introduce binderRpcBenchmark
VRRRRRRRRRRRRRRRRRMMMMMMM!!!
Some caveats:
- we are partially benchmarking the scheduler
- this is a microbenchmark
Results running on host (we'll want to watch and tune these on real
devices with core clamping and thermal throttling off):
-------------------------------------------------------------
Benchmark Time CPU Iterations
-------------------------------------------------------------
BM_getRootObject 45959 ns 39054 ns 18345
BM_pingTransaction 26128 ns 23360 ns 28705
BM_repeatString 394286 ns 200186 ns 3508
BM_repeatBinder 39357 ns 37437 ns 17705
Bug: 182940634
Test: run benchmark on host and on cuttlefish
Change-Id: I5084a00ddf207eaeb803da98587ca04df65add4d
diff --git a/libs/binder/RpcConnection.cpp b/libs/binder/RpcConnection.cpp
index 1bf3d88..c1ed93a 100644
--- a/libs/binder/RpcConnection.cpp
+++ b/libs/binder/RpcConnection.cpp
@@ -61,7 +61,8 @@
public:
explicit UnixSocketAddress(const char* path) : mAddr({.sun_family = AF_UNIX}) {
unsigned int pathLen = strlen(path) + 1;
- LOG_ALWAYS_FATAL_IF(pathLen > sizeof(mAddr.sun_path), "%u %s", pathLen, path);
+ LOG_ALWAYS_FATAL_IF(pathLen > sizeof(mAddr.sun_path), "Socket path is too long: %u %s",
+ pathLen, path);
memcpy(mAddr.sun_path, path, pathLen);
}
virtual ~UnixSocketAddress() {}