diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_5.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_5.c | 491 |
1 files changed, 356 insertions, 135 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_5.c b/drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_5.c index 78b74daf4..edf5bcdd2 100644 --- a/drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_5.c +++ b/drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_5.c @@ -34,7 +34,17 @@ #include "vcn/vcn_4_0_5_sh_mask.h" #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h" -#define regUVD_JPEG_PITCH_INTERNAL_OFFSET 0x401f +#define mmUVD_DPG_LMA_CTL regUVD_DPG_LMA_CTL +#define mmUVD_DPG_LMA_CTL_BASE_IDX regUVD_DPG_LMA_CTL_BASE_IDX +#define mmUVD_DPG_LMA_DATA regUVD_DPG_LMA_DATA +#define mmUVD_DPG_LMA_DATA_BASE_IDX regUVD_DPG_LMA_DATA_BASE_IDX + +#define regUVD_JPEG_PITCH_INTERNAL_OFFSET 0x401f +#define regJPEG_DEC_GFX10_ADDR_CONFIG_INTERNAL_OFFSET 0x4026 +#define regJPEG_SYS_INT_EN_INTERNAL_OFFSET 0x4141 +#define regJPEG_CGC_CTRL_INTERNAL_OFFSET 0x4161 +#define regJPEG_CGC_GATE_INTERNAL_OFFSET 0x4160 +#define regUVD_NO_OP_INTERNAL_OFFSET 0x0029 static void jpeg_v4_0_5_set_dec_ring_funcs(struct amdgpu_device *adev); static void jpeg_v4_0_5_set_irq_funcs(struct amdgpu_device *adev); @@ -43,6 +53,11 @@ static int jpeg_v4_0_5_set_powergating_state(void *handle, static void jpeg_v4_0_5_dec_ring_set_wptr(struct amdgpu_ring *ring); +static int amdgpu_ih_clientid_jpeg[] = { + SOC15_IH_CLIENTID_VCN, + SOC15_IH_CLIENTID_VCN1 +}; + /** * jpeg_v4_0_5_early_init - set function pointers * @@ -54,8 +69,20 @@ static int jpeg_v4_0_5_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + switch (amdgpu_ip_version(adev, UVD_HWIP, 0)) { + case IP_VERSION(4, 0, 5): + adev->jpeg.num_jpeg_inst = 1; + break; + case IP_VERSION(4, 0, 6): + adev->jpeg.num_jpeg_inst = 2; + break; + default: + DRM_DEV_ERROR(adev->dev, + "Failed to init vcn ip block(UVD_HWIP:0x%x)\n", + amdgpu_ip_version(adev, UVD_HWIP, 0)); + return -EINVAL; + } - adev->jpeg.num_jpeg_inst = 1; adev->jpeg.num_jpeg_rings = 1; jpeg_v4_0_5_set_dec_ring_funcs(adev); @@ -75,25 +102,30 @@ static int jpeg_v4_0_5_sw_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; struct amdgpu_ring *ring; - int r; - - /* JPEG TRAP */ - r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, - VCN_4_0__SRCID__JPEG_DECODE, &adev->jpeg.inst->irq); - if (r) - return r; - - /* JPEG DJPEG POISON EVENT */ - r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, - VCN_4_0__SRCID_DJPEG0_POISON, &adev->jpeg.inst->irq); - if (r) - return r; - - /* JPEG EJPEG POISON EVENT */ - r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, - VCN_4_0__SRCID_EJPEG0_POISON, &adev->jpeg.inst->irq); - if (r) - return r; + int r, i; + + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + /* JPEG TRAP */ + r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_jpeg[i], + VCN_4_0__SRCID__JPEG_DECODE, &adev->jpeg.inst[i].irq); + if (r) + return r; + + /* JPEG DJPEG POISON EVENT */ + r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_jpeg[i], + VCN_4_0__SRCID_DJPEG0_POISON, &adev->jpeg.inst[i].irq); + if (r) + return r; + + /* JPEG EJPEG POISON EVENT */ + r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_jpeg[i], + VCN_4_0__SRCID_EJPEG0_POISON, &adev->jpeg.inst[i].irq); + if (r) + return r; + } r = amdgpu_jpeg_sw_init(adev); if (r) @@ -103,21 +135,23 @@ static int jpeg_v4_0_5_sw_init(void *handle) if (r) return r; - ring = adev->jpeg.inst->ring_dec; - ring->use_doorbell = true; - ring->doorbell_index = amdgpu_sriov_vf(adev) ? - (((adev->doorbell_index.vcn.vcn_ring0_1) << 1) + 4) : - ((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1); - ring->vm_hub = AMDGPU_MMHUB0(0); - - sprintf(ring->name, "jpeg_dec"); - r = amdgpu_ring_init(adev, ring, 512, &adev->jpeg.inst->irq, 0, - AMDGPU_RING_PRIO_DEFAULT, NULL); - if (r) - return r; - - adev->jpeg.internal.jpeg_pitch[0] = regUVD_JPEG_PITCH_INTERNAL_OFFSET; - adev->jpeg.inst->external.jpeg_pitch[0] = SOC15_REG_OFFSET(JPEG, 0, regUVD_JPEG_PITCH); + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + ring = adev->jpeg.inst[i].ring_dec; + ring->use_doorbell = true; + ring->vm_hub = AMDGPU_MMHUB0(0); + ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1 + 8 * i; + sprintf(ring->name, "jpeg_dec_%d", i); + r = amdgpu_ring_init(adev, ring, 512, &adev->jpeg.inst[i].irq, + 0, AMDGPU_RING_PRIO_DEFAULT, NULL); + if (r) + return r; + + adev->jpeg.internal.jpeg_pitch[0] = regUVD_JPEG_PITCH_INTERNAL_OFFSET; + adev->jpeg.inst[i].external.jpeg_pitch[0] = SOC15_REG_OFFSET(JPEG, i, regUVD_JPEG_PITCH); + } return 0; } @@ -152,14 +186,27 @@ static int jpeg_v4_0_5_sw_fini(void *handle) static int jpeg_v4_0_5_hw_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct amdgpu_ring *ring = adev->jpeg.inst->ring_dec; - int r; + struct amdgpu_ring *ring; + int r, i; - r = amdgpu_ring_test_helper(ring); - if (r) - return r; + // TODO: Enable ring test with DPG support + if (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG) { + DRM_DEV_INFO(adev->dev, "JPEG decode initialized successfully under DPG Mode"); + return 0; + } + + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; - DRM_DEV_INFO(adev->dev, "JPEG decode initialized successfully.\n"); + ring = adev->jpeg.inst[i].ring_dec; + r = amdgpu_ring_test_helper(ring); + if (r) + return r; + } + + if (!r) + DRM_INFO("JPEG decode initialized successfully under SPG Mode\n"); return 0; } @@ -174,14 +221,20 @@ static int jpeg_v4_0_5_hw_init(void *handle) static int jpeg_v4_0_5_hw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int i; cancel_delayed_work_sync(&adev->vcn.idle_work); - if (!amdgpu_sriov_vf(adev)) { - if (adev->jpeg.cur_state != AMD_PG_STATE_GATE && - RREG32_SOC15(JPEG, 0, regUVD_JRBC_STATUS)) - jpeg_v4_0_5_set_powergating_state(adev, AMD_PG_STATE_GATE); - } + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + if (!amdgpu_sriov_vf(adev)) { + if (adev->jpeg.cur_state != AMD_PG_STATE_GATE && + RREG32_SOC15(JPEG, i, regUVD_JRBC_STATUS)) + jpeg_v4_0_5_set_powergating_state(adev, AMD_PG_STATE_GATE); + } + } return 0; } @@ -227,11 +280,11 @@ static int jpeg_v4_0_5_resume(void *handle) return r; } -static void jpeg_v4_0_5_disable_clock_gating(struct amdgpu_device *adev) +static void jpeg_v4_0_5_disable_clock_gating(struct amdgpu_device *adev, int inst) { uint32_t data = 0; - data = RREG32_SOC15(JPEG, 0, regJPEG_CGC_CTRL); + data = RREG32_SOC15(JPEG, inst, regJPEG_CGC_CTRL); if (adev->cg_flags & AMD_CG_SUPPORT_JPEG_MGCG) { data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; data &= (~JPEG_CGC_CTRL__JPEG_DEC_MODE_MASK); @@ -241,21 +294,21 @@ static void jpeg_v4_0_5_disable_clock_gating(struct amdgpu_device *adev) data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; - WREG32_SOC15(JPEG, 0, regJPEG_CGC_CTRL, data); + WREG32_SOC15(JPEG, inst, regJPEG_CGC_CTRL, data); - data = RREG32_SOC15(JPEG, 0, regJPEG_CGC_GATE); + data = RREG32_SOC15(JPEG, inst, regJPEG_CGC_GATE); data &= ~(JPEG_CGC_GATE__JPEG_DEC_MASK | JPEG_CGC_GATE__JPEG2_DEC_MASK | JPEG_CGC_GATE__JMCIF_MASK | JPEG_CGC_GATE__JRBBM_MASK); - WREG32_SOC15(JPEG, 0, regJPEG_CGC_GATE, data); + WREG32_SOC15(JPEG, inst, regJPEG_CGC_GATE, data); } -static void jpeg_v4_0_5_enable_clock_gating(struct amdgpu_device *adev) +static void jpeg_v4_0_5_enable_clock_gating(struct amdgpu_device *adev, int inst) { uint32_t data = 0; - data = RREG32_SOC15(JPEG, 0, regJPEG_CGC_CTRL); + data = RREG32_SOC15(JPEG, inst, regJPEG_CGC_CTRL); if (adev->cg_flags & AMD_CG_SUPPORT_JPEG_MGCG) { data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; data |= JPEG_CGC_CTRL__JPEG_DEC_MODE_MASK; @@ -265,47 +318,66 @@ static void jpeg_v4_0_5_enable_clock_gating(struct amdgpu_device *adev) data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; - WREG32_SOC15(JPEG, 0, regJPEG_CGC_CTRL, data); + WREG32_SOC15(JPEG, inst, regJPEG_CGC_CTRL, data); - data = RREG32_SOC15(JPEG, 0, regJPEG_CGC_GATE); + data = RREG32_SOC15(JPEG, inst, regJPEG_CGC_GATE); data |= (JPEG_CGC_GATE__JPEG_DEC_MASK |JPEG_CGC_GATE__JPEG2_DEC_MASK |JPEG_CGC_GATE__JMCIF_MASK |JPEG_CGC_GATE__JRBBM_MASK); - WREG32_SOC15(JPEG, 0, regJPEG_CGC_GATE, data); + WREG32_SOC15(JPEG, inst, regJPEG_CGC_GATE, data); } -static int jpeg_v4_0_5_disable_static_power_gating(struct amdgpu_device *adev) +static void jpeg_engine_4_0_5_dpg_clock_gating_mode(struct amdgpu_device *adev, + int inst_idx, uint8_t indirect) +{ + uint32_t data = 0; + + if (adev->cg_flags & AMD_CG_SUPPORT_JPEG_MGCG) + data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + else + data |= 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + + data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; + data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; + WREG32_SOC15_JPEG_DPG_MODE(inst_idx, regJPEG_CGC_CTRL_INTERNAL_OFFSET, data, indirect); + + data = 0; + WREG32_SOC15_JPEG_DPG_MODE(inst_idx, regJPEG_CGC_GATE_INTERNAL_OFFSET, + data, indirect); +} + +static int jpeg_v4_0_5_disable_static_power_gating(struct amdgpu_device *adev, int inst) { if (adev->pg_flags & AMD_PG_SUPPORT_JPEG) { - WREG32(SOC15_REG_OFFSET(JPEG, 0, regUVD_IPX_DLDO_CONFIG), + WREG32(SOC15_REG_OFFSET(JPEG, inst, regUVD_IPX_DLDO_CONFIG), 1 << UVD_IPX_DLDO_CONFIG__ONO1_PWR_CONFIG__SHIFT); - SOC15_WAIT_ON_RREG(JPEG, 0, regUVD_IPX_DLDO_STATUS, + SOC15_WAIT_ON_RREG(JPEG, inst, regUVD_IPX_DLDO_STATUS, 0, UVD_IPX_DLDO_STATUS__ONO1_PWR_STATUS_MASK); } /* disable anti hang mechanism */ - WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JPEG_POWER_STATUS), 0, + WREG32_P(SOC15_REG_OFFSET(JPEG, inst, regUVD_JPEG_POWER_STATUS), 0, ~UVD_JPEG_POWER_STATUS__JPEG_POWER_STATUS_MASK); /* keep the JPEG in static PG mode */ - WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JPEG_POWER_STATUS), 0, + WREG32_P(SOC15_REG_OFFSET(JPEG, inst, regUVD_JPEG_POWER_STATUS), 0, ~UVD_JPEG_POWER_STATUS__JPEG_PG_MODE_MASK); return 0; } -static int jpeg_v4_0_5_enable_static_power_gating(struct amdgpu_device *adev) +static int jpeg_v4_0_5_enable_static_power_gating(struct amdgpu_device *adev, int inst) { /* enable anti hang mechanism */ - WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JPEG_POWER_STATUS), + WREG32_P(SOC15_REG_OFFSET(JPEG, inst, regUVD_JPEG_POWER_STATUS), UVD_JPEG_POWER_STATUS__JPEG_POWER_STATUS_MASK, ~UVD_JPEG_POWER_STATUS__JPEG_POWER_STATUS_MASK); if (adev->pg_flags & AMD_PG_SUPPORT_JPEG) { - WREG32(SOC15_REG_OFFSET(JPEG, 0, regUVD_IPX_DLDO_CONFIG), + WREG32(SOC15_REG_OFFSET(JPEG, inst, regUVD_IPX_DLDO_CONFIG), 2 << UVD_IPX_DLDO_CONFIG__ONO1_PWR_CONFIG__SHIFT); - SOC15_WAIT_ON_RREG(JPEG, 0, regUVD_IPX_DLDO_STATUS, + SOC15_WAIT_ON_RREG(JPEG, inst, regUVD_IPX_DLDO_STATUS, 1 << UVD_IPX_DLDO_STATUS__ONO1_PWR_STATUS__SHIFT, UVD_IPX_DLDO_STATUS__ONO1_PWR_STATUS_MASK); } @@ -314,61 +386,153 @@ static int jpeg_v4_0_5_enable_static_power_gating(struct amdgpu_device *adev) } /** - * jpeg_v4_0_5_start - start JPEG block + * jpeg_v4_0_5_start_dpg_mode - Jpeg start with dpg mode * * @adev: amdgpu_device pointer + * @inst_idx: instance number index + * @indirect: indirectly write sram * - * Setup and start the JPEG block + * Start JPEG block with dpg mode */ -static int jpeg_v4_0_5_start(struct amdgpu_device *adev) +static void jpeg_v4_0_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect) { - struct amdgpu_ring *ring = adev->jpeg.inst->ring_dec; - int r; + struct amdgpu_ring *ring = adev->jpeg.inst[inst_idx].ring_dec; + uint32_t reg_data = 0; - if (adev->pm.dpm_enabled) - amdgpu_dpm_enable_jpeg(adev, true); + /* enable anti hang mechanism */ + reg_data = RREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS); + reg_data &= ~UVD_JPEG_POWER_STATUS__JPEG_POWER_STATUS_MASK; + reg_data |= 0x1; + WREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS, reg_data); - /* doorbell programming is done for every playback */ - adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell, - (adev->doorbell_index.vcn.vcn_ring0_1 << 1), 0); + if (adev->pg_flags & AMD_PG_SUPPORT_JPEG) { + WREG32(SOC15_REG_OFFSET(JPEG, inst_idx, regUVD_IPX_DLDO_CONFIG), + 2 << UVD_IPX_DLDO_CONFIG__ONO1_PWR_CONFIG__SHIFT); + SOC15_WAIT_ON_RREG(JPEG, inst_idx, regUVD_IPX_DLDO_STATUS, + 1 << UVD_IPX_DLDO_STATUS__ONO1_PWR_STATUS__SHIFT, + UVD_IPX_DLDO_STATUS__ONO1_PWR_STATUS_MASK); + } - WREG32_SOC15(VCN, 0, regVCN_JPEG_DB_CTRL, - ring->doorbell_index << VCN_JPEG_DB_CTRL__OFFSET__SHIFT | - VCN_JPEG_DB_CTRL__EN_MASK); + reg_data = RREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS); + reg_data |= UVD_JPEG_POWER_STATUS__JPEG_PG_MODE_MASK; + WREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS, reg_data); - /* disable power gating */ - r = jpeg_v4_0_5_disable_static_power_gating(adev); - if (r) - return r; + if (indirect) + adev->jpeg.inst[inst_idx].dpg_sram_curr_addr = + (uint32_t *)adev->jpeg.inst[inst_idx].dpg_sram_cpu_addr; - /* JPEG disable CGC */ - jpeg_v4_0_5_disable_clock_gating(adev); + jpeg_engine_4_0_5_dpg_clock_gating_mode(adev, inst_idx, indirect); /* MJPEG global tiling registers */ - WREG32_SOC15(JPEG, 0, regJPEG_DEC_GFX10_ADDR_CONFIG, - adev->gfx.config.gb_addr_config); - + WREG32_SOC15_JPEG_DPG_MODE(inst_idx, regJPEG_DEC_GFX10_ADDR_CONFIG_INTERNAL_OFFSET, + adev->gfx.config.gb_addr_config, indirect); + /* enable