Leading items

Reverse engineering: more than NVIDIA deserves?

Reverse engineering is a longstanding tradition in the free software community. It has often been the only way to get hardware to work when the manufacturer refuses to make documentation available, but there is more to it than that. Some of us, certainly, enjoy the challenge of figuring out how a particular device works. And our sense of freedom tells us that it is our right to understand the hardware which we have purchased and rightfully own. We, as a group, tend not to respond well to those who tell us that reverse engineering a product is not the right thing to do. But, increasingly, your editor is hearing voices within the community which are saying just that.

One of the most prominent reverse engineering projects at the moment is Nouveau, which is starting to have some real success in making NVIDIA graphics adapters work with free software; see this week's Kernel Page for an article on the state of this project. NVIDIA hardware has been a problem for a long time, of course. It is said to be nicely-designed, and it is certainly present in a significant percentage of new machines, but NVIDIA has had no interest in making free drivers (or documentation) available for some years. So the only way for owners of this hardware to use it with reasonable performance under Linux is to use NVIDIA's proprietary kernel module, and that is a price many of us are not willing to pay.

There are currently about eight developers working to make the Nouveau driver better. They have reached a point where their understanding of the hardware and their reverse engineering tools are quite good; that, in turn, is enabling fast progress toward the creation of a working driver. With this kind of developer attention, the Nouveau driver may reach a stable state over the course of the next year, at least for some versions of the hardware. And that, it seems, should be a good thing.

Except for one little issue. NVIDIA's competition in this market is provided mainly by Intel and AMD/ATI. Intel provides free drivers for its hardware as a matter of company policy, and AMD has pushed a much more friendly policy onto ATI since the middle of last year. So free drivers for Intel video adapters come with distributions, and the first ATI drivers are beginning to become available.

One rather perverse result of this situation is that there are almost no community developers working on the Intel drivers at all. The development and maintenance of those drivers is an expense carried by Intel alone. One could argue that the lack of hardware documentation from Intel has made it hard for other developers to participate; Intel is now beginning to address that problem by burying the community in comprehensive, Creative Commons-licensed hardware programming manuals. It will be interesting to see how much more community help Intel gets as a result of its documentation release.

ATI, which has not, to date, provided working, free drivers, is arguably getting more help from the community and, especially, from distributors who have an interest in working drivers. But that company, too, is putting in resources of its own toward that goal.

NVIDIA, instead, is giving us nothing - and, in return, we are giving it an eight-person development team dedicated to the production of free drivers for its hardware. Once Nouveau is in a working state, Linux users will be able to buy NVIDIA hardware in the knowledge that it will simply work without requiring them to download and use binary-only kernel modules. The result of that can only be higher sales for NVIDIA.

While talking to developers at linux.conf.au, your editor heard a number of them say that NVIDIA does not deserve a gift of this magnitude from the community. We are now quite close to having free support for video hardware at all performance levels, supplied by friendly companies. Rather than penalize those companies by making a free gift to their biggest competitor, some say, shouldn't NVIDIA be made to pay for its behavior by exclusion from our community until it comes around?

There is a point here. The biggest lever we have when talking with hardware companies (or any company, for that matter) is money. Companies which see themselves as missing out on the Linux market will find a strong incentive to change their behavior. So if NVIDIA finds that system resellers are not using its chipsets for Linux-based systems, it will have to reconsider its position with regard to free drivers.

In the past, there was no credible alternative to NVIDIA, so the company had no real reason to fear that it could lose money as a result of its uncooperative behavior. Now there are well-supported alternatives at the lower end of the market, and the prospect of the same for high-end graphics as well. So there will be no need to buy hardware from this particular vendor, and, since the alternatives will be well supported, every reason to buy from somebody else.

Unless NVIDIA's hardware, too, is made to work via a community-supported driver. Should this happen, one could well say that we, as a community, have taken a prize away from companies which have treated us well and handed it to their competitor (which has not). Arguably, the community should not pursue the creation of reverse-engineered drivers in situations where competing vendors are playing by our rules. Otherwise, we are sending a rather conflicted message to both types of companies. It may really be true that, in the long run, the Nouveau driver is harmful to our real interests.

