CVE-2023-53656

A logic error was found in the Linux kernel's HiSilicon PCIe performance monitoring driver in the CPU offline handling path. A local user can trigger this issue during CPU hotplug operations when the driver migrates performance monitoring context, potentially selecting the CPU being torn down as the migration target. This causes deadlock when the migration code attempts to use a CPU that is no longer operational, resulting in hung tasks and denial of service through system unresponsiveness.

The hisi_pcie_pmu_offline_cpu function is called during CPU hotplug teardown to migrate the perf context away from the offline CPU. However, the function uses cpumask_any_online which consults cpu_online_mask before it has been updated to exclude the teardown CPU. This creates a race where the function may select the very CPU being torn down as the migration destination. When perf_pmu_migrate_context subsequently attempts operations on this dying CPU, it encounters locks and resources that are no longer valid, causing the task to hang indefinitely in an uninterruptible sleep state.

The CVSS score(s) listed for this vulnerability do not reflect the associated product's status, and are included for informational purposes to better understand the severity of this vulnerability.