vm: controller: add an auto-detach flag for vm-objects

include/kernel/atomic.h

@@ -0,0 +1,66 @@
+#ifndef KERNEL_ATOMIC_H_
+#define KERNEL_ATOMIC_H_
+
+#include <stdbool.h>
+#include <stdint.h>
+
+typedef int64_t atomic_t;
+
+/* load and return the value pointed to by `v` */
+static inline atomic_t atomic_load(atomic_t *v)
+{
+	return __atomic_load_n(v, __ATOMIC_ACQUIRE);
+}
+/* store the value `v` to the pointer `dest` */
+static inline void atomic_store(atomic_t *dest, atomic_t v)
+{
+	__atomic_store_n(dest, v, __ATOMIC_ACQUIRE);
+}
+/* store the value `v` to the pointer `dest`, and return the value previously
+ * stored at `dest` */
+static inline atomic_t atomic_exchange(atomic_t *dest, atomic_t v)
+{
+	return __atomic_exchange_n(dest, v, __ATOMIC_ACQUIRE);
+}
+/* compare the contents of `ptr` to the contents of `expected`.
+ * if they match, store the value `desired` to the pointer `ptr` and return
+ * true. if the do NOT match, store the value `*ptr` to the pointer `desired`
+ * and return false.
+ */
+static inline bool atomic_compare_exchange(
+	atomic_t *ptr,
+	atomic_t *expected,
+	atomic_t desired)
+{
+	return __atomic_compare_exchange_n(
+		ptr,
+		expected,
+		desired,
+		false,
+		__ATOMIC_ACQUIRE,
+		__ATOMIC_ACQUIRE);
+}
+
+/* perform the operation *ptr += val, and return the result */
+static inline atomic_t atomic_add_fetch(atomic_t *ptr, atomic_t val)
+{
+	return __atomic_add_fetch(ptr, val, __ATOMIC_ACQUIRE);
+}
+/* perform the operation *ptr -= val, and return the result */
+static inline atomic_t atomic_sub_fetch(atomic_t *ptr, atomic_t val)
+{
+	return __atomic_sub_fetch(ptr, val, __ATOMIC_ACQUIRE);
+}
+
+/* perform the operation *ptr += val, and return the previous value of *ptr */
+static inline atomic_t atomic_fetch_add(atomic_t *ptr, atomic_t val)
+{
+	return __atomic_fetch_add(ptr, val, __ATOMIC_ACQUIRE);
+}
+/* perform the operation *ptr -= val, and return the previous value of *ptr */
+static inline atomic_t atomic_fetch_sub(atomic_t *ptr, atomic_t val)
+{
+	return __atomic_fetch_sub(ptr, val, __ATOMIC_ACQUIRE);
+}
+
+#endif

include/kernel/msg.h

@@ -13,6 +13,7 @@ enum kmsg_status {
 	KMSG_WAIT_RECEIVE,
 	KMSG_WAIT_REPLY,
 	KMSG_REPLY_SENT,
+	KMSG_ASYNC,
 };
 
 struct msg {
@@ -20,10 +21,26 @@ struct msg {
 	enum kmsg_status msg_status;
 	struct btree_node msg_node;
 	msgid_t msg_id;
-	kern_status_t msg_result;
 	struct port *msg_sender_port;
 	struct thread *msg_sender_thread;
-	kern_msg_t msg_req, msg_resp;
+	kern_status_t msg_result;
+	kern_msg_type_t msg_type;
+
+	union {
+		/* msg_type = KERN_MSG_TYPE_DATA */
+		struct {
+			kern_msg_t msg_req, msg_resp;
+		};
+
+		/* msg_type = KERN_MSG_TYPE_EVENT */
+		struct {
+			kern_msg_event_type_t msg_event;
+		};
+	};
 };
 
+extern void msg_init(void);
+extern struct msg *msg_alloc(void);
+extern void msg_free(struct msg *msg);
+
 #endif

include/kernel/object.h

@@ -1,6 +1,7 @@
 #ifndef KERNEL_OBJECT_H_
 #define KERNEL_OBJECT_H_
 
+#include <kernel/atomic.h>
 #include <kernel/flags.h>
 #include <kernel/locks.h>
 #include <kernel/vm.h>
@@ -79,10 +80,7 @@ enum object_type_flags {
 };
 
 struct object_ops {
-	kern_status_t (*destroy)(struct object *obj, struct queue *q);
-	kern_status_t (*destroy_recurse)(
-		struct queue_entry *entry,
-		struct object **out);
+	kern_status_t (*destroy)(struct object *obj);
 };
 
 struct object_type {
@@ -101,8 +99,7 @@ struct object {
 	struct object_type *ob_type;
 	spin_lock_t ob_lock;
 	uint32_t ob_signals;
-	unsigned int ob_refcount;
-	unsigned int ob_handles;
+	atomic_t ob_refcount;
 	struct queue_entry ob_list;
 	struct waitqueue ob_wq;
 } __aligned(sizeof(long));
@@ -114,8 +111,6 @@ extern kern_status_t object_type_unregister(struct object_type *p);
 extern struct object *object_create(struct object_type *type);
 extern struct object *object_ref(struct object *obj);
 extern void object_unref(struct object *obj);
-extern void object_add_handle(struct object *obj);
-extern void object_remove_handle(struct object *obj);
 extern void object_lock(struct object *obj);
 extern void object_unlock(struct object *obj);
 extern void object_lock_irqsave(struct object *obj, unsigned long *flags);

include/kernel/vm-controller.h

@@ -13,13 +13,11 @@ enum page_request_status {
 	PAGE_REQUEST_PENDING = 0,
 	PAGE_REQUEST_IN_PROGRESS,
 	PAGE_REQUEST_COMPLETE,
+	PAGE_REQUEST_ASYNC,
 };
 
 struct vm_controller {
 	struct object vc_base;
-	/* tree of struct vm_objects bound to this controller, keyed with the
-	 * equeue_key_t specified when the object(s) were created. */
-	struct btree vc_objects;
 	/* tree of pending page requests */
 	struct btree vc_requests;
 	/* the equeue to send async page requests to */
@@ -36,7 +34,7 @@ struct page_request {
 	enum page_request_status req_status;
 	kern_status_t req_result;
 	spin_lock_t req_lock;
-	struct vm_object *req_object;
+	equeue_key_t req_object;
 	struct thread *req_sender;
 	struct btree_node req_node;
 	off_t req_offset;

include/kernel/vm-object.h

@@ -14,19 +14,25 @@ enum vm_object_flags {
 	VMO_IN_PLACE = 0x01u,
 	/* this vm-object is/was attached to a vm-controller */
 	VMO_CONTROLLER = 0x02u,
+	/* when a vm-object is attached to a controller, and the ref-count of
+	 * the object falls to one (i.e. the last reference is the handle to
+	 * the object held by the server that created it), the object will
+	 * be detached, allowing the server to close the last handle to the
+	 * object and dispose of it. */
+	VMO_AUTO_DETACH = 0x04u,
 
 	/* these flags are for use with vm_object_get_page */
 	/**************************************************/
 
 	/* if the relevant page hasn't been allocated yet, it will be allocated
 	 * and returned. if this flag isn't specified, NULL will be returned. */
-	VMO_ALLOCATE_MISSING_PAGE = 0x04u,
+	VMO_ALLOCATE_MISSING_PAGE = 0x08u,
 	/* if the vm-object is attached to a vm-controller, and the relevant
 	 * page is uncommitted, send a request to the vm-controller to provide
 	 * the missing page. will result in the vm-object being unlocked and
 	 * the current thread sleeping until the request is fulfilled. the
 	 * vm-object will be re-locked before the function returns. */
-	VMO_REQUEST_MISSING_PAGE = 0x08u,
+	VMO_REQUEST_MISSING_PAGE = 0x10u,
 };
 
 struct vm_object {

init/main.c

@@ -78,6 +78,7 @@ void kernel_init(uintptr_t arg)
 
 	port_type_init();
 	channel_type_init();
+	msg_init();
 
 	struct boot_module bsp_image = {0};
 	bsp_get_location(&bsp_image);

kernel/channel.c

@@ -98,11 +98,18 @@ static struct msg *get_next_msg(
 	while (cur) {
 		struct msg *msg = BTREE_CONTAINER(struct msg, msg_node, cur);
 		spin_lock_irqsave(&msg->msg_lock, lock_flags);
-		if (msg->msg_status == KMSG_WAIT_RECEIVE) {
+		switch (msg->msg_status) {
+		case KMSG_WAIT_RECEIVE:
 			msg->msg_status = KMSG_WAIT_REPLY;
 			msg->msg_sender_port->p_status = PORT_REPLY_BLOCKED;
 			channel->c_msg_waiting--;
 			return msg;
+		case KMSG_ASYNC:
+			btree_delete(&channel->c_msg, &msg->msg_node);
+			channel->c_msg_waiting--;
+			return msg;
+		default:
+			break;
 		}
 
 		spin_unlock_irqrestore(&msg->msg_lock, *lock_flags);
@@ -146,24 +153,22 @@ extern kern_status_t channel_recv_msg(
 			&channel->c_base,
 			CHANNEL_SIGNAL_MSG_RECEIVED);
 	}
-#if 0
-	wait_item_init(&waiter, self);
-	for (;;) {
-		thread_wait_begin(&waiter, &channel->c_wq);
-		msg = get_next_msg(channel, &msg_lock_flags);
-		if (msg) {
-			break;
-		}
-
-		object_unlock_irqrestore(&channel->c_base, *irq_flags);
-		schedule(SCHED_NORMAL);
-		object_lock_irqsave(&channel->c_base, irq_flags);
-	}
-	thread_wait_end(&waiter, &channel->c_wq);
-#endif
 
 	/* msg is now set to the next message to process */
 
+	if (msg->msg_type != KERN_MSG_TYPE_DATA) {
+		/* event messages as asynchronous */
+		out_msg->msg_id = msg->msg_id;
+		out_msg->msg_type = msg->msg_type;
+		out_msg->msg_event = msg->msg_event;
+		out_msg->msg_sender = msg->msg_sender_thread->tr_parent->t_id;
+		out_msg->msg_endpoint = msg->msg_sender_port->p_base.ob_id;
+		spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
+		msg_free(msg);
+
+		return KERN_OK;
+	}
+
 	struct task *sender = msg->msg_sender_thread->tr_parent;
 	struct task *receiver = self->tr_parent;
 
@@ -218,6 +223,7 @@ extern kern_status_t channel_recv_msg(
 	}
 
 	out_msg->msg_id = msg->msg_id;
+	out_msg->msg_type = msg->msg_type;
 	out_msg->msg_sender = msg->msg_sender_thread->tr_parent->t_id;
 	out_msg->msg_endpoint = msg->msg_sender_port->p_base.ob_id;
 

kernel/handle.c

@@ -46,7 +46,7 @@ static void do_handle_table_destroy_leaf(struct handle_table *tab)
 		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);
+			object_unref(child->h_object);
 			child->h_object = NULL;
 		}
 	}
@@ -195,7 +195,7 @@ kern_status_t handle_table_free_handle(
 		= &tab->t_handles.t_handle_list[handle_index];
 
 	if (handle_entry->h_object) {
-		object_remove_handle(handle_entry->h_object);
+		object_unref(handle_entry->h_object);
 	}
 
 	memset(handle_entry, 0x0, sizeof *handle_entry);
@@ -307,7 +307,7 @@ kern_status_t handle_table_transfer(
 
 			dst_entry->h_object = src_entry->h_object;
 			dst_entry->h_flags = src_entry->h_flags;
-			object_add_handle(dst_entry->h_object);
+			object_ref(dst_entry->h_object);
 
 			handle_table_free_handle(src, src_handles[i].hnd_value);
 
