[Intel-gfx] [PATCH] drm/i915: restore ggtt double-bind avoidance

[Chris Wilson]

On Thu, Oct 15, 2015 at 02:23:01PM +0200, Daniel Vetter wrote:
> This was accidentally lost in
> 
> commit 75d04a3773ecee617847de963ae4195d6aa74c28
> Author: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> Date:   Tue Apr 28 17:56:17 2015 +0300
> 
>     drm/i915/gtt: Allocate va range only if vma is not bound
> 
> While at it implement an improved version suggested by Chris which
> avoids the double-bind irrespective of what type of bind is done
> first.
> 
> Note that this exact bug was already addressed in
> 
> commit d0e30adc42d979e4adc36b6c112b57337423b70c
> Author: Chris Wilson <chris at chris-wilson.co.uk>
> Date:   Wed Jul 29 20:02:48 2015 +0100
> 
>     drm/i915: Mark PIN_USER binding as GLOBAL_BIND without the aliasing ppgtt
> 
> but the problem is still that originally in
> 
> commit 0875546c5318c85c13d07014af5350e9000bc9e9
> Author: Daniel Vetter <daniel.vetter at ffwll.ch>
> Date:   Mon Apr 20 09:04:05 2015 -0700
> 
>     drm/i915: Fix up the vma aliasing ppgtt binding
> 
> if forgotten to take into account there case where we have a
> GLOBAL_BIND before a LOCAL_BIND. This patch here fixes that.
> 
> v2: Pimp commit message and revert the partial fix.
> 
> v3: Split into two functions to specialize on aliasing_ppgtt y/n.
> 
> v4: WARN_ON for paranoia in the init sequence, since the ggtt probe
> and aliasing ppgtt setup are far apart.
> 
> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Michel Thierry <michel.thierry at intel.com>
> Cc: Mika Kuoppala <mika.kuoppala at intel.com>
> Link: http://mid.gmane.org/1444894651-5332-1-git-send-email-daniel.vetter@ffwll.ch
> Signed-off-by: Daniel Vetter <daniel.vetter at intel.com>
> ---
>  drivers/gpu/drm/i915/i915_gem_gtt.c | 49 ++++++++++++++++++++++++++++---------
>  1 file changed, 37 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
> index ff1004876f56..419687135f41 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> @@ -2502,6 +2502,39 @@ static int ggtt_bind_vma(struct i915_vma *vma,
>  			 enum i915_cache_level cache_level,
>  			 u32 flags)
>  {
> +	struct drm_i915_gem_object *obj = vma->obj;
> +	struct sg_table *pages = obj->pages;

Time to drop this as it is now correctly set later from the ggtt view.
As we then only use obj in one place, lets use it directly. (Having
read-only and read-write vma is a posibility to consider, too bad the
read-only flag didn't come into greater use by hw.)

> +	u32 pte_flags = 0;
> +	int ret;
> +
> +	ret = i915_get_ggtt_vma_pages(vma);
> +	if (ret)
> +		return ret;
> +	pages = vma->ggtt_view.pages;
> +
> +	/* Currently applicable only to VLV */
> +	if (obj->gt_ro)
> +		pte_flags |= PTE_READ_ONLY;
> +
> +

Spare line we can trade in.

> +	vma->vm->insert_entries(vma->vm, pages,

And since we only use pages here we can just use vma->gtt_view.pages
instead.

> +				vma->node.start,
> +				cache_level, pte_flags);
> +
> +	/*
> +	 * Without aliasing PPGTT there's no difference between
> +	 * GLOBAL/LOCAL_BIND, it's all the same ptes. Hence unconditionally
> +	 * upgrade to both bound if we bind either to avoid double-binding.
> +	 */
> +	vma->bound |= GLOBAL_BIND | LOCAL_BIND;
> +
> +	return 0;
> +}
> +
> +static int aliasing_gtt_bind_vma(struct i915_vma *vma,
> +				 enum i915_cache_level cache_level,
> +				 u32 flags)
> +{
>  	struct drm_device *dev = vma->vm->dev;
>  	struct drm_i915_private *dev_priv = dev->dev_private;
>  	struct drm_i915_gem_object *obj = vma->obj;

Could make similar contractions here but meh.

Stylistic nitpicks aside,
Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre