[continued from previous message]

+ * Therefore we first check whether all_hint is zero, and, if it is,
+ * we skip trying to wake "all" waiters.
+ *
+ * Since wait requests must originate from user-space threads, we're
+ * limited here by PID_MAX_LIMIT, so there's no risk of overflow.
+ */
+ atomic_t all_hint;
+};
+
+struct ntsync_q_entry {
+ struct list_head node;
+ struct ntsync_q *q;
+ struct ntsync_obj *obj;
+ __u32 index;
+};
+
+struct ntsync_q {
+ struct task_struct *task;
+ __u32 owner;
+
+ /*
+ * Protected via atomic_try_cmpxchg(). Only the thread that wins the
+ * compare-and-swap may actually change object states and wake this
+ * task.
+ */
+ atomic_t signaled;
+
+ bool all;
+ bool ownerdead;
+ __u32 count;
+ struct ntsync_q_entry entries[];
};

struct ntsync_device {
+ /*
+ * Wait-all operations must atomically grab all objects, and be totally + * ordered with respect to each other and wait-any operations.
+ * If one thread is trying to acquire several objects, another thread
+ * cannot touch the object at the same time.
+ *
+ * This device-wide lock is used to serialize wait-for-all
+ * operations, and operations on an object that is involved in a
+ * wait-for-all.
+ */
+ struct mutex wait_all_lock;
+
struct file *file;
};

