[Intel-gfx] [PATCH v6 8/9] drm/i915: Support LRC context single submission

[Wang, Zhi A]

For now If we want to achieve that, we have to add a member in struct i915_gem_request like your code. :(

/* Assume in all host GEM request, req->vgpu == NULL*/
If (req0->vgpu = req1->vgpu)
	combine!

> -----Original Message-----
> From: Joonas Lahtinen [mailto:joonas.lahtinen at linux.intel.com]
> Sent: Friday, June 03, 2016 12:47 PM
> To: Wang, Zhi A <zhi.a.wang at intel.com>; intel-gfx at lists.freedesktop.org;
> tvrtko.ursulin at linux.intel.com; Tian, Kevin <kevin.tian at intel.com>; Lv, Zhiyuan
> <zhiyuan.lv at intel.com>; chris at chris-wilson.co.uk
> Subject: Re: [PATCH v6 8/9] drm/i915: Support LRC context single submission
> 
> On to, 2016-06-02 at 12:36 -0400, Zhi Wang wrote:
> > This patch introduces the support of LRC context signle submission.
> 
> "single"
> 
> > As GVT context may come from different guests, which requires
> > different
> 
> "require"
> 
> > configuration of render registers. It can't be combined into a dual
> > ELSP submission combo.
> >
> > Only GVT-g will create this kinds of GEM context currently.
> >
> > v6:
> > - Make GVT code as dead code when !CONFIG_DRM_I915_GVT. (Chris)
> >
> > v5:
> >
> > - Only compile this feature when CONFIG_DRM_I915_GVT=y. (Tvrtko)
> >
> > Signed-off-by: Zhi Wang <zhi.a.wang at intel.com>
> > ---
> >  drivers/gpu/drm/i915/i915_drv.h  |  1 +
> >  drivers/gpu/drm/i915/intel_lrc.c | 15 +++++++++++++++
> >  2 files changed, 16 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_drv.h
> > b/drivers/gpu/drm/i915/i915_drv.h index dee72d3..92d01e3 100644
> > --- a/drivers/gpu/drm/i915/i915_drv.h
> > +++ b/drivers/gpu/drm/i915/i915_drv.h
> > @@ -882,6 +882,7 @@ struct i915_gem_context {
> >  	u32 lrc_ring_buffer_size;
> >  	u32 lrc_addressing_mode_bits;
> >  	bool enable_lrc_status_change_notification;
> > +	bool enable_lrc_single_submission;
> >
> >  	struct list_head link;
> >
> > diff --git a/drivers/gpu/drm/i915/intel_lrc.c
> > b/drivers/gpu/drm/i915/intel_lrc.c
> > index 96d20c8..ce707ea 100644
> > --- a/drivers/gpu/drm/i915/intel_lrc.c
> > +++ b/drivers/gpu/drm/i915/intel_lrc.c
> > @@ -446,6 +446,21 @@ static void execlists_context_unqueue(struct
> intel_engine_cs *engine)
> >  			i915_gem_request_unreference(req0);
> >  			req0 = cursor;
> >  		} else {
> > +			/* Compiler will do the dead-code elimination */
> > +			if (IS_ENABLED(CONFIG_DRM_I915_GVT)) {
> > +				/*
> > +				 * req0 (after merged) ctx requires single
> > +				 * submission, stop picking
> > +				 */
> > +				if (req0->ctx->enable_lrc_single_submission)
> > +					break;
> > +				/*
> > +				 * req0 ctx doesn't require single submission,
> > +				 * but next req ctx requires, stop picking
> > +				 */
> > +				if (cursor->ctx->enable_lrc_single_submission)
> > +					break;
> > +			}
> 
> I remember discussing this on the F2F, we will be aware of the VM IDs at this
> point, so we could use that criterion, instead of just disabling it pessimistically.
> Is there some reason we could not do it?
> 
> Check would be rather simple;
> 
> #if IS_ENABLED(CONFIG_DRM_I915_GVT)
> if (req0->gvt.vmid != cursor->gvt.vmid)
> 	break;
> #endif
> 
> I'm not sure if it will be worth the #if when vmid would be zero in
> DOM0 always.
> 
> Regards, Joonas
> 
> >  			req1 = cursor;
> >  			WARN_ON(req1->elsp_submitted);
> >  			break;
> --
> Joonas Lahtinen
> Open Source Technology Center
> Intel Corporation