Intel Medfield Linux Support Gets Going

[phoronix]

Phoronix: Intel Medfield Linux Support Gets Going

Intel's next-generation MID (Mobile Internet Device) platform to succeed Moorestown is codenamed Medfield and is slated to be released next year. However, in usual Intel fashion, open-source patches for supporting this next-generation platform under Linux are beginning to make their way out there months in advance of the hardware's public availability...

http://www.phoronix.com/vr.php?view=ODc4NQ

[allquixotic]

The funny thing is that the PowerVR SGX series of tiny GPUs have been integrated in a tremendous number of very popular top-of-the-line smartphones! AFAIK, PowerVR is behind the iPad, iPhone, Motorola Droid, Droid 2, Droid X, the Palm Pre, Samsung Galaxy, Wave, and Galaxy Tab, many netbooks, etc.

PowerVR may easily sell as many GPUs as Intel these days (if not more), considering that mobile devices sell in far larger quantities than their more expensive and less convenient stationary counterparts. PowerVR had never been on my GPU vendor radar until this year, when I bought a Droid 2, and did some research, only to realize that TI OMAP and PowerVR are behind the surprisingly powerful graphics engine in this little smartphone.

I don't doubt that on Android devices such as Droid 2, there's a proprietary kernel module linked into the Linux kernel for the PowerVR SGX graphics. So I imagine it would be very interesting to a lot of people if a project were started to implement a gallium3d driver for some of the common PowerVR SGX cores. Admittedly, there would have to be quirks for the multitude of different applications of this core (e.g., there would have to be tweaks for differences between the SGX implementation in Apple A4, TI OMAP, and on the Atom netbooks), and the code would have to be cross-architecture, supporting at least x86, ARM v8, and Cortex/v9.

But -- if this could be accomplished -- it would be possible for open source enthusiasts to root their smartphone and possibly get an alternative graphics stack going based on mesa, gallium3d, KMS, etc. But then running the Android UI on top of that is another matter entirely....

I think the reason that Intel hasn't managed to ship a 100% working, open source, "Poulsbo" driver is that PowerVR/Imagination Technologies have been resistant to opening up a driver implementation on this popular core. Since they seem to have a relative monopoly in the small 3d-accelerated devices market, they probably want to hold on to as much of their unique "IP" (hate that term, btw) that gives them a competitive advantage, as possible.

[Svartalf]

The funny thing is that the PowerVR SGX series of tiny GPUs have been integrated in a tremendous number of very popular top-of-the-line smartphones! AFAIK, PowerVR is behind the iPad, iPhone, Motorola Droid, Droid 2, Droid X, the Palm Pre, Samsung Galaxy, Wave, and Galaxy Tab, many netbooks, etc.

Yep.

PowerVR may easily sell as many GPUs as Intel these days (if not more), considering that mobile devices sell in far larger quantities than their more expensive and less convenient stationary counterparts. PowerVR had never been on my GPU vendor radar until this year, when I bought a Droid 2, and did some research, only to realize that TI OMAP and PowerVR are behind the surprisingly powerful graphics engine in this little smartphone.

Heh... It's the SGX that empowers it, yes- but there are other GPUs in the mill, such as the Mali, that also are as capable. ImgTec's got the lead because they were there first along with the former AMD group that Qualcomm bought from AMD a while back.

I don't doubt that on Android devices such as Droid 2, there's a proprietary kernel module linked into the Linux kernel for the PowerVR SGX graphics. So I imagine it would be very interesting to a lot of people if a project were started to implement a gallium3d driver for some of the common PowerVR SGX cores. Admittedly, there would have to be quirks for the multitude of different applications of this core (e.g., there would have to be tweaks for differences between the SGX implementation in Apple A4, TI OMAP, and on the Atom netbooks), and the code would have to be cross-architecture, supporting at least x86, ARM v8, and Cortex/v9.

You should doubt it. There's an open sourced KERNEL module (else it'd be a GPL violation...) that exists in most of the bundles using ImgTec's and other player's OpenGL ES stuff. That's not the driver in the normal sense though. It's a command dispatch and memory management interface, mostly. The "driver" you normally associate with what you call a "driver", whether you're talking about Linux, Windows, or even MacOS, is in userspace in a proprietary blob there- where licensing isn't an issue.

Now...the code there can be cross-platform (and the guts thereof for AMD's and NVidia's X86 Linux drivers are pretty much that...) but it doesn't have to be. Now, as an aside, there HAS been a Gallium 3D driver made for the SGX 535. It just hasn't been published by the developer because of licensing- even for closed-source purposes. It's a sore subject because people with the Poulsbo chipset machines would very much like to have 3D with their Linux.

But -- if this could be accomplished -- it would be possible for open source enthusiasts to root their smartphone and possibly get an alternative graphics stack going based on mesa, gallium3d, KMS, etc. But then running the Android UI on top of that is another matter entirely....

Not currently possible. There are people within the Open Pandora community that're doing a Nouveau reverse engineer pass at the SGX, and they've gotten FAR, but they're not remotely close to where you'd be able to do that sort of thing. The only way you're going to get that sort of situation is if someone like Dell or HP bought out a 50-55% stake in ImgTec and TOLD them to FOSS their drivers. (Don't snicker...it's been done to another silicon vendor in recent times...)

I think the reason that Intel hasn't managed to ship a 100% working, open source, "Poulsbo" driver is that PowerVR/Imagination Technologies have been resistant to opening up a driver implementation on this popular core. Since they seem to have a relative monopoly in the small 3d-accelerated devices market, they probably want to hold on to as much of their unique "IP" (hate that term, btw) that gives them a competitive advantage, as possible.

You got it right in the last sentence. Past that, Intel's not able to publish a working FOSS Poulsbo driver because part of the licensing for any use of the programming info gives ImgTec a right to nix the work going out altogether. They don't want it out there- so it's not even available except under an NDA with Intel.