RISC-V Has A Nice Portable Performance Optimization With Linux 7.4

The RISC-V architecture changes were merged today for the Linux 7.4 kernel with one nice optimization standing out. Merged for the Linux 7.3 kernel by Tenstorrent was a RISC-V optimization to skip software aligning for kernels compiled for RISC-V hardware that natively supports fast misaligned memory copies. A nice win up to an 11% bandwidth improvement in tests conducted by Tenstorrent: But the Linux 7.3 kernel required the support to be compiled-in and thus not portable for use across RISC-V kernel builds seeing a mix of hardware, some of which may not support fast misaligned accesses. With Linux 7.4 that skipping of software aligning for RISC-V hardware with "HAVE_EFFICIENT_UNALIGNED_ACCESS" is now punted to run-time. This thus opens distribution RISC-V kernel builds and other more generic kernel builds to carry this nice optimization. The optimization will now be available to skip software aligning in user copy with fast unaligned access being detected at run-time. This improved user copy performance is one of the exciting changes of the RISC-V pull request for Linux 7.4, which was merged a short time ago to Git. RISC-V in Linux 7.4 also adds support for crash hot-plug, optimized code paths involving some page table functions, an optimization during context switching, the "B" RISC-V ISA extension is added to hwcap and hwprobe reporting, and support for reporting the RISC-V Zic64b ISA extension.