On Tue, 2008-05-27 at 21:37 +1000, Brett Nash wrote:
> > 
> > Actually that is rather odd. You shouldn't be getting bus errors on an
> > Intel platform as it can do non-aligned accesses.
> 
> Umm.. Only if the Unaligned bit is set in the flags register.  Any OS
> designed with half a clue or an application writer with half a clue and
> some knowledge of CPU usage will turn that bit off.

> Apple turn it off by default I do believe (god bless them).  The CPU
> itself boots with it off ;-)  

Well, they don't have fixups on PPC, so I guess there is not reason to
have them on Intel too :)

> Unaligned loads are handled by microcode BTW, they actually expand to
> two (independent) loads, two shifts and a register copy.  They are one
> of the reasons intel CPUs need stupid guarantees on TLB entries (the
> last _NINE_ entries are never flushed).

Any idea how to get a userspace application in Intel Linux to not have
fixups automatically? (And hence we can share the pain too.)

> > It will be interesting to find out what is going wrong.
> 
> /me notes it's probably unaligned access after a string.

Interestingly, it's not in a function I would have thought would have
these types of accesses.

Tim 'Mithro' Ansell

_______________________________________________
tp-devel mailing list
[email protected]
http://www.thousandparsec.net/tp/mailman.php/listinfo/tp-devel

Reply via email to