[CVE-2026-99148][MODERATE REGULAR] drm/amdgpu: keep the eviction fence until the last handle is closed [ Upstream

[AL-KERNEL <alkernel@kernelcve.org>]

CVE: CVE-2026-99148
Priority: MODERATE REGULAR
AL-KERNEL base severity: MODERATE
KPANIC flag: NO
Patch: drm/amdgpu: keep the eviction fence until the last handle is closed [ Upstream
Commit: dce8f7f03a47b1744dccf506268f9229b67fab47
Upstream patch: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=dce8f7f03a47b1744dccf506268f9229b67fab47
Original CVE announcement: https://lore.kernel.org/linux-cve-announce/?q=CVE-2026-99148
Analysis date: Fri, 30 Oct 2026 12:08:03 -0400
ActionableScore: 4
ActionableScore lower bound: 3
Actionable bucket: Actionable Moderate / Strong Important candidate requiring manual review
Manual review required: YES

Summary:
Premature eviction-fence detachment lets an active AMDGPU user queue consume stale BO mappings after eviction, with cross-workload impact unresolved.

======================================================================
ABOUT THIS REPORT
======================================================================

The original Linux kernel CVE announcement for CVE-2026-99148 is available here:
https://lore.kernel.org/linux-cve-announce/?q=CVE-2026-99148

The original announcement does not normally provide a security severity
estimate, CVSS assessment, or enough information to determine whether the
reported kernel bug represents a practically relevant security issue.

This report was generated by AL-KERNEL, an AI-assisted Linux kernel
vulnerability analysis system developed by Alexander Larkin. It combines
an autonomous classifier with LLM-assisted technical analysis and a
separate ActionableScore mechanism.

The purpose of this report is to prioritize Linux kernel CVEs before
manual review, identify cases that require prompt investigation, and
support automatic closure of issues that are unlikely to have meaningful
security impact.

Published priority for this report: MODERATE REGULAR
Manual review required: YES

A detailed explanation of the methodology and priority rules is included
at the end of this message.

======================================================================
AL-KERNEL CLASSIFICATION RESULT
======================================================================

CVE-2026-99148	MODERATE	https://www.kernelcve.org/list/?q=CVE-2026-99148 CHECK WITH IMPACT FROM ORIG NN LOW | AEGIS IMPACT MODERATE_KPANIC AEGIS_AS=1 LEGACY=MODERATE RESULT=DIFF (final Legacy and Aegis differ; FINAL: MODERATE, Aegis: MODERATE_KPANIC)	Maybe valid. Check manually. Hints by AL-KERNEL: The best (paranoid) CVSS is 'AV:L/AC:H/PR:L/UI:N/S:U/C:L/I:L/A:H';*CWE-664;CWE-362;CWE-667;Other CVSS 'AV:L/AC:H/PR:L/UI:N/S:U/C:N/I:N/A:L';BEST CVSS score: '5.8';DESCR 'amdgpu_gem_object_close detached the user queue eviction fence when any GEM handle was closed even though another handle could still retain the bo_va and its GPU mappings. The BO could then be evicted lacking suspension of the file user queues, allowing GPU access through mappings that were only marked for update. For the CVSS the PR:L value is selected because an attacker needs local access to the AMDGPU DRM interface and control of a user queue and GEM handles. AC:H reflects the need for duplicate handles, BO eviction, supported hardware and user queue behavior, and a useful timing window. Base C:N and I:N are used because the patch does not establish an information leak or write primitive. The paranoid score uses C:L and I:L because a stale GPU mapping may continue to reference VRAM after movement or reuse, while A:H covers a plausible GPU hang or reset affecting other workloads. Automated impact decision path. The original SimpleNet classifier selected LOW. The primary LLM estimated paranoid CVSS 5.8 and the production ActionableScore returned 4 with recommendation Actionable Moderate / Strong Important candidate requiring manual review. The production rules then evaluated CVSS thresholds and ActionableScore thresholds. Recorded rule transitions were CVSS and initial triage stage LOW to MODERATE_KPANIC because the manual review requirement was applied. KP and FD specialist checks stage MODERATE_KPANIC to MODERATE because KP specialist did not establish a qualifying panic. The legacy KP specialist produced a review artifact and its decision was processed by the applicable KP rules. The legacy FD or DH downgrade specialist has no review artifact for this run. Exact triggered rule markers were INCREASED_FROM_LOW_BASED_ON_REQUIREMANUALCHECK and DECREASED_TO_MODERATEREG_BASED_ON_FALSEPOSCHECKOFKP and V41_INHERITED_KPANIC_PROMOTION_BLOCKED and MANUAL_REVIEW_REQUIRED. The independent weaker Aegis shadow evaluated CVSS 2.5 and ActionableScore 1 with primitive class CLASS_C and manual review REQUIRED. Its independently reconciled impact was MODERATE_KPANIC. The models disagreed but the aggregation rules retained the production result. The final reported impact is MODERATE.';YES REQUIRES MANUAL CHECK; ,and ActionableScore result is Actionable Moderate / Strong Important candidate requiring manual review (with actual score 4)	MAYBE	LEAK HARDWARE  INCREASED_FROM_LOW_BASED_ON_REQUIREMANUALCHECK DECREASED_TO_MODERATEREG_BASED_ON_FALSEPOSCHECKOFKP V41_INHERITED_KPANIC_PROMOTION_BLOCKED MANUAL_REVIEW_REQUIRED	-	-	checked

