path: root/common/recipes-kernel/linux/linux-yocto-4.14.71/3316-Revert-drm-amdgpu-remove-now-superflous-_hdp-operati.patch

From d26a7965dbf9d601958f9e9cc2b98c3256b98b3b Mon Sep 17 00:00:00 2001
From: Jim Qu <Jim.Qu@amd.com>
Date: Thu, 1 Feb 2018 15:08:01 +0800
Subject: [PATCH 3316/4131] Revert "drm/amdgpu: remove now superflous *_hdp
 operation"

This reverts commit cccee96b9e3ab89fec51fcc1bce769989dde1f74.
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h      |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c   |  7 +++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h |  1 +
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c    | 10 +++++++-
 drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c    | 39 ++++++++++++++++++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c    | 26 +++++++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c    | 19 ++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c    | 14 ++++++++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c   | 10 +++++++-
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c   | 11 ++++++++-
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c   | 13 ++++++++++-
 drivers/gpu/drm/amd/amdgpu/si_dma.c      | 19 +++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c    | 30 ++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c    | 30 ++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c    | 35 +++++++++++++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c    | 37 ++++++++++++++++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c    | 18 ++++++++++++++-
 17 files changed, 302 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 81e6aa2..4a00efb 100755
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1851,6 +1851,7 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
 #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags))
 #define amdgpu_ring_emit_gds_switch(r, v, db, ds, wb, ws, ab, as) (r)->funcs->emit_gds_switch((r), (v), (db), (ds), (wb), (ws), (ab), (as))
 #define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r))
+#define amdgpu_ring_emit_hdp_invalidate(r) (r)->funcs->emit_hdp_invalidate((r))
 #define amdgpu_ring_emit_switch_buffer(r) (r)->funcs->emit_switch_buffer((r))
 #define amdgpu_ring_emit_cntxcntl(r, d) (r)->funcs->emit_cntxcntl((r), (d))
 #define amdgpu_ring_emit_rreg(r, d) (r)->funcs->emit_rreg((r), (d))
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index 5d79b43..1b2b7c1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -233,7 +233,12 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 #ifdef CONFIG_X86_64
 	if (!(adev->flags & AMD_IS_APU))
 #endif
-		amdgpu_asic_invalidate_hdp(adev, ring);
+	{
+		if (ring->funcs->emit_hdp_invalidate)
+			amdgpu_ring_emit_hdp_invalidate(ring);
+		else
+			amdgpu_asic_invalidate_hdp(adev, ring);
+	}
 
 	r = amdgpu_fence_emit(ring, f);
 	if (r) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index 115d523..668a1ed 100755
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -129,6 +129,7 @@ struct amdgpu_ring_funcs {
 	void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vmid,
 			      unsigned pasid, uint64_t pd_addr);
 	void (*emit_hdp_flush)(struct amdgpu_ring *ring);
+	void (*emit_hdp_invalidate)(struct amdgpu_ring *ring);
 	void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid,
 				uint32_t gds_base, uint32_t gds_size,
 				uint32_t gws_base, uint32_t gws_size,
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index d6a6871..6766f8c 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
@@ -261,6 +261,13 @@ static void cik_sdma_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, (0xfff << 16) | 10); /* retry count, poll interval */
 }
 
+static void cik_sdma_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
+	amdgpu_ring_write(ring, mmHDP_DEBUG0);
+	amdgpu_ring_write(ring, 1);
+}
+
 /**
  * cik_sdma_ring_emit_fence - emit a fence on the DMA ring
  *
@@ -1270,7 +1277,7 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
 	.set_wptr = cik_sdma_ring_set_wptr,
 	.emit_frame_size =
 		6 + /* cik_sdma_ring_emit_hdp_flush */
-		3 + /* hdp invalidate */
+		3 + /* cik_sdma_ring_emit_hdp_invalidate */
 		6 + /* cik_sdma_ring_emit_pipeline_sync */
 		CIK_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* cik_sdma_ring_emit_vm_flush */
 		9 + 9 + 9, /* cik_sdma_ring_emit_fence x3 for user fence, vm fence */
@@ -1280,6 +1287,7 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
 	.emit_pipeline_sync = cik_sdma_ring_emit_pipeline_sync,
 	.emit_vm_flush = cik_sdma_ring_emit_vm_flush,
 	.emit_hdp_flush = cik_sdma_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = cik_sdma_ring_emit_hdp_invalidate,
 	.test_ring = cik_sdma_ring_test_ring,
 	.test_ib = cik_sdma_ring_test_ib,
 	.insert_nop = cik_sdma_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c
index 3517fd9..e6c3a24 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c
@@ -1809,6 +1809,17 @@ static int gfx_v6_0_ring_test_ring(struct amdgpu_ring *ring)
 	return r;
 }
 
