Re: [PATCH] drm/i915: Don't list gens one by bone in the VMA creation function

[Daniel Vetter <daniel@ffwll.ch>]

On Fri, Jan 10, 2014 at 06:32:49PM +0100, Daniel Vetter wrote:
> On Fri, Jan 10, 2014 at 09:16:39AM -0800, Ben Widawsky wrote:
> > On Fri, Jan 10, 2014 at 06:09:15PM +0100, Daniel Vetter wrote:
> > > On Fri, Jan 10, 2014 at 03:47:52PM +0000, Damien Lespiau wrote:
> > > > There are only two cases here, pre and post SNB (PPGTT).
> > > > 
> > > > Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
> > > > ---
> > > >  drivers/gpu/drm/i915/i915_gem_gtt.c | 14 ++------------
> > > >  1 file changed, 2 insertions(+), 12 deletions(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
> > > > index 3192089..866ca90 100644
> > > > --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> > > > +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> > > > @@ -1795,10 +1795,7 @@ static struct i915_vma *__i915_gem_vma_create(struct drm_i915_gem_object *obj,
> > > >  	vma->vm = vm;
> > > >  	vma->obj = obj;
> > > >  
> > > > -	switch (INTEL_INFO(vm->dev)->gen) {
> > > > -	case 8:
> > > > -	case 7:
> > > > -	case 6:
> > > > +	if (INTEL_INFO(vm->dev)->gen >= 6) {
> > > >  		if (i915_is_ggtt(vm)) {
> > > >  			vma->unbind_vma = ggtt_unbind_vma;
> > > >  			vma->bind_vma = ggtt_bind_vma;
> > > 
> > > Imo we should move the ->bind/unbind_vma functions into the global
> > > dev_prive->gt vtable - The personality we set here neither changes with
> > > the vma, the address space but the hw generation. So I'll reject this
> > > cleanup as not going far enough ;-)
> > > -Daniel
> > 
> > Actually, it was wrong to begin with anyway. gen6 shouldn't have PPGTT
> > functions.
> 
> If you want to go this far you can coalesce much more. _Everyone_ whether
> ppgtt or ggtt can use the same functions, which just call the insert/clear
> entries of the right vm. The exception is gen6 which needs to bind
> according to flags.
> 
> All the few remaining checks for has_*_mapping we still have splattered
> over the code can be garbage-collected and converted to unconditional
> calls to ->vma_bind with the right flags set, since vma_bind will
> short-circuit correctly on gen6 (and not care anywhere else really).

s/gen6/aliasing_ppgtt/ in the above, i.e. it depends upon the runtime
option we're using in the driver ... Idea still holds.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch