thp: change pmd_trans_huge_lock() interface to return ptl
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Fri, 22 Jan 2016 00:40:25 +0000 (16:40 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 22 Jan 2016 01:20:51 +0000 (17:20 -0800)
After THP refcounting rework we have only two possible return values
from pmd_trans_huge_lock(): success and failure.  Return-by-pointer for
ptl doesn't make much sense in this case.

Let's convert pmd_trans_huge_lock() to return ptl on success and NULL on
failure.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/task_mmu.c
include/linux/huge_mm.h
mm/huge_memory.c
mm/memcontrol.c
mm/mincore.c

index 71ffc91060f6d6ad32081bfa53d2a3c7b3706e53..85d16c67c33eaa8b7a5ac65e38ec465814916619 100644 (file)
@@ -602,7 +602,8 @@ static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
        pte_t *pte;
        spinlock_t *ptl;
 
-       if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
+       ptl = pmd_trans_huge_lock(pmd, vma);
+       if (ptl) {
                smaps_pmd_entry(pmd, addr, walk);
                spin_unlock(ptl);
                return 0;
@@ -913,7 +914,8 @@ static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
        spinlock_t *ptl;
        struct page *page;
 
-       if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
+       ptl = pmd_trans_huge_lock(pmd, vma);
+       if (ptl) {
                if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
                        clear_soft_dirty_pmd(vma, addr, pmd);
                        goto out;
@@ -1187,7 +1189,8 @@ static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
        int err = 0;
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-       if (pmd_trans_huge_lock(pmdp, vma, &ptl)) {
+       ptl = pmd_trans_huge_lock(pmdp, vma);
+       if (ptl) {
                u64 flags = 0, frame = 0;
                pmd_t pmd = *pmdp;
 
@@ -1519,7 +1522,8 @@ static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
        pte_t *orig_pte;
        pte_t *pte;
 
-       if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
+       ptl = pmd_trans_huge_lock(pmd, vma);
+       if (ptl) {
                pte_t huge_pte = *(pte_t *)pmd;
                struct page *page;
 
index cfe81e10bd5429baf5ed0ccf50bd4613ff6ee3c6..459fd25b378e73cfd2e911761ad845076be547de 100644 (file)
@@ -120,15 +120,15 @@ extern void vma_adjust_trans_huge(struct vm_area_struct *vma,
                                    unsigned long start,
                                    unsigned long end,
                                    long adjust_next);
-extern bool __pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma,
-               spinlock_t **ptl);
+extern spinlock_t *__pmd_trans_huge_lock(pmd_t *pmd,
+               struct vm_area_struct *vma);
 /* mmap_sem must be held on entry */
-static inline bool pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma,
-               spinlock_t **ptl)
+static inline spinlock_t *pmd_trans_huge_lock(pmd_t *pmd,
+               struct vm_area_struct *vma)
 {
        VM_BUG_ON_VMA(!rwsem_is_locked(&vma->vm_mm->mmap_sem), vma);
        if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd))
-               return __pmd_trans_huge_lock(pmd, vma, ptl);
+               return __pmd_trans_huge_lock(pmd, vma);
        else
                return false;
 }
@@ -190,10 +190,10 @@ static inline void vma_adjust_trans_huge(struct vm_area_struct *vma,
                                         long adjust_next)
 {
 }
-static inline bool pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma,
-               spinlock_t **ptl)
+static inline spinlock_t *pmd_trans_huge_lock(pmd_t *pmd,
+               struct vm_area_struct *vma)
 {
-       return false;
+       return NULL;
 }
 
 static inline int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
index 8ad580273521046904ec13478e7b897839d955d7..2d1ffe9d0e2632295319f562a452eecd5e4fcd49 100644 (file)
@@ -1560,7 +1560,8 @@ int madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
        struct mm_struct *mm = tlb->mm;
        int ret = 0;
 
-       if (!pmd_trans_huge_lock(pmd, vma, &ptl))
+       ptl = pmd_trans_huge_lock(pmd, vma);
+       if (!ptl)
                goto out_unlocked;
 
        orig_pmd = *pmd;
@@ -1627,7 +1628,8 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
        pmd_t orig_pmd;
        spinlock_t *ptl;
 
-       if (!__pmd_trans_huge_lock(pmd, vma, &ptl))
+       ptl = __pmd_trans_huge_lock(pmd, vma);
+       if (!ptl)
                return 0;
        /*
         * For architectures like ppc64 we look at deposited pgtable
@@ -1690,7 +1692,8 @@ bool move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
         * We don't have to worry about the ordering of src and dst
         * ptlocks because exclusive mmap_sem prevents deadlock.
         */
-       if (__pmd_trans_huge_lock(old_pmd, vma, &old_ptl)) {
+       old_ptl = __pmd_trans_huge_lock(old_pmd, vma);
+       if (old_ptl) {
                new_ptl = pmd_lockptr(mm, new_pmd);
                if (new_ptl != old_ptl)
                        spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
@@ -1724,7 +1727,8 @@ int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
        spinlock_t *ptl;
        int ret = 0;
 
-       if (__pmd_trans_huge_lock(pmd, vma, &ptl)) {
+       ptl = __pmd_trans_huge_lock(pmd, vma);
+       if (ptl) {
                pmd_t entry;
                bool preserve_write = prot_numa && pmd_write(*pmd);
                ret = 1;
@@ -1760,14 +1764,14 @@ int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
  * Note that if it returns true, this routine returns without unlocking page
  * table lock. So callers must unlock it.
  */
-bool __pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma,
-               spinlock_t **ptl)
+spinlock_t *__pmd_trans_huge_lock(pmd_t *pmd, struct vm_area_struct *vma)
 {
-       *ptl = pmd_lock(vma->vm_mm, pmd);
+       spinlock_t *ptl;
+       ptl = pmd_lock(vma->vm_mm, pmd);
        if (likely(pmd_trans_huge(*pmd) || pmd_devmap(*pmd)))
-               return true;
-       spin_unlock(*ptl);
-       return false;
+               return ptl;
+       spin_unlock(ptl);
+       return NULL;
 }
 
 #define VM_NO_THP (VM_SPECIAL | VM_HUGETLB | VM_SHARED | VM_MAYSHARE)
index ca052f2a4a0b374eb9a2a6498b407be7da28ac48..d06cae2de783acf462162e27f8770ff7bf60f4ad 100644 (file)
@@ -4638,7 +4638,8 @@ static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
        pte_t *pte;
        spinlock_t *ptl;
 
-       if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
+       ptl = pmd_trans_huge_lock(pmd, vma);
+       if (ptl) {
                if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
                        mc.precharge += HPAGE_PMD_NR;
                spin_unlock(ptl);
@@ -4826,7 +4827,8 @@ static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
        union mc_target target;
        struct page *page;
 
-       if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
+       ptl = pmd_trans_huge_lock(pmd, vma);
+       if (ptl) {
                if (mc.precharge < HPAGE_PMD_NR) {
                        spin_unlock(ptl);
                        return 0;
index 2a565ed8bb4907398a0d3d2619fd4939df777970..563f320454902df04e782f73aa8d0473656034c9 100644 (file)
@@ -117,7 +117,8 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
        unsigned char *vec = walk->private;
        int nr = (end - addr) >> PAGE_SHIFT;
 
-       if (pmd_trans_huge_lock(pmd, vma, &ptl)) {
+       ptl = pmd_trans_huge_lock(pmd, vma);
+       if (ptl) {
                memset(vec, 1, nr);
                spin_unlock(ptl);
                goto out;