Homepage

Improper Update of Reference Count Affecting kernel6.18-debuginfo package, versions <1:6.18.25-55.108.amzn2023

Severity

 Recommended
high

Based on Amazon Linux security rating.

Threat Intelligence

EPSS
0.01% (3rd percentile)

Do your applications use this vulnerable package?

In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.

Test your applications
  • Snyk IDSNYK-AMZN2023-KERNEL618DEBUGINFO-16736629
  • published18 May 2026
  • disclosed24 Apr 2026
Report a new vulnerability Found a mistake?

Introduced: 24 Apr 2026

NewCVE-2026-31656  (opens in a new tab)
CWE-911  (opens in a new tab)

How to fix?

Upgrade Amazon-Linux:2023 kernel6.18-debuginfo to version 1:6.18.25-55.108.amzn2023 or higher.
This issue was patched in ALAS2023-2026-1693.

NVD Description

Note: Versions mentioned in the description apply only to the upstream kernel6.18-debuginfo package and not the kernel6.18-debuginfo package as distributed by Amazon-Linux. See How to fix? for Amazon-Linux:2023 relevant fixed versions and status.

In the Linux kernel, the following vulnerability has been resolved:

drm/i915/gt: fix refcount underflow in intel_engine_park_heartbeat

A use-after-free / refcount underflow is possible when the heartbeat worker and intel_engine_park_heartbeat() race to release the same engine->heartbeat.systole request.

The heartbeat worker reads engine->heartbeat.systole and calls i915_request_put() on it when the request is complete, but clears the pointer in a separate, non-atomic step. Concurrently, a request retirement on another CPU can drop the engine wakeref to zero, triggering __engine_park() -> intel_engine_park_heartbeat(). If the heartbeat timer is pending at that point, cancel_delayed_work() returns true and intel_engine_park_heartbeat() reads the stale non-NULL systole pointer and calls i915_request_put() on it again, causing a refcount underflow:

&lt;4&gt; [487.221889] Workqueue: i915-unordered engine_retire [i915]
&lt;4&gt; [487.222640] RIP: 0010:refcount_warn_saturate+0x68/0xb0
...
&lt;4&gt; [487.222707] Call Trace:
&lt;4&gt; [487.222711]  &lt;TASK&gt;
&lt;4&gt; [487.222716]  intel_engine_park_heartbeat.part.0+0x6f/0x80 [i915]
&lt;4&gt; [487.223115]  intel_engine_park_heartbeat+0x25/0x40 [i915]
&lt;4&gt; [487.223566]  __engine_park+0xb9/0x650 [i915]
&lt;4&gt; [487.223973]  ____intel_wakeref_put_last+0x2e/0xb0 [i915]
&lt;4&gt; [487.224408]  __intel_wakeref_put_last+0x72/0x90 [i915]
&lt;4&gt; [487.224797]  intel_context_exit_engine+0x7c/0x80 [i915]
&lt;4&gt; [487.225238]  intel_context_exit+0xf1/0x1b0 [i915]
&lt;4&gt; [487.225695]  i915_request_retire.part.0+0x1b9/0x530 [i915]
&lt;4&gt; [487.226178]  i915_request_retire+0x1c/0x40 [i915]
&lt;4&gt; [487.226625]  engine_retire+0x122/0x180 [i915]
&lt;4&gt; [487.227037]  process_one_work+0x239/0x760
&lt;4&gt; [487.227060]  worker_thread+0x200/0x3f0
&lt;4&gt; [487.227068]  ? __pfx_worker_thread+0x10/0x10
&lt;4&gt; [487.227075]  kthread+0x10d/0x150
&lt;4&gt; [487.227083]  ? __pfx_kthread+0x10/0x10
&lt;4&gt; [487.227092]  ret_from_fork+0x3d4/0x480
&lt;4&gt; [487.227099]  ? __pfx_kthread+0x10/0x10
&lt;4&gt; [487.227107]  ret_from_fork_asm+0x1a/0x30
&lt;4&gt; [487.227141]  &lt;/TASK&gt;

Fix this by replacing the non-atomic pointer read + separate clear with xchg() in both racing paths. xchg() is a single indivisible hardware instruction that atomically reads the old pointer and writes NULL. This guarantees only one of the two concurrent callers obtains the non-NULL pointer and performs the put, the other gets NULL and skips it.

(cherry picked from commit 13238dc0ee4f9ab8dafa2cca7295736191ae2f42)