Paolo Gai wrote: > Dear all, > > here is another small demo... this time of a periodic thread in POSIX. > > It works under Xenomai, but it fails under Posix Real-Time. > > The reason of the failure is the SIGUSR1 not being blocked by the main > thread. If in main() the comments below the note "// REMOVE THE COMMENTS > BELOW" are removed, then the signal is masked also in the main(), and as > a result the POSIX version works. > > The question I have is if the behavior in Xenomai (that is, not failing > like the POSIX counterpart) is correct or not...
The answer is in section 2.4.2 of the standard, delivery of real-time signals. http://www.opengroup.org/onlinepubs/009695399/functions/xsh_chap02_04.html#tag_02_04_02 real-time signals are marked pending to the process, not a particular thread. So, Linux implementation is correct, Xenomai is incorrect. When creating a timer with timer_create, Xenomai assume that signals will be delivered to the thread which created the thread. Linux supports an extension which allows timer notifications to be delivered to a particular thread; set the sigev_notify member of the sigevent structure to SIGEV_SIGNAL | SIGEV_THREAD_ID and set the member _sigev_un._tid to the thread ID of the target thread. -- Gilles Chanteperdrix.
