Just want to share this. First, at: http://tuxology.net/lectures/crash-and-burn-writing-linux-application-fault-handlers/ At page 6, is the common practise of capturing the stacktrace, whenever error occurred. But this is actually wrong, and will give rise to race condition. Inside the signal/fault handler, there are many scenario which may give rise to memory allocation. POSIX C has defined a list of async-safe function, and if implementation is correct, these functions are not supposed to allocate memory, and thus can be safely used inside a signal handler. Good examples of non-async safe functions are backtrace(), and printf(). But write(), and open() are async safe, and should be used for I/O instead - after taking care of locks etc in multi-core scenario. Essentially all the problems are summarized in page 17, and the rest of slides is expanding on that. Share your comments. -- Regards, Peter Teoh -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
