lkml.org 
[lkml]   [2014]   [Jul]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH] blkio: Release blkg infrastructure only after last policy is deactivated.
From
Tejun, I do not have another iteration of the patch yet.

we have two customers who reported this problem. One of them has not
responded to requests to test a/any patch and other one is not able to
recreate this
with the latest release. And I am not able to recreate this problem
on my own but
working on a setup. So if you have any patch, I would be happy to
test/deploy it on a
first setup I can get hold of.

I was also trying to figure out whether this is some refcounting issue
or not since
I was thinking that put by anybody other than blk_cleanup_queue() should
not result on freeing/blk_release_queue() i.e. put by blk_cleanup_queue() should
always be the last put.

Regards,

Shirish


On Fri, Jul 4, 2014 at 12:18 PM, Tejun Heo <tj@kernel.org> wrote:
> On Tue, Jun 24, 2014 at 05:43:40PM -0400, Tejun Heo wrote:
>> Hello,
>>
>> On Tue, Jun 24, 2014 at 04:37:58PM -0500, Shirish Pargaonkar wrote:
>> > When we start from blk_cleanup_queue(), we put request queue in bypass mode,
>> > drain it (and service queues), and then destroy blkcgs (explicitly)
>> >
>> > When we start from blk_release_queue(), we do not drain first and then
>> > destroy blkcgs. So if we destroy blkcg and then call (implicitly) and
>> > bail out of
>> > blk_drain_queue, we would not have drained the service queues which
>> > is not what we want.
>>
>> I'm not really following you. What do you mean "when we start from
>> blk_release_queue()"? blk_release_queue() is called after the last
>> put which can only follow blk_cleanup_queue() if the queue is fully
>> initialized. The queue is already in bypass mode and fully drained by
>> the time control reaches blk_release_queue(). Module [un]load
>> re-invoking the path doesn't change anything.
>>
>> > I do not see any harm in waiting till end to release blkcgs (as I understand).
>>
>> Well, the harm there is not freeing those blkgs unless all the blkcg
>> policies are unloaded which is usually never on most systems.
>
> Ping. We have a patch which makes this problem more visible. Are you
> still planning to re-spin the patch?
>
> Thanks.
>
> --
> tejun


\
 
 \ /
  Last update: 2014-07-04 21:41    [W:0.097 / U:0.240 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site