Syscalls reference (curated)
Source of truth for numbers and semantics: src/kernel/syscalls.rs (syscall_dispatcher).
libc / app stubs: SDK/lib/cstd/src/sys.rs, p1ui.rs.
This table is not every arm in the dispatcher (there are duplicates and legacy stubs). It lists what normal apps and daemons rely on.
POSIX-shaped I/O and files
# |
Name |
Notes |
|---|
0 |
read
|
Resolves process-local FD; stdin reads the PTY / pipe path only — no direct PS/2 drain in read(0) (compositor + apps use poll_ui_events and/or ps2d → IPC). P1SCode: io.read_line() in p1sc.py lowers to repeated read(0, …) until a newline or EOF, then String::from_utf8_lossy (strips a trailing CR before LF when present). |
1 |
write
|
Resolve FD; fd 1/2 also push trimmed UTF-8 to system log; int 0x81 yield after stdout/stderr. |
2 |
open
|
NUL path; cwd-relative; returns process-local FD. |
3 |
close
|
|
4 |
stat
|
Two implementations exist in match — first full path-based wins for early arms; later stub may be unreachable (debt). |
5 |
fstat
|
|
8 |
lseek
|
Duplicate arms — same semantics. |
21 |
access
|
|
78 |
getdents
|
|
83 |
mkdir
|
FAT backend. |
87 |
unlink
|
FAT delete. |
Memory
# |
Name |
Notes |
|---|
9 |
mmap
|
First 9 => arm: per-task mmap_current growth, anon, invlpg. Later duplicate arm in file is unreachable (Rust match) — dead second implementation using pages_allocated + 0x4000_0000 base; do not rely on it. |
227 |
iomem_map
|
rdi = physical base (bytes; kernel page-aligns down), rsi = span length, rdx = flags: bit1 set = user executable (clears NX); bit2 set = write-back cacheable (low RAM / DMA window); clear = uncached (MMIO). Maps into the caller’s mmap arena; returns user VA for rdi (same offset within first page) or !0.
|
230 |
dma_alloc
|
rdi = byte size (page-rounded up), rsi = *mut u64 physical base out. Allocates physically contiguous frames, maps RW + NX into mmap arena, returns user VA or !0. 228 is already clock_gettime-style uptime — do not use 228 for DMA.
|
231 |
net_io
|
Delegated RTL8139 port I/O only in legacy userspace NIC mode (network/legacy_kernel_driver == 1). rdi = port (u16), rsi = op (0 inb … 5 outl), rdx = value for out ops. Port must lie in [io_base, io_base+0x100). Returns the read value or 0 after a write; !0 on fault. In kernel-owned mode (legacy == 0), returns !0. |
234 |
iomem_map_user
|
NIC IRQ subscriber only (netd). Maps another task’s user pages into the caller’s mmap arena: rdi = remote PID, rsi = remote user VA, rdx = byte length (≤1514), r10 = flags (same as 227). Returns page-aligned mapping base VA or !0. Caller **munmap**s the mapped span after use. |
12 |
brk
|
Query with 0; extend only; no shrink. |
Process / thread
# |
Name |
Notes |
|---|
39 |
getpid
|
|
56 |
clone
|
rdi = flags, rsi = child RSP (stack top), rdx = entry, r10 = arg → RDI at entry. Returns child tid.
|
59 |
execve
|
Path NUL; argv/envp optional; ramdisk fast path via ramdisk_bytes_if_cached. Spawns a new task (fresh PML4); rax = child PID; caller continues (not Linux exec replace). Stray TERMINATED tasks are reaped before load; teardown_sender runs only after SCHEDULER is dropped (process-lifecycle.md §2.2, ipc_present_maps.rs). |
60 |
exit
|
Shared cleanup with kill — see process-lifecycle.md. |
61 |
waitpid stub
|
Sets BLOCKED + wait_pid. |
62 |
kill
|
Phased: kill_recursively → quiesce → cleanup_process_memory → reap_task_fields → purge_target_dead. |
63 |
set_foreground_pid
|
|
64 |
get_foreground_pid
|
|
158 |
arch_prctl
|
ARCH_SET_FS (0x1002) for TLS.
|
Scheduling / time
# |
Name |
Notes |
|---|
24 |
yield
|
int 0x81.
