Linux 7.4 begins CXL Dynamic Capacity Device support for hot-pluggable memory

The Linux 7.4 kernel merge window has accepted the first phase of Compute Express Link (CXL) subsystem changes targeting Dynamic Capacity Devices (DCD), a Type-3 CXL memory expansion technology that permits server memory capacity to scale up or down dynamically without rebooting or physical reconfiguration. This initial code drop establishes mailbox flagging, dynamic capacity configuration reading, DSMAS data gathering for DCD partitions, and interrupt configuration, though DCD remains disabled pending completion of the full patch series under review.

The changes also permit zero-sized HDM decoders with corresponding cxl_test support, harden switch decoder enumeration, refine latency calculations to account for link width with consistent units, and rework GPF timeout programming. The remaining DCD enablement patches are being staged in smaller chunks to ease review burden. This infrastructure targets data center and cloud environments where memory pooling across racks and workload-driven capacity bursting can reduce over-provisioning.

CXL DCD infrastructure changes in Linux 7.4

• Dynamic Capacity Device groundwork: The merged code flags DCD support in the CXL mailbox, reads dynamic capacity configuration from hardware, gathers DSMAS data for DCD partitions, and configures dynamic capacity interrupts, but does not yet enable DCD operation.
• HDM decoder flexibility: Zero-sized HDM decoders are now permitted, with cxl_test framework support added for validation of these edge-case decoder configurations.
• Switch decoder hardening: A set of patches strengthens switch decoder enumeration against potential misbehavior or malformed hardware responses.
• Latency and timing accuracy: Latency calculations now incorporate link width parameters and use consistent units across the subsystem; GPF timeout programming has been reworked for improved reliability.
• Review strategy: The maintainers deliberately split the full DCD patch set to reduce review burden, with this merge window taking only the foundational infrastructure pieces.

Why it matters

CXL DCD support positions the Linux kernel to support next-generation disaggregated memory architectures in enterprise and cloud data centers, where memory pooling and dynamic allocation can substantially improve utilization rates. For distribution maintainers and kernel engineers, this establishes the API and infrastructure baseline that future CXL memory-tiering and NUMA-aware scheduling work will build upon, with full enablement expected in subsequent kernel cycles once the remaining patches complete review.