Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6-microblaze.git] / mm / memory.c
index 0ec7bc6..109e986 100644 (file)
@@ -47,6 +47,7 @@
 #include <linux/pagemap.h>
 #include <linux/rmap.h>
 #include <linux/module.h>
+#include <linux/delayacct.h>
 #include <linux/init.h>
 
 #include <asm/pgalloc.h>
@@ -126,7 +127,7 @@ static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
        pmd_clear(pmd);
        pte_lock_deinit(page);
        pte_free_tlb(tlb, page);
-       dec_page_state(nr_page_table_pages);
+       dec_zone_page_state(page, NR_PAGETABLE);
        tlb->mm->nr_ptes--;
 }
 
@@ -311,7 +312,7 @@ int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
                pte_free(new);
        } else {
                mm->nr_ptes++;
-               inc_page_state(nr_page_table_pages);
+               inc_zone_page_state(new, NR_PAGETABLE);
                pmd_populate(mm, pmd, new);
        }
        spin_unlock(&mm->page_table_lock);
@@ -434,7 +435,9 @@ copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
        /* pte contains position in swap or file, so copy. */
        if (unlikely(!pte_present(pte))) {
                if (!pte_file(pte)) {
-                       swap_duplicate(pte_to_swp_entry(pte));
+                       swp_entry_t entry = pte_to_swp_entry(pte);
+
+                       swap_duplicate(entry);
                        /* make sure dst_mm is on swapoff's mmlist. */
                        if (unlikely(list_empty(&dst_mm->mmlist))) {
                                spin_lock(&mmlist_lock);
@@ -443,6 +446,16 @@ copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
                                                 &src_mm->mmlist);
                                spin_unlock(&mmlist_lock);
                        }
+                       if (is_write_migration_entry(entry) &&
+                                       is_cow_mapping(vm_flags)) {
+                               /*
+                                * COW mappings require pages in both parent
+                                * and child to be set to read.
+                                */
+                               make_migration_entry_read(&entry);
+                               pte = swp_entry_to_pte(entry);
+                               set_pte_at(src_mm, addr, src_pte, pte);
+                       }
                }
                goto out_set_pte;
        }
@@ -491,7 +504,7 @@ again:
                return -ENOMEM;
        src_pte = pte_offset_map_nested(src_pmd, addr);
        src_ptl = pte_lockptr(src_mm, src_pmd);
-       spin_lock(src_ptl);
+       spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
 
        do {
                /*
@@ -1445,25 +1458,60 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
 {
        struct page *old_page, *new_page;
        pte_t entry;
-       int ret = VM_FAULT_MINOR;
+       int reuse, ret = VM_FAULT_MINOR;
 
        old_page = vm_normal_page(vma, address, orig_pte);
        if (!old_page)
                goto gotten;
 
-       if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
-               int reuse = can_share_swap_page(old_page);
-               unlock_page(old_page);
-               if (reuse) {
-                       flush_cache_page(vma, address, pte_pfn(orig_pte));
-                       entry = pte_mkyoung(orig_pte);
-                       entry = maybe_mkwrite(pte_mkdirty(entry), vma);
-                       ptep_set_access_flags(vma, address, page_table, entry, 1);
-                       update_mmu_cache(vma, address, entry);
-                       lazy_mmu_prot_update(entry);
-                       ret |= VM_FAULT_WRITE;
-                       goto unlock;
+       if (unlikely((vma->vm_flags & (VM_SHARED|VM_WRITE)) ==
+                               (VM_SHARED|VM_WRITE))) {
+               if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
+                       /*
+                        * Notify the address space that the page is about to
+                        * become writable so that it can prohibit this or wait
+                        * for the page to get into an appropriate state.
+                        *
+                        * We do this without the lock held, so that it can
+                        * sleep if it needs to.
+                        */
+                       page_cache_get(old_page);
+                       pte_unmap_unlock(page_table, ptl);
+
+                       if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
+                               goto unwritable_page;
+
+                       page_cache_release(old_page);
+
+                       /*
+                        * Since we dropped the lock we need to revalidate
+                        * the PTE as someone else may have changed it.  If
+                        * they did, we just return, as we can count on the
+                        * MMU to tell us if they didn't also make it writable.
+                        */
+                       page_table = pte_offset_map_lock(mm, pmd, address,
+                                                        &ptl);
+                       if (!pte_same(*page_table, orig_pte))
+                               goto unlock;
                }
+
+               reuse = 1;
+       } else if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
+               reuse = can_share_swap_page(old_page);
+               unlock_page(old_page);
+       } else {
+               reuse = 0;
+       }
+
+       if (reuse) {
+               flush_cache_page(vma, address, pte_pfn(orig_pte));
+               entry = pte_mkyoung(orig_pte);
+               entry = maybe_mkwrite(pte_mkdirty(entry), vma);
+               ptep_set_access_flags(vma, address, page_table, entry, 1);
+               update_mmu_cache(vma, address, entry);
+               lazy_mmu_prot_update(entry);
+               ret |= VM_FAULT_WRITE;
+               goto unlock;
        }
 
        /*
@@ -1502,9 +1550,9 @@ gotten:
                flush_cache_page(vma, address, pte_pfn(orig_pte));
                entry = mk_pte(new_page, vma->vm_page_prot);
                entry = maybe_mkwrite(pte_mkdirty(entry), vma);
+               lazy_mmu_prot_update(entry);
                ptep_establish(vma, address, page_table, entry);
                update_mmu_cache(vma, address, entry);
-               lazy_mmu_prot_update(entry);
                lru_cache_add_active(new_page);
                page_add_new_anon_rmap(new_page, vma, address);
 
@@ -1523,6 +1571,10 @@ oom:
        if (old_page)
                page_cache_release(old_page);
        return VM_FAULT_OOM;
+
+unwritable_page:
+       page_cache_release(old_page);
+       return VM_FAULT_SIGBUS;
 }
 
 /*
@@ -1802,7 +1854,7 @@ int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
 
        return 0;
 }
-EXPORT_SYMBOL(vmtruncate_range);
+EXPORT_UNUSED_SYMBOL(vmtruncate_range);  /*  June 2006  */
 
 /* 
  * Primitive swap readahead code. We simply read an aligned block of
@@ -1879,7 +1931,11 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
                goto out;
 
        entry = pte_to_swp_entry(orig_pte);
-again:
+       if (is_migration_entry(entry)) {
+               migration_entry_wait(mm, pmd, address);
+               goto out;
+       }
+       delayacct_set_flag(DELAYACCT_PF_SWAPIN);
        page = lookup_swap_cache(entry);
        if (!page) {
                swapin_readahead(entry, address, vma);
@@ -1892,23 +1948,19 @@ again:
                        page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
                        if (likely(pte_same(*page_table, orig_pte)))
                                ret = VM_FAULT_OOM;
+                       delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
                        goto unlock;
                }
 
                /* Had to read the page from swap area: Major fault */
                ret = VM_FAULT_MAJOR;
-               inc_page_state(pgmajfault);
+               count_vm_event(PGMAJFAULT);
                grab_swap_token();
        }
 
+       delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
        mark_page_accessed(page);
        lock_page(page);
-       if (!PageSwapCache(page)) {
-               /* Page migration has occured */
-               unlock_page(page);
-               page_cache_release(page);
-               goto again;
-       }
 
        /*
         * Back out if somebody else already faulted in this pte.
@@ -2074,18 +2126,31 @@ retry:
        /*
         * Should we do an early C-O-W break?
         */
-       if (write_access && !(vma->vm_flags & VM_SHARED)) {
-               struct page *page;
+       if (write_access) {
+               if (!(vma->vm_flags & VM_SHARED)) {
+                       struct page *page;
 
-               if (unlikely(anon_vma_prepare(vma)))
-                       goto oom;
-               page = alloc_page_vma(GFP_HIGHUSER, vma, address);
-               if (!page)
-                       goto oom;
-               copy_user_highpage(page, new_page, address);
-               page_cache_release(new_page);
-               new_page = page;
-               anon = 1;
+                       if (unlikely(anon_vma_prepare(vma)))
+                               goto oom;
+                       page = alloc_page_vma(GFP_HIGHUSER, vma, address);
+                       if (!page)
+                               goto oom;
+                       copy_user_highpage(page, new_page, address);
+                       page_cache_release(new_page);
+                       new_page = page;
+                       anon = 1;
+
+               } else {
+                       /* if the page will be shareable, see if the backing
+                        * address space wants to know that the page is about
+                        * to become writable */
+                       if (vma->vm_ops->page_mkwrite &&
+                           vma->vm_ops->page_mkwrite(vma, new_page) < 0
+                           ) {
+                               page_cache_release(new_page);
+                               return VM_FAULT_SIGBUS;
+                       }
+               }
        }
 
        page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
@@ -2263,7 +2328,7 @@ int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 
        __set_current_state(TASK_RUNNING);
 
-       inc_page_state(pgfault);
+       count_vm_event(PGFAULT);
 
        if (unlikely(is_vm_hugetlb_page(vma)))
                return hugetlb_fault(mm, vma, address, write_access);