kernel/irq_work.c

   1 /*
   2  * Copyright (C) 2010 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
   3  *
   4  * Provides a framework for enqueueing and running callbacks from hardirq
   5  * context. The enqueueing is NMI-safe.
   6  */
   7
   8 #include <linux/kernel.h>
   9 #include <linux/module.h>
  10 #include <linux/irq_work.h>
  11 #include <linux/hardirq.h>
  12
  13 /*
  14  * An entry can be in one of four states:
  15  *
  16  * free      NULL, 0 -> {claimed}       : free to be used
  17  * claimed   NULL, 3 -> {pending}       : claimed to be enqueued
  18  * pending   next, 3 -> {busy}          : queued, pending callback
  19  * busy      NULL, 2 -> {free, claimed} : callback in progress, can be claimed
  20  *
  21  * We use the lower two bits of the next pointer to keep PENDING and BUSY
  22  * flags.
  23  */
  24
  25 #define IRQ_WORK_PENDING        1UL
  26 #define IRQ_WORK_BUSY           2UL
  27 #define IRQ_WORK_FLAGS          3UL
  28
  29 static inline bool irq_work_is_set(struct irq_work *entry, int flags)
  30 {
  31         return (unsigned long)entry->next & flags;
  32 }
  33
  34 static inline struct irq_work *irq_work_next(struct irq_work *entry)
  35 {
  36         unsigned long next = (unsigned long)entry->next;
  37         next &= ~IRQ_WORK_FLAGS;
  38         return (struct irq_work *)next;
  39 }
  40
  41 static inline struct irq_work *next_flags(struct irq_work *entry, int flags)
  42 {
  43         unsigned long next = (unsigned long)entry;
  44         next |= flags;
  45         return (struct irq_work *)next;
  46 }
  47
  48 static DEFINE_PER_CPU(struct irq_work *, irq_work_list);
  49
  50 /*
  51  * Claim the entry so that no one else will poke at it.
  52  */
  53 static bool irq_work_claim(struct irq_work *entry)
  54 {
  55         struct irq_work *next, *nflags;
  56
  57         do {
  58                 next = entry->next;
  59                 if ((unsigned long)next & IRQ_WORK_PENDING)
  60                         return false;
  61                 nflags = next_flags(next, IRQ_WORK_FLAGS);
  62         } while (cmpxchg(&entry->next, next, nflags) != next);
  63
  64         return true;
  65 }
  66
  67
  68 void __weak arch_irq_work_raise(void)
  69 {
  70         /*
  71          * Lame architectures will get the timer tick callback
  72          */
  73 }
  74
  75 /*
  76  * Queue the entry and raise the IPI if needed.
  77  */
  78 static void __irq_work_queue(struct irq_work *entry)
  79 {
  80         struct irq_work *next;
  81
  82         preempt_disable();
  83
  84         do {
  85                 next = __this_cpu_read(irq_work_list);
  86                 /* Can assign non-atomic because we keep the flags set. */
  87                 entry->next = next_flags(next, IRQ_WORK_FLAGS);
  88         } while (this_cpu_cmpxchg(irq_work_list, next, entry) != next);
  89
  90         /* The list was empty, raise self-interrupt to start processing. */
  91         if (!irq_work_next(entry))
  92                 arch_irq_work_raise();
  93
  94         preempt_enable();
  95 }
  96
  97 /*
  98  * Enqueue the irq_work @entry, returns true on success, failure when the
  99  * @entry was already enqueued by someone else.
 100  *
 101  * Can be re-enqueued while the callback is still in progress.
 102  */
 103 bool irq_work_queue(struct irq_work *entry)
 104 {
 105         if (!irq_work_claim(entry)) {
 106                 /*
 107                  * Already enqueued, can't do!
 108                  */
 109                 return false;
 110         }
 111
 112         __irq_work_queue(entry);
 113         return true;
 114 }
 115 EXPORT_SYMBOL_GPL(irq_work_queue);
 116
 117 /*
 118  * Run the irq_work entries on this cpu. Requires to be ran from hardirq
 119  * context with local IRQs disabled.
 120  */
 121 void irq_work_run(void)
 122 {
 123         struct irq_work *list;
 124
 125         if (this_cpu_read(irq_work_list) == NULL)
 126                 return;
 127
 128         BUG_ON(!in_irq());
 129         BUG_ON(!irqs_disabled());
 130
 131         list = this_cpu_xchg(irq_work_list, NULL);
 132
 133         while (list != NULL) {
 134                 struct irq_work *entry = list;
 135
 136                 list = irq_work_next(list);
 137
 138                 /*
 139                  * Clear the PENDING bit, after this point the @entry
 140                  * can be re-used.
 141                  */
 142                 entry->next = next_flags(NULL, IRQ_WORK_BUSY);
 143                 entry->func(entry);
 144                 /*
 145                  * Clear the BUSY bit and return to the free state if
 146                  * no-one else claimed it meanwhile.
 147                  */
 148                 (void)cmpxchg(&entry->next,
 149                               next_flags(NULL, IRQ_WORK_BUSY),
 150                               NULL);
 151         }
 152 }
 153 EXPORT_SYMBOL_GPL(irq_work_run);
 154
 155 /*
 156  * Synchronize against the irq_work @entry, ensures the entry is not
 157  * currently in use.
 158  */
 159 void irq_work_sync(struct irq_work *entry)
 160 {
 161         WARN_ON_ONCE(irqs_disabled());
 162
 163         while (irq_work_is_set(entry, IRQ_WORK_BUSY))
 164                 cpu_relax();
 165 }
 166 EXPORT_SYMBOL_GPL(irq_work_sync);