Phoronix reported that a Linux scheduler patch merged into Linux 7.4 Git on October 25, 2026, gives CPU priority precedence over cache locality when a task can move to a higher-priority idle CPU. The change addresses a reported Gentoo compilation problem on an AMD Ryzen AI 9 HX 370.
The reported slowdown involved Clang full-LTO builds
The reported case involved compiling Gentoo with LLVM Clang and full link-time optimization (LTO), a process that lets the compiler optimize across linked program components. Phoronix said tasks placed on the processor’s Zen 5C cores could remain there instead of moving to available, higher-priority Zen 5 cores.
The Ryzen AI 9 HX 370 configuration described in the report has 12 cores: four Zen 5 and eight Zen 5C. The patch rationale compared the Zen 5C cores’ maximum clock of up to 3.3 GHz and 8 MB of L3 cache with about 5.1 GHz and 16 MB for the larger-core cluster. Those differences help explain why the reported task placement mattered for this workload; they do not establish a general slowdown across Ryzen processors or applications.
The patch favors a higher-priority idle CPU
Cache Aware Scheduling favors keeping a process’s tasks close together in a shared last-level cache (LLC), the final cache level before data reaches memory. Asymmetric packing instead favors CPUs with higher scheduling priority. The two policies can therefore prefer different destinations for a task.
Under the reported change, asymmetric packing takes precedence over LLC balancing when it tries to move a task to an idle CPU with higher priority than the task’s current CPU. Phoronix reported that this rule applies to both regular and active load balancing.
Linux 7.2 and 7.3 were backport targets
Phoronix reported the patch merged into Linux 7.4 Git and marked for backporting to Linux 7.2 and Linux 7.3. The report describes those two stable branches as targets, not completed backports.
