FreeBSD Is Looking To Drop Many Of Its 10/100 Ethernet Drivers

[Mifune]

Solaris meanwhile has only Tribblix and OpenIndiana

Your ignorance on Illumos is disarming. SmartOS, Nexentastor, Delphix, OmniOSce and XStreamOS all fare really well at an enterprise level. Just to let you know, SmartOS and Delphix represent major contributors to OpenZFS project
Then there's more niche, yet active projects, like Unleashed, v9os and DilOS

[Mifune]

That got me thinking that perhaps FreeBSD will one day be considered the same, "largely irrelevant".

I see you drew some very rational conclusion here...or is rather just some sort of fallacious syllogism?

While that observation may be accurate, how many of those embedded devices can utilize the full Gigabit Ethernet bandwidth? Go ahead, do the research, I'll wait.

A lot of embedded devices do not make full use of Gigabit Ethernet speeds. Even USB-attached (USB3) Gigabit Ethernet devices get less than 500 megabits per second of speed in the real world, not those almost impossible and unrealistic lab tests that are used to advertise fantastic speeds.

Which translated means:

'Embedded devices do no make full use of Gigabit Ethernet speeds, so we must keep legacy 10/100 drivers to support legacy hardware which nowadays is by no means shipped with those same embedded SoCs, (instead of drivers supporting the 100/1000 NICs which those embedded systems are actually equipped with)'

I don't know enough about FreeBSD to comment with any accuracy as to it's internal issues, but everything I read suggests the following:

Which pretty much says it all about the noteworthiness of your observations

[aht0]

There is a reason FreeNAS guides (FreeBSD) recommend Intel NICs only.

Not only FreeNAS guides. You'd get identical recommendation also from pfSense forum and probably from OPNSense guys (haven't browsed that forum so often though)
My first pfSense box years a go used 3 RTL8139's and man, did it really bog the CPU down..

Thanks to those who helped with "decrypting" the driver names, I saw that xl corresponds to the 3COM PCI 10/100 NICs (3C905) and these were common in desktops too - in companies, institutional, not so much consumers who all got the Realtek stuff. So if you're an hobbyist who doesn't need/want to buy a NIC instead of using one from a junk pile, look for these pretty 3COM Etherlink III with yellow lines painted on green PCB.

My pleasure. Frankly, haven't seen RTL8139 in newer than Socket775 (included) machines. Be aware that RTL8139C+ and newer (like commonly integrated-into-motherboards RTL8168/8169) have in fact different driver in FreeBSD (re - https://www.freebsd.org/cgi/man.cgi?re(4)) and you would need to have really old hardware in order to need rl. RTL8169 is not specifically marked in manual but it works with re (my PC)
And that removal is planned few years down the road, it's not going to happen now.

[Gromit]

The discussion here is full of misunderstanding but OTOH the article does it best to add to this and muddies instead of clarifies.

...and for clarification, NO I don't work on FreeBSD. I work developing applications that run on Linux but I DO use both FreeBSD and Linux

First, this is first and foremost a call to action. A call to make the drivers on the list work with iflib. Those that fail to be fixed will first be marked for deprecation and, if still not fixed when it's time to release FreeBSD 13, be removed from the official builds. Likely it will still be possible to make some of them compile and work with FreeBSD 13. However, this allows the network stack plummers able to make important work without having to worry about also fixing drivers not updated to work with iflib. This is about drivers and NOT about 10/100Base support in itself which will remain OFC.

What they gain is "less drag" on the author of the various drivers.

No, the gain is for network stack developers

That got me thinking that perhaps FreeBSD will one day be considered the same, "largely irrelevant".

Right, Sony (PS3, PS4), Juniper (JunOS), Netflix, FreeNAS/TrueOS makes the future of FreeBSD uncertain... or do you think PS5 will run a Linux kernel!?

There there is this comment from the initial mailing list entry:

While that observation may be accurate, how many of those embedded devices can utilize the full Gigabit Ethernet bandwidth? Go ahead, do the research, I'll wait.

And yet, that is not what this is about in ANY way. This is dropping support for drivers not updated to use iflib and ONLY supporting 10/100Base not at what speed they support on the wire.

A lot of embedded devices do not make full use of Gigabit Ethernet speeds. Even USB-attached (USB3) Gigabit Ethernet devices get less than 500 megabits per second of speed in the real world, not those almost impossible and unrealistic lab tests that are used to advertise fantastic speeds.

Again, this will not in ANY way change this or be affected by this. The mailing post even EXPLICITLY states that USB NIC drivers are NOT in this round.

I don't know enough about FreeBSD to comment with any accuracy as to it's internal issues, but everything I read suggests the following:

FreeBSD internals are not sufficiently modular or contain adequate intermediate layers to permit major code changes without causing major breakage/rewrites elsewhere across the codebase.

Quite wrong. Again, this would not even be an issue if the drivers were rewritten to use iflib.

FreeBSD has an openly admitted (in that initial email in the mailing list) migration path (port to 'iflib') that they chose not to follow because it's too much work to port every 10/100 driver in the FreeBSD library. It almost sounds like that porting effort is an "either/or" proposition; we either have to do them all or we can't do any of them at all. So what is the real reason behind this attitude/approach? Bad initial coding? Bad initial design? Who is responsible for this current mess? Name names!

Wait what? From the POST:

The criteria for exception are: ... - Ported to use iflib (reducing future maintenance cost.)

In other words, if someone (i.e. you, me, anyone) port the driver to use iflib then it will NOT be removed from builds of FreeBSD 13

The author/maintainer of the driver is getting bored doing their work and/or simply does not see the value in these drivers that other see.

Right, you clearly have no clue as to how development work on big open source projects work. It's NOT an obligation to fix others code. You never see stuff removed from Linux? Riiiight.

FreeBSD might have a maintainer staff issue, like not enough of them that want/need to do the proper and required work to maintain their portion of the overall product.

Do you think someone get paid to do that? Your way is NOT how you get people, mostly unpaid volunteers, to do the work

FreeBSD claims they are eliminating no longer used or non-working drivers. I agree with eliminating "non-working" drivers, but only after an investigation as to what efforts were made to correct those issues. Did the maintainer simply give up? Did vendor support disappear? What's the reason Kenneth?

If they are not updated to use iflib they will be non-working in FreeBSD 13. This is about identifying those issues BEFORE it happens. You think they have vendor support? What vendor will pay someone to fix drivers not being sold for quite some time i.e. not generating any doe on the other end? If you need to get a clue, just look at how Android updates work for 3 year old phones.

Then there is the question of "no longer used " drivers. How was that conclusion arrived at? What data does FreeBSD have to back up it's claims? Perhaps those drivers simply work, have bugs that rarely impact anyone, and users simply use them without complaining about them. If that's the case then why not simply freeze the development of that driver and keep it in the library?

Because it will not work without support for iflib support. Also, from the post:

- 5 reports of uses in the wild on machines running FreeBSD 12 will be deemed satisfy the "popular" requirement.

If people using said drivers on FreeBSD 12 can't be bothers to do this then I deem it quite unused.

While I have used FreeBSD in the past to a limited extent I never dug into it's details. I kept finding more and more information about Linux compared to FreeBSD, and that was back in the days when you could find books in a bookstore when online resources were incredibly immature (compared to now). Now I strictly use Linux as my *nix of choice.

I honestly don't mean this as an insult but you are clearly worse for it. One of the main reasons I use FreeBSD is that it keeps me honest and not make assumptions about how things are supposed to work. I've been using both FreeBSD and Linux for more than 20 years and I like them both for different reasons. If Linux devs were to do a similar move then I would fully understand it as well. It does not in any way prevent existing software from working. And

If FreeBSD maintainers were to move forward with this divestiture of 10/100 drivers, then I wish a pox on their houses and their decendants for all eternity to come.

...and yet you do not use FreeBSD?! Wow, just, wow....
Suggestion is to change your diet to something less toxic