mm, oom: change type of oom_score_adj to short
[linux-3.10.git] / mm / ksm.c
index 942dfc7..b4d5a9d 100644 (file)
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -35,6 +35,7 @@
 #include <linux/ksm.h>
 #include <linux/hash.h>
 #include <linux/freezer.h>
+#include <linux/oom.h>
 
 #include <asm/tlbflush.h>
 #include "internal.h"
@@ -373,6 +374,20 @@ static int break_ksm(struct vm_area_struct *vma, unsigned long addr)
        return (ret & VM_FAULT_OOM) ? -ENOMEM : 0;
 }
 
+static struct vm_area_struct *find_mergeable_vma(struct mm_struct *mm,
+               unsigned long addr)
+{
+       struct vm_area_struct *vma;
+       if (ksm_test_exit(mm))
+               return NULL;
+       vma = find_vma(mm, addr);
+       if (!vma || vma->vm_start > addr)
+               return NULL;
+       if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
+               return NULL;
+       return vma;
+}
+
 static void break_cow(struct rmap_item *rmap_item)
 {
        struct mm_struct *mm = rmap_item->mm;
@@ -386,15 +401,9 @@ static void break_cow(struct rmap_item *rmap_item)
        put_anon_vma(rmap_item->anon_vma);
 
        down_read(&mm->mmap_sem);
-       if (ksm_test_exit(mm))
-               goto out;
-       vma = find_vma(mm, addr);
-       if (!vma || vma->vm_start > addr)
-               goto out;
-       if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
-               goto out;
-       break_ksm(vma, addr);
-out:
+       vma = find_mergeable_vma(mm, addr);
+       if (vma)
+               break_ksm(vma, addr);
        up_read(&mm->mmap_sem);
 }
 
@@ -420,12 +429,8 @@ static struct page *get_mergeable_page(struct rmap_item *rmap_item)
        struct page *page;
 
        down_read(&mm->mmap_sem);
-       if (ksm_test_exit(mm))
-               goto out;
-       vma = find_vma(mm, addr);
-       if (!vma || vma->vm_start > addr)
-               goto out;
-       if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
+       vma = find_mergeable_vma(mm, addr);
+       if (!vma)
                goto out;
 
        page = follow_page(vma, addr, FOLL_GET);
@@ -671,9 +676,9 @@ error:
 static u32 calc_checksum(struct page *page)
 {
        u32 checksum;
-       void *addr = kmap_atomic(page, KM_USER0);
+       void *addr = kmap_atomic(page);
        checksum = jhash2(addr, PAGE_SIZE / 4, 17);
-       kunmap_atomic(addr, KM_USER0);
+       kunmap_atomic(addr);
        return checksum;
 }
 
@@ -682,11 +687,11 @@ static int memcmp_pages(struct page *page1, struct page *page2)
        char *addr1, *addr2;
        int ret;
 
-       addr1 = kmap_atomic(page1, KM_USER0);
-       addr2 = kmap_atomic(page2, KM_USER1);
+       addr1 = kmap_atomic(page1);
+       addr2 = kmap_atomic(page2);
        ret = memcmp(addr1, addr2, PAGE_SIZE);
-       kunmap_atomic(addr2, KM_USER1);
-       kunmap_atomic(addr1, KM_USER0);
+       kunmap_atomic(addr2);
+       kunmap_atomic(addr1);
        return ret;
 }
 
@@ -704,15 +709,22 @@ static int write_protect_page(struct vm_area_struct *vma, struct page *page,
        spinlock_t *ptl;
        int swapped;
        int err = -EFAULT;
+       unsigned long mmun_start;       /* For mmu_notifiers */
+       unsigned long mmun_end;         /* For mmu_notifiers */
 
        addr = page_address_in_vma(page, vma);
        if (addr == -EFAULT)
                goto out;
 
        BUG_ON(PageTransCompound(page));
+
+       mmun_start = addr;
+       mmun_end   = addr + PAGE_SIZE;
+       mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
+
        ptep = page_check_address(page, mm, addr, &ptl, 0);
        if (!ptep)
-               goto out;
+               goto out_mn;
 
        if (pte_write(*ptep) || pte_dirty(*ptep)) {
                pte_t entry;
@@ -747,6 +759,8 @@ static int write_protect_page(struct vm_area_struct *vma, struct page *page,
 
 out_unlock:
        pte_unmap_unlock(ptep, ptl);
+out_mn:
+       mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 out:
        return err;
 }
@@ -764,35 +778,31 @@ static int replace_page(struct vm_area_struct *vma, struct page *page,
                        struct page *kpage, pte_t orig_pte)
 {
        struct mm_struct *mm = vma->vm_mm;
-       pgd_t *pgd;
-       pud_t *pud;
        pmd_t *pmd;
        pte_t *ptep;
        spinlock_t *ptl;
        unsigned long addr;
        int err = -EFAULT;
+       unsigned long mmun_start;       /* For mmu_notifiers */
+       unsigned long mmun_end;         /* For mmu_notifiers */
 
        addr = page_address_in_vma(page, vma);
        if (addr == -EFAULT)
                goto out;
 
-       pgd = pgd_offset(mm, addr);
-       if (!pgd_present(*pgd))
-               goto out;
-
-       pud = pud_offset(pgd, addr);
-       if (!pud_present(*pud))
+       pmd = mm_find_pmd(mm, addr);
+       if (!pmd)
                goto out;
-
-       pmd = pmd_offset(pud, addr);
        BUG_ON(pmd_trans_huge(*pmd));
-       if (!pmd_present(*pmd))
-               goto out;
+
+       mmun_start = addr;
+       mmun_end   = addr + PAGE_SIZE;
+       mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
 
        ptep = pte_offset_map_lock(mm, pmd, addr, &ptl);
        if (!pte_same(*ptep, orig_pte)) {
                pte_unmap_unlock(ptep, ptl);
-               goto out;
+               goto out_mn;
        }
 
        get_page(kpage);
@@ -809,6 +819,8 @@ static int replace_page(struct vm_area_struct *vma, struct page *page,
 
        pte_unmap_unlock(ptep, ptl);
        err = 0;
+out_mn:
+       mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
 out:
        return err;
 }
@@ -1301,6 +1313,12 @@ static struct rmap_item *scan_get_next_rmap_item(struct page **page)
                slot = list_entry(slot->mm_list.next, struct mm_slot, mm_list);
                ksm_scan.mm_slot = slot;
                spin_unlock(&ksm_mmlist_lock);
+               /*
+                * Although we tested list_empty() above, a racing __ksm_exit
+                * of the last mm on the list may have removed it since then.
+                */
+               if (slot == &ksm_mm_head)
+                       return NULL;
 next_mm:
                ksm_scan.address = 0;
                ksm_scan.rmap_list = &slot->rmap_list;
@@ -1458,10 +1476,14 @@ int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
                 */
                if (*vm_flags & (VM_MERGEABLE | VM_SHARED  | VM_MAYSHARE   |
                                 VM_PFNMAP    | VM_IO      | VM_DONTEXPAND |
-                                VM_RESERVED  | VM_HUGETLB | VM_INSERTPAGE |
-                                VM_NONLINEAR | VM_MIXEDMAP | VM_SAO))
+                                VM_HUGETLB | VM_NONLINEAR | VM_MIXEDMAP))
                        return 0;               /* just ignore the advice */
 
+#ifdef VM_SAO
+               if (*vm_flags & VM_SAO)
+                       return 0;
+#endif
+
                if (!test_bit(MMF_VM_MERGEABLE, &mm->flags)) {
                        err = __ksm_enter(mm);
                        if (err)
@@ -1571,7 +1593,7 @@ struct page *ksm_does_need_to_copy(struct page *page,
                SetPageSwapBacked(new_page);
                __set_page_locked(new_page);
 
-               if (page_evictable(new_page, vma))
+               if (!mlocked_vma_newpage(vma, new_page))
                        lru_cache_add_lru(new_page, LRU_ACTIVE_ANON);
                else
                        add_page_to_unevictable_list(new_page);
@@ -1603,7 +1625,8 @@ again:
                struct vm_area_struct *vma;
 
                anon_vma_lock(anon_vma);
-               list_for_each_entry(vmac, &anon_vma->head, same_anon_vma) {
+               anon_vma_interval_tree_foreach(vmac, &anon_vma->rb_root,
+                                              0, ULONG_MAX) {
                        vma = vmac->vma;
                        if (rmap_item->address < vma->vm_start ||
                            rmap_item->address >= vma->vm_end)
@@ -1656,7 +1679,8 @@ again:
                struct vm_area_struct *vma;
 
                anon_vma_lock(anon_vma);
-               list_for_each_entry(vmac, &anon_vma->head, same_anon_vma) {
+               anon_vma_interval_tree_foreach(vmac, &anon_vma->rb_root,
+                                              0, ULONG_MAX) {
                        vma = vmac->vma;
                        if (rmap_item->address < vma->vm_start ||
                            rmap_item->address >= vma->vm_end)
@@ -1708,7 +1732,8 @@ again:
                struct vm_area_struct *vma;
 
                anon_vma_lock(anon_vma);
-               list_for_each_entry(vmac, &anon_vma->head, same_anon_vma) {
+               anon_vma_interval_tree_foreach(vmac, &anon_vma->rb_root,
+                                              0, ULONG_MAX) {
                        vma = vmac->vma;
                        if (rmap_item->address < vma->vm_start ||
                            rmap_item->address >= vma->vm_end)
@@ -1894,9 +1919,12 @@ static ssize_t run_store(struct kobject *kobj, struct kobj_attribute *attr,
        if (ksm_run != flags) {
                ksm_run = flags;
                if (flags & KSM_RUN_UNMERGE) {
-                       current->flags |= PF_OOM_ORIGIN;
+                       short oom_score_adj;
+
+                       oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
                        err = unmerge_and_remove_all_rmap_items();
-                       current->flags &= ~PF_OOM_ORIGIN;
+                       compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX,
+                                                               oom_score_adj);
                        if (err) {
                                ksm_run = KSM_RUN_STOP;
                                count = err;