[CPUFREQ] Bail out of cpufreq_add_dev if the link for a managed CPU got created
Thomas Renninger [Fri, 24 Jul 2009 13:25:03 +0000 (15:25 +0200)]
Doing:
echo 0 >cpu1/online
echo 1 >cpu1/online

on a managed CPU will result in:
Jul 22 15:15:37 linux kernel: [   80.013864] WARNING: at fs/sysfs/dir.c:487 sysfs_add_one+0xcf/0xe6()
Jul 22 15:15:37 linux kernel: [   80.013866] Hardware name: To Be Filled By O.E.M.
Jul 22 15:15:37 linux kernel: [   80.013868] sysfs: cannot create duplicate filename '/devices/system/cpu/cpu1/cpufreq'
Jul 22 15:15:37 linux kernel: [   80.013870] Modules linked in: powernow_k8
Jul 22 15:15:37 linux kernel: [   80.013874] Pid: 5750, comm: bash Not tainted 2.6.31-rc2 #40
Jul 22 15:15:37 linux kernel: [   80.013876] Call Trace:
Jul 22 15:15:37 linux kernel: [   80.013879]  [<ffffffff8112ebda>] ? sysfs_add_one+0xcf/0xe6
Jul 22 15:15:37 linux kernel: [   80.013884]  [<ffffffff81041926>] warn_slowpath_common+0x77/0xa4
Jul 22 15:15:37 linux kernel: [   80.013888]  [<ffffffff810419a0>] warn_slowpath_fmt+0x3c/0x3e
Jul 22 15:15:37 linux kernel: [   80.013891]  [<ffffffff8112ebda>] sysfs_add_one+0xcf/0xe6
Jul 22 15:15:37 linux kernel: [   80.013894]  [<ffffffff8112f213>] create_dir+0x58/0x87
Jul 22 15:15:37 linux kernel: [   80.013898]  [<ffffffff8112f27a>] sysfs_create_dir+0x38/0x4f
Jul 22 15:15:37 linux kernel: [   80.013902]  [<ffffffff811ffb8a>] kobject_add_internal+0x11f/0x1de
Jul 22 15:15:37 linux kernel: [   80.013905]  [<ffffffff811ffd21>] kobject_add_varg+0x41/0x4e
Jul 22 15:15:37 linux kernel: [   80.013908]  [<ffffffff811ffd7a>] kobject_init_and_add+0x4c/0x57
Jul 22 15:15:37 linux kernel: [   80.013913]  [<ffffffff810667bc>] ? mark_lock+0x22/0x228
Jul 22 15:15:37 linux kernel: [   80.013918]  [<ffffffff813e8a3b>] cpufreq_add_dev_interface+0x40/0x1e4
...

This bug slipped in by git commit:
150b06f7f223cfd0f808737a5243cceca8ea47fa

When splitting up cpufreq_add_dev, the whole cpufreq_add_dev function
is not left anymore, only cpufreq_add_dev_policy.
This patch should reconstruct the identical functionality again as it
was before the split.

CC: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dave Jones <davej@redhat.com>

drivers/cpufreq/cpufreq.c

index 8456878..bbd5c21 100644 (file)
@@ -756,7 +756,12 @@ static struct kobj_type ktype_cpufreq = {
        .release        = cpufreq_sysfs_release,
 };
 
-
+/*
+ * Returns:
+ *   Negative: Failure
+ *   0:        Success
+ *   Positive: When we have a managed CPU and the sysfs got symlinked
+ */
 int cpufreq_add_dev_policy(unsigned int cpu, struct cpufreq_policy *policy,
                struct sys_device *sys_dev)
 {
@@ -817,7 +822,11 @@ int cpufreq_add_dev_policy(unsigned int cpu, struct cpufreq_policy *policy,
                         */
                        if (cpufreq_driver->exit)
                                cpufreq_driver->exit(policy);
-                       return ret;
+
+                       if (!ret)
+                               return 1;
+                       else
+                               return ret;
                }
        }
 #endif
@@ -1001,8 +1010,13 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
                                     CPUFREQ_START, policy);
 
        ret = cpufreq_add_dev_policy(cpu, policy, sys_dev);
-       if (ret)
+       if (ret) {
+               if (ret > 0)
+                       /* This is a managed cpu, symlink created,
+                          exit with 0 */
+                       ret = 0;
                goto err_unlock_policy;
+       }
 
        ret = cpufreq_add_dev_interface(cpu, policy, sys_dev);
        if (ret)