Hacker News Reader

Comment by dark-star | Hacker News Reader

Man, FreeBSD really needs some driver love...

* No Suspend/Resume? Linux has solved this for years, I have not come across a laptop in the last 10 years that had trouble with s2r or s2d (but I guess that there might be a few that don't fully support it, I mainly tested Dell laptops)

* WiFi driver and Ethernet over USB-C both highly unstable? That's an Intel chipset, those are usually the more stable ones (at least on Linux/Windows), and the wire ethernet is not a cheapo chipset either (Realtek)

* Bluetooth not supported???

* microSD slot on USB-C docking station doesn't work? This is just SPI, how hard can that be...

I wonder if there is any laptop from the last 5 years where FreeBSD fully works. I keep trying FreeBSD on and off on my two Laptops but I had similar experiences (plus I'm not used to the FreeBSD way of doing things). I'm ready to jump through some hoops but in my case, FreeBSD was always removed after a weekend or so, and replaced with Linux where everything just works.

I used to tinker a lot to get Linux and BSD running on hardware, but these days I kinda expect most things to just work out of the box ...

I really wish FreeBSD would get some more developer backing

View on HN

I wouldn't be so sure. My Dell XPS 13 and ThinkPad X1 Carbon both had issues with suspend/resume. Close the lid @ get surprised with a completely dead battery.

Ironically, when I used FreeBSD on X1 - S3 state worked perfectly fine. The issue in post is how bhyve handles wake up from S3 state, with devices being pass-through to the VM.

Not once had a laptop that isn't a MacBook with macOS with properly working hibernation.

> * Bluetooth not supported???

Yeah, BT stack is completely broken for years. No one has desire to fix it.

> Ethernet over USB-C both highly unstable?

- I had no issues with my no-name dongle.

> I wonder if there is any laptop from the last 5 years where FreeBSD fully works.

Define fully works? Graphics - you're limited to Nvidia and whatever Linux 5.4 supported. Wi-Fi - 802.11g is as fast as you can go without wifibox.

> but these days I kinda expect most things to just work out of the box ...

Even on Linux, you either have to deal with hardware that been out for some time or know how to install the latest kernel and maybe apply some patches. When I got my XPS, it was the first one with Intel Xe - had to use "unstable" kernel in NixOS with every other distro not having graphics output.

This is still an issue in 2022: https://01.org/blogs/qwang59/2018/how-achieve-s0ix-states-li...

I haven't used much of FreeBSD in quite a while and have largely settled for the simplicity that is OpenBSD, even with all its varied drawbacks. However, I was surprised with some other things you mentioned as I thought FreeBSD would do better than OpenBSD in these spaces:

> Not once had a laptop that isn't a MacBook with macOS with properly working hibernation.

OpenBSD can pull this off, though the process is not very performant.

> ...802.11g is as fast as you can go without wifibox.

There are a few different Wi-Fi card options that will support 11ac on OpenBSD. Yes, it's not the full 11ac implementation, but some is better than none.

> Graphics - you're limited to Nvidia and whatever Linux 5.4 supported.

There's full Xorg GPU acceleration on AMD cards with drm-kmod[1]

> Wi-Fi - 802.11g is as fast as you can go without wifibox.

Some 5GHz 80211n cards will work. There's no support for ac/ax but at least you're not limited to overcrowded 2.4GHz.

[1] https://github.com/freebsd/drm-kmod

Yes, and look at the version of that `drm-kmod` 5.10 and 5.4. Take a look which cards are supported on those versions. The only reason I stopped running FreeBSD on desktop is that my 6900 XT is still not supported.

On Windows the first thing I always do is turn off hibernation. It made a bigger difference back in the day since the swap was on a spinning disk but it still has a little performance improvement and nets you a bit more ram. With machines booting up so fast these days I personally avoid anything related to hibernate / sleep.

True, but I never turn off my Mac unless I'm installing updates. A few times a year, I have an issue from resume (from Suspend to RAM) on a desktop - GPU won't wake up.

When I had an Nvidia card, this issue would happen every other day, so I had to disable sleep and hibernate. Because it was a very shitty way to start your day.

Hibernate is the alternative to sleep if you don't want to overheat your laptop and kill your battery when you transport it in your carrying case (the "sleep" of Windows 10 will do that as it's made for tablets).

It's true that some hardware is not Linux compatible, but if you get stuff that is (like the Framework 12th gen I'm typing this on), it does in fact just work out of the box. Wifi, sound, GPU, fingerprint reader, suspend, the works.

I had a laptop on the Ubuntu Certified hardware list (https://ubuntu.com/certified) some HP model.

Suspend still didn't work. Is there a list I have not found?

Were you running Ubuntu and the specific Ubuntu Release + Device model combo they outline on the website? Anything outside of that would be outside of the scope of that list.

Much better resource is reading various wikis and user forums, I found.

Was actually also debating recently between the thinkpad and the dell - ended up based on cost with the XPS 15 - any other general tips or big gotchas running linux on it?

Not really, everything[1] works out of the box on new distros. I had to switch NixOS to Fedora, so I can run daemons required for compliance - at the time of F34 (or was it F35?) release thing worked well.

Sometimes you need to enable legacy/linux ACPI state in BIOS. I think it maps S0ix to S3 or something like that.

There is (was?) an issue with suspend on some intels and I had to tweak the behavior otherwise the laptop would keep going after you close the lid.

I think both are solid laptops for linux.

[1]: Everything I used and tried

> No Suspend/Resume? Linux has solved this for years, I have not come across a laptop in the last 10 years that had trouble with s2r or s2d (but I guess that there might be a few that don't fully support it, I mainly tested Dell laptops)

I believe the article author calls this "Sleep" and got it to work well. It's hibernation to disk that's not working.

WiFi and Bluetooth seem indeed to be issues but I don't really need them on my desktop. On a laptop it would be a dealbreaker indeed.

I have that same intel wifi nic. On Linux, it also caused crashes on resume from suspend to disk (worked OK on resume from S3 suspend to ram). I added a hook script, that on suspend to disk, downs the interface, if it was up, and unloads the kernel module before suspend. It reloads the module on resume, and if interface was up before suspend, it brings the interface back up. Now resume from suspend to disk is reliable.

Not sure if the hook script is still necessary (kernel version has updated a couple times since), but it doesn't cause any issues, so I haven't taken the time to check.

Not exactly "worked out of the box". Easy fix, but the same fix is available on *bsd too.

> * No Suspend/Resume? Linux has solved this for years, I have not come across a laptop in the last 10 years that had trouble with s2r or s2d (but I guess that there might be a few that don't fully support it, I mainly tested Dell laptops) I have an older Thinkpad that I swtiched back to Windows after getting fed up with suspend and hibernate on linux. It had like a 20% chance to suspend or hibernate, other times it did nothing. Screen went black and after 2-3s I was back at the login screen. Shutdown was reliable so when in a hurry I needed to do a full shutdown.

> and the wire ethernet is not a cheapo chipset either (Realtek)

Buying Realtek does not guarantee getting good hardware, the RTL8153 is infamous [0], and I remember there were reports about Realtek chips flooding the network with corrupt packets if they were powered on via the charger but not connected to a laptop.

[0] https://overengineer.dev/blog/2021/04/25/usb-c-hub-madness.h...

Interesting, I wasn't aware of that. I have used Realtek chips in various computers over the last 15 years or so and they were all quite stable, that's why I jumped to that conclusion

Ethernet is likely a Realtek chipset (on the docking station). Intel does not do Ethernet over USB chipsets.

The Ethernet expansion card uses a Realtek 2.5Gb/s controller with a USB-C connection to the laptop.

This is a fairly new controller that may not have a FreeBSD driver yet.

if FreeBSD had a copyleft license, maybe companies would have a reason to publish their drivers, or you could port drivers from Linux.

That’s probably true but it cuts both ways too.

I don’t know if the lack of copy left license is the problem as much as the momentum.

The various Unix and BSD wars (that were predicated on a much different licensing challenge) and splintering of the communities has meant there was never a big way for developer or even manufacturer interest to really focus.

OpenBSD has a very different philosophy related to drivers than FreeBSD and for very understandable reasons, most of the BSDs don’t want to use or touch GPL code.

The copyleft nature of the GPL definitely helped Linux, but I don’t think it’s the only reason Linux was successful.

But Linux got its shit together first and was able to benefit from that. Apple and to a lesser-extent, Solaris, show that you can still have broader hardware support and drivers without a copyleft license.

I do think that if companies could easily port their Linux drivers, you’d see that happen. That’s absolutely a consequence of the license choice (and it’s a consequence from either license decision). But if there is a market for the BSDs in a particular area, I do think you see companies making drivers.

The reality is, the market for people running FreeBSD on laptops is pretty small and probably not worth it for a lot of WiFi chipset makers. But if you had a specific industry where FreeBSD was widely deployed, I do think you’d have broader hardware support for things important to that industry, if nothing else.

From what I heard, at least some companies that use FreeBSD in their products do give back and upstream at least bits of their code (like Sony, NetApp, ...). Probably mostly improvements to existing kernel subsystems, maybe not drivers, if the hardware isn't freely available... but they still participate in development. At the end of the day, it is in their interest as it lessens the maintenance burden of forward-porting everything to every new kernel version.

Correct! And that’s sort of my point. But I don’t think a lot of those companies care about wifi drivers working on laptops, as an example.

It is absolutely in their interest to help. But if people expect companies to just write a driver for X operating system, that’s not realistic. And if you could just port the Linux drivers, sure, that would be easier, but it wouldn’t change the fact that device makers still have the be able to prioritize where to put their resources. And laptop FreeBSD users is probably really low on most lists.

netflix definitely gives back to the freebsd project. They run their appliance that delivers netflix content on FreeBSD-current, the development branch

Well, it's not that simple.

FreeBSD has some contribution from businesses like Netflix and Nginx and those are very welcome, but they have nothing to do with desktop usage. Those have deadlines and budgets, features get implemented and tested. That's why FreeBSD has superb network stack, good NUMA support, files from disk to nic go brrr. That's why the move to OpenZFS was so fast.

Then, they're contributions without company backing it. This is where desktop feature live. That's why FreeBSD's Wi-Fi stack is stuck on 802.11g - no company depends on this. GPU drivers stuck at Linux 5.4 level - no company cares about it[1]. Those get discussed to death - there is no deadline, no money.

Note, that solution to graphics card driver is to use Linux's drivers, so they don't shy about GPL. However, FreeBSD folks LOVE to bikeshed and polish things. This is why PR to port linux 5.11 drm was open when 5.11 was already EoL for an entire year. Instead of "just" porting 5.11, they decided to refactor how things are done. There are like 3 people working on that PR. It still not merged and not even in review yet.

[1]: Bless Nvidia for having FreeBSD driver available. Would be nice to have CUDA too...

Seriously - more than anything I feel like FreeBSD is a server OS first and foremost. Or more realistically something with lower flexibility that makes it ideal for more niche use cases as a pseudo "lower level" OS. The way I see it the two things you want MOST reliable are storages and network - and all the other features of most operating systems are just unnecessary fluff for those.

What I would like to see however is maybe some specific wifi chip or USB/thunderbolt Ethernet adapter that is certified to work well with FreeBSD - for testing purposes and as a mobile network tool it would be useful to have a laptop running FreeBSD in a VM that could communicate over that dedicated piece of hardware. This would make it super convenient to test a bunch of physical hardware or do all kinds of other infrastructure work while remote.

Well PlayStation 4 and 5 are on FreeBSD. Modified of course.

That's really irrelevant for multiple reasons: they don't contribute back, and their graphics libraries are irrelevant to your Wayland and X11.

> But Linux got its shit together first and was able to benefit from that.

Linux is not a company or person. It's the people behind Linux who got it together, but sadly BSD folks have left behind.

> Apple and to a lesser-extent, Solaris, show that you can still have broader hardware support and drivers without a copyleft license.

You're comparing Apple (a company, and a FAANG of them no less) with community-driven projects. Apples and oranges, almost literally.

This is a question of pure ignorance at this layer, but is or why isn't there a standard for drivers that can follow a similar way of building an interface? Or maybe its not even possible due to entirely different purposes of the device.

What company do you think is writing FreeBSD drivers for the framework - or any laptop - and not sharing them?

If FreeBSD was encumbered with copyleft then half the contributors would leave, and the sectors in which it is widely used would seek an alternative.

Pretty sure any “port” of Linux drivers would be in effect a substantial rewrite, for the dissimilar kernel structures and coding standards.

I am glad FreeBSD exists. Not every commercial venture has the luxury to release their source code, even if they wanted to. If you're writing software for banks or defense, the source needs to be closed.

FreeBSD enables these companies to exist and while consumer grade drivers are not available, many things in FreeBSD are rock solid. Thanks to commercial contributions.

I disagree. Work on permissively licensed software typically has the net effect of decreasing global freedom by reducing the costs of corporations to produce proprietary software. I try to avoid contributing to permissively-licensed software when possible, and I prefer to co-operate with businesses that use free software to avoid vendor lock-in as much as possible.

Why should we all do free work for corporations? At least with the GPL you can negotiate a commercial license in exchange for money, but with a permissive license you forfeit your ability to do that.

This article accurately sums up my point of view: https://lukesmith.xyz/articles/why-i-use-the-gpl-and-not-cuc...

Also, the GPL does not require you to share the source code for programs that you do not distribute outside your company.

