Reference
System call ABI
How SlopOS numbers its system calls, which registers they use, and what they return.
This page lists the rules of the SlopOS system call interface: numbering,
registers, return values and the families of calls. The authoritative table of
numbers is abi/src/syscall/numbers.rs (crate slopos_abi); errno values are
in abi/src/errno.rs. System calls explains how
the kernel handles a call.
Numbering
| Range | Contents | Rule |
|---|---|---|
0 to 472 (SYSCALL_LINUX_MAX) | Linux-numbered calls | Number, name, arguments and behaviour are those of the Linux x86-64 call |
1024 (SYSCALL_PRIVATE_BASE) upwards | SlopOS-only calls | Numbered 1024 + k, contiguous from k = 0; the private table holds 48 slots (SYSCALL_PRIVATE_TABLE_SIZE) |
- A number below 1024 always means the Linux call with that number. If SlopOS
does not implement it, the number returns
-ENOSYS; it is never reused for something else. - 1024 is above every number Linux has assigned and below the x32 marker bit
(bit 30), so a private number cannot be mistaken for either. ARM uses the same
scheme with
__ARM_NR_BASE. - SlopRing uses private numbers, not io_uring's (425 to 427), because it does not implement io_uring's interface.
scripts/check_syscall_abi.shenforces these rules. It compares every number below 1024 with a vendored copy of Linux'ssyscall_64.tbland fails if a name sits on a number Linux gave to another call, if the private range has a gap, if a number or name is used twice, or if a private name matches a Linux call without an allowlist entry explaining why.- Numbers changed in September 2026 (a breaking change). Programs built against older numbers do not run on current kernels.
Registers
The calling convention is the Linux x86-64 one.
| Register | Use |
|---|---|
rax | System call number on entry, return value on exit |
rdi, rsi, rdx, r10, r8, r9 | Arguments 1 to 6 |
rcx, r11 | Overwritten by the syscall instruction |
The fourth argument is in r10, not rcx, because the syscall instruction
stores the return address in rcx.
Return values and errors
- Success returns a value of zero or more.
- Failure returns a negated errno, so any value from
-4095to-1is an error. - Errno values use Linux's numbers (
EPERMis 1,ENOSYSis 38). - A number with no handler, in either range or outside both, returns
-ENOSYS. ERESTARTSYS(512) andERESTARTNOHAND(514) are kernel-internal and never reach a program. When a signal interrupts a call that returned one of them, the kernel either restarts the call after the signal handler runs or returns-EINTR, following Linux'sSA_RESTARTrules.
Families
Linux-numbered calls:
| Family | Calls |
|---|---|
| File I/O | read, write, open, openat, close, pread64, pwrite64, readv, writev, lseek, fcntl, ioctl, flock, fsync, fdatasync, truncate, ftruncate |
| Paths and metadata | stat, fstat, lstat, newfstatat, access, faccessat, faccessat2, getdents64, getcwd, chdir, fchdir, mkdir, mkdirat, rmdir, rename, renameat, link, linkat, unlink, unlinkat, symlink, symlinkat, readlink, readlinkat, chmod, fchmod, fchmodat, fchmodat2, mknod, mknodat, umask, utimensat, statfs, fstatfs |
| Descriptors and polling | dup, dup2, dup3, pipe, pipe2, poll, ppoll, select, pselect6, memfd_create, pidfd_open, signalfd4 |
| Memory | mmap, munmap, mprotect, msync, brk |
| Processes and threads | fork, clone, execve, wait4, exit, exit_group, getpid, getppid, gettid, getpgid, setpgid, getsid, setsid, getuid, getgid, geteuid, getegid, arch_prctl, futex, prlimit64, vhangup |
| Signals | rt_sigaction, rt_sigprocmask, rt_sigreturn, rt_sigsuspend, rt_sigtimedwait, rt_sigpending, rt_sigqueueinfo, rt_tgsigqueueinfo, kill, tgkill, sigaltstack |
| Scheduling and time | sched_yield, sched_setaffinity, sched_getaffinity, getcpu, getpriority, setpriority, nanosleep, clock_gettime, clock_settime |
| Networking | socket, bind, listen, accept, accept4, connect, sendto, recvfrom, sendmsg, recvmsg, getsockname, getpeername, setsockopt, getsockopt, shutdown |
| System | uname, sysinfo, getrandom, mount, umount2, sync, reboot |
SlopOS-only calls:
| Family | Calls |
|---|---|
| Introspection and logging | klog_write, sys_info, process_list, cpu_info, percpu_stats |
| Terminal | ctty_read |
| Process launch | spawn_path, sigdefault |
| Network control | resolve, net_query, net_iface_ctl, net_addr_ctl, net_route_ctl, net_resolver_set, net_monitor |
| Display and input | fb_info, fb_flip, cursor_set_image, cursor_move, set_display_mode, screen_acquire, input_sink_acquire, input_poll_batch, clipboard_copy, clipboard_paste, font_set, keymap_load, keymap_get_name |
| SlopRing | ring_setup, ring_enter, ring_register (see SlopRing) |
| Wheel of Fate | roulette, roulette_result, roulette_draw |
| Testing | test_report, run_userland_tests, test_panic |
| Firmware variables | efivar_get, efivar_set |
Each call also has a capability class fixed at compile time, which decides who may make it. Permissions lists the classes.
Rules for changing the ABI
abi/src/syscall/numbers.rsis the only place a number is written down. Userland imports the constant; documentation does not repeat numbers.- A Linux-numbered slot holds only the Linux call of that number, with Linux's arguments.
- A new SlopOS-only call takes the next free
1024 + k. When the private table is full,SYSCALL_PRIVATE_TABLE_SIZEgrows. - Every struct passed between a program and the kernel is
#[repr(C)]and has compile-time checks on its size and field offsets.
Add a syscall walks through adding one.
The Linux syscall(2)
and errno(3) man pages
document the conventions SlopOS follows, and Linux's
syscall_64.tbl
is the numbering it follows.