[Intel-gfx] [PATCH 2/3] drm/i915: Infrastructure for supporting different GGTT views per object

[Daniel Vetter]

On Thu, Dec 4, 2014 at 1:17 PM, Tvrtko Ursulin
<tvrtko.ursulin at linux.intel.com> wrote:
> I looked at propagating errors from i915_vma_bind() out to callers and it is
> mostly all fine apart from the i915_gem_restore_gtt_mappings during
> i915_drm_resume.

Bummer.

> I don't see how this is fixable apart by going back and having sgls stay
> around for the lifetime of their VMAs. It shouldn't be such a big deal -
> they are not so big even with non-coalesced entries.
>
> Thoughts?

I guess we'll just have to eat the additional pointer in vma structs
and that's it. If you want to go fancy we could have a new subtype:

struct i915_vma_special_view {
    struct i915_vma base;
    struct sg_table *sgt;
}

And then typecast accordingly for non-normal views. Feels a bit like
overkill though, at least until we start using vmas for other crazy
things which need special handling.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch