Tests: Rank the ctx-field case on a scalar field, not the packet pointer

xdp_md.data is a u32 in C, but the verifier tracks it as a packet
pointer and rejects 32-bit arithmetic on it: "R0 32-bit pointer
arithmetic prohibited", the same for a C program, which is why C casts
it through (void *)(long) first. The u32 ranking the test pins is the
correct one; it just has to be pinned on a field that is a scalar to the
verifier too. ingress_ifindex is.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Author: r41k0u

tests/passing_tests/vmlinux/ctx_field_rank.py

@@ -1,7 +1,12 @@
 # A sub-register context field is loaded widened to i64, but C ranks it by
-# its declared width: xdp_md.data is a u32, so data - k with a c_int32 k is
-# a u32 subtraction (the result is cut to 32 bits and zero-extended), not a
-# 64-bit one.
+# its declared width: xdp_md.ingress_ifindex is a u32, so ifindex - k with a
+# c_int32 k is a u32 subtraction (the result is cut to 32 bits and
+# zero-extended), not a 64-bit one.
+#
+# Not xdp_md.data: it is a u32 in C too, but the verifier tracks it as a
+# packet pointer and rejects 32-bit arithmetic on it ("R0 32-bit pointer
+# arithmetic prohibited"), for C programs as much as for this one. That is
+# why C casts it through (void *)(long) before doing anything with it.
 from ctypes import c_int32, c_int64
 from pythonbpf import bpf, section, bpfglobal, compile
 from vmlinux import struct_xdp_md
@@ -11,7 +16,7 @@
 @section("xdp")
 def prog(ctx: struct_xdp_md) -> c_int64:
     k = c_int32(1)
-    d = ctx.data - k
+    d = ctx.ingress_ifindex - k
     return c_int64(d)