Some more progress with debugging pager exits

conts/posix/mm0/main.c

@@ -140,14 +140,13 @@ void handle_requests(void)
 		break;
 	}
 	case L4_IPC_TAG_EXIT: {
-		/* Zombie test for kernel
 		struct task_ids ids;
 
 		l4_getid(&ids);
 
 		printf("\n%s: Destroying self (%d), along with any tasks.\n", __TASKNAME__, self_tid());
 		l4_thread_control(THREAD_DESTROY, &ids);
-		*/
+
 		/* An exiting task has no receive phase */
 		sys_exit(sender, (int)mr[0]);
 		return;

src/arch/arm/exception.c

@@ -123,7 +123,12 @@ void fault_ipc_to_pager(u32 faulty_pc, u32 fsr, u32 far)
 		       "returned error (%d). Suspend and exiting thread.\n",
 		       current->tid, err);
 		BUG_ON(current->nlocks);
-		sched_die_sync();
+
+		/* Declare our exit, currently only used by bug_on checks */
+		current->flags |= TASK_EXITING;
+		/* Try to die forever */
+		while (1)
+			sched_die_sync();
 	}
 }
 

src/generic/container.c

@@ -102,7 +102,8 @@ int init_pager(struct pager *pager,
 		address_space_attach(task, space);
 	} else {
 		/* Otherwise allocate conventionally */
-		task->space = address_space_create(0);
+		space = address_space_create(0);
+		address_space_attach(task, space);
 	}
 
 	/* Initialize ktcb */
@@ -115,6 +116,13 @@ int init_pager(struct pager *pager,
 	task->tgid = task->tid;
 	task->container = cont;
 
+	/*
+	 * Setup dummy container pointer so that curcont works,
+	 * and add the address space to container space list
+	 */
+	current->container = cont;
+	address_space_add(task->space);
+
 	/* Initialize uninitialized capability fields while on dummy */
 	list_foreach_struct(cap, &current->cap_list.caps, list) {
 		/* Initialize owner */
@@ -156,7 +164,6 @@ int init_pager(struct pager *pager,
 
 	/* Container list that keeps all tasks */
 	tcb_add(task);
-
 	return 0;
 }
 
@@ -169,16 +176,19 @@ int container_init_pagers(struct kernel_resources *kres,
 {
 	struct container *cont;
 	struct pager *pager;
+	int first = 1;
 
 	list_foreach_struct(cont, &kres->containers.list, list) {
 		for (int i = 0; i < cont->npagers; i++) {
 			pager = &cont->pager[i];
 
 			/* First pager initializes specially */
-			if (i == 0)
+			if (first) {
 				init_pager(pager, cont, current_pgd);
-			else
+				first = 0;
+			} else {
 				init_pager(pager, cont, 0);
+			}
 		}
 	}
 

src/generic/scheduler.c

@@ -301,6 +301,8 @@ void sched_die_pager(void)
  */
 void sched_die_child(void)
 {
+	int err;
+
 	/*
 	 * Find pager, he _must_ be there because he never
 	 * quits before quitting us
@@ -308,7 +310,8 @@ void sched_die_child(void)
 	struct ktcb *pager = tcb_find(current->pagerid);
 
 	/* Lock its task_dead queue */
-	mutex_lock(&pager->task_dead.list_lock);
+	if ((err = mutex_lock(&pager->task_dead.list_lock)) < 0)
+		return err;
 
 	/* Remove from container task list,
 	 * callers get -ESRCH */
@@ -332,7 +335,7 @@ void sched_die_child(void)
 	 * Add self to pager's dead tasks list,
 	 * to be deleted by pager
 	 */
-	ktcb_list_add(current, &pager->task_dead);
+	__ktcb_list_add_nolock(current, &pager->task_dead);
 
 	/* Now quit the scheduler */
 	preempt_disable();
@@ -354,15 +357,20 @@ void sched_die_child(void)
 	 * pager can safely delete us
 	 */
 	mutex_unlock(&pager->task_dead.list_lock);
+	schedule();
 	BUG();
 }
 
 void sched_die_sync(void)
 {
-	if (current->tid == current->pagerid)
-		sched_die_pager();
-	else
-		sched_die_child();
+	/*
+	 * Infinitely retry if mutexes get interrupted
+	 */
+	while (1)
+		if (current->tid == current->pagerid)
+			sched_die_pager();
+		else
+			sched_die_child();
 }
 
