Intel Proposes Using Netlink For Linux Graphics Driver RAS & Telemetry

Intel Linux kernel graphics driver developers are looking at making use of Netlink for exposing RAS (reliability, availability, serviceability) and telemetry features of kernel graphics drivers to user-space for their modern GPUs.

Netlink is a generic means of transferring data between the kernel and user-space processes by way of a sockets-based interface. Netlink supports a variety of services already from network routing/firewall/IPSec to SELinux notifications and crypto. The proposal out of Intel is to leverage Netlink for RAS and telemetry communication in place of their existing PMU counter and sysfs interface exposure they use internally. The hope too is that other Direct Rendering Manager (DRM) drivers will make use of this Netlink interface for RAS/telemetry functionality.

Making use of Netlink to be able to subscribe and be automatically notified of new hardware errors should be more efficient than user-space continually polling via sysfs or DebugFS files to check on any counters.

Intel has been working on this Netlink RAS/telemetry support as part of their work on the new Xe DRM kernel driver. Their request for comments on this new functionality can be found via the dri-devel mailing list.