+static void gfx_v6_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
+{
+	/* flush hdp cache */
+	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
+	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
+				 WRITE_DATA_DST_SEL(0)));
+	amdgpu_ring_write(ring, mmHDP_MEM_COHERENCY_FLUSH_CNTL);
+	amdgpu_ring_write(ring, 0);
+	amdgpu_ring_write(ring, 0x1);
+}
+
 static void gfx_v6_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
 {
 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
@@ -1816,6 +1827,24 @@ static void gfx_v6_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
 		EVENT_INDEX(0));
 }
 
+/**
+ * gfx_v6_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp
+ *
+ * @adev: amdgpu_device pointer
+ * @ridx: amdgpu ring index
+ *
+ * Emits an hdp invalidate on the cp.
+ */
+static void gfx_v6_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
+	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
+				 WRITE_DATA_DST_SEL(0)));
+	amdgpu_ring_write(ring, mmHDP_DEBUG0);
+	amdgpu_ring_write(ring, 0);
+	amdgpu_ring_write(ring, 0x1);
+}
+
 static void gfx_v6_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
 				     u64 seq, unsigned flags)
 {
@@ -3478,7 +3507,8 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_gfx = {
 	.get_wptr = gfx_v6_0_ring_get_wptr,
 	.set_wptr = gfx_v6_0_ring_set_wptr_gfx,
 	.emit_frame_size =
-		5 + 5 + /* hdp flush / invalidate */
+		5 + /* gfx_v6_0_ring_emit_hdp_flush */
+		5 + /* gfx_v6_0_ring_emit_hdp_invalidate */
 		14 + 14 + 14 + /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */
 		7 + 4 + /* gfx_v6_0_ring_emit_pipeline_sync */
 		SI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v6_0_ring_emit_vm_flush */
@@ -3488,6 +3518,8 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_gfx = {
 	.emit_fence = gfx_v6_0_ring_emit_fence,
 	.emit_pipeline_sync = gfx_v6_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = gfx_v6_0_ring_emit_vm_flush,
+	.emit_hdp_flush = gfx_v6_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v6_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v6_0_ring_test_ring,
 	.test_ib = gfx_v6_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
@@ -3503,7 +3535,8 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_compute = {
 	.get_wptr = gfx_v6_0_ring_get_wptr,
 	.set_wptr = gfx_v6_0_ring_set_wptr_compute,
 	.emit_frame_size =
-		5 + 5 + /* hdp flush / invalidate */
+		5 + /* gfx_v6_0_ring_emit_hdp_flush */
+		5 + /* gfx_v6_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v6_0_ring_emit_pipeline_sync */
 		SI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v6_0_ring_emit_vm_flush */
 		14 + 14 + 14, /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */
@@ -3512,6 +3545,8 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_compute = {
 	.emit_fence = gfx_v6_0_ring_emit_fence,
 	.emit_pipeline_sync = gfx_v6_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = gfx_v6_0_ring_emit_vm_flush,
+	.emit_hdp_flush = gfx_v6_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v6_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v6_0_ring_test_ring,
 	.test_ib = gfx_v6_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index ff71f7c..aeeced9 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -2147,6 +2147,26 @@ static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
 		EVENT_INDEX(0));
 }
 
