Merge to Fedora kernel-2.6.18-1.2224_FC5 patched with stable patch-2.6.18.1-vs2.0...
[linux-2.6.git] / mm / rmap.c
index a4658b4..bca42f9 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -53,6 +53,7 @@
 #include <linux/rmap.h>
 #include <linux/rcupdate.h>
 #include <linux/module.h>
+#include <linux/kallsyms.h>
 #include <linux/vs_memory.h>
 
 #include <asm/tlbflush.h>
@@ -104,7 +105,7 @@ int anon_vma_prepare(struct vm_area_struct *vma)
                spin_lock(&mm->page_table_lock);
                if (likely(!vma->anon_vma)) {
                        vma->anon_vma = anon_vma;
-                       list_add(&vma->anon_vma_node, &anon_vma->head);
+                       list_add_tail(&vma->anon_vma_node, &anon_vma->head);
                        allocated = NULL;
                }
                spin_unlock(&mm->page_table_lock);
@@ -128,7 +129,7 @@ void __anon_vma_link(struct vm_area_struct *vma)
        struct anon_vma *anon_vma = vma->anon_vma;
 
        if (anon_vma) {
-               list_add(&vma->anon_vma_node, &anon_vma->head);
+               list_add_tail(&vma->anon_vma_node, &anon_vma->head);
                validate_anon_vma(vma);
        }
 }
@@ -139,7 +140,7 @@ void anon_vma_link(struct vm_area_struct *vma)
 
        if (anon_vma) {
                spin_lock(&anon_vma->lock);
-               list_add(&vma->anon_vma_node, &anon_vma->head);
+               list_add_tail(&vma->anon_vma_node, &anon_vma->head);
                validate_anon_vma(vma);
                spin_unlock(&anon_vma->lock);
        }
@@ -206,44 +207,6 @@ out:
        return anon_vma;
 }
 
-#ifdef CONFIG_MIGRATION
-/*
- * Remove an anonymous page from swap replacing the swap pte's
- * through real pte's pointing to valid pages and then releasing
- * the page from the swap cache.
- *
- * Must hold page lock on page and mmap_sem of one vma that contains
- * the page.
- */
-void remove_from_swap(struct page *page)
-{
-       struct anon_vma *anon_vma;
-       struct vm_area_struct *vma;
-       unsigned long mapping;
-
-       if (!PageSwapCache(page))
-               return;
-
-       mapping = (unsigned long)page->mapping;
-
-       if (!mapping || (mapping & PAGE_MAPPING_ANON) == 0)
-               return;
-
-       /*
-        * We hold the mmap_sem lock. So no need to call page_lock_anon_vma.
-        */
-       anon_vma = (struct anon_vma *) (mapping - PAGE_MAPPING_ANON);
-       spin_lock(&anon_vma->lock);
-
-       list_for_each_entry(vma, &anon_vma->head, anon_vma_node)
-               remove_vma_swap(vma, page);
-
-       spin_unlock(&anon_vma->lock);
-       delete_from_swap_cache(page);
-}
-EXPORT_SYMBOL(remove_from_swap);
-#endif
-
 /*
  * At what user virtual address is page expected in vma?
  */
@@ -473,6 +436,71 @@ int page_referenced(struct page *page, int is_locked)
        return referenced;
 }
 
+static int page_mkclean_one(struct page *page, struct vm_area_struct *vma)
+{
+       struct mm_struct *mm = vma->vm_mm;
+       unsigned long address;
+       pte_t *pte, entry;
+       spinlock_t *ptl;
+       int ret = 0;
+
+       address = vma_address(page, vma);
+       if (address == -EFAULT)
+               goto out;
+
+       pte = page_check_address(page, mm, address, &ptl);
+       if (!pte)
+               goto out;
+
+       if (!pte_dirty(*pte) && !pte_write(*pte))
+               goto unlock;
+
+       entry = ptep_get_and_clear(mm, address, pte);
+       entry = pte_mkclean(entry);
+       entry = pte_wrprotect(entry);
+       ptep_establish(vma, address, pte, entry);
+       lazy_mmu_prot_update(entry);
+       ret = 1;
+
+unlock:
+       pte_unmap_unlock(pte, ptl);
+out:
+       return ret;
+}
+
+static int page_mkclean_file(struct address_space *mapping, struct page *page)
+{
+       pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
+       struct vm_area_struct *vma;
+       struct prio_tree_iter iter;
+       int ret = 0;
+
+       BUG_ON(PageAnon(page));
+
+       spin_lock(&mapping->i_mmap_lock);
+       vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
+               if (vma->vm_flags & VM_SHARED)
+                       ret += page_mkclean_one(page, vma);
+       }
+       spin_unlock(&mapping->i_mmap_lock);
+       return ret;
+}
+
+int page_mkclean(struct page *page)
+{
+       int ret = 0;
+
+       WARN_ON_ONCE(!PageLocked(page));
+
+       if (page_mapped(page)) {
+               struct address_space *mapping = page_mapping(page);
+               if (mapping)
+                       ret = page_mkclean_file(mapping, page);
+       }
+
+       return ret;
+}
+
 /**
  * page_set_anon_rmap - setup new anonymous rmap
  * @page:      the page to add the mapping to
@@ -494,7 +522,7 @@ static void __page_set_anon_rmap(struct page *page,
         * nr_mapped state can be updated without turning off
         * interrupts because it is not modified via interrupt.
         */
-       __inc_page_state(nr_mapped);
+       __inc_zone_page_state(page, NR_ANON_PAGES);
 }
 
