Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | |
| 3 | /* |
| 4 | * Copyright 2016-2020 HabanaLabs, Ltd. |
| 5 | * All Rights Reserved. |
| 6 | */ |
| 7 | |
| 8 | #include "gaudiP.h" |
Greg Kroah-Hartman | 7b16a15 | 2020-07-28 19:18:51 +0200 | [diff] [blame] | 9 | #include "../include/hw_ip/mmu/mmu_general.h" |
| 10 | #include "../include/hw_ip/mmu/mmu_v1_1.h" |
| 11 | #include "../include/gaudi/gaudi_masks.h" |
| 12 | #include "../include/gaudi/gaudi_fw_if.h" |
| 13 | #include "../include/gaudi/gaudi_reg_map.h" |
| 14 | #include "../include/gaudi/gaudi_async_ids_map_extended.h" |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 15 | |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/pci.h> |
| 18 | #include <linux/firmware.h> |
| 19 | #include <linux/hwmon.h> |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 20 | #include <linux/iommu.h> |
| 21 | #include <linux/seq_file.h> |
| 22 | |
| 23 | /* |
| 24 | * Gaudi security scheme: |
| 25 | * |
| 26 | * 1. Host is protected by: |
| 27 | * - Range registers |
| 28 | * - MMU |
| 29 | * |
| 30 | * 2. DDR is protected by: |
| 31 | * - Range registers (protect the first 512MB) |
| 32 | * |
| 33 | * 3. Configuration is protected by: |
| 34 | * - Range registers |
| 35 | * - Protection bits |
| 36 | * |
| 37 | * MMU is always enabled. |
| 38 | * |
Ofir Bitton | 0940cab | 2020-08-31 08:52:56 +0300 | [diff] [blame] | 39 | * QMAN DMA channels 0,1 (PCI DMAN): |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 40 | * - DMA is not secured. |
| 41 | * - PQ and CQ are secured. |
| 42 | * - CP is secured: The driver needs to parse CB but WREG should be allowed |
| 43 | * because of TDMA (tensor DMA). Hence, WREG is always not |
| 44 | * secured. |
| 45 | * |
| 46 | * When the driver needs to use DMA it will check that Gaudi is idle, set DMA |
| 47 | * channel 0 to be secured, execute the DMA and change it back to not secured. |
| 48 | * Currently, the driver doesn't use the DMA while there are compute jobs |
| 49 | * running. |
| 50 | * |
| 51 | * The current use cases for the driver to use the DMA are: |
| 52 | * - Clear SRAM on context switch (happens on context switch when device is |
| 53 | * idle) |
| 54 | * - MMU page tables area clear (happens on init) |
| 55 | * |
Ofir Bitton | 0940cab | 2020-08-31 08:52:56 +0300 | [diff] [blame] | 56 | * QMAN DMA 2-7, TPC, MME, NIC: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 57 | * PQ is secured and is located on the Host (HBM CON TPC3 bug) |
| 58 | * CQ, CP and the engine are not secured |
| 59 | * |
| 60 | */ |
| 61 | |
| 62 | #define GAUDI_BOOT_FIT_FILE "habanalabs/gaudi/gaudi-boot-fit.itb" |
| 63 | #define GAUDI_LINUX_FW_FILE "habanalabs/gaudi/gaudi-fit.itb" |
| 64 | #define GAUDI_TPC_FW_FILE "habanalabs/gaudi/gaudi_tpc.bin" |
| 65 | |
| 66 | #define GAUDI_DMA_POOL_BLK_SIZE 0x100 /* 256 bytes */ |
| 67 | |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 68 | #define GAUDI_RESET_TIMEOUT_MSEC 2000 /* 2000ms */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 69 | #define GAUDI_RESET_WAIT_MSEC 1 /* 1ms */ |
| 70 | #define GAUDI_CPU_RESET_WAIT_MSEC 200 /* 200ms */ |
| 71 | #define GAUDI_TEST_QUEUE_WAIT_USEC 100000 /* 100ms */ |
| 72 | |
| 73 | #define GAUDI_PLDM_RESET_WAIT_MSEC 1000 /* 1s */ |
| 74 | #define GAUDI_PLDM_HRESET_TIMEOUT_MSEC 20000 /* 20s */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 75 | #define GAUDI_PLDM_TEST_QUEUE_WAIT_USEC 1000000 /* 1s */ |
| 76 | #define GAUDI_PLDM_MMU_TIMEOUT_USEC (MMU_CONFIG_TIMEOUT_USEC * 100) |
| 77 | #define GAUDI_PLDM_QMAN0_TIMEOUT_USEC (HL_DEVICE_TIMEOUT_USEC * 30) |
| 78 | #define GAUDI_PLDM_TPC_KERNEL_WAIT_USEC (HL_DEVICE_TIMEOUT_USEC * 30) |
| 79 | #define GAUDI_BOOT_FIT_REQ_TIMEOUT_USEC 1000000 /* 1s */ |
Oded Gabbay | 788cacf | 2020-07-07 17:30:13 +0300 | [diff] [blame] | 80 | #define GAUDI_MSG_TO_CPU_TIMEOUT_USEC 4000000 /* 4s */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 81 | |
| 82 | #define GAUDI_QMAN0_FENCE_VAL 0x72E91AB9 |
| 83 | |
| 84 | #define GAUDI_MAX_STRING_LEN 20 |
| 85 | |
| 86 | #define GAUDI_CB_POOL_CB_CNT 512 |
| 87 | #define GAUDI_CB_POOL_CB_SIZE 0x20000 /* 128KB */ |
| 88 | |
| 89 | #define GAUDI_ALLOC_CPU_MEM_RETRY_CNT 3 |
| 90 | |
| 91 | #define GAUDI_NUM_OF_TPC_INTR_CAUSE 20 |
| 92 | |
| 93 | #define GAUDI_NUM_OF_QM_ERR_CAUSE 16 |
| 94 | |
| 95 | #define GAUDI_NUM_OF_QM_ARB_ERR_CAUSE 3 |
| 96 | |
Oded Gabbay | 647e835 | 2020-06-07 11:26:48 +0300 | [diff] [blame] | 97 | #define GAUDI_ARB_WDT_TIMEOUT 0x1000000 |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 98 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 99 | #define GAUDI_CLK_GATE_DEBUGFS_MASK (\ |
| 100 | BIT(GAUDI_ENGINE_ID_MME_0) |\ |
| 101 | BIT(GAUDI_ENGINE_ID_MME_2) |\ |
| 102 | GENMASK_ULL(GAUDI_ENGINE_ID_TPC_7, GAUDI_ENGINE_ID_TPC_0)) |
| 103 | |
farah kassabri | 03df136 | 2020-05-06 11:17:38 +0300 | [diff] [blame] | 104 | #define HBM_SCRUBBING_TIMEOUT_US 1000000 /* 1s */ |
| 105 | |
Alon Mizrahi | 4147864 | 2020-11-17 14:25:14 +0200 | [diff] [blame] | 106 | #define GAUDI_PLL_MAX 10 |
| 107 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 108 | static const char gaudi_irq_name[GAUDI_MSI_ENTRIES][GAUDI_MAX_STRING_LEN] = { |
| 109 | "gaudi cq 0_0", "gaudi cq 0_1", "gaudi cq 0_2", "gaudi cq 0_3", |
| 110 | "gaudi cq 1_0", "gaudi cq 1_1", "gaudi cq 1_2", "gaudi cq 1_3", |
| 111 | "gaudi cq 5_0", "gaudi cq 5_1", "gaudi cq 5_2", "gaudi cq 5_3", |
| 112 | "gaudi cpu eq" |
| 113 | }; |
| 114 | |
| 115 | static const u8 gaudi_dma_assignment[GAUDI_DMA_MAX] = { |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 116 | [GAUDI_PCI_DMA_1] = GAUDI_ENGINE_ID_DMA_0, |
| 117 | [GAUDI_PCI_DMA_2] = GAUDI_ENGINE_ID_DMA_1, |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 118 | [GAUDI_HBM_DMA_1] = GAUDI_ENGINE_ID_DMA_2, |
| 119 | [GAUDI_HBM_DMA_2] = GAUDI_ENGINE_ID_DMA_3, |
| 120 | [GAUDI_HBM_DMA_3] = GAUDI_ENGINE_ID_DMA_4, |
Ofir Bitton | 0940cab | 2020-08-31 08:52:56 +0300 | [diff] [blame] | 121 | [GAUDI_HBM_DMA_4] = GAUDI_ENGINE_ID_DMA_5, |
| 122 | [GAUDI_HBM_DMA_5] = GAUDI_ENGINE_ID_DMA_6, |
| 123 | [GAUDI_HBM_DMA_6] = GAUDI_ENGINE_ID_DMA_7 |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 124 | }; |
| 125 | |
| 126 | static const u8 gaudi_cq_assignment[NUMBER_OF_CMPLT_QUEUES] = { |
| 127 | [0] = GAUDI_QUEUE_ID_DMA_0_0, |
| 128 | [1] = GAUDI_QUEUE_ID_DMA_0_1, |
| 129 | [2] = GAUDI_QUEUE_ID_DMA_0_2, |
| 130 | [3] = GAUDI_QUEUE_ID_DMA_0_3, |
| 131 | [4] = GAUDI_QUEUE_ID_DMA_1_0, |
| 132 | [5] = GAUDI_QUEUE_ID_DMA_1_1, |
| 133 | [6] = GAUDI_QUEUE_ID_DMA_1_2, |
| 134 | [7] = GAUDI_QUEUE_ID_DMA_1_3, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 135 | }; |
| 136 | |
| 137 | static const u16 gaudi_packet_sizes[MAX_PACKET_ID] = { |
| 138 | [PACKET_WREG_32] = sizeof(struct packet_wreg32), |
| 139 | [PACKET_WREG_BULK] = sizeof(struct packet_wreg_bulk), |
| 140 | [PACKET_MSG_LONG] = sizeof(struct packet_msg_long), |
| 141 | [PACKET_MSG_SHORT] = sizeof(struct packet_msg_short), |
| 142 | [PACKET_CP_DMA] = sizeof(struct packet_cp_dma), |
| 143 | [PACKET_REPEAT] = sizeof(struct packet_repeat), |
| 144 | [PACKET_MSG_PROT] = sizeof(struct packet_msg_prot), |
| 145 | [PACKET_FENCE] = sizeof(struct packet_fence), |
| 146 | [PACKET_LIN_DMA] = sizeof(struct packet_lin_dma), |
| 147 | [PACKET_NOP] = sizeof(struct packet_nop), |
| 148 | [PACKET_STOP] = sizeof(struct packet_stop), |
| 149 | [PACKET_ARB_POINT] = sizeof(struct packet_arb_point), |
| 150 | [PACKET_WAIT] = sizeof(struct packet_wait), |
| 151 | [PACKET_LOAD_AND_EXE] = sizeof(struct packet_load_and_exe) |
| 152 | }; |
| 153 | |
Ofir Bitton | bc75be2 | 2020-07-30 14:56:38 +0300 | [diff] [blame] | 154 | static inline bool validate_packet_id(enum packet_id id) |
| 155 | { |
| 156 | switch (id) { |
| 157 | case PACKET_WREG_32: |
| 158 | case PACKET_WREG_BULK: |
| 159 | case PACKET_MSG_LONG: |
| 160 | case PACKET_MSG_SHORT: |
| 161 | case PACKET_CP_DMA: |
| 162 | case PACKET_REPEAT: |
| 163 | case PACKET_MSG_PROT: |
| 164 | case PACKET_FENCE: |
| 165 | case PACKET_LIN_DMA: |
| 166 | case PACKET_NOP: |
| 167 | case PACKET_STOP: |
| 168 | case PACKET_ARB_POINT: |
| 169 | case PACKET_WAIT: |
| 170 | case PACKET_LOAD_AND_EXE: |
| 171 | return true; |
| 172 | default: |
| 173 | return false; |
| 174 | } |
| 175 | } |
| 176 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 177 | static const char * const |
| 178 | gaudi_tpc_interrupts_cause[GAUDI_NUM_OF_TPC_INTR_CAUSE] = { |
| 179 | "tpc_address_exceed_slm", |
| 180 | "tpc_div_by_0", |
| 181 | "tpc_spu_mac_overflow", |
| 182 | "tpc_spu_addsub_overflow", |
| 183 | "tpc_spu_abs_overflow", |
| 184 | "tpc_spu_fp_dst_nan_inf", |
| 185 | "tpc_spu_fp_dst_denorm", |
| 186 | "tpc_vpu_mac_overflow", |
| 187 | "tpc_vpu_addsub_overflow", |
| 188 | "tpc_vpu_abs_overflow", |
| 189 | "tpc_vpu_fp_dst_nan_inf", |
| 190 | "tpc_vpu_fp_dst_denorm", |
| 191 | "tpc_assertions", |
| 192 | "tpc_illegal_instruction", |
| 193 | "tpc_pc_wrap_around", |
| 194 | "tpc_qm_sw_err", |
| 195 | "tpc_hbw_rresp_err", |
| 196 | "tpc_hbw_bresp_err", |
| 197 | "tpc_lbw_rresp_err", |
| 198 | "tpc_lbw_bresp_err" |
| 199 | }; |
| 200 | |
| 201 | static const char * const |
| 202 | gaudi_qman_error_cause[GAUDI_NUM_OF_QM_ERR_CAUSE] = { |
| 203 | "PQ AXI HBW error", |
| 204 | "CQ AXI HBW error", |
| 205 | "CP AXI HBW error", |
| 206 | "CP error due to undefined OPCODE", |
| 207 | "CP encountered STOP OPCODE", |
| 208 | "CP AXI LBW error", |
| 209 | "CP WRREG32 or WRBULK returned error", |
| 210 | "N/A", |
| 211 | "FENCE 0 inc over max value and clipped", |
| 212 | "FENCE 1 inc over max value and clipped", |
| 213 | "FENCE 2 inc over max value and clipped", |
| 214 | "FENCE 3 inc over max value and clipped", |
| 215 | "FENCE 0 dec under min value and clipped", |
| 216 | "FENCE 1 dec under min value and clipped", |
| 217 | "FENCE 2 dec under min value and clipped", |
| 218 | "FENCE 3 dec under min value and clipped" |
| 219 | }; |
| 220 | |
| 221 | static const char * const |
| 222 | gaudi_qman_arb_error_cause[GAUDI_NUM_OF_QM_ARB_ERR_CAUSE] = { |
| 223 | "Choice push while full error", |
| 224 | "Choice Q watchdog error", |
| 225 | "MSG AXI LBW returned with error" |
| 226 | }; |
| 227 | |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 228 | enum gaudi_sm_sei_cause { |
| 229 | GAUDI_SM_SEI_SO_OVERFLOW, |
| 230 | GAUDI_SM_SEI_LBW_4B_UNALIGNED, |
| 231 | GAUDI_SM_SEI_AXI_RESPONSE_ERR |
| 232 | }; |
| 233 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 234 | static enum hl_queue_type gaudi_queue_type[GAUDI_QUEUE_ID_SIZE] = { |
| 235 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_0_0 */ |
| 236 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_0_1 */ |
| 237 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_0_2 */ |
| 238 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_0_3 */ |
| 239 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_1_0 */ |
| 240 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_1_1 */ |
| 241 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_1_2 */ |
| 242 | QUEUE_TYPE_EXT, /* GAUDI_QUEUE_ID_DMA_1_3 */ |
| 243 | QUEUE_TYPE_CPU, /* GAUDI_QUEUE_ID_CPU_PQ */ |
| 244 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_2_0 */ |
| 245 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_2_1 */ |
| 246 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_2_2 */ |
| 247 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_2_3 */ |
| 248 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_3_0 */ |
| 249 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_3_1 */ |
| 250 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_3_2 */ |
| 251 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_3_3 */ |
| 252 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_4_0 */ |
| 253 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_4_1 */ |
| 254 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_4_2 */ |
| 255 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_4_3 */ |
Ofir Bitton | 0940cab | 2020-08-31 08:52:56 +0300 | [diff] [blame] | 256 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_5_0 */ |
| 257 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_5_1 */ |
| 258 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_5_2 */ |
| 259 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_5_3 */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 260 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_6_0 */ |
| 261 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_6_1 */ |
| 262 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_6_2 */ |
| 263 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_6_3 */ |
| 264 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_7_0 */ |
| 265 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_7_1 */ |
| 266 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_7_2 */ |
| 267 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_DMA_7_3 */ |
| 268 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_0_0 */ |
| 269 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_0_1 */ |
| 270 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_0_2 */ |
| 271 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_0_3 */ |
| 272 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_1_0 */ |
| 273 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_1_1 */ |
| 274 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_1_2 */ |
| 275 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_MME_1_3 */ |
| 276 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_0_0 */ |
| 277 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_0_1 */ |
| 278 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_0_2 */ |
| 279 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_0_3 */ |
| 280 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_1_0 */ |
| 281 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_1_1 */ |
| 282 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_1_2 */ |
| 283 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_1_3 */ |
| 284 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_2_0 */ |
| 285 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_2_1 */ |
| 286 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_2_2 */ |
| 287 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_2_3 */ |
| 288 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_3_0 */ |
| 289 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_3_1 */ |
| 290 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_3_2 */ |
| 291 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_3_3 */ |
| 292 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_4_0 */ |
| 293 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_4_1 */ |
| 294 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_4_2 */ |
| 295 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_4_3 */ |
| 296 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_5_0 */ |
| 297 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_5_1 */ |
| 298 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_5_2 */ |
| 299 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_5_3 */ |
| 300 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_6_0 */ |
| 301 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_6_1 */ |
| 302 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_6_2 */ |
| 303 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_6_3 */ |
| 304 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_7_0 */ |
| 305 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_7_1 */ |
| 306 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_7_2 */ |
| 307 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_TPC_7_3 */ |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 308 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_0_0 */ |
| 309 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_0_1 */ |
| 310 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_0_2 */ |
| 311 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_0_3 */ |
| 312 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_1_0 */ |
| 313 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_1_1 */ |
| 314 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_1_2 */ |
| 315 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_1_3 */ |
| 316 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_2_0 */ |
| 317 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_2_1 */ |
| 318 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_2_2 */ |
| 319 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_2_3 */ |
| 320 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_3_0 */ |
| 321 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_3_1 */ |
| 322 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_3_2 */ |
| 323 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_3_3 */ |
| 324 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_4_0 */ |
| 325 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_4_1 */ |
| 326 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_4_2 */ |
| 327 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_4_3 */ |
| 328 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_5_0 */ |
| 329 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_5_1 */ |
| 330 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_5_2 */ |
| 331 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_5_3 */ |
| 332 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_6_0 */ |
| 333 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_6_1 */ |
| 334 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_6_2 */ |
| 335 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_6_3 */ |
| 336 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_7_0 */ |
| 337 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_7_1 */ |
| 338 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_7_2 */ |
| 339 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_7_3 */ |
| 340 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_8_0 */ |
| 341 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_8_1 */ |
| 342 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_8_2 */ |
| 343 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_8_3 */ |
| 344 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_9_0 */ |
