| 1 | #include <smp/core.h> |
| 2 | #include <smp/percpu.h> |
| 3 | #include <stdint.h> |
| 4 | #include <string.h> |
| 5 | |
| 6 | struct prng_core { |
| 7 | uint32_t state[16]; |
| 8 | uint8_t buffer[64]; // keystream buffer |
| 9 | size_t pos; |
| 10 | }; |
| 11 | |
| 12 | void prng_build(struct prng_core *this_one, size_t cpu) { |
| 13 | (void) this_one, (void) cpu; |
| 14 | } |
| 15 | |
| 16 | PERCPU_DECLARE(pcs, struct prng_core, prng_build); |
| 17 | #define prng_core_state PERCPU_READ(pcs) |
| 18 | |
| 19 | static void prng_seed_core(uint64_t seed) { |
| 20 | const char *sigma = "expand 32-byte k" ; |
| 21 | uint8_t key[32]; |
| 22 | |
| 23 | uint64_t tsc = smp_core()->last_tsc; |
| 24 | uint32_t core_id = smp_core()->id; |
| 25 | |
| 26 | for (int i = 0; i < 4; i++) { |
| 27 | key[i * 8 + 0] = (seed >> (i * 8 + 0)) & 0xff; |
| 28 | key[i * 8 + 1] = (seed >> (i * 8 + 8)) & 0xff; |
| 29 | |
| 30 | key[i * 8 + 2] = (tsc >> (i * 8 + 32)) & 0xff; |
| 31 | key[i * 8 + 3] = (tsc >> (i * 8 + 32)) & 0xff; |
| 32 | |
| 33 | key[i * 8 + 4] = (core_id >> (i * 8 + 0)) & 0xff; |
| 34 | key[i * 8 + 5] = (core_id >> (i * 8 + 4)) & 0xff; |
| 35 | |
| 36 | key[i * 8 + 6] = (tsc >> (i * 8 + 16)) & 0xff; |
| 37 | key[i * 8 + 7] = (tsc >> (i * 8 + 24)) & 0xff; |
| 38 | } |
| 39 | |
| 40 | for (int i = 0; i < 4; i++) |
| 41 | prng_core_state.state[i] = ((uint32_t *) sigma)[i]; |
| 42 | for (int i = 0; i < 8; i++) |
| 43 | prng_core_state.state[4 + i] = ((uint32_t *) key)[i]; |
| 44 | |
| 45 | prng_core_state.state[12] = 0; |
| 46 | prng_core_state.state[13] = 0; |
| 47 | prng_core_state.state[14] = 0; |
| 48 | prng_core_state.state[15] = 0; |
| 49 | |
| 50 | prng_core_state.pos = 64; |
| 51 | } |
| 52 | |
| 53 | #define QR(a, b, c, d) \ |
| 54 | a += b; \ |
| 55 | d ^= a; \ |
| 56 | d = (d << 16) | (d >> 16); \ |
| 57 | c += d; \ |
| 58 | b ^= c; \ |
| 59 | b = (b << 12) | (b >> 20); \ |
| 60 | a += b; \ |
| 61 | d ^= a; \ |
| 62 | d = (d << 8) | (d >> 24); \ |
| 63 | c += d; \ |
| 64 | b ^= c; \ |
| 65 | b = (b << 7) | (b >> 25) |
| 66 | |
| 67 | static void chacha20_generate_block(uint8_t out[64], uint32_t state[16]) { |
| 68 | uint32_t x[16]; |
| 69 | memcpy(x, state, sizeof(x)); |
| 70 | |
| 71 | for (int i = 0; i < 10; i++) { |
| 72 | QR(x[0], x[4], x[8], x[12]); |
| 73 | QR(x[1], x[5], x[9], x[13]); |
| 74 | QR(x[2], x[6], x[10], x[14]); |
| 75 | QR(x[3], x[7], x[11], x[15]); |
| 76 | QR(x[0], x[5], x[10], x[15]); |
| 77 | QR(x[1], x[6], x[11], x[12]); |
| 78 | QR(x[2], x[7], x[8], x[13]); |
| 79 | QR(x[3], x[4], x[9], x[14]); |
| 80 | } |
| 81 | |
| 82 | for (int i = 0; i < 16; i++) { |
| 83 | x[i] += state[i]; |
| 84 | out[i * 4 + 0] = x[i] & 0xff; |
| 85 | out[i * 4 + 1] = (x[i] >> 8) & 0xff; |
| 86 | out[i * 4 + 2] = (x[i] >> 16) & 0xff; |
| 87 | out[i * 4 + 3] = (x[i] >> 24) & 0xff; |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | uint64_t prng_next(void) { |
| 92 | if (prng_core_state.pos >= 64) { |
| 93 | chacha20_generate_block(prng_core_state.buffer, prng_core_state.state); |
| 94 | prng_core_state.state[12]++; |
| 95 | prng_core_state.pos = 0; |
| 96 | } |
| 97 | |
| 98 | uint64_t val; |
| 99 | memcpy(&val, prng_core_state.buffer + prng_core_state.pos, |
| 100 | sizeof(uint64_t)); |
| 101 | prng_core_state.pos += 8; |
| 102 | return val; |
| 103 | } |
| 104 | |
| 105 | void prng_seed(uint64_t seed) { |
| 106 | if (seed == 0) |
| 107 | seed = smp_core()->last_tsc; |
| 108 | prng_seed_core(seed); |
| 109 | } |
| 110 | |