This Week in Spectrum, 2025-W23

This Week in Spectrum, 2025-W23

[Alyssa Ross]

[-- Attachment #1: Type: text/plain, Size: 3590 bytes --]

Once again, a lot of focus this week was on arranging future funding for
Spectrum.  Everything is submitted for now, so hopefully I'll have more
time for fun things now.  One thing I did get done was fixing a test
that I noticed had accidentally been removed from the release build a
while ago.  It had bitrotted quite a bit in the meantime, so I spent a
while gradually fixing it on increasingly more recent versions of
Spectrum until it worked again on the main branch[1], and now it's part
of the release build again[2].

[1]: https://spectrum-os.org/git/spectrum/commit/?id=88acfe0da9f9255948e3f7c83205bc9cf3d0aaf0
[2]: https://spectrum-os.org/git/spectrum/commit/?id=a4e3a407a62319b44619cc7b939560d2784dbf41

A team presented a Spectrum-based demo at the Direction
interministérielle du Numérique Hack Days[3], showing several instances
of an application running concurrently in different VMs.  They booted
Spectrum on an assortment of hardware, and ran into an issue I'd seen
before where Intel wireless devices were not fully configured.  Yureka
was able to debug this enough to determine that the iwlmvm driver was
not being automatically loaded, and that was enough for me to figure out
the underlying problem, which was that the kernel was not able to
dynamically load modules using modprobe as a usermode helper, because it
tries to run /sbin/modprobe by default and Spectrum VMs had no /sbin.
This is now fixed[4], although I don't think anybody has had a chance to
retest Intel wireless devices — I'm travelling and don't have the
hardware with me.  If anybody wants to try and report back:

 1. Boot Spectrum[5] on a system with Intel wireless hardware.
 2. In a terminal, run: vm-start sys.netvm && vm-console sys.netvm
 3. In the sys.netvm console window that opens, run: ip link

If a wireless device shows up, the bug is fixed.

[3]: https://hackdays.numerique.gouv.fr/en
[4]: https://spectrum-os.org/git/spectrum/commit/?id=96adb87dfbf00accdc1af86befb6e7be6d5e87cd
[5]: https://spectrum-os.org/doc/installation/getting-spectrum.html#installing-spectrum

Demi has been continuing working on trying to solve the x86_64
virtio-iommu interrupt remapping, by beginning a proof of concept
implementation of her draft spec[6] in Cloud Hypervisor.  She also spent
some time investigating how we might provide camera access to
applications in future, with the aim to both protect applications
against malicious hardware, and hardware against malicious applications;
and wrote it up on the list[7].  As the number of people working on
Spectrum grows, we're going to need to be doing a lot more planning in
advance for collaboration (and funding) to go smoothly, so it's great to
see this research being done.  Even if it will be a while before it's
time to focus on cameras, thinking about how it will work now helps
identify dependencies on fundamentals that we might want to get the ball
rolling on quite soon, especially those that require standards
discussions or other work with upstreams and might have a longer lead time.

[6]: https://lore.kernel.org/virtio-comment/ebeef135-1b87-4834-ad8d-5bb76554baa2@gmail.com/
[7]: https://spectrum-os.org/lists/archives/spectrum-devel/878qm6cngn.fsf@alyssa.is/

I'll be at DevConf.CZ in Brno next week, and then OW2'con[8] in Paris
and NGI Forum[9] in Brussels the following week.  Get in touch with me
on Matrix (@qyliss:fairydust.space) if we should talk at any of
those. :)

[8]: https://www.ow2con.org/view/2025/
[9]: https://ngi.eu/ngi-forum25/

Until next week!

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 227 bytes --]

Re: This Week in Spectrum, 2025-W23

[Yureka]

Hi everyone,


I can confirm that wlan0 device is now visible in the netvm! So now 
someone just needs to add some kind of wpa_supplicant/iwd in the netvm 
(I believe there is already a task dedicated to it in the NLnet grant(s) 
planned).

