drm/amdkfd: Add XCC instance to kgd2kfd interface (v3)

Gfx 9 starts to have multiple XCC instances in one device. Add instance
parameter to kgd2kfd functions where XCC instance was hard coded as 0.
Also, update code to pass the correct instance number when running
on a multi-XCC setup.

v2: introduce the XCC instance to gfx v11 (Morris)
v3: rebase (Alex)

Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Tested-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Morris Zhang <Shiwu.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 82 insertions, 78 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
index d36219ecd3dd..7918a00cbb5a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -46,26 +46,26 @@ enum hqd_dequeue_request_type {
 	SAVE_WAVES
 };
 
-static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
-			uint32_t queue, uint32_t vmid)
+static void kgd_gfx_v9_lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid, uint32_t inst)
 {
 	mutex_lock(&adev->srbm_mutex);
-	soc15_grbm_select(adev, mec, pipe, queue, vmid, 0);
+	soc15_grbm_select(adev, mec, pipe, queue, vmid, inst);
 }
 
-static void unlock_srbm(struct amdgpu_device *adev)
+static void kgd_gfx_v9_unlock_srbm(struct amdgpu_device *adev, uint32_t inst)
 {
-	soc15_grbm_select(adev, 0, 0, 0, 0, 0);
+	soc15_grbm_select(adev, 0, 0, 0, 0, inst);
 	mutex_unlock(&adev->srbm_mutex);
 }
 
 void kgd_gfx_v9_acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
-				uint32_t queue_id)
+				uint32_t queue_id, uint32_t inst)
 {
 	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
 	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
 
-	lock_srbm(adev, mec, pipe, queue_id, 0);
+	kgd_gfx_v9_lock_srbm(adev, mec, pipe, queue_id, 0, inst);
 }
 
 uint64_t kgd_gfx_v9_get_queue_mask(struct amdgpu_device *adev,
@@ -77,28 +77,28 @@ uint64_t kgd_gfx_v9_get_queue_mask(struct amdgpu_device *adev,
 	return 1ull << bit;
 }
 
-void kgd_gfx_v9_release_queue(struct amdgpu_device *adev)
+void kgd_gfx_v9_release_queue(struct amdgpu_device *adev, uint32_t inst)
 {
-	unlock_srbm(adev);
+	kgd_gfx_v9_unlock_srbm(adev, inst);
 }
 
 void kgd_gfx_v9_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
 					uint32_t sh_mem_config,
 					uint32_t sh_mem_ape1_base,
 					uint32_t sh_mem_ape1_limit,
-					uint32_t sh_mem_bases)
+					uint32_t sh_mem_bases, uint32_t inst)
 {
-	lock_srbm(adev, 0, 0, 0, vmid);
+	kgd_gfx_v9_lock_srbm(adev, 0, 0, 0, vmid, inst);
 
-	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmSH_MEM_CONFIG), sh_mem_config);
-	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmSH_MEM_BASES), sh_mem_bases);
+	WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmSH_MEM_CONFIG), sh_mem_config);
+	WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmSH_MEM_BASES), sh_mem_bases);
 	/* APE1 no longer exists on GFX9 */
 
-	unlock_srbm(adev);
+	kgd_gfx_v9_unlock_srbm(adev, inst);
 }
 
 int kgd_gfx_v9_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
-					unsigned int vmid)
+					unsigned int vmid, uint32_t inst)
 {
 	/*
 	 * We have to assume that there is no outstanding mapping.
@@ -156,7 +156,8 @@ int kgd_gfx_v9_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
  * but still works
  */
 
-int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id)
+int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst)
 {
 	uint32_t mec;
 	uint32_t pipe;
@@ -164,13 +165,13 @@ int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id)
 	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
 	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
 
-	lock_srbm(adev, mec, pipe, 0, 0);
+	kgd_gfx_v9_lock_srbm(adev, mec, pipe, 0, 0, inst);
 
