The timer migration expiry check should prevent the migration of a
timer to another CPU when the timer expires before the next event is
scheduled on the other CPU. Migrating the timer might delay it because
we can not reprogram the clock event device on the other CPU. But the
code implementing that check has two flaws:

- for !HIGHRES the check compares the expiry value with the clock
  events device expiry value which is wrong for CLOCK_REALTIME based
  timers.

- the check is racy. It holds the hrtimer base lock of the target CPU,
  but the clock event device expiry value can be modified
  nevertheless, e.g. by an timer interrupt firing.

The !HIGHRES case is easy to fix as we can enqueue the timer on the
cpu which was selected by the load balancer. It runs the idle
balancing code once per jiffy anyway. So the maximum delay for the
timer is the same as when we keep the tick on the current cpu going.

In the HIGHRES case we can get the next expiry value from the hrtimer
cpu_base of the target CPU and serialize the update with the cpu_base
lock. This moves the lock section in hrtimer_interrupt() so we can set
next_event to KTIME_MAX while we are handling the expired timers and
set it to the next expiry value after we handled the timers under the
base lock. While the expired timers are processed timer migration is
blocked because the expiry time of the timer is always <= KTIME_MAX.

Also remove the now useless clockevents_get_next_event() function.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

The timer migration code needs to check whether the expiry time of the
timer is before the programmed clock event expiry time when the timer
is enqueued on another CPU because we can not reprogram the timer
device on the other CPU. The current logic checks the expiry time even
if we enqueue on the current CPU when nohz_get_load_balancer() returns
current CPU. This might lead to an endless loop in the expiry check
code when the expiry time of the timer is before the current
programmed next event.

Check whether nohz_get_load_balancer() returns current CPU and skip
the expiry check if this is the case.

The bug was triggered from the networking code. The patch fixes the
regression http://bugzilla.kernel.org/show_bug.cgi?id=13738
(Soft-Lockup/Race in networking in 2.6.31-rc1+195)

Cc: Arun Bharadwaj <arun@linux.vnet.ibm.com
Tested-by: Joao Correia <joaomiguelcorreia@gmail.com>
Tested-by: Andres Freund <andres@anarazel.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

* 'linux-next' of git://git.infradead.org/ubifs-2.6:
  UBIFS: start using hrtimers
  hrtimer: export ktime_add_safe
  UBIFS: do not forget to register BDI device
  UBIFS: allow sync option in rootflags
  UBIFS: remove dead code
  UBIFS: use anonymous device
  UBIFS: return proper error code if the compr is not present
  UBIFS: return error if link and unlink race
  UBIFS: reset no_space flag after inode deletion

We want to use hrtimers in UBIFS (for write-buffer write-back timer).
We need the 'hrtimer_set_expires_range_ns()', which is an in-line
function which uses 'ktime_add_safe()'.

Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Acked-by: Ingo Molnar <mingo@elte.hu>

* Arun R Bharadwaj <arun@linux.vnet.ibm.com> [2009-04-16 12:11:36]:

This patch migrates all non pinned timers and hrtimers to the current
idle load balancer, from all the idle CPUs. Timers firing on busy CPUs
are not migrated.

While migrating hrtimers, care should be taken to check if migrating
a hrtimer would result in a latency or not. So we compare the expiry of the
hrtimer with the next timer interrupt on the target cpu and migrate the
hrtimer only if it expires *after* the next interrupt on the target cpu.
So, added a clockevents_get_next_event() helper function to return the
next_event on the target cpu's clock_event_device.

[ tglx: cleanups and simplifications ]

Signed-off-by: Arun R Bharadwaj <arun@linux.vnet.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

* Arun R Bharadwaj <arun@linux.vnet.ibm.com> [2009-04-16 12:11:36]:

This patch creates a new framework for identifying cpu-pinned timers
and hrtimers.

This framework is needed because pinned timers are expected to fire on
the same CPU on which they are queued. So it is essential to identify
these and not migrate them, in case there are any.

For regular timers, the currently existing add_timer_on() can be used
queue pinned timers and subsequently mod_timer_pinned() can be used
to modify the 'expires' field.

For hrtimers, new modes HRTIMER_ABS_PINNED and HRTIMER_REL_PINNED are
added to queue cpu-pinned hrtimer.

[ tglx: use .._PINNED mode argument instead of creating tons of new
functions ]

Signed-off-by: Arun R Bharadwaj <arun@linux.vnet.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

It appears I inadvertly introduced rq->lock recursion to the
hrtimer_start() path when I delegated running already expired
timers to softirq context.

This patch fixes it by introducing a __hrtimer_start_range_ns()
method that will not use raise_softirq_irqoff() but
__raise_softirq_irqoff() which avoids the wakeup.

It then also changes schedule() to check for pending softirqs and
do the wakeup then, I'm not quite sure I like this last bit, nor
am I convinced its really needed.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: paulus@samba.org
LKML-Reference: <20090313112301.096138802@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Impact: prevent false positive WARN_ON() in clockevents_program_event()

clock_was_set() changes the base->offset of CLOCK_REALTIME and
enforces the reprogramming of the clockevent device to expire timers
which are based on CLOCK_REALTIME. If the clock change is large enough
then the subtraction of the timer expiry value and base->offset can
become negative which triggers the warning in
clockevents_program_event().

Check the subtraction result and set a negative value to 0.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Impact: fix CPU hotplug hang on Power6 testbox

On architectures that support offlining all cpus (at least powerpc/pseries),
hot-unpluging the tick_do_timer_cpu can result in a system hang.

This comes from the fact that if the cpu going down happens to be the
cpu doing the tick, then as the tick_do_timer_cpu handover happens after the
cpu is dead (via the CPU_DEAD notification), we're left without ticks,
jiffies are frozen and any task relying on timers (msleep, ...) is stuck.
That's particularly the case for the cpu looping in __cpu_die() waiting
for the dying cpu to be dead.

This patch addresses this by having the tick_do_timer_cpu handover happen
earlier during the CPU_DYING notification. For this, a new clockevent
notification type is introduced (CLOCK_EVT_NOTIFY_CPU_DYING) which is triggered
in hrtimer_cpu_notify().

Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Impact: avoid timer IRQ hanging slow systems

While using the function graph tracer on a virtualized system, the
hrtimer_interrupt can hang the system on an infinite loop.

This can be caused in several situations:

 - the hardware is very slow and HZ is set too high

 - something intrusive is slowing the system down (tracing under emulation)

... and the next clock events to program are always before the current time.

This patch implements a reasonable compromise: if such a situation is
detected, we share the CPUs time in 1/4 to process the hrtimer interrupts.
This is enough to let the system running without serious starvation.

It has been successfully tested under VirtualBox with 1000 HZ and 100 HZ
with function graph tracer launched. On both cases, the clock events were
increased until about 25 ms periodic ticks, which means 40 HZ.

So we change a hard to debug hang into a warning message and a system that
still manages to limp along.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>