Re: [RFC][PATCH 2/5] add SWAP_HAS_CACHE flag to swap_map

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Thu, 28 May 2009 09:41:57 +0900
Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote:

> > @@ -1969,17 +2017,33 @@ int swap_duplicate(swp_entry_t entry)
> >  	offset = swp_offset(entry);
> >  
> >  	spin_lock(&swap_lock);
> > -	if (offset < p->max && p->swap_map[offset]) {
> > -		if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
> > -			p->swap_map[offset]++;
> > +
> > +	if (unlikely(offset >= p->max))
> > +		goto unlock_out;
> > +
> > +	count = swap_count(p->swap_map[offset]);
> > +	has_cache = swap_has_cache(p->swap_map[offset]);
> > +	if (cache) {
> > +		/* set SWAP_HAS_CACHE if there is no cache and entry is used */
> > +		if (!has_cache && count) {
> Should we check !has_cache here ?
I added !has_cache to return 0 in racy case.

> 
> Concurrent read_swap_cache_async() might have set SWAP_HAS_CACHE, but not have added
> a page to swap cache yet when find_get_page() was called.
yes.

> add_to_swap_cache() would handle the race of concurrent read_swap_cache_async(),
> but considering more, swapcache_free() at the end of the loop might dangerous in this case...

I can't catch what you mean.

I think swapcache_prepare() returns 0 in racy case and no add_to_swap_cache() happens.
wrong ?

> So I think it should be like:
> 
> 	read_swap_cache_async()
> 		:
> 		valid = swapcache_prepare(entry);
> 		if (!valid)
> 			break;
> 		if (valid == -EAGAIN);
> 			continue;
> 
> to let the context that succeeded in swapcache_prepare() do add_to_swap_cache().
> 

What you reccomend is code like this ?

==
	ret = swapcache_prapare(entry);
	if (ret == -ENOENT)
		break;    /* unused swap entry */
	if (ret == -EBUSY)
		continue; /* to call find_get_page() again */
==

-Kame





--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>