Arm64 Drops Big-Endian and Boots CPUs in Parallel
Will Deacon proposes patches to remove big-endian kernel support and implement parallel secondary CPU booting via PSCI for Linux 7.4.
Stock photo for illustration only, not from the actual event
- arm64 drops big-endian kernel support targeting Linux 7.4
- Parallel CPU bringup cuts 192-core boot time by 41 percent
- Embedded teams must verify PSCI firmware and BSP configs
Platform direction for arm64 dominated the week as maintainer Will Deacon posted version two of a patch series removing big-endian kernel support, targeted at Linux 7.4, alongside a new 19-patch series bringing parallel secondary CPU onlining through the PSCI v0.2 CPU_ON argument. Concurrently, Linux 7.3-rc2 arrived with scheduler fixes, Buildroot 2026.08 shipped supporting GCC 16 and glibc 2.44, and kernel.org reported that AI crawlers consume roughly a fifth of its git infrastructure capacity.
The removal of big-endian support marks the end of a staged deprecation rather than an abrupt choice. The CONFIG_CPU_BIG_ENDIAN option was made to depend on BROKEN back in version 6.18, rendering the code unbuildable while dead components gradually left the tree. The v2 series posted on 7 September contains 14 patches deleting around 540 lines of big-endian handling from the arm64 tree, removing paths from BPF JIT, crypto assembly, optimised string routines, compat vDSO builds, and the KVM hypervisor while forcing little-endian explicitly during early boot.
Stock photo for illustration only, not from the actual event
Two specific details matter for embedded teams. First, SETEND emulation for 32-bit user tasks remains intact, ensuring mixed-endian 32-bit userspace continues executing. Second, during review discussions, Arnd Bergmann confirmed that a parallel process applies to 32-bit Arm, where big-endian mode is deprecated in 7.3, the ixp4xx platform runs little-endian, and arm32 removal is anticipated in a subsequent long-term release.
The ongoing removal of legacy architectures and unmaintained configurations from the Linux kernel highlights a broader maintenance strategy aimed at reducing technical debt and improving codebase security. By deprecating features with minimal active testers, maintainers can concentrate engineering efforts on performance enhancements and modern standards that serve active production environments, particularly as little-endian remains the dominant industry baseline.
On the same day, Will Deacon published "arm64: Implement parallel CPU onlining with PSCI v0.2+", a 19-patch series enabling arm64 to bring secondary CPUs into the kernel concurrently instead of sequentially. While architectures like x86, MIPS, and RISC-V already feature parallel bringup, arm64 remained a holdout. This new series supersedes earlier work on HOTPLUG_PARALLEL submitted by Jinjie Ruan of Huawei in June.
The underlying mechanism utilizes PSCI v0.2, which allows the CPU_ON call to pass a context argument directly to incoming CPUs via firmware. This eliminates the serialization bottleneck where each CPU had to identify itself against shared global state, allowing arm64 to drop custom synchronization in favor of generic machinery like HOTPLUG_PARALLEL, HOTPLUG_SPLIT_STARTUP, and core-syncing used by other architectures. Measurements from the predecessor series showed a 192-core HiSilicon server booting its CPUs in 8.6 seconds instead of 14.6, delivering a 41 percent reduction in bringup time.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment