CVE-2026-64309: Linux Kernel Crypto CCP DoS Vulnerability

CVE-2026-64309 Overview

CVE-2026-64309 affects the Linux kernel Cryptographic Coprocessor (CCP) driver used for AMD Secure Encrypted Virtualization with Secure Nested Paging (SEV-SNP). The vulnerability involves incorrect firmware initialization when handling the ioctl(SNP_COMMIT) command through /dev/sev. The kernel unnecessarily initialized SNP firmware for a command that does not require it, creating a code path where userspace could disrupt active virtualization state.

Critical Impact
If SEV initialization fails while KVM hosts active virtual machines, a userspace process invoking /dev/sev ioctls could zero out MSR_VM_HSAVE_PA globally, potentially causing a general protection fault on the next VMRUN and crashing the host.

Affected Products

Linux kernel CCP (Cryptographic Coprocessor) driver
AMD SEV-SNP firmware interface via /dev/sev
KVM hypervisor hosts running AMD EPYC processors with SEV-SNP

Discovery Timeline

2026-07-25 - CVE-2026-64309 published to NVD
2026-07-25 - Last updated in NVD database

Technical Details for CVE-2026-64309

Vulnerability Analysis

The issue resides in the AMD CCP driver's handling of the SNP_COMMIT ioctl command. The SNP_COMMIT command does not require the SEV-SNP firmware to be in any particular initialization state. However, the kernel path attempted to initialize SNP even when the firmware was previously uninitialized. If that initialization failed while KVM was already running non-SNP guests, the failure handler could zero out MSR_VM_HSAVE_PA on all CPUs. The next VMRUN executed by an active VM would then take a general protection fault, crashing the host kernel.

Root Cause

The root cause is an incorrect state assumption in the CCP driver. The SEV-SNP firmware specification (document 56860) did not list SNP_COMMIT in Table 5 as a command permitted in the UNINIT state, yet the command is in fact allowed and a documentation update will reflect that. The driver treated SNP_COMMIT as if it required initialization, triggering unnecessary firmware state transitions and cleanup side effects on failure paths.

Attack Vector

A local userspace process with access to /dev/sev can issue SEV ioctls such as SEV_PDH_GEN or SNP_COMMIT to trigger the vulnerable initialization path. On a system where SEV initialization fails, this triggers global MSR modification affecting all CPUs. Active KVM guests unrelated to SEV-SNP would then crash the host at the next VMRUN, producing a denial-of-service condition against the hypervisor and every tenant workload it hosts.

The fix skips SNP initialization for SNP_COMMIT when the firmware was previously uninitialized. See the upstream kernel commit for the patch implementation.

Detection Methods for CVE-2026-64309

Indicators of Compromise

Unexpected host kernel panics or general protection faults referencing VMRUN on AMD EPYC hosts running KVM
Kernel log entries showing SEV or SNP initialization failures followed by VM instability
Userspace processes opening /dev/sev and issuing SEV_PDH_GEN, SNP_COMMIT, or related ioctls outside of expected virtualization management workflows

Detection Strategies

Audit access to /dev/sev and correlate ioctl invocations with the process identity and parent workload
Monitor dmesg and journal output for ccp, sev, and snp subsystem errors, especially failure paths that touch MSR_VM_HSAVE_PA
Track host crash frequency on AMD SEV-capable hypervisors and compare against baseline uptime metrics

Monitoring Recommendations

Ingest kernel logs from AMD EPYC KVM hosts into a centralized logging platform and alert on SEV subsystem errors
Restrict device node permissions on /dev/sev and log any process outside the virtualization stack that opens the device
Alert on VMRUN-related general protection faults, which typically indicate host-level MSR corruption

How to Mitigate CVE-2026-64309

Immediate Actions Required

Apply the upstream Linux kernel patches referenced in the stable tree commits for the CCP driver
Restrict /dev/sev access to trusted virtualization management components only
Reboot affected KVM hosts after patching to ensure the corrected initialization path is active

Patch Information

The fix is available in the mainline and stable kernel trees. Refer to the following commits: 5a1364da2f04, 67ed191b4c8b, 74768f73854d, and 7a361c74bb12. Distribution vendors will backport these commits to their supported kernel branches.

Workarounds

Tighten permissions on /dev/sev so only the virtualization control plane can issue ioctls
Disable SEV-SNP in firmware or kernel boot parameters on hosts that do not require confidential VM workloads
Isolate confidential-computing tenants from standard KVM workloads to limit blast radius if the host crashes

bash

# Restrict /dev/sev access to the kvm group only
chown root:kvm /dev/sev
chmod 0660 /dev/sev

# Optionally disable SEV-SNP at boot if not required
# Append to kernel cmdline in /etc/default/grub:
# kvm_amd.sev=0 kvm_amd.sev_snp=0
update-grub