espressif/esp-idf
1
0
Fork 1
mirror of https://github.com/espressif/esp-idf.git synced 2025-11-02 08:01:43 +01:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'feature/add_bignum_ll_layer' into 'master'

Browse Source 
bignum: added bignum hal and ll layer

Closes IDF-7071

See merge request espressif/esp-idf!22823
This commit is contained in:
Mahavir Jain
2023-04-25 19:50:25 +08:00
parent e2a5c092f7 fbb8937758
commit a8b6a70620
51 changed files with 2345 additions and 1475 deletions
Download Patch File Download Diff File Expand all files Collapse all files

227
components/mbedtls/port/bignum/bignum_alt.c Normal file
View File

@@ -0,0 +1,227 @@
/*
* SPDX-FileCopyrightText: 2023 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
#include "esp_crypto_lock.h"
#include "esp_private/periph_ctrl.h"
#include "bignum_impl.h"
#include "mbedtls/bignum.h"
#include "hal/mpi_hal.h"
void esp_mpi_enable_hardware_hw_op( void )
{
esp_crypto_mpi_lock_acquire();
/* Enable RSA hardware */
periph_module_enable(PERIPH_RSA_MODULE);
mpi_hal_enable_hardware_hw_op();
}
void esp_mpi_disable_hardware_hw_op( void )
{
mpi_hal_disable_hardware_hw_op();
/* Disable RSA hardware */
periph_module_disable(PERIPH_RSA_MODULE);
esp_crypto_mpi_lock_release();
}
size_t esp_mpi_hardware_words(size_t words)
{
return mpi_hal_calc_hardware_words(words);
}
void esp_mpi_interrupt_enable(bool enable)
{
mpi_hal_interrupt_enable(enable);
}
void esp_mpi_interrupt_clear(void)
{
mpi_hal_clear_interrupt();
}
/* Z = (X * Y) mod M */
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)
{
#if CONFIG_IDF_TARGET_ESP32
/* "mode" register loaded with number of 512-bit blocks, minus 1 */
mpi_hal_set_mode((num_words / 16) - 1);
#else
mpi_hal_set_mode(num_words - 1);
#endif
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_hal_write_to_mem_block(MPI_PARAM_M, 0, M->MBEDTLS_PRIVATE(p), M->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_write_to_mem_block(MPI_PARAM_X, 0, X->MBEDTLS_PRIVATE(p), X->MBEDTLS_PRIVATE(n), num_words);
#if !CONFIG_IDF_TARGET_ESP32
mpi_hal_write_to_mem_block(MPI_PARAM_Y, 0, Y->MBEDTLS_PRIVATE(p), Y->MBEDTLS_PRIVATE(n), num_words);
#endif
mpi_hal_write_to_mem_block(MPI_PARAM_Z, 0, Rinv->MBEDTLS_PRIVATE(p), Rinv->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_write_m_prime(Mprime);
#if CONFIG_IDF_TARGET_ESP32
mpi_hal_start_op(MPI_MULT);
mpi_hal_wait_op_complete();
/* execute second stage */
/* Load Y to X input memory block, rerun */
mpi_hal_write_to_mem_block(MPI_PARAM_X, 0, Y->MBEDTLS_PRIVATE(p), Y->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_start_op(MPI_MULT);
#else
mpi_hal_start_op(MPI_MODMULT);
#endif
}
/* Z = X * Y */
void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* Copy X (right-extended) & Y (left-extended) to memory block */
mpi_hal_write_to_mem_block(MPI_PARAM_X, 0, X->MBEDTLS_PRIVATE(p), X->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_write_to_mem_block(MPI_PARAM_Z, num_words * 4, Y->MBEDTLS_PRIVATE(p), Y->MBEDTLS_PRIVATE(n), num_words);
/* NB: as Y is left-exte, we don't zero the bottom words_mult words of Y block.
This is OK for now bec zeroing is done by hardware when we do esp_mpi_acquire_hardware().
*/
#if CONFIG_IDF_TARGET_ESP32
mpi_hal_write_m_prime(0);
/* "mode" register loaded with number of 512-bit blocks in result,
plus 7 (for range 9-12). (this is ((N~ / 32) - 1) + 8))
*/
mpi_hal_set_mode(((num_words * 2) / 16) + 7);
#else
mpi_hal_set_mode(num_words * 2 - 1);
#endif
mpi_hal_start_op(MPI_MULT);
}
/* Special-case of mbedtls_mpi_mult_mpi(), where we use hardware montgomery mod
multiplication to calculate an mbedtls_mpi_mult_mpi result where either
A or B are >2048 bits so can't use the standard multiplication method.
Result (number of words, based on A bits + B bits) must still be less than 4096 bits.
This case is simpler than the general case modulo multiply of
esp_mpi_mul_mpi_mod() because we can control the other arguments:
* Modulus is chosen with M=(2^num_bits - 1) (ie M=R-1), so output
* Mprime and Rinv are therefore predictable as follows:
isn't actually modulo anything.
Mprime 1
Rinv 1
(See RSA Accelerator section in Technical Reference for more about Mprime, Rinv)
*/
void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* M = 2^num_words - 1, so block is entirely FF */
for (int i = 0; i < num_words; i++) {
mpi_hal_write_at_offset(MPI_PARAM_M, i * 4, UINT32_MAX);
}
/* Mprime = 1 */
mpi_hal_write_m_prime(1);
#if CONFIG_IDF_TARGET_ESP32
/* "mode" register loaded with number of 512-bit blocks, minus 1 */
mpi_hal_set_mode((num_words / 16) - 1);
#else
mpi_hal_set_mode(num_words - 1);
#endif
/* Load X & Y */
mpi_hal_write_to_mem_block(MPI_PARAM_X, 0, X->MBEDTLS_PRIVATE(p), X->MBEDTLS_PRIVATE(n), num_words);
#if !CONFIG_IDF_TARGET_ESP32
mpi_hal_write_to_mem_block(MPI_PARAM_Y, 0, Y->MBEDTLS_PRIVATE(p), Y->MBEDTLS_PRIVATE(n), num_words);
#endif
/* Rinv = 1, write first word */
mpi_hal_write_rinv(1);
/* Zero out rest of the Rinv words */
for (int i = 1; i < num_words; i++) {
mpi_hal_write_at_offset(MPI_PARAM_Z, i * 4, 0);
}
#if CONFIG_IDF_TARGET_ESP32
mpi_hal_start_op(MPI_MULT);
mpi_hal_wait_op_complete();
mpi_hal_write_to_mem_block(MPI_PARAM_X, 0, Y->MBEDTLS_PRIVATE(p), Y->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_start_op(MPI_MULT);
#else
mpi_hal_start_op(MPI_MODMULT);
#endif
}
#ifdef ESP_MPI_USE_MONT_EXP
int esp_mont_hw_op(mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_mpi *Y, const mbedtls_mpi *M, mbedtls_mpi_uint Mprime, size_t hw_words, bool again)
{
// Note Z may be the same pointer as X or Y
int ret = 0;
// montgomery mult prepare
if (again == false) {
mpi_hal_write_to_mem_block(MPI_PARAM_M, 0, M->MBEDTLS_PRIVATE(p), M->MBEDTLS_PRIVATE(n), hw_words);
mpi_hal_write_m_prime(Mprime);
mpi_hal_set_mode((hw_words / 16) - 1);
}
mpi_hal_write_to_mem_block(MPI_PARAM_X, 0, X->MBEDTLS_PRIVATE(p), X->MBEDTLS_PRIVATE(n), hw_words);
mpi_hal_write_to_mem_block(MPI_PARAM_Z, 0, Y->MBEDTLS_PRIVATE(p), Y->MBEDTLS_PRIVATE(n), hw_words);
mpi_hal_start_op(MPI_MULT);
Z->MBEDTLS_PRIVATE(s) = 1; // The sign of Z will be = M->s (but M->s is always 1)
MBEDTLS_MPI_CHK( mbedtls_mpi_grow(Z, hw_words) );
/* Read back the result */
mpi_hal_read_result_hw_op(Z->MBEDTLS_PRIVATE(p), Z->MBEDTLS_PRIVATE(n), hw_words);
/* from HAC 14.36 - 3. If Z >= M then Z = Z - M */
if (mbedtls_mpi_cmp_mpi(Z, M) >= 0) {
MBEDTLS_MPI_CHK(mbedtls_mpi_sub_mpi(Z, Z, M));
}
cleanup:
return ret;
}
#else
/* Z = (X ^ Y) mod M
*/
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)
{
size_t y_bits = mbedtls_mpi_bitlen(Y);
mpi_hal_set_mode(num_words - 1);
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_hal_write_to_mem_block(MPI_PARAM_X, 0, X->MBEDTLS_PRIVATE(p), X->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_write_to_mem_block(MPI_PARAM_Y, 0, Y->MBEDTLS_PRIVATE(p), Y->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_write_to_mem_block(MPI_PARAM_M, 0, M->MBEDTLS_PRIVATE(p), M->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_write_to_mem_block(MPI_PARAM_Z, 0, Rinv->MBEDTLS_PRIVATE(p), Rinv->MBEDTLS_PRIVATE(n), num_words);
mpi_hal_write_m_prime(Mprime);
/* Enable acceleration options */
mpi_hal_enable_constant_time(false);
mpi_hal_enable_search(true);
mpi_hal_set_search_position(y_bits - 1);
/* Execute first stage montgomery multiplication */
mpi_hal_start_op(MPI_MODEXP);
mpi_hal_enable_search(false);
}
#endif //ESP_MPI_USE_MONT_EXP

