To support grbm select for multiple XCD case.
v2: unify naming style
Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
                        uint32_t queue, uint32_t vmid)
 {
        mutex_lock(&adev->srbm_mutex);
-       soc15_grbm_select(adev, mec, pipe, queue, vmid);
+       soc15_grbm_select(adev, mec, pipe, queue, vmid, 0);
 }
 
 static void unlock_srbm(struct amdgpu_device *adev)
 {
-       soc15_grbm_select(adev, 0, 0, 0, 0);
+       soc15_grbm_select(adev, 0, 0, 0, 0, 0);
        mutex_unlock(&adev->srbm_mutex);
 }
 
        *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);
+       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) +
                         queue_slot);
        *wave_cnt = reg_val & SPI_CSQ_WF_ACTIVE_COUNT_0__COUNT_MASK;
        DECLARE_BITMAP(cp_queue_bitmap, KGD_MAX_QUEUES);
 
        lock_spi_csq_mutexes(adev);
-       soc15_grbm_select(adev, 1, 0, 0, 0);
+       soc15_grbm_select(adev, 1, 0, 0, 0, 0);
 
        /*
         * Iterate through the shader engines and arrays of the device
        }
 
        amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-       soc15_grbm_select(adev, 0, 0, 0, 0);
+       soc15_grbm_select(adev, 0, 0, 0, 0, 0);
        unlock_spi_csq_mutexes(adev);
 
        /* Update the output parameters and return */
 
 static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
                                  u32 me, u32 pipe, u32 q, u32 vm)
 {
-       soc15_grbm_select(adev, me, pipe, q, vm);
+       soc15_grbm_select(adev, me, pipe, q, vm, 0);
 }
 
 static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
 
        mutex_lock(&adev->srbm_mutex);
        for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
-               soc15_grbm_select(adev, 0, 0, 0, i);
+               soc15_grbm_select(adev, 0, 0, 0, i, 0);
                /* CP and shaders */
                WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
                WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
        }
-       soc15_grbm_select(adev, 0, 0, 0, 0);
+       soc15_grbm_select(adev, 0, 0, 0, 0, 0);
        mutex_unlock(&adev->srbm_mutex);
 
        /* Initialize all compute VMIDs to have no GDS, GWS, or OA
        /* where to put LDS, scratch, GPUVM in FSA64 space */
        mutex_lock(&adev->srbm_mutex);
        for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
-               soc15_grbm_select(adev, 0, 0, 0, i);
+               soc15_grbm_select(adev, 0, 0, 0, i, 0);
                /* CP and shaders */
                if (i == 0) {
                        tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
                        WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp);
                }
        }
-       soc15_grbm_select(adev, 0, 0, 0, 0);
+       soc15_grbm_select(adev, 0, 0, 0, 0, 0);
 
        mutex_unlock(&adev->srbm_mutex);
 
                amdgpu_ring_clear_ring(ring);
 
                mutex_lock(&adev->srbm_mutex);
-               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
                gfx_v9_0_kiq_init_register(ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
        } else {
                memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
                if (amdgpu_sriov_vf(adev) && adev->in_suspend)
                        amdgpu_ring_clear_ring(ring);
                mutex_lock(&adev->srbm_mutex);
-               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
                gfx_v9_0_mqd_init(ring);
                gfx_v9_0_kiq_init_register(ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
 
                if (adev->gfx.kiq[0].mqd_backup)
                ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
                ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
                mutex_lock(&adev->srbm_mutex);
-               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
                gfx_v9_0_mqd_init(ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
 
                if (adev->gfx.mec.mqd_backup[mqd_idx])
                mutex_lock(&adev->srbm_mutex);
                soc15_grbm_select(adev, adev->gfx.kiq[0].ring.me,
                                adev->gfx.kiq[0].ring.pipe,
-                               adev->gfx.kiq[0].ring.queue, 0);
+                               adev->gfx.kiq[0].ring.queue, 0, 0);
                gfx_v9_0_kiq_fini_register(&adev->gfx.kiq[0].ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
        }
 
 
 
        for (i = first_vmid; i < last_vmid; i++) {
                data = 0;
-               soc15_grbm_select(adev, 0, 0, 0, i);
+               soc15_grbm_select(adev, 0, 0, 0, i, 0);
                data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
                data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
                data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE,
                WREG32(SOC15_REG_OFFSET(GC, 0, regSPI_GDBG_PER_VMID_CNTL), data);
        }
 
-       soc15_grbm_select(adev, 0, 0, 0, 0);
+       soc15_grbm_select(adev, 0, 0, 0, 0, 0);
        mutex_unlock(&adev->srbm_mutex);
 }
 
 
 static void gfx_v9_4_3_select_me_pipe_q(struct amdgpu_device *adev,
                                        u32 me, u32 pipe, u32 q, u32 vm)
 {
-       soc15_grbm_select(adev, me, pipe, q, vm);
+       soc15_grbm_select(adev, me, pipe, q, vm, 0);
 }
 
 static const struct amdgpu_gfx_funcs gfx_v9_4_3_gfx_funcs = {
 
        mutex_lock(&adev->srbm_mutex);
        for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
-               soc15_grbm_select(adev, 0, 0, 0, i);
+               soc15_grbm_select(adev, 0, 0, 0, i, 0);
                /* CP and shaders */
                WREG32_SOC15_RLC(GC, 0, regSH_MEM_CONFIG, sh_mem_config);
                WREG32_SOC15_RLC(GC, 0, regSH_MEM_BASES, sh_mem_bases);
        }
-       soc15_grbm_select(adev, 0, 0, 0, 0);
+       soc15_grbm_select(adev, 0, 0, 0, 0, 0);
        mutex_unlock(&adev->srbm_mutex);
 
        /* Initialize all compute VMIDs to have no GDS, GWS, or OA
        /* where to put LDS, scratch, GPUVM in FSA64 space */
        mutex_lock(&adev->srbm_mutex);
        for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
-               soc15_grbm_select(adev, 0, 0, 0, i);
+               soc15_grbm_select(adev, 0, 0, 0, i, 0);
                /* CP and shaders */
                if (i == 0) {
                        tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
                        WREG32_SOC15_RLC(GC, 0, regSH_MEM_BASES, tmp);
                }
        }
-       soc15_grbm_select(adev, 0, 0, 0, 0);
+       soc15_grbm_select(adev, 0, 0, 0, 0, 0);
 
        mutex_unlock(&adev->srbm_mutex);
 
                amdgpu_ring_clear_ring(ring);
 
                mutex_lock(&adev->srbm_mutex);
-               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
                gfx_v9_4_3_kiq_init_register(ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
        } else {
                memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
                ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
                ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
                mutex_lock(&adev->srbm_mutex);
-               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
                gfx_v9_4_3_mqd_init(ring);
                gfx_v9_4_3_kiq_init_register(ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
 
                if (adev->gfx.kiq[0].mqd_backup)
                ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
                ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
                mutex_lock(&adev->srbm_mutex);
-               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+               soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
                gfx_v9_4_3_mqd_init(ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
 
                if (adev->gfx.mec.mqd_backup[mqd_idx])
                mutex_lock(&adev->srbm_mutex);
                soc15_grbm_select(adev, adev->gfx.kiq[0].ring.me,
                                adev->gfx.kiq[0].ring.pipe,
-                               adev->gfx.kiq[0].ring.queue, 0);
+                               adev->gfx.kiq[0].ring.queue, 0, 0);
                gfx_v9_4_3_kiq_fini_register(&adev->gfx.kiq[0].ring);
-               soc15_grbm_select(adev, 0, 0, 0, 0);
+               soc15_grbm_select(adev, 0, 0, 0, 0, 0);
                mutex_unlock(&adev->srbm_mutex);
        }
 
 
 
 
 void soc15_grbm_select(struct amdgpu_device *adev,
-                    u32 me, u32 pipe, u32 queue, u32 vmid)
+                    u32 me, u32 pipe, u32 queue, u32 vmid, int xcc_id)
 {
        u32 grbm_gfx_cntl = 0;
        grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, PIPEID, pipe);
        grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, VMID, vmid);
        grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, QUEUEID, queue);
 
-       WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_CNTL, grbm_gfx_cntl);
+       WREG32_SOC15_RLC_SHADOW(GC, xcc_id, mmGRBM_GFX_CNTL, grbm_gfx_cntl);
 }
 
 static void soc15_vga_set_state(struct amdgpu_device *adev, bool state)
 
 #define SOC15_RAS_REG_FIELD_VAL(val, entry, field) SOC15_REG_FIELD_VAL((val), (entry).field##_count_mask, (entry).field##_count_shift)
 
 void soc15_grbm_select(struct amdgpu_device *adev,
-                   u32 me, u32 pipe, u32 queue, u32 vmid);
+                   u32 me, u32 pipe, u32 queue, u32 vmid, int xcc_id);
 void soc15_set_virt_ops(struct amdgpu_device *adev);
 
 void soc15_program_register_sequence(struct amdgpu_device *adev,