[SCSI] qla2xxx: correct locking while call starget_for_each_device()

Removed spin_unlock_irq()/spin_lock_irq() pairs surrounding
starget_for_each_device() calls.
As Matthew W. pointed out, starget_for_each_device() can be called under
a spinlock being held.
The change has been tested and verified on qla2xxx.ko module.
Thanks Matthew W. and Hisashi H. for help.

Signed-off-by: Andrew Vasquez <Andrew.vasquez@qlogic.com>
Signed-off-by: Seokmann Ju <Seokmann.ju@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

drivers/scsi/qla2xxx/qla_isr.c

@@ -650,10 +650,8 @@ qla2x00_ramp_up_queue_depth(scsi_qla_host_t *ha, srb_t *sp)
 	    fcport->last_queue_full + ql2xqfullrampup * HZ))
 		return;
 
-	spin_unlock_irq(&ha->hardware_lock);
 	starget_for_each_device(sdev->sdev_target, fcport,
 	    qla2x00_adjust_sdev_qdepth_up);
-	spin_lock_irq(&ha->hardware_lock);
 }
 
 /**
@@ -923,10 +921,8 @@ qla2x00_status_entry(scsi_qla_host_t *ha, void *pkt)
 
 			/* Adjust queue depth for all luns on the port. */
 			fcport->last_queue_full = jiffies;
-			spin_unlock_irq(&ha->hardware_lock);
 			starget_for_each_device(cp->device->sdev_target,
 			    fcport, qla2x00_adjust_sdev_qdepth_down);
-			spin_lock_irq(&ha->hardware_lock);
 			break;
 		}
 		if (lscsi_status != SS_CHECK_CONDITION)