WorkStruct: Merge the pending bit into the wq_data pointer

Reclaim a word from the size of the work_struct by folding the pending bit and
the wq_data pointer together. This shouldn't cause misalignment problems as
all pointers should be at least 4-byte aligned.

Signed-Off-By: David Howells <[e-mail]>

drivers/block/floppy.c

···
 	printk("fdc_busy=%lu\n", fdc_busy);
 	if (do_floppy)
 		printk("do_floppy=%p\n", do_floppy);
-	if (floppy_work.pending)
+	if (work_pending(&floppy_work))
 		printk("floppy_work.func=%p\n", floppy_work.func);
 	if (timer_pending(&fd_timer))
 		printk("fd_timer.function=%p\n", fd_timer.function);
···
 		printk("floppy timer still active:%s\n", timeout_message);
 	if (timer_pending(&fd_timer))
 		printk("auxiliary floppy timer still active\n");
-	if (floppy_work.pending)
+	if (work_pending(&floppy_work))
 		printk("work still pending\n");
 #endif
 	old_fdc = fdc;

include/linux/workqueue.h

···
 typedef void (*work_func_t)(void *data);
 
 struct work_struct {
-	unsigned long pending;
+	/* the first word is the work queue pointer and the pending flag
+	 * rolled into one */
+	unsigned long management;
+#define WORK_STRUCT_PENDING 0		/* T if work item pending execution */
+#define WORK_STRUCT_FLAG_MASK (3UL)
+#define WORK_STRUCT_WQ_DATA_MASK (~WORK_STRUCT_FLAG_MASK)
 	struct list_head entry;
 	work_func_t func;
 	void *data;
-	void *wq_data;
 };
 
 struct delayed_work {
···
 #define INIT_WORK(_work, _func, _data)				\
 	do {							\
 		INIT_LIST_HEAD(&(_work)->entry);		\
-		(_work)->pending = 0;				\
+		(_work)->management = 0;			\
 		PREPARE_WORK((_work), (_func), (_data));	\
 	} while (0)
 
···
 		INIT_WORK(&(_work)->work, (_func), (_data));	\
 		init_timer(&(_work)->timer);			\
 	} while (0)
+
+/**
+ * work_pending - Find out whether a work item is currently pending
+ * @work: The work item in question
+ */
+#define work_pending(work) \
+	test_bit(WORK_STRUCT_PENDING, &(work)->management)
+
+/**
+ * delayed_work_pending - Find out whether a delayable work item is currently
+ * pending
+ * @work: The work item in question
+ */
+#define delayed_work_pending(work) \
+	test_bit(WORK_STRUCT_PENDING, &(work)->work.management)
 
 
 extern struct workqueue_struct *__create_workqueue(const char *name,
···
 
 	ret = del_timer_sync(&work->timer);
 	if (ret)
-		clear_bit(0, &work->work.pending);
+		clear_bit(WORK_STRUCT_PENDING, &work->work.management);
 	return ret;
 }
 

kernel/workqueue.c

···
 	return list_empty(&wq->list);
 }
 
+static inline void set_wq_data(struct work_struct *work, void *wq)
+{
+	unsigned long new, old, res;
+
+	/* assume the pending flag is already set and that the task has already
+	 * been queued on this workqueue */
+	new = (unsigned long) wq | (1UL << WORK_STRUCT_PENDING);
+	res = work->management;
+	if (res != new) {
+		do {
+			old = res;
+			new = (unsigned long) wq;
+			new |= (old & WORK_STRUCT_FLAG_MASK);
+			res = cmpxchg(&work->management, old, new);
+		} while (res != old);
+	}
+}
+
+static inline void *get_wq_data(struct work_struct *work)
+{
+	return (void *) (work->management & WORK_STRUCT_WQ_DATA_MASK);
+}
+
 /* Preempt must be disabled. */
 static void __queue_work(struct cpu_workqueue_struct *cwq,
 			 struct work_struct *work)
···
 	unsigned long flags;
 
 	spin_lock_irqsave(&cwq->lock, flags);
-	work->wq_data = cwq;
+	set_wq_data(work, cwq);
 	list_add_tail(&work->entry, &cwq->worklist);
 	cwq->insert_sequence++;
 	wake_up(&cwq->more_work);
···
 {
 	int ret = 0, cpu = get_cpu();
 
-	if (!test_and_set_bit(0, &work->pending)) {
+	if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
 		if (unlikely(is_single_threaded(wq)))
 			cpu = singlethread_cpu;
 		BUG_ON(!list_empty(&work->entry));
···
 static void delayed_work_timer_fn(unsigned long __data)
 {
 	struct delayed_work *dwork = (struct delayed_work *)__data;
-	struct workqueue_struct *wq = dwork->work.wq_data;
+	struct workqueue_struct *wq = get_wq_data(&dwork->work);
 	int cpu = smp_processor_id();
 
 	if (unlikely(is_single_threaded(wq)))
···
 	if (delay == 0)
 		return queue_work(wq, work);
 
-	if (!test_and_set_bit(0, &work->pending)) {
+	if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
 		BUG_ON(timer_pending(timer));
 		BUG_ON(!list_empty(&work->entry));
 
 		/* This stores wq for the moment, for the timer_fn */
-		work->wq_data = wq;
+		set_wq_data(work, wq);
 		timer->expires = jiffies + delay;
 		timer->data = (unsigned long)dwork;
 		timer->function = delayed_work_timer_fn;
···
 	struct timer_list *timer = &dwork->timer;
 	struct work_struct *work = &dwork->work;
 
-	if (!test_and_set_bit(0, &work->pending)) {
+	if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
 		BUG_ON(timer_pending(timer));
 		BUG_ON(!list_empty(&work->entry));
 
 		/* This stores wq for the moment, for the timer_fn */
-		work->wq_data = wq;
+		set_wq_data(work, wq);
 		timer->expires = jiffies + delay;
 		timer->data = (unsigned long)dwork;
 		timer->function = delayed_work_timer_fn;
···
 		list_del_init(cwq->worklist.next);
 		spin_unlock_irqrestore(&cwq->lock, flags);
 
-		BUG_ON(work->wq_data != cwq);
-		clear_bit(0, &work->pending);
+		BUG_ON(get_wq_data(work) != cwq);
+		clear_bit(WORK_STRUCT_PENDING, &work->management);
 		f(data);
 
 		spin_lock_irqsave(&cwq->lock, flags);