Re: aa-110-zone_accounting

[Andrew Morton <akpm@zip.com.au>]

Rik van Riel wrote:
> 
> On Tue, 19 Mar 2002, Andrew Morton wrote:
> 
> > 1: page_cache_size is no longer an atomic type - it's now just an
> >    unsigned long.  It's always altered under pagecache_lock.
> 
> Is this change worth it ?   This code will have to be changed
> back in any patch trying to fine-grain the pagecache lock...

Yes, it's the right change.  Changes to page_cache_size always
occur at the same time as changes to the per-zone accounting.
So the locking is shared and it all nestles nicely.

In the longer-term we don't want that atomic_t either.  We
need to avoid global and even per-zone atomic_t's and locks.

> Alternatively, maybe we should hide the page_cache_size behind
> magic macros too, so it's easier to change the underlying data
> structure(s).

I currently have:

+extern struct page_state {
+       unsigned long nr_dirty;
+       unsigned long nr_locked;
+       unsigned long nr_pagecache;
+} ____cacheline_aligned page_states[NR_CPUS];
+
+extern void get_page_state(struct page_state *ret);



-