exit() almost there.

- Implemented reasonable way to suspend task.
  - A task that has a pending suspend would be interrupted
    from its sleep via the suspender task.
  - If suspend was raised and right after, task became about to sleep,
    then scheduler wakes it up.
  - If suspend was raised when task was in user mode, then an irq suspends it.
  - Also suspends are checked at the end of a syscall so that if suspend was
    raised because of a syscall from the task, the task is suspended before it
    goes back to user mode.

  - This mechanism is very similar to signals, and it may lead as a base for
    implementing signal handling.

- Implemented common vma dropping for shadow vm object dropping and task exiting.

include/l4/generic/scheduler.h

@@ -36,8 +36,11 @@ static inline struct ktcb *current_task(void)
 #define current			current_task()
 #define need_resched		(current->ts_need_resched)
 
+
 void sched_init_task(struct ktcb *task, int priority);
 void sched_prepare_sleep(void);
+void sched_suspend_sync(void);
+void sched_suspend_async(void);
 void sched_resume_sync(struct ktcb *task);
 void sched_resume_async(struct ktcb *task);
 void scheduler_start(void);

include/l4/generic/tcb.h

@@ -148,5 +148,7 @@ extern struct id_pool *thread_id_pool;
 extern struct id_pool *space_id_pool;
 extern struct id_pool *tgroup_id_pool;
 
+void task_process_pending_flags(void);
+
 #endif /* __TCB_H__ */
 

include/l4/lib/wait.h

@@ -10,6 +10,11 @@ struct waitqueue {
 	struct ktcb *task;
 };
 
+enum wakeup_flags {
+	WAKEUP_INTERRUPT = (1 << 0),
+	WAKEUP_SYNC	 = (1 << 1)
+};
+
 #define CREATE_WAITQUEUE_ON_STACK(wq, tsk)		\
 struct waitqueue wq = {					\
 	.task_list = { &wq.task_list, &wq.task_list },	\
@@ -65,9 +70,9 @@ do {								\
 	}							\
 } while(0);
 
-void wake_up(struct waitqueue_head *wqh, int sync);
-int wake_up_task(struct ktcb *task, int sync);
-void wake_up_all(struct waitqueue_head *wqh, int sync);
+void wake_up(struct waitqueue_head *wqh, unsigned int flags);
+int wake_up_task(struct ktcb *task, unsigned int flags);
+void wake_up_all(struct waitqueue_head *wqh, unsigned int flags);
 
 #endif /* __LIB_WAIT_H__ */
 

src/api/thread.c

@@ -28,7 +28,7 @@ int sys_thread_switch(syscall_context_t *regs)
 int thread_suspend(struct task_ids *ids)
 {
 	struct ktcb *task;
-	int ret;
+	int ret = 0;
 
 	if (!(task = find_task(ids->tid)))
 		return -ESRCH;
@@ -36,17 +36,13 @@ int thread_suspend(struct task_ids *ids)
 	if (task->state == TASK_INACTIVE)
 		return 0;
 
-	/* First show our intention to suspend thread */
+	/* Signify we want to suspend the thread */
 	task->flags |= TASK_SUSPENDING;
 
-	/*
-	 * Interrupt the task in case it was sleeping
-	 * so that it will be caught and suspended by
-	 * the scheduler.
-	 */
-	wake_up_task(task, 1);
+	/* Wake it up if it's sleeping */
+	wake_up_task(task, WAKEUP_INTERRUPT | WAKEUP_SYNC);
 
-	/* Wait until scheduler wakes us up */
+	/* Wait until task suspends itself */
 	WAIT_EVENT(&task->wqh_pager,
 		   task->state == TASK_INACTIVE, ret);
 

src/generic/irq.c

@@ -7,10 +7,11 @@
  */
 #include <l4/config.h>
 #include <l4/macros.h>
+#include <l4/generic/scheduler.h>
 #include <l4/generic/platform.h>
+#include <l4/generic/tcb.h>
 #include <l4/generic/irq.h>
 #include <l4/lib/mutex.h>
-#include <l4/generic/scheduler.h>
 #include <l4/lib/printk.h>
 #include INC_PLAT(irq.h)
 #include INC_ARCH(exception.h)
@@ -73,5 +74,11 @@ void do_irq(void)
 		printk("Spurious or broken irq\n"); BUG();
 	}
 	irq_enable(irq_index);
+
+	/* Process any pending flags for currently runnable task */
+	task_process_pending_flags();
 }
 