63
components/mbedtls/port/esp_bignum.c → components/mbedtls/port/bignum/esp_bignum.c
View File

@@ -6,7 +6,7 @@
*
* SPDX-License-Identifier: Apache-2.0
*
* SPDX-FileContributor: 2016-2022 Espressif Systems (Shanghai) CO LTD
* SPDX-FileContributor: 2016-2023 Espressif Systems (Shanghai) CO LTD
*/
#include <stdio.h>
#include <string.h>
@@ -24,17 +24,16 @@
#include "esp_pm.h"
#endif
#include "esp_private/periph_ctrl.h"
#include "freertos/FreeRTOS.h"
#include "freertos/semphr.h"
#include "soc/hwcrypto_periph.h"
#include "soc/periph_defs.h"
#include "soc/soc_caps.h"
#include "bignum_impl.h"
#include <mbedtls/bignum.h>
#include "mbedtls/bignum.h"
#include "hal/mpi_hal.h"
/* Some implementation notes:
*
@@ -65,7 +64,7 @@ static esp_pm_lock_handle_t s_pm_sleep_lock;
static IRAM_ATTR void esp_mpi_complete_isr(void *arg)
{
BaseType_t higher_woken;
esp_mpi_interrupt_clear();
mpi_hal_clear_interrupt();
xSemaphoreGiveFromISR(op_complete_sem, &higher_woken);
if (higher_woken) {
@@ -76,8 +75,8 @@ static IRAM_ATTR void esp_mpi_complete_isr(void *arg)
static esp_err_t esp_mpi_isr_initialise(void)
{
esp_mpi_interrupt_clear();
esp_mpi_interrupt_enable(true);
mpi_hal_clear_interrupt();
mpi_hal_interrupt_enable(true);
if (op_complete_sem == NULL) {
op_complete_sem = xSemaphoreCreateBinary();
@@ -120,7 +119,7 @@ static int esp_mpi_wait_intr(void)
esp_pm_lock_release(s_pm_sleep_lock);
#endif // CONFIG_PM_ENABLE
esp_mpi_interrupt_enable(false);
mpi_hal_interrupt_enable(false);
return 0;
}
@@ -208,8 +207,6 @@ cleanup:
/* Z = (X * Y) mod M
Not an mbedTLS function
@@ -226,7 +223,7 @@ int esp_mpi_mul_mpi_mod(mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_mpi
size_t y_words = bits_to_words(y_bits);
size_t m_words = bits_to_words(m_bits);
size_t z_words = bits_to_words(z_bits);
size_t hw_words = esp_mpi_hardware_words(MAX(x_words, MAX(y_words, m_words))); /* longest operand */
size_t hw_words = mpi_hal_calc_hardware_words(MAX(x_words, MAX(y_words, m_words))); /* longest operand */
mbedtls_mpi Rinv;
mbedtls_mpi_uint Mprime;
@@ -241,7 +238,8 @@ int esp_mpi_mul_mpi_mod(mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_mpi
MBEDTLS_MPI_CHK(mbedtls_mpi_grow(Z, z_words));
esp_mpi_read_result_hw_op(Z, z_words);
/* Read back the result */
mpi_hal_read_result_hw_op(Z->MBEDTLS_PRIVATE(p), Z->MBEDTLS_PRIVATE(n), z_words);
Z->MBEDTLS_PRIVATE(s) = X->MBEDTLS_PRIVATE(s) * Y->MBEDTLS_PRIVATE(s);
cleanup:
@@ -274,6 +272,7 @@ static size_t mbedtls_mpi_msb( const mbedtls_mpi *X )
return 0;
}
/*
* Montgomery exponentiation: Z = X ^ Y mod M (HAC 14.94)
*/
@@ -335,6 +334,7 @@ cleanup2:
#endif //USE_MONT_EXPONENATIATION
/*
* Z = X ^ Y mod M
*
@@ -358,7 +358,7 @@ static int esp_mpi_exp_mod( mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_
/* "all numbers must be the same length", so choose longest number
as cardinal length of operation...
*/
size_t num_words = esp_mpi_hardware_words(MAX(m_words, MAX(x_words, y_words)));
size_t num_words = mpi_hal_calc_hardware_words(MAX(m_words, MAX(x_words, y_words)));
if (num_words * 32 > SOC_RSA_MAX_BIT_LEN) {
return MBEDTLS_ERR_MPI_NOT_ACCEPTABLE;
@@ -420,7 +420,9 @@ static int esp_mpi_exp_mod( mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_
}
#endif //CONFIG_MBEDTLS_MPI_USE_INTERRUPT
esp_mpi_read_result_hw_op(Z, m_words);
/* Read back the result */
mpi_hal_read_result_hw_op(Z->MBEDTLS_PRIVATE(p), Z->MBEDTLS_PRIVATE(n), m_words);
esp_mpi_disable_hardware_hw_op();
#endif
@@ -479,7 +481,7 @@ int mbedtls_mpi_mul_mpi( mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_mpi
size_t x_words = bits_to_words(x_bits);
size_t y_words = bits_to_words(y_bits);
size_t z_words = bits_to_words(x_bits + y_bits);
size_t hw_words = esp_mpi_hardware_words(MAX(x_words, y_words)); // length of one operand in hardware
size_t hw_words = mpi_hal_calc_hardware_words(MAX(x_words, y_words)); // length of one operand in hardware
/* Short-circuit eval if either argument is 0 or 1.
@@ -534,7 +536,9 @@ int mbedtls_mpi_mul_mpi( mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_mpi
esp_mpi_enable_hardware_hw_op();
esp_mpi_mul_mpi_hw_op(X, Y, hw_words);
esp_mpi_read_result_hw_op(Z, z_words);
/* Read back the result */
mpi_hal_read_result_hw_op(Z->MBEDTLS_PRIVATE(p), Z->MBEDTLS_PRIVATE(n), z_words);
esp_mpi_disable_hardware_hw_op();
@@ -612,34 +616,19 @@ cleanup:
return ret;
}
/* Special-case of mbedtls_mpi_mult_mpi(), where we use hardware montgomery mod
multiplication to calculate an mbedtls_mpi_mult_mpi result where either
A or B are >2048 bits so can't use the standard multiplication method.
Result (number of words, based on A bits + B bits) must still be less than 4096 bits.
This case is simpler than the general case modulo multiply of
esp_mpi_mul_mpi_mod() because we can control the other arguments:
* Modulus is chosen with M=(2^num_bits - 1) (ie M=R-1), so output
* Mprime and Rinv are therefore predictable as follows:
isn't actually modulo anything.
Mprime 1
Rinv 1
(See RSA Accelerator section in Technical Reference for more about Mprime, Rinv)
*/
static int mpi_mult_mpi_failover_mod_mult( mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t z_words)
{
int ret;
size_t hw_words = esp_mpi_hardware_words(z_words);
size_t hw_words = mpi_hal_calc_hardware_words(z_words);
esp_mpi_enable_hardware_hw_op();
esp_mpi_mult_mpi_failover_mod_mult_hw_op(X, Y, hw_words );
MBEDTLS_MPI_CHK( mbedtls_mpi_grow(Z, hw_words) );
esp_mpi_read_result_hw_op(Z, hw_words);
/* Read back the result */
mpi_hal_read_result_hw_op(Z->MBEDTLS_PRIVATE(p), Z->MBEDTLS_PRIVATE(n), hw_words);
Z->MBEDTLS_PRIVATE(s) = X->MBEDTLS_PRIVATE(s) * Y->MBEDTLS_PRIVATE(s);
/*

296
components/mbedtls/port/esp32/bignum.c
View File

@@ -1,296 +0,0 @@
/*
* Multi-precision integer library
* ESP32 hardware accelerated parts based on mbedTLS implementation
*
* SPDX-FileCopyrightText: The Mbed TLS Contributors
*
* SPDX-License-Identifier: Apache-2.0
*
* SPDX-FileContributor: 2016-2022 Espressif Systems (Shanghai) CO LTD
*/
#include "soc/hwcrypto_periph.h"
#include "soc/dport_reg.h"
#include "esp_private/periph_ctrl.h"
#include <mbedtls/bignum.h>
#include "bignum_impl.h"
#include <sys/param.h>
#include <sys/lock.h>
static _lock_t mpi_lock;
/* Round up number of words to nearest
512 bit (16 word) block count.
*/
size_t esp_mpi_hardware_words(size_t words)
{
return (words + 0xF) & ~0xF;
}
void esp_mpi_enable_hardware_hw_op( void )
{
/* newlib locks lazy initialize on ESP-IDF */
_lock_acquire(&mpi_lock);
/* Enable RSA hardware */
periph_module_enable(PERIPH_RSA_MODULE);
DPORT_REG_CLR_BIT(DPORT_RSA_PD_CTRL_REG, DPORT_RSA_PD);
while (DPORT_REG_READ(RSA_CLEAN_REG) != 1)
{ }
// Note: from enabling RSA clock to here takes about 1.3us
}
void esp_mpi_disable_hardware_hw_op( void )
{
DPORT_REG_SET_BIT(DPORT_RSA_PD_CTRL_REG, DPORT_RSA_PD);
/* Disable RSA hardware */
periph_module_disable(PERIPH_RSA_MODULE);
_lock_release(&mpi_lock);
}
void esp_mpi_interrupt_enable( bool enable )
{
DPORT_REG_WRITE(RSA_INTERRUPT_REG, enable);
}
void esp_mpi_interrupt_clear( void )
{
DPORT_REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
}
/* Copy mbedTLS MPI bignum 'mpi' to hardware memory block at 'mem_base'.
If hw_words is higher than the number of words in the bignum then
these additional words will be zeroed in the memory buffer.
*/
/* Please see detailed note inside the function body below.
* Relevant: IDF-6029
https://github.com/espressif/esp-idf/issues/8710
https://github.com/espressif/esp-idf/issues/10403
*/
static inline void mpi_to_mem_block(uint32_t mem_base, const mbedtls_mpi *mpi, size_t hw_words)
{
uint32_t copy_words = MIN(hw_words, mpi->MBEDTLS_PRIVATE(n));
/* Copy MPI data to memory block registers */
for (uint32_t i = 0; i < copy_words; i++) {
DPORT_REG_WRITE(mem_base + i * 4, mpi->MBEDTLS_PRIVATE(p[i]));
}
/* Zero any remaining memory block data */
for (uint32_t i = copy_words; i < hw_words; i++) {
DPORT_REG_WRITE(mem_base + i * 4, 0);
}
#if _INTERNAL_DEBUG_PURPOSE
/*
* With Xtensa GCC 11.2.0 (from ESP-IDF v5.x), it was observed that above zero initialization
* loop gets optimized to `memset` call from the ROM library. This was causing an issue that
* specific write (store) operation to the MPI peripheral block was getting lost erroneously.
* Following data re-verify loop could catch it during runtime.
*
* As a workaround, we are using DPORT_WRITE_REG (volatile writes) wrappers to write to
* the MPI peripheral.
*
*/
//for (uint32_t i = copy_words; i < hw_words; i++) { assert(pbase[i] == 0); }
#endif
}
/* Read mbedTLS MPI bignum back from hardware memory block.
Reads num_words words from block.
Bignum 'x' should already be grown to at least num_words by caller (can be done while
calculation is in progress, to save some cycles)
*/
static inline void mem_block_to_mpi(mbedtls_mpi *x, uint32_t mem_base, size_t num_words)
{
assert(x->MBEDTLS_PRIVATE(n) >= num_words);
/* Copy data from memory block registers */
esp_dport_access_read_buffer(x->MBEDTLS_PRIVATE(p), mem_base, num_words);
/* Zero any remaining limbs in the bignum, if the buffer is bigger
than num_words */
for (size_t i = num_words; i < x->MBEDTLS_PRIVATE(n); i++) {
x->MBEDTLS_PRIVATE(p[i]) = 0;
}
}
/* Begin an RSA operation. op_reg specifies which 'START' register
to write to.
*/
static inline void start_op(uint32_t op_reg)
{
/* Clear interrupt status */
DPORT_REG_WRITE(RSA_INTERRUPT_REG, 1);
/* Note: above REG_WRITE includes a memw, so we know any writes
to the memory blocks are also complete. */
DPORT_REG_WRITE(op_reg, 1);
}
/* Wait for an RSA operation to complete.
*/
static inline void wait_op_complete(void)
{
while (DPORT_REG_READ(RSA_INTERRUPT_REG) != 1)
{ }
/* clear the interrupt */
DPORT_REG_WRITE(RSA_INTERRUPT_REG, 1);
}
/* Read result from last MPI operation */
void esp_mpi_read_result_hw_op(mbedtls_mpi *Z, size_t z_words)
{
wait_op_complete();
mem_block_to_mpi(Z, RSA_MEM_Z_BLOCK_BASE, z_words);
}
/* Z = (X * Y) mod M */
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 hw_words)
{
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, hw_words);
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, hw_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, hw_words);
DPORT_REG_WRITE(RSA_M_DASH_REG, (uint32_t)Mprime);
/* "mode" register loaded with number of 512-bit blocks, minus 1 */
DPORT_REG_WRITE(RSA_MULT_MODE_REG, (hw_words / 16) - 1);
/* Execute first stage montgomery multiplication */
start_op(RSA_MULT_START_REG);
wait_op_complete();
/* execute second stage */
/* Load Y to X input memory block, rerun */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, Y, hw_words);
start_op(RSA_MULT_START_REG);
}
/* Z = X * Y */
void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t hw_words)
{
/* Copy X (right-extended) & Y (left-extended) to memory block */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, hw_words);
mpi_to_mem_block(RSA_MEM_Z_BLOCK_BASE + hw_words * 4, Y, hw_words);
/* NB: as Y is left-extended, we don't zero the bottom words_mult words of Y block.
This is OK for now because zeroing is done by hardware when we do esp_mpi_acquire_hardware().
*/
DPORT_REG_WRITE(RSA_M_DASH_REG, 0);
/* "mode" register loaded with number of 512-bit blocks in result,
plus 7 (for range 9-12). (this is ((N~ / 32) - 1) + 8))
*/
DPORT_REG_WRITE(RSA_MULT_MODE_REG, ((hw_words * 2) / 16) + 7);
start_op(RSA_MULT_START_REG);
}
int esp_mont_hw_op(mbedtls_mpi *Z, const mbedtls_mpi *X, const mbedtls_mpi *Y, const mbedtls_mpi *M,
mbedtls_mpi_uint Mprime,
size_t hw_words,
bool again)
{
// Note Z may be the same pointer as X or Y
int ret = 0;
// montgomery mult prepare
if (again == false) {
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, hw_words);
DPORT_REG_WRITE(RSA_M_DASH_REG, Mprime);
DPORT_REG_WRITE(RSA_MULT_MODE_REG, hw_words / 16 - 1);
}
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, hw_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Y, hw_words);
start_op(RSA_MULT_START_REG);
Z->MBEDTLS_PRIVATE(s) = 1; // The sign of Z will be = M->s (but M->s is always 1)
MBEDTLS_MPI_CHK( mbedtls_mpi_grow(Z, hw_words) );
wait_op_complete();
/* Read back the result */
mem_block_to_mpi(Z, RSA_MEM_Z_BLOCK_BASE, hw_words);
/* from HAC 14.36 - 3. If Z >= M then Z = Z - M */
if (mbedtls_mpi_cmp_mpi(Z, M) >= 0) {
MBEDTLS_MPI_CHK(mbedtls_mpi_sub_mpi(Z, Z, M));
}
cleanup:
return ret;
}
/* Special-case of mbedtls_mpi_mult_mpi(), where we use hardware montgomery mod
multiplication to calculate an mbedtls_mpi_mult_mpi result where either
A or B are >2048 bits so can't use the standard multiplication method.
Result (z_words, based on A bits + B bits) must still be less than 4096 bits.
This case is simpler than the general case modulo multiply of
esp_mpi_mul_mpi_mod() because we can control the other arguments:
* Modulus is chosen with M=(2^num_bits - 1) (ie M=R-1), so output
isn't actually modulo anything.
* Mprime and Rinv are therefore predictable as follows:
Mprime = 1
Rinv = 1
(See RSA Accelerator section in Technical Reference for more about Mprime, Rinv)
*/
void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
size_t hw_words = num_words;
/* M = 2^num_words - 1, so block is entirely FF */
for (size_t i = 0; i < hw_words; i++) {
DPORT_REG_WRITE(RSA_MEM_M_BLOCK_BASE + i * 4, UINT32_MAX);
}
/* Mprime = 1 */
DPORT_REG_WRITE(RSA_M_DASH_REG, 1);
/* "mode" register loaded with number of 512-bit blocks, minus 1 */
DPORT_REG_WRITE(RSA_MULT_MODE_REG, (hw_words / 16) - 1);
/* Load X */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, hw_words);
/* Rinv = 1, write first word */
DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE, 1);
/* Zero out rest of the Rinv words */
for (size_t i = 1; i < hw_words; i++) {
DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE + i * 4, 0);
}
start_op(RSA_MULT_START_REG);
wait_op_complete();
/* finish the modular multiplication */
/* Load Y to X input memory block, rerun */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, Y, hw_words);
start_op(RSA_MULT_START_REG);
}

