---
id: "CVE-2026-99150"
kind: "vulnerability"
aliases: ["CVE-2026-99150"]
title: "Linux amdgpu: debugfs GPR read drops runtime PM reference twice"
severity: "unknown"
cvss: null
epss: null
exploited: false
status: "open"
category: "runtime"
products: []
ecosystems: []
tags: ["amdgpu", "linux", "navi31", "runtime"]
not_affected_if: []
published_at: "2026-10-28T11:17:35+00:00"
first_seen_at: "2026-10-28T11:32:35+00:00"
updated_at: "2026-10-28T11:35:56+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-99150", "first_seen_at": "2026-10-28T11:32:35+00:00"}]
---

# Linux amdgpu: debugfs GPR read drops runtime PM reference twice

In the amdgpu driver's GPR debugfs read path, a failed copy to userspace drops the runtime power-management reference a second time, which corrupts the usage count. The sources describe a refcount underflow or stolen reference on Navi31 GPUs, with no stated attacker capability.

_Summary, check and fix written by a model from the sources; labelled as such._

## Timeline

- 2026-10-28T11:32:35+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-99150

## From the sources

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

drm/amdgpu: don't drop the runtime PM reference twice in amdgpu_gpr

amdgpu_debugfs_gpr_read() drops its runtime PM reference right after
reading the wave registers. When copying them to userspace fails, it
jumps to the err label, which drops the reference a second time. A read
into a bad buffer on Navi31 either steals a reference somebody else
holds (power/runtime_usage goes from 1 to 0) or underflows the count:

  amdgpu 0000:03:00.0: Runtime PM usage count underflow!

The registers are already in the kernel buffer at that point, so finish
the hardware access first and copy the buffer with a single
copy_to_user() instead of the put_user() loop with its own error path.

(cherry picked from commit 31035bbd74f7c3e10c163db875b8e67f8b50ee6a)

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_