yuzu-emu/mbedtls
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Prevent perf regressions in mbedtls_xor

Browse source 
Signed-off-by: Dave Rodgman <dave.rodgman@arm.com>
This commit is contained in:
Dave Rodgman 2022-12-30 21:32:03 +00:00
parent 051225d07a
commit b9cd19bc8c
3 changed files with 17 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
library/alignment.h
View file

@ -29,6 +29,19 @@
#include "mbedtls/build_info.h"
/*
* Define MBEDTLS_EFFICIENT_UNALIGNED_ACCESS for architectures where unaligned memory
* accesses are known to be safe and efficient.
*/
#if defined(__ARM_FEATURE_UNALIGNED) \
|| defined(__i386__) || defined(__amd64__) || defined(__x86_64__)
/*
* __ARM_FEATURE_UNALIGNED is defined where appropriate by armcc, gcc 7, clang 9
* (and later versions); all x86 platforms should have efficient unaligned access.
*/
#define MBEDTLS_EFFICIENT_UNALIGNED_ACCESS
#endif
/**
* Read the unsigned 16 bits integer from the given address, which need not
* be aligned.

6
library/common.h
View file

@ -122,11 +122,13 @@ static inline const unsigned char *mbedtls_buffer_offset_const(
*/
inline void mbedtls_xor(unsigned char *r, const unsigned char *a, const unsigned char *b, size_t n)
{
size_t i;
for (i = 0; (i + 4) <= n; i += 4) {
size_t i = 0;
#if defined(MBEDTLS_EFFICIENT_UNALIGNED_ACCESS)
for (; (i + 4) <= n; i += 4) {
uint32_t x = mbedtls_get_unaligned_uint32(a + i) ^ mbedtls_get_unaligned_uint32(b + i);
mbedtls_put_unaligned_uint32(r + i, x);
}
#endif
for (; i < n; i++) {
r[i] = a[i] ^ b[i];
}

9
library/constant_time.c
View file

@ -47,15 +47,6 @@
#include <string.h>
/*
* Define MBEDTLS_EFFICIENT_UNALIGNED_ACCESS for architectures where unaligned memory
* accesses are known to be safe and efficient.
*/
#if defined(__ARM_FEATURE_UNALIGNED)
/* __ARM_FEATURE_UNALIGNED is defined by armcc, gcc 7, clang 9 and later versions */
#define MBEDTLS_EFFICIENT_UNALIGNED_ACCESS
#endif
/*
* Define MBEDTLS_EFFICIENT_UNALIGNED_VOLATILE_ACCESS where assembly is present to
* perform fast unaligned access to volatile data.