+
+/**
+ * gfx_v7_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp
+ *
+ * @adev: amdgpu_device pointer
+ * @ridx: amdgpu ring index
+ *
+ * Emits an hdp invalidate on the cp.
+ */
+static void gfx_v7_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
+	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
+				 WRITE_DATA_DST_SEL(0) |
+				 WR_CONFIRM));
+	amdgpu_ring_write(ring, mmHDP_DEBUG0);
+	amdgpu_ring_write(ring, 0);
+	amdgpu_ring_write(ring, 1);
+}
+
 /**
  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
  *
@@ -5064,7 +5084,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
 	.emit_frame_size =
 		20 + /* gfx_v7_0_ring_emit_gds_switch */
 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
-		5 + /* hdp invalidate */
+		5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
@@ -5076,6 +5096,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v7_0_ring_test_ring,
 	.test_ib = gfx_v7_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
@@ -5095,7 +5116,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 	.emit_frame_size =
 		20 + /* gfx_v7_0_ring_emit_gds_switch */
 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
-		5 + /* hdp invalidate */
+		5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
 		7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
@@ -5106,6 +5127,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v7_0_ring_test_ring,
 	.test_ib = gfx_v7_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 8a334aa..918cefc 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -6233,6 +6233,19 @@ static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
 		EVENT_INDEX(0));
 }
 
+
+static void gfx_v8_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
+	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
+				 WRITE_DATA_DST_SEL(0) |
+				 WR_CONFIRM));
+	amdgpu_ring_write(ring, mmHDP_DEBUG0);
+	amdgpu_ring_write(ring, 0);
+	amdgpu_ring_write(ring, 1);
+
+}
+
 static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
 				      struct amdgpu_ib *ib,
 				      unsigned vmid, bool ctx_switch)
@@ -6877,6 +6890,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v8_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v8_0_ring_test_ring,
 	.test_ib = gfx_v8_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
@@ -6899,7 +6913,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.emit_frame_size =
 		20 + /* gfx_v8_0_ring_emit_gds_switch */
 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
-		5 + /* hdp_invalidate */
+		5 + /* gfx_v8_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
@@ -6910,6 +6924,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v8_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v8_0_ring_test_ring,
 	.test_ib = gfx_v8_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
@@ -6929,7 +6944,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
 	.emit_frame_size =
 		20 + /* gfx_v8_0_ring_emit_gds_switch */
 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
-		5 + /* hdp_invalidate */
+		5 + /* gfx_v8_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
 		17 + /* gfx_v8_0_ring_emit_vm_flush */
 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 61728ab..7fd44fcc 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -3591,6 +3591,14 @@ static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 			      ref_and_mask, ref_and_mask, 0x20);
 }
 
+static void gfx_v9_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+
+	gfx_v9_0_write_data_to_reg(ring, 0, true,
+				   SOC15_REG_OFFSET(HDP, 0, mmHDP_READ_CACHE_INVALIDATE), 1);
+}
+
 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
                                       struct amdgpu_ib *ib,
                                       unsigned vmid, bool ctx_switch)
@@ -4317,6 +4325,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
 	.emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v9_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v9_0_ring_test_ring,
 	.test_ib = gfx_v9_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
@@ -4341,7 +4350,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
 	.emit_frame_size =
 		20 + /* gfx_v9_0_ring_emit_gds_switch */
 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
-		5 + /* hdp invalidate */
+		5 + /* gfx_v9_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* gfx_v9_0_ring_emit_vm_flush */
 		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
@@ -4352,6 +4361,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
 	.emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
 	.emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
 	.emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = gfx_v9_0_ring_emit_hdp_invalidate,
 	.test_ring = gfx_v9_0_ring_test_ring,
 	.test_ib = gfx_v9_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
@@ -4372,7 +4382,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
 	.emit_frame_size =
 		20 + /* gfx_v9_0_ring_emit_gds_switch */
 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
-		5 + /* hdp invalidate */
+		5 + /* gfx_v9_0_ring_emit_hdp_invalidate */
 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* gfx_v9_0_ring_emit_vm_flush */
 		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 4a05c85..5e9bea0 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
@@ -289,6 +289,13 @@ static void sdma_v2_4_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */
 }
 
