fix joins & detaches

arch/sysv_x86-64/thread-entry.asm

@@ -2,7 +2,9 @@
 bits 64
 
 section .text
-global _cia_start_thread
+global _rt_thread_start
+
+extern _rt_thread_finish
 
 ; flags, &stack[-2], &parent_tid, &child_tid, 0
 
@@ -23,7 +25,7 @@ global _cia_start_thread
 ;     0 if returning as a parent
 ;     1 if returning as a child
 ;     negative value if there was an error making the thread
-_cia_start_thread:
+_rt_thread_start:
     mov r10, rcx
     ; Setup child stack
     sub rsi, 24
@@ -43,10 +45,9 @@ _cia_start_thread:
     pop rax ; thread_fn
     pop rdi ; ctx
     call rax
-    ; Make return value the first arg and call exit syscall
+    ; Make return value the first arg and call thread finish routine
     mov rdi, rax
-    mov rax, 60
-    syscall
+    call _rt_thread_finish
     ; Just to be sure...    
     hlt
 .exit:

cia.c

@@ -3,7 +3,6 @@
 
 // Base includes
 #include <cia/def.h>
-#include <cia/internal.h>
 
 // Platform-dependent sources
 #include _CIA_OS_CONF

include/cia/internal.h

@@ -11,13 +11,28 @@ struct Cia_CRT_Params {
     void *tls_image_base;
 };
 
+// These control how the thread should behave once it's terminated
+// if the main thread didn't call join nor detach, the thread
+// would just wait until one of them is set.
+// If the thread is joined,
+// would signal main thread that it has finished and exit without
+// cleaning up the resources (thrd_join does that for us)
+// If the thread is detached, it would clear it's resources and exit
+#define _CIA_THREAD_BEHAVIOUR_NOT_SET 0x0
+#define _CIA_THREAD_BEHAVIOUR_JOIN    0x1
+#define _CIA_THREAD_BEHAVIOUR_DETACH  0x2
+#define _CIA_THREAD_BEHAVIOUR_FINISH  0x3
+
 struct Cia_TCB typedef Cia_TCB;
 struct Cia_TCB {
-    u64 thread_id;
-    u64 parent_id;
-    u64 pad0[3];
-    u64 stack_canary;
-    u64 pad1[2];
+    /* +0x00 */ u64 thread_id;
+    /* +0x08 */ u64 parent_id;
+    /* +0x10 */ u32 thread_behaviour; /* One of the CIA_THREAD_BEHAVIOR_* constants */
+    /* +0x14 */ u32 thread_finished;
+    /* +0x18 */ u32 pad0;
+    /* +0x1c */ i32 exit_code;
+    /* +0x20 */ u64 pad1;
+    /* +0x28 */ u64 stack_canary;
 };
 
 

include/threads.h

@@ -17,3 +17,5 @@ enum {
 };
 
 int thrd_create(thrd_t *thr, thrd_start_t func, void *arg);
+int thrd_join(thrd_t thr, int *out_exit_code);
+int thrd_detach(thrd_t thr);

include/tinyrt.h

@@ -55,7 +55,7 @@ struct _RT_Thread {
 };
 static _RT_Status _rt_thread_current(_RT_Thread *thread);
 static _RT_Status _rt_thread_create(_RT_Thread *thread, int (*thread_fn)(void *ctx), void *ctx);
-static _RT_Status _rt_thread_join(_RT_Thread *thread);
+static _RT_Status _rt_thread_join(_RT_Thread *thread, int *out_exit_code);
 static _RT_Status _rt_thread_detach(_RT_Thread *thread);
 static _RT_Status _rt_thread_terminate(_RT_Thread *thread);
 static _RT_Status _rt_thread_sleep(u64 time);

os/linux/conf.h

@@ -5,6 +5,8 @@ static u64 cia_stack_size;
 static u64 cia_tls_image_size;
 static void *cia_tls_image_base;
 
+#include <cia/internal.h>
+
 #include <linux/futex.h>
 #include <sys/mman.h>
 #include <errno.h>

os/linux/tinyrt-threads.c

@@ -1,5 +1,5 @@
 
-extern i64 _cia_start_thread(
+extern i64 _rt_thread_start(
     u64 flags,
     void *stack_base,
     int *parent_tid,
@@ -33,6 +33,7 @@ static _RT_Status _rt_thread_create(_RT_Thread *thread, int (*thread_fn)(void *c
     // Initialize the _RT_Thread handle, which would point to
     // the TCB
     thread->handle = tcb;
+    tcb->thread_finished = 0;
     // Create the new thread
     u64 flags = 0;
     // flags |= CLONE_CHILD_CLEARTID;
@@ -47,19 +48,52 @@ static _RT_Status _rt_thread_create(_RT_Thread *thread, int (*thread_fn)(void *c
     int *parent_tid = (int *)&tcb->parent_id;
     *child_tid = 1;
     *parent_tid = 0;
-    i64 ret = _cia_start_thread(flags, stack, parent_tid, child_tid, 0, thread_fn, ctx);
+    i64 ret = _rt_thread_start(flags, stack, parent_tid, child_tid, 0, thread_fn, ctx);
     if(ret < 0) {
         return _RT_ERROR_GENERIC;
     }
     return _RT_STATUS_OK;
 }
 