-	WREG32_SOC15(GC, 0, mmCPC_INT_CNTL,
+	WREG32_SOC15(GC, inst, mmCPC_INT_CNTL,
 		CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
 		CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
 
-	unlock_srbm(adev);
+	kgd_gfx_v9_unlock_srbm(adev, inst);
 
 	return 0;
 }
@@ -220,7 +221,8 @@ static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
 int kgd_gfx_v9_hqd_load(struct amdgpu_device *adev, void *mqd,
 			uint32_t pipe_id, uint32_t queue_id,
 			uint32_t __user *wptr, uint32_t wptr_shift,
-			uint32_t wptr_mask, struct mm_struct *mm)
+			uint32_t wptr_mask, struct mm_struct *mm,
+			uint32_t inst)
 {
 	struct v9_mqd *m;
 	uint32_t *mqd_hqd;
@@ -228,21 +230,22 @@ int kgd_gfx_v9_hqd_load(struct amdgpu_device *adev, void *mqd,
 
 	m = get_mqd(mqd);
 
-	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id);
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
 
 	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
 	mqd_hqd = &m->cp_mqd_base_addr_lo;
-	hqd_base = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
+	hqd_base = SOC15_REG_OFFSET(GC, inst, mmCP_MQD_BASE_ADDR);
 
 	for (reg = hqd_base;
-	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+	     reg <= SOC15_REG_OFFSET(GC, inst, mmCP_HQD_PQ_WPTR_HI); reg++)
 		WREG32_RLC(reg, mqd_hqd[reg - hqd_base]);
 
 
 	/* Activate doorbell logic before triggering WPTR poll. */
 	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
 			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
-	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL), data);
+	WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_PQ_DOORBELL_CONTROL),
+					data);
 
 	if (wptr) {
 		/* Don't read wptr with get_user because the user
@@ -271,43 +274,43 @@ int kgd_gfx_v9_hqd_load(struct amdgpu_device *adev, void *mqd,
 		guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
 		guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
 
-		WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_LO),
+		WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_PQ_WPTR_LO),
 		       lower_32_bits(guessed_wptr));
-		WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI),
+		WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_PQ_WPTR_HI),
 		       upper_32_bits(guessed_wptr));
-		WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR),
+		WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_PQ_WPTR_POLL_ADDR),
 		       lower_32_bits((uintptr_t)wptr));
-		WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI),
+		WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI),
 		       upper_32_bits((uintptr_t)wptr));
 		WREG32_SOC15(GC, 0, mmCP_PQ_WPTR_POLL_CNTL1,
 		       (uint32_t)kgd_gfx_v9_get_queue_mask(adev, pipe_id, queue_id));
 	}
 
 	/* Start the EOP fetcher */
-	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_EOP_RPTR),
+	WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_EOP_RPTR),
 	       REG_SET_FIELD(m->cp_hqd_eop_rptr,
 			     CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
 
 	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
-	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_ACTIVE), data);
+	WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_ACTIVE), data);
 
-	kgd_gfx_v9_release_queue(adev);
+	kgd_gfx_v9_release_queue(adev, inst);
 
 	return 0;
 }
 
 int kgd_gfx_v9_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
 			    uint32_t pipe_id, uint32_t queue_id,
-			    uint32_t doorbell_off)
+			    uint32_t doorbell_off, uint32_t inst)
 {
-	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
+	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[inst].ring;
 	struct v9_mqd *m;
 	uint32_t mec, pipe;
 	int r;
 
 	m = get_mqd(mqd);
 
-	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id);
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
 
 	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
 	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
@@ -315,7 +318,7 @@ int kgd_gfx_v9_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
 	pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
 		 mec, pipe, queue_id);
 
-	spin_lock(&adev->gfx.kiq[0].ring_lock);
+	spin_lock(&adev->gfx.kiq[inst].ring_lock);
 	r = amdgpu_ring_alloc(kiq_ring, 7);
 	if (r) {
 		pr_err("Failed to alloc KIQ (%d).\n", r);
@@ -342,15 +345,15 @@ int kgd_gfx_v9_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
 	amdgpu_ring_commit(kiq_ring);
 
 out_unlock:
-	spin_unlock(&adev->gfx.kiq[0].ring_lock);
-	kgd_gfx_v9_release_queue(adev);
+	spin_unlock(&adev->gfx.kiq[inst].ring_lock);
+	kgd_gfx_v9_release_queue(adev, inst);
 
 	return r;
 }
 
 int kgd_gfx_v9_hqd_dump(struct amdgpu_device *adev,
 			uint32_t pipe_id, uint32_t queue_id,
-			uint32_t (**dump)[2], uint32_t *n_regs)
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
 {
 	uint32_t i = 0, reg;
 #define HQD_N_REGS 56
@@ -365,13 +368,13 @@ int kgd_gfx_v9_hqd_dump(struct amdgpu_device *adev,
 	if (*dump == NULL)
 		return -ENOMEM;
 
-	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id);
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
 
-	for (reg = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
-	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+	for (reg = SOC15_REG_OFFSET(GC, inst, mmCP_MQD_BASE_ADDR);
+	     reg <= SOC15_REG_OFFSET(GC, inst, mmCP_HQD_PQ_WPTR_HI); reg++)
 		DUMP_REG(reg);
 
-	kgd_gfx_v9_release_queue(adev);
+	kgd_gfx_v9_release_queue(adev, inst);
 
 	WARN_ON_ONCE(i != HQD_N_REGS);
 	*n_regs = i;
@@ -481,23 +484,23 @@ static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
 
 bool kgd_gfx_v9_hqd_is_occupied(struct amdgpu_device *adev,
 				uint64_t queue_address, uint32_t pipe_id,
-				uint32_t queue_id)
+				uint32_t queue_id, uint32_t inst)
 {
 	uint32_t act;
 	bool retval = false;
 	uint32_t low, high;
 
-	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id);
-	act = RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE);
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+	act = RREG32_SOC15(GC, inst, mmCP_HQD_ACTIVE);
 	if (act) {
 		low = lower_32_bits(queue_address >> 8);
 		high = upper_32_bits(queue_address >> 8);
 
-		if (low == RREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE) &&
-		   high == RREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE_HI))
+		if (low == RREG32_SOC15(GC, inst, mmCP_HQD_PQ_BASE) &&
+		   high == RREG32_SOC15(GC, inst, mmCP_HQD_PQ_BASE_HI))
 			retval = true;
 	}
-	kgd_gfx_v9_release_queue(adev);
+	kgd_gfx_v9_release_queue(adev, inst);
 	return retval;
 }
 
@@ -522,7 +525,7 @@ static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
 int kgd_gfx_v9_hqd_destroy(struct amdgpu_device *adev, void *mqd,
 				enum kfd_preempt_type reset_type,
 				unsigned int utimeout, uint32_t pipe_id,
-				uint32_t queue_id)
+				uint32_t queue_id, uint32_t inst)
 {
 	enum hqd_dequeue_request_type type;
 	unsigned long end_jiffies;
@@ -532,10 +535,10 @@ int kgd_gfx_v9_hqd_destroy(struct amdgpu_device *adev, void *mqd,
 	if (amdgpu_in_reset(adev))
 		return -EIO;
 
-	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id);
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
 
 	if (m->cp_hqd_vmid == 0)
-		WREG32_FIELD15_RLC(GC, 0, RLC_CP_SCHEDULERS, scheduler1, 0);
+		WREG32_FIELD15_RLC(GC, inst, RLC_CP_SCHEDULERS, scheduler1, 0);
 
 	switch (reset_type) {
 	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
@@ -552,22 +555,22 @@ int kgd_gfx_v9_hqd_destroy(struct amdgpu_device *adev, void *mqd,
 		break;
 	}
 
