sched: implement various ways to end tasks and threads

arch/x86_64/asm-offset.c

@@ -28,4 +28,6 @@
 static void __used common(void)
 {
 	OFFSET(THREAD_sp, struct thread, tr_sp);
+	OFFSET(THREAD_fsbase, struct thread, tr_ml.tr_fsbase);
+	OFFSET(THREAD_gsbase, struct thread, tr_ml.tr_gsbase);
 }

arch/x86_64/include/arch/msr.h

@@ -3,6 +3,7 @@
 
 #include <stdint.h>
 
+#define MSR_FS_BASE        0xC0000100
 #define MSR_GS_BASE        0xC0000101
 #define MSR_KERNEL_GS_BASE 0xC0000102
 

arch/x86_64/include/kernel/machine/thread.h

@@ -3,6 +3,10 @@
 
 #include <kernel/sched.h>
 
+struct ml_thread {
+	virt_addr_t tr_gsbase, tr_fsbase;
+};
+
 struct ml_cpu_context;
 
 /* switch from one thread to another. the stack of the `to` thread must have
@@ -28,4 +32,15 @@ extern kern_status_t ml_thread_prepare_user_context(
 	const uintptr_t *args,
 	size_t nr_args);
 
+extern kern_status_t ml_thread_config_get(
+	struct thread *thread,
+	kern_config_key_t key,
+	void *out,
+	size_t max);
+extern kern_status_t ml_thread_config_set(
+	struct thread *thread,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len);
+
 #endif

arch/x86_64/panic.c

@@ -175,7 +175,7 @@ static bool read_stack_frame(
 	struct stack_frame *out)
 {
 	if (bp >= VM_PAGEMAP_BASE) {
-		*out = *(struct stack_frame *)out;
+		*out = *(struct stack_frame *)bp;
 		return true;
 	}
 

arch/x86_64/pmap.c

@@ -100,7 +100,7 @@ static void delete_ptab(phys_addr_t pt)
 		return;
 	}
 
-	pt &= ~VM_PAGE_MASK;
+	pt = ENTRY_TO_PTR(pt);
 	if (!pt) {
 		/* physical address of 0x0, nothing to delete */
 		return;
@@ -117,7 +117,7 @@ static void delete_pdir(phys_addr_t pd)
 		return;
 	}
 
-	pd &= ~0x1FFFFFULL;
+	pd &= ENTRY_TO_PTR(pd);
 	if (!pd) {
 		/* physical address of 0x0, nothing to delete */
 		return;
@@ -412,8 +412,46 @@ pmap_t pmap_create(void)
 	return pmap;
 }
 
+static void delete_pdpt(phys_addr_t pd)
+{
+	if (pd & PTE_PAGESIZE) {
+		/* this entry points to a hugepage, nothing to delete */
+		return;
+	}
+
+	pd &= ENTRY_TO_PTR(pd);
+	if (!pd) {
+		/* physical address of 0x0, nothing to delete */
+		return;
+	}
+
+	struct pdpt *pdpt = vm_phys_to_virt(ENTRY_TO_PTR(pd));
+	for (int i = 0; i < 512; i++) {
+		if (pdpt->p_pages[i] & PTE_PAGESIZE) {
+			/* this is a hugepage, there is nothing to delete */
+			continue;
+		}
+
+		if (!pdpt->p_entries[i]) {
+			continue;
+		}
+
+		delete_ptab(pdpt->p_entries[i]);
+	}
+
+	kfree(pdpt);
+}
+
 void pmap_destroy(pmap_t pmap)
 {
+	struct pml4t *pml4t = vm_phys_to_virt(ENTRY_TO_PTR(pmap));
+	for (unsigned int i = 0; i < 256; i++) {
+		if (pml4t->p_entries[i]) {
+			delete_pdpt(pml4t->p_entries[i]);
+		}
+	}
+
+	kfree(pml4t);
 }
 
 static void log_fault(virt_addr_t fault_addr, enum pmap_fault_flags flags)

arch/x86_64/thread.c

@@ -1,5 +1,7 @@
+#include <arch/msr.h>
 #include <kernel/machine/cpu.h>
 #include <kernel/machine/thread.h>
+#include <kernel/thread.h>
 
 #define MAX_REG_ARGS 6
 #define REG_ARG_0    rdi
@@ -77,3 +79,52 @@ extern kern_status_t ml_thread_prepare_user_context(
 
 	return KERN_OK;
 }
+
+kern_status_t ml_thread_config_get(
+	struct thread *thread,
+	kern_config_key_t key,
+	void *out,
+	size_t max)
+{
+	return KERN_OK;
+}
+
+kern_status_t ml_thread_config_set(
+	struct thread *thread,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len)
+{
+	switch (key) {
+	case THREAD_CFG_FSBASE:
+		if (len != sizeof(thread->tr_ml.tr_fsbase)) {
+			return KERN_INVALID_ARGUMENT;
+		}
+
+		thread->tr_ml.tr_fsbase = *(virt_addr_t *)ptr;
+		if (thread == current_thread()) {
+			wrmsr(MSR_FS_BASE, thread->tr_ml.tr_fsbase);
+		}
+
+		break;
+	case THREAD_CFG_GSBASE:
+		if (len != sizeof(thread->tr_ml.tr_gsbase)) {
+			return KERN_INVALID_ARGUMENT;
+		}
+
+		thread->tr_ml.tr_gsbase = *(virt_addr_t *)ptr;
+		if (thread == current_thread()) {
+			/* we're in the kernel right now, so the user and kernel
+			 * gs-base registers are swapped. when we return to
+			 * usermode, this value will be swapped back into
+			 * the user gs-base register */
+			wrmsr(MSR_KERNEL_GS_BASE, thread->tr_ml.tr_gsbase);
+		}
+
+		break;
+	default:
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	return KERN_OK;
+}

arch/x86_64/thread_switch.S

@@ -13,6 +13,22 @@ ml_thread_switch:
 	push %rax
 	push %rcx
 	push %rdx
+
+	// set fs-base
+	mov $0xC0000100, %rcx
+	movq THREAD_fsbase(%rsi), %rax
+	movq THREAD_fsbase(%rsi), %rdx
+	shr $32, %rdx
+	wrmsr
+
+	// set (kernel) gs-base (it will be swapped back into user-gs-base at
+	// the end of this function)
+	mov $0xC0000102, %rcx
+	movq THREAD_gsbase(%rsi), %rax
+	movq THREAD_gsbase(%rsi), %rdx
+	shr $32, %rdx
+	wrmsr
+
 	push %rbx
 	pushq $0
 	push %rbp

include/kernel/address-space.h

@@ -6,6 +6,7 @@
 #include <kernel/vm.h>
 
 #define ADDRESS_SPACE_COPY_ALL ((size_t)-1)
+#define ADDRESS_SPACE_F_
 
 struct address_space;
 struct vm_object;
@@ -158,6 +159,12 @@ extern kern_status_t address_space_memmove_v(
 	size_t bytes_to_move,
 	size_t *nr_bytes_moved);
 
+extern kern_status_t address_space_translate(
+	struct address_space *space,
+	virt_addr_t in,
+	phys_addr_t *out,
+	unsigned long *irq_flags);
+
 void address_space_dump(struct address_space *region);
 
 DEFINE_OBJECT_LOCK_FUNCTION(address_space, s_base)

include/kernel/futex.h

@@ -0,0 +1,34 @@
+#ifndef KERNEL_FUTEX_H_
+#define KERNEL_FUTEX_H_
+
+#include <kernel/btree.h>
+#include <kernel/wait.h>
+#include <mango/types.h>
+
+struct task;
+struct address_space;
+
+typedef uintptr_t futex_key_t;
+
+struct futex {
+	struct btree_node f_node;
+
+	futex_key_t f_key;
+	struct waitqueue f_waiters;
+};
+
+extern kern_status_t futex_init(void);
+extern kern_status_t futex_get(
+	kern_futex_t *futex,
+	futex_key_t *out,
+	unsigned int flags);
+extern kern_status_t futex_wait(
+	futex_key_t futex,
+	kern_futex_t new_val,
+	unsigned int flags);
+extern kern_status_t futex_wake(
+	futex_key_t futex,
+	size_t nwaiters,
+	unsigned int flags);
+
+#endif

include/kernel/syscall.h

@@ -56,8 +56,29 @@ extern kern_status_t sys_task_create_thread(
 extern kern_status_t sys_task_get_address_space(
 	kern_handle_t task,
 	kern_handle_t *out);
+extern kern_status_t sys_task_config_get(
+	kern_handle_t task,
+	kern_config_key_t key,
+	void *ptr,
+	size_t len);
+extern kern_status_t sys_task_config_set(
+	kern_handle_t task,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len);
 
 extern kern_status_t sys_thread_start(kern_handle_t thread);
+extern kern_status_t sys_thread_exit(void);
+extern kern_status_t sys_thread_config_get(
+	kern_handle_t thread,
+	kern_config_key_t key,
+	void *ptr,
+	size_t len);
+extern kern_status_t sys_thread_config_set(
+	kern_handle_t thread,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len);
 
 extern kern_status_t sys_vm_object_create(
 	const char *name,
@@ -197,6 +218,15 @@ extern kern_status_t sys_vm_controller_supply_pages(
 	off_t src_offset,
 	size_t count);
 
+extern kern_status_t sys_futex_wait(
+	kern_futex_t *futex,
+	kern_futex_t new_val,
+	unsigned int flags);
+extern kern_status_t sys_futex_wake(
+	kern_futex_t *futex,
+	unsigned int nr_waiters,
+	unsigned int flags);
+
 extern virt_addr_t syscall_get_function(unsigned int sysid);
 
 #endif

include/kernel/task.h

@@ -27,7 +27,7 @@ struct task {
 	struct address_space *t_address_space;
 	spin_lock_t t_handles_lock;
 	struct handle_table *t_handles;
-	struct btree b_channels;
+	struct btree t_channels, t_futex;
 
 	struct btree_node t_tasklist;
 	struct queue_entry t_child_entry;
@@ -48,6 +48,7 @@ static inline void task_unref(struct task *task)
 {
 	object_unref(&task->t_base);
 }
+extern void task_exit(int status);
 extern kern_status_t task_add_child(struct task *parent, struct task *child);
 extern kern_status_t task_add_channel(
 	struct task *task,

include/kernel/thread.h

@@ -1,6 +1,7 @@
 #ifndef KERNEL_THREAD_H_
 #define KERNEL_THREAD_H_
 
+#include <kernel/machine/thread.h>
 #include <kernel/msg.h>
 #include <kernel/object.h>
 #include <kernel/vm-controller.h>
@@ -22,7 +23,7 @@ enum thread_flags {
 };
 
 struct thread {
-	struct object thr_base;
+	struct object tr_base;
 
 	enum thread_state tr_state;
 	enum thread_flags tr_flags;
@@ -38,15 +39,13 @@ struct thread {
 	virt_addr_t tr_ip, tr_sp, tr_bp;
 	virt_addr_t tr_cpu_user_sp, tr_cpu_kernel_sp;
 
+	struct ml_thread tr_ml;
 	struct runqueue *tr_rq;
-	struct msg tr_msg;
-	struct page_request tr_page_req;
 
 	struct queue_entry tr_parent_entry;
 	struct queue_entry tr_rqentry;
 
 	struct vm_page *tr_kstack;
-	struct vm_object *tr_ustack;
 };
 
 extern struct thread *thread_alloc(void);
@@ -60,8 +59,24 @@ extern kern_status_t thread_init_user(
 	size_t nr_args);
 extern int thread_priority(struct thread *thr);
 extern void thread_awaken(struct thread *thr);
+extern void thread_exit(void);
+extern void thread_join(struct thread *thread, unsigned long *irq_flags);
+extern void thread_kill(struct thread *thread);
 extern void idle(void);
 extern struct thread *create_kernel_thread(void (*fn)(void));
 extern struct thread *create_idle_thread(void);
 
+extern kern_status_t thread_config_get(
+	struct thread *thread,
+	kern_config_key_t key,
+	void *out,
+	size_t max);
+extern kern_status_t thread_config_set(
+	struct thread *thread,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len);
+
+DEFINE_OBJECT_LOCK_FUNCTION(thread, tr_base)
+
 #endif

include/kernel/wait.h

@@ -40,6 +40,15 @@ extern void thread_wait_end_nosleep(
 	struct waitqueue *q);
 extern void wait_on_queue(struct waitqueue *q);
 extern void wakeup_queue(struct waitqueue *q);
+extern void wakeup_n(struct waitqueue *q, size_t n);
 extern void wakeup_one(struct waitqueue *q);
+static inline bool waitqueue_empty(struct waitqueue *wq)
+{
+	unsigned long flags;
+	spin_lock_irqsave(&wq->wq_lock, &flags);
+	bool result = queue_empty(&wq->wq_waiters);
+	spin_unlock_irqrestore(&wq->wq_lock, flags);
+	return result;
+}
 
 #endif

kernel/futex.c

@@ -0,0 +1,211 @@
+#include <kernel/address-space.h>
+#include <kernel/futex.h>
+#include <kernel/sched.h>
+#include <kernel/task.h>
+#include <mango/status.h>
+
+#define FUTEX_CREATE 0x40u
+
+static struct btree shared_futex_list = {0};
+static spin_lock_t shared_futex_list_lock = SPIN_LOCK_INIT;
+
+static struct vm_cache futex_cache = {
+	.c_name = "futex",
+	.c_obj_size = sizeof(struct futex),
+};
+
+BTREE_DEFINE_SIMPLE_INSERT(struct futex, f_node, f_key, put_futex)
+BTREE_DEFINE_SIMPLE_GET(struct futex, uintptr_t, f_node, f_key, get_futex)
+
+kern_status_t futex_init(void)
+{
+	vm_cache_init(&futex_cache);
+	return KERN_OK;
+}
+
+static kern_status_t get_data(
+	futex_key_t key,
+	unsigned int flags,
+	struct futex **out,
+	spin_lock_t **out_lock,
+	unsigned long *irq_flags)
+{
+	spin_lock_t *lock = NULL;
+	struct btree *futex_list = NULL;
+	if (flags & FUTEX_PRIVATE) {
+		struct task *self = current_task();
+		lock = &self->t_base.ob_lock;
+		futex_list = &self->t_futex;
+	} else if (flags & FUTEX_SHARED) {
+		lock = &shared_futex_list_lock;
+		futex_list = &shared_futex_list;
+	} else {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock_irqsave(lock, irq_flags);
+	struct futex *futex = get_futex(futex_list, key);
+
+	if (!futex && !(flags & FUTEX_CREATE)) {
+		spin_unlock_irqrestore(lock, *irq_flags);
+		return KERN_NO_ENTRY;
+	}
+
+	futex = vm_cache_alloc(&futex_cache, VM_NORMAL);
+	if (!futex) {
+		spin_unlock_irqrestore(lock, *irq_flags);
+		return KERN_NO_MEMORY;
+	}
+
+	futex->f_key = key;
+
+	put_futex(futex_list, futex);
+
+	*out = futex;
+	*out_lock = lock;
+	return KERN_OK;
+}
+
+static kern_status_t cleanup_data(struct futex *futex, unsigned int flags)
+{
+	struct btree *futex_list = NULL;
+	if (flags & FUTEX_PRIVATE) {
+		struct task *self = current_task();
+		futex_list = &self->t_futex;
+	} else if (flags & FUTEX_SHARED) {
+		futex_list = &shared_futex_list;
+	} else {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	btree_delete(futex_list, &futex->f_node);
+	vm_cache_free(&futex_cache, futex);
+
+	return KERN_OK;
+}
+
+static kern_status_t futex_get_shared(kern_futex_t *futex, futex_key_t *out)
+{
+	struct task *self = current_task();
+	struct address_space *space = self->t_address_space;
+
+	unsigned long flags;
+	address_space_lock_irqsave(space, &flags);
+	kern_status_t status = address_space_translate(
+		space,
+		(virt_addr_t)futex,
+		out,
+		&flags);
+	address_space_unlock_irqrestore(space, flags);
+	return status;
+}
+
+static kern_status_t futex_get_private(kern_futex_t *futex, futex_key_t *out)
+{
+	*out = (futex_key_t)futex;
+	return KERN_OK;
+}
+
+kern_status_t futex_get(
+	kern_futex_t *futex,
+	futex_key_t *out,
+	unsigned int flags)
+{
+	if (flags & FUTEX_PRIVATE) {
+		return futex_get_private(futex, out);
+	}
+
+	if (flags & FUTEX_SHARED) {
+		return futex_get_shared(futex, out);
+	}
+
+	return KERN_INVALID_ARGUMENT;
+}
+
+static kern_status_t futex_read(
+	struct futex *futex,
+	unsigned int flags,
+	kern_futex_t *out)
+{
+	if (flags & FUTEX_PRIVATE) {
+		virt_addr_t addr = futex->f_key;
+		*out = *(kern_futex_t *)addr;
+		return KERN_OK;
+	}
+
+	if (flags & FUTEX_SHARED) {
+		phys_addr_t paddr = futex->f_key;
+		virt_addr_t vaddr = (virt_addr_t)vm_phys_to_virt(paddr);
+		if (!vaddr) {
+			return KERN_MEMORY_FAULT;
+		}
+
+		*out = *(kern_futex_t *)vaddr;
+		return KERN_OK;
+	}
+
+	return KERN_INVALID_ARGUMENT;
+}
+
+kern_status_t futex_wait(
+	futex_key_t key,
+	kern_futex_t new_val,
+	unsigned int flags)
+{
+	spin_lock_t *lock = NULL;
+	unsigned long irq_flags = 0;
+	struct futex *futex = NULL;
+	kern_status_t status = get_data(key, flags, &futex, &lock, &irq_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	kern_futex_t current_val = 0;
+	status = futex_read(futex, flags, &current_val);
+	if (status != KERN_OK) {
+		spin_unlock_irqrestore(lock, irq_flags);
+		return status;
+	}
+
+	if (current_val != new_val) {
+		spin_unlock_irqrestore(lock, irq_flags);
+		return KERN_BAD_STATE;
+	}
+
+	struct wait_item waiter;
+	thread_wait_begin(&waiter, &futex->f_waiters);
+	spin_unlock_irqrestore(lock, irq_flags);
+
+	schedule(SCHED_NORMAL);
+
+	spin_lock_irqsave(lock, &irq_flags);
+	thread_wait_end(&waiter, &futex->f_waiters);
+
+	if (waitqueue_empty(&futex->f_waiters)) {
+		cleanup_data(futex, flags);
+	}
+
+	spin_unlock_irqrestore(lock, irq_flags);
+
+	return KERN_OK;
+}
+
+kern_status_t futex_wake(futex_key_t key, size_t nwaiters, unsigned int flags)
+{
+	spin_lock_t *lock = NULL;
+	unsigned long irq_flags = 0;
+	struct futex *futex = NULL;
+	kern_status_t status = get_data(key, flags, &futex, &lock, &irq_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	if (nwaiters == FUTEX_WAKE_ALL) {
+		wakeup_queue(&futex->f_waiters);
+	} else {
+		wakeup_n(&futex->f_waiters, nwaiters);
+	}
+
+	spin_unlock_irqrestore(lock, irq_flags);
+	return KERN_OK;
+}

kernel/handle.c

@@ -12,7 +12,7 @@
 #define RESERVED_HANDLES 64
 
 static struct vm_cache handle_table_cache = {
-	.c_name = "handle_table",
+	.c_name = "handle-table",
 	.c_obj_size = sizeof(struct handle_table),
 };
 
@@ -33,8 +33,48 @@ struct handle_table *handle_table_create(void)
 	return out;
 }
 
+static void do_handle_table_destroy_leaf(struct handle_table *tab)
+{
+	while (1) {
+		unsigned int index = bitmap_lowest_set(
+			tab->t_handles.t_handle_map,
+			HANDLES_PER_TABLE);
+		if (index == BITMAP_NPOS) {
+			break;
+		}
+
+		struct handle *child = &tab->t_handles.t_handle_list[index];
+		bitmap_clear(tab->t_subtables.t_subtable_map, index);
+		if (child->h_object) {
+			object_remove_handle(child->h_object);
+			child->h_object = NULL;
+		}
+	}
+}
+
+static void do_handle_table_destroy(
+	struct handle_table *tab,
+	unsigned int depth)
+{
+	if (depth == MAX_TABLE_DEPTH - 1) {
+		do_handle_table_destroy_leaf(tab);
+		return;
+	}
+
+	for (size_t i = 0; i < REFS_PER_TABLE; i++) {
+		struct handle_table *child
+			= tab->t_subtables.t_subtable_list[i];
+		if (child) {
+			do_handle_table_destroy(child, depth + 1);
+		}
+	}
+
+	vm_cache_free(&handle_table_cache, tab);
+}
+
 void handle_table_destroy(struct handle_table *tab)
 {
+	do_handle_table_destroy(tab, 0);
 }
 
 static kern_status_t decode_handle_indices(

kernel/port.c

@@ -88,25 +88,25 @@ kern_status_t port_send_msg(
 	}
 
 	struct thread *self = current_thread();
-	struct msg *msg = &self->tr_msg;
-	memset(msg, 0x0, sizeof *msg);
-	msg->msg_status = KMSG_WAIT_RECEIVE;
-	msg->msg_sender_thread = self;
-	msg->msg_sender_port = port;
-	memcpy(&msg->msg_req, in_msg, sizeof msg->msg_req);
-	memcpy(&msg->msg_resp, out_reply, sizeof msg->msg_req);
+	struct msg msg;
+	memset(&msg, 0x0, sizeof msg);
+	msg.msg_status = KMSG_WAIT_RECEIVE;
+	msg.msg_sender_thread = self;
+	msg.msg_sender_port = port;
+	memcpy(&msg.msg_req, in_msg, sizeof msg.msg_req);
+	memcpy(&msg.msg_resp, out_reply, sizeof msg.msg_req);
 
 	unsigned long flags;
 	channel_lock_irqsave(port->p_remote, &flags);
 	port->p_status = PORT_SEND_BLOCKED;
-	channel_enqueue_msg(port->p_remote, msg);
+	channel_enqueue_msg(port->p_remote, &msg);
 	channel_unlock_irqrestore(port->p_remote, flags);
 
-	wait_for_reply(msg, lock_flags);
+	wait_for_reply(&msg, lock_flags);
 
 	channel_lock_irqsave(port->p_remote, &flags);
-	btree_delete(&port->p_remote->c_msg, &msg->msg_node);
+	btree_delete(&port->p_remote->c_msg, &msg.msg_node);
 	channel_unlock_irqrestore(port->p_remote, flags);
 
-	return msg->msg_result;
+	return msg.msg_result;
 }

libmango/arch/x86_64/syscall.S

@@ -60,8 +60,13 @@ SYSCALL_GATE task_self SYS_TASK_SELF 0
 SYSCALL_GATE task_create SYS_TASK_CREATE 5
 SYSCALL_GATE task_create_thread SYS_TASK_CREATE_THREAD 6
 SYSCALL_GATE task_get_address_space SYS_TASK_GET_ADDRESS_SPACE 1
+SYSCALL_GATE task_config_get SYS_TASK_CONFIG_GET 4
+SYSCALL_GATE task_config_set SYS_TASK_CONFIG_SET 4
 
 SYSCALL_GATE thread_start SYS_THREAD_START 1
+SYSCALL_GATE thread_exit SYS_THREAD_EXIT 0
+SYSCALL_GATE thread_config_get SYS_THREAD_CONFIG_GET 4
+SYSCALL_GATE thread_config_set SYS_THREAD_CONFIG_SET 4
 
 SYSCALL_GATE vm_object_create SYS_VM_OBJECT_CREATE 5
 SYSCALL_GATE vm_object_read SYS_VM_OBJECT_READ 5
@@ -100,3 +105,6 @@ SYSCALL_GATE vm_controller_supply_pages SYS_VM_CONTROLLER_SUPPLY_PAGES 6
 
 SYSCALL_GATE kern_object_wait SYS_KERN_OBJECT_WAIT 2
 
+SYSCALL_GATE futex_wait SYS_FUTEX_WAIT 3
+SYSCALL_GATE futex_wake SYS_FUTEX_WAKE 3
+

libmango/include-user/mango/futex.h

@@ -0,0 +1,16 @@
+#ifndef MANGO_FUTEX_H_
+#define MANGO_FUTEX_H_
+
+#include <mango/status.h>
+#include <mango/types.h>
+
+extern kern_status_t futex_wait(
+	kern_futex_t *futex,
+	kern_futex_t new_val,
+	unsigned int flags);
+extern kern_status_t futex_wake(
+	kern_futex_t *futex,
+	unsigned int nr_waiters,
+	unsigned int flags);
+
+#endif

libmango/include-user/mango/task.h

@@ -23,7 +23,28 @@ extern kern_status_t task_create_thread(
 extern kern_status_t task_get_address_space(
 	kern_handle_t task,
 	kern_handle_t *out);
+extern kern_status_t task_config_get(
+	kern_handle_t task,
+	kern_config_key_t key,
+	void *ptr,
+	size_t len);
+extern kern_status_t task_config_set(
+	kern_handle_t task,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len);
 
 extern kern_status_t thread_start(kern_handle_t thread);
+extern kern_status_t thread_exit(void);
+extern kern_status_t thread_config_get(
+	kern_handle_t thread,
+	kern_config_key_t key,
+	void *ptr,
+	size_t len);
+extern kern_status_t thread_config_set(
+	kern_handle_t thread,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len);
 
 #endif

libmango/include/mango/signal.h

@@ -1,6 +1,8 @@
 #ifndef MANGO_SIGNAL_H_
 #define MANGO_SIGNAL_H_
 
+#define THREAD_SIGNAL_STOPPED                 0x01u
+
 #define CHANNEL_SIGNAL_MSG_RECEIVED           0x01u
 
 #define VM_CONTROLLER_SIGNAL_REQUEST_RECEIVED 0x01u

libmango/include/mango/syscall.h

@@ -13,7 +13,12 @@
 #define SYS_TASK_CREATE                 0x09u
 #define SYS_TASK_CREATE_THREAD          0x0Au
 #define SYS_TASK_GET_ADDRESS_SPACE      0x0Bu
+#define SYS_TASK_CONFIG_GET             0x2Au
+#define SYS_TASK_CONFIG_SET             0x2Bu
 #define SYS_THREAD_START                0x0Cu
+#define SYS_THREAD_EXIT                 0x2Eu
+#define SYS_THREAD_CONFIG_GET           0x2Cu
+#define SYS_THREAD_CONFIG_SET           0x2Du
 #define SYS_VM_OBJECT_CREATE            0x0Du
 #define SYS_VM_OBJECT_READ              0x0Eu
 #define SYS_VM_OBJECT_WRITE             0x0Fu
@@ -41,5 +46,7 @@
 #define SYS_VM_CONTROLLER_CREATE_OBJECT 0x25u
 #define SYS_VM_CONTROLLER_DETACH_OBJECT 0x26u
 #define SYS_VM_CONTROLLER_SUPPLY_PAGES  0x27u
+#define SYS_FUTEX_WAIT                  0x28u
+#define SYS_FUTEX_WAKE                  0x29u
 
 #endif

libmango/include/mango/types.h

@@ -16,8 +16,17 @@
 #define MAP_ADDRESS_INVALID     ((virt_addr_t)0)
 #define KERN_HANDLE_INVALID     ((kern_handle_t)0xFFFFFFFF)
 
-#define KERN_CFG_INVALID        0x00u
-#define KERN_CFG_PAGE_SIZE      0x01u
+/* config keys for use with kern_config_get/kern_config_set */
+#define KERN_CFG_INVALID        0x00000u
+#define KERN_CFG_PAGE_SIZE      0x00001u
+
+/* config keys for use with task_config_get/task_config_set */
+#define TASK_CFG_INVALID        0x00000u
+
+/* config keys for use with thread_config_get/thread_config_set */
+#define THREAD_CFG_INVALID      0x00000u
+#define THREAD_CFG_FSBASE       0x20001u
+#define THREAD_CFG_GSBASE       0x20002u
 
 /* maximum number of handles that can be sent in a single message */
 #define KERN_MSG_MAX_HANDLES    64
@@ -44,6 +53,13 @@
 #define PAGE_REQUEST_DIRTY      0x02u
 #define PAGE_REQUEST_DETACH     0x03u
 
+/* futex special values */
+#define FUTEX_WAKE_ALL          ((size_t)-1)
+
+/* futex flags */
+#define FUTEX_PRIVATE           0x01u
+#define FUTEX_SHARED            0x02u
+
 #define IOVEC(p, len)                                                          \
 	{                                                                      \
 		.io_base = (virt_addr_t)(p),                                   \
@@ -75,6 +91,7 @@ typedef uint32_t kern_handle_t;
 typedef uint32_t kern_config_key_t;
 typedef uint32_t vm_prot_t;
 typedef int64_t ssize_t;
+typedef uint32_t kern_futex_t;
 
 typedef unsigned short equeue_packet_type_t;
 

sched/core.c

@@ -123,8 +123,16 @@ void __schedule(enum sched_mode mode)
 
 	enum thread_state prev_state = READ_ONCE(prev->tr_state);
 
-	if ((mode == SCHED_IRQ || prev_state == THREAD_READY)
-	    && prev != rq->rq_idle) {
+	bool reschedule = false;
+	if (prev_state == THREAD_READY || mode == SCHED_IRQ) {
+		reschedule = true;
+	}
+
+	if (prev == rq->rq_idle || prev_state == THREAD_STOPPED) {
+		reschedule = false;
+	}
+
+	if (reschedule) {
 		rq_enqueue(rq, prev);
 	}
 

sched/task.c

@@ -222,13 +222,13 @@ kern_status_t task_add_channel(
 {
 	channel->c_id = id;
 
-	if (!task->b_channels.b_root) {
-		task->b_channels.b_root = &channel->c_node;
-		btree_insert_fixup(&task->b_channels, &channel->c_node);
+	if (!task->t_channels.b_root) {
+		task->t_channels.b_root = &channel->c_node;
+		btree_insert_fixup(&task->t_channels, &channel->c_node);
 		return KERN_OK;
 	}
 
-	struct btree_node *cur = task->b_channels.b_root;
+	struct btree_node *cur = task->t_channels.b_root;
 	while (1) {
 		struct channel *cur_node
 			= BTREE_CONTAINER(struct channel, c_node, cur);
@@ -255,7 +255,7 @@ kern_status_t task_add_channel(
 		cur = next;
 	}
 
-	btree_insert_fixup(&task->b_channels, &channel->c_node);
+	btree_insert_fixup(&task->t_channels, &channel->c_node);
 	return KERN_OK;
 }
 
@@ -268,7 +268,7 @@ BTREE_DEFINE_SIMPLE_GET(
 
 struct channel *task_get_channel(struct task *task, unsigned int id)
 {
-	return get_channel_with_id(&task->b_channels, id);
+	return get_channel_with_id(&task->t_channels, id);
 }
 
 struct task *task_from_tid(tid_t id)
@@ -280,6 +280,72 @@ struct task *task_from_tid(tid_t id)
 	return t;
 }
 
+void task_exit(int status)
+{
+	struct task *self = current_task();
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+	struct task *parent = self->t_parent;
+
+	if (parent) {
+		task_unlock_irqrestore(self, flags);
+		task_lock_irqsave(parent, &flags);
+		task_lock(self);
+		queue_delete(&parent->t_children, &self->t_child_entry);
+		task_unlock(parent);
+	}
+
+	struct thread *cur_thread = current_thread();
+
+	self->t_state = TASK_STOPPED;
+	cur_thread->tr_state = THREAD_STOPPED;
+
+	struct queue_entry *cur = queue_first(&self->t_threads);
+	while (cur) {
+		struct queue_entry *next = queue_next(cur);
+		struct thread *thread
+			= QUEUE_CONTAINER(struct thread, tr_parent_entry, cur);
+
+		if (thread == cur_thread) {
+			cur = next;
+			continue;
+		}
+
+		thread_lock(thread);
+		thread_kill(thread);
+		queue_delete(&self->t_threads, cur);
+		thread_unlock(thread);
+
+		cur = next;
+	}
+
+	object_unref(&self->t_address_space->s_base);
+	spin_lock_t *handles_lock = &self->t_handles_lock;
+	struct handle_table *handles = self->t_handles;
+	spin_lock(&self->t_handles_lock);
+
+	pmap_switch(get_kernel_pmap());
+	pmap_destroy(self->t_pmap);
+
+	task_unlock(self);
+	handle_table_destroy(handles);
+
+	printk("thread %s[%u.%u] killed",
+	       self->t_name,
+	       self->t_id,
+	       cur_thread->tr_id);
+	printk("task %s[%u] killed (%u, %u)",
+	       self->t_name,
+	       self->t_id,
+	       self->t_base.ob_refcount,
+	       self->t_base.ob_handles);
+	spin_unlock_irqrestore(handles_lock, flags);
+
+	while (1) {
+		schedule(SCHED_NORMAL);
+	}
+}
+
 kern_status_t task_open_handle(
 	struct task *task,
 	struct object *obj,

sched/thread.c

@@ -2,15 +2,17 @@
 #include <kernel/cpu.h>
 #include <kernel/machine/thread.h>
 #include <kernel/object.h>
+#include <kernel/printk.h>
 #include <kernel/task.h>
 #include <kernel/thread.h>
+#include <mango/signal.h>
 
-#define THREAD_CAST(p) OBJECT_C_CAST(struct thread, thr_base, &thread_type, p)
+#define THREAD_CAST(p) OBJECT_C_CAST(struct thread, tr_base, &thread_type, p)
 
 static struct object_type thread_type = {
 	.ob_name = "thread",
 	.ob_size = sizeof(struct thread),
-	.ob_header_offset = offsetof(struct thread, thr_base),
+	.ob_header_offset = offsetof(struct thread, tr_base),
 };
 
 kern_status_t thread_object_type_init(void)
@@ -63,9 +65,6 @@ kern_status_t thread_init_user(
 	const uintptr_t *args,
 	size_t nr_args)
 {
-	thr->tr_id = thr->tr_parent->t_next_thread_id++;
-
-	thr->tr_prio = PRIO_NORMAL;
 	thr->tr_state = THREAD_READY;
 	thr->tr_quantum_target = default_quantum();
 
@@ -142,6 +141,55 @@ void thread_awaken(struct thread *thr)
 	rq_unlock(rq, flags);
 }
 
