[LWN Logo]

Bringing you the latest news from the Linux World.
Dedicated to keeping Linux users up-to-date, with concise news for all interests

 Main page
 Linux in the news
 Back page
All in one big page

Other stuff:
Contact us
Daily Updates
Notify list

Here is the permanent site for this page.

Leading items

This one needs to be looked at carefully. This week the Uniform Driver Interface (UDI) project burst on the scene in a big way. The Linux community is being told that UDI will be a great thing for it. But a closer look is called for before any real decisions are made.

UDI itself is not all that new: they first got going in 1994. The core members of the group include HP, SCO, Compaq, Interphase, Adaptec, and IBM, among others. Their goal has been to write device drivers that are source-compatible across many architectures. To this end, they have defined a set of minimal entry points that a driver must have, and they have created an extensive environment in which the driver runs. An individual device driver is well shielded from most aspects of the system it is running on; things like bus architectures and byte ordering are hidden within the driver environment. They plan to release a 1.0 version of the standard early next year.

What has brought UDI to the forefront now is Intel's endorsement of the upcoming standard. Intel has suddenly decided that supporting Unix on Intel systems is in its interest. Better late than never. Interpretations of this move vary, but it seems possible that Intel has concluded that NT will never win over the server market. One can imagine the folks at Intel reasoning that if the servers of the future will run Unix, they might as well make it easy to run Unix on Intel chips. That, after all, is where the high-margin chips are sold.

Perhaps most interestingly, Intel and UDI seem to have decided that adoption of UDI by Linux will be an important part of its success. See, for example, this ZDNet UK article on the subject. It merits a longish quote:

However, writing new drivers for the thousands of peripherals on the market is a daunting task. Hence, Project UDI is hoping the Linux community will help. Linux will be, said Quick, key to the adoption of the UDI initiative. A reference platform will be distibributed as freeware for Linux, and the Project UDI members will be counting on the Linux community to work on device drivers...

"The advantage of releasing to the Linux community is that their work will give Unix OS vendors a basis to work from"...

In other words, the Linux community, with its formidable ability to produce code in a hurry, is supposed to crank out the UDI drivers to make this whole thing work, to "give Unix OS vendors a basis to work from". This approach is simultaneously flattering and a bit presumptuous. It is not clear what Linux will get in return for providing the commercial OS vendors with a base of free drivers. Nonetheless, UDI should not be quickly rejected either. Here are a few points to consider:

  • One potential promise of UDI is that of drivers for hardware whose vendors will not release programming information. Of course, that is (1) not necessarily an entirely good thing, and (2) not guaranteed. For point (1), consider running a Linux system dependent on a base of binary-only drivers. It is hard to imagine that being a better system. A free system must have free drivers.

    For point (2), it is important to realize that UDI is a source-level standard. Just because a vendor has written a UDI driver for their hardware does not mean that they will provide one that has been built for Linux. Section 8.3 of their "Functional Requirements" document seems to address this point ("Drivers must not be required to be recompiled to run on different OSes which support the same ISA (Instruction Set Architecture), binary file format, and calling conventions"), but it is easy to see that not working out well in practice.

    It is also easy to imagine a binary-only driver that just does not work quite right under Linux, for whatever reason.

  • UDI promises to make drivers easier to write by providing a comprehensive support environment.

  • UDI claims that their drivers will not suffer a performance penalty. Others who have looked at the UDI specifications are not convinced. Extra layers of emulation code rarely are helpful in this regard.

  • It is not clear that hordes of Linux hackers will have any interest in creating lots of UDI drivers. The existing drivers work, there is little reward in converting them over to the UDI interface, perhaps slowing them down in the process. Somebody writing a new driver may choose the UDI route, but it's equally likely that they will want to write a native Linux driver. After all, native drivers have full access to the capabilities of the Linux kernel, will provide the best performance, and have a tremendous number of examples to work from.

  • Under what conditions can GPL'd Linux drivers be incorporated into a proprietary, binary-only Unix system? Conflicts over proper use of GPL code seem almost certain.

  • The image of Linux could be hurt if UDI is not supported. The cries of "they aren't serious about enterprise computing" and such can be easily imagined. It may be that some sort of UDI support will be needed just to avoid giving that sort of weapon to those who would attack Linux.

Some folks, notably Eric Raymond and Alan Cox, have suggested that the price for Linux UDI support should be an immediate opening up of the (currently nondisclosure) I2O bus architecture. Given the large degree of overlap in membership between UDI and I2O, such a deal could perhaps be possible.

Given the standards issue and the promise from the UDI backers that they will create a free implementation for Linux, some sort of incorporation of UDI seems likely at some point. That assumes, of course, that their implementation is reasonably well done, and could go into the kernel without making a mess. Nothing like UDI could possibly go in the kernel before 2.3 development starts, in any case, so there will be plenty of time to think about it.

For further information on UDI, including white papers and architecture definitions, have a look at the UDI web page. Hype can be found in Intel's and SCO's press releases.

LinuxWorld is coming. LinuxWorld is a new Linux-oriented web magazine; its editorial director will be Nicholas Petreley, the source of much good Linux commentary in the mainstream trade press. The senior editor is Robert McMillan. LinuxWorld was described to us as a "portal" site with an emphasis on professional and corporate users. There will also be original content, news articles, and the ability for readers to post items of interest. A press release should be forthcoming within a week or so; they expect to go online sometime in October.

LinuxWorld is thus the first foray into the Linux world by the mainstream technical press. Mr. Petreley is a figure with a lot of credibility in the Linux community; the chances are that he will produce something good. If this venture succeeds, there will likely be others - the world is discovering us.

Users of the Acorn Risc PCs were shocked when Acorn announced its plans to cancel their upcoming "Phoebe Risc PC 2" and concentrate instead on digital TV. For those unfamiliar with Acorn, Richard Simpson aptly summed them up as the last independent UK desktop computer design company. Comparable to Apple within the United States, Acorn has had a strong presence in UK education, but suffered from a dwindling market share. The demise of the Acorn Risc PC line definitely places Acorn among the many sufferers from the Microsoft and Intel monopolies.

Of what interest is this to the Linux community? Richard sent us a note mentioning that the Acorn announcements produced a wailing and gnashing of teeth and that the most common answer to the oft-asked question, "What now?" is to move to Linux. A quick perusal of the comp.sys.acorn.misc list backs him up. Here is a post by Phil Norman, outlining why he will be moving towards Linux and there are many, many others. The upshot is that another pool of talented and dedicated computer enthusiasts and programmers may be joining the Linux community. Of course, perhaps the very first reason they should join is the fact that they can count on Linux not just "disappearing" one day as a result of a management decision outside their control ...

We have set up a new address for letters to the editor. If you have a comment, correction, or dissertation that you would like to see published in LWN, send it to editor@lwn.net. We will assume that anything sent to that address is meant for publication. We can not, of course, promise to include everything sent to us; our goal is to include a not-too-large number of well-written items with each issue. For now, the "letters to the editor" section is in the back page, along with the links of the week.

As always, any other sort of correspondence should go to lwn@lwn.net.

September 24, 1998



Next: Linux in the news

Eklektix, Inc. Linux powered! Copyright © 1998 Eklektix, Inc., all rights reserved
Linux ® is a registered trademark of Linus Torvalds