@@ -326,7 +326,7 @@ kern_status_t handle_table_transfer(
 
 			dst_entry->h_object = src_entry->h_object;
 			dst_entry->h_flags = src_entry->h_flags;
-			object_add_handle(dst_entry->h_object);
+			object_ref(dst_entry->h_object);
 
 			dst_handle.hnd_mode = src_handles[i].hnd_mode;
 			dst_handle.hnd_value = dst_value;
@@ -371,7 +371,7 @@ kern_status_t handle_table_transfer(
 		struct handle *src_entry
 			= handle_table_get_handle(src, handle.hnd_value);
 		if (src_entry) {
-			object_remove_handle(src_entry->h_object);
+			object_unref(src_entry->h_object);
 			handle_table_free_handle(src, handle.hnd_value);
 		}
 	}

kernel/msg.c

@@ -0,0 +1,22 @@
+#include <kernel/msg.h>
+#include <kernel/vm.h>
+
+static struct vm_cache msg_cache = {
+	.c_name = "msg",
+	.c_obj_size = sizeof(struct msg),
+};
+
+void msg_init(void)
+{
+	vm_cache_init(&msg_cache);
+}
+
+struct msg *msg_alloc(void)
+{
+	return vm_cache_alloc(&msg_cache, VM_NORMAL);
+}
+
+void msg_free(struct msg *msg)
+{
+	vm_cache_free(&msg_cache, msg);
+}

kernel/object.c

@@ -74,90 +74,28 @@ struct object *object_create(struct object_type *type)
 	obj->ob_lock = SPIN_LOCK_INIT;
 	obj->ob_magic = OBJECT_MAGIC;
 	obj->ob_refcount = 1;
-	obj->ob_handles = 0;
 
 	return obj;
 }
 
 struct object *object_ref(struct object *obj)
 {
-	obj->ob_refcount++;
+	atomic_add_fetch(&obj->ob_refcount, 1);
 	return obj;
 }
 
-static void __cleanup(struct object *obj, struct queue *queue)
-{
-	if (HAS_OP(obj, destroy)) {
-		obj->ob_type->ob_ops.destroy(obj, queue);
-	}
-
-	vm_cache_free(&obj->ob_type->ob_cache, obj);
-}
-
-static void object_cleanup(struct object *obj, unsigned long flags)
-{
-	if (obj->ob_refcount > 0 || obj->ob_handles > 0) {
-		spin_unlock_irqrestore(&obj->ob_lock, flags);
-		return;
-	}
-
-	struct queue queue = QUEUE_INIT;
-	__cleanup(obj, &queue);
-
-	if (!HAS_OP(obj, destroy_recurse)) {
-		return;
-	}
-
-	while (!queue_empty(&queue)) {
-		struct queue_entry *entry = queue_pop_front(&queue);
-		struct object *child = NULL;
-		obj->ob_type->ob_ops.destroy_recurse(entry, &child);
-		if (!child) {
-			continue;
-		}
-
-		if (child->ob_refcount > 1) {
-			child->ob_refcount--;
-			continue;
-		}
-
-		if (child->ob_refcount == 0 && child->ob_handles == 0) {
-			__cleanup(child, &queue);
-		}
-	}
-}
-
 void object_unref(struct object *obj)
 {
-	unsigned long flags;
-	spin_lock_irqsave(&obj->ob_lock, &flags);
-
-	if (obj->ob_refcount == 0) {
-		spin_unlock_irqrestore(&obj->ob_lock, flags);
+	int ref = atomic_sub_fetch(&obj->ob_refcount, 1);
+	if (ref > 0) {
 		return;
 	}
 
-	obj->ob_refcount--;
-	object_cleanup(obj, flags);
-}
-
-void object_add_handle(struct object *obj)
-{
-	obj->ob_handles++;
-}
-
-void object_remove_handle(struct object *obj)
-{
-	unsigned long flags;
-	spin_lock_irqsave(&obj->ob_lock, &flags);
-
-	if (obj->ob_handles == 0) {
-		spin_unlock_irqrestore(&obj->ob_lock, flags);
-		return;
+	if (HAS_OP(obj, destroy)) {
+		obj->ob_type->ob_ops.destroy(obj);
 	}
 
-	obj->ob_handles--;
-	object_cleanup(obj, flags);
+	vm_cache_free(&obj->ob_type->ob_cache, obj);
 }
 
 void object_lock(struct object *obj)

kernel/port.c

@@ -1,16 +1,29 @@
 #include <kernel/channel.h>
 #include <kernel/port.h>
+#include <kernel/printk.h>
 #include <kernel/thread.h>
 #include <kernel/util.h>
 
 #define PORT_CAST(p) OBJECT_C_CAST(struct port, p_base, &port_type, p)
 
+static kern_status_t port_cleanup(struct object *obj);
+
 static struct object_type port_type = {
 	.ob_name = "port",
 	.ob_size = sizeof(struct port),
 	.ob_header_offset = offsetof(struct port, p_base),
+	.ob_ops = {
+		.destroy = port_cleanup,
+	},
 };
 
