Re: [PATCH 0/7] wifi: brcmfmac: AWDL for AirDrop on Apple T2 Macs

[Arend van Spriel <arend.vanspriel@broadcom.com>]

On 04/11/2026 16:15, Andre Eikmeyer wrote:
> This series adds AWDL support to brcmfmac, so Linux on the Apple T2 Macs
> can send and receive files over AirDrop with Macs, iPhones and iPads.
> 
> AirDrop runs over Apple Wireless Direct Link (AWDL). On these Macs the
> Apple firmware runs AWDL itself. The host creates an AWDL interface and
> sets its parameters. From then on the firmware keeps the AWDL schedule,
> sends the AWDL frames it gets from the host and carries data frames to
> and from the peers. Election, peer discovery and AirDrop run in user
> space.
> 
> The driver enables AWDL on any firmware that announces "awdl" in its
> capabilities and supports interface_create version 3. The linux-firmware
> image for the BCM4350 (7.35.180.133) on a pre-T2 MacBookPro14,1
> announces "awdl" and answers interface_create with BCME_UNSUPPORTED, so
> it stays without the feature. Other Apple chips, such as the BCM4355,
> are untested.
> 
> The KaiT2en community, about 280 members, runs this series with its
> kernels. Users reported it working with the BCM4364 and BCM4377 on the
> MacBookPro15,1, 15,2, 15,3, 15,4, 16,1, 16,4, MacBookAir9,1 and
> iMac20,1, and nobody reported a regression. AirDrop transfers on the
> MacBookPro16,1 and 16,4 are slower than on the other models, the cause
> is still open.
> 
>    1    match firmware capabilities as whole words
>    2    move the interface_create definitions to fwil_types.h
>    3    detect AWDL support of the firmware
>    4    create and remove the AWDL interface
>    5    add the AWDL data path
>    6    add vendor commands for AWDL
>    7    pass AWDL frames and role changes to user space
> 
> Patch 1 is a fix on its own and carries a Fixes tag. It is here because
> patch 3 adds an entry next to the lines it changes.
> 
> The AWDL interface is a network device "awdl%d" outside cfg80211. User
> space talks to the driver through typed vendor commands and events of
> the Broadcom OUI: start, stop, frames, sync, peer add and remove, and
> events for received frames and role changes. Creating the interface
> needs RTNL and vendor commands run with the wiphy lock only, so
> AWDL_START and AWDL_STOP run in a work item and report their result as
> vendor events. The driver sets the firmware parameters itself.
> 
> The firmware interface was worked out by tracing. A dtrace script
> recorded the iovar calls macOS 15.4.1 made to the firmware on a
> MacBookAir9,1 (BCM4377) while AirDrop was opened and closed in Finder.
> The other way round, the MacBookAir9,1 was traced as AWDL peer while a
> MacBookPro15,1 (BCM4364) brought AWDL up under Linux, together with the
> AWDL frames macOS received and the console and events of the BCM4364
> firmware. The BCM4364 handling in patch 6 comes from these runs. The
> iovars, their order, their values and the AWDL frames come from these
> traces and from the frames the firmware passes up.
> 
> The user space side is t2-drop, part of KaiT2en-Fedora and published
> under the GPL:
> 
>    https://github.com/kaiT2en/KaiT2en-Fedora/tree/main/apps/t2-drop
> 
> With it, AirDrop works in both directions.
> 
> Comments on the vendor command interface are welcome in particular.
> nl80211 has no AWDL interface type, and the AWDL schedule and frame
> exchange run in the firmware here.

I really have no idea how to deal with this series. AWDL is not part of 
the IEEE 802.11 so I can imagine the kernel wireless subsystem does not 
care to support this. From driver perspective I am not keen on 
supporting this upstream either. My limited understanding of AWDL was 
that it was Apple's variant of TDLS (802.11z).

So what does AWDL provide to user-space and could the same be 
accomplished by TDLS? From the explanation above it seems AWDL provides 
more knobs to user-space, but not sure whether the awdl%d interface is 
merely a control interface or actually passing data through it (patch 5 
implies it does). If my memory does not fail me then adding vendor 
commands is generally pushed back.

Regards,
Arend