On 22/02/19 07:36, Peter Xu wrote: > On Fri, Feb 22, 2019 at 07:25:16AM +0100, Marc-André Lureau wrote: >> Hi >> >> On Fri, Feb 22, 2019 at 4:14 AM Peter Xu <pet...@redhat.com> wrote: >>> >>> Only sending an init-done message using lock+cond seems an overkill to >>> me. Replacing it with a simpler semaphore. >>> >>> Meanwhile, init the semaphore unconditionally, then we can destroy it >>> unconditionally too in finalize which seems cleaner. >>> >>> Signed-off-by: Peter Xu <pet...@redhat.com> >> >> The lock is also protecting thread_id. > > IMHO it's fine because thread_id is only changed at the beginning of > iothread_run where the caller will definitely wait for the thread_id > to be generated. Here qemu_sem_post() should at least contain one > write memory barrier there to make sure the waker will read the > correct value after sem_wait() and then later on thread_id is never > changed.
Yes, qemu_sem_post() is a "release" operation. Anything that happens before it is visible to the thread that does qemu_sem_wait(). In fact, thread_id is accessed outside the lock in iothread_instance_finalize, so Peter's change is overall an improvement. Paolo
