magicalgoat (magicalgoat.delete@this.goatland.com) on January 9, 2020 4:58 pm wrote:
>
> Bit off topic, but could you give an example or two of something you've had to remove in this
> fashion? Just curious what things had serious enough reasons to get axed in this manner. Were
> most of these removals earlier on in the kernel history, or more evenly spaced out?

Most of the "feature removals" tend to be about hardware.

An example of this is removal of support for the original 80386 CPU's (and some non-intel clones that sometimes claimed they were 486 or 586) - the fact that it doesn't honor WP at CPL0 and we had to hack around it (badly, and insecurely due to fundamental races) in software instead just got to be too painful to maintain, for no real upside. Nobody is running original 80386 (or clones) any more, and even if they were (iirc there were RAD-hardened versions that might be used in very embedded places) they'll just have to use old kernels.

And we've actually removed a.out format support on x86 entirely, and nobody seems to have noticed. ELF has been the preferred format for 25+ years, but some people did have old a.out binaries for testing.

The code still exists, because the old alpha ECOFF binary format is really just "extended a.out", and people who ran alpha machines still used that. But it's disabled for x86, and nobody seems to have figured that out in a year, so we can pretty much consider it dead.

There's been a few random system calls too over the years. Sometimes we've found things that have been broken enough to "that can simply not have worked at all" for a long enough time that when we find the breakage, we go "let's remove it entirely" rather than try to fix it.

And it's been pretty evenly spaced out, I think. It's never been common.

Even early on, when we noticed that we had bad system call interfaces (think "struct stat" with 32-bit fields etc), we'd just create a new system call with the better interface. And it might happen a few times. I think we have four different versions of "stat()" system call, and a three different readdir/getdents system calls because we iterated over them..

Linus