[Intel-gfx] [PATCH v6 7/9] drm/i915: Introduce execlist context status change notification

[Wang, Zhi A]

> -----Original Message-----
> From: Joonas Lahtinen [mailto:joonas.lahtinen at linux.intel.com]
> Sent: Friday, June 03, 2016 12:40 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 7/9] drm/i915: Introduce execlist context status change
> notification
> 
> On to, 2016-06-02 at 12:36 -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.
> >
> > v6:
> >
> > - When !CONFIG_DRM_I915_GVT, make GVT code as dead code then
> compiler
> > could automatically eliminate them for us. (Chris)
> > - Always initialize the notifier header, so it could be switched
> > on/off at runtime. (Chris)
> >
> > 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  |  2 ++
> >  drivers/gpu/drm/i915/intel_lrc.c | 24 ++++++++++++++++++++++++
> >  drivers/gpu/drm/i915/intel_lrc.h |  5 +++++
> >  3 files changed, 31 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_drv.h
> > b/drivers/gpu/drm/i915/i915_drv.h index f6cb60a..dee72d3 100644
> > --- a/drivers/gpu/drm/i915/i915_drv.h
> > +++ b/drivers/gpu/drm/i915/i915_drv.h
> > @@ -877,9 +877,11 @@ struct i915_gem_context {
> >  		u64 lrc_desc;
> >  		int pin_count;
> >  		bool initialised;
> > +		struct atomic_notifier_head status_notifier;
> 
> I think this could be outside engine block, just one per context.
> 
> >  	} engine[I915_NUM_ENGINES];
> >  	u32 lrc_ring_buffer_size;
> >  	u32 lrc_addressing_mode_bits;
> > +	bool enable_lrc_status_change_notification;
> >
> >  	struct list_head link;
> >
> > diff --git a/drivers/gpu/drm/i915/intel_lrc.c
> > b/drivers/gpu/drm/i915/intel_lrc.c
> > index ffb436c..96d20c8 100644
> > --- a/drivers/gpu/drm/i915/intel_lrc.c
> > +++ b/drivers/gpu/drm/i915/intel_lrc.c
> > @@ -404,6 +404,22 @@ static void execlists_submit_requests(struct
> drm_i915_gem_request *rq0,
> >  	spin_unlock_irq(&dev_priv->uncore.lock);
> >  }
> >
> > +static inline void execlists_context_status_change(
> > +		struct drm_i915_gem_request *rq,
> > +		unsigned long status)
> > +{
> > +	/* The compiler should be fine with the dead-code elimination */
> 
> The comment could rather be "Currently, only GVT-g code uses status
> notifications" :)
> 
> > +	if (!IS_ENABLED(CONFIG_DRM_I915_GVT))
> > +		return;
> > +
> > +	if (!rq->ctx->enable_lrc_status_change_notification)
> > +		return;
> > +
> 
> I think above line should be dropped, just don't register notifier for context that
> does not want notifications, and if the chain is empty, no call is made.
> 
I keep this in v7 as I think a "if" here is much cheaper than the rcu stuff in atomic_notifier_call_chain() with a lot of "if" even when the chain is empty. :)
> > +	atomic_notifier_call_chain(
> > +			&rq->ctx->engine[rq->engine->id].status_notifier,
> > +			status, rq);
> > +}
> > +
> >  static void execlists_context_unqueue(struct intel_engine_cs *engine)
> >  {
> >  	struct drm_i915_gem_request *req0 = NULL, *req1 = NULL; @@ -439,6
> > +455,11 @@ static void execlists_context_unqueue(struct intel_engine_cs
> *engine)
> >  	if (unlikely(!req0))
> >  		return;
> >
> > +	execlists_context_status_change(req0, CONTEXT_SCHEDULE_IN);
> > +
> > +	if (req1)
> > +		execlists_context_status_change(req1, CONTEXT_SCHEDULE_IN);
> > +
> >  	if (req0->elsp_submitted & engine->idle_lite_restore_wa) {
> >  		/*
> >  		 * WaIdleLiteRestore: make sure we never cause a lite restore @@
> > -477,6 +498,8 @@ execlists_check_remove_request(struct intel_engine_cs
> *engine, u32 ctx_id)
> >  	if (--head_req->elsp_submitted > 0)
> >  		return 0;
> >
> > +	execlists_context_status_change(head_req, CONTEXT_SCHEDULE_OUT);
> > +
> >  	list_del(&head_req->execlist_link);
> >  	i915_gem_request_unreference(head_req);
> >
> > @@ -2489,6 +2512,7 @@ static int execlists_context_deferred_alloc(struct
> i915_gem_context *ctx,
> >  	ce->ringbuf = ringbuf;
> >  	ce->state = ctx_obj;
> >  	ce->initialised = engine->init_context == NULL;
> > +	ATOMIC_INIT_NOTIFIER_HEAD(&ce->status_notifier);
> >
> >  	return 0;
> >
> > diff --git a/drivers/gpu/drm/i915/intel_lrc.h
> > b/drivers/gpu/drm/i915/intel_lrc.h
> > index e133c33..7a7ae8d 100644
> > --- a/drivers/gpu/drm/i915/intel_lrc.h
> > +++ b/drivers/gpu/drm/i915/intel_lrc.h
> > @@ -68,6 +68,11 @@ enum {
> >  #define GEN8_CSB_READ_PTR(csb_status) \
> >  	(((csb_status) & GEN8_CSB_READ_PTR_MASK) >> 8)
> >
> > +enum {
> > +	CONTEXT_SCHEDULE_IN = 0,
> > +	CONTEXT_SCHEDULE_OUT,
> > +};
> > +
> 
> Again, prefixes? At least INTEL_
> 
> With above;
> 
> Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> 
> Regards, Joonas
> 
> >  /* Logical Rings */
> >  int intel_logical_ring_alloc_request_extras(struct
> > drm_i915_gem_request *request);
> >  int intel_logical_ring_reserve_space(struct drm_i915_gem_request
> > *request);
> --
> Joonas Lahtinen
> Open Source Technology Center
> Intel Corporation