[Intel-gfx] [PATCH 02/26] dma-buf: add dma_resv_for_each_fence

[Daniel Vetter]

On Fri, Sep 17, 2021 at 03:27:55PM +0200, Daniel Vetter wrote:
> On Fri, Sep 17, 2021 at 02:34:49PM +0200, Christian König wrote:
> > A simpler version of the iterator to be used when the dma_resv object is
> > locked.
> > 
> > Signed-off-by: Christian König <christian.koenig at amd.com>
> > ---
> >  drivers/dma-buf/dma-resv.c | 33 +++++++++++++++++++++++++++++++++
> >  include/linux/dma-resv.h   | 17 +++++++++++++++++
> >  2 files changed, 50 insertions(+)
> > 
> > diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c
> > index 3e77cad2c9d4..a3c79a99fb44 100644
> > --- a/drivers/dma-buf/dma-resv.c
> > +++ b/drivers/dma-buf/dma-resv.c
> > @@ -384,6 +384,39 @@ struct dma_fence *dma_resv_iter_walk_unlocked(struct dma_resv_iter *cursor,
> >  }
> >  EXPORT_SYMBOL_GPL(dma_resv_iter_walk_unlocked);
> >  
> > +/**
> > + * dma_resv_iter_walk - walk over fences in a dma_resv obj
> > + * @cursor: cursor to record the current position
> > + * @first: if we should start over
> > + *
> > + * Return all the fences in the dma_resv object while holding the
> > + * dma_resv::lock.

I think we should document here that the fence is valid for as long as the
dma_resv_lock is held, which is not like the _unlocked version, where the
fence stops being valid either on the next call to iter_next() or on the
call to iter_end() to clean up everything.

Might be good to clarify that also for the unlocked version to be really
precise here when the fence is valid and when not.
-Daniel

> > + */
> > +struct dma_fence *dma_resv_iter_walk(struct dma_resv_iter *cursor, bool first)
> > +{
> > +	dma_resv_assert_held(cursor->obj);
> > +
> > +	cursor->is_first = first;
> > +	if (first) {
> > +		struct dma_fence *fence;
> > +
> > +		cursor->index = -1;
> > +		cursor->fences = dma_resv_shared_list(cursor->obj);
> > +
> > +		fence = dma_resv_excl_fence(cursor->obj);
> > +		if (fence)
> > +			return fence;
> > +	}
> 
> I think you can still use the shared iter_begin/end functions even with my
> suggestions for patch 1, but would mean changes here too.
> 
> > +
> > +	if (!cursor->all_fences || !cursor->fences ||
> > +	    ++cursor->index >= cursor->fences->shared_count)
> > +		return NULL;
> > +
> > +	return rcu_dereference_protected(cursor->fences->shared[cursor->index],
> > +					 dma_resv_held(cursor->obj));
> > +}
> > +EXPORT_SYMBOL_GPL(dma_resv_iter_walk);
> > +
> >  /**
> >   * dma_resv_copy_fences - Copy all fences from src to dst.
> >   * @dst: the destination reservation object
> > diff --git a/include/linux/dma-resv.h b/include/linux/dma-resv.h
> > index 693d16117153..8c968f8c9d33 100644
> > --- a/include/linux/dma-resv.h
> > +++ b/include/linux/dma-resv.h
> > @@ -179,6 +179,7 @@ struct dma_resv_iter {
> >  
> >  struct dma_fence *dma_resv_iter_walk_unlocked(struct dma_resv_iter *cursor,
> >  					      bool first);
> > +struct dma_fence *dma_resv_iter_walk(struct dma_resv_iter *cursor, bool first);
> >  
> >  /**
> >   * dma_resv_iter_begin - initialize a dma_resv_iter object
> > @@ -233,6 +234,22 @@ static inline bool dma_resv_iter_is_exclusive(struct dma_resv_iter *cursor)
> >  	for (fence = dma_resv_iter_walk_unlocked(cursor, true);		\
> >  	     fence; fence = dma_resv_iter_walk_unlocked(cursor, false))
> >  
> > +/**
> > + * dma_resv_for_each_fence - fence iterator
> > + * @cursor: a struct dma_resv_iter pointer
> > + * @obj: a dma_resv object pointer
> > + * @all_fences: true if all fences should be returned
> > + * @fence: the current fence
> > + *
> > + * Iterate over the fences in a struct dma_resv object while holding the
> > + * dma_resv::lock. @all_fences controls if the shared fences are returned as
> 
> &dma_resv.lock is how you reference struct members in kerneldoc. I think
> you had this also in patch 1.
> 
> > + * well. The cursor initialisation is part of the iterator.
> 
> Please also link to the iter_begin/end functions here.
> 
> Aside from doc nits and obviously changes due to changes in patch 1 (if we
> do them), this looks good.
> -Daniel
> 
> > + */
> > +#define dma_resv_for_each_fence(cursor, obj, all_fences, fence)	\
> > +	for (dma_resv_iter_begin(cursor, obj, all_fences),	\
> > +	     fence = dma_resv_iter_walk(cursor, true); fence;	\
> > +	     fence = dma_resv_iter_walk(cursor, false))
> > +
> >  #define dma_resv_held(obj) lockdep_is_held(&(obj)->lock.base)
> >  #define dma_resv_assert_held(obj) lockdep_assert_held(&(obj)->lock.base)
> >  
> > -- 
> > 2.25.1
> > 
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch