Skip to contentSkip to conversation

selftests/bpf: Account for zero extensions in patch list expectations

PATCH Applied Hungyu Lin 4 participants 4 messages Nov 3 – Nov 4

Subsystems (3)
lore
Share & tools
lore

Hungyu Lin

author
#0 lore raw

PATCH 1/1 selftests/bpf: Account for zero extensions in patch list expectations

patch_list_insert_in_front() checks the translated instructions around the call of an inlined bpf_loop() and its callback. On s390, the verifier adds zero extensions after the two 32-bit register writes in the loop. The call is therefore at instruction 25 rather than 23, and the later instructions checked by the test are also shifted by two. The test fails despite the call and jump offsets being correct.

Use the corresponding instruction indices on s390 and RISC-V, whose JITs request explicit zero extensions. Keep the original expectations for the other enabled architectures and retain all branch offset and callback entry checks.

Fixes 57fae4059380 ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf 7.4-rc1")
---
 .../selftests/bpf/progs/verifier_patch_list.c  | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)
tools/testing/selftests/bpf/progs/verifier_patch_list.c +18 -0
index 581b75b7b5c5..a9ad1dc56c3a 100644
@@ -298,6 +298,23 @@ __arch_arm64
 __arch_riscv64
 __arch_loongarch
 __success
+/* The verifier adds two zero extensions for these JITs. */
+#if defined(__TARGET_ARCH_s390) || defined(__TARGET_ARCH_riscv)
+__xlated("25: call pc+7")
+__xlated("...")
+__xlated("32: exit")
+__xlated("33: r0 = ")
+__xlated("35: r0 = &(void __percpu *)(r0)")
+__xlated("36: r0 = *(u64 *)(r0 +0)")
+__xlated("37: if r0 != 0x0 goto pc+2")
+__xlated("38: r0 = 1")
+__xlated("39: exit")
+__xlated("40: r1 += -1")
+__xlated("...")
+__xlated("51: if r1 != 0x0 goto pc-12")
+__xlated("52: r0 = 0")
+__xlated("53: exit")
+#else
 __xlated("23: call pc+7")
 __xlated("...")
 __xlated("30: exit")
@@ -312,6 +329,7 @@ __xlated("...")
 __xlated("49: if r1 != 0x0 goto pc-12")
 __xlated("50: r0 = 0")
 __xlated("51: exit")
+#endif
 __naked void patch_list_insert_in_front(void)
 {
 	asm volatile (

base-commit: d4bf9e34e299e45af8da2b0457b0ac3efbd7ae8d

Signature · 1 line
2.43.0
#1 lore raw
TL;DR: Commit message only: the Fixes: tag names a merge commit rather than the commit that introduced patch_list_insert_in_front(), and it lacks a preceding blank line.
Quoted 1 line

selftests/bpf: Account for zero extensions in patch list expectations

TL;DR: Commit message only: the Fixes: tag names a merge commit rather than the commit that introduced patch_list_insert_in_front(), and it lacks a preceding blank line.

Quoted 12 lines

selftests/bpf: Account for zero extensions in patch list expectations

patch_list_insert_in_front() checks the translated instructions around the call of an inlined bpf_loop() and its callback. On s390, the verifier adds zero extensions after the two 32-bit register writes in the loop. The call is therefore at instruction 25 rather than 23, and the later instructions checked by the test are also shifted by two. The test fails despite the call and jump offsets being correct.
Use the corresponding instruction indices on s390 and RISC-V, whose JITs request explicit zero extensions. Keep the original expectations for the other enabled architectures and retain all branch offset and callback entry checks.

Quoted 3 lines
Fixes 57fae4059380 ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf 7.4-rc1")

The Fixes: tag names 57fae4059380, which is a merge commit. Should it name the commit that introduced patch_list_insert_in_front() instead?

Fixes d86e6695cb2f ("selftests/bpf: Test the tables that follow patched and removed instructions")

Also, the Fixes: line follows the body paragraph directly. Could a blank line be added between them?

---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/41728315768

#2 lore raw
In reply to Hungyu Lin
Quoted 3 lines · Hungyu Lin

patch_list_insert_in_front() checks the translated instructions around the call of an inlined bpf_loop() and its callback. On s390, the verifier adds zero extensions after the two 32-bit register writes in the loop.

Thanks a bunch. Applied.

But could you follow up to
remove the two 32-bit writes from inline_bpf_loop().
'w7 = 0' and 'r7 = 0' do the same thing.
'w0 = -E2BIG' is an old bug. bpf_loop() returns long. The helper returns 0xfffffffffffffff9.
The inlined code returns 0xfffffff9.
So 'long ret = bpf_loop(...); if (ret < 0)' misses the error. The "over max limit" check in prog_tests/bpf_loop.c doesn't see it. The prog keeps the result in an int.

No other insn of this prog writes 32 bits.
With 64-bit writes the xlated code is the same on all five archs. Then no need for the #if and the second copy of the expectations.

Something like this (compile tested only):

kernel/bpf/fixups.c +2 -2
index 6b719a2..321624a 100644
@@ -2783,7 +2783,7 @@ static int inline_bpf_loop(struct bpf_verifier_env *env, int position,
 	 * expected number of iterations is too big.
 	 */
 	insn_buf[cnt++] = BPF_JMP_IMM(BPF_JLE, BPF_REG_1, BPF_MAX_LOOPS, 2);
-	insn_buf[cnt++] = BPF_MOV32_IMM(BPF_REG_0, -E2BIG);
+	insn_buf[cnt++] = BPF_MOV64_IMM(BPF_REG_0, -E2BIG);
 	insn_buf[cnt++] = BPF_JMP_IMM(BPF_JA, 0, 0, 16);
 	/* spill R6, R7, R8 to use these as loop vars */
 	insn_buf[cnt++] = BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_6, r6_offset);
@@ -2791,7 +2791,7 @@ static int inline_bpf_loop(struct bpf_verifier_env *env, int position,
 	insn_buf[cnt++] = BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_8, r8_offset);
 	/* initialize loop vars */
 	insn_buf[cnt++] = BPF_MOV64_REG(reg_loop_max, BPF_REG_1);
-	insn_buf[cnt++] = BPF_MOV32_IMM(reg_loop_cnt, 0);
+	insn_buf[cnt++] = BPF_MOV64_IMM(reg_loop_cnt, 0);
 	insn_buf[cnt++] = BPF_MOV64_REG(reg_loop_ctx, BPF_REG_3);
 	/* loop header,
 	 * if reg_loop_cnt >= reg_loop_max skip the loop body
#3 lore raw
In reply to Hungyu Lin
This patch was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

This patch was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:

Quoted 8 lines · On Tue, 3 Nov 2026 21:53:09 +0000 you wrote:

patch_list_insert_in_front() checks the translated instructions around the call of an inlined bpf_loop() and its callback. On s390, the verifier adds zero extensions after the two 32-bit register writes in the loop. The call is therefore at instruction 25 rather than 23, and the later instructions checked by the test are also shifted by two. The test fails despite the call and jump offsets being correct.

[...]

Here is the summary with links:
- [bpf-next] selftests/bpf: Account for zero extensions in patch list expectations

    https://git.kernel.org/bpf/bpf-next/c/8fb43ffc3327

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html