Linux plumbing of CXL error reporting
summary
Explains how CXL memory errors map to existing Linux mechanisms and what is actually missing: it distinguishes protocol vs component errors and shows protocol errors use PCI ER while component errors require a mailbox readout; it contrasts firmware-first vs OS-first flows and notes RCD (restricted) mode needs new PCI/RAS support; kernel patches for mailbox/interrupt handling are proposed, and practical choices (extend PCI, add event driver, or push parsing to userland) determine what gets built and who must act.
AI-written summary · about
chapters (13)
- 0:00Introduction and overview
- 1:10CXL system design basics
- 3:14Impact on RAS and error detection
- 5:24Firmware-first vs OS-first modes
- 7:14Error types and reporting locations
- 11:03Protocol and component errors
- 14:40VH mode vs restricted CXL mode
- 17:15Restricted CXL device mode details
- 19:33Missing pieces in RCD mode
- 21:21Extending existing subsystems
- 22:23User space considerations
- 23:29Summary and future work
- 24:33Q&A and discussion
related talks
39mCompute Express Link(CXL), the Next Generation Interconnect - Overview and the Status of Linux
24mAdding RAS Support for CXL Port Devices - Terry Bowman
29mCXL HDM-DB support for Linux
15mCXL Type-2 core support
26mCXL Memory Tiering for heterogenous computing
43mExpanding the Memory Limits with CXL Software Stack
29mPlumbing challenges in Dynamic capacity device
25mShared CXL 3 memory what will be required