Re: [PATCH v12 3/6] CPM/QE: use genalloc to manage CPM/QE muram

[Scott Wood <scottwood@freescale.com>]

Don't send HTML e-mail.

On Fri, 2015-10-23 at 02:06 -0500, Zhao Qiang-B45475 wrote:
> On Fri, 2015-10-23 at 11:00 AM, Wood Scott-B07421 <scottwood@freescale.com> 
> wrote:
> > -----Original Message-----
> > From: Wood Scott-B07421
> > Sent: Friday, October 23, 2015 11:00 AM
> > To: Zhao Qiang-B45475 <qiang.zhao@freescale.com>
> > Cc: linux-kernel@vger.kernel.org; linuxppc-dev@lists.ozlabs.org;
> > lauraa@codeaurora.org; Xie Xiaobo-R63061 <X.Xie@freescale.com>;
> > benh@kernel.crashing.org; Li Yang-Leo-R58472 <LeoLi@freescale.com>;
> > paulus@samba.org
> > Subject: Re: [PATCH v12 3/6] CPM/QE: use genalloc to manage CPM/QE muram
> >
> > On Wed, 2015-10-14 at 15:16 +0800, Zhao Qiang wrote:
> > > -/**
> > > +/*
> > >   * cpm_muram_alloc - allocate the requested size worth of multi-user 
> ram
> > >   * @size: number of bytes to allocate
> > >   * @align: requested alignment, in bytes @@ -141,59 +151,102 @@ out:
> > >   */
> > >  unsigned long cpm_muram_alloc(unsigned long size, unsigned long
> > > align)  {
> > > -     unsigned long start;
> > >       unsigned long flags;
> > > -
> > > +     unsigned long start;
> > > +     static struct genpool_data_align muram_pool_data;
> > >       spin_lock_irqsave(&cpm_muram_lock, flags);
> > > -     cpm_muram_info.alignment = align;
> > > -     start = rh_alloc(&cpm_muram_info, size, "commproc");
> > > -     memset(cpm_muram_addr(start), 0, size);
> > > +     muram_pool_data.align = align;
> > > +     gen_pool_set_algo(muram_pool, gen_pool_first_fit_align,
> > > +                       &muram_pool_data);
> > > +     start = cpm_muram_alloc_common(size, &muram_pool_data);
> > >       spin_unlock_irqrestore(&cpm_muram_lock, flags);
> > > -
> > >       return start;
> > >  }
> > >  EXPORT_SYMBOL(cpm_muram_alloc);
> >
> > Why is muram_pool_data static?  Why is it being passed to
> > gen_pool_set_algo()? 
> Cpm_muram use both align algo and fixed algo, so we need to set 
> corresponding algo and
> Algo data.

The data gets passed in via gen_pool_alloc_data().  The point was to allow it 
to be on the caller's stack, not a long-lived data structure shared by all 
callers and needing synchronization.

> >The whole reason we're adding gen_pool_alloc_data()
> > is to avoid that.  Do we need gen_pool_alloc_algo() too?
>  
> We add gen_pool_alloc_data() to pass data to algo, because align algo and 
> fixed algo,
> Because align and fixed algos need specific data.

And my point is that because of that, it seems like we need a version that 
accepts an algorithm as well.

-Scott