[Intel-gfx] [PATCH 3/4] drm/i915: Cleanup handling of last_fenced_seqno

[Daniel Vetter]

A few nitpicks below.

> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 5201f82..2dbf8f7 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -2461,13 +2461,15 @@ i915_gem_object_flush_fence(struct drm_i915_gem_object *obj,
>  		obj->fenced_gpu_access = false;
>  	}
>  
> +	if (obj->last_fenced_seqno &&
> +	    ring_passed_seqno(obj->ring, obj->last_fenced_seqno))
> +		obj->last_fenced_seqno = 0;

This only avoids running the request retiring logic. Without this there are a
few more things to simplify, I think:

>  	if (obj->last_fenced_seqno && pipelined != obj->ring) {
> -		if (!ring_passed_seqno(obj->ring, obj->last_fenced_seqno)) {
> -			ret = i915_wait_request(obj->ring,
> -						obj->last_fenced_seqno);
> -			if (ret)
> -				return ret;
> -		}
> +		ret = i915_wait_request(obj->ring,
> +					obj->last_fenced_seqno);
> +		if (ret)
> +			return ret;
>  
>  		obj->last_fenced_seqno = 0;

Can't we move that (and the other copies) to move_off_active?

> @@ -2648,15 +2650,18 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj,
>  						       old->last_fenced_seqno);
>  		}

Above here is the imo unnecessary clause mentioned in a previous patch.

>  
> +		obj->last_fenced_seqno = old->last_fenced_seqno;
>  		drm_gem_object_unreference(&old->base);
> -	} else if (obj->last_fenced_seqno == 0)
> -		pipelined = NULL;
> +	}
>  
>  	reg->obj = obj;
>  	list_move_tail(&reg->lru_list, &dev_priv->mm.fence_list);
>  	obj->fence_reg = reg - dev_priv->fence_regs;
>  
>  update:
> +	if (obj->last_fenced_seqno == 0)
> +		pipelined = NULL;
> +
>  	reg->setup_seqno =
>  		pipelined ? i915_gem_next_request_seqno(pipelined) : 0;
>  	reg->setup_ring = pipelined;
-- 
Daniel Vetter
Mail: daniel at ffwll.ch
Mobile: +41 (0)79 365 57 48