thp: madvise(MADV_HUGEPAGE)
authorAndrea Arcangeli <aarcange@redhat.com>
Thu, 13 Jan 2011 23:46:55 +0000 (15:46 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 14 Jan 2011 01:32:42 +0000 (17:32 -0800)
Add madvise MADV_HUGEPAGE to mark regions that are important to be
hugepage backed.  Return -EINVAL if the vma is not of an anonymous type,
or the feature isn't built into the kernel.  Never silently return
success.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/huge_mm.h
mm/huge_memory.c
mm/madvise.c

index 9301824c7491ae72f9ac49826eda850ecc1804bd..d9ab70d776e295473f146bb1a61ea7d5d50a8566 100644 (file)
@@ -97,6 +97,7 @@ extern void __split_huge_page_pmd(struct mm_struct *mm, pmd_t *pmd);
 #if HPAGE_PMD_ORDER > MAX_ORDER
 #error "hugepages can't be allocated by the buddy allocator"
 #endif
+extern int hugepage_madvise(unsigned long *vm_flags);
 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 #define HPAGE_PMD_SHIFT ({ BUG(); 0; })
 #define HPAGE_PMD_MASK ({ BUG(); 0; })
@@ -113,6 +114,11 @@ static inline int split_huge_page(struct page *page)
        do { } while (0)
 #define wait_split_huge_page(__anon_vma, __pmd)        \
        do { } while (0)
+static inline int hugepage_madvise(unsigned long *vm_flags)
+{
+       BUG();
+       return 0;
+}
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
 #endif /* _LINUX_HUGE_MM_H */
index 7635079328983883e483332be82fda788bfc3add..620891f4e54f4ee6543b1cc82bcf3d810804ee7c 100644 (file)
@@ -896,6 +896,22 @@ out:
        return ret;
 }
 
+int hugepage_madvise(unsigned long *vm_flags)
+{
+       /*
+        * Be somewhat over-protective like KSM for now!
+        */
+       if (*vm_flags & (VM_HUGEPAGE | VM_SHARED  | VM_MAYSHARE   |
+                        VM_PFNMAP   | VM_IO      | VM_DONTEXPAND |
+                        VM_RESERVED | VM_HUGETLB | VM_INSERTPAGE |
+                        VM_MIXEDMAP | VM_SAO))
+               return -EINVAL;
+
+       *vm_flags |= VM_HUGEPAGE;
+
+       return 0;
+}
+
 void __split_huge_page_pmd(struct mm_struct *mm, pmd_t *pmd)
 {
        struct page *page;
index 319528b8db74c8d0c7cd10c8468acb5539aea7dd..ecde40a401c1911df933949c26e35529e446cbcc 100644 (file)
@@ -71,6 +71,11 @@ static long madvise_behavior(struct vm_area_struct * vma,
                if (error)
                        goto out;
                break;
+       case MADV_HUGEPAGE:
+               error = hugepage_madvise(&new_flags);
+               if (error)
+                       goto out;
+               break;
        }
 
        if (new_flags == vma->vm_flags) {
@@ -282,6 +287,9 @@ madvise_behavior_valid(int behavior)
 #ifdef CONFIG_KSM
        case MADV_MERGEABLE:
        case MADV_UNMERGEABLE:
+#endif
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+       case MADV_HUGEPAGE:
 #endif
                return 1;