]> git.karo-electronics.de Git - mv-sheeva.git/blobdiff - mm/filemap.c
[PATCH] x86: convert bigsmp to use flat physical mode
[mv-sheeva.git] / mm / filemap.c
index 88611928e71fc928a89b5b319044d759ea5f612d..4ef24a397684f7b9a51c70e94843f4cff35ed281 100644 (file)
 #include <asm/uaccess.h>
 #include <asm/mman.h>
 
+static ssize_t
+generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
+       loff_t offset, unsigned long nr_segs);
+
 /*
  * Shared mappings implemented 30.11.1994. It's not fully working yet,
  * though.
@@ -62,7 +66,7 @@
  *
  *  ->mmap_sem
  *    ->i_mmap_lock
- *      ->page_table_lock      (various places, mainly in mmap.c)
+ *      ->page_table_lock or pte_lock  (various, mainly in memory.c)
  *        ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
  *
  *  ->mmap_sem
@@ -82,9 +86,9 @@
  *    ->anon_vma.lock          (vma_adjust)
  *
  *  ->anon_vma.lock
- *    ->page_table_lock                (anon_vma_prepare and various)
+ *    ->page_table_lock or pte_lock    (anon_vma_prepare and various)
  *
- *  ->page_table_lock
+ *  ->page_table_lock or pte_lock
  *    ->swap_lock              (try_to_unmap_one)
  *    ->private_lock           (try_to_unmap_one)
  *    ->tree_lock              (try_to_unmap_one)
@@ -130,7 +134,7 @@ static int sync_page(void *word)
        struct address_space *mapping;
        struct page *page;
 
-       page = container_of((page_flags_t *)word, struct page, flags);
+       page = container_of((unsigned long *)word, struct page, flags);
 
        /*
         * page_mapping() is being called without PG_locked held.
@@ -148,7 +152,7 @@ static int sync_page(void *word)
         * in the ->sync_page() methods make essential use of the
         * page_mapping(), merely passing the page down to the backing
         * device's unplug functions when it's non-NULL, which in turn
-        * ignore it for all cases but swap, where only page->private is
+        * ignore it for all cases but swap, where only page_private(page) is
         * of interest. When page_mapping() does go NULL, the entire
         * call stack gracefully ignores the page and returns.
         * -- wli
@@ -301,8 +305,9 @@ EXPORT_SYMBOL(sync_page_range);
  * as it forces O_SYNC writers to different parts of the same file
  * to be serialised right until io completion.
  */
-int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
-                       loff_t pos, size_t count)
+static int sync_page_range_nolock(struct inode *inode,
+                                 struct address_space *mapping,
+                                 loff_t pos, size_t count)
 {
        pgoff_t start = pos >> PAGE_CACHE_SHIFT;
        pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
@@ -317,7 +322,6 @@ int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
                ret = wait_on_page_writeback_range(mapping, start, end);
        return ret;
 }
-EXPORT_SYMBOL(sync_page_range_nolock);
 
 /**
  * filemap_fdatawait - walk the list of under-writeback pages of the given
@@ -373,7 +377,7 @@ int filemap_write_and_wait_range(struct address_space *mapping,
  * This function does not add the page to the LRU.  The caller must do that.
  */
 int add_to_page_cache(struct page *page, struct address_space *mapping,
-               pgoff_t offset, int gfp_mask)
+               pgoff_t offset, gfp_t gfp_mask)
 {
        int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
 
@@ -397,7 +401,7 @@ int add_to_page_cache(struct page *page, struct address_space *mapping,
 EXPORT_SYMBOL(add_to_page_cache);
 
 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
-                               pgoff_t offset, int gfp_mask)
+                               pgoff_t offset, gfp_t gfp_mask)
 {
        int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
        if (ret == 0)
@@ -551,11 +555,12 @@ repeat:
                page_cache_get(page);
                if (TestSetPageLocked(page)) {
                        read_unlock_irq(&mapping->tree_lock);
-                       lock_page(page);
+                       __lock_page(page);
                        read_lock_irq(&mapping->tree_lock);
 
