Development problem: uninterruptable processes

Development problem: uninterruptable processes

[Anselm Busse]

Hello everyone,

I am not sure whether I am right in this mailing list or not, but I hope
so...

I am relative new to kernel development and especially to the device mapper.
Currently I am developing a driver using the device mapper API. It works
perfectly with one exception. After some time of heavy load every process
that tries to access the mapped block device gets stuck in the
uninterruptable state.

I already checked every entry and callback routine in my driver, but it gets
not stuck there. I also checked every lock I am using in my driver and every
one of them gets released properly. I am pretty stuck at the moment, because
I don't know where to look else for the problem and was hoping that some one
of you might have experienced similar problems or an idea where to look for
it. 
I don't know if it is important but I've noticed (using the kernel debug
option for detecting hung tasks) that every process seems to hang in a write
call and I've tried kernel versions 2.6.31 and 2.6.33.

I would appreciate every bit of help because I am completely out of ideas!

Regards, Anselm

Re: Development problem: uninterruptable processes

[Heinz Mauelshagen]

Hi Anselm,

you're not explaining the design and functions of your driver here, so
it's hard to tell. You're telling though, starvation happens in the
write path so you want to look if you're in an OOM situation, presumably
the locking is correct as you claim. Another possibility may be that
you're scheduling a thread handling your write not properly.

You better provide the source code of your driver to allow for better
advice.

Regards,
Heinz

On Sat, 2010-04-24 at 18:00 +0200, Anselm Busse wrote:
> Hello everyone,
> 
> I am not sure whether I am right in this mailing list or not, but I hope
> so...
> 
> I am relative new to kernel development and especially to the device mapper.
> Currently I am developing a driver using the device mapper API. It works
> perfectly with one exception. After some time of heavy load every process
> that tries to access the mapped block device gets stuck in the
> uninterruptable state.
> 
> I already checked every entry and callback routine in my driver, but it gets
> not stuck there. I also checked every lock I am using in my driver and every
> one of them gets released properly. I am pretty stuck at the moment, because
> I don't know where to look else for the problem and was hoping that some one
> of you might have experienced similar problems or an idea where to look for
> it. 
> I don't know if it is important but I've noticed (using the kernel debug
> option for detecting hung tasks) that every process seems to hang in a write
> call and I've tried kernel versions 2.6.31 and 2.6.33.
> 
> I would appreciate every bit of help because I am completely out of ideas!
> 
> Regards, Anselm
> 
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel