diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 1162 |
1 files changed, 935 insertions, 227 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c index b53c8fd4e8cf..c233edf60569 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c @@ -37,11 +37,18 @@ #include "gc/gc_9_4_3_sh_mask.h" #include "gfx_v9_4_3.h" +#include "gfx_v9_4_3_cleaner_shader.h" #include "amdgpu_xcp.h" #include "amdgpu_aca.h" MODULE_FIRMWARE("amdgpu/gc_9_4_3_mec.bin"); +MODULE_FIRMWARE("amdgpu/gc_9_4_4_mec.bin"); +MODULE_FIRMWARE("amdgpu/gc_9_5_0_mec.bin"); MODULE_FIRMWARE("amdgpu/gc_9_4_3_rlc.bin"); +MODULE_FIRMWARE("amdgpu/gc_9_4_4_rlc.bin"); +MODULE_FIRMWARE("amdgpu/gc_9_5_0_rlc.bin"); +MODULE_FIRMWARE("amdgpu/gc_9_4_3_sjt_mec.bin"); +MODULE_FIRMWARE("amdgpu/gc_9_4_4_sjt_mec.bin"); #define GFX9_MEC_HPD_SIZE 4096 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L @@ -49,9 +56,110 @@ MODULE_FIRMWARE("amdgpu/gc_9_4_3_rlc.bin"); #define GOLDEN_GB_ADDR_CONFIG 0x2a114042 #define CP_HQD_PERSISTENT_STATE_DEFAULT 0xbe05301 -#define mmSMNAID_XCD0_MCA_SMU 0x36430400 /* SMN AID XCD0 */ -#define mmSMNAID_XCD1_MCA_SMU 0x38430400 /* SMN AID XCD1 */ -#define mmSMNXCD_XCD0_MCA_SMU 0x40430400 /* SMN XCD XCD0 */ +#define XCC_REG_RANGE_0_LOW 0x2000 /* XCC gfxdec0 lower Bound */ +#define XCC_REG_RANGE_0_HIGH 0x3400 /* XCC gfxdec0 upper Bound */ +#define XCC_REG_RANGE_1_LOW 0xA000 /* XCC gfxdec1 lower Bound */ +#define XCC_REG_RANGE_1_HIGH 0x10000 /* XCC gfxdec1 upper Bound */ + +#define NORMALIZE_XCC_REG_OFFSET(offset) \ + (offset & 0xFFFF) + +static const struct amdgpu_hwip_reg_entry gc_reg_list_9_4_3[] = { + SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS2), + SOC15_REG_ENTRY_STR(GC, 0, regCP_STALLED_STAT1), + SOC15_REG_ENTRY_STR(GC, 0, regCP_STALLED_STAT2), + SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_STALLED_STAT1), + SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_STALLED_STAT1), + SOC15_REG_ENTRY_STR(GC, 0, regCP_BUSY_STAT), + SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_BUSY_STAT), + SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_BUSY_STAT), + SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_ERROR), + SOC15_REG_ENTRY_STR(GC, 0, regCPF_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regCPC_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regCPG_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regGDS_PROTECTION_FAULT), + SOC15_REG_ENTRY_STR(GC, 0, regGDS_VM_PROTECTION_FAULT), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regRMI_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regSQC_DCACHE_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regSQC_ICACHE_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regSQ_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regTCP_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regWD_UTCL1_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regVM_L2_PROTECTION_FAULT_CNTL), + SOC15_REG_ENTRY_STR(GC, 0, regVM_L2_PROTECTION_FAULT_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regCP_DEBUG), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_CNTL), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC1_INSTR_PNTR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC2_INSTR_PNTR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_STAT), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_COMMAND), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_MESSAGE), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_ARGUMENT_1), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_ARGUMENT_2), + SOC15_REG_ENTRY_STR(GC, 0, regSMU_RLC_RESPONSE), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_SAFE_MODE), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_SAFE_MODE), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_INT_STAT), + SOC15_REG_ENTRY_STR(GC, 0, regRLC_GPM_GENERAL_6), + /* SE status registers */ + SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE0), + SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE1), + SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE2), + SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE3) +}; + +static const struct amdgpu_hwip_reg_entry gc_cp_reg_list_9_4_3[] = { + /* compute queue registers */ + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_VMID), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_ACTIVE), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PERSISTENT_STATE), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PIPE_PRIORITY), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_QUEUE_PRIORITY), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_QUANTUM), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_BASE), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_BASE_HI), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_RPTR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_CONTROL), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_BASE_ADDR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_BASE_ADDR_HI), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_RPTR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_CONTROL), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_DEQUEUE_REQUEST), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_BASE_ADDR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_BASE_ADDR_HI), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_CONTROL), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_RPTR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_WPTR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_EVENTS), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_BASE_ADDR_LO), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_BASE_ADDR_HI), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_CONTROL), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CNTL_STACK_OFFSET), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CNTL_STACK_SIZE), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_WG_STATE_OFFSET), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_SIZE), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_GDS_RESOURCE_STATE), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_ERROR), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_WPTR_MEM), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_LO), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_HI), + SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_GFX_STATUS), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), + SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), +}; struct amdgpu_gfx_ras gfx_v9_4_3_ras; @@ -61,10 +169,18 @@ static void gfx_v9_4_3_set_gds_init(struct amdgpu_device *adev); static void gfx_v9_4_3_set_rlc_funcs(struct amdgpu_device *adev); static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev, struct amdgpu_cu_info *cu_info); +static void gfx_v9_4_3_xcc_set_safe_mode(struct amdgpu_device *adev, int xcc_id); +static void gfx_v9_4_3_xcc_unset_safe_mode(struct amdgpu_device *adev, int xcc_id); static void gfx_v9_4_3_kiq_set_resources(struct amdgpu_ring *kiq_ring, uint64_t queue_mask) { + struct amdgpu_device *adev = kiq_ring->adev; + u64 shader_mc_addr; + + /* Cleaner shader MC address */ + shader_mc_addr = adev->gfx.cleaner_shader_gpu_addr >> 8; + amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6)); amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) | @@ -74,8 +190,8 @@ static void gfx_v9_4_3_kiq_set_resources(struct amdgpu_ring *kiq_ring, lower_32_bits(queue_mask)); /* queue mask lo */ amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask)); /* queue mask hi */ - amdgpu_ring_write(kiq_ring, 0); /* gws mask lo */ - amdgpu_ring_write(kiq_ring, 0); /* gws mask hi */ + amdgpu_ring_write(kiq_ring, lower_32_bits(shader_mc_addr)); /* cleaner shader addr lo */ + amdgpu_ring_write(kiq_ring, upper_32_bits(shader_mc_addr)); /* cleaner shader addr hi */ amdgpu_ring_write(kiq_ring, 0); /* oac mask */ amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */ } @@ -172,12 +288,46 @@ static void gfx_v9_4_3_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring, PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type)); } +static void gfx_v9_4_3_kiq_reset_hw_queue(struct amdgpu_ring *kiq_ring, uint32_t queue_type, + uint32_t me_id, uint32_t pipe_id, uint32_t queue_id, + uint32_t xcc_id, uint32_t vmid) +{ + struct amdgpu_device *adev = kiq_ring->adev; + unsigned i; + + /* enter save mode */ + amdgpu_gfx_rlc_enter_safe_mode(adev, xcc_id); + mutex_lock(&adev->srbm_mutex); + soc15_grbm_select(adev, me_id, pipe_id, queue_id, 0, xcc_id); + + if (queue_type == AMDGPU_RING_TYPE_COMPUTE) { + WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 0x2); + WREG32_SOC15(GC, GET_INST(GC, xcc_id), regSPI_COMPUTE_QUEUE_RESET, 0x1); + /* wait till dequeue take effects */ + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + if (i >= adev->usec_timeout) + dev_err(adev->dev, "fail to wait on hqd deactive\n"); + } else { + dev_err(adev->dev, "reset queue_type(%d) not supported\n\n", queue_type); + } + + soc15_grbm_select(adev, 0, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); + /* exit safe mode */ + amdgpu_gfx_rlc_exit_safe_mode(adev, xcc_id); +} + static const struct kiq_pm4_funcs gfx_v9_4_3_kiq_pm4_funcs = { .kiq_set_resources = gfx_v9_4_3_kiq_set_resources, .kiq_map_queues = gfx_v9_4_3_kiq_map_queues, .kiq_unmap_queues = gfx_v9_4_3_kiq_unmap_queues, .kiq_query_status = gfx_v9_4_3_kiq_query_status, .kiq_invalidate_tlbs = gfx_v9_4_3_kiq_invalidate_tlbs, + .kiq_reset_hw_queue = gfx_v9_4_3_kiq_reset_hw_queue, .set_resources_size = 8, .map_queues_size = 7, .unmap_queues_size = 6, @@ -204,20 +354,28 @@ static void gfx_v9_4_3_init_golden_registers(struct amdgpu_device *adev) WREG32_SOC15(GC, dev_inst, regGB_ADDR_CONFIG, GOLDEN_GB_ADDR_CONFIG); - /* Golden settings applied by driver for ASIC with rev_id 0 */ - if (adev->rev_id == 0) { - WREG32_FIELD15_PREREG(GC, dev_inst, TCP_UTCL1_CNTL1, - REDUCE_FIFO_DEPTH_BY_2, 2); - } else { - WREG32_FIELD15_PREREG(GC, dev_inst, TCP_UTCL1_CNTL2, - SPARE, 0x1); - } + WREG32_FIELD15_PREREG(GC, dev_inst, TCP_UTCL1_CNTL2, SPARE, 0x1); } } +static uint32_t gfx_v9_4_3_normalize_xcc_reg_offset(uint32_t reg) +{ + uint32_t normalized_reg = NORMALIZE_XCC_REG_OFFSET(reg); + + /* If it is an XCC reg, normalize the reg to keep + lower 16 bits in local xcc */ + + if (((normalized_reg >= XCC_REG_RANGE_0_LOW) && (normalized_reg < XCC_REG_RANGE_0_HIGH)) || + ((normalized_reg >= XCC_REG_RANGE_1_LOW) && (normalized_reg < XCC_REG_RANGE_1_HIGH))) + return normalized_reg; + else + return reg; +} + static void gfx_v9_4_3_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel, bool wc, uint32_t reg, uint32_t val) { + reg = gfx_v9_4_3_normalize_xcc_reg_offset(reg); amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) | WRITE_DATA_DST_SEL(0) | @@ -232,6 +390,12 @@ static void gfx_v9_4_3_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel, uint32_t addr1, uint32_t ref, uint32_t mask, uint32_t inv) { + /* Only do the normalization on regspace */ + if (mem_space == 0) { + addr0 = gfx_v9_4_3_normalize_xcc_reg_offset(addr0); + addr1 = gfx_v9_4_3_normalize_xcc_reg_offset(addr1); + } + amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); amdgpu_ring_write(ring, /* memory (1) or register (0) */ @@ -333,7 +497,7 @@ static int gfx_v9_4_3_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = -EINVAL; err2: - amdgpu_ib_free(adev, &ib, NULL); + amdgpu_ib_free(&ib, NULL); dma_fence_put(f); err1: amdgpu_device_wb_free(adev, index); @@ -370,15 +534,15 @@ static void gfx_v9_4_3_free_microcode(struct amdgpu_device *adev) static int gfx_v9_4_3_init_rlc_microcode(struct amdgpu_device *adev, const char *chip_name) { - char fw_name[30]; int err; const struct rlc_firmware_header_v2_0 *rlc_hdr; uint16_t version_major; uint16_t version_minor; - snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name); - err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name); + err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, + AMDGPU_UCODE_REQUIRED, + "amdgpu/%s_rlc.bin", chip_name); if (err) goto out; rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data; @@ -393,26 +557,24 @@ out: return err; } -static bool gfx_v9_4_3_should_disable_gfxoff(struct pci_dev *pdev) -{ - return true; -} - -static void gfx_v9_4_3_check_if_need_gfxoff(struct amdgpu_device *adev) -{ - if (gfx_v9_4_3_should_disable_gfxoff(adev->pdev)) - adev->pm.pp_feature &= ~PP_GFXOFF_MASK; -} - static int gfx_v9_4_3_init_cp_compute_microcode(struct amdgpu_device *adev, const char *chip_name) { - char fw_name[30]; int err; - snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name); - - err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name); + if (amdgpu_sriov_vf(adev)) { + err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, + AMDGPU_UCODE_REQUIRED, + "amdgpu/%s_sjt_mec.bin", chip_name); + + if (err) + err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, + AMDGPU_UCODE_REQUIRED, + "amdgpu/%s_mec.bin", chip_name); + } else + err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, + AMDGPU_UCODE_REQUIRED, + "amdgpu/%s_mec.bin", chip_name); if (err) goto out; amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1); @@ -421,8 +583,6 @@ static int gfx_v9_4_3_init_cp_compute_microcode(struct amdgpu_device *adev, adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version; adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version; - gfx_v9_4_3_check_if_need_gfxoff(adev); - out: if (err) amdgpu_ucode_release(&adev->gfx.mec_fw); @@ -431,16 +591,16 @@ out: static int gfx_v9_4_3_init_microcode(struct amdgpu_device *adev) { - const char *chip_name; + char ucode_prefix[15]; int r; - chip_name = "gc_9_4_3"; + amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix)); - r = gfx_v9_4_3_init_rlc_microcode(adev, chip_name); + r = gfx_v9_4_3_init_rlc_microcode(adev, ucode_prefix); if (r) return r; - r = gfx_v9_4_3_init_cp_compute_microcode(adev, chip_name); + r = gfx_v9_4_3_init_cp_compute_microcode(adev, ucode_prefix); if (r) return r; @@ -624,6 +784,15 @@ static void gfx_v9_4_3_select_me_pipe_q(struct amdgpu_device *adev, soc15_grbm_select(adev, me, pipe, q, vm, GET_INST(GC, xcc_id)); } +static int gfx_v9_4_3_get_xccs_per_xcp(struct amdgpu_device *adev) +{ + u32 xcp_ctl; + + /* Value is expected to be the same on all, fetch from first instance */ + xcp_ctl = RREG32_SOC15(GC, GET_INST(GC, 0), regCP_HYP_XCP_CTL); + + return