+void thread_exit(void)
+{
+	struct thread *self = current_thread();
+	unsigned long flags;
+	thread_lock_irqsave(self, &flags);
+	self->tr_state = THREAD_STOPPED;
+	object_assert_signal(&self->tr_base, THREAD_SIGNAL_STOPPED);
+	printk("thread %s[%u.%u] exited",
+	       self->tr_parent->t_name,
+	       self->tr_parent->t_id,
+	       self->tr_id);
+	thread_unlock_irqrestore(self, flags);
+
+	while (1) {
+		schedule(SCHED_NORMAL);
+	}
+}
+
+void thread_join(struct thread *thread, unsigned long *irq_flags)
+{
+	while (1) {
+		if (thread->tr_state == THREAD_STOPPED) {
+			break;
+		}
+
+		object_wait_signal(
+			&thread->tr_base,
+			THREAD_SIGNAL_STOPPED,
+			irq_flags);
+	}
+}
+
+void thread_kill(struct thread *thread)
+{
+	thread->tr_state = THREAD_STOPPED;
+	if (thread->tr_rq) {
+		unsigned long flags;
+		rq_lock(thread->tr_rq, &flags);
+		rq_remove_thread(thread->tr_rq, thread);
+		rq_unlock(thread->tr_rq, flags);
+	}
+
+	object_assert_signal(&thread->tr_base, THREAD_SIGNAL_STOPPED);
+	printk("thread %s[%u.%u] killed",
+	       thread->tr_parent->t_name,
+	       thread->tr_parent->t_id,
+	       thread->tr_id);
+}
+
 struct thread *create_kernel_thread(void (*fn)(void))
 {
 	struct task *kernel = kernel_task();
@@ -185,3 +233,31 @@ struct thread *create_idle_thread(void)
 
 	return thr;
 }
+
+kern_status_t thread_config_get(
+	struct thread *thread,
+	kern_config_key_t key,
+	void *out,
+	size_t max)
+{
+	switch (key) {
+	default:
+		break;
+	}
+
+	return ml_thread_config_get(thread, key, out, max);
+}
+
+kern_status_t thread_config_set(
+	struct thread *thread,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len)
+{
+	switch (key) {
+	default:
+		break;
+	}
+
+	return ml_thread_config_set(thread, key, ptr, len);
+}

sched/wait.c

@@ -69,6 +69,24 @@ void wakeup_queue(struct waitqueue *q)
 	spin_unlock_irqrestore(&q->wq_lock, flags);
 }
 
+void wakeup_n(struct waitqueue *q, size_t nr_waiters)
+{
+	unsigned long flags;
+	spin_lock_irqsave(&q->wq_lock, &flags);
+	struct queue_entry *ent = queue_pop_front(&q->wq_waiters);
+	while (ent && nr_waiters > 0) {
+		struct wait_item *waiter
+			= QUEUE_CONTAINER(struct wait_item, w_entry, ent);
+		struct thread *thr = waiter->w_thread;
+		thread_awaken(thr);
+
+		ent = queue_pop_front(&q->wq_waiters);
+		nr_waiters--;
+	}
+
+	spin_unlock_irqrestore(&q->wq_lock, flags);
+}
+
 void wakeup_one(struct waitqueue *q)
 {
 	unsigned long flags;

syscall/dispatch.c

@@ -11,6 +11,9 @@ static const virt_addr_t syscall_table[] = {
 	SYSCALL_TABLE_ENTRY(TASK_CREATE_THREAD, task_create_thread),
 	SYSCALL_TABLE_ENTRY(TASK_GET_ADDRESS_SPACE, task_get_address_space),
 	SYSCALL_TABLE_ENTRY(THREAD_START, thread_start),
+	SYSCALL_TABLE_ENTRY(THREAD_EXIT, thread_exit),
+	SYSCALL_TABLE_ENTRY(THREAD_CONFIG_GET, thread_config_get),
+	SYSCALL_TABLE_ENTRY(THREAD_CONFIG_SET, thread_config_set),
 	SYSCALL_TABLE_ENTRY(VM_OBJECT_CREATE, vm_object_create),
 	SYSCALL_TABLE_ENTRY(VM_OBJECT_READ, vm_object_read),
 	SYSCALL_TABLE_ENTRY(VM_OBJECT_WRITE, vm_object_write),
@@ -48,6 +51,8 @@ static const virt_addr_t syscall_table[] = {
 		VM_CONTROLLER_SUPPLY_PAGES,
 		vm_controller_supply_pages),
 	SYSCALL_TABLE_ENTRY(KERN_OBJECT_WAIT, kern_object_wait),
+	SYSCALL_TABLE_ENTRY(FUTEX_WAIT, futex_wait),
+	SYSCALL_TABLE_ENTRY(FUTEX_WAKE, futex_wake),
 };
 static const size_t syscall_table_count
 	= sizeof syscall_table / sizeof syscall_table[0];

syscall/futex.c

@@ -0,0 +1,37 @@
+#include <kernel/futex.h>
+#include <kernel/sched.h>
+#include <kernel/syscall.h>
+#include <kernel/task.h>
+
+kern_status_t sys_futex_wait(
+	kern_futex_t *futex,
+	kern_futex_t new_val,
+	unsigned int flags)
+{
+	struct task *self = current_task();
+	if (!validate_access_r(self, futex, sizeof *futex)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	futex_key_t key;
+	kern_status_t status = futex_get(futex, &key, flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	return futex_wait(key, new_val, flags);
+}
+
+kern_status_t sys_futex_wake(
+	kern_futex_t *futex,
+	unsigned int nr_waiters,
+	unsigned int flags)
+{
+	futex_key_t key;
+	kern_status_t status = futex_get(futex, &key, flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	return futex_wake(key, nr_waiters, flags);
+}

syscall/log.c

@@ -1,10 +1,12 @@
 #include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/task.h>
+#include <kernel/thread.h>
 
 kern_status_t sys_kern_log(const char *s)
 {
 	struct task *task = current_task();
-	printk("%s[%d]: %s", task->t_name, task->t_id, s);
+	struct thread *thread = current_thread();
+	printk("%s[%d.%d]: %s", task->t_name, task->t_id, thread->tr_id, s);
 	return KERN_OK;
 }

syscall/task.c

@@ -10,11 +10,8 @@ extern kern_status_t sys_task_exit(int status)
 {
 	struct task *self = current_task();
 	printk("%s[%d]: task_exit(%d)", self->t_name, self->t_id, status);
-	while (1) {
-		milli_sleep(5000);
-	}
-
-	return KERN_UNIMPLEMENTED;
+	task_exit(status);
+	return KERN_FATAL_ERROR;
 }
 
 kern_status_t sys_task_self(kern_handle_t *out)
@@ -179,6 +176,7 @@ kern_status_t sys_task_create_thread(
 		&target_handle,
 		&out_handle);
 	if (status != KERN_OK) {
+		object_unref(target_obj);
 		task_unlock_irqrestore(self, flags);
 		return status;
 	}
@@ -198,10 +196,11 @@ kern_status_t sys_task_create_thread(
 	}
 
 	thread_init_user(thread, ip, sp, args, nr_args);
-	target_handle->h_object = &thread->thr_base;
-	object_add_handle(&thread->thr_base);
+	target_handle->h_object = &thread->tr_base;
+	object_add_handle(&thread->tr_base);
 
 	task_unlock_irqrestore(target, flags);
+	object_unref(target_obj);
 
