Weakness in libusb_exit cleanup codes

[mcuee]

From here:

• windows: hotplug implementation #1406 (comment)

@sonatique mentioned the following.

I guess that for some reason the faulting device is correctly unref'ed down to zero and freed, but still in the list at the end, causing the segfault when log wants to access its properties.

Code in question:

libusb/libusb/core.c

Lines 2629 to 2633 in 0929a2b

	for_each_device(_ctx, dev) {
	usbi_warn(_ctx, "device %d.%d still referenced",
	dev->bus_number, dev->device_address);
	DEVICE_CTX(dev) = NULL;
	}

[mcuee]

BTW, macOS backend seems to do it within the backend.

libusb/libusb/os/darwin_usb.c

Lines 837 to 847 in 0929a2b

	/* cleanup function to destroy cached devices. must be called with a lock on darwin_cached_devices_mutex */
	static void darwin_cleanup_devices(void) {
	struct darwin_cached_device *dev, *next;
	list_for_each_entry_safe(dev, next, &darwin_cached_devices, list, struct darwin_cached_device) {
	if (dev->refcount > 1) {
	usbi_err(NULL, "device still referenced at libusb_exit");
	}
	darwin_deref_cached_device(dev);
	}
	}

[mcuee]

#1452 may be part of the fix here.

• hotplug_exit: removes all parents with a single child remaining in parent tree + lock device list #1452