 /**
@@ -538,7 +566,7 @@ void page_add_new_anon_rmap(struct page *page,
 void page_add_file_rmap(struct page *page)
 {
        if (atomic_inc_and_test(&page->_mapcount))
-               __inc_page_state(nr_mapped);
+               __inc_zone_page_state(page, NR_FILE_MAPPED);
 }
 
 /**
@@ -547,18 +575,18 @@ void page_add_file_rmap(struct page *page)
  *
  * The caller needs to hold the pte lock.
  */
-void page_remove_rmap(struct page *page)
+void page_remove_rmap(struct page *page, struct vm_area_struct *vma)
 {
        if (atomic_add_negative(-1, &page->_mapcount)) {
-#ifdef CONFIG_DEBUG_VM
                if (unlikely(page_mapcount(page) < 0)) {
                        printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page));
                        printk (KERN_EMERG "  page->flags = %lx\n", page->flags);
                        printk (KERN_EMERG "  page->count = %x\n", page_count(page));
                        printk (KERN_EMERG "  page->mapping = %p\n", page->mapping);
+                       if (vma->vm_ops)
+                               print_symbol (KERN_EMERG "  vma->vm_ops->nopage = %s\n", (unsigned long)vma->vm_ops->nopage);
+                       BUG();
                }
-#endif
-               BUG_ON(page_mapcount(page) < 0);
                /*
                 * It would be tidy to reset the PageAnon mapping here,
                 * but that might overwrite a racing page_add_anon_rmap
@@ -570,7 +598,8 @@ void page_remove_rmap(struct page *page)
                 */
                if (page_test_and_clear_dirty(page))
                        set_page_dirty(page);
-               __dec_page_state(nr_mapped);
+               __dec_zone_page_state(page,
+                               PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
        }
 }
 
@@ -579,7 +608,7 @@ void page_remove_rmap(struct page *page)
  * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
  */
 static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
-                               int ignore_refs)
+                               int migration)
 {
        struct mm_struct *mm = vma->vm_mm;
        unsigned long address;
@@ -601,9 +630,8 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
         * If it's recently referenced (perhaps page_referenced
         * skipped over this mm) then we should reactivate it.
         */
-       if ((vma->vm_flags & VM_LOCKED) ||
-                       (ptep_clear_flush_young(vma, address, pte)
-                               && !ignore_refs)) {
+       if (!migration && ((vma->vm_flags & VM_LOCKED) ||
+                       (ptep_clear_flush_young(vma, address, pte)))) {
                ret = SWAP_FAIL;
                goto out_unmap;
        }
@@ -621,25 +649,46 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 
        if (PageAnon(page)) {
                swp_entry_t entry = { .val = page_private(page) };
-               /*
-                * Store the swap location in the pte.
-                * See handle_pte_fault() ...
-                */
-               BUG_ON(!PageSwapCache(page));
-               swap_duplicate(entry);
-               if (list_empty(&mm->mmlist)) {
-                       spin_lock(&mmlist_lock);
-                       if (list_empty(&mm->mmlist))
-                               list_add(&mm->mmlist, &init_mm.mmlist);
-                       spin_unlock(&mmlist_lock);
+
+               if (PageSwapCache(page)) {
+                       /*
+                        * Store the swap location in the pte.
+                        * See handle_pte_fault() ...
+                        */
+                       swap_duplicate(entry);
+                       if (list_empty(&mm->mmlist)) {
+                               spin_lock(&mmlist_lock);
+                               if (list_empty(&mm->mmlist))
+                                       list_add(&mm->mmlist, &init_mm.mmlist);
+                               spin_unlock(&mmlist_lock);
+                       }
+                       dec_mm_counter(mm, anon_rss);
+#ifdef CONFIG_MIGRATION
+               } else {
+                       /*
+                        * Store the pfn of the page in a special migration
+                        * pte. do_swap_page() will wait until the migration
+                        * pte is removed and then restart fault handling.
+                        */
+                       BUG_ON(!migration);
+                       entry = make_migration_entry(page, pte_write(pteval));
+#endif
                }
                set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
                BUG_ON(pte_file(*pte));
-               dec_mm_counter(mm, anon_rss);
        } else
+#ifdef CONFIG_MIGRATION
+       if (migration) {
+               /* Establish migration entry for a file page */
+               swp_entry_t entry;
+               entry = make_migration_entry(page, pte_write(pteval));
+               set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
+       } else
+#endif
                dec_mm_counter(mm, file_rss);
 
