Archive for September, 2013

Linux Plumbers 2013

Posted: September 26, 2013 in Uncategorized

I attended 2013 Linux Plumbers in New Orleans.  The Graphics and Display mini-conf and the Android and Display mini-conf were the two items of biggest interest to me. Besides myself Sumit Semwal and Ross Oldfield from the Linaro Graphics Working Group were also there. I’m sure they’ll be posting their own thoughts as well.

I won’t cover every topic nor will I dive too deep for all that was part of the mini-confs.

Graphics and Display – Rob Clark

Rob Clark gave a good rundown of all the various completely Open Source driver development activities going on and demostrated his own project the open source Freedreno project. Vivante, Adreno, Mali, and Tegra are all seeing various attempts. Some like Freedreno and Lima are getting to be impressively good.

Media Decode and Composition: Bridging the Gap  – Daniel Stone

This was good to see especially since it’s an issue especially for set top boxes. It does point the way that EGL probably isn’t the tech of the feature here but rather it’s Wayland.

Common Display Framework – Laurent Pinchart

I had high hopes for Laurent’s session as given that make up of the room I was optimistic we might see some concrete agreements that might help things move along with Laurent’s proposal. From my perspective Laurent received some general guidance after he covered the current state of the design but general direction such as asking for something more incremental and more based on KMS didn’t give me hope that CDF is close to being merged. I share the opinion that fitting into the KMS infrastructure with appropriate proposals to change KMS seems like the right thing. There wasn’t (or I missed it) any discussion touching on the various protocols which panels use and how to make them into something that would be KMS based.

Sync/dmabuf-fences/dmabuf-sync

The sync discussion seemed quite productive. After setting the context of where and why sync is useful on Android,  there was agreement that for dma-buf and with dma-fences make sense and should go forward.

Atomic Display Framework (ADF)

ADF is a “new” set of patch recently posted that is sort of a next generation hwcomposer. Essentially how to atomically get a number of buffers put together and up on the screen. It’s entirely android specific which is one of it’s problems. I wouldn’t be the least bit surprised if this showed up in a future version of android which if that happened before it was upstreamed would be unfortunate. :-/

ION

The ion discussion was centered various issues that should be address on the road towards upstreaming, least that was the goal. Between how constraints should (or shouldn’t) be handled, how to keep performance good when the slow case especially on android is never acceptable and quite a number of complaints there didn’t seem to be an “ah ha” moment and as such it seemed more like the discussion was a review of things previously touched on but without exact consensus. It IS a complicated problem. In the spirit of increment improvement that kind of complexity makes it difficult to divide and conquer through a serious of smaller acceptable patches.

John Stultz’s lwn.net acticle on ion if you haven’t read, you should. http://lwn.net/Articles/565469/