First stab at having the GC scan the shadow stack.

[vext01]

This comes with the caveat that each thread scans all other threads shadow stacks, which may not be completely safe(?)

With this, and recent ykllvm/yk changes, this mandelbrot benchmark now runs:

https://benchmarksgame-team.pages.debian.net/benchmarksgame/program/fannkuchredux-python3-6.html

(If you run with YKD_SERIALISE_COMPILATION=1 however, I get an unimplemented ptrtoint constant expression -- this isn't hard to implement)

[vext01]

There's a part of the code where micropython signals another thread (literally a UNIX signal) to scan it's own stack, so we might take that as a cue to do this properly? Not sure.

[ltratt]

Scanning other threads' stacks isn't safe in general, but it's fine for now. I think we should undraft this and get it in.

[vext01]

FWIW, here's the code I was referencing:

ykmicropython/ports/unix/mpthreadport.c

Lines 188 to 212 in 900ae4a

	// This function scans all pointers that are external to the current thread.
	// It does this by signalling all other threads and getting them to scan their
	// own registers and stack. Note that there may still be some edge cases left
	// with race conditions and root-pointer scanning: a given thread may manipulate
	// the global root pointers (in mp_state_ctx) while another thread is doing a
	// garbage collection and tracing these pointers.
	void mp_thread_gc_others(void) {
	mp_thread_unix_begin_atomic_section();
	for (mp_thread_t *th = thread; th != NULL; th = th->next) {
	gc_collect_root(&th->arg, 1);
	if (th->id == pthread_self()) {
	continue;
	}
	if (!th->ready) {
	continue;
	}
	pthread_kill(th->id, MP_THREAD_GC_SIGNAL);
	#if defined(__APPLE__)
	sem_wait(thread_signal_done_p);
	#else
	sem_wait(&thread_signal_done);
	#endif
	}
	mp_thread_unix_end_atomic_section();
	}

[ltratt]

I'm assuming this is probably a stop-the-world GC, so this should be simple, especially given that code segment. But we can finesse this next week.

[vext01]

I think we just need a yk API to get the current thread's shadow stack memory range.

At the moment we can only get all threads shadow stacks.

But I'm not a GC guy :)

[vext01]

Anyway, undrafted