uvm ncolors [was: CVS commit: src/sys/arch/x86/x86]

[Simon Burge <simonb%NetBSD.org@localhost>]

Thor Lancelot Simon wrote:

> Module Name:  src
> Committed By: tls
> Date:         Tue Apr 22 02:23:05 UTC 2008
> 
> Modified Files:
> 
>       src/sys/arch/x86/x86: cpu.c
> 
> Log Message:
> 
> Commit a quick workaround for the not-power-of-two cache colors problem
> pointed out by Simon (Simon's option #3): use the greatest power of two
> which is a divisor of the desired number of cache colors.
> 
> This code might want to stay even after the cache probing code is fixed.

I've got a simpler patch (below) that fixes this in uvm itself - this
change doesn't help any non-x86 if they ever try to do something stupid.

OK to back this out and apply this one instead?  I'll probably change
the wording in the comment slightly.  I also had the panic in an #ifdef
DIAGNOSTIC, but now think that it's worth enabling all the time - the
lossage mode when this is wrong isn't obvious.

Cheers,
Simon.
--
Index: uvm_page.c
===================================================================
RCS file: /cvsroot/src/sys/uvm/uvm_page.c,v
retrieving revision 1.131
diff -d -p -u -r1.131 uvm_page.c
--- uvm_page.c  24 Mar 2008 08:53:25 -0000      1.131
+++ uvm_page.c  21 Apr 2008 00:30:07 -0000
@@ -942,6 +942,16 @@ uvm_page_recolor(int newncolors)
        if (newncolors <= uvmexp.ncolors)
                return;
 
+       /*
+        * check if newncolors is a power-of-two, and adjust if not
+        * to the largest power-of-two that divides cleanly into the
+        * original requested number of colours.
+        */
+       newncolors = 1 << (ffs(newncolors) - 1);
+       if ((newncolors & (newncolors - 1)) != 0)
+               panic("%s: ncolors not a power of two %d", __func__,
+                   newncolors);
+
        if (uvm.page_init_done == false) {
                uvmexp.ncolors = newncolors;
                return;