Skip to main content
Vulnerability Database/CVE-2026-93220

CVE-2026-93220: Linux Kernel Race Condition Vulnerability

CVE-2026-93220 is a race condition flaw in the Linux kernel sched_ext module that can cause deadlocks when balance callbacks run on foreign CPUs. This article covers the technical details, affected versions, and mitigation strategies.

Published:

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

CVE-2026-93220 Overview

CVE-2026-93220 is a Linux kernel vulnerability in the sched_ext extensible scheduler class. The flaw affects kick_sync_wait_bal_cb(), which assumes it executes on the runqueue's own CPU. Dispatch paths can now drop the runqueue lock while the callback is queued, allowing other lock takers to flush pending balance callbacks on release. This causes the callback to run on a foreign CPU, where it compares against unrelated per-CPU snapshots. When the executing CPU is itself a wait target, the condition produces a deadlock.

Critical Impact

A local workload that triggers scheduler class changes or scx task iteration while kick_sync callbacks are queued can deadlock a CPU, resulting in denial of service.

Affected Products

  • Linux kernel builds with sched_ext (SCX) enabled
  • Stable kernel branches referenced in the upstream fix commits
  • Distributions shipping extensible scheduler support

Discovery Timeline

  • 2026-09-24 - CVE-2026-93220 published to the National Vulnerability Database (NVD)
  • 2026-09-24 - Last updated in NVD database

Technical Details for CVE-2026-93220

Vulnerability Analysis

The vulnerability resides in the sched_ext scheduler class, specifically in the kick_sync_wait_bal_cb() balance callback. This callback is designed to run from the tail of __schedule() on the runqueue's own CPU. The busy-wait loop operates with the runqueue lock dropped and interrupts enabled, relying on per-CPU snapshot data owned by the executing CPU.

Recent changes allow dispatch to drop the runqueue lock while the callback is still queued. During that window, other runqueue lock takers, including scheduler class change paths and the scx task iterator, flush pending balance callbacks when they release the lock. The flush can execute kick_sync_wait_bal_cb() on a CPU other than the one that queued it.

Running the callback on a foreign CPU causes it to inspect unrelated per-CPU snapshots. When the executing CPU happens to be one of the wait targets encoded in cpus_to_sync, the wait never observes progress, producing a deadlock.

Root Cause

The root cause is a broken locality assumption. kick_sync_wait_bal_cb() depends on executing on the runqueue's own CPU to access the correct percpu snapshot area. The dispatch path's ability to drop the runqueue lock violates that invariant by letting foreign CPUs flush the queued callback. This is a race condition [CWE-362] in kernel scheduling code.

Attack Vector

Exploitation requires local workload execution that drives the scheduler through paths which flush balance callbacks after acquiring and releasing the runqueue lock. These paths include scheduler class transitions and scx task iteration. A triggered deadlock halts progress on the affected CPU and can propagate to dependent workloads, resulting in a local denial of service. No remote attack vector is documented.

The upstream fix makes kick_sync_wait_bal_cb() bail out when invoked on a foreign CPU, leaving the wait state untouched. Resched kicks already guarantee progress observation, and the runqueue's next wait picks up the stale cpus_to_sync bits. Technical details are available in the upstream commit c736ea0fe7b4 and the stable backport e0253dd04beb.

Detection Methods for CVE-2026-93220

Indicators of Compromise

  • Soft lockup or RCU stall messages in dmesg referencing kick_sync_wait_bal_cb or sched_ext call paths
  • Hung task warnings on CPUs executing scx scheduler workloads during class transitions
  • Unresponsive processes bound to CPUs participating in sched_ext dispatch

Detection Strategies

  • Monitor kernel logs for lockup traces that include __schedule, balance_callbacks, and sched_ext symbols
  • Inventory hosts running kernels with CONFIG_SCHED_CLASS_EXT enabled and correlate against patched versions
  • Review scheduler telemetry for stalled runqueues and extended wait times on CPUs running SCX policies

Monitoring Recommendations

  • Enable hung_task detection and softlockup panics on hosts running sched_ext workloads to surface events quickly
  • Collect /proc/sched_debug and ftrace scheduler events during reproduction windows to confirm callback locality
  • Aggregate kernel log events in a centralized data lake and alert on sched_ext stall signatures

How to Mitigate CVE-2026-93220

Immediate Actions Required

  • Apply the upstream fix commits c736ea0fe7b4 and e0253dd04beb by updating to a patched kernel from your distribution
  • Audit production kernels for sched_ext usage and prioritize patching on hosts running custom BPF schedulers
  • Test patched kernels against representative scx workloads before broad rollout

Patch Information

The fix modifies kick_sync_wait_bal_cb() to detect execution on a foreign CPU and return without touching the wait state. The resched kicks already guarantee the progress that the wait observes, and the next wait on the runqueue picks up any stale cpus_to_sync bits. Patched kernels are available from upstream stable trees and will propagate through distribution security updates.

Workarounds

  • Disable sched_ext BPF schedulers by unloading the active SCX program and reverting workloads to the default CFS or EEVDF scheduler
  • Avoid scheduler class change operations and scx task iteration on production hosts until a patched kernel is deployed
  • Where kernel configuration is controlled, build without CONFIG_SCHED_CLASS_EXT on hosts that do not require extensible scheduling

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.