| 345 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_9_1 */ |
| 346 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_9_2 */ |
| 347 | QUEUE_TYPE_INT, /* GAUDI_QUEUE_ID_NIC_9_3 */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 348 | }; |
| 349 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 350 | struct ecc_info_extract_params { |
| 351 | u64 block_address; |
| 352 | u32 num_memories; |
| 353 | bool derr; |
| 354 | bool disable_clock_gating; |
| 355 | }; |
| 356 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 357 | static int gaudi_mmu_update_asid_hop0_addr(struct hl_device *hdev, u32 asid, |
| 358 | u64 phys_addr); |
| 359 | static int gaudi_send_job_on_qman0(struct hl_device *hdev, |
| 360 | struct hl_cs_job *job); |
| 361 | static int gaudi_memset_device_memory(struct hl_device *hdev, u64 addr, |
| 362 | u32 size, u64 val); |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 363 | static int gaudi_memset_registers(struct hl_device *hdev, u64 reg_base, |
| 364 | u32 num_regs, u32 val); |
| 365 | static int gaudi_schedule_register_memset(struct hl_device *hdev, |
| 366 | u32 hw_queue_id, u64 reg_base, u32 num_regs, u32 val); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 367 | static int gaudi_run_tpc_kernel(struct hl_device *hdev, u64 tpc_kernel, |
| 368 | u32 tpc_id); |
| 369 | static int gaudi_mmu_clear_pgt_range(struct hl_device *hdev); |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 370 | static int gaudi_cpucp_info_get(struct hl_device *hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 371 | static void gaudi_disable_clock_gating(struct hl_device *hdev); |
| 372 | static void gaudi_mmu_prepare(struct hl_device *hdev, u32 asid); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 373 | static u32 gaudi_gen_signal_cb(struct hl_device *hdev, void *data, u16 sob_id, |
Alon Mizrahi | 72ab9ca5 | 2020-12-02 19:55:30 +0200 | [diff] [blame] | 374 | u32 size, bool eb); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 375 | static u32 gaudi_gen_wait_cb(struct hl_device *hdev, |
| 376 | struct hl_gen_wait_properties *prop); |
| 377 | |
| 378 | static inline enum hl_collective_mode |
| 379 | get_collective_mode(struct hl_device *hdev, u32 queue_id) |
| 380 | { |
| 381 | if (gaudi_queue_type[queue_id] == QUEUE_TYPE_EXT) |
| 382 | return HL_COLLECTIVE_MASTER; |
| 383 | |
| 384 | if (queue_id >= GAUDI_QUEUE_ID_DMA_5_0 && |
| 385 | queue_id <= GAUDI_QUEUE_ID_DMA_5_3) |
| 386 | return HL_COLLECTIVE_SLAVE; |
| 387 | |
| 388 | if (queue_id >= GAUDI_QUEUE_ID_TPC_7_0 && |
| 389 | queue_id <= GAUDI_QUEUE_ID_TPC_7_3) |
| 390 | return HL_COLLECTIVE_SLAVE; |
| 391 | |
| 392 | if (queue_id >= GAUDI_QUEUE_ID_NIC_0_0 && |
| 393 | queue_id <= GAUDI_QUEUE_ID_NIC_9_3) |
| 394 | return HL_COLLECTIVE_SLAVE; |
| 395 | |
| 396 | return HL_COLLECTIVE_NOT_SUPPORTED; |
| 397 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 398 | |
Koby Elbaz | cd5def8 | 2021-02-23 21:31:27 +0200 | [diff] [blame] | 399 | static inline void set_default_power_values(struct hl_device *hdev) |
| 400 | { |
| 401 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 402 | |
| 403 | if (hdev->card_type == cpucp_card_type_pmc) { |
| 404 | prop->max_power_default = MAX_POWER_DEFAULT_PMC; |
| 405 | prop->dc_power_default = DC_POWER_DEFAULT_PMC; |
| 406 | } else { |
| 407 | prop->max_power_default = MAX_POWER_DEFAULT_PCI; |
| 408 | prop->dc_power_default = DC_POWER_DEFAULT_PCI; |
| 409 | } |
| 410 | } |
| 411 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 412 | static int gaudi_get_fixed_properties(struct hl_device *hdev) |
| 413 | { |
| 414 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
Ofir Bitton | 843839b | 2020-07-19 11:08:09 +0300 | [diff] [blame] | 415 | u32 num_sync_stream_queues = 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 416 | int i; |
| 417 | |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 418 | prop->max_queues = GAUDI_QUEUE_ID_SIZE; |
| 419 | prop->hw_queues_props = kcalloc(prop->max_queues, |
| 420 | sizeof(struct hw_queue_properties), |
| 421 | GFP_KERNEL); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 422 | |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 423 | if (!prop->hw_queues_props) |
| 424 | return -ENOMEM; |
| 425 | |
| 426 | for (i = 0 ; i < prop->max_queues ; i++) { |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 427 | if (gaudi_queue_type[i] == QUEUE_TYPE_EXT) { |
| 428 | prop->hw_queues_props[i].type = QUEUE_TYPE_EXT; |
| 429 | prop->hw_queues_props[i].driver_only = 0; |
Ofir Bitton | 21e7a34 | 2020-05-14 18:25:47 +0300 | [diff] [blame] | 430 | prop->hw_queues_props[i].supports_sync_stream = 1; |
Tal Cohen | 4bb1f2f | 2020-06-03 09:25:27 +0300 | [diff] [blame] | 431 | prop->hw_queues_props[i].cb_alloc_flags = |
| 432 | CB_ALLOC_KERNEL; |
Ofir Bitton | 843839b | 2020-07-19 11:08:09 +0300 | [diff] [blame] | 433 | num_sync_stream_queues++; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 434 | } else if (gaudi_queue_type[i] == QUEUE_TYPE_CPU) { |
| 435 | prop->hw_queues_props[i].type = QUEUE_TYPE_CPU; |
| 436 | prop->hw_queues_props[i].driver_only = 1; |
Ofir Bitton | 21e7a34 | 2020-05-14 18:25:47 +0300 | [diff] [blame] | 437 | prop->hw_queues_props[i].supports_sync_stream = 0; |
Tal Cohen | 4bb1f2f | 2020-06-03 09:25:27 +0300 | [diff] [blame] | 438 | prop->hw_queues_props[i].cb_alloc_flags = |
| 439 | CB_ALLOC_KERNEL; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 440 | } else if (gaudi_queue_type[i] == QUEUE_TYPE_INT) { |
| 441 | prop->hw_queues_props[i].type = QUEUE_TYPE_INT; |
| 442 | prop->hw_queues_props[i].driver_only = 0; |
Tal Cohen | 4bb1f2f | 2020-06-03 09:25:27 +0300 | [diff] [blame] | 443 | prop->hw_queues_props[i].supports_sync_stream = 0; |
| 444 | prop->hw_queues_props[i].cb_alloc_flags = |
| 445 | CB_ALLOC_USER; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 446 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 447 | } |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 448 | prop->hw_queues_props[i].collective_mode = |
| 449 | get_collective_mode(hdev, i); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 450 | } |
| 451 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 452 | prop->completion_queues_count = NUMBER_OF_CMPLT_QUEUES; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 453 | prop->collective_first_sob = 0; |
| 454 | prop->collective_first_mon = 0; |
| 455 | |
| 456 | /* 2 SOBs per internal queue stream are reserved for collective */ |
| 457 | prop->sync_stream_first_sob = |
| 458 | ALIGN(NUMBER_OF_SOBS_IN_GRP, HL_MAX_SOBS_PER_MONITOR) |
| 459 | * QMAN_STREAMS * HL_RSVD_SOBS; |
| 460 | |
| 461 | /* 1 monitor per internal queue stream are reserved for collective |
| 462 | * 2 monitors per external queue stream are reserved for collective |
| 463 | */ |
| 464 | prop->sync_stream_first_mon = |
| 465 | (NUMBER_OF_COLLECTIVE_QUEUES * QMAN_STREAMS) + |
| 466 | (NUMBER_OF_EXT_HW_QUEUES * 2); |
| 467 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 468 | prop->dram_base_address = DRAM_PHYS_BASE; |
| 469 | prop->dram_size = GAUDI_HBM_SIZE_32GB; |
| 470 | prop->dram_end_address = prop->dram_base_address + |
| 471 | prop->dram_size; |
| 472 | prop->dram_user_base_address = DRAM_BASE_ADDR_USER; |
| 473 | |
| 474 | prop->sram_base_address = SRAM_BASE_ADDR; |
| 475 | prop->sram_size = SRAM_SIZE; |
| 476 | prop->sram_end_address = prop->sram_base_address + |
| 477 | prop->sram_size; |
| 478 | prop->sram_user_base_address = prop->sram_base_address + |
| 479 | SRAM_USER_BASE_OFFSET; |
| 480 | |
| 481 | prop->mmu_pgt_addr = MMU_PAGE_TABLES_ADDR; |
| 482 | if (hdev->pldm) |
| 483 | prop->mmu_pgt_size = 0x800000; /* 8MB */ |
| 484 | else |
| 485 | prop->mmu_pgt_size = MMU_PAGE_TABLES_SIZE; |
| 486 | prop->mmu_pte_size = HL_PTE_SIZE; |
| 487 | prop->mmu_hop_table_size = HOP_TABLE_SIZE; |
| 488 | prop->mmu_hop0_tables_total_size = HOP0_TABLES_TOTAL_SIZE; |
| 489 | prop->dram_page_size = PAGE_SIZE_2MB; |
Oded Gabbay | 7f070c9 | 2020-11-09 09:48:31 +0200 | [diff] [blame] | 490 | prop->dram_supports_virtual_memory = false; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 491 | |
| 492 | prop->pmmu.hop0_shift = HOP0_SHIFT; |
| 493 | prop->pmmu.hop1_shift = HOP1_SHIFT; |
| 494 | prop->pmmu.hop2_shift = HOP2_SHIFT; |
| 495 | prop->pmmu.hop3_shift = HOP3_SHIFT; |
| 496 | prop->pmmu.hop4_shift = HOP4_SHIFT; |
| 497 | prop->pmmu.hop0_mask = HOP0_MASK; |
| 498 | prop->pmmu.hop1_mask = HOP1_MASK; |
| 499 | prop->pmmu.hop2_mask = HOP2_MASK; |
| 500 | prop->pmmu.hop3_mask = HOP3_MASK; |
| 501 | prop->pmmu.hop4_mask = HOP4_MASK; |
| 502 | prop->pmmu.start_addr = VA_HOST_SPACE_START; |
| 503 | prop->pmmu.end_addr = |
| 504 | (VA_HOST_SPACE_START + VA_HOST_SPACE_SIZE / 2) - 1; |
| 505 | prop->pmmu.page_size = PAGE_SIZE_4KB; |
Moti Haimovski | 7edf341 | 2020-08-23 13:23:13 +0300 | [diff] [blame] | 506 | prop->pmmu.num_hops = MMU_ARCH_5_HOPS; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 507 | |
| 508 | /* PMMU and HPMMU are the same except of page size */ |
| 509 | memcpy(&prop->pmmu_huge, &prop->pmmu, sizeof(prop->pmmu)); |
| 510 | prop->pmmu_huge.page_size = PAGE_SIZE_2MB; |
| 511 | |
| 512 | /* shifts and masks are the same in PMMU and DMMU */ |
| 513 | memcpy(&prop->dmmu, &prop->pmmu, sizeof(prop->pmmu)); |
| 514 | prop->dmmu.start_addr = (VA_HOST_SPACE_START + VA_HOST_SPACE_SIZE / 2); |
| 515 | prop->dmmu.end_addr = VA_HOST_SPACE_END; |
| 516 | prop->dmmu.page_size = PAGE_SIZE_2MB; |
| 517 | |
| 518 | prop->cfg_size = CFG_SIZE; |
| 519 | prop->max_asid = MAX_ASID; |
| 520 | prop->num_of_events = GAUDI_EVENT_SIZE; |
| 521 | prop->tpc_enabled_mask = TPC_ENABLED_MASK; |
| 522 | |
Koby Elbaz | cd5def8 | 2021-02-23 21:31:27 +0200 | [diff] [blame] | 523 | set_default_power_values(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 524 | |
| 525 | prop->cb_pool_cb_cnt = GAUDI_CB_POOL_CB_CNT; |
| 526 | prop->cb_pool_cb_size = GAUDI_CB_POOL_CB_SIZE; |
| 527 | |
| 528 | prop->pcie_dbi_base_address = mmPCIE_DBI_BASE; |
| 529 | prop->pcie_aux_dbi_reg_addr = CFG_BASE + mmPCIE_AUX_DBI; |
| 530 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 531 | strncpy(prop->cpucp_info.card_name, GAUDI_DEFAULT_CARD_NAME, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 532 | CARD_NAME_MAX_LEN); |
| 533 | |
Ofir Bitton | c16d45f | 2020-06-02 12:28:27 +0300 | [diff] [blame] | 534 | prop->max_pending_cs = GAUDI_MAX_PENDING_CS; |
| 535 | |
Ofir Bitton | 843839b | 2020-07-19 11:08:09 +0300 | [diff] [blame] | 536 | prop->first_available_user_sob[HL_GAUDI_WS_DCORE] = |
Ofir Bitton | 3cf74b3 | 2020-09-10 09:17:50 +0300 | [diff] [blame] | 537 | prop->sync_stream_first_sob + |
| 538 | (num_sync_stream_queues * HL_RSVD_SOBS); |
Ofir Bitton | 843839b | 2020-07-19 11:08:09 +0300 | [diff] [blame] | 539 | prop->first_available_user_mon[HL_GAUDI_WS_DCORE] = |
Ofir Bitton | 3cf74b3 | 2020-09-10 09:17:50 +0300 | [diff] [blame] | 540 | prop->sync_stream_first_mon + |
| 541 | (num_sync_stream_queues * HL_RSVD_MONS); |
Ofir Bitton | 843839b | 2020-07-19 11:08:09 +0300 | [diff] [blame] | 542 | |
Ofir Bitton | e1fa724 | 2021-01-06 15:40:37 +0200 | [diff] [blame] | 543 | prop->first_available_user_msix_interrupt = USHRT_MAX; |
| 544 | |
Ofir Bitton | e52606d | 2021-01-27 16:34:37 +0200 | [diff] [blame] | 545 | for (i = 0 ; i < HL_MAX_DCORES ; i++) |
| 546 | prop->first_available_cq[i] = USHRT_MAX; |
| 547 | |
Ofir Bitton | 323b726 | 2020-10-04 09:09:19 +0300 | [diff] [blame] | 548 | prop->fw_security_status_valid = false; |
Ofir Bitton | d611b9f | 2020-11-08 13:10:09 +0200 | [diff] [blame] | 549 | prop->hard_reset_done_by_fw = false; |
Ofir Bitton | 323b726 | 2020-10-04 09:09:19 +0300 | [diff] [blame] | 550 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 551 | return 0; |
| 552 | } |
| 553 | |
| 554 | static int gaudi_pci_bars_map(struct hl_device *hdev) |
| 555 | { |
| 556 | static const char * const name[] = {"SRAM", "CFG", "HBM"}; |
| 557 | bool is_wc[3] = {false, false, true}; |
| 558 | int rc; |
| 559 | |
| 560 | rc = hl_pci_bars_map(hdev, name, is_wc); |
| 561 | if (rc) |
| 562 | return rc; |
| 563 | |
| 564 | hdev->rmmio = hdev->pcie_bar[CFG_BAR_ID] + |
| 565 | (CFG_BASE - SPI_FLASH_BASE_ADDR); |
| 566 | |
| 567 | return 0; |
| 568 | } |
| 569 | |
| 570 | static u64 gaudi_set_hbm_bar_base(struct hl_device *hdev, u64 addr) |
| 571 | { |
| 572 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ofir Bitton | f4cbfd2 | 2020-06-15 17:45:12 +0300 | [diff] [blame] | 573 | struct hl_inbound_pci_region pci_region; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 574 | u64 old_addr = addr; |
| 575 | int rc; |
| 576 | |
| 577 | if ((gaudi) && (gaudi->hbm_bar_cur_addr == addr)) |
| 578 | return old_addr; |
| 579 | |
| 580 | /* Inbound Region 2 - Bar 4 - Point to HBM */ |
Ofir Bitton | f4cbfd2 | 2020-06-15 17:45:12 +0300 | [diff] [blame] | 581 | pci_region.mode = PCI_BAR_MATCH_MODE; |
| 582 | pci_region.bar = HBM_BAR_ID; |
| 583 | pci_region.addr = addr; |
| 584 | rc = hl_pci_set_inbound_region(hdev, 2, &pci_region); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 585 | if (rc) |
| 586 | return U64_MAX; |
| 587 | |
| 588 | if (gaudi) { |
| 589 | old_addr = gaudi->hbm_bar_cur_addr; |
| 590 | gaudi->hbm_bar_cur_addr = addr; |
| 591 | } |
| 592 | |
| 593 | return old_addr; |
| 594 | } |
| 595 | |
| 596 | static int gaudi_init_iatu(struct hl_device *hdev) |
| 597 | { |
Ofir Bitton | f4cbfd2 | 2020-06-15 17:45:12 +0300 | [diff] [blame] | 598 | struct hl_inbound_pci_region inbound_region; |
| 599 | struct hl_outbound_pci_region outbound_region; |
| 600 | int rc; |
| 601 | |
Ofir Bitton | 41f458f | 2021-03-10 15:08:44 +0200 | [diff] [blame] | 602 | if (hdev->asic_prop.iatu_done_by_fw) { |
| 603 | hdev->asic_funcs->set_dma_mask_from_fw(hdev); |
| 604 | return 0; |
| 605 | } |
| 606 | |
Ofir Bitton | f4cbfd2 | 2020-06-15 17:45:12 +0300 | [diff] [blame] | 607 | /* Inbound Region 0 - Bar 0 - Point to SRAM + CFG */ |
| 608 | inbound_region.mode = PCI_BAR_MATCH_MODE; |
| 609 | inbound_region.bar = SRAM_BAR_ID; |
| 610 | inbound_region.addr = SRAM_BASE_ADDR; |
| 611 | rc = hl_pci_set_inbound_region(hdev, 0, &inbound_region); |
| 612 | if (rc) |
| 613 | goto done; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 614 | |
| 615 | /* Inbound Region 1 - Bar 2 - Point to SPI FLASH */ |
Ofir Bitton | f4cbfd2 | 2020-06-15 17:45:12 +0300 | [diff] [blame] | 616 | inbound_region.mode = PCI_BAR_MATCH_MODE; |
| 617 | inbound_region.bar = CFG_BAR_ID; |
| 618 | inbound_region.addr = SPI_FLASH_BASE_ADDR; |
| 619 | rc = hl_pci_set_inbound_region(hdev, 1, &inbound_region); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 620 | if (rc) |
Ofir Bitton | f4cbfd2 | 2020-06-15 17:45:12 +0300 | [diff] [blame] | 621 | goto done; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 622 | |
Ofir Bitton | f4cbfd2 | 2020-06-15 17:45:12 +0300 | [diff] [blame] | 623 | /* Inbound Region 2 - Bar 4 - Point to HBM */ |
| 624 | inbound_region.mode = PCI_BAR_MATCH_MODE; |
| 625 | inbound_region.bar = HBM_BAR_ID; |
| 626 | inbound_region.addr = DRAM_PHYS_BASE; |
| 627 | rc = hl_pci_set_inbound_region(hdev, 2, &inbound_region); |
| 628 | if (rc) |
| 629 | goto done; |
| 630 | |
| 631 | hdev->asic_funcs->set_dma_mask_from_fw(hdev); |
| 632 | |
| 633 | /* Outbound Region 0 - Point to Host */ |
| 634 | outbound_region.addr = HOST_PHYS_BASE; |
| 635 | outbound_region.size = HOST_PHYS_SIZE; |
| 636 | rc = hl_pci_set_outbound_region(hdev, &outbound_region); |
| 637 | |
| 638 | done: |
| 639 | return rc; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 640 | } |
| 641 | |
Ofir Bitton | d1ddd90 | 2020-10-19 17:04:20 +0300 | [diff] [blame] | 642 | static enum hl_device_hw_state gaudi_get_hw_state(struct hl_device *hdev) |
| 643 | { |
| 644 | return RREG32(mmHW_STATE); |
| 645 | } |
| 646 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 647 | static int gaudi_early_init(struct hl_device *hdev) |
| 648 | { |
| 649 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 650 | struct pci_dev *pdev = hdev->pdev; |
Ofir Bitton | 41f458f | 2021-03-10 15:08:44 +0200 | [diff] [blame] | 651 | u32 fw_boot_status; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 652 | int rc; |
| 653 | |
| 654 | rc = gaudi_get_fixed_properties(hdev); |
| 655 | if (rc) { |
| 656 | dev_err(hdev->dev, "Failed to get fixed properties\n"); |
| 657 | return rc; |
| 658 | } |
| 659 | |
| 660 | /* Check BAR sizes */ |
| 661 | if (pci_resource_len(pdev, SRAM_BAR_ID) != SRAM_BAR_SIZE) { |
| 662 | dev_err(hdev->dev, |
| 663 | "Not " HL_NAME "? BAR %d size %llu, expecting %llu\n", |
| 664 | SRAM_BAR_ID, |
| 665 | (unsigned long long) pci_resource_len(pdev, |
| 666 | SRAM_BAR_ID), |
| 667 | SRAM_BAR_SIZE); |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 668 | rc = -ENODEV; |
| 669 | goto free_queue_props; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 670 | } |
| 671 | |
| 672 | if (pci_resource_len(pdev, CFG_BAR_ID) != CFG_BAR_SIZE) { |
| 673 | dev_err(hdev->dev, |
| 674 | "Not " HL_NAME "? BAR %d size %llu, expecting %llu\n", |
| 675 | CFG_BAR_ID, |
| 676 | (unsigned long long) pci_resource_len(pdev, |
| 677 | CFG_BAR_ID), |
| 678 | CFG_BAR_SIZE); |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 679 | rc = -ENODEV; |
| 680 | goto free_queue_props; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 681 | } |
| 682 | |
| 683 | prop->dram_pci_bar_size = pci_resource_len(pdev, HBM_BAR_ID); |
| 684 | |
Ofir Bitton | 41f458f | 2021-03-10 15:08:44 +0200 | [diff] [blame] | 685 | /* If FW security is enabled at this point it means no access to ELBI */ |
| 686 | if (!hdev->asic_prop.fw_security_disabled) { |
| 687 | hdev->asic_prop.iatu_done_by_fw = true; |
| 688 | goto pci_init; |
| 689 | } |
| 690 | |
| 691 | rc = hl_pci_elbi_read(hdev, CFG_BASE + mmCPU_BOOT_DEV_STS0, |
| 692 | &fw_boot_status); |
| 693 | if (rc) |
| 694 | goto free_queue_props; |
| 695 | |
| 696 | /* Check whether FW is configuring iATU */ |
| 697 | if ((fw_boot_status & CPU_BOOT_DEV_STS0_ENABLED) && |
| 698 | (fw_boot_status & CPU_BOOT_DEV_STS0_FW_IATU_CONF_EN)) |
| 699 | hdev->asic_prop.iatu_done_by_fw = true; |
| 700 | |
| 701 | pci_init: |
Ofir Bitton | d1ddd90 | 2020-10-19 17:04:20 +0300 | [diff] [blame] | 702 | rc = hl_pci_init(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 703 | if (rc) |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 704 | goto free_queue_props; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 705 | |
