On Thu, Jul 16, 2009 at 05:42:27PM +0200, Wolfgang Denk wrote:
Dear Scott Wood,
In message 4A5F4913.5030808@freescale.com you wrote:
So how do you propose that illegal divide operations be reported to the application?
In the same way as Linux is doing it (i. e. usually nothing at all) ?
Yay bugs.
What is so unreasonable about having a function to print a message and dump registers?
We didn't need one for the so far. What exactly do we need it for now?
For the same reason we have cpu/*/traps.c.
You could even make it a weak symbol that stays at NULL, so any attempt to call it will trap that way (assuming NULL pointers are trapped in U-Boot on that architecture...).
As you know, they are not.
If you're happy with not making it debuggable, then there's no need to trap on the NULL -- just make sure it never happens. :-)
I'd much rather spend the handful of bytes on at least a __builtin_trap(), though.
Or do compiler implementation specific things such as reimplementing libgcc functions or carefully avoiding generating calls to them, if you prefer. Whatever.
-Scott