Re: more dput lock contentions in 2.6.38-rc?

From: Nick Piggin
Date: Mon Jan 24 2011 - 21:09:55 EST


On Tue, Jan 25, 2011 at 1:01 PM, Shaohua Li <shaohua.li@xxxxxxxxx> wrote:
> On Tue, Jan 25, 2011 at 09:44:45AM +0800, Nick Piggin wrote:
>> On Tue, Jan 25, 2011 at 12:34 PM, Shaohua Li <shaohua.li@xxxxxxxxx> wrote:
>> > On Tue, 2011-01-25 at 09:26 +0800, Nick Piggin wrote:
>> >> On Tue, Jan 25, 2011 at 12:11 PM, Shaohua Li <shaohua.li@xxxxxxxxx> wrote:
>> >> > On Tue, 2011-01-25 at 09:04 +0800, Nick Piggin wrote:
>> >> >> On Tue, Jan 25, 2011 at 11:35 AM, Shaohua Li <shaohua.li@xxxxxxxxx> wrote:
>> >> >> > Hi,
>> >> >> > we are testing dbench benchmark and see big drop of 2.6.38-rc compared
>> >> >> > to 2.6.37 in several machines with 2 sockets or 4 sockets. We have 12
>> >> >> > disks mount to /mnt/stp/dbenchdata/sd*/ and dbench runs against data of
>> >> >> > the disks. According to perf, we saw more lock contentions:
>> >> >> > In 2.6.37: 13.00%        dbench  [kernel.kallsyms]   [k] _raw_spin_lock
>> >> >> > In 2.6.38-rc: 69.45%        dbench  [kernel.kallsyms]   [k]_raw_spin_lock
>> >> >> > -     69.45%        dbench  [kernel.kallsyms]   [k] _raw_spin_lock
>> >> >> >   - _raw_spin_lock
>> >> >> >      - 48.41% dput
>> >> >> >         - 61.17% path_put
>> >> >> >            - 60.47% do_path_lookup
>> >> >> >               + 53.18% user_path_at
>> >> >> >               + 42.13% do_filp_open
>> >> >> >               + 4.69% user_path_parent
>> >> >>
>> >> >> What filesystems are mounted on the path?
>> >> > ext3 or ext4
>> >>
>> >> ext3 or 4 along every step of the path? Are there
>> >> any acls loaded, or security policy running?
>> > all disks are formated with the same fs, just some machines use ext3 and
>> > others ext4. no we don't have acl or security policy.
>> >> It may be possible that they're all coming from
>> >> /proc/ access.
>> > I added trace in dput just after the lock taken. and most files are '/',
>> > 'mnt', 'stp'. the percentage of 'proc' is small actually.
>>
>> Hm, OK well I could send you a patch to gather some statistics for
>> why rcu-walk gets dropped. It'll have to wait until I get home, though.
> Sure, I'm still at office tomorrow. I can test before that day, otherwise
> maybe Tim can help.

There _are_ a lot of dput contentions coming from d_path. Some of the
dentries you're seeing in dput could be coming from path_put in
d_path (mountpoints, eg. 'mnt').

What does an actual snippet from perf with callgraphs look like?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/