My application uses libunwind as part of an in vivo profiling tool that is integrated into the core scripting language of the app. It's an interrupt driven stack sampler eerily similar to the google profiler.
I keep finding that rather large numbers of stack samples fail and I am missing out on significant profiling data. This happens when unw_step() and/or unw_get_reg() are returning negative values. What are the conditions that cause this? Is there anything I can do about it? We are running with x86_64 on AMD Opterons. I'm just doing local sampling in one thread (although there are 2 pthreads to the entire app). I'm using libunwind from libunwind-snap-070410, and can easily update to something newer as long it's stable. -stephen _______________________________________________ Libunwind-devel mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/libunwind-devel