-static _RT_Status _rt_thread_join(_RT_Thread *thread) {
-    return _RT_ERROR_NOT_IMPLEMENTED;
+void _rt_thread_finish(int exit_code) {
+    Cia_TCB *tcb = (void *)((u64)__builtin_frame_address(0) & ~(cia_stack_size - 1));
+    // Wait until the main thread decides what to do with the child thread
+    while(tcb->thread_behaviour == _CIA_THREAD_BEHAVIOUR_NOT_SET) {
+        syscall(SYS_futex, &tcb->thread_behaviour, FUTEX_WAIT, _CIA_THREAD_BEHAVIOUR_NOT_SET, NULL, 0, 0);
+    }
+    if(tcb->thread_behaviour == _CIA_THREAD_BEHAVIOUR_JOIN) {
+        tcb->exit_code = exit_code;
+        // Idk if a memory barrier should be here, because we don't want the compiler
+        // to reorder these two lines. If that happens, and we get a spurious wake up
+        // after the thread_finished is set and before exit_code is set, the main thread
+        // will proceed to fetch an invalid exit code from the thread.
+        tcb->thread_finished = 1;
+        syscall(SYS_futex, &tcb->thread_finished, FUTEX_WAKE, 0, NULL, 0, 0);
+        sys_exit(exit_code);
+    }
+    else if(tcb->thread_behaviour == _CIA_THREAD_BEHAVIOUR_DETACH) {
+        // TODO: clean up the thread resources
+        sys_exit(exit_code);
+    }
+}
+
+static _RT_Status _rt_thread_join(_RT_Thread *thread, int *out_exit_code) {
+    Cia_TCB *tcb = thread->handle;
+    // Signal the thread that we want it to be joined
+    tcb->thread_behaviour = _CIA_THREAD_BEHAVIOUR_JOIN;
+    // Wait until the thread signals that it has completed the execution
+    while(tcb->thread_finished != 1) {
+        syscall(SYS_futex, &tcb->thread_finished, FUTEX_WAIT, 0, NULL, 0, 0);
+    }
+    // Set the exit code
+    *out_exit_code = tcb->exit_code;
+    return _RT_STATUS_OK;
 }
 
 static _RT_Status _rt_thread_detach(_RT_Thread *thread) {
-    return _RT_ERROR_NOT_IMPLEMENTED;
+    Cia_TCB *tcb = thread->handle;
+    tcb->thread_behaviour = _CIA_THREAD_BEHAVIOUR_DETACH;
+    return _RT_STATUS_OK;
 }
 
 static _RT_Status _rt_thread_terminate(_RT_Thread *thread) {

src/stdlib-thread/thread.c

@@ -5,4 +5,20 @@ int thrd_create(thrd_t *thr, thrd_start_t func, void *arg) {
         return thrd_success;
     }
     return thrd_error;
-}
+}
+
+int thrd_detach(thrd_t thr) {
+    _RT_Status status = _rt_thread_detach(&thr.thread);
+    if(status == _RT_STATUS_OK) {
+        return thrd_success;
+    }
+    return thrd_error;
+}
+
+int thrd_join(thrd_t thr, int *out_exit_code) {
+    _RT_Status status = _rt_thread_join(&thr.thread, out_exit_code);
+    if(status == _RT_STATUS_OK) {
+        return thrd_success;
+    }
+    return thrd_error;
+}

tests/threaded.c

@@ -36,7 +36,6 @@ static void print_int(i64 number) {
 
 static Cia_Mutex g_print_mutex;
 static Cia_Mutex g_mutex;
-static volatile _Atomic i64 n_completed = 0;
 static volatile i64 counter = 0;
 
 int thrd_func(void *arg) {
@@ -46,12 +45,6 @@ int thrd_func(void *arg) {
         counter += 1;
         cia_mutex_unlock(&g_mutex);
     }
-    atomic_fetch_add(&n_completed, 1);
-    for(;;) {
-        if(n_completed == 2) {
-            break;
-        }
-    }
     cia_mutex_lock(&g_print_mutex);
     print_string("child thread: counter = ");
     print_int(counter);
@@ -76,12 +69,8 @@ int main() {
         counter += 1;
         cia_mutex_unlock(&g_mutex);
     }
-    atomic_fetch_add(&n_completed, 1);
-    for(;;) {
-        if(n_completed == 2) {
-            break;
-        }
-    }
+    int exit_code;
+    thrd_join(thrd, &exit_code);
     cia_mutex_lock(&g_print_mutex);
     print_string("main thread: counter = ");
     print_int(counter);