CXL HDM-DB support for Linux
summary
Linux gets upstreamable CXL HDM-DB plumbing: a kernel-side BI discovery flag lets the stack detect device-coherent Back-Invalidate capability, and QEMU receives a flit-mode extension to exercise PCIe 6 topology. Current implementation supports Type 3 regions created via cctl (device must be offline to free its BID), allows multiple coherence models per device, and shows a practical cost: converting a 2 GB region with software CL flush takes about 1 minute, whereas terabyte-scale flushes risk hours.
AI-written summary · about
chapters (7)
- 0:00Introduction to CXL HDM-DB
- 3:16CXL coherency models overview
- 10:05Back-invalidate use cases
- 16:22Linux support for HDM-DB
- 21:53Discussion on DAX and type 2 support
- 23:56Future challenges and research
- 27:12Accelerator perspective and conclusion
related talks
28mCXL Type-3 device discovery, configuration in firmware and prepare ACPI tables...
39mCompute Express Link(CXL), the Next Generation Interconnect - Overview and the Status of Linux
30mCXL Dynamic Capacity MM
28mCXL Emulation in QEMU - Progress, status and most importantly what next?
26mCXL Memory Tiering for heterogenous computing
15mCXL Type-2 core support
31mHMSDK: Enhancing Software Support for CXL Memory
43mExpanding the Memory Limits with CXL Software Stack