Ofir Bitton | d1ddd90 | 2020-10-19 17:04:20 +0300 | [diff] [blame] | 706 | /* Before continuing in the initialization, we need to read the preboot |
| 707 | * version to determine whether we run with a security-enabled firmware |
| 708 | */ |
| 709 | rc = hl_fw_read_preboot_status(hdev, mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS, |
| 710 | mmCPU_BOOT_DEV_STS0, mmCPU_BOOT_ERR0, |
| 711 | GAUDI_BOOT_FIT_REQ_TIMEOUT_USEC); |
| 712 | if (rc) { |
| 713 | if (hdev->reset_on_preboot_fail) |
| 714 | hdev->asic_funcs->hw_fini(hdev, true); |
| 715 | goto pci_fini; |
| 716 | } |
| 717 | |
Ofir Bitton | 9c9013c | 2020-12-01 10:39:54 +0200 | [diff] [blame] | 718 | if (gaudi_get_hw_state(hdev) == HL_DEVICE_HW_STATE_DIRTY) { |
| 719 | dev_info(hdev->dev, |
| 720 | "H/W state is dirty, must reset before initializing\n"); |
| 721 | hdev->asic_funcs->hw_fini(hdev, true); |
| 722 | } |
| 723 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 724 | return 0; |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 725 | |
Ofir Bitton | d1ddd90 | 2020-10-19 17:04:20 +0300 | [diff] [blame] | 726 | pci_fini: |
| 727 | hl_pci_fini(hdev); |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 728 | free_queue_props: |
| 729 | kfree(hdev->asic_prop.hw_queues_props); |
| 730 | return rc; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 731 | } |
| 732 | |
| 733 | static int gaudi_early_fini(struct hl_device *hdev) |
| 734 | { |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 735 | kfree(hdev->asic_prop.hw_queues_props); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 736 | hl_pci_fini(hdev); |
| 737 | |
| 738 | return 0; |
| 739 | } |
| 740 | |
| 741 | /** |
| 742 | * gaudi_fetch_psoc_frequency - Fetch PSOC frequency values |
| 743 | * |
| 744 | * @hdev: pointer to hl_device structure |
| 745 | * |
| 746 | */ |
Ofir Bitton | 1cbca89 | 2020-10-05 11:36:00 +0300 | [diff] [blame] | 747 | static int gaudi_fetch_psoc_frequency(struct hl_device *hdev) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 748 | { |
| 749 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
Alon Mizrahi | 6585489 | 2020-11-19 16:34:19 +0200 | [diff] [blame] | 750 | u32 nr = 0, nf = 0, od = 0, div_fctr = 0, pll_clk, div_sel; |
| 751 | u16 pll_freq_arr[HL_PLL_NUM_OUTPUTS], freq; |
Ofir Bitton | 1cbca89 | 2020-10-05 11:36:00 +0300 | [diff] [blame] | 752 | int rc; |
| 753 | |
Alon Mizrahi | 6585489 | 2020-11-19 16:34:19 +0200 | [diff] [blame] | 754 | if (hdev->asic_prop.fw_security_disabled) { |
| 755 | /* Backward compatibility */ |
| 756 | div_fctr = RREG32(mmPSOC_CPU_PLL_DIV_FACTOR_2); |
| 757 | div_sel = RREG32(mmPSOC_CPU_PLL_DIV_SEL_2); |
| 758 | nr = RREG32(mmPSOC_CPU_PLL_NR); |
| 759 | nf = RREG32(mmPSOC_CPU_PLL_NF); |
| 760 | od = RREG32(mmPSOC_CPU_PLL_OD); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 761 | |
Alon Mizrahi | 6585489 | 2020-11-19 16:34:19 +0200 | [diff] [blame] | 762 | if (div_sel == DIV_SEL_REF_CLK || |
| 763 | div_sel == DIV_SEL_DIVIDED_REF) { |
| 764 | if (div_sel == DIV_SEL_REF_CLK) |
| 765 | freq = PLL_REF_CLK; |
| 766 | else |
| 767 | freq = PLL_REF_CLK / (div_fctr + 1); |
| 768 | } else if (div_sel == DIV_SEL_PLL_CLK || |
| 769 | div_sel == DIV_SEL_DIVIDED_PLL) { |
| 770 | pll_clk = PLL_REF_CLK * (nf + 1) / |
| 771 | ((nr + 1) * (od + 1)); |
| 772 | if (div_sel == DIV_SEL_PLL_CLK) |
| 773 | freq = pll_clk; |
| 774 | else |
| 775 | freq = pll_clk / (div_fctr + 1); |
| 776 | } else { |
| 777 | dev_warn(hdev->dev, |
| 778 | "Received invalid div select value: %d", |
| 779 | div_sel); |
| 780 | freq = 0; |
| 781 | } |
| 782 | } else { |
Bharat Jauhari | 285c0fa | 2021-03-25 18:15:40 +0200 | [diff] [blame] | 783 | rc = hl_fw_cpucp_pll_info_get(hdev, HL_GAUDI_CPU_PLL, pll_freq_arr); |
Alon Mizrahi | 6585489 | 2020-11-19 16:34:19 +0200 | [diff] [blame] | 784 | |
| 785 | if (rc) |
| 786 | return rc; |
| 787 | |
| 788 | freq = pll_freq_arr[2]; |
| 789 | } |
| 790 | |
| 791 | prop->psoc_timestamp_frequency = freq; |
| 792 | prop->psoc_pci_pll_nr = nr; |
| 793 | prop->psoc_pci_pll_nf = nf; |
| 794 | prop->psoc_pci_pll_od = od; |
| 795 | prop->psoc_pci_pll_div_factor = div_fctr; |
Ofir Bitton | 1cbca89 | 2020-10-05 11:36:00 +0300 | [diff] [blame] | 796 | |
| 797 | return 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 798 | } |
| 799 | |
| 800 | static int _gaudi_init_tpc_mem(struct hl_device *hdev, |
| 801 | dma_addr_t tpc_kernel_src_addr, u32 tpc_kernel_size) |
| 802 | { |
| 803 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 804 | struct packet_lin_dma *init_tpc_mem_pkt; |
| 805 | struct hl_cs_job *job; |
| 806 | struct hl_cb *cb; |
| 807 | u64 dst_addr; |
| 808 | u32 cb_size, ctl; |
| 809 | u8 tpc_id; |
| 810 | int rc; |
| 811 | |
Ofir Bitton | a04b7cd | 2020-07-13 13:36:55 +0300 | [diff] [blame] | 812 | cb = hl_cb_kernel_create(hdev, PAGE_SIZE, false); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 813 | if (!cb) |
| 814 | return -EFAULT; |
| 815 | |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 816 | init_tpc_mem_pkt = cb->kernel_address; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 817 | cb_size = sizeof(*init_tpc_mem_pkt); |
| 818 | memset(init_tpc_mem_pkt, 0, cb_size); |
| 819 | |
| 820 | init_tpc_mem_pkt->tsize = cpu_to_le32(tpc_kernel_size); |
| 821 | |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 822 | ctl = FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_LIN_DMA); |
| 823 | ctl |= FIELD_PREP(GAUDI_PKT_LIN_DMA_CTL_LIN_MASK, 1); |
| 824 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 825 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 826 | |
| 827 | init_tpc_mem_pkt->ctl = cpu_to_le32(ctl); |
| 828 | |
| 829 | init_tpc_mem_pkt->src_addr = cpu_to_le64(tpc_kernel_src_addr); |
| 830 | dst_addr = (prop->sram_user_base_address & |
| 831 | GAUDI_PKT_LIN_DMA_DST_ADDR_MASK) >> |
| 832 | GAUDI_PKT_LIN_DMA_DST_ADDR_SHIFT; |
| 833 | init_tpc_mem_pkt->dst_addr |= cpu_to_le64(dst_addr); |
| 834 | |
| 835 | job = hl_cs_allocate_job(hdev, QUEUE_TYPE_EXT, true); |
| 836 | if (!job) { |
| 837 | dev_err(hdev->dev, "Failed to allocate a new job\n"); |
| 838 | rc = -ENOMEM; |
| 839 | goto release_cb; |
| 840 | } |
| 841 | |
| 842 | job->id = 0; |
| 843 | job->user_cb = cb; |
Tomer Tayar | f0748674 | 2020-08-02 22:51:31 +0300 | [diff] [blame] | 844 | atomic_inc(&job->user_cb->cs_cnt); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 845 | job->user_cb_size = cb_size; |
| 846 | job->hw_queue_id = GAUDI_QUEUE_ID_DMA_0_0; |
| 847 | job->patched_cb = job->user_cb; |
| 848 | job->job_cb_size = job->user_cb_size + sizeof(struct packet_msg_prot); |
| 849 | |
| 850 | hl_debugfs_add_job(hdev, job); |
| 851 | |
| 852 | rc = gaudi_send_job_on_qman0(hdev, job); |
| 853 | |
| 854 | if (rc) |
| 855 | goto free_job; |
| 856 | |
| 857 | for (tpc_id = 0 ; tpc_id < TPC_NUMBER_OF_ENGINES ; tpc_id++) { |
| 858 | rc = gaudi_run_tpc_kernel(hdev, dst_addr, tpc_id); |
| 859 | if (rc) |
| 860 | break; |
| 861 | } |
| 862 | |
| 863 | free_job: |
| 864 | hl_userptr_delete_list(hdev, &job->userptr_list); |
| 865 | hl_debugfs_remove_job(hdev, job); |
| 866 | kfree(job); |
Tomer Tayar | f0748674 | 2020-08-02 22:51:31 +0300 | [diff] [blame] | 867 | atomic_dec(&cb->cs_cnt); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 868 | |
| 869 | release_cb: |
| 870 | hl_cb_put(cb); |
| 871 | hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, cb->id << PAGE_SHIFT); |
| 872 | |
| 873 | return rc; |
| 874 | } |
| 875 | |
| 876 | /* |
| 877 | * gaudi_init_tpc_mem() - Initialize TPC memories. |
| 878 | * @hdev: Pointer to hl_device structure. |
| 879 | * |
| 880 | * Copy TPC kernel fw from firmware file and run it to initialize TPC memories. |
| 881 | * |
| 882 | * Return: 0 for success, negative value for error. |
| 883 | */ |
| 884 | static int gaudi_init_tpc_mem(struct hl_device *hdev) |
| 885 | { |
| 886 | const struct firmware *fw; |
| 887 | size_t fw_size; |
| 888 | void *cpu_addr; |
| 889 | dma_addr_t dma_handle; |
Oded Gabbay | 98e8781 | 2020-12-09 23:07:58 +0200 | [diff] [blame] | 890 | int rc, count = 5; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 891 | |
Oded Gabbay | 98e8781 | 2020-12-09 23:07:58 +0200 | [diff] [blame] | 892 | again: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 893 | rc = request_firmware(&fw, GAUDI_TPC_FW_FILE, hdev->dev); |
Oded Gabbay | 98e8781 | 2020-12-09 23:07:58 +0200 | [diff] [blame] | 894 | if (rc == -EINTR && count-- > 0) { |
| 895 | msleep(50); |
| 896 | goto again; |
| 897 | } |
| 898 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 899 | if (rc) { |
Oded Gabbay | 98e8781 | 2020-12-09 23:07:58 +0200 | [diff] [blame] | 900 | dev_err(hdev->dev, "Failed to load firmware file %s\n", |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 901 | GAUDI_TPC_FW_FILE); |
| 902 | goto out; |
| 903 | } |
| 904 | |
| 905 | fw_size = fw->size; |
| 906 | cpu_addr = hdev->asic_funcs->asic_dma_alloc_coherent(hdev, fw_size, |
| 907 | &dma_handle, GFP_KERNEL | __GFP_ZERO); |
| 908 | if (!cpu_addr) { |
| 909 | dev_err(hdev->dev, |
| 910 | "Failed to allocate %zu of dma memory for TPC kernel\n", |
| 911 | fw_size); |
| 912 | rc = -ENOMEM; |
| 913 | goto out; |
| 914 | } |
| 915 | |
| 916 | memcpy(cpu_addr, fw->data, fw_size); |
| 917 | |
| 918 | rc = _gaudi_init_tpc_mem(hdev, dma_handle, fw_size); |
| 919 | |
| 920 | hdev->asic_funcs->asic_dma_free_coherent(hdev, fw->size, cpu_addr, |
| 921 | dma_handle); |
| 922 | |
| 923 | out: |
| 924 | release_firmware(fw); |
| 925 | return rc; |
| 926 | } |
| 927 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 928 | static void gaudi_collective_map_sobs(struct hl_device *hdev, u32 stream) |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 929 | { |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 930 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 931 | struct gaudi_collective_properties *prop = &gaudi->collective_props; |
| 932 | struct hl_hw_queue *q; |
| 933 | u32 i, sob_id, sob_group_id, queue_id; |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 934 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 935 | /* Iterate through SOB groups and assign a SOB for each slave queue */ |
| 936 | sob_group_id = |
| 937 | stream * HL_RSVD_SOBS + prop->curr_sob_group_idx[stream]; |
| 938 | sob_id = prop->hw_sob_group[sob_group_id].base_sob_id; |
| 939 | |
| 940 | queue_id = GAUDI_QUEUE_ID_NIC_0_0 + stream; |
| 941 | for (i = 0 ; i < NIC_NUMBER_OF_ENGINES ; i++) { |
| 942 | q = &hdev->kernel_queues[queue_id + (4 * i)]; |
| 943 | q->sync_stream_prop.collective_sob_id = sob_id + i; |
| 944 | } |
| 945 | |
| 946 | /* Both DMA5 and TPC7 use the same resources since only a single |
| 947 | * engine need to participate in the reduction process |
| 948 | */ |
| 949 | queue_id = GAUDI_QUEUE_ID_DMA_5_0 + stream; |
| 950 | q = &hdev->kernel_queues[queue_id]; |
| 951 | q->sync_stream_prop.collective_sob_id = |
| 952 | sob_id + NIC_NUMBER_OF_ENGINES; |
| 953 | |
| 954 | queue_id = GAUDI_QUEUE_ID_TPC_7_0 + stream; |
| 955 | q = &hdev->kernel_queues[queue_id]; |
| 956 | q->sync_stream_prop.collective_sob_id = |
| 957 | sob_id + NIC_NUMBER_OF_ENGINES; |
| 958 | } |
| 959 | |
| 960 | static void gaudi_sob_group_hw_reset(struct kref *ref) |
| 961 | { |
| 962 | struct gaudi_hw_sob_group *hw_sob_group = |
| 963 | container_of(ref, struct gaudi_hw_sob_group, kref); |
| 964 | struct hl_device *hdev = hw_sob_group->hdev; |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 965 | u64 base_addr; |
| 966 | int rc; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 967 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 968 | base_addr = CFG_BASE + mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0 + |
| 969 | hw_sob_group->base_sob_id * 4; |
| 970 | rc = gaudi_schedule_register_memset(hdev, hw_sob_group->queue_id, |
| 971 | base_addr, NUMBER_OF_SOBS_IN_GRP, 0); |
| 972 | if (rc) |
| 973 | dev_err(hdev->dev, |
| 974 | "failed resetting sob group - sob base %u, count %u", |
| 975 | hw_sob_group->base_sob_id, NUMBER_OF_SOBS_IN_GRP); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 976 | |
| 977 | kref_init(&hw_sob_group->kref); |
| 978 | } |
| 979 | |
| 980 | static void gaudi_sob_group_reset_error(struct kref *ref) |
| 981 | { |
| 982 | struct gaudi_hw_sob_group *hw_sob_group = |
| 983 | container_of(ref, struct gaudi_hw_sob_group, kref); |
| 984 | struct hl_device *hdev = hw_sob_group->hdev; |
| 985 | |
| 986 | dev_crit(hdev->dev, |
| 987 | "SOB release shouldn't be called here, base_sob_id: %d\n", |
| 988 | hw_sob_group->base_sob_id); |
| 989 | } |
| 990 | |
| 991 | static int gaudi_collective_init(struct hl_device *hdev) |
| 992 | { |
| 993 | u32 i, master_monitor_sobs, sob_id, reserved_sobs_per_group; |
| 994 | struct gaudi_collective_properties *prop; |
| 995 | struct gaudi_device *gaudi; |
| 996 | |
| 997 | gaudi = hdev->asic_specific; |
| 998 | prop = &gaudi->collective_props; |
| 999 | sob_id = hdev->asic_prop.collective_first_sob; |
| 1000 | |
| 1001 | /* First sob in group must be aligned to HL_MAX_SOBS_PER_MONITOR */ |
| 1002 | reserved_sobs_per_group = |
| 1003 | ALIGN(NUMBER_OF_SOBS_IN_GRP, HL_MAX_SOBS_PER_MONITOR); |
| 1004 | |
| 1005 | /* Init SOB groups */ |
| 1006 | for (i = 0 ; i < NUM_SOB_GROUPS; i++) { |
| 1007 | prop->hw_sob_group[i].hdev = hdev; |
| 1008 | prop->hw_sob_group[i].base_sob_id = sob_id; |
| 1009 | sob_id += reserved_sobs_per_group; |
| 1010 | gaudi_sob_group_hw_reset(&prop->hw_sob_group[i].kref); |
| 1011 | } |
| 1012 | |
| 1013 | for (i = 0 ; i < QMAN_STREAMS; i++) { |
| 1014 | prop->next_sob_group_val[i] = 1; |
| 1015 | prop->curr_sob_group_idx[i] = 0; |
| 1016 | gaudi_collective_map_sobs(hdev, i); |
| 1017 | } |
| 1018 | |
| 1019 | prop->mstr_sob_mask[0] = 0; |
| 1020 | master_monitor_sobs = HL_MAX_SOBS_PER_MONITOR; |
| 1021 | for (i = 0 ; i < master_monitor_sobs ; i++) |
| 1022 | if (gaudi->hw_cap_initialized & BIT(HW_CAP_NIC_SHIFT + i)) |
| 1023 | prop->mstr_sob_mask[0] |= BIT(i); |
| 1024 | |
| 1025 | prop->mstr_sob_mask[1] = 0; |
| 1026 | master_monitor_sobs = |
| 1027 | NIC_NUMBER_OF_ENGINES - HL_MAX_SOBS_PER_MONITOR; |
| 1028 | for (i = 0 ; i < master_monitor_sobs; i++) { |
| 1029 | if (gaudi->hw_cap_initialized & BIT(HW_CAP_NIC_SHIFT + i)) |
| 1030 | prop->mstr_sob_mask[1] |= BIT(i); |
| 1031 | } |
| 1032 | |
| 1033 | /* Set collective engine bit */ |
| 1034 | prop->mstr_sob_mask[1] |= BIT(i); |
| 1035 | |
| 1036 | return 0; |
| 1037 | } |
| 1038 | |
| 1039 | static void gaudi_reset_sob_group(struct hl_device *hdev, u16 sob_group) |
| 1040 | { |
| 1041 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1042 | struct gaudi_collective_properties *cprop = &gaudi->collective_props; |
| 1043 | |
| 1044 | kref_put(&cprop->hw_sob_group[sob_group].kref, |
| 1045 | gaudi_sob_group_hw_reset); |
| 1046 | } |
| 1047 | |
| 1048 | static void gaudi_collective_master_init_job(struct hl_device *hdev, |
| 1049 | struct hl_cs_job *job, u32 stream, u32 sob_group_offset) |
| 1050 | { |
| 1051 | u32 master_sob_base, master_monitor, queue_id, cb_size = 0; |
| 1052 | struct gaudi_collective_properties *cprop; |
| 1053 | struct hl_gen_wait_properties wait_prop; |
| 1054 | struct hl_sync_stream_properties *prop; |
| 1055 | struct gaudi_device *gaudi; |
| 1056 | |
| 1057 | gaudi = hdev->asic_specific; |
| 1058 | cprop = &gaudi->collective_props; |
| 1059 | queue_id = job->hw_queue_id; |
| 1060 | prop = &hdev->kernel_queues[queue_id].sync_stream_prop; |
| 1061 | |
| 1062 | master_sob_base = |
| 1063 | cprop->hw_sob_group[sob_group_offset].base_sob_id; |
| 1064 | master_monitor = prop->collective_mstr_mon_id[0]; |
| 1065 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 1066 | cprop->hw_sob_group[sob_group_offset].queue_id = queue_id; |
| 1067 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1068 | dev_dbg(hdev->dev, |
| 1069 | "Generate master wait CBs, sob %d (mask %#x), val:0x%x, mon %u, q %d\n", |
| 1070 | master_sob_base, cprop->mstr_sob_mask[0], |
| 1071 | cprop->next_sob_group_val[stream], |
| 1072 | master_monitor, queue_id); |
| 1073 | |
| 1074 | wait_prop.data = (void *) job->patched_cb; |
| 1075 | wait_prop.sob_base = master_sob_base; |
| 1076 | wait_prop.sob_mask = cprop->mstr_sob_mask[0]; |
| 1077 | wait_prop.sob_val = cprop->next_sob_group_val[stream]; |
| 1078 | wait_prop.mon_id = master_monitor; |
| 1079 | wait_prop.q_idx = queue_id; |
| 1080 | wait_prop.size = cb_size; |
| 1081 | cb_size += gaudi_gen_wait_cb(hdev, &wait_prop); |
| 1082 | |
| 1083 | master_sob_base += HL_MAX_SOBS_PER_MONITOR; |
| 1084 | master_monitor = prop->collective_mstr_mon_id[1]; |
| 1085 | |
| 1086 | dev_dbg(hdev->dev, |
| 1087 | "Generate master wait CBs, sob %d (mask %#x), val:0x%x, mon %u, q %d\n", |
| 1088 | master_sob_base, cprop->mstr_sob_mask[1], |
| 1089 | cprop->next_sob_group_val[stream], |
| 1090 | master_monitor, queue_id); |
| 1091 | |
| 1092 | wait_prop.sob_base = master_sob_base; |
| 1093 | wait_prop.sob_mask = cprop->mstr_sob_mask[1]; |
| 1094 | wait_prop.mon_id = master_monitor; |
| 1095 | wait_prop.size = cb_size; |
| 1096 | cb_size += gaudi_gen_wait_cb(hdev, &wait_prop); |
| 1097 | } |
| 1098 | |
| 1099 | static void gaudi_collective_slave_init_job(struct hl_device *hdev, |
| 1100 | struct hl_cs_job *job, struct hl_cs_compl *cs_cmpl) |
| 1101 | { |
| 1102 | struct hl_gen_wait_properties wait_prop; |
| 1103 | struct hl_sync_stream_properties *prop; |
| 1104 | u32 queue_id, cb_size = 0; |
| 1105 | |
| 1106 | queue_id = job->hw_queue_id; |
| 1107 | prop = &hdev->kernel_queues[queue_id].sync_stream_prop; |
| 1108 | |
| 1109 | /* Add to wait CBs using slave monitor */ |
| 1110 | wait_prop.data = (void *) job->user_cb; |
| 1111 | wait_prop.sob_base = cs_cmpl->hw_sob->sob_id; |
| 1112 | wait_prop.sob_mask = 0x1; |
| 1113 | wait_prop.sob_val = cs_cmpl->sob_val; |
| 1114 | wait_prop.mon_id = prop->collective_slave_mon_id; |
| 1115 | wait_prop.q_idx = queue_id; |
| 1116 | wait_prop.size = cb_size; |
| 1117 | |
| 1118 | dev_dbg(hdev->dev, |
| 1119 | "Generate slave wait CB, sob %d, val:0x%x, mon %d, q %d\n", |
| 1120 | cs_cmpl->hw_sob->sob_id, cs_cmpl->sob_val, |
| 1121 | prop->collective_slave_mon_id, queue_id); |
| 1122 | |
| 1123 | cb_size += gaudi_gen_wait_cb(hdev, &wait_prop); |
| 1124 | |
| 1125 | dev_dbg(hdev->dev, |
| 1126 | "generate signal CB, sob_id: %d, sob val: 1, q_idx: %d\n", |
| 1127 | prop->collective_sob_id, queue_id); |
| 1128 | |
| 1129 | cb_size += gaudi_gen_signal_cb(hdev, job->user_cb, |
Alon Mizrahi | 72ab9ca5 | 2020-12-02 19:55:30 +0200 | [diff] [blame] | 1130 | prop->collective_sob_id, cb_size, false); |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 1131 | } |
| 1132 | |
| 1133 | static void gaudi_collective_wait_init_cs(struct hl_cs *cs) |
| 1134 | { |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1135 | struct hl_cs_compl *signal_cs_cmpl = |
| 1136 | container_of(cs->signal_fence, struct hl_cs_compl, base_fence); |
| 1137 | struct hl_cs_compl *cs_cmpl = |
| 1138 | container_of(cs->fence, struct hl_cs_compl, base_fence); |
| 1139 | struct gaudi_collective_properties *cprop; |
| 1140 | u32 stream, queue_id, sob_group_offset; |
| 1141 | struct gaudi_device *gaudi; |
| 1142 | struct hl_device *hdev; |
| 1143 | struct hl_cs_job *job; |
| 1144 | struct hl_ctx *ctx; |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 1145 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1146 | ctx = cs->ctx; |
| 1147 | hdev = ctx->hdev; |
| 1148 | gaudi = hdev->asic_specific; |
| 1149 | cprop = &gaudi->collective_props; |
| 1150 | |
| 1151 | /* copy the SOB id and value of the signal CS */ |
| 1152 | cs_cmpl->hw_sob = signal_cs_cmpl->hw_sob; |
| 1153 | cs_cmpl->sob_val = signal_cs_cmpl->sob_val; |
| 1154 | |
| 1155 | /* Calculate the stream from collective master queue (1st job) */ |
| 1156 | job = list_first_entry(&cs->job_list, struct hl_cs_job, cs_node); |
| 1157 | stream = job->hw_queue_id % 4; |
| 1158 | sob_group_offset = |
| 1159 | stream * HL_RSVD_SOBS + cprop->curr_sob_group_idx[stream]; |
| 1160 | |
| 1161 | list_for_each_entry(job, &cs->job_list, cs_node) { |
| 1162 | queue_id = job->hw_queue_id; |
| 1163 | |
| 1164 | if (hdev->kernel_queues[queue_id].collective_mode == |
| 1165 | HL_COLLECTIVE_MASTER) |
| 1166 | gaudi_collective_master_init_job(hdev, job, stream, |
| 1167 | sob_group_offset); |
| 1168 | else |
| 1169 | gaudi_collective_slave_init_job(hdev, job, cs_cmpl); |
| 1170 | } |
| 1171 | |
| 1172 | cs_cmpl->sob_group = sob_group_offset; |
| 1173 | |
| 1174 | /* Handle sob group kref and wraparound */ |
| 1175 | kref_get(&cprop->hw_sob_group[sob_group_offset].kref); |
| 1176 | cprop->next_sob_group_val[stream]++; |
| 1177 | |
| 1178 | if (cprop->next_sob_group_val[stream] == HL_MAX_SOB_VAL) { |
| 1179 | /* |
| 1180 | * Decrement as we reached the max value. |
| 1181 | * The release function won't be called here as we've |
| 1182 | * just incremented the refcount. |
| 1183 | */ |
| 1184 | kref_put(&cprop->hw_sob_group[sob_group_offset].kref, |
| 1185 | gaudi_sob_group_reset_error); |
| 1186 | cprop->next_sob_group_val[stream] = 1; |
| 1187 | /* only two SOBs are currently in use */ |
| 1188 | cprop->curr_sob_group_idx[stream] = |
| 1189 | (cprop->curr_sob_group_idx[stream] + 1) & |
| 1190 | (HL_RSVD_SOBS - 1); |
| 1191 | |
| 1192 | gaudi_collective_map_sobs(hdev, stream); |
| 1193 | |
| 1194 | dev_dbg(hdev->dev, "switched to SOB group %d, stream: %d\n", |
| 1195 | cprop->curr_sob_group_idx[stream], stream); |
| 1196 | } |
| 1197 | |
| 1198 | /* Increment kref since all slave queues are now waiting on it */ |
| 1199 | kref_get(&cs_cmpl->hw_sob->kref); |
| 1200 | /* |
| 1201 | * Must put the signal fence after the SOB refcnt increment so |
| 1202 | * the SOB refcnt won't turn 0 and reset the SOB before the |
| 1203 | * wait CS was submitted. |
| 1204 | */ |
| 1205 | mb(); |
| 1206 | hl_fence_put(cs->signal_fence); |
| 1207 | cs->signal_fence = NULL; |
| 1208 | } |
| 1209 | |
| 1210 | static int gaudi_collective_wait_create_job(struct hl_device *hdev, |
| 1211 | struct hl_ctx *ctx, struct hl_cs *cs, |
| 1212 | enum hl_collective_mode mode, u32 queue_id, u32 wait_queue_id) |
| 1213 | { |
| 1214 | struct hw_queue_properties *hw_queue_prop; |
| 1215 | struct hl_cs_counters_atomic *cntr; |
| 1216 | struct hl_cs_job *job; |
| 1217 | struct hl_cb *cb; |
| 1218 | u32 cb_size; |
| 1219 | bool patched_cb; |
| 1220 | |
| 1221 | cntr = &hdev->aggregated_cs_counters; |
| 1222 | |
| 1223 | if (mode == HL_COLLECTIVE_MASTER) { |
| 1224 | /* CB size of collective master queue contains |
| 1225 | * 4 msg short packets for monitor 1 configuration |
| 1226 | * 1 fence packet |
| 1227 | * 4 msg short packets for monitor 2 configuration |
| 1228 | * 1 fence packet |
| 1229 | * 2 msg prot packets for completion and MSI-X |
| 1230 | */ |
| 1231 | cb_size = sizeof(struct packet_msg_short) * 8 + |
| 1232 | sizeof(struct packet_fence) * 2 + |
| 1233 | sizeof(struct packet_msg_prot) * 2; |
| 1234 | patched_cb = true; |
| 1235 | } else { |
| 1236 | /* CB size of collective slave queues contains |
| 1237 | * 4 msg short packets for monitor configuration |
| 1238 | * 1 fence packet |
| 1239 | * 1 additional msg short packet for sob signal |
| 1240 | */ |
| 1241 | cb_size = sizeof(struct packet_msg_short) * 5 + |
| 1242 | sizeof(struct packet_fence); |
| 1243 | patched_cb = false; |
| 1244 | } |
| 1245 | |
| 1246 | hw_queue_prop = &hdev->asic_prop.hw_queues_props[queue_id]; |
| 1247 | job = hl_cs_allocate_job(hdev, hw_queue_prop->type, true); |
| 1248 | if (!job) { |
farah kassabri | e753643 | 2020-10-12 14:30:26 +0300 | [diff] [blame] | 1249 | atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1250 | atomic64_inc(&cntr->out_of_mem_drop_cnt); |
| 1251 | dev_err(hdev->dev, "Failed to allocate a new job\n"); |
| 1252 | return -ENOMEM; |
| 1253 | } |
| 1254 | |
| 1255 | /* Allocate internal mapped CB for non patched CBs */ |
| 1256 | cb = hl_cb_kernel_create(hdev, cb_size, |
| 1257 | hdev->mmu_enable && !patched_cb); |
| 1258 | if (!cb) { |
farah kassabri | e753643 | 2020-10-12 14:30:26 +0300 | [diff] [blame] | 1259 | atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1260 | atomic64_inc(&cntr->out_of_mem_drop_cnt); |
| 1261 | kfree(job); |
| 1262 | return -EFAULT; |
| 1263 | } |
| 1264 | |
| 1265 | job->id = 0; |
| 1266 | job->cs = cs; |
| 1267 | job->user_cb = cb; |
Tomer Tayar | f0748674 | 2020-08-02 22:51:31 +0300 | [diff] [blame] | 1268 | atomic_inc(&job->user_cb->cs_cnt); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1269 | job->user_cb_size = cb_size; |
| 1270 | job->hw_queue_id = queue_id; |
| 1271 | |
| 1272 | /* |
| 1273 | * No need in parsing, user CB is the patched CB. |
| 1274 | * We call hl_cb_destroy() out of two reasons - we don't need |
| 1275 | * the CB in the CB idr anymore and to decrement its refcount as |
| 1276 | * it was incremented inside hl_cb_kernel_create(). |
| 1277 | */ |
| 1278 | if (patched_cb) |
| 1279 | job->patched_cb = job->user_cb; |
| 1280 | else |
| 1281 | job->patched_cb = NULL; |
| 1282 | |
| 1283 | job->job_cb_size = job->user_cb_size; |
| 1284 | hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, cb->id << PAGE_SHIFT); |
| 1285 | |
| 1286 | /* increment refcount as for external queues we get completion */ |
| 1287 | if (hw_queue_prop->type == QUEUE_TYPE_EXT) |
| 1288 | cs_get(cs); |
| 1289 | |
| 1290 | cs->jobs_in_queue_cnt[job->hw_queue_id]++; |
| 1291 | |
| 1292 | list_add_tail(&job->cs_node, &cs->job_list); |
| 1293 | |
| 1294 | hl_debugfs_add_job(hdev, job); |
| 1295 | |
| 1296 | return 0; |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 1297 | } |
| 1298 | |
| 1299 | static int gaudi_collective_wait_create_jobs(struct hl_device *hdev, |
| 1300 | struct hl_ctx *ctx, struct hl_cs *cs, u32 wait_queue_id, |
| 1301 | u32 collective_engine_id) |
| 1302 | { |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1303 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1304 | struct hw_queue_properties *hw_queue_prop; |
| 1305 | u32 queue_id, collective_queue, num_jobs; |
| 1306 | u32 stream, nic_queue, nic_idx = 0; |
| 1307 | bool skip; |
Ofir Bitton | 266cdfa | 2020-12-22 15:56:12 +0200 | [diff] [blame] | 1308 | int i, rc = 0; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1309 | |
| 1310 | /* Verify wait queue id is configured as master */ |
| 1311 | hw_queue_prop = &hdev->asic_prop.hw_queues_props[wait_queue_id]; |
| 1312 | if (!(hw_queue_prop->collective_mode == HL_COLLECTIVE_MASTER)) { |
| 1313 | dev_err(hdev->dev, |
| 1314 | "Queue %d is not configured as collective master\n", |
| 1315 | wait_queue_id); |
| 1316 | return -EINVAL; |
| 1317 | } |
| 1318 | |
| 1319 | /* Verify engine id is supported */ |
| 1320 | if (collective_engine_id != GAUDI_ENGINE_ID_DMA_5 && |
| 1321 | collective_engine_id != GAUDI_ENGINE_ID_TPC_7) { |
| 1322 | dev_err(hdev->dev, |
| 1323 | "Collective wait does not support engine %u\n", |
| 1324 | collective_engine_id); |
| 1325 | return -EINVAL; |
| 1326 | } |
| 1327 | |
| 1328 | stream = wait_queue_id % 4; |
| 1329 | |
| 1330 | if (collective_engine_id == GAUDI_ENGINE_ID_DMA_5) |
| 1331 | collective_queue = GAUDI_QUEUE_ID_DMA_5_0 + stream; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1332 | else |
Ofir Bitton | 71a984f | 2020-10-19 16:52:00 +0300 | [diff] [blame] | 1333 | collective_queue = GAUDI_QUEUE_ID_TPC_7_0 + stream; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1334 | |
| 1335 | num_jobs = NUMBER_OF_SOBS_IN_GRP + 1; |
| 1336 | nic_queue = GAUDI_QUEUE_ID_NIC_0_0 + stream; |
| 1337 | |
| 1338 | /* First job goes to the collective master queue, it will wait for |
| 1339 | * the collective slave queues to finish execution. |
| 1340 | * The synchronization is done using two monitors: |
| 1341 | * First monitor for NICs 0-7, second monitor for NICs 8-9 and the |
| 1342 | * reduction engine (DMA5/TPC7). |
| 1343 | * |
| 1344 | * Rest of the jobs goes to the collective slave queues which will |
| 1345 | * all wait for the user to signal sob 'cs_cmpl->sob_val'. |
| 1346 | */ |
| 1347 | for (i = 0 ; i < num_jobs ; i++) { |
| 1348 | if (i == 0) { |
| 1349 | queue_id = wait_queue_id; |
| 1350 | rc = gaudi_collective_wait_create_job(hdev, ctx, cs, |
| 1351 | HL_COLLECTIVE_MASTER, queue_id, wait_queue_id); |
| 1352 | } else { |
| 1353 | if (nic_idx < NIC_NUMBER_OF_ENGINES) { |
| 1354 | if (gaudi->hw_cap_initialized & |
| 1355 | BIT(HW_CAP_NIC_SHIFT + nic_idx)) |
| 1356 | skip = false; |
| 1357 | else |
| 1358 | skip = true; |
| 1359 | |
| 1360 | queue_id = nic_queue; |
| 1361 | nic_queue += 4; |
| 1362 | nic_idx++; |
| 1363 | |
| 1364 | if (skip) |
| 1365 | continue; |
| 1366 | } else { |
| 1367 | queue_id = collective_queue; |
| 1368 | } |
| 1369 | |
| 1370 | rc = gaudi_collective_wait_create_job(hdev, ctx, cs, |
| 1371 | HL_COLLECTIVE_SLAVE, queue_id, wait_queue_id); |
| 1372 | } |
| 1373 | |
| 1374 | if (rc) |
| 1375 | return rc; |
| 1376 | } |
| 1377 | |
| 1378 | return rc; |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 1379 | } |
| 1380 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1381 | static int gaudi_late_init(struct hl_device *hdev) |
| 1382 | { |
| 1383 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1384 | int rc; |
| 1385 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 1386 | rc = gaudi->cpucp_info_get(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1387 | if (rc) { |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 1388 | dev_err(hdev->dev, "Failed to get cpucp info\n"); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1389 | return rc; |
| 1390 | } |
| 1391 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 1392 | if ((hdev->card_type == cpucp_card_type_pci) && |
| 1393 | (hdev->nic_ports_mask & 0x3)) { |
| 1394 | dev_info(hdev->dev, |
| 1395 | "PCI card detected, only 8 ports are enabled\n"); |
| 1396 | hdev->nic_ports_mask &= ~0x3; |
| 1397 | |
| 1398 | /* Stop and disable unused NIC QMANs */ |
| 1399 | WREG32(mmNIC0_QM0_GLBL_CFG1, NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 1400 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 1401 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 1402 | |
| 1403 | WREG32(mmNIC0_QM1_GLBL_CFG1, NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 1404 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 1405 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 1406 | |
| 1407 | WREG32(mmNIC0_QM0_GLBL_CFG0, 0); |
| 1408 | WREG32(mmNIC0_QM1_GLBL_CFG0, 0); |
| 1409 | |
| 1410 | gaudi->hw_cap_initialized &= ~(HW_CAP_NIC0 | HW_CAP_NIC1); |
| 1411 | } |
| 1412 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 1413 | rc = hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_ENABLE_PCI_ACCESS); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1414 | if (rc) { |
| 1415 | dev_err(hdev->dev, "Failed to enable PCI access from CPU\n"); |
| 1416 | return rc; |
| 1417 | } |
| 1418 | |
Ofir Bitton | 1cbca89 | 2020-10-05 11:36:00 +0300 | [diff] [blame] | 1419 | rc = gaudi_fetch_psoc_frequency(hdev); |
| 1420 | if (rc) { |
| 1421 | dev_err(hdev->dev, "Failed to fetch psoc frequency\n"); |
| 1422 | goto disable_pci_access; |
| 1423 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1424 | |
| 1425 | rc = gaudi_mmu_clear_pgt_range(hdev); |
| 1426 | if (rc) { |
| 1427 | dev_err(hdev->dev, "Failed to clear MMU page tables range\n"); |
| 1428 | goto disable_pci_access; |
| 1429 | } |
| 1430 | |
| 1431 | rc = gaudi_init_tpc_mem(hdev); |
| 1432 | if (rc) { |
| 1433 | dev_err(hdev->dev, "Failed to initialize TPC memories\n"); |
| 1434 | goto disable_pci_access; |
| 1435 | } |
| 1436 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 1437 | rc = gaudi_collective_init(hdev); |
| 1438 | if (rc) { |
| 1439 | dev_err(hdev->dev, "Failed to init collective\n"); |
| 1440 | goto disable_pci_access; |
| 1441 | } |
| 1442 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1443 | return 0; |
| 1444 | |
| 1445 | disable_pci_access: |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 1446 | hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_DISABLE_PCI_ACCESS); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1447 | |
| 1448 | return rc; |
| 1449 | } |
| 1450 | |
| 1451 | static void gaudi_late_fini(struct hl_device *hdev) |
| 1452 | { |
| 1453 | const struct hwmon_channel_info **channel_info_arr; |
| 1454 | int i = 0; |
| 1455 | |
| 1456 | if (!hdev->hl_chip_info->info) |
| 1457 | return; |
| 1458 | |
| 1459 | channel_info_arr = hdev->hl_chip_info->info; |
| 1460 | |
| 1461 | while (channel_info_arr[i]) { |
| 1462 | kfree(channel_info_arr[i]->config); |
| 1463 | kfree(channel_info_arr[i]); |
| 1464 | i++; |
| 1465 | } |
| 1466 | |
| 1467 | kfree(channel_info_arr); |
| 1468 | |
| 1469 | hdev->hl_chip_info->info = NULL; |
| 1470 | } |
| 1471 | |
| 1472 | static int gaudi_alloc_cpu_accessible_dma_mem(struct hl_device *hdev) |
| 1473 | { |
| 1474 | dma_addr_t dma_addr_arr[GAUDI_ALLOC_CPU_MEM_RETRY_CNT] = {}, end_addr; |
| 1475 | void *virt_addr_arr[GAUDI_ALLOC_CPU_MEM_RETRY_CNT] = {}; |
| 1476 | int i, j, rc = 0; |
| 1477 | |
| 1478 | /* |
| 1479 | * The device CPU works with 40-bits addresses, while bit 39 must be set |
| 1480 | * to '1' when accessing the host. |
| 1481 | * Bits 49:39 of the full host address are saved for a later |
| 1482 | * configuration of the HW to perform extension to 50 bits. |
| 1483 | * Because there is a single HW register that holds the extension bits, |
| 1484 | * these bits must be identical in all allocated range. |
| 1485 | */ |
| 1486 | |
| 1487 | for (i = 0 ; i < GAUDI_ALLOC_CPU_MEM_RETRY_CNT ; i++) { |
| 1488 | virt_addr_arr[i] = |
| 1489 | hdev->asic_funcs->asic_dma_alloc_coherent(hdev, |
| 1490 | HL_CPU_ACCESSIBLE_MEM_SIZE, |
| 1491 | &dma_addr_arr[i], |
| 1492 | GFP_KERNEL | __GFP_ZERO); |
| 1493 | if (!virt_addr_arr[i]) { |
| 1494 | rc = -ENOMEM; |
| 1495 | goto free_dma_mem_arr; |
| 1496 | } |
| 1497 | |
| 1498 | end_addr = dma_addr_arr[i] + HL_CPU_ACCESSIBLE_MEM_SIZE - 1; |
| 1499 | if (GAUDI_CPU_PCI_MSB_ADDR(dma_addr_arr[i]) == |
| 1500 | GAUDI_CPU_PCI_MSB_ADDR(end_addr)) |
| 1501 | break; |
| 1502 | } |
| 1503 | |
| 1504 | if (i == GAUDI_ALLOC_CPU_MEM_RETRY_CNT) { |
| 1505 | dev_err(hdev->dev, |
| 1506 | "MSB of CPU accessible DMA memory are not identical in all range\n"); |
| 1507 | rc = -EFAULT; |
| 1508 | goto free_dma_mem_arr; |
| 1509 | } |
| 1510 | |
| 1511 | hdev->cpu_accessible_dma_mem = virt_addr_arr[i]; |
| 1512 | hdev->cpu_accessible_dma_address = dma_addr_arr[i]; |
| 1513 | hdev->cpu_pci_msb_addr = |
| 1514 | GAUDI_CPU_PCI_MSB_ADDR(hdev->cpu_accessible_dma_address); |
| 1515 | |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 1516 | if (hdev->asic_prop.fw_security_disabled) |
| 1517 | GAUDI_PCI_TO_CPU_ADDR(hdev->cpu_accessible_dma_address); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1518 | |
| 1519 | free_dma_mem_arr: |
| 1520 | for (j = 0 ; j < i ; j++) |
| 1521 | hdev->asic_funcs->asic_dma_free_coherent(hdev, |
| 1522 | HL_CPU_ACCESSIBLE_MEM_SIZE, |
| 1523 | virt_addr_arr[j], |
| 1524 | dma_addr_arr[j]); |
| 1525 | |
| 1526 | return rc; |
| 1527 | } |
| 1528 | |
| 1529 | static void gaudi_free_internal_qmans_pq_mem(struct hl_device *hdev) |
| 1530 | { |
| 1531 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1532 | struct gaudi_internal_qman_info *q; |
| 1533 | u32 i; |
| 1534 | |
| 1535 | for (i = 0 ; i < GAUDI_QUEUE_ID_SIZE ; i++) { |
| 1536 | q = &gaudi->internal_qmans[i]; |
| 1537 | if (!q->pq_kernel_addr) |
| 1538 | continue; |
| 1539 | hdev->asic_funcs->asic_dma_free_coherent(hdev, q->pq_size, |
| 1540 | q->pq_kernel_addr, |
| 1541 | q->pq_dma_addr); |
| 1542 | } |
| 1543 | } |
| 1544 | |
| 1545 | static int gaudi_alloc_internal_qmans_pq_mem(struct hl_device *hdev) |
| 1546 | { |
| 1547 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1548 | struct gaudi_internal_qman_info *q; |
| 1549 | int rc, i; |
| 1550 | |
| 1551 | for (i = 0 ; i < GAUDI_QUEUE_ID_SIZE ; i++) { |
| 1552 | if (gaudi_queue_type[i] != QUEUE_TYPE_INT) |
| 1553 | continue; |
| 1554 | |
| 1555 | q = &gaudi->internal_qmans[i]; |
| 1556 | |
| 1557 | switch (i) { |
Ofir Bitton | 0940cab | 2020-08-31 08:52:56 +0300 | [diff] [blame] | 1558 | case GAUDI_QUEUE_ID_DMA_2_0 ... GAUDI_QUEUE_ID_DMA_7_3: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1559 | q->pq_size = HBM_DMA_QMAN_SIZE_IN_BYTES; |
| 1560 | break; |
| 1561 | case GAUDI_QUEUE_ID_MME_0_0 ... GAUDI_QUEUE_ID_MME_1_3: |
