Merge branch 'bpf-improve-linked-register-tracking'

Puranjay Mohan says:

====================
bpf: Improve linked register tracking

V3: https://lore.kernel.org/all/20260203222643.994713-1-puranjay@kernel.org/
Changes in v3->v4:
- Add a call to reg_bounds_sync() in sync_linked_regs() to sync bounds after alu op.
- Add a __sink(path[0]); in the C selftest so compiler doesn't say  "error: variable 'path' set but
  not used"

V2: https://lore.kernel.org/all/20260113152529.3217648-1-puranjay@kernel.org/
Changes in v2->v3:
- Added another selftest showing a real usage pattern
- Rebased on bpf-next/master

v1: https://lore.kernel.org/bpf/20260107203941.1063754-1-puranjay@kernel.org/
Changes in v1->v2:
- Add support for alu32 operations in linked register tracking (Alexei)
- Squash the selftest fix with the first patch (Eduard)
- Add more selftests to detect edge cases

This series extends the BPF verifier's linked register tracking to handle
negative offsets, BPF_SUB operations, and alu32 operations, enabling better bounds propagation for
common arithmetic patterns.

The verifier previously only tracked positive constant deltas between linked
registers using BPF_ADD. This meant patterns using negative offsets or
subtraction couldn't benefit from bounds propagation:

  void alu32_negative_offset(void)
  {
          volatile char path[5];
          volatile int offset = bpf_get_prandom_u32();
          int off = offset;

          if (off >= 5 && off < 10)
                  path[off - 5] = '.';
  }

this gets compiled to:

0000000000000478 <alu32_negative_offset>:
     143:       call 0x7
     144:       *(u32 *)(r10 - 0xc) = w0
     145:       w1 = *(u32 *)(r10 - 0xc)
     146:       w2 = w1				// w2 and w1 share the same id
     147:       w2 += -0x5			// verifier knows w1 = w2 + 5
     148:       if w2 > 0x4 goto +0x5 <L0>	// in fall-through: verifier knows w2 ∈ [0,4] => w1 ∈ [5, 9]
     149:       r2 = r10
     150:       r2 += -0x5			// r2 = fp - 5
     151:       r2 += r1			// r2 = fp - 5 + r1 (∈ [5, 9]) => r2 ∈ [fp, fp + 4]
     152:       w1 = 0x2e
     153:       *(u8 *)(r2 - 0x5) = w1		// r2 ∈ [fp, fp + 4] => r2 - 5 ∈ [fp - 5, fp - 1]
<L0>:
     154:       exit

After the changes, the verifier could link 32-bit scalars and also supported -ve offsets for linking:

	146:       w2 = w1
	147:       w2 += -0x5

It allowed the verifier to correctly propagate bounds, without the
changes in this patchset, verifier would reject this program with:

invalid unbounded variable-offset write to stack R2

This program has been added as a selftest in the second patch.

Veristat comparison on programs from sched_ext, selftests, and some meta internal programs:

Scx Progs
File               Program           Verdict (A)  Verdict (B)  Verdict (DIFF)  Insns (A)  Insns (B)  Insns  (DIFF)
-----------------  ----------------  -----------  -----------  --------------  ---------  ---------  -------------
scx_layered.bpf.o  layered_runnable  success      success      MATCH                5674       6077  +403 (+7.10%)

FB Progs
File          Program           Verdict (A)  Verdict (B)  Verdict (DIFF)  Insns (A)  Insns (B)  Insns      (DIFF)
------------  ----------------  -----------  -----------  --------------  ---------  ---------  -----------------
bpf232.bpf.o  layered_dump      success      success      MATCH                1151       1218       +67 (+5.82%)
bpf257.bpf.o  layered_runnable  success      success      MATCH                5743       6143      +400 (+6.97%)
bpf252.bpf.o  layered_runnable  success      success      MATCH                5677       6075      +398 (+7.01%)
bpf227.bpf.o  layered_dump      success      success      MATCH                 915        982       +67 (+7.32%)
bpf239.bpf.o  layered_runnable  success      success      MATCH                5459       5861      +402 (+7.36%)
bpf246.bpf.o  layered_runnable  success      success      MATCH                5562       6008      +446 (+8.02%)
bpf229.bpf.o  layered_runnable  success      success      MATCH                2559       3011     +452 (+17.66%)
bpf231.bpf.o  layered_runnable  success      success      MATCH                2559       3011     +452 (+17.66%)
bpf234.bpf.o  layered_runnable  success      success      MATCH                2549       3001     +452 (+17.73%)
bpf019.bpf.o  do_sendmsg        success      success      MATCH              124823     153523   +28700 (+22.99%)
bpf019.bpf.o  do_parse          success      success      MATCH              124809     153509   +28700 (+23.00%)
bpf227.bpf.o  layered_runnable  success      success      MATCH                1915       2356     +441 (+23.03%)
bpf228.bpf.o  layered_runnable  success      success      MATCH                1700       2152     +452 (+26.59%)
bpf232.bpf.o  layered_runnable  success      success      MATCH                1499       1951     +452 (+30.15%)

