Updates from Linaro [LWN.net]

Updates from Linaro

Posted May 26, 2011 18:46 UTC (Thu) by kiko (subscriber, #69905) [Link]

We definitely are trying. I've had multiple meetings with directors from both ARM SoC vendors and GPU IP vendors, but so far it's been meetings and not much else. I've discussed sharing of specifications, independent driver development and the marketing (and quality) advantage of a fully open source solution, but they still don't seem to understand the integration challenges open source presents, and are paranoid about IP lawsuits which might be well-founded given the amount of GPU-related patents approved and the amount of IP that's gone around in mergers acquisitions.

What sort of argument should we present them, and are there other things we could do to improve the GPU situation on ARM?

Note that David Airlie and I ran a session on this at Southern Plumbers; there's a video of it at http://blip.tv/linuxconfau/embedded-gpus-for-arm-open-dis...

Updates from Linaro

Posted May 27, 2011 12:21 UTC (Fri) by mjr (guest, #6979) [Link]

Thanks, glad to hear that it's at least pursued. Sadly, I don't have any magic bullet arguments to suggest.

Updates from Linaro

Posted Jun 7, 2011 7:19 UTC (Tue) by alison (✭ supporter ✭, #63752) [Link]

Arnd Bergmann of Linaro/IBM said at Linux Collaboration Summit that he expects to see an open GPU driver released in 2011. He doesn't strike me as a wild prognosticator, so he must know that one of the major vendors is considering a release their driver code. Once one vendor has done so, if they aren't sued to bits by others, perhaps we'll see real progress with GPUs.

I wonder how releasing the drivers would affect the GPU designers' business. On the one hand, they can monopolize the support and tools business around their closed products. On the other hand, the closed driver model may limit adoption of their products and will prevent community contribution of patches and other valuable ideas.

I read somewhere that Nvidia was actively collaborating with the developers of the Nouveau open-source drivers for their products. And not too long ago, Broadcom joined the Linux Foundation. TI is all in on open-source, so they may be pressuring Imagination Technologies about their drivers. So I think there's a glimmer of hope on the GPU front if no good news just yet.

Updates from Linaro

Posted May 26, 2011 23:06 UTC (Thu) by wookey (subscriber, #5501) [Link]

People are exerting pressure. Engineers understand what a PITA non-free drivers are, and what the vendors are missing out on by only providing an API, not instruction-level info, and I (and others) agitate within ARM every chance we get, so the people with the ability to do something about it are hearing the message regularly. But none of them have changed their minds yet. It'll take a while, same as it did on the desktop, and change is not inevitable.

Some people are scared about patents and loss of 'IP' control. Some believe their software is better than everyone else's so they don't want people copying it (failing to understand the difference between driver code and instruction set specs there). Several don't believe that outsiders could write drivers that were any good.

Updates from Linaro

Posted May 27, 2011 1:29 UTC (Fri) by aryonoco (subscriber, #55563) [Link]

You could point out to them that if we've been able to write a kernel that's good enough for them, we should be able to write drivers that are good enough as well.

As for patents, you could point out that whether their product has open source drivers or not has no bearing on whether it's infringing any patents. They can keep their drivers as close as they want and still be hit with patent lawsuits.

In the desktop and server space, there are many huge open source programs these days, programs that are essential in the enterprise market, and none of them have been hit with patent lawsuits any more than proprietary software of similar nature has been. I fail to understand the logic that assumes opening up code suddenly makes one more vulnerable to patent lawsuits.

Updates from Linaro

Posted May 27, 2011 11:00 UTC (Fri) by stevem (subscriber, #1512) [Link]

The usual argument is that a lot of the patented stuff is not immediately visible with closed drivers, whereas if people give out source showing how various features of a GPU work then it makes it easier for others to spot infringements.

The GPU area is a total mess because of the patent bollocks: as always, the only people benefiting are the lawyers.

Updates from Linaro

Posted May 27, 2011 11:53 UTC (Fri) by wookey (subscriber, #5501) [Link]

Indeed - that is excactly the argument given. Clearly code licencing has exactly zero effect on infringement status, but nevertheless people feel that at least if they only ever ship binaries it's harder to point the finger at any particular infringement. In practice there is probably some truth in that, but it seems a pretty ropey scheme to base a major plank of your business on.

Of course whilst we'd like open drivers we're not actually asking for them to hand over the source - just some specs so we can write our own. I think the fear there is that writing down enough detail to program the device also shows fairly clearly which patents might be being infringed. I have no idea of the truth or otherwise of that as my expertise in not in GPUs and their architecture/instruction sets.