mm: add orig_pte field into vm_fault
authorJan Kara <jack@suse.cz>
Wed, 14 Dec 2016 23:07:16 +0000 (15:07 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 15 Dec 2016 00:04:09 +0000 (16:04 -0800)
Add orig_pte field to vm_fault structure to allow ->page_mkwrite
handlers to fully handle the fault.

This also allows us to save some passing of extra arguments around.

Link: http://lkml.kernel.org/r/1479460644-25076-8-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/mm.h
mm/internal.h
mm/khugepaged.c
mm/memory.c

index 75fda0de64bf577c47f6fedb990a71acf881d240..39c17a2efcea085a76a1296ec1ff1814fd680bc5 100644 (file)
@@ -298,8 +298,8 @@ struct vm_fault {
        pgoff_t pgoff;                  /* Logical page offset based on vma */
        unsigned long address;          /* Faulting virtual address */
        pmd_t *pmd;                     /* Pointer to pmd entry matching
-                                        * the 'address'
-                                        */
+                                        * the 'address' */
+       pte_t orig_pte;                 /* Value of PTE at the time of fault */
 
        struct page *cow_page;          /* Handler may choose to COW */
        struct page *page;              /* ->fault handlers should return a
index 093b1eacc91b09090f8d04ae613bdef52f30d1c7..44d68895a9b9b29deefbed2be37e52eb94ba9f33 100644 (file)
@@ -36,7 +36,7 @@
 /* Do not use these with a slab allocator */
 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
 
-int do_swap_page(struct vm_fault *vmf, pte_t orig_pte);
+int do_swap_page(struct vm_fault *vmf);
 
 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
                unsigned long floor, unsigned long ceiling);
index f062d54c968353111e30dc2ceb4f48baf249de99..7434a63cac941756b4febe09f0e7de2fb07b3a28 100644 (file)
@@ -875,7 +875,6 @@ static bool __collapse_huge_page_swapin(struct mm_struct *mm,
                                        unsigned long address, pmd_t *pmd,
                                        int referenced)
 {
-       pte_t pteval;
        int swapped_in = 0, ret = 0;
        struct vm_fault vmf = {
                .vma = vma,
@@ -893,11 +892,11 @@ static bool __collapse_huge_page_swapin(struct mm_struct *mm,
        vmf.pte = pte_offset_map(pmd, address);
        for (; vmf.address < address + HPAGE_PMD_NR*PAGE_SIZE;
                        vmf.pte++, vmf.address += PAGE_SIZE) {
-               pteval = *vmf.pte;
-               if (!is_swap_pte(pteval))
+               vmf.orig_pte = *vmf.pte;
+               if (!is_swap_pte(vmf.orig_pte))
                        continue;
                swapped_in++;
-               ret = do_swap_page(&vmf, pteval);
+               ret = do_swap_page(&vmf);
 
                /* do_swap_page returns VM_FAULT_RETRY with released mmap_sem */
                if (ret & VM_FAULT_RETRY) {
index 7ba9cc58ddddbfcbd8a1b9c674e21709d90fa734..cf74f7ca911b23b8f8525ce1c9230a89c7fbf822 100644 (file)
@@ -2070,8 +2070,8 @@ static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page,
  * case, all we need to do here is to mark the page as writable and update
  * any related book-keeping.
  */
-static inline int wp_page_reuse(struct vm_fault *vmf, pte_t orig_pte,
-                       struct page *page, int page_mkwrite, int dirty_shared)
+static inline int wp_page_reuse(struct vm_fault *vmf, struct page *page,
+                               int page_mkwrite, int dirty_shared)
        __releases(vmf->ptl)
 {
        struct vm_area_struct *vma = vmf->vma;
@@ -2084,8 +2084,8 @@ static inline int wp_page_reuse(struct vm_fault *vmf, pte_t orig_pte,
        if (page)
                page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
 
-       flush_cache_page(vma, vmf->address, pte_pfn(orig_pte));
-       entry = pte_mkyoung(orig_pte);
+       flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
+       entry = pte_mkyoung(vmf->orig_pte);
        entry = maybe_mkwrite(pte_mkdirty(entry), vma);
        if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1))
                update_mmu_cache(vma, vmf->address, vmf->pte);
@@ -2135,8 +2135,7 @@ static inline int wp_page_reuse(struct vm_fault *vmf, pte_t orig_pte,
  *   held to the old page, as well as updating the rmap.
  * - In any case, unlock the PTL and drop the reference we took to the old page.
  */
-static int wp_page_copy(struct vm_fault *vmf, pte_t orig_pte,
-               struct page *old_page)
+static int wp_page_copy(struct vm_fault *vmf, struct page *old_page)
 {
        struct vm_area_struct *vma = vmf->vma;
        struct mm_struct *mm = vma->vm_mm;
@@ -2150,7 +2149,7 @@ static int wp_page_copy(struct vm_fault *vmf, pte_t orig_pte,
        if (unlikely(anon_vma_prepare(vma)))
                goto oom;
 
-       if (is_zero_pfn(pte_pfn(orig_pte))) {
+       if (is_zero_pfn(pte_pfn(vmf->orig_pte))) {
                new_page = alloc_zeroed_user_highpage_movable(vma,
                                                              vmf->address);
                if (!new_page)
@@ -2174,7 +2173,7 @@ static int wp_page_copy(struct vm_fault *vmf, pte_t orig_pte,
         * Re-check the pte - we dropped the lock
         */
        vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl);
-       if (likely(pte_same(*vmf->pte, orig_pte))) {
+       if (likely(pte_same(*vmf->pte, vmf->orig_pte))) {
                if (old_page) {
                        if (!PageAnon(old_page)) {
                                dec_mm_counter_fast(mm,
@@ -2184,7 +2183,7 @@ static int wp_page_copy(struct vm_fault *vmf, pte_t orig_pte,
                } else {
                        inc_mm_counter_fast(mm, MM_ANONPAGES);
                }
-               flush_cache_page(vma, vmf->address, pte_pfn(orig_pte));
+               flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
                entry = mk_pte(new_page, vma->vm_page_prot);
                entry = maybe_mkwrite(pte_mkdirty(entry), vma);
                /*
@@ -2268,7 +2267,7 @@ oom:
  * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
  * mapping
  */
-static int wp_pfn_shared(struct vm_fault *vmf, pte_t orig_pte)
+static int wp_pfn_shared(struct vm_fault *vmf)
 {
        struct vm_area_struct *vma = vmf->vma;
 
@@ -2286,16 +2285,15 @@ static int wp_pfn_shared(struct vm_fault *vmf, pte_t orig_pte)
                 * We might have raced with another page fault while we
                 * released the pte_offset_map_lock.
                 */
-               if (!pte_same(*vmf->pte, orig_pte)) {
+               if (!pte_same(*vmf->pte, vmf->orig_pte)) {
                        pte_unmap_unlock(vmf->pte, vmf->ptl);
                        return 0;
                }
        }
-       return wp_page_reuse(vmf, orig_pte, NULL, 0, 0);
+       return wp_page_reuse(vmf, NULL, 0, 0);
 }
 
-static int wp_page_shared(struct vm_fault *vmf, pte_t orig_pte,
-               struct page *old_page)
+static int wp_page_shared(struct vm_fault *vmf, struct page *old_page)
        __releases(vmf->ptl)
 {
        struct vm_area_struct *vma = vmf->vma;
@@ -2321,7 +2319,7 @@ static int wp_page_shared(struct vm_fault *vmf, pte_t orig_pte,
                 */
                vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
                                                vmf->address, &vmf->ptl);
-               if (!pte_same(*vmf->pte, orig_pte)) {
+               if (!pte_same(*vmf->pte, vmf->orig_pte)) {
                        unlock_page(old_page);
                        pte_unmap_unlock(vmf->pte, vmf->ptl);
                        put_page(old_page);
@@ -2330,7 +2328,7 @@ static int wp_page_shared(struct vm_fault *vmf, pte_t orig_pte,
                page_mkwrite = 1;
        }
 
-       return wp_page_reuse(vmf, orig_pte, old_page, page_mkwrite, 1);
+       return wp_page_reuse(vmf, old_page, page_mkwrite, 1);
 }
 
