ENGR00142351 mxc_gpu: autogating: add pending flag

pending indicate the timer has been fired but clock not yet disabled.

This patch fixs the bug that sometimes in irq handle it tries to enable
clock and cause BUG.
Signed-off-by: Richard Zhao <richard.zhao@freescale.com>

drivers/staging/amd-gpu/platform/hal/linux/misc.c

@@ -30,6 +30,8 @@ typedef struct _gsl_autogate_t {
     struct timer_list timer;	
     spinlock_t lock;
     int active;
+    /* pending indicate the timer has been fired but clock not yet disabled. */
+    int pending;
     int timeout;
     gsl_device_t *dev;
     struct work_struct dis_task;
@@ -47,6 +49,7 @@ static void clk_disable_task(struct work_struct *work)
 	if (autogate->dev->ftbl.device_idle)
 		autogate->dev->ftbl.device_idle(autogate->dev, GSL_TIMEOUT_DEFAULT);
 	kgsl_clock(autogate->dev->id, 0);
+	autogate->pending = 0;
 }
 
 static int _kgsl_device_active(gsl_device_t *dev, int all)
@@ -62,7 +65,7 @@ static int _kgsl_device_active(gsl_device_t *dev, int all)
 
 	spin_lock_irqsave(&autogate->lock, flags);
 	if (in_interrupt()) {
-		if (!autogate->active)
+		if (!autogate->active && !autogate->pending)
 			BUG();
 	} else {
 		to_active = !autogate->active;
@@ -99,6 +102,7 @@ static void kgsl_device_inactive(unsigned long data)
 	WARN(!autogate->active, "GPU Device %d is already inactive\n", autogate->dev->id);
 	if (autogate->active) {
 		autogate->active = 0;
+		autogate->pending = 1;
 		schedule_work(&autogate->dis_task);
 	}
 	spin_unlock_irqrestore(&autogate->lock, flags);
@@ -128,7 +132,7 @@ int kgsl_device_autogate_init(gsl_device_t *dev)
 	gsl_autogate_t *autogate;
 
 //	printk(KERN_ERR "%s:%d id %d\n", __func__, __LINE__, dev->id);
-	autogate = kmalloc(sizeof(gsl_autogate_t), GFP_KERNEL);
+	autogate = kzalloc(sizeof(gsl_autogate_t), GFP_KERNEL);
 	if (!autogate) {
 		printk(KERN_ERR "%s: out of memory!\n", __func__);
 		return -ENOMEM;