bpf312.bpf.o  mount_exit        success      success      MATCH               19253      62883  +43630 (+226.61%)
bpf312.bpf.o  umount_exit       success      success      MATCH               19253      62883  +43630 (+226.61%)
bpf311.bpf.o  mount_exit        success      success      MATCH               19226      62863  +43637 (+226.97%)
bpf311.bpf.o  umount_exit       success      success      MATCH               19226      62863  +43637 (+226.97%)

The above four programs have specific patters that make the verifier explore a lot more states:

    for (; depth < MAX_DIR_DEPTH; depth++) {
        const unsigned char* name = BPF_CORE_READ(dentry, d_name.name);
        if (offset >= MAX_PATH_LEN - MAX_DIR_LEN) {
            return depth;
        }
        int len = bpf_probe_read_kernel_str(&path[offset], MAX_DIR_LEN, name);
        offset += len;
        if (len == MAX_DIR_LEN) {
            if (offset - 2 < MAX_PATH_LEN) {   // <---- (a)
                path[offset - 2] = '.';
            }
            if (offset - 3 < MAX_PATH_LEN) {   // <---- (b)
                path[offset - 3] = '.';
            }
            if (offset - 4 < MAX_PATH_LEN) {   // <---- (c)
                path[offset - 4] = '.';
            }
        }
    }

When at some depth == N false branches of conditions (a), (b) and (c) are scheduled for
verification, constraints for offset at depth == N+1 are:
1. offset >= MAX_PATH_LEN + 2
2. offset >= MAX_PATH_LEN + 3
3. offset >= MAX_PATH_LEN + 4 (visited before others)

And after offset += len it becomes:
1. offset >= MAX_PATH_LEN - 4093
2. offset >= MAX_PATH_LEN - 4092
3. offset >= MAX_PATH_LEN - 4091 (visited before others)

Because of the DFS states exploration logic, the states above are visited in order 3, 2, 1; 3 is not
a subset of 2 and 1 is not a subset of 2, so pruning logic does not kick in.
Previously this was not a problem, because range for offset was not propagated through the
statements (a), (b), (c).

As the root cause of this regression is understood, this is not a blocker for this change.

Selftest Progs
File                                Program                   Verdict (A)  Verdict (B)  Verdict (DIFF)  Insns (A)  Insns (B)  Insns   (DIFF)
----------------------------------  ------------------------  -----------  -----------  --------------  ---------  ---------  --------------
linked_list_peek.bpf.o              list_peek                 success      success      MATCH                 152         88   -64 (-42.11%)
verifier_iterating_callbacks.bpf.o  cond_break2               success      success      MATCH                 110         88   -22 (-20.00%)

These are the added selftests that failed earlier but are passing now:

verifier_linked_scalars.bpf.o       alu32_negative_offset     failure      success      MISMATCH               11         13    +2 (+18.18%)
verifier_linked_scalars.bpf.o       scalars_alu32_big_offset  failure      success      MISMATCH                7         10    +3 (+42.86%)
verifier_linked_scalars.bpf.o       scalars_neg_alu32_add     failure      success      MISMATCH                7         10    +3 (+42.86%)
verifier_linked_scalars.bpf.o       scalars_neg_alu32_sub     failure      success      MISMATCH                7         10    +3 (+42.86%)
verifier_linked_scalars.bpf.o       scalars_neg               failure      success      MISMATCH                7         10    +3 (+42.86%)
verifier_linked_scalars.bpf.o       scalars_neg_sub           failure      success      MISMATCH                7         10    +3 (+42.86%)
verifier_linked_scalars.bpf.o       scalars_sub_neg_imm       failure      success      MISMATCH                7         10    +3 (+42.86%)

iters.bpf.o                         iter_obfuscate_counter    success      success      MATCH                  83        119   +36 (+43.37%)
bpf_cubic.bpf.o                     bpf_cubic_acked           success      success      MATCH                 243        430  +187 (+76.95%)
====================

