Weekly Edition
Return to the Press pageSponsored link
Serve your customers, not your servers, with VERIO Linux VPS. Full-access test-drive here.
| Weekly edition | Kernel | Security | Distributions | Search |
| Archives | Calendar | Subscribe | Write for LWN | LWN.net FAQ |
LinuxBIOS ready to go mainstream (Linux.com)
(Log in to post comments)
Er, not so much
Posted Dec 7, 2006 18:14 UTC (Thu) by bos (subscriber, #6154) [Link]
If you haven't dealt with BIOS issues before (lucky you!), beware of this article. There's a degree of sunshine and happiness in it that I would deem a little, er, excessive.
In my experience, the LinuxBIOS tree is broken the majority of the time for most notionally supported motherboards. The source has a history of sweeping changes that are only lightly tested, which I posit doesn't help with stability. Cut-and-paste code abounds. There's never been any attempt to converge on a stable release. And like any BIOS, it's just about impossible to debug; usually the best you can hope for is winking a two-digit number on a POST card (if you can get far enough to actually talk to the PCI bus at all).
Don't get me wrong: all PC-class BIOSes are horrible beyond belief. LinuxBIOS is better than the average in that it's (a) open and (b) mostly written in C. The developers are helpful and responsive, two words you'll never see associated with proprietary BIOS teams. But the project is a long way from being usable by anyone without a lot of specialised knowledge and a high tolerance for pain.
Er, not so much
Posted Dec 7, 2006 18:50 UTC (Thu) by AJWM (subscriber, #15888) [Link]
> usually the best you can hope for is winking a two-digit number on a POST card
<yorkshire_accent> We used to _dream_ of having winking two-digit numbers on POST card...</yorkshire_accent>
Seriously, yes, bit-banging the hardware before there's a decent OS loaded is always difficult. At least back in the old days we had toggle switches ;-) I'm glad to see manufacturer interest; even though I'm not likely to want to mess with the BIOS code, it's nice to know that I can if I want.
Uhhh, actually "yes, very much"
Posted Dec 7, 2006 19:16 UTC (Thu) by horen (subscriber, #2514) [Link]
Unfortunately, when I tried, several years ago, I found it impossible to implement LinuxBIOS on my ASUS CUV4X motherboard; sadly, it was/is not among the supported ones.
HOWEVER, the following link lists vendors and their products which do run LinuxBIOS. I'd say that's pretty darn "mainstream", wouldn't you?
Er, not so much
Posted Dec 8, 2006 5:42 UTC (Fri) by drag (subscriber, #31333) [Link]
That code messiness is one of the things that are being worked on apparently.
Google is now providing funding and resources for testing code. If AMD takes a active interest then it would be good also.
Especially when you throw virtualization into the mix then you start to have something interesting.
Say it's like this..
Imagine you buy your 'server' motherboard and it has Xen/Linux or other hypervisor kernel combo for the bios along with a 32MB flash for simple linux install.
You get the motherboard, throw on some harddrives and stick it into the case. No video card, no keyboard, no monitor, no serial ports, no nothing.
Plug the case into the power strip, plug in the network cable. Walk over to your desk and pop open your laptop or cell phone or whatever.
Open the web browser to the motherboard's http interface. Update the firmware image to the latest version. Kexec into it.
Partition (or maybe LVM) and format the harddrives. Maybe some software raid, or Open-iSCSI. who knows. Pull down your operating system image, or multiple images, and install them on your machine, then boot them up.
Various Open-source bios emulation stuff (such as is used in Qemu) and legacy I/O (vnc, serial ports, keyboard, mouse, parrellel ports) is emulated and accessable via ssh, or vpn, or whatever. Something.
Your finished.
Doesn't matter if the OS being installed is Linux or Windows or Dos or SCO or anything like that.
To me if I was Asus or Tyan or other motherboad maker I would be VERY interested in this technology. I know most of the above is possible nowadays, it may not be that easy to perfect, but think of the marketing possiblities!