                        /* Has the page been truncated while we slept? */
-                       if (page->mapping != mapping || page->index != offset) {
+                       if (unlikely(page->mapping != mapping ||
+                                    page->index != offset)) {
                                unlock_page(page);
                                page_cache_release(page);
                                goto repeat;
@@ -587,7 +592,7 @@ EXPORT_SYMBOL(find_lock_page);
  * memory exhaustion.
  */
 struct page *find_or_create_page(struct address_space *mapping,
-               unsigned long index, unsigned int gfp_mask)
+               unsigned long index, gfp_t gfp_mask)
 {
        struct page *page, *cached_page = NULL;
        int err;
@@ -679,7 +684,7 @@ struct page *
 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
 {
        struct page *page = find_get_page(mapping, index);
-       unsigned int gfp_mask;
+       gfp_t gfp_mask;
 
        if (page) {
                if (!TestSetPageLocked(page))
@@ -827,8 +832,13 @@ readpage:
                /* Start the actual read. The read will unlock the page. */
                error = mapping->a_ops->readpage(filp, page);
 
-               if (unlikely(error))
+               if (unlikely(error)) {
+                       if (error == AOP_TRUNCATED_PAGE) {
+                               page_cache_release(page);
+                               goto find_page;
+                       }
                        goto readpage_error;
+               }
 
                if (!PageUptodate(page)) {
                        lock_page(page);
@@ -1026,8 +1036,8 @@ __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
                        desc.error = 0;
                        do_generic_file_read(filp,ppos,&desc,file_read_actor);
                        retval += desc.written;
-                       if (!retval) {
-                               retval = desc.error;
+                       if (desc.error) {
+                               retval = retval ?: desc.error;
                                break;
                        }
                }
@@ -1148,26 +1158,24 @@ static int fastcall page_cache_read(struct file * file, unsigned long offset)
 {
        struct address_space *mapping = file->f_mapping;
        struct page *page; 
-       int error;
+       int ret;
 
-       page = page_cache_alloc_cold(mapping);
-       if (!page)
-               return -ENOMEM;
+       do {
+               page = page_cache_alloc_cold(mapping);
+               if (!page)
+                       return -ENOMEM;
+
+               ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
+               if (ret == 0)
+                       ret = mapping->a_ops->readpage(file, page);
+               else if (ret == -EEXIST)
+                       ret = 0; /* losing race to add is OK */
 
-       error = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
-       if (!error) {
-               error = mapping->a_ops->readpage(file, page);
                page_cache_release(page);
-               return error;
-       }
 
-       /*
-        * We arrive here in the unlikely event that someone 
-        * raced with us and added our page to the cache first
-        * or we are out of memory for radix-tree nodes.
-        */
-       page_cache_release(page);
-       return error == -EEXIST ? 0 : error;
+       } while (ret == AOP_TRUNCATED_PAGE);
+               
+       return ret;
 }
 
 #define MMAP_LOTSAMISS  (100)
@@ -1327,10 +1335,14 @@ page_not_uptodate:
                goto success;
        }
 
-       if (!mapping->a_ops->readpage(file, page)) {
+       error = mapping->a_ops->readpage(file, page);
+       if (!error) {
                wait_on_page_locked(page);
                if (PageUptodate(page))
                        goto success;
+       } else if (error == AOP_TRUNCATED_PAGE) {
+               page_cache_release(page);
+               goto retry_find;
        }
 
        /*
@@ -1354,10 +1366,14 @@ page_not_uptodate:
                goto success;
        }
        ClearPageError(page);
-       if (!mapping->a_ops->readpage(file, page)) {
+       error = mapping->a_ops->readpage(file, page);
+       if (!error) {
                wait_on_page_locked(page);
                if (PageUptodate(page))
                        goto success;
+       } else if (error == AOP_TRUNCATED_PAGE) {
+               page_cache_release(page);
+               goto retry_find;
        }
 
