MT string, to do next

[Ulrich Drepper]

Nathan Myers <ncm@nospam.cantrip.org> writes:

> I had understood earlier, from you, that the 386 (like SPARC-1 and
> HPPA) lacked an atomic synchronization primitive instruction suitable 
> for MT.  Was that understanding wrong?

These processors don't have the cmpxchg instruction.  Atomic increment
is available.

> It's not up to me, or you.  Red Hat and Debian distributions are both 
> built with flags for 386, not 486.  The Red Hat box says it runs on
> a 'x86', and the Debian box says "from 386 up to ...".

X86 certainly can exclude NexGen.  You know how incompatible they are.

> Anyway, it may be moot, if in fact the assembly code in atomicity.h 
> runs on any x86.  Are you sure about this?

Yes.

> Again, is it established that pre-486 hardware won't trap when 
> it sees the "lock" prefix?

Yes.

> Is it a no-op?

No.  It raises a external pin on the processor (namely #LOCK).  But
this has no consequence unless you are running one of those
multi-processor Sequent machines with 386.

-- 
---------------.      drepper at gnu.org  ,-.   1325 Chesapeake Terrace
Ulrich Drepper  \    ,-------------------'   \  Sunnyvale, CA 94089 USA
Red Hat          `--' drepper at redhat.com   `------------------------