Skip to main content
Vulnerability Database/CVE-2026-93809

CVE-2026-93809: Linux Kernel Use-After-Free Vulnerability

CVE-2026-93809 is a use-after-free flaw in the Linux kernel drm/amdgpu module that occurs during module unload. This post explains the technical details, affected versions, impact, and mitigation steps.

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-2026-93809 Overview

CVE-2026-93809 is a Linux kernel vulnerability in the AMD GPU driver (drm/amdgpu). The flaw occurs when the module is unloaded while RCU (Read-Copy-Update) callbacks remain pending. The kernel frees module text before call_rcu() callbacks execute, producing a page fault at an address inside freed module memory. The resolution adds an rcu_barrier() call during module exit to wait for outstanding callbacks before memory is released. The oops trace shows the fault triggered from rcu_do_batch during softirq processing.

Critical Impact

A pending RCU callback executes against freed amdgpu module text, causing a kernel oops and a potential use-after-free condition that can crash affected systems.

Affected Products

  • Linux kernel builds containing the vulnerable drm/amdgpu driver prior to the fix commit
  • Systems using AMD GPU hardware with the in-tree amdgpu kernel module
  • Distributions that permit amdgpu module unload or reload workflows

Discovery Timeline

  • 2026-09-24 - CVE-2026-93809 published to NVD
  • 2026-09-25 - Last updated in NVD database

Technical Details for CVE-2026-93809

Vulnerability Analysis

The amdgpu driver schedules deferred work through call_rcu(), which executes callback functions after a grace period. When the module is unloaded, the kernel removes the module's text and data sections. If any call_rcu() callback remains pending at unload time, the callback pointer still references a function inside the now-freed module text.

When RCU subsequently processes its batch via rcu_do_batch(), it dereferences the stale function pointer. The resulting instruction fetch targets unmapped memory, producing the page fault shown at RIP: 0010:0xffffffffc1d59c40. The faulting address lies in the module load region (0xffffffffc...), confirming the callback belonged to code that has been removed.

The fix adds rcu_barrier() to the module exit path. This call blocks until all previously scheduled RCU callbacks have completed, ensuring no deferred work references module text that is about to be freed. The upstream fix is tracked at kernel commit 67a654b41cfa73c3b83402c4a01b2689cad5b9bc.

Root Cause

The root cause is a missing synchronization barrier between call_rcu() scheduling and module teardown. The amdgpu exit routine releases module memory without draining pending RCU callbacks, creating a use-after-free of module text.

Attack Vector

Triggering requires the ability to unload the amdgpu module while callbacks are pending. On typical systems, module unload requires CAP_SYS_MODULE, limiting the attack surface to privileged contexts or automated workflows that reload graphics drivers. Successful triggering produces a kernel oops and system instability.

No verified proof-of-concept code is published. Refer to the Kernel Git Commit Reference for the complete patch details.

Detection Methods for CVE-2026-93809

Indicators of Compromise

  • Kernel oops entries referencing rcu_do_batch+0x163/0x450 and rcu_core in the call trace
  • Page fault messages with faulting RIP addresses in the module region (0xffffffffc...)
  • Unexpected amdgpu module unload or reload events in system logs

Detection Strategies

  • Parse dmesg and /var/log/kern.log for oops signatures matching the RCU callback fault pattern
  • Monitor kernel crash dumps on systems running AMD GPU hardware and correlate with module lifecycle events
  • Review systemd journal entries for modprobe or rmmod invocations targeting amdgpu

Monitoring Recommendations

  • Enable kdump or equivalent crash collection to capture full stack traces when the fault occurs
  • Alert on repeated kernel soft lockups or oops events on graphics-capable hosts
  • Track kernel version inventory to confirm fixed builds are deployed across the fleet

How to Mitigate CVE-2026-93809

Immediate Actions Required

  • Update to a Linux kernel build that includes commit 67a654b41cfa73c3b83402c4a01b2689cad5b9bc
  • Avoid unloading the amdgpu module on unpatched systems, including during driver upgrade workflows
  • Restrict CAP_SYS_MODULE to administrative accounts and audit automated module management scripts

Patch Information

The fix adds rcu_barrier() to the amdgpu module exit function to flush pending RCU callbacks before module text is freed. See the Kernel Git Commit Reference for the upstream commit. Distribution kernel maintainers backport this fix into stable release trains; verify your vendor's errata before deployment.

Workarounds

  • Keep the amdgpu module permanently loaded and reboot the host rather than unloading the driver
  • Blacklist unload operations via administrative controls on multi-tenant or shared systems
  • Schedule kernel upgrades during maintenance windows to apply the patched build
bash
# Verify running kernel version and amdgpu module status
uname -r
lsmod | grep amdgpu

# Check whether the patched commit is present in the installed kernel source
git -C /usr/src/linux log --oneline | grep 67a654b41cfa

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.