 	*out_thread = out_handle;
 	return KERN_OK;
@@ -287,3 +286,82 @@ kern_status_t sys_thread_start(kern_handle_t thread_handle)
 
 	return KERN_OK;
 }
+
+kern_status_t sys_thread_exit(void)
+{
+	thread_exit();
+	/* unreachable */
+	return KERN_FATAL_ERROR;
+}
+
+kern_status_t sys_thread_config_get(
+	kern_handle_t thread_handle,
+	kern_config_key_t key,
+	void *ptr,
+	size_t len)
+{
+	unsigned long flags;
+	struct task *self = current_task();
+
+	if (!validate_access_w(self, ptr, len)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	struct object *thread_obj;
+	handle_flags_t thread_flags;
+	task_lock_irqsave(self, &flags);
+	kern_status_t status = task_resolve_handle(
+		self,
+		thread_handle,
+		&thread_obj,
+		&thread_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		return status;
+	}
+
+	struct thread *thread = thread_cast(thread_obj);
+	task_unlock_irqrestore(self, flags);
+
+	status = thread_config_get(thread, key, ptr, len);
+
+	object_unref(thread_obj);
+
+	return status;
+}
+
+kern_status_t sys_thread_config_set(
+	kern_handle_t thread_handle,
+	kern_config_key_t key,
+	const void *ptr,
+	size_t len)
+{
+	unsigned long flags;
+	struct task *self = current_task();
+
+	if (!validate_access_w(self, ptr, len)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	struct object *thread_obj;
+	handle_flags_t thread_flags;
+	task_lock_irqsave(self, &flags);
+	kern_status_t status = task_resolve_handle(
+		self,
+		thread_handle,
+		&thread_obj,
+		&thread_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		return status;
+	}
+
+	struct thread *thread = thread_cast(thread_obj);
+	task_unlock_irqrestore(self, flags);
+
+	status = thread_config_set(thread, key, ptr, len);
+
+	object_unref(thread_obj);
+
+	return status;
+}

syscall/vm-object.c

@@ -29,12 +29,9 @@ kern_status_t sys_vm_object_create(
 
 	kern_status_t status
 		= task_open_handle(self, &obj->vo_base, 0, out_handle);
-	if (status != KERN_OK) {
-		object_unref(&obj->vo_base);
-		return status;
-	}
+	object_unref(&obj->vo_base);
 
-	return KERN_OK;
+	return status;
 }
 
 kern_status_t sys_vm_object_read(

vm/address-space.c

@@ -34,6 +34,7 @@ struct vm_iterator {
 /* iterates over the areas in an address space */
 struct area_iterator {
 	struct address_space *it_root;
+	struct btree *it_list;
 	struct vm_area *it_area;
 	virt_addr_t it_search_base, it_search_limit;
 	virt_addr_t it_base, it_limit;
@@ -44,12 +45,15 @@ enum search_direction {
 	SEARCH_RIGHT,
 };
 
-static kern_status_t address_space_object_destroy(struct object *obj);
+static kern_status_t address_space_cleanup(struct object *obj, struct queue *q);
 
 static struct object_type address_space_type = {
 	.ob_name = "address-space",
 	.ob_size = sizeof(struct address_space),
 	.ob_header_offset = offsetof(struct address_space, s_base),
+	.ob_ops = {
+		.destroy = address_space_cleanup,
+	},
 };
 
 static struct vm_cache vm_area_cache = {
@@ -59,6 +63,10 @@ static struct vm_cache vm_area_cache = {
 
 /*** INTERNAL UTILITY FUNCTION ************************************************/
 
+static kern_status_t delete_area(
+	struct vm_area *mapping,
+	struct address_space *root);
+
 /* this function must be called with `parent` locked */
 static void put_entry(struct btree *tree, struct vm_area *child)
 {
@@ -105,7 +113,7 @@ static void put_entry(struct btree *tree, struct vm_area *child)
 }
 
 static struct vm_area *get_entry(
-	struct address_space *region,
+	struct btree *list,
 	virt_addr_t address,
 	enum get_entry_flags flags)
 {
@@ -114,7 +122,7 @@ static struct vm_area *get_entry(
 	/* `x` must be to the right of `y` */
 #define RIGHT_DIFF(x, y) ((y) ? ((y)->vma_limit - (x)) : ((size_t)-1))
 
-	struct btree_node *cur = region->s_mappings.b_root;
+	struct btree_node *cur = list->b_root;
 	if (!cur) {
 		return NULL;
 	}
@@ -345,7 +353,7 @@ static void vm_iterator_begin(
 	it->it_region = region;
 	it->it_prot = prot;
 
-	it->it_mapping = get_entry(region, base, GET_ENTRY_EXACT);
+	it->it_mapping = get_entry(&region->s_mappings, base, GET_ENTRY_EXACT);
 	if (!it->it_mapping) {
 		return;
 	}
@@ -409,8 +417,10 @@ static kern_status_t vm_iterator_seek(struct vm_iterator *it, size_t nr_bytes)
 
 	it->it_base += nr_bytes;
 
-	struct vm_area *next_mapping
-		= get_entry(it->it_region, it->it_base, GET_ENTRY_EXACT);
+	struct vm_area *next_mapping = get_entry(
+		&it->it_region->s_mappings,
+		it->it_base,
+		GET_ENTRY_EXACT);
 	if (!next_mapping) {
 		it->it_buf = NULL;
 		it->it_max = 0;
@@ -478,12 +488,14 @@ static void vm_iterator_finish(struct vm_iterator *it)
 static void area_iterator_begin(
 	struct area_iterator *it,
 	struct address_space *space,
+	struct btree *area_list,
 	virt_addr_t base,
 	virt_addr_t limit)
 {
 	memset(it, 0x0, sizeof *it);
 
-	struct vm_area *area = get_entry(space, base, GET_ENTRY_CLOSEST_RIGHT);
+	struct vm_area *area
+		= get_entry(area_list, base, GET_ENTRY_CLOSEST_RIGHT);
 	if (!area) {
 		return;
 	}
@@ -552,8 +564,46 @@ end:
 	return KERN_NO_ENTRY;
 }
 
