[PATCH] drm/amdgpu: Flush tlb by VM_INVALIDATION packet in sdma_v5_2

[YuanShang]

To flush tlb by one single packet, Use VM_INVALIDATION packet
in function sdma_v5_2_ring_emit_vm_flush.

Signed-off-by: YuanShang <YuanShang.Mao at amd.com>
---
 .../gpu/drm/amd/amdgpu/navi10_sdma_pkt_open.h | 64 +++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c        | 24 ++++++-
 2 files changed, 87 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/navi10_sdma_pkt_open.h b/drivers/gpu/drm/amd/amdgpu/navi10_sdma_pkt_open.h
index a5b60c9a2418..c88284ff92d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/navi10_sdma_pkt_open.h
+++ b/drivers/gpu/drm/amd/amdgpu/navi10_sdma_pkt_open.h
@@ -68,6 +68,7 @@
 #define SDMA_SUBOP_POLL_REG_WRITE_MEM  1
 #define SDMA_SUBOP_POLL_DBIT_WRITE_MEM  2
 #define SDMA_SUBOP_POLL_MEM_VERIFY  3
+#define SDMA_SUBOP_VM_INVALIDATION  4
 #define HEADER_AGENT_DISPATCH  4
 #define HEADER_BARRIER  5
 #define SDMA_OP_AQL_COPY  0
@@ -4040,6 +4041,69 @@
 #define SDMA_PKT_POLL_MEM_VERIFY_RESERVED_RESERVED(x) (((x) & SDMA_PKT_POLL_MEM_VERIFY_RESERVED_reserved_mask) << SDMA_PKT_POLL_MEM_VERIFY_RESERVED_reserved_shift)
 