/*
+ * Single objects are locked using obj->lock.
+ *
+ * Multiple objects are 'locked' while holding dev->wait_all_lock.
+ * In this case however, individual objects are not locked by holding
+ * obj->lock, but by setting obj->dev_locked.
+ *
+ * This means that in order to lock a single object, the sequence is slightly + * more complicated than usual. Specifically it needs to check obj->dev_locked + * after acquiring obj->lock, if set, it needs to drop the lock and acquire
+ * dev->wait_all_lock in order to serialize against the multi-object operation.
+ */
+
+static void dev_lock_obj(struct ntsync_device *dev, struct ntsync_obj *obj)
+{
+ lockdep_assert_held(&dev->wait_all_lock);
+ lockdep_assert(obj->dev == dev);
+ spin_lock(&obj->lock);
+ /*
+ * By setting obj->dev_locked inside obj->lock, it is ensured that
+ * anyone holding obj->lock must see the value.
+ */
+ obj->dev_locked = 1;
+ spin_unlock(&obj->lock);
+}
+
+static void dev_unlock_obj(struct ntsync_device *dev, struct ntsync_obj *obj) +{
+ lockdep_assert_held(&dev->wait_all_lock);
+ lockdep_assert(obj->dev == dev);
+ spin_lock(&obj->lock);
+ obj->dev_locked = 0;
+ spin_unlock(&obj->lock);
+}
+
+static void obj_lock(struct ntsync_obj *obj)
+{
+ struct ntsync_device *dev = obj->dev;
+
+ for (;;) {
+ spin_lock(&obj->lock);
+ if (likely(!obj->dev_locked))
+ break;
+
+ spin_unlock(&obj->lock);
+ mutex_lock(&dev->wait_all_lock);
+ spin_lock(&obj->lock);
+ /*
+ * obj->dev_locked should be set and released under the same
+ * wait_all_lock section, since we now own this lock, it should + * be clear.
+ */
+ lockdep_assert(!obj->dev_locked);
+ spin_unlock(&obj->lock);
+ mutex_unlock(&dev->wait_all_lock);
+ }
+}
+
+static void obj_unlock(struct ntsync_obj *obj)
+{
+ spin_unlock(&obj->lock);
+}
+
+static bool ntsync_lock_obj(struct ntsync_device *dev, struct ntsync_obj *obj) +{
+ bool all;
+
+ obj_lock(obj);
+ all = atomic_read(&obj->all_hint);
+ if (unlikely(all)) {
+ obj_unlock(obj);
+ mutex_lock(&dev->wait_all_lock);
+ dev_lock_obj(dev, obj);
+ }
+
+ return all;
+}
+
+static void ntsync_unlock_obj(struct ntsync_device *dev, struct ntsync_obj *obj, bool all)
+{
+ if (all) {
+ dev_unlock_obj(dev, obj);
+ mutex_unlock(&dev->wait_all_lock);
+ } else {
+ obj_unlock(obj);
+ }
+}
+
+#define ntsync_assert_held(obj) \
+ lockdep_assert((lockdep_is_held(&(obj)->lock) != LOCK_STATE_NOT_HELD) || \
+ ((lockdep_is_held(&(obj)->dev->wait_all_lock) != LOCK_STATE_NOT_HELD) && \
+ (obj)->dev_locked))
+
+static bool is_signaled(struct ntsync_obj *obj, __u32 owner)
+{
+ ntsync_assert_held(obj);
+
+ switch (obj->type) {
+ case NTSYNC_TYPE_SEM:
+ return !!obj->u.sem.count;
+ case NTSYNC_TYPE_MUTEX:
+ if (obj->u.mutex.owner && obj->u.mutex.owner != owner)
+ return false;
+ return obj->u.mutex.count < UINT_MAX;
+ case NTSYNC_TYPE_EVENT:
+ return obj->u.event.signaled;
+ }
+
+ WARN(1, "bad object type %#x\n", obj->type);
+ return false;
+}
+
+/*
+ * "locked_obj" is an optional pointer to an object which is already locked and
+ * should not be locked again. This is necessary so that changing an object's + * state and waking it can be a single atomic operation.
+ */
+static void try_wake_all(struct ntsync_device *dev, struct ntsync_q *q,
+ struct ntsync_obj *locked_obj)
+{
+ __u32 count = q->count;
+ bool can_wake = true;
+ int signaled = -1;
+ __u32 i;
+
+ lockdep_assert_held(&dev->wait_all_lock);
+ if (locked_obj)
+ lockdep_assert(locked_obj->dev_locked);
+
+ for (i = 0; i < count; i++) {
+ if (q->entries[i].obj != locked_obj)
+ dev_lock_obj(dev, q->entries[i].obj);
+ }
+
+ for (i = 0; i < count; i++) {
+ if (!is_signaled(q->entries[i].obj, q->owner)) {
+ can_wake = false;
+ break;
+ }
+ }
+
+ if (can_wake && atomic_try_cmpxchg(&q->signaled, &signaled, 0)) {
+ for (i = 0; i < count; i++) {
+ struct ntsync_obj *obj = q->entries[i].obj;
+
+ switch (obj->type) {
+ case NTSYNC_TYPE_SEM:
+ obj->u.sem.count--;
+ break;
+ case NTSYNC_TYPE_MUTEX:
+ if (obj->u.mutex.ownerdead)
+ q->ownerdead = true;
+ obj->u.mutex.ownerdead = false;
+ obj->u.mutex.count++;
+ obj->u.mutex.owner = q->owner;
+ break;
+ case NTSYNC_TYPE_EVENT:
+ if (!obj->u.event.manual)
+ obj->u.event.signaled = false;
+ break;
+ }
+ }
+ wake_up_process(q->task);
+ }
+
+ for (i = 0; i < count; i++) {
+ if (q->entries[i].obj != locked_obj)
+ dev_unlock_obj(dev, q->entries[i].obj);
+ }
+}
+
+static void try_wake_all_obj(struct ntsync_device *dev, struct ntsync_obj *obj)
+{
+ struct ntsync_q_entry *entry;
+
+ lockdep_assert_held(&dev->wait_all_lock);
+ lockdep_assert(obj->dev_locked);
+
+ list_for_each_entry(entry, &obj->all_waiters, node)
+ try_wake_all(dev, entry->q, obj);
+}
+
+static void try_wake_any_sem(struct ntsync_obj *sem)
+{
+ struct ntsync_q_entry *entry;
+
+ ntsync_assert_held(sem);
+ lockdep_assert(sem->type == NTSYNC_TYPE_SEM);
+
+ list_for_each_entry(entry, &sem->any_waiters, node) {
+ struct ntsync_q *q = entry->q;
+ int signaled = -1;
+
+ if (!sem->u.sem.count)
+ break;
+
+ if (atomic_try_cmpxchg(&q->signaled, &signaled, entry->index)) {
+ sem->u.sem.count--;
+ wake_up_process(q->task);
+ }
+ }
+}
+
+static void try_wake_any_mutex(struct ntsync_obj *mutex)
+{
+ struct ntsync_q_entry *entry;
+
+ ntsync_assert_held(mutex);
+ lockdep_assert(mutex->type == NTSYNC_TYPE_MUTEX);
+
+ list_for_each_entry(entry, &mutex->any_waiters, node) {
+ struct ntsync_q *q = entry->q;
+ int signaled = -1;
+
+ if (mutex->u.mutex.count == UINT_MAX)
+ break;
+ if (mutex->u.mutex.owner && mutex->u.mutex.owner != q->owner)
+ continue;
+
+ if (atomic_try_cmpxchg(&q->signaled, &signaled, entry->index)) {
+ if (mutex->u.mutex.ownerdead)
+ q->ownerdead = true;
+ mutex->u.mutex.ownerdead = false;
+ mutex->u.mutex.count++;
+ mutex->u.mutex.owner = q->owner;
+ wake_up_process(q->task);
+ }
+ }
+}
+
+static void try_wake_any_event(struct ntsync_obj *event)
+{
+ struct ntsync_q_entry *entry;
+
+ ntsync_assert_held(event);
+ lockdep_assert(event->type == NTSYNC_TYPE_EVENT);
+
+ list_for_each_entry(entry, &event->any_waiters, node) {

[continued in next message]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)