+static kern_status_t port_cleanup(struct object *obj)
+{
+	struct port *port = PORT_CAST(obj);
+	port_disconnect(port);
+	return KERN_OK;
+}
+
 kern_status_t port_type_init(void)
 {
 	return object_type_register(&port_type);
@@ -58,9 +71,26 @@ struct port *port_create(void)
 kern_status_t port_connect(struct port *port, struct channel *remote)
 {
 	if (port->p_status != PORT_OFFLINE) {
+		tracek("port_connect: port in bad state (%d)", port->p_status);
 		return KERN_BAD_STATE;
 	}
 
+	struct msg *msg = msg_alloc();
+	if (!msg) {
+		return KERN_NO_MEMORY;
+	}
+
+	msg->msg_status = KMSG_ASYNC;
+	msg->msg_type = KERN_MSG_TYPE_EVENT;
+	msg->msg_event = KERN_MSG_EVENT_CONNECTION;
+	msg->msg_sender_thread = current_thread();
+	msg->msg_sender_port = port;
+
+	unsigned long flags;
+	channel_lock_irqsave(remote, &flags);
+	channel_enqueue_msg(remote, msg);
+	channel_unlock_irqrestore(remote, flags);
+
 	port->p_remote = remote;
 	port->p_status = PORT_READY;
 	return KERN_OK;
@@ -69,9 +99,27 @@ kern_status_t port_connect(struct port *port, struct channel *remote)
 kern_status_t port_disconnect(struct port *port)
 {
 	if (port->p_status != PORT_READY) {
+		tracek("port_disconnect: port in bad state (%d)",
+		       port->p_status);
 		return KERN_BAD_STATE;
 	}
 
+	struct msg *msg = msg_alloc();
+	if (!msg) {
+		return KERN_NO_MEMORY;
+	}
+
+	msg->msg_status = KMSG_ASYNC;
+	msg->msg_type = KERN_MSG_TYPE_EVENT;
+	msg->msg_event = KERN_MSG_EVENT_DISCONNECTION;
+	msg->msg_sender_thread = current_thread();
+	msg->msg_sender_port = port;
+
+	unsigned long flags;
+	channel_lock_irqsave(port->p_remote, &flags);
+	channel_enqueue_msg(port->p_remote, msg);
+	channel_unlock_irqrestore(port->p_remote, flags);
+
 	port->p_remote = NULL;
 	port->p_status = PORT_OFFLINE;
 	return KERN_OK;
@@ -84,12 +132,14 @@ kern_status_t port_send_msg(
 	unsigned long *lock_flags)
 {
 	if (port->p_status != PORT_READY) {
+		tracek("port_send_msg: port in bad state (%d)", port->p_status);
 		return KERN_BAD_STATE;
 	}
 
 	struct thread *self = current_thread();
 	struct msg msg;
 	memset(&msg, 0x0, sizeof msg);
+	msg.msg_type = KERN_MSG_TYPE_DATA;
 	msg.msg_status = KMSG_WAIT_RECEIVE;
 	msg.msg_sender_thread = self;
 	msg.msg_sender_port = port;

libmango/include/mango/types.h

@@ -4,61 +4,71 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#define VM_PROT_READ            0x01u
-#define VM_PROT_WRITE           0x02u
-#define VM_PROT_EXEC            0x04u
-#define VM_PROT_USER            0x08u
-#define VM_PROT_SVR             0x10u
-#define VM_PROT_NOCACHE         0x10u
-#define VM_PROT_MAP_SPECIFIC    0x40u
+#define VM_PROT_READ                 0x01u
+#define VM_PROT_WRITE                0x02u
+#define VM_PROT_EXEC                 0x04u
+#define VM_PROT_USER                 0x08u
+#define VM_PROT_SVR                  0x10u
+#define VM_PROT_NOCACHE              0x10u
+#define VM_PROT_MAP_SPECIFIC         0x40u
 
-#define MAP_ADDRESS_ANY         ((virt_addr_t) - 1)
-#define MAP_ADDRESS_INVALID     ((virt_addr_t)0)
-#define KERN_HANDLE_INVALID     ((kern_handle_t)0xFFFFFFFF)
+#define MAP_ADDRESS_ANY              ((virt_addr_t) - 1)
+#define MAP_ADDRESS_INVALID          ((virt_addr_t)0)
+#define KERN_HANDLE_INVALID          ((kern_handle_t)0xFFFFFFFF)
 
 /* config keys for use with kern_config_get/kern_config_set */
-#define KERN_CFG_INVALID        0x00000u
-#define KERN_CFG_PAGE_SIZE      0x00001u
+#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
+#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
+#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
+#define KERN_MSG_MAX_HANDLES         64
 
 /* the corresponding handle should be ignored */
-#define KERN_MSG_HANDLE_IGNORE  0
+#define KERN_MSG_HANDLE_IGNORE       0
 /* the corresponding handle should be moved to the recipient task. the handle
  * will be closed. */
-#define KERN_MSG_HANDLE_MOVE    1
+#define KERN_MSG_HANDLE_MOVE         1
 /* the corresponding handle should be copied to the recipient task. the handle
  * will remain valid for the sending task. */
-#define KERN_MSG_HANDLE_COPY    2
+#define KERN_MSG_HANDLE_COPY         2
 
 /* maximum number of objects that can be waited on in a single call to
  * kern_object_wait */
-#define KERN_WAIT_MAX_ITEMS     64
+#define KERN_WAIT_MAX_ITEMS          64
+
+/* message types */
+#define KERN_MSG_TYPE_NONE           0
+#define KERN_MSG_TYPE_DATA           1
+#define KERN_MSG_TYPE_EVENT          2
+
+/* event message types */
+#define KERN_MSG_EVENT_NONE          0
+#define KERN_MSG_EVENT_CONNECTION    1
+#define KERN_MSG_EVENT_DISCONNECTION 2
 
 /* equeue packet types */
-#define EQUEUE_PKT_PAGE_REQUEST 0x01u
-#define EQUEUE_PKT_ASYNC_SIGNAL 0x02u
+#define EQUEUE_PKT_PAGE_REQUEST      0x01u
+#define EQUEUE_PKT_ASYNC_SIGNAL      0x02u
 
 /* page request types */
-#define PAGE_REQUEST_READ       0x01u
-#define PAGE_REQUEST_DIRTY      0x02u
-#define PAGE_REQUEST_DETACH     0x03u
+#define PAGE_REQUEST_READ            0x01u
+#define PAGE_REQUEST_DIRTY           0x02u
+#define PAGE_REQUEST_DETACH          0x03u
 
 /* futex special values */
-#define FUTEX_WAKE_ALL          ((size_t)-1)
+#define FUTEX_WAKE_ALL               ((size_t)-1)
 
 /* futex flags */
-#define FUTEX_PRIVATE           0x01u
-#define FUTEX_SHARED            0x02u
+#define FUTEX_PRIVATE                0x01u
+#define FUTEX_SHARED                 0x02u
 
 #define IOVEC(p, len)                                                          \
 	{                                                                      \
@@ -92,6 +102,8 @@ 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 uint32_t kern_msg_type_t;
+typedef uint32_t kern_msg_event_type_t;
 
 typedef unsigned short equeue_packet_type_t;
 
@@ -122,14 +134,27 @@ typedef struct {
 	tid_t msg_sender;
 	/* the id of the port or channel used to send a particular message. */
 	koid_t msg_endpoint;
-	/* a list of iovecs that point to the buffers that make up the main
-	 * message data. */
-	kern_iovec_t *msg_data;
-	size_t msg_data_count;
-	/* a list of handle entries that contain the kernel handles included
-	 * in a message. */
-	kern_msg_handle_t *msg_handles;
-	size_t msg_handles_count;
+	/* the message type */
+	kern_msg_type_t msg_type;
+
+	union {
+		/* msg_type = KERN_MSG_TYPE_DATA */
+		struct {
+			/* a list of iovecs that point to the buffers that make
+			 * up the main message data. */
+			kern_iovec_t *msg_data;
+			size_t msg_data_count;
+			/* a list of handle entries that contain the kernel
+			 * handles included in a message. */
+			kern_msg_handle_t *msg_handles;
+			size_t msg_handles_count;
+		};
+
+		/* msg_type = KERN_MSG_TYPE_EVENT */
+		struct {
+			kern_msg_event_type_t msg_event;
+		};
+	};
 } kern_msg_t;
 
 typedef struct {

sched/task.c

@@ -334,11 +334,10 @@ void task_exit(int status)
 	       self->t_name,
 	       self->t_id,
 	       cur_thread->tr_id);
-	tracek("task %s[%u] killed (%u, %u)",
+	tracek("task %s[%u] killed (%u)",
 	       self->t_name,
 	       self->t_id,
-	       self->t_base.ob_refcount,
-	       self->t_base.ob_handles);
+	       self->t_base.ob_refcount);
 	spin_unlock_irqrestore(handles_lock, flags);
 
 	while (1) {
@@ -359,7 +358,7 @@ kern_status_t task_open_handle(
 		return status;
 	}
 
-	object_add_handle(obj);
+	object_ref(obj);
 	handle_data->h_object = obj;
 	handle_data->h_flags = flags;
 

syscall/msg.c

@@ -60,9 +60,9 @@ kern_status_t sys_port_create(kern_handle_t *out)
 	kern_status_t status
 		= task_open_handle(self, &port->p_base, 0, &handle);
 	task_unlock_irqrestore(self, irq_flags);
+	object_unref(&port->p_base);
 
 	if (status != KERN_OK) {
-		object_unref(&port->p_base);
 		return status;
 	}
 
@@ -114,6 +114,7 @@ kern_status_t sys_port_connect(
 	status = port_connect(port, remote);
 	port_unlock_irqrestore(port, flags);
 	object_unref(&remote->c_base);
+	object_unref(port_obj);
 
 	return KERN_OK;
 }

syscall/task.c

@@ -38,7 +38,7 @@ kern_status_t sys_task_self(kern_handle_t *out)
 		return status;
 	}
 
-	object_add_handle(&self->t_base);
+	object_ref(&self->t_base);
 	handle_slot->h_object = &self->t_base;
 
 	*out = handle;
@@ -129,8 +129,8 @@ kern_status_t sys_task_create(
 	child_handle_slot->h_object = &child->t_base;
 	space_handle_slot->h_object = &child->t_address_space->s_base;
 
-	object_add_handle(&child->t_base);
-	object_add_handle(&child->t_address_space->s_base);
+	object_ref(&child->t_base);
+	object_ref(&child->t_address_space->s_base);
 
 	object_unref(parent_obj);
 
@@ -199,7 +199,7 @@ kern_status_t sys_task_create_thread(
 
 	thread_init_user(thread, ip, sp, args, nr_args);
 	target_handle->h_object = &thread->tr_base;
-	object_add_handle(&thread->tr_base);
+	object_ref(&thread->tr_base);
 
 	task_unlock_irqrestore(target, flags);
 	object_unref(target_obj);
@@ -254,7 +254,7 @@ kern_status_t sys_task_get_address_space(
 	}
 
 	handle_slot->h_object = &task->t_address_space->s_base;
-	object_add_handle(&task->t_address_space->s_base);
+	object_ref(&task->t_address_space->s_base);
 	task_unlock_irqrestore(self, flags);
 	object_unref(task_obj);
 
@@ -286,7 +286,7 @@ kern_status_t sys_thread_self(kern_handle_t *out)
 		return status;
 	}
 
-	object_add_handle(&self_thread->tr_base);
+	object_ref(&self_thread->tr_base);
 	handle_slot->h_object = &self_thread->tr_base;
 
