On Thu, Jun 09, 2011 at 09:23:57PM +0800, Christoph Hellwig wrote: > Here is a patch that removes the XFS-internal writing of data in sync > and instead trusts the VFS code to get it right, and also does the > ioend_wait under a lock, to protect against new pending I/Os showing up. > > It passed a single xfsqa run so far, so at least it won't blow up > completely. > > Can you check if that helps with your livelock test case? (And yes, I > should go and try it myself. Will do as soon as I'll get a bit more > time). My pleasure. It's easier to compare results on the same test bed and I've made it very convenient to test patches :) The patch does improve the sync time. before patch after patch -------------------------------------------------- sync time: 18 sync time: 18 sync time: 21 sync time: 15 sync time: 14 sync time: 14 sync time: 20 sync time: 16 sync time: 21 sync time: 14 avg 18.800 avg 15.400 stddev 2.638 stddev 1.497 I also did extra test runs and collect sync(1) call traces by doing "echo t > /proc/sysrq-trigger" randomly. The ones in XFS are: [ 78.395664] sync D 0000000000000002 5296 3024 3003 0x00000000 [ 78.396162] ffff8800b4229de8 0000000000000046 ffffffff819252fa 0000000000000000 [ 78.396850] ffff8800b4228000 00000000001d1a40 00000000001d1a40 ffff8800aefa22f0 [ 78.397519] 00000000001d1a40 ffff8800b4229fd8 00000000001d1a40 ffff8800b4229fd8 [ 78.398189] Call Trace: [ 78.398431] [<ffffffff819252fa>] ? schedule+0x374/0x992 [ 78.398778] [<ffffffff812e906f>] ? _xfs_log_force+0x1de/0x226 [ 78.399143] [<ffffffff81092be5>] ? local_clock+0x41/0x5a [ 78.399493] [<ffffffff8109d329>] ? lock_release_holdtime+0xa3/0xab [ 78.399875] [<ffffffff812e906f>] ? _xfs_log_force+0x1de/0x226 [ 78.400239] [<ffffffff812e9074>] _xfs_log_force+0x1e3/0x226 [ 78.400599] [<ffffffff81067c10>] ? try_to_wake_up+0x28b/0x28b [ 78.400964] [<ffffffff813052f9>] ? xfs_inode_ag_iterator+0x6d/0x8f [ 78.401347] [<ffffffff81164771>] ? __sync_filesystem+0x7a/0x7a [ 78.401715] [<ffffffff812e90cf>] xfs_log_force+0x18/0x3a [ 78.402064] [<ffffffff813053e8>] xfs_quiesce_data+0x3a/0x96 [ 78.402424] [<ffffffff81302f82>] xfs_fs_sync_fs+0x21/0x48 [ 78.402777] [<ffffffff8116475d>] __sync_filesystem+0x66/0x7a [ 78.403138] [<ffffffff81164787>] sync_one_sb+0x16/0x18 [ 78.403482] [<ffffffff81142d37>] iterate_supers+0x72/0xc8 [ 78.403835] [<ffffffff811646bc>] sync_filesystems+0x20/0x22 [ 78.404198] [<ffffffff811647fa>] sys_sync+0x21/0x33 [ 78.404533] [<ffffffff8192f742>] system_call_fastpath+0x16/0x1b [ 221.106751] sync D 0000000000000007 5296 3136 3116 0x00000000 [ 221.107229] ffff8800aee7bb40 0000000000000046 ffffffff819252fa 0000000000000000 [ 221.107903] ffff8800aee7a000 00000000001d1a40 00000000001d1a40 ffff88009c6a45e0 [ 221.108574] 00000000001d1a40 ffff8800aee7bfd8 00000000001d1a40 ffff8800aee7bfd8 [ 221.109246] Call Trace: [ 221.109486] [<ffffffff819252fa>] ? schedule+0x374/0x992 [ 221.109831] [<ffffffff81928012>] ? rwsem_down_failed_common+0xc7/0x103 [ 221.110225] [<ffffffff81092be5>] ? local_clock+0x41/0x5a [ 