export swapper_space

This is now used as part of the page_mapping() macro. However, certain filesystems, such as ext3, make use of this. If it's not exported, they can't be compiled as modules.

James

===== mm/swap_state.c 1.77 vs edited =====

@@ -14,6 +14,7 @@
 #include <linux/pagemap.h>
 #include <linux/buffer_head.h>
 #include <linux/backing-dev.h>
+#include <linux/module.h>
 
 #include <asm/pgtable.h>
 

@@ -38,6 +39,7 @@
 	.a_ops		= &swap_aops,
 	.backing_dev_info = &swap_backing_dev_info,
 };
+EXPORT_SYMBOL(swapper_space);
 
 #define INC_CACHE_INFO(x)	do { swap_cache_info.x++; } while (0)
 

A filesystem (except for tmpfs of course) using page_mapping() looks
like a bug to me. I can't find ext3 using it anyway..

I'd be a bit reluctant to do this. filesystems actually have no need for page_mapping() - page->mapping is always correct in that context and page_mapping() has additional overhead. So if poss we should avoid this export so as to force filesystems to avoid page_mapping().

parisc broke because its flush_dcache_page() is inlined, and it uses page_mapping(). I'd suggest that parisc and arm uninline that function - it's quite large anyway.

You mean our fast path for flush_dcache_page

static inline void flush_dcache_page(struct page *page)

{
        struct address_space *mapping = page_mapping(page);

        if (mapping && !mapping_mapped(mapping)) {
                set_bit(PG_dcache_dirty, &page->flags);
        } else {
                __flush_dcache_page(page);
        }
}

Yes, that's the culprit...I suppose we could uninline it.

James