page allocation failure. order:0, mode:0x0

[David F Barrera]

Tried the lru-removal.patch on 2.5.27 but the problem still occurs.
Many of these displayed on dmesg:  pdflush: page allocation failure.
order:0, mode:0x0

David Barrera

                                                                                                                                       
                      Andrew Morton                                                                                                    
                      <[e-mail]>        To:       Andrea Arcangeli <[e-mail]>                                             
                      Sent by:                 cc:       David F Barrera/Austin/IBM@IBMUS, [e-mail]                
                      [e-mail]        Subject:  Re: kernel BUG at page_alloc.c:92! & page allocation failure. order:0,        
                      .com                      mode:0x0                                                                               
                                                                                                                                       
                                                                                                                                       
                      07/24/2002 03:11                                                                                                 
                      PM                                                                                                               
                                                                                                                                       
                                                                                                                                       

Andrea Arcangeli wrote:
>
> On Wed, Jul 24, 2002 at 12:35:54PM -0700, Andrew Morton wrote:
> > And please drop the ptrace.c change and use
> > http://www.zip.com.au/~akpm/linux/patches/2.5/2.5.27/lru-removal.patch
> > instead.
>
> page_cache_release() can return a #define to __free_page().
>

Man, it can do a ton more than that.  This patch is just a stopgap
to prevent the oops.

page_cache_release() goes out onto the bus for the PageReserved()
test and then immediately goes out onto the bus again to perform the
atomic_dec_and_test().  Plus it tends to do all this inside
a global lock.  That PageReserved thing needs to go away.

Seriously, this stuff needs a truck driven through it.  See
http://mail.nl.linux.org/linux-mm/2002-07/msg00009.html and things
like pagevec_release().  It still needs quite some work, but the
optimisations which are available here are considerable.

-