Chris Wilson | cf4a720 | 2016-12-22 14:45:14 +0000 | [diff] [blame^] | 1 | #define pr_fmt(fmt) "prime numbers: " fmt "\n" |
| 2 | |
| 3 | #include <linux/module.h> |
| 4 | #include <linux/mutex.h> |
| 5 | #include <linux/prime_numbers.h> |
| 6 | #include <linux/slab.h> |
| 7 | |
| 8 | #define bitmap_size(nbits) (BITS_TO_LONGS(nbits) * sizeof(unsigned long)) |
| 9 | |
| 10 | struct primes { |
| 11 | struct rcu_head rcu; |
| 12 | unsigned long last, sz; |
| 13 | unsigned long primes[]; |
| 14 | }; |
| 15 | |
| 16 | #if BITS_PER_LONG == 64 |
| 17 | static const struct primes small_primes = { |
| 18 | .last = 61, |
| 19 | .sz = 64, |
| 20 | .primes = { |
| 21 | BIT(2) | |
| 22 | BIT(3) | |
| 23 | BIT(5) | |
| 24 | BIT(7) | |
| 25 | BIT(11) | |
| 26 | BIT(13) | |
| 27 | BIT(17) | |
| 28 | BIT(19) | |
| 29 | BIT(23) | |
| 30 | BIT(29) | |
| 31 | BIT(31) | |
| 32 | BIT(37) | |
| 33 | BIT(41) | |
| 34 | BIT(43) | |
| 35 | BIT(47) | |
| 36 | BIT(53) | |
| 37 | BIT(59) | |
| 38 | BIT(61) |
| 39 | } |
| 40 | }; |
| 41 | #elif BITS_PER_LONG == 32 |
| 42 | static const struct primes small_primes = { |
| 43 | .last = 31, |
| 44 | .sz = 32, |
| 45 | .primes = { |
| 46 | BIT(2) | |
| 47 | BIT(3) | |
| 48 | BIT(5) | |
| 49 | BIT(7) | |
| 50 | BIT(11) | |
| 51 | BIT(13) | |
| 52 | BIT(17) | |
| 53 | BIT(19) | |
| 54 | BIT(23) | |
| 55 | BIT(29) | |
| 56 | BIT(31) |
| 57 | } |
| 58 | }; |
| 59 | #else |
| 60 | #error "unhandled BITS_PER_LONG" |
| 61 | #endif |
| 62 | |
| 63 | static DEFINE_MUTEX(lock); |
| 64 | static const struct primes __rcu *primes = RCU_INITIALIZER(&small_primes); |
| 65 | |
| 66 | static unsigned long selftest_max; |
| 67 | |
| 68 | static bool slow_is_prime_number(unsigned long x) |
| 69 | { |
| 70 | unsigned long y = int_sqrt(x); |
| 71 | |
| 72 | while (y > 1) { |
| 73 | if ((x % y) == 0) |
| 74 | break; |
| 75 | y--; |
| 76 | } |
| 77 | |
| 78 | return y == 1; |
| 79 | } |
| 80 | |
| 81 | static unsigned long slow_next_prime_number(unsigned long x) |
| 82 | { |
| 83 | while (x < ULONG_MAX && !slow_is_prime_number(++x)) |
| 84 | ; |
| 85 | |
| 86 | return x; |
| 87 | } |
| 88 | |
| 89 | static unsigned long clear_multiples(unsigned long x, |
| 90 | unsigned long *p, |
| 91 | unsigned long start, |
| 92 | unsigned long end) |
| 93 | { |
| 94 | unsigned long m; |
| 95 | |
| 96 | m = 2 * x; |
| 97 | if (m < start) |
| 98 | m = roundup(start, x); |
| 99 | |
| 100 | while (m < end) { |
| 101 | __clear_bit(m, p); |
| 102 | m += x; |
| 103 | } |
| 104 | |
| 105 | return x; |
| 106 | } |
| 107 | |
| 108 | static bool expand_to_next_prime(unsigned long x) |
| 109 | { |
| 110 | const struct primes *p; |
| 111 | struct primes *new; |
| 112 | unsigned long sz, y; |
| 113 | |
| 114 | /* Betrand's Postulate (or Chebyshev's theorem) states that if n > 3, |
| 115 | * there is always at least one prime p between n and 2n - 2. |
| 116 | * Equivalently, if n > 1, then there is always at least one prime p |
| 117 | * such that n < p < 2n. |
| 118 | * |
| 119 | * http://mathworld.wolfram.com/BertrandsPostulate.html |
| 120 | * https://en.wikipedia.org/wiki/Bertrand's_postulate |
