drm/i915/gvt: Audit and shadow workload during ELSP writing

Let the workload audit and shadow ahead of vGPU scheduling, that
will eliminate GPU idle time and improve performance for multi-VM.

The performance of Heaven running simultaneously in 3VMs has
improved 20% after this patch.

v2:Remove condition current->vgpu==vgpu when shadow during ELSP
writing.
Signed-off-by: Ping Gao <ping.a.gao@intel.com>
Reviewed-by: Zhi Wang <zhi.a.wang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>

drivers/gpu/drm/i915/gvt/execlist.c

@@ -605,6 +605,7 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 	struct list_head *q = workload_q_head(vgpu, ring_id);
 	struct intel_vgpu_workload *last_workload = get_last_workload(q);
 	struct intel_vgpu_workload *workload = NULL;
+	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
 	u64 ring_context_gpa;
 	u32 head, tail, start, ctl, ctx_ctl, per_ctx, indirect_ctx;
 	int ret;
@@ -668,6 +669,7 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 	workload->complete = complete_execlist_workload;
 	workload->status = -EINPROGRESS;
 	workload->emulate_schedule_in = emulate_schedule_in;
+	workload->shadowed = false;
 
 	if (ring_id == RCS) {
 		intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
@@ -701,6 +703,15 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 		return ret;
 	}
 
+	/* Only scan and shadow the first workload in the queue
+	 * as there is only one pre-allocated buf-obj for shadow.
+	 */
+	if (list_empty(workload_q_head(vgpu, ring_id))) {
+		mutex_lock(&dev_priv->drm.struct_mutex);
+		intel_gvt_scan_and_shadow_workload(workload);
+		mutex_unlock(&dev_priv->drm.struct_mutex);
+	}
+
 	queue_workload(workload);
 	return 0;
 }

drivers/gpu/drm/i915/gvt/scheduler.c

@@ -201,6 +201,9 @@ int intel_gvt_scan_and_shadow_workload(struct intel_vgpu_workload *workload)
 	struct intel_vgpu *vgpu = workload->vgpu;
 	int ret;
 
+	if (workload->shadowed)
+		return 0;
+
 	shadow_ctx->desc_template &= ~(0x3 << GEN8_CTX_ADDRESSING_MODE_SHIFT);
 	shadow_ctx->desc_template |= workload->ctx_desc.addressing_mode <<
 				    GEN8_CTX_ADDRESSING_MODE_SHIFT;
@@ -228,6 +231,10 @@ int intel_gvt_scan_and_shadow_workload(struct intel_vgpu_workload *workload)
 	}
 
 	ret = populate_shadow_context(workload);
+	if (ret)
+		goto out;
+
+	workload->shadowed = true;
 
 out:
 	return ret;

drivers/gpu/drm/i915/gvt/scheduler.h

@@ -82,6 +82,7 @@ struct intel_vgpu_workload {
 	struct drm_i915_gem_request *req;
 	/* if this workload has been dispatched to i915? */
 	bool dispatched;
+	bool shadowed;
 	int status;
 
 	struct intel_vgpu_mm *shadow_mm;