Re: [PATCH] 2/2 2.5.45 cleanup & add original copy_ro/from_user

[Andrew Morton <akpm@digeo.com>]

Akira Tsukamoto wrote:
> 
> This consists mainly of the optimized copy routine for PIII/P4.
> 
> It is basically identical to what was introduced in 2.5.45.

But you've inlined them again.  Your patches increase my kernel
size by 17 kbytes, which is larger than my entire Layer 1 instruction
cache!

I'd prefer that we have these functions in .c, and laid out with
a minimum of C tricks.  Because more work needs to be done on the
memory copy functions, and doing that in header files is a pain.
(That is, using the movnta instructions for well-aligned copies
and clears so that we don't read the destination memory while overwriting
it).

Hopefully, yes, we can end up removing the runtime-selectable alignment
mask.  I left that in at present because it provides the infrastructure
for making other runtime-selectable decisions about how to perform
copies and clears.  Distributors like to be able to ship a minimum
number of kernels (say, just a PII-compiled kernel) and we want those
to run as well as possible on PIII and P4.