As far as I know perf is able to sample a full callstack for every process/thread even if it currently is inside the kernel. Since perf is running in userspace I assume, I want to know how it gets access to the kernel part of the stack of a different process/thread.
The purpose is to write an application that can sample callstacks of other processes/threads even if they currently are in kernel.
Regards
perf consists of two parts. The userspace tools, and the perf_event_open syscall.
Generally, the stack sampling is done by the kernel through the syscall. You can write an application with also uses perf_event_open - by default kernel samples are included (exclude_kernel==0).