+static void sdma_v2_4_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
+			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
+	amdgpu_ring_write(ring, mmHDP_DEBUG0);
+	amdgpu_ring_write(ring, 1);
+}
 /**
  * sdma_v2_4_ring_emit_fence - emit a fence on the DMA ring
  *
@@ -1193,7 +1200,7 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = {
 	.set_wptr = sdma_v2_4_ring_set_wptr,
 	.emit_frame_size =
 		6 + /* sdma_v2_4_ring_emit_hdp_flush */
-		3 + /* hdp invalidate */
+		3 + /* sdma_v2_4_ring_emit_hdp_invalidate */
 		6 + /* sdma_v2_4_ring_emit_pipeline_sync */
 		VI_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* sdma_v2_4_ring_emit_vm_flush */
 		10 + 10 + 10, /* sdma_v2_4_ring_emit_fence x3 for user fence, vm fence */
@@ -1203,6 +1210,7 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = {
 	.emit_pipeline_sync = sdma_v2_4_ring_emit_pipeline_sync,
 	.emit_vm_flush = sdma_v2_4_ring_emit_vm_flush,
 	.emit_hdp_flush = sdma_v2_4_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = sdma_v2_4_ring_emit_hdp_invalidate,
 	.test_ring = sdma_v2_4_ring_test_ring,
 	.test_ib = sdma_v2_4_ring_test_ib,
 	.insert_nop = sdma_v2_4_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index 586c55b..3c2195c 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
@@ -460,6 +460,14 @@ static void sdma_v3_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */
 }
 
+static void sdma_v3_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
+			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
+	amdgpu_ring_write(ring, mmHDP_DEBUG0);
+	amdgpu_ring_write(ring, 1);
+}
+
 /**
  * sdma_v3_0_ring_emit_fence - emit a fence on the DMA ring
  *
@@ -1626,7 +1634,7 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = {
 	.set_wptr = sdma_v3_0_ring_set_wptr,
 	.emit_frame_size =
 		6 + /* sdma_v3_0_ring_emit_hdp_flush */
-		3 + /* hdp invalidate */
+		3 + /* sdma_v3_0_ring_emit_hdp_invalidate */
 		6 + /* sdma_v3_0_ring_emit_pipeline_sync */
 		12 + /* sdma_v3_0_ring_emit_vm_flush */
 		10 + 10 + 10, /* sdma_v3_0_ring_emit_fence x3 for user fence, vm fence */
@@ -1636,6 +1644,7 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = {
 	.emit_pipeline_sync = sdma_v3_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = sdma_v3_0_ring_emit_vm_flush,
 	.emit_hdp_flush = sdma_v3_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = sdma_v3_0_ring_emit_hdp_invalidate,
 	.test_ring = sdma_v3_0_ring_test_ring,
 	.test_ib = sdma_v3_0_ring_test_ib,
 	.insert_nop = sdma_v3_0_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
index 4344490..54a58b2 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
@@ -375,6 +375,16 @@ static void sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */
 }
 
+static void sdma_v4_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+
+	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
+			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
+	amdgpu_ring_write(ring, SOC15_REG_OFFSET(HDP, 0, mmHDP_READ_CACHE_INVALIDATE));
+	amdgpu_ring_write(ring, 1);
+}
+
 /**
  * sdma_v4_0_ring_emit_fence - emit a fence on the DMA ring
  *
@@ -1573,7 +1583,7 @@ static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
 	.set_wptr = sdma_v4_0_ring_set_wptr,
 	.emit_frame_size =
 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
-		3 + /* hdp invalidate */
+		3 + /* sdma_v4_0_ring_emit_hdp_invalidate */
 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* sdma_v4_0_ring_emit_vm_flush */
 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
@@ -1583,6 +1593,7 @@ static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = sdma_v4_0_ring_emit_hdp_invalidate,
 	.test_ring = sdma_v4_0_ring_test_ring,
 	.test_ib = sdma_v4_0_ring_test_ib,
 	.insert_nop = sdma_v4_0_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/si_dma.c b/drivers/gpu/drm/amd/amdgpu/si_dma.c
index 2db5bfb..0275e6f 100644
--- a/drivers/gpu/drm/amd/amdgpu/si_dma.c
+++ b/drivers/gpu/drm/amd/amdgpu/si_dma.c
@@ -75,6 +75,20 @@ static void si_dma_ring_emit_ib(struct amdgpu_ring *ring,
 