System Interrupt for JRBC */ + WREG32_SOC15_JPEG_DPG_MODE(inst_idx, regJPEG_SYS_INT_EN_INTERNAL_OFFSET, + JPEG_SYS_INT_EN__DJRBC_MASK, indirect); - /* enable JMI channel */ - WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JMI_CNTL), 0, - ~UVD_JMI_CNTL__SOFT_RESET_MASK); + /* add nop to workaround PSP size check */ + WREG32_SOC15_JPEG_DPG_MODE(inst_idx, regUVD_NO_OP_INTERNAL_OFFSET, 0, indirect); - /* enable System Interrupt for JRBC */ - WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regJPEG_SYS_INT_EN), - JPEG_SYS_INT_EN__DJRBC_MASK, - ~JPEG_SYS_INT_EN__DJRBC_MASK); + if (indirect) + amdgpu_jpeg_psp_update_sram(adev, inst_idx, 0); - WREG32_SOC15(JPEG, 0, regUVD_LMI_JRBC_RB_VMID, 0); - WREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_CNTL, (0x00000001L | 0x00000002L)); - WREG32_SOC15(JPEG, 0, regUVD_LMI_JRBC_RB_64BIT_BAR_LOW, + WREG32_SOC15(JPEG, inst_idx, regUVD_LMI_JRBC_RB_VMID, 0); + WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_CNTL, (0x00000001L | 0x00000002L)); + WREG32_SOC15(JPEG, inst_idx, regUVD_LMI_JRBC_RB_64BIT_BAR_LOW, lower_32_bits(ring->gpu_addr)); - WREG32_SOC15(JPEG, 0, regUVD_LMI_JRBC_RB_64BIT_BAR_HIGH, + WREG32_SOC15(JPEG, inst_idx, regUVD_LMI_JRBC_RB_64BIT_BAR_HIGH, upper_32_bits(ring->gpu_addr)); - WREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_RPTR, 0); - WREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_WPTR, 0); - WREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_CNTL, 0x00000002L); - WREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_SIZE, ring->ring_size / 4); - ring->wptr = RREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_WPTR); + WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_RPTR, 0); + WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_WPTR, 0); + WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_CNTL, 0x00000002L); + WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_SIZE, ring->ring_size / 4); + ring->wptr = RREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_WPTR); +} + +/** + * jpeg_v4_0_5_stop_dpg_mode - Jpeg stop with dpg mode + * + * @adev: amdgpu_device pointer + * @inst_idx: instance number index + * + * Stop JPEG block with dpg mode + */ +static void jpeg_v4_0_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx) +{ + uint32_t reg_data = 0; + + reg_data = RREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS); + reg_data &= ~UVD_JPEG_POWER_STATUS__JPEG_PG_MODE_MASK; + WREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS, reg_data); + +} + +/** + * jpeg_v4_0_5_start - start JPEG block + * + * @adev: amdgpu_device pointer + * + * Setup and start the JPEG block + */ +static int jpeg_v4_0_5_start(struct amdgpu_device *adev) +{ + struct amdgpu_ring *ring; + int r, i; + + if (adev->pm.dpm_enabled) + amdgpu_dpm_enable_jpeg(adev, true); + + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + ring = adev->jpeg.inst[i].ring_dec; + /* doorbell programming is done for every playback */ + adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell, + (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i, i); + + WREG32_SOC15(VCN, i, regVCN_JPEG_DB_CTRL, + ring->doorbell_index << VCN_JPEG_DB_CTRL__OFFSET__SHIFT | + VCN_JPEG_DB_CTRL__EN_MASK); + + if (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG) { + jpeg_v4_0_5_start_dpg_mode(adev, i, adev->jpeg.indirect_sram); + continue; + } + + /* disable power gating */ + r = jpeg_v4_0_5_disable_static_power_gating(adev, i); + if (r) + return r; + + /* JPEG disable CGC */ + jpeg_v4_0_5_disable_clock_gating(adev, i); + + /* MJPEG global tiling registers */ + WREG32_SOC15(JPEG, i, regJPEG_DEC_GFX10_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + + /* enable JMI channel */ + WREG32_P(SOC15_REG_OFFSET(JPEG, i, regUVD_JMI_CNTL), 0, + ~UVD_JMI_CNTL__SOFT_RESET_MASK); + + /* enable System Interrupt for JRBC */ + WREG32_P(SOC15_REG_OFFSET(JPEG, i, regJPEG_SYS_INT_EN), + JPEG_SYS_INT_EN__DJRBC_MASK, + ~JPEG_SYS_INT_EN__DJRBC_MASK); + + WREG32_SOC15(JPEG, i, regUVD_LMI_JRBC_RB_VMID, 0); + WREG32_SOC15(JPEG, i, regUVD_JRBC_RB_CNTL, (0x00000001L | 0x00000002L)); + WREG32_SOC15(JPEG, i, regUVD_LMI_JRBC_RB_64BIT_BAR_LOW, + lower_32_bits(ring->gpu_addr)); + WREG32_SOC15(JPEG, i, regUVD_LMI_JRBC_RB_64BIT_BAR_HIGH, + upper_32_bits(ring->gpu_addr)); + WREG32_SOC15(JPEG, i, regUVD_JRBC_RB_RPTR, 0); + WREG32_SOC15(JPEG, i, regUVD_JRBC_RB_WPTR, 0); + WREG32_SOC15(JPEG, i, regUVD_JRBC_RB_CNTL, 0x00000002L); + WREG32_SOC15(JPEG, i, regUVD_JRBC_RB_SIZE, ring->ring_size / 4); + ring->wptr = RREG32_SOC15(JPEG, i, regUVD_JRBC_RB_WPTR); + } return 0; } @@ -382,20 +546,29 @@ static int jpeg_v4_0_5_start(struct amdgpu_device *adev) */ static int jpeg_v4_0_5_stop(struct amdgpu_device *adev) { - int r; + int r, i; - /* reset JMI */ - WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JMI_CNTL), - UVD_JMI_CNTL__SOFT_RESET_MASK, - ~UVD_JMI_CNTL__SOFT_RESET_MASK); + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; - jpeg_v4_0_5_enable_clock_gating(adev); + if (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG) { + jpeg_v4_0_5_stop_dpg_mode(adev, i); + continue; + } - /* enable power gating */ - r = jpeg_v4_0_5_enable_static_power_gating(adev); - if (r) - return r; + /* reset JMI */ + WREG32_P(SOC15_REG_OFFSET(JPEG, i, regUVD_JMI_CNTL), + UVD_JMI_CNTL__SOFT_RESET_MASK, + ~UVD_JMI_CNTL__SOFT_RESET_MASK); + + jpeg_v4_0_5_enable_clock_gating(adev, i); + /* enable power gating */ + r = jpeg_v4_0_5_enable_static_power_gating(adev, i); + if (r) + return r; + } if (adev->pm.dpm_enabled) amdgpu_dpm_enable_jpeg(adev, false); @@ -413,7 +586,7 @@ static uint64_t jpeg_v4_0_5_dec_ring_get_rptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - return RREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_RPTR); + return RREG32_SOC15(JPEG, ring->me, regUVD_JRBC_RB_RPTR); } /** @@ -430,7 +603,7 @@ static uint64_t jpeg_v4_0_5_dec_ring_get_wptr(struct amdgpu_ring *ring) if (ring->use_doorbell) return *ring->wptr_cpu_addr; else - return RREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_WPTR); + return RREG32_SOC15(JPEG, ring->me, regUVD_JRBC_RB_WPTR); } /** @@ -448,29 +621,41 @@ static void jpeg_v4_0_5_dec_ring_set_wptr(struct amdgpu_ring *ring) *ring->wptr_cpu_addr = lower_32_bits(ring->wptr); WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); } else { - WREG32_SOC15(JPEG, 0, regUVD_JRBC_RB_WPTR, lower_32_bits(ring->wptr)); + WREG32_SOC15(JPEG, ring->me, regUVD_JRBC_RB_WPTR, lower_32_bits(ring->wptr)); } } static bool