Asahi Linux 7.3: M3 without EXPERT, macaudio and AVD drivers head upstream

⚡ The gist in 5 seconds

• Asahi Linux dropped the EXPERT flag for M3 support — it's now a regular install, no warnings.
• DCP is now enabled on M3: sleep works, the battery lasts longer, and the area around the notch on the MacBook has come alive.
• The macaudio (~3,100 lines) and AVD (~13,000 lines) drivers have gone out for review in the mainline kernel; the first Devicetrees for M4 Pro/Max and M5 have appeared — they boot up to a TTY.

🔍 What happened

The Asahi Linux team published another progress report — this time covering Linux kernel 7.3. The main news for owners of recent Macs: M3 support is out of EXPERT status. The flag no longer needs to be set in the installer.

DCP has also been enabled on M3. In practice, that means three things: it's faster, the battery lasts longer, and sleep works. Plus, the area around the notch on the MacBook is now usable — it used to be a dead rectangle.

Now, about upstream. This cycle, two major drivers went out for review: macaudio (about 3,100 lines including Devicetree) and AVD (about 13,000 lines). AVD spent less than three months in Asahi's downstream tree — the fastest turnaround for a large SoC block in the project's history.

Audio turned out to be the hardest part. The Apple Silicon audio subsystem required notable changes to the kernel's ALSA, and the speaker and headphone amplifiers are variants of already-supported chips, but with serious rework.

The first Devicetrees for M4 Pro, M4 Max, and M5 have appeared (thanks to Yureka). They're bare-bones: they bring the SoC up to a TTY, and that's it. The next step is expanding them.

The team has also started moving to an "upstream-first" approach: changes to TI USB-PD, SPMI, and initial Thunderbolt support went straight into the mainline kernel.

Meanwhile, m1n1 learned to virtualize macOS with SPTM — on both M4 and M5. On M5 the task is harder: it requires Exclaves (isolated apps managed by the Secure Kernel), and XNU talks to them via SPTM and the Tightbeam protocol. Without working Exclaves, the kernel panics. Sven put the Secure Kernel and Exclaves userspace into a separate VM — and m1n1 now runs three VMs in EL1. The authors joke: "maybe we should start selling m1n1 to companies fleeing VMware."

The GPU remains. Apple's GPU follows a tick-tock cadence, and the M3 chip differs significantly from the M1 and M2. A new group of contributors is doing the reverse engineering; at XDC, Alyssa Millburn, Anna Maniscalco, and TellowKrinkle gave an update on how things are going.

🧠 Our take

Dropping EXPERT isn't about a checkbox in the installer — it's about the project believing in itself. "Almost no problems from early testers" is a rare statement in the hardware world. From experience, people say that when there really are few bugs, not when they're embarrassed to show them.

The line about AVD and 13,000 lines in three months is the most interesting part of the report. Usually a major driver lives in a fork for years, accumulates hacks, and then takes months to get into mainline. Here, the cycle has shrunk. Our assessment: this is the first real sign that Asahi is ceasing to be a "patch project" and becoming part of the normal kernel process.

Who wins? M3 users — right away. M4 and M5 users — only in theory for now: a TTY in the log is a starting point, not a working laptop. Does anyone lose? Perhaps only those used to the idea that "Linux will never run on new Macs." That idea seems to be aging faster than the drivers.

⚠️ What we don't know

There are no numbers at all. How many benchmarks the M3 GPU passes, what performance gain DCP brings, how many hours the battery lasts — the report contains not a single measurement. "Almost no problems" is also without numbers: it's unknown how many people actually tested the M3. There's no GA timeline for M4 and M5, and no list of what still doesn't work. Verify things yourself: installing it on a machine you need for work is not a great idea.

Prediction: by the end of the quarter, Asahi's report will include the first public numbers for the M3 GPU — at least one benchmark. If reverse engineering stalls, the next major release won't be the GPU but expanded Devicetree support for M4 Pro and M5.

Source: asahilinux.org