[Intel-gfx] [RFC 6/8] drm/i915: Insert commands for capture of OA counters in the ring

[Chris Wilson]

On Wed, Jul 15, 2015 at 02:17:01PM +0530, sourab.gupta at intel.com wrote:
> +void i915_oa_insert_cmd(struct intel_ringbuffer *ringbuf, u32 ctx_id)
You need to pass in the request here instead.

A better name would be i915_oa_emit_perf_report(). insert_cmd() is a
little too generic (i.e. which cmd do you mean?).

> +{
> +	struct intel_engine_cs *ring = ringbuf->ring;
> +	struct drm_i915_private *dev_priv = ring->dev->dev_private;
> +	struct drm_i915_gem_object *obj = dev_priv->oa_pmu.oa_rcs_buffer.obj;
> +	struct i915_oa_rcs_node *entry;
> +	unsigned long lock_flags;
> +	u32 addr = 0;
> +	int ret;
> +
> +	/* OA counters are only supported on the render ring */
> +	if (ring->id != RCS)
> +		return;
> +
> +	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
> +	if (entry == NULL) {
> +		DRM_ERROR("alloc failed\n");
> +		return;
> +	}
> +	entry->ctx_id = ctx_id;
> +	i915_gem_request_assign(&entry->req, ring->outstanding_lazy_request);
> +
> +	spin_lock_irqsave(&dev_priv->oa_pmu.lock, lock_flags);
> +	if (list_empty(&dev_priv->oa_pmu.node_list))
> +		entry->offset = 0;
> +	else {
> +		struct i915_oa_rcs_node *last_entry;
> +		int max_offset = dev_priv->oa_pmu.oa_rcs_buffer.node_count *
> +				dev_priv->oa_pmu.oa_rcs_buffer.node_size;
> +
> +		last_entry = list_last_entry(&dev_priv->oa_pmu.node_list,
> +					struct i915_oa_rcs_node, head);
> +		entry->offset = last_entry->offset +
> +				dev_priv->oa_pmu.oa_rcs_buffer.node_size;
> +
> +		if (entry->offset > max_offset)
> +			entry->offset = 0;
> +	}
> +	list_add_tail(&entry->head, &dev_priv->oa_pmu.node_list);
> +	spin_unlock_irqrestore(&dev_priv->oa_pmu.lock, lock_flags);
> +
> +	addr = i915_gem_obj_ggtt_offset(obj) + entry->offset;

Don't do more than one i915_gem_obj_to_ggtt() please (preferably none
and just keep hold of your pinned vma from the start).

> +	/* addr should be 64 byte aligned */
> +	BUG_ON(addr & 0x3f);
> +
> +	ret = intel_ring_begin(ring, 4);
> +	if (ret)
> +		return;

You've commited the request to the sample list, but have just erred out.

> +
> +	intel_ring_emit(ring, MI_REPORT_PERF_COUNT | (1<<0));
> +	intel_ring_emit(ring, addr | MI_REPORT_PERF_COUNT_GGTT);
> +	intel_ring_emit(ring, ring->outstanding_lazy_request->seqno);
> +	intel_ring_emit(ring, MI_NOOP);
> +	intel_ring_advance(ring);
> +
> +	obj->base.write_domain = I915_GEM_DOMAIN_RENDER;
> +	i915_vma_move_to_active(i915_gem_obj_to_ggtt(obj), ring);

That's the magic I have been looking for.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre