When a task is part of a numa_group, the comparison should always use
the group weight, in order to make workloads converge.
Signed-off-by: NRik van Riel <riel@redhat.com>
Cc: chegu_vinod@hp.com
Cc: mgorman@suse.de
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1403538378-31571-4-git-send-email-riel@redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

When CONFIG_FAIR_GROUP_SCHED is enabled, the load that a task places
on a CPU is determined by the group the task is in. The active groups
on the source and destination CPU can be different, resulting in a
different load contribution by the same task at its source and at its
destination. As a result, the load needs to be calculated separately
for each CPU, instead of estimated once with task_h_load().

Getting this calculation right allows some workloads to converge,
where previously the last thread could get stuck on another node,
without being able to migrate to its final destination.
Signed-off-by: NRik van Riel <riel@redhat.com>
Cc: mgorman@suse.de
Cc: chegu_vinod@hp.com
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1403538378-31571-3-git-send-email-riel@redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

Currently the NUMA code scales the load on each node with the
amount of CPU power available on that node, but it does not
apply any adjustment to the load of the task that is being
moved over.

On systems with SMT/HT, this results in a task being weighed
much more heavily than a CPU core, and a task move that would
even out the load between nodes being disallowed.

The correct thing is to apply the power correction to the
numbers after we have first applied the move of the tasks'
loads to them.

This also allows us to do the power correction with a multiplication,
rather than a division.

Also drop two function arguments for load_too_unbalanced, since it
takes various factors from env already.
Signed-off-by: NRik van Riel <riel@redhat.com>
Cc: chegu_vinod@hp.com
Cc: mgorman@suse.de
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1403538378-31571-2-git-send-email-riel@redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

From task_numa_placement, always try to consolidate the tasks
in a group on the group's top nid.

In case this task is part of a group that is interleaved over
multiple nodes, task_numa_migrate will set the task's preferred
nid to the best node it could find for the task, so this patch
will cause at most one run through task_numa_migrate.
Signed-off-by: NRik van Riel <riel@redhat.com>
Cc: mgorman@suse.de
Cc: chegu_vinod@hp.com
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1403538095-31256-2-git-send-email-riel@redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

When a system is lightly loaded (i.e. no more than 1 job per cpu),
attempt to pull job to a cpu before putting it to idle is unnecessary and
can be skipped.  This patch adds an indicator so the scheduler can know
when there's no more than 1 active job is on any CPU in the system to
skip needless job pulls.

On a 4 socket machine with a request/response kind of workload from
clients, we saw about 0.13 msec delay when we go through a full load
balance to try pull job from all the other cpus.  While 0.1 msec was
spent on processing the request and generating a response, the 0.13 msec
load balance overhead was actually more than the actual work being done.
This overhead can be skipped much of the time for lightly loaded systems.

With this patch, we tested with a netperf request/response workload that
has the server busy with half the cpus in a 4 socket system.  We found
the patch eliminated 75% of the load balance attempts before idling a cpu.

The overhead of setting/clearing the indicator is low as we already gather
the necessary info while we call add_nr_running() and update_sd_lb_stats.()
We switch to full load balance load immediately if any cpu got more than
one job on its run queue in add_nr_running.  We'll clear the indicator
to avoid load balance when we detect no cpu's have more than one job
when we scan the work queues in update_sg_lb_stats().  We are aggressive
in turning on the load balance and opportunistic in skipping the load
balance.
Signed-off-by: NTim Chen <tim.c.chen@linux.intel.com>
Acked-by: NRik van Riel <riel@redhat.com>
Acked-by: NJason Low <jason.low2@hp.com>
Cc: "Paul E.McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <davidlohr@hp.com>
Cc: Alex Shi <alex.shi@linaro.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: Peter Hurley <peter@hurleysoftware.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1403551009.2970.613.camel@schen9-DESKSigned-off-by: NIngo Molnar <mingo@kernel.org>

distribute_cfs_runtime() intentionally only hands out enough runtime to
bring each cfs_rq to 1 ns of runtime, expecting the cfs_rqs to then take
the runtime they need only once they actually get to run. However, if
they get to run sufficiently quickly, the period timer is still in
distribute_cfs_runtime() and no runtime is available, causing them to
throttle. Then distribute has to handle them again, and this can go on
until distribute has handed out all of the runtime 1ns at a time, which
takes far too long.

