nommu: fix a number of issues with the per-MM VMA patch
David Howells [Thu, 2 Apr 2009 23:56:32 +0000 (16:56 -0700)]
Fix a number of issues with the per-MM VMA patch:

 (1) Make mmap_pages_allocated an atomic_long_t, just in case this is used on
     a NOMMU system with more than 2G pages.  Makes no difference on a 32-bit
     system.

 (2) Report vma->vm_pgoff * PAGE_SIZE as a 64-bit value, not a 32-bit value,
     lest it overflow.

 (3) Move the allocation of the vm_area_struct slab back for fork.c.

 (4) Use KMEM_CACHE() for both vm_area_struct and vm_region slabs.

 (5) Use BUG_ON() rather than if () BUG().

 (6) Make the default validate_nommu_regions() a static inline rather than a
     #define.

 (7) Make free_page_series()'s objection to pages with a refcount != 1 more
     informative.

 (8) Adjust the __put_nommu_region() banner comment to indicate that the
     semaphore must be held for writing.

 (9) Limit the number of warnings about munmaps of non-mmapped regions.

Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Greg Ungerer <gerg@snapgear.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/proc/meminfo.c
fs/proc/task_nommu.c
include/linux/mm.h
kernel/fork.c
mm/mmap.c
mm/nommu.c

index 43d2394..74ea974 100644 (file)
@@ -120,7 +120,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v)
                K(i.freeram-i.freehigh),
 #endif
 #ifndef CONFIG_MMU
-               K((unsigned long) atomic_read(&mmap_pages_allocated)),
+               K((unsigned long) atomic_long_read(&mmap_pages_allocated)),
 #endif
                K(i.totalswap),
                K(i.freeswap),
index 343ea12..370be0a 100644 (file)
@@ -136,14 +136,14 @@ static int nommu_vma_show(struct seq_file *m, struct vm_area_struct *vma)
        }
 
        seq_printf(m,
-                  "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
+                  "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
                   vma->vm_start,
                   vma->vm_end,
                   flags & VM_READ ? 'r' : '-',
                   flags & VM_WRITE ? 'w' : '-',
                   flags & VM_EXEC ? 'x' : '-',
                   flags & VM_MAYSHARE ? flags & VM_SHARED ? 'S' : 's' : 'p',
-                  vma->vm_pgoff << PAGE_SHIFT,
+                  (unsigned long long) vma->vm_pgoff << PAGE_SHIFT,
                   MAJOR(dev), MINOR(dev), ino, &len);
 
        if (file) {
index aeabe95..bff1f0d 100644 (file)
@@ -1079,7 +1079,7 @@ static inline void setup_per_cpu_pageset(void) {}
 #endif
 
 /* nommu.c */
-extern atomic_t mmap_pages_allocated;
+extern atomic_long_t mmap_pages_allocated;
 
 /* prio_tree.c */
 void vma_prio_tree_add(struct vm_area_struct *, struct vm_area_struct *old);
index 47c1584..51d1aa2 100644 (file)
@@ -1488,6 +1488,7 @@ void __init proc_caches_init(void)
        mm_cachep = kmem_cache_create("mm_struct",
                        sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
                        SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
+       vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
        mmap_init();
 }
 
index 1abb918..4a38411 100644 (file)
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2481,7 +2481,4 @@ void mm_drop_all_locks(struct mm_struct *mm)
  */
 void __init mmap_init(void)
 {
-       vm_area_cachep = kmem_cache_create("vm_area_struct",
-                       sizeof(struct vm_area_struct), 0,
-                       SLAB_PANIC, NULL);
 }
index 2fcf47d..72eda4a 100644 (file)
@@ -69,7 +69,7 @@ int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
 int sysctl_nr_trim_pages = 1; /* page trimming behaviour */
 int heap_stack_gap = 0;
 
-atomic_t mmap_pages_allocated;
+atomic_long_t mmap_pages_allocated;
 
