Re: [RFC][PATCH 3/9] actual generic PAGE_SIZE infrastructure

[Dave Hansen <haveblue@us.ibm.com>]

On Wed, 2006-08-30 at 17:08 -0700, Christoph Lameter wrote:
> On Wed, 30 Aug 2006, Dave Hansen wrote:
> 
> >  #endif	/* CONFIG_ARCH_HAVE_GET_ORDER */
> > -#endif /*  __ASSEMBLY__ */
> > +#endif  /* __ASSEMBLY__ */
>          ^^^ Extra blank.

OK.  I'll fix that.

> > +	prompt "Kernel Page Size"
> 	               page size?

Sure.

> > +	  This lets you select the page size of the kernel.  For best
> > +	  32-bit compatibility on 64-bit architectures, a page size of 4KB
> > +	  should be selected (although most binaries work perfectly fine with
> > +	  a larger page size).  For best performance, a page size of larger
> > +	  than 4KB is recommended.  However, there are a number of
> > +	  side-effects of larger page sizes, like small files fitting poorly
> > +	  into the page cache.
> 
> Could we change this somewhat? Avoid the direct address and maybe say:
> 
>  The kernel page size determines the basic chunk of memory handled
>  by the Linux VM. The bigger the page size the less page objects
>  have to be managed by the kernel which reduces the VM overhead in
>  handling large amounts of data. However, larger pages also lead
>  to memory being wasted by the kernel since small files will
>  at mininum require one page of memory. A 4K pagesize is fairly standard 
>  and may be required for 32 bit compatibility on many platforms.
> 
>  It is usually not wise to select another page size than the default
>  unless one knows what one is doing or has some time to spend on
>  getting to know the kernel.

This is very nice.  I'll incorporate it.  

> Note that the default pagesize on IA64 is 16K and some important things 
> would change if a lesser size is selected. I have never run a 4K kernel. 
> I do not think we can just say that 4KB is okay. There may be other 
> platforms that have other default page sizes.

If we can't just say that it is OK, then we should probably disable it
in Kconfig.  Should we do that?

-- Dave