| 1562 | q->pq_size = MME_QMAN_SIZE_IN_BYTES; |
| 1563 | break; |
| 1564 | case GAUDI_QUEUE_ID_TPC_0_0 ... GAUDI_QUEUE_ID_TPC_7_3: |
| 1565 | q->pq_size = TPC_QMAN_SIZE_IN_BYTES; |
| 1566 | break; |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 1567 | case GAUDI_QUEUE_ID_NIC_0_0 ... GAUDI_QUEUE_ID_NIC_9_3: |
| 1568 | q->pq_size = NIC_QMAN_SIZE_IN_BYTES; |
| 1569 | break; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1570 | default: |
| 1571 | dev_err(hdev->dev, "Bad internal queue index %d", i); |
| 1572 | rc = -EINVAL; |
| 1573 | goto free_internal_qmans_pq_mem; |
| 1574 | } |
| 1575 | |
| 1576 | q->pq_kernel_addr = hdev->asic_funcs->asic_dma_alloc_coherent( |
| 1577 | hdev, q->pq_size, |
| 1578 | &q->pq_dma_addr, |
| 1579 | GFP_KERNEL | __GFP_ZERO); |
| 1580 | if (!q->pq_kernel_addr) { |
| 1581 | rc = -ENOMEM; |
| 1582 | goto free_internal_qmans_pq_mem; |
| 1583 | } |
| 1584 | } |
| 1585 | |
| 1586 | return 0; |
| 1587 | |
| 1588 | free_internal_qmans_pq_mem: |
| 1589 | gaudi_free_internal_qmans_pq_mem(hdev); |
| 1590 | return rc; |
| 1591 | } |
| 1592 | |
| 1593 | static int gaudi_sw_init(struct hl_device *hdev) |
| 1594 | { |
| 1595 | struct gaudi_device *gaudi; |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 1596 | u32 i, event_id = 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1597 | int rc; |
| 1598 | |
| 1599 | /* Allocate device structure */ |
| 1600 | gaudi = kzalloc(sizeof(*gaudi), GFP_KERNEL); |
| 1601 | if (!gaudi) |
| 1602 | return -ENOMEM; |
| 1603 | |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 1604 | for (i = 0 ; i < ARRAY_SIZE(gaudi_irq_map_table) ; i++) { |
| 1605 | if (gaudi_irq_map_table[i].valid) { |
| 1606 | if (event_id == GAUDI_EVENT_SIZE) { |
| 1607 | dev_err(hdev->dev, |
| 1608 | "Event array exceeds the limit of %u events\n", |
| 1609 | GAUDI_EVENT_SIZE); |
| 1610 | rc = -EINVAL; |
| 1611 | goto free_gaudi_device; |
| 1612 | } |
| 1613 | |
| 1614 | gaudi->events[event_id++] = |
| 1615 | gaudi_irq_map_table[i].fc_id; |
| 1616 | } |
| 1617 | } |
| 1618 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 1619 | gaudi->cpucp_info_get = gaudi_cpucp_info_get; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1620 | |
| 1621 | gaudi->max_freq_value = GAUDI_MAX_CLK_FREQ; |
| 1622 | |
| 1623 | hdev->asic_specific = gaudi; |
| 1624 | |
| 1625 | /* Create DMA pool for small allocations */ |
| 1626 | hdev->dma_pool = dma_pool_create(dev_name(hdev->dev), |
| 1627 | &hdev->pdev->dev, GAUDI_DMA_POOL_BLK_SIZE, 8, 0); |
| 1628 | if (!hdev->dma_pool) { |
| 1629 | dev_err(hdev->dev, "failed to create DMA pool\n"); |
| 1630 | rc = -ENOMEM; |
| 1631 | goto free_gaudi_device; |
| 1632 | } |
| 1633 | |
| 1634 | rc = gaudi_alloc_cpu_accessible_dma_mem(hdev); |
| 1635 | if (rc) |
| 1636 | goto free_dma_pool; |
| 1637 | |
| 1638 | hdev->cpu_accessible_dma_pool = gen_pool_create(ilog2(32), -1); |
| 1639 | if (!hdev->cpu_accessible_dma_pool) { |
| 1640 | dev_err(hdev->dev, |
| 1641 | "Failed to create CPU accessible DMA pool\n"); |
| 1642 | rc = -ENOMEM; |
| 1643 | goto free_cpu_dma_mem; |
| 1644 | } |
| 1645 | |
| 1646 | rc = gen_pool_add(hdev->cpu_accessible_dma_pool, |
| 1647 | (uintptr_t) hdev->cpu_accessible_dma_mem, |
| 1648 | HL_CPU_ACCESSIBLE_MEM_SIZE, -1); |
| 1649 | if (rc) { |
| 1650 | dev_err(hdev->dev, |
| 1651 | "Failed to add memory to CPU accessible DMA pool\n"); |
| 1652 | rc = -EFAULT; |
| 1653 | goto free_cpu_accessible_dma_pool; |
| 1654 | } |
| 1655 | |
| 1656 | rc = gaudi_alloc_internal_qmans_pq_mem(hdev); |
| 1657 | if (rc) |
| 1658 | goto free_cpu_accessible_dma_pool; |
| 1659 | |
| 1660 | spin_lock_init(&gaudi->hw_queues_lock); |
| 1661 | mutex_init(&gaudi->clk_gate_mutex); |
| 1662 | |
| 1663 | hdev->supports_sync_stream = true; |
| 1664 | hdev->supports_coresight = true; |
Ofir Bitton | 2795c88 | 2020-12-08 13:47:05 +0200 | [diff] [blame] | 1665 | hdev->supports_staged_submission = true; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1666 | |
| 1667 | return 0; |
| 1668 | |
| 1669 | free_cpu_accessible_dma_pool: |
| 1670 | gen_pool_destroy(hdev->cpu_accessible_dma_pool); |
| 1671 | free_cpu_dma_mem: |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 1672 | if (hdev->asic_prop.fw_security_disabled) |
| 1673 | GAUDI_CPU_TO_PCI_ADDR(hdev->cpu_accessible_dma_address, |
| 1674 | hdev->cpu_pci_msb_addr); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1675 | hdev->asic_funcs->asic_dma_free_coherent(hdev, |
| 1676 | HL_CPU_ACCESSIBLE_MEM_SIZE, |
| 1677 | hdev->cpu_accessible_dma_mem, |
| 1678 | hdev->cpu_accessible_dma_address); |
| 1679 | free_dma_pool: |
| 1680 | dma_pool_destroy(hdev->dma_pool); |
| 1681 | free_gaudi_device: |
| 1682 | kfree(gaudi); |
| 1683 | return rc; |
| 1684 | } |
| 1685 | |
| 1686 | static int gaudi_sw_fini(struct hl_device *hdev) |
| 1687 | { |
| 1688 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1689 | |
| 1690 | gaudi_free_internal_qmans_pq_mem(hdev); |
| 1691 | |
| 1692 | gen_pool_destroy(hdev->cpu_accessible_dma_pool); |
| 1693 | |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 1694 | if (hdev->asic_prop.fw_security_disabled) |
| 1695 | GAUDI_CPU_TO_PCI_ADDR(hdev->cpu_accessible_dma_address, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1696 | hdev->cpu_pci_msb_addr); |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 1697 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1698 | hdev->asic_funcs->asic_dma_free_coherent(hdev, |
| 1699 | HL_CPU_ACCESSIBLE_MEM_SIZE, |
| 1700 | hdev->cpu_accessible_dma_mem, |
| 1701 | hdev->cpu_accessible_dma_address); |
| 1702 | |
| 1703 | dma_pool_destroy(hdev->dma_pool); |
| 1704 | |
| 1705 | mutex_destroy(&gaudi->clk_gate_mutex); |
| 1706 | |
| 1707 | kfree(gaudi); |
| 1708 | |
| 1709 | return 0; |
| 1710 | } |
| 1711 | |
| 1712 | static irqreturn_t gaudi_irq_handler_single(int irq, void *arg) |
| 1713 | { |
| 1714 | struct hl_device *hdev = arg; |
| 1715 | int i; |
| 1716 | |
| 1717 | if (hdev->disabled) |
| 1718 | return IRQ_HANDLED; |
| 1719 | |
| 1720 | for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++) |
| 1721 | hl_irq_handler_cq(irq, &hdev->completion_queue[i]); |
| 1722 | |
| 1723 | hl_irq_handler_eq(irq, &hdev->event_queue); |
| 1724 | |
| 1725 | return IRQ_HANDLED; |
| 1726 | } |
| 1727 | |
| 1728 | /* |
| 1729 | * For backward compatibility, new MSI interrupts should be set after the |
| 1730 | * existing CPU and NIC interrupts. |
| 1731 | */ |
| 1732 | static int gaudi_pci_irq_vector(struct hl_device *hdev, unsigned int nr, |
| 1733 | bool cpu_eq) |
| 1734 | { |
| 1735 | int msi_vec; |
| 1736 | |
| 1737 | if ((nr != GAUDI_EVENT_QUEUE_MSI_IDX) && (cpu_eq)) |
| 1738 | dev_crit(hdev->dev, "CPU EQ must use IRQ %d\n", |
| 1739 | GAUDI_EVENT_QUEUE_MSI_IDX); |
| 1740 | |
| 1741 | msi_vec = ((nr < GAUDI_EVENT_QUEUE_MSI_IDX) || (cpu_eq)) ? nr : |
| 1742 | (nr + NIC_NUMBER_OF_ENGINES + 1); |
| 1743 | |
| 1744 | return pci_irq_vector(hdev->pdev, msi_vec); |
| 1745 | } |
| 1746 | |
| 1747 | static int gaudi_enable_msi_single(struct hl_device *hdev) |
| 1748 | { |
| 1749 | int rc, irq; |
| 1750 | |
Oded Gabbay | 3b82c34 | 2020-11-27 18:10:20 +0200 | [diff] [blame] | 1751 | dev_dbg(hdev->dev, "Working in single MSI IRQ mode\n"); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1752 | |
| 1753 | irq = gaudi_pci_irq_vector(hdev, 0, false); |
| 1754 | rc = request_irq(irq, gaudi_irq_handler_single, 0, |
| 1755 | "gaudi single msi", hdev); |
| 1756 | if (rc) |
| 1757 | dev_err(hdev->dev, |
| 1758 | "Failed to request single MSI IRQ\n"); |
| 1759 | |
| 1760 | return rc; |
| 1761 | } |
| 1762 | |
| 1763 | static int gaudi_enable_msi_multi(struct hl_device *hdev) |
| 1764 | { |
| 1765 | int cq_cnt = hdev->asic_prop.completion_queues_count; |
| 1766 | int rc, i, irq_cnt_init, irq; |
| 1767 | |
| 1768 | for (i = 0, irq_cnt_init = 0 ; i < cq_cnt ; i++, irq_cnt_init++) { |
| 1769 | irq = gaudi_pci_irq_vector(hdev, i, false); |
| 1770 | rc = request_irq(irq, hl_irq_handler_cq, 0, gaudi_irq_name[i], |
| 1771 | &hdev->completion_queue[i]); |
| 1772 | if (rc) { |
| 1773 | dev_err(hdev->dev, "Failed to request IRQ %d", irq); |
| 1774 | goto free_irqs; |
| 1775 | } |
| 1776 | } |
| 1777 | |
| 1778 | irq = gaudi_pci_irq_vector(hdev, GAUDI_EVENT_QUEUE_MSI_IDX, true); |
| 1779 | rc = request_irq(irq, hl_irq_handler_eq, 0, gaudi_irq_name[cq_cnt], |
| 1780 | &hdev->event_queue); |
| 1781 | if (rc) { |
| 1782 | dev_err(hdev->dev, "Failed to request IRQ %d", irq); |
| 1783 | goto free_irqs; |
| 1784 | } |
| 1785 | |
| 1786 | return 0; |
| 1787 | |
| 1788 | free_irqs: |
| 1789 | for (i = 0 ; i < irq_cnt_init ; i++) |
| 1790 | free_irq(gaudi_pci_irq_vector(hdev, i, false), |
| 1791 | &hdev->completion_queue[i]); |
| 1792 | return rc; |
| 1793 | } |
| 1794 | |
| 1795 | static int gaudi_enable_msi(struct hl_device *hdev) |
| 1796 | { |
| 1797 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1798 | int rc; |
| 1799 | |
| 1800 | if (gaudi->hw_cap_initialized & HW_CAP_MSI) |
| 1801 | return 0; |
| 1802 | |
Oded Gabbay | 12e66a1 | 2021-03-08 18:06:57 +0200 | [diff] [blame] | 1803 | rc = pci_alloc_irq_vectors(hdev->pdev, 1, 1, PCI_IRQ_MSI); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1804 | if (rc < 0) { |
| 1805 | dev_err(hdev->dev, "MSI: Failed to enable support %d\n", rc); |
| 1806 | return rc; |
| 1807 | } |
| 1808 | |
| 1809 | if (rc < NUMBER_OF_INTERRUPTS) { |
| 1810 | gaudi->multi_msi_mode = false; |
| 1811 | rc = gaudi_enable_msi_single(hdev); |
| 1812 | } else { |
| 1813 | gaudi->multi_msi_mode = true; |
| 1814 | rc = gaudi_enable_msi_multi(hdev); |
| 1815 | } |
| 1816 | |
| 1817 | if (rc) |
| 1818 | goto free_pci_irq_vectors; |
| 1819 | |
| 1820 | gaudi->hw_cap_initialized |= HW_CAP_MSI; |
| 1821 | |
| 1822 | return 0; |
| 1823 | |
| 1824 | free_pci_irq_vectors: |
| 1825 | pci_free_irq_vectors(hdev->pdev); |
| 1826 | return rc; |
| 1827 | } |
| 1828 | |
| 1829 | static void gaudi_sync_irqs(struct hl_device *hdev) |
| 1830 | { |
| 1831 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1832 | int i, cq_cnt = hdev->asic_prop.completion_queues_count; |
| 1833 | |
| 1834 | if (!(gaudi->hw_cap_initialized & HW_CAP_MSI)) |
| 1835 | return; |
| 1836 | |
| 1837 | /* Wait for all pending IRQs to be finished */ |
| 1838 | if (gaudi->multi_msi_mode) { |
| 1839 | for (i = 0 ; i < cq_cnt ; i++) |
| 1840 | synchronize_irq(gaudi_pci_irq_vector(hdev, i, false)); |
| 1841 | |
| 1842 | synchronize_irq(gaudi_pci_irq_vector(hdev, |
| 1843 | GAUDI_EVENT_QUEUE_MSI_IDX, |
| 1844 | true)); |
| 1845 | } else { |
| 1846 | synchronize_irq(gaudi_pci_irq_vector(hdev, 0, false)); |
| 1847 | } |
| 1848 | } |
| 1849 | |
| 1850 | static void gaudi_disable_msi(struct hl_device *hdev) |
| 1851 | { |
| 1852 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1853 | int i, irq, cq_cnt = hdev->asic_prop.completion_queues_count; |
| 1854 | |
| 1855 | if (!(gaudi->hw_cap_initialized & HW_CAP_MSI)) |
| 1856 | return; |
| 1857 | |
| 1858 | gaudi_sync_irqs(hdev); |
| 1859 | |
| 1860 | if (gaudi->multi_msi_mode) { |
| 1861 | irq = gaudi_pci_irq_vector(hdev, GAUDI_EVENT_QUEUE_MSI_IDX, |
| 1862 | true); |
| 1863 | free_irq(irq, &hdev->event_queue); |
| 1864 | |
| 1865 | for (i = 0 ; i < cq_cnt ; i++) { |
| 1866 | irq = gaudi_pci_irq_vector(hdev, i, false); |
| 1867 | free_irq(irq, &hdev->completion_queue[i]); |
| 1868 | } |
| 1869 | } else { |
| 1870 | free_irq(gaudi_pci_irq_vector(hdev, 0, false), hdev); |
| 1871 | } |
| 1872 | |
| 1873 | pci_free_irq_vectors(hdev->pdev); |
| 1874 | |
| 1875 | gaudi->hw_cap_initialized &= ~HW_CAP_MSI; |
| 1876 | } |
| 1877 | |
| 1878 | static void gaudi_init_scrambler_sram(struct hl_device *hdev) |
| 1879 | { |
| 1880 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1881 | |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 1882 | if (!hdev->asic_prop.fw_security_disabled) |
| 1883 | return; |
| 1884 | |
| 1885 | if (hdev->asic_prop.fw_security_status_valid && |
| 1886 | (hdev->asic_prop.fw_app_security_map & |
| 1887 | CPU_BOOT_DEV_STS0_SRAM_SCR_EN)) |
| 1888 | return; |
| 1889 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1890 | if (gaudi->hw_cap_initialized & HW_CAP_SRAM_SCRAMBLER) |
| 1891 | return; |
| 1892 | |
| 1893 | if (!hdev->sram_scrambler_enable) |
| 1894 | return; |
| 1895 | |
| 1896 | WREG32(mmNIF_RTR_CTRL_0_SCRAM_SRAM_EN, |
| 1897 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1898 | WREG32(mmNIF_RTR_CTRL_1_SCRAM_SRAM_EN, |
| 1899 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1900 | WREG32(mmNIF_RTR_CTRL_2_SCRAM_SRAM_EN, |
| 1901 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1902 | WREG32(mmNIF_RTR_CTRL_3_SCRAM_SRAM_EN, |
| 1903 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1904 | WREG32(mmNIF_RTR_CTRL_4_SCRAM_SRAM_EN, |
| 1905 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1906 | WREG32(mmNIF_RTR_CTRL_5_SCRAM_SRAM_EN, |
| 1907 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1908 | WREG32(mmNIF_RTR_CTRL_6_SCRAM_SRAM_EN, |
| 1909 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1910 | WREG32(mmNIF_RTR_CTRL_7_SCRAM_SRAM_EN, |
| 1911 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1912 | |
| 1913 | WREG32(mmSIF_RTR_CTRL_0_SCRAM_SRAM_EN, |
| 1914 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1915 | WREG32(mmSIF_RTR_CTRL_1_SCRAM_SRAM_EN, |
| 1916 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1917 | WREG32(mmSIF_RTR_CTRL_2_SCRAM_SRAM_EN, |
| 1918 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1919 | WREG32(mmSIF_RTR_CTRL_3_SCRAM_SRAM_EN, |
| 1920 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1921 | WREG32(mmSIF_RTR_CTRL_4_SCRAM_SRAM_EN, |
| 1922 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1923 | WREG32(mmSIF_RTR_CTRL_5_SCRAM_SRAM_EN, |
| 1924 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1925 | WREG32(mmSIF_RTR_CTRL_6_SCRAM_SRAM_EN, |
| 1926 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1927 | WREG32(mmSIF_RTR_CTRL_7_SCRAM_SRAM_EN, |
| 1928 | 1 << IF_RTR_CTRL_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1929 | |
| 1930 | WREG32(mmDMA_IF_E_N_DOWN_CH0_SCRAM_SRAM_EN, |
| 1931 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1932 | WREG32(mmDMA_IF_E_N_DOWN_CH1_SCRAM_SRAM_EN, |
| 1933 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1934 | WREG32(mmDMA_IF_E_S_DOWN_CH0_SCRAM_SRAM_EN, |
| 1935 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1936 | WREG32(mmDMA_IF_E_S_DOWN_CH1_SCRAM_SRAM_EN, |
| 1937 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1938 | WREG32(mmDMA_IF_W_N_DOWN_CH0_SCRAM_SRAM_EN, |
| 1939 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1940 | WREG32(mmDMA_IF_W_N_DOWN_CH1_SCRAM_SRAM_EN, |
| 1941 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1942 | WREG32(mmDMA_IF_W_S_DOWN_CH0_SCRAM_SRAM_EN, |
| 1943 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1944 | WREG32(mmDMA_IF_W_S_DOWN_CH1_SCRAM_SRAM_EN, |
| 1945 | 1 << DMA_IF_DOWN_CHX_SCRAM_SRAM_EN_VAL_SHIFT); |
| 1946 | |
| 1947 | gaudi->hw_cap_initialized |= HW_CAP_SRAM_SCRAMBLER; |
| 1948 | } |
| 1949 | |
| 1950 | static void gaudi_init_scrambler_hbm(struct hl_device *hdev) |
| 1951 | { |
| 1952 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 1953 | |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 1954 | if (!hdev->asic_prop.fw_security_disabled) |
| 1955 | return; |
| 1956 | |
| 1957 | if (hdev->asic_prop.fw_security_status_valid && |
| 1958 | (hdev->asic_prop.fw_boot_cpu_security_map & |
| 1959 | CPU_BOOT_DEV_STS0_DRAM_SCR_EN)) |
| 1960 | return; |
| 1961 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 1962 | if (gaudi->hw_cap_initialized & HW_CAP_HBM_SCRAMBLER) |
| 1963 | return; |
| 1964 | |
| 1965 | if (!hdev->dram_scrambler_enable) |
| 1966 | return; |
| 1967 | |
| 1968 | WREG32(mmNIF_RTR_CTRL_0_SCRAM_HBM_EN, |
| 1969 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1970 | WREG32(mmNIF_RTR_CTRL_1_SCRAM_HBM_EN, |
| 1971 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1972 | WREG32(mmNIF_RTR_CTRL_2_SCRAM_HBM_EN, |
| 1973 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1974 | WREG32(mmNIF_RTR_CTRL_3_SCRAM_HBM_EN, |
| 1975 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1976 | WREG32(mmNIF_RTR_CTRL_4_SCRAM_HBM_EN, |
| 1977 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1978 | WREG32(mmNIF_RTR_CTRL_5_SCRAM_HBM_EN, |
| 1979 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1980 | WREG32(mmNIF_RTR_CTRL_6_SCRAM_HBM_EN, |
| 1981 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1982 | WREG32(mmNIF_RTR_CTRL_7_SCRAM_HBM_EN, |
| 1983 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1984 | |
| 1985 | WREG32(mmSIF_RTR_CTRL_0_SCRAM_HBM_EN, |
| 1986 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1987 | WREG32(mmSIF_RTR_CTRL_1_SCRAM_HBM_EN, |
| 1988 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1989 | WREG32(mmSIF_RTR_CTRL_2_SCRAM_HBM_EN, |
| 1990 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1991 | WREG32(mmSIF_RTR_CTRL_3_SCRAM_HBM_EN, |
| 1992 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1993 | WREG32(mmSIF_RTR_CTRL_4_SCRAM_HBM_EN, |
| 1994 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1995 | WREG32(mmSIF_RTR_CTRL_5_SCRAM_HBM_EN, |
| 1996 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1997 | WREG32(mmSIF_RTR_CTRL_6_SCRAM_HBM_EN, |
| 1998 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 1999 | WREG32(mmSIF_RTR_CTRL_7_SCRAM_HBM_EN, |
| 2000 | 1 << IF_RTR_CTRL_SCRAM_HBM_EN_VAL_SHIFT); |
| 2001 | |
| 2002 | WREG32(mmDMA_IF_E_N_DOWN_CH0_SCRAM_HBM_EN, |
| 2003 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2004 | WREG32(mmDMA_IF_E_N_DOWN_CH1_SCRAM_HBM_EN, |
| 2005 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2006 | WREG32(mmDMA_IF_E_S_DOWN_CH0_SCRAM_HBM_EN, |
| 2007 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2008 | WREG32(mmDMA_IF_E_S_DOWN_CH1_SCRAM_HBM_EN, |
| 2009 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2010 | WREG32(mmDMA_IF_W_N_DOWN_CH0_SCRAM_HBM_EN, |
| 2011 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2012 | WREG32(mmDMA_IF_W_N_DOWN_CH1_SCRAM_HBM_EN, |
| 2013 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2014 | WREG32(mmDMA_IF_W_S_DOWN_CH0_SCRAM_HBM_EN, |
| 2015 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2016 | WREG32(mmDMA_IF_W_S_DOWN_CH1_SCRAM_HBM_EN, |
