[Top] [All Lists]

[PATCH v2 0/2] fix spinlock recursion on xa_lock in xfs_buf_item_push

To: xfs@xxxxxxxxxxx
Subject: [PATCH v2 0/2] fix spinlock recursion on xa_lock in xfs_buf_item_push
From: Brian Foster <bfoster@xxxxxxxxxx>
Date: Wed, 6 Feb 2013 07:44:39 -0500
Delivered-to: xfs@xxxxxxxxxxx
Hi all,

Here is v2 of the xa_lock recursion fix[1]. This has survived several days of 
reproduction workload and an xfstests run shows no regressions. Thoughts


- Patch 2 is reworked to detect the potential race and defer the log force to
  xfsaild by returning XFS_ITEM_PINNED.

[1] - recursion stack trace

BUG: spinlock recursion on CPU#5, xfsaild/dm-3/2690
Pid: 2690, comm: xfsaild/dm-3 Not tainted 3.8.0-rc1+ #46
Call Trace:
 [<ffffffff8163211c>] spin_dump+0x8a/0x8f
 [<ffffffff81632142>] spin_bug+0x21/0x26
 [<ffffffff812f66a1>] do_raw_spin_lock+0x101/0x150
 [<ffffffff816378ce>] _raw_spin_lock+0xe/0x10
 [<ffffffffa0522c85>] xlog_assign_tail_lsn+0x25/0x50 [xfs]
 [<ffffffffa0523286>] xlog_state_release_iclog+0x86/0xd0 [xfs]
 [<ffffffffa0523c89>] xlog_write+0x569/0x710 [xfs]
 [<ffffffffa052555c>] xlog_cil_push+0x29c/0x3c0 [xfs]
 [<ffffffffa04cbfe2>] ? xfs_buf_get_map+0xf2/0x1b0 [xfs]
 [<ffffffffa0525d17>] xlog_cil_force_lsn+0x157/0x160 [xfs]
 [<ffffffffa04cced1>] ? xfs_buf_read_map+0x31/0x130 [xfs]
 [<ffffffffa0529e99>] ? xfs_trans_read_buf_map+0x279/0x4b0 [xfs]
 [<ffffffff8117e45d>] ? __kmalloc+0x15d/0x1b0
 [<ffffffffa0523f7d>] _xfs_log_force+0x6d/0x290 [xfs]
 [<ffffffffa051450f>] ? xfs_iflush_cluster+0x25f/0x3d0 [xfs]
 [<ffffffffa05241d9>] xfs_log_force+0x39/0xc0 [xfs]
 [<ffffffffa04cbaa0>] xfs_buf_trylock+0xd0/0xe0 [xfs]
 [<ffffffffa0526369>] xfs_buf_item_push+0x39/0xd0 [xfs]
 [<ffffffffa0527bdf>] ? xfs_inode_item_push+0x8f/0x140 [xfs]
 [<ffffffffa0528c01>] xfsaild+0x2e1/0x6e0 [xfs]
 [<ffffffff8108aa08>] ? __wake_up_common+0x58/0x90
 [<ffffffffa0528920>] ? xfs_trans_ail_cursor_first+0xc0/0xc0 [xfs]
 [<ffffffff81081708>] kthread+0xd8/0xe0
 [<ffffffff81081630>] ? flush_kthread_work+0x150/0x150
 [<ffffffff816400ac>] ret_from_fork+0x7c/0xb0
 [<ffffffff81081630>] ? flush_kthread_work+0x150/0x150

Brian Foster (2):
  xfs: conditionally force log on trylock failure of pinned/stale buf
  xfs: disable log force in xfs_buf_item_push() to avoid xa_lock

 fs/xfs/xfs_buf.c      |    8 +++++---
 fs/xfs/xfs_buf.h      |    3 ++-
 fs/xfs/xfs_buf_item.c |   14 +++++++++++++-
 3 files changed, 20 insertions(+), 5 deletions(-)


<Prev in Thread] Current Thread [Next in Thread>