 	*out = handle;

syscall/vm-controller.c

@@ -186,8 +186,6 @@ kern_status_t sys_vm_controller_create_object(
 	}
 
 	out_slot->h_object = &out_vmo->vo_base;
-	object_add_handle(&out_vmo->vo_base);
-	object_unref(&out_vmo->vo_base);
 
 	*out = out_handle;
 	return KERN_OK;

vm/address-space.c

@@ -7,6 +7,7 @@
 #include <kernel/sched.h>
 #include <kernel/task.h>
 #include <kernel/util.h>
+#include <kernel/vm-controller.h>
 #include <kernel/vm-object.h>
 #include <mango/status.h>
 
@@ -47,7 +48,7 @@ enum search_direction {
 	SEARCH_RIGHT,
 };
 
-static kern_status_t address_space_cleanup(struct object *obj, struct queue *q);
+static kern_status_t address_space_cleanup(struct object *obj);
 
 static struct object_type address_space_type = {
 	.ob_name = "address-space",
@@ -141,16 +142,16 @@ static struct vm_area *get_entry(
 
 		if (address < child->vma_base) {
 			next = btree_left(cur);
-			if (LEFT_DIFF(address, child)
-			    < LEFT_DIFF(address, closest_left)) {
-				closest_left = child;
-			}
-		} else if (address > child->vma_limit) {
-			next = btree_right(cur);
 			if (RIGHT_DIFF(address, child)
 			    < RIGHT_DIFF(address, closest_right)) {
 				closest_right = child;
 			}
+		} else if (address > child->vma_limit) {
+			next = btree_right(cur);
+			if (LEFT_DIFF(address, child)
+			    < LEFT_DIFF(address, closest_left)) {
+				closest_left = child;
+			}
 		} else {
 			result = child;
 			break;
@@ -664,9 +665,10 @@ static void area_unmap(struct vm_area *area)
 	}
 }
 
-static kern_status_t address_space_cleanup(struct object *obj, struct queue *q)
+static kern_status_t address_space_cleanup(struct object *obj)
 {
 	struct address_space *space = ADDRESS_SPACE_CAST(obj);
+	tracek("begin address space cleanup %p", space);
 	struct btree_node *cur = btree_first(&space->s_mappings);
 	while (cur) {
 		struct btree_node *next = btree_next(cur);
@@ -680,6 +682,20 @@ static kern_status_t address_space_cleanup(struct object *obj, struct queue *q)
 		cur = next;
 	}
 
+	cur = btree_first(&space->s_reserved);
+	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_reserved, cur);
+
+		delete_area(area, space);
+		vm_cache_free(&vm_area_cache, area);
+
+		cur = next;
+	}
+
+	tracek("end address space cleanup %p", space);
 	return KERN_OK;
 }
 
@@ -706,6 +722,9 @@ kern_status_t address_space_map(
 	}
 
 	tracek("address_space_map(%zx, %zx)", map_address, length);
+	if (map_address == 0xc6a55000) {
+		printk("break");
+	}
 
 	if (!root || !object) {
 		tracek("null pointer");
@@ -823,7 +842,7 @@ static kern_status_t split_area(
 
 	put_entry(&root->s_mappings, right);
 	for (size_t i = unmap_base; i < unmap_limit; i += VM_PAGE_SIZE) {
-		tracek("unmapping %zx", i);
+		tracek("pmap_remove %zx", i);
 		pmap_remove(root->s_pmap, i);
 	}
 
@@ -860,7 +879,7 @@ static kern_status_t left_reduce_area(
 		return KERN_OK;
 	}
 
-	tracek(" unmapping %zx-%zx (%zx bytes)", base, base + length, length);
+	tracek(" pmap_remove %zx-%zx (%zx bytes)", base, base + length, length);
 	for (size_t i = base; i < limit; i += VM_PAGE_SIZE) {
 		pmap_remove(root->s_pmap, i);
 	}
@@ -897,7 +916,7 @@ static kern_status_t right_reduce_area(
 		return KERN_OK;
 	}
 
-	tracek(" unmapping %zx-%zx (%zx bytes)", base, base + length, length);
+	tracek(" pmap_remove %zx-%zx (%zx bytes)", base, base + length, length);
 	for (size_t i = base; i < limit; i += VM_PAGE_SIZE) {
 		pmap_remove(root->s_pmap, i);
 	}
@@ -915,9 +934,10 @@ static kern_status_t delete_area(
 		return KERN_OK;
 	}
 
-	tracek("delete mapping [%zx-%zx]",
+	tracek("delete mapping [%zx-%zx] (%zx bytes)",
 	       mapping->vma_base,
-	       mapping->vma_limit);
+	       mapping->vma_limit,
+	       mapping->vma_limit - mapping->vma_base);
 
 	for (size_t i = mapping->vma_base; i < mapping->vma_limit;
 	     i += VM_PAGE_SIZE) {
@@ -931,7 +951,23 @@ static kern_status_t delete_area(
 		&mapping->vma_object->vo_mappings,
 		&mapping->vma_object_entry);
 	mapping->vma_object = NULL;
+	/* if the object is attached to a controller and the ref-count is 2,
+	 * then the only other remaining reference to this object is held by
+	 * the controller. */
+	struct vm_controller *ctrl = object->vo_ctrl;
+	bool detach = ctrl != NULL && object->vo_base.ob_refcount == 2
+	           && (object->vo_flags & VMO_AUTO_DETACH);
 	vm_object_unlock_irqrestore(object, flags);
+
+	if (detach) {
+		/* TODO find a better way to achieve this, and/or give the
+		 * server that created the object more control over when it
+		 * should be detached */
+		vm_controller_lock_irqsave(ctrl, &flags);
+		vm_controller_detach_object(ctrl, object);
+		vm_controller_unlock_irqrestore(ctrl, flags);
+	}
+
 	object_unref(&object->vo_base);
 
