[Intel-gfx] [PATCH 10/12] drm/i915: Introduce dedicated object VMA iterator

[Chris Wilson]

On Tue, Feb 02, 2016 at 11:06:28AM +0000, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> 
> Purpose is to catch places which iterate the object VMA list
> without holding the big lock.
> 
> Implemented by open coding list_for_each_entry to make the
> macro compatible with existing call sites.
> 
> v2: Error capture runs without the mutex so iterate directly from there.
> v3: Replace WARN_ON with lockdep_assert_held. (Chris Wilson, Daniel Vetter)
> v4: Moved under dedicated CONFIG_DRM_I915_DEBUG and back to WARN_ON.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/i915_debugfs.c      |  8 ++++----
>  drivers/gpu/drm/i915/i915_drv.h          | 11 +++++++++++
>  drivers/gpu/drm/i915/i915_gem.c          | 24 ++++++++++++------------
>  drivers/gpu/drm/i915/i915_gem_gtt.c      |  2 +-
>  drivers/gpu/drm/i915/i915_gem_shrinker.c |  2 +-
>  5 files changed, 29 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> index 863012a2602e..ff444f09ea98 100644
> --- a/drivers/gpu/drm/i915/i915_debugfs.c
> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> @@ -117,7 +117,7 @@ static u64 i915_gem_obj_total_ggtt_size(struct drm_i915_gem_object *obj)
>  	u64 size = 0;
>  	struct i915_vma *vma;
>  
> -	list_for_each_entry(vma, &obj->vma_list, vma_link) {
> +	i915_gem_obj_for_each_vma(vma, obj) {
>  		if (i915_is_ggtt(vma->vm) &&
>  		    drm_mm_node_allocated(&vma->node))
>  			size += vma->node.size;
> @@ -155,7 +155,7 @@ describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
>  		   obj->madv == I915_MADV_DONTNEED ? " purgeable" : "");
>  	if (obj->base.name)
>  		seq_printf(m, " (name: %d)", obj->base.name);
> -	list_for_each_entry(vma, &obj->vma_list, vma_link) {
> +	i915_gem_obj_for_each_vma(vma, obj) {
>  		if (vma->pin_count > 0)
>  			pin_count++;
>  	}
> @@ -164,7 +164,7 @@ describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
>  		seq_printf(m, " (display)");
>  	if (obj->fence_reg != I915_FENCE_REG_NONE)
>  		seq_printf(m, " (fence: %d)", obj->fence_reg);
> -	list_for_each_entry(vma, &obj->vma_list, vma_link) {
> +	i915_gem_obj_for_each_vma(vma, obj) {
>  		seq_printf(m, " (%sgtt offset: %08llx, size: %08llx",
>  			   i915_is_ggtt(vma->vm) ? "g" : "pp",
>  			   vma->node.start, vma->node.size);
> @@ -342,7 +342,7 @@ static int per_file_stats(int id, void *ptr, void *data)
>  		stats->shared += obj->base.size;
>  
>  	if (USES_FULL_PPGTT(obj->base.dev)) {
> -		list_for_each_entry(vma, &obj->vma_list, vma_link) {
> +		i915_gem_obj_for_each_vma(vma, obj) {
>  			struct i915_hw_ppgtt *ppgtt;
>  
>  			if (!drm_mm_node_allocated(&vma->node))
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 905e90f25957..05ef750386df 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -2861,6 +2861,17 @@ struct drm_i915_gem_object *i915_gem_object_create_from_data(
>  void i915_gem_free_object(struct drm_gem_object *obj);
>  void i915_gem_vma_destroy(struct i915_vma *vma);
>  
> +#ifdef CONFIG_DRM_I915_DEBUG
> +  #define i915_gem_obj_for_each_vma(vma, obj) \

This is the wrong name, i915_gem_object is the object on which we are
iterating over, so i915_gem_object_.

> +	for (WARN_ON_ONCE(!mutex_is_locked(&(obj)->base.dev->struct_mutex)), \

I disagree that we should be doing i915-specific lockdep debugging and
*ignoring* the core facilities. If we locked at RCU, they introduced a
new set of _protected/_check interfaces, an idea that could also be used
here to add this or something like this to the core.
Something like

#define list_for_each_entry_check(pos, list, member, lock) \
for (typeof(*lock) == typeof(struct mutex) ? assert_lockdep_held(lock) : assert_spin_locked(lock); \
     pos = list_first_entry(head, typeof(*pos), member); \
     &pos->member != (head); \
     pos = list_next_entry(pos, member))

#define i915_gem_object_for_each_vma(vma, obj) \
	list_for_each_entry_check(vma, &(obj)->vma_list, vma_link, &(obj)->base.dev->struct_mutex)
	
could be lifted easily, and makes i915_gem_object_for_each_vma() easier
to understand (i.e. serves better as documentation).

> +	     vma = list_first_entry(&(obj)->vma_list, typeof(*vma), vma_link);\
> +	     &vma->vma_link != (&(obj)->vma_list); \
> +	     vma = list_next_entry(vma, vma_link))
> +#else
> +  #define i915_gem_obj_for_each_vma(vma, obj) \
> +		list_for_each_entry((vma), &(obj)->vma_list, vma_link)
> +#endif
> +
>  /* Flags used by pin/bind&friends. */
>  #define PIN_MAPPABLE	(1<<0)
>  #define PIN_NONBLOCK	(1<<1)
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index c558887b2084..ce9d0544b42c 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -2454,7 +2454,7 @@ i915_gem_object_retire__read(struct drm_i915_gem_object *obj, int ring)
>  	list_move_tail(&obj->global_list,
>  		       &to_i915(obj->base.dev)->mm.bound_list);
>  
> -	list_for_each_entry(vma, &obj->vma_list, vma_link) {
> +	i915_gem_obj_for_each_vma(vma, obj) {

This and the majority of the conversions simply should not exist and
only do so because of what I consider to be bad API. After they are
removed, there are only a few list iterators left. That said, there is
value in documenting the current locking.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre