cpu: make sure that cpu/online file created before KOBJ_ADD is emitted
authorIgor Mammedov <imammedo@redhat.com>
Tue, 14 May 2013 14:46:07 +0000 (16:46 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 21 May 2013 17:15:40 +0000 (10:15 -0700)
It fixes race between udev and hotplugged CPU registration by defining
"online" attribute statically, so that device_add() would create it
before notifying udev about new CPU.

Original issue report is at https://lkml.org/lkml/2012/4/30/198
"
> On Mon, Apr 30, 2012 at 11:36:23AM -0400, Konrad Rzeszutek Wilk wrote:
> > Hey Greg,
> >
> > Hoping you can help with some guidance on how to fix this.
> >
> > The issue is with CPU hotplug is that when a CPU goes up
> > it calls 'arch_register_cpu' which eventually calls
> > register_cpu. That function does these two things:
> >
> > 251         error = device_register(&cpu->dev);
> > 252         if (!error && cpu->hotpluggable)
> > 253                 register_cpu_control(cpu);
> >
> > and the device_register creates a nice little SysFS directory:
> >
> > /sys/devices/system/cpu/cpu2/ which at line 251 has the 'add' attribute
> > but no 'online' attribute. udev then tries to echo 1 to the 'online'
> > and it we get:
> > udevd-work[2421]: error opening ATTR{/sys/devices/system/cpu/cpu2/online} for writing: No such file or directory
> >
> > Line 253 creates said 'online' and at that time udev [or the system admin]
> > can write 1 to 'online' and the CPU goes up.
> >
> > So .. any thoughts? Is there some way to inhibit from uevent being sent
> > until line 253 has run?
"

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Tested-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/cpu.c

index 8f9e264..c377673 100644 (file)
@@ -85,18 +85,21 @@ static ssize_t __ref store_online(struct device *dev,
 }
 static DEVICE_ATTR(online, 0644, show_online, store_online);
 
-static void __cpuinit register_cpu_control(struct cpu *cpu)
-{
-       device_create_file(&cpu->dev, &dev_attr_online);
-}
+static struct attribute *hotplug_cpu_attrs[] = {
+       &dev_attr_online.attr,
+       NULL
+};
+
+static struct attribute_group hotplug_cpu_attr_group = {
+       .attrs = hotplug_cpu_attrs,
+};
+
 void unregister_cpu(struct cpu *cpu)
 {
        int logical_cpu = cpu->dev.id;
 
        unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
 
-       device_remove_file(&cpu->dev, &dev_attr_online);
-
        device_unregister(&cpu->dev);
        per_cpu(cpu_sys_devices, logical_cpu) = NULL;
        return;
@@ -122,11 +125,6 @@ static ssize_t cpu_release_store(struct device *dev,
 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
-
-#else /* ... !CONFIG_HOTPLUG_CPU */
-static inline void register_cpu_control(struct cpu *cpu)
-{
-}
 #endif /* CONFIG_HOTPLUG_CPU */
 
 #ifdef CONFIG_KEXEC
@@ -183,6 +181,16 @@ static const struct attribute_group *common_cpu_attr_groups[] = {
        NULL
 };
 
+static const struct attribute_group *hotplugable_cpu_attr_groups[] = {
+#ifdef CONFIG_KEXEC
+       &crash_note_cpu_attr_group,
+#endif
+#ifdef CONFIG_HOTPLUG_CPU
+       &hotplug_cpu_attr_group,
+#endif
+       NULL
+};
+
 /*
  * Print cpu online, possible, present, and system maps
  */
@@ -298,9 +306,9 @@ int __cpuinit register_cpu(struct cpu *cpu, int num)
        cpu->dev.bus->uevent = arch_cpu_uevent;
 #endif
        cpu->dev.groups = common_cpu_attr_groups;
+       if (cpu->hotpluggable)
+               cpu->dev.groups = hotplugable_cpu_attr_groups;
        error = device_register(&cpu->dev);
-       if (!error && cpu->hotpluggable)
-               register_cpu_control(cpu);
        if (!error)
                per_cpu(cpu_sys_devices, num) = &cpu->dev;
        if (!error)