+
+
+

src/generic/scheduler.c

@@ -228,6 +228,44 @@ static inline int sched_recalc_ticks(struct ktcb *task, int prio_total)
 		SCHED_TICKS * task->priority / prio_total;
 }
 
+/*
+ * NOTE: Could do these as sched_prepare_suspend()
+ * + schedule() or need_resched = 1
+ */
+void sched_suspend_sync(void)
+{
+	preempt_disable();
+	sched_rq_remove_task(current);
+	current->state = TASK_INACTIVE;
+	current->flags &= ~TASK_SUSPENDING;
+	prio_total -= current->priority;
+	BUG_ON(prio_total <= 0);
+	preempt_enable();
+
+	/* Async wake up any waiters */
+	wake_up_task(find_task(current->pagerid), 0);
+	schedule();
+}
+
+void sched_suspend_async(void)
+{
+	preempt_disable();
+	sched_rq_remove_task(current);
+	current->state = TASK_INACTIVE;
+	current->flags &= ~TASK_SUSPENDING;
+	prio_total -= current->priority;
+	BUG_ON(prio_total <= 0);
+
+	/* This will make sure we yield soon */
+	preempt_enable();
+
+	/* Async wake up any waiters */
+	wake_up_task(find_task(current->pagerid), 0);
+	need_resched = 1;
+}
+
+
+
 /*
  * Tasks come here, either by setting need_resched (via next irq),
  * or by directly calling it (in process context).
@@ -270,17 +308,6 @@ void schedule()
 	/* Cannot have any irqs that schedule after this */
 	preempt_disable();
 
-#if 0
-	/* NOTE:
-	 * We could avoid unnecessary scheduling by detecting
-	 * a task that has been just woken up.
-	 */
-	if ((task->flags & TASK_WOKEN_UP) && in_process_context()) {
-		preempt_enable();
-		return 0;
-	}
-#endif
-
 	/* Reset schedule flag */
 	need_resched = 0;
 
@@ -293,43 +320,13 @@ void schedule()
 			sched_rq_add_task(current, rq_expired, RQ_ADD_BEHIND);
 	}
 
-	/* Check if there's a pending suspend for thread */
-	if (current->flags & TASK_SUSPENDING) {
-		/*
-		 * The task should have no locks and be in a runnable state.
-		 * (e.g. properly woken up by the suspender)
-		 */
-		if (current->nlocks == 0 &&
-		    current->state == TASK_RUNNABLE) {
-			/* Suspend it if suitable */
-			current->state = TASK_INACTIVE;
-			current->flags &= ~TASK_SUSPENDING;
-
-			/*
-			 * The task has been made inactive here.
-			 * A suspended task affects timeslices whereas
-			 * a sleeping task doesn't as it is believed
-			 * sleepers would become runnable soon.
-			 */
-			prio_total -= current->priority;
-			BUG_ON(prio_total <= 0);
-
-			/* Prepare to wake up any waiters */
-			wake_up(&current->wqh_pager, 0);
-		} else {
-			if (current->state == TASK_RUNNABLE)
-				sched_rq_remove_task(current);
-
-			/*
-			 * Top up task's ticks temporarily, and
-			 * wait for it to release its locks.
-			 */
-			current->state = TASK_RUNNABLE;
-			current->ticks_left = max(current->ticks_left,
-						  SCHED_GRANULARITY);
-			sched_rq_add_task(current, rq_runnable, RQ_ADD_FRONT);
-		}
-	}
+	/*
+	 * If task is about to sleep and
+	 * it has pending events, wake it up.
+	 */
+	if (current->flags & TASK_SUSPENDING &&
+	    current->state == TASK_SLEEPING)
+		wake_up_task(current, WAKEUP_INTERRUPT);
 
 	/* Determine the next task to be run */
 	if (rq_runnable->total > 0) {

src/generic/tcb.c

@@ -5,7 +5,10 @@
  */
 #include <l4/generic/tcb.h>
 #include <l4/generic/space.h>
+#include <l4/generic/scheduler.h>
+#include <l4/generic/preempt.h>
 #include <l4/lib/idpool.h>
+#include INC_ARCH(exception.h)
 
 /* ID pools for threads and spaces. */
 struct id_pool *thread_id_pool;
@@ -20,6 +23,25 @@ unsigned int need_resched_offset = offsetof(struct ktcb, ts_need_resched);
 unsigned int syscall_regs_offset = offsetof(struct ktcb, syscall_regs);
 
 
+/*
+ * When there is an asynchronous pending event to be handled by
+ * the task (e.g. task is suspending), normally it is processed
+ * when the task is returning to user mode from the kernel. If
+ * the event is raised when the task is in userspace, this call
+ * in irq context makes sure it is handled.
+ */
+void task_process_pending_flags(void)
+{
+	if (TASK_IN_USER(current)) {
+		if (current->flags & TASK_SUSPENDING) {
+			if (in_irq_context())
+				sched_suspend_async();
+			else
+				sched_suspend_sync();
+		}
+	}
+}
+
 #if 0
 int task_suspend(struct ktcb *task)
 {

src/glue/arm/systable.c

@@ -6,6 +6,7 @@
 #include <l4/lib/mutex.h>
 #include <l4/lib/printk.h>
 #include <l4/generic/space.h>
+#include <l4/generic/scheduler.h>
 #include <l4/api/errno.h>
 #include INC_GLUE(memlayout.h)
 #include INC_GLUE(syscall.h)
@@ -60,13 +61,15 @@ void syscall_init()
 /* Checks a syscall is legitimate and dispatches to appropriate handler. */
 int syscall(syscall_context_t *regs, unsigned long swi_addr)
 {
+	int ret = 0;
+
 	/* Check if genuine system call, coming from the syscall page */
 	if ((swi_addr & ARM_SYSCALL_PAGE) == ARM_SYSCALL_PAGE) {
 		/* Check within syscall offset boundary */
 		if (((swi_addr & syscall_offset_mask) >= 0) &&
 		    ((swi_addr & syscall_offset_mask) <= syscalls_end_offset)) {
 			/* Quick jump, rather than compare each */
-			return (*syscall_table[(swi_addr & 0xFF) >> 2])(regs);
+			ret = (*syscall_table[(swi_addr & 0xFF) >> 2])(regs);
 		} else {
 			printk("System call received from call @ 0x%lx."
 			       "Instruction: 0x%lx.\n", swi_addr,
@@ -78,5 +81,12 @@ int syscall(syscall_context_t *regs, unsigned long swi_addr)
 		       "Discarding.\n", swi_addr);
 		return -ENOSYS;
 	}
+
+	if (current->flags & TASK_SUSPENDING) {
+		BUG_ON(current->nlocks);
+		sched_suspend_sync();
+	}
+
+	return ret;
 }
 

src/lib/wait.c

@@ -59,7 +59,7 @@ int wait_on(struct waitqueue_head *wqh)
 }
 
 /* Wake up all */
-void wake_up_all(struct waitqueue_head *wqh, int sync)
+void wake_up_all(struct waitqueue_head *wqh, unsigned int flags)
 {
 	BUG_ON(wqh->sleepers < 0);
 	spin_lock(&wqh->slock);
@@ -72,11 +72,12 @@ void wake_up_all(struct waitqueue_head *wqh, int sync)
 		BUG_ON(list_empty(&wqh->task_list));
 		list_del_init(&wq->task_list);
 		wqh->sleepers--;
-		sleeper->flags |= TASK_INTERRUPTED;
+		if (flags & WAKEUP_INTERRUPT)
+			sleeper->flags |= TASK_INTERRUPTED;
 		printk("(%d) Waking up (%d)\n", current->tid, sleeper->tid);
 		spin_unlock(&wqh->slock);
 
-		if (sync)
+		if (flags & WAKEUP_SYNC)
 			sched_resume_sync(sleeper);
 		else
 			sched_resume_async(sleeper);
@@ -85,7 +86,7 @@ void wake_up_all(struct waitqueue_head *wqh, int sync)
 }
 
