Make instrumentation trampoline able to jump to jit code
In order to get the benefit of the instrumentation trampoline it must
be able to jump to jit code. This patch adds support for doing that
and adds code to ensure that we will not be jumping back and forth
between the trampoline and the interpreter all the time if the jit has
not yet compiled the current method.
Note we also disable the jit-gc when turning on these trampolines
since otherwise we could end up either sending instrumentation events
multiple times or running uninitialized memory.
Bug: 110263880
Test: ./test/testrunner/testrunner.py --host --runtime-option=-Xplugin:libtracefast-trampolined.so
Test: ./test/testrunner/testrunner.py --host --runtime-option=-Xplugin:libtracefast-trampolined.so --runtime-option=-Xjitthreshhold:0
Test: ./test.py --host
Change-Id: Ie6e92ec6367452fe4fde24d520d808b7af91d1b5
diff --git a/runtime/jit/jit_code_cache.h b/runtime/jit/jit_code_cache.h
index b056bc3..f7829de 100644
--- a/runtime/jit/jit_code_cache.h
+++ b/runtime/jit/jit_code_cache.h
@@ -154,6 +154,10 @@
// Return true if the code cache contains this pc.
bool ContainsPc(const void* pc) const;
+ // Returns true if either the method's entrypoint is JIT compiled code or it is the
+ // instrumentation entrypoint and we can jump to jit code for this method. For testing use only.
+ bool WillExecuteJitCode(ArtMethod* method) REQUIRES(!lock_);
+
// Return true if the code cache contains this method.
bool ContainsMethod(ArtMethod* method) REQUIRES(!lock_);
@@ -269,6 +273,16 @@
garbage_collect_code_ = value;
}
+ bool GetGarbageCollectCode() const {
+ return garbage_collect_code_;
+ }
+
+ // If Jit-gc has been disabled (and instrumentation has been enabled) this will return the
+ // jit-compiled entrypoint for this method. Otherwise it will return null.
+ const void* FindCompiledCodeForInstrumentation(ArtMethod* method)
+ REQUIRES(!lock_)
+ REQUIRES_SHARED(Locks::mutator_lock_);
+
private:
// Take ownership of maps.
JitCodeCache(MemMap* code_map,