Instead allow access to the same runtime that distribute is handing out,
accepting that corner cases with very low quota may be able to spend the
entire cfs_b->runtime during distribute_cfs_runtime, meaning that the
runtime directly handed out by distribute_cfs_runtime was over quota. In
addition, if a cfs_rq does manage to throttle like this, make sure the
existing distribute_cfs_runtime no longer loops over it again.
Signed-off-by: NBen Segall <bsegall@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20140620222120.13814.21652.stgit@sword-of-the-dawn.mtv.corp.google.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

When computing cache hot, we should check if the migration dst cpu is idle,
instead of the current cpu. Though they are same in normal balancing, that
is false nowadays in nohz idle balancing at least.
Signed-off-by: NHillf Danton <dhillf@gmail.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Mike Galbraith <mgalbraith@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20140607090452.4696E301D2@webmail.sinamail.sina.com.cnSigned-off-by: NIngo Molnar <mingo@kernel.org>

It is possible that at task_numa_placement() time, the task's
numa_preferred_nid does not change, but the task is not
actually running on the preferred node at the time.

In that case, we still want to attempt migration to the
preferred node.
Signed-off-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: mgorman@suse.de
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20140604163315.1dbc7b56@cuia.bos.redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

The first thing task_numa_migrate() does is check to see if there is
CPU capacity available on the preferred node, in order to move the
task there.

However, if the preferred node is all busy, we would skip considering
that node for tasks swaps in the subsequent loop. This prevents NUMA
convergence of tasks on busy systems.

However, swapping locations with a task on our preferred nid, when
the preferred nid is busy, is perfectly fine.

The fix is to also look for a CPU on our preferred nid when it is
totally busy.

This changes "perf bench numa mem -p 4 -t 20 -m -0 -P 1000" from
not converging in 15 minutes on my 4 node system, to converging in
10-20 seconds.
Signed-off-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: mgorman@suse.de
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20140604160942.6969b101@cuia.bos.redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

This function is supposed to return true if the new load imbalance is
worse than the old one.  It didn't.  I can only hope brown paper bags
are in style.

Now things converge much better on both the 4 node and 8 node systems.

I am not sure why this did not seem to impact specjbb performance on the
4 node system, which is the system I have full-time access to.

This bug was introduced recently, with commit e63da036 ("sched/numa:
Allow task switch if load imbalance improves")
Signed-off-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is better not to think about compute capacity as being equivalent
to "CPU power".  The upcoming "power aware" scheduler work may create
confusion with the notion of energy consumption if "power" is used too
liberally.

Let's rename the following feature flags since they do relate to capacity:

	SD_SHARE_CPUPOWER  -> SD_SHARE_CPUCAPACITY
	ARCH_POWER         -> ARCH_CAPACITY
	NONTASK_POWER      -> NONTASK_CAPACITY
Signed-off-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Andy Fleming <afleming@freescale.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Grant Likely <grant.likely@linaro.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: devicetree@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Link: http://lkml.kernel.org/n/tip-e93lpnxb87owfievqatey6b5@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

It is better not to think about compute capacity as being equivalent
to "CPU power".  The upcoming "power aware" scheduler work may create
confusion with the notion of energy consumption if "power" is used too
liberally.

This contains the architecture visible changes.  Incidentally, only ARM
takes advantage of the available pow^H^H^Hcapacity scaling hooks and
therefore those changes outside kernel/sched/ are confined to one ARM
specific file.  The default arch_scale_smt_power() hook is not overridden
by anyone.

Replacements are as follows:

	arch_scale_freq_power  --> arch_scale_freq_capacity
	arch_scale_smt_power   --> arch_scale_smt_capacity
	SCHED_POWER_SCALE      --> SCHED_CAPACITY_SCALE
	SCHED_POWER_SHIFT      --> SCHED_CAPACITY_SHIFT

The local usage of "power" in arch/arm/kernel/topology.c is also changed
to "capacity" as appropriate.
Signed-off-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Grant Likely <grant.likely@linaro.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mark Brown <broonie@linaro.org>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Sudeep KarkadaNagesha <sudeep.karkadanagesha@arm.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: devicetree@vger.kernel.org
Cc: linux-arm-kernel@lists.infradead.org
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/n/tip-48zba9qbznvglwelgq2cfygh@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

It is better not to think about compute capacity as being equivalent
to "CPU power".  The upcoming "power aware" scheduler work may create
confusion with the notion of energy consumption if "power" is used too
liberally.

This is the remaining "power" -> "capacity" rename for local symbols.
Those symbols visible to the rest of the kernel are not included yet.
Signed-off-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/n/tip-yyyhohzhkwnaotr3lx8zd5aa@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

It is better not to think about compute capacity as being equivalent
to "CPU power".  The upcoming "power aware" scheduler work may create
confusion with the notion of energy consumption if "power" is used too
liberally.

Since struct sched_group_power is really about compute capacity of sched
groups, let's rename it to struct sched_group_capacity. Similarly sgp
becomes sgc. Related variables and functions dealing with groups are also
adjusted accordingly.
Signed-off-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/n/tip-5yeix833vvgf2uyj5o36hpu9@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

We have "power" (which should actually become "capacity") and "capacity"
which is a scaled down "capacity factor" in terms of unitary tasks.
Let's use "capacity_factor" to make room for proper usage of "capacity"
later.
Signed-off-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/n/tip-gk1co8sqdev3763opqm6ovml@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

The capacity of a CPU/group should be some intrinsic value that doesn't
change with task placement.  It is like a container which capacity is
stable regardless of the amount of liquid in it (its "utilization")...
unless the container itself is crushed that is, but that's another story.

Therefore let's rename "has_capacity" to "has_free_capacity" in order to
better convey the wanted meaning.
Signed-off-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/n/tip-djzkk027jm0e8x8jxy70opzh@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

It is better not to think about compute capacity as being equivalent
to "CPU power".  The upcoming "power aware" scheduler work may create
confusion with the notion of energy consumption if "power" is used too
liberally.

To make things explicit and not create more confusion with the existing
"capacity" member, let's rename things as follows:

	power    -> compute_capacity
	capacity -> task_capacity

Note: none of those fields are actually used outside update_numa_stats().
Signed-off-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/n/tip-2e2ndymj5gyshyjq8am79f20@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

To be future-proof and for better readability the time comparisons are modified
to use time_after() instead of plain, error-prone math.
Signed-off-by: NManuel Schölling <manuel.schoelling@gmx.de>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1400780723-24626-1-git-send-email-manuel.schoelling@gmx.deSigned-off-by: NIngo Molnar <mingo@kernel.org>

The current no_hz idle load balancer do load balancing for *all* idle cpus,
even though the time due to load balance for a particular
idle cpu could be still a while in the future.  This introduces a much
higher load balancing rate than what is necessary.  The patch
changes the behavior by only doing idle load balancing on
behalf of an idle cpu only when it is due for load balancing.

On SGI's systems with over 3000 cores, the cpu responsible for idle balancing
got overwhelmed with idle balancing, and introduces a lot of OS noise
to workloads.  This patch fixes the issue.
Signed-off-by: NTim Chen <tim.c.chen@linux.intel.com>
Acked-by: NRuss Anderson <rja@sgi.com>
Reviewed-by: NRik van Riel <riel@redhat.com>
Reviewed-by: NJason Low <jason.low2@hp.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Len Brown <len.brown@intel.com>
Cc: Dimitri Sivanich <sivanich@sgi.com>
Cc: Hedi Berriche <hedi@sgi.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: MichelLespinasse <walken@google.com>
Cc: Peter Hurley <peter@hurleysoftware.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1400621967.2970.280.camel@schen9-DESKSigned-off-by: NIngo Molnar <mingo@kernel.org>

sched_cfs_period_timer() reads cfs_b->period without locks before calling
do_sched_cfs_period_timer(), and similarly unthrottle_offline_cfs_rqs()
would read cfs_b->period without the right lock. Thus a simultaneous
change of bandwidth could cause corruption on any platform where ktime_t
or u64 writes/reads are not atomic.

Extend cfs_b->lock from do_sched_cfs_period_timer() to include the read of
cfs_b->period to solve that issue; unthrottle_offline_cfs_rqs() can just
use 1 rather than the exact quota, much like distribute_cfs_runtime()
does.

There is also an unlocked read of cfs_b->runtime_expires, but a race
there would only delay runtime expiry by a tick. Still, the comparison
should just be != anyway, which clarifies even that problem.
Signed-off-by: NBen Segall <bsegall@google.com>
Tested-by: NRoman Gushchin <klamm@yandex-team.ru>
[peterz: Fix compile warn]
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20140519224945.20303.93530.stgit@sword-of-the-dawn.mtv.corp.google.com
Cc: pjt@google.com
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NIngo Molnar <mingo@kernel.org>

tg_set_cfs_bandwidth() sets cfs_b->timer_active to 0 to
force the period timer restart. It's not safe, because
can lead to deadlock, described in commit 927b54fc:
"__start_cfs_bandwidth calls hrtimer_cancel while holding rq->lock,
waiting for the hrtimer to finish. However, if sched_cfs_period_timer
runs for another loop iteration, the hrtimer can attempt to take
rq->lock, resulting in deadlock."

Three CPUs must be involved:

  CPU0               CPU1                         CPU2
  take rq->lock      period timer fired
  ...                take cfs_b lock
  ...                ...                          tg_set_cfs_bandwidth()
  throttle_cfs_rq()  release cfs_b lock           take cfs_b lock
  ...                distribute_cfs_runtime()     timer_active = 0
  take cfs_b->lock   wait for rq->lock            ...
  __start_cfs_bandwidth()
  {wait for timer callback
   break if timer_active == 1}

So, CPU0 and CPU1 are deadlocked.

Instead of resetting cfs_b->timer_active, tg_set_cfs_bandwidth can
wait for period timer callbacks (ignoring cfs_b->timer_active) and
restart the timer explicitly.
Signed-off-by: NRoman Gushchin <klamm@yandex-team.ru>
Reviewed-by: NBen Segall <bsegall@google.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/87wqdi9g8e.wl\%klamm@yandex-team.ru
Cc: pjt@google.com
Cc: chris.j.arges@canonical.com
Cc: gregkh@linuxfoundation.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NIngo Molnar <mingo@kernel.org>

As Peter Zijlstra told me, we have the following path:

do_exit()
  exit_itimers()
    itimer_delete()
      spin_lock_irqsave(&timer->it_lock, &flags);
      timer_delete_hook(timer);
        kc->timer_del(timer) := posix_cpu_timer_del()
          put_task_struct()
            __put_task_struct()
              task_numa_free()
                spin_lock(&grp->lock);

Which means that task_numa_free() can be called with interrupts
disabled, which means that we should not be using spin_lock_irq() but
spin_lock_irqsave() instead. Otherwise we are enabling interrupts while
holding an interrupt unsafe lock!
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner<tglx@linutronix.de>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20140527182541.GH11096@twins.programming.kicks-ass.netSigned-off-by: NIngo Molnar <mingo@kernel.org>

Affine wakeups have the potential to interfere with NUMA placement.
If a task wakes up too many other tasks, affine wakeups will get
disabled.

However, regardless of how many other tasks it wakes up, it gets
re-enabled once a second, potentially interfering with NUMA
placement of other tasks.

By decaying wakee_wakes in half instead of zeroing it, we can avoid
that problem for some workloads.
Signed-off-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: chegu_vinod@hp.com
Cc: umgwanakikbuti@gmail.com
Link: http://lkml.kernel.org/r/20140516001332.67f91af2@annuminas.surriel.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

Update the migrate_improves/degrades_locality() functions with
knowledge of pseudo-interleaving.

Do not consider moving tasks around within the set of group's active
nodes as improving or degrading locality. Instead, leave the load
balancer free to balance the load between a numa_group's active nodes.

Also, switch from the group/task_weight functions to the group/task_fault
functions. The "weight" functions involve a division, but both calls use
the same divisor, so there's no point in doing that from these functions.

On a 4 node (x10 core) system, performance of SPECjbb2005 seems
unaffected, though the number of migrations with 2 8-warehouse wide
instances seems to have almost halved, due to the scheduler running
each instance on a single node.
Signed-off-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: mgorman@suse.de
Cc: chegu_vinod@hp.com
Link: http://lkml.kernel.org/r/20140515130306.61aae7db@cuia.bos.redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

Currently the NUMA balancing code only allows moving tasks between NUMA
nodes when the load on both nodes is in balance. This breaks down when
the load was imbalanced to begin with.

Allow tasks to be moved between NUMA nodes if the imbalance is small,
or if the new imbalance is be smaller than the original one.
Suggested-by: NPeter Zijlstra <peterz@infradead.org>
Signed-off-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: mgorman@suse.de
Cc: chegu_vinod@hp.com
Signed-off-by: NIngo Molnar <mingo@kernel.org>
Link: http://lkml.kernel.org/r/20140514132221.274b3463@annuminas.surriel.com

Sometimes ->nr_running may cross 2 but interrupt is not being
sent to rq's cpu. In this case we don't reenable the timer.
Looks like this may be the reason for rare unexpected effects,
if nohz is enabled.

Patch replaces all places of direct changing of nr_running
and makes add_nr_running() caring about crossing border.
Signed-off-by: NKirill Tkhai <tkhai@yandex.ru>
Acked-by: NFrederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20140508225830.2469.97461.stgit@localhostSigned-off-by: NIngo Molnar <mingo@kernel.org>

Currently, in idle_balance(), we update rq->next_balance when we pull_tasks.
However, it is also important to update this in the !pulled_tasks case too.

When the CPU is "busy" (the CPU isn't idle), rq->next_balance gets computed
using sd->busy_factor (so we increase the balance interval when the CPU is
busy). However, when the CPU goes idle, rq->next_balance could still be set
to a large value that was computed with the sd->busy_factor.

Thus, we need to also update rq->next_balance in idle_balance() in the cases
where !pulled_tasks too, so that rq->next_balance gets updated without taking
the busy_factor into account when the CPU is about to go idle.

This patch makes rq->next_balance get updated independently of whether or
not we pulled_task. Also, we add logic to ensure that we always traverse
at least 1 of the sched domains to get a proper next_balance value for
updating rq->next_balance.

Additionally, since load_balance() modifies the sd->balance_interval, we
need to re-obtain the sched domain's interval after the call to
load_balance() in rebalance_domains() before we update rq->next_balance.

This patch adds and uses 2 new helper functions, update_next_balance() and
get_sd_balance_interval() to update next_balance and obtain the sched
domain's balance_interval.
Signed-off-by: NJason Low <jason.low2@hp.com>
Reviewed-by: NPreeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: daniel.lezcano@linaro.org
Cc: alex.shi@linaro.org
Cc: efault@gmx.de
Cc: vincent.guittot@linaro.org
Cc: morten.rasmussen@arm.com
Cc: aswin@hp.com
Link: http://lkml.kernel.org/r/1399596562.2200.7.camel@j-VirtualBoxSigned-off-by: NIngo Molnar <mingo@kernel.org>

On smaller systems, the top level sched domain will be an affine
domain, and select_idle_sibling is invoked for every SD_WAKE_AFFINE
wakeup. This seems to be working well.

