xfs: use scalable vmap API
authorAlex Elder <aelder@sgi.com>
Tue, 16 Mar 2010 18:55:56 +0000 (18:55 +0000)
committerAlex Elder <aelder@sgi.com>
Tue, 16 Mar 2010 20:40:36 +0000 (15:40 -0500)
Re-apply a commit that had been reverted due to regressions
that have since been fixed.

    From 95f8e302c04c0b0c6de35ab399a5551605eeb006 Mon Sep 17 00:00:00 2001
    From: Nick Piggin <npiggin@suse.de>
    Date: Tue, 6 Jan 2009 14:43:09 +1100

    Implement XFS's large buffer support with the new vmap APIs. See the vmap
    rewrite (db64fe02) for some numbers. The biggest improvement that comes from
    using the new APIs is avoiding the global KVA allocation lock on every call.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Reviewed-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Only modifications here were a minor reformat, plus making the patch
apply given the new use of xfs_buf_is_vmapped().

Modified-by: Alex Elder <aelder@sgi.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
fs/xfs/linux-2.6/xfs_buf.c

index 81f4ef27de3e852fa804c5114dfd9e2a408d92af..bd111b7e1daa5812b8b1d68907ecc390caecd8e0 100644 (file)
@@ -268,7 +268,8 @@ xfs_buf_free(
                uint            i;
 
                if (xfs_buf_is_vmapped(bp))
-                       vunmap(bp->b_addr - bp->b_offset);
+                       vm_unmap_ram(bp->b_addr - bp->b_offset,
+                                       bp->b_page_count);
 
                for (i = 0; i < bp->b_page_count; i++) {
                        struct page     *page = bp->b_pages[i];
@@ -388,8 +389,8 @@ _xfs_buf_map_pages(
                bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
                bp->b_flags |= XBF_MAPPED;
        } else if (flags & XBF_MAPPED) {
-               bp->b_addr = vmap(bp->b_pages, bp->b_page_count,
-                                       VM_MAP, PAGE_KERNEL);
+               bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
+                                       -1, PAGE_KERNEL);
                if (unlikely(bp->b_addr == NULL))
                        return -ENOMEM;
                bp->b_addr += bp->b_offset;