queue-4.19/genirq-avoid-summation-loops-for-proc-stat.patch

   1 From a07167b5aa2226adf4d633069c055b9197f914fa Mon Sep 17 00:00:00 2001
   2 From: Thomas Gleixner <tglx@linutronix.de>
   3 Date: Fri, 8 Feb 2019 14:48:03 +0100
   4 Subject: genirq: Avoid summation loops for /proc/stat
   5
   6 [ Upstream commit 1136b0728969901a091f0471968b2b76ed14d9ad ]
   7
   8 Waiman reported that on large systems with a large amount of interrupts the
   9 readout of /proc/stat takes a long time to sum up the interrupt
  10 statistics. In principle this is not a problem. but for unknown reasons
  11 some enterprise quality software reads /proc/stat with a high frequency.
  12
  13 The reason for this is that interrupt statistics are accounted per cpu. So
  14 the /proc/stat logic has to sum up the interrupt stats for each interrupt.
  15
  16 This can be largely avoided for interrupts which are not marked as
  17 'PER_CPU' interrupts by simply adding a per interrupt summation counter
  18 which is incremented along with the per interrupt per cpu counter.
  19
  20 The PER_CPU interrupts need to avoid that and use only per cpu accounting
  21 because they share the interrupt number and the interrupt descriptor and
  22 concurrent updates would conflict or require unwanted synchronization.
  23
  24 Reported-by: Waiman Long <longman@redhat.com>
  25 Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
  26 Reviewed-by: Waiman Long <longman@redhat.com>
  27 Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
  28 Reviewed-by: Davidlohr Bueso <dbueso@suse.de>
  29 Cc: Matthew Wilcox <willy@infradead.org>
  30 Cc: Andrew Morton <akpm@linux-foundation.org>
  31 Cc: Alexey Dobriyan <adobriyan@gmail.com>
  32 Cc: Kees Cook <keescook@chromium.org>
  33 Cc: linux-fsdevel@vger.kernel.org
  34 Cc: Davidlohr Bueso <dave@stgolabs.net>
  35 Cc: Miklos Szeredi <miklos@szeredi.hu>
  36 Cc: Daniel Colascione <dancol@google.com>
  37 Cc: Dave Chinner <david@fromorbit.com>
  38 Cc: Randy Dunlap <rdunlap@infradead.org>
  39 Link: https://lkml.kernel.org/r/20190208135020.925487496@linutronix.de
  40
  41 8<-------------
  42
  43 v2: Undo the unintentional layout change of struct irq_desc.
  44
  45  include/linux/irqdesc.h |    1 +
  46  kernel/irq/chip.c       |   12 ++++++++++--
  47  kernel/irq/internals.h  |    8 +++++++-
  48  kernel/irq/irqdesc.c    |    7 ++++++-
  49  4 files changed, 24 insertions(+), 4 deletions(-)
  50
  51 Signed-off-by: Sasha Levin <sashal@kernel.org>
  52 ---
  53  include/linux/irqdesc.h |  1 +
  54  kernel/irq/chip.c       | 12 ++++++++++--
  55  kernel/irq/internals.h  |  8 +++++++-
  56  kernel/irq/irqdesc.c    |  7 ++++++-
  57  4 files changed, 24 insertions(+), 4 deletions(-)
  58
  59 diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
  60 index dd1e40ddac7d..875c41b23f20 100644
  61 --- a/include/linux/irqdesc.h
  62 +++ b/include/linux/irqdesc.h
  63 @@ -65,6 +65,7 @@ struct irq_desc {
  64         unsigned int            core_internal_state__do_not_mess_with_it;
  65         unsigned int            depth;          /* nested irq disables */
  66         unsigned int            wake_depth;     /* nested wake enables */
  67 +       unsigned int            tot_count;
  68         unsigned int            irq_count;      /* For detecting broken IRQs */
  69         unsigned long           last_unhandled; /* Aging timer for unhandled count */
  70         unsigned int            irqs_unhandled;
  71 diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
  72 index a2b3d9de999c..811009ebacd4 100644
  73 --- a/kernel/irq/chip.c
  74 +++ b/kernel/irq/chip.c
  75 @@ -855,7 +855,11 @@ void handle_percpu_irq(struct irq_desc *desc)
  76  {
  77         struct irq_chip *chip = irq_desc_get_chip(desc);
  78
  79 -       kstat_incr_irqs_this_cpu(desc);
  80 +       /*
  81 +        * PER CPU interrupts are not serialized. Do not touch
  82 +        * desc->tot_count.