| 121 | */ |
| 122 | sz = 2 * x; |
| 123 | if (sz < x) |
| 124 | return false; |
| 125 | |
| 126 | sz = round_up(sz, BITS_PER_LONG); |
| 127 | new = kmalloc(sizeof(*new) + bitmap_size(sz), GFP_KERNEL); |
| 128 | if (!new) |
| 129 | return false; |
| 130 | |
| 131 | mutex_lock(&lock); |
| 132 | p = rcu_dereference_protected(primes, lockdep_is_held(&lock)); |
| 133 | if (x < p->last) { |
| 134 | kfree(new); |
| 135 | goto unlock; |
| 136 | } |
| 137 | |
| 138 | /* Where memory permits, track the primes using the |
| 139 | * Sieve of Eratosthenes. The sieve is to remove all multiples of known |
| 140 | * primes from the set, what remains in the set is therefore prime. |
| 141 | */ |
| 142 | bitmap_fill(new->primes, sz); |
| 143 | bitmap_copy(new->primes, p->primes, p->sz); |
| 144 | for (y = 2UL; y < sz; y = find_next_bit(new->primes, sz, y + 1)) |
| 145 | new->last = clear_multiples(y, new->primes, p->sz, sz); |
| 146 | new->sz = sz; |
| 147 | |
| 148 | BUG_ON(new->last <= x); |
| 149 | |
| 150 | rcu_assign_pointer(primes, new); |
| 151 | if (p != &small_primes) |
| 152 | kfree_rcu((struct primes *)p, rcu); |
| 153 | |
| 154 | unlock: |
| 155 | mutex_unlock(&lock); |
| 156 | return true; |
| 157 | } |
| 158 | |
| 159 | static void free_primes(void) |
| 160 | { |
| 161 | const struct primes *p; |
| 162 | |
| 163 | mutex_lock(&lock); |
| 164 | p = rcu_dereference_protected(primes, lockdep_is_held(&lock)); |
| 165 | if (p != &small_primes) { |
| 166 | rcu_assign_pointer(primes, &small_primes); |
| 167 | kfree_rcu((struct primes *)p, rcu); |
| 168 | } |
| 169 | mutex_unlock(&lock); |
| 170 | } |
| 171 | |
| 172 | /** |
| 173 | * next_prime_number - return the next prime number |
| 174 | * @x: the starting point for searching to test |
| 175 | * |
| 176 | * A prime number is an integer greater than 1 that is only divisible by |
| 177 | * itself and 1. The set of prime numbers is computed using the Sieve of |
| 178 | * Eratoshenes (on finding a prime, all multiples of that prime are removed |
| 179 | * from the set) enabling a fast lookup of the next prime number larger than |
| 180 | * @x. If the sieve fails (memory limitation), the search falls back to using |
| 181 | * slow trial-divison, up to the value of ULONG_MAX (which is reported as the |
| 182 | * final prime as a sentinel). |
| 183 | * |
| 184 | * Returns: the next prime number larger than @x |
| 185 | */ |
| 186 | unsigned long next_prime_number(unsigned long x) |
| 187 | { |
| 188 | const struct primes *p; |
| 189 | |
| 190 | rcu_read_lock(); |
| 191 | p = rcu_dereference(primes); |
| 192 | while (x >= p->last) { |
| 193 | rcu_read_unlock(); |
| 194 | |
| 195 | if (!expand_to_next_prime(x)) |
| 196 | return slow_next_prime_number(x); |
| 197 | |
| 198 | rcu_read_lock(); |
| 199 | p = rcu_dereference(primes); |
| 200 | } |
| 201 | x = find_next_bit(p->primes, p->last, x + 1); |
| 202 | rcu_read_unlock(); |
| 203 | |
| 204 | return x; |
| 205 | } |
| 206 | EXPORT_SYMBOL(next_prime_number); |
| 207 | |
| 208 | /** |
| 209 | * is_prime_number - test whether the given number is prime |