Link: https://patch.msgid.link/20260204151741.2678118-1-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

4 files changed, 346 insertions, 15 deletions

diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h
index 746025df82c8..ef8e45a362d9 100644
--- a/include/linux/bpf_verifier.h
+++ b/include/linux/bpf_verifier.h
@@ -147,8 +147,12 @@ struct bpf_reg_state {
 	 * registers. Example:
 	 * r1 = r2;    both will have r1->id == r2->id == N
 	 * r1 += 10;   r1->id == N | BPF_ADD_CONST and r1->off == 10
+	 * r3 = r2;    both will have r3->id == r2->id == N
+	 * w3 += 10;   r3->id == N | BPF_ADD_CONST32 and r3->off == 10
 	 */
-#define BPF_ADD_CONST (1U << 31)
+#define BPF_ADD_CONST64 (1U << 31)
+#define BPF_ADD_CONST32 (1U << 30)
+#define BPF_ADD_CONST (BPF_ADD_CONST64 | BPF_ADD_CONST32)
 	u32 id;
 	/* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
 	 * from a pointer-cast helper, bpf_sk_fullsock() and
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 92e03a5a50f5..edf5342b982f 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -16209,6 +16209,13 @@ static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
 		verbose(env, "verifier internal error: no src_reg\n");
 		return -EFAULT;
 	}
+	/*
+	 * For alu32 linked register tracking, we need to check dst_reg's
+	 * umax_value before the ALU operation. After adjust_scalar_min_max_vals(),
+	 * alu32 ops will have zero-extended the result, making umax_value <= U32_MAX.
+	 */
+	u64 dst_umax = dst_reg->umax_value;
+
 	err = adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg);
 	if (err)
 		return err;
@@ -16218,26 +16225,44 @@ static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
 	 * r1 += 0x1
 	 * if r2 < 1000 goto ...
 	 * use r1 in memory access
-	 * So for 64-bit alu remember constant delta between r2 and r1 and
-	 * update r1 after 'if' condition.
+	 * So remember constant delta between r2 and r1 and update r1 after
+	 * 'if' condition.
 	 */
 	if (env->bpf_capable &&
-	    BPF_OP(insn->code) == BPF_ADD && !alu32 &&
-	    dst_reg->id && is_reg_const(src_reg, false)) {
-		u64 val = reg_const_value(src_reg, false);
+	    (BPF_OP(insn->code) == BPF_ADD || BPF_OP(insn->code) == BPF_SUB) &&
+	    dst_reg->id && is_reg_const(src_reg, alu32)) {
+		u64 val = reg_const_value(src_reg, alu32);
+		s32 off;
+
+		if (!alu32 && ((s64)val < S32_MIN || (s64)val > S32_MAX))
+			goto clear_id;
+
+		if (alu32 && (dst_umax > U32_MAX))
+			goto clear_id;
 
-		if ((dst_reg->id & BPF_ADD_CONST) ||
-		    /* prevent overflow in sync_linked_regs() later */
-		    val > (u32)S32_MAX) {
+		off = (s32)val;
+
+		if (BPF_OP(insn->code) == BPF_SUB) {
+			/* Negating S32_MIN would overflow */
+			if (off == S32_MIN)
+				goto clear_id;
+			off = -off;
+		}
+
+		if (dst_reg->id & BPF_ADD_CONST) {
 			/*
 			 * If the register already went through rX += val
 			 * we cannot accumulate another val into rx->off.
 			 */
+clear_id:
 			dst_reg->off = 0;
 			dst_reg->id = 0;
 		} else {
-			dst_reg->id |= BPF_ADD_CONST;
-			dst_reg->off = val;
+			if (alu32)
+				dst_reg->id |= BPF_ADD_CONST32;
+			else
+				dst_reg->id |= BPF_ADD_CONST64;
+			dst_reg->off = off;
 		}
 	} else {
 		/*
@@ -17334,7 +17359,7 @@ static void sync_linked_regs(struct bpf_verifier_env *env, struct bpf_verifier_s
 			u32 saved_id = reg->id;
 
 			fake_reg.type = SCALAR_VALUE;
-			__mark_reg_known(&fake_reg, (s32)reg->off - (s32)known_reg->off);
+			__mark_reg_known(&fake_reg, (s64)reg->off - (s64)known_reg->off);
 