  83 +        */
  84 +       __kstat_incr_irqs_this_cpu(desc);
  85
  86         if (chip->irq_ack)
  87                 chip->irq_ack(&desc->irq_data);
  88 @@ -884,7 +888,11 @@ void handle_percpu_devid_irq(struct irq_desc *desc)
  89         unsigned int irq = irq_desc_get_irq(desc);
  90         irqreturn_t res;
  91
  92 -       kstat_incr_irqs_this_cpu(desc);
  93 +       /*
  94 +        * PER CPU interrupts are not serialized. Do not touch
  95 +        * desc->tot_count.
  96 +        */
  97 +       __kstat_incr_irqs_this_cpu(desc);
  98
  99         if (chip->irq_ack)
 100                 chip->irq_ack(&desc->irq_data);
 101 diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h
 102 index ca6afa267070..e74e7eea76cf 100644
 103 --- a/kernel/irq/internals.h
 104 +++ b/kernel/irq/internals.h
 105 @@ -242,12 +242,18 @@ static inline void irq_state_set_masked(struct irq_desc *desc)
 106
 107  #undef __irqd_to_state
 108
 109 -static inline void kstat_incr_irqs_this_cpu(struct irq_desc *desc)
 110 +static inline void __kstat_incr_irqs_this_cpu(struct irq_desc *desc)
 111  {
 112         __this_cpu_inc(*desc->kstat_irqs);
 113         __this_cpu_inc(kstat.irqs_sum);
 114  }
 115
 116 +static inline void kstat_incr_irqs_this_cpu(struct irq_desc *desc)
 117 +{
 118 +       __kstat_incr_irqs_this_cpu(desc);
 119 +       desc->tot_count++;
 120 +}
 121 +
 122  static inline int irq_desc_get_node(struct irq_desc *desc)
 123  {
 124         return irq_common_data_get_node(&desc->irq_common_data);
 125 diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
 126 index 578d0e5f1b5b..ba454cba4069 100644
 127 --- a/kernel/irq/irqdesc.c
 128 +++ b/kernel/irq/irqdesc.c
 129 @@ -119,6 +119,7 @@ static void desc_set_defaults(unsigned int irq, struct irq_desc *desc, int node,
 130         desc->depth = 1;
 131         desc->irq_count = 0;
 132         desc->irqs_unhandled = 0;
 133 +       desc->tot_count = 0;
 134         desc->name = NULL;
 135         desc->owner = owner;
 136         for_each_possible_cpu(cpu)
 137 @@ -915,11 +916,15 @@ unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
 138  unsigned int kstat_irqs(unsigned int irq)
 139  {
 140         struct irq_desc *desc = irq_to_desc(irq);
 141 -       int cpu;
 142         unsigned int sum = 0;
 143 +       int cpu;
 144
 145         if (!desc || !desc->kstat_irqs)
 146                 return 0;
 147 +       if (!irq_settings_is_per_cpu_devid(desc) &&
 148 +           !irq_settings_is_per_cpu(desc))
 149 +           return desc->tot_count;
 150 +
 151         for_each_possible_cpu(cpu)
 152                 sum += *per_cpu_ptr(desc->kstat_irqs, cpu);
 153         return sum;
 154 --
 155 2.19.1
 156