure
they rely on.
Change-Id: Icd14c79298a3c268abffa06ed17a79dececf423a
* When sys.use_fifo_ui is enabled, ActivityManager tracks the UI thread and RenderThread (the two most UI-critical threads) of the top application and makes those threads SCHED_FIFO instead of SCHED_OTHER. This effectively eliminates jitter from UI and RenderThreads; the traces that've gathered with this option enabled show runnable times on the order of microseconds instead of milliseconds.
* However, it should be enabled only when we have EAS i.e a capacity-aware RT scheduler because if RT load balancer is not capacity-aware, then there would be a 30% reduction in application startup performance.
Reference: https://source.android.com/devices/tech/debug/jank_jitter
Signed-off-by: Saurabh Charde <saurabhchardereal@gmail.com>
Signed-off-by: Aston-Martinn <advaithbhat9@gmail.com>
Change-Id: I8b55ad1899be985fc7357abaf06d68d6a130fb75
Those values seem off. CAF maybe messed them up as those values look possibly wrong.
Those values are used as powerhint values on pixel 2 and 3.
Signed-off-by: Dusan Uveric <dusan.uveric9@gmail.com>
Change-Id: I1eebb1f57a7e91f26b9900cda65ec151b05ecbd7
* No matter how hard QC tried to improve these features, they still suffer
from stability issues, mainly due to the unpredictable nature of vmpressure
they rely on.
Signed-off-by: PIPIPIG233666 <2212848813@qq.com>
Change-Id: I8256abb91b7e3160670ec45a22fdceeecc2a928c
Removing the check of 1GB RAM while enabling
low_ram target specific configurations to enable them
for 2GB RAM targets as well.
Updating ZRAM size to 75% of RAM for <2GB devices
Change-Id: I8d5341a4397cd20f2c59e7396801fcbea7fe5b73
- SnapdragonVoiceActivation feature for Far Field Voice version
affines its threads to 0-3 CPUs and so prevent them from isolation
for qcs605 target
Change-Id: I09ef5cc97ee245133315edc0821f3a2afc54e99b
ZRAM compression uses the linux slub allocator its internal data
structures. If these internal data structure consume a lot of memory,
ZRAM compression may increase memory usage, rather than decrease it
as expected. Decrease the chance of this occuring by ensuring the
SLAB_STORE_USER debug option is disabled.
Change-Id: I4684feda94b7bebacb6218b7f979c95771b0ee10
Signed-off-by: Patrick Daly <pdaly@codeaurora.org>
check populated dm-* nodes and Update read_ahead_kb
values to them. This is needed because on targets
with verity disabled dm-0 to dm-6 are populated,
while on enabled targets dm-3 to dm-6 are populated.
Change-Id: Ia4e262d7487120695db53b562b5764598fd40a87