REG_GET_FIELD(xcp_ctl, CP_HYP_XCP_CTL, NUM_XCC_IN_XCP); +} static int gfx_v9_4_3_switch_compute_partition(struct amdgpu_device *adev, int num_xccs_per_xcp) @@ -678,40 +847,48 @@ static const struct amdgpu_gfx_funcs gfx_v9_4_3_gfx_funcs = { .select_me_pipe_q = &gfx_v9_4_3_select_me_pipe_q, .switch_partition_mode = &gfx_v9_4_3_switch_compute_partition, .ih_node_to_logical_xcc = &gfx_v9_4_3_ih_to_xcc_inst, + .get_xccs_per_xcp = &gfx_v9_4_3_get_xccs_per_xcp, }; -static int gfx_v9_4_3_aca_bank_generate_report(struct aca_handle *handle, - struct aca_bank *bank, enum aca_error_type type, - struct aca_bank_report *report, void *data) +static int gfx_v9_4_3_aca_bank_parser(struct aca_handle *handle, + struct aca_bank *bank, enum aca_smu_type type, + void *data) { - u64 status, misc0; + struct aca_bank_info info; + u64 misc0; u32 instlo; int ret; - status = bank->regs[ACA_REG_IDX_STATUS]; - if ((type == ACA_ERROR_TYPE_UE && - ACA_REG__STATUS__ERRORCODEEXT(status) == ACA_EXTERROR_CODE_FAULT) || - (type == ACA_ERROR_TYPE_CE && - ACA_REG__STATUS__ERRORCODEEXT(status) == ACA_EXTERROR_CODE_CE)) { + ret = aca_bank_info_decode(bank, &info); + if (ret) + return ret; - ret = aca_bank_info_decode(bank, &report->info); - if (ret) - return ret; + /* NOTE: overwrite info.die_id with xcd id for gfx */ + instlo = ACA_REG__IPID__INSTANCEIDLO(bank->regs[ACA_REG_IDX_IPID]); + instlo &= GENMASK(31, 1); + info.die_id = instlo == mmSMNAID_XCD0_MCA_SMU ? 0 : 1; - /* NOTE: overwrite info.die_id with xcd id for gfx */ - instlo = ACA_REG__IPID__INSTANCEIDLO(bank->regs[ACA_REG_IDX_IPID]); - instlo &= GENMASK(31, 1); - report->info.die_id = instlo == mmSMNAID_XCD0_MCA_SMU ? 0 : 1; + misc0 = bank->regs[ACA_REG_IDX_MISC0]; - misc0 = bank->regs[ACA_REG_IDX_MISC0]; - report->count[type] = ACA_REG__MISC0__ERRCNT(misc0); + switch (type) { + case ACA_SMU_TYPE_UE: + bank->aca_err_type = ACA_ERROR_TYPE_UE; + ret = aca_error_cache_log_bank_error(handle, &info, bank->aca_err_type, 1ULL); + break; + case ACA_SMU_TYPE_CE: + bank->aca_err_type = ACA_ERROR_TYPE_CE; + ret = aca_error_cache_log_bank_error(handle, &info, bank->aca_err_type, + ACA_REG__MISC0__ERRCNT(misc0)); + break; + default: + return -EINVAL; } - return 0; + return ret; } static bool gfx_v9_4_3_aca_bank_is_valid(struct aca_handle *handle, struct aca_bank *bank, - enum aca_error_type type, void *data) + enum aca_smu_type type, void *data) { u32 instlo; @@ -730,7 +907,7 @@ static bool gfx_v9_4_3_aca_bank_is_valid(struct aca_handle *handle, struct aca_b } static const struct aca_bank_ops gfx_v9_4_3_aca_bank_ops = { - .aca_bank_generate_report = gfx_v9_4_3_aca_bank_generate_report, + .aca_bank_parser = gfx_v9_4_3_aca_bank_parser, .aca_bank_is_valid = gfx_v9_4_3_aca_bank_is_valid, }; @@ -742,26 +919,15 @@ static const struct aca_info gfx_v9_4_3_aca_info = { static int gfx_v9_4_3_gpu_early_init(struct amdgpu_device *adev) { - u32 gb_addr_config; - adev->gfx.funcs = &gfx_v9_4_3_gfx_funcs; adev->gfx.ras = &gfx_v9_4_3_ras; - switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { - case IP_VERSION(9, 4, 3): - adev->gfx.config.max_hw_contexts = 8; - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; - gb_addr_config = RREG32_SOC15(GC, GET_INST(GC, 0), regGB_ADDR_CONFIG); - break; - default: - BUG(); - break; - } - - adev->gfx.config.gb_addr_config = gb_addr_config; + adev->gfx.config.max_hw_contexts = 8; + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; + adev->gfx.config.gb_addr_config = GOLDEN_GB_ADDR_CONFIG; adev->gfx.config.gb_addr_config_fields.num_pipes = 1 << REG_GET_FIELD( @@ -840,10 +1006,58 @@ static int gfx_v9_4_3_compute_ring_init(struct amdgpu_device *adev, int ring_id, hw_prio, NULL); } -static int gfx_v9_4_3_sw_init(void *handle) +static void gfx_v9_4_3_alloc_ip_dump(struct amdgpu_device *adev) +{ + uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9_4_3); + uint32_t *ptr, num_xcc, inst; + + num_xcc = NUM_XCC(adev->gfx.xcc_mask); + + ptr = kcalloc(reg_count * num_xcc, sizeof(uint32_t), GFP_KERNEL); + if (!ptr) { + DRM_ERROR("Failed to allocate memory for GFX IP Dump\n"); + adev->gfx.ip_dump_core = NULL; + } else { + adev->gfx.ip_dump_core = ptr; + } + + /* Allocate memory for compute queue registers for all the instances */ + reg_count = ARRAY_SIZE(gc_cp_reg_list_9_4_3); + inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec * + adev->gfx.mec.num_queue_per_pipe; + + ptr = kcalloc(reg_count * inst * num_xcc, sizeof(uint32_t), GFP_KERNEL); + if (!ptr) { + DRM_ERROR("Failed to allocate memory for Compute Queues IP Dump\n"); + adev->gfx.ip_dump_compute_queues = NULL; + } else { + adev->gfx.ip_dump_compute_queues = ptr; + } +} + +static int gfx_v9_4_3_sw_init(struct amdgpu_ip_block *ip_block) { int i, j, k, r, ring_id, xcc_id, num_xcc; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; + + switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { + case IP_VERSION(9, 4, 3): + case IP_VERSION(9, 4, 4): + adev->gfx.cleaner_shader_ptr = gfx_9_4_3_cleaner_shader_hex; + adev->gfx.cleaner_shader_size = sizeof(gfx_9_4_3_cleaner_shader_hex); + if (adev->gfx.mec_fw_version >= 153) { + adev->gfx.enable_cleaner_shader = true; + r = amdgpu_gfx_cleaner_shader_sw_init(adev, adev->gfx.cleaner_shader_size); + if (r) { + adev->gfx.enable_cleaner_shader = false; + dev_err(adev->dev, "Failed to initialize cleaner shader\n"); + } + } + break; + default: + adev->gfx.enable_cleaner_shader = false; + break; + } adev->gfx.mec.num_mec = 2; adev->gfx.mec.num_pipe_per_mec = 4; @@ -856,6 +1070,13 @@ static int gfx_v9_4_3_sw_init(void *handle) if (r) return r; + /* Bad opcode Event */ + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, + GFX_9_0__SRCID__CP_BAD_OPCODE_ERROR, + &adev->gfx.bad_op_irq); + if (r) + return r; + /* Privileged reg */ r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT, &adev->gfx.priv_reg_irq); @@ -922,6 +1143,25 @@ static int gfx_v9_4_3_sw_init(void *handle) return r; } + adev->gfx.compute_supported_reset = + amdgpu_get_soft_full_reset_mask(&adev->gfx.compute_ring[0]); + switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { + case IP_VERSION(9, 4, 3): + case IP_VERSION(9, 4, 4): + if (adev->gfx.mec_fw_version >= 155) { + adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; + adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_PIPE; + } + break; + case IP_VERSION(9, 5, 0): + if (adev->gfx.mec_fw_version >= 21) { + adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; + adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_PIPE; + } + break; + default: + break; + } r = gfx_v9_4_3_gpu_early_init(adev); if (r) return r; @@ -930,17 +1170,19 @@ static int gfx_v9_4_3_sw_init(void *handle) if (r) return r; + r = amdgpu_gfx_sysfs_init(adev); + if (r) + return r; - if (!amdgpu_sriov_vf(adev)) - r = amdgpu_gfx_sysfs_init(adev); + gfx_v9_4_3_alloc_ip_dump(adev); - return r; + return 0; } -static int gfx_v9_4_3_sw_fini(void *handle) +static int gfx_v9_4_3_sw_fini(struct amdgpu_ip_block *ip_block) { int i, num_xcc; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; num_xcc = NUM_XCC(adev->gfx.xcc_mask); for (i = 0; i < adev->gfx.num_compute_rings * num_xcc; i++) @@ -952,11 +1194,15 @@ static int gfx_v9_4_3_sw_fini(void *handle) amdgpu_gfx_kiq_fini(adev, i); } + amdgpu_gfx_cleaner_shader_sw_fini(adev); + gfx_v9_4_3_mec_fini(adev); amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj); gfx_v9_4_3_free_microcode(adev); - if (!amdgpu_sriov_vf(adev)) - amdgpu_gfx_sysfs_fini(adev); + amdgpu_gfx_sysfs_fini(adev); + + kfree(adev->gfx.ip_dump_core); + kfree(adev->gfx.ip_dump_compute_queues); return 0; } @@ -997,8 +1243,10 @@ static void gfx_v9_4_3_xcc_init_compute_vmid(struct amdgpu_device *adev, soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); mutex_unlock(&adev->srbm_mutex); - /* Initialize all compute VMIDs to have no GDS, GWS, or OA - acccess. These should be enabled by FW for target VMIDs. */ + /* + * Initialize all compute VMIDs to have no GDS, GWS, or OA + * access. These should be enabled by FW for target VMIDs. + */ for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_BASE, 2 * i, 0); WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_SIZE, 2 * i, 0); @@ -1025,6 +1273,22 @@ static void gfx_v9_4_3_xcc_init_gds_vmid(struct amdgpu_device *adev, int xcc_id) } } +/* For ASICs that needs xnack chain and MEC version supports, set SG_CONFIG1 + * DISABLE_XNACK_CHECK_IN_RETRY_DISABLE bit and inform KFD to set xnack_chain + * bit in SET_RESOURCES + */ +static void gfx_v9_4_3_xcc_init_sq(struct amdgpu_device *adev, int xcc_id) +{ + uint32_t data; + + if (!(adev->gmc.xnack_flags & AMDGPU_GMC_XNACK_FLAG_CHAIN)) + return; + + data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_CONFIG1); + data = REG_SET_FIELD(data, SQ_CONFIG1, DISABLE_XNACK_CHECK_IN_RETRY_DISABLE, 1); + WREG32_SOC15(GC, xcc_id, regSQ_CONFIG1, data); +} + static void gfx_v9_4_3_xcc_constants_init(struct amdgpu_device *adev, int xcc_id) { @@ -1069,6 +1333,7 @@ static void gfx_v9_4_3_xcc_constants_init(struct amdgpu_device *adev, gfx_v9_4_3_xcc_init_compute_vmid(adev, xcc_id); gfx_v9_4_3_xcc_init_gds_vmid(adev, xcc_id); + gfx_v9_4_3_xcc_init_sq(adev, xcc_id); } static void gfx_v9_4_3_constants_init(struct amdgpu_device *adev) @@ -1081,6 +1346,20 @@ static void gfx_v9_4_3_constants_init(struct amdgpu_device *adev) adev->gfx.config.db_debug2 = RREG32_SOC15(GC, GET_INST(GC, 0), regDB_DEBUG2); + switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { + /* ToDo: GC 9.4.4 */ + case IP_VERSION(9, 4, 3): + if (adev->gfx.mec_fw_version >= 184) + adev->gmc.xnack_flags |= AMDGPU_GMC_XNACK_FLAG_CHAIN; + break; + case IP_VERSION(9, 5, 0): + if (adev->gfx.mec_fw_version >= 23) + adev->gmc.xnack_flags |= AMDGPU_GMC_XNACK_FLAG_CHAIN; + break; + default: + break; + } + for (i = 0; i < num_xcc; i++) gfx_v9_4_3_xcc_constants_init(adev, i); } @@ -1096,10 +1375,8 @@ static void gfx_v9_4_3_xcc_init_pg(struct amdgpu_device *adev, int xcc_id) { /* * Rlc save restore list is workable since v2_1. - * And it's needed by gfxoff feature. */ - if (adev->gfx.rlc.is_rlc_v2_1) - gfx_v9_4_3_xcc_enable_save_restore_machine(adev, xcc_id); + gfx_v9_4_3_xcc_enable_save_restore_machine(adev, xcc_id); } static void gfx_v9_4_3_xcc_disable_gpa_mode(struct amdgpu_device *adev, int xcc_id) @@ -1386,21 +1663,23 @@ static int gfx_v9_4_3_rlc_resume(struct amdgpu_device *adev) static void gfx_v9_4_3_update_spm_vmid(struct amdgpu_device *adev, struct amdgpu_ring *ring, unsigned vmid) { - u32 reg, data; + u32 reg, pre_data, data; reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL); - if (amdgpu_sriov_is_pp_one_vf(adev)) - data = RREG32_NO_KIQ(reg); + if (amdgpu_sriov_is_pp_one_vf(adev) && !amdgpu_sriov_runtime(adev)) + pre_data = RREG32_NO_KIQ(reg); else - data = RREG32(reg); + pre_data = RREG32(reg); - data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK; + data = pre_data & (~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK); data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT; - if (amdgpu_sriov_is_pp_one_vf(adev)) - WREG32_SOC15_NO_KIQ(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data); - else - WREG32_SOC15(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data); + if (pre_data != data) { + if (amdgpu_sriov_is_pp_one_vf(adev) && !amdgpu_sriov_runtime(adev)) { + WREG32_SOC15_NO_KIQ(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data); + } else + WREG32_SOC15(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data); + } } static const struct soc15_reg_rlcg rlcg_access_gc_9_4_3[] = { @@ -1449,7 +1728,15 @@ static void gfx_v9_4_3_xcc_cp_compute_enable(struct amdgpu_device *adev, WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MEC_CNTL, 0); } else { WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MEC_CNTL, - (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK)); + (CP_MEC_CNTL__MEC_INVALIDATE_ICACHE_MASK | + CP_MEC_CNTL__MEC_ME1_PIPE0_RESET_MASK | + CP_MEC_CNTL__MEC_ME1_PIPE1_RESET_MASK | + CP_MEC_CNTL__MEC_ME1_PIPE2_RESET_MASK | + CP_MEC_CNTL__MEC_ME1_PIPE3_RESET_MASK | + CP_MEC_CNTL__MEC_ME2_PIPE0_RESET_MASK | + CP_MEC_CNTL__MEC_ME2_PIPE1_RESET_MASK | + CP_MEC_CNTL__MEC_ME1_HALT_MASK | + CP_MEC_CNTL__MEC_ME2_HALT_MASK)); adev->gfx.kiq[xcc_id].ring.sched.ready = false; } udelay(50); @@ -1513,9 +1800,7 @@ static void gfx_v9_4_3_xcc_kiq_setting(struct amdgpu_ring *ring, int xcc_id) tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS); tmp &= 0xffffff00; tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue); - WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp); - tmp |= 0x80; - WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp); + WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp | 0x80); } static void gfx_v9_4_3_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd) @@ -1576,6 +1861,9 @@ static int gfx_v9_4_3_xcc_mqd_init(struct amdgpu_ring *ring, int xcc_id) DOORBELL_SOURCE, 0); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_HIT, 0); + if (amdgpu_sriov_multi_vf_mode(adev)) + tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, + DOORBELL_MODE, 1); } else { tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 0); @@ -1860,7 +2148,7 @@ static int gfx_v9_4_3_xcc_kiq_init_queue(struct amdgpu_ring *ring, int xcc_id) return 0; } -static int gfx_v9_4_3_xcc_kcq_init_queue(struct amdgpu_ring *ring, int xcc_id) +static int gfx_v9_4_3_xcc_kcq_init_queue(struct amdgpu_ring *ring, int xcc_id, bool restore) { struct amdgpu_device *adev = ring->adev; struct v9_mqd *mqd = ring->mqd_ptr; @@ -1872,8 +2160,8 @@ static int gfx_v9_4_3_xcc_kcq_init_queue(struct amdgpu_ring *ring, int xcc_id) */ tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx]; - if (!tmp_mqd->cp_hqd_pq_control || - (!amdgpu_in_reset(adev) && !adev->in_suspend)) { + if (!restore && (!tmp_mqd->cp_hqd_pq_control || + (!amdgpu_in_reset(adev) && !adev->in_suspend))) { 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; @@ -1921,55 +2209,27 @@ static int gfx_v9_4_3_xcc_kcq_fini_register(struct amdgpu_device *adev, int xcc_ static int gfx_v9_4_3_xcc_kiq_resume(struct amdgpu_device *adev, int xcc_id) { - struct amdgpu_ring *ring; - int r; - - ring = &adev->gfx.kiq[xcc_id].ring; - - r = amdgpu_bo_reserve(ring->mqd_obj, false); - if (unlikely(r != 0)) - return r; - - r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); - if (unlikely(r != 0)) { - amdgpu_bo_unreserve(ring->mqd_obj); - return r; - } - - gfx_v9_4_3_xcc_kiq_init_queue(ring, xcc_id); - amdgpu_bo_kunmap(ring->mqd_obj); - ring->mqd_ptr = NULL; - amdgpu_bo_unreserve(ring->mqd_obj); + gfx_v9_4_3_xcc_kiq_init_queue(&adev->gfx.kiq[xcc_id].ring, xcc_id); return 0; } static int gfx_v9_4_3_xcc_kcq_resume(struct amdgpu_device *adev, int xcc_id) { - struct amdgpu_ring *ring = NULL; - int r = 0, i; + struct amdgpu_ring *ring; + int i, r; gfx_v9_4_3_xcc_cp_compute_enable(adev, true, xcc_id); for (i = 0; i < adev->gfx.num_compute_rings; i++) { - ring = &adev->gfx.compute_ring[i + xcc_id * adev->gfx.num_compute_rings]; - - r = amdgpu_bo_reserve(ring->mqd_obj, false); - if (unlikely(r != 0)) - goto done; - r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); - if (!r) { - r = gfx_v9_4_3_xcc_kcq_init_queue(ring, xcc_id); - amdgpu_bo_kunmap(ring->mqd_obj); - ring->mqd_ptr = NULL; - } - amdgpu_bo_unreserve(ring->mqd_obj); + ring = &adev->gfx.compute_ring[i + xcc_id * + adev->gfx.num_compute_rings]; + + r = gfx_v9_4_3_xcc_kcq_init_queue(ring, xcc_id, false); if (r) - goto done; + return r; } - r = amdgpu_gfx_enable_kcq(adev, xcc_id); -done: - return r; + return amdgpu_gfx_enable_kcq(adev, xcc_id); } static int gfx_v9_4_3_xcc_cp_resume(struct amdgpu_device *adev, int xcc_id) @@ -1985,6 +2245,8 @@ static int gfx_v9_4_3_xcc_cp_resume(struct amdgpu_device *adev, int xcc_id) r = gfx_v9_4_3_xcc_cp_compute_load_microcode(adev, xcc_id); if (r) return r; + } else { + gfx_v9_4_3_xcc_cp_compute_enable(adev, false, xcc_id); } r = gfx_v9_4_3_xcc_kiq_resume(adev, xcc_id); @@ -2010,18 +2272,31 @@ static int gfx_v9_4_3_xcc_cp_resume(struct amdgpu_device *adev, int xcc_id) static int gfx_v9_4_3_cp_resume(struct amdgpu_device *adev) { - int r = 0, i, num_xcc; + int r = 0, i, num_xcc, num_xcp, num_xcc_per_xcp; + + num_xcc = NUM_XCC(adev->gfx.xcc_mask); + if (amdgpu_sriov_vf(adev)) { + enum amdgpu_gfx_partition mode; - if (amdgpu_xcp_query_partition_mode(adev->xcp_mgr, - AMDGPU_XCP_FL_NONE) == - AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE) - r = amdgpu_xcp_switch_partition_mode(adev->xcp_mgr, - amdgpu_user_partt_mode); + mode = amdgpu_xcp_query_partition_mode(adev->xcp_mgr, + AMDGPU_XCP_FL_NONE); + if (mode == AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE) + return -EINVAL; + num_xcc_per_xcp = gfx_v9_4_3_get_xccs_per_xcp(adev); + adev->gfx.num_xcc_per_xcp = num_xcc_per_xcp; + num_xcp = num_xcc / num_xcc_per_xcp; + r = amdgpu_xcp_init(adev->xcp_mgr, num_xcp, mode); + } else { + if (amdgpu_xcp_query_partition_mode(adev->xcp_mgr, + AMDGPU_XCP_FL_NONE) == + AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE) + r = amdgpu_xcp_switch_partition_mode( + adev->xcp_mgr, amdgpu_user_partt_mode); + } if (r) return r; - num_xcc = NUM_XCC(adev->gfx.xcc_mask); for (i = 0; i < num_xcc; i++) { r = gfx_v9_4_3_xcc_cp_resume(adev, i); if (r) @@ -2031,12 +2306,6 @@ static int gfx_v9_4_3_cp_resume(struct amdgpu_device *adev) return 0; } -static void gfx_v9_4_3_xcc_cp_enable(struct amdgpu_device *adev, bool enable, - int xcc_id) -{ - gfx_v9_4_3_xcc_cp_compute_enable(adev, enable, xcc_id); -} - static void gfx_v9_4_3_xcc_fini(struct amdgpu_device *adev, int xcc_id) { if (amdgpu_gfx_disable_kcq(adev, xcc_id)) @@ -2068,13 +2337,16 @@ static void gfx_v9_4_3_xcc_fini(struct amdgpu_device *adev, int xcc_id) } gfx_v9_4_3_xcc_kcq_fini_register(adev, xcc_id); - gfx_v9_4_3_xcc_cp_enable(adev, false, xcc_id); + gfx_v9_4_3_xcc_cp_compute_enable(adev, false, xcc_id); } -static int gfx_v9_4_3_hw_init(void *handle) +static int gfx_v9_4_3_hw_init(struct amdgpu_ip_block *ip_block) { int r; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; + + amdgpu_gfx_cleaner_shader_init(adev, adev->gfx.cleaner_shader_size, + adev->gfx.cleaner_shader_ptr); if (!amdgpu_sriov_vf(adev)) gfx_v9_4_3_init_golden_registers(adev); @@ -2092,13 +2364,14 @@ static int gfx_v9_4_3_hw_init(void *handle) return r; } -static int gfx_v9_4_3_hw_fini(void *handle) +static int gfx_v9_4_3_hw_fini(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; int i, num_xcc; amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); + amdgpu_irq_put(adev, &adev->gfx.bad_op_irq, 0); num_xcc = NUM_XCC(adev->gfx.xcc_mask); for (i = 0; i < num_xcc; i++) { @@ -2108,19 +2381,19 @@ static int gfx_v9_4_3_hw_fini(void *handle) return 0; } -static int gfx_v9_4_3_suspend(void *handle) +static int gfx_v9_4_3_suspend(struct amdgpu_ip_block *ip_block) { - return gfx_v9_4_3_hw_fini(handle); + return gfx_v9_4_3_hw_fini(ip_block); } -static int gfx_v9_4_3_resume(void *handle) +static int gfx_v9_4_3_resume(struct amdgpu_ip_block *ip_block) { - return gfx_v9_4_3_hw_init(handle); + return gfx_v9_4_3_hw_init(ip_block); } -static bool gfx_v9_4_3_is_idle(void *handle) +static bool gfx_v9_4_3_is_idle(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; int i, num_xcc; num_xcc = NUM_XCC(adev->gfx.xcc_mask); @@ -2132,24 +2405,24 @@ static bool gfx_v9_4_3_is_idle(void *handle) return true; } -static int gfx_v9_4_3_wait_for_idle(void *handle) +static int gfx_v9_4_3_wait_for_idle(struct amdgpu_ip_block *ip_block) { unsigned i; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; for (i = 0; i < adev->usec_timeout; i++) { - if (gfx_v9_4_3_is_idle(handle)) + if (gfx_v9_4_3_is_idle(ip_block)) return 0; udelay(1); } return -ETIMEDOUT; } -static int gfx_v9_4_3_soft_reset(void *handle) +static int gfx_v9_4_3_soft_reset(struct amdgpu_ip_block *ip_block) { u32 grbm_soft_reset = 0; u32 tmp; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; /* GRBM_STATUS */ tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_STATUS); @@ -2233,9 +2506,9 @@ static void gfx_v9_4_3_ring_emit_gds_switch(struct amdgpu_ring *ring, (1 << (oa_size + oa_base)) - (1 << oa_base)); } -static int gfx_v9_4_3_early_init(void *handle) +static int gfx_v9_4_3_early_init(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev), AMDGPU_MAX_COMPUTE_RINGS); @@ -2251,9 +2524,9 @@ static int gfx_v9_4_3_early_init(void *handle) return gfx_v9_4_3_init_microcode(adev); } -static int gfx_v9_4_3_late_init(void *handle) +static int gfx_v9_4_3_late_init(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; int r; r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); @@ -2264,6 +2537,10 @@ static int gfx_v9_4_3_late_init(void *handle) if (r) return r; + r = amdgpu_irq_get(adev, &adev->gfx.bad_op_irq, 0); + if (r) + return r; + if (adev->gfx.ras && adev->gfx.ras->enable_watchdog_timer) adev->gfx.ras->enable_watchdog_timer(adev); @@ -2398,10 +2675,10 @@ gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(struct amdgpu_device *adev, if (def != data) WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE, data); - /* enable cgcg FSM(0x0000363F) */ + /* CGCG Hysteresis: 400us */ def = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL); - data = (0x36 + data = (0x2710 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) @@ -2410,10 +2687,10 @@ gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(struct amdgpu_device *adev, if (def != data) WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL, data); - /* set IDLE_POLL_COUNT(0x00900100) */ + /* set IDLE_POLL_COUNT(0x33450100)*/ def = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_RB_WPTR_POLL_CNTL); data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) | - (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); + (0x3345 