drm/amdgpu/vcn: fix race condition issue for dpg unpause mode switch

Couldn't only rely on enc fence to decide switching to dpg unpaude mode.
Since a enc thread may not schedule a fence in time during multiple
threads running situation.

v3: 1. Rename enc_submission_cnt to dpg_enc_submission_cnt
    2. Add dpg_enc_submission_cnt check in idle_work_handler

v4:  Remove extra counter check, and reduce counter before idle
    work schedule
Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c

@@ -65,6 +65,8 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
 	INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
 	mutex_init(&adev->vcn.vcn_pg_lock);
 	atomic_set(&adev->vcn.total_submission_cnt, 0);
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
+		atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
 
 	switch (adev->asic_type) {
 	case CHIP_RAVEN:
@@ -298,7 +300,8 @@ static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
 			struct dpg_pause_state new_state;
 
-			if (fence[j])
+			if (fence[j] ||
+				unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
 			else
 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
@@ -333,19 +336,22 @@ void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
 
 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
 		struct dpg_pause_state new_state;
-		unsigned int fences = 0;
-		unsigned int i;
 
-		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
-			fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
-		}
-		if (fences)
+		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
+			atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
-		else
-			new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
+		} else {
+			unsigned int fences = 0;
+			unsigned int i;
 
-		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
-			new_state.fw_based = VCN_DPG_STATE__PAUSE;
+			for (i = 0; i < adev->vcn.num_enc_rings; ++i)
+				fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
+
+			if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
+				new_state.fw_based = VCN_DPG_STATE__PAUSE;
+			else
+				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
+		}
 
 		adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
 	}
@@ -354,6 +360,10 @@ void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
 
 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
 {
+	if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
+		ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
+		atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
+
 	atomic_dec(&ring->adev->vcn.total_submission_cnt);
 
 	schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);

drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h

@@ -183,6 +183,7 @@ struct amdgpu_vcn_inst {
 	void			*dpg_sram_cpu_addr;
 	uint64_t		dpg_sram_gpu_addr;
 	uint32_t		*dpg_sram_curr_addr;
+	atomic_t		dpg_enc_submission_cnt;
 };
 
 struct amdgpu_vcn {