[Intel-gfx] [PATCH 7/9] drm/i915: Introduce execlist context status change notification

[Wang, Zhi A]

Oh. Thanks Chris!

> -----Original Message-----
> From: Chris Wilson [mailto:chris at chris-wilson.co.uk]
> Sent: Friday, May 20, 2016 2:17 PM
> To: Wang, Zhi A <zhi.a.wang at intel.com>
> Cc: intel-gfx at lists.freedesktop.org; tvrtko.ursulin at linux.intel.com;
> joonas.lahtinen at linux.intel.com; Tian, Kevin <kevin.tian at intel.com>; Lv,
> Zhiyuan <zhiyuan.lv at intel.com>
> Subject: Re: [PATCH 7/9] drm/i915: Introduce execlist context status change
> notification
> 
> On Tue, May 17, 2016 at 04:19:07AM -0400, Zhi Wang wrote:
> > This patch introduces an approach to track the execlist context status
> > change.
> >
> > GVT-g uses GVT context as the "shadow context". The content inside GVT
> > context will be copied back to guest after the context is idle. So
> > GVT-g has to know the status of the execlist context.
> >
> > This function is configurable in the context creation service.
> > Currently, Only GVT-g will create the "status-change-notification"
> > enabled GEM context.
> >
> > v5:
> >
> > - Only compile this feature when CONFIG_DRM_I915_GVT is
> > enabled.(Tvrtko)
> >
> > Signed-off-by: Zhi Wang <zhi.a.wang at intel.com>
> > ---
> >  drivers/gpu/drm/i915/i915_drv.h  |  6 ++++++
> > drivers/gpu/drm/i915/intel_lrc.c | 30 ++++++++++++++++++++++++++++++
> > drivers/gpu/drm/i915/intel_lrc.h |  7 +++++++
> >  3 files changed, 43 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_drv.h
> > b/drivers/gpu/drm/i915/i915_drv.h index 91f69e5..9688006 100644
> > --- a/drivers/gpu/drm/i915/i915_drv.h
> > +++ b/drivers/gpu/drm/i915/i915_drv.h
> > @@ -882,9 +882,15 @@ struct intel_context {
> >  		u64 lrc_desc;
> >  		uint32_t *lrc_reg_state;
> >  		bool initialised;
> > +#if IS_ENABLED(CONFIG_DRM_I915_GVT)
> > +		struct atomic_notifier_head status_notifier_head;
> 
> struct atomic_notifier_head status_notifier;
> 
> > +#endif
> >  	} engine[I915_NUM_ENGINES];
> >  	u32 ring_buffer_size;
> >  	bool use_48bit_addressing_mode;
> > +#if IS_ENABLED(CONFIG_DRM_I915_GVT)
> > +	bool enable_status_change_notification;
> 
> unsigned status_notify;
> 
> > +#endif
> >
> >  	struct list_head link;
> >  };
> > diff --git a/drivers/gpu/drm/i915/intel_lrc.c
> > b/drivers/gpu/drm/i915/intel_lrc.c
> > index d97623f..9069836 100644
> > --- a/drivers/gpu/drm/i915/intel_lrc.c
> > +++ b/drivers/gpu/drm/i915/intel_lrc.c
> > @@ -415,6 +415,20 @@ static void execlists_submit_requests(struct
> drm_i915_gem_request *rq0,
> >  	spin_unlock_irq(&dev_priv->uncore.lock);
> >  }
> >
> > +#if IS_ENABLED(CONFIG_DRM_I915_GVT)
> 
> Actually, I think there will be other use cases for this notifier.
> So let's not clutter the code with #ifdefs. Remove the ones from the struct and
> do
> 


OK. I guess the reason you want to turn a "bool" to "unsigned" is:

if (status_notify & NOTFIY_CONTEXT_STATUS)
	notify context status change
if (status_notify & NOTIFY_OTHER_THINGS)
	notify other things
.... extend it in future.

Or is there any other reason? Glad to know. :)

> > +static inline void execlists_context_status_change(
> > +		struct drm_i915_gem_request *req,
> > +		unsigned long status)
> > +{
> 
> 	if (!IS_ENABLED(CONFIG_DRM_I915_GVT))
> 		return;
> 
> instead. The compiler should be fine with the dead-code elimination.
> 
OK. Will do.
> 
> > +	if (!req->ctx->enable_status_change_notification)
> 
> What locks are you holding here?
> 

I'm holding engine->execlist_lock at this path.

> if (!READ_ONCE(req->ctx->status_notify))
> 	return;
> 
> > +		return;
> > +
> > +	atomic_notifier_call_chain(
> > +			&req->ctx->engine[req->engine->id].status_notifier_head,
> 
> I think we have enough justification for req->ctx_engine;
>

Are you saying you're doing some refactors on req->ctx, or you want to get some local pointers in this function, so that the req->xxxxx could be much shorter? :)
 
> > +			status, req);
> > +}
> > +#endif
> > +
> >  static void execlists_context_unqueue(struct intel_engine_cs *engine)
> > {
> >  	struct drm_i915_gem_request *req0 = NULL, *req1 = NULL; @@ -450,6
> > +464,13 @@ static void execlists_context_unqueue(struct intel_engine_cs
> *engine)
> >  	if (unlikely(!req0))
> >  		return;
> >
> > +#if IS_ENABLED(CONFIG_DRM_I915_GVT)
> > +	execlists_context_status_change(req0, CONTEXT_SCHEDULE_IN);
> > +
> > +	if (req1)
> > +		execlists_context_status_change(req1, CONTEXT_SCHEDULE_IN);
> 
> > +#endif
> > +
> >  	if (req0->elsp_submitted & engine->idle_lite_restore_wa) {
> >  		/*
> >  		 * WaIdleLiteRestore: make sure we never cause a lite restore @@
> > -488,6 +509,10 @@ execlists_check_remove_request(struct intel_engine_cs
> *engine, u32 ctx_id)
> >  	if (--head_req->elsp_submitted > 0)
> >  		return 0;
> >
> > +#if IS_ENABLED(CONFIG_DRM_I915_GVT)
> > +	execlists_context_status_change(head_req, CONTEXT_SCHEDULE_OUT);
> > +#endif
> > +
> >  	list_del(&head_req->execlist_link);
> >  	i915_gem_request_unreference(head_req);
> >
> > @@ -2534,6 +2559,11 @@ static int execlists_context_deferred_alloc(struct
> intel_context *ctx,
> >  	ctx->engine[engine->id].state = ctx_obj;
> >  	ctx->engine[engine->id].initialised = engine->init_context == NULL;
> >
> > +#if IS_ENABLED(CONFIG_DRM_I915_GVT)
> > +	if (ctx->enable_status_change_notification)
> 
> Always do the init. That allows us to start switching the notify on/off at
> runtime.
Good idea. Will do :)
> -Chris
> 
> --
> Chris Wilson, Intel Open Source Technology Centre