OpenWrt users get working Broadcom Wi-Fi on aging Northstar routers
A community b43-next driver brings 2.4 and 5 GHz up on boards long stuck with closed Broadcom code, while hardware offload is still waiting for review.
Open-source wireless for a class of Broadcom Northstar routers is far enough along that owners report both bands working, and the work has been submitted for inclusion in regular OpenWrt builds.
Developer egc began with a Linksys EA6900 port and, with others, assembled b43-next, a driver kit aimed at ARM BCM4709-class boards such as the R6400. egc wrote that the drivers are still a work in progress but that his own Broadcom routers "run very well with 2.4 and 5 GHz wireless." Other boards, he warned, may need extra tweaking, with no guarantee they will work. Early reports back that caution and the progress: one user said an R6300v2 in access-point mode had both radios up; others have reported success on variants of the Asus RT-AC68U. A TP-Link Archer C9 v1 needed an ugly workaround because its NVRAM does not sit where the bcm53xx code expects it.
The practical stake is the closed-source Broadcom stack these boards have depended on. Asuswrt-Merlin has dropped older Asus targets. DD-WRT still supports them, but remains on an older kernel because of those proprietary drivers. An open driver would let the same hardware track current OpenWrt kernels. Radio support is not just a chip ID: 5 GHz parts such as the 4360 still depend on board wiring, external amplifiers, and whether setup data comes from NVRAM or SPROM. One 5 GHz failure on femctrl 3 boards turned out to be a single chip-control bit the proprietary wl driver sets on every bring-up and the open port never wrote. After a cold boot the access point looked enabled and no client heard a beacon; a warm reboot after wl had run kept the bit and restored hundreds of megabits. Clearing that bit alone took the radio off the air.
egc has also submitted Broadcom flow acceleration, the Northstar hardware offload path, for boards that support it. Without it, routing on these SoCs stays CPU-bound: testers still see a ceiling around 600 Mbit/s with the cores overloaded, far short of the wire-speed figures associated with the offload path. Neither change is in official images yet.