 /*

src/generic/space.c

@@ -68,7 +68,12 @@ void address_space_add(struct address_space *space)
 void address_space_remove(struct address_space *space)
 {
 	spin_lock(&curcont->space_list.list_lock);
-	BUG_ON(list_empty(&space->list));
+
+	/*
+	 * If current is quitting as the last task of this space,
+	 * its tcb may already be removed, and this is fine
+	 */
+	BUG_ON(list_empty(&space->list) && space != current->space);
 	BUG_ON(--curcont->space_list.count < 0);
 	list_remove_init(&space->list);
 	spin_unlock(&curcont->space_list.list_lock);

src/generic/tcb.c

@@ -70,7 +70,8 @@ void tcb_delete(struct ktcb *tcb)
 	BUG_ON(tcb->wqh_pager.sleepers > 0);
 	BUG_ON(tcb->wqh_send.sleepers > 0);
 	BUG_ON(tcb->wqh_recv.sleepers > 0);
-	BUG_ON(!list_empty(&tcb->task_list));
+	BUG_ON(!list_empty(&tcb->task_list) &&
+	       !(tcb->flags & TASK_EXITING));
 	BUG_ON(!list_empty(&tcb->rq_list) && tcb != current);
 	BUG_ON(tcb->rq && tcb != current);
 	BUG_ON(tcb->nlocks);

src/glue/arm/init.c

@@ -283,12 +283,6 @@ void init_finalize(struct kernel_resources *kres)
 	 */
 	kip.utcb = (u32)current->utcb_address;
 
-	/*
-	 * Added the first pager's space area to
-	 * the address space list of the related container
-	 */
-	address_space_add(current->space);
-
 	/*
 	 * Start the scheduler, jumping to task
 	 */

src/lib/mutex.c

@@ -117,6 +117,7 @@ int mutex_lock(struct mutex *mutex)
 	 * undeterministic as to how many retries will result in success.
 	 * We may need to add priority-based locking.
 	 */
+	printk("Thread (%d) locking (%p) nlocks: %d\n", current->tid, mutex, current->nlocks);
 	for (;;) {
 		spin_lock(&mutex->wqh.slock);
 		if (!__mutex_lock(&mutex->lock)) { /* Could not lock, sleep. */
@@ -131,6 +132,7 @@ int mutex_lock(struct mutex *mutex)
 
 			/* Did we wake up normally or get interrupted */
 			if (current->flags & TASK_INTERRUPTED) {
+				printk("Thread (%d) interrupted on mutex sleep\n", current->tid);
 				current->flags &= ~TASK_INTERRUPTED;
 				return -EINTR;
 			}
@@ -140,11 +142,14 @@ int mutex_lock(struct mutex *mutex)
 		}
 	}
 	spin_unlock(&mutex->wqh.slock);
+	printk("Thread (%d) locked (%p) nlocks: %d\n", current->tid, mutex, current->nlocks);
 	return 0;
 }
 
 static inline void mutex_unlock_common(struct mutex *mutex, int sync)
 {
+	struct ktcb *c = current; if (c);
+	printk("Thread (%d) unlocking (%p) nlocks: %d\n", c->tid, mutex, c->nlocks);
 	spin_lock(&mutex->wqh.slock);
 	__mutex_unlock(&mutex->lock);
 	current->nlocks--;

src/lib/wait.c

@@ -38,7 +38,7 @@ void task_unset_wqh(struct ktcb *task)
 /*
  * Initiate wait on current task that
  * has already been placed in a waitqueue
- * 
+ *
  * NOTE: This enables preemption and wait_on_prepare()
  * should be called first.
  */
@@ -63,7 +63,7 @@ int wait_on_prepared_wait(void)
  * Do all preparations to sleep but return without sleeping.
  * This is useful if the task needs to get in the waitqueue before
  * it releases a lock.
- * 
+ *
  * NOTE: This disables preemption and it should be enabled by a
  * call to wait_on_prepared_wait() - the other function of the pair.
  */