pm_qos: remove BKL
Jonathan Corbet [Thu, 6 Aug 2009 19:22:40 +0000 (13:22 -0600)]
pm_qos_power_open got its lock_kernel() calls from the open() pushdown.  A
look at the code shows that the only global resources accessed are
pm_qos_array and "name".  pm_qos_array doesn't change (things pointed to
therein do change, but they are atomics and/or are protected by
pm_qos_lock).  Accesses to "name" are totally unprotected with or without
the BKL; that will be fixed shortly.  The BKL is not helpful here; take it
out.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>
LKML-Reference: <20091010153349.071381158@linutronix.de>
Acked-by: Mark Gross <mgross@linux.intel.com>
Reviewed-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

kernel/pm_qos_params.c

index dfdec52..d96b83e 100644 (file)
@@ -29,7 +29,6 @@
 
 #include <linux/pm_qos_params.h>
 #include <linux/sched.h>
-#include <linux/smp_lock.h>
 #include <linux/spinlock.h>
 #include <linux/slab.h>
 #include <linux/time.h>
@@ -352,20 +351,15 @@ static int pm_qos_power_open(struct inode *inode, struct file *filp)
        int ret;
        long pm_qos_class;
 
-       lock_kernel();
        pm_qos_class = find_pm_qos_object_by_minor(iminor(inode));
        if (pm_qos_class >= 0) {
                filp->private_data = (void *)pm_qos_class;
                sprintf(name, "process_%d", current->pid);
                ret = pm_qos_add_requirement(pm_qos_class, name,
                                        PM_QOS_DEFAULT_VALUE);
-               if (ret >= 0) {
-                       unlock_kernel();
+               if (ret >= 0)
                        return 0;
-               }
        }
-       unlock_kernel();
-
        return -EPERM;
 }