 }
 
+static void si_dma_ring_emit_hdp_flush(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
+	amdgpu_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL));
+	amdgpu_ring_write(ring, 1);
+}
+
+static void si_dma_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
+	amdgpu_ring_write(ring, (0xf << 16) | (HDP_DEBUG0));
+	amdgpu_ring_write(ring, 1);
+}
+
 /**
  * si_dma_ring_emit_fence - emit a fence on the DMA ring
  *
@@ -758,7 +772,8 @@ static const struct amdgpu_ring_funcs si_dma_ring_funcs = {
 	.get_wptr = si_dma_ring_get_wptr,
 	.set_wptr = si_dma_ring_set_wptr,
 	.emit_frame_size =
-		3 + 3 + /* hdp flush / invalidate */
+		3 + /* si_dma_ring_emit_hdp_flush */
+		3 + /* si_dma_ring_emit_hdp_invalidate */
 		6 + /* si_dma_ring_emit_pipeline_sync */
 		SI_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* si_dma_ring_emit_vm_flush */
 		9 + 9 + 9, /* si_dma_ring_emit_fence x3 for user fence, vm fence */
@@ -767,6 +782,8 @@ static const struct amdgpu_ring_funcs si_dma_ring_funcs = {
 	.emit_fence = si_dma_ring_emit_fence,
 	.emit_pipeline_sync = si_dma_ring_emit_pipeline_sync,
 	.emit_vm_flush = si_dma_ring_emit_vm_flush,
+	.emit_hdp_flush = si_dma_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = si_dma_ring_emit_hdp_invalidate,
 	.test_ring = si_dma_ring_test_ring,
 	.test_ib = si_dma_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c
index 948bb943..8ab10c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c
+++ b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c
@@ -464,6 +464,32 @@ static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq
 }
 
 /**
+ * uvd_v4_2_ring_emit_hdp_flush - emit an hdp flush
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp flush.
+ */
+static void uvd_v4_2_ring_emit_hdp_flush(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0));
+	amdgpu_ring_write(ring, 0);
+}
+
+/**
+ * uvd_v4_2_ring_hdp_invalidate - emit an hdp invalidate
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp invalidate.
+ */
+static void uvd_v4_2_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0));
+	amdgpu_ring_write(ring, 1);
+}
+
+/**
  * uvd_v4_2_ring_test_ring - register write test
  *
  * @ring: amdgpu_ring pointer
@@ -739,10 +765,14 @@ static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
 	.set_wptr = uvd_v4_2_ring_set_wptr,
 	.parse_cs = amdgpu_uvd_ring_parse_cs,
 	.emit_frame_size =
+		2 + /* uvd_v4_2_ring_emit_hdp_flush */
+		2 + /* uvd_v4_2_ring_emit_hdp_invalidate */
 		14, /* uvd_v4_2_ring_emit_fence  x1 no user fence */
 	.emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
 	.emit_ib = uvd_v4_2_ring_emit_ib,
 	.emit_fence = uvd_v4_2_ring_emit_fence,
+	.emit_hdp_flush = uvd_v4_2_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = uvd_v4_2_ring_emit_hdp_invalidate,
 	.test_ring = uvd_v4_2_ring_test_ring,
 	.test_ib = amdgpu_uvd_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c
index 6445d55..c1fe30c 100644
--- a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c
@@ -479,6 +479,32 @@ static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq
 }
 
 /**
+ * uvd_v5_0_ring_emit_hdp_flush - emit an hdp flush
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp flush.
+ */
+static void uvd_v5_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0));
+	amdgpu_ring_write(ring, 0);
+}
+
+/**
+ * uvd_v5_0_ring_hdp_invalidate - emit an hdp invalidate
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp invalidate.
+ */
+static void uvd_v5_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0));
+	amdgpu_ring_write(ring, 1);
+}
+
+/**
  * uvd_v5_0_ring_test_ring - register write test
  *
  * @ring: amdgpu_ring pointer
@@ -847,10 +873,14 @@ static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
 	.set_wptr = uvd_v5_0_ring_set_wptr,
 	.parse_cs = amdgpu_uvd_ring_parse_cs,
 	.emit_frame_size =
+		2 + /* uvd_v5_0_ring_emit_hdp_flush */
+		2 + /* uvd_v5_0_ring_emit_hdp_invalidate */
 		14, /* uvd_v5_0_ring_emit_fence  x1 no user fence */
 	.emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */
 	.emit_ib = uvd_v5_0_ring_emit_ib,
 	.emit_fence = uvd_v5_0_ring_emit_fence,
+	.emit_hdp_flush = uvd_v5_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = uvd_v5_0_ring_emit_hdp_invalidate,
 	.test_ring = uvd_v5_0_ring_test_ring,
 	.test_ib = amdgpu_uvd_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
index ac76a76..84c558d 100644
--- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
@@ -964,6 +964,32 @@ static void uvd_v6_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
 }
 
