releases/2.6.36.2/nohz-s390-fix-arch_needs_cpu-return-value-on-offline-cpus.patch

   1 From 398812159e328478ae49b4bd01f0d71efea96c39 Mon Sep 17 00:00:00 2001
   2 From: Heiko Carstens <heiko.carstens@de.ibm.com>
   3 Date: Wed, 1 Dec 2010 10:08:01 +0100
   4 Subject: [S390] nohz/s390: fix arch_needs_cpu() return value on offline cpus
   5
   6 From: Heiko Carstens <heiko.carstens@de.ibm.com>
   7
   8 commit 398812159e328478ae49b4bd01f0d71efea96c39 upstream.
   9
  10 This fixes the same problem as described in the patch "nohz: fix
  11 printk_needs_cpu() return value on offline cpus" for the arch_needs_cpu()
  12 primitive:
  13
  14 arch_needs_cpu() may return 1 if called on offline cpus. When a cpu gets
  15 offlined it schedules the idle process which, before killing its own cpu,
  16 will call tick_nohz_stop_sched_tick().
  17 That function in turn will call arch_needs_cpu() in order to check if the
  18 local tick can be disabled. On offline cpus this function should naturally
  19 return 0 since regardless if the tick gets disabled or not the cpu will be
  20 dead short after. That is besides the fact that __cpu_disable() should already
  21 have made sure that no interrupts on the offlined cpu will be delivered anyway.
  22
  23 In this case it prevents tick_nohz_stop_sched_tick() to call
  24 select_nohz_load_balancer(). No idea if that really is a problem. However what
  25 made me debug this is that on 2.6.32 the function get_nohz_load_balancer() is
  26 used within __mod_timer() to select a cpu on which a timer gets enqueued.
  27 If arch_needs_cpu() returns 1 then the nohz_load_balancer cpu doesn't get
  28 updated when a cpu gets offlined. It may contain the cpu number of an offline
  29 cpu. In turn timers get enqueued on an offline cpu and not very surprisingly
  30 they never expire and cause system hangs.
  31
  32 This has been observed 2.6.32 kernels. On current kernels __mod_timer() uses
  33 get_nohz_timer_target() which doesn't have that problem. However there might
  34 be other problems because of the too early exit tick_nohz_stop_sched_tick()
  35 in case a cpu goes offline.
  36
  37 This specific bug was indrocuded with 3c5d92a0 "nohz: Introduce
  38 arch_needs_cpu".
  39
  40 In this case a cpu hotplug notifier is used to fix the issue in order to keep
  41 the normal/fast path small. All we need to do is to clear the condition that
  42 makes arch_needs_cpu() return 1 since it is just a performance improvement
  43 which is supposed to keep the local tick running for a short period if a cpu
  44 goes idle. Nothing special needs to be done except for clearing the condition.
  45
  46 Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
  47 Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
  48 Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
  49 Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
  50
  51 ---
  52  arch/s390/kernel/vtime.c |   19 +++++++++++++++++++
  53  1 file changed, 19 insertions(+)
  54
  55 --- a/arch/s390/kernel/vtime.c
  56 +++ b/arch/s390/kernel/vtime.c
  57 @@ -19,6 +19,7 @@
  58  #include <linux/kernel_stat.h>
  59  #include <linux/rcupdate.h>
  60  #include <linux/posix-timers.h>
  61 +#include <linux/cpu.h>
  62
  63  #include <asm/s390_ext.h>
  64  #include <asm/timer.h>
  65 @@ -565,6 +566,23 @@ void init_cpu_vtimer(void)
  66         __ctl_set_bit(0,10);
  67  }
  68
  69 +static int __cpuinit s390_nohz_notify(struct notifier_block *self,
  70 +                                     unsigned long action, void *hcpu)
  71 +{
  72 +       struct s390_idle_data *idle;
  73 +       long cpu = (long) hcpu;
  74 +
  75 +       idle = &per_cpu(s390_idle, cpu);
  76 +       switch (action) {
  77 +       case CPU_DYING:
  78 +       case CPU_DYING_FROZEN:
  79 +               idle->nohz_delay = 0;
  80 +       default:
  81 +               break;
  82 +       }
  83 +       return NOTIFY_OK;
  84 +}
  85 +
  86  void __init vtime_init(void)
  87  {
  88         /* request the cpu timer external interrupt */
  89 @@ -573,5 +591,6 @@ void __init vtime_init(void)
  90
  91         /* Enable cpu timer interrupts on the boot cpu. */
  92         init_cpu_vtimer();
  93 +       cpu_notifier(s390_nohz_notify, 0);
  94  }
  95