| 2017 | 1 << DMA_IF_DOWN_CHX_SCRAM_HBM_EN_VAL_SHIFT); |
| 2018 | |
| 2019 | gaudi->hw_cap_initialized |= HW_CAP_HBM_SCRAMBLER; |
| 2020 | } |
| 2021 | |
| 2022 | static void gaudi_init_e2e(struct hl_device *hdev) |
| 2023 | { |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 2024 | if (!hdev->asic_prop.fw_security_disabled) |
| 2025 | return; |
| 2026 | |
| 2027 | if (hdev->asic_prop.fw_security_status_valid && |
| 2028 | (hdev->asic_prop.fw_boot_cpu_security_map & |
| 2029 | CPU_BOOT_DEV_STS0_E2E_CRED_EN)) |
| 2030 | return; |
| 2031 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2032 | WREG32(mmSIF_RTR_CTRL_0_E2E_HBM_WR_SIZE, 247 >> 3); |
| 2033 | WREG32(mmSIF_RTR_CTRL_0_E2E_HBM_RD_SIZE, 785 >> 3); |
| 2034 | WREG32(mmSIF_RTR_CTRL_0_E2E_PCI_WR_SIZE, 49); |
| 2035 | WREG32(mmSIF_RTR_CTRL_0_E2E_PCI_RD_SIZE, 101); |
| 2036 | |
| 2037 | WREG32(mmSIF_RTR_CTRL_1_E2E_HBM_WR_SIZE, 275 >> 3); |
| 2038 | WREG32(mmSIF_RTR_CTRL_1_E2E_HBM_RD_SIZE, 614 >> 3); |
| 2039 | WREG32(mmSIF_RTR_CTRL_1_E2E_PCI_WR_SIZE, 1); |
| 2040 | WREG32(mmSIF_RTR_CTRL_1_E2E_PCI_RD_SIZE, 39); |
| 2041 | |
| 2042 | WREG32(mmSIF_RTR_CTRL_2_E2E_HBM_WR_SIZE, 1); |
| 2043 | WREG32(mmSIF_RTR_CTRL_2_E2E_HBM_RD_SIZE, 1); |
| 2044 | WREG32(mmSIF_RTR_CTRL_2_E2E_PCI_WR_SIZE, 1); |
| 2045 | WREG32(mmSIF_RTR_CTRL_2_E2E_PCI_RD_SIZE, 32); |
| 2046 | |
| 2047 | WREG32(mmSIF_RTR_CTRL_3_E2E_HBM_WR_SIZE, 176 >> 3); |
| 2048 | WREG32(mmSIF_RTR_CTRL_3_E2E_HBM_RD_SIZE, 32 >> 3); |
| 2049 | WREG32(mmSIF_RTR_CTRL_3_E2E_PCI_WR_SIZE, 19); |
| 2050 | WREG32(mmSIF_RTR_CTRL_3_E2E_PCI_RD_SIZE, 32); |
| 2051 | |
| 2052 | WREG32(mmSIF_RTR_CTRL_4_E2E_HBM_WR_SIZE, 176 >> 3); |
| 2053 | WREG32(mmSIF_RTR_CTRL_4_E2E_HBM_RD_SIZE, 32 >> 3); |
| 2054 | WREG32(mmSIF_RTR_CTRL_4_E2E_PCI_WR_SIZE, 19); |
| 2055 | WREG32(mmSIF_RTR_CTRL_4_E2E_PCI_RD_SIZE, 32); |
| 2056 | |
| 2057 | WREG32(mmSIF_RTR_CTRL_5_E2E_HBM_WR_SIZE, 1); |
| 2058 | WREG32(mmSIF_RTR_CTRL_5_E2E_HBM_RD_SIZE, 1); |
| 2059 | WREG32(mmSIF_RTR_CTRL_5_E2E_PCI_WR_SIZE, 1); |
| 2060 | WREG32(mmSIF_RTR_CTRL_5_E2E_PCI_RD_SIZE, 32); |
| 2061 | |
| 2062 | WREG32(mmSIF_RTR_CTRL_6_E2E_HBM_WR_SIZE, 275 >> 3); |
| 2063 | WREG32(mmSIF_RTR_CTRL_6_E2E_HBM_RD_SIZE, 614 >> 3); |
| 2064 | WREG32(mmSIF_RTR_CTRL_6_E2E_PCI_WR_SIZE, 1); |
| 2065 | WREG32(mmSIF_RTR_CTRL_6_E2E_PCI_RD_SIZE, 39); |
| 2066 | |
| 2067 | WREG32(mmSIF_RTR_CTRL_7_E2E_HBM_WR_SIZE, 297 >> 3); |
| 2068 | WREG32(mmSIF_RTR_CTRL_7_E2E_HBM_RD_SIZE, 908 >> 3); |
| 2069 | WREG32(mmSIF_RTR_CTRL_7_E2E_PCI_WR_SIZE, 19); |
| 2070 | WREG32(mmSIF_RTR_CTRL_7_E2E_PCI_RD_SIZE, 19); |
| 2071 | |
| 2072 | WREG32(mmNIF_RTR_CTRL_0_E2E_HBM_WR_SIZE, 318 >> 3); |
| 2073 | WREG32(mmNIF_RTR_CTRL_0_E2E_HBM_RD_SIZE, 956 >> 3); |
| 2074 | WREG32(mmNIF_RTR_CTRL_0_E2E_PCI_WR_SIZE, 79); |
| 2075 | WREG32(mmNIF_RTR_CTRL_0_E2E_PCI_RD_SIZE, 163); |
| 2076 | |
| 2077 | WREG32(mmNIF_RTR_CTRL_1_E2E_HBM_WR_SIZE, 275 >> 3); |
| 2078 | WREG32(mmNIF_RTR_CTRL_1_E2E_HBM_RD_SIZE, 614 >> 3); |
| 2079 | WREG32(mmNIF_RTR_CTRL_1_E2E_PCI_WR_SIZE, 1); |
| 2080 | WREG32(mmNIF_RTR_CTRL_1_E2E_PCI_RD_SIZE, 39); |
| 2081 | |
| 2082 | WREG32(mmNIF_RTR_CTRL_2_E2E_HBM_WR_SIZE, 1); |
| 2083 | WREG32(mmNIF_RTR_CTRL_2_E2E_HBM_RD_SIZE, 1); |
| 2084 | WREG32(mmNIF_RTR_CTRL_2_E2E_PCI_WR_SIZE, 1); |
| 2085 | WREG32(mmNIF_RTR_CTRL_2_E2E_PCI_RD_SIZE, 32); |
| 2086 | |
| 2087 | WREG32(mmNIF_RTR_CTRL_3_E2E_HBM_WR_SIZE, 176 >> 3); |
| 2088 | WREG32(mmNIF_RTR_CTRL_3_E2E_HBM_RD_SIZE, 32 >> 3); |
| 2089 | WREG32(mmNIF_RTR_CTRL_3_E2E_PCI_WR_SIZE, 19); |
| 2090 | WREG32(mmNIF_RTR_CTRL_3_E2E_PCI_RD_SIZE, 32); |
| 2091 | |
| 2092 | WREG32(mmNIF_RTR_CTRL_4_E2E_HBM_WR_SIZE, 176 >> 3); |
| 2093 | WREG32(mmNIF_RTR_CTRL_4_E2E_HBM_RD_SIZE, 32 >> 3); |
| 2094 | WREG32(mmNIF_RTR_CTRL_4_E2E_PCI_WR_SIZE, 19); |
| 2095 | WREG32(mmNIF_RTR_CTRL_4_E2E_PCI_RD_SIZE, 32); |
| 2096 | |
| 2097 | WREG32(mmNIF_RTR_CTRL_5_E2E_HBM_WR_SIZE, 1); |
| 2098 | WREG32(mmNIF_RTR_CTRL_5_E2E_HBM_RD_SIZE, 1); |
| 2099 | WREG32(mmNIF_RTR_CTRL_5_E2E_PCI_WR_SIZE, 1); |
| 2100 | WREG32(mmNIF_RTR_CTRL_5_E2E_PCI_RD_SIZE, 32); |
| 2101 | |
| 2102 | WREG32(mmNIF_RTR_CTRL_6_E2E_HBM_WR_SIZE, 275 >> 3); |
| 2103 | WREG32(mmNIF_RTR_CTRL_6_E2E_HBM_RD_SIZE, 614 >> 3); |
| 2104 | WREG32(mmNIF_RTR_CTRL_6_E2E_PCI_WR_SIZE, 1); |
| 2105 | WREG32(mmNIF_RTR_CTRL_6_E2E_PCI_RD_SIZE, 39); |
| 2106 | |
| 2107 | WREG32(mmNIF_RTR_CTRL_7_E2E_HBM_WR_SIZE, 318 >> 3); |
| 2108 | WREG32(mmNIF_RTR_CTRL_7_E2E_HBM_RD_SIZE, 956 >> 3); |
| 2109 | WREG32(mmNIF_RTR_CTRL_7_E2E_PCI_WR_SIZE, 79); |
| 2110 | WREG32(mmNIF_RTR_CTRL_7_E2E_PCI_RD_SIZE, 79); |
| 2111 | |
| 2112 | WREG32(mmDMA_IF_E_N_DOWN_CH0_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2113 | WREG32(mmDMA_IF_E_N_DOWN_CH0_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2114 | WREG32(mmDMA_IF_E_N_DOWN_CH0_E2E_PCI_WR_SIZE, 162); |
| 2115 | WREG32(mmDMA_IF_E_N_DOWN_CH0_E2E_PCI_RD_SIZE, 338); |
| 2116 | |
| 2117 | WREG32(mmDMA_IF_E_N_DOWN_CH1_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2118 | WREG32(mmDMA_IF_E_N_DOWN_CH1_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2119 | WREG32(mmDMA_IF_E_N_DOWN_CH1_E2E_PCI_WR_SIZE, 162); |
| 2120 | WREG32(mmDMA_IF_E_N_DOWN_CH1_E2E_PCI_RD_SIZE, 338); |
| 2121 | |
| 2122 | WREG32(mmDMA_IF_E_S_DOWN_CH0_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2123 | WREG32(mmDMA_IF_E_S_DOWN_CH0_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2124 | WREG32(mmDMA_IF_E_S_DOWN_CH0_E2E_PCI_WR_SIZE, 162); |
| 2125 | WREG32(mmDMA_IF_E_S_DOWN_CH0_E2E_PCI_RD_SIZE, 338); |
| 2126 | |
| 2127 | WREG32(mmDMA_IF_E_S_DOWN_CH1_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2128 | WREG32(mmDMA_IF_E_S_DOWN_CH1_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2129 | WREG32(mmDMA_IF_E_S_DOWN_CH1_E2E_PCI_WR_SIZE, 162); |
| 2130 | WREG32(mmDMA_IF_E_S_DOWN_CH1_E2E_PCI_RD_SIZE, 338); |
| 2131 | |
| 2132 | WREG32(mmDMA_IF_W_N_DOWN_CH0_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2133 | WREG32(mmDMA_IF_W_N_DOWN_CH0_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2134 | WREG32(mmDMA_IF_W_N_DOWN_CH0_E2E_PCI_WR_SIZE, 162); |
| 2135 | WREG32(mmDMA_IF_W_N_DOWN_CH0_E2E_PCI_RD_SIZE, 338); |
| 2136 | |
| 2137 | WREG32(mmDMA_IF_W_N_DOWN_CH1_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2138 | WREG32(mmDMA_IF_W_N_DOWN_CH1_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2139 | WREG32(mmDMA_IF_W_N_DOWN_CH1_E2E_PCI_WR_SIZE, 162); |
| 2140 | WREG32(mmDMA_IF_W_N_DOWN_CH1_E2E_PCI_RD_SIZE, 338); |
| 2141 | |
| 2142 | WREG32(mmDMA_IF_W_S_DOWN_CH0_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2143 | WREG32(mmDMA_IF_W_S_DOWN_CH0_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2144 | WREG32(mmDMA_IF_W_S_DOWN_CH0_E2E_PCI_WR_SIZE, 162); |
| 2145 | WREG32(mmDMA_IF_W_S_DOWN_CH0_E2E_PCI_RD_SIZE, 338); |
| 2146 | |
| 2147 | WREG32(mmDMA_IF_W_S_DOWN_CH1_E2E_HBM_WR_SIZE, 344 >> 3); |
| 2148 | WREG32(mmDMA_IF_W_S_DOWN_CH1_E2E_HBM_RD_SIZE, 1000 >> 3); |
| 2149 | WREG32(mmDMA_IF_W_S_DOWN_CH1_E2E_PCI_WR_SIZE, 162); |
| 2150 | WREG32(mmDMA_IF_W_S_DOWN_CH1_E2E_PCI_RD_SIZE, 338); |
| 2151 | |
| 2152 | if (!hdev->dram_scrambler_enable) { |
| 2153 | WREG32(mmSIF_RTR_CTRL_0_NL_HBM_SEL_0, 0x21); |
| 2154 | WREG32(mmSIF_RTR_CTRL_0_NL_HBM_SEL_1, 0x22); |
| 2155 | WREG32(mmSIF_RTR_CTRL_0_NL_HBM_OFFSET_18, 0x1F); |
| 2156 | WREG32(mmSIF_RTR_CTRL_0_NL_HBM_PC_SEL_3, 0x20); |
| 2157 | |
| 2158 | WREG32(mmSIF_RTR_CTRL_1_NL_HBM_SEL_0, 0x21); |
| 2159 | WREG32(mmSIF_RTR_CTRL_1_NL_HBM_SEL_1, 0x22); |
| 2160 | WREG32(mmSIF_RTR_CTRL_1_NL_HBM_OFFSET_18, 0x1F); |
| 2161 | WREG32(mmSIF_RTR_CTRL_1_NL_HBM_PC_SEL_3, 0x20); |
| 2162 | |
| 2163 | WREG32(mmSIF_RTR_CTRL_2_NL_HBM_SEL_0, 0x21); |
| 2164 | WREG32(mmSIF_RTR_CTRL_2_NL_HBM_SEL_1, 0x22); |
| 2165 | WREG32(mmSIF_RTR_CTRL_2_NL_HBM_OFFSET_18, 0x1F); |
| 2166 | WREG32(mmSIF_RTR_CTRL_2_NL_HBM_PC_SEL_3, 0x20); |
| 2167 | |
| 2168 | WREG32(mmSIF_RTR_CTRL_3_NL_HBM_SEL_0, 0x21); |
| 2169 | WREG32(mmSIF_RTR_CTRL_3_NL_HBM_SEL_1, 0x22); |
| 2170 | WREG32(mmSIF_RTR_CTRL_3_NL_HBM_OFFSET_18, 0x1F); |
| 2171 | WREG32(mmSIF_RTR_CTRL_3_NL_HBM_PC_SEL_3, 0x20); |
| 2172 | |
| 2173 | WREG32(mmSIF_RTR_CTRL_4_NL_HBM_SEL_0, 0x21); |
| 2174 | WREG32(mmSIF_RTR_CTRL_4_NL_HBM_SEL_1, 0x22); |
| 2175 | WREG32(mmSIF_RTR_CTRL_4_NL_HBM_OFFSET_18, 0x1F); |
| 2176 | WREG32(mmSIF_RTR_CTRL_4_NL_HBM_PC_SEL_3, 0x20); |
| 2177 | |
| 2178 | WREG32(mmSIF_RTR_CTRL_5_NL_HBM_SEL_0, 0x21); |
| 2179 | WREG32(mmSIF_RTR_CTRL_5_NL_HBM_SEL_1, 0x22); |
| 2180 | WREG32(mmSIF_RTR_CTRL_5_NL_HBM_OFFSET_18, 0x1F); |
| 2181 | WREG32(mmSIF_RTR_CTRL_5_NL_HBM_PC_SEL_3, 0x20); |
| 2182 | |
| 2183 | WREG32(mmSIF_RTR_CTRL_6_NL_HBM_SEL_0, 0x21); |
| 2184 | WREG32(mmSIF_RTR_CTRL_6_NL_HBM_SEL_1, 0x22); |
| 2185 | WREG32(mmSIF_RTR_CTRL_6_NL_HBM_OFFSET_18, 0x1F); |
| 2186 | WREG32(mmSIF_RTR_CTRL_6_NL_HBM_PC_SEL_3, 0x20); |
| 2187 | |
| 2188 | WREG32(mmSIF_RTR_CTRL_7_NL_HBM_SEL_0, 0x21); |
| 2189 | WREG32(mmSIF_RTR_CTRL_7_NL_HBM_SEL_1, 0x22); |
| 2190 | WREG32(mmSIF_RTR_CTRL_7_NL_HBM_OFFSET_18, 0x1F); |
| 2191 | WREG32(mmSIF_RTR_CTRL_7_NL_HBM_PC_SEL_3, 0x20); |
| 2192 | |
| 2193 | WREG32(mmNIF_RTR_CTRL_0_NL_HBM_SEL_0, 0x21); |
| 2194 | WREG32(mmNIF_RTR_CTRL_0_NL_HBM_SEL_1, 0x22); |
| 2195 | WREG32(mmNIF_RTR_CTRL_0_NL_HBM_OFFSET_18, 0x1F); |
| 2196 | WREG32(mmNIF_RTR_CTRL_0_NL_HBM_PC_SEL_3, 0x20); |
| 2197 | |
| 2198 | WREG32(mmNIF_RTR_CTRL_1_NL_HBM_SEL_0, 0x21); |
| 2199 | WREG32(mmNIF_RTR_CTRL_1_NL_HBM_SEL_1, 0x22); |
| 2200 | WREG32(mmNIF_RTR_CTRL_1_NL_HBM_OFFSET_18, 0x1F); |
| 2201 | WREG32(mmNIF_RTR_CTRL_1_NL_HBM_PC_SEL_3, 0x20); |
| 2202 | |
| 2203 | WREG32(mmNIF_RTR_CTRL_2_NL_HBM_SEL_0, 0x21); |
| 2204 | WREG32(mmNIF_RTR_CTRL_2_NL_HBM_SEL_1, 0x22); |
| 2205 | WREG32(mmNIF_RTR_CTRL_2_NL_HBM_OFFSET_18, 0x1F); |
| 2206 | WREG32(mmNIF_RTR_CTRL_2_NL_HBM_PC_SEL_3, 0x20); |
| 2207 | |
| 2208 | WREG32(mmNIF_RTR_CTRL_3_NL_HBM_SEL_0, 0x21); |
| 2209 | WREG32(mmNIF_RTR_CTRL_3_NL_HBM_SEL_1, 0x22); |
| 2210 | WREG32(mmNIF_RTR_CTRL_3_NL_HBM_OFFSET_18, 0x1F); |
| 2211 | WREG32(mmNIF_RTR_CTRL_3_NL_HBM_PC_SEL_3, 0x20); |
| 2212 | |
| 2213 | WREG32(mmNIF_RTR_CTRL_4_NL_HBM_SEL_0, 0x21); |
| 2214 | WREG32(mmNIF_RTR_CTRL_4_NL_HBM_SEL_1, 0x22); |
| 2215 | WREG32(mmNIF_RTR_CTRL_4_NL_HBM_OFFSET_18, 0x1F); |
| 2216 | WREG32(mmNIF_RTR_CTRL_4_NL_HBM_PC_SEL_3, 0x20); |
| 2217 | |
| 2218 | WREG32(mmNIF_RTR_CTRL_5_NL_HBM_SEL_0, 0x21); |
| 2219 | WREG32(mmNIF_RTR_CTRL_5_NL_HBM_SEL_1, 0x22); |
| 2220 | WREG32(mmNIF_RTR_CTRL_5_NL_HBM_OFFSET_18, 0x1F); |
| 2221 | WREG32(mmNIF_RTR_CTRL_5_NL_HBM_PC_SEL_3, 0x20); |
| 2222 | |
| 2223 | WREG32(mmNIF_RTR_CTRL_6_NL_HBM_SEL_0, 0x21); |
| 2224 | WREG32(mmNIF_RTR_CTRL_6_NL_HBM_SEL_1, 0x22); |
| 2225 | WREG32(mmNIF_RTR_CTRL_6_NL_HBM_OFFSET_18, 0x1F); |
| 2226 | WREG32(mmNIF_RTR_CTRL_6_NL_HBM_PC_SEL_3, 0x20); |
| 2227 | |
| 2228 | WREG32(mmNIF_RTR_CTRL_7_NL_HBM_SEL_0, 0x21); |
| 2229 | WREG32(mmNIF_RTR_CTRL_7_NL_HBM_SEL_1, 0x22); |
| 2230 | WREG32(mmNIF_RTR_CTRL_7_NL_HBM_OFFSET_18, 0x1F); |
| 2231 | WREG32(mmNIF_RTR_CTRL_7_NL_HBM_PC_SEL_3, 0x20); |
| 2232 | |
| 2233 | WREG32(mmDMA_IF_E_N_DOWN_CH0_NL_HBM_SEL_0, 0x21); |
| 2234 | WREG32(mmDMA_IF_E_N_DOWN_CH0_NL_HBM_SEL_1, 0x22); |
| 2235 | WREG32(mmDMA_IF_E_N_DOWN_CH0_NL_HBM_OFFSET_18, 0x1F); |
| 2236 | WREG32(mmDMA_IF_E_N_DOWN_CH0_NL_HBM_PC_SEL_3, 0x20); |
| 2237 | |
| 2238 | WREG32(mmDMA_IF_E_N_DOWN_CH1_NL_HBM_SEL_0, 0x21); |
| 2239 | WREG32(mmDMA_IF_E_N_DOWN_CH1_NL_HBM_SEL_1, 0x22); |
| 2240 | WREG32(mmDMA_IF_E_N_DOWN_CH1_NL_HBM_OFFSET_18, 0x1F); |
| 2241 | WREG32(mmDMA_IF_E_N_DOWN_CH1_NL_HBM_PC_SEL_3, 0x20); |
| 2242 | |
| 2243 | WREG32(mmDMA_IF_E_S_DOWN_CH0_NL_HBM_SEL_0, 0x21); |
| 2244 | WREG32(mmDMA_IF_E_S_DOWN_CH0_NL_HBM_SEL_1, 0x22); |
| 2245 | WREG32(mmDMA_IF_E_S_DOWN_CH0_NL_HBM_OFFSET_18, 0x1F); |
| 2246 | WREG32(mmDMA_IF_E_S_DOWN_CH0_NL_HBM_PC_SEL_3, 0x20); |
| 2247 | |
| 2248 | WREG32(mmDMA_IF_E_S_DOWN_CH1_NL_HBM_SEL_0, 0x21); |
| 2249 | WREG32(mmDMA_IF_E_S_DOWN_CH1_NL_HBM_SEL_1, 0x22); |
| 2250 | WREG32(mmDMA_IF_E_S_DOWN_CH1_NL_HBM_OFFSET_18, 0x1F); |
| 2251 | WREG32(mmDMA_IF_E_S_DOWN_CH1_NL_HBM_PC_SEL_3, 0x20); |
| 2252 | |
| 2253 | WREG32(mmDMA_IF_W_N_DOWN_CH0_NL_HBM_SEL_0, 0x21); |
| 2254 | WREG32(mmDMA_IF_W_N_DOWN_CH0_NL_HBM_SEL_1, 0x22); |
| 2255 | WREG32(mmDMA_IF_W_N_DOWN_CH0_NL_HBM_OFFSET_18, 0x1F); |
| 2256 | WREG32(mmDMA_IF_W_N_DOWN_CH0_NL_HBM_PC_SEL_3, 0x20); |
| 2257 | |
| 2258 | WREG32(mmDMA_IF_W_N_DOWN_CH1_NL_HBM_SEL_0, 0x21); |
| 2259 | WREG32(mmDMA_IF_W_N_DOWN_CH1_NL_HBM_SEL_1, 0x22); |
| 2260 | WREG32(mmDMA_IF_W_N_DOWN_CH1_NL_HBM_OFFSET_18, 0x1F); |
| 2261 | WREG32(mmDMA_IF_W_N_DOWN_CH1_NL_HBM_PC_SEL_3, 0x20); |
| 2262 | |
| 2263 | WREG32(mmDMA_IF_W_S_DOWN_CH0_NL_HBM_SEL_0, 0x21); |
| 2264 | WREG32(mmDMA_IF_W_S_DOWN_CH0_NL_HBM_SEL_1, 0x22); |
| 2265 | WREG32(mmDMA_IF_W_S_DOWN_CH0_NL_HBM_OFFSET_18, 0x1F); |
| 2266 | WREG32(mmDMA_IF_W_S_DOWN_CH0_NL_HBM_PC_SEL_3, 0x20); |
| 2267 | |
| 2268 | WREG32(mmDMA_IF_W_S_DOWN_CH1_NL_HBM_SEL_0, 0x21); |
| 2269 | WREG32(mmDMA_IF_W_S_DOWN_CH1_NL_HBM_SEL_1, 0x22); |
| 2270 | WREG32(mmDMA_IF_W_S_DOWN_CH1_NL_HBM_OFFSET_18, 0x1F); |
| 2271 | WREG32(mmDMA_IF_W_S_DOWN_CH1_NL_HBM_PC_SEL_3, 0x20); |
| 2272 | } |
| 2273 | |
| 2274 | WREG32(mmSIF_RTR_CTRL_0_E2E_HBM_EN, |
| 2275 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2276 | WREG32(mmSIF_RTR_CTRL_0_E2E_PCI_EN, |
| 2277 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2278 | |
| 2279 | WREG32(mmSIF_RTR_CTRL_1_E2E_HBM_EN, |
| 2280 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2281 | WREG32(mmSIF_RTR_CTRL_1_E2E_PCI_EN, |
| 2282 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2283 | |
| 2284 | WREG32(mmSIF_RTR_CTRL_2_E2E_HBM_EN, |
| 2285 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2286 | WREG32(mmSIF_RTR_CTRL_2_E2E_PCI_EN, |
| 2287 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2288 | |
| 2289 | WREG32(mmSIF_RTR_CTRL_3_E2E_HBM_EN, |
| 2290 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2291 | WREG32(mmSIF_RTR_CTRL_3_E2E_PCI_EN, |
| 2292 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2293 | |
| 2294 | WREG32(mmSIF_RTR_CTRL_4_E2E_HBM_EN, |
| 2295 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2296 | WREG32(mmSIF_RTR_CTRL_4_E2E_PCI_EN, |
| 2297 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2298 | |
| 2299 | WREG32(mmSIF_RTR_CTRL_5_E2E_HBM_EN, |
| 2300 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2301 | WREG32(mmSIF_RTR_CTRL_5_E2E_PCI_EN, |
| 2302 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2303 | |
| 2304 | WREG32(mmSIF_RTR_CTRL_6_E2E_HBM_EN, |
| 2305 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2306 | WREG32(mmSIF_RTR_CTRL_6_E2E_PCI_EN, |
| 2307 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2308 | |
| 2309 | WREG32(mmSIF_RTR_CTRL_7_E2E_HBM_EN, |
| 2310 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2311 | WREG32(mmSIF_RTR_CTRL_7_E2E_PCI_EN, |
| 2312 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2313 | |
| 2314 | WREG32(mmNIF_RTR_CTRL_0_E2E_HBM_EN, |
| 2315 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2316 | WREG32(mmNIF_RTR_CTRL_0_E2E_PCI_EN, |
| 2317 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2318 | |
| 2319 | WREG32(mmNIF_RTR_CTRL_1_E2E_HBM_EN, |
| 2320 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2321 | WREG32(mmNIF_RTR_CTRL_1_E2E_PCI_EN, |
| 2322 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2323 | |
| 2324 | WREG32(mmNIF_RTR_CTRL_2_E2E_HBM_EN, |
| 2325 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2326 | WREG32(mmNIF_RTR_CTRL_2_E2E_PCI_EN, |
| 2327 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2328 | |
| 2329 | WREG32(mmNIF_RTR_CTRL_3_E2E_HBM_EN, |
| 2330 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2331 | WREG32(mmNIF_RTR_CTRL_3_E2E_PCI_EN, |
| 2332 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2333 | |
| 2334 | WREG32(mmNIF_RTR_CTRL_4_E2E_HBM_EN, |
| 2335 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2336 | WREG32(mmNIF_RTR_CTRL_4_E2E_PCI_EN, |
| 2337 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2338 | |
| 2339 | WREG32(mmNIF_RTR_CTRL_5_E2E_HBM_EN, |
| 2340 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2341 | WREG32(mmNIF_RTR_CTRL_5_E2E_PCI_EN, |
| 2342 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2343 | |
| 2344 | WREG32(mmNIF_RTR_CTRL_6_E2E_HBM_EN, |
| 2345 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2346 | WREG32(mmNIF_RTR_CTRL_6_E2E_PCI_EN, |
| 2347 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2348 | |
| 2349 | WREG32(mmNIF_RTR_CTRL_7_E2E_HBM_EN, |
| 2350 | 1 << IF_RTR_CTRL_E2E_HBM_EN_VAL_SHIFT); |
| 2351 | WREG32(mmNIF_RTR_CTRL_7_E2E_PCI_EN, |
| 2352 | 1 << IF_RTR_CTRL_E2E_PCI_EN_VAL_SHIFT); |
| 2353 | |
| 2354 | WREG32(mmDMA_IF_E_N_DOWN_CH0_E2E_HBM_EN, |
| 2355 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2356 | WREG32(mmDMA_IF_E_N_DOWN_CH0_E2E_PCI_EN, |
| 2357 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2358 | |
| 2359 | WREG32(mmDMA_IF_E_N_DOWN_CH1_E2E_HBM_EN, |
| 2360 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2361 | WREG32(mmDMA_IF_E_N_DOWN_CH1_E2E_PCI_EN, |
| 2362 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2363 | |
| 2364 | WREG32(mmDMA_IF_E_S_DOWN_CH0_E2E_HBM_EN, |
| 2365 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2366 | WREG32(mmDMA_IF_E_S_DOWN_CH0_E2E_PCI_EN, |
| 2367 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2368 | |
| 2369 | WREG32(mmDMA_IF_E_S_DOWN_CH1_E2E_HBM_EN, |
| 2370 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2371 | WREG32(mmDMA_IF_E_S_DOWN_CH1_E2E_PCI_EN, |
| 2372 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2373 | |
| 2374 | WREG32(mmDMA_IF_W_N_DOWN_CH0_E2E_HBM_EN, |
| 2375 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2376 | WREG32(mmDMA_IF_W_N_DOWN_CH0_E2E_PCI_EN, |
| 2377 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2378 | |
| 2379 | WREG32(mmDMA_IF_W_N_DOWN_CH1_E2E_HBM_EN, |
| 2380 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2381 | WREG32(mmDMA_IF_W_N_DOWN_CH1_E2E_PCI_EN, |
| 2382 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2383 | |
| 2384 | WREG32(mmDMA_IF_W_S_DOWN_CH0_E2E_HBM_EN, |
| 2385 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2386 | WREG32(mmDMA_IF_W_S_DOWN_CH0_E2E_PCI_EN, |
| 2387 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2388 | |
| 2389 | WREG32(mmDMA_IF_W_S_DOWN_CH1_E2E_HBM_EN, |
| 2390 | 1 << DMA_IF_DOWN_CHX_E2E_HBM_EN_VAL_SHIFT); |
