]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - mm/hugetlb.c
PCI: remove unused resource assignment in pci_read_bridge_bases()
[linux-2.6-omap-h63xx.git] / mm / hugetlb.c
index d237a02eb2289774ded6ea2c0767f779d5b76e9b..ce8cbb29860bd1b867454014195fe497332e2f61 100644 (file)
@@ -17,7 +17,7 @@
 #include <linux/mutex.h>
 #include <linux/bootmem.h>
 #include <linux/sysfs.h>
-#include <asm/io.h>
+
 #include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/io.h>
@@ -262,7 +262,7 @@ struct resv_map {
        struct list_head regions;
 };
 
-struct resv_map *resv_map_alloc(void)
+static struct resv_map *resv_map_alloc(void)
 {
        struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL);
        if (!resv_map)
@@ -274,7 +274,7 @@ struct resv_map *resv_map_alloc(void)
        return resv_map;
 }
 
-void resv_map_release(struct kref *ref)
+static void resv_map_release(struct kref *ref)
 {
        struct resv_map *resv_map = container_of(ref, struct resv_map, refs);
 
@@ -289,7 +289,7 @@ static struct resv_map *vma_resv_map(struct vm_area_struct *vma)
        if (!(vma->vm_flags & VM_SHARED))
                return (struct resv_map *)(get_vma_private_data(vma) &
                                                        ~HPAGE_RESV_MASK);
-       return 0;
+       return NULL;
 }
 
 static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map)
@@ -565,7 +565,7 @@ static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid)
                huge_page_order(h));
        if (page) {
                if (arch_prepare_hugepage(page)) {
-                       __free_pages(page, HUGETLB_PAGE_ORDER);
+                       __free_pages(page, huge_page_order(h));
                        return NULL;
                }
                prep_new_huge_page(h, page, nid);
@@ -665,6 +665,11 @@ static struct page *alloc_buddy_huge_page(struct hstate *h,
                                        __GFP_REPEAT|__GFP_NOWARN,
                                        huge_page_order(h));
 
+       if (page && arch_prepare_hugepage(page)) {
+               __free_pages(page, huge_page_order(h));
+               return NULL;
+       }
+
        spin_lock(&hugetlb_lock);
        if (page) {
                /*
@@ -1283,7 +1288,12 @@ module_exit(hugetlb_exit);
 
 static int __init hugetlb_init(void)
 {
-       BUILD_BUG_ON(HPAGE_SHIFT == 0);
+       /* Some platform decide whether they support huge pages at boot
+        * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when
+        * there is no such support
+        */
+       if (HPAGE_SHIFT == 0)
+               return 0;
 
        if (!size_to_hstate(default_hstate_size)) {
                default_hstate_size = HPAGE_SIZE;
@@ -1449,11 +1459,11 @@ int hugetlb_report_meminfo(char *buf)
 {
        struct hstate *h = &default_hstate;
        return sprintf(buf,
-                       "HugePages_Total: %5lu\n"
-                       "HugePages_Free:  %5lu\n"
-                       "HugePages_Rsvd:  %5lu\n"
-                       "HugePages_Surp:  %5lu\n"
-                       "Hugepagesize:    %5lu kB\n",
+                       "HugePages_Total:   %5lu\n"
+                       "HugePages_Free:    %5lu\n"
+                       "HugePages_Rsvd:    %5lu\n"
+                       "HugePages_Surp:    %5lu\n"
+                       "Hugepagesize:   %8lu kB\n",
                        h->nr_huge_pages,
                        h->free_huge_pages,
                        h->resv_huge_pages,
@@ -1737,10 +1747,8 @@ void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start,
  * from other VMAs and let the children be SIGKILLed if they are faulting the
  * same region.
  */
-int unmap_ref_private(struct mm_struct *mm,
-                                       struct vm_area_struct *vma,
-                                       struct page *page,
-                                       unsigned long address)
+static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma,
+                               struct page *page, unsigned long address)
 {
        struct vm_area_struct *iter_vma;
        struct address_space *mapping;
@@ -1932,6 +1940,18 @@ retry:
                        lock_page(page);
        }
 
+       /*
+        * If we are going to COW a private mapping later, we examine the
+        * pending reservations for this page now. This will ensure that
+        * any allocations necessary to record that reservation occur outside
+        * the spinlock.
+        */
+       if (write_access && !(vma->vm_flags & VM_SHARED))
+               if (vma_needs_reservation(h, vma, address) < 0) {
+                       ret = VM_FAULT_OOM;
+                       goto backout_unlocked;
+               }
+
        spin_lock(&mm->page_table_lock);
        size = i_size_read(mapping->host) >> huge_page_shift(h);
        if (idx >= size)
@@ -1957,6 +1977,7 @@ out:
 
 backout:
        spin_unlock(&mm->page_table_lock);
+backout_unlocked:
        unlock_page(page);
        put_page(page);
        goto out;
@@ -1968,6 +1989,7 @@ int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        pte_t *ptep;
        pte_t entry;
        int ret;
+       struct page *pagecache_page = NULL;
        static DEFINE_MUTEX(hugetlb_instantiation_mutex);
        struct hstate *h = hstate_vma(vma);
 