 	/* don't actually delete the mapping yet. that will be done by
@@ -980,9 +1016,11 @@ kern_status_t address_space_unmap(
 			= (area_base <= unmap_base
 		           && area_limit >= unmap_limit);
 		bool left_reduce
-			= (unmap_base <= area_base && unmap_limit < area_limit);
+			= (unmap_base <= area_base && unmap_limit > area_base
+		           && unmap_limit < area_limit);
 		bool right_reduce
-			= (unmap_base > area_base && unmap_limit >= area_limit);
+			= (unmap_base > area_base && unmap_base < area_limit
+		           && unmap_limit >= area_limit);
 
 		if (split) {
 			status = split_area(
@@ -1120,9 +1158,10 @@ kern_status_t address_space_release(
 		           && area_limit >= release_limit);
 		bool left_reduce
 			= (release_base <= area_base
+		           && release_limit > area_base
 		           && release_limit < area_limit);
 		bool right_reduce
-			= (release_base > area_base
+			= (release_base > area_base && release_base < area_limit
 		           && release_limit >= area_limit);
 
 		if (split) {

vm/vm-controller.c

@@ -23,8 +23,14 @@ static struct object_type vm_controller_type = {
 	.ob_header_offset = offsetof(struct vm_controller, vc_base),
 };
 
+static struct vm_cache page_request_cache = {
+	.c_name = "page-request",
+	.c_obj_size = sizeof(struct page_request),
+};
+
 kern_status_t vm_controller_type_init(void)
 {
+	vm_cache_init(&page_request_cache);
 	return object_type_register(&vm_controller_type);
 }
 
@@ -52,10 +58,17 @@ static struct page_request *get_next_request(struct vm_controller *ctrl)
 		struct page_request *req
 			= BTREE_CONTAINER(struct page_request, req_node, cur);
 		spin_lock(&req->req_lock);
-		if (req->req_status == PAGE_REQUEST_PENDING) {
+		switch (req->req_status) {
+		case PAGE_REQUEST_PENDING:
 			req->req_status = PAGE_REQUEST_IN_PROGRESS;
 			ctrl->vc_requests_waiting--;
 			return req;
+		case PAGE_REQUEST_ASYNC:
+			btree_delete(&ctrl->vc_requests, &req->req_node);
+			ctrl->vc_requests_waiting--;
+			return req;
+		default:
+			break;
 		}
 
 		spin_unlock(&req->req_lock);
@@ -65,98 +78,6 @@ static struct page_request *get_next_request(struct vm_controller *ctrl)
 	return NULL;
 }
 
-kern_status_t vm_controller_recv(
-	struct vm_controller *ctrl,
-	equeue_packet_page_request_t *out)
-{
-	struct page_request *req = NULL;
-
-	req = get_next_request(ctrl);
-	if (!req) {
-		return KERN_NO_ENTRY;
-	}
-
-	if (ctrl->vc_requests_waiting == 0) {
-		object_clear_signal(
-			&ctrl->vc_base,
-			VM_CONTROLLER_SIGNAL_REQUEST_RECEIVED);
-	}
-
-	out->req_vmo = req->req_object->vo_key;
-	out->req_type = req->req_type;
-	out->req_offset = req->req_offset;
-	out->req_length = req->req_length;
-
-	spin_unlock(&req->req_lock);
-	return KERN_OK;
-}
-
-kern_status_t vm_controller_recv_async(
-	struct vm_controller *ctrl,
-	struct equeue *eq,
-	equeue_key_t key)
-{
-	if (ctrl->vc_eq) {
-		object_unref(&ctrl->vc_eq->eq_base);
-	}
-
-	object_ref(&eq->eq_base);
-	ctrl->vc_eq = eq;
-	ctrl->vc_eq_key = key;
-
-	return KERN_OK;
-}
-
-kern_status_t vm_controller_create_object(
-	struct vm_controller *ctrl,
-	const char *name,
-	size_t name_len,
-	equeue_key_t key,
-	size_t data_len,
-	vm_prot_t prot,
-	struct vm_object **out)
-{
-	struct vm_object *vmo = get_object(&ctrl->vc_objects, key);
-	if (vmo) {
-		return KERN_NAME_EXISTS;
-	}
-
-	vmo = vm_object_create(name, name_len, data_len, prot);
-	if (!vmo) {
-		return KERN_NO_MEMORY;
-	}
-
-	object_ref(&ctrl->vc_base);
-	object_ref(&vmo->vo_base);
-
-	vmo->vo_flags |= VMO_CONTROLLER;
-	vmo->vo_ctrl = ctrl;
-	vmo->vo_key = key;
-
-	put_object(&ctrl->vc_objects, vmo);
-
-	*out = vmo;
-	return KERN_OK;
-}
-
-kern_status_t vm_controller_detach_object(
-	struct vm_controller *ctrl,
-	struct vm_object *vmo)
-{
-	if (vmo->vo_ctrl != ctrl) {
-		return KERN_INVALID_ARGUMENT;
-	}
-
-	vmo->vo_ctrl = NULL;
-	vmo->vo_key = 0;
-	btree_delete(&ctrl->vc_objects, &vmo->vo_ctrl_node);
-
-	object_unref(&ctrl->vc_base);
-	object_unref(&vmo->vo_base);
-
-	return KERN_OK;
-}
-
 static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
 {
 	if (!tree->b_root) {
@@ -196,6 +117,119 @@ static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
 	return true;
 }
 
