1 /**
2 * \brief Multi-precision integer library, ESP32 S2 hardware accelerated parts
3 *
4 * based on mbedTLS implementation
5 *
6 * Copyright (C) 2006-2015, ARM Limited, All Rights Reserved
7 * Additions Copyright (C) 2016-2020, Espressif Systems (Shanghai) PTE Ltd
8 * SPDX-License-Identifier: Apache-2.0
9 *
10 * Licensed under the Apache License, Version 2.0 (the "License"); you may
11 * not use this file except in compliance with the License.
12 * You may obtain a copy of the License at
13 *
14 * http://www.apache.org/licenses/LICENSE-2.0
15 *
16 * Unless required by applicable law or agreed to in writing, software
17 * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
18 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
19 * See the License for the specific language governing permissions and
20 * limitations under the License.
21 *
22 */
23 #include "soc/hwcrypto_periph.h"
24 #include "driver/periph_ctrl.h"
25 #include <mbedtls/bignum.h>
26 #include "bignum_impl.h"
27 #include "soc/dport_reg.h"
28 #include "soc/periph_defs.h"
29 #include <sys/param.h>
30 #include "esp_crypto_lock.h"
31
esp_mpi_hardware_words(size_t words)32 size_t esp_mpi_hardware_words(size_t words)
33 {
34 return words;
35 }
36
esp_mpi_enable_hardware_hw_op(void)37 void esp_mpi_enable_hardware_hw_op( void )
38 {
39 esp_crypto_mpi_lock_acquire();
40
41 /* Enable RSA hardware */
42 periph_module_enable(PERIPH_RSA_MODULE);
43
44 DPORT_REG_CLR_BIT(DPORT_RSA_PD_CTRL_REG, DPORT_RSA_MEM_PD);
45
46 while (DPORT_REG_READ(RSA_QUERY_CLEAN_REG) != 1) {
47 }
48 // Note: from enabling RSA clock to here takes about 1.3us
49 }
50
esp_mpi_disable_hardware_hw_op(void)51 void esp_mpi_disable_hardware_hw_op( void )
52 {
53 DPORT_REG_SET_BIT(DPORT_RSA_PD_CTRL_REG, DPORT_RSA_PD);
54
55 /* Disable RSA hardware */
56 periph_module_disable(PERIPH_RSA_MODULE);
57
58 esp_crypto_mpi_lock_release();
59 }
60
61
62 /* Copy mbedTLS MPI bignum 'mpi' to hardware memory block at 'mem_base'.
63
64 If num_words is higher than the number of words in the bignum then
65 these additional words will be zeroed in the memory buffer.
66 */
mpi_to_mem_block(uint32_t mem_base,const mbedtls_mpi * mpi,size_t num_words)67 static inline void mpi_to_mem_block(uint32_t mem_base, const mbedtls_mpi *mpi, size_t num_words)
68 {
69 uint32_t *pbase = (uint32_t *)mem_base;
70 uint32_t copy_words = MIN(num_words, mpi->n);
71
72 /* Copy MPI data to memory block registers */
73 for (uint32_t i = 0; i < copy_words; i++) {
74 pbase[i] = mpi->p[i];
75 }
76
77 /* Zero any remaining memory block data */
78 for (uint32_t i = copy_words; i < num_words; i++) {
79 pbase[i] = 0;
80 }
81 }
82
83 /* Read mbedTLS MPI bignum back from hardware memory block.
84
85 Reads num_words words from block.
86 */
mem_block_to_mpi(mbedtls_mpi * x,uint32_t mem_base,int num_words)87 static inline void mem_block_to_mpi(mbedtls_mpi *x, uint32_t mem_base, int num_words)
88 {
89
90 /* Copy data from memory block registers */
91 esp_dport_access_read_buffer(x->p, mem_base, num_words);
92 /* Zero any remaining limbs in the bignum, if the buffer is bigger
93 than num_words */
94 for (size_t i = num_words; i < x->n; i++) {
95 x->p[i] = 0;
96 }
97 }
98
99
100
101 /* Begin an RSA operation. op_reg specifies which 'START' register
102 to write to.
103 */
start_op(uint32_t op_reg)104 static inline void start_op(uint32_t op_reg)
105 {
106 /* Clear interrupt status */
107 DPORT_REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
108
109 /* Note: above REG_WRITE includes a memw, so we know any writes
110 to the memory blocks are also complete. */
111
112 DPORT_REG_WRITE(op_reg, 1);
113 }
114
115 /* Wait for an RSA operation to complete.