@@ -1984,25 +2006,57 @@ int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        entry = huge_ptep_get(ptep);
        if (huge_pte_none(entry)) {
                ret = hugetlb_no_page(mm, vma, address, ptep, write_access);
-               mutex_unlock(&hugetlb_instantiation_mutex);
-               return ret;
+               goto out_mutex;
        }
 
        ret = 0;
 
+       /*
+        * If we are going to COW the mapping later, we examine the pending
+        * reservations for this page now. This will ensure that any
+        * allocations necessary to record that reservation occur outside the
+        * spinlock. For private mappings, we also lookup the pagecache
+        * page now as it is used to determine if a reservation has been
+        * consumed.
+        */
+       if (write_access && !pte_write(entry)) {
+               if (vma_needs_reservation(h, vma, address) < 0) {
+                       ret = VM_FAULT_OOM;
+                       goto out_mutex;
+               }
+
+               if (!(vma->vm_flags & VM_SHARED))
+                       pagecache_page = hugetlbfs_pagecache_page(h,
+                                                               vma, address);
+       }
+
        spin_lock(&mm->page_table_lock);
        /* Check for a racing update before calling hugetlb_cow */
-       if (likely(pte_same(entry, huge_ptep_get(ptep))))
-               if (write_access && !pte_write(entry)) {
-                       struct page *page;
-                       page = hugetlbfs_pagecache_page(h, vma, address);
-                       ret = hugetlb_cow(mm, vma, address, ptep, entry, page);
-                       if (page) {
-                               unlock_page(page);
-                               put_page(page);
-                       }
+       if (unlikely(!pte_same(entry, huge_ptep_get(ptep))))
+               goto out_page_table_lock;
+
+
+       if (write_access) {
+               if (!pte_write(entry)) {
+                       ret = hugetlb_cow(mm, vma, address, ptep, entry,
+                                                       pagecache_page);
+                       goto out_page_table_lock;
                }
+               entry = pte_mkdirty(entry);
+       }
+       entry = pte_mkyoung(entry);
+       if (huge_ptep_set_access_flags(vma, address, ptep, entry, write_access))
+               update_mmu_cache(vma, address, entry);
+
+out_page_table_lock:
        spin_unlock(&mm->page_table_lock);
+
+       if (pagecache_page) {
+               unlock_page(pagecache_page);
+               put_page(pagecache_page);
+       }
+
+out_mutex:
        mutex_unlock(&hugetlb_instantiation_mutex);
 
        return ret;
@@ -2017,6 +2071,14 @@ follow_huge_pud(struct mm_struct *mm, unsigned long address,
        return NULL;
 }
 
+static int huge_zeropage_ok(pte_t *ptep, int write, int shared)
+{
+       if (!ptep || write || shared)
+               return 0;
+       else
+               return huge_pte_none(huge_ptep_get(ptep));
+}
+
 int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
                        struct page **pages, struct vm_area_struct **vmas,
                        unsigned long *position, int *length, int i,
@@ -2026,6 +2088,8 @@ int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
        unsigned long vaddr = *position;
        int remainder = *length;
        struct hstate *h = hstate_vma(vma);
+       int zeropage_ok = 0;
+       int shared = vma->vm_flags & VM_SHARED;
 
        spin_lock(&mm->page_table_lock);
        while (vaddr < vma->vm_end && remainder) {
@@ -2038,8 +2102,11 @@ int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
                 * first, for the page indexing below to work.
                 */
                pte = huge_pte_offset(mm, vaddr & huge_page_mask(h));
+               if (huge_zeropage_ok(pte, write, shared))
+                       zeropage_ok = 1;
 
-               if (!pte || huge_pte_none(huge_ptep_get(pte)) ||
+               if (!pte ||
+                   (huge_pte_none(huge_ptep_get(pte)) && !zeropage_ok) ||
                    (write && !pte_write(huge_ptep_get(pte)))) {
                        int ret;
 
@@ -2059,8 +2126,11 @@ int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
                page = pte_page(huge_ptep_get(pte));
 same_page:
                if (pages) {
-                       get_page(page);
-                       pages[i] = page + pfn_offset;
+                       if (zeropage_ok)
+                               pages[i] = ZERO_PAGE(0);
+                       else
+                               pages[i] = page + pfn_offset;
+                       get_page(pages[i]);
                }
 
                if (vmas)