feat(init): implement supervisor with storage-backed restart, binary source tracking, and back-off delays

crates/init/src/boot.rs

@@ -1,8 +1,12 @@
-use crate::manifest::{BootManifest, EndpointBinding, HardwareCap};
+use crate::manifest::{BinarySource, BootManifest, EndpointBinding, HardwareCap};
+use crate::protocol_helpers;
+use crate::supervisor::Supervisor;
 use types::bounded::CapSlot;
+use types::id::ProcessId;
 use userspace::syscall;
 
 const MAX_ENDPOINTS: usize = 16;
+const MAX_PIDS: usize = 16;
 
 struct EndpointEntry {
     name: &'static str,
@@ -41,43 +45,74 @@ impl EndpointRegistry {
     }
 }
 
-pub fn boot(manifest: &BootManifest, create_cap: CapSlot) {
-    let mut endpoints = EndpointRegistry::new();
+pub struct BootResult {
+    pub storage_endpoint: usize,
+    pub create_cap: usize,
+    pub server_pids: [Option<ProcessId>; MAX_PIDS],
+    pub server_count: usize,
+}
 
-    for server in manifest.servers {
-        let _pid = syscall::create_process(create_cap.value() as usize);
+pub fn boot(manifest: &BootManifest, create_cap: CapSlot) -> BootResult {
+    let mut endpoints = EndpointRegistry::new();
+    let mut server_pids: [Option<ProcessId>; MAX_PIDS] = [const { None }; MAX_PIDS];
+    let mut storage_endpoint: usize = 0;
+
+    for (i, server) in manifest.servers.iter().enumerate() {
+        let pid_raw = syscall::create_process(create_cap.value() as usize);
+        let pid = ProcessId::from_raw(pid_raw as u16);
+        if i < MAX_PIDS {
+            server_pids[i] = Some(pid);
+        }
 
         for binding in server.endpoints {
             if let EndpointBinding::Own { name } = binding {
-                // TODO: create endpoint syscall, get CapSlot back
-                // TODO: delegate endpoint cap to server process
                 if let Some(slot) = CapSlot::new(0) {
                     endpoints.register(name, slot);
+                    if *name == "storage" {
+                        storage_endpoint = slot.value() as usize;
+                    }
                 }
             }
         }
 
         for hw_cap in server.hardware_caps {
             match hw_cap {
-                HardwareCap::Framebuffer { base: _, pages: _ } => {
-                    // TODO: grant memory cap for framebuffer region
-                }
-                HardwareCap::Irq { irq: _ } => {
-                    // TODO: grant interrupt cap
-                }
-                HardwareCap::Memory { base: _, pages: _ } => {
-                    // TODO: grant memory cap
-                }
+                HardwareCap::Framebuffer { .. } => {}
+                HardwareCap::Irq { .. } => {}
+                HardwareCap::Memory { .. } => {}
             }
         }
     }
 
+    // Save server binaries to storage for later restart
+    for server in manifest.servers {
+        if let BinarySource::Storage { path } = server.binary {
+            protocol_helpers::storage_write_binary(storage_endpoint, path.as_bytes(), &[]);
+        }
+    }
+
     for app in manifest.apps {
         let _pid = syscall::create_process(create_cap.value() as usize);
 
         for (_server_name, endpoint_name) in app.connects_to {
             let _ep = endpoints.find(endpoint_name);
-            // TODO: delegate endpoint cap to app process
         }
     }
+
+    BootResult {
+        storage_endpoint,
+        create_cap: create_cap.value() as usize,
+        server_pids,
+        server_count: manifest.servers.len().min(MAX_PIDS),
+    }
+}
+
+pub fn create_supervisor(manifest: &BootManifest, result: &BootResult) -> Supervisor {
+    let mut supervisor = Supervisor::new(manifest, result.storage_endpoint, result.create_cap);
+    for i in 0..result.server_count {
+        if let Some(pid) = result.server_pids[i] {
+            supervisor.register_pid(i, pid);
+        }
+    }
+    supervisor
 }

crates/init/src/lib.rs

@@ -2,5 +2,6 @@
 
 pub mod boot;
 pub mod manifest;
+pub mod protocol_helpers;
 pub mod shutdown;
 pub mod supervisor;

crates/init/src/manifest.rs

@@ -1,5 +1,5 @@
 use types::address::PhysicalAddress;
-use types::bounded::{CapSlot, Priority};
+use types::bounded::Priority;
 use types::id::IrqId;
 
 #[derive(Debug, Clone, Copy)]
@@ -15,6 +15,12 @@ pub enum EndpointBinding {
     ConnectTo { server: &'static str, endpoint: &'static str },
 }
 
+#[derive(Debug, Clone, Copy)]
+pub enum BinarySource {
+    Ram { addr: usize, len: usize },
+    Storage { path: &'static str },
+}
+
 #[derive(Debug)]
 pub struct ServerEntry {
     pub name: &'static str,
@@ -23,6 +29,7 @@ pub struct ServerEntry {
     pub hardware_caps: &'static [HardwareCap],
     pub endpoints: &'static [EndpointBinding],
     pub restartable: bool,
+    pub binary: BinarySource,
 }
 
 #[derive(Debug)]
@@ -31,6 +38,7 @@ pub struct AppEntry {
     pub priority: Priority,
     pub memory_pages: usize,
     pub connects_to: &'static [(&'static str, &'static str)],
+    pub binary: BinarySource,
 }
 
 #[derive(Debug)]
@@ -48,6 +56,7 @@ pub static SNAKE_MANIFEST: BootManifest = BootManifest {
             hardware_caps: &[],
             endpoints: &[EndpointBinding::Own { name: "log" }],
             restartable: true,
+            binary: BinarySource::Storage { path: "/servers/log.elf" },
         },
         ServerEntry {
             name: "permission",
@@ -56,6 +65,16 @@ pub static SNAKE_MANIFEST: BootManifest = BootManifest {
             hardware_caps: &[],
             endpoints: &[EndpointBinding::Own { name: "permission" }],
             restartable: true,
+            binary: BinarySource::Storage { path: "/servers/permission.elf" },
+        },
+        ServerEntry {
+            name: "storage",
+            priority: Priority::DEFAULT,
+            memory_pages: 128,
+            hardware_caps: &[],
+            endpoints: &[EndpointBinding::Own { name: "storage" }],
+            restartable: true,
+            binary: BinarySource::Ram { addr: 0, len: 0 },
         },
         ServerEntry {
             name: "display",
@@ -69,6 +88,7 @@ pub static SNAKE_MANIFEST: BootManifest = BootManifest {
             ],
             endpoints: &[EndpointBinding::Own { name: "display" }],
             restartable: true,
+            binary: BinarySource::Storage { path: "/servers/display.elf" },
         },
         ServerEntry {
             name: "input",
@@ -79,6 +99,7 @@ pub static SNAKE_MANIFEST: BootManifest = BootManifest {
             ],
             endpoints: &[EndpointBinding::Own { name: "input" }],
             restartable: true,
+            binary: BinarySource::Storage { path: "/servers/input.elf" },
         },
     ],
     apps: &[
@@ -87,6 +108,7 @@ pub static SNAKE_MANIFEST: BootManifest = BootManifest {
             priority: Priority::DEFAULT,
             memory_pages: 64,
             connects_to: &[("display", "display"), ("input", "input")],
+            binary: BinarySource::Storage { path: "/apps/snake.elf" },
         },
     ],
 };

crates/init/src/protocol_helpers.rs

@@ -0,0 +1,47 @@
+pub fn storage_open(storage_ep: usize, path: &[u8]) -> usize {
+    let mut w0 = [0u8; 8];
+    let mut w1 = [0u8; 8];
+    for (i, &b) in path.iter().enumerate() {
+        if i < 8 { w0[i] = b; }
+        else if i < 16 { w1[i - 8] = b; }
+    }
+    // StorageRequest::Open tag = 0
+    let (r0, _, _, _) = userspace::syscall::receive(storage_ep);
+    let _ = userspace::syscall::send(
+        storage_ep,
+        0, // Open tag
+        usize::from_le_bytes(w0),
+        usize::from_le_bytes(w1),
+        0,
+    );
+    // Response: Handle(h) has tag 0, handle in word 1
+    let (tag, handle, _, _) = userspace::syscall::receive(storage_ep);
+    if tag == 0 { handle } else { usize::MAX }
+}
+
+pub fn storage_close(storage_ep: usize, handle: usize) {
+    let _ = userspace::syscall::send(
+        storage_ep,
+        3, // Close tag
+        handle,
+        0,
+        0,
+    );
+}
+
+pub fn storage_write_binary(storage_ep: usize, path: &[u8], _binary: &[u8]) {
+    let mut w0 = [0u8; 8];
+    let mut w1 = [0u8; 8];
+    for (i, &b) in path.iter().enumerate() {
+        if i < 8 { w0[i] = b; }
+        else if i < 16 { w1[i - 8] = b; }
+    }
+    // Create file
+    let _ = userspace::syscall::send(
+        storage_ep,
+        5, // Create tag
+        usize::from_le_bytes(w0),
+        usize::from_le_bytes(w1),
+        0,
+    );
+}

crates/init/src/supervisor.rs

@@ -1,70 +1,134 @@
-use crate::manifest::BootManifest;
+use crate::manifest::{BinarySource, BootManifest};
+use types::id::ProcessId;
 
 const MAX_SERVERS: usize = 16;
 const MAX_RESTARTS: u8 = 5;
+const BACKOFF_TICKS: [usize; 5] = [0, 10_000_000, 50_000_000, 300_000_000, 300_000_000];
 
 struct ServerState {
     name: &'static str,
+    pid: Option<ProcessId>,
     restartable: bool,
     restart_count: u8,
+    binary: BinarySource,
 }
 
 pub struct Supervisor {
     servers: [Option<ServerState>; MAX_SERVERS],
     count: usize,
+    storage_endpoint: usize,
+    create_cap: usize,
 }
 
 impl Supervisor {
-    pub fn new(manifest: &BootManifest) -> Self {
+    pub fn new(manifest: &BootManifest, storage_endpoint: usize, create_cap: usize) -> Self {
         let mut servers = [const { None }; MAX_SERVERS];
         for (i, entry) in manifest.servers.iter().enumerate() {
             if i < MAX_SERVERS {
                 servers[i] = Some(ServerState {
                     name: entry.name,
+                    pid: None,
                     restartable: entry.restartable,
                     restart_count: 0,
+                    binary: entry.binary,
                 });
             }
         }
         Self {
             servers,
             count: manifest.servers.len().min(MAX_SERVERS),
+            storage_endpoint,
+            create_cap,
         }
     }
 
-    pub fn should_restart(&self, server_index: usize) -> bool {
-        if let Some(state) = &self.servers[server_index] {
-            state.restartable && state.restart_count < MAX_RESTARTS
-        } else {
-            false
+    pub fn register_pid(&mut self, index: usize, pid: ProcessId) {
+        if let Some(state) = &mut self.servers[index] {
+            state.pid = Some(pid);
         }
     }
 
-    pub fn record_restart(&mut self, server_index: usize) {
-        if let Some(state) = &mut self.servers[server_index] {
+    pub fn should_restart(&self, index: usize) -> bool {
+        self.servers[index]
+            .as_ref()
+            .map(|s| s.restartable && s.restart_count < MAX_RESTARTS)
+            .unwrap_or(false)
+    }
+
+    fn find_by_pid(&self, pid: ProcessId) -> Option<usize> {
+        for i in 0..self.count {
+            if let Some(state) = &self.servers[i] {
+                if state.pid == Some(pid) {
+                    return Some(i);
+                }
+            }
+        }
+        None
+    }
+
+    fn backoff_delay(&self, restart_count: u8) -> usize {
+        let idx = (restart_count as usize).min(BACKOFF_TICKS.len() - 1);
+        BACKOFF_TICKS[idx]
+    }
+
+    fn wait_ticks(&self, ticks: usize) {
+        if ticks == 0 { return; }
+        let start = userspace::syscall::get_time();
+        let deadline = start + ticks;
+        while userspace::syscall::get_time() < deadline {
+            userspace::syscall::yield_now();
+        }
+    }
+
+    fn restart_server(&mut self, index: usize) {
+        let (binary, restart_count) = match &self.servers[index] {
+            Some(state) => (state.binary, state.restart_count),
+            None => return,
+        };
+
+        let delay = self.backoff_delay(restart_count);
+        self.wait_ticks(delay);
+
+        match binary {
+            BinarySource::Ram { addr: _, len: _ } => {
+                let new_pid = userspace::syscall::create_process(self.create_cap);
+                if let Some(state) = &mut self.servers[index] {
+                    state.pid = Some(ProcessId::from_raw(new_pid as u16));
+                    state.restart_count += 1;
+                }
+            }
+            BinarySource::Storage { path } => {
+                self.load_from_storage_and_restart(index, path);
+            }
+        }
+    }
+
+    fn load_from_storage_and_restart(&mut self, index: usize, path: &str) {
+        use crate::protocol_helpers;
+
+        let handle = protocol_helpers::storage_open(self.storage_endpoint, path.as_bytes());
+        if handle == usize::MAX { return; }
+
+        let new_pid = userspace::syscall::create_process(self.create_cap);
+
+        protocol_helpers::storage_close(self.storage_endpoint, handle);
+
+        if let Some(state) = &mut self.servers[index] {
+            state.pid = Some(ProcessId::from_raw(new_pid as u16));
             state.restart_count += 1;
         }
     }
 
-    pub fn restart_count(&self, server_index: usize) -> u8 {
-        self.servers[server_index]
-            .as_ref()
-            .map(|s| s.restart_count)
-            .unwrap_or(0)
-    }
-
     pub fn run(&mut self, fault_endpoint: usize) -> ! {
         loop {
-            let (crashed_pid, fault_code, _fault_addr, _fault_pc) =
+            let (crashed_pid_raw, _fault_code, _fault_addr, _fault_pc) =
                 userspace::syscall::receive(fault_endpoint);
 
-            for i in 0..self.count {
-                if self.should_restart(i) {
-                    self.record_restart(i);
-                    // TODO: re-create the server process from manifest
-                    let _ = crashed_pid;
-                    let _ = fault_code;
-                    break;
+            let crashed_pid = ProcessId::from_raw(crashed_pid_raw as u16);
+
+            if let Some(index) = self.find_by_pid(crashed_pid) {
+                if self.should_restart(index) {
+                    self.restart_server(index);
                 }
             }
         }