Programe exited after call pthread_completejoin() many times on ESP32

[HongxiaWangSSSS]

Hi, @anchao

When I tested a multi-threaded stress test case, I found that after introducing (#11898), my program would exit with status code = 12.

The test log as below:

nsh> iwasm --max-threads=12 stress_test_threads_creation_wasi.aot
Spawning stress test is 10% finished
Spawning stress test is 20% finished
exit status=12
exit status=12
nsh>

After debugging I found in the function pthread_findjoininfo it will malloc some memory, but doesn't free it, like: https://github.com/apache/nuttx/blob/master/sched/pthread/pthread_completejoin.c#L100C1-L101C1 (there are other places similar to this),

So I'm a little curious about the reason and hope if you could help me~ Thank U

BR,
Hongxia

[anchao]

Hi, @HongxiaWangSSSS
Do you have other test cases that can reproduce this issue?

joininfo will not be released until the task group is destroyed:
https://github.com/apache/nuttx/blob/master/sched/pthread/pthread_release.c#L60-L80

[anchao]

I tried to run the test code in the C native, the test case works as expected:

https://github.com/bytecodealliance/wasm-micro-runtime/blob/main/core/iwasm/libraries/lib-wasi-threads/stress-test/stress_test_threads_creation.c#L52

1. replace the test code:

$ git status
HEAD detached at origin/master
Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
	modified:   examples/hello/hello_main.c

no changes added to commit (use "git add" and/or "git commit -a")

2. remove the wasi api:

https://github.com/bytecodealliance/wasm-micro-runtime/blob/main/core/iwasm/libraries/lib-wasi-threads/stress-test/stress_test_threads_creation.c#L67-L71

3. run the test demo:

./tools/configure.sh sabre-6quad/nsh

$ qemu-system-arm -semihosting -M sabrelite -m 1024 -smp 4 -nographic -kernel ./nuttx

NuttShell (NSH) NuttX-10.4.0
nsh> hello
Spawning stress test is 10% finished
Spawning stress test is 20% finished
Spawning stress test is 30% finished
Spawning stress test is 40% finished
Spawning stress test is 50% finished
Spawning stress test is 60% finished
Spawning stress test is 70% finished
Spawning stress test is 80% finished
Spawning stress test is 89% finished
Spawning stress test finished successfully executed 50000 threads with retry ratio 1.000160
nsh> free
                   total       used       free    maxused    maxfree  nused  nfree
        Umem: 1065190036      10132 1065179904    1247684 1065166264     24      3
nsh> hello
Spawning stress test is 10% finished
Spawning stress test is 20% finished
Spawning stress test is 30% finished
Spawning stress test is 40% finished
Spawning stress test is 50% finished
Spawning stress test is 60% finished
Spawning stress test is 70% finished
Spawning stress test is 80% finished
Spawning stress test is 89% finished
Spawning stress test finished successfully executed 100000 threads with retry ratio 1.000120
nsh> free
                   total       used       free    maxused    maxfree  nused  nfree
        Umem: 1065190036      10132 1065179904    1247684 1065166264     24      3
nsh> 

[HongxiaWangSSSS]

Hi,

Thank you for the test !!!🙇‍
Yes, I also found the exit and memory leak only occurs when built this sample into WASM/AOT, and during the run time, it continued malloc until exited and never released.
Maybe I should investigate the pthread_join's implementation in WASI-LIBC side.
Anyway, thank you for your explanation~ If there is any progress I will update.

BR,
Hongxia

[anchao]

@HongxiaWangSSSS

I probably know the root cause. after the introduction of #11898, the detached thread will not destroy the joininfo. In your test case, a large number of pthreads will be created with detached attribute, so there will be many joininfo pending in the task group，each of joininfo is to ensure that pthread_join() could gets the correct return value:

joining a detached/canceled thread should return EINVAL, not ESRCH

apache/nuttx-apps#2329

https://github.com/apache/nuttx/blob/master/sched/pthread/pthread_join.c#L92-L102

since many of joininfo will consume too much memory, I think this is not friendly to embedded MCU devices，maybe we can keep the semantics before #11898 was introduced, @xiaoxiang781216 how do you think?

[xiaoxiang781216]

thread should release joininfo after detaching, since nobody will call pthread_join.

[anchao]

@HongxiaWangSSSS please try PR #12106

[HongxiaWangSSSS]

yes, it worked!!
Thank you @anchao for the quick action.👍
Let me close the issue~