Re: [NFS] [PATCH 009 of 11] knfsd: Allow max size of NFSd payload to be configured.

[Hugh Dickins <hugh@veritas.com>]

On Thu, 28 Sep 2006, Neil Brown wrote:
> But are the pages that totalram is measure in, normal pages, of
> page_cache pages?  And is there a difference?

There's never yet been a difference, outside of some patches by bcrl.
But totalram_pages comes "before" any idea of page cache, so it's in
normal pages.

> Should we use PAGE_CACHE_SHIFT, or PAGE_SHIFT?

PAGE_SHIFT.

> And why do we have both if they are numerically identical?

Very irritating: the time I've wasted on "correcting" code for the
"difference" between them!  Yet there's still plenty wrong and I've
largely given up on it.

Probably never will be a difference: but the idea was that the page
cache might use >0-order pages (unclear what happens to swap cache).

I wish they'd waited for a working implementation before introducing
the distinction; but never quite felt like deleting all trace of it.

> 
> I'll submit a patch which uses
>         12 - PAGE_SHIFT
> in a little while.

I haven't seen your context; but "12 - PAGE_SHIFT" sounds like a
bad idea on all those architectures with PAGE_SHIFT 13 or more;
you'll be on much safer ground working with "PAGE_SHIFT - 12".

Hugh