| /* |
| * Copyright (c) 2021-2022, ARM Limited. All rights reserved. |
| * |
| * SPDX-License-Identifier: BSD-3-Clause |
| */ |
| |
| #include <assert.h> |
| #include <stdbool.h> |
| #include <stdint.h> |
| #include <lib/spinlock.h> |
| #include <plat/common/plat_trng.h> |
| |
| /* |
| * # Entropy pool |
| * Note that the TRNG Firmware interface can request up to 192 bits of entropy |
| * in a single call or three 64bit words per call. We have 4 words in the pool |
| * so that when we have 1-63 bits in the pool, and we have a request for |
| * 192 bits of entropy, we don't have to throw out the leftover 1-63 bits of |
| * entropy. |
| */ |
| #define WORDS_IN_POOL (4) |
| static uint64_t entropy[WORDS_IN_POOL]; |
| /* index in bits of the first bit of usable entropy */ |
| static uint32_t entropy_bit_index; |
| /* then number of valid bits in the entropy pool */ |
| static uint32_t entropy_bit_size; |
| |
| static spinlock_t trng_pool_lock; |
| |
| #define BITS_PER_WORD (sizeof(entropy[0]) * 8) |
| #define BITS_IN_POOL (WORDS_IN_POOL * BITS_PER_WORD) |
| #define ENTROPY_MIN_WORD (entropy_bit_index / BITS_PER_WORD) |
| #define ENTROPY_FREE_BIT (entropy_bit_size + entropy_bit_index) |
| #define _ENTROPY_FREE_WORD (ENTROPY_FREE_BIT / BITS_PER_WORD) |
| #define ENTROPY_FREE_INDEX (_ENTROPY_FREE_WORD % WORDS_IN_POOL) |
| /* ENTROPY_WORD_INDEX(0) includes leftover bits in the lower bits */ |
| #define ENTROPY_WORD_INDEX(i) ((ENTROPY_MIN_WORD + i) % WORDS_IN_POOL) |
| |
| /* |
| * Fill the entropy pool until we have at least as many bits as requested. |
| * Returns true after filling the pool, and false if the entropy source is out |
| * of entropy and the pool could not be filled. |
| * Assumes locks are taken. |
| */ |
| static bool trng_fill_entropy(uint32_t nbits) |
| { |
| while (nbits > entropy_bit_size) { |
| bool valid = plat_get_entropy(&entropy[ENTROPY_FREE_INDEX]); |
| |
| if (valid) { |
| entropy_bit_size += BITS_PER_WORD; |
| assert(entropy_bit_size <= BITS_IN_POOL); |
| } else { |
| return false; |
| } |
| } |
| return true; |
| } |
| |
| /* |
| * Pack entropy into the out buffer, filling and taking locks as needed. |
| * Returns true on success, false on failure. |
| * |
| * Note: out must have enough space for nbits of entropy |
| */ |
| bool trng_pack_entropy(uint32_t nbits, uint64_t *out) |
| { |
| bool ret = true; |
| uint32_t bits_to_discard = nbits; |
| spin_lock(&trng_pool_lock); |
| |
| if (!trng_fill_entropy(nbits)) { |
| ret = false; |
| goto out; |
| } |
| |
| const unsigned int rshift = entropy_bit_index % BITS_PER_WORD; |
| const unsigned int lshift = BITS_PER_WORD - rshift; |
| const int to_fill = ((nbits + BITS_PER_WORD - 1) / BITS_PER_WORD); |
| int word_i; |
| |
| for (word_i = 0; word_i < to_fill; word_i++) { |
| /* |
| * Repack the entropy from the pool into the passed in out |
| * buffer. This takes lesser bits from the valid upper bits |
| * of word_i and more bits from the lower bits of (word_i + 1). |
| * |
| * I found the following diagram useful. note: `e` represents |
| * valid entropy, ` ` represents invalid bits (not entropy) and |
| * `x` represents valid entropy that must not end up in the |
| * packed word. |
| * |
| * |---------entropy pool----------| |
| * C var |--(word_i + 1)-|----word_i-----| |
| * bit idx |7 6 5 4 3 2 1 0|7 6 5 4 3 2 1 0| |
| * [x,x,e,e,e,e,e,e|e,e, , , , , , ] |
| * | [e,e,e,e,e,e,e,e] | |
| * | |--out[word_i]--| | |
| * lshift|---| |--rshift---| |
| * |
| * ==== Which is implemented as ==== |
| * |
| * |---------entropy pool----------| |
| * C var |--(word_i + 1)-|----word_i-----| |
| * bit idx |7 6 5 4 3 2 1 0|7 6 5 4 3 2 1 0| |
| * [x,x,e,e,e,e,e,e|e,e, , , , , , ] |
| * C expr << lshift >> rshift |
| * bit idx 5 4 3 2 1 0 7 6 |
| * [e,e,e,e,e,e,0,0|0,0,0,0,0,0,e,e] |
| * ==== bit-wise or ==== |
| * 5 4 3 2 1 0 7 6 |
| * [e,e,e,e,e,e,e,e] |
| */ |
| out[word_i] |= entropy[ENTROPY_WORD_INDEX(word_i)] >> rshift; |
| |
| /** |
| * Discarding the used/packed entropy bits from the respective |
| * words, (word_i) and (word_i+1) as applicable. |
