Comment 8 for bug 1856256

looking into this more. Profiles with hats, are what will trigger this because the list of children is transferred whole sale from the old parent to the new. This forces an rcu synchronization on the profile replacement. But this is inside of a mutex lock blocking other replacement. Hence one profile gets replaced, and then there is a pause waiting for the rcu synch, and then the lock is released and the next profile replacement happens and blocks in the same way.

This part of the profile replacement is going to need to be redesigned. Because rcu is used to provide a lockless read side. This probably means creating new children instead of moving the children. This isn't as bad as it sounds as they can share the same state machines etc (the majority of the profile) via reference counting.