1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Fast batching percpu counters.
4 */
5
6#include <linux/percpu_counter.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 const struct debug_obj_descr percpu_counter_debug_descr;
21
22static bool 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 true;
31	default:
32		return false;
33	}
34}
35
36static const 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
75/**
76 * This function is both preempt and irq safe. The former is due to explicit
77 * preemption disable. The latter is guaranteed by the fact that the slow path
78 * is explicitly protected by an irq-safe spinlock whereas the fast patch uses
79 * this_cpu_add which is irq-safe by definition. Hence there is no need muck
80 * with irq state before calling this one
81 */
82void percpu_counter_add_batch(struct percpu_counter *fbc, s64 amount, s32 batch)
83{
84	s64 count;
85
86	preempt_disable();
87	count = __this_cpu_read(*fbc->counters) + amount;
88	if (abs(count) >= batch) {
89		unsigned long flags;
90		raw_spin_lock_irqsave(&fbc->lock, flags);
91		fbc->count += count;
92		__this_cpu_sub(*fbc->counters, count - amount);
93		raw_spin_unlock_irqrestore(&fbc->lock, flags);
94	} else {
95		this_cpu_add(*fbc->counters, amount);
96	}
97	preempt_enable();
98}
99EXPORT_SYMBOL(percpu_counter_add_batch);
100
101/*
102 * For percpu_counter with a big batch, the devication of its count could
103 * be big, and there is requirement to reduce the deviation, like when the
104 * counter's batch could be runtime decreased to get a better accuracy,
105 * which can be achieved by running this sync function on each CPU.
106 */
107void percpu_counter_sync(struct percpu_counter *fbc)
108{
109	unsigned long flags;
110	s64 count;
111
112	raw_spin_lock_irqsave(&fbc->lock, flags);
113	count = __this_cpu_read(*fbc->counters);
114	fbc->count += count;
115	__this_cpu_sub(*fbc->counters, count);
116	raw_spin_unlock_irqrestore(&fbc->lock, flags);
117}
118EXPORT_SYMBOL(percpu_counter_sync);
119
120/*
121 * Add up all the per-cpu counts, return the result.  This is a more accurate
122 * but much slower version of percpu_counter_read_positive()
123 */
124s64 __percpu_counter_sum(struct percpu_counter *fbc)
125{
126	s64 ret;
127	int cpu;
128	unsigned long flags;
129
130	raw_spin_lock_irqsave(&fbc->lock, flags);
131	ret = fbc->count;
132	for_each_online_cpu(cpu) {
133		s32 *pcount = per_cpu_ptr(fbc->counters, cpu);
134		ret += *pcount;
135	}
136	raw_spin_unlock_irqrestore(&fbc->lock, flags);
137	return ret;
138}
139EXPORT_SYMBOL(__percpu_counter_sum);
140
141int __percpu_counter_init(struct percpu_counter *fbc, s64 amount, gfp_t gfp,
142			  struct lock_class_key *key)
143{
144	unsigned long flags __maybe_unused;
145
146	raw_spin_lock_init(&fbc->lock);
147	lockdep_set_class(&fbc->lock, key);
148	fbc->count = amount;
149	fbc->counters = alloc_percpu_gfp(s32, gfp);
150	if (!fbc->counters)
151		return -ENOMEM;
152
153	debug_percpu_counter_activate(fbc);
154
155#ifdef CONFIG_HOTPLUG_CPU
156	INIT_LIST_HEAD(&fbc->list);
157	spin_lock_irqsave(&percpu_counters_lock, flags);
158	list_add(&fbc->list, &percpu_counters);
159	spin_unlock_irqrestore(&percpu_counters_lock, flags);
160#endif
161	return 0;
162}
163EXPORT_SYMBOL(__percpu_counter_init);
164
165void percpu_counter_destroy(struct percpu_counter *fbc)
166{
167	unsigned long flags __maybe_unused;
168
169	if (!fbc->counters)
170		return;
171
172	debug_percpu_counter_deactivate(fbc);
173
174#ifdef CONFIG_HOTPLUG_CPU
175	spin_lock_irqsave(&percpu_counters_lock, flags);
176	list_del(&fbc->list);
177	spin_unlock_irqrestore(&percpu_counters_lock, flags);
178#endif
179	free_percpu(fbc->counters);
180	fbc->counters = NULL;
181}
182EXPORT_SYMBOL(percpu_counter_destroy);
183
184int percpu_counter_batch __read_mostly = 32;
185EXPORT_SYMBOL(percpu_counter_batch);
186
187static int compute_batch_value(unsigned int cpu)
188{
189	int nr = num_online_cpus();
190
191	percpu_counter_batch = max(32, nr*2);
192	return 0;
193}
194
195static int percpu_counter_cpu_dead(unsigned int cpu)
196{
197#ifdef CONFIG_HOTPLUG_CPU
198	struct percpu_counter *fbc;
199
200	compute_batch_value(cpu);
201
202	spin_lock_irq(&percpu_counters_lock);
203	list_for_each_entry(fbc, &percpu_counters, list) {
204		s32 *pcount;
205
206		raw_spin_lock(&fbc->lock);
207		pcount = per_cpu_ptr(fbc->counters, cpu);
208		fbc->count += *pcount;
209		*pcount = 0;
210		raw_spin_unlock(&fbc->lock);
211	}
212	spin_unlock_irq(&percpu_counters_lock);
213#endif
214	return 0;
215}
216
217/*
218 * Compare counter against given value.
219 * Return 1 if greater, 0 if equal and -1 if less
220 */
221int __percpu_counter_compare(struct percpu_counter *fbc, s64 rhs, s32 batch)
222{
223	s64	count;
224
225	count = percpu_counter_read(fbc);
226	/* Check to see if rough count will be sufficient for comparison */
227	if (abs(count - rhs) > (batch * num_online_cpus())) {
228		if (count > rhs)
229			return 1;
230		else
231			return -1;
232	}
233	/* Need to use precise count */
234	count = percpu_counter_sum(fbc);
235	if (count > rhs)
236		return 1;
237	else if (count < rhs)
238		return -1;
239	else
240		return 0;
241}
242EXPORT_SYMBOL(__percpu_counter_compare);
243
244static int __init percpu_counter_startup(void)
245{
246	int ret;
247
248	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "lib/percpu_cnt:online",
249				compute_batch_value, NULL);
250	WARN_ON(ret < 0);
251	ret = cpuhp_setup_state_nocalls(CPUHP_PERCPU_CNT_DEAD,
252					"lib/percpu_cnt:dead", NULL,
253					percpu_counter_cpu_dead);
254	WARN_ON(ret < 0);
255	return 0;
256}
257module_init(percpu_counter_startup);
258