| * In each iteration of the loop, we pack 64bits of entropy to |
| * the output buffer. The bits are picked linearly starting from |
| * 1st word (entropy[0]) till 4th word (entropy[3]) and then |
| * rolls back (entropy[0]). Discarding of bits is managed |
| * similarly. |
| * |
| * The following diagram illustrates the logic: |
| * |
| * |---------entropy pool----------| |
| * C var |--(word_i + 1)-|----word_i-----| |
| * bit idx |7 6 5 4 3 2 1 0|7 6 5 4 3 2 1 0| |
| * [e,e,e,e,e,e,e,e|e,e,0,0,0,0,0,0] |
| * | [e,e,e,e,e,e,e,e] | |
| * | |--out[word_i]--| | |
| * lshift|---| |--rshift---| |
| * |e,e|0,0,0,0,0,0,0,0|0,0,0,0,0,0| |
| * |<== || ==>| |
| * bits_to_discard (from these bytes) |
| * |
| * variable(bits_to_discard): Tracks the amount of bits to be |
| * discarded and is updated accordingly in each iteration. |
| * |
| * It monitors these packed bits from respective word_i and |
| * word_i+1 and overwrites them with zeros accordingly. |
| * It discards linearly from the lowest index and moves upwards |
| * until bits_to_discard variable becomes zero. |
| * |
| * In the above diagram,for example, we pack 2bytes(7th and 6th |
| * from word_i) and 6bytes(0th till 5th from word_i+1), combine |
| * and pack them as 64bit to output buffer out[i]. |
| * Depending on the number of bits requested, we discard the |
| * bits from these packed bytes by overwriting them with zeros. |
| */ |
| |
| /* |
| * If the bits to be discarded is lesser than the amount of bits |
| * copied to the output buffer from word_i, we discard that much |
| * amount of bits only. |
| */ |
| if (bits_to_discard < (BITS_PER_WORD - rshift)) { |
| entropy[ENTROPY_WORD_INDEX(word_i)] &= |
| (~0ULL << ((bits_to_discard+rshift) % BITS_PER_WORD)); |
| bits_to_discard = 0; |
| } else { |
| /* |
| * If the bits to be discarded is more than the amount of valid |
| * upper bits from word_i, which has been copied to the output |
| * buffer, we just set the entire word_i to 0, as the lower bits |
| * will be already zeros from previous operations, and the |
| * bits_to_discard is updated precisely. |
| */ |
| entropy[ENTROPY_WORD_INDEX(word_i)] = 0; |
| bits_to_discard -= (BITS_PER_WORD - rshift); |
| } |
| |
| /* |
| * Note that a shift of 64 bits is treated as a shift of 0 bits. |
| * When the shift amount is the same as the BITS_PER_WORD, we |
| * don't want to include the next word of entropy, so we skip |
| * the `|=` operation. |
| */ |
| if (lshift != BITS_PER_WORD) { |
| out[word_i] |= entropy[ENTROPY_WORD_INDEX(word_i + 1)] |
| << lshift; |
| /** |
| * Discarding the remaining packed bits from upperword |
| * (word[i+1]) which was copied to output buffer by |
| * overwriting with zeros. |
| * |
| * If the remaining bits to be discarded is lesser than |
| * the amount of bits from [word_i+1], which has been |
| * copied to the output buffer, we overwrite that much |
| * amount of bits only. |
| */ |
| if (bits_to_discard < (BITS_PER_WORD - lshift)) { |
| entropy[ENTROPY_WORD_INDEX(word_i+1)] &= |
| (~0ULL << ((bits_to_discard) % BITS_PER_WORD)); |
| bits_to_discard = 0; |
| } else { |
| /* |
| * If bits to discard is more than the bits from word_i+1 |
| * which got packed into the output, then we discard all |
| * those copied bits. |
| * |
| * Note: we cannot set the entire word_i+1 to 0, as |
| * there are still some unused valid entropy bits at the |
| * upper end for future use. |
| */ |
| entropy[ENTROPY_WORD_INDEX(word_i+1)] &= |
| (~0ULL << ((BITS_PER_WORD - lshift) % BITS_PER_WORD)); |
| bits_to_discard -= (BITS_PER_WORD - lshift); |
| } |
| |
| } |
| } |
| const uint64_t mask = ~0ULL >> (BITS_PER_WORD - (nbits % BITS_PER_WORD)); |
| |
| out[to_fill - 1] &= mask; |
| |
| entropy_bit_index = (entropy_bit_index + nbits) % BITS_IN_POOL; |
| entropy_bit_size -= nbits; |
| |
| out: |
| spin_unlock(&trng_pool_lock); |
| |
| return ret; |
| } |
| |
| void trng_entropy_pool_setup(void) |
| { |
| int i; |
| |
| for (i = 0; i < WORDS_IN_POOL; i++) { |
| entropy[i] = 0; |
| } |
| entropy_bit_index = 0; |
| entropy_bit_size = 0; |
| } |