Merge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/dvrabel/uwb
[pandora-kernel.git] / mm / hugetlb.c
index e83ad2c..d0351e3 100644 (file)
@@ -578,41 +578,6 @@ static void free_huge_page(struct page *page)
                hugetlb_put_quota(mapping, 1);
 }
 
-/*
- * Increment or decrement surplus_huge_pages.  Keep node-specific counters
- * balanced by operating on them in a round-robin fashion.
- * Returns 1 if an adjustment was made.
- */
-static int adjust_pool_surplus(struct hstate *h, int delta)
-{
-       static int prev_nid;
-       int nid = prev_nid;
-       int ret = 0;
-
-       VM_BUG_ON(delta != -1 && delta != 1);
-       do {
-               nid = next_node(nid, node_online_map);
-               if (nid == MAX_NUMNODES)
-                       nid = first_node(node_online_map);
-
-               /* To shrink on this node, there must be a surplus page */
-               if (delta < 0 && !h->surplus_huge_pages_node[nid])
-                       continue;
-               /* Surplus cannot exceed the total number of pages */
-               if (delta > 0 && h->surplus_huge_pages_node[nid] >=
-                                               h->nr_huge_pages_node[nid])
-                       continue;
-
-               h->surplus_huge_pages += delta;
-               h->surplus_huge_pages_node[nid] += delta;
-               ret = 1;
-               break;
-       } while (nid != prev_nid);
-
-       prev_nid = nid;
-       return ret;
-}
-
 static void prep_new_huge_page(struct hstate *h, struct page *page, int nid)
 {
        set_compound_page_dtor(page, free_huge_page);
@@ -623,6 +588,34 @@ static void prep_new_huge_page(struct hstate *h, struct page *page, int nid)
        put_page(page); /* free it into the hugepage allocator */
 }
 
+static void prep_compound_gigantic_page(struct page *page, unsigned long order)
+{
+       int i;
+       int nr_pages = 1 << order;
+       struct page *p = page + 1;
+
+       /* we rely on prep_new_huge_page to set the destructor */
+       set_compound_order(page, order);
+       __SetPageHead(page);
+       for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) {
+               __SetPageTail(p);
+               p->first_page = page;
+       }
+}
+
+int PageHuge(struct page *page)
+{
+       compound_page_dtor *dtor;
+
+       if (!PageCompound(page))
+               return 0;
+
+       page = compound_head(page);
+       dtor = get_compound_page_dtor(page);
+
+       return dtor == free_huge_page;
+}
+
 static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid)
 {
        struct page *page;
@@ -630,7 +623,7 @@ static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid)
        if (h->order >= MAX_ORDER)
                return NULL;
 
-       page = alloc_pages_node(nid,
+       page = alloc_pages_exact_node(nid,
                htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE|
                                                __GFP_REPEAT|__GFP_NOWARN,
                huge_page_order(h));
@@ -649,7 +642,7 @@ static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid)
  * Use a helper variable to find the next node and then
  * copy it back to hugetlb_next_nid afterwards:
  * otherwise there's a window in which a racer might
- * pass invalid nid MAX_NUMNODES to alloc_pages_node.
+ * pass invalid nid MAX_NUMNODES to alloc_pages_exact_node.
  * But we don't need to use a spin_lock here: it really
  * doesn't matter if occasionally a racer chooses the
  * same nid as we do.  Move nid forward in the mask even
@@ -875,7 +868,7 @@ static void return_unused_surplus_pages(struct hstate *h,
         * can no longer free unreserved surplus pages. This occurs when
         * the nodes with surplus pages have no free pages.
         */
-       unsigned long remaining_iterations = num_online_nodes();
+       unsigned long remaining_iterations = nr_online_nodes;
 
        /* Uncommit the reservation */
        h->resv_huge_pages -= unused_resv_pages;
@@ -904,7 +897,7 @@ static void return_unused_surplus_pages(struct hstate *h,
                        h->surplus_huge_pages--;
                        h->surplus_huge_pages_node[nid]--;
                        nr_pages--;
-                       remaining_iterations = num_online_nodes();
+                       remaining_iterations = nr_online_nodes;
                }
        }
 }
@@ -1140,6 +1133,41 @@ static inline void try_to_free_low(struct hstate *h, unsigned long count)
 }
 #endif
 
+/*
+ * Increment or decrement surplus_huge_pages.  Keep node-specific counters
+ * balanced by operating on them in a round-robin fashion.
+ * Returns 1 if an adjustment was made.
+ */
+static int adjust_pool_surplus(struct hstate *h, int delta)
+{
+       static int prev_nid;
+       int nid = prev_nid;
+       int ret = 0;
+
+       VM_BUG_ON(delta != -1 && delta != 1);
+       do {
+               nid = next_node(nid, node_online_map);
+               if (nid == MAX_NUMNODES)
+                       nid = first_node(node_online_map);
+
+               /* To shrink on this node, there must be a surplus page */
+               if (delta < 0 && !h->surplus_huge_pages_node[nid])
+                       continue;
+               /* Surplus cannot exceed the total number of pages */
+               if (delta > 0 && h->surplus_huge_pages_node[nid] >=
+                                               h->nr_huge_pages_node[nid])
+                       continue;
+
+               h->surplus_huge_pages += delta;
+               h->surplus_huge_pages_node[nid] += delta;
+               ret = 1;
+               break;
+       } while (nid != prev_nid);
+
+       prev_nid = nid;
+       return ret;
+}
+
 #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages)
 static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count)
 {
@@ -1957,7 +1985,7 @@ static struct page *hugetlbfs_pagecache_page(struct hstate *h,
 }
 
 static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
-                       unsigned long address, pte_t *ptep, int write_access)
+                       unsigned long address, pte_t *ptep, unsigned int flags)
 {
        struct hstate *h = hstate_vma(vma);
        int ret = VM_FAULT_SIGBUS;
@@ -2025,7 +2053,7 @@ retry:
         * any allocations necessary to record that reservation occur outside
         * the spinlock.
         */
-       if (write_access && !(vma->vm_flags & VM_SHARED))
+       if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED))
                if (vma_needs_reservation(h, vma, address) < 0) {
                        ret = VM_FAULT_OOM;
                        goto backout_unlocked;
@@ -2044,7 +2072,7 @@ retry:
                                && (vma->vm_flags & VM_SHARED)));
        set_huge_pte_at(mm, address, ptep, new_pte);
 
-       if (write_access && !(vma->vm_flags & VM_SHARED)) {
+       if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) {
                /* Optimization, do the COW without a second fault */
                ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page);
        }
@@ -2063,7 +2091,7 @@ backout_unlocked:
 }
 
 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
-                       unsigned long address, int write_access)
+                       unsigned long address, unsigned int flags)
 {
        pte_t *ptep;
        pte_t entry;
@@ -2084,7 +2112,7 @@ int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        mutex_lock(&hugetlb_instantiation_mutex);
        entry = huge_ptep_get(ptep);
        if (huge_pte_none(entry)) {
-               ret = hugetlb_no_page(mm, vma, address, ptep, write_access);
+               ret = hugetlb_no_page(mm, vma, address, ptep, flags);
                goto out_mutex;
        }
 
@@ -2098,7 +2126,7 @@ int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
         * page now as it is used to determine if a reservation has been
         * consumed.
         */
-       if (write_access && !pte_write(entry)) {
+       if ((flags & FAULT_FLAG_WRITE) && !pte_write(entry)) {
                if (vma_needs_reservation(h, vma, address) < 0) {
                        ret = VM_FAULT_OOM;
                        goto out_mutex;
@@ -2115,7 +2143,7 @@ int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
                goto out_page_table_lock;
 
 
-       if (write_access) {
+       if (flags & FAULT_FLAG_WRITE) {
                if (!pte_write(entry)) {
                        ret = hugetlb_cow(mm, vma, address, ptep, entry,
                                                        pagecache_page);
@@ -2124,7 +2152,8 @@ int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
                entry = pte_mkdirty(entry);
        }
        entry = pte_mkyoung(entry);
-       if (huge_ptep_set_access_flags(vma, address, ptep, entry, write_access))
+       if (huge_ptep_set_access_flags(vma, address, ptep, entry,
+                                               flags & FAULT_FLAG_WRITE))
                update_mmu_cache(vma, address, entry);
 
 out_page_table_lock: