irq_work.c 4.44 KB
Newer Older
1
/*
2
 * Copyright (C) 2010 Red Hat, Inc., Peter Zijlstra
3 4 5 6 7
 *
 * Provides a framework for enqueueing and running callbacks from hardirq
 * context. The enqueueing is NMI-safe.
 */

8
#include <linux/bug.h>
9
#include <linux/kernel.h>
10
#include <linux/export.h>
11
#include <linux/irq_work.h>
12
#include <linux/percpu.h>
13
#include <linux/hardirq.h>
14
#include <linux/irqflags.h>
15 16
#include <linux/sched.h>
#include <linux/tick.h>
17 18
#include <linux/cpu.h>
#include <linux/notifier.h>
19
#include <linux/smp.h>
20
#include <asm/processor.h>
21 22


23 24
static DEFINE_PER_CPU(struct llist_head, raised_list);
static DEFINE_PER_CPU(struct llist_head, lazy_list);
25 26 27 28

/*
 * Claim the entry so that no one else will poke at it.
 */
29
static bool irq_work_claim(struct irq_work *work)
30
{
31
	unsigned long flags, oflags, nflags;
32

33 34 35 36 37
	/*
	 * Start with our best wish as a premise but only trust any
	 * flag value after cmpxchg() result.
	 */
	flags = work->flags & ~IRQ_WORK_PENDING;
38 39
	for (;;) {
		nflags = flags | IRQ_WORK_FLAGS;
40 41
		oflags = cmpxchg(&work->flags, flags, nflags);
		if (oflags == flags)
42
			break;
43 44 45
		if (oflags & IRQ_WORK_PENDING)
			return false;
		flags = oflags;
46 47
		cpu_relax();
	}
48 49 50 51 52 53 54 55 56 57 58

	return true;
}

void __weak arch_irq_work_raise(void)
{
	/*
	 * Lame architectures will get the timer tick callback
	 */
}

59
#ifdef CONFIG_SMP
60
/*
61
 * Enqueue the irq_work @work on @cpu unless it's already pending
62 63 64
 * somewhere.
 *
 * Can be re-enqueued while the callback is still in progress.
65
 */
66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
bool irq_work_queue_on(struct irq_work *work, int cpu)
{
	/* All work should have been flushed before going offline */
	WARN_ON_ONCE(cpu_is_offline(cpu));

	/* Arch remote IPI send/receive backend aren't NMI safe */
	WARN_ON_ONCE(in_nmi());

	/* Only queue if not already pending */
	if (!irq_work_claim(work))
		return false;

	if (llist_add(&work->llnode, &per_cpu(raised_list, cpu)))
		arch_send_call_function_single_ipi(cpu);

	return true;
}
EXPORT_SYMBOL_GPL(irq_work_queue_on);
#endif

/* Enqueue the irq work @work on the current CPU */
87
bool irq_work_queue(struct irq_work *work)
88
{
89 90
	/* Only queue if not already pending */
	if (!irq_work_claim(work))
91
		return false;
92 93

	/* Queue the entry and raise the IPI if needed. */
94
	preempt_disable();
95

96 97
	/* If the work is "lazy", handle it from next tick if any */
	if (work->flags & IRQ_WORK_LAZY) {
98
		if (llist_add(&work->llnode, this_cpu_ptr(&lazy_list)) &&
99 100 101
		    tick_nohz_tick_stopped())
			arch_irq_work_raise();
	} else {
102
		if (llist_add(&work->llnode, this_cpu_ptr(&raised_list)))
103 104
			arch_irq_work_raise();
	}
105

106
	preempt_enable();
107 108

	return true;
109 110 111
}
EXPORT_SYMBOL_GPL(irq_work_queue);

112 113
bool irq_work_needs_cpu(void)
{
114
	struct llist_head *raised, *lazy;
115

116 117
	raised = this_cpu_ptr(&raised_list);
	lazy = this_cpu_ptr(&lazy_list);
118 119 120 121

	if (llist_empty(raised) || arch_irq_work_has_interrupt())
		if (llist_empty(lazy))
			return false;
122

123 124 125
	/* All work should have been flushed before going offline */
	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));

126 127 128
	return true;
}

129
static void irq_work_run_list(struct llist_head *list)
130
{
131
	unsigned long flags;
132 133
	struct irq_work *work;
	struct llist_node *llnode;
134

135
	BUG_ON(!irqs_disabled());
136

137
	if (llist_empty(list))
138 139
		return;

140
	llnode = llist_del_all(list);
141 142
	while (llnode != NULL) {
		work = llist_entry(llnode, struct irq_work, llnode);
143

Peter Zijlstra's avatar
Peter Zijlstra committed
144
		llnode = llist_next(llnode);
145 146

		/*
147
		 * Clear the PENDING bit, after this point the @work
148
		 * can be re-used.
149 150 151
		 * Make it immediately visible so that other CPUs trying
		 * to claim that work don't rely on us to handle their data
		 * while we are in the middle of the func.
152
		 */
153 154 155
		flags = work->flags & ~IRQ_WORK_PENDING;
		xchg(&work->flags, flags);

156
		work->func(work);
157 158 159 160
		/*
		 * Clear the BUSY bit and return to the free state if
		 * no-one else claimed it meanwhile.
		 */
161
		(void)cmpxchg(&work->flags, flags, flags & ~IRQ_WORK_BUSY);
162 163
	}
}
164 165

/*
166 167
 * hotplug calls this through:
 *  hotplug_cfd() -> flush_smp_call_function_queue()
168 169 170
 */
void irq_work_run(void)
{
171 172
	irq_work_run_list(this_cpu_ptr(&raised_list));
	irq_work_run_list(this_cpu_ptr(&lazy_list));
173
}
174 175
EXPORT_SYMBOL_GPL(irq_work_run);

176 177
void irq_work_tick(void)
{
178
	struct llist_head *raised = this_cpu_ptr(&raised_list);
179 180 181

	if (!llist_empty(raised) && !arch_irq_work_has_interrupt())
		irq_work_run_list(raised);
182
	irq_work_run_list(this_cpu_ptr(&lazy_list));
183 184
}

185 186 187 188
/*
 * Synchronize against the irq_work @entry, ensures the entry is not
 * currently in use.
 */
189
void irq_work_sync(struct irq_work *work)
190 191 192
{
	WARN_ON_ONCE(irqs_disabled());

193
	while (work->flags & IRQ_WORK_BUSY)
194 195 196
		cpu_relax();
}
EXPORT_SYMBOL_GPL(irq_work_sync);