Executive brief
A vulnerability in the Linux kernel's Mellanox (mlx5) network driver could cause a system deadlock. This occurs during specific error recovery operations where the system's internal locking mechanisms are triggered in the wrong order. If triggered, the network interface or the entire system could become unresponsive, potentially leading to a denial of service.
Technical details
A deadlock vulnerability exists in the net/mlx5e component of the Linux kernel due to circular locking dependencies. The issue arises when work tasks triggering devlink health reporter recovery (such as mlx5e_tx_err_cqe_work) acquire the netdev instance lock before the devlink lock. This contradicts the established initialization flow where the devlink lock must be acquired before the netdev lock. An attacker or a system error triggering recovery flows (RX timeout, TX error CQE, or PTP SQ unhealthy states) can cause the system to hang. The fix involves moving netdev_trylock calls lower in the call stack into the specific recovery functions where they are required, ensuring proper lock hierarchy.
Affected products
- Linux Linux kernel mlx5e driver versions using instance locking
Timeline
- 2026-02-18: patched: Initial patch submitted by Cosmin Ratiu
- 2026-05-27: disclosed: CVE-2026-45907 published