| 2391 | WREG32(mmDMA_IF_W_S_DOWN_CH1_E2E_PCI_EN, |
| 2392 | 1 << DMA_IF_DOWN_CHX_E2E_PCI_EN_VAL_SHIFT); |
| 2393 | } |
| 2394 | |
| 2395 | static void gaudi_init_hbm_cred(struct hl_device *hdev) |
| 2396 | { |
| 2397 | uint32_t hbm0_wr, hbm1_wr, hbm0_rd, hbm1_rd; |
| 2398 | |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 2399 | if (!hdev->asic_prop.fw_security_disabled) |
| 2400 | return; |
| 2401 | |
| 2402 | if (hdev->asic_prop.fw_security_status_valid && |
| 2403 | (hdev->asic_prop.fw_boot_cpu_security_map & |
| 2404 | CPU_BOOT_DEV_STS0_HBM_CRED_EN)) |
| 2405 | return; |
| 2406 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2407 | hbm0_wr = 0x33333333; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2408 | hbm0_rd = 0x77777777; |
Oded Gabbay | 0b168c8 | 2020-06-15 19:25:57 +0300 | [diff] [blame] | 2409 | hbm1_wr = 0x55555555; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2410 | hbm1_rd = 0xDDDDDDDD; |
| 2411 | |
| 2412 | WREG32(mmDMA_IF_E_N_HBM0_WR_CRED_CNT, hbm0_wr); |
| 2413 | WREG32(mmDMA_IF_E_N_HBM1_WR_CRED_CNT, hbm1_wr); |
| 2414 | WREG32(mmDMA_IF_E_N_HBM0_RD_CRED_CNT, hbm0_rd); |
| 2415 | WREG32(mmDMA_IF_E_N_HBM1_RD_CRED_CNT, hbm1_rd); |
| 2416 | |
| 2417 | WREG32(mmDMA_IF_E_S_HBM0_WR_CRED_CNT, hbm0_wr); |
| 2418 | WREG32(mmDMA_IF_E_S_HBM1_WR_CRED_CNT, hbm1_wr); |
| 2419 | WREG32(mmDMA_IF_E_S_HBM0_RD_CRED_CNT, hbm0_rd); |
| 2420 | WREG32(mmDMA_IF_E_S_HBM1_RD_CRED_CNT, hbm1_rd); |
| 2421 | |
| 2422 | WREG32(mmDMA_IF_W_N_HBM0_WR_CRED_CNT, hbm0_wr); |
| 2423 | WREG32(mmDMA_IF_W_N_HBM1_WR_CRED_CNT, hbm1_wr); |
| 2424 | WREG32(mmDMA_IF_W_N_HBM0_RD_CRED_CNT, hbm0_rd); |
| 2425 | WREG32(mmDMA_IF_W_N_HBM1_RD_CRED_CNT, hbm1_rd); |
| 2426 | |
| 2427 | WREG32(mmDMA_IF_W_S_HBM0_WR_CRED_CNT, hbm0_wr); |
| 2428 | WREG32(mmDMA_IF_W_S_HBM1_WR_CRED_CNT, hbm1_wr); |
| 2429 | WREG32(mmDMA_IF_W_S_HBM0_RD_CRED_CNT, hbm0_rd); |
| 2430 | WREG32(mmDMA_IF_W_S_HBM1_RD_CRED_CNT, hbm1_rd); |
| 2431 | |
| 2432 | WREG32(mmDMA_IF_E_N_HBM_CRED_EN_0, |
| 2433 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2434 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2435 | WREG32(mmDMA_IF_E_S_HBM_CRED_EN_0, |
| 2436 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2437 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2438 | WREG32(mmDMA_IF_W_N_HBM_CRED_EN_0, |
| 2439 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2440 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2441 | WREG32(mmDMA_IF_W_S_HBM_CRED_EN_0, |
| 2442 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2443 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2444 | |
| 2445 | WREG32(mmDMA_IF_E_N_HBM_CRED_EN_1, |
| 2446 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2447 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2448 | WREG32(mmDMA_IF_E_S_HBM_CRED_EN_1, |
| 2449 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2450 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2451 | WREG32(mmDMA_IF_W_N_HBM_CRED_EN_1, |
| 2452 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2453 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2454 | WREG32(mmDMA_IF_W_S_HBM_CRED_EN_1, |
| 2455 | (1 << DMA_IF_HBM_CRED_EN_READ_CREDIT_EN_SHIFT) | |
| 2456 | (1 << DMA_IF_HBM_CRED_EN_WRITE_CREDIT_EN_SHIFT)); |
| 2457 | } |
| 2458 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2459 | static void gaudi_init_golden_registers(struct hl_device *hdev) |
| 2460 | { |
| 2461 | u32 tpc_offset; |
| 2462 | int tpc_id, i; |
| 2463 | |
| 2464 | gaudi_init_e2e(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2465 | gaudi_init_hbm_cred(hdev); |
| 2466 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2467 | for (tpc_id = 0, tpc_offset = 0; |
| 2468 | tpc_id < TPC_NUMBER_OF_ENGINES; |
| 2469 | tpc_id++, tpc_offset += TPC_CFG_OFFSET) { |
| 2470 | /* Mask all arithmetic interrupts from TPC */ |
| 2471 | WREG32(mmTPC0_CFG_TPC_INTR_MASK + tpc_offset, 0x8FFF); |
| 2472 | /* Set 16 cache lines */ |
| 2473 | WREG32_FIELD(TPC0_CFG_MSS_CONFIG, tpc_offset, |
| 2474 | ICACHE_FETCH_LINE_NUM, 2); |
| 2475 | } |
| 2476 | |
| 2477 | /* Make sure 1st 128 bytes in SRAM are 0 for Tensor DMA */ |
| 2478 | for (i = 0 ; i < 128 ; i += 8) |
| 2479 | writeq(0, hdev->pcie_bar[SRAM_BAR_ID] + i); |
| 2480 | |
| 2481 | WREG32(mmMME0_CTRL_EUS_ROLLUP_CNT_ADD, 3); |
| 2482 | WREG32(mmMME1_CTRL_EUS_ROLLUP_CNT_ADD, 3); |
| 2483 | WREG32(mmMME2_CTRL_EUS_ROLLUP_CNT_ADD, 3); |
| 2484 | WREG32(mmMME3_CTRL_EUS_ROLLUP_CNT_ADD, 3); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2485 | } |
| 2486 | |
| 2487 | static void gaudi_init_pci_dma_qman(struct hl_device *hdev, int dma_id, |
| 2488 | int qman_id, dma_addr_t qman_pq_addr) |
| 2489 | { |
| 2490 | u32 mtr_base_en_lo, mtr_base_en_hi, mtr_base_ws_lo, mtr_base_ws_hi; |
| 2491 | u32 so_base_en_lo, so_base_en_hi, so_base_ws_lo, so_base_ws_hi; |
| 2492 | u32 q_off, dma_qm_offset; |
| 2493 | u32 dma_qm_err_cfg; |
| 2494 | |
| 2495 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 2496 | |
| 2497 | mtr_base_en_lo = lower_32_bits(CFG_BASE + |
| 2498 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2499 | mtr_base_en_hi = upper_32_bits(CFG_BASE + |
| 2500 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2501 | so_base_en_lo = lower_32_bits(CFG_BASE + |
| 2502 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2503 | so_base_en_hi = upper_32_bits(CFG_BASE + |
| 2504 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2505 | mtr_base_ws_lo = lower_32_bits(CFG_BASE + |
| 2506 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2507 | mtr_base_ws_hi = upper_32_bits(CFG_BASE + |
| 2508 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2509 | so_base_ws_lo = lower_32_bits(CFG_BASE + |
| 2510 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2511 | so_base_ws_hi = upper_32_bits(CFG_BASE + |
| 2512 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2513 | |
| 2514 | q_off = dma_qm_offset + qman_id * 4; |
| 2515 | |
| 2516 | WREG32(mmDMA0_QM_PQ_BASE_LO_0 + q_off, lower_32_bits(qman_pq_addr)); |
| 2517 | WREG32(mmDMA0_QM_PQ_BASE_HI_0 + q_off, upper_32_bits(qman_pq_addr)); |
| 2518 | |
| 2519 | WREG32(mmDMA0_QM_PQ_SIZE_0 + q_off, ilog2(HL_QUEUE_LENGTH)); |
| 2520 | WREG32(mmDMA0_QM_PQ_PI_0 + q_off, 0); |
| 2521 | WREG32(mmDMA0_QM_PQ_CI_0 + q_off, 0); |
| 2522 | |
Ofir Bitton | 25121d9 | 2020-09-24 08:22:58 +0300 | [diff] [blame] | 2523 | WREG32(mmDMA0_QM_CP_LDMA_TSIZE_OFFSET_0 + q_off, QMAN_LDMA_SIZE_OFFSET); |
| 2524 | WREG32(mmDMA0_QM_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 2525 | QMAN_LDMA_SRC_OFFSET); |
| 2526 | WREG32(mmDMA0_QM_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
| 2527 | QMAN_LDMA_DST_OFFSET); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2528 | |
| 2529 | WREG32(mmDMA0_QM_CP_MSG_BASE0_ADDR_LO_0 + q_off, mtr_base_en_lo); |
| 2530 | WREG32(mmDMA0_QM_CP_MSG_BASE0_ADDR_HI_0 + q_off, mtr_base_en_hi); |
| 2531 | WREG32(mmDMA0_QM_CP_MSG_BASE1_ADDR_LO_0 + q_off, so_base_en_lo); |
| 2532 | WREG32(mmDMA0_QM_CP_MSG_BASE1_ADDR_HI_0 + q_off, so_base_en_hi); |
| 2533 | WREG32(mmDMA0_QM_CP_MSG_BASE2_ADDR_LO_0 + q_off, mtr_base_ws_lo); |
| 2534 | WREG32(mmDMA0_QM_CP_MSG_BASE2_ADDR_HI_0 + q_off, mtr_base_ws_hi); |
| 2535 | WREG32(mmDMA0_QM_CP_MSG_BASE3_ADDR_LO_0 + q_off, so_base_ws_lo); |
| 2536 | WREG32(mmDMA0_QM_CP_MSG_BASE3_ADDR_HI_0 + q_off, so_base_ws_hi); |
| 2537 | |
Omer Shpigelman | ce04326 | 2020-06-16 17:56:27 +0300 | [diff] [blame] | 2538 | WREG32(mmDMA0_QM_CP_BARRIER_CFG_0 + q_off, 0x100); |
| 2539 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2540 | /* The following configuration is needed only once per QMAN */ |
| 2541 | if (qman_id == 0) { |
| 2542 | /* Configure RAZWI IRQ */ |
| 2543 | dma_qm_err_cfg = PCI_DMA_QMAN_GLBL_ERR_CFG_MSG_EN_MASK; |
| 2544 | if (hdev->stop_on_err) { |
| 2545 | dma_qm_err_cfg |= |
| 2546 | PCI_DMA_QMAN_GLBL_ERR_CFG_STOP_ON_ERR_EN_MASK; |
| 2547 | } |
| 2548 | |
| 2549 | WREG32(mmDMA0_QM_GLBL_ERR_CFG + dma_qm_offset, dma_qm_err_cfg); |
| 2550 | WREG32(mmDMA0_QM_GLBL_ERR_ADDR_LO + dma_qm_offset, |
| 2551 | lower_32_bits(CFG_BASE + |
| 2552 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2553 | WREG32(mmDMA0_QM_GLBL_ERR_ADDR_HI + dma_qm_offset, |
| 2554 | upper_32_bits(CFG_BASE + |
| 2555 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2556 | WREG32(mmDMA0_QM_GLBL_ERR_WDATA + dma_qm_offset, |
| 2557 | gaudi_irq_map_table[GAUDI_EVENT_DMA0_QM].cpu_id + |
| 2558 | dma_id); |
| 2559 | |
| 2560 | WREG32(mmDMA0_QM_ARB_ERR_MSG_EN + dma_qm_offset, |
| 2561 | QM_ARB_ERR_MSG_EN_MASK); |
| 2562 | |
| 2563 | /* Increase ARB WDT to support streams architecture */ |
| 2564 | WREG32(mmDMA0_QM_ARB_SLV_CHOISE_WDT + dma_qm_offset, |
| 2565 | GAUDI_ARB_WDT_TIMEOUT); |
| 2566 | |
| 2567 | WREG32(mmDMA0_QM_GLBL_PROT + dma_qm_offset, |
| 2568 | QMAN_EXTERNAL_MAKE_TRUSTED); |
| 2569 | |
| 2570 | WREG32(mmDMA0_QM_GLBL_CFG1 + dma_qm_offset, 0); |
| 2571 | } |
| 2572 | } |
| 2573 | |
| 2574 | static void gaudi_init_dma_core(struct hl_device *hdev, int dma_id) |
| 2575 | { |
| 2576 | u32 dma_offset = dma_id * DMA_CORE_OFFSET; |
| 2577 | u32 dma_err_cfg = 1 << DMA0_CORE_ERR_CFG_ERR_MSG_EN_SHIFT; |
| 2578 | |
| 2579 | /* Set to maximum possible according to physical size */ |
| 2580 | WREG32(mmDMA0_CORE_RD_MAX_OUTSTAND + dma_offset, 0); |
| 2581 | WREG32(mmDMA0_CORE_RD_MAX_SIZE + dma_offset, 0); |
| 2582 | |
Oded Gabbay | d1f3633 | 2020-09-14 09:26:54 +0300 | [diff] [blame] | 2583 | /* WA for H/W bug H3-2116 */ |
| 2584 | WREG32(mmDMA0_CORE_LBW_MAX_OUTSTAND + dma_offset, 15); |
| 2585 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2586 | /* STOP_ON bit implies no completion to operation in case of RAZWI */ |
| 2587 | if (hdev->stop_on_err) |
| 2588 | dma_err_cfg |= 1 << DMA0_CORE_ERR_CFG_STOP_ON_ERR_SHIFT; |
| 2589 | |
| 2590 | WREG32(mmDMA0_CORE_ERR_CFG + dma_offset, dma_err_cfg); |
| 2591 | WREG32(mmDMA0_CORE_ERRMSG_ADDR_LO + dma_offset, |
| 2592 | lower_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2593 | WREG32(mmDMA0_CORE_ERRMSG_ADDR_HI + dma_offset, |
| 2594 | upper_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2595 | WREG32(mmDMA0_CORE_ERRMSG_WDATA + dma_offset, |
| 2596 | gaudi_irq_map_table[GAUDI_EVENT_DMA0_CORE].cpu_id + dma_id); |
| 2597 | WREG32(mmDMA0_CORE_PROT + dma_offset, |
| 2598 | 1 << DMA0_CORE_PROT_ERR_VAL_SHIFT); |
| 2599 | /* If the channel is secured, it should be in MMU bypass mode */ |
| 2600 | WREG32(mmDMA0_CORE_SECURE_PROPS + dma_offset, |
| 2601 | 1 << DMA0_CORE_SECURE_PROPS_MMBP_SHIFT); |
| 2602 | WREG32(mmDMA0_CORE_CFG_0 + dma_offset, 1 << DMA0_CORE_CFG_0_EN_SHIFT); |
| 2603 | } |
| 2604 | |
| 2605 | static void gaudi_enable_qman(struct hl_device *hdev, int dma_id, |
| 2606 | u32 enable_mask) |
| 2607 | { |
| 2608 | u32 dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 2609 | |
| 2610 | WREG32(mmDMA0_QM_GLBL_CFG0 + dma_qm_offset, enable_mask); |
| 2611 | } |
| 2612 | |
| 2613 | static void gaudi_init_pci_dma_qmans(struct hl_device *hdev) |
| 2614 | { |
| 2615 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 2616 | struct hl_hw_queue *q; |
| 2617 | int i, j, dma_id, cpu_skip, nic_skip, cq_id = 0, q_idx, msi_vec = 0; |
| 2618 | |
| 2619 | if (gaudi->hw_cap_initialized & HW_CAP_PCI_DMA) |
| 2620 | return; |
| 2621 | |
| 2622 | for (i = 0 ; i < PCI_DMA_NUMBER_OF_CHNLS ; i++) { |
| 2623 | dma_id = gaudi_dma_assignment[i]; |
| 2624 | /* |
| 2625 | * For queues after the CPU Q need to add 1 to get the correct |
| 2626 | * queue. In addition, need to add the CPU EQ and NIC IRQs in |
| 2627 | * order to get the correct MSI register. |
| 2628 | */ |
| 2629 | if (dma_id > 1) { |
| 2630 | cpu_skip = 1; |
| 2631 | nic_skip = NIC_NUMBER_OF_ENGINES; |
| 2632 | } else { |
| 2633 | cpu_skip = 0; |
| 2634 | nic_skip = 0; |
| 2635 | } |
| 2636 | |
| 2637 | for (j = 0 ; j < QMAN_STREAMS ; j++) { |
| 2638 | q_idx = 4 * dma_id + j + cpu_skip; |
| 2639 | q = &hdev->kernel_queues[q_idx]; |
| 2640 | q->cq_id = cq_id++; |
| 2641 | q->msi_vec = nic_skip + cpu_skip + msi_vec++; |
| 2642 | gaudi_init_pci_dma_qman(hdev, dma_id, j, |
| 2643 | q->bus_address); |
| 2644 | } |
| 2645 | |
| 2646 | gaudi_init_dma_core(hdev, dma_id); |
| 2647 | |
| 2648 | gaudi_enable_qman(hdev, dma_id, PCI_DMA_QMAN_ENABLE); |
| 2649 | } |
| 2650 | |
| 2651 | gaudi->hw_cap_initialized |= HW_CAP_PCI_DMA; |
| 2652 | } |
| 2653 | |
| 2654 | static void gaudi_init_hbm_dma_qman(struct hl_device *hdev, int dma_id, |
| 2655 | int qman_id, u64 qman_base_addr) |
| 2656 | { |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2657 | u32 mtr_base_en_lo, mtr_base_en_hi, mtr_base_ws_lo, mtr_base_ws_hi; |
| 2658 | u32 so_base_en_lo, so_base_en_hi, so_base_ws_lo, so_base_ws_hi; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2659 | u32 q_off, dma_qm_offset; |
| 2660 | u32 dma_qm_err_cfg; |
| 2661 | |
| 2662 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 2663 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2664 | mtr_base_en_lo = lower_32_bits(CFG_BASE + |
| 2665 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2666 | mtr_base_en_hi = upper_32_bits(CFG_BASE + |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2667 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2668 | so_base_en_lo = lower_32_bits(CFG_BASE + |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2669 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2670 | so_base_en_hi = upper_32_bits(CFG_BASE + |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2671 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2672 | mtr_base_ws_lo = lower_32_bits(CFG_BASE + |
| 2673 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2674 | mtr_base_ws_hi = upper_32_bits(CFG_BASE + |
| 2675 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2676 | so_base_ws_lo = lower_32_bits(CFG_BASE + |
| 2677 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2678 | so_base_ws_hi = upper_32_bits(CFG_BASE + |
| 2679 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2680 | |
| 2681 | q_off = dma_qm_offset + qman_id * 4; |
| 2682 | |
| 2683 | if (qman_id < 4) { |
| 2684 | WREG32(mmDMA0_QM_PQ_BASE_LO_0 + q_off, |
| 2685 | lower_32_bits(qman_base_addr)); |
| 2686 | WREG32(mmDMA0_QM_PQ_BASE_HI_0 + q_off, |
| 2687 | upper_32_bits(qman_base_addr)); |
| 2688 | |
| 2689 | WREG32(mmDMA0_QM_PQ_SIZE_0 + q_off, ilog2(HBM_DMA_QMAN_LENGTH)); |
| 2690 | WREG32(mmDMA0_QM_PQ_PI_0 + q_off, 0); |
| 2691 | WREG32(mmDMA0_QM_PQ_CI_0 + q_off, 0); |
| 2692 | |
Ofir Bitton | 25121d9 | 2020-09-24 08:22:58 +0300 | [diff] [blame] | 2693 | WREG32(mmDMA0_QM_CP_LDMA_TSIZE_OFFSET_0 + q_off, |
| 2694 | QMAN_CPDMA_SIZE_OFFSET); |
| 2695 | WREG32(mmDMA0_QM_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 2696 | QMAN_CPDMA_SRC_OFFSET); |
| 2697 | WREG32(mmDMA0_QM_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
| 2698 | QMAN_CPDMA_DST_OFFSET); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2699 | } else { |
Ofir Bitton | 25121d9 | 2020-09-24 08:22:58 +0300 | [diff] [blame] | 2700 | WREG32(mmDMA0_QM_CP_LDMA_TSIZE_OFFSET_0 + q_off, |
| 2701 | QMAN_LDMA_SIZE_OFFSET); |
| 2702 | WREG32(mmDMA0_QM_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 2703 | QMAN_LDMA_SRC_OFFSET); |
| 2704 | WREG32(mmDMA0_QM_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
Oded Gabbay | 5b94d6e | 2020-09-25 20:14:15 +0300 | [diff] [blame] | 2705 | QMAN_LDMA_DST_OFFSET); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2706 | |
| 2707 | /* Configure RAZWI IRQ */ |
| 2708 | dma_qm_err_cfg = HBM_DMA_QMAN_GLBL_ERR_CFG_MSG_EN_MASK; |
| 2709 | if (hdev->stop_on_err) { |
| 2710 | dma_qm_err_cfg |= |
| 2711 | HBM_DMA_QMAN_GLBL_ERR_CFG_STOP_ON_ERR_EN_MASK; |
| 2712 | } |
| 2713 | WREG32(mmDMA0_QM_GLBL_ERR_CFG + dma_qm_offset, dma_qm_err_cfg); |
| 2714 | |
| 2715 | WREG32(mmDMA0_QM_GLBL_ERR_ADDR_LO + dma_qm_offset, |
| 2716 | lower_32_bits(CFG_BASE + |
| 2717 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2718 | WREG32(mmDMA0_QM_GLBL_ERR_ADDR_HI + dma_qm_offset, |
| 2719 | upper_32_bits(CFG_BASE + |
| 2720 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2721 | WREG32(mmDMA0_QM_GLBL_ERR_WDATA + dma_qm_offset, |
| 2722 | gaudi_irq_map_table[GAUDI_EVENT_DMA0_QM].cpu_id + |
| 2723 | dma_id); |
| 2724 | |
| 2725 | WREG32(mmDMA0_QM_ARB_ERR_MSG_EN + dma_qm_offset, |
| 2726 | QM_ARB_ERR_MSG_EN_MASK); |
| 2727 | |
| 2728 | /* Increase ARB WDT to support streams architecture */ |
| 2729 | WREG32(mmDMA0_QM_ARB_SLV_CHOISE_WDT + dma_qm_offset, |
| 2730 | GAUDI_ARB_WDT_TIMEOUT); |
| 2731 | |
| 2732 | WREG32(mmDMA0_QM_GLBL_CFG1 + dma_qm_offset, 0); |
| 2733 | WREG32(mmDMA0_QM_GLBL_PROT + dma_qm_offset, |
| 2734 | QMAN_INTERNAL_MAKE_TRUSTED); |
| 2735 | } |
| 2736 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2737 | WREG32(mmDMA0_QM_CP_MSG_BASE0_ADDR_LO_0 + q_off, mtr_base_en_lo); |
| 2738 | WREG32(mmDMA0_QM_CP_MSG_BASE0_ADDR_HI_0 + q_off, mtr_base_en_hi); |
| 2739 | WREG32(mmDMA0_QM_CP_MSG_BASE1_ADDR_LO_0 + q_off, so_base_en_lo); |
| 2740 | WREG32(mmDMA0_QM_CP_MSG_BASE1_ADDR_HI_0 + q_off, so_base_en_hi); |
| 2741 | |
| 2742 | /* Configure DMA5 CP_MSG_BASE 2/3 for sync stream collective */ |
| 2743 | if (gaudi_dma_assignment[dma_id] == GAUDI_ENGINE_ID_DMA_5) { |
| 2744 | WREG32(mmDMA0_QM_CP_MSG_BASE2_ADDR_LO_0 + q_off, |
| 2745 | mtr_base_ws_lo); |
| 2746 | WREG32(mmDMA0_QM_CP_MSG_BASE2_ADDR_HI_0 + q_off, |
| 2747 | mtr_base_ws_hi); |
| 2748 | WREG32(mmDMA0_QM_CP_MSG_BASE3_ADDR_LO_0 + q_off, |
| 2749 | so_base_ws_lo); |
| 2750 | WREG32(mmDMA0_QM_CP_MSG_BASE3_ADDR_HI_0 + q_off, |
| 2751 | so_base_ws_hi); |
| 2752 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2753 | } |
| 2754 | |
| 2755 | static void gaudi_init_hbm_dma_qmans(struct hl_device *hdev) |
| 2756 | { |
| 2757 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 2758 | struct gaudi_internal_qman_info *q; |
| 2759 | u64 qman_base_addr; |
| 2760 | int i, j, dma_id, internal_q_index; |
| 2761 | |
| 2762 | if (gaudi->hw_cap_initialized & HW_CAP_HBM_DMA) |
| 2763 | return; |
| 2764 | |
| 2765 | for (i = 0 ; i < HBM_DMA_NUMBER_OF_CHNLS ; i++) { |
| 2766 | dma_id = gaudi_dma_assignment[GAUDI_HBM_DMA_1 + i]; |
| 2767 | |
| 2768 | for (j = 0 ; j < QMAN_STREAMS ; j++) { |
| 2769 | /* |
| 2770 | * Add the CPU queue in order to get the correct queue |
| 2771 | * number as all internal queue are placed after it |
| 2772 | */ |
| 2773 | internal_q_index = dma_id * QMAN_STREAMS + j + 1; |