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); if (def != data) WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_RB_WPTR_POLL_CNTL, data); } else { @@ -2478,37 +2755,32 @@ static const struct amdgpu_rlc_funcs gfx_v9_4_3_rlc_funcs = { .is_rlcg_access_range = gfx_v9_4_3_is_rlcg_access_range, }; -static int gfx_v9_4_3_set_powergating_state(void *handle, +static int gfx_v9_4_3_set_powergating_state(struct amdgpu_ip_block *ip_block, enum amd_powergating_state state) { return 0; } -static int gfx_v9_4_3_set_clockgating_state(void *handle, +static int gfx_v9_4_3_set_clockgating_state(struct amdgpu_ip_block *ip_block, enum amd_clockgating_state state) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; int i, num_xcc; if (amdgpu_sriov_vf(adev)) return 0; num_xcc = NUM_XCC(adev->gfx.xcc_mask); - switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { - case IP_VERSION(9, 4, 3): - for (i = 0; i < num_xcc; i++) - gfx_v9_4_3_xcc_update_gfx_clock_gating( - adev, state == AMD_CG_STATE_GATE, i); - break; - default: - break; - } + for (i = 0; i < num_xcc; i++) + gfx_v9_4_3_xcc_update_gfx_clock_gating( + adev, state == AMD_CG_STATE_GATE, i); + return 0; } -static void gfx_v9_4_3_get_clockgating_state(void *handle, u64 *flags) +static void gfx_v9_4_3_get_clockgating_state(struct amdgpu_ip_block *ip_block, u64 *flags) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; int data; if (amdgpu_sriov_vf(adev)) @@ -2716,6 +2988,8 @@ static void gfx_v9_4_3_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg, { struct amdgpu_device *adev = ring->adev; + reg = gfx_v9_4_3_normalize_xcc_reg_offset(reg); + amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); amdgpu_ring_write(ring, 0 | /* src: register*/ (5 << 8) | /* dst: memory */ @@ -2733,6 +3007,8 @@ static void gfx_v9_4_3_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, { uint32_t cmd = 0; + reg = gfx_v9_4_3_normalize_xcc_reg_offset(reg); + switch (ring->funcs->type) { case AMDGPU_RING_TYPE_GFX: cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM; @@ -2765,6 +3041,21 @@ static void gfx_v9_4_3_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, ref, mask); } +static void gfx_v9_4_3_ring_soft_recovery(struct amdgpu_ring *ring, + unsigned vmid) +{ + struct amdgpu_device *adev = ring->adev; + uint32_t value = 0; + + value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03); + value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); + value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); + value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); + amdgpu_gfx_rlc_enter_safe_mode(adev, ring->xcc_id); + WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regSQ_CMD, value); + amdgpu_gfx_rlc_exit_safe_mode(adev, ring->xcc_id); +} + static void gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( struct amdgpu_device *adev, int me, int pipe, enum amdgpu_interrupt_state state, int xcc_id) @@ -2818,21 +3109,103 @@ static void gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( } } +static u32 gfx_v9_4_3_get_cpc_int_cntl(struct amdgpu_device *adev, + int xcc_id, int me, int pipe) +{ + /* + * amdgpu controls only the first MEC. That's why this function only + * handles the setting of interrupts for this specific MEC. All other + * pipes' interrupts are set by amdkfd. + */ + if (me != 1) + return 0; + + switch (pipe) { + case 0: + return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE0_INT_CNTL); + case 1: + return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE1_INT_CNTL); + case 2: + return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE2_INT_CNTL); + case 3: + return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE3_INT_CNTL); + default: + return 0; + } +} + static int gfx_v9_4_3_set_priv_reg_fault_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source, unsigned type, enum amdgpu_interrupt_state state) { - int i, num_xcc; + u32 mec_int_cntl_reg, mec_int_cntl; + int i, j, k, num_xcc; num_xcc = NUM_XCC(adev->gfx.xcc_mask); switch (state) { case AMDGPU_IRQ_STATE_DISABLE: case AMDGPU_IRQ_STATE_ENABLE: - for (i = 0; i < num_xcc; i++) + for (i = 0; i < num_xcc; i++) { WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0, - PRIV_REG_INT_ENABLE, - state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + PRIV_REG_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + for (j = 0; j < adev->gfx.mec.num_mec; j++) { + for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) { + /* MECs start at 1 */ + mec_int_cntl_reg = gfx_v9_4_3_get_cpc_int_cntl(adev, i, j + 1, k); + + if (mec_int_cntl_reg) { + mec_int_cntl = RREG32_XCC(mec_int_cntl_reg, i); + mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, + PRIV_REG_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? + 1 : 0); + WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, i); + } + } + } + } + break; + default: + break; + } + + return 0; +} + +static int gfx_v9_4_3_set_bad_op_fault_state(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + unsigned type, + enum amdgpu_interrupt_state state) +{ + u32 mec_int_cntl_reg, mec_int_cntl; + int i, j, k, num_xcc; + + num_xcc = NUM_XCC(adev->gfx.xcc_mask); + switch (state) { + case AMDGPU_IRQ_STATE_DISABLE: + case AMDGPU_IRQ_STATE_ENABLE: + for (i = 0; i < num_xcc; i++) { + WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0, + OPCODE_ERROR_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + for (j = 0; j < adev->gfx.mec.num_mec; j++) { + for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) { + /* MECs start at 1 */ + mec_int_cntl_reg = gfx_v9_4_3_get_cpc_int_cntl(adev, i, j + 1, k); + + if (mec_int_cntl_reg) { + mec_int_cntl = RREG32_XCC(mec_int_cntl_reg, i); + mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, + OPCODE_ERROR_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? + 1 : 0); + WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, i); + } + } + } + } break; default: break; @@ -2993,6 +3366,15 @@ static int gfx_v9_4_3_priv_reg_irq(struct amdgpu_device *adev, return 0; } +static int gfx_v9_4_3_bad_op_irq(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + struct amdgpu_iv_entry *entry) +{ + DRM_ERROR("Illegal opcode in command stream\n"); + gfx_v9_4_3_fault(adev, entry); + return 0; +} + static int gfx_v9_4_3_priv_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) @@ -3079,6 +3461,167 @@ static void gfx_v9_4_3_emit_wave_limit(struct amdgpu_ring *ring, bool enable) } } +static int gfx_v9_4_3_unmap_done(struct amdgpu_device *adev, uint32_t me, + uint32_t pipe, uint32_t queue, + uint32_t xcc_id) +{ + int i, r; + /* make sure dequeue is complete*/ + gfx_v9_4_3_xcc_set_safe_mode(adev, xcc_id); + mutex_lock(&adev->srbm_mutex); + soc15_grbm_select(adev, me, pipe, queue, 0, GET_INST(GC, xcc_id)); + for (i = 0; i < adev->usec_timeout; i++) { + if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1)) + break; + udelay(1); + } + if (i >= adev->usec_timeout) + r = -ETIMEDOUT; + else + r = 0; + soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); + mutex_unlock(&adev->srbm_mutex); + gfx_v9_4_3_xcc_unset_safe_mode(adev, xcc_id); + + return r; + +} + +static bool gfx_v9_4_3_pipe_reset_support(struct amdgpu_device *adev) +{ + if (!!