rseq_slice_yield(2)

Purpose

rseq_slice_yield() is the query/clear side of the rseq time-slice-extension feature: a task that was granted a slice extension yields the CPU once it has finished the critical work, and this syscall reports – and resets – whether the yield actually happened in the granted window (kernel/rseq.c, comment above SYSCALL_DEFINE0(rseq_slice_yield)). It exists so the extension stays side-effect free: the kernel needs a cheap, explicit handshake to know the task gave the CPU back inside the extension rather than being forced out.

Synopsis

/* no glibc declaration; called directly */
long syscall(SYS_rseq_slice_yield);

Arguments and behavior

No arguments. The kernel reads current->rseq.slice.yielded, clears it, and returns !!yielded (kernel/rseq.c, body). The comment in the source states the contract: returns 1 if the task successfully yielded the CPU within the granted slice; returns 0 if the slice extension was never granted, was revoked by going over the granted extension, by issuing a syscall other than this one, or by being scheduled out earlier due to an interrupt. The syscall itself does not schedule – the syscall-entry work immediately relinquishes the CPU and reschedules if required (kernel/rseq.c, comment). The extension length is tunable through the rseq_slice_ext sysfs knob, bounded by rseq_slice_ext_nsecs_min/max (kernel/rseq.c, rseq_slice_ext_show/rseq_slice_ext_write).

Return value and errors

1 (yielded inside the granted slice) or 0 (extension not granted/revoked). No error returns are defined in the implementation – the syscall cannot fail (kernel/rseq.c, body).

Flags and constants

None.

Security and notes

This is a cooperative-scheduling knob, not a correctness primitive: misreporting cannot corrupt the kernel, and the 0/1 contract is deliberately trivial. The feature as a whole is gated by the rseq slice-extension config/sysfs knob, so ENOSYS is possible on kernels without it – callers must treat any nonzero negative return as “feature absent”. This page is source-limited: the sole reachable primary source is kernel/rseq.c (master, fall 2026); no man page exists yet, so the behavior section quotes the implementation’s own contract comment. Cross-reference: rseq, sched_yield.

References

• Kernel source: kernel/rseq.c (SYSCALL_DEFINE0(rseq_slice_yield) and the contract comment above it, torvalds/linux master, fetched 2026-10-17)
• asm-generic numbering: scripts/syscall.tbl line 471 common rseq_slice_yield sys_rseq_slice_yield
• worklog: no glibc wrapper symbol (nm -D, glibc 2.41)