> Just a general comment - i'm not too familiar with this code, but it looks > like all of these failures will only happen on thread init. Can we make > sure it starts? > > You can use similar approach to Olivier's (recently merged) thread > affinity patches, with pthread barriers etc. to ensure the thread has > initialized properly, pthread_cancel() it if it didn't, and return -1 on > thread init.
Thanks for catching this one. I am now reverting it as well from this set, to be properly handled in a set dedicated to the threading issue
