1/*
2 * Fast batching percpu counters.
3 */
4
5#include <linux/percpu_counter.h>
6#include <linux/notifier.h>
7#include <linux/mutex.h>
8#include <linux/init.h>
9#include <linux/cpu.h>
10#include <linux/module.h>
11#include <linux/debugobjects.h>
12
13#ifdef CONFIG_HOTPLUG_CPU
14static LIST_HEAD(percpu_counters);
15static DEFINE_SPINLOCK(percpu_counters_lock);
16#endif
17
18#ifdef CONFIG_DEBUG_OBJECTS_PERCPU_COUNTER
19
20static struct debug_obj_descr percpu_counter_debug_descr;
21
22static int percpu_counter_fixup_free(void *addr, enum debug_obj_state state)
23{
24	struct percpu_counter *fbc = addr;
25
26	switch (state) {
27	case ODEBUG_STATE_ACTIVE:
28		percpu_counter_destroy(fbc);
29		debug_object_free(fbc, &percpu_counter_debug_descr);
30		return 1;
31	default:
32		return 0;
33	}
34}
35
36static struct debug_obj_descr percpu_counter_debug_descr = {
37	.name		= "percpu_counter",
38	.fixup_free	= percpu_counter_fixup_free,
39};
40
41static inline void debug_percpu_counter_activate(struct percpu_counter *fbc)
42{
43	debug_object_init(fbc, &percpu_counter_debug_descr);
44	debug_object_activate(fbc, &percpu_counter_debug_descr);
45}
46
47static inline void debug_percpu_counter_deactivate(struct percpu_counter *fbc)
48{
49	debug_object_deactivate(fbc, &percpu_counter_debug_descr);
50	debug_object_free(fbc, &percpu_counter_debug_descr);
51}
52
53#else	/* CONFIG_DEBUG_OBJECTS_PERCPU_COUNTER */
54static inline void debug_percpu_counter_activate(struct percpu_counter *fbc)
55{ }
56static inline void debug_percpu_counter_deactivate(struct percpu_counter *fbc)
57{ }
58#endif	/* CONFIG_DEBUG_OBJECTS_PERCPU_COUNTER */
59
60void percpu_counter_set(struct percpu_counter *fbc, s64 amount)
61{
62	int cpu;
63	unsigned long flags;
64
65	raw_spin_lock_irqsave(&fbc->lock, flags);
66	for_each_possible_cpu(cpu) {
67		s32 *pcount = per_cpu_ptr(fbc->counters, cpu);
68		*pcount = 0;
69	}
70	fbc->count = amount;
71	raw_spin_unlock_irqrestore(&fbc->lock, flags);
72}
73EXPORT_SYMBOL(percpu_counter_set);
74
75void __percpu_counter_add(struct percpu_counter *fbc, s64 amount, s32 batch)
76{
77	s64 count;
78
79	preempt_disable();
80	count = __this_cpu_read(*fbc->counters) + amount;
81	if (count >= batch || count <= -batch) {
82		unsigned long flags;
83		raw_spin_lock_irqsave(&fbc->lock, flags);
84		fbc->count += count;
85		__this_cpu_sub(*fbc->counters, count - amount);
86		raw_spin_unlock_irqrestore(&fbc->lock, flags);
87	} else {
88		this_cpu_add(*fbc->counters, amount);
89	}
90	preempt_enable();
91}
92EXPORT_SYMBOL(__percpu_counter_add);
93
94/*
95 * Add up all the per-cpu counts, return the result.  This is a more accurate
96 * but much slower version of percpu_counter_read_positive()
97 */
98s64 __percpu_counter_sum(struct percpu_counter *fbc)
99{
100	s64 ret;
101	int cpu;
102	unsigned long flags;
103
104	raw_spin_lock_irqsave(&fbc->lock, flags);
105	ret = fbc->count;
106	for_each_online_cpu(cpu) {
107		s32 *pcount = per_cpu_ptr(fbc->counters, cpu);
108		ret += *pcount;
109	}
110	raw_spin_unlock_irqrestore(&fbc->lock, flags);
111	return ret;
112}
