]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
LoongArch: BPF: Sign-extend struct ops return values properly
authorHengqi Chen <hengqi.chen@gmail.com>
Thu, 2 Oct 2025 14:39:53 +0000 (22:39 +0800)
committerHuacai Chen <chenhuacai@loongson.cn>
Thu, 2 Oct 2025 14:39:53 +0000 (22:39 +0800)
The ns_bpf_qdisc selftest triggers a kernel panic:

  Oops[#1]:
  CPU 0 Unable to handle kernel paging request at virtual address 0000000000741d58, era == 90000000851b5ac0, ra == 90000000851b5aa4
  CPU: 0 UID: 0 PID: 449 Comm: test_progs Tainted: G           OE       6.16.0+ #3 PREEMPT(full)
  Tainted: [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
  Hardware name: QEMU QEMU Virtual Machine, BIOS unknown 2/2/2022
  pc 90000000851b5ac0 ra 90000000851b5aa4 tp 90000001076b8000 sp 90000001076bb600
  a0 0000000000741ce8 a1 0000000000000001 a2 90000001076bb5c0 a3 0000000000000008
  a4 90000001004c4620 a5 9000000100741ce8 a6 0000000000000000 a7 0100000000000000
  t0 0000000000000010 t1 0000000000000000 t2 9000000104d24d30 t3 0000000000000001
  t4 4f2317da8a7e08c4 t5 fffffefffc002f00 t6 90000001004c4620 t7 ffffffffc61c5b3d
  t8 0000000000000000 u0 0000000000000001 s9 0000000000000050 s0 90000001075bc800
  s1 0000000000000040 s2 900000010597c400 s3 0000000000000008 s4 90000001075bc880
  s5 90000001075bc8f0 s6 0000000000000000 s7 0000000000741ce8 s8 0000000000000000
     ra: 90000000851b5aa4 __qdisc_run+0xac/0x8d8
    ERA: 90000000851b5ac0 __qdisc_run+0xc8/0x8d8
   CRMD: 000000b0 (PLV0 -IE -DA +PG DACF=CC DACM=CC -WE)
   PRMD: 00000004 (PPLV0 +PIE -PWE)
   EUEN: 00000007 (+FPE +SXE +ASXE -BTE)
   ECFG: 00071c1d (LIE=0,2-4,10-12 VS=7)
  ESTAT: 00010000 [PIL] (IS= ECode=1 EsubCode=0)
   BADV: 0000000000741d58
   PRID: 0014c010 (Loongson-64bit, Loongson-3A5000)
  Modules linked in: bpf_testmod(OE) [last unloaded: bpf_testmod(OE)]
  Process test_progs (pid: 449, threadinfo=000000009af02b3a, task=00000000e9ba4956)
  Stack : 0000000000000000 90000001075bc8ac 90000000869524a8 9000000100741ce8
          90000001075bc800 9000000100415300 90000001075bc8ac 0000000000000000
          900000010597c400 900000008694a000 0000000000000000 9000000105b59000
          90000001075bc800 9000000100741ce8 0000000000000050 900000008513000c
          9000000086936000 0000000100094d4c fffffff400676208 0000000000000000
          9000000105b59000 900000008694a000 9000000086bf0dc0 9000000105b59000
          9000000086bf0d68 9000000085147010 90000001075be788 0000000000000000
          9000000086bf0f98 0000000000000001 0000000000000010 9000000006015840
          0000000000000000 9000000086be6c40 0000000000000000 0000000000000000
          0000000000000000 4f2317da8a7e08c4 0000000000000101 4f2317da8a7e08c4
          ...
  Call Trace:
  [<90000000851b5ac0>] __qdisc_run+0xc8/0x8d8
  [<9000000085130008>] __dev_queue_xmit+0x578/0x10f0
  [<90000000853701c0>] ip6_finish_output2+0x2f0/0x950
  [<9000000085374bc8>] ip6_finish_output+0x2b8/0x448
  [<9000000085370b24>] ip6_xmit+0x304/0x858
  [<90000000853c4438>] inet6_csk_xmit+0x100/0x170
  [<90000000852b32f0>] __tcp_transmit_skb+0x490/0xdd0
  [<90000000852b47fc>] tcp_connect+0xbcc/0x1168
  [<90000000853b9088>] tcp_v6_connect+0x580/0x8a0
  [<90000000852e7738>] __inet_stream_connect+0x170/0x480
  [<90000000852e7a98>] inet_stream_connect+0x50/0x88
  [<90000000850f2814>] __sys_connect+0xe4/0x110
  [<90000000850f2858>] sys_connect+0x18/0x28
  [<9000000085520c94>] do_syscall+0x94/0x1a0
  [<9000000083df1fb8>] handle_syscall+0xb8/0x158

