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

[Akira Tsukamoto]

On Sat, 02 Nov 2002 10:13:46 -0800
Andrew Morton <[e-mail]> mentioned:
> Akira Tsukamoto wrote:
> > > size by 17 kbytes, which is larger than my entire Layer 1 instruction
> > > cache!
> > 
> > It is because I was working on this patch based on 2.5.44 :)

Oh, Oh,
I meant the above that I *do* agree you. I just wanted to say,
two function was kept inline because from before 2.4 until though 2.5.44,
it was inlined, and when I received the faster-intel-copy from
Takahashi, I was working my patch against 2.5.44.

Just move __copy_to/from_user to usercopy.c will decrease the 
kernel size and should be almost the same as before. 
Am I missing something?

> - cache misses are slow
> - kernel has no right to be evicting user code from the CPU cache

Is this relevant to this patch?
I did not change any in my patch about it.

> - subroutine calls are fast

You mean almost no overhead.

> - smaller is faster

It could be said more efficient but faster?
The code or binary size directly connected to this issue?

> - inlining to the point of increasing code size is probably wrong

I agree, as my first comment.