diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 662 |
1 files changed, 644 insertions, 18 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c index f67fca38c1a9..2d64ba1adf99 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c @@ -29,6 +29,7 @@ #include "soc15.h" #include "soc15d.h" #include "vcn_v2_0.h" +#include "mmsch_v1_0.h" #include "vcn/vcn_2_5_offset.h" #include "vcn/vcn_2_5_sh_mask.h" @@ -54,6 +55,9 @@ static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev); static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev); static int vcn_v2_5_set_powergating_state(void *handle, enum amd_powergating_state state); +static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev, + int inst_idx, struct dpg_pause_state *new_state); +static int vcn_v2_5_sriov_start(struct amdgpu_device *adev); static int amdgpu_ih_clientid_vcns[] = { SOC15_IH_CLIENTID_VCN, @@ -88,7 +92,13 @@ static int vcn_v2_5_early_init(void *handle) } else adev->vcn.num_vcn_inst = 1; - adev->vcn.num_enc_rings = 2; + if (amdgpu_sriov_vf(adev)) { + adev->vcn.num_vcn_inst = 2; + adev->vcn.harvest_config = 0; + adev->vcn.num_enc_rings = 1; + } else { + adev->vcn.num_enc_rings = 2; + } vcn_v2_5_set_dec_ring_funcs(adev); vcn_v2_5_set_enc_ring_funcs(adev); @@ -176,7 +186,9 @@ static int vcn_v2_5_sw_init(void *handle) ring = &adev->vcn.inst[j].ring_dec; ring->use_doorbell = true; - ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8*j; + + ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + + (amdgpu_sriov_vf(adev) ? 2*j : 8*j); sprintf(ring->name, "vcn_dec_%d", j); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 0); if (r) @@ -185,7 +197,10 @@ static int vcn_v2_5_sw_init(void *handle) for (i = 0; i < adev->vcn.num_enc_rings; ++i) { ring = &adev->vcn.inst[j].ring_enc[i]; ring->use_doorbell = true; - ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + i + 8*j; + + ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + + (amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j)); + sprintf(ring->name, "vcn_enc_%d.%d", j, i); r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 0); if (r) @@ -193,6 +208,15 @@ static int vcn_v2_5_sw_init(void *handle) } } + if (amdgpu_sriov_vf(adev)) { + r = amdgpu_virt_alloc_mm_table(adev); + if (r) + return r; + } + + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) + adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode; + return 0; } @@ -208,6 +232,9 @@ static int vcn_v2_5_sw_fini(void *handle) int r; struct amdgpu_device *adev = (struct amdgpu_device *)handle; + if (amdgpu_sriov_vf(adev)) + amdgpu_virt_free_mm_table(adev); + r = amdgpu_vcn_suspend(adev); if (r) return r; @@ -228,31 +255,44 @@ static int vcn_v2_5_hw_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; struct amdgpu_ring *ring; - int i, j, r; + int i, j, r = 0; + + if (amdgpu_sriov_vf(adev)) + r = vcn_v2_5_sriov_start(adev); for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { if (adev->vcn.harvest_config & (1 << j)) continue; - ring = &adev->vcn.inst[j].ring_dec; - adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell, - ring->doorbell_index, j); + if (amdgpu_sriov_vf(adev)) { + adev->vcn.inst[j].ring_enc[0].sched.ready = true; + adev->vcn.inst[j].ring_enc[1].sched.ready = false; + adev->vcn.inst[j].ring_enc[2].sched.ready = false; + adev->vcn.inst[j].ring_dec.sched.ready = true; + } else { - r = amdgpu_ring_test_helper(ring); - if (r) - goto done; + ring = &adev->vcn.inst[j].ring_dec; + + adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell, + ring->doorbell_index, j); - for (i = 0; i < adev->vcn.num_enc_rings; ++i) { - ring = &adev->vcn.inst[j].ring_enc[i]; r = amdgpu_ring_test_helper(ring); if (r) goto done; + + for (i = 0; i < adev->vcn.num_enc_rings; ++i) { + ring = &adev->vcn.inst[j].ring_enc[i]; + r = amdgpu_ring_test_helper(ring); + if (r) + goto done; + } } } done: if (!r) - DRM_INFO("VCN decode and encode initialized successfully.\n"); + DRM_INFO("VCN decode and encode initialized successfully(under %s).\n", + (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode"); return r; } @@ -275,7 +315,9 @@ static int vcn_v2_5_hw_fini(void *handle) continue; ring = &adev->vcn.inst[i].ring_dec; - if (RREG32_SOC15(VCN, i, mmUVD_STATUS)) + if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || + (adev->vcn.cur_state != AMD_PG_STATE_GATE && + RREG32_SOC15(VCN, i, mmUVD_STATUS))) vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE); ring->sched.ready = false; @@ -350,9 +392,9 @@ static void vcn_v2_5_mc_resume(struct amdgpu_device *adev) /* cache window 0: fw */ if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, - (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo)); + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo)); WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, - (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi)); + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi)); WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_OFFSET0, 0); offset = 0; } else { @@ -384,6 +426,99 @@ static void vcn_v2_5_mc_resume(struct amdgpu_device *adev) } } +static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect) +{ + uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); + uint32_t offset; + + /* cache window 0: fw */ + if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { + if (!indirect) { + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); + } else { + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); + } + offset = 0; + } else { + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), + lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), + upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect); + offset = size; + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), + AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect); + } + + if (!indirect) + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect); + else + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect); + + /* cache window 1: stack */ + if (!indirect) { + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), + lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), + upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); + } else { + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); + } + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect); + + /* cache window 2: context */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), + lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), + upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect); + + /* non-cache window */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect); + + /* VCN global tiling registers */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect); +} + /** * vcn_v2_5_disable_clock_gating - disable VCN clock gating * @@ -502,6 +637,54 @@ static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev) } } +static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev, + uint8_t sram_sel, int inst_idx, uint8_t indirect) +{ + uint32_t reg_data = 0; + + /* enable sw clock gating control */ + if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) + reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + else + reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; + reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; + reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | + UVD_CGC_CTRL__UDEC_CM_MODE_MASK | + UVD_CGC_CTRL__UDEC_IT_MODE_MASK | + UVD_CGC_CTRL__UDEC_DB_MODE_MASK | + UVD_CGC_CTRL__UDEC_MP_MODE_MASK | + UVD_CGC_CTRL__SYS_MODE_MASK | + UVD_CGC_CTRL__UDEC_MODE_MASK | + UVD_CGC_CTRL__MPEG2_MODE_MASK | + UVD_CGC_CTRL__REGS_MODE_MASK | + UVD_CGC_CTRL__RBC_MODE_MASK | + UVD_CGC_CTRL__LMI_MC_MODE_MASK | + UVD_CGC_CTRL__LMI_UMC_MODE_MASK | + UVD_CGC_CTRL__IDCT_MODE_MASK | + UVD_CGC_CTRL__MPRD_MODE_MASK | + UVD_CGC_CTRL__MPC_MODE_MASK | + UVD_CGC_CTRL__LBSI_MODE_MASK | + UVD_CGC_CTRL__LRBBM_MODE_MASK | + UVD_CGC_CTRL__WCB_MODE_MASK | + UVD_CGC_CTRL__VCPU_MODE_MASK | + UVD_CGC_CTRL__MMSCH_MODE_MASK); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect); + + /* turn off clock gating */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect); + + /* turn on SUVD clock gating */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect); + + /* turn on sw mode in UVD_SUVD_CGC_CTRL */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect); +} + /** * vcn_v2_5_enable_clock_gating - enable VCN clock gating * @@ -564,6 +747,138 @@ static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev) } } +static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect) +{ + struct amdgpu_ring *ring; + uint32_t rb_bufsz, tmp; + + /* disable register anti-hang mechanism */ + WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 1, + ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); + /* enable dynamic power gating mode */ + tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS); + tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK; + tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK; + WREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS, tmp); + + if (indirect) + adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t*)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr; + + /* enable clock gating */ + vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect); + + /* enable VCPU clock */ + tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); + tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; + tmp |= UVD_VCPU_CNTL__BLK_RST_MASK; + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect); + + /* disable master interupt */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect); + + /* setup mmUVD_LMI_CTRL */ + tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | + UVD_LMI_CTRL__REQ_MODE_MASK | + UVD_LMI_CTRL__CRC_RESET_MASK | + UVD_LMI_CTRL__MASK_MC_URGENT_MASK | + UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | + UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | + (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | + 0x00100000L); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect); + + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_MPC_CNTL), + 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect); + + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_MPC_SET_MUXA0), + ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | + (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | + (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect); + + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_MPC_SET_MUXB0), + ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | + (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | + (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect); + + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_MPC_SET_MUX), + ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | + (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect); + + vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect); + + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect); + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect); + + /* enable LMI MC and UMC channels */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_LMI_CTRL2), 0, 0, indirect); + + /* unblock VCPU register access */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect); + + tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); + tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect); + + /* enable master interrupt */ + WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0( + UVD, 0, mmUVD_MASTINT_EN), + UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect); + + if (indirect) + psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr, + (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr - + (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr)); + + ring = &adev->vcn.inst[inst_idx].ring_dec; + /* force RBC into idle state */ + rb_bufsz = order_base_2(ring->ring_size); + tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); + WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_CNTL, tmp); + + /* set the write pointer delay */ + WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0); + + /* set the wb address */ + WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR_ADDR, + (upper_32_bits(ring->gpu_addr) >> 2)); + + /* programm the RB_BASE for ring buffer */ + WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, + lower_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, + upper_32_bits(ring->gpu_addr)); + + /* Initialize the ring buffer's read and write pointers */ + WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR, 0); + + WREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2, 0); + + ring->wptr = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR); + WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR, + lower_32_bits(ring->wptr)); + + return 0; +} + static int vcn_v2_5_start(struct amdgpu_device *adev) { struct amdgpu_ring *ring; @@ -576,6 +891,11 @@ static int vcn_v2_5_start(struct amdgpu_device *adev) for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { if (adev->vcn.harvest_config & (1 << i)) continue; + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { + r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram); + continue; + } + /* disable register anti-hang mechanism */ WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_POWER_STATUS), 0, ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); @@ -585,6 +905,9 @@ static int vcn_v2_5_start(struct amdgpu_device *adev) WREG32_SOC15(UVD, i, mmUVD_STATUS, tmp); } + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) + return 0; + /*SW clock gating */ vcn_v2_5_disable_clock_gating(adev); @@ -741,6 +1064,233 @@ static int vcn_v2_5_start(struct amdgpu_device *adev) return 0; } +static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev, + struct amdgpu_mm_table *table) +{ + uint32_t data = 0, loop = 0, size = 0; + uint64_t addr = table->gpu_addr; + struct mmsch_v1_1_init_header *header = NULL;; + + header = (struct mmsch_v1_1_init_header *)table->cpu_addr; + size = header->total_size; + + /* + * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of + * memory descriptor location + */ + WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr)); + WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr)); + + /* 2, update vmid of descriptor */ + data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID); + data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK; + /* use domain0 for MM scheduler */ + data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT); + WREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID, data); + + /* 3, notify mmsch about the size of this descriptor */ + WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_SIZE, size); + + /* 4, set resp to zero */ + WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP, 0); + + /* + * 5, kick off the initialization and wait until + * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero + */ + WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001); + + data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP); + loop = 10; + while ((data & 0x10000002) != 0x10000002) { + udelay(100); + data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP); + loop--; + if (!loop) + break; + } + + if (!loop) { + dev_err(adev->dev, + "failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n", + data); + return -EBUSY; + } + + return 0; +} + +static int vcn_v2_5_sriov_start(struct amdgpu_device *adev) +{ + struct amdgpu_ring *ring; + uint32_t offset, size, tmp, i, rb_bufsz; + uint32_t table_size = 0; + struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } }; + struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } }; + struct mmsch_v1_0_cmd_direct_polling direct_poll = { { 0 } }; + struct mmsch_v1_0_cmd_end end = { { 0 } }; + uint32_t *init_table = adev->virt.mm_table.cpu_addr; + struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table; + + direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE; + direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE; + direct_poll.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_POLLING; + end.cmd_header.command_type = MMSCH_COMMAND__END; + + header->version = MMSCH_VERSION; + header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2; + init_table += header->total_size; + + for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { + header->eng[i].table_offset = header->total_size; + header->eng[i].init_status = 0; + header->eng[i].table_size = 0; + + table_size = 0; + + MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_STATUS), + ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY); + + size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); + /* mc resume*/ + if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), + adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), + adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi); + offset = 0; + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0), 0); + } else { + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), + lower_32_bits(adev->vcn.inst[i].gpu_addr)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), + upper_32_bits(adev->vcn.inst[i].gpu_addr)); + offset = size; + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0), + AMDGPU_UVD_FIRMWARE_OFFSET >> 3); + } + + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE0), + size); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), + lower_32_bits(adev->vcn.inst[i].gpu_addr + offset)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), + upper_32_bits(adev->vcn.inst[i].gpu_addr + offset)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET1), + 0); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE1), + AMDGPU_VCN_STACK_SIZE); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), + lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + + AMDGPU_VCN_STACK_SIZE)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), + upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + + AMDGPU_VCN_STACK_SIZE)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET2), + 0); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE2), + AMDGPU_VCN_CONTEXT_SIZE); + + ring = &adev->vcn.inst[i].ring_enc[0]; + ring->wptr = 0; + + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_LO), + lower_32_bits(ring->gpu_addr)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_HI), + upper_32_bits(ring->gpu_addr)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_RB_SIZE), + ring->ring_size / 4); + + ring = &adev->vcn.inst[i].ring_dec; + ring->wptr = 0; + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_RBC_RB_64BIT_BAR_LOW), + lower_32_bits(ring->gpu_addr)); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, + mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH), + upper_32_bits(ring->gpu_addr)); + + /* force RBC into idle state */ + rb_bufsz = order_base_2(ring->ring_size); + tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); + MMSCH_V1_0_INSERT_DIRECT_WT( + SOC15_REG_OFFSET(UVD, i, mmUVD_RBC_RB_CNTL), tmp); + + /* add end packet */ + memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end)); + table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4; + init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4; + + /* refine header */ + header->eng[i].table_size = table_size; + header->total_size += table_size; + } + + return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table); +} + +static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx) +{ + int ret_code = 0; + uint32_t tmp; + + /* Wait for power status to be 1 */ + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); + + /* wait for read ptr to be equal to write ptr */ + tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR); + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code); + + tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2); + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code); + + tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code); + + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); + + /* disable dynamic power gating mode */ + WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 0, + ~UVD_POWER_STATUS__UVD_PG_MODE_MASK); + + return 0; +} + static int vcn_v2_5_stop(struct amdgpu_device *adev) { uint32_t tmp; @@ -749,6 +1299,11 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev) for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { if (adev->vcn.harvest_config & (1 << i)) continue; + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { + r = vcn_v2_5_stop_dpg_mode(adev, i); + continue; + } + /* wait for vcn idle */ SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7, r); if (r) @@ -804,6 +1359,67 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev) return 0; } +static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev, + int inst_idx, struct dpg_pause_state *new_state) +{ + struct amdgpu_ring *ring; + uint32_t reg_data = 0; + int ret_code; + + /* pause/unpause if state is changed */ + if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) { + DRM_DEBUG("dpg pause state changed %d -> %d", + adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based); + reg_data = RREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE) & + (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); + + if (new_state->fw_based == VCN_DPG_STATE__PAUSE) { + ret_code = 0; + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 0x1, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); + + if (!ret_code) { + /* pause DPG */ + reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; + WREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE, reg_data); + + /* wait for ACK */ + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_DPG_PAUSE, + UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, + UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, ret_code); + + /* Restore */ + ring = &adev->vcn.inst[inst_idx].ring_enc[0]; + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); + + ring = &adev->vcn.inst[inst_idx].ring_enc[1]; + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); + WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); + + WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR, + RREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2) & 0x7FFFFFFF); + + SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, + UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code); + } + } else { + /* unpause dpg, no need to wait */ + reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; + WREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE, reg_data); + } + adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based; + } + + return 0; +} + /** * vcn_v2_5_dec_ring_get_rptr - get read pointer * @@ -846,6 +1462,10 @@ static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) + WREG32_SOC15(UVD, ring->me, mmUVD_SCRATCH2, + lower_32_bits(ring->wptr) | 0x80000000); + if (ring->use_doorbell) { adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr); WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); @@ -871,7 +1491,7 @@ static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = { .emit_ib = vcn_v2_0_dec_ring_emit_ib, .emit_fence = vcn_v2_0_dec_ring_emit_fence, .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush, - .test_ring = amdgpu_vcn_dec_ring_test_ring, + .test_ring = vcn_v2_0_dec_ring_test_ring, .test_ib = amdgpu_vcn_dec_ring_test_ib, .insert_nop = vcn_v2_0_dec_ring_insert_nop, .insert_start = vcn_v2_0_dec_ring_insert_start, @@ -1046,7 +1666,10 @@ static int vcn_v2_5_set_clockgating_state(void *handle, enum amd_clockgating_state state) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - bool enable = (state == AMD_CG_STATE_GATE) ? true : false; + bool enable = (state == AMD_CG_STATE_GATE); + + if (amdgpu_sriov_vf(adev)) + return 0; if (enable) { if (vcn_v2_5_is_idle(handle)) @@ -1065,6 +1688,9 @@ static int vcn_v2_5_set_powergating_state(void *handle, struct amdgpu_device *adev = (struct amdgpu_device *)handle; int ret; + if (amdgpu_sriov_vf(adev)) + return 0; + if(state == adev->vcn.cur_state) return 0; |