======================================================================
ACTIONABLESCORE ANALYSIS
======================================================================

ActionableScore=4
ActionableScoreLower=3
PrimitiveClass=CLASS_B
RealUAF=NO
HostAdminRequired=UNKNOWN
NamespaceScopedPrivilege=UNKNOWN
ImportantCandidate=YES
ManualReview=REQUIRED
AutoDowngradeAllowed=NO
RequiresHostUID0=UNKNOWN
AuthorizationDecisionBypass=NO
ProtectedResourceImpact=NONE
B4PendingReview=NO
B4PendingConfirmed=NO

1. ActionableScore

* Raw base score: 3.
* Conservative score: 3; no floor or cap applied.
* Raw paranoid score: 4.
* Paranoid score: 4; no floor or cap applied.
* Final recommended bucket: **Actionable Moderate / Strong Important candidate requiring manual review**::

2. Signal breakdown

* +1 weak/indirect active-state corruption: closing one duplicate GEM handle prematurely detaches the BO eviction fence while another handle, `bo_va`, and GPU mappings remain active.
* +1 active synchronization bypass: without the fence, eviction can proceed without suspending user queues that still access the BO.
* +1 secondary mapping staleness: the surviving queue consumes GPU mappings that are only marked for update, rather than this being passive bookkeeping.
* Paranoid +1 plausible device-wide DoS: stale GPU translation use can credibly cause a GPU hang/reset affecting other workloads, as identified by the supplied triage, but no kernel panic or reliable host crash is shown.

Not awarded: no UAF, heap OOB write, controlled reclaim, arbitrary write, semantic-domain transition, confidentiality impact, protected-data integrity impact, or proven cross-process memory access. The same possible hang is not counted again as availability. B5-R, B3-PENDING, and B4-PENDING do not apply.

3. Reachability analysis

The required sequence uses a DRM file, duplicate GEM handles, an AMDGPU user queue, and BO eviction. The duplicate-handle close is directly user-driven, and queue use is attacker-influenced. However, the evidence does not establish render-node permissions or prove that the complete user-queue setup and eviction sequence is available to an ordinary user, a delegated graphics service, DRM master, or only an administrator. Therefore host-admin and namespace privilege remain unknown, with no privilege penalty or unprivileged bonus.

Navi31 and SDMA provide the observed configuration, but the patch does not establish a rare-hardware-only restriction. Eviction/timing requirements may reduce reproducibility, but no sufficiently narrow or uncontrollable race is proven for a timing penalty.

4. Primitive and severity reading

The primitive is premature removal of synchronization protecting a live BO mapping: attacker-influenced handle closure removes the eviction fence, eviction proceeds, and a GPU queue later consumes stale translation state. This is active structured state, not fixed stale storage, a UAF, or an authorization bypass.

It is CLASS_B because a concrete active mapping/synchronization invariant is violated and the stale state reaches a privileged GPU/kernel-managed consumer. A delegated DRM-interface-to-GPU/other-workload trust boundary is credible but unresolved. No specific wrong-memory target, cross-process victim, host-memory overwrite, or LPE is proven.

ImportantCandidate is YES under the active-state/trust-boundary review rule. Auto-downgrade is disallowed because limiting the effect to the originating BO or process has not been demonstrated.

5. One-sentence report phrase

Premature eviction-fence detachment lets an active AMDGPU user queue consume stale BO mappings after eviction, with cross-workload impact unresolved.

6. Manual review advice

MANUAL CHECK REQUIRED

Confirm render-node/user-queue privileges, eviction reproducibility, the resulting GPU address translation, and whether stale accesses can reach storage belonging to another workload.

7. Call-site confidence

Call-site confidence: medium

======================================================================
UPSTREAM PATCH SUMMARY
======================================================================

Patch: drm/amdgpu: keep the eviction fence until the last handle is closed [ Upstream
Commit: dce8f7f03a47b1744dccf506268f9229b67fab47
Upstream URL: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=dce8f7f03a47b1744dccf506268f9229b67fab47

Changed files:
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c

Diff excerpt:

Not included in this email. See the upstream URL for the full patch.

