wasmfs_create_opfs_backend() hangs on low end devices

[velzie]

Calls to wasmfs_create_opfs_backend() seem to not return on certain low end devices, or only return once every x reloads.

Tested on a dell 3100 chromebook (2 core machine)

Version of emscripten/emsdk:

emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 3.1.71 (4171ae200b77a6c266b0e1ebb507d61d1ade3501)
clang version 20.0.0git (https:/github.com/llvm/llvm-project d6344c1cd0d099f8d99ee320f33fc9254dbe8288)
Target: wasm32-unknown-emscripten
Thread model: posix
InstalledDir: /usr/lib/emsdk/upstream/bin

Compiling with emcc Emscripten.c -o index.html -s WASM=1 -s USE_PTHREADS=1 -sWASMFS -sPTHREAD_POOL_SIZE=2

#include <assert.h>
#include <emscripten/console.h>
#include <emscripten/proxying.h>
#include <emscripten/threading.h>
#include <emscripten/wasmfs.h>
#include <pthread.h>
#include <stdio.h>

int mount_opfs() {
  emscripten_console_log("mount_opfs: starting");
  backend_t opfs = wasmfs_create_opfs_backend();
  emscripten_console_log("mount_opfs: created opfs backend");
  int ret = wasmfs_create_directory("/libsdl", 0777, opfs);
  emscripten_console_log("mount_opfs: mounted opfs");
  return ret;
}

void *thread_main(void *arg) {
  emscripten_console_log("trying to mount\n");
  mount_opfs();
  emscripten_console_log("finished mount\n");
}

int main() {
  int arg = 42;

  pthread_t thread;
  if (pthread_create(&thread, NULL, thread_main, &arg) != 0) {
    perror("pthread_create failed");
    return 1;
  }

  emscripten_console_log("Main thread continues\n");

  /* pthread_join(thread, NULL); */

  emscripten_console_log("Main thread done\n");
  return 0;
}

On my development PC, the log reaches finished mount, but on a lower end device it only hits mount_opfs: starting . The main thread continues however.

From my understanding it's spawning the proxy worker that fails, but I'm not sure how I would fix or work around this

[velzie]

Reproduced on 4.0.9 (latest git)

[sbc100]

@tlively any ideas?

[tlively]

Perhaps a problem trying to synchronously start a new thread and wait for it to start executing without returning to the main thread spawning Worker's event loop?

[sbc100]

Do you always see Main thread done ?

[velzie]

Do you always see Main thread done ?

Yes, I can also run a loop in main thread and it continues with the same behavior

[tlively]

Presumably this could be worked around if we had an asynchonous opfs backend constructor.

[sbc100]

But it would be good to know that cause.. I can't think of one in this case.

[r58Playz]

I can reproduce on latest emscripten and chromium 136.0.7103.113 by throttling my CPU to 800mhz and launching with taskset -c 0-1 chromium. Sometimes it happens on the first load, sometimes it takes multiple loads to get stuck.

The main thread doesn't seem to be doing anything, as DevTools never enters a paused state for that thread.
The em-pthread-1 only shows being in __wasmfs_thread_utils_heartbeat (presumably after ProxyWorker's code called it after notifying the condvar):

The em-pthread-2 (which called wasmfs_create_opfs_backend()) is stuck waiting for the worker to spawn:

[velzie]

Worth noting, creating the backend succeeds 100% of the time if -sPTHREAD_POOL_SIZE=0 or I exhaust the thread pool before calling it. This isn't a viable workaround for me though because the app will deadlock later on

[sbc100]

That is very odd.. I would have expected and empty thread pool to have the opposite effect here.