 			/* reg = known_reg; reg += delta */
 			copy_register_state(reg, known_reg);
@@ -17349,6 +17374,9 @@ static void sync_linked_regs(struct bpf_verifier_env *env, struct bpf_verifier_s
 			scalar32_min_max_add(reg, &fake_reg);
 			scalar_min_max_add(reg, &fake_reg);
 			reg->var_off = tnum_add(reg->var_off, fake_reg.var_off);
+			if (known_reg->id & BPF_ADD_CONST32)
+				zext_32_to_64(reg);
+			reg_bounds_sync(reg);
 		}
 		if (e->is_reg)
 			mark_reg_scratched(env, e->regno);
diff --git a/tools/testing/selftests/bpf/progs/verifier_bounds.c b/tools/testing/selftests/bpf/progs/verifier_bounds.c
index 411a18437d7e..560531404bce 100644
--- a/tools/testing/selftests/bpf/progs/verifier_bounds.c
+++ b/tools/testing/selftests/bpf/progs/verifier_bounds.c
@@ -1477,7 +1477,7 @@ __naked void sub64_full_overflow(void)
 SEC("socket")
 __description("64-bit subtraction, partial overflow, result in unbounded reg")
 __success __log_level(2)
-__msg("3: (1f) r3 -= r2 {{.*}} R3=scalar()")
+__msg("3: (1f) r3 -= r2 {{.*}} R3=scalar(id=1-1)")
 __retval(0)
 __naked void sub64_partial_overflow(void)
 {
diff --git a/tools/testing/selftests/bpf/progs/verifier_linked_scalars.c b/tools/testing/selftests/bpf/progs/verifier_linked_scalars.c
index 5f41bbb730a7..2ef346c827c2 100644
--- a/tools/testing/selftests/bpf/progs/verifier_linked_scalars.c
+++ b/tools/testing/selftests/bpf/progs/verifier_linked_scalars.c
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include <linux/bpf.h>
+#include <limits.h>
 #include <bpf/bpf_helpers.h>
 #include "bpf_misc.h"
 
@@ -18,9 +19,9 @@ __naked void scalars(void)
 	r4 = r1;				\
 	w2 += 0x7FFFFFFF;			\
 	w4 += 0;				\
-	if r2 == 0 goto l1;			\
+	if r2 == 0 goto l0_%=;			\
 	exit;					\
-l1:						\
+l0_%=:						\
 	r4 >>= 63;				\
 	r3 = 1;					\
 	r3 -= r4;				\
@@ -64,4 +65,302 @@ l0_%=:								\
 	: __clobber_all);
 }
 