 
+/*
+** Definitions for SDMA_PKT_VM_INVALIDATION packet
+*/
+
+/*define for HEADER word*/
+/*define for op field*/
+#define SDMA_PKT_VM_INVALIDATION_HEADER_op_offset 0
+#define SDMA_PKT_VM_INVALIDATION_HEADER_op_mask   0x000000FF
+#define SDMA_PKT_VM_INVALIDATION_HEADER_op_shift  0
+#define SDMA_PKT_VM_INVALIDATION_HEADER_OP(x) (((x) & SDMA_PKT_VM_INVALIDATION_HEADER_op_mask) << SDMA_PKT_VM_INVALIDATION_HEADER_op_shift)
+
+/*define for sub_op field*/
+#define SDMA_PKT_VM_INVALIDATION_HEADER_sub_op_offset 0
+#define SDMA_PKT_VM_INVALIDATION_HEADER_sub_op_mask   0x000000FF
+#define SDMA_PKT_VM_INVALIDATION_HEADER_sub_op_shift  8
+#define SDMA_PKT_VM_INVALIDATION_HEADER_SUB_OP(x) (((x) & SDMA_PKT_VM_INVALIDATION_HEADER_sub_op_mask) << SDMA_PKT_VM_INVALIDATION_HEADER_sub_op_shift)
+
+/*define for gfx_eng_id field*/
+#define SDMA_PKT_VM_INVALIDATION_HEADER_gfx_eng_id_offset 0
+#define SDMA_PKT_VM_INVALIDATION_HEADER_gfx_eng_id_mask   0x0000001F
+#define SDMA_PKT_VM_INVALIDATION_HEADER_gfx_eng_id_shift  16
+#define SDMA_PKT_VM_INVALIDATION_HEADER_GFX_ENG_ID(x) (((x) & SDMA_PKT_VM_INVALIDATION_HEADER_gfx_eng_id_mask) << SDMA_PKT_VM_INVALIDATION_HEADER_gfx_eng_id_shift)
+
+/*define for mm_eng_id field*/
+#define SDMA_PKT_VM_INVALIDATION_HEADER_mm_eng_id_offset 0
+#define SDMA_PKT_VM_INVALIDATION_HEADER_mm_eng_id_mask   0x0000001F
+#define SDMA_PKT_VM_INVALIDATION_HEADER_mm_eng_id_shift  24
+#define SDMA_PKT_VM_INVALIDATION_HEADER_MM_ENG_ID(x) (((x) & SDMA_PKT_VM_INVALIDATION_HEADER_mm_eng_id_mask) << SDMA_PKT_VM_INVALIDATION_HEADER_mm_eng_id_shift)
+
+/*define for INVALIDATEREQ word*/
+/*define for invalidatereq field*/
+#define SDMA_PKT_VM_INVALIDATION_INVALIDATEREQ_invalidatereq_offset 1
+#define SDMA_PKT_VM_INVALIDATION_INVALIDATEREQ_invalidatereq_mask   0xFFFFFFFF
+#define SDMA_PKT_VM_INVALIDATION_INVALIDATEREQ_invalidatereq_shift  0
+#define SDMA_PKT_VM_INVALIDATION_INVALIDATEREQ_INVALIDATEREQ(x) (((x) & SDMA_PKT_VM_INVALIDATION_INVALIDATEREQ_invalidatereq_mask) << SDMA_PKT_VM_INVALIDATION_INVALIDATEREQ_invalidatereq_shift)
+
+/*define for ADDRESSRANGELO word*/
+/*define for addressrangelo field*/
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGELO_addressrangelo_offset 2
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGELO_addressrangelo_mask   0xFFFFFFFF
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGELO_addressrangelo_shift  0
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGELO_ADDRESSRANGELO(x) (((x) & SDMA_PKT_VM_INVALIDATION_ADDRESSRANGELO_addressrangelo_mask) << SDMA_PKT_VM_INVALIDATION_ADDRESSRANGELO_addressrangelo_shift)
+
+/*define for ADDRESSRANGEHI word*/
+/*define for invalidateack field*/
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_invalidateack_offset 3
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_invalidateack_mask   0x0000FFFF
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_invalidateack_shift  0
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_INVALIDATEACK(x) (((x) & SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_invalidateack_mask) << SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_invalidateack_shift)
+
+/*define for addressrangehi field*/
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_addressrangehi_offset 3
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_addressrangehi_mask   0x0000001F
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_addressrangehi_shift  16
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_ADDRESSRANGEHI(x) (((x) & SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_addressrangehi_mask) << SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_addressrangehi_shift)
+
+/*define for reserved field*/
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_reserved_offset 3
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_reserved_mask   0x000001FF
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_reserved_shift  23
+#define SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_RESERVED(x) (((x) & SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_reserved_mask) << SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_reserved_shift)
+
+
 /*
 ** Definitions for SDMA_PKT_ATOMIC packet
 */
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c b/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
index cc9e961f0078..a112ea3749c1 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
@@ -1121,7 +1121,29 @@ static void sdma_v5_2_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
 static void sdma_v5_2_ring_emit_vm_flush(struct amdgpu_ring *ring,
 					 unsigned vmid, uint64_t pd_addr)
 {
-	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
+	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub];
+	uint32_t req = hub->vmhub_funcs->get_invalidate_req(vmid, 0);
+
+	/* Update the PD address for this VMID. */
+	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_lo32 +
+			      (hub->ctx_addr_distance * vmid),
+			      lower_32_bits(pd_addr));
+	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_hi32 +
+			      (hub->ctx_addr_distance * vmid),
+			      upper_32_bits(pd_addr));
+
+	/* Trigger invalidation. */
+	amdgpu_ring_write(ring,
+			  SDMA_PKT_VM_INVALIDATION_HEADER_OP(SDMA_OP_POLL_REGMEM) |
+			  SDMA_PKT_VM_INVALIDATION_HEADER_SUB_OP(SDMA_SUBOP_VM_INVALIDATION) |
+			  SDMA_PKT_VM_INVALIDATION_HEADER_GFX_ENG_ID(ring->vm_inv_eng) |
+			  SDMA_PKT_VM_INVALIDATION_HEADER_MM_ENG_ID(0x1f));
+	amdgpu_ring_write(ring, req);
+	amdgpu_ring_write(ring, 0xFFFFFFFF);
+	amdgpu_ring_write(ring,
+			  SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_INVALIDATEACK(1 << vmid) |
+			  SDMA_PKT_VM_INVALIDATION_ADDRESSRANGEHI_ADDRESSRANGEHI(0x1F));
+	//amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
 }
 
 static void sdma_v5_2_ring_emit_wreg(struct amdgpu_ring *ring,
-- 
2.25.1