Re: [PATCH bpf-next v2 05/15] bpf: Support new signed div/mod instructions.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 7/18/23 7:44 PM, Alexei Starovoitov wrote:
On Tue, Jul 18, 2023 at 7:31 PM Yonghong Song <yhs@xxxxxxxx> wrote:



On 7/18/23 4:00 PM, Eduard Zingerman wrote:
On Wed, 2023-07-12 at 23:07 -0700, Yonghong Song wrote:
Add interpreter/jit support for new signed div/mod insns.
The new signed div/mod instructions are encoded with
unsigned div/mod instructions plus insn->off == 1.
Also add basic verifier support to ensure new insns get
accepted.

Signed-off-by: Yonghong Song <yhs@xxxxxx>
---
   arch/x86/net/bpf_jit_comp.c | 27 +++++++----
   kernel/bpf/core.c           | 96 ++++++++++++++++++++++++++++++-------
   kernel/bpf/verifier.c       |  6 ++-
   3 files changed, 103 insertions(+), 26 deletions(-)

diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index adda5e7626b4..3176b60d25c7 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -1194,15 +1194,26 @@ static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image, u8 *rw_image
                              /* mov rax, dst_reg */
                              emit_mov_reg(&prog, is64, BPF_REG_0, dst_reg);

-                    /*
-                     * xor edx, edx
-                     * equivalent to 'xor rdx, rdx', but one byte less
-                     */
-                    EMIT2(0x31, 0xd2);
+                    if (insn->off == 0) {
+                            /*
+                             * xor edx, edx
+                             * equivalent to 'xor rdx, rdx', but one byte less
+                             */
+                            EMIT2(0x31, 0xd2);

-                    /* div src_reg */
-                    maybe_emit_1mod(&prog, src_reg, is64);
-                    EMIT2(0xF7, add_1reg(0xF0, src_reg));
+                            /* div src_reg */
+                            maybe_emit_1mod(&prog, src_reg, is64);
+                            EMIT2(0xF7, add_1reg(0xF0, src_reg));
+                    } else {
+                            if (BPF_CLASS(insn->code) == BPF_ALU)
+                                    EMIT1(0x99); /* cltd */
+                            else
+                                    EMIT2(0x48, 0x99); /* cqto */

Nitpick: I can't find names cltd/cqto in the Intel instruction manual,
           instead it uses names cdq/cqo for these instructions.
           (See Vol. 2A pages 3-315 and 3-497)

I got these asm names from
    https://defuse.ca/online-x86-assembler.htm
I will check the Intel insn manual and make the change
accordingly.

Heh. I've been using the same.
Most of the comments in the x86 JIT code are from there :)

and it actually returns 99 -> cdq, 4899 -> cqo

cltd/cqto must be aliases ?

Yes, cltd/cqto are aliases.

From llvm-project repo:

[~/work/llvm-project/llvm/lib/Target/X86 (ni23)]$ egrep -r cltd
egrep: warning: egrep is obsolescent; using grep -E
README.txt:        cltd
X86InstrAsmAlias.td:def : MnemonicAlias<"cdq",  "cltd", "att">;
X86InstrExtension.td: "{cltd|cdq}", []>, OpSize32, Sched<[WriteALU]>;

[~/work/llvm-project/llvm/lib/Target/X86 (ni23)]$ egrep -r cqto
egrep: warning: egrep is obsolescent; using grep -E
X86InstrAsmAlias.td:def : MnemonicAlias<"cqo",  "cqto", "att">;
X86InstrExtension.td: "{cqto|cqo}", []>, Sched<[WriteALU]>, Requires<[In64BitMode]>;

I will change to use cdq and cgo.




[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux