[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <20230818224006.a611cd1a73e00ca1a48478bc@kernel.org>
Date: Fri, 18 Aug 2023 22:40:06 +0900
From: Masami Hiramatsu (Google) <mhiramat@...nel.org>
To: Florent Revest <revest@...omium.org>
Cc: Alexei Starovoitov <alexei.starovoitov@...il.com>,
Steven Rostedt <rostedt@...dmis.org>,
linux-trace-kernel@...r.kernel.org,
LKML <linux-kernel@...r.kernel.org>,
Martin KaFai Lau <martin.lau@...ux.dev>,
bpf <bpf@...r.kernel.org>, Sven Schnelle <svens@...ux.ibm.com>,
Alexei Starovoitov <ast@...nel.org>,
Jiri Olsa <jolsa@...nel.org>,
Arnaldo Carvalho de Melo <acme@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>,
Alan Maguire <alan.maguire@...cle.com>,
Mark Rutland <mark.rutland@....com>,
Peter Zijlstra <peterz@...radead.org>,
Thomas Gleixner <tglx@...utronix.de>
Subject: Re: [PATCH v3 0/8] bpf: fprobe: rethook: Use ftrace_regs instead of
pt_regs
On Thu, 17 Aug 2023 10:57:13 +0200
Florent Revest <revest@...omium.org> wrote:
> On Sat, Aug 12, 2023 at 7:36 AM Masami Hiramatsu (Google)
> <mhiramat@...nel.org> wrote:
> >
> > Hi,
> >
> > Here is the 3rd version of RFC series to use ftrace_regs instead of pt_regs.
> > The previous version is here;
> >
> > https://lore.kernel.org/all/169139090386.324433.6412259486776991296.stgit@devnote2/
> >
> > This also includes the generic part and minimum modifications of arch
> > dependent code. (e.g. not including rethook for arm64.)
>
> I think that one aspect that's missing from the discussion (and maybe
> the series) so far is plans to actually save partial registers in the
> existing rethook trampolines.
Yes, it is arch-dependent part. We have to recheck what registers are
required for the rethook, and that is saved correctly on partial pt_regs
on each architecture.
> For now the series makes everything called by the rethook trampolines
> handle the possibility of having a sparse ftrace_regs but the rethook
> trampolines still save full ftrace_regs. I think that to rip the full
> benefits of this series, we should have the rethook trampolines save
> the equivalent ftrace_regs as the light "args" version of the ftrace
> trampoline.
I think this part depends on the architecture implementation, but yes.
Arm64 can *add* the rethook implementation but not enable KRETPROBE_ON_RETHOOK.
(do not remove kretprobe trampoline)
For this perpose, we need HAVE_RETHOOK_WITH_REGS;
config KRETPROBE_ON_RETHOOK
def_bool y
- depends on HAVE_RETHOOK
+ depends on HAVE_RETHOOK_WITH_REGS
depends on KRETPROBES
select RETHOOK
So there will be pt_regs rethook and ftrace_regs (partial regs) rethook.
I would like to replace rethook's pt_regs with ftrace_regs too. However the
most problematic part is kretprobe. If CONFIG_KRETPROBE_ON_RETHOOK=y, the
rethook must use pt_regs instead of ftrace_regs for API compatibility.
But it makes hard to integrate the rethook and function-graph trace return
hook. (I will discuss this in LPC)
Thank you,
--
Masami Hiramatsu (Google) <mhiramat@...nel.org>
Powered by blists - more mailing lists