116 */
wait_op_complete(void)117 static inline void wait_op_complete(void)
118 {
119 while (DPORT_REG_READ(RSA_QUERY_INTERRUPT_REG) != 1)
120 { }
121
122 /* clear the interrupt */
123 DPORT_REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
124 }
125
126
127 /* Read result from last MPI operation */
esp_mpi_read_result_hw_op(mbedtls_mpi * Z,size_t z_words)128 void esp_mpi_read_result_hw_op(mbedtls_mpi *Z, size_t z_words)
129 {
130 wait_op_complete();
131 mem_block_to_mpi(Z, RSA_MEM_Z_BLOCK_BASE, z_words);
132 }
133
134
135 /* Z = (X * Y) mod M
136
137 Not an mbedTLS function
138 */
esp_mpi_mul_mpi_mod_hw_op(const mbedtls_mpi * X,const mbedtls_mpi * Y,const mbedtls_mpi * M,const mbedtls_mpi * Rinv,mbedtls_mpi_uint Mprime,size_t num_words)139 void esp_mpi_mul_mpi_mod_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, const mbedtls_mpi *M, const mbedtls_mpi *Rinv, mbedtls_mpi_uint Mprime, size_t num_words)
140 {
141 DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
142
143 /* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
144 mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
145 mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
146 mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
147 mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
148 DPORT_REG_WRITE(RSA_M_DASH_REG, Mprime);
149
150 start_op(RSA_MOD_MULT_START_REG);
151 }
152
153 /* Z = (X ^ Y) mod M
154 */
esp_mpi_exp_mpi_mod_hw_op(const mbedtls_mpi * X,const mbedtls_mpi * Y,const mbedtls_mpi * M,const mbedtls_mpi * Rinv,mbedtls_mpi_uint Mprime,size_t num_words)155 void esp_mpi_exp_mpi_mod_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, const mbedtls_mpi *M, const mbedtls_mpi *Rinv, mbedtls_mpi_uint Mprime, size_t num_words)
156 {
157 size_t y_bits = mbedtls_mpi_bitlen(Y);
158
159 DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
160
161 /* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
162 mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
163 mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
164 mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
165 mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
166 DPORT_REG_WRITE(RSA_M_DASH_REG, Mprime);
167
168 /* Enable acceleration options */
169 DPORT_REG_WRITE(RSA_CONSTANT_TIME_REG, 0);
170 DPORT_REG_WRITE(RSA_SEARCH_OPEN_REG, 1);
171 DPORT_REG_WRITE(RSA_SEARCH_POS_REG, y_bits - 1);
172
173 /* Execute first stage montgomery multiplication */
174 start_op(RSA_MODEXP_START_REG);
175
176 DPORT_REG_WRITE(RSA_SEARCH_OPEN_REG, 0);
177 }
178
179
180 /* Z = X * Y */
esp_mpi_mul_mpi_hw_op(const mbedtls_mpi * X,const mbedtls_mpi * Y,size_t num_words)181 void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
182 {
183 /* Copy X (right-extended) & Y (left-extended) to memory block */
184 mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
185 mpi_to_mem_block(RSA_MEM_Z_BLOCK_BASE + num_words * 4, Y, num_words);
186 /* NB: as Y is left-extended, we don't zero the bottom words_mult words of Y block.
187 This is OK for now because zeroing is done by hardware when we do esp_mpi_acquire_hardware().
188 */
189 DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words * 2 - 1));
190 start_op(RSA_MULT_START_REG);
191 }
192
193
194
195 /**
196 * @brief Special-case of (X * Y), where we use hardware montgomery mod
197 multiplication to calculate result where either A or B are >2048 bits so
198 can't use the standard multiplication method.
199 *
200 */
esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi * X,const mbedtls_mpi * Y,size_t num_words)201 void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
202 {
203 /* M = 2^num_words - 1, so block is entirely FF */
204 for (size_t i = 0; i < num_words; i++) {
205 DPORT_REG_WRITE(RSA_MEM_M_BLOCK_BASE + i * 4, UINT32_MAX);
206 }
207
208 /* Mprime = 1 */
209 DPORT_REG_WRITE(RSA_M_DASH_REG, 1);
210 DPORT_REG_WRITE(RSA_LENGTH_REG, num_words - 1);
211
212 /* Load X & Y */
213 mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
214 mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
215
216 /* Rinv = 1, write first word */
217 DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE, 1);
218
219 /* Zero out rest of the Rinv words */
220 for (size_t i = 1; i < num_words; i++) {
221 DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE + i * 4, 0);
222 }
223
224 start_op(RSA_MOD_MULT_START_REG);
225 }
226