jpeg_v4_0_5_is_idle(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; - int ret = 1; + int i, ret = 1; - ret &= (((RREG32_SOC15(JPEG, 0, regUVD_JRBC_STATUS) & - UVD_JRBC_STATUS__RB_JOB_DONE_MASK) == - UVD_JRBC_STATUS__RB_JOB_DONE_MASK)); + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + ret &= (((RREG32_SOC15(JPEG, i, regUVD_JRBC_STATUS) & + UVD_JRBC_STATUS__RB_JOB_DONE_MASK) == + UVD_JRBC_STATUS__RB_JOB_DONE_MASK)); + } return ret; } static int jpeg_v4_0_5_wait_for_idle(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int i; + + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + return SOC15_WAIT_ON_RREG(JPEG, i, regUVD_JRBC_STATUS, + UVD_JRBC_STATUS__RB_JOB_DONE_MASK, + UVD_JRBC_STATUS__RB_JOB_DONE_MASK); + } - return SOC15_WAIT_ON_RREG(JPEG, 0, regUVD_JRBC_STATUS, - UVD_JRBC_STATUS__RB_JOB_DONE_MASK, - UVD_JRBC_STATUS__RB_JOB_DONE_MASK); + return 0; } static int jpeg_v4_0_5_set_clockgating_state(void *handle, @@ -478,13 +663,20 @@ static int jpeg_v4_0_5_set_clockgating_state(void *handle, { struct amdgpu_device *adev = (struct amdgpu_device *)handle; bool enable = (state == AMD_CG_STATE_GATE) ? true : false; + int i; - if (enable) { - if (!jpeg_v4_0_5_is_idle(handle)) - return -EBUSY; - jpeg_v4_0_5_enable_clock_gating(adev); - } else { - jpeg_v4_0_5_disable_clock_gating(adev); + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + if (enable) { + if (!jpeg_v4_0_5_is_idle(handle)) + return -EBUSY; + + jpeg_v4_0_5_enable_clock_gating(adev, i); + } else { + jpeg_v4_0_5_disable_clock_gating(adev, i); + } } return 0; @@ -519,11 +711,25 @@ static int jpeg_v4_0_5_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { + uint32_t ip_instance; + DRM_DEBUG("IH: JPEG TRAP\n"); + switch (entry->client_id) { + case SOC15_IH_CLIENTID_VCN: + ip_instance = 0; + break; + case SOC15_IH_CLIENTID_VCN1: + ip_instance = 1; + break; + default: + DRM_ERROR("Unhandled client id: %d\n", entry->client_id); + return 0; + } + switch (entry->src_id) { case VCN_4_0__SRCID__JPEG_DECODE: - amdgpu_fence_process(adev->jpeg.inst->ring_dec); + amdgpu_fence_process(adev->jpeg.inst[ip_instance].ring_dec); break; case VCN_4_0__SRCID_DJPEG0_POISON: case VCN_4_0__SRCID_EJPEG0_POISON: @@ -589,8 +795,16 @@ static const struct amdgpu_ring_funcs jpeg_v4_0_5_dec_ring_vm_funcs = { static void jpeg_v4_0_5_set_dec_ring_funcs(struct amdgpu_device *adev) { - adev->jpeg.inst->ring_dec->funcs = &jpeg_v4_0_5_dec_ring_vm_funcs; - DRM_DEV_INFO(adev->dev, "JPEG decode is enabled in VM mode\n"); + int i; + + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + adev->jpeg.inst[i].ring_dec->funcs = &jpeg_v4_0_5_dec_ring_vm_funcs; + adev->jpeg.inst[i].ring_dec->me = i; + DRM_DEV_INFO(adev->dev, "JPEG%d decode is enabled in VM mode\n", i); + } } static const struct amdgpu_irq_src_funcs jpeg_v4_0_5_irq_funcs = { @@ -599,8 +813,15 @@ static const struct amdgpu_irq_src_funcs jpeg_v4_0_5_irq_funcs = { static void jpeg_v4_0_5_set_irq_funcs(struct amdgpu_device *adev) { - adev->jpeg.inst->irq.num_types = 1; - adev->jpeg.inst->irq.funcs = &jpeg_v4_0_5_irq_funcs; + int i; + + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + + adev->jpeg.inst[i].irq.num_types = 1; + adev->jpeg.inst[i].irq.funcs = &jpeg_v4_0_5_irq_funcs; + } } const struct amdgpu_ip_block_version jpeg_v4_0_5_ip_block = { |