  Code: 4001ad80  2400873f  2400832d <240073cc001137ff  001133ff  6407b41f  001503cc  0280041d

  ---[ end trace 0000000000000000 ]---

The bpf_fifo_dequeue prog returns a skb which is a pointer. The pointer
is treated as a 32bit value and sign extend to 64bit in epilogue. This
behavior is right for most bpf prog types but wrong for struct ops which
requires LoongArch ABI.

So let's sign extend struct ops return values according to the LoongArch
ABI ([1]) and return value spec in function model.

[1]: https://loongson.github.io/LoongArch-Documentation/LoongArch-ELF-ABI-EN.html

Cc: stable@vger.kernel.org
Fixes: 6abf17d690d8 ("LoongArch: BPF: Add struct ops support for trampoline")
Signed-off-by: Hengqi Chen <hengqi.chen@gmail.com>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
arch/loongarch/net/bpf_jit.c

index fa1a3234e9a60a3108f35e01fe77fff2a30deaa2..cbe53d0b7fb063351ce774948acf41a7cb2d2d3c 100644 (file)
@@ -1448,6 +1448,37 @@ void arch_free_bpf_trampoline(void *image, unsigned int size)
        bpf_prog_pack_free(image, size);
 }
 
+/*
+ * Sign-extend the register if necessary
+ */
+static void sign_extend(struct jit_ctx *ctx, int rd, int rj, u8 size, bool sign)
+{
+       /* ABI requires unsigned char/short to be zero-extended */
+       if (!sign && (size == 1 || size == 2)) {
+               if (rd != rj)
+                       move_reg(ctx, rd, rj);
+               return;
+       }
+
+       switch (size) {
+       case 1:
+               emit_insn(ctx, extwb, rd, rj);
+               break;
+       case 2:
+               emit_insn(ctx, extwh, rd, rj);
+               break;
+       case 4:
+               emit_insn(ctx, addiw, rd, rj, 0);
+               break;
+       case 8:
+               if (rd != rj)
+                       move_reg(ctx, rd, rj);
+               break;
+       default:
+               pr_warn("bpf_jit: invalid size %d for sign_extend\n", size);
+       }
+}
+
 static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_image *im,
                                         const struct btf_func_model *m, struct bpf_tramp_links *tlinks,
                                         void *func_addr, u32 flags)
@@ -1655,8 +1686,12 @@ static int __arch_prepare_bpf_trampoline(struct jit_ctx *ctx, struct bpf_tramp_i
                restore_args(ctx, m->nr_args, args_off);
 
        if (save_ret) {
-               emit_insn(ctx, ldd, LOONGARCH_GPR_A0, LOONGARCH_GPR_FP, -retval_off);
                emit_insn(ctx, ldd, regmap[BPF_REG_0], LOONGARCH_GPR_FP, -(retval_off - 8));
+               if (is_struct_ops)
+                       sign_extend(ctx, LOONGARCH_GPR_A0, regmap[BPF_REG_0],
+                                   m->ret_size, m->ret_flags & BTF_FMODEL_SIGNED_ARG);
+               else
+                       emit_insn(ctx, ldd, LOONGARCH_GPR_A0, LOONGARCH_GPR_FP, -retval_off);
        }
 
        emit_insn(ctx, ldd, LOONGARCH_GPR_S1, LOONGARCH_GPR_FP, -sreg_off);