+SEC("socket")
+__success
+__naked void scalars_neg(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r0 &= 0xff;					\
+	r1 = r0;					\
+	r1 += -4;					\
+	if r1 s< 0 goto l0_%=;				\
+	if r0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/* Same test but using BPF_SUB instead of BPF_ADD with negative immediate */
+SEC("socket")
+__success
+__naked void scalars_neg_sub(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r0 &= 0xff;					\
+	r1 = r0;					\
+	r1 -= 4;					\
+	if r1 s< 0 goto l0_%=;				\
+	if r0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/* alu32 with negative offset */
+SEC("socket")
+__success
+__naked void scalars_neg_alu32_add(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	w0 &= 0xff;					\
+	w1 = w0;					\
+	w1 += -4;					\
+	if w1 s< 0 goto l0_%=;				\
+	if w0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/* alu32 with negative offset using SUB */
+SEC("socket")
+__success
+__naked void scalars_neg_alu32_sub(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	w0 &= 0xff;					\
+	w1 = w0;					\
+	w1 -= 4;					\
+	if w1 s< 0 goto l0_%=;				\
+	if w0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/* Positive offset: r1 = r0 + 4, then if r1 >= 6, r0 >= 2, so r0 != 0 */
+SEC("socket")
+__success
+__naked void scalars_pos(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r0 &= 0xff;					\
+	r1 = r0;					\
+	r1 += 4;					\
+	if r1 < 6 goto l0_%=;				\
+	if r0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/* SUB with negative immediate: r1 -= -4 is equivalent to r1 += 4 */
+SEC("socket")
+__success
+__naked void scalars_sub_neg_imm(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r0 &= 0xff;					\
+	r1 = r0;					\
+	r1 -= -4;					\
+	if r1 < 6 goto l0_%=;				\
+	if r0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/* Double ADD clears the ID (can't accumulate offsets) */
+SEC("socket")
+__failure
+__msg("div by zero")
+__naked void scalars_double_add(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r0 &= 0xff;					\
+	r1 = r0;					\
+	r1 += 2;					\
+	r1 += 2;					\
+	if r1 < 6 goto l0_%=;				\
+	if r0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/*
+ * Test that sync_linked_regs() correctly handles large offset differences.
+ * r1.off = S32_MIN, r2.off = 1, delta = S32_MIN - 1 requires 64-bit math.
+ */
+SEC("socket")
+__success
+__naked void scalars_sync_delta_overflow(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r0 &= 0xff;					\
+	r1 = r0;					\
+	r2 = r0;					\
+	r1 += %[s32_min];				\
+	r2 += 1;					\
+	if r2 s< 100 goto l0_%=;			\
+	if r1 s< 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32),
+	  [s32_min]"i"(INT_MIN)
+	: __clobber_all);
+}
+
+/*
+ * Another large delta case: r1.off = S32_MAX, r2.off = -1.
+ * delta = S32_MAX - (-1) = S32_MAX + 1 requires 64-bit math.
+ */
+SEC("socket")
+__success
+__naked void scalars_sync_delta_overflow_large_range(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r0 &= 0xff;					\
+	r1 = r0;					\
+	r2 = r0;					\
+	r1 += %[s32_max];				\
+	r2 += -1;					\
+	if r2 s< 0 goto l0_%=;				\
+	if r1 s>= 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32),
+	  [s32_max]"i"(INT_MAX)
+	: __clobber_all);
+}
+
+/*
+ * Test linked scalar tracking with alu32 and large positive offset (0x7FFFFFFF).
+ * After w1 += 0x7FFFFFFF, w1 wraps to negative for any r0 >= 1.
+ * If w1 is signed-negative, then r0 >= 1, so r0 != 0.
+ */
+SEC("socket")
+__success
+__naked void scalars_alu32_big_offset(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	w0 &= 0xff;					\
+	w1 = w0;					\
+	w1 += 0x7FFFFFFF;				\
+	if w1 s>= 0 goto l0_%=;				\
+	if w0 != 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+SEC("socket")
+__failure
+__msg("div by zero")
+__naked void scalars_alu32_basic(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	r1 = r0;					\
+	w1 += 1;					\
+	if r1 > 10 goto 1f;				\
+	r0 >>= 32;					\
+	if r0 == 0 goto 1f;				\
+	r0 /= 0;					\
+1:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+/*
+ * Test alu32 linked register tracking with wrapping.
+ * R0 is bounded to [0xffffff00, 0xffffffff] (high 32-bit values)
+ * w1 += 0x100 causes R1 to wrap to [0, 0xff]
+ *
+ * After sync_linked_regs, if bounds are computed correctly:
+ *   R0 should be [0x00000000_ffffff00, 0x00000000_ffffff80]
+ *   R0 >> 32 == 0, so div by zero is unreachable
+ *
+ * If bounds are computed incorrectly (64-bit underflow):
+ *   R0 becomes [0xffffffff_ffffff00, 0xffffffff_ffffff80]
+ *   R0 >> 32 == 0xffffffff != 0, so div by zero is reachable
+ */
+SEC("socket")
+__success
+__naked void scalars_alu32_wrap(void)
+{
+	asm volatile ("					\
+	call %[bpf_get_prandom_u32];			\
+	w0 |= 0xffffff00;				\
+	r1 = r0;					\
+	w1 += 0x100;					\
+	if r1 > 0x80 goto l0_%=;			\
+	r2 = r0;					\
+	r2 >>= 32;					\
+	if r2 == 0 goto l0_%=;				\
+	r0 /= 0;					\
+l0_%=:							\
+	r0 = 0;						\
+	exit;						\
+"	:
+	: __imm(bpf_get_prandom_u32)
+	: __clobber_all);
+}
+
+SEC("socket")
+__success
+void alu32_negative_offset(void)
+{
+	volatile char path[5];
+	volatile int offset = bpf_get_prandom_u32();
+	int off = offset;
+
+	if (off >= 5 && off < 10)
+		path[off - 5] = '.';
+
+	/* So compiler doesn't say: error: variable 'path' set but not used */
+	__sink(path[0]);
+}
+
 char _license[] SEC("license") = "GPL";