drm/amdgpu: Add XCC inst to PASID TLB flushing

Add XCC instance to select the correct KIQ ring when
flushing TLBs on a multi-XCC setup.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Tested-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -743,7 +743,9 @@ int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct amdgpu_device *adev,
 }
 
 int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
-				      uint16_t pasid, enum TLB_FLUSH_TYPE flush_type)
+				      uint16_t pasid,
+				      enum TLB_FLUSH_TYPE flush_type,
+				      uint32_t inst)
 {
 	bool all_hub = false;
 
@@ -751,7 +753,7 @@ int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 	    adev->family == AMDGPU_FAMILY_RV)
 		all_hub = true;
 
-	return amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, flush_type, all_hub);
+	return amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, flush_type, all_hub, inst);
 }
 
 bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev)

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

@@ -160,7 +160,8 @@ bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev);
 int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct amdgpu_device *adev,
 				uint16_t vmid);
 int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
-				uint16_t pasid, enum TLB_FLUSH_TYPE flush_type);
+				uint16_t pasid, enum TLB_FLUSH_TYPE flush_type,
+				uint32_t inst);
 
 bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid);
 

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

@@ -119,7 +119,8 @@ struct amdgpu_gmc_funcs {
 				uint32_t vmhub, uint32_t flush_type);
 	/* flush the vm tlb via pasid */
 	int (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid,
-					uint32_t flush_type, bool all_hub);
+					uint32_t flush_type, bool all_hub,
+					uint32_t inst);
 	/* flush the vm tlb via ring */
 	uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
 				       uint64_t pd_addr);
@@ -296,9 +297,9 @@ struct amdgpu_gmc {
 };
 
 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, vmhub, type) ((adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (vmhub), (type)))
-#define amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, type, allhub) \
+#define amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, type, allhub, inst) \
 	((adev)->gmc.gmc_funcs->flush_gpu_tlb_pasid \
-	((adev), (pasid), (type), (allhub)))
+	((adev), (pasid), (type), (allhub), (inst)))
 #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
 #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
 #define amdgpu_gmc_map_mtype(adev, flags) (adev)->gmc.gmc_funcs->map_mtype((adev),(flags))

drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c

@@ -419,7 +419,7 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
  */
 static int gmc_v10_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 					uint16_t pasid, uint32_t flush_type,
-					bool all_hub)
+					bool all_hub, uint32_t inst)
 {
 	int vmid, i;
 	signed long r;

drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c

@@ -324,7 +324,7 @@ static void gmc_v11_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
  */
 static int gmc_v11_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 					uint16_t pasid, uint32_t flush_type,
-					bool all_hub)
+					bool all_hub, uint32_t inst)
 {
 	int vmid, i;
 	signed long r;

drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c

@@ -424,7 +424,7 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev)
  */
 static int gmc_v7_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 					uint16_t pasid, uint32_t flush_type,
-					bool all_hub)
+					bool all_hub, uint32_t inst)
 {
 	int vmid;
 	unsigned int tmp;

drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c

@@ -622,7 +622,7 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
  */
 static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 					uint16_t pasid, uint32_t flush_type,
-					bool all_hub)
+					bool all_hub, uint32_t inst)
 {
 	int vmid;
 	unsigned int tmp;

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c

@@ -924,7 +924,7 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
  */
 static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 					uint16_t pasid, uint32_t flush_type,
-					bool all_hub)
+					bool all_hub, uint32_t inst)
 {
 	int vmid, i;
 	signed long r;
@@ -932,8 +932,8 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 	uint16_t queried_pasid;
 	bool ret;
 	u32 usec_timeout = amdgpu_sriov_vf(adev) ? SRIOV_USEC_TIMEOUT : adev->usec_timeout;
-	struct amdgpu_ring *ring = &adev->gfx.kiq[0].ring;
-	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
+	struct amdgpu_ring *ring = &adev->gfx.kiq[inst].ring;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[inst];
 
 	if (amdgpu_in_reset(adev))
 		return -EIO;
@@ -953,7 +953,7 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 		if (vega20_xgmi_wa)
 			ndw += kiq->pmf->invalidate_tlbs_size;
 
-		spin_lock(&adev->gfx.kiq[0].ring_lock);
+		spin_lock(&adev->gfx.kiq[inst].ring_lock);
 		/* 2 dwords flush + 8 dwords fence */
 		amdgpu_ring_alloc(ring, ndw);
 		if (vega20_xgmi_wa)
@@ -964,13 +964,13 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 		r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
 		if (r) {
 			amdgpu_ring_undo(ring);
-			spin_unlock(&adev->gfx.kiq[0].ring_lock);
+			spin_unlock(&adev->gfx.kiq[inst].ring_lock);
 			up_read(&adev->reset_domain->sem);
 			return -ETIME;
 		}
 
 		amdgpu_ring_commit(ring);
-		spin_unlock(&adev->gfx.kiq[0].ring_lock);
+		spin_unlock(&adev->gfx.kiq[inst].ring_lock);
 		r = amdgpu_fence_wait_polling(ring, seq, usec_timeout);
 		if (r < 1) {
 			dev_err(adev->dev, "wait for kiq fence error: %ld.\n", r);

drivers/gpu/drm/amd/amdkfd/kfd_process.c

@@ -2052,6 +2052,7 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type)
 	struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
 	uint64_t tlb_seq = amdgpu_vm_tlb_seq(vm);
 	struct kfd_node *dev = pdd->dev;
+	int xcc = 0;
 
 	/*
 	 * It can be that we race and lose here, but that is extremely unlikely
@@ -2069,8 +2070,10 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type)
 			amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->adev,
 							pdd->qpd.vmid);
 	} else {
+		for (xcc = 0; xcc < dev->num_xcc_per_node; xcc++)
 			amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->adev,
-					pdd->process->pasid, type);
+					pdd->process->pasid, type,
+					dev->start_xcc_id + xcc);
 	}
 }