Mark common android.os.Parcel methods as FastNative
Mark the marshaling functions for non-primitive parameters as being
@FastNative: they're all expected to complete quickly.
According to the table in CriticalNative.java, the difference between
a regular JNI call and a FastNative call is about 80ns, so this change
should shave about 160ns per non-primitive parameter off the latency
of a binder transaction. The actual savings will be greater as we also
use these Parcel methods to marshal and unmarshal complex objects like
PackageInfo sent as objects over Binder.
Bug: 142271139
Test: boots
Change-Id: I85c3ee8498640193b9c0777c8c5e7eb349c7b23c
diff --git a/core/jni/android_util_Binder.cpp b/core/jni/android_util_Binder.cpp
index 0afbaa0e..e406e22 100644
--- a/core/jni/android_util_Binder.cpp
+++ b/core/jni/android_util_Binder.cpp
@@ -697,6 +697,9 @@
// same IBinder, and the original BinderProxy is still alive, return the same BinderProxy.
jobject javaObjectForIBinder(JNIEnv* env, const sp<IBinder>& val)
{
+ // N.B. This function is called from a @FastNative JNI method, so don't take locks around
+ // calls to Java code or block the calling thread for a long time for any reason.
+
if (val == NULL) return NULL;
if (val->checkSubclass(&gBinderOffsets)) {