Jason Thorpe <thorpej@nas.nasa.gov> writes:
> Are you sure it's not something arm32 specific?  Alignment problems on
> the arm32 cause the data to be incorrect; it doesn't cause a fault.

Uh, to be a bit more correct:

alignment problems on the arm32 often cause the data to be not what
you intended, not "incorrect."  The behaviour of unaligned accesses on
the ARM architecture is well defined... and not what most people
expect.

FWIW, it is possible on at least some ARM-family CPUs (i'm pretty sure
all of the StrongARM family) to turn on faults for unaligned accesses.
The relevant bit seems to be coprocessor 15, register 1, bit 1.

However, depending on the compiler's setting for architecture and
optimiziations, it might intentionally generate them.  Somebody should
probably play around with it and see what can be done...


cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.