On Mon, Sep 27, 2021 at 08:29:38PM +0530, Kumar Kartikeya Dwivedi wrote:
> +static void emit_relo_kfunc_btf(struct bpf_gen *gen, struct ksym_relo_desc *relo, int insn)
> +{
> + struct kfunc_desc *kdesc;
> + int btf_fd_idx;
> +
> + kdesc = get_kfunc_desc(gen, relo->name);
> + if (!kdesc)
> + return;
> +
> + btf_fd_idx = kdesc->ref > 1 ? kdesc->off : add_kfunc_btf_fd(gen);
> + if (btf_fd_idx > INT16_MAX) {
> + pr_warn("BTF fd off %d for kfunc %s exceeds INT16_MAX, cannot process relocation\n",
> + btf_fd_idx, relo->name);
> + gen->error = -E2BIG;
> + return;
> + }
> + /* load slot pointer */
> + emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_8, BPF_PSEUDO_MAP_IDX_VALUE,
> + 0, 0, 0, blob_fd_array_off(gen, btf_fd_idx)));
> + /* Try to map one insn->off to one insn->imm */
> + if (kdesc->ref > 1) {
> + emit(gen, BPF_MOV64_REG(BPF_REG_9, BPF_REG_7));
> + goto skip_btf_fd;
> + } else {
> + /* cannot use index == 0 */
> + if (!btf_fd_idx) {
> + btf_fd_idx = add_kfunc_btf_fd(gen);
> + /* shift to next slot */
> + emit(gen, BPF_ALU64_IMM(BPF_ADD, BPF_REG_8, sizeof(int)));
> + }
> + kdesc->off = btf_fd_idx;
> + }
> +
> + /* set a default value */
> + emit(gen, BPF_ST_MEM(BPF_W, BPF_REG_8, 0, 0));
> + emit(gen, BPF_MOV64_REG(BPF_REG_9, BPF_REG_7));
> + /* store BTF fd if ret < 0 */
> + emit(gen, BPF_JMP_IMM(BPF_JSLT, BPF_REG_7, 0, 3));
> + emit(gen, BPF_ALU64_IMM(BPF_RSH, BPF_REG_7, 32));
> + /* store BTF fd in slot */
> + emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_8, BPF_REG_7, 0));
> + emit(gen, BPF_MOV64_REG(BPF_REG_7, BPF_REG_9));
> +skip_btf_fd:
> + /* remember BTF fd to skip insn->off store for vmlinux case */
> + emit(gen, BPF_ALU64_IMM(BPF_RSH, BPF_REG_9, 32));
> + /* set a default value */
> + emit(gen, BPF_ST_MEM(BPF_H, BPF_REG_0, offsetof(struct bpf_insn, off), 0));
> + /* skip insn->off store if ret < 0 */
> + emit(gen, BPF_JMP_IMM(BPF_JSLT, BPF_REG_7, 0, 2));
> + /* skip if vmlinux BTF */
> + emit(gen, BPF_JMP_IMM(BPF_JEQ, BPF_REG_9, 0, 1));
> + /* store index into insn[insn_idx].off */
> + emit(gen, BPF_ST_MEM(BPF_H, BPF_REG_0, offsetof(struct bpf_insn, off), btf_fd_idx));
> + /* close fd that we skipped storing in fd_array */
> + if (kdesc->ref > 1) {
> + emit(gen, BPF_MOV64_REG(BPF_REG_1, BPF_REG_9));
> + __emit_sys_close(gen);
> + }
kdesc->ref optimization is neat, but I wonder why it's done half way.
The generated code still calls bpf_btf_find_by_name_kind() and the kernel
allocates the new FD. Then the loader prog simply ignores that FD and closes it.
May be don't call the helper at all and just copy imm+off pair from
already populated insn?
I was thinking to do this optimization for emit_relo() for other cases,
but didn't have time to do it.
Since the code is doing this optimization I think it's worth doing it cleanly.
Or just don't do it at all.
It's great that there is a test for it in the patch 12 :)
