>This 'mistake' only happened because someone decided to split the I and
>D caches. It wasn't a mistake before that happened. Therefore, by
>definition, it cannot be called a mistake at all, but a 'misdesign
>of the ARM architecture' as a whole.
Well, "mistake" was a bit strong. But even with a unified cache it is still
marginally more expensive to load and mask the SWI instruction than to have
user space load the value into a register.
>However, if you really think about it, not to put the syscall number in
>the comment field would have been an even bigger mistake to have made
>(since then you have to copy the user r1..r3,[sp],[sp,#4] to kernel
Well, you could put the syscall number in R12 or something, and do the args
precisely as they are done at the moment. As I mentioned, this is the route I
plan to take for Thumb, partly because the comment field is too small to be of
any real use and partly because it seems like a generally better plan with the
current state of the world.
p.
unsubscribe: body of `unsubscribe linux-arm' to [EMAIL PROTECTED]
