Re: [PATCH 1/3] padata: separate serial and parallel cpumasks

[Steffen Klassert <steffen.klassert@secunet.com>]

On Fri, Jul 02, 2010 at 01:07:44PM +0400, Dan Kruchinin wrote:
> >
> > The above is not save against cpu hotplug. You initialize ctx->cb_cpu
> > to -1 when the transformation is initialized. So you choose your
> > callback cpu with the first call to pcrypt_do_parallel() and then never
> > again. What if the coosen callback cpu goes offline?
> >
> > Also I don't understand why you changed the prototype of pcrypt_do_parallel
> > and added a 'count' to pcrypt_aead_ctx. The only change you have to do, is
> > to
> > check whether the choosen callback cpu is still in the cpu_active_mask
> > _and_
> > in your new padata callback cpumask and update the callback cpu accordingly
> > if not. Checking whether the callback cpu is in the callback cpumask needs
> > some special care in your other patch of course, because you can change
> > this
> > cpumask from userspace then.
> >
> 
> Well, my point was to reduce the code and select callback CPU only once
> according to
> serial cpumask of given data instance. In case when  I modify serial cpumask
> of given padata
> instance I have to do callback cpu calculation twice in worst case. The
> first time in pcrypt_aead_init_tfm
> and the second time in pcrypt_do_parallel if cb_cpu is not set in my serial
> cpumask.
> So I decided to do it only once in pcrypt_do_parallel.
> 

But the active cpumask, and now also your serial cpumask might change.
We need to catch this changes somehow, that's why I checked the active
cpumask against the callback cpu.

Steffen