On 01/04/2015 13:37, Robert Stupp wrote:
:
Just want to understand ”the magic” here (sure, NIO is the way to go)…
Where’s the ”trick” to avoid GC pauses for socketWrite0 and similar
methods?
Is it something hard coded like ”if JNI method is from jdk source tree”?
AFAIK there’s always a safe point w/ JNI calls. socketWrite0 is simply
declared as ”JNIEXPORT void JNICALL
Java_java_net_SocketOutputStream_socketWrite0” and I could not find
anything in the source tree declaring ”socketWrite0” or
”SocketOutputStream” to be special (in jdk8u tree).
There isn't anything special here. In HotSpot then threads in JNI
methods are considered to be a safepoint and so continue to execute even
if the VM is brought to a safepoint, say for a stop-the-world GC. If the
thread attempts to call back into VM or return from the native method
then it will block until the safepoint completes. So nothing special for
socketWrite0.
-Alan.
