Re: [linuxsh-dev] cache-sh4.c

[NIIBE Y. <gn...@m1...>]

M. R. Brown wrote:
 > Wow ... were you planning on merging these with linux-2_4-branch also?

Yes.  Besides this, we have important bug fixes (FPU thing, syscall
restart) and feature (kernel profiling support), which should be merged
into 2.4.

 > For this and your other changes, do you have numbers to show your
 > improvements?

We need more information, that's the reason I've implemented kernel
profiling (at last ;-).

I have only rough one.  I do:
	$ time for x in 0 1 2 3 4 5 6 7 8 9; do fo y in 0 1 2 3 4 5 6 7 8 9; \
	  do sh -c "echo >/dev/null"; done; done

to see how exec/exit goes.  flush_cache_range was the bad guy which
heavily calls flush_cache_4096 or __flush_cache_page.  It took about
4sec user, 25sec system with old implementation, while 4sec user,
17sec system with new implementation.  It is because of number of
instructions spent for cache flushing.

On SolutionEngine 7750. 

BTW, when we change flush_cache_range to just call flush_cache_all, it
goes 4sec user and 4sec system.  It seems for me that we have no way
other than flush_cache_all...
--