Re: [PATCH bpf-next v2 2/3] bpf: Remove 'may_goto 0' instruction in opt_remove_nops()

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

 






On 1/20/25 7:29 AM, Daniel Borkmann wrote:
On 1/18/25 8:20 PM, Yonghong Song wrote:
Since 'may_goto 0' insns are actually no-op, let us remove them.
Otherwise, verifier will generate code like
    /* r10 - 8 stores the implicit loop count */
    r11 = *(u64 *)(r10 -8)
    if r11 == 0x0 goto pc+2
    r11 -= 1
    *(u64 *)(r10 -8) = r11

which is the pure overhead.

The following code patterns (from the previous commit) are also
handled:
    may_goto 2
    may_goto 1
    may_goto 0

With this commit, the above three 'may_goto' insns are all
eliminated.

Signed-off-by: Yonghong Song <yonghong.song@xxxxxxxxx>
---
  kernel/bpf/verifier.c | 9 +++++++--
  1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 963dfda81c06..784547aa40a8 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -20187,20 +20187,25 @@ static const struct bpf_insn NOP = BPF_JMP_IMM(BPF_JA, 0, 0, 0);
    static int opt_remove_nops(struct bpf_verifier_env *env)
  {
+    const struct bpf_insn may_goto_0 = BPF_RAW_INSN(BPF_JMP | BPF_JCOND, 0, 0, 0, 0);
      const struct bpf_insn ja = NOP;
      struct bpf_insn *insn = env->prog->insnsi;
      int insn_cnt = env->prog->len;
+    bool is_may_goto_0, is_ja;
      int i, err;
        for (i = 0; i < insn_cnt; i++) {
-        if (memcmp(&insn[i], &ja, sizeof(ja)))
+        is_may_goto_0 = !memcmp(&insn[i], &may_goto_0, sizeof(may_goto_0));
+        is_ja = !memcmp(&insn[i], &ja, sizeof(ja));
+
+        if (!is_may_goto_0 && !is_ja)
              continue;

Why the extra may_goto_0 stack var?

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 245f1f3f1aec..16ba26295ec7 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -20185,16 +20185,19 @@ static int opt_remove_dead_code(struct bpf_verifier_env *env)
 }

 static const struct bpf_insn NOP = BPF_JMP_IMM(BPF_JA, 0, 0, 0);
+static const struct bpf_insn MAY_GOTO_0 = BPF_RAW_INSN(BPF_JMP | BPF_JCOND, 0, 0, 0, 0);

This actually is what I did initially. I changed to use the stack var because
NOP is used in other functions too while MAY_GOTO_0 is only used in
opt_remove_nops(). Certainly, using MAY_GOTO_0 as static variable works too.


 static int opt_remove_nops(struct bpf_verifier_env *env)
 {
-       const struct bpf_insn ja = NOP;
        struct bpf_insn *insn = env->prog->insnsi;
        int insn_cnt = env->prog->len;
+       bool is_ja, is_may_goto_0;
        int i, err;

        for (i = 0; i < insn_cnt; i++) {
-               if (memcmp(&insn[i], &ja, sizeof(ja)))
+               is_may_goto_0 = !memcmp(&insn[i], &MAY_GOTO_0, sizeof(MAY_GOTO_0));
+               is_ja         = !memcmp(&insn[i], &NOP, sizeof(NOP));
+               if (!is_may_goto_0 && !is_ja)
                        continue;

          err = verifier_remove_insns(env, i, 1);
          if (err)
              return err;
          insn_cnt--;
-        i--;
+        i -= (is_may_goto_0 && i > 0) ? 2 : 1;

Maybe add a comment for this logic?

Thanks Alexei for adding comments before merging!


Thanks,
Daniel





[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