Commit a26ac2455ffcf3(rcu: move TREE_RCU from softirq to kthread)
introduced performance regression. In an AIM7 test, this commit degraded
performance by about 40%.

The commit runs rcu callbacks in a kthread instead of softirq. We observed
high rate of context switch which is caused by this. Out test system has
64 CPUs and HZ is 1000, so we saw more than 64k context switch per second
which is caused by RCU's per-CPU kthread.  A trace showed that most of
the time the RCU per-CPU kthread doesn't actually handle any callbacks,
but instead just does a very small amount of work handling grace periods.
This means that RCU's per-CPU kthreads are making the scheduler do quite
a bit of work in order to allow a very small amount of RCU-related
processing to be done.

Alex Shi's analysis determined that this slowdown is due to lock
contention within the scheduler.  Unfortunately, as Peter Zijlstra points
out, the scheduler's real-time semantics require global action, which
means that this contention is inherent in real-time scheduling.  (Yes,
perhaps someone will come up with a workaround -- otherwise, -rt is not
going to do well on large SMP systems -- but this patch will work around
this issue in the meantime.  And "the meantime" might well be forever.)

This patch therefore re-introduces softirq processing to RCU, but only
for core RCU work.  RCU callbacks are still executed in kthread context,
so that only a small amount of RCU work runs in softirq context in the
common case.  This should minimize ksoftirqd execution, allowing us to
skip boosting of ksoftirqd for CONFIG_RCU_BOOST=y kernels.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Tested-by: "Alex,Shi" <alex.shi@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

If RCU priority boosting is to be meaningful, callback invocation must
be boosted in addition to preempted RCU readers.  Otherwise, in presence
of CPU real-time threads, the grace period ends, but the callbacks don't
get invoked.  If the callbacks don't get invoked, the associated memory
doesn't get freed, so the system is still subject to OOM.

But it is not reasonable to priority-boost RCU_SOFTIRQ, so this commit
moves the callback invocations to a kthread, which can be boosted easily.

Also add comments and properly synchronized all accesses to
rcu_cpu_kthread_task, as suggested by Lai Jiangshan.

Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>

Fixes generated by 'codespell' and manually reviewed.

Signed-off-by: Lucas De Marchi <lucas.demarchi@profusion.mobi>

ksoftirqd, kworker, migration, and pktgend kthreads can be created with
kthread_create_on_node(), to get proper NUMA affinities for their stack and
task_struct.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: David S. Miller <davem@davemloft.net>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: David Howells <dhowells@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

* 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (116 commits)
  x86: Enable forced interrupt threading support
  x86: Mark low level interrupts IRQF_NO_THREAD
  x86: Use generic show_interrupts
  x86: ioapic: Avoid redundant lookup of irq_cfg
  x86: ioapic: Use new move_irq functions
  x86: Use the proper accessors in fixup_irqs()
  x86: ioapic: Use irq_data->state
  x86: ioapic: Simplify irq chip and handler setup
  x86: Cleanup the genirq name space
  genirq: Add chip flag to force mask on suspend
  genirq: Add desc->irq_data accessor
  genirq: Add comments to Kconfig switches
  genirq: Fixup fasteoi handler for oneshot mode
  genirq: Provide forced interrupt threading
  sched: Switch wait_task_inactive to schedule_hrtimeout()
  genirq: Add IRQF_NO_THREAD
  genirq: Allow shared oneshot interrupts
  genirq: Prepare the handling of shared oneshot interrupts
  genirq: Make warning in handle_percpu_event useful
  x86: ioapic: Move trigger defines to io_apic.h
  ...

Fix up trivial(?) conflicts in arch/x86/pci/xen.c due to genirq name
space changes clashing with the Xen cleanups.  The set_irq_msi() had
moved to xen_bind_pirq_msi_to_irq().

Add a commandline parameter "threadirqs" which forces all interrupts except
those marked IRQF_NO_THREAD to run threaded. That's mostly a debug option to
allow retrieving better debug data from crashing interrupt handlers. If
"threadirqs" is not enabled on the kernel command line, then there is no
impact in the interrupt hotpath.

