[External] Re: [PATCH bpf-next v2 1/9] bpf: tracing: add support to record and check the accessed args

Jiri Olsa olsajiri at gmail.com
Sat Mar 30 10:52:30 PDT 2024


On Sat, Mar 30, 2024 at 08:27:55AM -0400, Steven Rostedt wrote:
> On Fri, 29 Mar 2024 16:28:33 -0700
> Andrii Nakryiko <andrii.nakryiko at gmail.com> wrote:
> 
> > I thought I'll just ask instead of digging through code, sorry for
> > being lazy :) Is there any way to pass pt_regs/ftrace_regs captured
> > before function execution to a return probe (fexit/kretprobe)? I.e.,
> > how hard is it to pass input function arguments to a kretprobe? That's
> > the biggest advantage of fexit over kretprobe, and if we can make
> > these original pt_regs/ftrace_regs available to kretprobe, then
> > multi-kretprobe will effectively be this multi-fexit.
> 
> This should be possible with the updates that Masami is doing with the
> fgraph code.

yes, I have bpf kprobe-multi link support for that [0] (it's on top of
Masami's fprobe-over-fgraph changes) we discussed that in [1]

jirka

[0] https://git.kernel.org/pub/scm/linux/kernel/git/jolsa/perf.git/log/?h=bpf/session_data
[1] https://lore.kernel.org/bpf/20240228090242.4040210-1-jolsa@kernel.org/



More information about the linux-riscv mailing list