Re: MM patches against 2.5.31

["Christian Ehrhardt" <ehrhardt@mathematik.uni-ulm.de>]

On Mon, Aug 26, 2002 at 10:09:38PM +0200, Daniel Phillips wrote:
> On Monday 26 August 2002 22:00, Christian Ehrhardt wrote:
> > On Mon, Aug 26, 2002 at 07:56:52PM +0200, Daniel Phillips wrote:
> > > On Monday 26 August 2002 17:29, Christian Ehrhardt wrote:
> > > > On Mon, Aug 26, 2002 at 04:22:50PM +0200, Daniel Phillips wrote:
> > > > > On Monday 26 August 2002 11:10, Christian Ehrhardt wrote:
> > > > > > + * A special Problem is the lru lists. Presence on one of these lists
> > > > > > + * does not increase the page count.
> > > > > 
> > > > > Please remind me... why should it not?
> > > > 
> > > > Pages that are only on the lru but not reference by anyone are of no
> > > > use and we want to free them immediatly. If we leave them on the lru
> > > > list with a page count of 1, someone else will have to walk the lru
> > > > list and remove pages that are only on the lru.
> > > 
> > > I don't understand this argument.  Suppose lru list membership is worth a 
> > > page count of one.  Then anyone who finds a page by way of the lru list can 
> > 
> > This does fix the double free problem but think of a typical anonymous
> > page at exit. The page is on the lru list and there is one reference held
> > by the pte. According to your scheme the pte reference would be freed
> > (obviously due to the exit) but the page would remain on the lru list.
> > However, there is no point in leaving the page on the lru list at all.
> 
> If you want the page off the lru list at that point (which you probably do)
> then you take the lru lock and put_page_testzero.

Could you clarify what you mean with "at that point"? Especially how
do you plan to test for "this point". Besides it is illegal to use
the page after put_page_testzero (unless put_page_testzero returns true).

> > If you think about who is going to remove the page from the lru you'll
> > see the problem.
> 
> Nope, still don't see it.  Whoever hits put_page_testzero frees the page,
> secure in the knowlege that there are no other references to it.

Well yes, but we cannot remove the page from the lru atomatically
at page_cache_release time if we follow your proposal. If you think we can,
show me your implementation of page_cache_release and I'll show
you where the races are (unless you do everything under the lru_lock
of course).

    regards   Christian

-- 
THAT'S ALL FOLKS!
--
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/