On larger systems, with the node distance between far away NUMA nodes
being > RECLAIM_DISTANCE, select_idle_sibling is only called if the
waker and the wakee are on nodes less than RECLAIM_DISTANCE apart.

This patch leaves in place the policy of not pulling the task across
nodes on such systems, while fixing the issue that select_idle_sibling
is not called at all in certain circumstances.

The code will look for an idle CPU in the same CPU package as the
CPU where the task ran previously.
Signed-off-by: NRik van Riel <riel@redhat.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: morten.rasmussen@arm.com
Cc: george.mccollister@gmail.com
Cc: ktkhai@parallels.com
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Link: http://lkml.kernel.org/r/20140514114037.2d93266f@annuminas.surriel.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

task_hot checks exec_start on any runnable task, but if it has been
migrated since the it last ran, then exec_start is a clock_task from
another cpu. If the old cpu's clock_task was sufficiently far ahead of
this cpu's then the task will not be considered for another migration
until it has run. Instead reset exec_start whenever a task is migrated,
since it is presumably no longer hot anyway.
Signed-off-by: NBen Segall <bsegall@google.com>
[ Made it compile. ]
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20140515225920.7179.13924.stgit@sword-of-the-dawn.mtv.corp.google.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

It was found that when running some workloads (such as AIM7) on large
systems with many cores, CPUs do not remain idle for long. Thus, tasks
can wake/get enqueued while doing idle balancing.

In this patch, while traversing the domains in idle balance, in
addition to checking for pulled_task, we add an extra check for
this_rq->nr_running for determining if we should stop searching for
tasks to pull. If there are runnable tasks on this rq, then we will
stop traversing the domains. This reduces the chance that idle balance
delays a task from running.

This patch resulted in approximately a 6% performance improvement when
running a Java Server workload on an 8 socket machine.
Signed-off-by: NJason Low <jason.low2@hp.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: daniel.lezcano@linaro.org
Cc: alex.shi@linaro.org
Cc: preeti@linux.vnet.ibm.com
Cc: efault@gmx.de
Cc: vincent.guittot@linaro.org
Cc: morten.rasmussen@arm.com
Cc: aswin@hp.com
Cc: chegu_vinod@hp.com
Link: http://lkml.kernel.org/r/1398303035-18255-4-git-send-email-jason.low2@hp.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

Setting the numa_preferred_node for a task in task_numa_migrate
does nothing on a 2-node system. Either we migrate to the node
that already was our preferred node, or we stay where we were.

On a 4-node system, it can slightly decrease overhead, by not
calling the NUMA code as much. Since every node tends to be
directly connected to every other node, running on the wrong
node for a while does not do much damage.

However, on an 8 node system, there are far more bad nodes
than there are good ones, and pretending that a second choice
is actually the preferred node can greatly delay, or even
prevent, a workload from converging.

The only time we can safely pretend that a second choice
node is the preferred node is when the task is part of a
workload that spans multiple NUMA nodes.
Signed-off-by: NRik van Riel <riel@redhat.com>
Tested-by: NVinod Chegu <chegu_vinod@hp.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1397235629-16328-4-git-send-email-riel@redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

When tasks have not converged on their preferred nodes yet, we want
to retry fairly often, to make sure we do not migrate a task's memory
to an undesirable location, only to have to move it again later.

This patch reduces the interval at which migration is retried,
when the task's numa_scan_period is small.
Signed-off-by: NRik van Riel <riel@redhat.com>
Tested-by: NVinod Chegu <chegu_vinod@hp.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1397235629-16328-3-git-send-email-riel@redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

The NUMA code is smart enough to distribute the memory of workloads
that span multiple NUMA nodes across those NUMA nodes.

However, it still has a pretty high scan rate for such workloads,
because any memory that is left on a node other than the node of
the CPU that faulted on the memory is counted as non-local, which
causes the scan rate to go up.

