Nobody can know everything, but...

Posted May 1, 2012 9:14 UTC (Tue) by etrusco (guest, #4227)
In reply to: Nobody can know everything, but... by bdale
Parent article: Fixing the unfixable autofs ABI

After reading Linus' comments I can't avoid being biased against automount (well, this and maybe the automount's old-age fame of being hacky and buggy).
As I see it systemd simply "discovered" a bug in the kernel and the maintainers decided a general solution in the kernel was the best. And why do you say it was "fast-tracked"? 5 moths don't seem that rushed to me - but of course it's a bit unpleasant that they failed to notice that the other obvious user of the API had a different "fix" during all that time.

Nobody can know everything, but...

Posted May 1, 2012 10:05 UTC (Tue) by cortana (subscriber, #24596) [Link] (7 responses)

AIUI the problem only crops up when i386 userland is used on an amd64 kernel. Perhaps I'm being hopelessly naive, but I'd hope that such a setup is not particularly commonplace.

Nobody can know everything, but...

Posted May 1, 2012 10:29 UTC (Tue) by Fowl (subscriber, #65667) [Link]

Why? Seems like a perfectly reasonable situation. No bloated 64bit pointers but 4gb per process.

Nobody can know everything, but...

Posted May 1, 2012 15:09 UTC (Tue) by bdale (subscriber, #6829) [Link] (2 responses)

FWIW, 64 bit kernel and 32 bit userspace is what I run on my notebook.

Nobody can know everything, but...

Posted May 4, 2012 14:37 UTC (Fri) by bronson (subscriber, #4806) [Link] (1 responses)

How do you set that up?

Nobody can know everything, but...

Posted May 10, 2012 12:36 UTC (Thu) by nye (subscriber, #51576) [Link]

>How do you set that up?

aptitude install linux-image-amd64

Actually, it might even be the default these days if you install the i686 version of Debian on a 64-bit machine.

Nobody can know everything, but...

Posted May 1, 2012 16:45 UTC (Tue) by drag (guest, #31333) [Link]

I am running 64bit kernel with 32bit userland on a couple of my systems.

It's a very good option.

Nobody can know everything, but...

Posted May 11, 2012 13:21 UTC (Fri) by WolfWings (subscriber, #56790) [Link]

It's becoming the norm actually, lots of distro's do that by default or support it on their install/rescue media since it means they no longer need to build two separate userspaces, they just include two kernels to boot from on one disc, so for 99% of actual users there's only one specific disc image to download as a result.

Especially since a LOT of the time you _need_ a 32-bit userland to deal with a lot of binary packages out there or to get stuff running under WINE properly, it's easier for the "end user" images to stick to 32-bit userland even on 64-bit platforms. And even if you have a 64-bit userland you'll find it littered with various 32-bit-compat libraries.

Servers? Those are a different story, but many of those folks don't even need 32-bit compatibility so they can end up with a cruft-free pure-64-bit userland without all the 'compat' crap bolted in. So it reduces the size of the true-64-bit install media too, since there's no 32-bit cruft now.

Nobody can know everything, but...

Posted May 16, 2012 23:04 UTC (Wed) by steffen780 (guest, #68142) [Link]

IIRC that is what Linus recommends for many situations.

Nobody can know everything, but...

Posted May 1, 2012 17:14 UTC (Tue) by jzbiciak (guest, #5246) [Link] (1 responses)

I guess that's the real issue here: Linux accidentally broke the user-space ABI, and user-space hacked around it. (For me, that raises the question: Did they ever report this breakage to the kernel folks? If not, shame on them. If so, why didn't we fix this then?)

In any case, that effectively enshrined the bug as a new facet of the Linux kernel ABI.

Nobody can know everything, but...

Posted May 1, 2012 17:21 UTC (Tue) by raven667 (subscriber, #5198) [Link]

That communication and feedback loop from user space developers is really a critical factor. I would theorize that this explains a lot in the internals of older proprietary software such as Oracle or Solaris or Windows where the development team is kept away from customers. Raymond Chen has many curious examples of weird compatibility hacks in Windows for example which closely mirror this situation.