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: <20230607125911.145345-1-imagedong@tencent.com>
 <20230607125911.145345-3-imagedong@tencent.com> <20230607200348.dprmfvpzdvk5ldpp@macbook-pro-8.dhcp.thefacebook.com>
In-Reply-To: <20230607200348.dprmfvpzdvk5ldpp@macbook-pro-8.dhcp.thefacebook.com>
From:   Menglong Dong <menglong8.dong@gmail.com>
Date:   Thu, 8 Jun 2023 12:38:23 +0800
Message-ID: <CADxym3a+5t9tMun6Pid+38UmFgcQkMYC4esWdENGs2E24zornA@mail.gmail.com>
Subject: Re: [PATCH bpf-next v3 2/3] bpf, x86: clean garbage value in the
 stack of trampoline
To:     Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc:     davem@davemloft.net, dsahern@kernel.org, ast@kernel.org,
        daniel@iogearbox.net, andrii@kernel.org, martin.lau@linux.dev,
        song@kernel.org, yhs@fb.com, john.fastabend@gmail.com,
        kpsingh@kernel.org, sdf@google.com, x86@kernel.org,
        imagedong@tencent.com, benbjiang@tencent.com,
        netdev@vger.kernel.org, bpf@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-kselftest@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Jun 8, 2023 at 4:03=E2=80=AFAM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Wed, Jun 07, 2023 at 08:59:10PM +0800, menglong8.dong@gmail.com wrote:
> > From: Menglong Dong <imagedong@tencent.com>
> >
> > There are garbage values in upper bytes when we store the arguments
> > into stack in save_regs() if the size of the argument less then 8.
> >
> > As we already reserve 8 byte for the arguments in regs and stack,
> > it is ok to store/restore the regs in BPF_DW size. Then, the garbage
> > values in upper bytes will be cleaned.
> >
> > Reviewed-by: Jiang Biao <benbjiang@tencent.com>
> > Signed-off-by: Menglong Dong <imagedong@tencent.com>
> > ---
> >  arch/x86/net/bpf_jit_comp.c | 19 ++++++-------------
> >  1 file changed, 6 insertions(+), 13 deletions(-)
> >
> > diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
> > index 413b986b5afd..e9bc0b50656b 100644
> > --- a/arch/x86/net/bpf_jit_comp.c
> > +++ b/arch/x86/net/bpf_jit_comp.c
> > @@ -1878,20 +1878,16 @@ static void save_regs(const struct btf_func_mod=
el *m, u8 **prog, int nr_regs,
> >
> >               if (i <=3D 5) {
> >                       /* copy function arguments from regs into stack *=
/
> > -                     emit_stx(prog, bytes_to_bpf_size(arg_size),
> > -                              BPF_REG_FP,
> > +                     emit_stx(prog, BPF_DW, BPF_REG_FP,
> >                                i =3D=3D 5 ? X86_REG_R9 : BPF_REG_1 + i,
> >                                -(stack_size - i * 8));
>
> This is ok,
>
> >               } else {
> >                       /* copy function arguments from origin stack fram=
e
> >                        * into current stack frame.
> >                        */
> > -                     emit_ldx(prog, bytes_to_bpf_size(arg_size),
> > -                              BPF_REG_0, BPF_REG_FP,
> > +                     emit_ldx(prog, BPF_DW, BPF_REG_0, BPF_REG_FP,
> >                                (i - 6) * 8 + 0x18);
> > -                     emit_stx(prog, bytes_to_bpf_size(arg_size),
> > -                              BPF_REG_FP,
> > -                              BPF_REG_0,
> > +                     emit_stx(prog, BPF_DW, BPF_REG_FP, BPF_REG_0,
> >                                -(stack_size - i * 8));
>
> But this is not.
> See https://godbolt.org/z/qW17f6cYe
> mov dword ptr [rsp], 6
>
> the compiler will store 32-bit only. The upper 32-bit are still garbage.

Enn......I didn't expect this case, and it seems
that this only happens on clang. With gcc,
"push 6" is used.

I haven't found a solution for this case, and it seems
not worth it to add an extra insn to clean the garbage
values.

Does anyone have any ideas here?

Thanks!
Menglong Dong

>
> >               }
> >
> > @@ -1918,7 +1914,7 @@ static void restore_regs(const struct btf_func_mo=
del *m, u8 **prog, int nr_regs,
> >                       next_same_struct =3D !next_same_struct;
> >               }
> >
> > -             emit_ldx(prog, bytes_to_bpf_size(arg_size),
> > +             emit_ldx(prog, BPF_DW,
> >                        i =3D=3D 5 ? X86_REG_R9 : BPF_REG_1 + i,
> >                        BPF_REG_FP,
> >                        -(stack_size - i * 8));
> > @@ -1949,12 +1945,9 @@ static void prepare_origin_stack(const struct bt=
f_func_model *m, u8 **prog,
> >               }
> >
> >               if (i > 5) {
> > -                     emit_ldx(prog, bytes_to_bpf_size(arg_size),
> > -                              BPF_REG_0, BPF_REG_FP,
> > +                     emit_ldx(prog, BPF_DW, BPF_REG_0, BPF_REG_FP,
> >                                (i - 6) * 8 + 0x18);
> > -                     emit_stx(prog, bytes_to_bpf_size(arg_size),
> > -                              BPF_REG_FP,
> > -                              BPF_REG_0,
> > +                     emit_stx(prog, BPF_DW, BPF_REG_FP, BPF_REG_0,
> >                                -(stack_size - (i - 6) * 8));
> >               }
> >
> > --
> > 2.40.1
> >