diff options
| author | Chris Mason <chris.mason@oracle.com> | 2008-06-25 22:14:04 +0200 |
|---|---|---|
| committer | Chris Mason <chris.mason@oracle.com> | 2008-09-25 17:04:04 +0200 |
| commit | f9efa9c784aa3b801feb367f72c6867d26fb348e (patch) | |
| tree | d9f678310c6c8a4daf2003fa4f6b4164111dbbd7 /fs/btrfs/ctree.c | |
| parent | Btrfs: Online btree defragmentation fixes (diff) | |
| download | linux-f9efa9c784aa3b801feb367f72c6867d26fb348e.tar.xz linux-f9efa9c784aa3b801feb367f72c6867d26fb348e.zip | |
Btrfs: Reduce contention on the root node
This calls unlock_up sooner in btrfs_search_slot in order to decrease the
amount of work done with the higher level tree locks held.
Also, it changes btrfs_tree_lock to spin for a big against the page lock
before scheduling. This makes a big difference in context switch rate under
highly contended workloads.
Longer term, a better locking structure is needed than the page lock.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Diffstat (limited to 'fs/btrfs/ctree.c')
| -rw-r--r-- | fs/btrfs/ctree.c | 17 |
1 files changed, 11 insertions, 6 deletions
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c index 0cb80f32a9c7..c6759fc1004a 100644 --- a/fs/btrfs/ctree.c +++ b/fs/btrfs/ctree.c @@ -1313,16 +1313,13 @@ again: slot = p->slots[level]; BUG_ON(btrfs_header_nritems(b) == 1); } + unlock_up(p, level, lowest_unlock); + /* this is only true while dropping a snapshot */ if (level == lowest_level) { - unlock_up(p, level, lowest_unlock); break; } - if (should_reada) - reada_for_search(root, p, level, slot, - key->objectid); - blocknr = btrfs_node_blockptr(b, slot); gen = btrfs_node_ptr_generation(b, slot); blocksize = btrfs_level_size(root, level - 1); @@ -1340,6 +1337,11 @@ again: btrfs_release_path(NULL, p); if (tmp) free_extent_buffer(tmp); + if (should_reada) + reada_for_search(root, p, + level, slot, + key->objectid); + tmp = read_tree_block(root, blocknr, blocksize, gen); if (tmp) @@ -1348,12 +1350,15 @@ again: } else { if (tmp) free_extent_buffer(tmp); + if (should_reada) + reada_for_search(root, p, + level, slot, + key->objectid); b = read_node_slot(root, b, slot); } } if (!p->skip_locking) btrfs_tree_lock(b); - unlock_up(p, level, lowest_unlock); } else { p->slots[level] = slot; if (ins_len > 0 && btrfs_leaf_free_space(root, b) < |