Phoebe Chen

We (SiFive) are currently engaged in the development of RISC-V vector cryptography, and have implemented several critical cryptographic optimization patches for OpenSSL. We will contribute the source code, which can...

We at SiFive have upstreamed our RVV crypto optimization to OpenSSL and all patches have been merged into the mainline. https://github.com/openssl/openssl/pull/21923.

@prashanthswami Thank you for providing support for the Android RISC-V configuration! The main configuration file contains the riscv64 target configuration https://github.com/openssl/openssl/blob/master/Configurations/10-main.conf#L810-L816 may be useful to you. I'm not well-versed in...

Hi @fbarchard @alankelly, This PR only enables RVV x32-transpose microkernels. If you have free time, please help to review. Thanks

I have cross-compiled using riscv-unknown-linux-gnu-gcc toolchain as indicated by the following configuration: `Configure linux64-riscv64 --cross-compile-prefix=riscv64-unknown-linux-gnu- --prefix=/path/to/openssl/build -mabi=lp64d --strict-warnings enable-fips` and use openssl speed to test AES GCM for this PR...

> @phoebesv, which QEMU patchset did you use? > > My test process also uses cross-compilation and executes `make test`. I will try to get `openssl speed` running as well...

@cmuellner Thank you for testing, I have noticed that the issue has been resolved in the master branch.(https://github.com/openssl/openssl/pull/20306) openssl-speed issue could be resolved after rebasing to the latest master commit.

@cmuellner We appreciate your contribution to the RVV crypto optimization code. Recently, we tested AES-GCM implementations with RISC-V hardware capabilities enabled using `$OPENSSL_riscvcap=OPENSSL_riscvcap_v_zvbb_zvbc_zvkned` on qemu, but encountered unexpected results in...

@cmuellner Here are the parts of my log where all gcm-tests for evpciph_aes_common.txt failed. ``` # INFO: @ test/evp_test.c:3633 # openssl/test/recipes/30-test_evp_data/evpciph_aes_common.txt:946: Source of above error (aligned in-place, not fragmented); unexpected...

@cmuellner I think the root cause of the failure of the test is the incorrect mask policy in `ghash-riscv64-zvbb-zvbc.pl`. It should be set as `mask-undisturbed`. Tail/mask-agnostic setting should be aware,...