Executive brief
A vulnerability in the Linux kernel's performance monitoring subsystem for Intel processors could lead to system instability or incorrect performance data. On certain Intel server platforms, the system may fail to correctly identify hardware components when specific power-saving or memory configurations are used. This can cause the kernel to skip necessary hardware initialization, potentially leading to errors or crashes during performance analysis tasks.
Technical details
A vulnerability in the Linux kernel's perf/x86/intel/uncore driver stems from improper handling of die ID lookups in snbep_pci2phy_map_init() and spr_update_device_location(). When all CPUs associated with a UBOX device are offline, uncore_device_to_die() returns -1, causing the initialization loop to break prematurely and skip subsequent devices. Additionally, on Intel Sapphire Rapids (SPR) and Emerald Rapids (EMR) platforms with NUMA disabled, the driver fails to add PMON units to the RB tree because pcibus_to_node() returns NUMA_NO_NODE. The fix involves utilizing uncore_pcibus_to_dieid() to retrieve topology directly from the UBOX GIDNIDMAP register, ensuring correct mapping regardless of NUMA status or CPU affinity.
Affected products
- Linux Linux Kernel Intel SPR and EMR platforms
Timeline
- 2026-03-13: other: Patch authored
- 2026-04-22: patched: Patch committed to stable tree
- 2026-05-08: disclosed: CVE published