[Intel-gfx] [PATCH 11/15] drm/i915: Introduce execlist context status change notification

[Wang, Zhi A]

-----Original Message-----
From: Tvrtko Ursulin [mailto:tvrtko.ursulin at linux.intel.com] 
Sent: Monday, May 16, 2016 7:00 AM
To: Wang, Zhi A <zhi.a.wang at intel.com>; intel-gfx at lists.freedesktop.org; Gordon, David S <david.s.gordon at intel.com>; joonas.lahtinen at linux.intel.com; Tian, Kevin <kevin.tian at intel.com>; Lv, Zhiyuan <zhiyuan.lv at intel.com>
Subject: Re: [Intel-gfx] [PATCH 11/15] drm/i915: Introduce execlist context status change notification


On 15/05/16 18:32, 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.
>
> Signed-off-by: Zhi Wang <zhi.a.wang at intel.com>
> ---
>   drivers/gpu/drm/i915/i915_drv.h  |  5 +++++
>   drivers/gpu/drm/i915/intel_lrc.c | 23 +++++++++++++++++++++++
>   drivers/gpu/drm/i915/intel_lrc.h |  5 +++++
>   3 files changed, 33 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h 
> b/drivers/gpu/drm/i915/i915_drv.h index 7f050a3..1c9e865 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -884,6 +884,8 @@ struct intel_context {
>   		bool initialised;
>   		bool skip_init_context;
>   		u32 ring_buffer_size;
> +		bool enable_status_change_notification;

This flag will be per context and not per context-engine, correct? In which case it would be worth moving it to the context. Would make execlists_context_status_change cheaper as well.

OK. Will do.
> +		struct atomic_notifier_head status_notifier_head;
>   	} engine[I915_NUM_ENGINES];
>   	bool use_48bit_addressing_mode;
>
> @@ -2403,6 +2405,9 @@ struct drm_i915_gem_request {
>
>   	/** Execlists context hardware id. */
>   	unsigned ctx_hw_id;
> +
> +	/** GVT workload **/
> +	void *gvt_workload;

Why is this a void *, for what it will be used ? I can't see in this patch series what will be stored here so asking. Is it only temporary until further patches?

Also, is it really per-request? I am just wondering since other patches are mostly concerned about contexts.

Will remove that according to today discussion.
>   };
>
>   struct drm_i915_gem_request * __must_check diff --git 
> a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index 0a96d4a..aeaea2e 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -415,6 +415,19 @@ 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 *req,
> +		unsigned long status)
> +{
> +	if (!req->ctx->engine[req->engine->id].
> +			enable_status_change_notification)
> +		return;
> +
> +	atomic_notifier_call_chain(
> +			&req->ctx->engine[req->engine->id].status_notifier_head,
> +			status, req->gvt_workload);
> +}

I would really like to see this compiled out when !CONFIG_DRM_I915_GVT. 
Otherwise is just adding useless conditionals on submission fast paths.

Let's do that.
> +
>   static void execlists_context_unqueue(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_gem_request *req0 = NULL, *req1 = NULL; @@ -450,6 
> +463,11 @@ 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 @@ 
> -488,6 +506,8 @@ 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);
>
> @@ -2541,6 +2561,9 @@ static int execlists_context_deferred_alloc(struct intel_context *ctx,
>   	ctx->engine[engine->id].state = ctx_obj;
>   	ctx->engine[engine->id].initialised = engine_initialised(ctx, 
> engine);
>
> +	if (ctx->engine[engine->id].enable_status_change_notification)
> +		ATOMIC_INIT_NOTIFIER_HEAD(
> +				&ctx->engine[engine->id].status_notifier_head);
>   	return 0;
>
>   error_ringbuf:
> diff --git a/drivers/gpu/drm/i915/intel_lrc.h 
> b/drivers/gpu/drm/i915/intel_lrc.h
> index 1afba03..98c94ee 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.h
> +++ b/drivers/gpu/drm/i915/intel_lrc.h
> @@ -57,6 +57,11 @@
>   #define GEN8_CSB_READ_PTR(csb_status) \
>   	(((csb_status) & GEN8_CSB_READ_PTR_MASK) >> 8)
>
> +enum {
> +	CONTEXT_SCHEDULE_IN = 0,
> +	CONTEXT_SCHEDULE_OUT,
> +};
> +
>   /* 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);
>

Regards,

Tvrtko