Re: [PATCH] x86: fix cpu-hotplug regression

[Akinobu Mita]

> [PATCH] x86: fix cpu hotplug regression (don't call mce_create_device on CPU_UP_PREPARE)
>
> Fix regression introduced with d435d862baca3e25e5eec236762a43251b1e7ffc
> ("cpu hotplug: mce: fix cpu hotplug error handling").
>
> For CPUs not brought up during boot (using maxcpus and additional_cpus
> parameters) we don't know whether mce is supported or not at "CPU_UP_PREPARE"-time.
> Thus mce_cpu_callback should be called after the CPU is online.

Thank you for finding and fixing the problem.
I added two fixes to your patch:

- Avoid mce_remove_device() for the CPU that is not correctly initialized
  by mce_create_device() failure.

- make CPU_ONLINE callback always return NOTIFY_OK.
  Because CPU_ONLINE callback return value is always ignored.

> Signed-off-by: Andreas Herrmann <andreas.herrmann3@xxxxxxx>

[akinobu.mita@xxxxxxxxx: make CPU_ONLINE callback always return NOTIFY_OK]
[akinobu.mita@xxxxxxxxx: avoid mce_remove_device() for not initialized device]

Signed-off-by: Akinobu Mita <akinobu.mita@xxxxxxxxx>

---
 arch/x86/kernel/cpu/mcheck/mce_64.c |   18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

Index: 2.6-git/arch/x86/kernel/cpu/mcheck/mce_64.c
===================================================================
--- 2.6-git.orig/arch/x86/kernel/cpu/mcheck/mce_64.c
+++ 2.6-git/arch/x86/kernel/cpu/mcheck/mce_64.c
@@ -802,6 +802,8 @@ static struct sysdev_attribute *mce_attr
 	NULL
 };

+static cpumask_t mce_device_initialized = CPU_MASK_NONE;
+
 /* Per cpu sysdev init.  All of the cpus still share the same ctl bank */
 static __cpuinit int mce_create_device(unsigned int cpu)
 {
@@ -825,6 +827,7 @@ static __cpuinit int mce_create_device(u
 		if (err)
 			goto error;
 	}
+	cpu_set(cpu, mce_device_initialized);

 	return 0;
 error:
@@ -841,10 +844,14 @@ static void mce_remove_device(unsigned i
 {
 	int i;

+	if (!cpu_isset(cpu, mce_device_initialized))
+		return;
+
 	for (i = 0; mce_attributes[i]; i++)
 		sysdev_remove_file(&per_cpu(device_mce,cpu),
 			mce_attributes[i]);
 	sysdev_unregister(&per_cpu(device_mce,cpu));
+	cpu_clear(cpu, mce_device_initialized);
 }

 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
@@ -852,21 +859,18 @@ static int
 mce_cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
 {
 	unsigned int cpu = (unsigned long)hcpu;
-	int err = 0;

 	switch (action) {
-	case CPU_UP_PREPARE:
-	case CPU_UP_PREPARE_FROZEN:
-		err = mce_create_device(cpu);
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+		mce_create_device(cpu);
 		break;
-	case CPU_UP_CANCELED:
-	case CPU_UP_CANCELED_FROZEN:
 	case CPU_DEAD:
 	case CPU_DEAD_FROZEN:
 		mce_remove_device(cpu);
 		break;
 	}
-	return err ? NOTIFY_BAD : NOTIFY_OK;
+	return NOTIFY_OK;
 }

 static struct notifier_block mce_cpu_notifier = {
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/