Architecture code needs to select CONFIG_IRQ_FORCED_THREADING after
marking the interrupts which cant be threaded IRQF_NO_THREAD. All
interrupts which have IRQF_TIMER set are implict marked
IRQF_NO_THREAD. Also all PER_CPU interrupts are excluded.

Forced threading hard interrupts also forces all soft interrupt
handling into thread context.

When enabled it might slow down things a bit, but for debugging problems in
interrupt code it's a reasonable penalty as it does not immediately
crash and burn the machine when an interrupt handler is buggy.

Some test results on a Core2Duo machine:

Cache cold run of:
 # time git grep irq_desc

      non-threaded       threaded
 real 1m18.741s          1m19.061s
 user 0m1.874s           0m1.757s
 sys  0m5.843s           0m5.427s

 # iperf -c server
non-threaded
[  3]  0.0-10.0 sec  1.09 GBytes   933 Mbits/sec
[  3]  0.0-10.0 sec  1.09 GBytes   934 Mbits/sec
[  3]  0.0-10.0 sec  1.09 GBytes   933 Mbits/sec
threaded
[  3]  0.0-10.0 sec  1.09 GBytes   939 Mbits/sec
[  3]  0.0-10.0 sec  1.09 GBytes   934 Mbits/sec
[  3]  0.0-10.0 sec  1.09 GBytes   937 Mbits/sec

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <20110223234956.772668648@linutronix.de>

ksoftirqd() calls do_softirq() which switches stacks on several
architectures. That makes no sense at all. ksoftirqd's stack is
sufficient.

Call __do_softirq() directly.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: David Miller <davem@davemloft.net>
Cc: Paul Mundt <lethal@linux-sh.org>
Reviewed-by: Frank Rowand <frank.rowand@am.sony.com>
LKML-Reference: <alpine.LFD.2.00.1102021704530.31804@localhost6.localdomain6>

Cleanup patch, freeing up PF_KSOFTIRQD and use per_cpu ksoftirqd pointer
instead, as suggested by Eric Dumazet.

Tested-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1292980144-28796-2-git-send-email-venki@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

For arch which needs USE_GENERIC_SMP_HELPERS, it has to select
USE_GENERIC_SMP_HELPERS, rather than leaving a choice to user, since they
don't provide their own implementions.

Also, move on_each_cpu() to kernel/smp.c, it is strange to put it in
kernel/softirq.c.

For arch which doesn't use USE_GENERIC_SMP_HELPERS, e.g.  blackfin, only
on_each_cpu() is compiled.

Signed-off-by: Amerigo Wang <amwang@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

* 'for-2.6.38' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu: (30 commits)
  gameport: use this_cpu_read instead of lookup
  x86: udelay: Use this_cpu_read to avoid address calculation
  x86: Use this_cpu_inc_return for nmi counter
  x86: Replace uses of current_cpu_data with this_cpu ops
  x86: Use this_cpu_ops to optimize code
  vmstat: User per cpu atomics to avoid interrupt disable / enable
  irq_work: Use per cpu atomics instead of regular atomics
  cpuops: Use cmpxchg for xchg to avoid lock semantics
  x86: this_cpu_cmpxchg and this_cpu_xchg operations
  percpu: Generic this_cpu_cmpxchg() and this_cpu_xchg support
  percpu,x86: relocate this_cpu_add_return() and friends
  connector: Use this_cpu operations
  xen: Use this_cpu_inc_return
  taskstats: Use this_cpu_ops
  random: Use this_cpu_inc_return
  fs: Use this_cpu_inc_return in buffer.c
  highmem: Use this_cpu_xx_return() operations
  vmstat: Use this_cpu_inc_return for vm statistics
  x86: Support for this_cpu_add, sub, dec, inc_return
  percpu: Generic support for this_cpu_add, sub, dec, inc_return
  ...

Fixed up conflicts: in arch/x86/kernel/{apic/nmi.c, apic/x2apic_uv_x.c, process.c}
as per Tejun.