It is a fact that many companies from startups to corporations do not have the flexibility of working in open source.

There is nothing to disagree here.

You're shoving in GPL ideology where none was asked for. No one is asking to do free work for corporations; this is a flimsy straw man. You're welcome to do what you feel, I was stating the fact that FreeBSD has done the world a lot of good. That's hard to refute even in the worst of assumptions. And that's not saying anything about GPL. It has its place.

Folks that mindfully contribute BSD- and MIT-licensed software do not share the perspective that they are performing unpaid labor for corporations.

I personally regard copyleft as a severe curtailing of options and equate it with being forced to adopt a narrow, ideological worldview. Worse, the GPL perpetuates the fundamental atrocity of intellectual property law, by hacking it, and therefore intrinsically depending on it; in contrast, the BSD and MIT licenses are mostly a liability waiver under an assertion of attribution, ultimately creations more of necessity than ideology.

In the context of abjuring exploitative corporations, I hear an echo of trade unionism, a movement often oblivious to its own reinforcement of capitalism, again by depending upon the thing it purports to confront; few entities are more aligned with the moral sentiments of Adam Smith than a trade union engaged in collective bargaining. And so it is with the copyleft licenses and their implicit support for enforceable intellectual property rights.

The article linked appears to be derogatory, emotive, riddled with false equivalences, and relies on invective as a substitute for rhetoric. Rather than finding it compelling, it think it describes a myopic, narcissistic view of both intellectual advancement and the value of unfettered contribution.

Corporations that don’t want to share code don’t share it even when copyleft.

From the opinion article I linked:

>At the end of the day, though, what does it matter? What is a totally unenforced GPL? It's just a Cuck License—Isn't that what license cucks want? So why should they care? At their very best, BSD and MIT licenses are only what GPL might be at its very worst.

>The GPL is a permanent liability for any company that crosses it. Some companies might be so bold to lift GPL code and hide it, but there is always a risk and a worry that prevents its general violation.

Those comments are unrelated. The point they were making is that if companies don't want to release source, they will either use a permissively licensed open source project as the base, or they will roll their own (probably lower quality) proprietary version. Or they'll pull an nvidia and release some hacky shim/workaround that breaks all the time and leads to bad user experience.

> they will either use a permissively licensed open source project as the base

which will be less feasible if people don't contribute to permissive projects in the first place, and basically serve as free labor for corporations to take advantage of.

> or they will roll their own (probably lower quality) proprietary version

this will cost them time and money to produce, putting them at a disadvantage compared to their competitors.

>Or they'll pull an nvidia and release some hacky shim/workaround that breaks all the time and leads to bad user experience.

nvidia can pull that off with linux because pushover operating systems like BSD give them the leverage to make those sorts of demands of the linux ecosystem. If linux sues nvidia or makes the driver/shim incompatible that's fine by nvidia- they will be mostly content with to tell their userbase to switch over to BSD so they can distribute a closed-source driver. They can do this to the extent that BSD is a workable replacement for linux, so all work on BSD is really just leverage against the linux GPL.

If there were no feasible permissive competitors to linux, then linux developers would be quite free to make demands of companies who would be forced to comply with the GPL or write their own OS from scratch.

> nvidia can pull that off with linux because pushover operating systems like BSD give them the leverage to make those sorts of demands of the linux ecosystem

That is quite the assertion given that FreeBSD is much smaller than linux, which is already much smaller than windows. Honestly kind of laughable, FreeBSD is so small and insignificant, it's just an afterthought of an afterthought for companies, the only reason those drivers exist is because it's a minimal amount of extra work for them since they already have the linux drivers.

It would take a massive amount of work and nurturing to create a FreeBSD community large enough for it to become a palatable option for most linux users. Not to mention FreeBSD doesn't even have CUDA support, so nvidia themselves would have to do extra work to make it happen (probably other features too, that's just the main one that bit me). As for shims, linux already breaks them all the time, that's why everyone hates nvidia on linux. But that doesn't stop nvidia from doing it.

> If there were no feasible permissive competitors to linux, then linux developers would be quite free to make demands of companies who would be forced to comply with the GPL or write their own OS from scratch.

nvidia is already complying with the GPL. And no, if they wanted to skip shims they wouldn't be forced to write their own OS, they would just force their users to use windows. This comparison makes no sense.

For my hobby server of 1000+ days [1] uptime with ram failing. A raid controller on the brink. I've amazed at the solidarity of FreeBSD.

[1] https://i.redd.it/2omcrsz8ut691.png

* i am aware how in todays time you shouldn't, updates, and blah blah. But not required.