For the cost of some testing and making sure your Linux is very compatable with your hardware you can now advertise your hardware has:
built-in hypervisor..
raid 0/1/5/10/etc capabilities. Even works when you abratrially add new PCIe sata/scsi adapters.
some logical volume management.
built-in iSCSI initiator that is transparent to the installed operating systems.
advanced web-based configuration system
advanced hardware monitoring and notification system.
built-in ability to image volumes, restore systems from backups, etc etc
and all sorts of other possibilities.
No extra licensing costs. No need to license a bios or anything like that. All of it built into the motherboard, all it costs is compatability with Linux and the small on-board flash drive. A big possiblity for increase in value with very little increase in cost.
All of it starts off, of course, with LinuxBIOS. :-)
In fact I don't know why more servers aren't shipped with built-in embedded Linux for system management possiblities.
Er, not so much
Posted Dec 8, 2006 15:40 UTC (Fri) by horen (subscriber, #2514) [Link]
First of all, you've written an excellent and concretely thought-provoking post. Thank you.
Secondly, I'm embarrassed to admit that although I've heard of most of the items you mentioned (i.e., iScsi and PCIe), I'm not at all familiar with them.
Despite being a Linux/Unix sysadmin for 18 years, I've come to realize that even though I devote a good 1-2 hours/day reading articles in LWN and other online sources, the lion's-share of knowledge about what's out there in our line-of-work remains "context-driven" -- if you're not working in it, by-and-large, you're not going to be conversant in it.
Thanks for the "heads-up".
Er, not so much
Posted Dec 9, 2006 19:48 UTC (Sat) by danshearer (guest, #18686) [Link]
In terms of futures rather than the (fair enough) present you describe, seehttp://linuxbios.org/pipermail/linuxbios/2006-September/0... .
Booting a BIOS in a full-system simulator is a great way to avoid bit-bashing, in fact a truly good
simulator not only implements the hardware sufficiently well that there is no functional
difference [1] but the debugging facilities are sufficently advanced that the development
environment is better than most visual IDEs for high level software development. Alas there is no
such OSS simulator, QEMU is the very credible best we have but once you've worked with Simics
[2] there's no going back..
--
Dan Shearer
dan@shearer.org
[1] absent bugs in both the simulator and the hardware which differ from the spec, and other
issues of this class. If hasn't booted on production silicon it doesn't work :-)
[2] from Virtutech, former employee here and it's as good as I'm claiming. Develop your BIOS in a
virtual world containing target, the server the target talks to, simulated network equipment and a
full range of peripherals for the target -- all to the same virtual clock and all reversible.
Not quite.
Posted Dec 11, 2006 7:56 UTC (Mon) by jd (subscriber, #26381) [Link]
It's still not simple to install, the web pages aren't very accurately maintained, the exact level of support can be hard to find, the claim that it is a "replacement BIOS" is meaningless (Linux doesn't use BIOS calls, so you are not replacing the BIOS as much as the bootstrap), it's not included in any major distro that I know of, and documentation can be hard to find.On the flip-side, if it's good enough for Cray, the U.N., Lawrence-Livermore Laboratories and the NSA, it is likely good enough for your average Joe User. (Last I heard, Joe Average User has a much lower reliability and quality expectation than supercomputer labs, spy agencies and nuke facilities.) The primary developers do not appear to have acquired concrete boots or any fetishes for oxygen-free deep-sea diving, so it's fair to assume that those whose idea of "loose change" is anything under a million are at least moderately pleased with their purchase.
A major problem, though, is image. I have been trying very hard and very patiently (for me, at least) to get through to assorted PHB that "LinuxBIOS" is not a distribution for embedded computers and has more than one user on the planet. This has been extremely trying on my nerves, as the mere mention of the acronym "BIOS" seems to terrify them beyond all imagining.
LinuxBIOS may well be the next "new thing", but it won't be the current "new thing" until it can get itself firmly established in the minds of those who really should be using it. What people think is true is far more influential than what is actually true. Sure, promotion is much less fun than development, and development is key to actually producing something people would use, but an idea that nobody's heard of is not much better than an idea that nobody's thought of.
(What I'd love to see is a kit for common PCs in their standard initial state, where any idiot with totally standard kit could swap their out-the-door BIOS with LinuxBIOS and back again. I'd also like to see packages for common Linux distros, with the idea of getting them eventually folded into the standard set. These are the ways of corrupting the youth - errr, educating users.)