Re: Free pages leaking in 2.5.64?

[Nigel Cunningham]

Hi.

Thanks for the reply. I hadn't looked at the hot/cold stuff before. I
sussed it out this morning and added a condition to the test for
refilling the pcp arrays, stopping them from being refilled during a
suspend/resume cycle. Now everything works fine in that area for me.
I'll check that there aren't any other calls to refill the pcp arrays,
so I can be sure it will work with interrupts enabled and whenever smp
support is added to swsusp.

Now I just have to get the image written and read back and switch from
using page flags to dynamically allocated bitmaps, as I said I would.

Thanks again for the reply and regards,

Nigel

On Wed, 2003-03-12 at 13:25, Andrew Morton wrote:
> Nigel Cunningham <[e-mail]> wrote:
> >
> > Hi all.
> > 
> > I've come across the following problem in 2.5.64. Here's example output.
> > The header is one page - all messages only have a single call to
> > get_zeroed_page between the printings and the same code works as
> 
> nr_free_pages() does not account for the pages in the per-cpu head arrays. 
> 
> You can make the numbers look right via drain_local_pages(), but that is only
> 100% reliable on uniprocessor with interrupts disabled.
>