CVE-2022-50491

A context violation bug was found in the Linux kernel's CoreSight CTI (Cross Trigger Interface) driver in the hardware enable/disable functions. These functions are called from atomic context but attempt runtime PM operations that can sleep when communicating with firmware. This causes "sleeping function called from invalid context" errors and system hangs when enabling CoreSight tracing, particularly on ARM Juno platforms, resulting in denial of service.

The CTI component coordinates debugging across multiple cores. When perf enables ETM (Embedded Trace Macrocell) tracing, it builds a trace path and enables each component along that path—including CTI devices. This happens under spinlocks in the perf event scheduling code, creating an atomic context where sleeping is forbidden. The cti_enable_hw and cti_disable_hw functions were calling pm_runtime_get and pm_runtime_put to manage power. These runtime PM calls can communicate with firmware (like ARM SCMI) to control clocks and power domains. Firmware communication typically involves waiting for responses, which means sleeping. Sleeping in atomic context triggers kernel warnings and can cause complete system hangs. The issue became more visible after a revert of SCMI power domain changes made the sleep more likely. The solution: remove the PM calls entirely from these functions. The runtime PM has already been handled at a higher level when building the trace path, so these redundant calls were both wrong and unnecessary.

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.