hammerheadcaf: Disable cpu-boost sync_threshold to recover from regressionlp5.1

Load based syncs successfully avoid boosting of lightweight threads.
Unfortunately, cpus with high frequencies over sync_threshold will
be unnaturally throttled which becomes appearant in hackbench with
its high number of thread migrations. None of the newer targets sets
sync_threshold, either.

Pre patch:
root@hammerhead:/ # perf stat --repeat 10 hackbench 10
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.923
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 1.106
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.934
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.917
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.765
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.807
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.930
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.937
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.858
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.808

 Performance counter stats for 'hackbench 10  ' (10 runs):

   3575.129450 task-clock#3.498 CPUs utilized
 ( +-  3.95% )
 42637 context-switches  #0.012 M/sec
 ( +- 12.54% )
  6152 CPU-migrations#0.002 M/sec
 ( +-  9.98% )
 37874 page-faults   #0.011 M/sec
 ( +-  0.02% )
6208354174 cycles#1.737 GHz <--- should not happen!
 ( +-  1.78% ) [90.87%]
 0 stalled-cycles-frontend   #0.00% frontend cycles idle
 ( +-  0.00% ) [88.02%]
 0 stalled-cycles-backend#0.00% backend  cycles idle
 ( +-  0.00% ) [87.13%]
1948037598 instructions  #0.31  insns per cycle
 ( +-  1.53% ) [89.32%]
 203196136 branches  #   56.836 M/sec
 ( +-  1.46% ) [90.70%]
   5345440 branch-misses #2.63% of all branches
 ( +-  4.62% ) [85.52%]

   1.022038466 seconds time elapsed
 ( +-  4.09% )

Post patch:
root@hammerhead:/ # perf stat --repeat 10 hackbench 10
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.735
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.815
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.754
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.721
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.770
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.767
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.762
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.689
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.748
Running in process mode with 10 groups using 40 file descriptors each (== 400 tasks)
Each sender will pass 100 messages of 100 bytes
Time: 0.679

 Performance counter stats for 'hackbench 10  ' (10 runs):

   2838.930889 task-clock#3.343 CPUs utilized
 ( +-  1.81% )
 22301 context-switches  #0.008 M/sec
 ( +- 11.18% )
  3641 CPU-migrations#0.001 M/sec
 ( +- 11.13% )
 37950 page-faults   #0.013 M/sec
 ( +-  0.09% )
5714558403 cycles#2.013 GHz
 ( +-  1.59% ) [91.42%]
 0 stalled-cycles-frontend   #0.00% frontend cycles idle
 ( +-  0.00% ) [89.07%]
 0 stalled-cycles-backend#0.00% backend  cycles idle
 ( +-  0.00% ) [89.74%]
1868219180 instructions  #0.33  insns per cycle
 ( +-  0.82% ) [90.90%]
 193711678 branches  #   68.234 M/sec
 ( +-  1.44% ) [91.41%]
   4927373 branch-misses #2.54% of all branches
 ( +-  3.57% ) [87.20%]

   0.849242812 seconds time elapsed
 ( +-  1.58% )

Change-Id: I8744cc1f96fefa81149ded1c2dc54ff4d3b76665
(cherry picked from commit 8f45a2d638ec7f1511de06ebea84ed96f2b6cde7)

1 files changed, 0 insertions, 1 deletions

diff --git a/init.hammerhead.power.rc b/init.hammerhead.power.rc
index 709ea19..bbd021c 100644
--- a/init.hammerhead.power.rc
+++ b/init.hammerhead.power.rc
@@ -100,7 +100,6 @@ on property:init.svc.bootanim=stopped
 
     # cpu-boost
     write /sys/module/cpu_boost/parameters/boost_ms 20
-    write /sys/module/cpu_boost/parameters/sync_threshold 1728000
     write /sys/module/cpu_boost/parameters/input_boost_freq 1190400
     write /sys/module/cpu_boost/parameters/input_boost_ms 500
     write /sys/module/cpu_boost/parameters/load_based_syncs 1