113EXPORT_SYMBOL(__percpu_counter_sum);
114
115int __percpu_counter_init(struct percpu_counter *fbc, s64 amount, gfp_t gfp,
116			  struct lock_class_key *key)
117{
118	unsigned long flags __maybe_unused;
119
120	raw_spin_lock_init(&fbc->lock);
121	lockdep_set_class(&fbc->lock, key);
122	fbc->count = amount;
123	fbc->counters = alloc_percpu_gfp(s32, gfp);
124	if (!fbc->counters)
125		return -ENOMEM;
126
127	debug_percpu_counter_activate(fbc);
128
129#ifdef CONFIG_HOTPLUG_CPU
130	INIT_LIST_HEAD(&fbc->list);
131	spin_lock_irqsave(&percpu_counters_lock, flags);
132	list_add(&fbc->list, &percpu_counters);
133	spin_unlock_irqrestore(&percpu_counters_lock, flags);
134#endif
135	return 0;
136}
137EXPORT_SYMBOL(__percpu_counter_init);
138
139void percpu_counter_destroy(struct percpu_counter *fbc)
140{
141	unsigned long flags __maybe_unused;
142
143	if (!fbc->counters)
144		return;
145
146	debug_percpu_counter_deactivate(fbc);
147
148#ifdef CONFIG_HOTPLUG_CPU
149	spin_lock_irqsave(&percpu_counters_lock, flags);
150	list_del(&fbc->list);
151	spin_unlock_irqrestore(&percpu_counters_lock, flags);
152#endif
153	free_percpu(fbc->counters);
154	fbc->counters = NULL;
155}
156EXPORT_SYMBOL(percpu_counter_destroy);
157
158int percpu_counter_batch __read_mostly = 32;
159EXPORT_SYMBOL(percpu_counter_batch);
160
161static void compute_batch_value(void)
162{
163	int nr = num_online_cpus();
164
165	percpu_counter_batch = max(32, nr*2);
166}
167
168static int percpu_counter_hotcpu_callback(struct notifier_block *nb,
169					unsigned long action, void *hcpu)
170{
171#ifdef CONFIG_HOTPLUG_CPU
172	unsigned int cpu;
173	struct percpu_counter *fbc;
174
175	compute_batch_value();
176	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
177		return NOTIFY_OK;
178
179	cpu = (unsigned long)hcpu;
180	spin_lock_irq(&percpu_counters_lock);
181	list_for_each_entry(fbc, &percpu_counters, list) {
182		s32 *pcount;
183		unsigned long flags;
184
185		raw_spin_lock_irqsave(&fbc->lock, flags);
186		pcount = per_cpu_ptr(fbc->counters, cpu);
187		fbc->count += *pcount;
188		*pcount = 0;
189		raw_spin_unlock_irqrestore(&fbc->lock, flags);
190	}
191	spin_unlock_irq(&percpu_counters_lock);
192#endif
193	return NOTIFY_OK;
194}
195
196/*
197 * Compare counter against given value.
198 * Return 1 if greater, 0 if equal and -1 if less
199 */
200int __percpu_counter_compare(struct percpu_counter *fbc, s64 rhs, s32 batch)
201{
202	s64	count;
203
204	count = percpu_counter_read(fbc);
205	/* Check to see if rough count will be sufficient for comparison */
206	if (abs(count - rhs) > (batch * num_online_cpus())) {
207		if (count > rhs)
208			return 1;
209		else
210			return -1;
211	}
212	/* Need to use precise count */
213	count = percpu_counter_sum(fbc);
214	if (count > rhs)
215		return 1;
216	else if (count < rhs)
217		return -1;
218	else
219		return 0;
220}
221EXPORT_SYMBOL(__percpu_counter_compare);
222
223static int __init percpu_counter_startup(void)
224{
225	compute_batch_value();
226	hotcpu_notifier(percpu_counter_hotcpu_callback, 0);
227	return 0;
228}
229module_init(percpu_counter_startup);
230