Merge back earlier ACPICA changes for v4.11.
[platform/kernel/linux-starfive.git] / fs / buffer.c
index a3bfd57..d21771f 100644 (file)
@@ -43,6 +43,7 @@
 #include <linux/bitops.h>
 #include <linux/mpage.h>
 #include <linux/bit_spinlock.h>
+#include <linux/pagevec.h>
 #include <trace/events/block.h>
 
 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
@@ -1604,37 +1605,80 @@ void create_empty_buffers(struct page *page,
 }
 EXPORT_SYMBOL(create_empty_buffers);
 
-/*
- * We are taking a block for data and we don't want any output from any
- * buffer-cache aliases starting from return from that function and
- * until the moment when something will explicitly mark the buffer
- * dirty (hopefully that will not happen until we will free that block ;-)
- * We don't even need to mark it not-uptodate - nobody can expect
- * anything from a newly allocated buffer anyway. We used to used
- * unmap_buffer() for such invalidation, but that was wrong. We definitely
- * don't want to mark the alias unmapped, for example - it would confuse
- * anyone who might pick it with bread() afterwards...
- *
- * Also..  Note that bforget() doesn't lock the buffer.  So there can
- * be writeout I/O going on against recently-freed buffers.  We don't
- * wait on that I/O in bforget() - it's more efficient to wait on the I/O
- * only if we really need to.  That happens here.
- */
-void unmap_underlying_metadata(struct block_device *bdev, sector_t block)
+/**
+ * clean_bdev_aliases: clean a range of buffers in block device
+ * @bdev: Block device to clean buffers in
+ * @block: Start of a range of blocks to clean
+ * @len: Number of blocks to clean
+ *
+ * We are taking a range of blocks for data and we don't want writeback of any
+ * buffer-cache aliases starting from return from this function and until the
+ * moment when something will explicitly mark the buffer dirty (hopefully that
+ * will not happen until we will free that block ;-) We don't even need to mark
+ * it not-uptodate - nobody can expect anything from a newly allocated buffer
+ * anyway. We used to use unmap_buffer() for such invalidation, but that was
+ * wrong. We definitely don't want to mark the alias unmapped, for example - it
+ * would confuse anyone who might pick it with bread() afterwards...
+ *
+ * Also..  Note that bforget() doesn't lock the buffer.  So there can be
+ * writeout I/O going on against recently-freed buffers.  We don't wait on that
+ * I/O in bforget() - it's more efficient to wait on the I/O only if we really
+ * need to.  That happens here.
+ */
+void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
 {
-       struct buffer_head *old_bh;
+       struct inode *bd_inode = bdev->bd_inode;
+       struct address_space *bd_mapping = bd_inode->i_mapping;
+       struct pagevec pvec;
+       pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
+       pgoff_t end;
+       int i;
+       struct buffer_head *bh;
+       struct buffer_head *head;
 
-       might_sleep();
+       end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
+       pagevec_init(&pvec, 0);
+       while (index <= end && pagevec_lookup(&pvec, bd_mapping, index,
+                       min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
+               for (i = 0; i < pagevec_count(&pvec); i++) {
+                       struct page *page = pvec.pages[i];
 
-       old_bh = __find_get_block_slow(bdev, block);
-       if (old_bh) {
-               clear_buffer_dirty(old_bh);
-               wait_on_buffer(old_bh);
-               clear_buffer_req(old_bh);
-               __brelse(old_bh);
+                       index = page->index;
+                       if (index > end)
+                               break;
+                       if (!page_has_buffers(page))
+                               continue;
+                       /*
+                        * We use page lock instead of bd_mapping->private_lock
+                        * to pin buffers here since we can afford to sleep and
+                        * it scales better than a global spinlock lock.
+                        */
+                       lock_page(page);
+                       /* Recheck when the page is locked which pins bhs */
+                       if (!page_has_buffers(page))
+                               goto unlock_page;
+                       head = page_buffers(page);
+                       bh = head;
+                       do {
+                               if (!buffer_mapped(bh))
+                                       goto next;
+                               if (bh->b_blocknr >= block + len)
+                                       break;
+                               clear_buffer_dirty(bh);
+                               wait_on_buffer(bh);
+                               clear_buffer_req(bh);
+next:
+                               bh = bh->b_this_page;
+                       } while (bh != head);
+unlock_page:
+                       unlock_page(page);
+               }
+               pagevec_release(&pvec);
+               cond_resched();
+               index++;
        }
 }
-EXPORT_SYMBOL(unmap_underlying_metadata);
+EXPORT_SYMBOL(clean_bdev_aliases);
 
 /*
  * Size is a power-of-two in the range 512..PAGE_SIZE,
@@ -1745,8 +1789,7 @@ int __block_write_full_page(struct inode *inode, struct page *page,
                        if (buffer_new(bh)) {
                                /* blockdev mappings never come here */
                                clear_buffer_new(bh);
-                               unmap_underlying_metadata(bh->b_bdev,
-                                                       bh->b_blocknr);
+                               clean_bdev_bh_alias(bh);
                        }
                }
                bh = bh->b_this_page;
@@ -1992,8 +2035,7 @@ int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
                        }
 
                        if (buffer_new(bh)) {
-                               unmap_underlying_metadata(bh->b_bdev,
-                                                       bh->b_blocknr);
+                               clean_bdev_bh_alias(bh);
                                if (PageUptodate(page)) {
                                        clear_buffer_new(bh);
                                        set_buffer_uptodate(bh);
@@ -2633,7 +2675,7 @@ int nobh_write_begin(struct address_space *mapping,
                if (!buffer_mapped(bh))
                        is_mapped_to_disk = 0;
                if (buffer_new(bh))
-                       unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
+                       clean_bdev_bh_alias(bh);
                if (PageUptodate(page)) {
                        set_buffer_uptodate(bh);
                        continue;