Automotive Grade Linux and a distribution for cars

At the 2015 Automotive Linux Summit in Tokyo, Dan Cauchy from the Linux Foundation (LF) kicked off the first day's program with an announcement: that the LF's Automotive Grade Linux (AGL) workgroup has decided to build its own Linux distribution, which it plans to run as an ongoing, long-term project. While the desire for a workgroup to create a distribution tailored to its needs is nothing new, the announcement had several in the crowd wondering what this decision meant for Tizen IVI—which, up until now, has served as the reference distribution for AGL. Tizen, of course, is also an LF-hosted project, and it has made in-vehicle infotainment (IVI) one of its high-priority use cases.

The creation of the new distribution project was the second of Cauchy's announcements during the opening session; the first was the 1.0 release of the AGL specification, which defines a Linux-based platform for use in vehicles. The specification sets out a list of required components. Many are standard modules in any modern distribution, such as BlueZ, GStreamer, systemd, and so on. Others are more peculiar to the automotive use-case, such as a "lifecycle manager" that governs the systematic startup and shutdown of components in a data-preserving manner—which is important considering that a car's driver can choose to switch the system off at any moment.

After the announcements, Cauchy said that Tizen IVI had served well as the jumping off point for AGL, and that the AGL membership has had a good working relationship with the Tizen IVI team. Notably, AGL members had contributed code that Tizen IVI developers merged into the distribution—work that includes, for example, several of the components detailed in the AGL 1.0 specification.

Using Tizen IVI as an upstream project allowed AGL to gain momentum quickly and to bring new developers on board in member companies, Cauchy said. But, he continued, the AGL membership now felt that Tizen IVI was not advancing past the "proof of concept" stage, and the membership was eager to start working on a software platform that it could actually deploy.

As those who follow Tizen will know, the distinction between those two possible states is certainly true. Tizen (in all its various flavors: smartphone, smart TV, IVI, and others) has always been marketed as a base platform on which consumer electronics makers could build a full-fledged product, not as a deployable finished product itself. Tizen IVI is not easy for developers to install, nor is it a plug-and-play solution for most cars. Project co-sponsor Samsung has been using Tizen in its latest generation of products, but the project itself is intentionally not designed to be a drop-in system for commercial vendors. Cauchy also noted that Tizen IVI is focused on supporting Tizen's HTML5 application framework, while many automotive suppliers want support for native applications as well.

Despite the decision to move away from Tizen as a formal upstream, though, Cauchy added that the AGL distribution (while still a long way from its eventual release) will draw heavily on Tizen IVI's components. He showed a slide indicating that the new distribution would incorporate software modules from both Tizen IVI and from the GENIVI Alliance. "In the past," he said, "these have been the three big players. Our goal now is to unite them."

Furthermore, he said, while both GENIVI and Tizen IVI are focused solely on the infotainment platform—roughly defined as the software running on the in-dash head unit—AGL has broader goals. In the future, carmakers will be running Linux on embedded telematics (data-collection and analysis) systems, engine-control units, instrument clusters, advanced driver-assistance systems (ADAS), and other computers in the car, he said, and AGL wants to be the distribution running on all of those devices.

The same approach—standardizing on a single distribution across all vendors—has worked well in other "highly vertical" markets, Cauchy said. It is the tactic undertaken by telecommunications companies in Carrier Grade Linux and by financial institutions. The automotive sector is noticeably behind the times in this regard, he said.

Despite Cauchy's explanation about the background of the decision, a number of people in the audience clearly found the decision to break with Tizen IVI to be a surprise. One of the first questions from the audience at the end of the session was whether or not AGL and Tizen IVI sponsor Intel were discontinuing their collaboration. Cauchy replied that the relationship with Intel remained strong, and that several of the software projects started in Tizen IVI were expected to be key pieces of the AGL distribution, too. Despite these assurances, though, it was hard not to notice that, in a change from the past, Intel and Tizen had little to no presence at the event itself this year.

ARM work

A session later in the day may have shed additional light on the reasoning that went into the decision. Hisao Munakata from ARM vendor Renesas presented [PDF] the results of a recent study that assessed Tizen IVI's suitability for usage in real-world automotive systems based on ARM hardware. ARM support is critical, Munakata said, because many carmakers and equipment suppliers want to take advantage of low-cost system-on-chip (SoC) hardware.

The study identified a few problem areas that perhaps stem from Tizen IVI's historical emphasis on Intel hardware. Intel and ARM processors have different power-management requirements, for example; while most of the engineers working on Tizen IVI came from Intel, the ARM side simply got less testing. But the biggest distinction between the platforms was not the CPU itself—it was the graphics stack. Current automotive UIs, including Tizen IVI's, rely heavily on OpenGL and hardware acceleration. Intel GPUs are (thanks to Intel's open-source drivers) better supported in these areas than are the GPUs found on most ARM SoCs. Closing that gap will require additional engineering effort.

The study also looked at how several Tizen IVI components measured up with respect to automotive-specific use cases. It singled out Automotive Message Broker (AMB) and Crosswalk as components that were positioned to do what carmakers want, but that still need additional work. AMB is a message-passing bus used for low-level interprocess communication (IPC). Munakata said that it needed to be extended to cover several other IPC domains, such as message-passing between discrete engine-control units (ECUs). Crosswalk is Tizen's HTML5 application runtime. Munakata said that Crosswalk performed well on most points in ARM SoCs, but needed additional work on WebGL and Video4Linux2 support.

In both cases, Munakata said, he has presented his findings to the respective Tizen IVI team and they have been well-received. So it may be that the automotive-specific software projects initiated inside Tizen IVI will get more attention moving forward, even if the AGL distribution takes over Tizen's place as the central reference platform.

What next

The distinction between proof-of-concept code and deployable product would appear to be the root issue. But it is still a major undertaking to start a new distribution project. AGL may become an industry standard—it certainly has the backing of major players—but it has a lot of ground to cover.

For the developers who have been contributing to Tizen IVI, the news of a new automotive distribution from AGL may not be all bad news. There is clearly still interest in the work they have done, and there is (thus far) no real overlap between the software projects hosted at Tizen IVI and those at AGL and at GENIVI.

The more difficult question is assessing whether or not the announcement should be taken as a serious setback for Tizen. But, in reality, Tizen IVI has always been substantially different from the other Tizen device profiles: a smart watch may have fewer system resources than a smartphone, but it still has one screen, one simultaneous user, and one battery, for example. Between the differing hardware requirements, the automotive-specific layers, and the regulatory issues that are critical in automotive, perhaps maintaining a shared code base between IVI and other Tizen device profiles is not entirely realistic.

Perhaps it could be argued, though, that as Tizen matures, its various device profiles have diverged to the point where ensuring that they all offer the same stack is a significant strain. If so, though, providing a compatible application-level API through Crosswalk may be the only factor that genuinely needs to unite Tizen IVI and the other Tizen device profiles.

Time will tell how the Tizen project weighs these issues, of course. In the meantime, the prospect of an ongoing automotive distribution project managed by AGL certainly means there will be renewed development on a number of automotive software projects. With any luck, the result will also include something that more Linux users and developers can install, use, and contribute to.

[The author would like to thank the Linux Foundation for travel assistance to attend ALS 2015.]