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-2026-97960 Overview
CVE-2026-97960 is a Linux kernel vulnerability in the Intel x86 performance monitoring subsystem. The flaw affects the drain_pebs() function in perf/x86/intel, which handles the Precise Event Based Sampling (PEBS) buffer shared by all events on a CPU. Reentry into drain_pebs() can cause one instance to observe stale buffer state and potentially access out-of-bounds memory.
Critical Impact
Reentrant execution of drain_pebs() can lead to out-of-bounds memory access in kernel space, resulting in potential kernel memory corruption or information disclosure on affected systems.
Affected Products
- Linux kernel versions containing the vulnerable perf/x86/intel PEBS implementation
- Systems running Intel x86 processors with PEBS-capable Performance Monitoring Units (PMUs)
- Distributions shipping unpatched upstream kernels prior to the fix commits
Discovery Timeline
- 2026-09-25 - CVE-2026-97960 published to NVD
- 2026-09-25 - Last updated in NVD database
Technical Details for CVE-2026-97960
Vulnerability Analysis
The PEBS buffer is a per-CPU resource shared by all performance events scheduled on that CPU. The drain_pebs() routine processes entries from this buffer and must execute atomically with respect to itself. Most invocations originate in Non-Maskable Interrupt (NMI) context, where reentry is naturally prevented by hardware.
However, drain_pebs() is also reachable from process context through intel_pmu_drain_pebs_buffer(). In those paths, the PMU is often already disabled but not guaranteed to be. Specifically, __intel_pmu_pebs_disable() only disables the target counter, leaving other active counters free to raise a Performance Monitoring Interrupt (PMI).
A PMI raised on another active counter can interrupt an in-flight drain_pebs() call, triggering reentry. The second instance may then operate on stale buffer indices, leading to out-of-bounds reads or writes within kernel memory.
Root Cause
The root cause is insufficient synchronization around a shared per-CPU resource. The call path __perf_addr_filters_adjust() → perf_event_stop() → __perf_event_stop() → x86_pmu_stop() → intel_pmu_disable_event() → intel_pmu_pebs_disable() → __intel_pmu_pebs_disable() → intel_pmu_drain_large_pebs() → intel_pmu_drain_pebs_buffer() reaches the drain logic without disabling the full PMU, allowing PMIs on sibling counters to preempt the drain.
Attack Vector
Exploitation requires local code execution capable of invoking perf subsystem operations that trigger intel_pmu_drain_pebs_buffer() from process context while other counters remain active. An attacker with the ability to configure performance events may race PMI delivery against drain operations to induce the reentry condition. The vulnerability is a kernel-level race condition and does not expose a remote attack surface.
See the upstream commits for the complete code-level analysis: Git Kernel Commit a56c03a, Git Kernel Commit a5fe19d, and Git Kernel Commit c55599c.
Detection Methods for CVE-2026-97960
Indicators of Compromise
- Kernel warnings emitted from intel_pmu_drain_pebs_buffer() indicating the full PMU was not disabled when the drain executed
- Unexpected kernel oops or panic traces referencing drain_pebs, intel_pmu_drain_pebs_buffer, or intel_pmu_drain_large_pebs
- Memory corruption reports from KASAN (Kernel Address Sanitizer) pointing to PEBS buffer regions
Detection Strategies
- Audit installed kernel package versions against distribution advisories referencing the three upstream fix commits
- Enable CONFIG_DEBUG_KERNEL and KASAN in test environments to surface out-of-bounds accesses in the PEBS drain path
- Monitor dmesg for warnings introduced by the patch when the full PMU is not disabled during a drain
Monitoring Recommendations
- Collect kernel ring buffer logs centrally and alert on stack traces containing drain_pebs or intel_pmu_drain_pebs_buffer
- Track usage of the perf_event_open() syscall from unprivileged processes, particularly where address filters are configured
- Review which users and containers hold CAP_PERFMON or CAP_SYS_ADMIN and can exercise the vulnerable code path
How to Mitigate CVE-2026-97960
Immediate Actions Required
- Apply the upstream kernel patches that introduce __intel_pmu_quiesce() and __intel_pmu_resume() helpers to prevent reentry
- Update to a kernel build that incorporates commits a56c03a, a5fe19d, and c55599c
- Prioritize patching on multi-tenant systems where untrusted users can invoke the perf subsystem
Patch Information
The fix disables the full PMU around intel_pmu_drain_pebs_buffer() calls originating in process context, preventing sibling counters from raising a PMI during the drain. A warning is also added to intel_pmu_drain_pebs_buffer() to catch future regressions where the full PMU is not disabled. Patches are available at Git Kernel Commit a56c03a, Git Kernel Commit a5fe19d, and Git Kernel Commit c55599c.
Workarounds
- Restrict access to the perf subsystem by setting kernel.perf_event_paranoid to 3 to block unprivileged users from opening performance events
- Remove CAP_PERFMON and CAP_SYS_ADMIN from workloads that do not require performance monitoring
- Where feasible, disable PEBS-based profiling on affected hosts until the patched kernel is deployed
# Configuration example
# Restrict unprivileged access to the perf subsystem
sysctl -w kernel.perf_event_paranoid=3
echo 'kernel.perf_event_paranoid=3' >> /etc/sysctl.d/99-cve-2026-97960.conf
# Verify installed kernel version against patched builds
uname -r
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

