On Sun, Jan 08, 2017 at 11:58:09PM +0900, Masami Hiramatsu wrote: > Make __kernel_text_address()/kernel_text_address() returns > true if the given address is on a kprobe's instruction slot, > which is generated by kprobes as a trampoline code. > This can help stacktraces to determine the address is on a > text area or not. > > To implement this without any sleep in is_kprobe_*_slot(), > this also modify insn_cache page list as a rcu list. It may > increase processing deley (not processing time) for garbage > slot collection, because it requires to wait an additional > rcu grance period when freeing a page from the list. > However, since it is not a hot path, we may not take care of it. > > Note: this can give a small overhead to stack unwinders because > this adds 2 checks in __kernel_text_address(). However, the > impact should be very small, kprobe_insn_pages list has 1 entry > per 256 probes(on x86, on arm/arm64 it will be 1024 probes), > and kprobe_optinsn_pages has 1 entry per 32 probes(on x86). > In most use cases, the number of kprobe events may be less > than 20, this means is_kprobe_*_slot() will check just 1 entry. > > Signed-off-by: Masami Hiramatsu <[email protected]>
Thanks for doing this Masami! I verified that it works: Tested-by: Josh Poimboeuf <[email protected]> I suspect that BPF generated code also has the same issue, so I'll leave the unwinder warning disabled for now. BTW, I think we'll have more problems with generated code if/when we move to an x86 DWARF unwinder, because it won't have any idea how to unwind past generated code. Long term I wonder if it would make sense to create some kind of framework for creating or registering generated code, so we can solve these types of problems in a single place. -- Josh
