blob: e1bd60d9e978f6645ba55bccec90418663a6669d [file] [log] [blame]
/*
* Copyright (C) 2016 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package dalvik.annotation.optimization;
import java.lang.annotation.ElementType;
import java.lang.annotation.Retention;
import java.lang.annotation.RetentionPolicy;
import java.lang.annotation.Target;
/**
* An ART runtime built-in optimization for {@code native} methods to speed up JNI transitions:
* Methods that are annotated with {@literal @}{@code CriticalNative} use the fastest
* available JNI transitions from managed code to the native code and back.
* This annotation can be applied only to native methods that do not use managed
* objects (in parameters or return values, or as an implicit {@code this}).
*
* <p>
* The {@literal @}{@code CriticalNative} annotation changes the JNI transition ABI.
* The native implementation must exclude the {@code JNIEnv} and {@code jclass} parameters
* from its function signature.
* </p>
*
* <p>
* While executing a {@literal @}{@code CriticalNative} method, the garbage collection cannot
* suspend the thread for essential work and may become blocked. Use with caution. Do not use
* this annotation for long-running methods, including usually-fast, but generally unbounded,
* methods. In particular, the code should not perform significant I/O operations or acquire
* native locks that can be held for a long time. (Some logging or native allocations, which
* internally acquire native locks for a short time, are generally OK. However, as the cost
* of several such operations adds up, the {@literal @}{@code CriticalNative} performance gain
* can become insignificant and overshadowed by potential GC delays.)
* </p>
*
* <p>
* For performance critical methods that need this annotation, it is strongly recommended
* to explicitly register the method(s) with JNI {@code RegisterNatives} instead of relying
* on the built-in dynamic JNI linking.
* </p>
*
* <p>
* The {@literal @}{@code CriticalNative} optimization was implemented for system use since
* Android 8 and became CTS-tested public API in Android 14. Developers aiming for maximum
* compatibility should avoid calling {@literal @}{@code CriticalNative} methods on Android 13-.
* The optimization is likely to work also on Android 8-13 devices (after all, it was used
* in the system, albeit without the strong CTS guarantees), especially those that use
* unmodified versions of ART, such as Android 12+ devices with the official ART Module.
* The built-in dynamic JNI linking is working only in Android 12+, the explicit registration
* with JNI {@code RegisterNatives} is strictly required for running on Android versions 8-11.
* The annotation is ignored on Android 7-, so the ABI mismatch would lead to wrong argument
* marshalling and likely crashes.
* </p>
*
* <p>
* A similar annotation, {@literal @}{@link FastNative}, exists for methods that need fast
* transitions but absolutely need to use managed objects, whether as the implicit {@code this}
* for non-static methods, or method arguments, return values or to otherwise call back to
* managed code (say, static methods), or access managed heap objects (say, static fields).
* </p>
*
* <p>
* Performance of JNI transitions:
* <ul>
* <li>Regular JNI cost in nanoseconds: 115
* <li>Fast {@code (!)} JNI cost in nanoseconds: 60
* <li>{@literal @}{@link FastNative} cost in nanoseconds: 35
* <li>{@literal @}{@code CriticalNative} cost in nanoseconds: 25
* </ul>
* (Measured on angler-userdebug in 07/2016).
* </p>
*
* <p>
* <b>Deadlock Warning:</b> As a rule of thumb, any native locks acquired in a
* {@literal @}{@link CriticalNative} call (despite the above warning that this is an unbounded
* operation that can block GC for a long time) must be released before returning to managed code.
* </p>
*
* <p>
* Say some code does:
*
* <code>
* critical_native_call_to_grab_a_lock();
* does_some_java_work();
* critical_native_call_to_release_a_lock();
* </code>
*
* <p>
* This code can lead to deadlocks. Say thread 1 just finishes
* {@code critical_native_call_to_grab_a_lock()} and is in {@code does_some_java_work()}.
* GC kicks in and suspends thread 1. Thread 2 now is in
* {@code critical_native_call_to_grab_a_lock()} but is blocked on grabbing the
* native lock since it's held by thread 1. Now thread suspension can't finish
* since thread 2 can't be suspended since it's doing CriticalNative JNI.
* </p>
*
* <p>
* Normal natives don't have the issue since once it's executing in native code,
* it is considered suspended from the runtime's point of view.
* CriticalNative natives however don't do the state transition done by the normal natives.
* </p>
*
* <p>
* This annotation has no effect when used with non-native methods.
* </p>
*
* <p>
* The runtime shall throw a {@link java.lang.VerifyError} during verification if this annotation
* is present on a native method that is non-static, or contains object parameters, or returns an
* object.
* </p>
*/
@Retention(RetentionPolicy.CLASS) // Save memory, don't instantiate as an object at runtime.
@Target(ElementType.METHOD)
public @interface CriticalNative {}