Kbuild speedups, BPF callx and verifier hardening

Kernel traffic on this day concentrated on kbuild changes that promise large reductions in build times, together with BPF work on indirect subprogram calls and tighter frame scoping for callbacks. Several security-oriented patches also landed discussion around USB serial bounds checks, NOMMU pin handling, and io_uring identity handoff.

Kbuild series claims large build speedups

A 21-patch kbuild series proposes substantial acceleration of Linux kernel builds, reporting up to 90 percent faster noop builds by introducing parallel gzip and zstd handling plus lower .mod.c overhead. Linus Torvalds and other maintainers participated in the review. Faster builds matter because they shorten the edit-compile-test loop for the entire kernel community.

BPF gains verified indirect subprogram calls

Alexei posted a 17-patch series for bpf-next that adds a callx instruction enabling verified indirect calls to static BPF subprograms. The work includes libbpf and rodata support aimed at vtables and Rust. This expands the expressiveness of safe BPF programs without relaxing the verifier.

BPF verifier scopes callback arguments to their frame

A six-patch series introduces REF_TYPE_FRAME in the BPF verifier so that callback arguments cannot outlive their stack frame. The change closes kernel memory exposure paths previously reachable through ringbuf and map helpers. Verifier hardening of this kind protects the kernel from crafted BPF programs that try to retain dangling references.

Follow-on kbuild series reiterates speed goals

A separate 20-patch kbuild series (v3) likewise targets 36 to 90 percent faster builds through depcheck improvements, ELF parsing changes, and link-time adjustments. Maintainers are actively reviewing the patches. Continued iteration on these techniques keeps pressure on the long-standing cost of full kernel rebuilds.

RFC for orphaned virtual machines after host kexec

An RFC of 46 patches sketches standalone "Caretaker" execution paths for x86 and ARM virtual machines so they can continue running on preserved CPUs across host kexec or live update. The proof-of-concept explores survival of guest state when the host kernel is replaced. The approach would matter for high-availability setups that cannot tolerate full VM teardown during host updates.

Garmin GPS USB driver bounds fix

A patch corrects a signed integer underflow and out-of-bounds read in the garmin_gps USB serial driver that could be triggered by a userspace write to the corresponding tty device. The flaw allowed malformed packet lengths to reach unsafe buffer handling. Drivers exposed to untrusted userspace input remain a recurring source of local memory-safety bugs.

NOMMU GUP respects FOLL_PIN to avoid UAF

A patch updates NOMMU __get_user_pages_locked so that FOLL_PIN is honoured, eliminating a refcount mismatch. The mismatch could produce use-after-free when io_uring fixed buffers were in use. Correct pin accounting is required for any path that hands user pages to long-lived kernel consumers.

io_uring RFC for thread identity handoff

Jens Axboe posted a 15-patch RFC that lets io_uring hand off thread identity so blockable operations can be issued inline and only offloaded when they actually sleep. The design aims to reduce unnecessary worker scheduling for operations that complete without blocking. Lower overhead for common I/O paths benefits latency-sensitive users of io_uring.