|
25 |
sleep
|
Milliseconds; BLOCKED + yield. (Duplicate arms — same idea.) |
228 |
clock_gettime-style
|
clock_id=0 => realtime Unix epoch from software RTC (tv_nsec=0 today). Other IDs => monotonic uptime from kernel ticks.
|
300 |
get_time
|
Packed RTC fields (moved off 200 collision with raw net). |
IPC
# |
Name |
Notes |
|---|
30 |
legacy ipc_send |
Registers: target, type, p1–p3; no syscall-50 present remap. |
31 |
legacy ipc_recv |
|
50 |
ipc_send (P1-DBUS)
|
Pointer to IpcMessage; message_type == 260: netd RX injection (handled in-kernel, no compositor broker). 261: TX to netd — see 261 paragraph below (p1 / p2 / p3 / sender_pid). Otherwise target_pid == 0 → resolve deskgui.elf; types 2/5/18 cross-map framebuffer; then broker send. |
51 |
ipc_recv
|
rdi==0 → current PID; returns 1 if message, 0 if empty.
|
224 |
ipc_unmap_present_slot
|
Compositor: drop (current_pid, sender_pid) present map + unmap PTEs. |
225 |
cpu_core_load
|
Writes two u32 LE: online cores, busy mask. |
226 |
irq_subscribe
|
rdi = IRQ number (1 keyboard, 12 PS/2 mouse, NIC: PCI interrupt line, often 11 on QEMU pc + RTL8139), rsi = subscriber PID (must equal caller getpid), rdx = syscall-50 message_type. PS/2: raw inb(0x60) in p1. NIC: p1 = RTL8139 ISR bits after ACK (see kernel nic_interrupt_handler). One subscriber per IRQ. Delivery uses bounded spill queues when the IPC broker is contended (src/kernel/irq_subscribe.rs).
|
235 |
irq_unsubscribe
|
rdi = IRQ number. Drops the subscription if the caller is the current subscriber. 0 or !0. Used by netd when RegCube network/legacy_kernel_driver becomes 1 (release IRQ before any future kernel or alternate owner).
|
IPC message_type 250 / 251 (not syscall numbers)
On syscall 50, message_type == 250 is WIRE_PS2D_KEY_EVENT: ps2d → compositor decoded keyboard. p1 = legacy WM scancode byte; p2 = (modifiers << 32) | key; p3 = 1 if down (make), 0 if up (break). target_pid == 0 resolves to the compositor. Unrelated to syscall 250 (write_file) in the dispatcher table below.
message_type == 251 is WIRE_PS2D_MOUSE_EVENT: ps2d → compositor — p1 packs dx and dy as two i16 in the low 32 bits, p2 carries button bits and optional wheel dz. target_pid == 0 → compositor.
message_type == 260 (WIRE_NETD_RX_RAW): netd → kernel Ethernet payload. p1 = user VA of frame, p2 = length (14…1600, clamped). Kernel copies from sender address space into the normal RX path; this message is part of the legacy userspace-NIC mode.
message_type == 261 (WIRE_KERNEL_NETD_TX): TX to netd. p2 = Ethernet length (14…1514). p3 & 1 (phys mode): p1 = physical address of the first frame byte (kernel / staging); sender_pid usually 0 — netd uses sys_iomem_map (227) then munmap. Otherwise p1 = user VA in sender_pid (e.g. sys_net_send_raw) — netd uses sys_iomem_map_user (234). netd still copies into its DMA TX scratch for RTL8139 descriptors. No kernel bounce and no syscall 233.
message_type == 271 (WIRE_TIME_SVC): time-service control channel used by p1-timesvc. p1=1 requests sync-now; response uses p1=2 success or p1=3 failure.
Signals / PTY / job control
# |
Name |
Notes |
|---|
13 |
sigaction
|
Stub success. |
16 |
ioctl
|
VFS ioctl — PTY winsize, etc. |
220 |
openpty
|
Writes process-local master/slave fds. |
221 |
pty_rebind_std
|
Maps local 0,1,2 to slave global fd. |
223 |
signal_foreground
|
SIGINT/SIGKILL bits in sigmask; SIGKILL clears FOREGROUND_PID. |
Graphics / introspection (selected)
# |
Name |
Notes |
|---|
72 |
ptrace_peek
|
Identity-map assumption — security-sensitive. |
73 |
poll
|
TCP-centric pollfd scan. |
79 |
ptrace_poke
|
Identity-map assumption — was dead code when it shared number 73 with poll; use 79. |
100 |
init_graphics
|
Framebuffer info to user; heavy mapping path. |
101 |
poll_ui_events
|
|
102 |
ps
|
|
103 |
get_network_info
|
rdi = buffer, rsi = max length. 56 bytes: MAC, SLIRP-style IPv4/gateway/DNS, link, mask, TX/RX counters (legacy layout). If rsi ≥ 128, kernel also writes netd tail: u16 LE I/O base @56, PCI IRQ byte @58, last RX len @59 (≤64), preview @60–123, 0x813910EC @124–127 when RTL8139 present. Return value is bytes written (56 or 128) or !0.
