Re: [RFC][PATCH 0/3 v3] ftrace: Add access to function arguments for all callbacks

[Masami Hiramatsu]

On Wed, 28 Oct 2020 20:08:16 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> This is something I wanted to implement a long time ago, but held off until
> there was a good reason to do so. Now it appears that having access to the
> arguments of the function by default is very useful. As a bonus, because
> arguments must be saved regardless before calling a callback, because they
> need to be restored before returning back to the start of the traced
> function, there's not much work to do to have them always be there for
> normal function callbacks.
> 
> The basic idea is that if CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS is set, then
> all callbacks registered to ftrace can use the regs parameter for the stack
> and arguments (kernel_stack_pointer(regs), regs_get_kernel_argument(regs, n)),
> without the need to set REGS that causes overhead by saving all registers as
> REGS simulates a breakpoint.
> 
> This could be extended to move the REGS portion to kprobes itself, and
> remove the SAVE_REGS flag completely, but for now, kprobes still uses the
> full SAVE_REGS support.

Yeah, I would like to keep using the SAVE_REGS support until SAVE_ARGS
implemented on all arch. But in parallel, I will prepare generic
kprobe-on-ftrace handler.

> The last patch extends the WITH_ARGS to allow default function tracing to
> modify the instruction pointer, where livepatching for x86 no longer needs
> to save all registers.
> 
> The idea of this approach is to give enough information to a callback that
> it could retrieve all arguments, which includes the stack pointer and
> instruction pointer.
> 
> This can also be extended to modify the function graph tracer to use the
> function tracer instead of having a separate trampoline.

Anyway, this series (3 patches) looks good to me.

Reviewed-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>

Thank you!

> 
> Changes since v2:
> 
>  - Change all callback args to use struct ftrace_regs, and if something
>    requires the full regs, it must use the ftrace_get_regs() helper macro,
>    which will return NULL if the regs isn't full (FL_SAVE_REGS set).
>    This addresses a concern by both Peter Zijlsta and Thomas Gleixner that
>    a partially filled pt_regs may be used inappropriately.
> 
>  - No test patch in this version.
> 
>  - Use case using livepatching is added. This passes the test_livepatch.sh
>    selftest.
> 
> Steven Rostedt (VMware) (3):
>       ftrace: Have the callbacks receive a struct ftrace_regs instead of pt_regs
>       ftrace/x86: Allow for arguments to be passed in to ftrace_regs by default
>       livepatching: Use the default ftrace_ops instead of REGS when ARGS is available
> 
> ----
>  arch/powerpc/include/asm/livepatch.h |  4 +++-
>  arch/s390/include/asm/livepatch.h    |  5 ++++-
>  arch/x86/Kconfig                     |  1 +
>  arch/x86/include/asm/ftrace.h        | 18 ++++++++++++++++++
>  arch/x86/include/asm/livepatch.h     |  4 ++--
>  arch/x86/kernel/ftrace_64.S          | 15 +++++++++++++--
>  arch/x86/kernel/kprobes/ftrace.c     |  3 ++-
>  include/linux/ftrace.h               | 28 ++++++++++++++++++++++++++--
>  include/linux/kprobes.h              |  2 +-
>  kernel/livepatch/patch.c             | 10 ++++++----
>  kernel/trace/Kconfig                 |  9 +++++++++
>  kernel/trace/ftrace.c                | 27 +++++++++++++++------------
>  kernel/trace/trace_event_perf.c      |  2 +-
>  kernel/trace/trace_functions.c       |  9 ++++-----
>  kernel/trace/trace_irqsoff.c         |  2 +-
>  kernel/trace/trace_sched_wakeup.c    |  2 +-
>  kernel/trace/trace_selftest.c        | 20 +++++++++++---------
>  kernel/trace/trace_stack.c           |  2 +-
>  18 files changed, 119 insertions(+), 44 deletions(-)


-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>