All of this discussion may be moot. There's no way that any of us could keep others from reverse-engineering their hardware and writing drivers, even if we wanted to. Anybody arguing against the mainline inclusion of a GPL-licensed driver for popular hardware is likely to end up in a minority position, to say the least. So, as a community, we cannot make a collective decision to stop this kind of development. But, as individual developers, we may occasionally want to give a moment's thought to the question of whether our activities are truly beneficial in the long run.

Comments (94 posted)

Directions in UMPC-land

It is an exciting time for Linux users who are interested in ultra-mobile PCs (UMPCs). New models are being announced frequently with many—dare we say most?—coming with at least the option to have Linux pre-installed. The low-cost models probably require Linux in order to make their price point, but even higher-end UMPCs seem to be made with Linux firmly in mind. In many ways, the One Laptop Per Child (OLPC) project has driven the demand for low-cost machines for adults as well.

Commercial offerings from ASUS (Eee PC), Everex (Cloudbook), Elonex (One), along with a rumored UMPC from HP are giving both the OLPC and Intel's ClassmatePC some competition. Add in Nokia's N810 and you have a half-dozen very mobile solutions featuring Linux—though the ClassmatePC seems to be more geared towards Windows XP. None of them has quite the right set of features to be the ultimate UMPC, but we seem to be headed in the right direction, so it is worth contemplating what that machine might look like.

Battery life is the achilles heel of mobile devices; some kind of breakthrough in power consumption or energy storage needs to happen for big strides to be made in this area. Because of weight considerations, today's UMPCs tend to have small batteries and three hours or less of battery life. Something on the order of twelve hours—with a measurement in days being the real goal—is more like what is needed. Perhaps some kind of human-powered or alternative charging mechanism can play a role. It is probably the biggest challenge to reaching something approaching an ultimate device.

Part of the reason that battery life is so low is because of how much power the display consumes. With rotating media on its way out (at least for these kinds of devices), the display is one of the areas where power savings would be felt most strongly. The E-Ink displays, such as those used by the newer e-book readers, have some great properties in terms of power consumption, but the speed at which they update makes them undesirable for general computer use. Many of us spend a fair amount of time looking at a static screen for several to many seconds at a time. Web pages or e-books might be candidates for using E-Ink, perhaps, but not Wesnoth or typing a document.

Perhaps a dual-mode screen that combined an LED and E-Ink display could blend the best of both. OLPC has an innovative display with many of the characteristics needed which can also can be viewed in sunlit conditions. Former OLPC CTO Mary Lou Jepsen's startup is licensing the XO display technology, so we may see it in a UMPC before too long.

The size of the display will likely need to be larger than today's offerings as well. That will be a balancing act between size, weight, and cost which will be interesting to see play out. A touchscreen is another feature that will be necessary as the display should be usable separate from the keyboard. Some way of transforming a small laptop into a tablet PC and e-book reader would be very desirable, with bonus points awarded if that transformation is fast and seamless.

A full-sized or nearly so keyboard is also a necessity. Too much of the work that we do involves words and numbers that need to be input. If this device is to become an integral part of a day-to-day routine, thumb or child-sized keyboards just won't cut it.

Wifi and wired connectivity are obvious, while Bluetooth would seem to be a good addition to provide internet via cell phone. Some might want to integrate actual cell phone functionality into the device itself—to avoid the multiple device hassle. Given that the size of a UMPC won't ever reach that of a cell phone, that seems like a stretch, but for those who want it, an optional feature seems like the way to provide that.

Like the OLPC, the device should be ruggedized, able to withstand reasonable amounts of abuse without much more than a case scratch. This is another area where flash disks will help as there won't be the threat of losing data when the disk heads suffer rapid deceleration. The price per gigabyte for solid-state drives will drop to the point where a few hundred GB will be possible at a reasonable price. Carrying around one's favorite music as FLACs, rather than in some lossy format, should be possible.

A fairly modest and power-friendly processor with a GB or two of RAM should round out the basics of the hardware. The device will run Linux, of course, and might have a few other peripherals: camera, microphone, speakers, etc. All should be available for $500-700, at least in a very functional low-end configuration. When might we see such a device? Two to three years seems quite likely, certainly before five years have passed. When it's ready, please send one to LWN for review in care of the author.

Comments (31 posted)

SCO to continue the fight?

Just as it seemed the SCO saga was drawing to a close, a new player, with up to $100 million to risk, has come on the scene. Stephen Norris Capital Partners (SNCP) has made an offer to take SCO private while providing a line of credit to allow the company to continue its operations. If the bankruptcy court in Delaware agrees to the plan—which is not a foregone conclusion—SCO and its various legal cases could be with us for a long time to come.

SNCP will put up $5 million in cash to essentially purchase between 51 and 85% of SCO; the exact percentage is dependent upon how much of the $95 million credit line is used to pay off Novell and/or IBM. If there is no payment, because SCO eventually wins those cases, SNCP will get 51%. If the payment is over $30 million, SNCP gets 85%; in between those two, the percentage of ownership will be pro-rated between the two. The actual transaction would issue "Series A Preferred" stock to SNCP (and its investors), which would be convertible into SCO "New Common Stock"; the current common stockholders would be see their shares "extinguished" and a trust established for them. This deal would take SCO private, no longer publicly traded nor subject to SEC reporting requirements.

Under the proposed agreement, the credit line has an interest rate of the London Interbank Offered Rate (LIBOR) plus "1700 basis points"—17% for those without a high-finance background—which currently works out to be around 20%. This is clearly not cheap money, but it does provide a rather large war chest for SCO to continue the fight. The Memorandum of Understanding (MOU) [PDF] makes it clear that interest payments are part of what the line of credit is supposed to pay for:

SCO's bombastic CEO, Darl McBride, will be required to resign as a condition of the deal. The Series A stockholders would be entitled to elect four of the seven board members, ensuring that they control the day-to-day direction of the company. The CEO would hold another seat, as would an "outside executive with suitable industry expertise." The remaining seat would be open to anyone and voted on by the current common stockholders.

What do the current stockholders get from this deal? Not much in the short term, as the MOU would set up a trust with $2 million (from the $5 million cash investment) to be distributed amongst the current stockholders. The current common stock would be "extinguished" and the trust would hold "New Common Stock" equivalent to the 15-49% left over based on the amount of the credit line used. Shareholders would get a pro-rata interest in the trust based on their current percentage of ownership. Based on 22 million outstanding shares, the distribution will amount to around $0.09 per share.

Since SCO sued IBM in March 2003, most of the stock speculation has been based on some kind of monetary settlement from IBM. Investors in SCO since that time have essentially been betting on that outcome; the new arrangement still allows the current stockholders to hold onto their litigation lottery ticket. Any settlement money that comes to SCO as a result of the Novell and IBM cases would be paid to the trust in the percentage of ownership of the company that it holds (i.e. 15-49%). At that time, the trust would also get its percentage of four times the previous year's earnings. These would then be distributed to the members of the trust.

It's a fairly complicated deal, this just covers the high points; the curious are directed at the MOU itself. It is a bit premature to proclaim that SCO is going private or getting $100 million as some in the press have done. The bankruptcy court will have its say; Novell may have an objection or two as well though, as things currently stand, they would be the likely beneficiary of some substantial part of the line of credit. We may get a read on how confident Novell is based on what, if any, objections they raise.

It is hard to imagine that SNCP thinks SCO's business prospects are such that a large financial commitment is warranted. This is very clearly an attempt to wring money out of the current litigation—and perhaps start additional lawsuits. It is interesting to note that in addition to the Novell and IBM lawsuits, the MOU specifically mentions the Autozone case. There is speculation that the idea of a "Linux tax" on users is an outcome that SNCP and its investors covet.

The question is, does SNCP truly believe that the claims made by SCO—without much in the way of supporting evidence so far—are likely to succeed on their merits? Or do they think that by providing enough incentive—in the form of a further protracted legal battle—might cause someone to settle? The IBM case has been dragging on for almost five years now. With the kind of money SCO would have at its disposal if this deal goes through, dragging out for another five does not seem implausible. At some point IBM or Novell may tire of the whole thing and try to cut some kind of deal. One hopes not, but that may be exactly what SNCP is betting on. The other side of that coin is that if that doesn't happen, we may well get a real hearing on some of IBM's counterclaims, in particular the GPL-infringement claims. That could be very interesting to watch.

Comments (66 posted)

Page editor: Jake Edge
Next page: Security>>