 /*
@@ -2351,13 +2349,13 @@ static int wp_page_shared(struct vm_fault *vmf, pte_t orig_pte,
  * but allow concurrent faults), with pte both mapped and locked.
  * We return with mmap_sem still held, but pte unmapped and unlocked.
  */
-static int do_wp_page(struct vm_fault *vmf, pte_t orig_pte)
+static int do_wp_page(struct vm_fault *vmf)
        __releases(vmf->ptl)
 {
        struct vm_area_struct *vma = vmf->vma;
        struct page *old_page;
 
-       old_page = vm_normal_page(vma, vmf->address, orig_pte);
+       old_page = vm_normal_page(vma, vmf->address, vmf->orig_pte);
        if (!old_page) {
                /*
                 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
@@ -2368,10 +2366,10 @@ static int do_wp_page(struct vm_fault *vmf, pte_t orig_pte)
                 */
                if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
                                     (VM_WRITE|VM_SHARED))
-                       return wp_pfn_shared(vmf, orig_pte);
+                       return wp_pfn_shared(vmf);
 
                pte_unmap_unlock(vmf->pte, vmf->ptl);
-               return wp_page_copy(vmf, orig_pte, old_page);
+               return wp_page_copy(vmf, old_page);
        }
 
        /*
@@ -2386,7 +2384,7 @@ static int do_wp_page(struct vm_fault *vmf, pte_t orig_pte)
                        lock_page(old_page);
                        vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
                                        vmf->address, &vmf->ptl);
-                       if (!pte_same(*vmf->pte, orig_pte)) {
+                       if (!pte_same(*vmf->pte, vmf->orig_pte)) {
                                unlock_page(old_page);
                                pte_unmap_unlock(vmf->pte, vmf->ptl);
                                put_page(old_page);
@@ -2406,12 +2404,12 @@ static int do_wp_page(struct vm_fault *vmf, pte_t orig_pte)
                                page_move_anon_rmap(old_page, vma);
                        }
                        unlock_page(old_page);
-                       return wp_page_reuse(vmf, orig_pte, old_page, 0, 0);
+                       return wp_page_reuse(vmf, old_page, 0, 0);
                }
                unlock_page(old_page);
        } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
                                        (VM_WRITE|VM_SHARED))) {
-               return wp_page_shared(vmf, orig_pte, old_page);
+               return wp_page_shared(vmf, old_page);
        }
 
        /*
@@ -2420,7 +2418,7 @@ static int do_wp_page(struct vm_fault *vmf, pte_t orig_pte)
        get_page(old_page);
 
        pte_unmap_unlock(vmf->pte, vmf->ptl);
-       return wp_page_copy(vmf, orig_pte, old_page);
+       return wp_page_copy(vmf, old_page);
 }
 
 static void unmap_mapping_range_vma(struct vm_area_struct *vma,
@@ -2508,7 +2506,7 @@ EXPORT_SYMBOL(unmap_mapping_range);
  * We return with the mmap_sem locked or unlocked in the same cases
  * as does filemap_fault().
  */
-int do_swap_page(struct vm_fault *vmf, pte_t orig_pte)
+int do_swap_page(struct vm_fault *vmf)
 {
        struct vm_area_struct *vma = vmf->vma;
        struct page *page, *swapcache;
@@ -2519,10 +2517,10 @@ int do_swap_page(struct vm_fault *vmf, pte_t orig_pte)
        int exclusive = 0;
        int ret = 0;
 
-       if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, orig_pte))
+       if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte))
                goto out;
 