Full patch:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=dce8f7f03a47b1744dccf506268f9229b67fab47

======================================================================
DETAILED REPORT METHODOLOGY
======================================================================

The original Linux kernel CVE announcement for CVE-2026-99148 can be found here:
https://lore.kernel.org/linux-cve-announce/?q=CVE-2026-99148

The original CVE announcement normally does not include a security-level
estimate. In particular, it may not contain a CVSS assessment, an impact
level, or enough information to determine whether the reported bug is a
practically relevant security issue. One purpose of this parallel CVE list
is to provide that missing technical and prioritization information.

The original goal of the AL-KERNEL project was to prioritize Linux kernel
CVE analysis automatically before manual review. The system can also help
identify non-security issues that may be suitable for automatic closure.

This report was generated by AL-KERNEL, an AI-assisted Linux kernel
vulnerability analysis system developed by Alexander Larkin.

The first analysis stage combines an autonomous classifier with additional
LLM-based analysis. The autonomous classifier runs locally on a CPU and is
based on a backpropagation neural network. Together, these mechanisms
produce a technical vulnerability description, identify likely weakness
types, estimate CVSS severity, and provide input for ActionableScore.

Two CVSS estimates are retained because incomplete kernel vulnerability
information often permits more than one defensible interpretation:

  Conservative CVSS vector: AV:L/AC:H/PR:L/UI:N/S:U/C:N/I:N/A:L
  The Best / paranoid CVSS vector: AV:L/AC:H/PR:L/UI:N/S:U/C:L/I:L/A:H
  The Best / paranoid CVSS score: 5.8

The conservative vector represents a lower-impact interpretation.
The Best/paranoid vector intentionally represents a plausible upper-bound
interpretation and should not automatically be treated as demonstrated
real-world impact.

CVSS may also need to be adjusted for a particular Linux deployment,
because actual reachability, privileges, enabled kernel configuration,
hardware, namespaces, exposed device nodes, and other environmental
conditions can differ significantly between systems.

A separate ActionableScore mechanism evaluates practical remediation
urgency. Its analysis may include reachability, attack prerequisites,
subsystem exposure, memory-corruption characteristics, denial-of-service
reliability, and possible confidentiality, integrity, or
privilege-escalation impact.

  Conservative ActionableScore: 3
  Paranoid ActionableScore: 4

The final base severity is taken directly from the second tab-separated
field of the AL-KERNEL classification result. ActionableScore does not
replace or independently override that final AL-KERNEL decision, and
if ActionableScore adjusted impact level of ALKERNEL, then you would see
self-readable flags above like INCREASED_TO_HIGH_BASED_ON_ACTIONABLESCOREHIGHEREQTHAN7.

For an AL-KERNEL result of MODERATE, this report uses the following
additional presentation split:

  ActionableScore below 5   -> MODERATE REGULAR
  ActionableScore 5 or more -> MODERATE 7.0

The distinction between MODERATE REGULAR and MODERATE 7.0 makes it
possible to identify Moderate issues that should receive manual analysis
and fixes before lower-priority MODERATE REGULAR issues. In many cases,
MODERATE REGULAR fixes may wait for a later rebase or routine update.

There is one override in which MODERATE REGULAR becomes MODERATE 7.0
even when the ActionableScore is below 5. When the AL-KERNEL result
contains the KPANIC flag, a MODERATE result is always presented as
MODERATE 7.0. The KPANIC flag selected with few regexps without
usage of AI at all, so it helps to detect cases when Kernel Crash happens
and similar (to filter False-Negative results from the LLM usage).

KPANIC indicates that a reliable kernel crash, kernel panic, or similarly
serious kernel availability impact was identified by the classification
workflow.

AL-KERNEL base severity for this report: MODERATE
KPANIC detected for this report: NO
Published priority for this report (same as in Subject): MODERATE REGULAR

These results are intended to support engineering triage. They are
machine-generated estimates, and cases marked for manual review should
be validated by a human security engineer before final disposition.
For more info read docs linked from here: https://kernelcve.org/
(and you can submit you own patch there to generate such a report
for non-existant CVE-id yet).

Note that in many cases this AI tool selects higher severity, than
real is (means you can expect Importants instead of Moderate 7.0 or
Moderates 7.0 instead of regular Moderates). If you see such cases,
please use reply email interface to add additional manual analyses
info to this particular CVE.
And please, please, let me know when you see Lows instead of Importants
or Important instead of Low (because particular for such cases I
need to tune this AI tool to make it better for this one and next similar).
My contact email for such notifications is alexanjelausa@gmail.com (and both
send reply to CVE record itself too and see "reply" button below for howto reply).