 /* Wake up single waiter */
-void wake_up(struct waitqueue_head *wqh, int sync)
+void wake_up(struct waitqueue_head *wqh, unsigned int flags)
 {
 	BUG_ON(wqh->sleepers < 0);
 	spin_lock(&wqh->slock);
@@ -98,11 +99,12 @@ void wake_up(struct waitqueue_head *wqh, int sync)
 		BUG_ON(list_empty(&wqh->task_list));
 		list_del_init(&wq->task_list);
 		wqh->sleepers--;
-		sleeper->flags |= TASK_INTERRUPTED;
+		if (flags & WAKEUP_INTERRUPT)
+			sleeper->flags |= TASK_INTERRUPTED;
 		printk("(%d) Waking up (%d)\n", current->tid, sleeper->tid);
 		spin_unlock(&wqh->slock);
 
-		if (sync)
+		if (flags & WAKEUP_SYNC)
 			sched_resume_sync(sleeper);
 		else
 			sched_resume_async(sleeper);
@@ -116,7 +118,7 @@ void wake_up(struct waitqueue_head *wqh, int sync)
  * as we were peeking on it, returns -1. @sync makes us immediately
  * yield or else leave it to scheduler's discretion.
  */
-int wake_up_task(struct ktcb *task, int sync)
+int wake_up_task(struct ktcb *task, unsigned int flags)
 {
 	struct waitqueue_head *wqh;
 	struct waitqueue *wq;
@@ -158,7 +160,8 @@ int wake_up_task(struct ktcb *task, int sync)
 	wqh->sleepers--;
 	task->waiting_on = 0;
 	task->wq = 0;
-	task->flags |= TASK_INTERRUPTED;
+	if (flags & WAKEUP_INTERRUPT)
+		task->flags |= TASK_INTERRUPTED;
 	spin_unlock(&wqh->slock);
 	spin_unlock(&task->waitlock);
 
@@ -167,7 +170,7 @@ int wake_up_task(struct ktcb *task, int sync)
 	 * safely resume it without locks as this is the only
 	 * code path that can resume the task.
 	 */
-	if (sync)
+	if (flags & WAKEUP_SYNC)
 		sched_resume_sync(task);
 	else
 		sched_resume_async(task);
@@ -175,5 +178,3 @@ int wake_up_task(struct ktcb *task, int sync)
 	return 0;
 }
 
-
-

tasks/libposix/exit.c

@@ -5,20 +5,22 @@
 #include <unistd.h>
 #include <l4/macros.h>
 
-static inline void l4_exit(int status)
+static inline void __attribute__ ((noreturn)) l4_exit(int status)
 {
-	int err;
+	int ret;
 
 	write_mr(L4SYS_ARG0, status);
 
-	/* Call pager with exit() request. */
-	err = l4_send(PAGER_TID, L4_IPC_TAG_EXIT);
-	printf("%s: L4 IPC Error: %d.\n", __FUNCTION__, err);
+	/* Call pager with exit() request and block on its receive phase */
+	ret = l4_sendrecv(PAGER_TID, PAGER_TID, L4_IPC_TAG_EXIT);
+
 	/* This call should not fail or return */
+	printf("%s: L4 IPC returned: %d.\n", __FUNCTION__, ret);
 	BUG();
 }
 
-void exit(int status)
+void __attribute__ ((noreturn)) _exit(int status)
 {
 	l4_exit(status);
 }
+

tasks/mm0/main.c

@@ -123,6 +123,11 @@ void handle_requests(void)
 		ret = sys_fork(sender);
 		break;
 	}
+	case L4_IPC_TAG_EXIT: {
+		/* An exiting task has no receive phase */
+		sys_exit(sender, (int)mr[0]);
+		return;
+	}
 	case L4_IPC_TAG_BRK: {
 //		ret = sys_brk(sender, (void *)mr[0]);
 //		break;

tasks/mm0/src/fault.c

@@ -365,7 +365,20 @@ int vma_drop_merge_delete(struct vm_area *vma, struct vm_obj_link *link)
 /* This version is used when exiting. */
 int vma_drop_merge_delete_all(struct vm_area *vma)
 {
-	struct vm_obj_link *vmo_link;
+	struct vm_obj_link *vmo_link, *n;
+
+	/* Vma cannot be empty */
+	BUG_ON(list_empty(&vma->vm_obj_list));
+
+	/* Traverse and get rid of all links */
+	list_for_each_entry_safe(vmo_link, n, &vma->vm_obj_list, list)
+		vma_drop_merge_delete(vma, vmo_link);
+
+	return 0;
+}
+int vma_drop_merge_delete_all_old(struct vm_area *vma)
+{
+	struct vm_obj_link *vmo_link, *n;
 