        /*
@@ -1440,10 +1456,14 @@ page_not_uptodate:
                goto success;
        }
 
-       if (!mapping->a_ops->readpage(file, page)) {
+       error = mapping->a_ops->readpage(file, page);
+       if (!error) {
                wait_on_page_locked(page);
                if (PageUptodate(page))
                        goto success;
+       } else if (error == AOP_TRUNCATED_PAGE) {
+               page_cache_release(page);
+               goto retry_find;
        }
 
        /*
@@ -1466,10 +1486,14 @@ page_not_uptodate:
        }
 
        ClearPageError(page);
-       if (!mapping->a_ops->readpage(file, page)) {
+       error = mapping->a_ops->readpage(file, page);
+       if (!error) {
                wait_on_page_locked(page);
                if (PageUptodate(page))
                        goto success;
+       } else if (error == AOP_TRUNCATED_PAGE) {
+               page_cache_release(page);
+               goto retry_find;
        }
 
        /*
@@ -1516,7 +1540,7 @@ repeat:
                        page_cache_release(page);
                        return err;
                }
-       } else {
+       } else if (vma->vm_flags & VM_NONLINEAR) {
                /* No page was found just because we can't read it in now (being
                 * here implies nonblock != 0), but the page may exist, so set
                 * the PTE to fault it in later. */
@@ -1533,6 +1557,7 @@ repeat:
 
        return 0;
 }
+EXPORT_SYMBOL(filemap_populate);
 
 struct vm_operations_struct generic_file_vm_ops = {
        .nopage         = filemap_nopage,
@@ -1551,7 +1576,6 @@ int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
        vma->vm_ops = &generic_file_vm_ops;
        return 0;
 }
-EXPORT_SYMBOL(filemap_populate);
 
 /*
  * This is for filesystems which do not implement ->writepage.
@@ -1930,12 +1954,16 @@ generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
                status = a_ops->prepare_write(file, page, offset, offset+bytes);
                if (unlikely(status)) {
                        loff_t isize = i_size_read(inode);
+
+                       if (status != AOP_TRUNCATED_PAGE)
+                               unlock_page(page);
+                       page_cache_release(page);
+                       if (status == AOP_TRUNCATED_PAGE)
+                               continue;
                        /*
                         * prepare_write() may have instantiated a few blocks
                         * outside i_size.  Trim these off again.
                         */
-                       unlock_page(page);
-                       page_cache_release(page);
                        if (pos + bytes > isize)
                                vmtruncate(inode, isize);
                        break;
@@ -1948,6 +1976,10 @@ generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
                                                cur_iov, iov_base, bytes);
                flush_dcache_page(page);
                status = a_ops->commit_write(file, page, offset, offset+bytes);
+               if (status == AOP_TRUNCATED_PAGE) {
+                       page_cache_release(page);
+                       continue;
+               }
                if (likely(copied > 0)) {
                        if (!status)
                                status = copied;
@@ -2008,7 +2040,7 @@ generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
 }
 EXPORT_SYMBOL(generic_file_buffered_write);
 
-ssize_t
+static ssize_t
 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
                                unsigned long nr_segs, loff_t *ppos)
 {
@@ -2108,7 +2140,7 @@ generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
        return ret;
 }
 
-ssize_t
+static ssize_t
 __generic_file_write_nolock(struct file *file, const struct iovec *iov,
                                unsigned long nr_segs, loff_t *ppos)
 {
@@ -2229,7 +2261,7 @@ EXPORT_SYMBOL(generic_file_writev);
  * Called under i_sem for writes to S_ISREG files.   Returns -EIO if something
  * went wrong during pagecache shootdown.
  */
-ssize_t
+static ssize_t
 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
        loff_t offset, unsigned long nr_segs)
 {
@@ -2264,4 +2296,3 @@ generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
        }
        return retval;
 }
-EXPORT_SYMBOL_GPL(generic_file_direct_IO);