Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221123142025.1504030-1-suagrfillet@gmail.com>
 <20221123142025.1504030-2-suagrfillet@gmail.com> <CAJF2gTTwvApRaGhZJxOrdcjWUC9DN-WF_7EKcHXx5HZ9Jkd5EA@mail.gmail.com>
 <CAAYs2=ghhMmmrM4Bktvw9iGiajS5PLd1=Z61j2ZfhbhGUi4ekg@mail.gmail.com> <CAJF2gTTgTfGAYTWbqqrdZg9bMjbpQfx8GzFzxoq3C4r9HDpz8Q@mail.gmail.com>
In-Reply-To: <CAJF2gTTgTfGAYTWbqqrdZg9bMjbpQfx8GzFzxoq3C4r9HDpz8Q@mail.gmail.com>
From:   Song Shuai <suagrfillet@gmail.com>
Date:   Thu, 24 Nov 2022 02:52:06 +0000
Message-ID: <CAAYs2=iJM3d6tqwLjCoE7S1p3JURMYVj54Vry8KiSz1bZakxcA@mail.gmail.com>
Subject: Re: [PATCH 1/2] riscv/ftrace: add DYNAMIC_FTRACE_WITH_DIRECT_CALLS support
To:     Guo Ren <guoren@kernel.org>
Cc:     paul.walmsley@sifive.com, palmer@dabbelt.com,
        aou@eecs.berkeley.edu, rostedt@goodmis.org, mhiramat@kernel.org,
        mark.rutland@arm.com, peterz@infradead.org, jolsa@redhat.com,
        bp@suse.de, jpoimboe@kernel.org, linux-riscv@lists.infradead.org,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Guo Ren <guoren@kernel.org> =E4=BA=8E2022=E5=B9=B411=E6=9C=8824=E6=97=A5=E5=
=91=A8=E5=9B=9B 02:08=E5=86=99=E9=81=93=EF=BC=9A
>
> On Thu, Nov 24, 2022 at 1:27 AM Song Shuai <suagrfillet@gmail.com> wrote:
> >
> > Guo Ren <guoren@kernel.org> =E4=BA=8E2022=E5=B9=B411=E6=9C=8823=E6=97=
=A5=E5=91=A8=E4=B8=89 23:02=E5=86=99=E9=81=93=EF=BC=9A
> > >
> > > Cool job, thx.
> > >
> > > On Wed, Nov 23, 2022 at 10:20 PM Song Shuai <suagrfillet@gmail.com> w=
rote:
> > >>
> > >> This patch adds DYNAMIC_FTRACE_WITH_DIRECT_CALLS support for RISC-V.
> > >>
> > >> select the DYNAMIC_FTRACE_WITH_DIRECT_CALLS to provide the
> > >> register_ftrace_direct[_multi] interfaces allowing users to register
> > >> the customed trampoline (direct_caller) as the mcount for one or
> > >> more target functions. And modify_ftrace_direct[_multi] are also
> > >> provided for modifying direct_caller.
> > >>
> > >> To make the direct_caller and the other ftrace hooks (eg. function/f=
graph
> > >> tracer, k[ret]probes) co-exist, a temporary register is nominated to
> > >> store the address of direct_caller in ftrace_regs_caller. After the
> > >> setting of the address direct_caller by direct_ops->func and the
> > >> RESTORE_REGS in ftrace_regs_caller, direct_caller will be jumped to
> > >> by the `jr` inst.
> > >>
> > >> Signed-off-by: Song Shuai <suagrfillet@gmail.com>
> > >> ---
> > >>  arch/riscv/Kconfig              | 1 +
> > >>  arch/riscv/include/asm/ftrace.h | 6 ++++++
> > >>  arch/riscv/kernel/mcount-dyn.S  | 4 ++++
> > >>  3 files changed, 11 insertions(+)
> > >>
> > >> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> > >> index 39ec8d628cf6..d083ec08d0b6 100644
> > >> --- a/arch/riscv/Kconfig
> > >> +++ b/arch/riscv/Kconfig
> > >> @@ -278,6 +278,7 @@ config ARCH_RV64I
> > >>         select ARCH_SUPPORTS_INT128 if CC_HAS_INT128
> > >>         select HAVE_DYNAMIC_FTRACE if !XIP_KERNEL && MMU && $(cc-opt=
ion,-fpatchable-function-entry=3D8)
> > >>         select HAVE_DYNAMIC_FTRACE_WITH_REGS if HAVE_DYNAMIC_FTRACE
> > >> +       select HAVE_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
> > >>         select HAVE_FTRACE_MCOUNT_RECORD if !XIP_KERNEL
> > >>         select HAVE_FUNCTION_GRAPH_TRACER
> > >>         select HAVE_FUNCTION_TRACER if !XIP_KERNEL && !PREEMPTION
> > >> diff --git a/arch/riscv/include/asm/ftrace.h b/arch/riscv/include/as=
m/ftrace.h
> > >> index 01bebb28eabe..be4d57566139 100644
> > >> --- a/arch/riscv/include/asm/ftrace.h
> > >> +++ b/arch/riscv/include/asm/ftrace.h
> > >> @@ -114,6 +114,12 @@ struct ftrace_regs;
> > >>  void ftrace_graph_func(unsigned long ip, unsigned long parent_ip,
> > >>                        struct ftrace_ops *op, struct ftrace_regs *fr=
egs);
> > >>  #define ftrace_graph_func ftrace_graph_func
> > >> +
> > >> +static inline void arch_ftrace_set_direct_caller(struct pt_regs *re=
gs, unsigned long addr)
> > >> +{
> > >> +               regs->t1 =3D addr;
> > >
> > > How about regs->t0 =3D addr; ?
> > > And delete all mcount-dyn.S modification.
> > >
> > The direct_caller has the same program layout as the ftrace_caller, whi=
ch means
> > the reg t0 will never be changed when direct_caller returns.
> >
> > If regs->t0 changes here and ftrace_regs_caller executes `jr t0`,
> > direct_caller will enter the dead loop.
> ?
>
> ftrace_regs_caller->call_direct_funcs->
> arch_ftrace_set_direct_caller-> ftrace_regs_caller jr t0.
>
> Only call_direct_funcs call arch_ftrace_set_direct_caller !
>
> static void call_direct_funcs(unsigned long ip, unsigned long pip,
>                               struct ftrace_ops *ops, struct ftrace_regs =
*fregs)
> {
>         struct pt_regs *regs =3D ftrace_get_regs(fregs);
>         unsigned long addr;
>
>         addr =3D ftrace_find_rec_direct(ip);
>         if (!addr)
>                 return;
>
>         arch_ftrace_set_direct_caller(regs, addr);
> }
>
When direct_caller and function tracer co-hook a function, the simple
diagram is like this:

```
func -> ftrace_regs_caller -> arch_ftrace_ops_list_func ->
function_trace_call // write ringbuffer
            |
                 -> call_direct_funcs // regs->t1 =3D direct_caller
           -> t1 !=3D0 && jr t1 // goto direct_caller
```

And the direct_caller has a similar implement as ftrace_caller.

```
direct_caller:
add sp,sp,-?
sd t0,?(sp)
sd ra,?(sp)
call foo
ld t0,?(sp)
ld ra,?(sp)
add sp,sp,?
jr t0 // <- back to function entry
```

If we change regs->t0 as direct_caller and execute `jr t0` directly,
the direct_caller will always jump to itself due to its last `jr` inst.

```
func -> ftrace_regs_caller -> arch_ftrace_ops_list_func ->
function_trace_call // write ringbuffer
            |
                 -> call_direct_funcs // regs->t0 =3D direct_caller
           -> jr t0 // goto direct_caller

direct_caller:
...
sd t0,?(sp)
...
ld t0,?(s0)
...
jr t0 // goto direct_caller always
```

Hope I made it clear.
> >
> > Actually the reg t0 always saves the address of function entry with 8B
> > offset, it should only
> > changed by the IPMODIFY ops instead of the direct_ops.
> > >>
> > >> +}
> > >> +
> > >>  #endif /* CONFIG_DYNAMIC_FTRACE_WITH_REGS */
> > >>
> > >>  #endif /* __ASSEMBLY__ */
> > >> diff --git a/arch/riscv/kernel/mcount-dyn.S b/arch/riscv/kernel/mcou=
nt-dyn.S
> > >> index 466c6ef217b1..b89c85a58569 100644
> > >> --- a/arch/riscv/kernel/mcount-dyn.S
> > >> +++ b/arch/riscv/kernel/mcount-dyn.S
> > >> @@ -233,6 +233,7 @@ ENDPROC(ftrace_caller)
> > >>  #else /* CONFIG_DYNAMIC_FTRACE_WITH_REGS */
> > >>  ENTRY(ftrace_regs_caller)
> > >>         SAVE_ABI_REGS 1
> > >> +       REG_S   x0, PT_T1(sp)
> > >>         PREPARE_ARGS
> > >>
> > >>  ftrace_regs_call:
> > >> @@ -241,7 +242,10 @@ ftrace_regs_call:
> > >>
> > >>
> > >>         RESTORE_ABI_REGS 1
> > >> +       bnez    t1,.Ldirect
> > >>         jr t0
> > >> +.Ldirect:
> > >> +       jr t1
> > >>  ENDPROC(ftrace_regs_caller)
> > >>
> > >>  ENTRY(ftrace_caller)
> > >> --
> > >> 2.20.1
> > >>
> > >
> > >
> > > --
> > > Best Regards
> > >  Guo Ren
>
>
>
> --
> Best Regards
>  Guo Ren
Thanks,
Song