 EXPORT_SYMBOL(mem_map);
 EXPORT_SYMBOL(num_physpages);
@@ -463,12 +463,7 @@ SYSCALL_DEFINE1(brk, unsigned long, brk)
  */
 void __init mmap_init(void)
 {
-       vm_region_jar = kmem_cache_create("vm_region_jar",
-                                         sizeof(struct vm_region), 0,
-                                         SLAB_PANIC, NULL);
-       vm_area_cachep = kmem_cache_create("vm_area_struct",
-                                          sizeof(struct vm_area_struct), 0,
-                                          SLAB_PANIC, NULL);
+       vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 }
 
 /*
@@ -486,27 +481,24 @@ static noinline void validate_nommu_regions(void)
                return;
 
        last = rb_entry(lastp, struct vm_region, vm_rb);
-       if (unlikely(last->vm_end <= last->vm_start))
-               BUG();
-       if (unlikely(last->vm_top < last->vm_end))
-               BUG();
+       BUG_ON(unlikely(last->vm_end <= last->vm_start));
+       BUG_ON(unlikely(last->vm_top < last->vm_end));
 
        while ((p = rb_next(lastp))) {
                region = rb_entry(p, struct vm_region, vm_rb);
                last = rb_entry(lastp, struct vm_region, vm_rb);
 
-               if (unlikely(region->vm_end <= region->vm_start))
-                       BUG();
-               if (unlikely(region->vm_top < region->vm_end))
-                       BUG();
-               if (unlikely(region->vm_start < last->vm_top))
-                       BUG();
+               BUG_ON(unlikely(region->vm_end <= region->vm_start));
+               BUG_ON(unlikely(region->vm_top < region->vm_end));
+               BUG_ON(unlikely(region->vm_start < last->vm_top));
 
                lastp = p;
        }
 }
 #else
-#define validate_nommu_regions() do {} while(0)
+static void validate_nommu_regions(void)
+{
+}
 #endif
 
 /*
@@ -563,16 +555,17 @@ static void free_page_series(unsigned long from, unsigned long to)
                struct page *page = virt_to_page(from);
 
                kdebug("- free %lx", from);
-               atomic_dec(&mmap_pages_allocated);
+               atomic_long_dec(&mmap_pages_allocated);
                if (page_count(page) != 1)
-                       kdebug("free page %p [%d]", page, page_count(page));
+                       kdebug("free page %p: refcount not one: %d",
+                              page, page_count(page));
                put_page(page);
        }
 }
 
 /*
  * release a reference to a region
- * - the caller must hold the region semaphore, which this releases
+ * - the caller must hold the region semaphore for writing, which this releases
  * - the region may not have been added to the tree yet, in which case vm_top
  *   will equal vm_start
  */
@@ -1096,7 +1089,7 @@ static int do_mmap_private(struct vm_area_struct *vma,
                goto enomem;
 
        total = 1 << order;
-       atomic_add(total, &mmap_pages_allocated);
+       atomic_long_add(total, &mmap_pages_allocated);
 
        point = rlen >> PAGE_SHIFT;
 
@@ -1107,7 +1100,7 @@ static int do_mmap_private(struct vm_area_struct *vma,
                        order = ilog2(total - point);
                        n = 1 << order;
                        kdebug("shave %lu/%lu @%lu", n, total - point, total);
-                       atomic_sub(n, &mmap_pages_allocated);
+                       atomic_long_sub(n, &mmap_pages_allocated);
                        total -= n;
                        set_page_refcounted(pages + total);
                        __free_pages(pages + total, order);
@@ -1536,10 +1529,15 @@ int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
        /* find the first potentially overlapping VMA */
        vma = find_vma(mm, start);
        if (!vma) {
-               printk(KERN_WARNING
-                      "munmap of memory not mmapped by process %d (%s):"
-                      " 0x%lx-0x%lx\n",
-                      current->pid, current->comm, start, start + len - 1);
+               static int limit = 0;
+               if (limit < 5) {
+                       printk(KERN_WARNING
+                              "munmap of memory not mmapped by process %d"
+                              " (%s): 0x%lx-0x%lx\n",
+                              current->pid, current->comm,
+                              start, start + len - 1);
+                       limit++;
+               }
                return -EINVAL;
        }