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_PIPE)) + return true; + else + dev_warn_once(adev->dev, "Please use the latest MEC version to see whether support pipe reset\n"); + + return false; +} + +static int gfx_v9_4_3_reset_hw_pipe(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + uint32_t reset_pipe, clean_pipe; + int r; + + if (!gfx_v9_4_3_pipe_reset_support(adev)) + return -EINVAL; + + gfx_v9_4_3_xcc_set_safe_mode(adev, ring->xcc_id); + mutex_lock(&adev->srbm_mutex); + + reset_pipe = RREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL); + clean_pipe = reset_pipe; + + if (ring->me == 1) { + switch (ring->pipe) { + case 0: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE0_RESET, 1); + break; + case 1: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE1_RESET, 1); + break; + case 2: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE2_RESET, 1); + break; + case 3: + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME1_PIPE3_RESET, 1); + break; + default: + break; + } + } else { + if (ring->pipe) + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME2_PIPE1_RESET, 1); + else + reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, + MEC_ME2_PIPE0_RESET, 1); + } + + WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL, reset_pipe); + WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL, clean_pipe); + mutex_unlock(&adev->srbm_mutex); + gfx_v9_4_3_xcc_unset_safe_mode(adev, ring->xcc_id); + + r = gfx_v9_4_3_unmap_done(adev, ring->me, ring->pipe, ring->queue, ring->xcc_id); + return r; +} + +static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring, + unsigned int vmid) +{ + struct amdgpu_device *adev = ring->adev; + struct amdgpu_kiq *kiq = &adev->gfx.kiq[ring->xcc_id]; + struct amdgpu_ring *kiq_ring = &kiq->ring; + unsigned long flags; + int r; + + if (amdgpu_sriov_vf(adev)) + return -EINVAL; + + if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) + return -EINVAL; + + spin_lock_irqsave(&kiq->ring_lock, flags); + + if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) { + spin_unlock_irqrestore(&kiq->ring_lock, flags); + return -ENOMEM; + } + + kiq->pmf->kiq_unmap_queues(kiq_ring, ring, RESET_QUEUES, + 0, 0); + amdgpu_ring_commit(kiq_ring); + + spin_unlock_irqrestore(&kiq->ring_lock, flags); + + r = amdgpu_ring_test_ring(kiq_ring); + if (r) { + dev_err(adev->dev, "kiq ring test failed after ring: %s queue reset\n", + ring->name); + goto pipe_reset; + } + + r = gfx_v9_4_3_unmap_done(adev, ring->me, ring->pipe, ring->queue, ring->xcc_id); + if (r) + dev_err(adev->dev, "fail to wait on hqd deactive and will try pipe reset\n"); + +pipe_reset: + if(r) { + r = gfx_v9_4_3_reset_hw_pipe(ring); + dev_info(adev->dev, "ring: %s pipe reset :%s\n", ring->name, + r ? "failed" : "successfully"); + if (r) + return r; + } + + r = gfx_v9_4_3_xcc_kcq_init_queue(ring, ring->xcc_id, true); + if (r) { + dev_err(adev->dev, "fail to init kcq\n"); + return r; + } + spin_lock_irqsave(&kiq->ring_lock, flags); + r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size); + if (r) { + spin_unlock_irqrestore(&kiq->ring_lock, flags); + return -ENOMEM; + } + kiq->pmf->kiq_map_queues(kiq_ring, ring); + amdgpu_ring_commit(kiq_ring); + spin_unlock_irqrestore(&kiq->ring_lock, flags); + + r = amdgpu_ring_test_ring(kiq_ring); + if (r) { + dev_err(adev->dev, "fail to remap queue\n"); + return r; + } + return amdgpu_ring_test_ring(ring); +} + enum amdgpu_gfx_cp_ras_mem_id { AMDGPU_GFX_CP_MEM1 = 1, AMDGPU_GFX_CP_MEM2, @@ -3891,8 +4434,8 @@ static void gfx_v9_4_3_inst_query_ras_err_count(struct amdgpu_device *adev, /* the caller should make sure initialize value of * err_data->ue_count and err_data->ce_count */ - amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, NULL, ue_count); - amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, NULL, ce_count); + amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count); + amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, ce_count); } static void gfx_v9_4_3_inst_reset_ras_err_count(struct amdgpu_device *adev, @@ -3994,6 +4537,161 @@ static void gfx_v9_4_3_enable_watchdog_timer(struct amdgpu_device *adev) amdgpu_gfx_ras_error_func(adev, NULL, gfx_v9_4_3_inst_enable_watchdog_timer); } +static void gfx_v9_4_3_ring_insert_nop(struct amdgpu_ring *ring, uint32_t num_nop) +{ + /* Header itself is a NOP packet */ + if (num_nop == 1) { + amdgpu_ring_write(ring, ring->funcs->nop); + return; + } + + /* Max HW optimization till 0x3ffe, followed by remaining one NOP at a time*/ + amdgpu_ring_write(ring, PACKET3(PACKET3_NOP, min(num_nop - 2, 0x3ffe))); + + /* Header is at index 0, followed by num_nops - 1 NOP packet's */ + amdgpu_ring_insert_nop(ring, num_nop - 1); +} + +static void gfx_v9_4_3_ip_print(struct amdgpu_ip_block *ip_block, struct drm_printer *p) +{ + struct amdgpu_device *adev = ip_block->adev; + uint32_t i, j, k; + uint32_t xcc_id, xcc_offset, inst_offset; + uint32_t num_xcc, reg, num_inst; + uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9_4_3); + + if (!adev->gfx.ip_dump_core) + return; + + num_xcc = NUM_XCC(adev->gfx.xcc_mask); + drm_printf(p, "Number of Instances:%d\n", num_xcc); + for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { + xcc_offset = xcc_id * reg_count; + drm_printf(p, "\nInstance id:%d\n", xcc_id); + for (i = 0; i < reg_count; i++) + drm_printf(p, "%-50s \t 0x%08x\n", + gc_reg_list_9_4_3[i].reg_name, + adev->gfx.ip_dump_core[xcc_offset + i]); + } + + /* print compute queue registers for all instances */ + if (!adev->gfx.ip_dump_compute_queues) + return; + + num_inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec * + adev->gfx.mec.num_queue_per_pipe; + + reg_count = ARRAY_SIZE(gc_cp_reg_list_9_4_3); + drm_printf(p, "\nnum_xcc: %d num_mec: %d num_pipe: %d num_queue: %d\n", + num_xcc, + adev->gfx.mec.num_mec, + adev->gfx.mec.num_pipe_per_mec, + adev->gfx.mec.num_queue_per_pipe); + + for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { + xcc_offset = xcc_id * reg_count * num_inst; + inst_offset = 0; + for (i = 0; i < adev->gfx.mec.num_mec; i++) { + for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) { + for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) { + drm_printf(p, + "\nxcc:%d mec:%d, pipe:%d, queue:%d\n", + xcc_id, i, j, k); + for (reg = 0; reg < reg_count; reg++) { + if (i && gc_cp_reg_list_9_4_3[reg].reg_offset == + regCP_MEC_ME1_HEADER_DUMP) + drm_printf(p, + "%-50s \t 0x%08x\n", + "regCP_MEC_ME2_HEADER_DUMP", + adev->gfx.ip_dump_compute_queues + [xcc_offset + inst_offset + + reg]); + else + drm_printf(p, + "%-50s \t 0x%08x\n", + gc_cp_reg_list_9_4_3[reg].reg_name, + adev->gfx.ip_dump_compute_queues + [xcc_offset + inst_offset + + reg]); + } + inst_offset += reg_count; + } + } + } + } +} + +static void gfx_v9_4_3_ip_dump(struct amdgpu_ip_block *ip_block) +{ + struct amdgpu_device *adev = ip_block->adev; + uint32_t i, j, k; + uint32_t num_xcc, reg, num_inst; + uint32_t xcc_id, xcc_offset, inst_offset; + uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9_4_3); + + if (!adev->gfx.ip_dump_core) + return; + + num_xcc = NUM_XCC(adev->gfx.xcc_mask); + + for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { + xcc_offset = xcc_id * reg_count; + for (i = 0; i < reg_count; i++) + adev->gfx.ip_dump_core[xcc_offset + i] = + RREG32(SOC15_REG_ENTRY_OFFSET_INST(gc_reg_list_9_4_3[i], + GET_INST(GC, xcc_id))); + } + + /* dump compute queue registers for all instances */ + if (!adev->gfx.ip_dump_compute_queues) + return; + + num_inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec * + adev->gfx.mec.num_queue_per_pipe; + reg_count = ARRAY_SIZE(gc_cp_reg_list_9_4_3); + mutex_lock(&adev->srbm_mutex); + for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { + xcc_offset = xcc_id * reg_count * num_inst; + inst_offset = 0; + for (i = 0; i < adev->gfx.mec.num_mec; i++) { + for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) { + for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) { + /* ME0 is for GFX so start from 1 for CP */ + soc15_grbm_select(adev, 1 + i, j, k, 0, + GET_INST(GC, xcc_id)); + + for (reg = 0; reg < reg_count; reg++) { + if (i && gc_cp_reg_list_9_4_3[reg].reg_offset == + regCP_MEC_ME1_HEADER_DUMP) + adev->gfx.ip_dump_compute_queues + [xcc_offset + + inst_offset + reg] = + RREG32(SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), + regCP_MEC_ME2_HEADER_DUMP)); + else + adev->gfx.ip_dump_compute_queues + [xcc_offset + + inst_offset + reg] = + RREG32(SOC15_REG_ENTRY_OFFSET_INST( + gc_cp_reg_list_9_4_3[reg], + GET_INST(GC, xcc_id))); + } + inst_offset += reg_count; + } + } + } + } + soc15_grbm_select(adev, 0, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); +} + +static void gfx_v9_4_3_ring_emit_cleaner_shader(struct amdgpu_ring *ring) +{ + /* Emit the cleaner shader */ + amdgpu_ring_write(ring, PACKET3(PACKET3_RUN_CLEANER_SHADER, 0)); + amdgpu_ring_write(ring, 0); /* RESERVED field, programmed to zero */ +} + static const struct amd_ip_funcs gfx_v9_4_3_ip_funcs = { .name = "gfx_v9_4_3", .early_init = gfx_v9_4_3_early_init, @@ -4010,6 +4708,8 @@ static const struct amd_ip_funcs gfx_v9_4_3_ip_funcs = { .set_clockgating_state = gfx_v9_4_3_set_clockgating_state, .set_powergating_state = gfx_v9_4_3_set_powergating_state, .get_clockgating_state = gfx_v9_4_3_get_clockgating_state, + .dump_ip_state = gfx_v9_4_3_ip_dump, + .print_ip_state = gfx_v9_4_3_ip_print, }; static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_compute = { @@ -4031,7 +4731,8 @@ static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_compute = { 8 + 8 + 8 + /* gfx_v9_4_3_ring_emit_fence x3 for user fence, vm fence */ 7 + /* gfx_v9_4_3_emit_mem_sync */ 5 + /* gfx_v9_4_3_emit_wave_limit for updating regSPI_WCL_PIPE_PERCENT_GFX register */ - 15, /* for updating 3 regSPI_WCL_PIPE_PERCENT_CS registers */ + 15 + /* for updating 3 regSPI_WCL_PIPE_PERCENT_CS registers */ + 2, /* gfx_v9_4_3_ring_emit_cleaner_shader */ .emit_ib_size = 7, /* gfx_v9_4_3_ring_emit_ib_compute */ .emit_ib = gfx_v9_4_3_ring_emit_ib_compute, .emit_fence = gfx_v9_4_3_ring_emit_fence, @@ -4041,13 +4742,18 @@ static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_compute = { .emit_hdp_flush = gfx_v9_4_3_ring_emit_hdp_flush, .test_ring = gfx_v9_4_3_ring_test_ring, .test_ib = gfx_v9_4_3_ring_test_ib, - .insert_nop = amdgpu_ring_insert_nop, + .insert_nop = gfx_v9_4_3_ring_insert_nop, .pad_ib = amdgpu_ring_generic_pad_ib, .emit_wreg = gfx_v9_4_3_ring_emit_wreg, .emit_reg_wait = gfx_v9_4_3_ring_emit_reg_wait, .emit_reg_write_reg_wait = gfx_v9_4_3_ring_emit_reg_write_reg_wait, + .soft_recovery = gfx_v9_4_3_ring_soft_recovery, .emit_mem_sync = gfx_v9_4_3_emit_mem_sync, .emit_wave_limit = gfx_v9_4_3_emit_wave_limit, + .reset = gfx_v9_4_3_reset_kcq, + .emit_cleaner_shader = gfx_v9_4_3_ring_emit_cleaner_shader, + .begin_use = amdgpu_gfx_enforce_isolation_ring_begin_use, + .end_use = amdgpu_gfx_enforce_isolation_ring_end_use, }; static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_kiq = { @@ -4102,6 +4808,11 @@ static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_reg_irq_funcs = { .process = gfx_v9_4_3_priv_reg_irq, }; +static const struct amdgpu_irq_src_funcs gfx_v9_4_3_bad_op_irq_funcs = { + .set = gfx_v9_4_3_set_bad_op_fault_state, + .process = gfx_v9_4_3_bad_op_irq, +}; + static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_inst_irq_funcs = { .set = gfx_v9_4_3_set_priv_inst_fault_state, .process = gfx_v9_4_3_priv_inst_irq, @@ -4115,6 +4826,9 @@ static void gfx_v9_4_3_set_irq_funcs(struct amdgpu_device *adev) adev->gfx.priv_reg_irq.num_types = 1; adev->gfx.priv_reg_irq.funcs = &gfx_v9_4_3_priv_reg_irq_funcs; + adev->gfx.bad_op_irq.num_types = 1; + adev->gfx.bad_op_irq.funcs = &gfx_v9_4_3_bad_op_irq_funcs; + adev->gfx.priv_inst_irq.num_types = 1; adev->gfx.priv_inst_irq.funcs = &gfx_v9_4_3_priv_inst_irq_funcs; } @@ -4127,30 +4841,13 @@ static void gfx_v9_4_3_set_rlc_funcs(struct amdgpu_device *adev) static void gfx_v9_4_3_set_gds_init(struct amdgpu_device *adev) { - /* init asci gds info */ - switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { - case IP_VERSION(9, 4, 3): - /* 9.4.3 removed all the GDS internal memory, - * only support GWS opcode in kernel, like barrier - * semaphore.etc */ - adev->gds.gds_size = 0; - break; - default: - adev->gds.gds_size = 0x10000; - break; - } - - switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { - case IP_VERSION(9, 4, 3): - /* deprecated for 9.4.3, no usage at all */ - adev->gds.gds_compute_max_wave_id = 0; - break; - default: - /* this really depends on the chip */ - adev->gds.gds_compute_max_wave_id = 0x7ff; - break; - } + /* 9.4.3 variants removed all the GDS internal memory, + * only support GWS opcode in kernel, like barrier + * semaphore.etc */ + /* init asic gds info */ + adev->gds.gds_size = 0; + adev->gds.gds_compute_max_wave_id = 0; adev->gds.gws_size = 64; adev->gds.oa_size = 16; } @@ -4187,9 +4884,10 @@ static u32 gfx_v9_4_3_get_cu_active_bitmap(struct amdgpu_device *adev, int xcc_i static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev, struct amdgpu_cu_info *cu_info) { - int i, j, k, counter, xcc_id, active_cu_number = 0; - u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; + int i, j, k, prev_counter, counter, xcc_id, active_cu_number = 0; + u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0, tmp; unsigned disable_masks[4 * 4]; + bool is_symmetric_cus; if (!adev || !cu_info) return -EINVAL; @@ -4207,6 +4905,7 @@ static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev, mutex_lock(&adev->grbm_idx_mutex); for (xcc_id = 0; xcc_id < NUM_XCC(adev->gfx.xcc_mask); xcc_id++) { + is_symmetric_cus = true; for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { mask = 1; @@ -4234,6 +4933,15 @@ static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev, ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); cu_info->ao_cu_bitmap[i][j] = ao_bitmap; } + if (i && is_symmetric_cus && prev_counter != counter) + is_symmetric_cus = false; + prev_counter = counter; + } + if (is_symmetric_cus) { + tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_DEBUG); + tmp = REG_SET_FIELD(tmp, CP_CPC_DEBUG, CPC_HARVESTING_RELAUNCH_DISABLE, 1); + tmp = REG_SET_FIELD(tmp, CP_CPC_DEBUG, CPC_HARVESTING_DISPATCH_DISABLE, 1); + WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_DEBUG, tmp); } gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, xcc_id); |