-       entry = pte_to_swp_entry(orig_pte);
+       entry = pte_to_swp_entry(vmf->orig_pte);
        if (unlikely(non_swap_entry(entry))) {
                if (is_migration_entry(entry)) {
                        migration_entry_wait(vma->vm_mm, vmf->pmd,
@@ -2530,7 +2528,7 @@ int do_swap_page(struct vm_fault *vmf, pte_t orig_pte)
                } else if (is_hwpoison_entry(entry)) {
                        ret = VM_FAULT_HWPOISON;
                } else {
-                       print_bad_pte(vma, vmf->address, orig_pte, NULL);
+                       print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL);
                        ret = VM_FAULT_SIGBUS;
                }
                goto out;
@@ -2547,7 +2545,7 @@ int do_swap_page(struct vm_fault *vmf, pte_t orig_pte)
                         */
                        vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
                                        vmf->address, &vmf->ptl);
-                       if (likely(pte_same(*vmf->pte, orig_pte)))
+                       if (likely(pte_same(*vmf->pte, vmf->orig_pte)))
                                ret = VM_FAULT_OOM;
                        delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
                        goto unlock;
@@ -2604,7 +2602,7 @@ int do_swap_page(struct vm_fault *vmf, pte_t orig_pte)
         */
        vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
                        &vmf->ptl);
-       if (unlikely(!pte_same(*vmf->pte, orig_pte)))
+       if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte)))
                goto out_nomap;
 
        if (unlikely(!PageUptodate(page))) {
@@ -2632,9 +2630,10 @@ int do_swap_page(struct vm_fault *vmf, pte_t orig_pte)
                exclusive = RMAP_EXCLUSIVE;
        }
        flush_icache_page(vma, page);
-       if (pte_swp_soft_dirty(orig_pte))
+       if (pte_swp_soft_dirty(vmf->orig_pte))
                pte = pte_mksoft_dirty(pte);
        set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte);
+       vmf->orig_pte = pte;
        if (page == swapcache) {
                do_page_add_anon_rmap(page, vma, vmf->address, exclusive);
                mem_cgroup_commit_charge(page, memcg, true, false);
@@ -2664,7 +2663,7 @@ int do_swap_page(struct vm_fault *vmf, pte_t orig_pte)
        }
 
        if (vmf->flags & FAULT_FLAG_WRITE) {
-               ret |= do_wp_page(vmf, pte);
+               ret |= do_wp_page(vmf);
                if (ret & VM_FAULT_ERROR)
                        ret &= VM_FAULT_ERROR;
                goto out;
@@ -3363,7 +3362,7 @@ static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
        return mpol_misplaced(page, vma, addr);
 }
 
-static int do_numa_page(struct vm_fault *vmf, pte_t pte)
+static int do_numa_page(struct vm_fault *vmf)
 {
        struct vm_area_struct *vma = vmf->vma;
        struct page *page = NULL;
@@ -3371,6 +3370,7 @@ static int do_numa_page(struct vm_fault *vmf, pte_t pte)
        int last_cpupid;
        int target_nid;
        bool migrated = false;
+       pte_t pte = vmf->orig_pte;
        bool was_writable = pte_write(pte);
        int flags = 0;
 
@@ -3521,8 +3521,7 @@ static int handle_pte_fault(struct vm_fault *vmf)
                 * So now it's safe to run pte_offset_map().
                 */
                vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
-
-               entry = *vmf->pte;
+               vmf->orig_pte = *vmf->pte;
 
                /*
                 * some architectures can have larger ptes than wordsize,
@@ -3533,7 +3532,7 @@ static int handle_pte_fault(struct vm_fault *vmf)
                 * ptl lock held. So here a barrier will do.
                 */
                barrier();
-               if (pte_none(entry)) {
+               if (pte_none(vmf->orig_pte)) {
                        pte_unmap(vmf->pte);
                        vmf->pte = NULL;
                }
@@ -3546,19 +3545,20 @@ static int handle_pte_fault(struct vm_fault *vmf)
                        return do_fault(vmf);
        }
 
-       if (!pte_present(entry))
-               return do_swap_page(vmf, entry);
+       if (!pte_present(vmf->orig_pte))
+               return do_swap_page(vmf);
 
-       if (pte_protnone(entry) && vma_is_accessible(vmf->vma))
-               return do_numa_page(vmf, entry);
+       if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma))
+               return do_numa_page(vmf);
 
        vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
        spin_lock(vmf->ptl);
+       entry = vmf->orig_pte;
        if (unlikely(!pte_same(*vmf->pte, entry)))
                goto unlock;
        if (vmf->flags & FAULT_FLAG_WRITE) {
                if (!pte_write(entry))
-                       return do_wp_page(vmf, entry);
+                       return do_wp_page(vmf);
                entry = pte_mkdirty(entry);
        }
        entry = pte_mkyoung(entry);