There was a report where a module-specific systemtap probe was not able to be computed (past -p2) without kernel-debuginfo installed, even though the module debuginfo was there. We should test whether this is still the case with current code, and fix it if so.
Embracing and extending this bug to cover the desired capability to do at least rudimentary function probing without debugging information. This could rely on Module.symvers (for exported symbols only), or some future mechanism to gather offline symbol data. /proc/kallsyms or /boot/Symbols.* is not quite good enough for cross-compilation.
*** Bug 4843 has been marked as a duplicate of this bug. ***
We've divided this feature into three phases: 1. Function boundary tracing using the kernel/module symbol table(s) (from vmlinux/module elf or from a text file like /proc/kallsyms) 2. Access to arg values by arg number rather than arg name, plus access to register values by name. 3. Function boundary tracing of user apps using the apps' symbol tables. Phase 1 is completed and merged. As far as I know, it's been explicitly tested only on i386 and x86_64, but the tests I added to testsuite/{semok,semko} should run on all architectures. This feature adds (restores?) the ability to probe assembly language functions (per Bug 4843). Phase 2 is completed on i386 and x86_64, and merged. The functional tests for the register feature are arch-specific and haven't been checked in yet. Phase 3 awaits completion of basic dwarf-based probing of user apps.
This feature is broken in current upstream systemtap. Needs fixing...
bug #6864 covers the restoration of this function