 /**
+ * uvd_v6_0_ring_emit_hdp_flush - emit an hdp flush
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp flush.
+ */
+static void uvd_v6_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0));
+	amdgpu_ring_write(ring, 0);
+}
+
+/**
+ * uvd_v6_0_ring_hdp_invalidate - emit an hdp invalidate
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp invalidate.
+ */
+static void uvd_v6_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0));
+	amdgpu_ring_write(ring, 1);
+}
+
+/**
  * uvd_v6_0_ring_test_ring - register write test
  *
  * @ring: amdgpu_ring pointer
@@ -1530,11 +1556,15 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_phys_funcs = {
 	.set_wptr = uvd_v6_0_ring_set_wptr,
 	.parse_cs = amdgpu_uvd_ring_parse_cs,
 	.emit_frame_size =
+		2 + /* uvd_v6_0_ring_emit_hdp_flush */
+		2 + /* uvd_v6_0_ring_emit_hdp_invalidate */
 		10 + /* uvd_v6_0_ring_emit_pipeline_sync */
 		14, /* uvd_v6_0_ring_emit_fence x1 no user fence */
 	.emit_ib_size = 8, /* uvd_v6_0_ring_emit_ib */
 	.emit_ib = uvd_v6_0_ring_emit_ib,
 	.emit_fence = uvd_v6_0_ring_emit_fence,
+	.emit_hdp_flush = uvd_v6_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = uvd_v6_0_ring_emit_hdp_invalidate,
 	.test_ring = uvd_v6_0_ring_test_ring,
 	.test_ib = amdgpu_uvd_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
@@ -1552,7 +1582,8 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_vm_funcs = {
 	.get_wptr = uvd_v6_0_ring_get_wptr,
 	.set_wptr = uvd_v6_0_ring_set_wptr,
 	.emit_frame_size =
-		6 + 6 + /* hdp flush / invalidate */
+		2 + /* uvd_v6_0_ring_emit_hdp_flush */
+		2 + /* uvd_v6_0_ring_emit_hdp_invalidate */
 		10 + /* uvd_v6_0_ring_emit_pipeline_sync */
 		20 + /* uvd_v6_0_ring_emit_vm_flush */
 		14 + 14, /* uvd_v6_0_ring_emit_fence x2 vm fence */
@@ -1561,6 +1592,8 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_vm_funcs = {
 	.emit_fence = uvd_v6_0_ring_emit_fence,
 	.emit_vm_flush = uvd_v6_0_ring_emit_vm_flush,
 	.emit_pipeline_sync = uvd_v6_0_ring_emit_pipeline_sync,
+	.emit_hdp_flush = uvd_v6_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = uvd_v6_0_ring_emit_hdp_invalidate,
 	.test_ring = uvd_v6_0_ring_test_ring,
 	.test_ib = amdgpu_uvd_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c
index d317c76..241e730 100644
--- a/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c
@@ -1136,6 +1136,37 @@ static void uvd_v7_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
 }
 
