One of the patch series going through nearly a dozen revisions now is for introducing a cold reset recovery method for the Direct Rendering Manager (DRM) drivers with an initial focus on the Intel Xe kernel graphics driver.
The cold reset recovery is for handling of errors that require a complete device power cycle in order to recover. For select hardware errors, a cold reset may be necessary to completely clear the state and allow for the device to recover without going through a complete system reboot / power cycle.
The main motivation currently for Intel engineers working on this is for Power Management Unit “PUNIT” errors that can persist across warm resets and driver reloads and thus requiring a complete cold reset to recover.
To the common DRM code this patch work introduces DRM_WEDGE_RECOVERY_COLD_RESET as a new DRM wedging recovery method. The DRM core in turn notifies user-space via a uevent that a cold reset is necessary and from there with udev it’s possible to take necessary action depending upon the system.
Those interested in this topic of cold reset recovery for kernel graphics drivers with initial infrastructure for the Xe driver can find the patches on dri-devel.