]>
Commit | Line | Data |
---|---|---|
e360adbe PZ |
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 | ||
83e3fa6f | 8 | #include <linux/bug.h> |
e360adbe | 9 | #include <linux/kernel.h> |
9984de1a | 10 | #include <linux/export.h> |
e360adbe | 11 | #include <linux/irq_work.h> |
967d1f90 | 12 | #include <linux/percpu.h> |
e360adbe | 13 | #include <linux/hardirq.h> |
ef1f0982 | 14 | #include <linux/irqflags.h> |
bc6679ae FW |
15 | #include <linux/sched.h> |
16 | #include <linux/tick.h> | |
c0e980a4 SR |
17 | #include <linux/cpu.h> |
18 | #include <linux/notifier.h> | |
967d1f90 | 19 | #include <asm/processor.h> |
e360adbe | 20 | |
e360adbe | 21 | |
b93e0b8f FW |
22 | static DEFINE_PER_CPU(struct llist_head, raised_list); |
23 | static DEFINE_PER_CPU(struct llist_head, lazy_list); | |
e360adbe PZ |
24 | |
25 | /* | |
26 | * Claim the entry so that no one else will poke at it. | |
27 | */ | |
38aaf809 | 28 | static bool irq_work_claim(struct irq_work *work) |
e360adbe | 29 | { |
e0bbe2d8 | 30 | unsigned long flags, oflags, nflags; |
e360adbe | 31 | |
e0bbe2d8 FW |
32 | /* |
33 | * Start with our best wish as a premise but only trust any | |
34 | * flag value after cmpxchg() result. | |
35 | */ | |
36 | flags = work->flags & ~IRQ_WORK_PENDING; | |
38aaf809 | 37 | for (;;) { |
38aaf809 | 38 | nflags = flags | IRQ_WORK_FLAGS; |
e0bbe2d8 FW |
39 | oflags = cmpxchg(&work->flags, flags, nflags); |
40 | if (oflags == flags) | |
38aaf809 | 41 | break; |
e0bbe2d8 FW |
42 | if (oflags & IRQ_WORK_PENDING) |
43 | return false; | |
44 | flags = oflags; | |
38aaf809 HY |
45 | cpu_relax(); |
46 | } | |
e360adbe PZ |
47 | |
48 | return true; | |
49 | } | |
50 | ||
e360adbe PZ |
51 | void __weak arch_irq_work_raise(void) |
52 | { | |
53 | /* | |
54 | * Lame architectures will get the timer tick callback | |
55 | */ | |
56 | } | |
57 | ||
58 | /* | |
c02cf5f8 | 59 | * Enqueue the irq_work @entry unless it's already pending |
60 | * somewhere. | |
61 | * | |
62 | * Can be re-enqueued while the callback is still in progress. | |
e360adbe | 63 | */ |
cd578abb | 64 | bool irq_work_queue(struct irq_work *work) |
e360adbe | 65 | { |
c02cf5f8 | 66 | /* Only queue if not already pending */ |
67 | if (!irq_work_claim(work)) | |
cd578abb | 68 | return false; |
c02cf5f8 | 69 | |
70 | /* Queue the entry and raise the IPI if needed. */ | |
20b87691 | 71 | preempt_disable(); |
e360adbe | 72 | |
b93e0b8f FW |
73 | /* If the work is "lazy", handle it from next tick if any */ |
74 | if (work->flags & IRQ_WORK_LAZY) { | |
75 | if (llist_add(&work->llnode, &__get_cpu_var(lazy_list)) && | |
76 | tick_nohz_tick_stopped()) | |
77 | arch_irq_work_raise(); | |
78 | } else { | |
79 | if (llist_add(&work->llnode, &__get_cpu_var(raised_list))) | |
bc6679ae FW |
80 | arch_irq_work_raise(); |
81 | } | |
e360adbe | 82 | |
20b87691 | 83 | preempt_enable(); |
cd578abb PZ |
84 | |
85 | return true; | |
e360adbe | 86 | } |
e360adbe PZ |
87 | EXPORT_SYMBOL_GPL(irq_work_queue); |
88 | ||
00b42959 FW |
89 | bool irq_work_needs_cpu(void) |
90 | { | |
b93e0b8f | 91 | struct llist_head *raised, *lazy; |
00b42959 | 92 | |
b93e0b8f FW |
93 | raised = &__get_cpu_var(raised_list); |
94 | lazy = &__get_cpu_var(lazy_list); | |
95 | if (llist_empty(raised) && llist_empty(lazy)) | |
00b42959 FW |
96 | return false; |
97 | ||
8aa2acce SR |
98 | /* All work should have been flushed before going offline */ |
99 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); | |
100 | ||
00b42959 FW |
101 | return true; |
102 | } | |
103 | ||
b93e0b8f | 104 | static void irq_work_run_list(struct llist_head *list) |
e360adbe | 105 | { |
bc6679ae | 106 | unsigned long flags; |
38aaf809 | 107 | struct irq_work *work; |
38aaf809 | 108 | struct llist_node *llnode; |
e360adbe | 109 | |
b93e0b8f | 110 | BUG_ON(!irqs_disabled()); |
bc6679ae | 111 | |
b93e0b8f | 112 | if (llist_empty(list)) |
e360adbe PZ |
113 | return; |
114 | ||
b93e0b8f | 115 | llnode = llist_del_all(list); |
38aaf809 HY |
116 | while (llnode != NULL) { |
117 | work = llist_entry(llnode, struct irq_work, llnode); | |
e360adbe | 118 | |
924f8f5a | 119 | llnode = llist_next(llnode); |
e360adbe PZ |
120 | |
121 | /* | |
38aaf809 | 122 | * Clear the PENDING bit, after this point the @work |
e360adbe | 123 | * can be re-used. |
c8446b75 FW |
124 | * Make it immediately visible so that other CPUs trying |
125 | * to claim that work don't rely on us to handle their data | |
126 | * while we are in the middle of the func. | |
e360adbe | 127 | */ |
bc6679ae FW |
128 | flags = work->flags & ~IRQ_WORK_PENDING; |
129 | xchg(&work->flags, flags); | |
130 | ||
38aaf809 | 131 | work->func(work); |
e360adbe PZ |
132 | /* |
133 | * Clear the BUSY bit and return to the free state if | |
134 | * no-one else claimed it meanwhile. | |
135 | */ | |
bc6679ae | 136 | (void)cmpxchg(&work->flags, flags, flags & ~IRQ_WORK_BUSY); |
e360adbe PZ |
137 | } |
138 | } | |
c0e980a4 | 139 | |
b93e0b8f FW |
140 | static void __irq_work_run(void) |
141 | { | |
142 | irq_work_run_list(&__get_cpu_var(raised_list)); | |
143 | irq_work_run_list(&__get_cpu_var(lazy_list)); | |
144 | } | |
145 | ||
c0e980a4 SR |
146 | /* |
147 | * Run the irq_work entries on this cpu. Requires to be ran from hardirq | |
148 | * context with local IRQs disabled. | |
149 | */ | |
150 | void irq_work_run(void) | |
151 | { | |
152 | BUG_ON(!in_irq()); | |
153 | __irq_work_run(); | |
154 | } | |
e360adbe PZ |
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 | */ | |
38aaf809 | 161 | void irq_work_sync(struct irq_work *work) |
e360adbe PZ |
162 | { |
163 | WARN_ON_ONCE(irqs_disabled()); | |
164 | ||
38aaf809 | 165 | while (work->flags & IRQ_WORK_BUSY) |
e360adbe PZ |
166 | cpu_relax(); |
167 | } | |
168 | EXPORT_SYMBOL_GPL(irq_work_sync); | |
c0e980a4 SR |
169 | |
170 | #ifdef CONFIG_HOTPLUG_CPU | |
171 | static int irq_work_cpu_notify(struct notifier_block *self, | |
172 | unsigned long action, void *hcpu) | |
173 | { | |
174 | long cpu = (long)hcpu; | |
175 | ||
176 | switch (action) { | |
177 | case CPU_DYING: | |
178 | /* Called from stop_machine */ | |
179 | if (WARN_ON_ONCE(cpu != smp_processor_id())) | |
180 | break; | |
181 | __irq_work_run(); | |
182 | break; | |
183 | default: | |
184 | break; | |
185 | } | |
186 | return NOTIFY_OK; | |
187 | } | |
188 | ||
189 | static struct notifier_block cpu_notify; | |
190 | ||
191 | static __init int irq_work_init_cpu_notifier(void) | |
192 | { | |
193 | cpu_notify.notifier_call = irq_work_cpu_notify; | |
194 | cpu_notify.priority = 0; | |
195 | register_cpu_notifier(&cpu_notify); | |
196 | return 0; | |
197 | } | |
198 | device_initcall(irq_work_init_cpu_notifier); | |
199 | ||
200 | #endif /* CONFIG_HOTPLUG_CPU */ |