Skip to content

Commit fa99571

Browse files
chadmedConan-Kudo
chadmed
authored and
Conan-Kudo
committed
blog: add Linux 6.14 progress report
Signed-off-by: James Calligeros <[email protected]>
1 parent 3ddbcb6 commit fa99571

File tree

1 file changed

+252
-0
lines changed

1 file changed

+252
-0
lines changed

‎content/blog/2025/03/21-progress-report-6-14.md

Lines changed: 252 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,252 @@
1+
+++
2+
date = "2025-03-21T09:00:00+10:00"
3+
draft = false
4+
title = "Progress Report: Linux 6.14"
5+
slug = "progress-report-6-14"
6+
author = "James Calligeros"
7+
+++
8+
9+
As March draws to a close and Linux 6.14 nears release, now is a good time to
10+
provide you all with our first major progress update since [taking the lead on
11+
the project](https://asahilinux.org/2025/02/passing-the-torch/). Going forward,
12+
we hope to keep these updates in sync with upstream kernel releases. We feel
13+
that this is a natural cadence given the focus on upstreaming, with enough time
14+
between posts for noteworthy downstream changes to accumulate.
15+
16+
After getting through all the administrative work required to keep the lights
17+
on after marcan's departure, we've hit the ground running with upstream patch
18+
submission. We held our first board meeting under interesting circumstances,
19+
and we've even managed to sneak a couple of new features in downstream
20+
while you weren't looking. So, without further ado, let's get into it.
21+
22+
## Our first board meeting
23+
We held our inaugural board meeting on 7 March, under challenging conditions
24+
for some. Neal and Davide battled crappy conference hall WiFi while at Southern
25+
California Linux Expo, while I sat in the dark with no power as Tropical Cyclone
26+
Alfred ravaged Southeast Queensland and northern New South Wales. Despite this,
27+
we managed to get through quite a bit. You can find the meeting minutes
28+
[here](https://asahilinux.org/docs/project/board/minutes/20250307).
29+
30+
## Lightening the load
31+
After four years of development, we have racked up a considerable patch
32+
set. Here are some quick-fire stats:
33+
34+
- 90,000+ lines of code added to the Linux kernel as of 6.13, across 1250 patches
35+
- A downstream U-Boot with a number of Apple Silicon and USB stack fixes
36+
- A downstream Mesa - required since our GPU uAPI is not upstream yet
37+
- A downstream virglrenderer - required since our GPU uAPI is not upstream yet
38+
- A downstream Flatpak runtime extension - required since we have a downstream Mesa
39+
40+
This is... less than ideal. Rebasing, testing, and releasing
41+
our forks is a massive drain on our time. This is especially true
42+
of Janne, who has the Sisyphean task of keeping our kernel patch set applying
43+
upstream and porting it to every stable point release. That is hours of work
44+
on an almost weekly basis, especially as the pace of Rust abstraction upstreaming
45+
increases. We can't keep doing this.
46+
47+
We want to bring you M3 and M4 support. We want to bring you Thunderbolt. We want
48+
to bring you DisplayPort Alt Mode. We want to bring you Variable Refresh Rate, and
49+
HDR, and hardware accelerated video playback, and better power management for the
50+
Pro/Max/Ultra Macs. To do that, however, we must start reducing the amount of patches
51+
we're carrying downstream. Most of what we're carrying is stable and has been
52+
for years. So after a month of tireless effort, how are we doing?
53+
54+
We have submitted three new drivers upstream - the Image Signal Processor (ISP)
55+
driver, which is necessary for webcam support, and drivers for the Touchbar's
56+
display controller and input digitiser. These three drivers total to almost
57+
8,000 lines of code, with the ISP driver being 6,000 of those. The good news is
58+
that both Touchbar drivers have already been accepted! Thanks to chaos_princess
59+
for taking on the responsibility of preparing and submitting all three.
60+
61+
Alyssa and Janne have been hard at work tidying up the GPU driver to prepare it for
62+
submission. This has involved some changes to the uAPI, which should slightly
63+
improve performance for some workloads.
64+
65+
Now that Rust for Linux abstractions are starting to be merged at a healthy
66+
pace, we are faced with an emerging challenge. It is rare for
67+
any kernel patch to survive the mailing list without at least a couple of
68+
non-trivial changes, and Rust abstractions are no exception. Every time an
69+
abstraction used by our driver is merged, we must drop our downstream version
70+
and rebase the driver atop the version accepted upstream. This is gruelling, menial,
71+
and unpleasant work, and Janne has our deepest gratitude for volunteering his
72+
time to get through it.
73+
74+
We have also been working to clean up and upstream other parts of the kernel.
75+
In addition to a number of miscellaneous fixes and changes for drivers already
76+
upstreamed such as the NVMe and I<sup>2</sup>C controllers, we have also submitted
77+
changes to the upstream Texas Instrumentes TAS2764 and TAS2770 speaker amplifier
78+
drivers which extend them to support the Apple-specific variants found in
79+
Apple Silicon Macs. Through this process, we found that the ASoC maintainers
80+
had already been cherry-picking some commits from our development branches!
81+
82+
We're far from done, but we are committed to getting through this vital work.
83+
