[Intel-gfx] [PATCH 5/5] drm/i915: Use partial view in mmap fault handler

[Joonas Lahtinen]

On ma, 2015-04-27 at 13:25 +0100, Chris Wilson wrote:
> On Mon, Apr 27, 2015 at 03:12:01PM +0300, Joonas Lahtinen wrote:
> > On ma, 2015-04-27 at 12:21 +0100, Chris Wilson wrote:
> > > On Mon, Apr 27, 2015 at 02:01:59PM +0300, Joonas Lahtinen wrote:
> > > > On pe, 2015-04-24 at 13:33 +0100, Chris Wilson wrote:
> > > > > On Fri, Apr 24, 2015 at 03:10:20PM +0300, Joonas Lahtinen wrote:
> > > > > > Use partial view for huge BOs (bigger than half the mappable aperture)
> > > > > > in fault handler so that they can be accessed withough trying to make
> > > > > > room for them by evicting other objects.
> > > > > > 
> > > > > > Signed-off-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> > > > > > ---
> > > > > >  drivers/gpu/drm/i915/i915_gem.c |   67 ++++++++++++++++++++++++++-------------
> > > > > >  1 file changed, 45 insertions(+), 22 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> > > > > > index c2c1819..eb30cee 100644
> > > > > > --- a/drivers/gpu/drm/i915/i915_gem.c
> > > > > > +++ b/drivers/gpu/drm/i915/i915_gem.c
> > > > > > @@ -1635,6 +1635,7 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
> > > > > >  	struct drm_i915_gem_object *obj = to_intel_bo(vma->vm_private_data);
> > > > > >  	struct drm_device *dev = obj->base.dev;
> > > > > >  	struct drm_i915_private *dev_priv = dev->dev_private;
> > > > > > +	struct i915_ggtt_view view = i915_ggtt_view_normal;
> > > > > >  	pgoff_t page_offset;
> > > > > >  	unsigned long pfn;
> > > > > >  	int ret = 0;
> > > > > > @@ -1667,8 +1668,21 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
> > > > > >  		goto unlock;
> > > > > >  	}
> > > > > >  
> > > > > > -	/* Now bind it into the GTT if needed */
> > > > > > -	ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE);
> > > > > > +	/* Use a partial view if the object is bigger than half the aperture. */
> > > > > > +	if (obj->base.size > dev_priv->gtt.mappable_end/2) {
> > > > > > +		static const size_t chunk_size = 256; // 1 MiB
> > > > > > +		memset(&view, 0, sizeof(view));
> > > > > > +		view.type = I915_GGTT_VIEW_PARTIAL;
> > > > > > +		view.params.partial.offset = rounddown(page_offset, chunk_size);
> > > > > > +		view.params.partial.size =
> > > > > > +			min_t(size_t,
> > > > > > +			      chunk_size,
> > > > > > +			      (vma->vm_end - vma->vm_start)/PAGE_SIZE -
> > > > > > +			      view.params.partial.offset);
> > > > > 
> > > > > This isn't what I was imagining.
> > > > > 
> > > > > I was expecting to see error handling inside the fault path if we could
> > > > > not pin the object. This way we could handle fragmentation or display
> > > > > objects pinned outside the aperture.
> > > > 
> > > > After discussion with Daniel, the idea was dropped due to high amount of
> > > > trashing which would occur if each object would be attempted to fit to
> > > > the mappable aperture for each fault to that object.
> > > 
> > > The point is that we fail to install a partial view for pinned objects
> > > outside of the aperture. Or did I miss how you handle them?
> > > 
> > 
> > That is true.
> > 
> > By changing the comparison to be against full aperture size, then the
> > patch will only bring new functionality to handle the cases when the
> > object was actually impossible to map previously (and was early
> > rejected).
> > 
> > I'd prefer to have this version in first (with change of removing the /2
> > from aperture size comparison), to get some feedback from the XenGT team
> > about the usability of it (speed-wise).
> 
> No. Daniel rejected a change just because we didn't have this series,
> only to find this series doesn't even provide the support Daniel
> wanted...
> 

Which change is that? If we're talking about the back-up path for failed
normal mapping, I can of course add the code for it, it will not cause
any worse trashing than it has before for objects smaller than aperture
size. Fancier logic to avoid trashing can be added later.

> I don't see this as a useful stepping point. The current users of large
> objects do not map through the GTT.

In XenGT, when the mappable aperture size is decreased due to slicing of
the aperture for different guests, it's not about large objects but
small aperture. And that is the reason why the feature was initially
implemented, and they've already been waiting for the feature for quite
a while.

All the series was intended to was to provide this support, so they can
proceed with testing with small aperture sizes. Everything else is
considered extra, and it was in the initial description of the task (by
either Jesse or Daniel) that half aperture size is most likely when the
current code will start to fail in real use cases.

I do not see why all possible features need to be implemented first to
get a useful feature for the XenGT team in. After all, before adding the
partial views, anything bigger than the mappable aperture was simply
rejected in mmap. So I do not follow the logic why the code path for
this case can not be added.

If some other change was not merged due to waiting for the partial views
support with more features than the code path for objects bigger than
aperture size, not merging the current code only causes two features to
keep waiting instead of just one. And as the XenGT team has been waiting
for the feature for months before the feature was even started on, I do
think it is not very reasonable.

The next revision will be started on immediately after this first one
has been merged, and intention is to keep this first one simple. Let me
know whatever feature is awaiting for the next revision, and I can take
it into consideration to have the support for that use case ready as
first priority.

Regards, Joonas

> -Chris
>