x86/mcheck: Move CPU_ONLINE and CPU_DOWN_PREPARE to hotplug state machine
authorSebastian Andrzej Siewior <bigeasy@linutronix.de>
Thu, 10 Nov 2016 17:44:46 +0000 (18:44 +0100)
committerThomas Gleixner <tglx@linutronix.de>
Wed, 16 Nov 2016 08:34:18 +0000 (09:34 +0100)
The CPU_ONLINE and CPU_DOWN_PREPARE look fully symmetrical and could be move
to the hotplug state machine.
On a failure during registration we have the tear down callback invoked
(mce_cpu_pre_down()) so there should be no timer around and so no need to need
keep notifier installed (this was the reason according to the comment why the
notifier was registered despite of errors).

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: rt@linutronix.de
Cc: linux-edac@vger.kernel.org
Link: http://lkml.kernel.org/r/20161110174447.11848-7-bigeasy@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
arch/x86/kernel/cpu/mcheck/mce.c

index a524faa5140034c39be1826e6eedf3c019506766..78955f501ff232f13e6bea6b139d29da3e9edc5c 100644 (file)
@@ -2511,21 +2511,8 @@ static int
 mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 {
        unsigned int cpu = (unsigned long)hcpu;
-       struct timer_list *t = &per_cpu(mce_timer, cpu);
 
        switch (action & ~CPU_TASKS_FROZEN) {
-       case CPU_ONLINE:
-       case CPU_DOWN_FAILED:
-
-               mce_device_create(cpu);
-
-               if (mce_threshold_create_device(cpu)) {
-                       mce_device_remove(cpu);
-                       return NOTIFY_BAD;
-               }
-               mce_reenable_cpu();
-               mce_start_timer(cpu, t);
-               break;
        case CPU_DEAD:
                mce_intel_hcpu_update(cpu);
 
@@ -2534,17 +2521,41 @@ mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
                        cmci_rediscover();
                break;
        case CPU_DOWN_PREPARE:
-               mce_disable_cpu();
-               del_timer_sync(t);
 
-               mce_threshold_remove_device(cpu);
-               mce_device_remove(cpu);
                break;
        }
 
        return NOTIFY_OK;
 }
 
+static int mce_cpu_online(unsigned int cpu)
+{
+       struct timer_list *t = &per_cpu(mce_timer, cpu);
+       int ret;
+
+       mce_device_create(cpu);
+
+       ret = mce_threshold_create_device(cpu);
+       if (ret) {
+               mce_device_remove(cpu);
+               return ret;
+       }
+       mce_reenable_cpu();
+       mce_start_timer(cpu, t);
+       return 0;
+}
+
+static int mce_cpu_pre_down(unsigned int cpu)
+{
+       struct timer_list *t = &per_cpu(mce_timer, cpu);
+
+       mce_disable_cpu();
+       del_timer_sync(t);
+       mce_threshold_remove_device(cpu);
+       mce_device_remove(cpu);
+       return 0;
+}
+
 static struct notifier_block mce_cpu_notifier = {
        .notifier_call = mce_cpu_callback,
 };
@@ -2569,8 +2580,8 @@ static __init void mce_init_banks(void)
 
 static __init int mcheck_init_device(void)
 {
+       enum cpuhp_state hp_online;
        int err;
-       int i = 0;
 
        if (!mce_available(&boot_cpu_data)) {
                err = -EIO;
@@ -2588,21 +2599,13 @@ static __init int mcheck_init_device(void)
        if (err)
                goto err_out_mem;
 
-       cpu_notifier_register_begin();
-       for_each_online_cpu(i) {
-               err = mce_device_create(i);
-               if (err) {
-                       /*
-                        * Register notifier anyway (and do not unreg it) so
-                        * that we don't leave undeleted timers, see notifier
-                        * callback above.
-                        */
-                       __register_hotcpu_notifier(&mce_cpu_notifier);
-                       cpu_notifier_register_done();
-                       goto err_device_create;
-               }
-       }
+       err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/mce:online",
+                               mce_cpu_online, mce_cpu_pre_down);
+       if (err < 0)
+               goto err_out_mem;
+       hp_online = err;
 
+       cpu_notifier_register_begin();
        __register_hotcpu_notifier(&mce_cpu_notifier);
        cpu_notifier_register_done();
 
@@ -2617,16 +2620,7 @@ static __init int mcheck_init_device(void)
 
 err_register:
        unregister_syscore_ops(&mce_syscore_ops);
-
-err_device_create:
-       /*
-        * We didn't keep track of which devices were created above, but
-        * even if we had, the set of online cpus might have changed.
-        * Play safe and remove for every possible cpu, since
-        * mce_device_remove() will do the right thing.
-        */
-       for_each_possible_cpu(i)
-               mce_device_remove(i);
+       cpuhp_remove_state(hp_online);
 
 err_out_mem:
        free_cpumask_var(mce_device_initialized);