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/export.h>
  10 #include <linux/irq_work.h>
  11 #include <linux/percpu.h>
  12 #include <linux/hardirq.h>
  13 #include <asm/processor.h>
  14
  15 /*
  16  * An entry can be in one of four states:
  17  *
  18  * free      NULL, 0 -> {claimed}       : free to be used
  19  * claimed   NULL, 3 -> {pending}       : claimed to be enqueued
  20  * pending   next, 3 -> {busy}          : queued, pending callback
  21  * busy      NULL, 2 -> {free, claimed} : callback in progress, can be claimed
  22  *
  23  * We use the lower two bits of the next pointer to keep PENDING and BUSY
  24  * flags.
  25  */
  26
  27 #define IRQ_WORK_PENDING        1UL
  28 #define IRQ_WORK_BUSY           2UL
  29 #define IRQ_WORK_FLAGS          3UL
  30
  31 static inline bool irq_work_is_set(struct irq_work *entry, int flags)
  32 {
  33         return (unsigned long)entry->next & flags;
  34 }
  35
  36 static inline struct irq_work *irq_work_next(struct irq_work *entry)
  37 {
  38         unsigned long next = (unsigned long)entry->next;
  39         next &= ~IRQ_WORK_FLAGS;
  40         return (struct irq_work *)next;
  41 }
  42
  43 static inline struct irq_work *next_flags(struct irq_work *entry, int flags)
  44 {
  45         unsigned long next = (unsigned long)entry;
  46         next |= flags;
  47         return (struct irq_work *)next;
  48 }
  49
  50 static DEFINE_PER_CPU(struct irq_work *, irq_work_list);
  51
  52 /*
  53  * Claim the entry so that no one else will poke at it.
  54  */
  55 static bool irq_work_claim(struct irq_work *entry)
  56 {
  57         struct irq_work *next, *nflags;
  58
  59         do {
  60                 next = entry->next;
  61                 if ((unsigned long)next & IRQ_WORK_PENDING)
  62                         return false;
  63                 nflags = next_flags(next, IRQ_WORK_FLAGS);
  64         } while (cmpxchg(&entry->next, next, nflags) != next);
  65
  66         return true;
  67 }
  68
  69
  70 void __weak arch_irq_work_raise(void)
  71 {
  72         /*
  73          * Lame architectures will get the timer tick callback
  74          */
  75 }
  76
  77 /*
  78  * Queue the entry and raise the IPI if needed.
  79  */
  80 static void __irq_work_queue(struct irq_work *entry)
  81 {
  82         struct irq_work *next;
  83
  84         preempt_disable();
  85
  86         do {
  87                 next = __this_cpu_read(irq_work_list);
  88                 /* Can assign non-atomic because we keep the flags set. */
  89                 entry->next = next_flags(next, IRQ_WORK_FLAGS);
  90         } while (this_cpu_cmpxchg(irq_work_list, next, entry) != next);
  91
  92         /* The list was empty, raise self-interrupt to start processing. */
  93         if (!irq_work_next(entry))
  94                 arch_irq_work_raise();
  95
  96         preempt_enable();
  97 }
  98
  99 /*
 100  * Enqueue the irq_work @entry, returns true on success, failure when the
 101  * @entry was already enqueued by someone else.
 102  *
 103  * Can be re-enqueued while the callback is still in progress.
 104  */
 105 bool irq_work_queue(struct irq_work *entry)
 106 {
 107         if (!irq_work_claim(entry)) {
 108                 /*
 109                  * Already enqueued, can't do!
 110                  */
 111                 return false;
 112         }
 113
 114         __irq_work_queue(entry);
 115         return true;
 116 }
 117 EXPORT_SYMBOL_GPL(irq_work_queue);
 118
 119 /*
 120  * Run the irq_work entries on this cpu. Requires to be ran from hardirq
 121  * context with local IRQs disabled.
 122  */
 123 void irq_work_run(void)
 124 {
 125         struct irq_work *list;
 126
 127         if (this_cpu_read(irq_work_list) == NULL)
 128                 return;
 129
 130         BUG_ON(!in_irq());
 131         BUG_ON(!irqs_disabled());
 132
 133         list = this_cpu_xchg(irq_work_list, NULL);
 134
 135         while (list != NULL) {
 136                 struct irq_work *entry = list;
 137
 138                 list = irq_work_next(list);
 139
 140                 /*
 141                  * Clear the PENDING bit, after this point the @entry
 142                  * can be re-used.
 143                  */
 144                 entry->next = next_flags(NULL, IRQ_WORK_BUSY);
 145                 entry->func(entry);
 146                 /*
 147                  * Clear the BUSY bit and return to the free state if
 148                  * no-one else claimed it meanwhile.
 149                  */
 150                 (void)cmpxchg(&entry->next,
 151                               next_flags(NULL, IRQ_WORK_BUSY),
 152                               NULL);
 153         }
 154 }
 155 EXPORT_SYMBOL_GPL(irq_work_run);
 156
 157 /*
 158  * Synchronize against the irq_work @entry, ensures the entry is not
 159  * currently in use.
 160  */
 161 void irq_work_sync(struct irq_work *entry)
 162 {
 163         WARN_ON_ONCE(irqs_disabled());
 164
 165         while (irq_work_is_set(entry, IRQ_WORK_BUSY))
 166                 cpu_relax();
 167 }
 168 EXPORT_SYMBOL_GPL(irq_work_sync);