On Thu, 1 Nov 2007, Hyon Lim wrote:
I think the reason of assembly implementation is processor context replacement.
Understood. Assembly may indeed be required for specific things like restoring register state or fiddling with the cache if there aren't already macros or functions in the kernel that do exactly what you need.
The second reason that I thinking is calling convention of C language.
It's not uncommon at all to have assembly language glue, say between a BIOS callback and the C language routine that does the work.
In your original post, you mentioned tracking variables and things that suggested a module that does much more that just load some odd registers or flip around a function call stack. If that is indeed the case, then for sake of maintainablility and readability, one would be strongly encouraged to write the core stuff in plain old C and sprinkle in assembly glue code as required.
Think about it this way. MIPS is a pretty large family of CPUs, each with it's own strange behaviors. Several of those people on this list spend a lot of time tweeking that assembly to make it work cleanly across various CPUs. It's a lot easier to understand 25 lines of assembly interface code and 200 lines of C code, than an entire 1000 line module written in assembly. It's also a lot easier when you can shove most of the work over to the compiler, especially if others like your work and want to generalize it for use on many other MIPS CPUs.
I guess the real question here is how complex do you think your code needs to be? That should determine your path.
Regards, -S-
