[Bug target/123502] xdp-tools fails tests for bpf (bitwise operator &= on pointer prohibited)

[jemarch at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123502

--- Comment #8 from Jose E. Marchesi <jemarch at gcc dot gnu.org> ---
It is the combiner that transforms:

  (insn 5 4 6 2 (set (reg:DI 22)
          (lshiftrt:DI (reg/f:DI 21)
              (const_int 6 [0x6]))) "foo.c":9:1 44 {lshrdi3}
       (expr_list:REG_DEAD (reg/f:DI 21)
          (nil)))
  (insn 6 5 7 2 (set (reg/f:DI 23)
          (ashift:DI (reg:DI 22)
              (const_int 6 [0x6]))) "foo.c":9:1 42 {ashldi3}
       (expr_list:REG_DEAD (reg:DI 22)
          (nil)))

into the AND.

However, in the kernel verifier:

        case BPF_AND:
        case BPF_OR:
        case BPF_XOR:
                /* bitwise ops on pointers are troublesome, prohibit. */
                verbose(env, "R%d bitwise operator %s on pointer prohibited\n",
                        dst, bpf_alu_string[opcode >> 4]);
                return -EACCES;
        default:
                /* other operators (e.g. MUL,LSH) produce non-pointer results
*/
                verbose(env, "R%d pointer arithmetic with %s operator
prohibited\n",
                        dst, bpf_alu_string[opcode >> 4]);
                return -EACCES;

So even if we somehow made the combiner to not transform into the AND,
the resulting code would still be non-verifiable.