Widely ported, sure. [LWN.net]

Widely ported, sure.

Posted Dec 30, 2005 22:10 UTC (Fri) by ncm (subscriber, #165)
Parent article: NetBSD 3.0

There's no question but that NetBSD is widely ported. But, is it still more widely ported than Linux? 57 systems is a lot, but how many distinct CPUs is that? How many of those can still be bought, even on Ebay? More practically, how many of them have ever actually been booted under NetBSD 3, not just built? I gather that, by all the most trivial measures, Linux has taken that crown, for all it's worth.

It's not my desire to start a flame war. Rather, I'd like to see NetBSD promoted for something more meaningful than the total number of porting projects ever undertaken. Surely there's plenty else for NetBSD developers to be proud of, such as UVM memory management that, with just a bit of care in coding, brings real zero-copy semantics to UNIX read() and write(), and much more cleanly and generally than sendfile(). (Yes, OpenBSD adopted NetBSD's UVM.) When last I heard, IPv6 code was being proven in NetBSD before being imported to Linux. What else?

I'd love to see an analog of UVM in Linux.

(Log in to post comments)

Widely ported, sure.

Posted Dec 30, 2005 22:44 UTC (Fri) by busterb (subscriber, #560) [Link]

I do think that it is possible to accurately know how many systems Linux or NetBSD has been ported to because a lot of work never gets put into the mainline, regardless of license.

Support for many embedded systems comes in the form of an out-of-tree patch, either from a vendor's board support package or some internal branch. Many of these will never make it into the mainline kernel due to lack of interest, low code quality, or both.

Wasabi Systems for instance probably has support for a lot of hardware not supported by NetBSD proper. MontaVista makes a lot of its work public, but not all; otherwise, why buy MontaVista Linux? Neither the GPL nor BSD licenses require the source to be distributed back to the originator, nor do they require any effort to get patches rolled into the mainline. I could easily take the kernel patches from several vendor-supplied board support packages that I am working with and post them on LKM (the're GPL-code derivative), but I really have no motivation to get the drivers and such up to proper Linux standards nor to support them. If I distribute a product with this code, I only have to supply the code to customers.

Widely ported, sure.

Posted Dec 30, 2005 23:57 UTC (Fri) by ncm (subscriber, #165) [Link]

That just means that out-of-tree ports Don't Count. If they're not maintained upstream, they're forks, instead.

Widely ported, sure.

Posted Dec 31, 2005 8:57 UTC (Sat) by danieldk (subscriber, #27876) [Link]

That is a ongoing discussion. But portability in the sense NetBSD uses does not mean that there is a maintained or undermaintained patchset somewhere that adds support for some platform. It means that on all supported platforms userland is almost identical, and that the system can be built for any platform on any platform with just one command.

We are trying to get rid of the 'oh, use NetBSD when you have an old obscure platform' mantra. NetBSD is very portable, but has many other merits. To name a few:

The cryptographic disk driver, a pseudo-device driver that can encrypt blocks on the way to a disk or filesystem.
Veriexec, a kernel subsystem that permits execution of binaries based on the correctness of the hash of the binary.
systrace, a program that can enforce system call policies for applications.
Good Xen support, with ready-to run domain0 and domainU sysinst images.
Binary emulations for many operating systems. Most notably Linux binary support on i386, which can be used to run Sun JDK, Opera, Acrobat Reader, Mathematica, and Matlab, to name a few.
Enforcement of non-exec bits for the stack and heap on many platforms. This is used by default since NetBSD 2.0.
The excellent pkgsrc package system, that is also ported to other operating systems.

Widely ported, sure.

Posted Dec 31, 2005 9:31 UTC (Sat) by ncm (subscriber, #165) [Link]

I recall, too, a topologically-sorted init process, in place of the baroque sysvinit scheme (with numbered symlinks in /etc/rc?.d) used in most Linux distros.

It seems worth mentioning that there's a Debian port with the NetBSD kernel underpinning a GNU userland, called Debian GNU/kNetBSD. This is no aberration; Debian has been intended to be kernel-agnostic from the beginning, originally to leave room for the Hurd. You can choose between exim and postfix to move your mail, between evolution and mutt to read it, between firefox and galeon to surf, and between Linux and the NetBSD kernel to operate the hardware.

Widely ported, sure.

Posted Dec 31, 2005 9:52 UTC (Sat) by tzafrir (subscriber, #11501) [Link]

Actually, topology-sorted init has been in SuSE since around 8.0, I believe. It is supposed to be part of the LSB since 1.3 but has remained widely unimplemented on other distroes (except, maybe, Gentoo).

Widely ported, sure.

Posted Dec 31, 2005 11:53 UTC (Sat) by khim (subscriber, #9252) [Link]

Gentoo does have topology-sorted init. I like it. But this kinda shows what's wrong with "*BSD vs Linux" comparision. Once you are comparing "*BSD kernel" and "Linux kernel" - comparision is more-or-less fair. But when you are trying to compare "*BSD" and "GNU/Linux" (and use just "Linux" to confuse everyone) - it becomes complicated. For every obscure feature found in *BSD you can find equally obscure GNU/Linux distribution where feature was "implemented for ages"! Not exactly fair comparision...

Widely ported, sure.

Posted Jan 1, 2006 21:11 UTC (Sun) by samb (subscriber, #32949) [Link]

Was Debian GNU/kNetBSD ever functional? It was basically a one-man show, and I believe it's defunct now.

Apples, oranges and BSDs

Posted Jan 2, 2006 15:30 UTC (Mon) by eru (subscriber, #2753) [Link]

Muddying the comparison about porting is additionally at least one more thing: When NetBSD (or any of the other BSDs) is considered ported, you don't get just a kernel. You also get the basic userland sufficient to run at least traditional text mode unix sessions. In other words you get what in Linux terms would be a minimalistic but usable distro. In Linux you still have to work on the userland after getting the kernel running. What would the comparison look like if this is taken into account, and we looked at comparable ported configurations? (I have no idea, just asking)

Apples, oranges and BSDs

Posted Jan 6, 2006 5:16 UTC (Fri) by roelofs (subscriber, #2599) [Link]

In Linux you still have to work on the userland after getting the kernel running.

Actually, once the kernel is there--which implicitly means GCC is--userspace stuff is relatively trivial. X is about the only "standard" thing that comes to mind as potentially difficult, and that only because it has its own drivers. If your kernel port includes a framebuffer and USB (at least USB keyboard/mouse), even X is pretty straightforward. The biggest hiccups actually come when you try to cross-compile things, but that has nothing to do with any given kernel, and it can be mitigated by getting a native filesystem and toolchain set up early.

Greg