On 2/7/26 03:31, Andrii Nakryiko wrote: > On Mon, Jun 29, 2026 at 8:24 AM Leon Hwang <[email protected]> wrote: [...] >> diff --git a/kernel/bpf/fixups.c b/kernel/bpf/fixups.c >> index 3cf2cc6e3ab6..4f84d087ca69 100644 >> --- a/kernel/bpf/fixups.c >> +++ b/kernel/bpf/fixups.c >> @@ -1819,6 +1819,38 @@ int bpf_do_misc_fixups(struct bpf_verifier_env *env) >> goto next_insn; >> } >> >> + if (env->prog->jit_requested && >> + bpf_jit_supports_percpu_insn() && >> + insn->code == (BPF_LD | BPF_IMM | BPF_DW) && >> + (insn->src_reg == BPF_PSEUDO_MAP_VALUE || >> + insn->src_reg == BPF_PSEUDO_MAP_IDX_VALUE)) { >> + struct bpf_map *map; >> + >> + aux = &env->insn_aux_data[i + delta]; >> + map = env->used_maps[aux->map_index]; >> + if (map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY) >> + goto next_insn; >> + >> + /* >> + * Reuse the original ld_imm64 insn, and add one >> + * mov64_percpu_reg insn. >> + */ >> + >> + insn_buf[0] = insn[1]; >> + insn_buf[1] = BPF_MOV64_PERCPU_REG(insn->dst_reg, >> insn->dst_reg); >> + cnt = 2; >> + >> + i++; > > oof, this was a subtle head scratcher for me.. that i++ is easy to > miss. let's update the comment to be more explicit: we are *skipping* > first half of ld_imm64, patching over second half of it with that same > half + percpu mov. All because bpf_patch_insn_data() can only replace > one 8-byte instruction, which doesn't work well for ldimm64. > > Anyways, this looks correct, it just took me a bit to figure this out > and while the above comment warned me about this, it didn't really > make it any easier for figure out what's going on. >
Ack. Will update the comment. Thanks, Leon
