Crashed at Inner::try_wake()

[black-binary]

My code crashed after switching from smol::lock::Mutex to fast-async-mutex::mutex::Mutex. I'm not familiar with the unsafe code so I'm not sure if it's a bug or I'm using it in the wrong way.

fast-async-mutex/src/inner.rs

Lines 98 to 109 in 817788e

	pub(crate) fn try_wake(&self, waker_ptr: *mut Waker) {
	let cur_waker_ptr = self.waker.swap(waker_ptr, Ordering::Relaxed);
	if !cur_waker_ptr.is_null() {
	let cur_waker = unsafe { Box::from_raw(cur_waker_ptr) };
	if waker_ptr.is_null() || !cur_waker.will_wake(unsafe { &*waker_ptr }) {
	cur_waker.wake();
	}
	} else if !waker_ptr.is_null() {
	unsafe { &*waker_ptr }.wake_by_ref();
	}
	}

I need to lock the mutex in a non-async function. So what I'm doing is

• calling lock.lock_owned() to generate a future fut

• calling fut.poll(cx) trying to lock the mutex immediately

• if poll() returns Pending, store the future in my struct, and wait for the next polling

Here is the original code

https://github.com/black-binary/ap-kcp/blob/ce0605523be63e8e68176847fc63c59a54b8918b/src/async_kcp.rs#L46-L73

And it crashed in the benchmark. I failed to reproduce this crash in debug mode, so I had to use gdb and IDA to find what happened.

The calling stack looks like this. The RIP register was zero (called a null pointer?)

IDA shows that the crash happened after calling Inner::try_wake()

[Mnwa]

Hey! Thank you for this issue. Can you try to connect this lib from the fix-try_wake branch?
Like are

fast-async-mutex = { git = "https://github.com/Mnwa/fast-async-mutex", branch = "fix-try_wake" }

in your Cargo.toml
And retry you benchmark

[black-binary]

Thanks for the fix and the reply. Everything goes well after updating to that branch.

Looking forward to the new version. :)

[black-binary]

But the patch looks a little bit weird. Is the waker simply get deleted?

[Mnwa]

Nice! I publish the new version 0.6.7 with this fix.
Now Weaker will wake a future on every call of the poll method. It does not affect the performance a lot. I updated the Benchmark results in the Readme.

[Mnwa]

And now there is a much less unsafe code.

[black-binary]

Nice! I publish the new version 0.6.7 with this fix.
Now Weaker will wake a future on every call of the poll method. It does not affect the performance a lot. I updated the Benchmark results in the Readme.

I think it's not the right way to fix this. Every task calling to poll will be woken immediately. If a task locked the mutex for a long time, other tasks will keep polling it until the mutex is released (busy waiting). That sounds like a spinning lock instead of a mutex.

[Mnwa]

You're right. This is a fast fix of the bug. Next I will think about how I can do calling that on every poll.