 /**
+ * uvd_v7_0_ring_emit_hdp_flush - emit an hdp flush
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp flush.
+ */
+static void uvd_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+
+	amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(NBIF, 0,
+		mmHDP_MEM_COHERENCY_FLUSH_CNTL), 0));
+	amdgpu_ring_write(ring, 0);
+}
+
+/**
+ * uvd_v7_0_ring_hdp_invalidate - emit an hdp invalidate
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp invalidate.
+ */
+static void uvd_v7_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+
+	amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(HDP, 0, mmHDP_READ_CACHE_INVALIDATE), 0));
+	amdgpu_ring_write(ring, 1);
+}
+
+/**
  * uvd_v7_0_ring_test_ring - register write test
  *
  * @ring: amdgpu_ring pointer
@@ -1662,13 +1693,16 @@ static const struct amdgpu_ring_funcs uvd_v7_0_ring_vm_funcs = {
 	.get_wptr = uvd_v7_0_ring_get_wptr,
 	.set_wptr = uvd_v7_0_ring_set_wptr,
 	.emit_frame_size =
-		6 + 6 + /* hdp flush / invalidate */
+		2 + /* uvd_v7_0_ring_emit_hdp_flush */
+		2 + /* uvd_v7_0_ring_emit_hdp_invalidate */
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 16 + /* uvd_v7_0_ring_emit_vm_flush */
 		14 + 14, /* uvd_v7_0_ring_emit_fence x2 vm fence */
 	.emit_ib_size = 8, /* uvd_v7_0_ring_emit_ib */
 	.emit_ib = uvd_v7_0_ring_emit_ib,
 	.emit_fence = uvd_v7_0_ring_emit_fence,
 	.emit_vm_flush = uvd_v7_0_ring_emit_vm_flush,
+	.emit_hdp_flush = uvd_v7_0_ring_emit_hdp_flush,
+	.emit_hdp_invalidate = uvd_v7_0_ring_emit_hdp_invalidate,
 	.test_ring = uvd_v7_0_ring_test_ring,
 	.test_ib = amdgpu_uvd_ring_test_ib,
 	.insert_nop = uvd_v7_0_ring_insert_nop,
@@ -1688,7 +1722,6 @@ static const struct amdgpu_ring_funcs uvd_v7_0_enc_ring_vm_funcs = {
 	.get_wptr = uvd_v7_0_enc_ring_get_wptr,
 	.set_wptr = uvd_v7_0_enc_ring_set_wptr,
 	.emit_frame_size =
-		3 + 3 + /* hdp flush / invalidate */
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + 8 + /* uvd_v7_0_enc_ring_emit_vm_flush */
 		5 + 5 + /* uvd_v7_0_enc_ring_emit_fence x2 vm fence */
 		1, /* uvd_v7_0_enc_ring_insert_end */
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
index d8bfeb2..bdc0052 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
@@ -832,6 +832,21 @@ static void vcn_v1_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64
 }
 
 /**
+ * vcn_v1_0_dec_ring_hdp_invalidate - emit an hdp invalidate
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Emits an hdp invalidate.
+ */
+static void vcn_v1_0_dec_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+
+	amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(HDP, 0, mmHDP_READ_CACHE_INVALIDATE), 0));
+	amdgpu_ring_write(ring, 1);
+}
+
+/**
  * vcn_v1_0_dec_ring_emit_ib - execute indirect buffer
  *
  * @ring: amdgpu_ring pointer
@@ -1103,7 +1118,7 @@ static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = {
 	.get_wptr = vcn_v1_0_dec_ring_get_wptr,
 	.set_wptr = vcn_v1_0_dec_ring_set_wptr,
 	.emit_frame_size =
-		6 + 6 + /* hdp invalidate / flush */
+		2 + /* vcn_v1_0_dec_ring_emit_hdp_invalidate */
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + 16 + /* vcn_v1_0_dec_ring_emit_vm_flush */
 		14 + 14 + /* vcn_v1_0_dec_ring_emit_fence x2 vm fence */
 		6,
@@ -1111,6 +1126,7 @@ static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = {
 	.emit_ib = vcn_v1_0_dec_ring_emit_ib,
 	.emit_fence = vcn_v1_0_dec_ring_emit_fence,
 	.emit_vm_flush = vcn_v1_0_dec_ring_emit_vm_flush,
+	.emit_hdp_invalidate = vcn_v1_0_dec_ring_emit_hdp_invalidate,
 	.test_ring = amdgpu_vcn_dec_ring_test_ring,
 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
 	.insert_nop = vcn_v1_0_ring_insert_nop,
-- 
2.7.4