+static kern_status_t send_request_async(
+	struct vm_controller *ctrl,
+	struct page_request *req)
+{
+	fill_random(&req->req_id, sizeof req->req_id);
+	while (!try_enqueue(&ctrl->vc_requests, req)) {
+		req->req_id++;
+	}
+
+	ctrl->vc_requests_waiting++;
+	object_assert_signal(
+		&ctrl->vc_base,
+		VM_CONTROLLER_SIGNAL_REQUEST_RECEIVED);
+
+	return KERN_OK;
+}
+
+kern_status_t vm_controller_recv(
+	struct vm_controller *ctrl,
+	equeue_packet_page_request_t *out)
+{
+	struct page_request *req = NULL;
+
+	req = get_next_request(ctrl);
+	if (!req) {
+		return KERN_NO_ENTRY;
+	}
+
+	if (ctrl->vc_requests_waiting == 0) {
+		object_clear_signal(
+			&ctrl->vc_base,
+			VM_CONTROLLER_SIGNAL_REQUEST_RECEIVED);
+	}
+
+	out->req_vmo = req->req_object;
+	out->req_type = req->req_type;
+	out->req_offset = req->req_offset;
+	out->req_length = req->req_length;
+
+	spin_unlock(&req->req_lock);
+
+	if (req->req_status == PAGE_REQUEST_ASYNC) {
+		vm_cache_free(&page_request_cache, req);
+	}
+
+	return KERN_OK;
+}
+
+kern_status_t vm_controller_recv_async(
+	struct vm_controller *ctrl,
+	struct equeue *eq,
+	equeue_key_t key)
+{
+	if (ctrl->vc_eq) {
+		object_unref(&ctrl->vc_eq->eq_base);
+	}
+
+	object_ref(&eq->eq_base);
+	ctrl->vc_eq = eq;
+	ctrl->vc_eq_key = key;
+
+	return KERN_OK;
+}
+
+kern_status_t vm_controller_create_object(
+	struct vm_controller *ctrl,
+	const char *name,
+	size_t name_len,
+	equeue_key_t key,
+	size_t data_len,
+	vm_prot_t prot,
+	struct vm_object **out)
+{
+	struct vm_object *vmo
+		= vm_object_create(name, name_len, data_len, prot);
+	if (!vmo) {
+		return KERN_NO_MEMORY;
+	}
+
+	object_ref(&ctrl->vc_base);
+
+	/* TODO expose the VMO_AUTO_DETACH flag to userspace */
+	vmo->vo_flags |= VMO_CONTROLLER | VMO_AUTO_DETACH;
+	vmo->vo_ctrl = ctrl;
+	vmo->vo_key = key;
+
+	*out = vmo;
+	return KERN_OK;
+}
+
+kern_status_t vm_controller_detach_object(
+	struct vm_controller *ctrl,
+	struct vm_object *vmo)
+{
+	if (vmo->vo_ctrl != ctrl) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	struct page_request *req
+		= vm_cache_alloc(&page_request_cache, VM_NORMAL);
+	req->req_type = PAGE_REQUEST_DETACH;
+	req->req_status = PAGE_REQUEST_ASYNC;
+	req->req_object = vmo->vo_key;
+	req->req_sender = current_thread();
+	send_request_async(ctrl, req);
+
+	vmo->vo_ctrl = NULL;
+	vmo->vo_key = 0;
+	object_unref(&ctrl->vc_base);
+
+	return KERN_OK;
+}
+
 static void wait_for_reply(
 	struct vm_controller *ctrl,
 	struct page_request *req,
@@ -221,7 +255,7 @@ static void wait_for_reply(
 
 static void fulfill_requests(
 	struct vm_controller *ctrl,
-	struct vm_object *obj,
+	equeue_key_t object,
 	off_t offset,
 	size_t length,
 	kern_status_t result)
@@ -242,7 +276,7 @@ static void fulfill_requests(
 			match = true;
 		}
 
-		if (req->req_object != obj) {
+		if (req->req_object != object) {
 			match = false;
 		}
 
@@ -280,7 +314,7 @@ kern_status_t vm_controller_supply_pages(
 		src_offset,
 		count,
 		NULL);
-	fulfill_requests(ctrl, dst, dst_offset, count, status);
+	fulfill_requests(ctrl, dst->vo_key, dst_offset, count, status);
 
 	return status;
 }

vm/vm-object.c

@@ -15,7 +15,7 @@
 		(p) += VM_PAGE_SIZE;                                           \
 	}
 
-static kern_status_t vm_object_cleanup(struct object *obj, struct queue *q)
+static kern_status_t vm_object_cleanup(struct object *obj)
 {
 	struct vm_object *vmo = vm_object_cast(obj);
 	struct btree_node *cur = btree_first(&vmo->vo_pages);
@@ -29,6 +29,13 @@ static kern_status_t vm_object_cleanup(struct object *obj, struct queue *q)
 		cur = next;
 	}
 
+	if (vmo->vo_ctrl) {
+		unsigned long flags;
+		vm_controller_lock_irqsave(vmo->vo_ctrl, &flags);
+		vm_controller_detach_object(vmo->vo_ctrl, vmo);
+		vm_controller_unlock_irqrestore(vmo->vo_ctrl, flags);
+	}
+
 	return KERN_OK;
 }
 
@@ -366,12 +373,13 @@ static kern_status_t request_page(
 	struct vm_controller *ctrl = vo->vo_ctrl;
 	struct page_request req = {0};
 	req.req_status = PAGE_REQUEST_PENDING;
+	req.req_type = PAGE_REQUEST_READ;
 	req.req_offset = offset;
 	req.req_length = vm_page_order_to_bytes(VM_PAGE_4K);
 	req.req_sender = current_thread();
 
 	object_ref(&vo->vo_base);
-	req.req_object = vo;
+	req.req_object = vo->vo_key;
 
 	vm_object_unlock_irqrestore(vo, *irq_flags);
 	vm_controller_lock_irqsave(ctrl, irq_flags);