Linux kernel crypto/ecc: missing carry in 128-bit accumulation corrupts ECC arithmetic at a boundary

Impact

The carry check in the kernel's ECC big-integer multiply only compares the high words and misses the case where they are equal and the low-word addition also carries, so results are wrong at a saturated-high-word boundary. crypto/ecc.c is the shared arithmetic behind the kernel's ECDH and ECDSA implementations, so this is a correctness defect in a primitive, not a memory-safety bug: the record shows no attacker-controlled memory corruption and no demonstrated exploit path. Practical exposure depends entirely on which consumer is compiled in and reachable on a headless server - the historical caller of this code is Bluetooth LE pairing, which datacenter hosts do not run, while the ECDSA side is used for signature verification. Treat this as a correctness fix to pick up on the normal cadence, not as a reason to open a window tonight.

Who can reach it

The kernel CNA scored it AV:A/PR:N, which points at an adjacent-network pairing path rather than anything a tenant can reach from a GPU pod. The record does not name a reachable path on a headless server, and states only that the carry condition is mishandled.

What to do

Pick up a stable kernel containing the five fix commits in the record at your next scheduled node reboot. No mitigation is needed or offered in the record, and no fixed release number is stated - map the commits through your distro's advisory. Nothing here justifies an out-of-band drain of GPU nodes.

References

Related entries

• Linux kernel QAT: use-after-free tearing down SR-IOV while VF2PF response work is in flightCVE-2026-64438 · Linux kernel Intel QAT driver (VF2PF work teardown in adf_disable_sriov)High
• Linux kernel (drivers/vfio/pci): If vfio-pci device registration fails after the device joined the VGA arbiter, theCVE-2026-64475 · Linux kernel (drivers/vfio/pci)High
• Linux kernel (arch/x86/kvm/mmu): If reclaiming shadow pages invalidates the root a fault is being serviced against, KVMCVE-2026-64561 · Linux kernel (arch/x86/kvm/mmu)High
• Linux kernel (arch/x86/kvm/vmx): Nested teardown freed the shadow VMCS page while vmcs01 still referenced it, andCVE-2026-64562 · Linux kernel (arch/x86/kvm/vmx)High
• Linux kernel (drivers/iommu/amd): Iommu_completion_wait() returned without waiting whenever another CPU had alreadyCVE-2026-68329 · Linux kernel (drivers/iommu/amd)High
• Linux kernel (net/xfrm): The rtnetlink changelink path for xfrm interfaces checked CAP_NET_ADMIN only against theCVE-2026-72136 · Linux kernel (net/xfrm)High