 	/* Get the first link on the vma */
 	BUG_ON(list_empty(&vma->vm_obj_list));
@@ -373,6 +386,9 @@ int vma_drop_merge_delete_all(struct vm_area *vma)
 			      struct vm_obj_link, list);
 
 	/* Traverse and get rid of all links */
+	list_for_each_entry_safe(vmo_link, n, &vma->vm_obj_list, list) {
+		vma_drop_merge_delete(vma, vmo_link);
+	}
 	do {
 		vma_drop_merge_delete(vma, vmo_link);
 
@@ -382,6 +398,7 @@ int vma_drop_merge_delete_all(struct vm_area *vma)
 	return 0;
 }
 
+
 /* TODO:
  * - Why not allocate a swap descriptor in vma_create_shadow() rather than
  *   a bare vm_object? It will be needed.

tasks/mm0/src/shm.c

@@ -106,19 +106,6 @@ static void *do_shmat(struct vm_file *shm_file, void *shm_addr, int shmflg,
 	return shm->shm_addr;
 }
 
-/* TODO: Do we need this?
- * MM0 never needs a task's utcb page. vfs needs it.
- * UTCBs get special treatment here. If the task
- * is attaching to its utcb, mm0 prefaults it so
- * that it can access it later on whether or not
- * the task makes a syscall to mm0 without first
- * faulting the utcb.
- */
-/*
-	if ((unsigned long)shmaddr == task->utcb_address)
-		utcb_prefault(task, VM_READ | VM_WRITE);
-*/
-
 void *sys_shmat(struct tcb *task, l4id_t shmid, void *shmaddr, int shmflg)
 {
 	struct vm_file *shm_file, *n;

tasks/mm0/src/utcb.c

@@ -1,9 +1,8 @@
 /*
- * Utcb address allocation for user tasks.
+ * utcb address allocation for user tasks.
  *
  * Copyright (C) 2008 Bahadir Balban
  */
-
 #include <stdio.h>
 #include <utcb.h>
 #include <lib/addr.h>
@@ -71,44 +70,3 @@ void *task_send_utcb_address(struct tcb *sender, l4id_t taskid)
 	return 0;
 }
 
-#if 0
-
-To be ditched
-/*
- * Triggered during a sys_shmat() by a client task when mapping its utcb.
- * This prefaults the utcb and maps it in to mm0 so that it can freely
- * access it anytime later.
- */
-int utcb_prefault(struct tcb *task, unsigned int vmflags)
-{
-	int err;
-	struct page *pg;
-
-	/* First map in the page to task with given flags, e.g. read/write */
-	if ((err = prefault_page(task, task->utcb_address, vmflags)) < 0) {
-		printf("%s: Failed: %d\n", __FUNCTION__, err);
-		return err;
-	}
-
-	/*
-	 * Get the topmost page. Since we did both a VM_READ and VM_WRITE
-	 * prefault, this gets a writeable instead of a read-only page.
-	 */
-	pg = task_virt_to_page(task, task->utcb_address);
-	if (!pg || IS_ERR(pg)) {
-		printf("%s: Cannot retrieve task %d's utcb page.\n",
-		       __FUNCTION__, task->tid);
-		BUG();
-	}
-
-	/* Map it in to self */
-	l4_map((void *)page_to_phys(pg), (void *)task->utcb_address, 1,
-	       MAP_USR_RW_FLAGS, self_tid());
-
-	/* Flag that says this task's utcb is mapped to mm0 as r/w */
-	task->utcb_mapped = 1;
-
-	return 0;
-}
-
-#endif

tasks/test0/src/forktest.c

@@ -32,7 +32,7 @@ int forktest(void)
 	printf("PID: %d, my global: %d\n", myid, global);
 	printf("-- PASSED --\n");
 out:
-	while(1)
-		;
+	printf("PID: %d exiting...\n", myid);
+	_exit(0);
 }