Ben Harris <bjh21@cam.ac.uk> writes:
> On 4 May 2000, Chris G. Demetriou wrote:
> > if arm26 is going to start out ELF, it should do so with the normal
> > alignment (really, structure padding) used by other arm systems.
> 
> thorpej seemed to say that this would cause problems with some network
> protocols.  How true is this?  Assuming the long-term aim is to get arm32
> using the standard padding, testing it on arm26 first (where fewer people
> will mind) is a good idea.

to be honest, I don't know.


My own experience with ARM and the extra padding has been in arm
linux, and there there were plenty of user-land bits which were broken
by the extra padding being added to structures which were improperly
packed.  In NetBSD, i'd expect this will be better, because e.g. each
program doesn't have its own copy of a bloody shared structure...

I'd expect the issues might show up in network and SCSI, and maybe
other places.  Really, the only way you can be sure you've killed
problems like this is to go and run the code and see what you get and
fix the problems you find.

it took a couple of years of "use system, notice bug, fix bug" to weed
out "the last" of the LP64 issues exposed by the Alpha port, and then
you keep fighting the battle of people implementing new bugs.  The ARM
structure padding issue is the same class of problem, but should be
less severe because people need/want to make assumptions about
structure packing in fewer cases, and because it's generally easy to
fix (via __attribute__((__packed__)) or whatever).



chris