bpf: Get better reg range with ldsx and 32bit compare

With latest llvm19, the selftest iters/iter_arr_with_actual_elem_count
failed with -mcpu=v4.

The following are the details:
  0: R1=ctx() R10=fp0
  ; int iter_arr_with_actual_elem_count(const void *ctx) @ iters.c:1420
  0: (b4) w7 = 0                        ; R7_w=0
  ; int i, n = loop_data.n, sum = 0; @ iters.c:1422
  1: (18) r1 = 0xffffc90000191478       ; R1_w=map_value(map=iters.bss,ks=4,vs=1280,off=1144)
  3: (61) r6 = *(u32 *)(r1 +128)        ; R1_w=map_value(map=iters.bss,ks=4,vs=1280,off=1144) R6_w=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
  ; if (n > ARRAY_SIZE(loop_data.data)) @ iters.c:1424
  4: (26) if w6 > 0x20 goto pc+27       ; R6_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=32,var_off=(0x0; 0x3f))
  5: (bf) r8 = r10                      ; R8_w=fp0 R10=fp0
  6: (07) r8 += -8                      ; R8_w=fp-8
  ; bpf_for(i, 0, n) { @ iters.c:1427
  7: (bf) r1 = r8                       ; R1_w=fp-8 R8_w=fp-8
  8: (b4) w2 = 0                        ; R2_w=0
  9: (bc) w3 = w6                       ; R3_w=scalar(id=1,smin=smin32=0,smax=umax=smax32=umax32=32,var_off=(0x0; 0x3f)) R6_w=scalar(id=1,smin=smin32=0,smax=umax=smax32=umax32=32,var_off=(0x0; 0x3f))
  10: (85) call bpf_iter_num_new#45179          ; R0=scalar() fp-8=iter_num(ref_id=2,state=active,depth=0) refs=2
  11: (bf) r1 = r8                      ; R1=fp-8 R8=fp-8 refs=2
  12: (85) call bpf_iter_num_next#45181 13: R0=rdonly_mem(id=3,ref_obj_id=2,sz=4) R6=scalar(id=1,smin=smin32=0,smax=umax=smax32=umax32=32,var_off=(0x0; 0x3f)) R7=0 R8=fp-8 R10=fp0 fp-8=iter_num(ref_id=2,state=active,depth=1) refs=2
  ; bpf_for(i, 0, n) { @ iters.c:1427
  13: (15) if r0 == 0x0 goto pc+2       ; R0=rdonly_mem(id=3,ref_obj_id=2,sz=4) refs=2
  14: (81) r1 = *(s32 *)(r0 +0)         ; R0=rdonly_mem(id=3,ref_obj_id=2,sz=4) R1_w=scalar(smin=0xffffffff80000000,smax=0x7fffffff) refs=2
  15: (ae) if w1 < w6 goto pc+4 20: R0=rdonly_mem(id=3,ref_obj_id=2,sz=4) R1=scalar(smin=0xffffffff80000000,smax=smax32=umax32=31,umax=0xffffffff0000001f,smin32=0,var_off=(0x0; 0xffffffff0000001f)) R6=scalar(id=1,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=32,var_off=(0x0; 0x3f)) R7=0 R8=fp-8 R10=fp0 fp-8=iter_num(ref_id=2,state=active,depth=1) refs=2
  ; sum += loop_data.data[i]; @ iters.c:1429
  20: (67) r1 <<= 2                     ; R1_w=scalar(smax=0x7ffffffc0000007c,umax=0xfffffffc0000007c,smin32=0,smax32=umax32=124,var_off=(0x0; 0xfffffffc0000007c)) refs=2
  21: (18) r2 = 0xffffc90000191478      ; R2_w=map_value(map=iters.bss,ks=4,vs=1280,off=1144) refs=2
  23: (0f) r2 += r1
  math between map_value pointer and register with unbounded min value is not allowed

The source code:
  int iter_arr_with_actual_elem_count(const void *ctx)
  {
        int i, n = loop_data.n, sum = 0;

        if (n > ARRAY_SIZE(loop_data.data))
                return 0;

        bpf_for(i, 0, n) {
                /* no rechecking of i against ARRAY_SIZE(loop_data.n) */
                sum += loop_data.data[i];
        }

        return sum;
  }

The insn #14 is a sign-extenstion load which is related to 'int i'.
The insn #15 did a subreg comparision. Note that smin=0xffffffff80000000 and this caused later
insn #23 failed verification due to unbounded min value.

Actually insn #15 R1 smin range can be better. Before insn #15, we have
  R1_w=scalar(smin=0xffffffff80000000,smax=0x7fffffff)
With the above range, we know for R1, upper 32bit can only be 0xffffffff or 0.
Otherwise, the value range for R1 could be beyond [smin=0xffffffff80000000,smax=0x7fffffff].

After insn #15, for the true patch, we know smin32=0 and smax32=32. With the upper 32bit 0xffffffff,
then the corresponding value is [0xffffffff00000000, 0xffffffff00000020]. The range is
obviously beyond the original range [smin=0xffffffff80000000,smax=0x7fffffff] and the
range is not possible. So the upper 32bit must be 0, which implies smin = smin32 and
smax = smax32.

This patch fixed the issue by adding additional register deduction after 32-bit compare
insn such that if the signed 32-bit register range is non-negative and 64-bit smin is
{S32/S16/S8}_MIN and 64-bit max is no greater than {U32/U16/U8}_MAX.
Here, we check smin with {S32/S16/S8}_MIN since this is the most common result related to
signed extension load.

With this patch, iters/iter_arr_with_actual_elem_count succeeded with better register range:

from 15 to 20: R0=rdonly_mem(id=7,ref_obj_id=2,sz=4) R1_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=31,var_off=(0x0; 0x1f)) R6=scalar(id=1,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=32,var_off=(0x0; 0x3f)) R7=scalar(id=9,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) R8=scalar(id=9,smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) R10=fp0 fp-8=iter_num(ref_id=2,state=active,depth=3) refs=2

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>

kernel/bpf/verifier.c

@@ -2182,6 +2182,21 @@ static void __reg_deduce_mixed_bounds(struct bpf_reg_state *reg)
 		reg->smin_value = max_t(s64, reg->smin_value, new_smin);
 		reg->smax_value = min_t(s64, reg->smax_value, new_smax);
 	}
+
+	/* if s32 range is non-negative and s64 range is in [S32/S16/S8_MIN, <= S32/S16/S8_MAX],
+	 * the s64/u64 range can be refined.
+	 */
+	if (reg->s32_min_value >= 0) {
+		if ((reg->smin_value == S32_MIN && reg->smax_value <= S32_MAX) ||
+		    (reg->smin_value == S16_MIN && reg->smax_value <= S16_MAX) ||
+		    (reg->smin_value == S8_MIN && reg->smax_value <= S8_MAX)) {
+			reg->smin_value = reg->umin_value = reg->s32_min_value;
+			reg->smax_value = reg->umax_value = reg->s32_max_value;
+			reg->var_off = tnum_intersect(reg->var_off,
+						      tnum_range(reg->smin_value,
+								 reg->smax_value));
+		}
+	}
 }
 
 static void __reg_deduce_bounds(struct bpf_reg_state *reg)