Re: [PATCH 03/16] dma-debug: add hash functions for dma_debug_entries

[Joerg Roedel <joro@8bytes.org>]

On Fri, Jan 09, 2009 at 09:23:39PM +0300, Evgeniy Polyakov wrote:
> On Fri, Jan 09, 2009 at 07:14:46PM +0100, Joerg Roedel (joro@8bytes.org) wrote:
> > > > +static struct hash_bucket *get_hash_bucket(struct dma_debug_entry *entry,
> > > > +					   unsigned long *flags)
> > > > +{
> > > > +	int idx = hash_fn(entry);
> > > > +	unsigned long __flags;
> > > > +
> > > > +	spin_lock_irqsave(&dma_entry_hash[idx].lock, __flags);
> > > > +	*flags = __flags;
> > > > +	return &dma_entry_hash[idx];
> > > > +}
> > > > +
> > > > +/*
> > > > + * Give up exclusive access to the hash bucket
> > > > + */
> > > > +static void put_hash_bucket(struct hash_bucket *bucket,
> > > > +			    unsigned long *flags)
> > > > +{
> > > > +	unsigned long __flags = *flags;
> > > > +
> > > > +	spin_unlock_irqrestore(&bucket->lock, __flags);
> > > > +}
> > > 
> > > Why do you need such ugly helpers?
> > 
> > Because everything else I thought about here was even more ugly. But
> > maybe you have a better idea? I tried to lock directly in the debug_
> > functions. But this is ugly and unnecessary code duplication.
> 
> I believe that having direct locking in the debug_ functions is not a
> duplication, anyone will have a direct vision on the locking and hash
> array dereference, and this will be just one additional line compared to
> the get_* call and the same number of lines for the put :)

Even more additional lines because of the additional variables needed in
every function. Anyway, I try it and if it does not look good I will
keep that change ;)

Joerg