Merge pull request #104 from sysprog21/refine-coroutine

Refine coroutine for hart scheduling

Author: jserv

.ci/common.sh

@@ -45,7 +45,7 @@ get_timeout() {
 
     case "${OS_TYPE}" in
         Darwin)
-            echo "600"
+            echo "900"
             ;;
         Linux)
             echo "90"

.ci/test-netdev.sh

@@ -6,10 +6,10 @@ export SCRIPT_DIR
 source "${SCRIPT_DIR}/common.sh"
 
 # Override timeout for netdev tests
-# Network tests need different timeout: 30s for Linux, 600s for macOS
+# Network tests need different timeout: 30s for Linux, 900s for macOS
 case "${OS_TYPE}" in
     Darwin)
-        TIMEOUT=600
+        TIMEOUT=900
         ;;
     Linux)
         TIMEOUT=30

.github/workflows/main.yml

@@ -103,7 +103,7 @@ jobs:
     - name: automated test
       run: .ci/autorun.sh
       shell: bash
-      timeout-minutes: 15
+      timeout-minutes: 20
     - name: netdev test
       # NOTE: vmnet requires sudo privileges which may not be available in GitHub Actions
       # This test is conditional and will be skipped if sudo is not available

coro.c

@@ -201,6 +201,7 @@ typedef struct {
 /* Internal coroutine structure */
 
 typedef struct {
+    uint32_t id;             /* Coroutine identifier */
     void (*func)(void *);    /* Entry point function (user-provided) */
     void *user_data;         /* User data (hart pointer) */
     coro_state_t state;      /* Current state */
@@ -213,7 +214,8 @@ typedef struct {
 
 static struct {
     coro_t **coroutines;   /* Array of coroutine pointers */
-    uint32_t n_hart;       /* Number of harts */
+    uint32_t total_slots;  /* Total number of coroutine slots */
+    uint32_t hart_slots;   /* Number of slots reserved for harts */
     uint32_t current_hart; /* Currently executing hart ID */
     bool initialized;      /* True if subsystem initialized */
     coro_t *running;       /* Currently running coroutine */
@@ -399,25 +401,27 @@ static void coro_entry_wrapper(void *arg)
 
 /* Public API implementation */
 
-bool coro_init(uint32_t n_hart)
+bool coro_init(uint32_t total_slots, uint32_t hart_slots)
 {
     if (coro_state.initialized) {
         fprintf(stderr, "coro_init: already initialized\n");
         return false;
     }
 
-    if (n_hart == 0 || n_hart > 32) {
-        fprintf(stderr, "coro_init: invalid n_hart=%u\n", n_hart);
+    if (total_slots == 0 || total_slots > 256 || hart_slots > total_slots) {
+        fprintf(stderr, "coro_init: invalid slots (total=%u, hart=%u)\n",
+                total_slots, hart_slots);
         return false;
     }
 
-    coro_state.coroutines = calloc(n_hart, sizeof(coro_t *));
+    coro_state.coroutines = calloc(total_slots, sizeof(coro_t *));
     if (!coro_state.coroutines) {
         fprintf(stderr, "coro_init: failed to allocate coroutines array\n");
         return false;
     }
 
-    coro_state.n_hart = n_hart;
+    coro_state.total_slots = total_slots;
+    coro_state.hart_slots = hart_slots;
     coro_state.current_hart = CORO_HART_ID_IDLE;
     coro_state.initialized = true;
     coro_state.running = NULL;
@@ -430,7 +434,7 @@ void coro_cleanup(void)
     if (!coro_state.initialized)
         return;
 
-    for (uint32_t i = 0; i < coro_state.n_hart; i++) {
+    for (uint32_t i = 0; i < coro_state.total_slots; i++) {
         if (coro_state.coroutines[i]) {
             coro_t *co = coro_state.coroutines[i];
             if (co->context) {
@@ -446,22 +450,23 @@ void coro_cleanup(void)
 
     free(coro_state.coroutines);
     coro_state.coroutines = NULL;
-    coro_state.n_hart = 0;
+    coro_state.total_slots = 0;
+    coro_state.hart_slots = 0;
     coro_state.current_hart = CORO_HART_ID_IDLE; /* Reset to idle state */
     coro_state.initialized = false;
     coro_state.running = NULL;
     tls_running_coro = NULL; /* Reset TLS as well */
 }
 
-bool coro_create_hart(uint32_t hart_id, void (*func)(void *), void *hart)
+bool coro_create_hart(uint32_t slot_id, void (*func)(void *), void *arg)
 {
     if (!coro_state.initialized) {
         fprintf(stderr, "coro_create_hart: not initialized\n");
         return false;
     }
 
-    if (hart_id >= coro_state.n_hart) {
-        fprintf(stderr, "coro_create_hart: invalid hart_id=%u\n", hart_id);
+    if (slot_id >= coro_state.total_slots) {
+        fprintf(stderr, "coro_create_hart: invalid slot_id=%u\n", slot_id);
         return false;
     }
 
@@ -470,9 +475,9 @@ bool coro_create_hart(uint32_t hart_id, void (*func)(void *), void *hart)
         return false;
     }
 
-    if (coro_state.coroutines[hart_id]) {
-        fprintf(stderr, "coro_create_hart: hart %u already has coroutine\n",
-                hart_id);
+    if (coro_state.coroutines[slot_id]) {
+        fprintf(stderr, "coro_create_hart: slot %u already has coroutine\n",
+                slot_id);
         return false;
     }
 
@@ -484,8 +489,9 @@ bool coro_create_hart(uint32_t hart_id, void (*func)(void *), void *hart)
     }
 
     /* Store user function and data */
+    co->id = slot_id;
     co->func = func;
-    co->user_data = hart;
+    co->user_data = arg;
     co->state = CORO_STATE_SUSPENDED;
 
     /* Allocate context */
@@ -529,34 +535,35 @@ bool coro_create_hart(uint32_t hart_id, void (*func)(void *), void *hart)
     }
 #endif
 
-    coro_state.coroutines[hart_id] = co;
+    coro_state.coroutines[slot_id] = co;
     return true;
 }
 
-void coro_resume_hart(uint32_t hart_id)
+void coro_resume_hart(uint32_t slot_id)
 {
-    if (!coro_state.initialized || hart_id >= coro_state.n_hart) {
-        fprintf(stderr, "coro_resume_hart: invalid hart_id=%u\n", hart_id);
+    if (!coro_state.initialized || slot_id >= coro_state.total_slots) {
+        fprintf(stderr, "coro_resume_hart: invalid slot_id=%u\n", slot_id);
         return;
     }
 
-    coro_t *co = coro_state.coroutines[hart_id];
+    coro_t *co = coro_state.coroutines[slot_id];
     if (!co || !co->context) {
-        fprintf(stderr, "coro_resume_hart: hart %u has no coroutine\n",
-                hart_id);
+        fprintf(stderr, "coro_resume_hart: slot %u has no coroutine\n",
+                slot_id);
         return;
     }
 
     if (co->state != CORO_STATE_SUSPENDED) {
-        fprintf(stderr, "coro_resume_hart: hart %u not suspended (state=%d)\n",
-                hart_id, co->state);
+        /* This may happen if a coroutine is waiting on I/O and the main loop
+         * tries to resume it. It is not a fatal error.
+         */
         return;
     }
 
     /* Check for stack overflow before resuming */
     coro_check_stack(co);
 
-    coro_state.current_hart = hart_id;
+    coro_state.current_hart = slot_id;
     co->state = CORO_STATE_RUNNING;
     jump_into(co);
 
@@ -586,12 +593,12 @@ void coro_yield(void)
     jump_out(co);
 }
 
-bool coro_is_suspended(uint32_t hart_id)
+bool coro_is_suspended(uint32_t slot_id)
 {
-    if (!coro_state.initialized || hart_id >= coro_state.n_hart)
+    if (!coro_state.initialized || slot_id >= coro_state.hart_slots)
         return false;
 
-    coro_t *co = coro_state.coroutines[hart_id];
+    coro_t *co = coro_state.coroutines[slot_id];
     if (!co || !co->context)
         return false;
 

coro.h

@@ -1,44 +1,39 @@
-/* Lightweight coroutine for multi-hart execution */
-
 #pragma once
 
 #include <stdbool.h>
+#include <stddef.h>
 #include <stdint.h>
 
 /* Forward declaration */
 typedef struct __hart_internal hart_t;
 
-/* Initialize coroutine subsystem for a VM with n_hart cores */
-bool coro_init(uint32_t n_hart);
+/* Sentinel value when no coroutine is executing */
+#define CORO_INVALID_ID UINT32_MAX
+
+/* Initialize coroutine subsystem for SMP hart scheduling.
+ * total_slots: total coroutine slots (usually n_hart)
+ * hart_slots: number of actual CPU harts
+ */
+bool coro_init(uint32_t total_slots, uint32_t hart_slots);
 
 /* Cleanup coroutine subsystem */
 void coro_cleanup(void);
 
-/* Create coroutine for a specific hart.
- * @hart_id: Hart identifier (0 to n_hart-1)
- * @func: Entry point function for the coroutine
- * @hart: User data (hart_t pointer) passed to the entry function
- *
- * Returns: true on success, false on failure
+/* Create coroutine for a hart.
+ * slot_id: Hart identifier
+ * func: Entry point (hart execution loop)
+ * arg: User data (hart_t pointer)
  */
-bool coro_create_hart(uint32_t hart_id, void (*func)(void *), void *hart);
+bool coro_create_hart(uint32_t slot_id, void (*func)(void *), void *arg);
 
-/* Resume execution of a specific hart's coroutine
- * The coroutine will execute until it yields or terminates.
- */
-void coro_resume_hart(uint32_t hart_id);
+/* Resume execution of a hart coroutine */
+void coro_resume_hart(uint32_t slot_id);
 
-/* Yield from current hart (called from WFI)
- * Suspends the current coroutine and returns control to the scheduler.
- */
+/* Yield from current hart (called from WFI) */
 void coro_yield(void);
 
-/* Check if a hart's coroutine is suspended (waiting in WFI)
- * Returns: true if suspended, false otherwise
- */
-bool coro_is_suspended(uint32_t hart_id);
+/* Check if hart is suspended (in WFI) */
+bool coro_is_suspended(uint32_t slot_id);
 
-/* Get the currently running hart ID
- * Returns: Hart ID of the currently executing coroutine, or UINT32_MAX if idle
- */
+/* Get currently executing hart ID */
 uint32_t coro_current_hart_id(void);