Sunday, July 10, 2022

[SOLVED] How does perf unwind the stack in newer kernel versions with ORC?

Issue

I'm running perf on a kernel 5.4.80 and explicitly use fp call graph:

perf record -g --call-graph fp ...

When running perf script it can profile the stack trace without any problem. This is while the kernel is not compiled with frame pointer enabled, rather with ORC:

CONFIG_UNWINDER_ORC=y
# CONFIG_UNWINDER_FRAME_POINTER is not set

The result is that the kernel symbols are available in /proc/kallsyms.

On the other hand, seems perf-script only knows about fp, dwarf and lbr: source.

I am wondering how perf script is able to unwind the stack and generate the report despite recording with fp call graph option on a kernel with ORC stack unwinding enabled?


Solution

According to perf-record(1),

--call-graph
           Setup and enable call-graph (stack chain/backtrace)
           recording, implies -g. Default is "fp" (for user space).

               The unwinding method used for kernel space is dependent on the
               unwinder used by the active kernel configuration, i.e
               CONFIG_UNWINDER_FRAME_POINTER (fp) or CONFIG_UNWINDER_ORC (orc)

               Any option specified here controls the method used for user space.

So, --call-graph only controls how the call-graph is unwound in userspace. In kernel space, whichever unwinder is actively enabled is used.



Answered By - Sagar
Answer Checked By - David Goodson (WPSolving Volunteer)
at