Hi, On Tue 04-09-12 11:54:54, Dirk Müller wrote:
starting since 3.6rcX in factory, I have regularly lockups in all code that tries to write to the filesystem. It doesn't happen immediately, but after a bit of an update of about a day, and with autobuild(or other builds) running in the background.
The symptom is that starting at one particular point in time, no new writes to the filessytem are possible:
All tasks are in "D" state with the backtrace ending in:
[87428.047819] [<ffffffff8116548b>] __sb_start_write+0xcb/0x110 [87428.047826] [<ffffffff811801db>] mnt_want_write+0x1b/0x50 [87428.047832] [<ffffffff81172142>] do_last+0xa12/0xe70 [87428.047839] [<ffffffff81172661>] path_openat+0xc1/0x4c0 [87428.047846] [<ffffffff81172ec4>] do_filp_open+0x44/0xb0 [87428.047852] [<ffffffff81162865>] do_sys_open+0xf5/0x1e0 [87428.047859] [<ffffffff8156ad3d>] system_call_fastpath+0x1a/0x1f
(basically mnt_want_write / __sb_start_write) Hmm, this is my new filesystem freezing (protection) code ;-). Can you attach output of sysrq-w (run echo w >/proc/sysrq-trigger, output is in dmesg)? And file a bug, assign to me please. Thanks.
the filesystem is ext4 mounted with data=writeback and noatime.
Any idea where to start hunting? I went back from kernel-desktop to kernel- default, and I'll try kernel-vanilla next.
Honza
--
Jan Kara