StackTraces / Callchains in Perf Events
Explore the callchains in perf from the implementation to the interface. Highlight their strengths and weaknesses and propose what we can do to improve that. This presentation is going to enter into some details on the kernel/user implementation of perf events callchains: exposing the data structures and the classical frame pointer based unwinding. Then we'll explore the possible improvements that can be made to make it more powerful like using Dwarf Call Frame Informations, LBR (last branch record) feature on some CPUs, BTS, etc.