Re: cmwq and dm-crypt devices?

[Tejun Heo <tj@kernel.org>]

Hello,

On 11/03/2010 05:13 PM, Andi Kleen wrote:
>> Please fix this regression the patch causes then or give some
>> fundamental arguments why we can't satisfy the two requirements
>> simultaneously: I just don't understand why it is necessary to cause a
>> regression in stacking/low memory support in order to provide the
>> performance boost in the way you desire.  The two things ought to be
>> independent of each other.
> 
> The old stacking relies on per device threads. Per device threads
> cannot be combined with per CPU threads, because that would lead
> to a thread explosion for large device counts.

Dunno the context here but creating WQ_NON_REENTRANT | WQ_MEM_RECLAIM
workqueue per device should achieve what you want.  It will be per-cpu
while reserving only one worker per workqueue, so there won't be
explosion of threads.

Thanks.

-- 
tejun