Hacker News new | past | comments | ask | show | jobs | submit login

As one of the people who has done work to keep Linux running on an neglected hardware platform (OpenSPARC T1 with an FPGA target) I agree with this article.

The patches for supporting various functions in the FPGA version of the T1 chip were never mainlined and have diverged from the current version of the kernel. The old version of the kernel they originally applied to no longer compiles without a number of changes. Automake doesn't work with the old makefiles anymore. So if you type make, you have to rewrite the makefile. Then if you do that, they still don't compile because of behavior changes in gcc. Some of these are the fault of the kernel developers making bad decisions, like including both -Wall and -Werror in their flags which cases compilation to fail with newer compilers when new warnings are introduced. (Compilers are always allowed to add warnings later, and indeed they do.) Along with a few other things that fail.

Since I didn't want to develop on a years old kernel anyway, I forward ported Sun's old patches to a newer version of the kernel and successfully moved the newest kernel code available for the platform forward a few years. But was never able to get the latest kernels working, because the newest kernels all fail on that platform during initialization.

I'm sure it's something I could fix if I needed to, but tracking down why a kernel crashes on an entire platform before it starts init is just no fun.

Maybe someone else fixed that bug and pushed it forward again, but as far as I know the current status of running Linux on an FPGA with the OpenSPARC T1 core is that the latest kernel available is 3.9 from my tree on github. [1]

Which is a shame, this is one of the very few open cores for a truly fast and well designed processor available in the world.

[1] https://github.com/djcapelis/linux-kernel-opensparc-fpga




> I'm sure it's something I could fix if I needed to, but ...

So in other words, no demand, no work. Had your job depended on it, you'd have done so. For everything else, if no hobbyist is willing to work on it then it gets de facto deprecated at some point down the read.

Plus, OpenSPARC T1 has been around for what... a decade? For all of its technical merits, that's about as old as dinosaurs in IT time. Deep Blue defeated Kasparov in 1997 and was the size of an oversized US fridge. Smart phones packed the same amount of computing power 13 years later. :-)


I mean. I'm not sure what your point is trying to explain my own motivations to me. It was part of what I was getting paid to do at the time. But I didn't need the newest kernel to run on it, so I never ported beyond 3.9. Arguably had it been a passionate side project, I may have gone further. Trying to use this as some kind of capitalistic example of how the market forces just didn't motivate it enough is... missing a lot of the point around so much of what happens here and why a lot of the people who do this work continue to do it.

Also, while the T1 core may be somewhat dated, it is still the newest general purpose ASIC design that's ever been fabricated as a major CPU that you can actually download the verilog for and put on an FPGA.

The advances in computing power haven't been driven by tons of changes in the core units since the gigahertz wars. They are by no means static, but an OpenSPARC T1 core has a surprising amount of similarity to a SPARC core shipping in new designs today. It's an extremely valuable model and opportunity to play with hardware modifications that is otherwise unparalleled outside the few organizations that regularly fabricate new ASICs.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: