drm/amdgpu: vcn and jpeg ring synchronization

Synchronize the ring usage for vcn1 and jpeg1 to workaround a hardware bug.
Signed-off-by: Veerabadhran Gopalakrishnan <veerabadhran.gopalakrishnan@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

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

@@ -68,6 +68,7 @@ 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);
+	mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
 	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);
@@ -237,6 +238,7 @@ int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
 	}
 
 	release_firmware(adev->vcn.fw);
+	mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
 	mutex_destroy(&adev->vcn.vcn_pg_lock);
 
 	return 0;

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

@@ -220,6 +220,7 @@ struct amdgpu_vcn {
 	struct amdgpu_vcn_inst	 inst[AMDGPU_MAX_VCN_INSTANCES];
 	struct amdgpu_vcn_reg	 internal;
 	struct mutex		 vcn_pg_lock;
+	struct mutex		vcn1_jpeg1_workaround;
 	atomic_t		 total_submission_cnt;
 
 	unsigned	harvest_config;

drivers/gpu/drm/amd/amdgpu/jpeg_v1_0.c

@@ -33,6 +33,7 @@
 
 static void jpeg_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev);
 static void jpeg_v1_0_set_irq_funcs(struct amdgpu_device *adev);
+static void jpeg_v1_0_ring_begin_use(struct amdgpu_ring *ring);
 
 static void jpeg_v1_0_decode_ring_patch_wreg(struct amdgpu_ring *ring, uint32_t *ptr, uint32_t reg_offset, uint32_t val)
 {
@@ -564,8 +565,8 @@ static const struct amdgpu_ring_funcs jpeg_v1_0_decode_ring_vm_funcs = {
 	.insert_start = jpeg_v1_0_decode_ring_insert_start,
 	.insert_end = jpeg_v1_0_decode_ring_insert_end,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
-	.begin_use = vcn_v1_0_ring_begin_use,
-	.end_use = amdgpu_vcn_ring_end_use,
+	.begin_use = jpeg_v1_0_ring_begin_use,
+	.end_use = vcn_v1_0_ring_end_use,
 	.emit_wreg = jpeg_v1_0_decode_ring_emit_wreg,
 	.emit_reg_wait = jpeg_v1_0_decode_ring_emit_reg_wait,
 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
@@ -586,3 +587,22 @@ static void jpeg_v1_0_set_irq_funcs(struct amdgpu_device *adev)
 {
 	adev->jpeg.inst->irq.funcs = &jpeg_v1_0_irq_funcs;
 }
+
+static void jpeg_v1_0_ring_begin_use(struct amdgpu_ring *ring)
+{
+	struct	amdgpu_device *adev = ring->adev;
+	bool	set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work);
+	int		cnt = 0;
+
+	mutex_lock(&adev->vcn.vcn1_jpeg1_workaround);
+
+	if (amdgpu_fence_wait_empty(&adev->vcn.inst->ring_dec))
+		DRM_ERROR("JPEG dec: vcn dec ring may not be empty\n");
+
+	for (cnt = 0; cnt < adev->vcn.num_enc_rings; cnt++) {
+		if (amdgpu_fence_wait_empty(&adev->vcn.inst->ring_enc[cnt]))
+			DRM_ERROR("JPEG dec: vcn enc ring[%d] may not be empty\n", cnt);
+	}
+
+	vcn_v1_0_set_pg_for_begin_use(ring, set_clocks);
+}

drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c

@@ -54,6 +54,7 @@ static int vcn_v1_0_pause_dpg_mode(struct amdgpu_device *adev,
 				int inst_idx, struct dpg_pause_state *new_state);
 
 static void vcn_v1_0_idle_work_handler(struct work_struct *work);
+static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring);
 
 /**
  * vcn_v1_0_early_init - set function pointers
@@ -1804,11 +1805,24 @@ static void vcn_v1_0_idle_work_handler(struct work_struct *work)
 	}
 }
 
-void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring)
+static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring)
 {
 	struct	amdgpu_device *adev = ring->adev;
 	bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work);
 
+	mutex_lock(&adev->vcn.vcn1_jpeg1_workaround);
+
+	if (amdgpu_fence_wait_empty(&ring->adev->jpeg.inst->ring_dec))
+		DRM_ERROR("VCN dec: jpeg dec ring may not be empty\n");
+
+	vcn_v1_0_set_pg_for_begin_use(ring, set_clocks);
+
+}
+
+void vcn_v1_0_set_pg_for_begin_use(struct amdgpu_ring *ring, bool set_clocks)
+{
+	struct amdgpu_device *adev = ring->adev;
+
 	if (set_clocks) {
 		amdgpu_gfx_off_ctrl(adev, false);
 		if (adev->pm.dpm_enabled)
@@ -1844,6 +1858,12 @@ void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring)
 	}
 }
 
+void vcn_v1_0_ring_end_use(struct amdgpu_ring *ring)
+{
+	schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
+	mutex_unlock(&ring->adev->vcn.vcn1_jpeg1_workaround);
+}
+
 static const struct amd_ip_funcs vcn_v1_0_ip_funcs = {
 	.name = "vcn_v1_0",
 	.early_init = vcn_v1_0_early_init,
@@ -1891,7 +1911,7 @@ static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = {
 	.insert_end = vcn_v1_0_dec_ring_insert_end,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
 	.begin_use = vcn_v1_0_ring_begin_use,
-	.end_use = amdgpu_vcn_ring_end_use,
+	.end_use = vcn_v1_0_ring_end_use,
 	.emit_wreg = vcn_v1_0_dec_ring_emit_wreg,
 	.emit_reg_wait = vcn_v1_0_dec_ring_emit_reg_wait,
 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
@@ -1923,7 +1943,7 @@ static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = {
 	.insert_end = vcn_v1_0_enc_ring_insert_end,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
 	.begin_use = vcn_v1_0_ring_begin_use,
-	.end_use = amdgpu_vcn_ring_end_use,
+	.end_use = vcn_v1_0_ring_end_use,
 	.emit_wreg = vcn_v1_0_enc_ring_emit_wreg,
 	.emit_reg_wait = vcn_v1_0_enc_ring_emit_reg_wait,
 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,

drivers/gpu/drm/amd/amdgpu/vcn_v1_0.h

@@ -24,7 +24,8 @@
 #ifndef __VCN_V1_0_H__
 #define __VCN_V1_0_H__
 
-void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring);
+void vcn_v1_0_ring_end_use(struct amdgpu_ring *ring);
+void vcn_v1_0_set_pg_for_begin_use(struct amdgpu_ring *ring, bool set_clocks);
 
 extern const struct amdgpu_ip_block_version vcn_v1_0_ip_block;