Re: [PATCH] Clear dirty bits etc on compound frees

> Guillaume noticed this on s390 whilst writing a driver that used
> compound pages. Seems correct to me, I've tested it on i386 as
> well. The patch just makes us call free_pages_check for each element
> of a compound page.
> 
> diff -purN -X /home/mbligh/.diff.exclude virgin/mm/page_alloc.c 
> clear_dirty/mm/page_alloc.c
> --- virgin/mm/page_alloc.c    2003-10-14 15:50:36.000000000 -0700
> +++ clear_dirty/mm/page_alloc.c       2003-11-26 10:36:04.000000000 -0800
> @@ -267,8 +267,11 @@ free_pages_bulk(struct zone *zone, int c
>  void __free_pages_ok(struct page *page, unsigned int order)
>  {
>       LIST_HEAD(list);
> +     int i;
>  
>       mod_page_state(pgfree, 1 << order);
> +     for (i = 0 ; i < (1 << order) ; ++i)
> +             free_pages_check(__FUNCTION__, page + i);
>       free_pages_check(__FUNCTION__, page);
>       list_add(&page->list, &list);
>       kernel_map_pages(page, 1<<order, 0);

Gah. Guillaume pointed out that in editing his patch, I left the old 
free pages check in as well. <beats head repeatedly against wall>. Sorry.

I think you can reproduce this without the driver he's playing with
by mmap'ing /dev/mem, and writing into any clustered page group (that
a driver might have created or whatever).

diff -purN -X /home/mbligh/.diff.exclude virgin/mm/page_alloc.c 
clear_dirty/mm/page_alloc.c
--- virgin/mm/page_alloc.c      2003-10-14 15:50:36.000000000 -0700
+++ clear_dirty/mm/page_alloc.c 2003-12-01 11:34:09.000000000 -0800
@@ -267,9 +267,11 @@ free_pages_bulk(struct zone *zone, int c
 void __free_pages_ok(struct page *page, unsigned int order)
 {
        LIST_HEAD(list);
+       int i;
 
        mod_page_state(pgfree, 1 << order);
-       free_pages_check(__FUNCTION__, page);
+       for (i = 0 ; i < (1 << order) ; ++i)
+               free_pages_check(__FUNCTION__, page + i);
        list_add(&page->list, &list);
        kernel_map_pages(page, 1<<order, 0);
        free_pages_bulk(page_zone(page), 1, &list, order);

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