1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* Copyright (c) 2016 Facebook |
3 | */ |
4 | #include "percpu_freelist.h" |
5 | |
6 | int pcpu_freelist_init(struct pcpu_freelist *s) |
7 | { |
8 | int cpu; |
9 | |
10 | s->freelist = alloc_percpu(struct pcpu_freelist_head); |
11 | if (!s->freelist) |
12 | return -ENOMEM; |
13 | |
14 | for_each_possible_cpu(cpu) { |
15 | struct pcpu_freelist_head *head = per_cpu_ptr(s->freelist, cpu); |
16 | |
17 | raw_spin_lock_init(&head->lock); |
18 | head->first = NULL; |
19 | } |
20 | raw_spin_lock_init(&s->extralist.lock); |
21 | s->extralist.first = NULL; |
22 | return 0; |
23 | } |
24 | |
25 | void pcpu_freelist_destroy(struct pcpu_freelist *s) |
26 | { |
27 | free_percpu(pdata: s->freelist); |
28 | } |
29 | |
30 | static inline void pcpu_freelist_push_node(struct pcpu_freelist_head *head, |
31 | struct pcpu_freelist_node *node) |
32 | { |
33 | node->next = head->first; |
34 | WRITE_ONCE(head->first, node); |
35 | } |
36 | |
37 | static inline void ___pcpu_freelist_push(struct pcpu_freelist_head *head, |
38 | struct pcpu_freelist_node *node) |
39 | { |
40 | raw_spin_lock(&head->lock); |
41 | pcpu_freelist_push_node(head, node); |
42 | raw_spin_unlock(&head->lock); |
43 | } |
44 | |
45 | static inline bool (struct pcpu_freelist *s, |
46 | struct pcpu_freelist_node *node) |
47 | { |
48 | if (!raw_spin_trylock(&s->extralist.lock)) |
49 | return false; |
50 | |
51 | pcpu_freelist_push_node(head: &s->extralist, node); |
52 | raw_spin_unlock(&s->extralist.lock); |
53 | return true; |
54 | } |
55 | |
56 | static inline void ___pcpu_freelist_push_nmi(struct pcpu_freelist *s, |
57 | struct pcpu_freelist_node *node) |
58 | { |
59 | int cpu, orig_cpu; |
60 | |
61 | orig_cpu = raw_smp_processor_id(); |
62 | while (1) { |
63 | for_each_cpu_wrap(cpu, cpu_possible_mask, orig_cpu) { |
64 | struct pcpu_freelist_head *head; |
65 | |
66 | head = per_cpu_ptr(s->freelist, cpu); |
67 | if (raw_spin_trylock(&head->lock)) { |
68 | pcpu_freelist_push_node(head, node); |
69 | raw_spin_unlock(&head->lock); |
70 | return; |
71 | } |
72 | } |
73 | |
74 | /* cannot lock any per cpu lock, try extralist */ |
75 | if (pcpu_freelist_try_push_extra(s, node)) |
76 | return; |
77 | } |
78 | } |
79 | |
80 | void __pcpu_freelist_push(struct pcpu_freelist *s, |
81 | struct pcpu_freelist_node *node) |
82 | { |
83 | if (in_nmi()) |
84 | ___pcpu_freelist_push_nmi(s, node); |
85 | else |
86 | ___pcpu_freelist_push(this_cpu_ptr(s->freelist), node); |
87 | } |
88 | |
89 | void pcpu_freelist_push(struct pcpu_freelist *s, |
90 | struct pcpu_freelist_node *node) |
91 | { |
92 | unsigned long flags; |
93 | |
94 | local_irq_save(flags); |
95 | __pcpu_freelist_push(s, node); |
96 | local_irq_restore(flags); |
97 | } |
98 | |
99 | void pcpu_freelist_populate(struct pcpu_freelist *s, void *buf, u32 elem_size, |
100 | u32 nr_elems) |
101 | { |
102 | struct pcpu_freelist_head *head; |
103 | unsigned int cpu, cpu_idx, i, j, n, m; |
104 | |
105 | n = nr_elems / num_possible_cpus(); |
106 | m = nr_elems % num_possible_cpus(); |
107 | |
108 | cpu_idx = 0; |
109 | for_each_possible_cpu(cpu) { |
110 | head = per_cpu_ptr(s->freelist, cpu); |
111 | j = n + (cpu_idx < m ? 1 : 0); |
112 | for (i = 0; i < j; i++) { |
113 | /* No locking required as this is not visible yet. */ |
114 | pcpu_freelist_push_node(head, node: buf); |
115 | buf += elem_size; |
116 | } |
117 | cpu_idx++; |
118 | } |
119 | } |
120 | |
121 | static struct pcpu_freelist_node *___pcpu_freelist_pop(struct pcpu_freelist *s) |
122 | { |
123 | struct pcpu_freelist_head *head; |
124 | struct pcpu_freelist_node *node; |
125 | int cpu; |
126 | |
127 | for_each_cpu_wrap(cpu, cpu_possible_mask, raw_smp_processor_id()) { |
128 | head = per_cpu_ptr(s->freelist, cpu); |
129 | if (!READ_ONCE(head->first)) |
130 | continue; |
131 | raw_spin_lock(&head->lock); |
132 | node = head->first; |
133 | if (node) { |
134 | WRITE_ONCE(head->first, node->next); |
135 | raw_spin_unlock(&head->lock); |
136 | return node; |
137 | } |
138 | raw_spin_unlock(&head->lock); |
139 | } |
140 | |
141 | /* per cpu lists are all empty, try extralist */ |
142 | if (!READ_ONCE(s->extralist.first)) |
143 | return NULL; |
144 | raw_spin_lock(&s->extralist.lock); |
145 | node = s->extralist.first; |
146 | if (node) |
147 | WRITE_ONCE(s->extralist.first, node->next); |
148 | raw_spin_unlock(&s->extralist.lock); |
149 | return node; |
150 | } |
151 | |
152 | static struct pcpu_freelist_node * |
153 | ___pcpu_freelist_pop_nmi(struct pcpu_freelist *s) |
154 | { |
155 | struct pcpu_freelist_head *head; |
156 | struct pcpu_freelist_node *node; |
157 | int cpu; |
158 | |
159 | for_each_cpu_wrap(cpu, cpu_possible_mask, raw_smp_processor_id()) { |
160 | head = per_cpu_ptr(s->freelist, cpu); |
161 | if (!READ_ONCE(head->first)) |
162 | continue; |
163 | if (raw_spin_trylock(&head->lock)) { |
164 | node = head->first; |
165 | if (node) { |
166 | WRITE_ONCE(head->first, node->next); |
167 | raw_spin_unlock(&head->lock); |
168 | return node; |
169 | } |
170 | raw_spin_unlock(&head->lock); |
171 | } |
172 | } |
173 | |
174 | /* cannot pop from per cpu lists, try extralist */ |
175 | if (!READ_ONCE(s->extralist.first) || !raw_spin_trylock(&s->extralist.lock)) |
176 | return NULL; |
177 | node = s->extralist.first; |
178 | if (node) |
179 | WRITE_ONCE(s->extralist.first, node->next); |
180 | raw_spin_unlock(&s->extralist.lock); |
181 | return node; |
182 | } |
183 | |
184 | struct pcpu_freelist_node *__pcpu_freelist_pop(struct pcpu_freelist *s) |
185 | { |
186 | if (in_nmi()) |
187 | return ___pcpu_freelist_pop_nmi(s); |
188 | return ___pcpu_freelist_pop(s); |
189 | } |
190 | |
191 | struct pcpu_freelist_node *pcpu_freelist_pop(struct pcpu_freelist *s) |
192 | { |
193 | struct pcpu_freelist_node *ret; |
194 | unsigned long flags; |
195 | |
196 | local_irq_save(flags); |
197 | ret = __pcpu_freelist_pop(s); |
198 | local_irq_restore(flags); |
199 | return ret; |
200 | } |
201 | |