From: Douglas Mayne on
This kernel reportedly fixes some XFS filesystem deadlock bugs. I
personally ran into deadlocks on 2.6.33.2, which made that version
unreliable for me. BTW, I was testing with XFS on a loopback file when I
ran into this bug. Anyway, my advice is to skip over 2.6.33.2 now that a
new patch is available. YMMV.
From: Grant on
On Wed, 28 Apr 2010 14:31:03 +0000 (UTC), Douglas Mayne <doug(a)localhost.localnet> wrote:

>This kernel reportedly fixes some XFS filesystem deadlock bugs. I
>personally ran into deadlocks on 2.6.33.2, which made that version
>unreliable for me. BTW, I was testing with XFS on a loopback file when I
>ran into this bug. Anyway, my advice is to skip over 2.6.33.2 now that a
>new patch is available. YMMV.

Always go for the latest .stable kernel in a particular release --
they're put out to fix known issues where the fix has also gone into
upstream.

Note also that 2.6.32.xx is slated for long term maintenance, for a
couple years going by how long it took to get to 2.6.27.46.

Grant.
--
http://bugs.id.au/
From: Douglas Mayne on
On Thu, 29 Apr 2010 06:22:26 +1000, Grant wrote:

> On Wed, 28 Apr 2010 14:31:03 +0000 (UTC), Douglas Mayne
> <doug(a)localhost.localnet> wrote:
>
>>This kernel reportedly fixes some XFS filesystem deadlock bugs. I
>>personally ran into deadlocks on 2.6.33.2, which made that version
>>unreliable for me. BTW, I was testing with XFS on a loopback file when I
>>ran into this bug. Anyway, my advice is to skip over 2.6.33.2 now that a
>>new patch is available. YMMV.
>
> Always go for the latest .stable kernel in a particular release --
>
I agree. Mainly, I mentioned it because 2.6.32.2 is currently available
precompiled for Slackware -current, and based on my experience caution
with it is warranted.

> they're put out to fix known issues where the fix has also gone into
> upstream.
>
> Note also that 2.6.32.xx is slated for long term maintenance, for a
> couple years going by how long it took to get to 2.6.27.46.
>
> Grant.
>
Note: comment inline.

Thanks for the infomation. I am curious where the period of extended
maintenance is documented. For example, I have some machines running
2.6.30.10, which is marked as "stable," but has not been updated since
December 2009. It is also good to know about 2.6.32 because there were
major changes to the scheduler options for 2.6.33 (most schedulers except
CFQ were dropped from future development, IIUC). Previously, I have used
"anticipatory," but will give CFQ another try, I guess.

--
Douglas Mayne
From: Grant on
On Thu, 29 Apr 2010 15:00:25 +0000 (UTC), Douglas Mayne <doug(a)localhost.localnet> wrote:

>On Thu, 29 Apr 2010 06:22:26 +1000, Grant wrote:
>
>> On Wed, 28 Apr 2010 14:31:03 +0000 (UTC), Douglas Mayne
>> <doug(a)localhost.localnet> wrote:
>>
>>>This kernel reportedly fixes some XFS filesystem deadlock bugs. I
>>>personally ran into deadlocks on 2.6.33.2, which made that version
>>>unreliable for me. BTW, I was testing with XFS on a loopback file when I
>>>ran into this bug. Anyway, my advice is to skip over 2.6.33.2 now that a
>>>new patch is available. YMMV.
>>
>> Always go for the latest .stable kernel in a particular release --
>>
>I agree. Mainly, I mentioned it because 2.6.32.2 is currently available
>precompiled for Slackware -current, and based on my experience caution
>with it is warranted.
>
>> they're put out to fix known issues where the fix has also gone into
>> upstream.
>>
>> Note also that 2.6.32.xx is slated for long term maintenance, for a
>> couple years going by how long it took to get to 2.6.27.46.
>>
>> Grant.
>>
>Note: comment inline.
>
>Thanks for the infomation. I am curious where the period of extended
>maintenance is documented. For example, I have some machines running
>2.6.30.10, which is marked as "stable," but has not been updated since
>December 2009.

Most recent statement: http://lkml.org/lkml/2010/1/18/377

And that was in response to a query similar to yours :) GregKH seems to be
the sole person running .stable these days. The policy for usual kernels is
for the most recent and second most recent release to receive stable patch
attention. So 2.6.30 goes without maintenance since 2.6.33 was released.

The extended maintenance kernel version is not formally documented on LKML
but has now turned up enough times that we may expect a new one to be
announced every couple years.

Kernels marked for longer term maintenance usually go into several distros,
though I've not seen a list of which distros. Slackware ignores these as
their release cycle is timed to other factors we know nothing of, "Released
when it's ready" :)

> It is also good to know about 2.6.32 because there were
>major changes to the scheduler options for 2.6.33 (most schedulers except
>CFQ were dropped from future development, IIUC). Previously, I have used
>"anticipatory," but will give CFQ another try, I guess.

Deadline is usually the best, on the KISS principle. The complex schedulers
may fail (to perform) in corner cases. Though I'm not using a linux desktop,
maybe the more complex schedulers do a perceivably better job there?

Grant.
--
http://bugs.id.au/
From: Sylvain Robitaille on
Grant wrote:

> ... The complex schedulers may fail (to perform) in corner cases.

Reference?

: charlotte[syl] ~; uptime
22:54:39 up 134 days, 4:11, 46 users, load average: 0.22, 0.16, 0.10
: charlotte[syl] ~; cat /etc/slackware-version
Slackware 13.0.0.0.0
: charlotte[syl] ~; uname -a
Linux charlotte 2.6.29.6-smp #2 SMP Mon Aug 17 00:52:54 CDT 2009 i686 Intel(R) Pentium(R) 4 CPU 3.00GHz GenuineIntel GNU/Linux
: charlotte[syl] ~; zgrep IOSCHED /proc/config.gz
CONFIG_IOSCHED_NOOP=y
CONFIG_IOSCHED_AS=y
CONFIG_IOSCHED_DEADLINE=y
CONFIG_IOSCHED_CFQ=y
CONFIG_DEFAULT_IOSCHED="cfq"

--
----------------------------------------------------------------------
Sylvain Robitaille syl(a)encs.concordia.ca

Systems analyst / AITS Concordia University
Faculty of Engineering and Computer Science Montreal, Quebec, Canada
----------------------------------------------------------------------