-static void area_iterator_erase(struct area_iterator *it)
+static kern_status_t area_iterator_erase(struct area_iterator *it)
 {
+	if (!it->it_root || !it->it_area) {
+		return KERN_NO_ENTRY;
+	}
+
+	struct btree_node *next = btree_next(&it->it_area->vma_node);
+	btree_delete(it->it_list, &it->it_area->vma_node);
+	vm_cache_free(&vm_area_cache, it->it_area);
+
+	if (!next) {
+		goto end;
+	}
+
+	struct vm_area *area = BTREE_CONTAINER(struct vm_area, vma_node, next);
+	if (!area) {
+		goto end;
+	}
+
+	if (area->vma_base > it->it_search_limit) {
+		goto end;
+	}
+
+	it->it_area = area;
+	it->it_base = area->vma_base;
+	it->it_limit = area->vma_base;
+
+	if (it->it_base < it->it_search_base) {
+		it->it_base = it->it_search_base;
+	}
+
+	if (it->it_limit > it->it_search_limit) {
+		it->it_limit = it->it_search_limit;
+	}
+
+	return KERN_OK;
+
+end:
+	memset(it, 0x0, sizeof *it);
+	return KERN_NO_ENTRY;
 }
 
 /*** PUBLIC API ***************************************************************/
@@ -602,6 +652,35 @@ kern_status_t address_space_create(
 	return KERN_OK;
 }
 
+static void area_unmap(struct vm_area *area)
+{
+	pmap_t pmap = area->vma_space->s_pmap;
+	virt_addr_t base = area->vma_base;
+	virt_addr_t limit = area->vma_limit;
+	for (virt_addr_t i = base; i < limit; i += VM_PAGE_SIZE) {
+		pmap_remove(pmap, i);
+	}
+}
+
+static kern_status_t address_space_cleanup(struct object *obj, struct queue *q)
+{
+	struct address_space *space = ADDRESS_SPACE_CAST(obj);
+	struct btree_node *cur = btree_first(&space->s_mappings);
+	while (cur) {
+		struct btree_node *next = btree_next(cur);
+		struct vm_area *area
+			= BTREE_CONTAINER(struct vm_area, vma_node, cur);
+		btree_delete(&space->s_mappings, cur);
+
+		delete_area(area, space);
+		vm_cache_free(&vm_area_cache, area);
+
+		cur = next;
+	}
+
+	return KERN_OK;
+}
+
 kern_status_t address_space_map(
 	struct address_space *root,
 	virt_addr_t map_address,
@@ -850,7 +929,7 @@ static kern_status_t delete_area(
 		&mapping->vma_object->vo_mappings,
 		&mapping->vma_object_entry);
 	mapping->vma_object = NULL;
-	vm_object_unlock_irqrestore(mapping->vma_object, flags);
+	vm_object_unlock_irqrestore(object, flags);
 	object_unref(&object->vo_base);
 
 	/* don't actually delete the mapping yet. that will be done by
@@ -882,7 +961,12 @@ kern_status_t address_space_unmap(
 	virt_addr_t unmap_limit = unmap_base + unmap_length - 1;
 	tracek("unmapping %zx-%zx", unmap_base, unmap_limit);
 
-	area_iterator_begin(&it, region, unmap_base, unmap_limit);
+	area_iterator_begin(
+		&it,
+		region,
+		&region->s_mappings,
+		unmap_base,
+		unmap_limit);
 	while (it.it_area) {
 		struct vm_area *area = it.it_area;
 		virt_addr_t area_base = area->vma_base;
@@ -1015,7 +1099,12 @@ kern_status_t address_space_release(
 	virt_addr_t release_limit = release_base + release_length - 1;
 	tracek("unreserving %zx-%zx", release_base, release_limit);
 
-	area_iterator_begin(&it, space, release_base, release_limit);
+	area_iterator_begin(
+		&it,
+		space,
+		&space->s_reserved,
+		release_base,
+		release_limit);
 	while (it.it_area) {
 		struct vm_area *area = it.it_area;
 		virt_addr_t area_base = area->vma_base;
@@ -1102,9 +1191,9 @@ bool address_space_validate_access(
 		limit -= 1;
 	}
 
-	/* TODO improve this to not require a per-page loop */
 	for (virt_addr_t i = base; i < limit;) {
-		struct vm_area *area = get_entry(region, i, GET_ENTRY_EXACT);
+		struct vm_area *area
+			= get_entry(&region->s_mappings, i, GET_ENTRY_EXACT);
 		if (!area) {
 			return false;
 		}
@@ -1176,7 +1265,8 @@ kern_status_t address_space_demand_map(
 	unsigned long irq_flags;
 	address_space_lock_irqsave(region, &irq_flags);
 
-	struct vm_area *area = get_entry(region, addr, GET_ENTRY_EXACT);
+	struct vm_area *area
+		= get_entry(&region->s_mappings, addr, GET_ENTRY_EXACT);
 	if (!area || !area->vma_object) {
 		address_space_unlock_irqrestore(region, irq_flags);
 		return KERN_NO_ENTRY;
@@ -1413,6 +1503,43 @@ extern kern_status_t address_space_memmove_v(
 	return KERN_OK;
 }
 
+kern_status_t address_space_translate(
+	struct address_space *space,
+	virt_addr_t in,
+	phys_addr_t *out,
+	unsigned long *irq_flags)
+{
+	if (in >= VM_KERNEL_BASE) {
+		return vm_virt_to_phys((const void *)in);
+	}
+
+	struct vm_area *area
+		= get_entry(&space->s_mappings, in, GET_ENTRY_EXACT);
+	if (!area || !area->vma_object) {
+		return KERN_NO_ENTRY;
+	}
+
+	off_t offset = in - area->vma_base + area->vma_object_offset;
+	struct vm_object *vmo = area->vma_object;
+	vm_object_lock(vmo);
+	address_space_unlock(space);
+	struct vm_page *pg = vm_object_get_page(
+		vmo,
+		offset,
+		VMO_ALLOCATE_MISSING_PAGE | VMO_REQUEST_MISSING_PAGE,
+		irq_flags);
+	if (!pg) {
+		return KERN_NO_ENTRY;
+	}
+
+	phys_addr_t paddr = vm_page_get_paddr(pg);
+	paddr += (in & VM_PAGE_MASK);
+	vm_object_unlock(vmo);
+	address_space_lock(space);
+
+	return paddr;
+}
+
 #ifdef TRACE
 void address_space_dump(struct address_space *region)
 {

vm/vm-object.c

@@ -1,4 +1,5 @@
 #include <kernel/address-space.h>
+#include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/util.h>
 #include <kernel/vm-controller.h>
@@ -14,10 +15,30 @@
 		(p) += VM_PAGE_SIZE;                                           \
 	}
 
+static kern_status_t vm_object_cleanup(struct object *obj, struct queue *q)
+{
+	struct vm_object *vmo = vm_object_cast(obj);
+	struct btree_node *cur = btree_first(&vmo->vo_pages);
+	while (cur) {
+		struct vm_page *pg
+			= BTREE_CONTAINER(struct vm_page, p_bnode, cur);
+		struct btree_node *next = btree_next(cur);
+		btree_delete(&vmo->vo_pages, cur);
+
+		vm_page_free(pg);
+		cur = next;
+	}
+
+	return KERN_OK;
+}
+
 static struct object_type vm_object_type = {
 	.ob_name = "vm-object",
 	.ob_size = sizeof(struct vm_object),
 	.ob_header_offset = offsetof(struct vm_object, vo_base),
+	.ob_ops = {
+		.destroy = vm_object_cleanup,
+	},
 };
 
 struct object_iterator {