[Intel-gfx] [PATCH 1/9] drm/i915/dsb: Replace HAS_DSB check with dsb->cmd_buf check

[Ville Syrjälä]

On Fri, Jan 31, 2020 at 05:36:15PM +0530, Manna, Animesh wrote:
> 
> On 31-01-2020 17:12, Ville Syrjälä wrote:
> > On Fri, Jan 31, 2020 at 03:04:17PM +0530, Manna, Animesh wrote:
> >> On 30-01-2020 23:43, Souza, Jose wrote:
> >>> On Wed, 2020-01-29 at 20:20 +0200, Ville Syrjala wrote:
> >>>> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> >>>>
> >>>> We may want to not use the DSB even if the platform has one.
> >>>> So replace the HAS_DSB check in the _put() with a cmd_buf check
> >>>> that will work in either case.
> >>> Reviewed-by: José Roberto de Souza <jose.souza at intel.com>
> >>>
> >>>> Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> >>>> ---
> >>>>    drivers/gpu/drm/i915/display/intel_dsb.c | 3 +--
> >>>>    1 file changed, 1 insertion(+), 2 deletions(-)
> >>>>
> >>>> diff --git a/drivers/gpu/drm/i915/display/intel_dsb.c
> >>>> b/drivers/gpu/drm/i915/display/intel_dsb.c
> >>>> index 9dd18144a664..12776f09f227 100644
> >>>> --- a/drivers/gpu/drm/i915/display/intel_dsb.c
> >>>> +++ b/drivers/gpu/drm/i915/display/intel_dsb.c
> >>>> @@ -160,9 +160,8 @@ intel_dsb_get(struct intel_crtc *crtc)
> >>>>    void intel_dsb_put(struct intel_dsb *dsb)
> >>>>    {
> >>>>    	struct intel_crtc *crtc = container_of(dsb, typeof(*crtc),
> >>>> dsb);
> >>>> -	struct drm_i915_private *i915 = to_i915(crtc->base.dev);
> >>>>    
> >>>> -	if (!HAS_DSB(i915))
> >>>> +	if (!dsb->cmd_buf)
> >> Ville and Jose,
> >>
> >> Have a concern here. In intel_dsb_get() if get failure during i915_gem_object_create_internal, i915_gem_object_ggtt_pin, i915_gem_object_pin_map then we may not have dsb->cmd_buf.
> >> Then ref-count mechanism will break.
> > Hmm. Yeah. The refcount WARN could easily be fixed by either
> > decrementung refcount on get() fail or doing the "let's never use
> > DSB" patch after the refcount inc.
> 
> Hmm, from design point get/put/ref-count mechanism introduced to check dsp-api are used properly or not.
> For erroneous case managing ref-count in get() itself void the purpose of put() call.
> For example,
> 
> intel_dsb_get()
> got error from i915_gem_object_create_internal, i915_gem_object_ggtt_pin, i915_gem_object_pin_map
> intel_dsb_put
> intel_dsb_put
> ...
> 
> Should throw warning but can not if we manage in get() itself.

None of this stuff should really exist in the guts of the dsb code
anyway. It's all just a hack to get the dsb code in without actually
taking advantage of the dsb. The real solution would involve doing the
dsb vs. mmio decision upfront at the start of the atomic commit, and
then using totally different codeepaths for those two cases. No real
need for refcounts in that case. But first we'd need to finish the
vblank workers so we'd have the mmio path sorted out.

-- 
Ville Syrjälä
Intel