Linux 7.0 Slab Fix On The Way For A "Severe Performance Regression"

Key Takeaways

• Linux 7.0 slab fixes address 64% IOPS drop
• Regression caused by sheaf refill restrictions in SLUB allocator
• Red Hat reported issue; SUSE engineer provided interim fix
• Fix enables sheaf refill when blocking not allowed
• Full resolution expected in upcoming patch before rc3

Summary

Today’s batch of Linux 7.0 slab allocator patches includes a fix for a severe performance regression that caused roughly a 64 % drop in IOPS on the ublk null target benchmark. The regression stemmed from unnecessary sheaf refill restrictions introduced by the new SLUB “sheaves” series merged last month. Red Hat engineer Ming Lei identified the issue, and SUSE developer Vlastimil Babka supplied an interim patch that permits sheaf refills when blocking is disallowed. The patch is slated for inclusion before the Linux 7.0‑rc3 release, with a further update expected to fully resolve the slowdown.

Pulse Analysis

The Linux kernel’s slab allocator is a core memory‑management subsystem that balances speed and fragmentation across CPUs. Recent enhancements for the 7.0 series introduced the SLUB “sheaves” concept, aiming to improve cross‑CPU allocation efficiency. However, an unintended side‑effect limited sheaf refills under non‑blocking conditions, creating a bottleneck that manifested as a steep 64 % IOPS decline in the ublk null target benchmark—a metric critical for high‑throughput storage stacks.

Industry stakeholders quickly noticed the anomaly after the merge window closed in February. Red Hat’s Ming Lei documented the performance drop, highlighting the regression’s impact on workloads with persistent cross‑CPU alloc/free patterns, a common scenario in cloud and hyperscale environments. The community’s rapid response, including a kernel panic that blocked precise bisecting, underscored the collaborative nature of open‑source debugging. SUSE engineer Vlastimil Babka’s interim patch relaxes the sheaf refill restriction when blocking is prohibited, restoring a portion of the lost throughput while a more comprehensive solution is prepared.

Looking ahead, the pending fix is expected to land before the Linux 7.0‑rc3 milestone, with an additional patch targeting memory‑less nodes slated for the subsequent release. Resolving this regression is vital for maintaining the kernel’s performance reputation, especially as enterprises adopt Linux 7.0 for next‑generation data‑center workloads. The episode also illustrates how early‑stage kernel changes can ripple through production environments, reinforcing the need for vigilant testing and swift upstream collaboration.