freenode

← digests

Kbuild speedups and verifier, buffer, RCU fixes

Kernel & Low-Level2026-09-11

Kernel traffic centered on a large kbuild performance series and several correctness and security-oriented fixes in eBPF, the buffer layer, Tasks RCU, and the module loader. Networking and KVM guest memory work also advanced with targeted patches.

kbuild series reports faster kernel builds

A 23-patch kbuild series on linux-kernel claims 20 to 36 percent faster allmodconfig builds, with larger gains for incremental and noop rebuilds on x86 and arm64. The changes aim to cut wall-clock time for full and partial kernel compiles. Developers and CI systems that rebuild frequently would benefit if the series is merged.

Logic bug in eBPF verifier error path

A public security report on bpf described a logic bug in the eBPF verifier function reg_bounds_sanity_check on its error path. Greg KH asked for the issue to be fixed through a normal patch. Verifier bugs matter because bounds checking underpins the safety guarantees for BPF programs running in the kernel.

Buffer layer stops clearing BH_Uptodate on write failure

A v2 series of 21 patches on linux-kernel changes buffer write-error handling so BH_Uptodate is no longer cleared when a write fails, and moves filesystems onto BH_Write_EIO instead. The discussion drew multiple filesystem participants. Clearer error state tracking helps prevent failed writes from being treated as valid cached data.

Tasks RCU RFC on preemption as quiescent state

A 13-patch RFC on bpf proposes treating preemption outside trampolines as a Tasks RCU quiescent state. The intent is to avoid stalls caused by long-running kthreads on PREEMPT_LAZY kernels. The change targets BPF trampoline and lazy-preemption combinations that can otherwise delay grace periods.

Module loader rejects out-of-range relocation indices

A patch on linux-kernel adds an sh_info bounds check in the module loader so out-of-range relocation target indices are rejected early. The goal is to stop out-of-bounds reads and crashes on ARM, ARM64, RISC-V, and similar architectures before architecture-specific code runs. Early validation hardens loading of malformed modules.

BPF SCC computation drops quadratic successor scan

A six-patch BPF series fixes an O(k^2) successor rescan in bpf_compute_scc and bounds total indirect edges by the instruction limit. The work addresses gotox-related paths in the verifier. Predictable verification time matters for complex programs that would otherwise hit quadratic cost.

taprio fix for infinite dequeue loop

A netdev patch fixes an infinite loop in taprio_dequeue_tc_priority that can lock up under concurrent mqprio queue mapping changes. The bug affects the time-aware priority qdisc used for deterministic scheduling. Operators running taprio would care about avoiding hard lockups during concurrent reconfiguration.

guest_memfd in-place conversion series at v13

Version 13 of a 44-patch guest_memfd in-place conversion series landed on linux-kernel with new KVM selftests for shared and private memory handling. The series continues work on converting guest memory between shared and private states in place. KVM and confidential computing developers tracking guest_memfd will want the updated tests and conversion paths.