Counting the memory on any node where the task's numa group is
actively running as local, allows the scan rate to slow down
once the application is settled in.

This should reduce the overhead of the automatic NUMA placement
code, when a workload spans multiple NUMA nodes.
Signed-off-by: NRik van Riel <riel@redhat.com>
Tested-by: NVinod Chegu <chegu_vinod@hp.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1397235629-16328-2-git-send-email-riel@redhat.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

The following commit:

  e5fc6611 ("sched: Fix race in idle_balance()")

can potentially cause rq->max_idle_balance_cost to not be updated,
even when load_balance(NEWLY_IDLE) is attempted and the per-sd
max cost value is updated.

Preeti noticed a similar issue with updating rq->next_balance.

In this patch, we fix this by making sure we still check/update those values
even if a task gets enqueued while browsing the domains.
Signed-off-by: NJason Low <jason.low2@hp.com>
Reviewed-by: NPreeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: morten.rasmussen@arm.com
Cc: aswin@hp.com
Cc: daniel.lezcano@linaro.org
Cc: alex.shi@linaro.org
Cc: efault@gmx.de
Cc: vincent.guittot@linaro.org
Link: http://lkml.kernel.org/r/1398725155-7591-2-git-send-email-jason.low2@hp.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

This reverts commit 4c6c4e38 ("sched/core: Fix endless loop in
pick_next_task()"), which is not necessary after ("sched/rt: Substract number
of tasks of throttled queues from rq->nr_running").
Signed-off-by: NKirill Tkhai <tkhai@yandex.ru>
Reviewed-by: NPreeti U Murthy <preeti@linux.vnet.ibm.com>
[conflict resolution with stop task checking patch]
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1394835307.18748.34.camel@HP-250-G1-Notebook-PC
Cc: linux-kernel@vger.kernel.org
Signed-off-by: NIngo Molnar <mingo@kernel.org>

Mike reported that, while unlikely, its entirely possible for
scale_rt_power() to see the time go backwards. This yields rather
'interesting' results.

So like all other sites that deal with clocks; make this one ignore
backward clock movement too.
Reported-by: NMike Galbraith <bitbucket@online.de>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20140227094035.GZ9987@twins.programming.kicks-ass.net
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: NIngo Molnar <mingo@kernel.org>

We need to do it like we do for the other higher priority classes..
Signed-off-by: NKirill Tkhai <tkhai@yandex.ru>
Cc: Michael wang <wangyun@linux.vnet.ibm.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/336561397137116@web27h.yandex.ruSigned-off-by: NIngo Molnar <mingo@kernel.org>

Sasha reported that lockdep claims that the following commit:
made numa_group.lock interrupt unsafe:

  156654f4 ("sched/numa: Move task_numa_free() to __put_task_struct()")

While I don't see how that could be, given the commit in question moved
task_numa_free() from one irq enabled region to another, the below does
make both gripes and lockups upon gripe with numa=fake=4 go away.
Reported-by: NSasha Levin <sasha.levin@oracle.com>
Fixes: 156654f4 ("sched/numa: Move task_numa_free() to __put_task_struct()")
Signed-off-by: NMike Galbraith <bitbucket@online.de>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: torvalds@linux-foundation.org
Cc: mgorman@suse.com
Cc: akpm@linux-foundation.org
Cc: Dave Jones <davej@redhat.com>
Link: http://lkml.kernel.org/r/1396860915.5170.5.camel@marge.simpson.netSigned-off-by: NIngo Molnar <mingo@kernel.org>

task_hot() doesn't need the 'sched_domain' parameter, so remove it.
Signed-off-by: NAlex Shi <alex.shi@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1394607111-1904-1-git-send-email-alex.shi@linaro.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

The tmp value has been already calculated in:

  scaled_busy_load_per_task =
		(busiest->load_per_task * SCHED_POWER_SCALE) /
		busiest->group_power;
Signed-off-by: NVincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1394555166-22894-1-git-send-email-vincent.guittot@linaro.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>