FreeBSD 11.4's libusb10_handle_events_sub ends every pass, a timed-out
one included, with pthread_cond_broadcast on the context's condition
variable, so a thread in libusb_wait_for_event wakes after each pass of
the handler. Here waiters woke only when the handler unlocked, so a
waiter could sleep through the handler's whole loop. Notify them when
the handler takes the event lock back after a pass.
The follower test that expected a timeout while the handler ran 20 ms
passes now races that wake-up, so guest_usbd checks three cases
instead: a follower with a long timeout returns 0 after a pass, a
follower times out during a longer pass, and a follower returns after
the handler held the event lock past its timeout.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The timed-out follower case joined the follower while the main thread
held the event lock, so a follower whose timeout had expired but which
had not yet reacquired the lock could not return. Keep handling events
in slices, which release the lock, until the follower has returned, and
run the case a second time with the lock held past the follower's
timeout.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
sceUsbdLockEvents, sceUsbdTryLockEvents, sceUsbdUnlockEvents,
sceUsbdLockEventWaiters, sceUsbdUnlockEventWaiters, sceUsbdWaitForEvent,
sceUsbdEventHandlerActive, sceUsbdHandleEvents and
sceUsbdHandleEventsLocked follow FreeBSD 11.4 libusb10_io.c: one lock
for events and their waiters, one handler thread that releases the lock
while the host libusb handles events, and waiters that are woken when
the handler unlocks and take over handling when there is none.
sceUsbdEventHandlingOk now reports whether the calling thread is the
handler, and sceUsbdHandleEventsTimeout takes the same path.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The device test raced sceUsbdCancelTransfer on the first interrupt IN endpoint of the device. On Windows the first one can belong to a mouse or keyboard collection, which the system opens exclusively, so libusb's HID backend cannot read it and the submission failed, reported as a failed cancel race. The test now tries each interrupt IN endpoint, requires a refusal to be the host's I/O error with nothing in flight, and races on the first one the host accepts, with a buffer sized from that endpoint.
sceUsbdGetDevice, sceUsbdGetDeviceSpeed, sceUsbdGetMaxPacketSize,
sceUsbdGetMaxIsoPacketSize, sceUsbdOpenDeviceWithVidPid,
sceUsbdSetInterfaceAltSetting, sceUsbdClearHalt and
sceUsbdDetachKernelDriver follow FreeBSD 11.4 libusb10.c over the host
libusb. The packet sizes and the alternate setting checks are done as
FreeBSD does them, so the library records which interfaces are claimed,
and a device stays known while a handle to it is open.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Dean