Besides that, I have encountered a bug on two devices (ThinkPad X270 and 
ThinkPad X390, incidentally both with i915 GPUs) where the device 
freezes on the tty at the point where weston would start, with no 
blinking cursor. It seems to happen reliably on some spectrum live 
images, but then it does reliably not happen on others without any real 
change in between except changes to the bundled AppVMs. Maybe it's a 
race condition depending on the way the image is packed? A workaround 
that Alyssa recommended and which worked was adding nomodeset to the 
command line options, but we're still working on tracking down the cause.


— Yureka

On 6/9/25 12:16, Alyssa Ross wrote:
> Once again, a lot of focus this week was on arranging future funding for
> Spectrum.  Everything is submitted for now, so hopefully I'll have more
> time for fun things now.  One thing I did get done was fixing a test
> that I noticed had accidentally been removed from the release build a
> while ago.  It had bitrotted quite a bit in the meantime, so I spent a
> while gradually fixing it on increasingly more recent versions of
> Spectrum until it worked again on the main branch[1], and now it's part
> of the release build again[2].
>
> [1]: https://spectrum-os.org/git/spectrum/commit/?id=88acfe0da9f9255948e3f7c83205bc9cf3d0aaf0
> [2]: https://spectrum-os.org/git/spectrum/commit/?id=a4e3a407a62319b44619cc7b939560d2784dbf41
>
> A team presented a Spectrum-based demo at the Direction
> interministérielle du Numérique Hack Days[3], showing several instances
> of an application running concurrently in different VMs.  They booted
> Spectrum on an assortment of hardware, and ran into an issue I'd seen
> before where Intel wireless devices were not fully configured.  Yureka
> was able to debug this enough to determine that the iwlmvm driver was
> not being automatically loaded, and that was enough for me to figure out
> the underlying problem, which was that the kernel was not able to
> dynamically load modules using modprobe as a usermode helper, because it
> tries to run /sbin/modprobe by default and Spectrum VMs had no /sbin.
> This is now fixed[4], although I don't think anybody has had a chance to
> retest Intel wireless devices — I'm travelling and don't have the
> hardware with me.  If anybody wants to try and report back:
>
>   1. Boot Spectrum[5] on a system with Intel wireless hardware.
>   2. In a terminal, run: vm-start sys.netvm && vm-console sys.netvm
>   3. In the sys.netvm console window that opens, run: ip link
>
> If a wireless device shows up, the bug is fixed.
>
> [3]: https://hackdays.numerique.gouv.fr/en
> [4]: https://spectrum-os.org/git/spectrum/commit/?id=96adb87dfbf00accdc1af86befb6e7be6d5e87cd
> [5]: https://spectrum-os.org/doc/installation/getting-spectrum.html#installing-spectrum
>
> Demi has been continuing working on trying to solve the x86_64
> virtio-iommu interrupt remapping, by beginning a proof of concept
> implementation of her draft spec[6] in Cloud Hypervisor.  She also spent
> some time investigating how we might provide camera access to
> applications in future, with the aim to both protect applications
> against malicious hardware, and hardware against malicious applications;
> and wrote it up on the list[7].  As the number of people working on
> Spectrum grows, we're going to need to be doing a lot more planning in
> advance for collaboration (and funding) to go smoothly, so it's great to
> see this research being done.  Even if it will be a while before it's
> time to focus on cameras, thinking about how it will work now helps
> identify dependencies on fundamentals that we might want to get the ball
> rolling on quite soon, especially those that require standards
> discussions or other work with upstreams and might have a longer lead time.
>
> [6]: https://lore.kernel.org/virtio-comment/ebeef135-1b87-4834-ad8d-5bb76554baa2@gmail.com/
> [7]: https://spectrum-os.org/lists/archives/spectrum-devel/878qm6cngn.fsf@alyssa.is/
>
> I'll be at DevConf.CZ in Brno next week, and then OW2'con[8] in Paris
> and NGI Forum[9] in Brussels the following week.  Get in touch with me
> on Matrix (@qyliss:fairydust.space) if we should talk at any of
> those. :)
>
> [8]: https://www.ow2con.org/view/2025/
> [9]: https://ngi.eu/ngi-forum25/
>
> Until next week!