-	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_DEQUEUE_REQUEST), type);
+	WREG32_RLC(SOC15_REG_OFFSET(GC, inst, mmCP_HQD_DEQUEUE_REQUEST), type);
 
 	end_jiffies = (utimeout * HZ / 1000) + jiffies;
 	while (true) {
-		temp = RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE);
+		temp = RREG32_SOC15(GC, inst, mmCP_HQD_ACTIVE);
 		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
 			break;
 		if (time_after(jiffies, end_jiffies)) {
 			pr_err("cp queue preemption time out.\n");
-			kgd_gfx_v9_release_queue(adev);
+			kgd_gfx_v9_release_queue(adev, inst);
 			return -ETIME;
 		}
 		usleep_range(500, 1000);
 	}
 
-	kgd_gfx_v9_release_queue(adev);
+	kgd_gfx_v9_release_queue(adev, inst);
 	return 0;
 }
 
@@ -624,14 +627,14 @@ bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
 
 int kgd_gfx_v9_wave_control_execute(struct amdgpu_device *adev,
 					uint32_t gfx_index_val,
-					uint32_t sq_cmd)
+					uint32_t sq_cmd, uint32_t inst)
 {
 	uint32_t data = 0;
 
 	mutex_lock(&adev->grbm_idx_mutex);
 
-	WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, gfx_index_val);
-	WREG32_SOC15(GC, 0, mmSQ_CMD, sq_cmd);
+	WREG32_SOC15_RLC_SHADOW(GC, inst, mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32_SOC15(GC, inst, mmSQ_CMD, sq_cmd);
 
 	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
 		INSTANCE_BROADCAST_WRITES, 1);
@@ -640,7 +643,7 @@ int kgd_gfx_v9_wave_control_execute(struct amdgpu_device *adev,
 	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
 		SE_BROADCAST_WRITES, 1);
 
-	WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data);
+	WREG32_SOC15_RLC_SHADOW(GC, inst, mmGRBM_GFX_INDEX, data);
 	mutex_unlock(&adev->grbm_idx_mutex);
 
 	return 0;
@@ -685,7 +688,7 @@ static void unlock_spi_csq_mutexes(struct amdgpu_device *adev)
  * is being collected
  */
 static void get_wave_count(struct amdgpu_device *adev, int queue_idx,
-		int *wave_cnt, int *vmid)
+		int *wave_cnt, int *vmid, uint32_t inst)
 {
 	int pipe_idx;
 	int queue_slot;
@@ -700,12 +703,12 @@ static void get_wave_count(struct amdgpu_device *adev, int queue_idx,
 	*wave_cnt = 0;
 	pipe_idx = queue_idx / adev->gfx.mec.num_queue_per_pipe;
 	queue_slot = queue_idx % adev->gfx.mec.num_queue_per_pipe;
-	soc15_grbm_select(adev, 1, pipe_idx, queue_slot, 0, 0);
-	reg_val = RREG32_SOC15_IP(GC, SOC15_REG_OFFSET(GC, 0, mmSPI_CSQ_WF_ACTIVE_COUNT_0) +
+	soc15_grbm_select(adev, 1, pipe_idx, queue_slot, 0, inst);
+	reg_val = RREG32_SOC15_IP(GC, SOC15_REG_OFFSET(GC, inst, mmSPI_CSQ_WF_ACTIVE_COUNT_0) +
 			 queue_slot);
 	*wave_cnt = reg_val & SPI_CSQ_WF_ACTIVE_COUNT_0__COUNT_MASK;
 	if (*wave_cnt != 0)
-		*vmid = (RREG32_SOC15(GC, 0, mmCP_HQD_VMID) &
+		*vmid = (RREG32_SOC15(GC, inst, mmCP_HQD_VMID) &
 			 CP_HQD_VMID__VMID_MASK) >> CP_HQD_VMID__VMID__SHIFT;
 }
 
