drm/i915/gvt: Trigger scheduling after context complete
authorPing Gao <ping.a.gao@intel.com>
Wed, 24 May 2017 01:14:11 +0000 (09:14 +0800)
committerZhenyu Wang <zhenyuw@linux.intel.com>
Thu, 8 Jun 2017 05:59:17 +0000 (13:59 +0800)
The time based scheduler poll context busy status at every
micro-second during vGPU switch, it will make GPU idle for a while
when the context is very small and completed before the next
micro-second arrival. Trigger scheduling immediately after context
complete will eliminate GPU idle and improve performance.

Create two vGPU with same type, run Heaven simultaneously:
Before this patch:
 +---------+----------+----------+
 |         |  vGPU1   |   vGPU2  |
 +---------+----------+----------+
 |  Heaven |  357     |    354   |
 +-------------------------------+

After this patch:
 +---------+----------+----------+
 |         |  vGPU1   |   vGPU2  |
 +---------+----------+----------+
 |  Heaven |  397     |    398   |
 +-------------------------------+

v2: Let need_reschedule protect by gvt-lock.

Signed-off-by: Ping Gao <ping.a.gao@intel.com>
Signed-off-by: Weinan Li <weinan.z.li@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
drivers/gpu/drm/i915/gvt/scheduler.c

index aa7e06df88b699f8a0f3c23b8169ffee4796041e..488fdea348a979e411258d6317748fe2bdd5e242 100644 (file)
@@ -452,6 +452,10 @@ static void complete_current_workload(struct intel_gvt *gvt, int ring_id)
 
        atomic_dec(&vgpu->running_workload_num);
        wake_up(&scheduler->workload_complete_wq);
+
+       if (gvt->scheduler.need_reschedule)
+               intel_gvt_request_service(gvt, INTEL_GVT_REQUEST_EVENT_SCHED);
+
        mutex_unlock(&gvt->lock);
 }