Re: ar "zero" flag

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Thu, Aug 07, 2008 at 09:55:54PM +0300, markucz%gmail.com@localhost wrote:
> That's why I use it. However, what I see makes me
> think these goals are being gradually replaced by something else. The actual
> i386 processor is practically desupported in recent versions.

There are a large number of good reasons to not support the i386 and
that don't apply to the successor and most of the non-Intel CPUs.

The biggest one is no support for cmpxchg. That means that i386 has to
use a different mutex implementation than all the other CPUs and even
then it normally has to done using either ras(9) or some other tricks.
In the kernel it might even be possible to patch around that, but that
doesn't help e.g. for rtld.

Other major issues are the lack of per-page TLB flushing and a number of
specific bugs to workaround in the critical code path.

This means additional branches and code that slows down everything else.
The code is almost completely untested and had some bugs for years,
noone can/wanted to test.

Joerg