@@ -756,7 +759,7 @@ static void get_wave_count(struct amdgpu_device *adev, int queue_idx,
  *  Reading registers referenced above involves programming GRBM appropriately
  */
 void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
-		int *pasid_wave_cnt, int *max_waves_per_cu)
+		int *pasid_wave_cnt, int *max_waves_per_cu, uint32_t inst)
 {
 	int qidx;
 	int vmid;
@@ -772,7 +775,7 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
 	DECLARE_BITMAP(cp_queue_bitmap, KGD_MAX_QUEUES);
 
 	lock_spi_csq_mutexes(adev);
-	soc15_grbm_select(adev, 1, 0, 0, 0, 0);
+	soc15_grbm_select(adev, 1, 0, 0, 0, inst);
 
 	/*
 	 * Iterate through the shader engines and arrays of the device
@@ -787,8 +790,8 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
 	for (se_idx = 0; se_idx < se_cnt; se_idx++) {
 		for (sh_idx = 0; sh_idx < sh_cnt; sh_idx++) {
 
-			amdgpu_gfx_select_se_sh(adev, se_idx, sh_idx, 0xffffffff, 0);
-			queue_map = RREG32_SOC15(GC, 0, mmSPI_CSQ_WF_ACTIVE_STATUS);
+			amdgpu_gfx_select_se_sh(adev, se_idx, sh_idx, 0xffffffff, inst);
+			queue_map = RREG32_SOC15(GC, inst, mmSPI_CSQ_WF_ACTIVE_STATUS);
 
 			/*
 			 * Assumption: queue map encodes following schema: four
@@ -808,10 +811,11 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
 					continue;
 
 				/* Get number of waves in flight and aggregate them */
-				get_wave_count(adev, qidx, &wave_cnt, &vmid);
+				get_wave_count(adev, qidx, &wave_cnt, &vmid,
+						inst);
 				if (wave_cnt != 0) {
 					pasid_tmp =
-					  RREG32(SOC15_REG_OFFSET(OSSSYS, 0,
+					  RREG32(SOC15_REG_OFFSET(OSSSYS, inst,
 						 mmIH_VMID_0_LUT) + vmid);
 					if (pasid_tmp == pasid)
 						vmid_wave_cnt += wave_cnt;
@@ -820,8 +824,8 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
 		}
 	}
 
-	amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
-	soc15_grbm_select(adev, 0, 0, 0, 0, 0);
+	amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, inst);
+	soc15_grbm_select(adev, 0, 0, 0, 0, inst);
 	unlock_spi_csq_mutexes(adev);
 
 	/* Update the output parameters and return */
@@ -831,27 +835,27 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
 }
 
 void kgd_gfx_v9_program_trap_handler_settings(struct amdgpu_device *adev,
-                        uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr)
+		uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr, uint32_t inst)
 {
-	lock_srbm(adev, 0, 0, 0, vmid);
+	kgd_gfx_v9_lock_srbm(adev, 0, 0, 0, vmid, inst);
 
 	/*
 	 * Program TBA registers
 	 */
-	WREG32_SOC15(GC, 0, mmSQ_SHADER_TBA_LO,
+	WREG32_SOC15(GC, inst, mmSQ_SHADER_TBA_LO,
                         lower_32_bits(tba_addr >> 8));
-	WREG32_SOC15(GC, 0, mmSQ_SHADER_TBA_HI,
+	WREG32_SOC15(GC, inst, mmSQ_SHADER_TBA_HI,
                         upper_32_bits(tba_addr >> 8));
 
 	/*
 	 * Program TMA registers
 	 */
-	WREG32_SOC15(GC, 0, mmSQ_SHADER_TMA_LO,
+	WREG32_SOC15(GC, inst, mmSQ_SHADER_TMA_LO,
 			lower_32_bits(tma_addr >> 8));
-	WREG32_SOC15(GC, 0, mmSQ_SHADER_TMA_HI,
+	WREG32_SOC15(GC, inst, mmSQ_SHADER_TMA_HI,
 			upper_32_bits(tma_addr >> 8));
 
-	unlock_srbm(adev);
+	kgd_gfx_v9_unlock_srbm(adev, inst);
 }
 
 const struct kfd2kgd_calls gfx_v9_kfd2kgd = {