Re: BE (Was: What's the plan for powerpc64 in FreeBSD 16)

On Monday, November 17th, 2025 at 1:57 PM, Vadim Goncharov <vadimnuclight@gmail.com> wrote:

> On Mon, 17 Nov 2025 18:29:20 +0000
> Minsoo Choo minsoochoo0122@proton.me wrote:
> 
> > I was thinking about this as well since the discussion on armv7. I think
> > big-endian powerpc should be removed from releases from FreeBSD 16 for the
> > following reasons.
> > 
> > - Big endian usage is not used widely anymore. Even most POWER systems on
> > linux run on little-endian, and most Linux distros like RHEL only provides
> > little-endian and not big-endian.
> > - As FreeBSD is a complete operating system, if the powerpc is causing
> > issues not only in kernel but also for utilities and ports due to lack of
> > developers and hardware, there is no need to maintain them. Linux is a
> > kernel, so Linux developers can maintain kernel only for powerpc64be and
> > other developers take the responsibility of implementing libraries and
> > utilities. FreeBSD cannot take this approach, and if the cost is bigger than
> > the benefit, we should remove it.
> > - This also applies to deprecation of 32-bit platforms, but there is no need
> > to consider big endian compatibility unless we have valid reason. So far,
> > all the major platforms we support are little-endian or bi-endian, and even
> > bi-endian platforms like aarch64 and POWER mostly run on little endian
> > operating systems. Is there a new architecture or even ongoing discussion of
> > them that will be based on big endian (or 32-bit) where FreeBSD can shine?
> > If not, there is no need to consider compatibility for future architectures.
> 
> 
> This does not mean that such will never arise in the future, especially given
> that big-endian is better; and there were such rumors/tries for RISC-V.
> 

Right, we don't know if big-endian will be trend in future. However, I'm against preparing next big-endian architecture too proactively. As I stated above, I don't see any big-endian architecture where FreeBSD can shine, and this includes RISC-V big-endian. A few weeks ago, someone tried to upstream riscvbe to Linux kernel, and Linus Torvalds was strongly against it [1], and I think although he talked bit aggressively (as always), his claim itself sounds reasonable.

[1] https://www.phoronix.com/news/Torvalds-No-RISC-V-BE.