News › performance
By Zayden R., August 22, 2026
Linux 7.3 introduces x86/mm patches that significantly reduce latency by addressing TLB flushing code. Engineers can expect improved real-time performance, crucial for latency-sensitive environments.
Linux 7.3 is set to bring a notable improvement for those running latency-sensitive workloads, thanks to recently merged patches focused on the x86/mm architecture. These changes, primarily aimed at reducing the time interrupts are disabled during TLB flushing, promise to enhance real-time performance significantly. This development is especially relevant for environments relying on low latency, such as those utilizing DPDK.
The work, spearheaded by Bytedance and detailed by engineer Chuyi Zhou, addresses the latency issues caused by the smp_call_function*() calls. These functions previously blocked until remote CPUs completed their tasks, hindering scheduling and increasing latency. The patches allow preemption during IPI completion, reducing the wait-induced scheduling latency that could reach up to 16ms on a 16-core machine.
In practical terms, after applying this patchset, engineers should notice that preemption is rarely disabled for more than 1ms on the critical path of arch_tlbbatch_flush/flush_tlb_mm_range. This is a substantial improvement from the previous state, with P99 of max preemption disabled events in a 30-second interval reduced to around 1.5ms. The remaining latency challenges are mainly due to lock contention.
While these patches have undergone extensive testing, including ten rounds of patch review, Intel engineer Dave Hansen has advised keeping an eye out for potential regressions. As always, it's prudent for engineers to test these changes in their specific environments to ensure compatibility and performance gains.
The Linux Camp teaches these topics as hands-on labs on real virtual machines, verified as you type.