It's Becoming Easier To Write Linux DRM Drivers

[phoronix]

Phoronix: It's Becoming Easier To Write Linux DRM Drivers

While writing DRM (Direct Rendering Manager) drivers were once a rather daunting task and not really considered much by ARM/embedded developers, over the past few years DRM has evolved a lot as it's picked up new drivers -- especially for today's many ARM SoCs -- and its core infrastructure has improved with picking up many new helpers and other improvements that lower the barrier of entry for DRM development...

http://www.phoronix.com/scan.php?page=news_item&px=Lower-Barrier-DRM-Drivers

[Jabberwocky]

I hope more people would realize how important code simplicity / maintainability improvements are. It's not as entertaining as performance improvements, but extremely valuable IMO.

[tildearrow]

Typo:

he was able to eliminate about two-thirdss of the code

[libv]

Now if only we knew about this hooks + helpers idea before...

Like when xfree86 moved from 3.x to 4.x. Sadly all that was thrown out of the window when RandR1.2 was invented, and no-one bothered to talk to display developers who had experience with the 4.x driver model.

[Ardje]

What is really sad is that his hardware had 4 years no upstream support, which is the state of most hardware...
A good thing, but also a bad thing... Can't submit complete drivers because there is hackish support for that things that kernel infra has no support for. Like GPIO on most usb-uart bridges, that can't be used.
But once a good infra structure starts, drivers are easy.

[ldo17]

But once a good infra structure starts, drivers are easy.

One thing that helps is Linux offers no guarantees of stable intra-kernel APIs. If they need to completely rework things and break backward compatibility to make it nicer, they will do it.

Of course, this means that out-of-tree drivers have a hard time of it. Which is why you should try to get them into the tree as much as possible.