229
components/mbedtls/port/esp32c3/bignum.c
View File

@@ -1,229 +0,0 @@
/*
* Multi-precision integer library
* ESP32 C3 hardware accelerated parts based on mbedTLS implementation
*
* SPDX-FileCopyrightText: The Mbed TLS Contributors
*
* SPDX-License-Identifier: Apache-2.0
*
* SPDX-FileContributor: 2016-2022 Espressif Systems (Shanghai) CO LTD
*/
#include <string.h>
#include <sys/param.h>
#include "soc/hwcrypto_periph.h"
#include "esp_private/periph_ctrl.h"
#include "mbedtls/bignum.h"
#include "bignum_impl.h"
#include "soc/system_reg.h"
#include "soc/periph_defs.h"
#include "esp_crypto_lock.h"
size_t esp_mpi_hardware_words(size_t words)
{
return words;
}
void esp_mpi_enable_hardware_hw_op( void )
{
esp_crypto_mpi_lock_acquire();
/* Enable RSA hardware */
periph_module_enable(PERIPH_RSA_MODULE);
REG_CLR_BIT(SYSTEM_RSA_PD_CTRL_REG, SYSTEM_RSA_MEM_PD);
while (REG_READ(RSA_QUERY_CLEAN_REG) != 1) {
}
// Note: from enabling RSA clock to here takes about 1.3us
REG_WRITE(RSA_INTERRUPT_REG, 0);
}
void esp_mpi_disable_hardware_hw_op( void )
{
REG_SET_BIT(SYSTEM_RSA_PD_CTRL_REG, SYSTEM_RSA_MEM_PD);
/* Disable RSA hardware */
periph_module_disable(PERIPH_RSA_MODULE);
esp_crypto_mpi_lock_release();
}
void esp_mpi_interrupt_enable( bool enable )
{
REG_WRITE(RSA_INTERRUPT_REG, enable);
}
void esp_mpi_interrupt_clear( void )
{
REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
}
/* Copy mbedTLS MPI bignum 'mpi' to hardware memory block at 'mem_base'.
If num_words is higher than the number of words in the bignum then
these additional words will be zeroed in the memory buffer.
*/
static inline void mpi_to_mem_block(uint32_t mem_base, const mbedtls_mpi *mpi, size_t num_words)
{
uint32_t *pbase = (uint32_t *)mem_base;
uint32_t copy_words = MIN(num_words, mpi->MBEDTLS_PRIVATE(n));
/* Copy MPI data to memory block registers */
for (int i = 0; i < copy_words; i++) {
pbase[i] = mpi->MBEDTLS_PRIVATE(p)[i];
}
/* Zero any remaining memory block data */
for (int i = copy_words; i < num_words; i++) {
pbase[i] = 0;
}
}
/* Read mbedTLS MPI bignum back from hardware memory block.
Reads num_words words from block.
*/
static inline void mem_block_to_mpi(mbedtls_mpi *x, uint32_t mem_base, int num_words)
{
/* Copy data from memory block registers */
const size_t REG_WIDTH = sizeof(uint32_t);
for (size_t i = 0; i < num_words; i++) {
x->MBEDTLS_PRIVATE(p)[i] = REG_READ(mem_base + (i * REG_WIDTH));
}
/* Zero any remaining limbs in the bignum, if the buffer is bigger
than num_words */
for (size_t i = num_words; i < x->MBEDTLS_PRIVATE(n); i++) {
x->MBEDTLS_PRIVATE(p)[i] = 0;
}
}
/* Begin an RSA operation. op_reg specifies which 'START' register
to write to.
*/
static inline void start_op(uint32_t op_reg)
{
/* Clear interrupt status */
REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
/* Note: above REG_WRITE includes a memw, so we know any writes
to the memory blocks are also complete. */
REG_WRITE(op_reg, 1);
}
/* Wait for an RSA operation to complete.
*/
static inline void wait_op_complete(void)
{
while (REG_READ(RSA_QUERY_INTERRUPT_REG) != 1)
{ }
/* clear the interrupt */
REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
}
/* Read result from last MPI operation */
void esp_mpi_read_result_hw_op(mbedtls_mpi *Z, size_t z_words)
{
wait_op_complete();
mem_block_to_mpi(Z, RSA_MEM_Z_BLOCK_BASE, z_words);
}
/* Z = (X * Y) mod M
Not an mbedTLS function
*/
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)
{
REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
REG_WRITE(RSA_M_DASH_REG, Mprime);
start_op(RSA_MOD_MULT_START_REG);
}
/* Z = (X ^ Y) mod M
*/
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)
{
size_t y_bits = mbedtls_mpi_bitlen(Y);
REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
REG_WRITE(RSA_M_DASH_REG, Mprime);
/* Enable acceleration options */
REG_WRITE(RSA_CONSTANT_TIME_REG, 0);
REG_WRITE(RSA_SEARCH_ENABLE_REG, 1);
REG_WRITE(RSA_SEARCH_POS_REG, y_bits - 1);
/* Execute first stage montgomery multiplication */
start_op(RSA_MODEXP_START_REG);
REG_WRITE(RSA_SEARCH_ENABLE_REG, 0);
}
/* Z = X * Y */
void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* Copy X (right-extended) & Y (left-extended) to memory block */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Z_BLOCK_BASE + num_words * 4, Y, num_words);
/* NB: as Y is left-extended, we don't zero the bottom words_mult words of Y block.
This is OK for now because zeroing is done by hardware when we do esp_mpi_acquire_hardware().
*/
REG_WRITE(RSA_LENGTH_REG, (num_words * 2 - 1));
start_op(RSA_MULT_START_REG);
}
/**
* @brief Special-case of (X * Y), where we use hardware montgomery mod
multiplication to calculate result where either A or B are >2048 bits so
can't use the standard multiplication method.
*
*/
void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* M = 2^num_words - 1, so block is entirely FF */
for (int i = 0; i < num_words; i++) {
REG_WRITE(RSA_MEM_M_BLOCK_BASE + i * 4, UINT32_MAX);
}
/* Mprime = 1 */
REG_WRITE(RSA_M_DASH_REG, 1);
REG_WRITE(RSA_LENGTH_REG, num_words - 1);
/* Load X & Y */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
/* Rinv = 1, write first word */
REG_WRITE(RSA_MEM_RB_BLOCK_BASE, 1);
/* Zero out rest of the Rinv words */
for (int i = 1; i < num_words; i++) {
REG_WRITE(RSA_MEM_RB_BLOCK_BASE + i * 4, 0);
}
start_op(RSA_MOD_MULT_START_REG);
}

