How Qualcomm is rearchitecting Wi-Fi for the Linux kernel — turning a generation-specific driver into a living platform for all future Qualcomm wireless silicon.

The Problem

A History Lesson Written in C

Open any recent Linux kernel tree and navigate to drivers/net/wireless/ath/. What you find is a generational timeline: ath9k for 802.11n, ath10k for Wi-Fi 5, ath11k for Wi-Fi 6, ath12k for Wi-Fi 7 — each a complete, self-contained driver, each solving the same problems its predecessor already solved.

This is the silent tax of generational hardware refreshes. ath10k and ath11k share an estimated 80% of their functional logic — identical WMI messaging model, identical HTC/HTT framing, identical firmware recovery philosophy — but share exactly zero lines of code. Every bug fix, every security patch, every protocol corner case must be manually evaluated and ported. The ath12k unified driver model is the attempt to break this cycle.

The Three-Tier Portability Model

Before any architectural decision, you need precision about what "shared" actually means — not all driver subsystems are equally portable across hardware generations.

Key Insight

The unified driver model is not simply a code-sharing exercise. It is a precise encoding of the three-tier portability classification into the module boundary itself. ath12k.ko owns Tier 1 entirely and the portable half of Tier 2. ath12k_wifi7.ko owns the hardware-specific half of Tier 2 and all of Tier 3.

Architecture

Two Modules, One Platform

The core change is a split of the monolithic ath12k.ko into two distinct kernel modules. The split precisely follows the Tier 1/2/3 portability classification. ath12k.ko owns everything hardware-generation-agnostic: the mac80211 interface, firmware messaging stack, shared data path, and core init sequences. ath12k_wifi7.ko owns everything tied to Wi-Fi 7 silicon: HAL register offsets, Copy Engine pipe assignments, MHI channel setup, Wi-Fi 7 firmware capability TLV negotiation.

Directory Structure

The module boundary is reflected directly in the filesystem layout. A new wifi7/ subdirectory is introduced, containing all device-specific code for the Wi-Fi 7 hardware generation.

Data Path & Monitor Mode

Where the Boundary Gets Subtle

Packet movement — enqueuing descriptors into TX rings, draining RX rings, managing buffer pools — follows the same structural pattern across hardware generations and lives in the shared dp.c, dp_rx.c, and dp_tx.c. What varies between generations is the interpretation layer: RX status TLVs, monitor mode frame annotation, and CE descriptor parsing are all hardware-format-specific.

Implementation

13 Patches, Sequenced with Intent

The implementation was delivered as 13 commits to the ath12k-ng branch. The commit ordering is deliberate: establish logical separation first, then draw the module boundary, then harden it. Commits 01–09 validate the separation within a still-compilable monolith. Commit 10 draws the actual module boundary. Commits 11–13 enforce it by eliminating cross-dependencies.

Future Extensibility

A Platform, Not a Driver

ath12k.ko is not a Wi-Fi 7 driver with shared code extracted into a library. It is a platform designed to host multiple device-specific modules. The extensibility model has three identified target classes and one implied long-term convergence path.

The most strategically significant implication is the potential to unify ath11k and ath12k under the same common infrastructure via a future ath12k_wifi6.ko. If realised, ath11k.ko could eventually be deprecated — collapsing two independently maintained drivers into one.

Upstream Path & Conclusion

What Needs to Be True for Mainline

The ath12k-ng branch designation signals this is a staging ground, not yet in Linus's tree. The phased delivery suggests a rollout strategy that aligns feature parity with the proprietary SDK before pushing to mainline. Three critical gates determine upstream acceptance.

The ath12k unified driver model is one of the more architecturally coherent proposals to emerge from the Qualcomm wireless driver team in recent times. It takes a real problem — the generational code duplication that has plagued the ath driver family since ath9k — and addresses it with a structural solution proportionate to the problem: 476 net lines of change, 13 commits, a clean module boundary, an explicit extensibility model, and a sequencing strategy that validates the separation before drawing the boundary.

The immediate value is reduced maintenance overhead for Wi-Fi 7 hardware. The longer-term value is a living driver platform that absorbs new Qualcomm Wi-Fi silicon by implementing only what is genuinely new — the HAL register layout, the CE configuration, the MHI channel setup — while inheriting a decade's worth of hardened firmware interface and 802.11 state machine code from the common layer.

That is the difference between a driver and a platform.