~5x Faster SM4 Cipher Performance With AVX/AES-NI Tuned Linux Kernel Code

Alibaba engineers are looking to mainline an x86_64 tuned version of the SM4 cipher that with making use of AVX and AES-NI can allow for a dramatic performance speed-up.

For those required to make use of China's SM4 block cipher or in use-cases such as supporting the country's WAPI wireless standard, an x86_64/AVX/AES-NI tuned version of SM4 is looking to get mainlined for the Linux kernel. There has already been an AArch64 tuned version of the SM4 code within the Linux kernel while now from Alibaba is the x86_64 accelerated version. The x86_64 tuning is based on existing work by libgcrypt and sm4ni.

Numbers shared by Alibaba on an Intel Xeon Cascade Lake server show the performance being improved by as much as five times faster compared to the existing generic code.

Should you be into SM4 usage, the tentative patches providing this modern Intel/AMD CPU performance speed-up can be found on the kernel mailing list.