CVE-2023-52453

What it actually does

The flaw is a missing update to saving and resuming data pointers based on file descriptor offset after PRE_COPY support was added, resulting in corrupted migration data streams for hisi_acc_vfio_pci devices.

Why it matters to the business

Live migration of virtual machines using affected HiSilicon accelerator hardware can fail with device errors on the destination, causing unplanned downtime or service disruption in environments that rely on kernel-based virtualization.

How an attack unfolds

An attacker or automated process initiates a VM live migration on a source host using the vulnerable driver; because data pointers are not adjusted for the fd offset, the migration stream delivered to the destination is corrupted, triggering SMMU and queue manager errors when the device is started.

How to fix it

Apply the upstream fix present in commit 45f80b2f230df10600e6fa1b83b28bf1c334185e and rebuild or upgrade to a kernel containing that commit; disable PRE_COPY migration as a temporary workaround if patching is not immediately possible.

How to confirm the fix worked

Confirm the running kernel contains commit 45f80b2f230df10600e6fa1b83b28bf1c334185e, then perform a test migration of a VM using the affected device and verify that no SMMU event 0x10 or hisi_zip qm_axi_rresp/qm_db_timeout errors occur.

Typical attack path

1. Identify a VM configured with a hisi_acc_vfio_pci device that supports PRE_COPY migration
2. Trigger a live migration to a destination host
3. Observe corrupted migration data due to incorrect pointer handling
4. Destination device initialization fails with memory or queue errors

How to detect it

Search kernel logs and dmesg output for the arm-smmu-v3 event 0x10 messages followed by hisi_zip qm_axi_rresp or qm_db_timeout errors that coincide with migration attempts; correlate with vfio migration sysfs or fd activity around the same timestamps.

• arm-smmu-v3.*event 0x10 received
• hisi_zip.*qm_axi_rresp [error status=0x1]
• hisi_zip.*qm_db_timeout [error status=0x400]
• qm sq doorbell timeout in function

What your security team should do

• Inventory systems running the affected Linux kernel with hisi_acc_vfio_pci devices and prioritize patching within normal change windows
• Add the fixed commit identifier to kernel build pipelines and CI tests that exercise VFIO migration
• Monitor kernel logs for the specific SMMU and hisi_zip error patterns during any planned migrations until patches are deployed