|
104–107 |
RegCube get/set/flush/load |
|
98 |
klog
|
|
99 |
get_logs
|
rdi = buffer, rsi = max length. Writes UTF-8 text, NUL-terminated if room. One line per log entry. P1E2 (preferred): P1E2, code, level, timestamp, category (Kernel / Application / …), component (subsystem: PCI, TCP, netd, …), message (all fields after P1E2 separated by U+001F \x1f; message escaped). Legacy P1E1: same without component (viewers may infer). Plain text may still appear for non-snapshot output.
|
318 |
getrandom
|
RDRAND-backed. |
400 |
get_sys_mem_info
|
Total / free frames. |
RegCube (selected)
Hive |
Key |
Notes |
|---|
HIVE_LOCAL_MACHINE
|
network/legacy_kernel_driver
|
0 (default) = kernel owns RTL8139 (MMIO/DMA/IRQ in-kernel). 1 = legacy userspace netd owner mode (sys_net_io, IRQ subscribe, IPC 260/261 path). \\boot.cfg: NET=USERSPACE / NETWORK=USERSPACE forces 1. NET=LEGACY / NETWORK=LEGACY forces kernel-owned mode (no userspace NIC delegation).
|
HIVE_LOCAL_MACHINE
|
time/ntp_server
|
Hostname for SNTP sync (p1-timesvc / timectl), default time.cloudflare.com. |
HIVE_LOCAL_MACHINE
|
time/auto_sync
|
1/0 toggle for daemon periodic sync behavior.
|
HIVE_LOCAL_MACHINE
|
time/sync_interval_sec
|
Periodic sync interval in seconds (daemon clamps minimum to 60). |
HIVE_LOCAL_MACHINE
|
time/last_sync_status, time/last_sync_unix
|
Last SNTP outcome diagnostics. |
Networking
# |
Name |
Notes |
|---|
41 |
socket
|
TCP unless type 2 (UDP) or 3 (raw-ish TCP ref). |
42 |
connect
|
|
44 |
send
|
|
45 |
recv
|
len==0 probes TCP state.
|
49 |
bind
|
|
52 |
listen
|
Stub no-op success. |
200 |
net_send_raw
|
Validates Ethernet frame (14..1514). In kernel-owned mode (legacy_kernel_driver == 0), kernel copies from caller memory and transmits via rtl8139::kernel_tx_packet (or optimized IPv4 path). In legacy userspace mode (legacy_kernel_driver == 1), forwards TX to netd via IPC 261 with caller sender_pid. |
201 |
net_recv_raw
|
|
202 |
dns_resolve
|
|
203 |
net_config_set
|
Copies 16 bytes from caller (`ip |
204 |
net_event_subscribe
|
rdi=1 subscribe caller as net-event daemon; rdi=0 unsubscribe caller. Used by p1-netmgr.
|
210 |
tcp_state
|
|
211 |
tcp_retransmit_syn
|
|
Misc
# |
Name |
Notes |
|---|
20 |
writev
|
stdout/stderr only → system log. |
22 |
set_time / RTC refresh
|
|
250 |
write_file
|
Path + bytes → FAT write (legacy name). |
int 0x81
Fast reschedule path used from syscall layer and IRQ return plumbing — treat as yield/reschedule, not a Linux syscall number.
Kernel NIC smoke test (default path)
Ensure \\boot.cfg does not contain NET=USERSPACE.
Boot and confirm logs show kernel-owned mode ([NET] boot mode: KERNEL OWNER default ... and init says netd not spawned).
Exercise ping / DNS / sockets and verify get_network_info (103) counters change as traffic flows.
Optional legacy check: run with NET=USERSPACE and confirm netd subscription logs and IPC 260/261 activity.
