On 07/11/2014 10:38 AM, Peter Zijlstra wrote:
On Fri, Jul 11, 2014 at 10:33:15AM +0200, Vlastimil Babka wrote:
Quoting Hugh from previous mail in this thread:
[ 363.600969] INFO: task trinity-c327:9203 blocked for more than 120 seconds.
[ 363.605359] Not tainted 3.16.0-rc4-next-20140708-sasha-00022-g94c7290-dirty #772
[ 363.609730] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 363.615861] trinity-c327 D 000000000000000b 13496 9203 8559 0x10000004
[ 363.620284] ffff8800b857bce8 0000000000000002 ffffffff9dc11b10 0000000000000001
[ 363.624468] ffff880104860000 ffff8800b857bfd8 00000000001d7740 00000000001d7740
[ 363.629118] ffff880104863000 ffff880104860000 ffff8800b857bcd8 ffff8801eaed8868
[ 363.633879] Call Trace:
[ 363.635442] [<ffffffff9a4dc535>] schedule+0x65/0x70
[ 363.638638] [<ffffffff9a4dc948>] schedule_preempt_disabled+0x18/0x30
[ 363.642833] [<ffffffff9a4df0a5>] mutex_lock_nested+0x2e5/0x550
[ 363.646599] [<ffffffff972a4d7c>] ? shmem_fallocate+0x6c/0x350
[ 363.651319] [<ffffffff9719b721>] ? get_parent_ip+0x11/0x50
[ 363.654683] [<ffffffff972a4d7c>] ? shmem_fallocate+0x6c/0x350
[ 363.658264] [<ffffffff972a4d7c>] shmem_fallocate+0x6c/0x350
So it's trying to acquire i_mutex at shmem_fallocate+0x6c...
[ 363.662010] [<ffffffff971bd96e>] ? put_lock_stats.isra.12+0xe/0x30
[ 363.665866] [<ffffffff9730c043>] do_fallocate+0x153/0x1d0
[ 363.669381] [<ffffffff972b472f>] SyS_madvise+0x33f/0x970
[ 363.672906] [<ffffffff9a4e3f13>] tracesys+0xe1/0xe6
[ 363.682900] 2 locks held by trinity-c327/9203:
[ 363.684928] #0: (sb_writers#12){.+.+.+}, at: [<ffffffff9730c02d>] do_fallocate+0x13d/0x1d0
[ 363.715102] #1: (&sb->s_type->i_mutex_key#16){+.+.+.}, at: [<ffffffff972a4d7c>] shmem_fallocate+0x6c/0x350
...but it already holds i_mutex, acquired at shmem_fallocate+0x6c.
Am I reading that correctly?
The output looks like mutex #1 is already taken, but actually the process is
sleeping when trying to take it. It appears that the output has taken
mutex_acquire_nest() action into account, but doesn't distinguish if
lock_acquired() already happened or not.
The call trace is very clear on it that its not. I've never found this
to be a problem in practise. You need to engage your brain anyhow, this
little bit extra isn't going to make a difference or not.
OK, but what about the case of "Showing all locks held in the system:"
output where you don't have the stacktraces? Wouldn't it be better if that
distinguished locks already taken and locks being taken?
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html