On linux using pthreads, if I have a 2-bytes variable x shared by pthreads used only in the following ways: (1) x=value; (2) or (separately) if (x==value) {do something (not with x)} (3) or (separately) other variable (not shared) = x; Is it correct to define it sig_atomic_t x; (no matter if 4 or 2 bytes) and AVOID to protect every instruction of kind (1) or (2) or (3) with pthread_mutex_lock()/trylock() before and pthread_mutex_unlock() after ? Can I be sure there is no race condition/unpredicted event/problem? In a few words: is it correct to speed up a software using sig_atomic_t data instead of mutexed data with the fundamental condition that data are as little as 2-4 bytes and are accessed only in the very simple way above? Thanks

Not sure if this answers your question: http://sourceware.org/ml/libc-alpha/2005-07/msg00004.html But I'll admit, you do ask some tough questions.