Comment 65 for bug 245779

Here is the change log entry from the link I posted above:

Date: Mon Aug 17 14:34:59 2009 -0700

    clockevent: Prevent dead lock on clockevents_lock

    Currently clockevents_notify() is called with interrupts enabled at
    some places and interrupts disabled at some other places.

    This results in a deadlock in this scenario.

    cpu A holds clockevents_lock in clockevents_notify() with irqs enabled
    cpu B waits for clockevents_lock in clockevents_notify() with irqs disabled
    cpu C doing set_mtrr() which will try to rendezvous of all the cpus.

    This will result in C and A come to the rendezvous point and waiting
    for B. B is stuck forever waiting for the spinlock and thus not
    reaching the rendezvous point.

    Fix the clockevents code so that clockevents_lock is taken with
    interrupts disabled and thus avoid the above deadlock.

    Also call lapic_timer_propagate_broadcast() on the destination cpu so
    that we avoid calling smp_call_function() in the clockevents notifier
    chain.

    This issue left us wondering if we need to change the MTRR rendezvous
    logic to use stop machine logic (instead of smp_call_function) or add
    a check in spinlock debug code to see if there are other spinlocks
    which gets taken under both interrupts enabled/disabled conditions.

    Signed-off-by: Suresh Siddha <email address hidden>
    Signed-off-by: Venkatesh Pallipadi <email address hidden>
    Cc: "Pallipadi Venkatesh" <email address hidden>
    Cc: "Brown Len" <email address hidden>
    LKML-Reference: <email address hidden>
    Signed-off-by: Thomas Gleixner <email address hidden>