DingliZhang
7e34543e9d
RISC-V points FP at the sender SP instead of at the saved FP, so the frame link pair lives below FP (frame_riscv.hpp: link_offset = -2, return_addr_offset = -1). The stack walker assumed the x86/AArch64 layout, and the interpreter frame offsets were hardcoded to the AArch64 values. _interpreter_frame_bcp_offset was never set on RISC-V, so hasStackStructs() returned false and cstack=vm was rejected. The profiler silently fell back to AsyncGetCallTrace, which cannot walk StubRoutines frames: on a SPECjbb-like workload 95% of the samples ended up in [unknown_Java], and inlined methods were never expanded from the nmethod scopes. Also teach unwindStub() about the two RISC-V stub prologues - the frame link pushed by MacroAssembler::enter() and the plain SP decrement used by leaf stubs - and implement unwindPrologue()/unwindEpilogue(). Measured with itimer sampling on qemu-riscv64, JDK 28: [unknown_Java] drops from 94.9% to 0%, and arraycopy/md5/sha/crc32 stubs now resolve to the same Java caller chains as x86_64. The x86_64 and AArch64 binaries are unchanged.
Async-profiler
This project is a low overhead sampling profiler for Java that does not suffer from the Safepoint bias problem. It features HotSpot-specific API to collect stack traces and to track memory allocations. The profiler works with OpenJDK and other Java runtimes based on the HotSpot JVM.
Unlike traditional Java profilers, async-profiler monitors non-Java threads (e.g., GC and JIT compiler threads) and shows native and kernel frames in stack traces.
What can be profiled:
- CPU time
- Allocations in Java Heap
- Native memory allocations and leaks
- Contended locks
- Hardware and software performance counters like cache misses, page faults, context switches
- and more.
See our 3 hours playlist to learn about more features.
Download
Stable release: 4.5
- Linux x64: async-profiler-4.5-linux-x64.tar.gz
- Linux arm64: async-profiler-4.5-linux-arm64.tar.gz
- macOS arm64/x64: async-profiler-4.5-macos.zip
- Profile converters: jfr-converter.jar
Nightly builds
The most recent binaries corresponding
to the latest successful commit in master.
For a build corresponding to one of the previous commits, go to Nightly Builds, click the desired build and scroll down to the artifacts section. These binaries are kept for 30 days.
Quick start
In a typical use case, profiling a Java application is just a matter of a running asprof with a PID of a
running Java process.
$ asprof -d 30 -f flamegraph.html <PID>
The above command translates to: run profiler for 30 seconds and save results to flamegraph.html
as an interactive Flame Graph that can be viewed in a browser.
Find more details in the Getting started guide.
Building
Build status
Minimum requirements
- make
- GCC 7.5.0+ or Clang 7.0.0+
- Static version of libstdc++ (e.g. on Amazon Linux 2023:
yum install libstdc++-static) - JDK 11+
How to build
Make sure gcc, g++ and java are available on the PATH.
Navigate to the root directory with async-profiler sources and run make.
async-profiler launcher will be available at build/bin/asprof.
Other Makefile targets:
make test- run unit and integration tests;make release- package async-profiler binaries as.tar.gz(Linux) or.zip(macOS).
Supported platforms
| Officially maintained builds | Other available ports | |
|---|---|---|
| Linux | x64, arm64 | x86, arm32, ppc64le, riscv64, loongarch64 |
| macOS | x64, arm64 |
Documentation
Basic usage
Profiler output
Advanced usage
- CPU Sampling Engines
- Stack Walking Modes
- Advanced Stacktrace Features
- Profiling Non-Java Applications
Troubleshooting
For known issues faced while running async-profiler and their detailed troubleshooting, please refer here.