-       page_remove_rmap(page);
+
+       page_remove_rmap(page, vma);
        page_cache_release(page);
 
 out_unmap:
@@ -729,7 +778,7 @@ static void try_to_unmap_cluster(unsigned long cursor,
                if (pte_dirty(pteval))
                        set_page_dirty(page);
 
-               page_remove_rmap(page);
+               page_remove_rmap(page, vma);
                page_cache_release(page);
                dec_mm_counter(mm, file_rss);
                (*mapcount)--;
@@ -737,7 +786,7 @@ static void try_to_unmap_cluster(unsigned long cursor,
        pte_unmap_unlock(pte - 1, ptl);
 }
 
-static int try_to_unmap_anon(struct page *page, int ignore_refs)
+static int try_to_unmap_anon(struct page *page, int migration)
 {
        struct anon_vma *anon_vma;
        struct vm_area_struct *vma;
@@ -748,7 +797,7 @@ static int try_to_unmap_anon(struct page *page, int ignore_refs)
                return ret;
 
        list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
-               ret = try_to_unmap_one(page, vma, ignore_refs);
+               ret = try_to_unmap_one(page, vma, migration);
                if (ret == SWAP_FAIL || !page_mapped(page))
                        break;
        }
@@ -765,7 +814,7 @@ static int try_to_unmap_anon(struct page *page, int ignore_refs)
  *
  * This function is only called from try_to_unmap for object-based pages.
  */
-static int try_to_unmap_file(struct page *page, int ignore_refs)
+static int try_to_unmap_file(struct page *page, int migration)
 {
        struct address_space *mapping = page->mapping;
        pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
@@ -779,7 +828,7 @@ static int try_to_unmap_file(struct page *page, int ignore_refs)
 
        spin_lock(&mapping->i_mmap_lock);
        vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
-               ret = try_to_unmap_one(page, vma, ignore_refs);
+               ret = try_to_unmap_one(page, vma, migration);
                if (ret == SWAP_FAIL || !page_mapped(page))
                        goto out;
        }
@@ -789,7 +838,7 @@ static int try_to_unmap_file(struct page *page, int ignore_refs)
 
        list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
                                                shared.vm_set.list) {
-               if (vma->vm_flags & VM_LOCKED)
+               if ((vma->vm_flags & VM_LOCKED) && !migration)
                        continue;
                cursor = (unsigned long) vma->vm_private_data;
                if (cursor > max_nl_cursor)
@@ -823,7 +872,7 @@ static int try_to_unmap_file(struct page *page, int ignore_refs)
        do {
                list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
                                                shared.vm_set.list) {
-                       if (vma->vm_flags & VM_LOCKED)
+                       if ((vma->vm_flags & VM_LOCKED) && !migration)
                                continue;
                        cursor = (unsigned long) vma->vm_private_data;
                        while ( cursor < max_nl_cursor &&
@@ -864,16 +913,16 @@ out:
  * SWAP_AGAIN  - we missed a mapping, try again later
  * SWAP_FAIL   - the page is unswappable
  */
-int try_to_unmap(struct page *page, int ignore_refs)
+int try_to_unmap(struct page *page, int migration)
 {
        int ret;
 
        BUG_ON(!PageLocked(page));
 
        if (PageAnon(page))
-               ret = try_to_unmap_anon(page, ignore_refs);
+               ret = try_to_unmap_anon(page, migration);
        else
-               ret = try_to_unmap_file(page, ignore_refs);
+               ret = try_to_unmap_file(page, migration);
 
        if (!page_mapped(page))
                ret = SWAP_SUCCESS;