| 210 | * @x: the number to test |
| 211 | * |
| 212 | * A prime number is an integer greater than 1 that is only divisible by |
| 213 | * itself and 1. Internally a cache of prime numbers is kept (to speed up |
| 214 | * searching for sequential primes, see next_prime_number()), but if the number |
| 215 | * falls outside of that cache, its primality is tested using trial-divison. |
| 216 | * |
| 217 | * Returns: true if @x is prime, false for composite numbers. |
| 218 | */ |
| 219 | bool is_prime_number(unsigned long x) |
| 220 | { |
| 221 | const struct primes *p; |
| 222 | bool result; |
| 223 | |
| 224 | rcu_read_lock(); |
| 225 | p = rcu_dereference(primes); |
| 226 | while (x >= p->sz) { |
| 227 | rcu_read_unlock(); |
| 228 | |
| 229 | if (!expand_to_next_prime(x)) |
| 230 | return slow_is_prime_number(x); |
| 231 | |
| 232 | rcu_read_lock(); |
| 233 | p = rcu_dereference(primes); |
| 234 | } |
| 235 | result = test_bit(x, p->primes); |
| 236 | rcu_read_unlock(); |
| 237 | |
| 238 | return result; |
| 239 | } |
| 240 | EXPORT_SYMBOL(is_prime_number); |
| 241 | |
| 242 | static void dump_primes(void) |
| 243 | { |
| 244 | const struct primes *p; |
| 245 | char *buf; |
| 246 | |
| 247 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
| 248 | |
| 249 | rcu_read_lock(); |
| 250 | p = rcu_dereference(primes); |
| 251 | |
| 252 | if (buf) |
| 253 | bitmap_print_to_pagebuf(true, buf, p->primes, p->sz); |
| 254 | pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...x%lx} = %s", |
| 255 | p->last, p->sz, p->primes[BITS_TO_LONGS(p->sz) - 1], buf); |
| 256 | |
| 257 | rcu_read_unlock(); |
| 258 | |
| 259 | kfree(buf); |
| 260 | } |
| 261 | |
| 262 | static int selftest(unsigned long max) |
| 263 | { |
| 264 | unsigned long x, last; |
| 265 | |
| 266 | if (!max) |
| 267 | return 0; |
| 268 | |
| 269 | for (last = 0, x = 2; x < max; x++) { |
| 270 | bool slow = slow_is_prime_number(x); |
| 271 | bool fast = is_prime_number(x); |
| 272 | |
| 273 | if (slow != fast) { |
| 274 | pr_err("inconsistent result for is-prime(%lu): slow=%s, fast=%s!", |
| 275 | x, slow ? "yes" : "no", fast ? "yes" : "no"); |
| 276 | goto err; |
| 277 | } |
| 278 | |
| 279 | if (!slow) |
| 280 | continue; |
| 281 | |
| 282 | if (next_prime_number(last) != x) { |
| 283 | pr_err("incorrect result for next-prime(%lu): expected %lu, got %lu", |
| 284 | last, x, next_prime_number(last)); |
| 285 | goto err; |
| 286 | } |
| 287 | last = x; |
| 288 | } |
| 289 | |
| 290 | pr_info("selftest(%lu) passed, last prime was %lu", x, last); |
| 291 | return 0; |
| 292 | |
| 293 | err: |
| 294 | dump_primes(); |
| 295 | return -EINVAL; |
| 296 | } |
| 297 | |
| 298 | static int __init primes_init(void) |
| 299 | { |
| 300 | return selftest(selftest_max); |
| 301 | } |
| 302 | |
| 303 | static void __exit primes_exit(void) |
| 304 | { |
| 305 | free_primes(); |
| 306 | } |
| 307 | |
| 308 | module_init(primes_init); |
| 309 | module_exit(primes_exit); |
| 310 | |
| 311 | module_param_named(selftest, selftest_max, ulong, 0400); |
| 312 | |
| 313 | MODULE_AUTHOR("Intel Corporation"); |
| 314 | MODULE_LICENSE("GPL"); |