diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/mes_v10_1.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/mes_v10_1.c | 119 |
1 files changed, 85 insertions, 34 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c b/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c index 18a129f36215..067d10073a56 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c @@ -87,21 +87,32 @@ static const struct amdgpu_ring_funcs mes_v10_1_ring_funcs = { }; static int mes_v10_1_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, - void *pkt, int size) + void *pkt, int size, + int api_status_off) { int ndw = size / 4; signed long r; union MESAPI__ADD_QUEUE *x_pkt = pkt; + struct MES_API_STATUS *api_status; struct amdgpu_device *adev = mes->adev; struct amdgpu_ring *ring = &mes->ring; + unsigned long flags; BUG_ON(size % 4 != 0); - if (amdgpu_ring_alloc(ring, ndw)) + spin_lock_irqsave(&mes->ring_lock, flags); + if (amdgpu_ring_alloc(ring, ndw)) { + spin_unlock_irqrestore(&mes->ring_lock, flags); return -ENOMEM; + } + + api_status = (struct MES_API_STATUS *)((char *)pkt + api_status_off); + api_status->api_completion_fence_addr = mes->ring.fence_drv.gpu_addr; + api_status->api_completion_fence_value = ++mes->ring.fence_drv.sync_seq; amdgpu_ring_write_multiple(ring, pkt, ndw); amdgpu_ring_commit(ring); + spin_unlock_irqrestore(&mes->ring_lock, flags); DRM_DEBUG("MES msg=%d was emitted\n", x_pkt->header.opcode); @@ -166,13 +177,9 @@ static int mes_v10_1_add_hw_queue(struct amdgpu_mes *mes, mes_add_queue_pkt.gws_size = input->gws_size; mes_add_queue_pkt.trap_handler_addr = input->tba_addr; - mes_add_queue_pkt.api_status.api_completion_fence_addr = - mes->ring.fence_drv.gpu_addr; - mes_add_queue_pkt.api_status.api_completion_fence_value = - ++mes->ring.fence_drv.sync_seq; - return mes_v10_1_submit_pkt_and_poll_completion(mes, - &mes_add_queue_pkt, sizeof(mes_add_queue_pkt)); + &mes_add_queue_pkt, sizeof(mes_add_queue_pkt), + offsetof(union MESAPI__ADD_QUEUE, api_status)); } static int mes_v10_1_remove_hw_queue(struct amdgpu_mes *mes, @@ -189,13 +196,9 @@ static int mes_v10_1_remove_hw_queue(struct amdgpu_mes *mes, mes_remove_queue_pkt.doorbell_offset = input->doorbell_offset; mes_remove_queue_pkt.gang_context_addr = input->gang_context_addr; - mes_remove_queue_pkt.api_status.api_completion_fence_addr = - mes->ring.fence_drv.gpu_addr; - mes_remove_queue_pkt.api_status.api_completion_fence_value = - ++mes->ring.fence_drv.sync_seq; - return mes_v10_1_submit_pkt_and_poll_completion(mes, - &mes_remove_queue_pkt, sizeof(mes_remove_queue_pkt)); + &mes_remove_queue_pkt, sizeof(mes_remove_queue_pkt), + offsetof(union MESAPI__REMOVE_QUEUE, api_status)); } static int mes_v10_1_unmap_legacy_queue(struct amdgpu_mes *mes, @@ -227,13 +230,9 @@ static int mes_v10_1_unmap_legacy_queue(struct amdgpu_mes *mes, mes_remove_queue_pkt.unmap_kiq_utility_queue = 1; } - mes_remove_queue_pkt.api_status.api_completion_fence_addr = - mes->ring.fence_drv.gpu_addr; - mes_remove_queue_pkt.api_status.api_completion_fence_value = - ++mes->ring.fence_drv.sync_seq; - return mes_v10_1_submit_pkt_and_poll_completion(mes, - &mes_remove_queue_pkt, sizeof(mes_remove_queue_pkt)); + &mes_remove_queue_pkt, sizeof(mes_remove_queue_pkt), + offsetof(union MESAPI__REMOVE_QUEUE, api_status)); } static int mes_v10_1_suspend_gang(struct amdgpu_mes *mes, @@ -258,13 +257,9 @@ static int mes_v10_1_query_sched_status(struct amdgpu_mes *mes) mes_status_pkt.header.opcode = MES_SCH_API_QUERY_SCHEDULER_STATUS; mes_status_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS; - mes_status_pkt.api_status.api_completion_fence_addr = - mes->ring.fence_drv.gpu_addr; - mes_status_pkt.api_status.api_completion_fence_value = - ++mes->ring.fence_drv.sync_seq; - return mes_v10_1_submit_pkt_and_poll_completion(mes, - &mes_status_pkt, sizeof(mes_status_pkt)); + &mes_status_pkt, sizeof(mes_status_pkt), + offsetof(union MESAPI__QUERY_MES_STATUS, api_status)); } static int mes_v10_1_set_hw_resources(struct amdgpu_mes *mes) @@ -299,7 +294,7 @@ static int mes_v10_1_set_hw_resources(struct amdgpu_mes *mes) for (i = 0; i < AMD_PRIORITY_NUM_LEVELS; i++) mes_set_hw_res_pkt.aggregated_doorbells[i] = - mes->agreegated_doorbells[i]; + mes->aggregated_doorbells[i]; for (i = 0; i < 5; i++) { mes_set_hw_res_pkt.gc_base[i] = adev->reg_offset[GC_HWIP][0][i]; @@ -313,13 +308,63 @@ static int mes_v10_1_set_hw_resources(struct amdgpu_mes *mes) mes_set_hw_res_pkt.disable_mes_log = 1; mes_set_hw_res_pkt.use_different_vmid_compute = 1; - mes_set_hw_res_pkt.api_status.api_completion_fence_addr = - mes->ring.fence_drv.gpu_addr; - mes_set_hw_res_pkt.api_status.api_completion_fence_value = - ++mes->ring.fence_drv.sync_seq; - return mes_v10_1_submit_pkt_and_poll_completion(mes, - &mes_set_hw_res_pkt, sizeof(mes_set_hw_res_pkt)); + &mes_set_hw_res_pkt, sizeof(mes_set_hw_res_pkt), + offsetof(union MESAPI_SET_HW_RESOURCES, api_status)); +} + +static void mes_v10_1_init_aggregated_doorbell(struct amdgpu_mes *mes) +{ + struct amdgpu_device *adev = mes->adev; + uint32_t data; + + data = RREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL1); + data &= ~(CP_MES_DOORBELL_CONTROL1__DOORBELL_OFFSET_MASK | + CP_MES_DOORBELL_CONTROL1__DOORBELL_EN_MASK | + CP_MES_DOORBELL_CONTROL1__DOORBELL_HIT_MASK); + data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_LOW] << + CP_MES_DOORBELL_CONTROL1__DOORBELL_OFFSET__SHIFT; + data |= 1 << CP_MES_DOORBELL_CONTROL1__DOORBELL_EN__SHIFT; + WREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL1, data); + + data = RREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL2); + data &= ~(CP_MES_DOORBELL_CONTROL2__DOORBELL_OFFSET_MASK | + CP_MES_DOORBELL_CONTROL2__DOORBELL_EN_MASK | + CP_MES_DOORBELL_CONTROL2__DOORBELL_HIT_MASK); + data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_NORMAL] << + CP_MES_DOORBELL_CONTROL2__DOORBELL_OFFSET__SHIFT; + data |= 1 << CP_MES_DOORBELL_CONTROL2__DOORBELL_EN__SHIFT; + WREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL2, data); + + data = RREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL3); + data &= ~(CP_MES_DOORBELL_CONTROL3__DOORBELL_OFFSET_MASK | + CP_MES_DOORBELL_CONTROL3__DOORBELL_EN_MASK | + CP_MES_DOORBELL_CONTROL3__DOORBELL_HIT_MASK); + data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_MEDIUM] << + CP_MES_DOORBELL_CONTROL3__DOORBELL_OFFSET__SHIFT; + data |= 1 << CP_MES_DOORBELL_CONTROL3__DOORBELL_EN__SHIFT; + WREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL3, data); + + data = RREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL4); + data &= ~(CP_MES_DOORBELL_CONTROL4__DOORBELL_OFFSET_MASK | + CP_MES_DOORBELL_CONTROL4__DOORBELL_EN_MASK | + CP_MES_DOORBELL_CONTROL4__DOORBELL_HIT_MASK); + data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_HIGH] << + CP_MES_DOORBELL_CONTROL4__DOORBELL_OFFSET__SHIFT; + data |= 1 << CP_MES_DOORBELL_CONTROL4__DOORBELL_EN__SHIFT; + WREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL4, data); + + data = RREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL5); + data &= ~(CP_MES_DOORBELL_CONTROL5__DOORBELL_OFFSET_MASK | + CP_MES_DOORBELL_CONTROL5__DOORBELL_EN_MASK | + CP_MES_DOORBELL_CONTROL5__DOORBELL_HIT_MASK); + data |= mes->aggregated_doorbells[AMDGPU_MES_PRIORITY_LEVEL_REALTIME] << + CP_MES_DOORBELL_CONTROL5__DOORBELL_OFFSET__SHIFT; + data |= 1 << CP_MES_DOORBELL_CONTROL5__DOORBELL_EN__SHIFT; + WREG32_SOC15(GC, 0, mmCP_MES_DOORBELL_CONTROL5, data); + + data = 1 << CP_HQD_GFX_CONTROL__DB_UPDATED_MSG_EN__SHIFT; + WREG32_SOC15(GC, 0, mmCP_HQD_GFX_CONTROL, data); } static const struct amdgpu_mes_funcs mes_v10_1_funcs = { @@ -1121,6 +1166,8 @@ static int mes_v10_1_hw_init(void *handle) if (r) goto failure; + mes_v10_1_init_aggregated_doorbell(&adev->mes); + r = mes_v10_1_query_sched_status(&adev->mes); if (r) { DRM_ERROR("MES is busy\n"); @@ -1133,6 +1180,7 @@ static int mes_v10_1_hw_init(void *handle) * with MES enabled. */ adev->gfx.kiq.ring.sched.ready = false; + adev->mes.ring.sched.ready = true; return 0; @@ -1145,6 +1193,8 @@ static int mes_v10_1_hw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + adev->mes.ring.sched.ready = false; + mes_v10_1_enable(adev, false); if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) { @@ -1183,7 +1233,8 @@ static int mes_v10_0_late_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - amdgpu_mes_self_test(adev); + if (!amdgpu_in_reset(adev)) + amdgpu_mes_self_test(adev); return 0; } |