cpu hotplug: cpu: deliver CPU_UP_CANCELED only to NOTIFY_OKed callbacks with CPU_UP_P...
authorAkinobu Mita <akinobu.mita@gmail.com>
Thu, 18 Oct 2007 10:05:12 +0000 (03:05 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 18 Oct 2007 21:37:21 +0000 (14:37 -0700)
The functions in a CPU notifier chain is called with CPU_UP_PREPARE event
before making the CPU online.  If one of the callback returns NOTIFY_BAD, it
stops to deliver CPU_UP_PREPARE event, and CPU online operation is canceled.
Then CPU_UP_CANCELED event is delivered to the functions in a CPU notifier
chain again.

This CPU_UP_CANCELED event is delivered to the functions which have been
called with CPU_UP_PREPARE, not delivered to the functions which haven't been
called with CPU_UP_PREPARE.

The problem that makes existing cpu hotplug error handlings complex is that
the CPU_UP_CANCELED event is delivered to the function that has returned
NOTIFY_BAD, too.

Usually we don't expect to call destructor function against the object that
has failed to initialize.  It is like:

err = register_something();
if (err) {
unregister_something();
return err;
}

So it is natural to deliver CPU_UP_CANCELED event only to the functions that
have returned NOTIFY_OK with CPU_UP_PREPARE event and not to call the function
that have returned NOTIFY_BAD.  This is what this patch is doing.

Otherwise, every cpu hotplug notifiler has to track whether notifiler event is
failed or not for each cpu.  (drivers/base/topology.c is doing this with
topology_dev_map)

Similary this patch makes same thing with CPU_DOWN_PREPARE and CPU_DOWN_FAILED
evnets.

Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/cpu.c

index 38033db8d8ec3d8f23af3973560dc995782efbda..a21f71af9d81fba3c0012afd8538d25a1f7c0c62 100644 (file)
@@ -150,6 +150,7 @@ static int _cpu_down(unsigned int cpu, int tasks_frozen)
        err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
                                        hcpu, -1, &nr_calls);
        if (err == NOTIFY_BAD) {
+               nr_calls--;
                __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
                                          hcpu, nr_calls, NULL);
                printk("%s: attempt to take down CPU %u failed\n",
@@ -233,6 +234,7 @@ static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
        ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
                                                        -1, &nr_calls);
        if (ret == NOTIFY_BAD) {
+               nr_calls--;
                printk("%s: attempt to bring up CPU %u failed\n",
                                __FUNCTION__, cpu);
                ret = -EINVAL;