230
components/mbedtls/port/esp32c6/bignum.c
View File

@@ -1,230 +0,0 @@
/*
* Multi-precision integer library
* ESP32 C6 hardware accelerated parts based on mbedTLS implementation
*
* SPDX-FileCopyrightText: The Mbed TLS Contributors
*
* SPDX-License-Identifier: Apache-2.0
*
* SPDX-FileContributor: 2023 Espressif Systems (Shanghai) CO LTD
*/
#include <string.h>
#include <sys/param.h>
#include "soc/hwcrypto_periph.h"
#include "esp_private/periph_ctrl.h"
#include "mbedtls/bignum.h"
#include "bignum_impl.h"
#include "soc/pcr_reg.h"
#include "soc/periph_defs.h"
#include "soc/system_reg.h"
#include "esp_crypto_lock.h"
size_t esp_mpi_hardware_words(size_t words)
{
return words;
}
void esp_mpi_enable_hardware_hw_op( void )
{
esp_crypto_mpi_lock_acquire();
/* Enable RSA hardware */
periph_module_enable(PERIPH_RSA_MODULE);
REG_CLR_BIT(PCR_RSA_PD_CTRL_REG, PCR_RSA_MEM_PD);
while (REG_READ(RSA_QUERY_CLEAN_REG) != 1) {
}
// Note: from enabling RSA clock to here takes about 1.3us
REG_WRITE(RSA_INT_ENA_REG, 0);
}
void esp_mpi_disable_hardware_hw_op( void )
{
REG_SET_BIT(PCR_RSA_PD_CTRL_REG, PCR_RSA_MEM_PD);
/* Disable RSA hardware */
periph_module_disable(PERIPH_RSA_MODULE);
esp_crypto_mpi_lock_release();
}
void esp_mpi_interrupt_enable( bool enable )
{
REG_WRITE(RSA_INT_ENA_REG, enable);
}
void esp_mpi_interrupt_clear( void )
{
REG_WRITE(RSA_INT_CLR_REG, 1);
}
/* Copy mbedTLS MPI bignum 'mpi' to hardware memory block at 'mem_base'.
If num_words is higher than the number of words in the bignum then
these additional words will be zeroed in the memory buffer.
*/
static inline void mpi_to_mem_block(uint32_t mem_base, const mbedtls_mpi *mpi, size_t num_words)
{
uint32_t *pbase = (uint32_t *)mem_base;
uint32_t copy_words = MIN(num_words, mpi->MBEDTLS_PRIVATE(n));
/* Copy MPI data to memory block registers */
for (int i = 0; i < copy_words; i++) {
pbase[i] = mpi->MBEDTLS_PRIVATE(p)[i];
}
/* Zero any remaining memory block data */
for (int i = copy_words; i < num_words; i++) {
pbase[i] = 0;
}
}
/* Read mbedTLS MPI bignum back from hardware memory block.
Reads num_words words from block.
*/
static inline void mem_block_to_mpi(mbedtls_mpi *x, uint32_t mem_base, int num_words)
{
/* Copy data from memory block registers */
const size_t REG_WIDTH = sizeof(uint32_t);
for (size_t i = 0; i < num_words; i++) {
x->MBEDTLS_PRIVATE(p)[i] = REG_READ(mem_base + (i * REG_WIDTH));
}
/* Zero any remaining limbs in the bignum, if the buffer is bigger
than num_words */
for (size_t i = num_words; i < x->MBEDTLS_PRIVATE(n); i++) {
x->MBEDTLS_PRIVATE(p)[i] = 0;
}
}
/* Begin an RSA operation. op_reg specifies which 'START' register
to write to.
*/
static inline void start_op(uint32_t op_reg)
{
/* Clear interrupt status */
REG_WRITE(RSA_INT_CLR_REG, 1);
/* Note: above REG_WRITE includes a memw, so we know any writes
to the memory blocks are also complete. */
REG_WRITE(op_reg, 1);
}
/* Wait for an RSA operation to complete.
*/
static inline void wait_op_complete(void)
{
while (REG_READ(RSA_QUERY_IDLE_REG) != 1)
{ }
/* clear the interrupt */
REG_WRITE(RSA_INT_CLR_REG, 1);
}
/* Read result from last MPI operation */
void esp_mpi_read_result_hw_op(mbedtls_mpi *Z, size_t z_words)
{
wait_op_complete();
mem_block_to_mpi(Z, RSA_Z_MEM, z_words);
}
/* Z = (X * Y) mod M
Not an mbedTLS function
*/
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)
{
REG_WRITE(RSA_MODE_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_X_MEM, X, num_words);
mpi_to_mem_block(RSA_Y_MEM, Y, num_words);
mpi_to_mem_block(RSA_M_MEM, M, num_words);
mpi_to_mem_block(RSA_Z_MEM, Rinv, num_words);
REG_WRITE(RSA_M_PRIME_REG, Mprime);
start_op(RSA_SET_START_MODMULT_REG);
}
/* Z = (X ^ Y) mod M
*/
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)
{
size_t y_bits = mbedtls_mpi_bitlen(Y);
REG_WRITE(RSA_MODE_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_X_MEM, X, num_words);
mpi_to_mem_block(RSA_Y_MEM, Y, num_words);
mpi_to_mem_block(RSA_M_MEM, M, num_words);
mpi_to_mem_block(RSA_Z_MEM, Rinv, num_words);
REG_WRITE(RSA_M_PRIME_REG, Mprime);
/* Enable acceleration options */
REG_WRITE(RSA_CONSTANT_TIME_REG, 0);
REG_WRITE(RSA_SEARCH_ENABLE_REG, 1);
REG_WRITE(RSA_SEARCH_POS_REG, y_bits - 1);
/* Execute first stage montgomery multiplication */
start_op(RSA_SET_START_MODEXP_REG);
REG_WRITE(RSA_SEARCH_ENABLE_REG, 0);
}
/* Z = X * Y */
void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* Copy X (right-extended) & Y (left-extended) to memory block */
mpi_to_mem_block(RSA_X_MEM, X, num_words);
mpi_to_mem_block(RSA_Z_MEM + num_words * 4, Y, num_words);
/* NB: as Y is left-exte, we don't zero the bottom words_mult words of Y block.
This is OK for now bec zeroing is done by hardware when we do esp_mpi_acquire_hardware().
*/
REG_WRITE(RSA_MODE_REG, (num_words * 2 - 1));
start_op(RSA_SET_START_MULT_REG);
}
/**
* @brief Special-case of (X * Y), where we use hardware montgomery mod
multiplication to calculate result where either A or B are >2048 bits so
can't use the standard multiplication method.
*
*/
void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* M = 2^num_words - 1, so block is entirely FF */
for (int i = 0; i < num_words; i++) {
REG_WRITE(RSA_M_MEM + i * 4, UINT32_MAX);
}
/* Mprime = 1 */
REG_WRITE(RSA_M_PRIME_REG, 1);
REG_WRITE(RSA_MODE_REG, num_words - 1);
/* Load X & Y */
mpi_to_mem_block(RSA_X_MEM, X, num_words);
mpi_to_mem_block(RSA_Y_MEM, Y, num_words);
/* Rinv = 1, write first word */
REG_WRITE(RSA_Z_MEM, 1);
/* Zero out rest of the Rinv words */
for (int i = 1; i < num_words; i++) {
REG_WRITE(RSA_Z_MEM + i * 4, 0);
}
start_op(RSA_SET_START_MODMULT_REG);
}

230
components/mbedtls/port/esp32h2/bignum.c
View File

@@ -1,230 +0,0 @@
/*
* Multi-precision integer library
* ESP32 H2 hardware accelerated parts based on mbedTLS implementation
*
* SPDX-FileCopyrightText: The Mbed TLS Contributors
*
* SPDX-License-Identifier: Apache-2.0
*
* SPDX-FileContributor: 2023 Espressif Systems (Shanghai) CO LTD
*/
#include <string.h>
#include <sys/param.h>
#include "soc/hwcrypto_periph.h"
#include "esp_private/periph_ctrl.h"
#include "mbedtls/bignum.h"
#include "bignum_impl.h"
#include "soc/pcr_reg.h"
#include "soc/periph_defs.h"
#include "soc/system_reg.h"
#include "esp_crypto_lock.h"
size_t esp_mpi_hardware_words(size_t words)
{
return words;
}
void esp_mpi_enable_hardware_hw_op( void )
{
esp_crypto_mpi_lock_acquire();
/* Enable RSA hardware */
periph_module_enable(PERIPH_RSA_MODULE);
REG_CLR_BIT(PCR_RSA_PD_CTRL_REG, PCR_RSA_MEM_PD);
while (REG_READ(RSA_QUERY_CLEAN_REG) != 1) {
}
// Note: from enabling RSA clock to here takes about 1.3us
REG_WRITE(RSA_INT_ENA_REG, 0);
}
void esp_mpi_disable_hardware_hw_op( void )
{
REG_SET_BIT(PCR_RSA_PD_CTRL_REG, PCR_RSA_MEM_PD);
/* Disable RSA hardware */
periph_module_disable(PERIPH_RSA_MODULE);
esp_crypto_mpi_lock_release();
}
void esp_mpi_interrupt_enable( bool enable )
{
REG_WRITE(RSA_INT_ENA_REG, enable);
}
void esp_mpi_interrupt_clear( void )
{
REG_WRITE(RSA_INT_CLR_REG, 1);
}
/* Copy mbedTLS MPI bignum 'mpi' to hardware memory block at 'mem_base'.
If num_words is higher than the number of words in the bignum then
these additional words will be zeroed in the memory buffer.
*/
static inline void mpi_to_mem_block(uint32_t mem_base, const mbedtls_mpi *mpi, size_t num_words)
{
uint32_t *pbase = (uint32_t *)mem_base;
uint32_t copy_words = MIN(num_words, mpi->MBEDTLS_PRIVATE(n));
/* Copy MPI data to memory block registers */
for (int i = 0; i < copy_words; i++) {
pbase[i] = mpi->MBEDTLS_PRIVATE(p)[i];
}
/* Zero any remaining memory block data */
for (int i = copy_words; i < num_words; i++) {
pbase[i] = 0;
}
}
/* Read mbedTLS MPI bignum back from hardware memory block.
Reads num_words words from block.
*/
static inline void mem_block_to_mpi(mbedtls_mpi *x, uint32_t mem_base, int num_words)
{
/* Copy data from memory block registers */
const size_t REG_WIDTH = sizeof(uint32_t);
for (size_t i = 0; i < num_words; i++) {
x->MBEDTLS_PRIVATE(p)[i] = REG_READ(mem_base + (i * REG_WIDTH));
}
/* Zero any remaining limbs in the bignum, if the buffer is bigger
than num_words */
for (size_t i = num_words; i < x->MBEDTLS_PRIVATE(n); i++) {
x->MBEDTLS_PRIVATE(p)[i] = 0;
}
}
/* Begin an RSA operation. op_reg specifies which 'START' register
to write to.
*/
static inline void start_op(uint32_t op_reg)
{
/* Clear interrupt status */
REG_WRITE(RSA_INT_CLR_REG, 1);
/* Note: above REG_WRITE includes a memw, so we know any writes
to the memory blocks are also complete. */
REG_WRITE(op_reg, 1);
}
/* Wait for an RSA operation to complete.
*/
static inline void wait_op_complete(void)
{
while (REG_READ(RSA_QUERY_IDLE_REG) != 1)
{ }
/* clear the interrupt */
REG_WRITE(RSA_INT_CLR_REG, 1);
}
/* Read result from last MPI operation */
void esp_mpi_read_result_hw_op(mbedtls_mpi *Z, size_t z_words)
{
wait_op_complete();
mem_block_to_mpi(Z, RSA_Z_MEM_REG, z_words);
}
/* Z = (X * Y) mod M
Not an mbedTLS function
*/
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)
{
REG_WRITE(RSA_MODE_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_X_MEM_REG, X, num_words);
mpi_to_mem_block(RSA_Y_MEM_REG, Y, num_words);
mpi_to_mem_block(RSA_M_MEM_REG, M, num_words);
mpi_to_mem_block(RSA_Z_MEM_REG, Rinv, num_words);
REG_WRITE(RSA_M_PRIME_REG, Mprime);
start_op(RSA_SET_START_MODMULT_REG);
}
/* Z = (X ^ Y) mod M
*/
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)
{
size_t y_bits = mbedtls_mpi_bitlen(Y);
REG_WRITE(RSA_MODE_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_X_MEM_REG, X, num_words);
mpi_to_mem_block(RSA_Y_MEM_REG, Y, num_words);
mpi_to_mem_block(RSA_M_MEM_REG, M, num_words);
mpi_to_mem_block(RSA_Z_MEM_REG, Rinv, num_words);
REG_WRITE(RSA_M_PRIME_REG, Mprime);
/* Enable acceleration options */
REG_WRITE(RSA_CONSTANT_TIME_REG, 0);
REG_WRITE(RSA_SEARCH_ENABLE_REG, 1);
REG_WRITE(RSA_SEARCH_POS_REG, y_bits - 1);
/* Execute first stage montgomery multiplication */
start_op(RSA_SET_START_MODEXP_REG);
REG_WRITE(RSA_SEARCH_ENABLE_REG, 0);
}
/* Z = X * Y */
void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* Copy X (right-extended) & Y (left-extended) to memory block */
mpi_to_mem_block(RSA_X_MEM_REG, X, num_words);
mpi_to_mem_block(RSA_Z_MEM_REG + num_words * 4, Y, num_words);
/* NB: as Y is left-exte, we don't zero the bottom words_mult words of Y block.
This is OK for now bec zeroing is done by hardware when we do esp_mpi_acquire_hardware().
*/
REG_WRITE(RSA_MODE_REG, (num_words * 2 - 1));
start_op(RSA_SET_START_MULT_REG);
}
/**
* @brief Special-case of (X * Y), where we use hardware montgomery mod
multiplication to calculate result where either A or B are >2048 bits so
can't use the standard multiplication method.
*
*/
void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* M = 2^num_words - 1, so block is entirely FF */
for (int i = 0; i < num_words; i++) {
REG_WRITE(RSA_M_MEM_REG + i * 4, UINT32_MAX);
}
/* Mprime = 1 */
REG_WRITE(RSA_M_PRIME_REG, 1);
REG_WRITE(RSA_MODE_REG, num_words - 1);
/* Load X & Y */
mpi_to_mem_block(RSA_X_MEM_REG, X, num_words);
mpi_to_mem_block(RSA_Y_MEM_REG, Y, num_words);
/* Rinv = 1, write first word */
REG_WRITE(RSA_Z_MEM_REG, 1);
/* Zero out rest of the Rinv words */
for (int i = 1; i < num_words; i++) {
REG_WRITE(RSA_Z_MEM_REG + i * 4, 0);
}
start_op(RSA_SET_START_MODMULT_REG);
}