221.110574] [<ffffffff8109d329>] ? lock_release_holdtime+0xa3/0xab [ 221.110956] [<ffffffff81928012>] ? rwsem_down_failed_common+0xc7/0x103 [ 221.111352] [<ffffffff8192801e>] rwsem_down_failed_common+0xd3/0x103 [ 221.111738] [<ffffffff81928075>] rwsem_down_read_failed+0x12/0x14 [ 221.112116] [<ffffffff813c81a4>] call_rwsem_down_read_failed+0x14/0x30 [ 221.112513] [<ffffffff812de648>] ? xfs_ilock+0x4d/0xba [ 221.112855] [<ffffffff812de62b>] ? xfs_ilock+0x30/0xba [ 221.113199] [<ffffffff8109153b>] ? down_read_nested+0x74/0x89 [ 221.113563] [<ffffffff812de648>] ? xfs_ilock+0x4d/0xba [ 221.113904] [<ffffffff812de648>] xfs_ilock+0x4d/0xba [ 221.114243] [<ffffffff81304ee0>] xfs_wait_ioend_cb+0x1b/0x36 [ 221.114604] [<ffffffff813051a9>] xfs_inode_ag_walk+0x1a5/0x288 [ 221.114973] [<ffffffff81305263>] ? xfs_inode_ag_walk+0x25f/0x288 [ 221.115348] [<ffffffff81304ec5>] ? xfs_sync_inode_data+0x9d/0x9d [ 221.115722] [<ffffffff8109d329>] ? lock_release_holdtime+0xa3/0xab [ 221.116103] [<ffffffff812efb94>] ? xfs_check_sizes+0x160/0x160 [ 221.116473] [<ffffffff812efbfc>] ? xfs_perag_get+0x68/0xd1 [ 221.116828] [<ffffffff812efc14>] ? xfs_perag_get+0x80/0xd1 [ 221.117185] [<ffffffff812efb94>] ? xfs_check_sizes+0x160/0x160 [ 221.117553] [<ffffffff813052f9>] ? xfs_inode_ag_iterator+0x6d/0x8f [ 221.117936] [<ffffffff81304ec5>] ? xfs_sync_inode_data+0x9d/0x9d [ 221.118312] [<ffffffff813052d3>] xfs_inode_ag_iterator+0x47/0x8f [ 221.118689] [<ffffffff81164771>] ? __sync_filesystem+0x7a/0x7a [ 221.119058] [<ffffffff813053d1>] xfs_quiesce_data+0x23/0x96 [ 221.119417] [<ffffffff81302f82>] xfs_fs_sync_fs+0x21/0x48 [ 221.119771] [<ffffffff8116475d>] __sync_filesystem+0x66/0x7a [ 221.120134] [<ffffffff81164787>] sync_one_sb+0x16/0x18 [ 221.120477] [<ffffffff81142d37>] iterate_supers+0x72/0xc8 [ 221.120828] [<ffffffff811646bc>] sync_filesystems+0x20/0x22 [ 221.121188] [<ffffffff811647fa>] sys_sync+0x21/0x33 [ 349.972430] sync D 0000000000000007 5296 3312 3292 0x00000000 [ 349.972907] ffff88009c77fc08 0000000000000046 ffffffff819252fa 0000000100000070 [ 349.973575] ffff88009c77e000 00000000001d1a40 00000000001d1a40 ffff8800b56b0000 [ 349.974250] 00000000001d1a40 ffff88009c77ffd8 00000000001d1a40 ffff88009c77ffd8 [ 349.974921] Call Trace: [ 349.975159] [<ffffffff819252fa>] ? schedule+0x374/0x992 [ 349.975504] [<ffffffff8109d329>] ? lock_release_holdtime+0xa3/0xab [ 349.975880] [<ffffffff8108d5c5>] ? prepare_to_wait+0x6c/0x79 [ 349.976238] [<ffffffff812fade3>] xfs_ioend_wait+0x87/0x9f [ 349.976588] [<ffffffff8108d332>] ? wake_up_bit+0x2a/0x2a [ 349.976932] [<ffffffff812de648>] ? xfs_ilock+0x4d/0xba [ 349.977272] [<ffffffff81304ee8>] xfs_wait_ioend_cb+0x23/0x36 [ 349.977631] [<ffffffff813051a9>] xfs_inode_ag_walk+0x1a5/0x288 [ 349.977997] [<ffffffff81305263>] ? xfs_inode_ag_walk+0x25f/0x288 [ 349.978371] [<ffffffff81304ec5>] ? xfs_sync_inode_data+0x9d/0x9d [ 349.978748] [<ffffffff8109d329>] ? lock_release_holdtime+0xa3/0xab [ 349.979129] [<ffffffff812efb94>] ? xfs_check_sizes+0x160/0x160 [ 349.979499] [<ffffffff812efbfc>] ? xfs_perag_get+0x68/0xd1 [ 349.979856] [<ffffffff812efc14>] ? xfs_perag_get+0x80/0xd1 [ 349.980211] [<ffffffff812efb94>] ? xfs_check_sizes+0x160/0x160 [ 349.980582] [<ffffffff813052f9>] ? xfs_inode_ag_iterator+0x6d/0x8f [ 349.980964] [<ffffffff81304ec5>] ? xfs_sync_inode_data+0x9d/0x9d [ 349.981338] [<ffffffff813052d3>] xfs_inode_ag_iterator+0x47/0x8f [ 349.981717] [<ffffffff81164771>] ? __sync_filesystem+0x7a/0x7a [ 349.982085] [<ffffffff813053d1>] xfs_quiesce_data+0x23/0x96 [ 349.982443] [<ffffffff81302f82>] xfs_fs_sync_fs+0x21/0x48 [ 349.982798] [<ffffffff8116475d>] __sync_filesystem+0x66/0x7a [ 349.983159] [<ffffffff81164787>] sync_one_sb+0x16/0x18 [ 349.983504] [<ffffffff81142d37>] iterate_supers+0x72/0xc8 [ 349.983857] [<ffffffff811646bc>] sync_filesystems+0x20/0x22 [ 349.984215] [<ffffffff811647fa>] sys_sync+0x21/0x33 [ 349.984552] [<ffffffff8192f742>] system_call_fastpath+0x16/0x1b [ 419.417496] sync D 0000000000000006 5296 3412 3392 0x00000000 [ 419.417972] ffff8800aef6dc08 0000000000000046 ffffffff819252fa 0000000100000070 [ 419.418645] ffff8800aef6c000 00000000001d1a40 00000000001d1a40 ffff8800b704a2f0 [ 419.419317] 00000000001d1a40 ffff8800aef6dfd8 00000000001d1a40 ffff8800aef6dfd8 [ 419.419987] Call Trace: [ 419.420228] [<ffffffff819252fa>] ? schedule+0x374/0x992 [ 419.420576] [<ffffffff8109d329>] ? lock_release_holdtime+0xa3/0xab [ 419.420957] [<ffffffff8108d5c5>] ? prepare_to_wait+0x6c/0x79 [ 419.421321] [<ffffffff812fade3>] xfs_ioend_wait+0x87/0x9f [ 419.421673] [<ffffffff8108d332>] ? wake_up_bit+0x2a/0x2a [ 419.422020] [<ffffffff812de648>] ? xfs_ilock+0x4d/0xba [ 419.422364] [<ffffffff81304ee8>] xfs_wait_ioend_cb+0x23/0x36 [ 419.422724] [<ffffffff813051a9>] xfs_inode_ag_walk+0x1a5/0x288 [ 419.423093] [<ffffffff81305263>] ? xfs_inode_ag_walk+0x25f/0x288 [ 419.423470] [<ffffffff81304ec5>] ? xfs_sync_inode_data+0x9d/0x9d [ 419.423844] [<ffffffff8109d329>] ? lock_release_holdtime+0xa3/0xab [ 419.424226] [<ffffffff812efb94>] ? xfs_check_sizes+0x160/0x160 [ 419.424596] [<ffffffff812efbfc>] ? xfs_perag_get+0x68/0xd1 [ 419.424951] [<ffffffff812efc14>] ? xfs_perag_get+0x80/0xd1 [ 419.425307] [<ffffffff812efb94>] ? xfs_check_sizes+0x160/0x160 [ 419.425676] [<ffffffff813052f9>] ? xfs_inode_ag_iterator+0x6d/0x8f [ 419.426058] [<ffffffff81304ec5>] ? xfs_sync_inode_data+0x9d/0x9d [ 419.426435] [<ffffffff813052d3>] xfs_inode_ag_iterator+0x47/0x8f [ 419.426812] [<ffffffff81164771>] ? __sync_filesystem+0x7a/0x7a [ 419.427180] [<ffffffff813053d1>] xfs_quiesce_data+0x23/0x96 [ 419.427542] [<ffffffff81302f82>] xfs_fs_sync_fs+0x21/0x48 [ 419.427895] [<ffffffff8116475d>] __sync_filesystem+0x66/0x7a [ 419.428258] [<ffffffff81164787>] sync_one_sb+0x16/0x18 [ 419.428601] [<ffffffff81142d37>] iterate_supers+0x72/0xc8 [ 419.428954] [<ffffffff811646bc>] sync_filesystems+0x20/0x22 [ 419.429315] [<ffffffff811647fa>] sys_sync+0x21/0x33 [ 419.429648] [<ffffffff8192f742>] system_call_fastpath+0x16/0x1b Thanks, Fengguang -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html