bug in order>0 page allocations with !CONFIG_MMU

[David Howells <dhowells@redhat.com>]

Hi,

I've found that this:

	[mm/page_alloc.c]
	static inline void set_page_refs(struct page *page, int order)
	{
	#ifdef CONFIG_MMU
		set_page_count(page, 1);
	#else
		int i;

		/*
		 * We need to reference all the pages for this order, otherwise if
		 * anyone accesses one of the pages with (get/put) it will be freed.
		 */
		for (i = 0; i < (1 << order); i++)
			set_page_count(page+i, 1);
	#endif /* CONFIG_MMU */
	}

Causes problems if !CONFIG_MMU because __free_pages_ok()/free_pages_check()
reports a bad page on the second page when it comes time to free it:

	Bad page state at __free_pages_ok (in process 'events/0', page c08132e0)
	flags:0x20000000 mapping:00000000 mapcount:0 count:1

Why is doing this necessary at all? No one should be touching the individual
pages of a block allocation. The kernel should defend itself against
userspace trying to munmap part of a multipage mmap.

I think this should be:

	static inline void set_page_refs(struct page *page, int order)
	{
		set_page_count(page, 1);
	}

It seems to work for me. If no one disagrees, I'll give akpm a patch for this.

David