x86, pci, amd-bus: Fix CPU hotplug callback registration
authorSrivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Mon, 10 Mar 2014 20:38:43 +0000 (02:08 +0530)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Thu, 20 Mar 2014 12:43:43 +0000 (13:43 +0100)
Subsystems that want to register CPU hotplug callbacks, as well as perform
initialization for the CPUs that are already online, often do it as shown
below:

get_online_cpus();

for_each_online_cpu(cpu)
init_cpu(cpu);

register_cpu_notifier(&foobar_cpu_notifier);

put_online_cpus();

This is wrong, since it is prone to ABBA deadlocks involving the
cpu_add_remove_lock and the cpu_hotplug.lock (when running concurrently
with CPU hotplug operations).

Instead, the correct and race-free way of performing the callback
registration is:

cpu_notifier_register_begin();

for_each_online_cpu(cpu)
init_cpu(cpu);

/* Note the use of the double underscored version of the API */
__register_cpu_notifier(&foobar_cpu_notifier);

cpu_notifier_register_done();

Fix the amd-bus code in x86 by using this latter form of callback
registration.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
arch/x86/pci/amd_bus.c

index a48be98e9ded5a48be34fd8139d37a97a3748cc4..f05cbf0764d34c674f4bc9356052d1c39d7cc99a 100644 (file)
@@ -380,10 +380,13 @@ static int __init pci_io_ecs_init(void)
        if (early_pci_allowed())
                pci_enable_pci_io_ecs();
 
-       register_cpu_notifier(&amd_cpu_notifier);
+       cpu_notifier_register_begin();
        for_each_online_cpu(cpu)
                amd_cpu_notify(&amd_cpu_notifier, (unsigned long)CPU_ONLINE,
                               (void *)(long)cpu);
+       __register_cpu_notifier(&amd_cpu_notifier);
+       cpu_notifier_register_done();
+
        pci_probe |= PCI_HAS_IO_ECS;
 
        return 0;