KMS For FreeBSD Is Still A Work In Progress
Phoronix: KMS For FreeBSD Is Still A Work In Progress
FreeBSD still lacks mainline support for kernel mode-setting (KMS) on modern hardware, but at least it's still being worked on...
The FreeBSD Intel-GPU Wiki states, that the developer doesn't care about an overall solution for KMS. "The power to serve" OS is personally the first choice of mine, when it comes to setup servers on modern hardware, but as member of a group of researchers, we are desperately dependend on high-performance computing, which also tend to be performed on modern GPU hardware with OpenCL or even CUDA.
AMD graphics cards are in most cases a better joke, even in GPGPU aspects and support. Well, the picture I see is: FreeBSD has defeated to an existence in a niche, mostly database related or in some cases webservice. ther the performance of FreeBSD is good and has been optimized for such purposes, but there is rarely more. As in networking or NUMA/architectural issues, FreeBSD is given away its leading role long ago and it is possibly worth looking at the "WHY", since the MIT/BSD licensing model is much more open and free, even for the use at universities as it is with the GPLv2/GPLv3 and beyond. Although this is a fact, I see, especially in Germany and rest of Europe, a very strong movement on Linux.
Well, the lack of suitable high performance compilers (C, C++ and Fortran95/2000, HPCF), optimized math libs and, even now, no GPU support via OpenCL/CUDA (I can not speak for most of the people, but I think this is simply a psychological issue, since Linux is popular on the desktop, so there are graphics cards needed and the development has gone far beyond that what FreeBSD still is en par with - so Linux has modern graphical substems with a high security aspect (KMS), GPU vendors offer for this market a lot of BLOBs and even working OpenCL implementations, amongst CUDA and therefore, there is a "market").
In the past I hoped there could be project related to implementing modern concept, based on BSD systems and funded by our government or the university. But since we dumped "Diploma" and adopted this "bachelor" and "masters" nonesense in germany, students do not have time to go deeper into the matter and those heading for masters, from my own perspective and experience, do not have the time to open their mind for the free academia sense, they're heading now for profit as fast as they can. Well, this is a social situation completely different to the time when X11, BSD and other projects were founded and I think this social and philosophical aspect is far beyond the target of the thread and focus of "FreeBSD lacks in KMS".
Watching the development on OpenBSD, which has a very strong security concern, reveals, that there is also not much going on for KMS, even KMS would hease the pain of having a root-UID X11-server. Development isn't "open" anymore, concepts are to harshly narrowed down to "Linux" and its siblings and incarnations and there is, already, a too strong business behind them.
In many cases, I regret that being one of the last "dinosaurs" in my department fighting for a BSD OS on our hardware platforms, that my colleaugues and opponents are right when they say it's a dead end relying on FreeBSD! ZFS, for instance, isn't a striking reason anymore since development in Linux has come en par with that in FreeBSD. And since our platforms, even the desktops, are tending more and more to get poerful graphics hardware, not for displaying, but for number crunching, we get more and more forced to switch over to Linux. And why keeping a heterogenous environment running, when one OS can provide everything we "need".
And then, reaching this point, FreeBSD is the only choice for at home. And with "at home", there is no voice in the public except for desktop systems with funny graphics - and here the deadly and develish circle closes ...
Do you think that if the LLVM infrastructure (with clang and let's say in the near future LLDB) is rolled in and successfully builds ports this could change considerably the picture in the BSD world? I see this happening over the next few iterations of LLVM and supplanting GCC-infrastructure in FreeBSD on a quasi-experimental level in 10. This, in my mind, could be a considerable boost for BSD.
Originally Posted by Eisnefaust