224
components/mbedtls/port/esp32s2/bignum.c
View File

@@ -1,224 +0,0 @@
/*
* Multi-precision integer library
* ESP32 S2 hardware accelerated parts based on mbedTLS implementation
*
* SPDX-FileCopyrightText: The Mbed TLS Contributors
*
* SPDX-License-Identifier: Apache-2.0
*
* SPDX-FileContributor: 2016-2022 Espressif Systems (Shanghai) CO LTD
*/
#include "soc/hwcrypto_periph.h"
#include "esp_private/periph_ctrl.h"
#include <mbedtls/bignum.h>
#include "bignum_impl.h"
#include "soc/dport_reg.h"
#include "soc/periph_defs.h"
#include <sys/param.h>
#include "esp_crypto_lock.h"
size_t esp_mpi_hardware_words(size_t words)
{
return words;
}
void esp_mpi_enable_hardware_hw_op( void )
{
esp_crypto_mpi_lock_acquire();
/* Enable RSA hardware */
periph_module_enable(PERIPH_RSA_MODULE);
DPORT_REG_CLR_BIT(DPORT_RSA_PD_CTRL_REG, DPORT_RSA_MEM_PD);
while (DPORT_REG_READ(RSA_QUERY_CLEAN_REG) != 1) {
}
// Note: from enabling RSA clock to here takes about 1.3us
REG_WRITE(RSA_INTERRUPT_REG, 0);
}
void esp_mpi_disable_hardware_hw_op( void )
{
DPORT_REG_SET_BIT(DPORT_RSA_PD_CTRL_REG, DPORT_RSA_PD);
/* Disable RSA hardware */
periph_module_disable(PERIPH_RSA_MODULE);
esp_crypto_mpi_lock_release();
}
void esp_mpi_interrupt_enable( bool enable )
{
REG_WRITE(RSA_INTERRUPT_REG, enable);
}
void esp_mpi_interrupt_clear( void )
{
REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
}
/* Copy mbedTLS MPI bignum 'mpi' to hardware memory block at 'mem_base'.
If num_words is higher than the number of words in the bignum then
these additional words will be zeroed in the memory buffer.
*/
static inline void mpi_to_mem_block(uint32_t mem_base, const mbedtls_mpi *mpi, size_t num_words)
{
uint32_t *pbase = (uint32_t *)mem_base;
uint32_t copy_words = MIN(num_words, mpi->MBEDTLS_PRIVATE(n));
/* Copy MPI data to memory block registers */
for (uint32_t i = 0; i < copy_words; i++) {
pbase[i] = mpi->MBEDTLS_PRIVATE(p)[i];
}
/* Zero any remaining memory block data */
for (uint32_t i = copy_words; i < num_words; i++) {
pbase[i] = 0;
}
}
/* Read mbedTLS MPI bignum back from hardware memory block.
Reads num_words words from block.
*/
static inline void mem_block_to_mpi(mbedtls_mpi *x, uint32_t mem_base, int num_words)
{
/* Copy data from memory block registers */
esp_dport_access_read_buffer(x->MBEDTLS_PRIVATE(p), mem_base, num_words);
/* Zero any remaining limbs in the bignum, if the buffer is bigger
than num_words */
for (size_t i = num_words; i < x->MBEDTLS_PRIVATE(n); i++) {
x->MBEDTLS_PRIVATE(p)[i] = 0;
}
}
/* Begin an RSA operation. op_reg specifies which 'START' register
to write to.
*/
static inline void start_op(uint32_t op_reg)
{
/* Clear interrupt status */
DPORT_REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
/* Note: above REG_WRITE includes a memw, so we know any writes
to the memory blocks are also complete. */
DPORT_REG_WRITE(op_reg, 1);
}
/* Wait for an RSA operation to complete.
*/
static inline void wait_op_complete(void)
{
while (DPORT_REG_READ(RSA_QUERY_INTERRUPT_REG) != 1)
{ }
/* clear the interrupt */
DPORT_REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
}
/* Read result from last MPI operation */
void esp_mpi_read_result_hw_op(mbedtls_mpi *Z, size_t z_words)
{
wait_op_complete();
mem_block_to_mpi(Z, RSA_MEM_Z_BLOCK_BASE, z_words);
}
/* Z = (X * Y) mod M
Not an mbedTLS function
*/
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)
{
DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
DPORT_REG_WRITE(RSA_M_DASH_REG, Mprime);
start_op(RSA_MOD_MULT_START_REG);
}
/* Z = (X ^ Y) mod M
*/
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)
{
size_t y_bits = mbedtls_mpi_bitlen(Y);
DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
DPORT_REG_WRITE(RSA_M_DASH_REG, Mprime);
/* Enable acceleration options */
DPORT_REG_WRITE(RSA_CONSTANT_TIME_REG, 0);
DPORT_REG_WRITE(RSA_SEARCH_OPEN_REG, 1);
DPORT_REG_WRITE(RSA_SEARCH_POS_REG, y_bits - 1);
/* Execute first stage montgomery multiplication */
start_op(RSA_MODEXP_START_REG);
DPORT_REG_WRITE(RSA_SEARCH_OPEN_REG, 0);
}
/* Z = X * Y */
void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* Copy X (right-extended) & Y (left-extended) to memory block */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Z_BLOCK_BASE + num_words * 4, Y, num_words);
/* NB: as Y is left-extended, we don't zero the bottom words_mult words of Y block.
This is OK for now because zeroing is done by hardware when we do esp_mpi_acquire_hardware().
*/
DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words * 2 - 1));
start_op(RSA_MULT_START_REG);
}
/**
* @brief Special-case of (X * Y), where we use hardware montgomery mod
multiplication to calculate result where either A or B are >2048 bits so
can't use the standard multiplication method.
*
*/
void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* M = 2^num_words - 1, so block is entirely FF */
for (size_t i = 0; i < num_words; i++) {
DPORT_REG_WRITE(RSA_MEM_M_BLOCK_BASE + i * 4, UINT32_MAX);
}
/* Mprime = 1 */
DPORT_REG_WRITE(RSA_M_DASH_REG, 1);
DPORT_REG_WRITE(RSA_LENGTH_REG, num_words - 1);
/* Load X & Y */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
/* Rinv = 1, write first word */
DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE, 1);
/* Zero out rest of the Rinv words */
for (size_t i = 1; i < num_words; i++) {
DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE + i * 4, 0);
}
start_op(RSA_MOD_MULT_START_REG);
}