More patches are being submitted all the time, so watch this space!
84+
85+
## Is this thing on?
86+
By the time you're reading this, we will have enabled microphone support on
87+
most laptops! This has been a long time coming, with Martin Povišer, Eileen
88+
Yoon, and chaos_princess having reverse engineered and developed a driver
89+
for the Always-On Processor (AOP) in Rust quite some time ago. This is Apple
90+
though. Nothing is ever simple.
91+
92+
Before even getting to play with the mics in Linux, we hit a hurdle.
93+
On certain machines, the Secure Enclave is in the path of the physical mic
94+
data lines. If SEP is unhappy, you don't get mic access. chaos_princess
95+
quickly reverse engineered the SEP endpoint responsible, and wrote a
96+
stub driver to simply toggle the hardware mic switch on. With data now
97+
coming in to ALSA on all machines, we can continue.
98+
99+
In all laptops released so far, Apple use three Pulse Density Modulation
100+
(PDM) mics wired up to an ADC and decimator in the AOP. All three mics are
101+
plumbed directly to userspace on separate channels, with no preamplification.
102+
Having three mics enables Apple to use beamforming, which is a signal processing
103+
technique to greatly enhance the directivity and sensitivity of an array of sensors.
104+
Originally developed for RADAR and military communications, it's now mostly known
105+
for being a marketing dot point for WiFi access points. It's also really, really,
106+
really hard.
107+
108+
Unfortunately, PDM mics are very omnidirectional and very sensitive. We cannot
109+
get by without some kind of beamforming. I initially tried
110+
a very basic delay-and-sum beamformer, which involves no advanced signal
111+
processing. However since the mics are only about 2cm apart and only sample at
112+
48 kHz, we cannot achieve the temporal resolution required to make this
113+
approach work. Signal processing it is, then.
114+
115+
The mysteries of signal processing in the acoustic domain are truly understood by
116+
very few. After consulting the scant available literature on the matter and my high
117+
school maths textbooks, I thought I had absolutely no hope of doing a good job of this.
118+
I put out a call for help on Mastodon, however no one offered to step up. Fine, I'll
119+
do it myself.
120+
121+
After reading a few more papers, I found an approach that looked familiar to me.
122+
A Minimum Variance Distortionless Response beamformer works by taking your complex
123+
sample data in the time domain and using optimisation techniques weighted in a
124+
particular direction relative to a fixed point (usually some mic in the array).
125+
At some point, it clicked that this is really just statistics!
126+
127+
After wrestling with Rust's immature scientific computing libraries for a few weeks,
128+
and even more reading of my high school maths and university statistics textbooks,
129+
the result was [Triforce](https://crates.io/crates/triforce-lv2) - an MVDR beamformer
130+
for the mics found in Apple laptops.
131+
132+
Thanks to the groundwork laid in PipeWire and WirePlumber for speaker support,
133+
wiring up a DSP chain including Triforce for the microphones was really simple.
134+
We just had to update the config files, and let WirePlumber figure out the rest!
135+
136+
Did it work? It's Rust, so it worked on the first try! When properly dialled in,
137+
the mic array is sensitive to signals coming from a "typical" seating position in
138+
front of the laptop, and mostly rejects noise from any other direction. Try it
139+
yourself - play some music out of your phone and move it around your laptop. It
140+
should become suppressed when moved out to the sides or behind the machine.
141+
142+
Once again, a huge thanks to Martin, Eileen, and chaos_princess for their incredible
143+
work reversing AOP and SEP.
144+
145+
## Fedora Asahi Remix 42 Beta
146+
Thanks to Davide's work validating builds, we are pleased to announce that Fedora
147+
Asahi Remix 42 is [now available](https://fedoramagazine.org/announcing-fedora-asahi-remix-42-beta/)
148+
as a beta! The final release is expected in about a month, bringing us closer than
149+
ever to releasing at the same time as Fedora. We hope that in the next cycle
150+
Fedora Asahi Remix 43 will release on the same day as Fedora Linux 43.
151+
152+
## Asahi Linux @ SCaLE
153+
While at [SCaLE](https://www.socallinuxexpo.org/scale/22x), Neal and Davide managed
154+
to set up a demonstration of Asahi Linux running on an M2 Mac mini and an M1 Mac Studio
155+
in the expo hall. Reception was extremely positive, and folks had a blast playing a whole
156+
bunch of Steam games through muvm and FEX! We even managed to get kids involved by firing
157+
up Nidhogg, which proved to be the most popular game by far.
158+
159+
## WoAoA
160+
Some astute users have noticed that ARM64 Windows VMs now work with KVM on
161+
Asahi! This is thanks to Oliver Upton, who submitted a rather large series of
162+
patches upstream to enable Arm PMUv3 emulation on Apple SoCs. WoA requires
163+
PMUv3, and until now we could not emulate it. We quietly cherry-picked those
164+
patches, and they are now incorporated into our latest kernels. Interested
165+
parties can try Windows on ARM on Asahi today!
166+
167+
## What does the dxdiag say about its Feature Level?
168+
It's 12_0! Alyssa recently added sparse binding support to our Vulkan driver,
169+
which was the last piece of the puzzle for Direct3D Feature Level 12_0 via vkd3d-proton.
170+
