For sqr and mul (28/32 bits) the sums of products can exceed maximum
64-bit signed value. Handle last sum value, that may cause overflow,
separately.
Fix formatting of overflow last line.
Added new x86 assembly for AES.
AES-CBC decrypt only 4 blocks at a time (not 6 or 8) due to reduces
register count.
GCM implementation for AVX2, AVX1 and AESNI only.
Disabled looking for other assembly files for x86.
For user_settings.h builds, .S assembly files need to include user_settings.h
in order to get the defines used by the build. However, a user_settings.h may
contain code only understood by a C compiler and not the assembler (e.g. a
typedef). This commit makes it so our autotools and CMake builds produce a file
user_settings_asm.h when doing a user_settings.h build. This generated header
contains only the preprocessor directives from the user_settings.h. As a result,
it can be safely included by our assembly code files.
* add Versal specific glue
The same structure of an "XSecure client" is used throughout the API's,
therefor define it once and re-use in all clients.
* integrate Versal AES-GCM engine
* integrate Versal SHA3-384 engine
* add versal support to tests
- There's no intermediate-hash API for Versal.
* add specific test with large AAD
Test only with `n*16 byte` wide chunks of AAD, so it gets processed in the
hardware engine.
* add specific test with misaligned AES-GCM arguments
* integrate Versal RSA engine
* disable failing RSA test-case when Xilinx Crypto is enabled
* introduce define `WOLFSSL_XILINX_CRYPT_VERSAL`
* integrate Versal TRNG engine
* allow using Versal TRNG w/o wolfcrypt DRBG
Versal TRNG already provides a HRNG mode which does the same as the
wolfcrypt DRBG implementation.
* add support for user-supplied nonce to Versal TRNG
* add `wc_XsecureErrorToString()` to map PLM error codes to messages.
* integrate Versal EcDSA engine
* update tests to work with Versal EcDSA
If deterministic K is enabled, the tests failed here since the Versal
EcDSA engine doesn't support the SECP256R1 curve yet.
* Xilinx crypto engines like aligned memory very much
Make this a default choice, not via the user configuration.
* add Xilinx-specific `WOLFSSL_MSG()` equivalent
`WOLFSSL_XIL_MSG()` does the same as `WOLFSSL_MSG()` besides waiting for
1 second before printing to stdout, since the PLM maybe prints to same and
outputs would be mixed up.
This waiting can be disabled by defining `WOLFSSL_XIL_MSG_NO_SLEEP`.
* add option to enable DPA CounterMeasures in AES-GCM crypto engine
* add "command mode" to Xilinx bare-metal example
* update Xilinx default user settings
* add script to execute benchmarks
* add scripts to create graphics
* add Vitis 2022.1 example projects
Signed-off-by: Steffen Jaeckel <jaeckel-floss@eyet-services.de>
* add option to run benchmarks with and w/o AAD
This adds the possibility to benchmark AES-GCM and AES-CCM both with and
w/o AAD, with the option to also run both.
The default behavior when wolfSSL provides the `main()` function is
unchanged.
The default behavior when wolfSSL doesn't provide the `main()` function has
been changed to "run both benchmarks - with and w/o ADD".
* add option to run benchmarks against 4096bit RSA&DH keys
* remove `BENCH_SIZE` macro from benchmark.c
* pre-define benchmark sizes in a single place, before it had to be done in two places
* improve `benchmark_static_init()`
- static variable doesn't need to be in global scope
- add option to force re-init
- add more static variables to be reset
* add `-blocks` option to benchmarks
* expose benchmark `main()` as `wolfcrypt_benchmark_main()`
* fix benchmark `-?` output
* use correct SI/Binary prefix in benchmarks
* use a separate column per detail in CSV output of benchmark
* add `-aad_size` option to benchmark
* don't always print symmetric CSV headers
* always use M[i]B/s when output format is CSV
* Versal specific patches for benchmarks
This also removes the default define for `COUNTS_PER_SECOND` for Xilinx
targets, since I prefer to have a build failure over wrongly calculated
output.
Signed-off-by: Steffen Jaeckel <jaeckel-floss@eyet-services.de>
* fix compilation warning
* adjust SHA3-384 test error-codes
The way the codes were constructed before, they were not unique.
* unify code
Instead of having `ifdef`'s in the code, define our own wrapper around
the keysource as required.
* add CMake option for help-text in wolfCrypt tests
* expose test `main()` as `wolfcrypt_test_main()`
* Don't overwrite previously set errors
* add FreeRTOS support for Xilinx demo
* move `fp_reverse` from `tfm.c` to `wolfmath.c` and rename to
`mp_reverse`.
Signed-off-by: Steffen Jaeckel <jaeckel-floss@eyet-services.de>
Generation scripts fixed for ARM32.
Regenerated output inlcudes:
- support for ARM32 architectures less than 7
- SP code also ensures parameters are in specific registers
Make logic to avoid masking return error conditionally compiled based on STSAFE configuration
Update logic at second crypto-callback location to return error code rather than mask it
