DRM Change Continues To Cause Debate

Kristian Høgsberg on the 6th of November had wrote a message on the DRI development list regarding the libdrm repository. With so much of the Direct Rendering Manager (DRM) work going straight into the Linux kernel -- thanks in large part to all of the work on memory management and kernel mode-setting -- Kristian proposed that the DRM driver code be removed from the separate DRM Git tree. With this message, Kristian created a new DRM repository that dropped all of the linux-core, bsd-core, and shared-core code. Seems simple and straightforward, right? Well, three weeks later with dozens of replies, this change is continuing to cause debate.

After some initial confusion was cleared up, some active DRM developers (such as Jerome Glisse) were in support of this change while others (such as Eric Anholt) thought it would do more harm than good with creating a larger burden on the developers and users wishing to test out this code. Some developers viewed this as a way to get greater testing of it living in the kernel tree that it would get better test coverage with each kernel release candidate and avoiding copying files, while others feared this would force users to upgrade their kernel to obtain simple DRM fixes.

These concerns led Kristian to revise his proposal to pull the latest DRM driver bits from a Linux kernel repository automatically into his libdrm Git repository. Kristian described this work as "[not] an attempt to change how we work, it was merely a re-organisation of the drm.git repo to better reflect and support the de facto workflow." But still, some developers objected.

A week ago, another can of works got opened when one user became disgruntled that the bsd-core code was removed from the tree. While the BSD driver bits were dropped from the tree, so were the Linux bits, but the BSD code was rarely worked on and not even used by the BSD maintainers. As can be seen from this DRM log, the last time the bsd-core was touched inside Mesa/DRM was in March and before that was only worked on every few months almost entirely by a lone FreeBSD developer, Robert Noland. Robert though wasn't even the one complaining about this code location change.

Robert Noland ended up responding to say "the code there is useless and only provided minor historical value." This though didn't settle things and led Vehemens to dispute the number of actual *BSD developers working on the DRM code, code not being shared, and other issues. Here though is a message that describes the FreeBSD DRM situation at least, by Robert himself, but really is just becoming into a public fight between two developers that both actually have the same goal of bettering the *BSD DRM code.

As of today, this mailing list thread remains very active. The changes though already went into effect and the linux/bsd/share-code folders were removed more than a week ago (commit) that reduced the number of lines of code by more than 130,000. Of course, this code remains available, but just not in the master DRM git tree. The linux-core code wasn't worked on in that location in months and the bsd-core code wasn't in even more months.

The DRM BSD code is continuing to be worked on, albeit in a different location and doesn't receive as much love as the Linux DRM. OpenSolaris or FreeBSD/NetBSD/OpenBSD have yet to support the Graphics Execution Manager or TTM for kernel memory management, let alone kernel mode-setting or other recent DRM changes.