Many more D3D12 games now run with the latest Mesa and FEX rootfs.
171+
Our Vulkan driver is not yet as mature as our OpenGL driver though, so performance
172+
and playability vary... but optimisations are coming soon.
173+
174+
## Magic pairing
175+
Apple Silicon Macs have a neat feature whereby Bluetooth devices remain
176+
synchronised across all macOS installs, even recoveryOS. Information required
177+
to pair devices is stored in the machine's NVRAM. macOS checks the NVRAM
178+
for devices that it doesn't know about, and copies that data. Since
179+
they see the same Bluetooth controller in the Mac regardless of
180+
which macOS install is booted, devices will happily connect automatically without
181+
having to go through the pairing process again. This is really useful for
182+
ensuring that Bluetooth keyboards and mice work seamlessly across all environments,
183+
especially in recoveryOS.
184+
185+
To replicate this in Linux, Janne developed [asahi-btsync](https://crates.io/crates/asahi-btsync).
186+
The tool reads the Bluetooth device information out of NVRAM and injects it into
187+
BlueZ's configuration. Up until a few weeks ago this was a manual process which
188+
required running asahi-btsync and connecting to the discovered devices manually.
189+
This is impractical for input devices which are required to be working for these
190+
actions.
191+
192+
As of version 2.4, asahi-btsync integrates with systemd and D-Bus to
193+
fully automate the Bluetooth device synchronisation process. Starting with Fedora
194+
Asahi Remix 42, all Bluetooth LE devices paired with macOS will be available from
195+
your first boot - including the initial Fedora Asahi Remix setup!
196+
197+
Do note however that this does not apply to Bluetooth 5.0 devices, which use a
198+
new authentication and encryption scheme.
199+
200+
## OpenCollective and GitHub Sponsors
201+
When we stood up our [OpenCollective](https://opencollective.com/asahilinux),
202+
none of us really knew what to expect. We had hoped to maybe scrape together
203+
enough support to keep the domains and CDN going, but we dared not hope any harder.
204+
205+
The sheer volume of support and the speed at which it flowed
206+
in left us floored and humbled beyond measure. Your faith in our work means more
207+
than can be expressed in words, and we seriously cannot thank you all enough.
208+
209+
The financial support provided via OpenCollective allows us to continue our work
210+
with confidence. Not only do we have the cash we need to keep the project's
211+
existing infrastructure online, but we have the resources we have always wanted
212+
to ensure the project's viability long into the future.
213+
214+
Until now, we have been mostly purchasing machines out of our own pockets. Given
215+
the nature of what we do, this can mean spending upwards of $10,000 a year on
216+
Macs. Not only is this unsustainable for us on an individual level, it
217+
also presents an enormous barrier to entry for talented folks who may not have
218+
the means to buy such expensive computers. The financial support provided by
219+
OpenCollective backers enables us to signficantly reduce that financial
220+
burden. This has already been invaluable for development work.
221+
222+
The M1 MacBook Air is our most popular Mac by far, with almost 14,000 installations
223+
and counting. This represents almost 30% of the total install base. We couldn't
224+
possibly release microphone support without supporting it. Being able to simply
225+
buy the machine and get the work done without worrying about food for the next
226+
two weeks is the only reason we were able to ship mics in a timely manner.
227+
228+
Buying hardware is not the only game-changing benefit afforded to us by your
229+
financial support. We can now actually build and maintain the CI farm we have
230+
sorely needed for years, which will accelerate development. We can afford email
231+
hosting and other collaborative infrastructure that greatly reduces friction.
232+
We can go to conferences and events without fear of going into financial stress.
233+
All of this is necessary to ensure the long-term sustainability of Asahi as a
234+
project, and none of it would be possible without you.
235+
236+
For those who would prefer to support us via GitHub Sponsors, it's in the works.
237+
We have put in the request with GitHub to link our Sponsors account to our OpenCollective,
238+
however this has not yet been approved. Stay tuned for more.
239+
240+
## What's next?
241+
While this month has been a whirlwind, literally in my case, we've started
242+
settling down into a rhythm that will work nicely in the near-term. Our primary
243+
focus is to continue submitting kernel patches upstream. Reducing our maintenance
244+
burden is the only way we will be able to get back to the fun stuff sustainably.
245+
246+
Neal and Davide will be [presenting](https://events.experiences.redhat.com/widget/redhat/sum25/SessionCatalog2025/session/1731519631980001Xort) at [Red Hat Summit](https://www.redhat.com/en/summit) on 19 May,
247+
demonstrating Fedora Asahi Remix and the upcoming CentOS Hyperscale Asahi Remix
248+
to showcase how Asahi has made it easier than ever for developers to start bringing
249+
their workloads to ARM64 Linux.
250+
251+
We hope to have even more good news for you when Linux 6.15 releases, so stay tuned
252+
for more updates! Once again, we sincerely thank everyone who has supported us so far.

0 commit comments

Comments
 (0)