226
components/mbedtls/port/esp32s3/bignum.c
View File

@@ -1,226 +0,0 @@
/*
* Multi-precision integer library
* ESP32 S3 hardware accelerated parts based on mbedTLS implementation
*
* SPDX-FileCopyrightText: The Mbed TLS Contributors
*
* SPDX-License-Identifier: Apache-2.0
*
* SPDX-FileContributor: 2016-2022 Espressif Systems (Shanghai) CO LTD
*/
#include "soc/hwcrypto_periph.h"
#include "esp_private/periph_ctrl.h"
#include <mbedtls/bignum.h>
#include "bignum_impl.h"
#include "soc/dport_reg.h"
#include "soc/system_reg.h"
#include "soc/periph_defs.h"
#include <sys/param.h>
#include "esp_crypto_lock.h"
size_t esp_mpi_hardware_words(size_t words)
{
return words;
}
void esp_mpi_enable_hardware_hw_op( void )
{
esp_crypto_mpi_lock_acquire();
/* Enable RSA hardware */
periph_module_enable(PERIPH_RSA_MODULE);
REG_CLR_BIT(SYSTEM_RSA_PD_CTRL_REG, SYSTEM_RSA_MEM_PD);
while (DPORT_REG_READ(RSA_QUERY_CLEAN_REG) != 1) {
}
// Note: from enabling RSA clock to here takes about 1.3us
REG_WRITE(RSA_INTERRUPT_REG, 0);
}
void esp_mpi_disable_hardware_hw_op( void )
{
REG_SET_BIT(SYSTEM_RSA_PD_CTRL_REG, SYSTEM_RSA_MEM_PD);
/* Disable RSA hardware */
periph_module_disable(PERIPH_RSA_MODULE);
esp_crypto_mpi_lock_release();
}
void esp_mpi_interrupt_enable( bool enable )
{
REG_WRITE(RSA_INTERRUPT_REG, enable);
}
void esp_mpi_interrupt_clear( void )
{
REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
}
/* Copy mbedTLS MPI bignum 'mpi' to hardware memory block at 'mem_base'.
If num_words is higher than the number of words in the bignum then
these additional words will be zeroed in the memory buffer.
*/
static inline void mpi_to_mem_block(uint32_t mem_base, const mbedtls_mpi *mpi, size_t num_words)
{
uint32_t *pbase = (uint32_t *)mem_base;
uint32_t copy_words = MIN(num_words, mpi->MBEDTLS_PRIVATE(n));
/* Copy MPI data to memory block registers */
for (uint32_t i = 0; i < copy_words; i++) {
pbase[i] = mpi->MBEDTLS_PRIVATE(p)[i];
}
/* Zero any remaining memory block data */
for (uint32_t i = copy_words; i < num_words; i++) {
pbase[i] = 0;
}
}
/* Read mbedTLS MPI bignum back from hardware memory block.
Reads num_words words from block.
*/
static inline void mem_block_to_mpi(mbedtls_mpi *x, uint32_t mem_base, int num_words)
{
/* Copy data from memory block registers */
esp_dport_access_read_buffer(x->MBEDTLS_PRIVATE(p), mem_base, num_words);
/* Zero any remaining limbs in the bignum, if the buffer is bigger
than num_words */
for (size_t i = num_words; i < x->MBEDTLS_PRIVATE(n); i++) {
x->MBEDTLS_PRIVATE(p)[i] = 0;
}
}
/* Begin an RSA operation. op_reg specifies which 'START' register
to write to.
*/
static inline void start_op(uint32_t op_reg)
{
/* Clear interrupt status */
DPORT_REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
/* Note: above REG_WRITE includes a memw, so we know any writes
to the memory blocks are also complete. */
DPORT_REG_WRITE(op_reg, 1);
}
/* Wait for an RSA operation to complete.
*/
static inline void wait_op_complete(void)
{
while (DPORT_REG_READ(RSA_QUERY_INTERRUPT_REG) != 1)
{ }
/* clear the interrupt */
DPORT_REG_WRITE(RSA_CLEAR_INTERRUPT_REG, 1);
}
/* Read result from last MPI operation */
void esp_mpi_read_result_hw_op(mbedtls_mpi *Z, size_t z_words)
{
wait_op_complete();
mem_block_to_mpi(Z, RSA_MEM_Z_BLOCK_BASE, z_words);
}
/* Z = (X * Y) mod M
Not an mbedTLS function
*/
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)
{
DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
DPORT_REG_WRITE(RSA_M_DASH_REG, Mprime);
start_op(RSA_MOD_MULT_START_REG);
}
/* Z = (X ^ Y) mod M
*/
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)
{
size_t y_bits = mbedtls_mpi_bitlen(Y);
DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words - 1));
/* Load M, X, Rinv, Mprime (Mprime is mod 2^32) */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
mpi_to_mem_block(RSA_MEM_M_BLOCK_BASE, M, num_words);
mpi_to_mem_block(RSA_MEM_RB_BLOCK_BASE, Rinv, num_words);
DPORT_REG_WRITE(RSA_M_DASH_REG, Mprime);
/* Enable acceleration options */
DPORT_REG_WRITE(RSA_CONSTANT_TIME_REG, 0);
DPORT_REG_WRITE(RSA_SEARCH_OPEN_REG, 1);
DPORT_REG_WRITE(RSA_SEARCH_POS_REG, y_bits - 1);
/* Execute first stage montgomery multiplication */
start_op(RSA_MODEXP_START_REG);
DPORT_REG_WRITE(RSA_SEARCH_OPEN_REG, 0);
}
/* Z = X * Y */
void esp_mpi_mul_mpi_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* Copy X (right-extended) & Y (left-extended) to memory block */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Z_BLOCK_BASE + num_words * 4, Y, num_words);
/* NB: as Y is left-extended, we don't zero the bottom words_mult words of Y block.
This is OK for now because zeroing is done by hardware when we do esp_mpi_acquire_hardware().
*/
DPORT_REG_WRITE(RSA_LENGTH_REG, (num_words * 2 - 1));
start_op(RSA_MULT_START_REG);
}
/**
* @brief Special-case of (X * Y), where we use hardware montgomery mod
multiplication to calculate result where either A or B are >2048 bits so
can't use the standard multiplication method.
*
*/
void esp_mpi_mult_mpi_failover_mod_mult_hw_op(const mbedtls_mpi *X, const mbedtls_mpi *Y, size_t num_words)
{
/* M = 2^num_words - 1, so block is entirely FF */
for (size_t i = 0; i < num_words; i++) {
DPORT_REG_WRITE(RSA_MEM_M_BLOCK_BASE + i * 4, UINT32_MAX);
}
/* Mprime = 1 */
DPORT_REG_WRITE(RSA_M_DASH_REG, 1);
DPORT_REG_WRITE(RSA_LENGTH_REG, num_words - 1);
/* Load X & Y */
mpi_to_mem_block(RSA_MEM_X_BLOCK_BASE, X, num_words);
mpi_to_mem_block(RSA_MEM_Y_BLOCK_BASE, Y, num_words);
/* Rinv = 1, write first word */
DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE, 1);
/* Zero out rest of the Rinv words */
for (size_t i = 1; i < num_words; i++) {
DPORT_REG_WRITE(RSA_MEM_RB_BLOCK_BASE + i * 4, 0);
}
start_op(RSA_MOD_MULT_START_REG);
}