nx_madvise_probe.nx

nx_madvise_probe.nx -- SYSCALL-IDENTITY PROBE: does raw __syscall(28) reach x86_64 madvise(2)? WHY RAW 28: the portable rv64/asm-generic number is 233 and x86ctx_rv64_to_x86_64_syscall carries no 233 row yet, so a portable const would fall through `return num` to x86_64 233 = epoll_ctl -- the wrong-syscall-not-an-error class (setpgid/flock precedent). Raw 28 is the sys_exit_group pattern (raw x86 231) and is conflict-checked: nothing in the tree calls __syscall(28,...) today. PROOF SHAPE (print the numbers, never just a verdict): T1 positive: MADV_WILLNEED(3) on a live 1MB anonymous mapping -> expect 0 T2 negative-control: MADV_WILLNEED on the unmapped page at 4096 -> expect -12 (ENOMEM) T3 identity-control: advice=9999 on the live mapping -> expect -22 (EINVAL) The TRIPLE (0,-12,-22) is a madvise signature; any other syscall answers differently (epoll_ctl would say EBADF/EINVAL on T1). exit 0 = MADVISE-LIVE, 1 = WRONG-SYSCALL with the numbers printed. license_tier: ORIGINAL No hw writes (Rule 26).