feat(kernel): seL4-style IPC, notifications, status ABI

- badges + Mint, Call/Reply/ReplyRecv w/ gen-checked reply slot
- cap transfer (1/msg, copy, needs Delegate, rechecked at delivery)
- notifications: signal/wait/poll/bind, irq -> notification routing
- WaitAny/WaitForInterrupt removed; a0=status, ErrorCode
- libkos: single ecall primitive, compat wrappers + typed new api
- fix: trap return clobbered user t0 on every trap
- fix: wait queues FIFO, delegate keeps badge, fault cap checked
- docs/ipc.md spec, ipc tests, ipc fuzz, kani ntfn + waitqueue

crates/adapter-riscv/src/trap/mod.rs

@@ -75,7 +75,6 @@ _restore_smode:
     ld x1,   8(sp)
     ld x3,  24(sp)
     ld x4,  32(sp)
-    ld x5,  40(sp)
     ld x6,  48(sp)
     ld x7,  56(sp)
     ld x8,  64(sp)
@@ -103,13 +102,19 @@ _restore_smode:
     ld x30, 240(sp)
     ld x31, 248(sp)
 
-    addi sp, sp, 272
+    # Every GPR except t0 (x5) is restored. Decide where we return to while
-
+    # t0 is still free, and restore it last on each path; using t0 after
+    # its restore would corrupt the interrupted program's t0.
     csrr t0, sstatus
     andi t0, t0, 0x100
     bnez t0, _sret_smode
+    ld x5,  40(sp)
+    addi sp, sp, 272
     csrrw sp, sscratch, sp
+    sret
 _sret_smode:
+    ld x5,  40(sp)
+    addi sp, sp, 272
     sret
 "#
 );
@@ -139,156 +144,138 @@ pub fn set_kernel_stack(kernel_sp: usize) {
     }
 }
 
+/// Decodes the syscall in `frame`, or `None` for an unknown number or an
+/// argument that is not representable (e.g. a slot out of range).
 fn decode_syscall(frame: &TrapFrame) -> Option<domain::syscall::Syscall> {
     use domain::message::Message;
-    use domain::syscall::{Syscall, WaitSet};
+    use domain::syscall::Syscall;
+    use types::address::{DataRef, EntryPoint, StackPointer, VirtualAddress};
+    use types::badge::Badge;
+    use types::bounded::{CapSlot, PageCount};
+    use types::permission::PermissionSet;
     use types::syscall_abi::*;
 
     let nr = frame.regs[17]; // a7
-    let a0 = frame.regs[10];
+    let a = |i: usize| frame.regs[10 + i];
-    let a1 = frame.regs[11];
+    let slot = |i: usize| CapSlot::new(u8::try_from(a(i)).ok()?);
-    let a2 = frame.regs[12];
+    let message = || -> Option<Message> {
-    let a3 = frame.regs[13];
+        let msg = Message::from_words([a(1), a(2), a(3), a(4)]);
-    let a4 = frame.regs[14];
+        Some(match a(5) {
+            NONE => msg,
+            _ => msg.with_cap(slot(5)?),
+        })
+    };
 
-    match nr {
+    Some(match nr {
-        SYS_EXIT => Some(Syscall::Exit),
+        SYS_EXIT => Syscall::Exit,
-        SYS_HALT => Some(Syscall::Halt),
+        SYS_HALT => Syscall::Halt,
-        SYS_EXIT_THREAD => Some(Syscall::ExitThread),
+        SYS_EXIT_THREAD => Syscall::ExitThread,
-        SYS_YIELD => Some(Syscall::Yield),
+        SYS_YIELD => Syscall::Yield,
-        SYS_GET_TIME => Some(Syscall::GetTime),
+        SYS_GET_TIME => Syscall::GetTime,
-        SYS_GET_PROCESS_ID => Some(Syscall::GetProcessId),
+        SYS_GET_PROCESS_ID => Syscall::GetProcessId,
-        SYS_SEND => {
+        SYS_SEND => Syscall::Send {
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+            endpoint: slot(0)?,
-            let msg = Message::from_words([a1, a2, a3, a4]);
+            message: message()?,
-            Some(Syscall::Send {
+        },
-                endpoint: slot,
+        SYS_CALL => Syscall::Call {
-                message: msg,
+            endpoint: slot(0)?,
-            })
+            message: message()?,
-        }
+        },
-        SYS_RECEIVE => {
+        SYS_RECEIVE => Syscall::Receive { endpoint: slot(0)? },
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+        SYS_REPLY => Syscall::Reply {
-            Some(Syscall::Receive { endpoint: slot })
+            message: message()?,
-        }
+        },
-        SYS_DELEGATE => {
+        SYS_REPLY_RECV => Syscall::ReplyRecv {
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+            endpoint: slot(0)?,
-            let to = types::bounded::CapSlot::new(a1 as u8)?;
+            message: message()?,
-            let perms = types::permission::PermissionSet::from_bits(a2 as u8);
+        },
-            Some(Syscall::Delegate {
+        SYS_DELEGATE => Syscall::Delegate {
-                slot,
+            slot: slot(0)?,
-                to_endpoint: to,
+            to_endpoint: slot(1)?,
-                permissions: perms,
+            permissions: PermissionSet::from_bits(u8::try_from(a(2)).ok()?),
-            })
+        },
-        }
+        SYS_MINT => Syscall::Mint {
-        SYS_WAIT_FOR_INTERRUPT => {
+            slot: slot(0)?,
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+            badge: Badge::new(a(1)),
-            Some(Syscall::WaitForInterrupt { irq: slot })
+            permissions: PermissionSet::from_bits(u8::try_from(a(2)).ok()?),
-        }
+        },
-        SYS_ACKNOWLEDGE_INTERRUPT => {
+        SYS_CREATE_NOTIFICATION => Syscall::CreateNotification,
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+        SYS_SIGNAL => Syscall::Signal {
-            Some(Syscall::AcknowledgeInterrupt { irq: slot })
+            notification: slot(0)?,
-        }
+        },
-        SYS_WAIT_ANY => {
+        SYS_WAIT => Syscall::Wait {
-            let ep0 = if a0 != usize::MAX {
+            notification: slot(0)?,
-                types::bounded::CapSlot::new(a0 as u8)
+        },
-            } else {
+        SYS_POLL => Syscall::Poll {
-                None
+            notification: slot(0)?,
-            };
+        },
-            let ep1 = if a1 != usize::MAX {
+        SYS_BIND_NOTIFICATION => Syscall::BindNotification {
-                types::bounded::CapSlot::new(a1 as u8)
+            notification: slot(0)?,
-            } else {
+        },
-                None
+        SYS_SET_IRQ_NOTIFICATION => Syscall::SetIrqNotification {
-            };
+            irq: slot(0)?,
-            let ep2 = if a2 != usize::MAX {
+            notification: slot(1)?,
-                types::bounded::CapSlot::new(a2 as u8)
+        },
-            } else {
+        SYS_ACKNOWLEDGE_INTERRUPT => Syscall::AcknowledgeInterrupt { irq: slot(0)? },
-                None
+        SYS_CREATE_PROCESS => Syscall::CreateProcess {
-            };
+            create_cap: slot(0)?,
-            let ep3 = if a3 != usize::MAX {
+            image: DataRef::empty(),
-                types::bounded::CapSlot::new(a3 as u8)
+        },
-            } else {
+        SYS_KILL => Syscall::Kill { process: slot(0)? },
-                None
+        SYS_REVOKE => Syscall::Revoke { slot: slot(0)? },
-            };
+        SYS_DROP_CAP => Syscall::DropCap { slot: slot(0)? },
-            let sources = WaitSet::new([ep0, ep1, ep2, ep3], [None, None, None, None]);
+        SYS_SPAWN_THREAD => Syscall::SpawnThread {
-            Some(Syscall::WaitAny { sources })
+            entry: EntryPoint::from_raw(a(0)),
-        }
+            stack: StackPointer::from_raw(a(1)),
-        SYS_CREATE_PROCESS => {
+        },
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+        SYS_REQUEST_PAGES => Syscall::RequestPages {
-            Some(Syscall::CreateProcess {
+            count: PageCount::new(u16::try_from(a(0)).ok()?)?,
-                create_cap: slot,
+        },
-                image: types::address::DataRef::empty(),
+        SYS_RELEASE_PAGES => Syscall::ReleasePages {
-            })
+            addr: VirtualAddress::from_raw(a(0)),
-        }
+            count: PageCount::new(u16::try_from(a(1)).ok()?)?,
-        SYS_KILL => {
+        },
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+        SYS_SPAWN_THREAD_IN => Syscall::SpawnThreadIn {
-            Some(Syscall::Kill { process: slot })
+            process: slot(0)?,
-        }
+            entry: EntryPoint::from_raw(a(1)),
-        SYS_REVOKE => {
+            stack: StackPointer::from_raw(a(2)),
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
+        },
-            Some(Syscall::Revoke { slot })
+        SYS_CREATE_ENDPOINT => Syscall::CreateEndpoint,
-        }
+        SYS_DEBUG_PUTCHAR => Syscall::DebugPutchar { byte: a(0) as u8 },
-        SYS_SPAWN_THREAD => Some(Syscall::SpawnThread {
+        SYS_MAP_PAGE => Syscall::MapPage {
-            entry: types::address::EntryPoint::from_raw(a0),
+            process: slot(0)?,
-            stack: types::address::StackPointer::from_raw(a1),
+            src_va: VirtualAddress::from_raw(a(1)),
-        }),
+            dst_va: VirtualAddress::from_raw(a(2)),
-        SYS_REQUEST_PAGES => {
+            executable: a(3) != 0,
-            let count = types::bounded::PageCount::new(a0 as u16)?;
+        },
-            Some(Syscall::RequestPages { count })
+        _ => return None,
-        }
+    })
-        SYS_RELEASE_PAGES => {
-            let count = types::bounded::PageCount::new(a1 as u16)?;
-            Some(Syscall::ReleasePages {
-                addr: types::address::VirtualAddress::from_raw(a0),
-                count,
-            })
-        }
-        SYS_SPAWN_THREAD_IN => {
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
-            Some(Syscall::SpawnThreadIn {
-                process: slot,
-                entry: types::address::EntryPoint::from_raw(a1),
-                stack: types::address::StackPointer::from_raw(a2),
-            })
-        }
-        SYS_CREATE_ENDPOINT => Some(Syscall::CreateEndpoint),
-        SYS_DROP_CAP => {
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
-            Some(Syscall::DropCap { slot })
-        }
-        SYS_DEBUG_PUTCHAR => Some(Syscall::DebugPutchar { byte: a0 as u8 }),
-        SYS_MAP_PAGE => {
-            let slot = types::bounded::CapSlot::new(a0 as u8)?;
-            Some(Syscall::MapPage {
-                process: slot,
-                src_va: types::address::VirtualAddress::from_raw(a1),
-                dst_va: types::address::VirtualAddress::from_raw(a2),
-                executable: a3 != 0,
-            })
-        }
-        _ => None,
-    }
 }
 
-fn write_result(frame: &mut TrapFrame, outcome: &kernel::SyscallOutcome) {
+/// Writes a completed syscall's result per the register ABI: status in
-    use kernel::{SyscallOutcome, SyscallResult};
+/// `a0`, results in `a1..`.
+fn write_result(frame: &mut TrapFrame, result: &kernel::SyscallResult) {
+    use kernel::SyscallResult;
+    use kernel::syscall::DeliveryKind;
+    use types::syscall_abi::{KIND_CALL, KIND_MESSAGE, KIND_SIGNAL, NONE};
 
-    match outcome {
+    let r = &mut frame.regs;
-        SyscallOutcome::Complete(result) => match result {
+    r[10] = 0;
-            SyscallResult::Ok => frame.regs[10] = 0,
+    match result {
-            SyscallResult::Value(v) => frame.regs[10] = *v,
+        SyscallResult::Ok => {}
-            SyscallResult::Slot(s) => frame.regs[10] = s.value() as usize,
+        SyscallResult::Value(v) => r[11] = *v,
-            SyscallResult::Message(m) => {
+        SyscallResult::Slot(s) => r[11] = s.value() as usize,
-                frame.regs[10] = m.word(0);
+        SyscallResult::Received(d) => {
-                frame.regs[11] = m.word(1);
+            r[11] = d.badge;
-                frame.regs[12] = m.word(2);
+            r[12..16].copy_from_slice(&d.words);
-                frame.regs[13] = m.word(3);
+            r[16] = d.cap.map_or(NONE, |s| s.value() as usize);
-            }
+            r[17] = match d.kind {
-            SyscallResult::Error(_) => frame.regs[10] = usize::MAX,
+                DeliveryKind::Message => KIND_MESSAGE,
-            SyscallResult::WakeReason(_) => frame.regs[10] = 0,
+                DeliveryKind::Call => KIND_CALL,
-        },
+                DeliveryKind::Signal => KIND_SIGNAL,
-        SyscallOutcome::BlockCaller(_) => {}
+            };
-        SyscallOutcome::SwitchTo(_) => frame.regs[10] = 0,
+        }
-        SyscallOutcome::TerminateCaller => {}
+        SyscallResult::Error(e) => r[10] = kernel::syscall::error_code(e) as usize,
     }
 }
 
@@ -310,24 +297,7 @@ fn load_context_to_frame(ctx: &domain::thread::SavedContext, frame: &mut TrapFra
 fn apply_pending_result(frame: &mut TrapFrame, tid: types::id::ThreadId) {
     let kern = unsafe { &mut *crate::kernel_ptr() };
     if let Some(result) = kern.take_pending_result(tid) {
-        write_result_direct(frame, &result);
+        write_result(frame, &result);
-    }
-}
-
-fn write_result_direct(frame: &mut TrapFrame, result: &kernel::SyscallResult) {
-    use kernel::SyscallResult;
-    match result {
-        SyscallResult::Ok => frame.regs[10] = 0,
-        SyscallResult::Value(v) => frame.regs[10] = *v,
-        SyscallResult::Slot(s) => frame.regs[10] = s.value() as usize,
-        SyscallResult::Message(m) => {
-            frame.regs[10] = m.word(0);
-            frame.regs[11] = m.word(1);
-            frame.regs[12] = m.word(2);
-            frame.regs[13] = m.word(3);
-        }
-        SyscallResult::Error(_) => frame.regs[10] = usize::MAX,
-        SyscallResult::WakeReason(_) => frame.regs[10] = 0,
     }
 }
 
@@ -441,14 +411,17 @@ fn handle_exception(frame: &mut TrapFrame, scause: usize) {
         CAUSE_ECALL_U => {
             frame.sepc += 4;
 
-            if let Some(syscall) = decode_syscall(frame) {
+            let Some(syscall) = decode_syscall(frame) else {
+                frame.regs[10] = types::syscall_abi::ErrorCode::UnknownSyscall as usize;
+                return;
+            };
+            {
                 let kern = unsafe { &mut *crate::kernel_ptr() };
                 let current_tid = kern.current_thread();
                 let outcome = kern.handle_syscall(syscall);
-                write_result(frame, &outcome);
 
                 match outcome {
-                    kernel::SyscallOutcome::Complete(_) => {}
+                    kernel::SyscallOutcome::Complete(result) => write_result(frame, &result),
                     kernel::SyscallOutcome::BlockCaller(_) => {
                         if let Some(tid) = current_tid {
                             switch_to_next(frame, tid);

crates/domain/src/capability/mod.rs

@@ -1,4 +1,4 @@
-use types::{bounded::CapSlot, generation::Generation, permission::PermissionSet};
+use types::{badge::Badge, bounded::CapSlot, generation::Generation, permission::PermissionSet};
 
 pub mod authority;
 
@@ -11,10 +11,12 @@ mod tests;
 // Sized so that every representable `CapSlot` indexes a real slot.
 pub const MAX_CAPS_PER_PROCESS: usize = CapSlot::MAX as usize + 1;
 
+#[derive(Clone)]
 pub struct Capability {
     resource: Resource,
     permissions: PermissionSet,
     generation: Generation,
+    badge: Badge,
 }
 
 impl Capability {
@@ -23,9 +25,20 @@ impl Capability {
             resource,
             permissions,
             generation,
+            badge: Badge::NONE,
         }
     }
 
+    /// The same capability with `badge` stamped on it.
+    pub fn with_badge(mut self, badge: Badge) -> Self {
+        self.badge = badge;
+        self
+    }
+
+    pub fn badge(&self) -> Badge {
+        self.badge
+    }
+
     pub fn resource(&self) -> &Resource {
         &self.resource
     }

crates/domain/src/endpoint/mod.rs

@@ -57,6 +57,14 @@ impl Endpoint {
         self.waiting_receivers.pop()
     }
 
+    pub fn is_sender(&self, thread: ThreadId) -> bool {
+        self.waiting_senders.contains(thread)
+    }
+
+    pub fn is_receiver(&self, thread: ThreadId) -> bool {
+        self.waiting_receivers.contains(thread)
+    }
+
     pub fn has_waiting_sender(&self) -> bool {
         !self.waiting_senders.is_empty()
     }

crates/domain/src/endpoint/tests.rs

@@ -19,21 +19,21 @@ fn test_endpoint() -> Endpoint {
 
 #[test]
 fn empty_queue_pop_returns_none() {
-    let mut queue = WaitQueue::new();
+    let mut queue: WaitQueue = WaitQueue::new();
     assert!(queue.pop().is_none());
     assert!(queue.is_empty());
 }
 
 #[test]
 fn push_then_pop_returns_same_thread() {
-    let mut queue = WaitQueue::new();
+    let mut queue: WaitQueue = WaitQueue::new();
     queue.push(thread(1)).unwrap();
     assert_eq!(queue.pop(), Some(thread(1)));
 }
 
 #[test]
 fn pop_order_is_fifo() {
-    let mut queue = WaitQueue::new();
+    let mut queue: WaitQueue = WaitQueue::new();
     queue.push(thread(1)).unwrap();
     queue.push(thread(2)).unwrap();
     queue.push(thread(3)).unwrap();
@@ -46,7 +46,7 @@ fn pop_order_is_fifo() {
 
 #[test]
 fn full_queue_rejects_push() {
-    let mut queue = WaitQueue::new();
+    let mut queue: WaitQueue = WaitQueue::new();
     for i in 0..MAX_WAITERS {
         assert!(queue.push(thread(i as u16)).is_some());
     }
@@ -55,7 +55,7 @@ fn full_queue_rejects_push() {
 
 #[test]
 fn is_empty_after_drain() {
-    let mut queue = WaitQueue::new();
+    let mut queue: WaitQueue = WaitQueue::new();
     queue.push(thread(1)).unwrap();
     queue.push(thread(2)).unwrap();
     queue.pop();
@@ -219,3 +219,40 @@ fn table_destroy_reuses_slot() {
     let new_id = table.create(owner()).unwrap();
     assert_eq!(id, new_id);
 }
+
+#[test]
+fn wait_queue_is_fifo_even_after_a_pop_leaves_a_hole() {
+    let mut q: WaitQueue = WaitQueue::new();
+    q.push(thread(1)).unwrap();
+    q.push(thread(2)).unwrap();
+    assert_eq!(q.pop(), Some(thread(1)));
+    q.push(thread(3)).unwrap();
+    assert_eq!(q.pop(), Some(thread(2)), "thread 3 jumped the queue");
+    assert_eq!(q.pop(), Some(thread(3)));
+}
+
+#[test]
+fn wait_queue_keeps_order_around_a_removed_thread() {
+    let mut q: WaitQueue = WaitQueue::new();
+    for n in 1..=3 {
+        q.push(thread(n)).unwrap();
+    }
+    assert!(q.remove_thread(thread(2)));
+    q.push(thread(4)).unwrap();
+    assert_eq!(q.pop(), Some(thread(1)));
+    assert_eq!(q.pop(), Some(thread(3)));
+    assert_eq!(q.pop(), Some(thread(4)));
+    assert_eq!(q.pop(), None);
+}
+
+#[test]
+fn wait_queue_kani_trace_push_push_remove() {
+    let mut q: WaitQueue<3> = WaitQueue::new();
+    q.push(thread(3)).unwrap();
+    q.push(thread(2)).unwrap();
+    assert!(q.remove_thread(thread(3)));
+    assert!(!q.remove_thread(thread(0)));
+    assert!(!q.remove_thread(thread(3)));
+    assert_eq!(q.pop(), Some(thread(2)));
+    assert_eq!(q.pop(), None);
+}

crates/domain/src/endpoint/wait_queue.rs

@@ -2,51 +2,127 @@ use types::id::ThreadId;
 
 pub const MAX_WAITERS: usize = 16;
 
-pub struct WaitQueue {
+/// Bounded FIFO of blocked threads. Order is preserved across pops and
-    entries: [Option<ThreadId>; MAX_WAITERS],
+/// removals, so no waiter can be overtaken.
-    count: u8,
+pub struct WaitQueue<const N: usize = MAX_WAITERS> {
+    entries: [Option<ThreadId>; N],
+    len: usize,
 }
 
-impl WaitQueue {
+impl<const N: usize> WaitQueue<N> {
     pub fn new() -> Self {
         Self {
-            entries: [None; MAX_WAITERS],
+            entries: [None; N],
-            count: 0,
+            len: 0,
         }
     }
 
     pub fn push(&mut self, thread: ThreadId) -> Option<()> {
-        let slot = self.entries.iter_mut().find(|s| s.is_none())?;
+        if self.len == N || self.contains(thread) {
-        *slot = Some(thread);
+            return None;
-        self.count += 1;
+        }
+        self.entries[self.len] = Some(thread);
+        self.len += 1;
         Some(())
     }
 
     pub fn pop(&mut self) -> Option<ThreadId> {
-        let slot = self.entries.iter_mut().find(|s| s.is_some())?;
+        let head = self.entries[0]?;
-        let thread = slot.take();
+        self.close_gap(0);
-        self.count -= 1;
+        Some(head)
-        thread
     }
 
     pub fn remove_thread(&mut self, thread: ThreadId) -> bool {
-        for slot in &mut self.entries {
+        match self.entries[..self.len]
-            if *slot == Some(thread) {
+            .iter()
-                *slot = None;
+            .position(|e| *e == Some(thread))
-                self.count -= 1;
+        {
-                return true;
+            Some(i) => {
+                self.close_gap(i);
+                true
             }
+            None => false,
         }
-        false
+    }
+
+    pub fn contains(&self, thread: ThreadId) -> bool {
+        self.entries[..self.len].contains(&Some(thread))
     }
 
     pub fn is_empty(&self) -> bool {
-        self.count == 0
+        self.len == 0
+    }
+
+    fn close_gap(&mut self, i: usize) {
+        for j in i..self.len - 1 {
+            self.entries[j] = self.entries[j + 1];
+        }
+        self.len -= 1;
+        self.entries[self.len] = None;
     }
 }
 
-impl Default for WaitQueue {
+impl<const N: usize> Default for WaitQueue<N> {
     fn default() -> Self {
         Self::new()
     }
 }
+
+#[cfg(kani)]
+mod proofs {
+    use super::*;
+
+    /// Against a simple reference model (a plain array kept in arrival
+    /// order), any sequence of pushes, pops and removals behaves the same:
+    /// FIFO order, no duplicates, nothing lost.
+    #[kani::proof]
+    #[kani::unwind(7)]
+    fn behaves_like_a_fifo_without_duplicates() {
+        const CAP: usize = 3;
+        let mut q: WaitQueue<CAP> = WaitQueue::new();
+        let mut model = [0u16; CAP];
+        let mut len = 0usize;
+
+        for _ in 0..5 {
+            let t: u16 = kani::any::<u16>() % 4;
+            match kani::any::<u8>() % 3 {
+                0 => {
+                    let ok = q.push(ThreadId::from_raw(t)).is_some();
+                    let dup = model[..len].contains(&t);
+                    assert!(ok == (!dup && len < CAP));
+                    if ok {
+                        model[len] = t;
+                        len += 1;
+                    }
+                }
+                1 => {
+                    let got = q.pop();
+                    if len == 0 {
+                        assert!(got.is_none());
+                    } else {
+                        assert!(got == Some(ThreadId::from_raw(model[0])));
+                        for j in 0..len - 1 {
+                            model[j] = model[j + 1];
+                        }
+                        len -= 1;
+                    }
+                }
+                _ => {
+                    let removed = q.remove_thread(ThreadId::from_raw(t));
+                    match model[..len].iter().position(|&m| m == t) {
+                        Some(i) => {
+                            assert!(removed);
+                            for j in i..len - 1 {
+                                model[j] = model[j + 1];
+                            }
+                            len -= 1;
+                        }
+                        None => assert!(!removed),
+                    }
+                }
+            }
+            assert!(q.len == len);
+            assert!(q.is_empty() == (len == 0));
+        }
+    }
+}

crates/domain/src/error.rs

@@ -33,4 +33,9 @@ pub enum KernelError {
     InvalidRegion,
     /// The object a capability would refer to no longer exists.
     NoSuchObject,
+    /// The capability names the wrong kind of object for this operation.
+    WrongObject,
+    /// `Reply` with an empty reply slot, or a call whose reply was dropped.
+    NoReply,
+    InvalidArgument,
 }

crates/domain/src/irq/mod.rs

@@ -1,6 +1,8 @@
-use types::id::{IrqId, ThreadId};
+use types::{
-
+    badge::Badge,
-use crate::error::KernelError;
+    generation::Generation,
+    id::{IrqId, NotificationId},
+};
 
 #[cfg(test)]
 #[path = "tests.rs"]
@@ -8,47 +10,47 @@ mod tests;
 
 const MAX_IRQS: usize = 256;
 
+/// Where an interrupt is delivered: a signal on a notification, through a
+/// capability snapshot (generation and badge) taken when the route was set.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct IrqRoute {
+    pub notification: NotificationId,
+    pub generation: Generation,
+    pub badge: Badge,
+}
+
 pub struct IrqRoutingTable {
-    owners: [Option<ThreadId>; MAX_IRQS],
+    routes: [Option<IrqRoute>; MAX_IRQS],
 }
 
 impl IrqRoutingTable {
     pub fn new() -> Self {
         Self {
-            owners: [None; MAX_IRQS],
+            routes: [None; MAX_IRQS],
         }
     }
 
-    pub fn register(&mut self, irq: IrqId, thread: ThreadId) -> Result<(), KernelError> {
+    /// Routes `irq`, replacing any previous route. Fails for lines beyond
-        let slot = self
+    /// the table.
-            .owners
+    pub fn set(&mut self, irq: IrqId, route: IrqRoute) -> bool {
-            .get_mut(irq.raw() as usize)
+        match self.routes.get_mut(irq.raw() as usize) {
-            .ok_or(KernelError::InvalidIrq(irq))?;
+            Some(slot) => {
-        if slot.is_some() {
+                *slot = Some(route);
-            return Err(KernelError::IrqAlreadyOwned(irq));
+                true
+            }
+            None => false,
         }
-        *slot = Some(thread);
-        Ok(())
     }
 
-    pub fn unregister(&mut self, irq: IrqId) {
+    pub fn clear(&mut self, irq: IrqId) {
-        if let Some(slot) = self.owners.get_mut(irq.raw() as usize) {
+        if let Some(slot) = self.routes.get_mut(irq.raw() as usize) {
             *slot = None;
         }
     }
 
-    /// Drops every route to `thread`, e.g. when it dies.
+    /// The route for `irq`; `None` for unknown or unrouted lines.
-    pub fn unregister_thread(&mut self, thread: ThreadId) {
+    pub fn route(&self, irq: IrqId) -> Option<IrqRoute> {
-        for slot in &mut self.owners {
+        *self.routes.get(irq.raw() as usize)?
-            if *slot == Some(thread) {
-                *slot = None;
-            }
-        }
-    }
-
-    /// The thread an interrupt is routed to; `None` for unknown lines.
-    pub fn owner(&self, irq: IrqId) -> Option<ThreadId> {
-        *self.owners.get(irq.raw() as usize)?
     }
 }
 

crates/domain/src/irq/tests.rs

@@ -1,50 +1,43 @@
 use super::*;
-use types::id::{IrqId, ThreadId};
+use types::badge::Badge;
+use types::generation::Generation;
+use types::id::{IrqId, NotificationId};
 
-fn irq(id: u16) -> IrqId {
+fn route(n: u16) -> IrqRoute {
-    IrqId::from_raw(id)
+    IrqRoute {
-}
+        notification: NotificationId::from_raw(n),
-
+        generation: Generation::initial(),
-fn thread(id: u16) -> ThreadId {
+        badge: Badge::new(1 << n),
-    ThreadId::from_raw(id)
+    }
 }
 
 #[test]
-fn new_table_has_no_owners() {
+fn new_table_routes_nothing() {
-    let table = IrqRoutingTable::new();
+    let t = IrqRoutingTable::new();
-    assert!(table.owner(irq(0)).is_none());
+    assert_eq!(t.route(IrqId::from_raw(10)), None);
-    assert!(table.owner(irq(10)).is_none());
 }
 
 #[test]
-fn register_and_lookup() {
+fn set_then_route() {
-    let mut table = IrqRoutingTable::new();
+    let mut t = IrqRoutingTable::new();
-    table.register(irq(5), thread(1)).unwrap();
+    assert!(t.set(IrqId::from_raw(10), route(3)));
-    assert_eq!(table.owner(irq(5)), Some(thread(1)));
+    assert_eq!(t.route(IrqId::from_raw(10)), Some(route(3)));
 }
 
 #[test]
-fn register_already_owned_fails() {
+fn set_replaces_and_clear_removes() {
-    let mut table = IrqRoutingTable::new();
+    let mut t = IrqRoutingTable::new();
-    table.register(irq(5), thread(1)).unwrap();
+    t.set(IrqId::from_raw(10), route(3));
-    assert!(table.register(irq(5), thread(2)).is_err());
+    t.set(IrqId::from_raw(10), route(4));
+    assert_eq!(t.route(IrqId::from_raw(10)), Some(route(4)));
+    t.clear(IrqId::from_raw(10));
+    assert_eq!(t.route(IrqId::from_raw(10)), None);
 }
 
 #[test]
-fn unregister_then_reregister() {
+fn lines_beyond_the_table_are_rejected_without_panicking() {
-    let mut table = IrqRoutingTable::new();
+    let mut t = IrqRoutingTable::new();
-    table.register(irq(5), thread(1)).unwrap();
+    assert!(!t.set(IrqId::from_raw(1000), route(1)));
-    table.unregister(irq(5));
+    assert_eq!(t.route(IrqId::from_raw(1000)), None);
-    assert!(table.owner(irq(5)).is_none());
+    t.clear(IrqId::from_raw(1000));
-    table.register(irq(5), thread(2)).unwrap();
-    assert_eq!(table.owner(irq(5)), Some(thread(2)));
-}
-
-#[test]
-fn different_irqs_independent() {
-    let mut table = IrqRoutingTable::new();
-    table.register(irq(0), thread(10)).unwrap();
-    table.register(irq(1), thread(20)).unwrap();
-    assert_eq!(table.owner(irq(0)), Some(thread(10)));
-    assert_eq!(table.owner(irq(1)), Some(thread(20)));
 }

crates/domain/src/lib.rs

@@ -11,6 +11,7 @@ pub mod generations;
 pub mod irq;
 pub mod memory;
 pub mod message;
+pub mod notification;
 pub mod process;
 pub mod resource;
 pub mod scheduler;

crates/domain/src/message/mod.rs

@@ -5,26 +5,26 @@ use types::bounded::CapSlot;
 mod tests;
 
 pub const MAX_MESSAGE_WORDS: usize = 4;
-pub const MAX_CAP_TRANSFERS: usize = 4;
 
+/// Four words and at most one capability (named by the sender's slot).
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub struct Message {
     words: [usize; MAX_MESSAGE_WORDS],
-    caps: [Option<CapSlot>; MAX_CAP_TRANSFERS],
+    cap: Option<CapSlot>,
 }
 
 impl Message {
     pub fn empty() -> Self {
-        Self {
+        Self::from_words([0; MAX_MESSAGE_WORDS])
-            words: [0; MAX_MESSAGE_WORDS],
-            caps: [None; MAX_CAP_TRANSFERS],
-        }
     }
 
     pub fn from_words(words: [usize; MAX_MESSAGE_WORDS]) -> Self {
-        Self {
+        Self { words, cap: None }
-            words,
+    }
-            caps: [None; MAX_CAP_TRANSFERS],
+
-        }
+    pub fn with_cap(mut self, slot: CapSlot) -> Self {
+        self.cap = Some(slot);
+        self
     }
 
     pub fn words(&self) -> &[usize; MAX_MESSAGE_WORDS] {
@@ -35,42 +35,7 @@ impl Message {
         self.words[index]
     }
 
-    pub fn caps(&self) -> &[Option<CapSlot>; MAX_CAP_TRANSFERS] {
+    pub fn cap(&self) -> Option<CapSlot> {
-        &self.caps
+        self.cap
-    }
-
-    pub fn cap(&self, index: usize) -> Option<CapSlot> {
-        self.caps[index]
-    }
-
-    pub fn builder() -> MessageBuilder {
-        MessageBuilder {
-            message: Message::empty(),
-        }
-    }
-}
-
-pub struct MessageBuilder {
-    message: Message,
-}
-
-impl MessageBuilder {
-    pub fn word(mut self, index: usize, value: usize) -> Self {
-        self.message.words[index] = value;
-        self
-    }
-
-    pub fn words(mut self, words: [usize; MAX_MESSAGE_WORDS]) -> Self {
-        self.message.words = words;
-        self
-    }
-
-    pub fn cap(mut self, index: usize, slot: CapSlot) -> Self {
-        self.message.caps[index] = Some(slot);
-        self
-    }
-
-    pub fn build(self) -> Message {
-        self.message
     }
 }

crates/domain/src/message/tests.rs

@@ -1,78 +1,23 @@
 use super::*;
 use types::bounded::CapSlot;
 
-fn slot(n: u8) -> CapSlot {
-    CapSlot::new(n).unwrap()
-}
-
-// --- Construction ---
-
 #[test]
-fn empty_message_has_zeroed_words_and_no_caps() {
+fn empty_message_has_zeroed_words_and_no_cap() {
     let msg = Message::empty();
     assert_eq!(msg.words(), &[0; MAX_MESSAGE_WORDS]);
-    for i in 0..MAX_CAP_TRANSFERS {
+    assert!(msg.cap().is_none());
-        assert!(msg.cap(i).is_none());
-    }
 }
 
 #[test]
-fn from_words_sets_words_and_no_caps() {
+fn from_words_keeps_every_word() {
     let msg = Message::from_words([10, 20, 30, 40]);
-    assert_eq!(msg.word(0), 10);
+    assert_eq!(msg.words(), &[10, 20, 30, 40]);
-    assert_eq!(msg.word(1), 20);
+    assert!(msg.cap().is_none());
-    assert_eq!(msg.word(2), 30);
-    assert_eq!(msg.word(3), 40);
-    assert!(msg.cap(0).is_none());
-}
-
-// --- Accessors ---
-
-#[test]
-fn words_returns_full_array() {
-    let msg = Message::from_words([1, 2, 3, 4]);
-    assert_eq!(msg.words(), &[1, 2, 3, 4]);
 }
 
 #[test]
-fn caps_returns_full_array() {
+fn with_cap_attaches_one_capability() {
-    let msg = Message::empty();
+    let msg = Message::from_words([1, 2, 3, 4]).with_cap(CapSlot::new(5).unwrap());
-    assert_eq!(msg.caps(), &[None; MAX_CAP_TRANSFERS]);
+    assert_eq!(msg.cap(), CapSlot::new(5));
-}
+    assert_eq!(msg.word(3), 4);
-
-// --- Builder ---
-
-#[test]
-fn builder_sets_individual_words() {
-    let msg = Message::builder().word(0, 42).word(2, 99).build();
-    assert_eq!(msg.word(0), 42);
-    assert_eq!(msg.word(1), 0);
-    assert_eq!(msg.word(2), 99);
-    assert_eq!(msg.word(3), 0);
-}
-
-#[test]
-fn builder_sets_caps() {
-    let msg = Message::builder().cap(0, slot(5)).cap(1, slot(10)).build();
-    assert_eq!(msg.cap(0), Some(slot(5)));
-    assert_eq!(msg.cap(1), Some(slot(10)));
-    assert!(msg.cap(2).is_none());
-}
-
-#[test]
-fn builder_sets_words_and_caps_together() {
-    let msg = Message::builder()
-        .word(0, 100)
-        .word(1, 200)
-        .cap(0, slot(3))
-        .build();
-    assert_eq!(msg.word(0), 100);
-    assert_eq!(msg.word(1), 200);
-    assert_eq!(msg.cap(0), Some(slot(3)));
-}
-
-#[test]
-fn builder_with_all_words() {
-    let msg = Message::builder().words([11, 22, 33, 44]).build();
-    assert_eq!(msg.words(), &[11, 22, 33, 44]);
 }

crates/domain/src/notification/mod.rs

@@ -0,0 +1,186 @@
+//! Asynchronous signals: a word of bits, a FIFO of waiters and at most one
+//! bound thread. See docs/ipc.md.
+
+use types::{
+    badge::Badge,
+    generation::Generation,
+    id::{NotificationId, ProcessId, ThreadId},
+};
+
+use crate::{endpoint::wait_queue::WaitQueue, generations::SlotGenerations};
+
+#[cfg(test)]
+#[path = "tests.rs"]
+mod tests;
+
+#[cfg(kani)]
+#[path = "proofs.rs"]
+mod proofs;
+
+pub const MAX_NOTIFICATIONS: usize = 128;
+
+/// What the kernel must do after a signal.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum SignalEffect {
+    /// Wake this blocked `Wait`er with these bits.
+    WakeWaiter(ThreadId, usize),
+    /// Wake the bound thread out of its `Receive` with these bits.
+    WakeBound(ThreadId, usize),
+    /// Nobody to wake; the bits were stored.
+    Stored,
+}
+
+pub struct Notification {
+    id: NotificationId,
+    owner: ProcessId,
+    bits: usize,
+    waiters: WaitQueue,
+    bound: Option<ThreadId>,
+}
+
+impl Notification {
+    pub fn new(id: NotificationId, owner: ProcessId) -> Self {
+        Self {
+            id,
+            owner,
+            bits: 0,
+            waiters: WaitQueue::new(),
+            bound: None,
+        }
+    }
+
+    pub fn id(&self) -> NotificationId {
+        self.id
+    }
+
+    pub fn owner(&self) -> ProcessId {
+        self.owner
+    }
+
+    pub fn bound(&self) -> Option<ThreadId> {
+        self.bound
+    }
+
+    /// Delivers a signal. `bound_is_receiving` says whether the bound
+    /// thread is currently blocked in `Receive` and can be woken.
+    pub fn signal(&mut self, badge: Badge, bound_is_receiving: bool) -> SignalEffect {
+        let bits = if badge.is_none() { 1 } else { badge.raw() };
+        if let Some(waiter) = self.waiters.pop() {
+            return SignalEffect::WakeWaiter(waiter, bits);
+        }
+        if let (Some(bound), true) = (self.bound, bound_is_receiving) {
+            return SignalEffect::WakeBound(bound, bits);
+        }
+        self.bits |= bits;
+        SignalEffect::Stored
+    }
+
+    /// Returns and clears the pending bits.
+    pub fn take_bits(&mut self) -> usize {
+        core::mem::take(&mut self.bits)
+    }
+
+    pub fn has_bits(&self) -> bool {
+        self.bits != 0
+    }
+
+    pub fn is_waiter(&self, thread: ThreadId) -> bool {
+        self.waiters.contains(thread)
+    }
+
+    pub fn add_waiter(&mut self, thread: ThreadId) -> bool {
+        self.waiters.push(thread).is_some()
+    }
+
+    pub fn bind(&mut self, thread: ThreadId) -> bool {
+        if self.bound.is_some() {
+            return false;
+        }
+        self.bound = Some(thread);
+        true
+    }
+
+    pub fn unbind(&mut self, thread: ThreadId) {
+        if self.bound == Some(thread) {
+            self.bound = None;
+        }
+    }
+
+    /// Forgets `thread` as waiter and as bound thread.
+    pub fn remove_thread(&mut self, thread: ThreadId) {
+        self.waiters.remove_thread(thread);
+        self.unbind(thread);
+    }
+
+    /// Drains every waiter, e.g. when the notification is destroyed.
+    pub fn pop_waiter(&mut self) -> Option<ThreadId> {
+        self.waiters.pop()
+    }
+}
+
+pub struct NotificationTable {
+    slots: [Option<Notification>; MAX_NOTIFICATIONS],
+    generations: SlotGenerations<MAX_NOTIFICATIONS>,
+}
+
+impl NotificationTable {
+    pub fn new() -> Self {
+        Self {
+            slots: [const { None }; MAX_NOTIFICATIONS],
+            generations: SlotGenerations::new(),
+        }
+    }
+
+    pub fn create(&mut self, owner: ProcessId) -> Option<NotificationId> {
+        let idx = (0..MAX_NOTIFICATIONS)
+            .find(|&i| self.slots[i].is_none() && self.generations.is_usable(i))?;
+        let id = NotificationId::from_raw(idx as u16);
+        self.slots[idx] = Some(Notification::new(id, owner));
+        Some(id)
+    }
+
+    pub fn get(&self, id: NotificationId) -> Option<&Notification> {
+        self.slots.get(id.raw() as usize)?.as_ref()
+    }
+
+    pub fn get_mut(&mut self, id: NotificationId) -> Option<&mut Notification> {
+        self.slots.get_mut(id.raw() as usize)?.as_mut()
+    }
+
+    pub fn generation(&self, id: NotificationId) -> Option<Generation> {
+        self.get(id)?;
+        self.generations.current(id.raw() as usize)
+    }
+
+    pub fn revoke(&mut self, id: NotificationId) {
+        if self.get(id).is_some() {
+            self.generations.advance(id.raw() as usize);
+        }
+    }
+
+    pub fn destroy(&mut self, id: NotificationId) -> Option<Notification> {
+        let n = self.slots.get_mut(id.raw() as usize)?.take()?;
+        self.generations.advance(id.raw() as usize);
+        Some(n)
+    }
+
+    pub fn ids_owned_by(&self, owner: ProcessId) -> impl Iterator<Item = NotificationId> + '_ {
+        self.slots
+            .iter()
+            .flatten()
+            .filter(move |n| n.owner == owner)
+            .map(|n| n.id)
+    }
+
+    pub fn remove_thread_from_all(&mut self, thread: ThreadId) {
+        for n in self.slots.iter_mut().flatten() {
+            n.remove_thread(thread);
+        }
+    }
+}
+
+impl Default for NotificationTable {
+    fn default() -> Self {
+        Self::new()
+    }
+}

crates/domain/src/notification/proofs.rs

@@ -0,0 +1,43 @@
+//! Signals are never lost: every signal either wakes exactly one thread
+//! with its bits, or leaves those bits set for the next `Wait`.
+
+use super::*;
+use types::badge::Badge;
+use types::id::{NotificationId, ProcessId, ThreadId};
+
+#[kani::proof]
+#[kani::unwind(5)]
+fn a_signal_is_delivered_or_stored() {
+    let mut n = Notification::new(NotificationId::from_raw(0), ProcessId::from_raw(0));
+    n.bits = kani::any();
+    if kani::any() {
+        n.add_waiter(ThreadId::from_raw(kani::any::<u16>() % 4));
+    }
+    if kani::any() {
+        n.bind(ThreadId::from_raw(kani::any::<u16>() % 4));
+    }
+    let before = n.bits;
+    let badge = Badge::new(kani::any());
+    let expected = if badge.is_none() { 1 } else { badge.raw() };
+
+    match n.signal(badge, kani::any()) {
+        SignalEffect::WakeWaiter(_, bits) | SignalEffect::WakeBound(_, bits) => {
+            assert!(bits == expected);
+            assert!(n.bits == before, "woken signals are not also stored");
+        }
+        SignalEffect::Stored => {
+            assert!(n.bits == before | expected);
+            assert!(n.bits & expected == expected);
+        }
+    }
+}
+
+#[kani::proof]
+#[kani::unwind(5)]
+fn take_bits_returns_everything_and_clears() {
+    let mut n = Notification::new(NotificationId::from_raw(0), ProcessId::from_raw(0));
+    let bits: usize = kani::any();
+    n.bits = bits;
+    assert!(n.take_bits() == bits);
+    assert!(n.bits == 0);
+}

crates/domain/src/notification/tests.rs

@@ -0,0 +1,116 @@
+use super::*;
+use types::badge::Badge;
+use types::id::{NotificationId, ProcessId, ThreadId};
+
+fn ntfn() -> Notification {
+    Notification::new(NotificationId::from_raw(0), ProcessId::from_raw(0))
+}
+
+fn t(n: u16) -> ThreadId {
+    ThreadId::from_raw(n)
+}
+
+#[test]
+fn signal_with_nobody_waiting_stores_the_badge() {
+    let mut n = ntfn();
+    assert_eq!(n.signal(Badge::new(0b100), false), SignalEffect::Stored);
+    assert_eq!(n.signal(Badge::new(0b001), false), SignalEffect::Stored);
+    assert_eq!(n.take_bits(), 0b101);
+    assert_eq!(n.take_bits(), 0, "taking clears");
+}
+
+#[test]
+fn unbadged_signal_sets_bit_zero() {
+    let mut n = ntfn();
+    n.signal(Badge::NONE, false);
+    assert_eq!(n.take_bits(), 1);
+}
+
+#[test]
+fn signal_wakes_the_first_waiter_with_its_badge() {
+    let mut n = ntfn();
+    assert!(n.add_waiter(t(1)));
+    assert!(n.add_waiter(t(2)));
+    assert_eq!(
+        n.signal(Badge::new(8), false),
+        SignalEffect::WakeWaiter(t(1), 8)
+    );
+    assert_eq!(n.take_bits(), 0, "delivered bits are not also stored");
+    assert_eq!(
+        n.signal(Badge::new(4), false),
+        SignalEffect::WakeWaiter(t(2), 4)
+    );
+}
+
+#[test]
+fn signal_wakes_a_receiving_bound_thread_when_no_waiter() {
+    let mut n = ntfn();
+    assert!(n.bind(t(5)));
+    assert_eq!(
+        n.signal(Badge::new(2), true),
+        SignalEffect::WakeBound(t(5), 2)
+    );
+    assert_eq!(n.signal(Badge::new(2), false), SignalEffect::Stored);
+}
+
+#[test]
+fn waiters_take_priority_over_the_bound_thread() {
+    let mut n = ntfn();
+    n.bind(t(5));
+    n.add_waiter(t(1));
+    assert_eq!(
+        n.signal(Badge::new(2), true),
+        SignalEffect::WakeWaiter(t(1), 2)
+    );
+}
+
+#[test]
+fn only_one_thread_can_be_bound() {
+    let mut n = ntfn();
+    assert!(n.bind(t(1)));
+    assert!(!n.bind(t(2)));
+    n.unbind(t(1));
+    assert!(n.bind(t(2)));
+}
+
+#[test]
+fn removed_waiter_is_not_woken() {
+    let mut n = ntfn();
+    n.add_waiter(t(1));
+    n.remove_thread(t(1));
+    assert_eq!(n.signal(Badge::new(1), false), SignalEffect::Stored);
+}
+
+#[test]
+fn remove_thread_also_unbinds() {
+    let mut n = ntfn();
+    n.bind(t(1));
+    n.remove_thread(t(1));
+    assert_eq!(n.bound(), None);
+}
+
+#[test]
+fn table_generations_follow_slot_reuse() {
+    let mut table = NotificationTable::new();
+    let owner = ProcessId::from_raw(3);
+    let id = table.create(owner).unwrap();
+    let g = table.generation(id).unwrap();
+    assert!(table.destroy(id).is_some());
+    assert_eq!(table.generation(id), None);
+    let again = table.create(owner).unwrap();
+    assert_eq!(again, id);
+    assert!(!g.is_valid_against(&table.generation(again).unwrap()));
+}
+
+#[test]
+fn table_lists_notifications_by_owner() {
+    let mut table = NotificationTable::new();
+    let a = table.create(ProcessId::from_raw(1)).unwrap();
+    table.create(ProcessId::from_raw(2)).unwrap();
+    let mut n = 0;
+    for id in table.ids_owned_by(ProcessId::from_raw(1)) {
+        assert_eq!(id, a);
+        n += 1;
+    }
+    assert_eq!(n, 1);
+}

crates/domain/src/proofs/thread.rs

@@ -22,7 +22,7 @@ fn dead_thread_rejects_all_transitions() {
 
     assert!(matches!(thread.state(), ThreadState::Dead));
     assert!(thread.set_running(CoreId::from_raw(0)).is_err());
-    assert!(thread.block(BlockReason::WaitingAny).is_err());
+    assert!(thread.block(BlockReason::AwaitingReply).is_err());
     assert!(thread.unblock().is_err());
     assert!(thread.preempt().is_err());
     assert!(thread.exit().is_err());
@@ -36,7 +36,7 @@ fn only_running_thread_can_block() {
     let mut thread = test_thread(Priority::new(prio_val).unwrap());
 
     assert!(matches!(thread.state(), ThreadState::Ready));
-    assert!(thread.block(BlockReason::WaitingAny).is_err());
+    assert!(thread.block(BlockReason::AwaitingReply).is_err());
 }
 
 #[kani::proof]
@@ -44,7 +44,7 @@ fn blocked_thread_cannot_run_directly() {
     let mut thread = test_thread(Priority::new(16).unwrap());
 
     thread.set_running(CoreId::from_raw(0)).unwrap();
-    thread.block(BlockReason::WaitingAny).unwrap();
+    thread.block(BlockReason::AwaitingReply).unwrap();
 
     assert!(thread.set_running(CoreId::from_raw(0)).is_err());
 }

crates/domain/src/resource.rs

@@ -1,9 +1,10 @@
-use types::id::{EndpointId, IrqId, ProcessId, RegionId};
+use types::id::{EndpointId, IrqId, NotificationId, ProcessId, RegionId};
 
 #[derive(Clone)]
 pub enum Resource {
     Memory(RegionId),
     Endpoint(EndpointId),
+    Notification(NotificationId),
     Process(ProcessId),
     Interrupt(IrqId),
     ProcessCreate,

crates/domain/src/scheduler/multi_core.rs

@@ -30,6 +30,13 @@ impl MultiCoreScheduler {
         }
     }
 
+    /// Whether `thread` waits in some core's run queue.
+    pub fn is_queued(&self, thread: ThreadId) -> bool {
+        self.cores[..self.num_cores as usize]
+            .iter()
+            .any(|c| c.is_queued(thread))
+    }
+
     pub fn current(&self, core: CoreId) -> Option<ThreadId> {
         self.cores[core.raw() as usize].current()
     }

crates/domain/src/scheduler/per_core.rs

@@ -29,6 +29,10 @@ impl PerCoreScheduler {
         self.queue.add(thread, priority);
     }
 
+    pub fn is_queued(&self, thread: ThreadId) -> bool {
+        self.queue.contains(thread)
+    }
+
     pub fn remove(&mut self, thread: ThreadId) {
         self.queue.remove(thread);
     }

crates/domain/src/syscall.rs

@@ -1,42 +1,12 @@
 use types::{
     address::{DataRef, EntryPoint, StackPointer, VirtualAddress},
+    badge::Badge,
     bounded::{CapSlot, PageCount},
     permission::PermissionSet,
 };
 
 use crate::message::Message;
 
-pub struct WaitSet {
-    endpoints: [Option<CapSlot>; 4],
-    interrupts: [Option<CapSlot>; 4],
-}
-
-impl WaitSet {
-    pub fn new(endpoints: [Option<CapSlot>; 4], interrupts: [Option<CapSlot>; 4]) -> Self {
-        Self {
-            endpoints,
-            interrupts,
-        }
-    }
-
-    pub fn endpoints(&self) -> &[Option<CapSlot>; 4] {
-        &self.endpoints
-    }
-
-    pub fn interrupts(&self) -> &[Option<CapSlot>; 4] {
-        &self.interrupts
-    }
-
-    pub fn is_empty(&self) -> bool {
-        self.endpoints.iter().all(|s| s.is_none()) && self.interrupts.iter().all(|s| s.is_none())
-    }
-}
-
-pub enum WakeReason {
-    MessageReady(CapSlot),
-    InterruptFired(CapSlot),
-}
-
 pub enum Syscall {
     Send {
         endpoint: CapSlot,
@@ -50,15 +20,47 @@ pub enum Syscall {
         to_endpoint: CapSlot,
         permissions: PermissionSet,
     },
-    WaitForInterrupt {
+    /// Send, then block for the reply (docs/ipc.md).
+    Call {
+        endpoint: CapSlot,
+        message: Message,
+    },
+    /// Answer the thread in the caller's reply slot.
+    Reply {
+        message: Message,
+    },
+    /// Reply (if a reply is pending), then receive.
+    ReplyRecv {
+        endpoint: CapSlot,
+        message: Message,
+    },
+    /// Copy an unbadged endpoint/notification cap with a badge and
+    /// narrower permissions into the caller's own table.
+    Mint {
+        slot: CapSlot,
+        badge: Badge,
+        permissions: PermissionSet,
+    },
+    CreateNotification,
+    Signal {
+        notification: CapSlot,
+    },
+    Wait {
+        notification: CapSlot,
+    },
+    Poll {
+        notification: CapSlot,
+    },
+    BindNotification {
+        notification: CapSlot,
+    },
+    SetIrqNotification {
         irq: CapSlot,
+        notification: CapSlot,
     },
     AcknowledgeInterrupt {
         irq: CapSlot,
     },
-    WaitAny {
-        sources: WaitSet,
-    },
     CreateProcess {
         create_cap: CapSlot,
         image: DataRef,

crates/domain/src/thread/mod.rs

@@ -1,10 +1,11 @@
 use types::{
-    bounded::{CapSlot, Priority},
+    bounded::Priority,
-    id::{CoreId, ProcessId, ThreadId},
+    generation::Generation,
+    id::{CoreId, EndpointId, NotificationId, ProcessId, ThreadId},
     time::TimeSlice,
 };
 
-use crate::error::KernelError;
+use crate::{error::KernelError, generations::SlotGenerations};
 
 #[cfg(test)]
 #[path = "tests.rs"]
@@ -80,10 +81,16 @@ impl SavedContext {
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum BlockReason {
-    Sending(CapSlot),
+    /// Queued as a sender on the endpoint.
-    Receiving(CapSlot),
+    Sending(EndpointId),
-    WaitingForInterrupt(CapSlot),
+    /// Queued as a `Call`er on the endpoint; will then await the reply.
-    WaitingAny,
+    Calling(EndpointId),
+    /// Queued as a receiver on the endpoint.
+    Receiving(EndpointId),
+    /// Delivered a `Call`; waiting for the server's reply.
+    AwaitingReply,
+    /// Blocked in `Wait` on the notification.
+    WaitingNotification(NotificationId),
 }
 
 impl Thread {
@@ -180,6 +187,17 @@ impl Thread {
         }
     }
 
+    /// Keeps a blocked thread blocked, waiting for something else.
+    pub fn rewait(&mut self, reason: BlockReason) -> Result<(), KernelError> {
+        match self.state {
+            ThreadState::Blocked(_) => {
+                self.state = ThreadState::Blocked(reason);
+                Ok(())
+            }
+            _ => Err(KernelError::InvalidStateTransition(self.id)),
+        }
+    }
+
     pub fn unblock(&mut self) -> Result<(), KernelError> {
         match self.state {
             ThreadState::Blocked(_) => {
@@ -215,15 +233,28 @@ pub const MAX_THREADS: usize = 256;
 
 pub struct ThreadTable {
     threads: [Option<Thread>; MAX_THREADS],
+    generations: SlotGenerations<MAX_THREADS>,
 }
 
 impl ThreadTable {
     pub fn new() -> Self {
         Self {
             threads: [const { None }; MAX_THREADS],
+            generations: SlotGenerations::new(),
         }
     }
 
+    /// Current generation of a live thread, so references to it (reply
+    /// slots) can detect that the id was reused.
+    pub fn generation(&self, id: ThreadId) -> Option<Generation> {
+        self.get(id).ok()?;
+        self.generations.current(id.raw() as usize)
+    }
+
+    fn free_slot(&self) -> Option<usize> {
+        (0..MAX_THREADS).find(|&i| self.threads[i].is_none() && self.generations.is_usable(i))
+    }
+
     pub fn spawn(
         &mut self,
         process: ProcessId,
@@ -231,11 +262,7 @@ impl ThreadTable {
         time_slice: TimeSlice,
         core_affinity: Option<CoreId>,
     ) -> Result<ThreadId, KernelError> {
-        let idx = self
+        let idx = self.free_slot().ok_or(KernelError::TooManyThreads)?;
-            .threads
-            .iter()
-            .position(|s| s.is_none())
-            .ok_or(KernelError::TooManyThreads)?;
 
         let id = ThreadId::from_raw(idx as u16);
         self.threads[idx] = Some(Thread::new(
@@ -256,11 +283,7 @@ impl ThreadTable {
         core_affinity: Option<CoreId>,
         context: SavedContext,
     ) -> Result<ThreadId, KernelError> {
-        let idx = self
+        let idx = self.free_slot().ok_or(KernelError::TooManyThreads)?;
-            .threads
-            .iter()
-            .position(|s| s.is_none())
-            .ok_or(KernelError::TooManyThreads)?;
 
         let id = ThreadId::from_raw(idx as u16);
         self.threads[idx] = Some(Thread::with_context(
@@ -292,10 +315,13 @@ impl ThreadTable {
 
     pub fn destroy(&mut self, id: ThreadId) -> Result<Thread, KernelError> {
         let idx = id.raw() as usize;
-        self.threads
+        let thread = self
+            .threads
             .get_mut(idx)
             .and_then(|s| s.take())
-            .ok_or(KernelError::InvalidThread(id))
+            .ok_or(KernelError::InvalidThread(id))?;
+        self.generations.advance(idx);
+        Ok(thread)
     }
 
     pub fn threads_for_process(&self, process: ProcessId) -> impl Iterator<Item = &Thread> {

crates/domain/src/thread/tests.rs

@@ -1,6 +1,6 @@
 use super::*;
 use types::bounded::Priority;
-use types::id::{CoreId, ProcessId, ThreadId};
+use types::id::{CoreId, EndpointId, ProcessId, ThreadId};
 use types::time::TimeSlice;
 
 fn thread(id: u16) -> ThreadId {
@@ -65,10 +65,10 @@ fn ready_to_running() {
 fn running_to_blocked() {
     let mut t = test_thread(0, 0);
     t.set_running(core(0)).unwrap();
-    t.block(BlockReason::WaitingAny).unwrap();
+    t.block(BlockReason::AwaitingReply).unwrap();
     assert!(matches!(
         t.state(),
-        ThreadState::Blocked(BlockReason::WaitingAny)
+        ThreadState::Blocked(BlockReason::AwaitingReply)
     ));
 }
 
@@ -76,7 +76,7 @@ fn running_to_blocked() {
 fn blocked_to_ready() {
     let mut t = test_thread(0, 0);
     t.set_running(core(0)).unwrap();
-    t.block(BlockReason::WaitingAny).unwrap();
+    t.block(BlockReason::AwaitingReply).unwrap();
     t.unblock().unwrap();
     assert!(matches!(t.state(), ThreadState::Ready));
 }
@@ -105,7 +105,7 @@ fn dead_cannot_transition() {
     t.set_running(core(0)).unwrap();
     t.exit().unwrap();
     assert!(t.set_running(core(0)).is_err());
-    assert!(t.block(BlockReason::WaitingAny).is_err());
+    assert!(t.block(BlockReason::AwaitingReply).is_err());
     assert!(t.unblock().is_err());
     assert!(t.preempt().is_err());
     assert!(t.exit().is_err());
@@ -114,14 +114,14 @@ fn dead_cannot_transition() {
 #[test]
 fn ready_cannot_block() {
     let mut t = test_thread(0, 0);
-    assert!(t.block(BlockReason::WaitingAny).is_err());
+    assert!(t.block(BlockReason::AwaitingReply).is_err());
 }
 
 #[test]
 fn blocked_cannot_run_directly() {
     let mut t = test_thread(0, 0);
     t.set_running(core(0)).unwrap();
-    t.block(BlockReason::WaitingAny).unwrap();
+    t.block(BlockReason::AwaitingReply).unwrap();
     assert!(t.set_running(core(0)).is_err());
 }
 
@@ -141,7 +141,7 @@ fn ready_cannot_exit() {
 fn blocked_cannot_preempt() {
     let mut t = test_thread(0, 0);
     t.set_running(core(0)).unwrap();
-    t.block(BlockReason::WaitingAny).unwrap();
+    t.block(BlockReason::AwaitingReply).unwrap();
     assert!(t.preempt().is_err());
 }
 
@@ -237,3 +237,32 @@ fn table_destroy_reuses_slot() {
         .unwrap();
     assert_eq!(id, new_id);
 }
+
+#[test]
+fn a_blocked_thread_can_change_what_it_waits_for() {
+    let mut t = Thread::new(
+        ThreadId::from_raw(1),
+        ProcessId::from_raw(0),
+        Priority::DEFAULT,
+        TimeSlice::DEFAULT,
+        None,
+    );
+    t.set_running(CoreId::from_raw(0)).unwrap();
+    t.block(BlockReason::Calling(EndpointId::from_raw(0)))
+        .unwrap();
+    t.rewait(BlockReason::AwaitingReply).unwrap();
+    assert_eq!(*t.state(), ThreadState::Blocked(BlockReason::AwaitingReply));
+}
+
+#[test]
+fn only_blocked_threads_can_rewait() {
+    let mut t = Thread::new(
+        ThreadId::from_raw(1),
+        ProcessId::from_raw(0),
+        Priority::DEFAULT,
+        TimeSlice::DEFAULT,
+        None,
+    );
+    assert!(t.rewait(BlockReason::AwaitingReply).is_err());
+    assert_eq!(*t.state(), ThreadState::Ready);
+}

crates/kernel/src/interrupt.rs

@@ -1,6 +1,6 @@
 use domain::scheduler::{Scheduler, SchedulerDecision};
 use platform::Platform;
-use types::id::{CoreId, IrqId, ThreadId};
+use types::id::CoreId;
 
 use crate::Kernel;
 
@@ -53,22 +53,4 @@ impl<P: Platform> Kernel<P> {
             self.scheduler.tick(core, current_tid)
         }
     }
-
-    pub fn handle_interrupt(&mut self, irq: IrqId) -> Option<ThreadId> {
-        self.platform.mask_irq(irq);
-
-        let tid = self.irqs.owner(irq)?;
-        let thread = self.threads.get_mut(tid).ok()?;
-
-        if !matches!(thread.state(), domain::thread::ThreadState::Blocked(_)) {
-            return None;
-        }
-
-        let priority = thread.priority();
-        let core = self.platform.core_id();
-        let _ = thread.unblock();
-        self.scheduler.add(tid, priority, core);
-
-        Some(tid)
-    }
 }

crates/kernel/src/lib.rs

@@ -18,11 +18,10 @@ use domain::{
     error::KernelError,
     irq::IrqRoutingTable,
     memory::{RegionTable, frame::FrameTable, quota::Quota},
-    message::Message,
+    notification::{MAX_NOTIFICATIONS, NotificationTable},
     process::{MAX_PROCESSES, ProcessTable},
     resource::Resource,
     scheduler::{Scheduler, multi_core::MultiCoreScheduler},
-    syscall::WaitSet,
     thread::{MAX_THREADS, ThreadState, ThreadTable},
 };
 use platform::Platform;
@@ -45,9 +44,9 @@ pub struct Kernel<P: Platform> {
     quotas: [Quota; MAX_PROCESSES],
     regions: RegionTable,
     irqs: IrqRoutingTable,
-    pending_messages: [Option<Message>; MAX_THREADS],
+    notifications: NotificationTable,
+    ipc: [syscall::IpcState; MAX_THREADS],
     pending_results: [Option<SyscallResult>; MAX_THREADS],
-    wait_sets: [Option<WaitSet>; MAX_THREADS],
 }
 
 impl<P: Platform> Kernel<P> {
@@ -63,9 +62,9 @@ impl<P: Platform> Kernel<P> {
             quotas: [Quota::default(); MAX_PROCESSES],
             regions: RegionTable::new(),
             irqs: IrqRoutingTable::new(),
-            pending_messages: [const { None }; MAX_THREADS],
+            notifications: NotificationTable::new(),
+            ipc: [syscall::IpcState::default(); MAX_THREADS],
             pending_results: [const { None }; MAX_THREADS],
-            wait_sets: [const { None }; MAX_THREADS],
         }
     }
 
@@ -95,9 +94,9 @@ impl<P: Platform> Kernel<P> {
             addr_of_mut!((*slot).quotas).write([Quota::default(); MAX_PROCESSES]);
             addr_of_mut!((*slot).regions).write(RegionTable::new());
             addr_of_mut!((*slot).irqs).write(IrqRoutingTable::new());
-            addr_of_mut!((*slot).pending_messages).write([const { None }; MAX_THREADS]);
+            addr_of_mut!((*slot).notifications).write(NotificationTable::new());
+            addr_of_mut!((*slot).ipc).write([syscall::IpcState::default(); MAX_THREADS]);
             addr_of_mut!((*slot).pending_results).write([const { None }; MAX_THREADS]);
-            addr_of_mut!((*slot).wait_sets).write([const { None }; MAX_THREADS]);
         }
     }
 
@@ -280,6 +279,7 @@ impl<P: Platform> Kernel<P> {
 
         self.cleanup_threads(&thread_ids);
         self.destroy_endpoints_owned_by(pid);
+        self.destroy_notifications_owned_by(pid);
         self.teardown_memory(pid);
 
         let _ = self.processes.destroy(pid);
@@ -289,10 +289,12 @@ impl<P: Platform> Kernel<P> {
         use domain::scheduler::Scheduler;
         for tid in thread_ids.iter().flatten() {
             self.endpoints.remove_thread_from_all(*tid);
-            self.cleanup_wait_any(*tid);
+            self.notifications.remove_thread_from_all(*tid);
-            self.irqs.unregister_thread(*tid);
+            let ipc = core::mem::take(&mut self.ipc[tid.raw() as usize]);
+            if let Some(caller) = ipc.reply_to {
+                self.release_caller(caller);
+            }
             self.scheduler.remove(*tid);
-            self.pending_messages[tid.raw() as usize] = None;
             self.pending_results[tid.raw() as usize] = None;
             let _ = self.threads.destroy(*tid);
         }
@@ -315,28 +317,31 @@ impl<P: Platform> Kernel<P> {
         }
     }
 
-    /// Makes a blocked thread runnable with `error` as its syscall result.
+    /// Destroys every notification `owner` created, waking its waiters
-    fn wake_with_error(&mut self, tid: ThreadId, error: KernelError) {
+    /// with an error and unbinding its bound thread.
-        self.cleanup_wait_any(tid);
+    pub(crate) fn destroy_notifications_owned_by(&mut self, owner: ProcessId) {
-        self.pending_messages[tid.raw() as usize] = None;
+        let mut owned = [None; MAX_NOTIFICATIONS];
-        let Ok(thread) = self.threads.get_mut(tid) else {
+        for (slot, id) in owned.iter_mut().zip(self.notifications.ids_owned_by(owner)) {
-            return;
+            *slot = Some(id);
-        };
-        if !matches!(thread.state(), ThreadState::Blocked(_)) {
-            return;
         }
-        let priority = thread.priority();
+        for id in owned.into_iter().flatten() {
-        if thread.unblock().is_ok() {
+            let Some(mut n) = self.notifications.destroy(id) else {
-            self.pending_results[tid.raw() as usize] = Some(SyscallResult::Error(error));
+                continue;
-            let core = self.platform.core_id();
+            };
-            self.scheduler.add(tid, priority, core);
+            if let Some(bound) = n.bound() {
+                self.ipc[bound.raw() as usize].bound = None;
+            }
+            while let Some(tid) = n.pop_waiter() {
+                self.wake_with_error(tid, KernelError::NoSuchObject);
+            }
         }
     }
 
-    pub(crate) fn cleanup_wait_any(&mut self, tid: ThreadId) {
+    /// Makes a blocked thread runnable with `error` as its syscall result,
-        if self.wait_sets[tid.raw() as usize].take().is_some() {
+    /// dropping whatever it was about to send.
-            self.endpoints.remove_thread_from_all(tid);
+    fn wake_with_error(&mut self, tid: ThreadId, error: KernelError) {
-        }
+        self.ipc[tid.raw() as usize].outgoing = None;
+        self.wake(tid, SyscallResult::Error(error));
     }
 
     fn check_cap(
@@ -362,6 +367,7 @@ impl<P: Platform> Kernel<P> {
         match resource {
             Resource::Endpoint(id) => self.endpoints.generation(*id),
             Resource::Memory(id) => self.regions.generation(*id),
+            Resource::Notification(id) => self.notifications.generation(*id),
             Resource::Process(pid) => self.processes.generation(*pid),
             Resource::Interrupt(_) | Resource::ProcessCreate => Some(Generation::initial()),
         }
@@ -371,6 +377,7 @@ impl<P: Platform> Kernel<P> {
         match resource {
             Resource::Endpoint(id) => self.endpoints.revoke(*id),
             Resource::Memory(id) => self.regions.revoke(*id),
+            Resource::Notification(id) => self.notifications.revoke(*id),
             _ => {}
         }
     }

crates/kernel/src/syscall/capability.rs

@@ -11,13 +11,17 @@ impl<P: Platform> Kernel<P> {
         to_endpoint: types::bounded::CapSlot,
         restrict: types::permission::PermissionSet,
     ) -> SyscallOutcome {
-        let (resource, permissions) =
+        // A copy of the capability itself, so its badge and generation
-            match self.check_cap(caller_pid, slot, types::permission::Permission::Delegate) {
+        // travel with it; only the permissions narrow.
-                Ok(cap) => (cap.resource().clone(), *cap.permissions()),
+        let copy = match self.check_cap(caller_pid, slot, types::permission::Permission::Delegate) {
-                Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
+            Ok(cap) => domain::capability::Capability::new(
-            };
+                cap.resource().clone(),
-
+                cap.permissions().restrict_to(&restrict),
-        let restricted = permissions.restrict_to(&restrict);
+                *cap.generation(),
+            )
+            .with_badge(cap.badge()),
+            Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
+        };
 
         let target_pid = match self.check_cap(
             caller_pid,
@@ -38,7 +42,11 @@ impl<P: Platform> Kernel<P> {
             Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
         };
 
-        if let Err(e) = self.grant_cap(target_pid, resource, restricted) {
+        let inserted = self
+            .processes
+            .get_mut(target_pid)
+            .and_then(|p| p.capabilities_mut().insert(copy));
+        if let Err(e) = inserted {
             return SyscallOutcome::Complete(SyscallResult::Error(e));
         }
 

crates/kernel/src/syscall/fault.rs

@@ -1,47 +1,42 @@
+use domain::{fault::Fault, message::Message};
 use platform::Platform;
+use types::{id::ThreadId, permission::Permission};
 
+use super::{Delivery, DeliveryKind, SyscallResult};
 use crate::Kernel;
 
 impl<P: Platform> Kernel<P> {
-    pub fn handle_fault(
+    /// A thread faulted. Report it to the process's fault endpoint (if one
-        &mut self,
+    /// is set, valid and has a receiver waiting), then destroy the process.
-        tid: types::id::ThreadId,
+    pub fn handle_fault(&mut self, tid: ThreadId, fault: Fault, fault_pc: usize) {
-        fault: domain::fault::Fault,
+        let Some(pid) = self.thread_process(tid) else {
-        fault_pc: usize,
+            return;
-    ) {
-        let pid = match self.threads.get(tid) {
-            Ok(t) => t.process(),
-            Err(_) => return,
         };
-
+        let handler = self
-        // Try to deliver fault to a fault handler endpoint (if configured)
+            .processes
-        if let Ok(process) = self.processes.get(pid)
+            .get(pid)
-            && let Some(handler_slot) = process.fault_handler()
+            .ok()
-            && let Ok(cap) = process.capabilities().get(handler_slot)
+            .and_then(|p| p.fault_handler())
-            && let domain::resource::Resource::Endpoint(ep_id) = cap.resource()
+            .and_then(|slot| self.check_cap(pid, slot, Permission::Execute).ok())
+            .and_then(|cap| match cap.resource() {
+                domain::resource::Resource::Endpoint(ep) => Some((*ep, cap.badge())),
+                _ => None,
+            });
+        if let Some((ep, badge)) = handler
+            && let Some(receiver) = self
+                .endpoints
+                .get_mut(ep)
+                .and_then(|e| e.dequeue_receiver())
         {
-            let msg = domain::message::Message::from_words([
+            let report =
-                pid.raw() as usize,
+                Message::from_words([pid.raw() as usize, fault.code(), fault.address(), fault_pc]);
-                fault.code(),
+            let delivery = Delivery {
-                fault.address(),
+                kind: DeliveryKind::Message,
-                fault_pc,
+                badge: badge.raw(),
-            ]);
+                words: *report.words(),
-
+                cap: None,
-            if let Some(ep) = self.endpoints.get_mut(*ep_id)
+            };
-                && ep.has_waiting_receiver()
+            self.wake(receiver, SyscallResult::Received(delivery));
-                && let Some(receiver_tid) = ep.dequeue_receiver()
-            {
-                self.pending_results[receiver_tid.raw() as usize] =
-                    Some(crate::SyscallResult::Message(msg));
-
-                if let Ok(receiver) = self.threads.get_mut(receiver_tid) {
-                    let priority = receiver.priority();
-                    let _ = receiver.unblock();
-                    let core = self.platform.core_id();
-                    use domain::scheduler::Scheduler;
-                    self.scheduler.add(receiver_tid, priority, core);
-                }
-            }
         }
 
         self.destroy_process(pid);

crates/kernel/src/syscall/ipc.rs

@@ -1,346 +1,553 @@
-use platform::Platform;
+//! Endpoint IPC, notifications and interrupt delivery. docs/ipc.md is the
+//! specification; the comments below only point at the non-obvious rules.
 
-use super::{SyscallOutcome, SyscallResult};
+use domain::{
+    capability::Capability,
+    error::KernelError,
+    message::Message,
+    notification::SignalEffect,
+    resource::Resource,
+    scheduler::Scheduler,
+    thread::{BlockReason, ThreadState},
+};
+use platform::Platform;
+use types::{
+    badge::Badge,
+    bounded::CapSlot,
+    generation::Generation,
+    id::{EndpointId, IrqId, NotificationId, ProcessId, ThreadId},
+    permission::{Permission, PermissionSet},
+};
+
+use super::{Delivery, DeliveryKind, SyscallOutcome, SyscallResult};
 use crate::Kernel;
 
+/// A thread named by id and generation, so a stale reference can be told
+/// apart from a new thread that reused the id.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct ThreadRef {
+    pub tid: ThreadId,
+    pub generation: Generation,
+}
+
+/// A message a blocked sender is waiting to hand over.
+#[derive(Debug, Clone, Copy)]
+pub(crate) struct Outgoing {
+    message: Message,
+    badge: Badge,
+    /// The sender's endpoint cap had `Delegate`, so a capability may travel.
+    grant: bool,
+}
+
+/// Per-thread IPC bookkeeping.
+#[derive(Debug, Clone, Copy, Default)]
+pub struct IpcState {
+    pub(crate) outgoing: Option<Outgoing>,
+    /// The caller this thread owes a reply to.
+    pub(crate) reply_to: Option<ThreadRef>,
+    pub(crate) bound: Option<NotificationId>,
+}
+
+fn rw_delegate() -> PermissionSet {
+    PermissionSet::empty()
+        .with(Permission::Read)
+        .with(Permission::Write)
+        .with(Permission::Delegate)
+}
+
 impl<P: Platform> Kernel<P> {
-    pub(crate) fn handle_create_endpoint(
+    fn thread_ref(&self, tid: ThreadId) -> Option<ThreadRef> {
+        Some(ThreadRef {
+            tid,
+            generation: self.threads.generation(tid)?,
+        })
+    }
+
+    fn is_live(&self, r: ThreadRef) -> bool {
+        self.threads.generation(r.tid) == Some(r.generation)
+    }
+
+    fn is_blocked_as(&self, tid: ThreadId, f: impl Fn(&BlockReason) -> bool) -> bool {
+        matches!(self.threads.get(tid).map(|t| *t.state()), Ok(ThreadState::Blocked(r)) if f(&r))
+    }
+
+    /// Blocks the running caller.
+    pub(crate) fn block_caller(&mut self, tid: ThreadId, reason: BlockReason) -> SyscallOutcome {
+        if let Ok(t) = self.threads.get_mut(tid) {
+            let _ = t.block(reason);
+        }
+        self.scheduler.remove(tid);
+        SyscallOutcome::BlockCaller(reason)
+    }
+
+    /// Makes a blocked thread runnable with `result` as its syscall result.
+    pub(crate) fn wake(&mut self, tid: ThreadId, result: SyscallResult) {
+        let Ok(thread) = self.threads.get_mut(tid) else {
+            return;
+        };
+        let priority = thread.priority();
+        if thread.unblock().is_ok() {
+            self.pending_results[tid.raw() as usize] = Some(result);
+            let core = self.platform.core_id();
+            self.scheduler.add(tid, priority, core);
+        }
+    }
+
+    /// Releases a caller waiting on `reply` with `NoReply`, if it is still
+    /// that caller and still waiting.
+    pub(crate) fn release_caller(&mut self, reply: ThreadRef) {
+        if self.is_live(reply)
+            && self.is_blocked_as(reply.tid, |r| *r == BlockReason::AwaitingReply)
+        {
+            self.wake(reply.tid, SyscallResult::Error(KernelError::NoReply));
+        }
+    }
+
+    fn endpoint_cap(
+        &self,
+        pid: ProcessId,
+        slot: CapSlot,
+        needed: Permission,
+    ) -> Result<(EndpointId, Badge, bool), KernelError> {
+        let cap = self.check_cap(pid, slot, needed)?;
+        match cap.resource() {
+            Resource::Endpoint(id) => Ok((
+                *id,
+                cap.badge(),
+                cap.permissions().allows(Permission::Delegate),
+            )),
+            _ => Err(KernelError::WrongObject),
+        }
+    }
+
+    fn notification_cap(
+        &self,
+        pid: ProcessId,
+        slot: CapSlot,
+        needed: Permission,
+    ) -> Result<(NotificationId, Badge, Generation), KernelError> {
+        let cap = self.check_cap(pid, slot, needed)?;
+        match cap.resource() {
+            Resource::Notification(id) => Ok((*id, cap.badge(), *cap.generation())),
+            _ => Err(KernelError::WrongObject),
+        }
+    }
+
+    /// Copies the capability in `slot` of `from` into `to`'s table, if the
+    /// transfer is allowed and still valid. Returns where it landed.
+    fn transfer_cap(
         &mut self,
-        caller_pid: types::id::ProcessId,
+        from: ProcessId,
+        slot: Option<CapSlot>,
+        grant: bool,
+        to: ProcessId,
+    ) -> Option<CapSlot> {
+        let slot = slot?;
+        if !grant {
+            return None;
+        }
+        let copy: Capability = self
+            .check_cap(from, slot, Permission::Delegate)
+            .ok()?
+            .clone();
+        self.processes
+            .get_mut(to)
+            .ok()?
+            .capabilities_mut()
+            .insert(copy)
+            .ok()
+    }
+
+    /// Hands `out` from `sender` to `receiver`. For a call, the receiver's
+    /// reply slot now names the sender.
+    fn deliver(
+        &mut self,
+        sender: ThreadId,
+        out: Outgoing,
+        receiver: ThreadId,
+        kind: DeliveryKind,
+    ) -> Delivery {
+        let from = self.thread_process(sender);
+        let to = self.thread_process(receiver);
+        let cap = match (from, to) {
+            (Some(from), Some(to)) => self.transfer_cap(from, out.message.cap(), out.grant, to),
+            _ => None,
+        };
+        if kind == DeliveryKind::Call {
+            self.ipc[receiver.raw() as usize].reply_to = self.thread_ref(sender);
+        }
+        Delivery {
+            kind,
+            badge: out.badge.raw(),
+            words: *out.message.words(),
+            cap,
+        }
+    }
+
+    // ---- endpoints ----
+
+    pub(crate) fn send(
+        &mut self,
+        tid: ThreadId,
+        pid: ProcessId,
+        slot: CapSlot,
+        message: Message,
+        is_call: bool,
     ) -> SyscallOutcome {
-        let ep_id = match self.create_endpoint(caller_pid) {
+        let (ep, badge, grant) = match self.endpoint_cap(pid, slot, Permission::Execute) {
+            Ok(v) => v,
+            Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
+        };
+        let out = Outgoing {
+            message,
+            badge,
+            grant,
+        };
+        let kind = if is_call {
+            DeliveryKind::Call
+        } else {
+            DeliveryKind::Message
+        };
+
+        let receiver = self
+            .endpoints
+            .get_mut(ep)
+            .and_then(|e| e.dequeue_receiver());
+        if let Some(receiver) = receiver {
+            let delivery = self.deliver(tid, out, receiver, kind);
+            self.wake(receiver, SyscallResult::Received(delivery));
+            if !is_call {
+                return SyscallOutcome::Complete(SyscallResult::Ok);
+            }
+            self.block_caller(tid, BlockReason::AwaitingReply);
+            // Run the server straight away instead of waiting for its turn.
+            let core = self.platform.core_id();
+            if self.scheduler.run_now(core, receiver)
+                && self
+                    .threads
+                    .get_mut(receiver)
+                    .and_then(|t| t.set_running(core))
+                    .is_ok()
+            {
+                return SyscallOutcome::SwitchTo(receiver);
+            }
+            return SyscallOutcome::BlockCaller(BlockReason::AwaitingReply);
+        }
+
+        let queued = self
+            .endpoints
+            .get_mut(ep)
+            .is_some_and(|e| e.enqueue_sender(tid).is_some());
+        if !queued {
+            return SyscallOutcome::Complete(SyscallResult::Error(KernelError::EndpointFull));
+        }
+        self.ipc[tid.raw() as usize].outgoing = Some(out);
+        let reason = if is_call {
+            BlockReason::Calling(ep)
+        } else {
+            BlockReason::Sending(ep)
+        };
+        self.block_caller(tid, reason)
+    }
+
+    /// `drop_reply`: a plain `Receive` abandons any pending reply.
+    pub(crate) fn receive(
+        &mut self,
+        tid: ThreadId,
+        pid: ProcessId,
+        slot: CapSlot,
+        drop_reply: bool,
+    ) -> SyscallOutcome {
+        let (ep, _, _) = match self.endpoint_cap(pid, slot, Permission::Read) {
+            Ok(v) => v,
+            Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
+        };
+        if drop_reply && let Some(pending) = self.ipc[tid.raw() as usize].reply_to.take() {
+            self.release_caller(pending);
+        }
+
+        if let Some(n) = self.ipc[tid.raw() as usize].bound
+            && let Some(ntfn) = self.notifications.get_mut(n)
+            && ntfn.has_bits()
+        {
+            let bits = ntfn.take_bits();
+            return SyscallOutcome::Complete(SyscallResult::Received(Delivery::signal(bits)));
+        }
+
+        let sender = self.endpoints.get_mut(ep).and_then(|e| e.dequeue_sender());
+        if let Some(sender) = sender {
+            let is_call = self.is_blocked_as(sender, |r| matches!(r, BlockReason::Calling(_)));
+            let Some(out) = self.ipc[sender.raw() as usize].outgoing.take() else {
+                // A queued sender always has a message; treat a missing one
+                // as a sender that vanished.
+                return SyscallOutcome::Complete(SyscallResult::Error(KernelError::WouldBlock));
+            };
+            let kind = if is_call {
+                DeliveryKind::Call
+            } else {
+                DeliveryKind::Message
+            };
+            let delivery = self.deliver(sender, out, tid, kind);
+            if is_call {
+                if let Ok(t) = self.threads.get_mut(sender) {
+                    let _ = t.rewait(BlockReason::AwaitingReply);
+                }
+            } else {
+                self.wake(sender, SyscallResult::Ok);
+            }
+            return SyscallOutcome::Complete(SyscallResult::Received(delivery));
+        }
+
+        let queued = self
+            .endpoints
+            .get_mut(ep)
+            .is_some_and(|e| e.enqueue_receiver(tid).is_some());
+        if !queued {
+            return SyscallOutcome::Complete(SyscallResult::Error(KernelError::EndpointFull));
+        }
+        self.block_caller(tid, BlockReason::Receiving(ep))
+    }
+
+    pub(crate) fn reply(
+        &mut self,
+        tid: ThreadId,
+        pid: ProcessId,
+        message: Message,
+    ) -> Result<(), KernelError> {
+        let caller = self.ipc[tid.raw() as usize]
+            .reply_to
+            .take()
+            .ok_or(KernelError::NoReply)?;
+        let waiting = self.is_live(caller)
+            && self.is_blocked_as(caller.tid, |r| *r == BlockReason::AwaitingReply);
+        if !waiting {
+            // The caller died; the reply has nowhere to go.
+            return Ok(());
+        }
+        let to = self.thread_process(caller.tid);
+        // A reply may always carry a capability the server can delegate.
+        let cap = to.and_then(|to| self.transfer_cap(pid, message.cap(), true, to));
+        let delivery = Delivery {
+            kind: DeliveryKind::Message,
+            badge: 0,
+            words: *message.words(),
+            cap,
+        };
+        self.wake(caller.tid, SyscallResult::Received(delivery));
+        Ok(())
+    }
+
+    pub(crate) fn reply_recv(
+        &mut self,
+        tid: ThreadId,
+        pid: ProcessId,
+        slot: CapSlot,
+        message: Message,
+    ) -> SyscallOutcome {
+        if let Err(e) = self.endpoint_cap(pid, slot, Permission::Read) {
+            return SyscallOutcome::Complete(SyscallResult::Error(e));
+        }
+        if self.ipc[tid.raw() as usize].reply_to.is_some() {
+            let _ = self.reply(tid, pid, message);
+        }
+        self.receive(tid, pid, slot, false)
+    }
+
+    pub(crate) fn mint(
+        &mut self,
+        pid: ProcessId,
+        slot: CapSlot,
+        badge: Badge,
+        permissions: PermissionSet,
+    ) -> Result<CapSlot, KernelError> {
+        let source = self.check_cap(pid, slot, Permission::Delegate)?;
+        if !matches!(
+            source.resource(),
+            Resource::Endpoint(_) | Resource::Notification(_)
+        ) {
+            return Err(KernelError::WrongObject);
+        }
+        if !source.badge().is_none() {
+            return Err(KernelError::InvalidArgument);
+        }
+        let minted = Capability::new(
+            source.resource().clone(),
+            source.permissions().restrict_to(&permissions),
+            *source.generation(),
+        )
+        .with_badge(badge);
+        self.processes
+            .get_mut(pid)?
+            .capabilities_mut()
+            .insert(minted)
+    }
+
+    pub(crate) fn handle_create_endpoint(&mut self, pid: ProcessId) -> SyscallOutcome {
+        let id = match self.create_endpoint(pid) {
             Ok(id) => id,
             Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
         };
-
+        let perms = PermissionSet::empty()
-        let all_perms = types::permission::PermissionSet::empty()
+            .with(Permission::Read)
-            .with(types::permission::Permission::Read)
+            .with(Permission::Execute)
-            .with(types::permission::Permission::Write)
+            .with(Permission::Delegate);
-            .with(types::permission::Permission::Execute)
+        match self.grant_cap(pid, Resource::Endpoint(id), perms) {
-            .with(types::permission::Permission::Delegate);
-
-        match self.grant_cap(
-            caller_pid,
-            domain::resource::Resource::Endpoint(ep_id),
-            all_perms,
-        ) {
             Ok(slot) => SyscallOutcome::Complete(SyscallResult::Slot(slot)),
             Err(e) => {
-                let _ = self.endpoints.destroy(ep_id);
+                self.endpoints.destroy(id);
                 SyscallOutcome::Complete(SyscallResult::Error(e))
             }
         }
     }
 
-    pub(crate) fn handle_send(
+    // ---- notifications ----
-        &mut self,
-        caller_tid: types::id::ThreadId,
-        caller_pid: types::id::ProcessId,
-        endpoint_slot: types::bounded::CapSlot,
-        message: domain::message::Message,
-    ) -> SyscallOutcome {
-        use domain::scheduler::Scheduler;
 
-        let ep_id = match self.check_cap(
+    pub fn create_notification(&mut self, owner: ProcessId) -> Result<NotificationId, KernelError> {
-            caller_pid,
+        self.processes.get(owner)?;
-            endpoint_slot,
+        self.notifications
-            types::permission::Permission::Execute,
+            .create(owner)
-        ) {
+            .ok_or(KernelError::NoFreeSlot)
-            Ok(cap) => match cap.resource() {
+    }
-                domain::resource::Resource::Endpoint(id) => *id,
+
-                _ => {
+    pub(crate) fn create_notification_cap(
-                    return SyscallOutcome::Complete(SyscallResult::Error(
+        &mut self,
-                        domain::error::KernelError::InvalidEndpoint(
+        pid: ProcessId,
-                            types::id::EndpointId::from_raw(0),
+    ) -> Result<CapSlot, KernelError> {
-                        ),
+        let id = self.create_notification(pid)?;
-                    ));
+        self.grant_cap(pid, Resource::Notification(id), rw_delegate())
-                }
+            .inspect_err(|_| {
-            },
+                self.notifications.destroy(id);
+            })
+    }
+
+    /// Signals `id` with `badge`; shared by `Signal` and interrupts.
+    fn signal_notification(&mut self, id: NotificationId, badge: Badge) {
+        let bound_receiving = self
+            .notifications
+            .get(id)
+            .and_then(|n| n.bound())
+            .is_some_and(|t| self.is_blocked_as(t, |r| matches!(r, BlockReason::Receiving(_))));
+        let Some(ntfn) = self.notifications.get_mut(id) else {
+            return;
+        };
+        match ntfn.signal(badge, bound_receiving) {
+            SignalEffect::WakeWaiter(t, bits) => {
+                self.wake(t, SyscallResult::Received(Delivery::signal(bits)));
+            }
+            SignalEffect::WakeBound(t, bits) => {
+                self.endpoints.remove_thread_from_all(t);
+                self.wake(t, SyscallResult::Received(Delivery::signal(bits)));
+            }
+            SignalEffect::Stored => {}
+        }
+    }
+
+    pub(crate) fn signal(&mut self, pid: ProcessId, slot: CapSlot) -> Result<(), KernelError> {
+        let (id, badge, _) = self.notification_cap(pid, slot, Permission::Write)?;
+        self.signal_notification(id, badge);
+        Ok(())
+    }
+
+    pub(crate) fn wait(&mut self, tid: ThreadId, pid: ProcessId, slot: CapSlot) -> SyscallOutcome {
+        let id = match self.notification_cap(pid, slot, Permission::Read) {
+            Ok((id, _, _)) => id,
             Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
         };
-
+        let Some(ntfn) = self.notifications.get_mut(id) else {
-        let endpoint = match self.endpoints.get_mut(ep_id) {
+            return SyscallOutcome::Complete(SyscallResult::Error(KernelError::NoSuchObject));
-            Some(ep) => ep,
-            None => {
-                return SyscallOutcome::Complete(SyscallResult::Error(
-                    domain::error::KernelError::InvalidEndpoint(ep_id),
-                ));
-            }
         };
+        if ntfn.has_bits() {
+            let bits = ntfn.take_bits();
+            return SyscallOutcome::Complete(SyscallResult::Received(Delivery::signal(bits)));
+        }
+        if !ntfn.add_waiter(tid) {
+            return SyscallOutcome::Complete(SyscallResult::Error(KernelError::EndpointFull));
+        }
+        self.block_caller(tid, BlockReason::WaitingNotification(id))
+    }
 
-        if endpoint.has_waiting_receiver() {
+    pub(crate) fn poll(&mut self, pid: ProcessId, slot: CapSlot) -> Result<usize, KernelError> {
-            let receiver_tid = match endpoint.dequeue_receiver() {
+        let (id, _, _) = self.notification_cap(pid, slot, Permission::Read)?;
-                Some(t) => t,
+        self.notifications
-                None => {
+            .get_mut(id)
-                    return SyscallOutcome::Complete(SyscallResult::Error(
+            .map(|n| n.take_bits())
-                        domain::error::KernelError::WouldBlock,
+            .ok_or(KernelError::NoSuchObject)
-                    ));
+    }
-                }
-            };
 
-            self.pending_results[receiver_tid.raw() as usize] =
+    pub(crate) fn bind_notification(
-                Some(SyscallResult::Message(message));
+        &mut self,
+        tid: ThreadId,
+        pid: ProcessId,
+        slot: CapSlot,
+    ) -> Result<(), KernelError> {
+        let (id, _, _) = self.notification_cap(pid, slot, Permission::Read)?;
+        if self.ipc[tid.raw() as usize].bound.is_some() {
+            return Err(KernelError::InvalidArgument);
+        }
+        let ntfn = self
+            .notifications
+            .get_mut(id)
+            .ok_or(KernelError::NoSuchObject)?;
+        if !ntfn.bind(tid) {
+            return Err(KernelError::InvalidArgument);
+        }
+        self.ipc[tid.raw() as usize].bound = Some(id);
+        Ok(())
+    }
 
-            let core = self.platform.core_id();
+    // ---- interrupts ----
 
-            let receiver = match self.threads.get_mut(receiver_tid) {
+    fn interrupt_cap(&self, pid: ProcessId, slot: CapSlot) -> Result<IrqId, KernelError> {
-                Ok(t) => t,
+        match self.check_cap(pid, slot, Permission::Execute)?.resource() {
-                Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
+            Resource::Interrupt(irq) => Ok(*irq),
-            };
+            _ => Err(KernelError::WrongObject),
-            let _ = receiver.unblock();
+        }
-            let _ = receiver.set_running(core);
+    }
 
-            let caller = match self.threads.get_mut(caller_tid) {
+    pub(crate) fn set_irq_notification(
-                Ok(t) => t,
+        &mut self,
-                Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
+        pid: ProcessId,
-            };
+        irq_slot: CapSlot,
-            let caller_priority = caller.priority();
+        ntfn_slot: CapSlot,
-            let _ = caller.preempt();
+    ) -> Result<(), KernelError> {
-            self.scheduler.add(caller_tid, caller_priority, core);
+        let irq = self.interrupt_cap(pid, irq_slot)?;
-
+        let (notification, badge, generation) =
-            self.scheduler.set_current(core, receiver_tid);
+            self.notification_cap(pid, ntfn_slot, Permission::Write)?;
-
+        let route = domain::irq::IrqRoute {
-            SyscallOutcome::SwitchTo(receiver_tid)
+            notification,
+            generation,
+            badge,
+        };
+        if self.irqs.set(irq, route) {
+            Ok(())
         } else {
-            self.pending_messages[caller_tid.raw() as usize] = Some(message);
+            Err(KernelError::InvalidIrq(irq))
-
-            let endpoint = match self.endpoints.get_mut(ep_id) {
-                Some(ep) => ep,
-                None => {
-                    return SyscallOutcome::Complete(SyscallResult::Error(
-                        domain::error::KernelError::InvalidEndpoint(ep_id),
-                    ));
-                }
-            };
-            if endpoint.enqueue_sender(caller_tid).is_none() {
-                self.pending_messages[caller_tid.raw() as usize] = None;
-                return SyscallOutcome::Complete(SyscallResult::Error(
-                    domain::error::KernelError::WouldBlock,
-                ));
-            }
-
-            let caller = match self.threads.get_mut(caller_tid) {
-                Ok(t) => t,
-                Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
-            };
-            let _ = caller.block(domain::thread::BlockReason::Sending(endpoint_slot));
-            self.scheduler.remove(caller_tid);
-
-            SyscallOutcome::BlockCaller(domain::thread::BlockReason::Sending(endpoint_slot))
         }
     }
 
-    pub(crate) fn handle_receive(
+    pub(crate) fn ack_interrupt(
         &mut self,
-        caller_tid: types::id::ThreadId,
+        pid: ProcessId,
-        caller_pid: types::id::ProcessId,
+        slot: CapSlot,
-        endpoint_slot: types::bounded::CapSlot,
+    ) -> Result<(), KernelError> {
-    ) -> SyscallOutcome {
+        let irq = self.interrupt_cap(pid, slot)?;
-        use domain::scheduler::Scheduler;
+        self.platform.unmask_irq(irq);
-
+        Ok(())
-        let ep_id = match self.check_cap(
-            caller_pid,
-            endpoint_slot,
-            types::permission::Permission::Read,
-        ) {
-            Ok(cap) => match cap.resource() {
-                domain::resource::Resource::Endpoint(id) => *id,
-                _ => {
-                    return SyscallOutcome::Complete(SyscallResult::Error(
-                        domain::error::KernelError::InvalidEndpoint(
-                            types::id::EndpointId::from_raw(0),
-                        ),
-                    ));
-                }
-            },
-            Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
-        };
-
-        let endpoint = match self.endpoints.get_mut(ep_id) {
-            Some(ep) => ep,
-            None => {
-                return SyscallOutcome::Complete(SyscallResult::Error(
-                    domain::error::KernelError::InvalidEndpoint(ep_id),
-                ));
-            }
-        };
-
-        if endpoint.has_waiting_sender() {
-            let sender_tid = match endpoint.dequeue_sender() {
-                Some(t) => t,
-                None => {
-                    return SyscallOutcome::Complete(SyscallResult::Error(
-                        domain::error::KernelError::WouldBlock,
-                    ));
-                }
-            };
-
-            let message = self.pending_messages[sender_tid.raw() as usize].take();
-            debug_assert!(
-                message.is_some(),
-                "sender in wait queue had no pending message"
-            );
-            let message = message.unwrap_or_else(domain::message::Message::empty);
-
-            self.pending_results[sender_tid.raw() as usize] = Some(SyscallResult::Ok);
-
-            let sender = match self.threads.get_mut(sender_tid) {
-                Ok(t) => t,
-                Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
-            };
-            let priority = sender.priority();
-            let _ = sender.unblock();
-            let core = self.platform.core_id();
-            self.scheduler.add(sender_tid, priority, core);
-
-            SyscallOutcome::Complete(SyscallResult::Message(message))
-        } else {
-            let endpoint = match self.endpoints.get_mut(ep_id) {
-                Some(ep) => ep,
-                None => {
-                    return SyscallOutcome::Complete(SyscallResult::Error(
-                        domain::error::KernelError::InvalidEndpoint(ep_id),
-                    ));
-                }
-            };
-            if endpoint.enqueue_receiver(caller_tid).is_none() {
-                return SyscallOutcome::Complete(SyscallResult::Error(
-                    domain::error::KernelError::WouldBlock,
-                ));
-            }
-
-            let caller = match self.threads.get_mut(caller_tid) {
-                Ok(t) => t,
-                Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
-            };
-            let _ = caller.block(domain::thread::BlockReason::Receiving(endpoint_slot));
-            self.scheduler.remove(caller_tid);
-
-            SyscallOutcome::BlockCaller(domain::thread::BlockReason::Receiving(endpoint_slot))
-        }
     }
 
-    pub(crate) fn handle_wait_for_interrupt(
+    /// An interrupt fired: mask the line and signal its notification. The
-        &mut self,
+    /// line stays masked until the handler acknowledges it.
-        caller_tid: types::id::ThreadId,
+    pub fn handle_interrupt(&mut self, irq: IrqId) {
-        caller_pid: types::id::ProcessId,
+        self.platform.mask_irq(irq);
-        irq_slot: types::bounded::CapSlot,
+        let Some(route) = self.irqs.route(irq) else {
-    ) -> SyscallOutcome {
+            return;
-        use domain::scheduler::Scheduler;
-
-        let irq_id = match self.check_cap(caller_pid, irq_slot, types::permission::Permission::Read)
-        {
-            Ok(cap) => match cap.resource() {
-                domain::resource::Resource::Interrupt(id) => *id,
-                _ => {
-                    return SyscallOutcome::Complete(SyscallResult::Error(
-                        domain::error::KernelError::InvalidIrq(types::id::IrqId::from_raw(0)),
-                    ));
-                }
-            },
-            Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
         };
-
+        // A route set through a capability that has since been revoked, or
-        let _ = self.irqs.register(irq_id, caller_tid);
+        // to a notification whose slot was reused, is dead.
-
+        if self.notifications.generation(route.notification) != Some(route.generation) {
-        let caller = match self.threads.get_mut(caller_tid) {
+            self.irqs.clear(irq);
-            Ok(t) => t,
+            return;
-            Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
-        };
-        let _ = caller.block(domain::thread::BlockReason::WaitingForInterrupt(irq_slot));
-        self.scheduler.remove(caller_tid);
-
-        SyscallOutcome::BlockCaller(domain::thread::BlockReason::WaitingForInterrupt(irq_slot))
-    }
-
-    pub(crate) fn handle_ack_interrupt(
-        &mut self,
-        caller_pid: types::id::ProcessId,
-        irq_slot: types::bounded::CapSlot,
-    ) -> SyscallOutcome {
-        let irq_id =
-            match self.check_cap(caller_pid, irq_slot, types::permission::Permission::Write) {
-                Ok(cap) => match cap.resource() {
-                    domain::resource::Resource::Interrupt(id) => *id,
-                    _ => {
-                        return SyscallOutcome::Complete(SyscallResult::Error(
-                            domain::error::KernelError::InvalidIrq(types::id::IrqId::from_raw(0)),
-                        ));
-                    }
-                },
-                Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
-            };
-
-        self.platform.unmask_irq(irq_id);
-        SyscallOutcome::Complete(SyscallResult::Ok)
-    }
-
-    pub(crate) fn handle_wait_any(
-        &mut self,
-        caller_tid: types::id::ThreadId,
-        caller_pid: types::id::ProcessId,
-        sources: domain::syscall::WaitSet,
-    ) -> SyscallOutcome {
-        use domain::scheduler::Scheduler;
-
-        if sources.is_empty() {
-            return SyscallOutcome::Complete(SyscallResult::Error(
-                domain::error::KernelError::WouldBlock,
-            ));
         }
-
+        self.signal_notification(route.notification, route.badge);
-        for slot in sources.endpoints().iter().flatten() {
-            let ep_id = match self.check_cap(caller_pid, *slot, types::permission::Permission::Read)
-            {
-                Ok(cap) => match cap.resource() {
-                    domain::resource::Resource::Endpoint(id) => *id,
-                    _ => continue,
-                },
-                Err(_) => continue,
-            };
-
-            if let Some(ep) = self.endpoints.get_mut(ep_id)
-                && ep.has_waiting_sender()
-            {
-                return SyscallOutcome::Complete(SyscallResult::WakeReason(
-                    domain::syscall::WakeReason::MessageReady(*slot),
-                ));
-            }
-        }
-
-        for irq_slot in sources.interrupts() {
-            if let Some(slot) = irq_slot
-                && let Ok(cap) =
-                    self.check_cap(caller_pid, *slot, types::permission::Permission::Read)
-                && let domain::resource::Resource::Interrupt(irq_id) = cap.resource()
-                && self.irqs.owner(*irq_id).is_some()
-            {
-                return SyscallOutcome::Complete(SyscallResult::WakeReason(
-                    domain::syscall::WakeReason::InterruptFired(*slot),
-                ));
-            }
-        }
-
-        for ep_slot in sources.endpoints() {
-            if let Some(slot) = ep_slot
-                && let Ok(cap) =
-                    self.check_cap(caller_pid, *slot, types::permission::Permission::Read)
-                && let domain::resource::Resource::Endpoint(id) = cap.resource()
-                && let Some(ep) = self.endpoints.get_mut(*id)
-            {
-                let _ = ep.enqueue_receiver(caller_tid);
-            }
-        }
-
-        self.wait_sets[caller_tid.raw() as usize] = Some(sources);
-
-        let caller = match self.threads.get_mut(caller_tid) {
-            Ok(t) => t,
-            Err(e) => return SyscallOutcome::Complete(SyscallResult::Error(e)),
-        };
-        let _ = caller.block(domain::thread::BlockReason::WaitingAny);
-        self.scheduler.remove(caller_tid);
-
-        SyscallOutcome::BlockCaller(domain::thread::BlockReason::WaitingAny)
     }
 }

crates/kernel/src/syscall/mod.rs

@@ -5,28 +5,95 @@ mod lifecycle;
 mod memory;
 mod simple;
 
-use domain::{error::KernelError, message::Message, syscall::Syscall};
+use domain::{error::KernelError, syscall::Syscall};
 use platform::Platform;
-use types::{bounded::CapSlot, id::ThreadId};
+use types::{bounded::CapSlot, id::ThreadId, syscall_abi::ErrorCode};
 
 use crate::Kernel;
 
+pub use ipc::{IpcState, ThreadRef};
+
 pub enum SyscallOutcome {
     Complete(SyscallResult),
     BlockCaller(domain::thread::BlockReason),
+    /// The caller blocked and `ThreadId` should run next.
     SwitchTo(ThreadId),
     TerminateCaller,
 }
 
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum DeliveryKind {
+    /// A `Send`; no reply expected.
+    Message,
+    /// A `Call`; the receiver's reply slot now names the caller.
+    Call,
+    /// Notification bits.
+    Signal,
+}
+
+/// What a receiving syscall hands back (docs/ipc.md, register ABI).
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct Delivery {
+    pub kind: DeliveryKind,
+    /// The sender's badge, or the signal bits.
+    pub badge: usize,
+    pub words: [usize; 4],
+    /// Where a transferred capability landed in the receiver's table.
+    pub cap: Option<CapSlot>,
+}
+
+impl Delivery {
+    pub fn signal(bits: usize) -> Self {
+        Self {
+            kind: DeliveryKind::Signal,
+            badge: bits,
+            words: [0; 4],
+            cap: None,
+        }
+    }
+}
+
+#[derive(Debug)]
 pub enum SyscallResult {
     Ok,
     Value(usize),
     Slot(CapSlot),
-    Message(Message),
+    Received(Delivery),
-    WakeReason(domain::syscall::WakeReason),
     Error(KernelError),
 }
 
+/// The status code userspace sees for a kernel error.
+pub fn error_code(error: &KernelError) -> ErrorCode {
+    use KernelError::*;
+    match error {
+        InvalidSlot(_) | RevokedCapability(_) | NoSuchObject | InvalidEndpoint(_) => {
+            ErrorCode::InvalidCap
+        }
+        PermissionDenied { .. } | ProcessCreateDenied => ErrorCode::PermissionDenied,
+        WrongObject => ErrorCode::WrongObject,
+        NoFreeSlot => ErrorCode::NoFreeSlot,
+        OutOfMemory => ErrorCode::OutOfMemory,
+        InvalidAddress => ErrorCode::InvalidAddress,
+        AddressInUse => ErrorCode::AddressInUse,
+        NoReply => ErrorCode::NoReply,
+        TooManyThreadsInProcess | TooManyProcesses | TooManyThreads | EndpointFull => {
+            ErrorCode::LimitReached
+        }
+        InvalidProcess(_)
+        | InvalidThread(_)
+        | InvalidStateTransition(_)
+        | InvalidIrq(_)
+        | InvalidRegion
+        | IrqAlreadyOwned(_)
+        | WouldBlock
+        | InvalidArgument => ErrorCode::InvalidArgument,
+    }
+}
+
+fn complete(result: Result<SyscallResult, KernelError>) -> SyscallOutcome {
+    SyscallOutcome::Complete(result.unwrap_or_else(SyscallResult::Error))
+}
+
 impl<P: Platform> Kernel<P> {
     pub fn handle_syscall(&mut self, syscall: Syscall) -> SyscallOutcome {
         let (tid, pid) = match self.caller() {
@@ -54,11 +121,45 @@ impl<P: Platform> Kernel<P> {
                 to_endpoint,
                 permissions,
             } => self.handle_delegate(pid, slot, to_endpoint, permissions),
-            Syscall::Send { endpoint, message } => self.handle_send(tid, pid, endpoint, message),
+            Syscall::Mint {
-            Syscall::Receive { endpoint } => self.handle_receive(tid, pid, endpoint),
+                slot,
-            Syscall::WaitForInterrupt { irq } => self.handle_wait_for_interrupt(tid, pid, irq),
+                badge,
-            Syscall::AcknowledgeInterrupt { irq } => self.handle_ack_interrupt(pid, irq),
+                permissions,
-            Syscall::WaitAny { sources } => self.handle_wait_any(tid, pid, sources),
+            } => complete(
+                self.mint(pid, slot, badge, permissions)
+                    .map(SyscallResult::Slot),
+            ),
+            Syscall::Send { endpoint, message } => self.send(tid, pid, endpoint, message, false),
+            Syscall::Call { endpoint, message } => self.send(tid, pid, endpoint, message, true),
+            Syscall::Receive { endpoint } => self.receive(tid, pid, endpoint, true),
+            Syscall::Reply { message } => {
+                complete(self.reply(tid, pid, message).map(|()| SyscallResult::Ok))
+            }
+            Syscall::ReplyRecv { endpoint, message } => {
+                self.reply_recv(tid, pid, endpoint, message)
+            }
+            Syscall::CreateNotification => {
+                complete(self.create_notification_cap(pid).map(SyscallResult::Slot))
+            }
+            Syscall::Signal { notification } => {
+                complete(self.signal(pid, notification).map(|()| SyscallResult::Ok))
+            }
+            Syscall::Wait { notification } => self.wait(tid, pid, notification),
+            Syscall::Poll { notification } => complete(
+                self.poll(pid, notification)
+                    .map(|bits| SyscallResult::Received(Delivery::signal(bits))),
+            ),
+            Syscall::BindNotification { notification } => complete(
+                self.bind_notification(tid, pid, notification)
+                    .map(|()| SyscallResult::Ok),
+            ),
+            Syscall::SetIrqNotification { irq, notification } => complete(
+                self.set_irq_notification(pid, irq, notification)
+                    .map(|()| SyscallResult::Ok),
+            ),
+            Syscall::AcknowledgeInterrupt { irq } => {
+                complete(self.ack_interrupt(pid, irq).map(|()| SyscallResult::Ok))
+            }
             Syscall::Halt => self.handle_halt(pid),
             Syscall::CreateEndpoint => self.handle_create_endpoint(pid),
             Syscall::DebugPutchar { byte } => self.handle_debug_putchar(byte),

crates/kernel/src/tests/helpers.rs

@@ -5,7 +5,7 @@ use platform::Platform;
 use types::{
     address::PhysicalAddress,
     bounded::{CapSlot, Priority},
-    id::{EndpointId, IrqId, ProcessId, ThreadId},
+    id::{EndpointId, ProcessId, ThreadId},
     permission::PermissionSet,
     time::TimeSlice,
 };
@@ -41,10 +41,6 @@ impl Kernel<MockPlatform> {
         tid
     }
 
-    pub fn register_irq_for_test(&mut self, irq: IrqId, thread: ThreadId) {
-        let _ = self.irqs.register(irq, thread);
-    }
-
     pub fn block_thread_for_test(&mut self, tid: ThreadId, reason: domain::thread::BlockReason) {
         let core = self.platform.core_id();
         if let Ok(t) = self.threads.get_mut(tid) {
@@ -128,6 +124,10 @@ impl Kernel<MockPlatform> {
         }
     }
 
+    pub fn destroy_process_notifications_for_test(&mut self, owner: ProcessId) {
+        self.destroy_notifications_owned_by(owner);
+    }
+
     pub fn thread_state_for_test(&self, tid: ThreadId) -> Option<ThreadState> {
         self.threads.get(tid).ok().map(|t| *t.state())
     }

crates/kernel/src/tests/ipc_invariants.rs

@@ -0,0 +1,350 @@
+//! Randomized check of the IPC state machine. Six threads in three
+//! processes send, call, receive, reply and signal at random over two
+//! endpoints and a notification; one process is periodically killed and
+//! re-created. After every step the kernel's queues, block reasons, reply
+//! slots and run queue must agree, and no message may be duplicated.
+
+use super::*;
+use crate::syscall::{Delivery, DeliveryKind};
+use crate::{SyscallOutcome, SyscallResult};
+use domain::message::Message;
+use domain::resource::Resource;
+use domain::syscall::Syscall;
+use domain::thread::{BlockReason, ThreadState};
+use types::bounded::CapSlot;
+use types::id::{EndpointId, NotificationId, ThreadId};
+use types::permission::{Permission, PermissionSet};
+
+const STEPS: usize = 4000;
+const THREADS_PER_PROC: usize = 2;
+const PROCS: usize = 3;
+
+struct Rng(u64);
+
+impl Rng {
+    fn next(&mut self) -> u64 {
+        self.0 ^= self.0 << 13;
+        self.0 ^= self.0 >> 7;
+        self.0 ^= self.0 << 17;
+        self.0
+    }
+
+    fn below(&mut self, n: usize) -> usize {
+        (self.next() % n as u64) as usize
+    }
+}
+
+fn all_perms() -> PermissionSet {
+    PermissionSet::empty()
+        .with(Permission::Read)
+        .with(Permission::Write)
+        .with(Permission::Execute)
+        .with(Permission::Delegate)
+}
+
+struct Proc {
+    pid: ProcessId,
+    threads: [ThreadId; THREADS_PER_PROC],
+    eps: [CapSlot; 2],
+    ntfn: CapSlot,
+}
+
+struct World {
+    k: Kernel<MockPlatform>,
+    procs: [Proc; PROCS],
+    eps: [EndpointId; 2],
+    ntfn: NotificationId,
+    next_id: usize,
+    delivered: [bool; 2 * STEPS + 16],
+    sent: [bool; 2 * STEPS + 16],
+    calls_answered: usize,
+    signals_seen: usize,
+    ran: usize,
+    rebirths: usize,
+}
+
+impl World {
+    fn new() -> Self {
+        let (mut k, p0) = kernel_with_process();
+        let t0 = k.current_thread().unwrap();
+        let p1 = k.create_process_internal().unwrap();
+        let p2 = k.create_process_internal().unwrap();
+        // Endpoints and the notification belong to init, which never dies.
+        let eps = [
+            k.create_endpoint(p0).unwrap(),
+            k.create_endpoint(p0).unwrap(),
+        ];
+        let ntfn = k.create_notification(p0).unwrap();
+        let mut procs = [p0, p1, p2].map(|pid| Proc {
+            pid,
+            threads: [t0; THREADS_PER_PROC],
+            eps: [CapSlot::new(0).unwrap(); 2],
+            ntfn: CapSlot::new(0).unwrap(),
+        });
+        for (i, p) in procs.iter_mut().enumerate() {
+            Self::equip(&mut k, p, eps, ntfn, i == 0, t0);
+        }
+        Self {
+            k,
+            procs,
+            eps,
+            ntfn,
+            next_id: 1,
+            delivered: [false; 2 * STEPS + 16],
+            sent: [false; 2 * STEPS + 16],
+            calls_answered: 0,
+            signals_seen: 0,
+            ran: 0,
+            rebirths: 0,
+        }
+    }
+
+    fn equip(
+        k: &mut Kernel<MockPlatform>,
+        p: &mut Proc,
+        eps: [EndpointId; 2],
+        ntfn: NotificationId,
+        is_init: bool,
+        init_thread: ThreadId,
+    ) {
+        for j in 0..THREADS_PER_PROC {
+            p.threads[j] = if is_init && j == 0 {
+                init_thread
+            } else {
+                k.spawn_thread_for_test(p.pid, Priority::DEFAULT, TimeSlice::DEFAULT)
+            };
+        }
+        for (e, ep) in eps.iter().enumerate() {
+            p.eps[e] = k.grant_cap_for_test(p.pid, Resource::Endpoint(*ep), all_perms());
+        }
+        p.ntfn = k.grant_cap_for_test(p.pid, Resource::Notification(ntfn), all_perms());
+        // The second thread of every process gets the notification bound.
+        k.switch_to_thread_for_test(p.threads[1]);
+        let _ = k.handle_syscall(Syscall::BindNotification {
+            notification: p.ntfn,
+        });
+    }
+
+    fn fresh_id(&mut self) -> usize {
+        let id = self.next_id;
+        self.next_id += 1;
+        self.sent[id] = true;
+        id
+    }
+
+    fn record(&mut self, d: &Delivery) {
+        match d.kind {
+            DeliveryKind::Signal => self.signals_seen += 1,
+            _ => {
+                let id = d.words[0];
+                assert!(
+                    id < self.sent.len() && self.sent[id],
+                    "delivered id {id} was never sent"
+                );
+                assert!(!self.delivered[id], "message {id} delivered twice");
+                self.delivered[id] = true;
+            }
+        }
+    }
+
+    fn record_outcome(&mut self, outcome: &SyscallOutcome) {
+        if let SyscallOutcome::Complete(SyscallResult::Received(d)) = outcome {
+            let d = *d;
+            self.record(&d);
+        }
+    }
+
+    fn runnable(&self) -> impl Iterator<Item = (usize, ThreadId)> + '_ {
+        self.procs.iter().enumerate().flat_map(move |(pi, p)| {
+            p.threads
+                .into_iter()
+                .filter(|&t| {
+                    matches!(
+                        self.k.thread_state_for_test(t),
+                        Some(ThreadState::Ready | ThreadState::Running(_))
+                    )
+                })
+                .map(move |t| (pi, t))
+        })
+    }
+
+    /// Kills process `pi` and brings it back with fresh threads and caps.
+    fn reincarnate(&mut self, pi: usize) {
+        self.k.destroy_process(self.procs[pi].pid);
+        self.procs[pi].pid = self.k.create_process_internal().unwrap();
+        let t0 = self.procs[0].threads[0];
+        Self::equip(
+            &mut self.k,
+            &mut self.procs[pi],
+            self.eps,
+            self.ntfn,
+            false,
+            t0,
+        );
+        self.rebirths += 1;
+    }
+
+    fn step(&mut self, rng: &mut Rng) {
+        let count = self.runnable().count();
+        if count == 0 || rng.below(100) == 0 {
+            // Random clients deadlock; killing someone is what breaks it.
+            self.reincarnate(1 + rng.below(PROCS - 1));
+            return;
+        }
+        let pick = rng.below(count);
+        let (pi, tid) = self.runnable().nth(pick).unwrap();
+        self.ran += 1;
+        self.k.switch_to_thread_for_test(tid);
+        // What the adapter would hand the thread when it resumes.
+        if let Some(SyscallResult::Received(d)) = self.k.take_pending_result(tid) {
+            self.record(&d);
+        }
+        let ep = self.procs[pi].eps[rng.below(2)];
+        let ntfn = self.procs[pi].ntfn;
+
+        let call = match rng.below(9) {
+            0 => Syscall::Send {
+                endpoint: ep,
+                message: Message::from_words([self.fresh_id(), 0, 0, 0]),
+            },
+            1 | 2 => Syscall::Call {
+                endpoint: ep,
+                message: Message::from_words([self.fresh_id(), 1, 0, 0]),
+            },
+            3 | 4 => Syscall::Receive { endpoint: ep },
+            5 => Syscall::Reply {
+                message: Message::from_words([self.fresh_id(), 2, 0, 0]),
+            },
+            6 => Syscall::ReplyRecv {
+                endpoint: ep,
+                message: Message::from_words([self.fresh_id(), 2, 0, 0]),
+            },
+            7 => Syscall::Signal { notification: ntfn },
+            _ => match rng.below(2) {
+                0 => Syscall::Wait { notification: ntfn },
+                _ => Syscall::Poll { notification: ntfn },
+            },
+        };
+        let is_reply = matches!(call, Syscall::Reply { .. } | Syscall::ReplyRecv { .. });
+        let had_reply = self.k.ipc[tid.raw() as usize].reply_to.is_some();
+        let outcome = self.k.handle_syscall(call);
+        if is_reply && had_reply {
+            self.calls_answered += 1;
+        }
+        self.record_outcome(&outcome);
+    }
+
+    fn check(&self) {
+        let k = &self.k;
+        for p in &self.procs {
+            for &t in &p.threads {
+                let Some(state) = k.thread_state_for_test(t) else {
+                    continue;
+                };
+                let ipc = &k.ipc[t.raw() as usize];
+                let queued_on = |e: EndpointId| {
+                    let ep = k.endpoints.get(e).unwrap();
+                    (ep.is_sender(t), ep.is_receiver(t))
+                };
+                let in_any_queue = self.eps.iter().any(|&e| {
+                    let (s, r) = queued_on(e);
+                    s || r
+                }) || k.notifications.get(self.ntfn).unwrap().is_waiter(t);
+
+                match state {
+                    ThreadState::Blocked(BlockReason::Sending(e) | BlockReason::Calling(e)) => {
+                        assert_eq!(
+                            queued_on(e),
+                            (true, false),
+                            "{t:?} blocked sending but not queued"
+                        );
+                        assert!(
+                            ipc.outgoing.is_some(),
+                            "{t:?} queued sender without a message"
+                        );
+                    }
+                    ThreadState::Blocked(BlockReason::Receiving(e)) => {
+                        assert_eq!(
+                            queued_on(e),
+                            (false, true),
+                            "{t:?} blocked receiving but not queued"
+                        );
+                    }
+                    ThreadState::Blocked(BlockReason::WaitingNotification(n)) => {
+                        assert!(k.notifications.get(n).unwrap().is_waiter(t));
+                    }
+                    ThreadState::Blocked(BlockReason::AwaitingReply) => {
+                        assert!(!in_any_queue, "{t:?} awaiting reply but still queued");
+                        let me = crate::syscall::ThreadRef {
+                            tid: t,
+                            generation: k.threads.generation(t).unwrap(),
+                        };
+                        let holders = self
+                            .procs
+                            .iter()
+                            .flat_map(|p| p.threads)
+                            .filter(|&s| k.thread_state_for_test(s).is_some())
+                            .filter(|s| k.ipc[s.raw() as usize].reply_to == Some(me))
+                            .count();
+                        assert_eq!(holders, 1, "{t:?} awaits a reply nobody can send");
+                    }
+                    ThreadState::Ready | ThreadState::Running(_) | ThreadState::Dead => {
+                        assert!(!in_any_queue, "{t:?} runnable but queued on an IPC object");
+                        assert!(ipc.outgoing.is_none());
+                    }
+                }
+                let blocked = matches!(state, ThreadState::Blocked(_));
+                assert_eq!(
+                    k.scheduler.is_queued(t),
+                    matches!(state, ThreadState::Ready),
+                    "{t:?} in state {state:?} has wrong run-queue membership"
+                );
+                if blocked {
+                    assert_ne!(k.current_thread(), Some(t));
+                }
+
+                // A live reply slot always points at a thread waiting for it.
+                if let Some(r) = ipc.reply_to
+                    && k.threads.generation(r.tid) == Some(r.generation)
+                {
+                    assert_eq!(
+                        k.thread_state_for_test(r.tid),
+                        Some(ThreadState::Blocked(BlockReason::AwaitingReply)),
+                        "{t:?} holds a reply for {:?} which is not waiting",
+                        r.tid
+                    );
+                }
+            }
+        }
+        for &e in &self.eps {
+            let ep = k.endpoints.get(e).unwrap();
+            assert!(!(ep.has_waiting_sender() && ep.has_waiting_receiver()));
+        }
+    }
+}
+
+#[test]
+fn ipc_state_machine_holds_under_random_operations() {
+    for seed in [3u64, 0xC0FFEE, 0xFEED_FACE, 99, 123_456_789] {
+        let mut world = World::new();
+        let mut rng = Rng(seed);
+        for _ in 0..STEPS {
+            world.step(&mut rng);
+            world.check();
+        }
+        assert!(
+            world.calls_answered > 20,
+            "seed {seed:#x}: too few replies: answered={} signals={} delivered={} ran={}",
+            world.calls_answered,
+            world.signals_seen,
+            world.delivered.iter().filter(|d| **d).count(),
+            world.ran
+        );
+        assert!(
+            world.rebirths > 5,
+            "seed {seed:#x}: kill/re-create barely exercised"
+        );
+        assert!(world.signals_seen > 20, "seed {seed:#x}: too few signals");
+        assert!(world.delivered.iter().filter(|d| **d).count() > 200);
+    }
+}

crates/kernel/src/tests/lifetimes.rs

@@ -6,10 +6,10 @@ use crate::{SyscallOutcome, SyscallResult};
 use domain::message::Message;
 use domain::resource::Resource;
 use domain::syscall::Syscall;
-use domain::thread::{BlockReason, ThreadState};
+use domain::thread::ThreadState;
 use types::address::DataRef;
 use types::bounded::CapSlot;
-use types::id::{IrqId, ThreadId};
+use types::id::ThreadId;
 use types::permission::{Permission, PermissionSet};
 
 fn create_process(k: &mut Kernel<MockPlatform>) -> CapSlot {
@@ -177,57 +177,6 @@ fn endpoint_slots_are_reclaimed_when_processes_die() {
     }
 }
 
-#[test]
-fn interrupt_routing_does_not_survive_the_owning_thread() {
-    let (mut k, pid_a) = kernel_with_process();
-    let tid_a = k.current_thread().unwrap();
-    let cap_b = create_process(&mut k);
-    let pid_b = ProcessId::from_raw(1);
-    let tid_b = k.spawn_thread_for_test(pid_b, Priority::DEFAULT, TimeSlice::DEFAULT);
-    let irq = IrqId::from_raw(10);
-    k.register_irq_for_test(irq, tid_b);
-
-    k.switch_to_thread_for_test(tid_a);
-    k.handle_syscall(Syscall::Kill { process: cap_b });
-
-    // A new thread reuses B's thread id and happens to be blocked.
-    let reused = k.spawn_thread_for_test(pid_a, Priority::DEFAULT, TimeSlice::DEFAULT);
-    assert_eq!(reused, tid_b);
-    k.block_thread_for_test(reused, BlockReason::Receiving(CapSlot::new(1).unwrap()));
-
-    assert_eq!(
-        k.handle_interrupt(irq),
-        None,
-        "IRQ delivered to an unrelated thread"
-    );
-}
-
-#[test]
-fn interrupt_numbers_beyond_the_table_are_ignored() {
-    let (mut k, _) = kernel_with_process();
-    // The PLIC can report sources up to 1023.
-    assert_eq!(k.handle_interrupt(IrqId::from_raw(1000)), None);
-}
-
-#[test]
-fn exiting_a_thread_releases_its_interrupt_route() {
-    let (mut k, pid_a) = kernel_with_process();
-    let tid_a = k.current_thread().unwrap();
-    let irq = IrqId::from_raw(10);
-    k.register_irq_for_test(irq, tid_a);
-    k.spawn_thread_for_test(pid_a, Priority::DEFAULT, TimeSlice::DEFAULT);
-
-    assert!(matches!(
-        k.handle_syscall(Syscall::ExitThread),
-        SyscallOutcome::TerminateCaller
-    ));
-
-    let reused = k.spawn_thread_for_test(pid_a, Priority::DEFAULT, TimeSlice::DEFAULT);
-    assert_eq!(reused, tid_a);
-    k.block_thread_for_test(reused, BlockReason::Receiving(CapSlot::new(1).unwrap()));
-    assert_eq!(k.handle_interrupt(irq), None);
-}
-
 #[test]
 fn drop_cap_frees_the_slot() {
     let (mut k, pid_a) = kernel_with_process();
@@ -253,3 +202,43 @@ fn killing_yourself_terminates_the_caller() {
     assert!(matches!(outcome, SyscallOutcome::TerminateCaller));
     assert!(!k.process_exists(pid_a));
 }
+
+#[test]
+fn exiting_a_thread_unbinds_its_notification() {
+    let (mut k, pid_a) = kernel_with_process();
+    let tid_a = k.current_thread().unwrap();
+    let n = k.create_notification(pid_a).unwrap();
+    let rw = PermissionSet::empty()
+        .with(Permission::Read)
+        .with(Permission::Write);
+    let ncap = k.grant_cap_for_test(pid_a, Resource::Notification(n), rw);
+    assert!(!is_error(&k.handle_syscall(Syscall::BindNotification {
+        notification: ncap
+    })));
+    let other = k.spawn_thread_for_test(pid_a, Priority::DEFAULT, TimeSlice::DEFAULT);
+    assert!(matches!(
+        k.handle_syscall(Syscall::ExitThread),
+        SyscallOutcome::TerminateCaller
+    ));
+
+    // A new thread reuses the id and blocks in an unrelated receive.
+    let reused = k.spawn_thread_for_test(pid_a, Priority::DEFAULT, TimeSlice::DEFAULT);
+    assert_eq!(reused, tid_a);
+    let ep = k.create_endpoint_for_test();
+    let recv = k.grant_cap_for_test(pid_a, Resource::Endpoint(ep), ipc_perms());
+    k.switch_to_thread_for_test(reused);
+    assert!(matches!(
+        k.handle_syscall(Syscall::Receive { endpoint: recv }),
+        SyscallOutcome::BlockCaller(_)
+    ));
+
+    k.switch_to_thread_for_test(other);
+    k.handle_syscall(Syscall::Signal { notification: ncap });
+    assert!(
+        matches!(
+            k.thread_state_for_test(reused),
+            Some(ThreadState::Blocked(_))
+        ),
+        "signal reached a thread that never bound the notification"
+    );
+}

crates/kernel/src/tests/mod.rs

@@ -5,6 +5,7 @@ mod capability;
 mod exit;
 mod invariants;
 mod ipc;
+mod ipc_invariants;
 mod isolation;
 mod lifecycle;
 mod lifetimes;

crates/kernel/src/tests/tick.rs

@@ -1,6 +1,6 @@
 use super::*;
 use domain::scheduler::SchedulerDecision;
-use types::id::{CoreId, IrqId, ThreadId};
+use types::id::CoreId;
 use types::time::TimeSlice;
 
 #[test]
@@ -30,26 +30,3 @@ fn tick_preempts_when_time_expired_and_other_thread_ready() {
     let d2 = k.handle_tick(CoreId::from_raw(0));
     assert!(matches!(d2, SchedulerDecision::Preempt(_)));
 }
-
-#[test]
-fn interrupt_unknown_irq_returns_none() {
-    let (mut k, _) = kernel_with_process();
-    assert!(k.handle_interrupt(IrqId::from_raw(99)).is_none());
-}
-
-#[test]
-fn interrupt_unblocks_registered_thread() {
-    let (mut k, _) = kernel_with_process();
-    let tid = ThreadId::from_raw(0);
-    let irq = IrqId::from_raw(5);
-
-    // Register and block the thread on this IRQ
-    k.register_irq_for_test(irq, tid);
-    k.block_thread_for_test(
-        tid,
-        domain::thread::BlockReason::WaitingForInterrupt(types::bounded::CapSlot::new(0).unwrap()),
-    );
-
-    let result = k.handle_interrupt(irq);
-    assert_eq!(result, Some(tid));
-}

crates/libkos/src/alloc_impl.rs

@@ -80,13 +80,17 @@ impl AllocatorInner {
                     if prev.is_null() {
                         self.free_list = new_block;
                     } else {
-                        unsafe { (*prev).next = new_block; }
+                        unsafe {
+                            (*prev).next = new_block;
+                        }
                     }
                 } else {
                     if prev.is_null() {
                         self.free_list = block.next;
                     } else {
-                        unsafe { (*prev).next = block.next; }
+                        unsafe {
+                            (*prev).next = block.next;
+                        }
                     }
                 }
 

crates/libkos/src/arch/mock.rs

@@ -1,27 +1,4 @@
-pub fn syscall0(_nr: usize) -> usize {
+/// Host builds have no kernel; every call reports success with zeroes.
-    0
+pub fn syscall(_nr: usize, _args: [usize; 6]) -> [usize; 8] {
-}
+    [0; 8]
-
-pub fn syscall1(_nr: usize, _a0: usize) -> usize {
-    0
-}
-
-pub fn syscall2(_nr: usize, _a0: usize, _a1: usize) -> usize {
-    0
-}
-
-pub fn syscall3(_nr: usize, _a0: usize, _a1: usize, _a2: usize) -> usize {
-    0
-}
-
-pub fn syscall4(_nr: usize, _a0: usize, _a1: usize, _a2: usize, _a3: usize) -> usize {
-    0
-}
-
-pub fn syscall5(_nr: usize, _a0: usize, _a1: usize, _a2: usize, _a3: usize, _a4: usize) -> usize {
-    0
-}
-
-pub fn syscall_receive(_nr: usize, _a0: usize) -> (usize, usize, usize, usize) {
-    (0, 0, 0, 0)
 }

crates/libkos/src/arch/riscv.rs

@@ -1,106 +1,20 @@
+/// One `ecall`: syscall number in `a7`, arguments in `a0..a5`. Returns the
+/// eight result registers `a0..a7` (docs/ipc.md, register ABI).
 #[inline(always)]
-pub fn syscall0(nr: usize) -> usize {
+pub fn syscall(nr: usize, args: [usize; 6]) -> [usize; 8] {
-    let ret: usize;
+    let mut r = [0usize; 8];
     unsafe {
         core::arch::asm!(
             "ecall",
-            in("a7") nr,
+            inlateout("a0") args[0] => r[0],
-            lateout("a0") ret,
+            inlateout("a1") args[1] => r[1],
+            inlateout("a2") args[2] => r[2],
+            inlateout("a3") args[3] => r[3],
+            inlateout("a4") args[4] => r[4],
+            inlateout("a5") args[5] => r[5],
+            lateout("a6") r[6],
+            inlateout("a7") nr => r[7],
         );
     }
-    ret
+    r
-}
-
-#[inline(always)]
-pub fn syscall1(nr: usize, a0: usize) -> usize {
-    let ret: usize;
-    unsafe {
-        core::arch::asm!(
-            "ecall",
-            in("a7") nr,
-            inlateout("a0") a0 => ret,
-        );
-    }
-    ret
-}
-
-#[inline(always)]
-pub fn syscall2(nr: usize, a0: usize, a1: usize) -> usize {
-    let ret: usize;
-    unsafe {
-        core::arch::asm!(
-            "ecall",
-            in("a7") nr,
-            inlateout("a0") a0 => ret,
-            in("a1") a1,
-        );
-    }
-    ret
-}
-
-#[inline(always)]
-pub fn syscall3(nr: usize, a0: usize, a1: usize, a2: usize) -> usize {
-    let ret: usize;
-    unsafe {
-        core::arch::asm!(
-            "ecall",
-            in("a7") nr,
-            inlateout("a0") a0 => ret,
-            in("a1") a1,
-            in("a2") a2,
-        );
-    }
-    ret
-}
-
-#[inline(always)]
-pub fn syscall4(nr: usize, a0: usize, a1: usize, a2: usize, a3: usize) -> usize {
-    let ret: usize;
-    unsafe {
-        core::arch::asm!(
-            "ecall",
-            in("a7") nr,
-            inlateout("a0") a0 => ret,
-            in("a1") a1,
-            in("a2") a2,
-            in("a3") a3,
-        );
-    }
-    ret
-}
-
-#[inline(always)]
-pub fn syscall5(nr: usize, a0: usize, a1: usize, a2: usize, a3: usize, a4: usize) -> usize {
-    let ret: usize;
-    unsafe {
-        core::arch::asm!(
-            "ecall",
-            in("a7") nr,
-            inlateout("a0") a0 => ret,
-            in("a1") a1,
-            in("a2") a2,
-            in("a3") a3,
-            in("a4") a4,
-        );
-    }
-    ret
-}
-
-#[inline(always)]
-pub fn syscall_receive(nr: usize, a0: usize) -> (usize, usize, usize, usize) {
-    let r0: usize;
-    let r1: usize;
-    let r2: usize;
-    let r3: usize;
-    unsafe {
-        core::arch::asm!(
-            "ecall",
-            in("a7") nr,
-            inlateout("a0") a0 => r0,
-            lateout("a1") r1,
-            lateout("a2") r2,
-            lateout("a3") r3,
-        );
-    }
-    (r0, r1, r2, r3)
 }

crates/libkos/src/display.rs

@@ -1,13 +1,20 @@
+use crate::syscall;
 use types::bounded::CapSlot;
 use types::display::{Color, Dimension, ScreenCoord};
-use crate::syscall;
 
 fn ipc(ep: CapSlot, w0: usize, w1: usize, w2: usize, w3: usize) -> (usize, usize, usize, usize) {
     syscall::send(ep, w0, w1, w2, w3);
     syscall::receive(ep)
 }
 
-pub fn draw_rect(ep: CapSlot, x: ScreenCoord, y: ScreenCoord, w: Dimension, h: Dimension, color: Color) {
+pub fn draw_rect(
+    ep: CapSlot,
+    x: ScreenCoord,
+    y: ScreenCoord,
+    w: Dimension,
+    h: Dimension,
+    color: Color,
+) {
     let xy = (x.raw() as usize) | ((y.raw() as usize) << 16);
     let wh = (w.raw() as usize) | ((h.raw() as usize) << 16);
     ipc(ep, 0, xy, wh, color.raw() as usize);

crates/libkos/src/fs.rs

@@ -1,19 +1,24 @@
-use types::bounded::CapSlot;
 use crate::syscall;
+use types::bounded::CapSlot;
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub struct FileHandle(u16);
 
 impl FileHandle {
-    pub const fn raw(&self) -> u16 { self.0 }
+    pub const fn raw(&self) -> u16 {
+        self.0
+    }
 }
 
 fn pack_path(path: &[u8]) -> (usize, usize) {
     let mut w0 = [0u8; 8];
     let mut w1 = [0u8; 8];
     for (i, &b) in path.iter().enumerate() {
-        if i < 8 { w0[i] = b; }
+        if i < 8 {
-        else if i < 16 { w1[i - 8] = b; }
+            w0[i] = b;
+        } else if i < 16 {
+            w1[i - 8] = b;
+        }
     }
     (usize::from_le_bytes(w0), usize::from_le_bytes(w1))
 }
@@ -26,7 +31,11 @@ fn ipc(ep: CapSlot, w0: usize, w1: usize, w2: usize, w3: usize) -> (usize, usize
 pub fn open(ep: CapSlot, path: &[u8]) -> Option<FileHandle> {
     let (p0, p1) = pack_path(path);
     let (tag, handle, _, _) = ipc(ep, 0, p0, p1, 0);
-    if tag == 0 { Some(FileHandle(handle as u16)) } else { None }
+    if tag == 0 {
+        Some(FileHandle(handle as u16))
+    } else {
+        None
+    }
 }
 
 pub fn read(ep: CapSlot, handle: FileHandle, max_bytes: usize) -> Option<usize> {
@@ -52,7 +61,11 @@ pub fn list_dir(ep: CapSlot, path: &[u8]) -> Option<usize> {
 pub fn create(ep: CapSlot, path: &[u8]) -> Option<FileHandle> {
     let (p0, p1) = pack_path(path);
     let (tag, handle, _, _) = ipc(ep, 5, p0, p1, 0);
-    if tag == 0 { Some(FileHandle(handle as u16)) } else { None }
+    if tag == 0 {
+        Some(FileHandle(handle as u16))
+    } else {
+        None
+    }
 }
 
 pub fn delete(ep: CapSlot, path: &[u8]) -> bool {

crates/libkos/src/input.rs

@@ -1,5 +1,5 @@
-use types::bounded::CapSlot;
 use crate::syscall;
+use types::bounded::CapSlot;
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum Key {

crates/libkos/src/lib.rs

@@ -2,14 +2,14 @@
 
 extern crate alloc;
 
-mod arch;
-pub mod syscall;
 pub mod alloc_impl;
-pub mod io;
+mod arch;
-pub mod fs;
 pub mod display;
+pub mod fs;
 pub mod input;
+pub mod io;
 mod rt;
+pub mod syscall;
 
 #[cfg(target_os = "none")]
 #[global_allocator]

crates/libkos/src/syscall.rs

@@ -1,90 +1,335 @@
+//! Thin wrappers over the kernel's register ABI (docs/ipc.md). Every call
+//! returns a status in `a0`; results follow in `a1..`.
+
 use crate::arch;
-use types::syscall_abi::*;
 use types::address::{EntryPoint, StackPointer, VirtualAddress};
+use types::badge::Badge;
 use types::bounded::{CapSlot, PageCount};
 use types::id::ProcessId;
 use types::permission::PermissionSet;
+use types::syscall_abi::*;
 use types::time::Ticks;
 
