On Thursday 27 August 2009, Michael Schwingen wrote:
> > Why? That creates an additional code path and is harder to test.
>
> Is the simulator 100% correct in all cases? What about swi, exceptions
> and other corner cases?
There are a *boatload* of instructions added after ARMv4T, which
can be used on ARM1136 cores (like OMAP2420 or i.MX31) when the
compiler is allowed to do so.
Including VFP floating point ... which even some ARM9 cores support.
Right now, it seems common for Linux distros to target ARMv5TE,
which means only the "EDSP" instructions are likely to be trouble.
At least in userspace; kernel code is likely to use a lot more of
the instruction set.
So I'd say: use hardware single step whereever possible. We have
no particular reason to trust this simulator, since it's hardly
used; and right now we KNOW it's incomplete.
- Dave
p.s. I'll send a patch soonish teaching how to disassemble a
couple dozen of those instructions. I'd expect the
simulation
_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development