Are you saying FreeBSD is not robust? Did you confirm that it is the operating system that is leaking memory and not your application? If that's true, that's huge and you should file a report. Better yet, write a blog post about the details and how to recreate it.

I read that as "the hardware is failing but FreeBSD is still trucking right along"

My bad, I misunderstood as sarcasm.

no, this shows incomplete understanding of layering in a modern operating system. Closed-by-default software projects, over time, are less secure; common source code components make a stronger system over time when executed well; secrets can be handled in a secure manner in security-sensitive applications; well factored software using modern layers are more secure over time with common, tested software components.

Yeah, because Linux’s copyleft is what forced nVidia to publish decent open source drivers 15 years ago.

I would say that linux isn't able to strong-arm nvidia into complying because nvidia can just go back to distributing their closed-source driver on permissively-licensed OSes such as BSD.

I have a hard time making the parent comment even making logical sense.

So the reason we are not seeing say open CUDA drivers from NVIDIA for Linux is that they instead distribute closed ones for permissively licensed operating systems such as FreeBSD, OpenBSD, Illumos, MINIX, Plan9, etc.? Which (apart from all lacking CUDA drivers) all universally have less support from hardware manufacturers both when it comes to open and closed drivers across the board? ERRNOLOGIC.

You are aware that FreeBSD got closed NVIDIA graphics driver support about a decade after Linux? It is also the only operating system with a permissive license that I am aware of NVIDIA ever supporting.

Alternatively, are you trying to refer to that some game consoles used FreeBSD as a basis and had proprietary NVIDIA drivers written for them? You are aware that there is nothing stopping them from doing exactly the same thing with Linux?

You are free to believe in “licensing magic” being a solution as much as you want, but I think that you will find that it all has much more to do with what NVIDIA is ultimately all about: Shipping as many cards as is humanly possible while maintaining a borderline monopoly maximising profits.

> FreeBSD got closed NVIDIA graphics driver support about a decade after Linux

Ahem, https://cgit.freebsd.org/ports/commit/?id=94fc89fbfbc7cd8886...

> some game consoles used FreeBSD as a basis and had proprietary NVIDIA drivers written for them

You mean proprietary AMD drivers? (Only PlayStation 4 and 5 are based on FreeBSD.)

Thank you for correcting me. Nothing to add in terms of AMD. But I could have sworn that there was a set of “demands” in terms of features from NVIDIA around the time I used FreeBSD on my desktop (~2010) in order to produce binary drivers back then. Perhaps I am misremembering?

That would be the amd64 driver port: https://wiki.freebsd.org/NvidiaFeatureRequests. For comparison, the first x86_64 driver version for Linux seems to be 1.0-4499 (https://download.nvidia.com/XFree86/Linux-x86-64/1.0-4499/) from 2003-09-02.

Aren't most Linux drivers closed source binary blobs?

No? The vast majority of hardware has OSS drivers (and they're usually even the better choice!) - the big exception here for average machines is NVIDIA.

AMD cards depend on non-free firmware.

And both Intel and AMD processors use non-free microcode. Firmware and drivers are distinct units, as only one of them runs on the CPU.

Not on libre distros. I mean, not more than loading the burned one.

The nvidia drivers sure are. I don't know about most drivers though. As for wifi, I own this ath9k card which I'm told is completely free, but I also own this intel 3160 chip that apparently requires some firmware to be loaded.

See https://en.wikipedia.org/wiki/Comparison_of_open-source_wire...

Funny, a decade ago the story was "atheros is evil" and Intel was the Linux darling. How things change..

No. You're thinking firmware.

The driver layer (code around firmware that initializes the hardware and marshals/unmarshals comms to/from the device).tends to be GPL lest it taint the kernel. The firmware on the other hand, is the last unconquered vestige of the Open Source landscape. Logic locking, and the prevalence/nature of fabless semiconductor IP tends to make that endeavor incredibly difficult.

"most" and "drivers" are things that need specific usages to be true, but if you drop them and restate, you're absolutely right.

It is really hard to have a complete personal computer on Linux or any other OS that doesn't require closed source / proprietary binary blobs.

Not at all, my Intel NUC except for the tiny wireless dongle totally runs with libre drivers.

Firmware blob free?

Yep, with Linux libre, I tried it.

Now I run OpenBSD, and I could purge /etc/firmware with no issues if I wanted.

> This is just SPI

It isn't "just SPI". That is a fallback protocol for the full SD interface.

SPI is the other interface. Both are legit and official.

Reply on news.ycombinator.com