+pub use types::syscall_abi::ErrorCode as Error;
+
+pub type Result<T> = core::result::Result<T, Error>;
+
+fn raw(nr: usize, args: [usize; 6]) -> Result<[usize; 8]> {
+    let r = arch::syscall(nr, args);
+    match r[0] {
+        0 => Ok(r),
+        code => Err(Error::from_raw(code).unwrap_or(Error::InvalidArgument)),
+    }
+}
+
+fn call0(nr: usize) -> Result<[usize; 8]> {
+    raw(nr, [0; 6])
+}
+
+fn slot_arg(slot: Option<CapSlot>) -> usize {
+    slot.map_or(NONE, |s| s.value() as usize)
+}
+
+fn slot_result(r: usize) -> Option<CapSlot> {
+    if r == NONE {
+        None
+    } else {
+        CapSlot::new(r as u8)
+    }
+}
+
+fn slot_of(r: Result<[usize; 8]>) -> Result<CapSlot> {
+    r.and_then(|r| slot_result(r[1]).ok_or(Error::InvalidArgument))
+}
+
+/// What a receiving call got.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum Received {
+    /// A `Send`. `badge` identifies the sender's capability.
+    Message {
+        badge: usize,
+        words: [usize; 4],
+        cap: Option<CapSlot>,
+    },
+    /// A `Call`; answer it with `reply` or `reply_recv`.
+    Call {
+        badge: usize,
+        words: [usize; 4],
+        cap: Option<CapSlot>,
+    },
+    /// Bits from a notification.
+    Signal { bits: usize },
+}
+
+impl Received {
+    fn decode(r: [usize; 8]) -> Self {
+        let words = [r[2], r[3], r[4], r[5]];
+        let cap = slot_result(r[6]);
+        match r[7] {
+            KIND_CALL => Received::Call {
+                badge: r[1],
+                words,
+                cap,
+            },
+            KIND_SIGNAL => Received::Signal { bits: r[1] },
+            _ => Received::Message {
+                badge: r[1],
+                words,
+                cap,
+            },
+        }
+    }
+
+    pub fn words(&self) -> [usize; 4] {
+        match self {
+            Received::Message { words, .. } | Received::Call { words, .. } => *words,
+            Received::Signal { .. } => [0; 4],
+        }
+    }
+}
+
+fn message_args(ep: usize, words: [usize; 4], cap: Option<CapSlot>) -> [usize; 6] {
+    [ep, words[0], words[1], words[2], words[3], slot_arg(cap)]
+}
+
+// ---- threads, processes, time ----
+
 pub fn exit() -> ! {
-    arch::syscall0(SYS_EXIT);
+    let _ = call0(SYS_EXIT);
     loop {}
 }
 
 pub fn exit_thread() -> ! {
-    arch::syscall0(SYS_EXIT_THREAD);
+    let _ = call0(SYS_EXIT_THREAD);
     loop {}
 }
 
 pub fn yield_now() {
-    arch::syscall0(SYS_YIELD);
+    let _ = call0(SYS_YIELD);
 }
 
 pub fn halt() -> ! {
-    arch::syscall0(SYS_HALT);
+    let _ = call0(SYS_HALT);
     loop {}
 }
 
 pub fn get_time() -> Ticks {
-    Ticks::from_raw(arch::syscall0(SYS_GET_TIME) as u64)
+    Ticks::from_raw(call0(SYS_GET_TIME).map_or(0, |r| r[1]) as u64)
 }
 
 pub fn get_process_id() -> ProcessId {
-    ProcessId::from_raw(arch::syscall0(SYS_GET_PROCESS_ID) as u16)
+    ProcessId::from_raw(call0(SYS_GET_PROCESS_ID).map_or(0, |r| r[1]) as u16)
 }
 
 pub fn create_process(create_cap: CapSlot) -> Option<CapSlot> {
-    let r = arch::syscall1(SYS_CREATE_PROCESS, create_cap.value() as usize);
+    slot_of(raw(
-    if r == usize::MAX { None } else { CapSlot::new(r as u8) }
+        SYS_CREATE_PROCESS,
-}
+        [create_cap.value() as usize, 0, 0, 0, 0, 0],
-
+    ))
-pub fn create_endpoint() -> Option<CapSlot> {
+    .ok()
-    let r = arch::syscall0(SYS_CREATE_ENDPOINT);
-    if r == usize::MAX { None } else { CapSlot::new(r as u8) }
 }
 
 pub fn spawn_thread(entry: EntryPoint, stack: StackPointer) -> usize {
-    arch::syscall2(SYS_SPAWN_THREAD, entry.raw(), stack.raw())
+    raw(SYS_SPAWN_THREAD, [entry.raw(), stack.raw(), 0, 0, 0, 0]).map_or(NONE, |r| r[1])
 }
 
 pub fn spawn_thread_in(process_cap: CapSlot, entry: EntryPoint, stack: StackPointer) -> usize {
-    arch::syscall3(SYS_SPAWN_THREAD_IN, process_cap.value() as usize, entry.raw(), stack.raw())
+    raw(
-}
+        SYS_SPAWN_THREAD_IN,
-
+        [
-pub fn request_pages(count: PageCount) -> Option<VirtualAddress> {
+            process_cap.value() as usize,
-    let r = arch::syscall1(SYS_REQUEST_PAGES, count.value() as usize);
+            entry.raw(),
-    if r == usize::MAX { None } else { Some(VirtualAddress::from_raw(r)) }
+            stack.raw(),
-}
+            0,
-
+            0,
-pub fn map_page(process_cap: CapSlot, src_va: VirtualAddress, dst_va: VirtualAddress, executable: bool) -> bool {
+            0,
-    arch::syscall4(SYS_MAP_PAGE, process_cap.value() as usize, src_va.raw(), dst_va.raw(), executable as usize) == 0
+        ],
-}
+    )
-
+    .map_or(NONE, |r| r[1])
-pub fn send(endpoint: CapSlot, w0: usize, w1: usize, w2: usize, w3: usize) -> usize {
-    arch::syscall5(SYS_SEND, endpoint.value() as usize, w0, w1, w2, w3)
-}
-
-pub fn receive(endpoint: CapSlot) -> (usize, usize, usize, usize) {
-    arch::syscall_receive(SYS_RECEIVE, endpoint.value() as usize)
-}
-
-pub fn delegate(slot: CapSlot, to: CapSlot, permissions: PermissionSet) -> bool {
-    arch::syscall3(SYS_DELEGATE, slot.value() as usize, to.value() as usize, permissions.bits() as usize) == 0
 }
 
 pub fn kill(process_cap: CapSlot) -> bool {
-    arch::syscall1(SYS_KILL, process_cap.value() as usize) == 0
+    raw(SYS_KILL, [process_cap.value() as usize, 0, 0, 0, 0, 0]).is_ok()
+}
+
+// ---- memory ----
+
+pub fn request_pages(count: PageCount) -> Option<VirtualAddress> {
+    raw(SYS_REQUEST_PAGES, [count.value() as usize, 0, 0, 0, 0, 0])
+        .ok()
+        .map(|r| VirtualAddress::from_raw(r[1]))
+}
+
+pub fn map_page(
+    process_cap: CapSlot,
+    src_va: VirtualAddress,
+    dst_va: VirtualAddress,
+    executable: bool,
+) -> bool {
+    raw(
+        SYS_MAP_PAGE,
+        [
+            process_cap.value() as usize,
+            src_va.raw(),
+            dst_va.raw(),
+            executable as usize,
+            0,
+            0,
+        ],
+    )
+    .is_ok()
+}
+
+// ---- capabilities ----
+
+pub fn delegate(slot: CapSlot, to: CapSlot, permissions: PermissionSet) -> bool {
+    raw(
+        SYS_DELEGATE,
+        [
+            slot.value() as usize,
+            to.value() as usize,
+            permissions.bits() as usize,
+            0,
+            0,
+            0,
+        ],
+    )
+    .is_ok()
+}
+
+/// Copies an unbadged endpoint or notification capability with `badge`
+/// and at most `permissions`.
+pub fn mint(slot: CapSlot, badge: Badge, permissions: PermissionSet) -> Result<CapSlot> {
+    slot_of(raw(
+        SYS_MINT,
+        [
+            slot.value() as usize,
+            badge.raw(),
+            permissions.bits() as usize,
+            0,
+            0,
+            0,
+        ],
+    ))
 }
 
 pub fn revoke(slot: CapSlot) -> bool {
-    arch::syscall1(SYS_REVOKE, slot.value() as usize) == 0
+    raw(SYS_REVOKE, [slot.value() as usize, 0, 0, 0, 0, 0]).is_ok()
 }
 
 /// Removes a capability from this process's own table, freeing the slot.
 pub fn drop_cap(slot: CapSlot) -> bool {
-    arch::syscall1(SYS_DROP_CAP, slot.value() as usize) == 0
+    raw(SYS_DROP_CAP, [slot.value() as usize, 0, 0, 0, 0, 0]).is_ok()
+}
+
+// ---- endpoints ----
+
+pub fn create_endpoint() -> Option<CapSlot> {
+    slot_of(call0(SYS_CREATE_ENDPOINT)).ok()
+}
+
+/// Sends four words; returns 0 on success (kept for existing callers).
+pub fn send(endpoint: CapSlot, w0: usize, w1: usize, w2: usize, w3: usize) -> usize {
+    match send_msg(endpoint, [w0, w1, w2, w3], None) {
+        Ok(()) => 0,
+        Err(e) => e as usize,
+    }
+}
+
+pub fn send_msg(endpoint: CapSlot, words: [usize; 4], cap: Option<CapSlot>) -> Result<()> {
+    raw(
+        SYS_SEND,
+        message_args(endpoint.value() as usize, words, cap),
+    )
+    .map(|_| ())
+}
+
+/// Receives and returns just the words (kept for existing callers).
+pub fn receive(endpoint: CapSlot) -> (usize, usize, usize, usize) {
+    let w = receive_msg(endpoint).map_or([0; 4], |r| r.words());
+    (w[0], w[1], w[2], w[3])
+}
+
+pub fn receive_msg(endpoint: CapSlot) -> Result<Received> {
+    raw(SYS_RECEIVE, [endpoint.value() as usize, 0, 0, 0, 0, 0]).map(Received::decode)
+}
+
+/// Sends and waits for the reply.
+pub fn call(endpoint: CapSlot, words: [usize; 4], cap: Option<CapSlot>) -> Result<Received> {
+    raw(
+        SYS_CALL,
+        message_args(endpoint.value() as usize, words, cap),
+    )
+    .map(Received::decode)
+}
+
+/// Answers the last `Call` this thread received.
+pub fn reply(words: [usize; 4], cap: Option<CapSlot>) -> Result<()> {
+    raw(SYS_REPLY, message_args(0, words, cap)).map(|_| ())
+}
+
+/// Replies to the pending `Call` (if any), then receives the next message.
+pub fn reply_recv(endpoint: CapSlot, words: [usize; 4], cap: Option<CapSlot>) -> Result<Received> {
+    raw(
+        SYS_REPLY_RECV,
+        message_args(endpoint.value() as usize, words, cap),
+    )
+    .map(Received::decode)
+}
+
+// ---- notifications and interrupts ----
+
+pub fn create_notification() -> Result<CapSlot> {
+    slot_of(call0(SYS_CREATE_NOTIFICATION))
+}
+
+pub fn signal(notification: CapSlot) -> Result<()> {
+    raw(SYS_SIGNAL, [notification.value() as usize, 0, 0, 0, 0, 0]).map(|_| ())
+}
+
+/// Blocks until the notification has bits, then returns and clears them.
+pub fn wait(notification: CapSlot) -> Result<usize> {
+    raw(SYS_WAIT, [notification.value() as usize, 0, 0, 0, 0, 0]).map(|r| r[1])
+}
+
+/// Returns and clears the bits without blocking.
+pub fn poll(notification: CapSlot) -> Result<usize> {
+    raw(SYS_POLL, [notification.value() as usize, 0, 0, 0, 0, 0]).map(|r| r[1])
+}
+
+/// After this, `receive` also returns when the notification is signalled.
+pub fn bind_notification(notification: CapSlot) -> Result<()> {
+    raw(
+        SYS_BIND_NOTIFICATION,
+        [notification.value() as usize, 0, 0, 0, 0, 0],
+    )
+    .map(|_| ())
+}
+
+/// Delivers interrupt `irq` as a signal on `notification`.
+pub fn set_irq_notification(irq: CapSlot, notification: CapSlot) -> Result<()> {
+    raw(
+        SYS_SET_IRQ_NOTIFICATION,
+        [
+            irq.value() as usize,
+            notification.value() as usize,
+            0,
+            0,
+            0,
+            0,
+        ],
+    )
+    .map(|_| ())
+}
+
+pub fn ack_interrupt(irq: CapSlot) -> Result<()> {
+    raw(
+        SYS_ACKNOWLEDGE_INTERRUPT,
+        [irq.value() as usize, 0, 0, 0, 0, 0],
+    )
+    .map(|_| ())
 }
 
 pub fn debug_putchar(byte: u8) {
-    arch::syscall1(SYS_DEBUG_PUTCHAR, byte as usize);
+    let _ = raw(SYS_DEBUG_PUTCHAR, [byte as usize, 0, 0, 0, 0, 0]);
 }

crates/types/src/badge.rs

@@ -0,0 +1,21 @@
+/// A word stamped on an endpoint or notification capability by `Mint`.
+/// Receivers see the badge of the capability a sender used, which is how a
+/// server tells its clients apart. Zero means "unbadged".
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub struct Badge(usize);
+
+impl Badge {
+    pub const NONE: Self = Self(0);
+
+    pub const fn new(value: usize) -> Self {
+        Self(value)
+    }
+
+    pub const fn raw(&self) -> usize {
+        self.0
+    }
+
+    pub const fn is_none(&self) -> bool {
+        self.0 == 0
+    }
+}

crates/types/src/id.rs

@@ -4,5 +4,6 @@ kernel_id!(ProcessId, u16);
 kernel_id!(ThreadId, u16);
 kernel_id!(EndpointId, u16);
 kernel_id!(RegionId, u16);
+kernel_id!(NotificationId, u16);
 kernel_id!(IrqId, u16);
 kernel_id!(CoreId, u8);

crates/types/src/lib.rs

@@ -1,6 +1,7 @@
 #![no_std]
 
 pub mod address;
+pub mod badge;
 pub mod bounded;
 pub mod display;
 pub mod generation;

crates/types/src/syscall_abi.rs

@@ -1,9 +1,12 @@
+//! Register-level syscall contract shared by the kernel and libkos.
+//! See docs/ipc.md for the meaning of each call.
+
 pub const SYS_SEND: usize = 0;
 pub const SYS_RECEIVE: usize = 1;
 pub const SYS_DELEGATE: usize = 2;
-pub const SYS_WAIT_FOR_INTERRUPT: usize = 3;
+// 3 was WaitForInterrupt; interrupts are delivered through notifications.
 pub const SYS_ACKNOWLEDGE_INTERRUPT: usize = 4;
-pub const SYS_WAIT_ANY: usize = 5;
+// 5 was WaitAny; superseded by notifications.
 pub const SYS_CREATE_PROCESS: usize = 6;
 pub const SYS_KILL: usize = 7;
 pub const SYS_REVOKE: usize = 8;
@@ -21,3 +24,63 @@ pub const SYS_DEBUG_PUTCHAR: usize = 19;
 pub const SYS_MAP_PAGE: usize = 20;
 pub const SYS_SPAWN_THREAD_IN: usize = 21;
 pub const SYS_DROP_CAP: usize = 22;
+pub const SYS_CALL: usize = 23;
+pub const SYS_REPLY: usize = 24;
+pub const SYS_REPLY_RECV: usize = 25;
+pub const SYS_MINT: usize = 26;
+pub const SYS_CREATE_NOTIFICATION: usize = 27;
+pub const SYS_SIGNAL: usize = 28;
+pub const SYS_WAIT: usize = 29;
+pub const SYS_POLL: usize = 30;
+pub const SYS_BIND_NOTIFICATION: usize = 31;
+pub const SYS_SET_IRQ_NOTIFICATION: usize = 32;
+
+/// "No capability" in a capability argument or result register.
+pub const NONE: usize = usize::MAX;
+
+/// What a receiving syscall got, returned in `a7`.
+pub const KIND_MESSAGE: usize = 0;
+pub const KIND_CALL: usize = 1;
+pub const KIND_SIGNAL: usize = 2;
+
+/// Status returned in `a0`; `0` is success.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+#[repr(usize)]
+pub enum ErrorCode {
+    /// No capability in that slot, or it names a dead or revoked object.
+    InvalidCap = 1,
+    /// The capability lacks the permission the operation needs.
+    PermissionDenied = 2,
+    /// The capability names the wrong kind of object.
+    WrongObject = 3,
+    NoFreeSlot = 4,
+    OutOfMemory = 5,
+    InvalidAddress = 6,
+    AddressInUse = 7,
+    InvalidArgument = 8,
+    /// `Reply` with nothing to reply to, or a call whose reply was dropped.
+    NoReply = 9,
+    /// A table (threads, processes, queue) is full.
+    LimitReached = 10,
+    UnknownSyscall = 11,
+}
+
+impl ErrorCode {
+    pub fn from_raw(code: usize) -> Option<Self> {
+        use ErrorCode::*;
+        Some(match code {
+            1 => InvalidCap,
+            2 => PermissionDenied,
+            3 => WrongObject,
+            4 => NoFreeSlot,
+            5 => OutOfMemory,
+            6 => InvalidAddress,
+            7 => AddressInUse,
+            8 => InvalidArgument,
+            9 => NoReply,
+            10 => LimitReached,
+            11 => UnknownSyscall,
+            _ => return None,
+        })
+    }
+}

docs/ipc.md

@@ -0,0 +1,105 @@
+# IPC
+
+Synchronous message passing modelled on classic (non-MCS) seL4: endpoints
+for rendezvous messages, notifications for asynchronous signals, an
+implicit per-thread reply slot for `Call`. This document is the contract the
+kernel tests and proofs check.
+
+## Objects
+
+**Endpoint.** A rendezvous point. Holds a FIFO of blocked senders and a
+FIFO of blocked receivers; never both non-empty. Owned by its creator and
+destroyed with it.
+
+**Notification.** A word of signal bits plus a FIFO of blocked waiters and
+at most one bound thread. Owned by its creator and destroyed with it.
+
+Both are named by capabilities whose generation must match (see P2).
+
+## Capabilities
+
+A capability to an endpoint or notification carries a **badge** (a word;
+`0` means unbadged). `Mint` copies an unbadged capability with a badge and
+narrower permissions. A badge cannot be changed once set.
+
+| Object       | Permission | Allows                                     |
+|--------------|------------|--------------------------------------------|
+| Endpoint     | Execute    | `Send`, `Call`                             |
+| Endpoint     | Read       | `Receive`, `ReplyRecv`                     |
+| Endpoint     | Delegate   | transfer a capability through it; `Mint`   |
+| Notification | Write      | `Signal`                                   |
+| Notification | Read       | `Wait`, `Poll`, `BindNotification`         |
+| Notification | Delegate   | `Mint`                                     |
+| Interrupt    | Execute    | `SetIrqNotification`, `AckInterrupt`       |
+
+## Messages
+
+A message is 4 words and at most one capability. The capability is
+transferred only if the sender's endpoint capability has `Delegate` and the
+transferred capability itself has `Delegate`; it is **copied** (the sender
+keeps its own) with its permissions and badge unchanged. If the receiver's
+table is full, or the capability is no longer valid at delivery time, the
+message is delivered without it.
+
+The receiver learns the badge of the capability the sender used. The kernel
+never reveals the sender's process or thread id.
+
+## Operations
+
+- `Send(ep, msg)`: if a receiver waits, deliver and continue; otherwise block
+  until one arrives.
+- `Receive(ep)`: if the caller has a bound notification with bits set,
+  return those bits (and clear them). Else if a sender waits, take its
+  message. Else block until a sender arrives or the bound notification is
+  signalled.
+- `Call(ep, msg)`: like `Send`, but the caller then blocks awaiting a reply.
+  The receiving thread's reply slot now names the caller.
+- `Reply(msg)`: deliver `msg` to the thread in the caller's reply slot and
+  clear the slot. Error if the slot is empty. Never blocks.
+- `ReplyRecv(ep, msg)`: `Reply` if the slot is full, then `Receive(ep)`.
+- `Signal(ntfn)`: let `b` be the capability's badge, or `1` if unbadged. If
+  a waiter is blocked, wake the first with `b`. Else if the bound thread is
+  blocked in `Receive`, wake it with `b`. Else `bits |= b`.
+- `Wait(ntfn)`: if `bits != 0` return and clear them; else block.
+- `Poll(ntfn)`: return and clear `bits` (possibly 0); never blocks.
+- `BindNotification(ntfn)`: bind to the calling thread. One binding per
+  thread and per notification.
+- `SetIrqNotification(irq, ntfn)`: deliver interrupt `irq` as `Signal`
+  through the given (badged) notification capability.
+- `AckInterrupt(irq)`: unmask the line after handling.
+
+An interrupt masks its line and signals its notification. If no route is
+set it stays masked.
+
+## Reply slot rules
+
+- A server thread has at most one pending reply. `Receive` (not
+  `ReplyRecv`) with a pending reply drops it: the caller wakes with
+  `NoReply` instead of blocking forever.
+- A reply slot names the caller by thread id **and generation**, so a reply
+  can never reach a different thread that reused the id.
+- If the server thread dies, its pending caller wakes with `NoReply`.
+- If the caller dies, the reply slot is simply stale and `Reply` is a
+  successful no-op.
+
+## Death
+
+Destroying an endpoint or notification wakes every thread blocked on it with
+`InvalidCap`. Destroying a thread removes it from every queue, binding,
+reply slot and IRQ route.
+
+## Register ABI (RISC-V)
+
+Inputs: `a7` = syscall number, arguments in `a0..a5`. A capability argument
+of `NONE` (`usize::MAX`) means "no capability".
+
+Outputs: `a0` = status (`0` = ok, else an `ErrorCode`), `a1..` = results.
+Receiving operations (`Receive`, `Call`, `ReplyRecv`, `Wait`) return:
+
+| Register | Value                                                  |
+|----------|--------------------------------------------------------|
+| a0       | status                                                 |
+| a1       | badge (message) or signal bits (notification)          |
+| a2..a5   | message words 0..3                                     |
+| a6       | received capability slot, or `NONE`                    |
+| a7       | kind: 0 = message, 1 = call (reply expected), 2 = signal |