mm: use vm_unmapped_area() in hugetlbfs on ia64 architecture
Michel Lespinasse [Thu, 21 Feb 2013 23:10:29 +0000 (15:10 -0800)]
Update the ia64 hugetlb_get_unmapped_area function to make use of
vm_unmapped_area() instead of implementing a brute force search.

Signed-off-by: Michel Lespinasse <walken@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>

arch/ia64/mm/hugetlbpage.c

index 5ca674b..76069c1 100644 (file)
@@ -148,7 +148,7 @@ void hugetlb_free_pgd_range(struct mmu_gather *tlb,
 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
                unsigned long pgoff, unsigned long flags)
 {
-       struct vm_area_struct *vmm;
+       struct vm_unmapped_area_info info;
 
        if (len > RGN_MAP_LIMIT)
                return -ENOMEM;
@@ -165,16 +165,14 @@ unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, u
        /* This code assumes that RGN_HPAGE != 0. */
        if ((REGION_NUMBER(addr) != RGN_HPAGE) || (addr & (HPAGE_SIZE - 1)))
                addr = HPAGE_REGION_BASE;
-       else
-               addr = ALIGN(addr, HPAGE_SIZE);
-       for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) {
-               /* At this point:  (!vmm || addr < vmm->vm_end). */
-               if (REGION_OFFSET(addr) + len > RGN_MAP_LIMIT)
-                       return -ENOMEM;
-               if (!vmm || (addr + len) <= vmm->vm_start)
-                       return addr;
-               addr = ALIGN(vmm->vm_end, HPAGE_SIZE);
-       }
+
+       info.flags = 0;
+       info.length = len;
+       info.low_limit = addr;
+       info.high_limit = HPAGE_REGION_BASE + RGN_MAP_LIMIT;
+       info.align_mask = PAGE_MASK & (HPAGE_SIZE - 1);
+       info.align_offset = 0;
+       return vm_unmapped_area(&info);
 }
 
 static int __init hugetlb_setup_sz(char *str)