RFR: 8374469: Mixed jstack does not work on Windows [v2]
Yasumasa Suenaga
ysuenaga at openjdk.org
Thu Jan 15 08:17:00 UTC 2026
On Thu, 15 Jan 2026 04:52:19 GMT, Chris Plummer <cjplummer at openjdk.org> wrote:
>> I don't know why these call stacks were produced... It's strange.
>> I tested this PR with cl.exe 19.50.35721 by VS 2026 18.1.1, I didn't see such call stacks.
>>
>> `callJNI()` makes recursive call, but it should include Java upcall (you can see it as LingeredAppWithNativeMethod.callback()`)
>>
>> Top of call stacks is `callJNI()`, thus I believe it has valid RSP/RBP/RIP (passed by `WindbgCDebugger::topFrameForThread`). `callJNI` is JNI function in NoFramePointer.dll generated by cl.exe (in usual (I believe) on OpenJDK build environment), thus I think `GetStackTrace` API works.
>> Each `NoFramePointer!Java_LingeredAppWithNativeMethod_callJNI` has instruction address, it means they were unwinded by the API, thus I think we need to investigate assembly code in NoFramePointer.dll to know what happens...
>
> The binary I have was built with VS2022-17.13.2.
I saw same lines on [GHA results](https://github.com/YaSuenag/jdk/actions/runs/20984170537/artifacts/5123260970) - it used VS 2022 to build. Repeats of `NoFramePointer!Java_LingeredAppWithNativeMethod_callJNI` is expected. It is inlined `fib()` as you said.
I saw callJNI + 0xdb and 0xd0 are repeated as following:
0x00007ffa9b0e10db NoFramePointer!Java_LingeredAppWithNativeMethod_callJNI + 0xdb
0x00007ffa9b0e10d0 NoFramePointer!Java_LingeredAppWithNativeMethod_callJNI + 0xd0
0x00007ffa9b0e10db NoFramePointer!Java_LingeredAppWithNativeMethod_callJNI + 0xdb
0x00007ffa9b0e10d0 NoFramePointer!Java_LingeredAppWithNativeMethod_callJNI + 0xd0
I got [GHA artifacts of debug build](https://github.com/YaSuenag/jdk/actions/runs/20984170537/artifacts/5122848054) and disassembled NoFramePointer.dll, then I saw following code around callJNI + 0xdb and 0xd0:
00000001800010A8: 48 89 5C 24 08 mov qword ptr [rsp+8],rbx
00000001800010AD: 57 push rdi
00000001800010AE: 48 83 EC 20 sub rsp,20h
00000001800010B2: 8B F9 mov edi,ecx
00000001800010B4: 85 C9 test ecx,ecx
00000001800010B6: 75 04 jne 00000001800010BC
00000001800010B8: 33 C0 xor eax,eax
00000001800010BA: EB 22 jmp 00000001800010DE
00000001800010BC: 83 FF 02 cmp edi,2
00000001800010BF: 7F 07 jg 00000001800010C8
00000001800010C1: B8 01 00 00 00 mov eax,1
00000001800010C6: EB 16 jmp 00000001800010DE
00000001800010C8: 83 C1 FE add ecx,0FFFFFFFEh
00000001800010CB: E8 D8 FF FF FF call 00000001800010A8
00000001800010D0: 8D 4F FF lea ecx,[rdi-1]
00000001800010D3: 48 8B D8 mov rbx,rax
00000001800010D6: E8 CD FF FF FF call 00000001800010A8
00000001800010DB: 48 03 C3 add rax,rbx
It makes recursive call to 0x1800010A8 (callJNI + 0xa8) until the condition is met. According to NoFramePointer.dll.map in build artifact, callJNI + 0xa8 is `fib()` in libNoFramePointer.c.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/29019#discussion_r2693380875
More information about the serviceability-dev
mailing list