| 2774 | |
| 2775 | q = &gaudi->internal_qmans[internal_q_index]; |
| 2776 | qman_base_addr = (u64) q->pq_dma_addr; |
| 2777 | gaudi_init_hbm_dma_qman(hdev, dma_id, j, |
| 2778 | qman_base_addr); |
| 2779 | } |
| 2780 | |
| 2781 | /* Initializing lower CP for HBM DMA QMAN */ |
| 2782 | gaudi_init_hbm_dma_qman(hdev, dma_id, 4, 0); |
| 2783 | |
| 2784 | gaudi_init_dma_core(hdev, dma_id); |
| 2785 | |
| 2786 | gaudi_enable_qman(hdev, dma_id, HBM_DMA_QMAN_ENABLE); |
| 2787 | } |
| 2788 | |
| 2789 | gaudi->hw_cap_initialized |= HW_CAP_HBM_DMA; |
| 2790 | } |
| 2791 | |
| 2792 | static void gaudi_init_mme_qman(struct hl_device *hdev, u32 mme_offset, |
| 2793 | int qman_id, u64 qman_base_addr) |
| 2794 | { |
| 2795 | u32 mtr_base_lo, mtr_base_hi; |
| 2796 | u32 so_base_lo, so_base_hi; |
| 2797 | u32 q_off, mme_id; |
| 2798 | u32 mme_qm_err_cfg; |
| 2799 | |
| 2800 | mtr_base_lo = lower_32_bits(CFG_BASE + |
| 2801 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2802 | mtr_base_hi = upper_32_bits(CFG_BASE + |
| 2803 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2804 | so_base_lo = lower_32_bits(CFG_BASE + |
| 2805 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2806 | so_base_hi = upper_32_bits(CFG_BASE + |
| 2807 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2808 | |
| 2809 | q_off = mme_offset + qman_id * 4; |
| 2810 | |
| 2811 | if (qman_id < 4) { |
| 2812 | WREG32(mmMME0_QM_PQ_BASE_LO_0 + q_off, |
| 2813 | lower_32_bits(qman_base_addr)); |
| 2814 | WREG32(mmMME0_QM_PQ_BASE_HI_0 + q_off, |
| 2815 | upper_32_bits(qman_base_addr)); |
| 2816 | |
| 2817 | WREG32(mmMME0_QM_PQ_SIZE_0 + q_off, ilog2(MME_QMAN_LENGTH)); |
| 2818 | WREG32(mmMME0_QM_PQ_PI_0 + q_off, 0); |
| 2819 | WREG32(mmMME0_QM_PQ_CI_0 + q_off, 0); |
| 2820 | |
Ofir Bitton | 25121d9 | 2020-09-24 08:22:58 +0300 | [diff] [blame] | 2821 | WREG32(mmMME0_QM_CP_LDMA_TSIZE_OFFSET_0 + q_off, |
| 2822 | QMAN_CPDMA_SIZE_OFFSET); |
| 2823 | WREG32(mmMME0_QM_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 2824 | QMAN_CPDMA_SRC_OFFSET); |
| 2825 | WREG32(mmMME0_QM_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
| 2826 | QMAN_CPDMA_DST_OFFSET); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2827 | } else { |
Ofir Bitton | 25121d9 | 2020-09-24 08:22:58 +0300 | [diff] [blame] | 2828 | WREG32(mmMME0_QM_CP_LDMA_TSIZE_OFFSET_0 + q_off, |
| 2829 | QMAN_LDMA_SIZE_OFFSET); |
| 2830 | WREG32(mmMME0_QM_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 2831 | QMAN_LDMA_SRC_OFFSET); |
| 2832 | WREG32(mmMME0_QM_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
| 2833 | QMAN_LDMA_DST_OFFSET); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2834 | |
| 2835 | /* Configure RAZWI IRQ */ |
| 2836 | mme_id = mme_offset / |
| 2837 | (mmMME1_QM_GLBL_CFG0 - mmMME0_QM_GLBL_CFG0); |
| 2838 | |
| 2839 | mme_qm_err_cfg = MME_QMAN_GLBL_ERR_CFG_MSG_EN_MASK; |
| 2840 | if (hdev->stop_on_err) { |
| 2841 | mme_qm_err_cfg |= |
| 2842 | MME_QMAN_GLBL_ERR_CFG_STOP_ON_ERR_EN_MASK; |
| 2843 | } |
| 2844 | WREG32(mmMME0_QM_GLBL_ERR_CFG + mme_offset, mme_qm_err_cfg); |
| 2845 | WREG32(mmMME0_QM_GLBL_ERR_ADDR_LO + mme_offset, |
| 2846 | lower_32_bits(CFG_BASE + |
| 2847 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2848 | WREG32(mmMME0_QM_GLBL_ERR_ADDR_HI + mme_offset, |
| 2849 | upper_32_bits(CFG_BASE + |
| 2850 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2851 | WREG32(mmMME0_QM_GLBL_ERR_WDATA + mme_offset, |
| 2852 | gaudi_irq_map_table[GAUDI_EVENT_MME0_QM].cpu_id + |
| 2853 | mme_id); |
| 2854 | |
| 2855 | WREG32(mmMME0_QM_ARB_ERR_MSG_EN + mme_offset, |
| 2856 | QM_ARB_ERR_MSG_EN_MASK); |
| 2857 | |
| 2858 | /* Increase ARB WDT to support streams architecture */ |
| 2859 | WREG32(mmMME0_QM_ARB_SLV_CHOISE_WDT + mme_offset, |
| 2860 | GAUDI_ARB_WDT_TIMEOUT); |
| 2861 | |
| 2862 | WREG32(mmMME0_QM_GLBL_CFG1 + mme_offset, 0); |
| 2863 | WREG32(mmMME0_QM_GLBL_PROT + mme_offset, |
| 2864 | QMAN_INTERNAL_MAKE_TRUSTED); |
| 2865 | } |
| 2866 | |
| 2867 | WREG32(mmMME0_QM_CP_MSG_BASE0_ADDR_LO_0 + q_off, mtr_base_lo); |
| 2868 | WREG32(mmMME0_QM_CP_MSG_BASE0_ADDR_HI_0 + q_off, mtr_base_hi); |
| 2869 | WREG32(mmMME0_QM_CP_MSG_BASE1_ADDR_LO_0 + q_off, so_base_lo); |
| 2870 | WREG32(mmMME0_QM_CP_MSG_BASE1_ADDR_HI_0 + q_off, so_base_hi); |
| 2871 | } |
| 2872 | |
| 2873 | static void gaudi_init_mme_qmans(struct hl_device *hdev) |
| 2874 | { |
| 2875 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 2876 | struct gaudi_internal_qman_info *q; |
| 2877 | u64 qman_base_addr; |
| 2878 | u32 mme_offset; |
| 2879 | int i, internal_q_index; |
| 2880 | |
| 2881 | if (gaudi->hw_cap_initialized & HW_CAP_MME) |
| 2882 | return; |
| 2883 | |
| 2884 | /* |
| 2885 | * map GAUDI_QUEUE_ID_MME_0_X to the N_W_MME (mmMME2_QM_BASE) |
| 2886 | * and GAUDI_QUEUE_ID_MME_1_X to the S_W_MME (mmMME0_QM_BASE) |
| 2887 | */ |
| 2888 | |
| 2889 | mme_offset = mmMME2_QM_GLBL_CFG0 - mmMME0_QM_GLBL_CFG0; |
| 2890 | |
| 2891 | for (i = 0 ; i < MME_NUMBER_OF_QMANS ; i++) { |
| 2892 | internal_q_index = GAUDI_QUEUE_ID_MME_0_0 + i; |
| 2893 | q = &gaudi->internal_qmans[internal_q_index]; |
| 2894 | qman_base_addr = (u64) q->pq_dma_addr; |
| 2895 | gaudi_init_mme_qman(hdev, mme_offset, (i & 0x3), |
| 2896 | qman_base_addr); |
| 2897 | if (i == 3) |
| 2898 | mme_offset = 0; |
| 2899 | } |
| 2900 | |
| 2901 | /* Initializing lower CP for MME QMANs */ |
| 2902 | mme_offset = mmMME2_QM_GLBL_CFG0 - mmMME0_QM_GLBL_CFG0; |
| 2903 | gaudi_init_mme_qman(hdev, mme_offset, 4, 0); |
| 2904 | gaudi_init_mme_qman(hdev, 0, 4, 0); |
| 2905 | |
| 2906 | WREG32(mmMME2_QM_GLBL_CFG0, QMAN_MME_ENABLE); |
| 2907 | WREG32(mmMME0_QM_GLBL_CFG0, QMAN_MME_ENABLE); |
| 2908 | |
| 2909 | gaudi->hw_cap_initialized |= HW_CAP_MME; |
| 2910 | } |
| 2911 | |
| 2912 | static void gaudi_init_tpc_qman(struct hl_device *hdev, u32 tpc_offset, |
| 2913 | int qman_id, u64 qman_base_addr) |
| 2914 | { |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2915 | u32 mtr_base_en_lo, mtr_base_en_hi, mtr_base_ws_lo, mtr_base_ws_hi; |
| 2916 | u32 so_base_en_lo, so_base_en_hi, so_base_ws_lo, so_base_ws_hi; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2917 | u32 q_off, tpc_id; |
| 2918 | u32 tpc_qm_err_cfg; |
| 2919 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2920 | mtr_base_en_lo = lower_32_bits(CFG_BASE + |
| 2921 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2922 | mtr_base_en_hi = upper_32_bits(CFG_BASE + |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2923 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2924 | so_base_en_lo = lower_32_bits(CFG_BASE + |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2925 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2926 | so_base_en_hi = upper_32_bits(CFG_BASE + |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2927 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2928 | mtr_base_ws_lo = lower_32_bits(CFG_BASE + |
| 2929 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2930 | mtr_base_ws_hi = upper_32_bits(CFG_BASE + |
| 2931 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 2932 | so_base_ws_lo = lower_32_bits(CFG_BASE + |
| 2933 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 2934 | so_base_ws_hi = upper_32_bits(CFG_BASE + |
| 2935 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2936 | |
| 2937 | q_off = tpc_offset + qman_id * 4; |
| 2938 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2939 | tpc_id = tpc_offset / |
| 2940 | (mmTPC1_QM_GLBL_CFG0 - mmTPC0_QM_GLBL_CFG0); |
| 2941 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2942 | if (qman_id < 4) { |
| 2943 | WREG32(mmTPC0_QM_PQ_BASE_LO_0 + q_off, |
| 2944 | lower_32_bits(qman_base_addr)); |
| 2945 | WREG32(mmTPC0_QM_PQ_BASE_HI_0 + q_off, |
| 2946 | upper_32_bits(qman_base_addr)); |
| 2947 | |
| 2948 | WREG32(mmTPC0_QM_PQ_SIZE_0 + q_off, ilog2(TPC_QMAN_LENGTH)); |
| 2949 | WREG32(mmTPC0_QM_PQ_PI_0 + q_off, 0); |
| 2950 | WREG32(mmTPC0_QM_PQ_CI_0 + q_off, 0); |
| 2951 | |
Ofir Bitton | 25121d9 | 2020-09-24 08:22:58 +0300 | [diff] [blame] | 2952 | WREG32(mmTPC0_QM_CP_LDMA_TSIZE_OFFSET_0 + q_off, |
| 2953 | QMAN_CPDMA_SIZE_OFFSET); |
| 2954 | WREG32(mmTPC0_QM_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 2955 | QMAN_CPDMA_SRC_OFFSET); |
| 2956 | WREG32(mmTPC0_QM_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
| 2957 | QMAN_CPDMA_DST_OFFSET); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2958 | } else { |
Ofir Bitton | 25121d9 | 2020-09-24 08:22:58 +0300 | [diff] [blame] | 2959 | WREG32(mmTPC0_QM_CP_LDMA_TSIZE_OFFSET_0 + q_off, |
| 2960 | QMAN_LDMA_SIZE_OFFSET); |
| 2961 | WREG32(mmTPC0_QM_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 2962 | QMAN_LDMA_SRC_OFFSET); |
| 2963 | WREG32(mmTPC0_QM_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
| 2964 | QMAN_LDMA_DST_OFFSET); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2965 | |
| 2966 | /* Configure RAZWI IRQ */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 2967 | tpc_qm_err_cfg = TPC_QMAN_GLBL_ERR_CFG_MSG_EN_MASK; |
| 2968 | if (hdev->stop_on_err) { |
| 2969 | tpc_qm_err_cfg |= |
| 2970 | TPC_QMAN_GLBL_ERR_CFG_STOP_ON_ERR_EN_MASK; |
| 2971 | } |
| 2972 | |
| 2973 | WREG32(mmTPC0_QM_GLBL_ERR_CFG + tpc_offset, tpc_qm_err_cfg); |
| 2974 | WREG32(mmTPC0_QM_GLBL_ERR_ADDR_LO + tpc_offset, |
| 2975 | lower_32_bits(CFG_BASE + |
| 2976 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2977 | WREG32(mmTPC0_QM_GLBL_ERR_ADDR_HI + tpc_offset, |
| 2978 | upper_32_bits(CFG_BASE + |
| 2979 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 2980 | WREG32(mmTPC0_QM_GLBL_ERR_WDATA + tpc_offset, |
| 2981 | gaudi_irq_map_table[GAUDI_EVENT_TPC0_QM].cpu_id + |
| 2982 | tpc_id); |
| 2983 | |
| 2984 | WREG32(mmTPC0_QM_ARB_ERR_MSG_EN + tpc_offset, |
| 2985 | QM_ARB_ERR_MSG_EN_MASK); |
| 2986 | |
| 2987 | /* Increase ARB WDT to support streams architecture */ |
| 2988 | WREG32(mmTPC0_QM_ARB_SLV_CHOISE_WDT + tpc_offset, |
| 2989 | GAUDI_ARB_WDT_TIMEOUT); |
| 2990 | |
| 2991 | WREG32(mmTPC0_QM_GLBL_CFG1 + tpc_offset, 0); |
| 2992 | WREG32(mmTPC0_QM_GLBL_PROT + tpc_offset, |
| 2993 | QMAN_INTERNAL_MAKE_TRUSTED); |
| 2994 | } |
| 2995 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 2996 | WREG32(mmTPC0_QM_CP_MSG_BASE0_ADDR_LO_0 + q_off, mtr_base_en_lo); |
| 2997 | WREG32(mmTPC0_QM_CP_MSG_BASE0_ADDR_HI_0 + q_off, mtr_base_en_hi); |
| 2998 | WREG32(mmTPC0_QM_CP_MSG_BASE1_ADDR_LO_0 + q_off, so_base_en_lo); |
| 2999 | WREG32(mmTPC0_QM_CP_MSG_BASE1_ADDR_HI_0 + q_off, so_base_en_hi); |
| 3000 | |
| 3001 | /* Configure TPC7 CP_MSG_BASE 2/3 for sync stream collective */ |
| 3002 | if (tpc_id == 6) { |
| 3003 | WREG32(mmTPC0_QM_CP_MSG_BASE2_ADDR_LO_0 + q_off, |
| 3004 | mtr_base_ws_lo); |
| 3005 | WREG32(mmTPC0_QM_CP_MSG_BASE2_ADDR_HI_0 + q_off, |
| 3006 | mtr_base_ws_hi); |
| 3007 | WREG32(mmTPC0_QM_CP_MSG_BASE3_ADDR_LO_0 + q_off, |
| 3008 | so_base_ws_lo); |
| 3009 | WREG32(mmTPC0_QM_CP_MSG_BASE3_ADDR_HI_0 + q_off, |
| 3010 | so_base_ws_hi); |
| 3011 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3012 | } |
| 3013 | |
| 3014 | static void gaudi_init_tpc_qmans(struct hl_device *hdev) |
| 3015 | { |
| 3016 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3017 | struct gaudi_internal_qman_info *q; |
| 3018 | u64 qman_base_addr; |
| 3019 | u32 so_base_hi, tpc_offset = 0; |
| 3020 | u32 tpc_delta = mmTPC1_CFG_SM_BASE_ADDRESS_HIGH - |
| 3021 | mmTPC0_CFG_SM_BASE_ADDRESS_HIGH; |
| 3022 | int i, tpc_id, internal_q_index; |
| 3023 | |
| 3024 | if (gaudi->hw_cap_initialized & HW_CAP_TPC_MASK) |
| 3025 | return; |
| 3026 | |
| 3027 | so_base_hi = upper_32_bits(CFG_BASE + |
| 3028 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 3029 | |
| 3030 | for (tpc_id = 0 ; tpc_id < TPC_NUMBER_OF_ENGINES ; tpc_id++) { |
| 3031 | for (i = 0 ; i < QMAN_STREAMS ; i++) { |
| 3032 | internal_q_index = GAUDI_QUEUE_ID_TPC_0_0 + |
| 3033 | tpc_id * QMAN_STREAMS + i; |
| 3034 | q = &gaudi->internal_qmans[internal_q_index]; |
| 3035 | qman_base_addr = (u64) q->pq_dma_addr; |
| 3036 | gaudi_init_tpc_qman(hdev, tpc_offset, i, |
| 3037 | qman_base_addr); |
| 3038 | |
| 3039 | if (i == 3) { |
| 3040 | /* Initializing lower CP for TPC QMAN */ |
| 3041 | gaudi_init_tpc_qman(hdev, tpc_offset, 4, 0); |
| 3042 | |
| 3043 | /* Enable the QMAN and TPC channel */ |
| 3044 | WREG32(mmTPC0_QM_GLBL_CFG0 + tpc_offset, |
| 3045 | QMAN_TPC_ENABLE); |
| 3046 | } |
| 3047 | } |
| 3048 | |
| 3049 | WREG32(mmTPC0_CFG_SM_BASE_ADDRESS_HIGH + tpc_id * tpc_delta, |
| 3050 | so_base_hi); |
| 3051 | |
| 3052 | tpc_offset += mmTPC1_QM_GLBL_CFG0 - mmTPC0_QM_GLBL_CFG0; |
| 3053 | |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 3054 | gaudi->hw_cap_initialized |= |
| 3055 | FIELD_PREP(HW_CAP_TPC_MASK, 1 << tpc_id); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3056 | } |
| 3057 | } |
| 3058 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3059 | static void gaudi_init_nic_qman(struct hl_device *hdev, u32 nic_offset, |
| 3060 | int qman_id, u64 qman_base_addr, int nic_id) |
| 3061 | { |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 3062 | u32 mtr_base_en_lo, mtr_base_en_hi, mtr_base_ws_lo, mtr_base_ws_hi; |
| 3063 | u32 so_base_en_lo, so_base_en_hi, so_base_ws_lo, so_base_ws_hi; |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3064 | u32 q_off; |
| 3065 | u32 nic_qm_err_cfg; |
| 3066 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 3067 | mtr_base_en_lo = lower_32_bits(CFG_BASE + |
| 3068 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 3069 | mtr_base_en_hi = upper_32_bits(CFG_BASE + |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3070 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 3071 | so_base_en_lo = lower_32_bits(CFG_BASE + |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3072 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 3073 | so_base_en_hi = upper_32_bits(CFG_BASE + |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3074 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 3075 | mtr_base_ws_lo = lower_32_bits(CFG_BASE + |
| 3076 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 3077 | mtr_base_ws_hi = upper_32_bits(CFG_BASE + |
| 3078 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0); |
| 3079 | so_base_ws_lo = lower_32_bits(CFG_BASE + |
| 3080 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
| 3081 | so_base_ws_hi = upper_32_bits(CFG_BASE + |
| 3082 | mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0); |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3083 | |
| 3084 | q_off = nic_offset + qman_id * 4; |
| 3085 | |
| 3086 | WREG32(mmNIC0_QM0_PQ_BASE_LO_0 + q_off, lower_32_bits(qman_base_addr)); |
| 3087 | WREG32(mmNIC0_QM0_PQ_BASE_HI_0 + q_off, upper_32_bits(qman_base_addr)); |
| 3088 | |
| 3089 | WREG32(mmNIC0_QM0_PQ_SIZE_0 + q_off, ilog2(NIC_QMAN_LENGTH)); |
| 3090 | WREG32(mmNIC0_QM0_PQ_PI_0 + q_off, 0); |
| 3091 | WREG32(mmNIC0_QM0_PQ_CI_0 + q_off, 0); |
| 3092 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 3093 | WREG32(mmNIC0_QM0_CP_LDMA_TSIZE_OFFSET_0 + q_off, |
| 3094 | QMAN_LDMA_SIZE_OFFSET); |
| 3095 | WREG32(mmNIC0_QM0_CP_LDMA_SRC_BASE_LO_OFFSET_0 + q_off, |
| 3096 | QMAN_LDMA_SRC_OFFSET); |
| 3097 | WREG32(mmNIC0_QM0_CP_LDMA_DST_BASE_LO_OFFSET_0 + q_off, |
| 3098 | QMAN_LDMA_DST_OFFSET); |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3099 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 3100 | WREG32(mmNIC0_QM0_CP_MSG_BASE0_ADDR_LO_0 + q_off, mtr_base_en_lo); |
| 3101 | WREG32(mmNIC0_QM0_CP_MSG_BASE0_ADDR_HI_0 + q_off, mtr_base_en_hi); |
| 3102 | WREG32(mmNIC0_QM0_CP_MSG_BASE1_ADDR_LO_0 + q_off, so_base_en_lo); |
| 3103 | WREG32(mmNIC0_QM0_CP_MSG_BASE1_ADDR_HI_0 + q_off, so_base_en_hi); |
| 3104 | |
| 3105 | /* Configure NIC CP_MSG_BASE 2/3 for sync stream collective */ |
| 3106 | WREG32(mmNIC0_QM0_CP_MSG_BASE2_ADDR_LO_0 + q_off, mtr_base_ws_lo); |
| 3107 | WREG32(mmNIC0_QM0_CP_MSG_BASE2_ADDR_HI_0 + q_off, mtr_base_ws_hi); |
| 3108 | WREG32(mmNIC0_QM0_CP_MSG_BASE3_ADDR_LO_0 + q_off, so_base_ws_lo); |
| 3109 | WREG32(mmNIC0_QM0_CP_MSG_BASE3_ADDR_HI_0 + q_off, so_base_ws_hi); |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3110 | |
| 3111 | if (qman_id == 0) { |
| 3112 | /* Configure RAZWI IRQ */ |
| 3113 | nic_qm_err_cfg = NIC_QMAN_GLBL_ERR_CFG_MSG_EN_MASK; |
| 3114 | if (hdev->stop_on_err) { |
| 3115 | nic_qm_err_cfg |= |
| 3116 | NIC_QMAN_GLBL_ERR_CFG_STOP_ON_ERR_EN_MASK; |
| 3117 | } |
| 3118 | |
| 3119 | WREG32(mmNIC0_QM0_GLBL_ERR_CFG + nic_offset, nic_qm_err_cfg); |
| 3120 | WREG32(mmNIC0_QM0_GLBL_ERR_ADDR_LO + nic_offset, |
| 3121 | lower_32_bits(CFG_BASE + |
| 3122 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 3123 | WREG32(mmNIC0_QM0_GLBL_ERR_ADDR_HI + nic_offset, |
| 3124 | upper_32_bits(CFG_BASE + |
| 3125 | mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR)); |
| 3126 | WREG32(mmNIC0_QM0_GLBL_ERR_WDATA + nic_offset, |
| 3127 | gaudi_irq_map_table[GAUDI_EVENT_NIC0_QM0].cpu_id + |
| 3128 | nic_id); |
| 3129 | |
| 3130 | WREG32(mmNIC0_QM0_ARB_ERR_MSG_EN + nic_offset, |
| 3131 | QM_ARB_ERR_MSG_EN_MASK); |
| 3132 | |
| 3133 | /* Increase ARB WDT to support streams architecture */ |
| 3134 | WREG32(mmNIC0_QM0_ARB_SLV_CHOISE_WDT + nic_offset, |
| 3135 | GAUDI_ARB_WDT_TIMEOUT); |
| 3136 | |
| 3137 | WREG32(mmNIC0_QM0_GLBL_CFG1 + nic_offset, 0); |
| 3138 | WREG32(mmNIC0_QM0_GLBL_PROT + nic_offset, |
| 3139 | QMAN_INTERNAL_MAKE_TRUSTED); |
| 3140 | } |
| 3141 | } |
| 3142 | |
| 3143 | static void gaudi_init_nic_qmans(struct hl_device *hdev) |
| 3144 | { |
| 3145 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3146 | struct gaudi_internal_qman_info *q; |
| 3147 | u64 qman_base_addr; |
| 3148 | u32 nic_offset = 0; |
| 3149 | u32 nic_delta_between_qmans = |
| 3150 | mmNIC0_QM1_GLBL_CFG0 - mmNIC0_QM0_GLBL_CFG0; |
| 3151 | u32 nic_delta_between_nics = |
| 3152 | mmNIC1_QM0_GLBL_CFG0 - mmNIC0_QM0_GLBL_CFG0; |
| 3153 | int i, nic_id, internal_q_index; |
| 3154 | |
| 3155 | if (!hdev->nic_ports_mask) |
| 3156 | return; |
| 3157 | |
| 3158 | if (gaudi->hw_cap_initialized & HW_CAP_NIC_MASK) |
| 3159 | return; |
| 3160 | |
| 3161 | dev_dbg(hdev->dev, "Initializing NIC QMANs\n"); |
| 3162 | |
| 3163 | for (nic_id = 0 ; nic_id < NIC_NUMBER_OF_ENGINES ; nic_id++) { |
| 3164 | if (!(hdev->nic_ports_mask & (1 << nic_id))) { |
| 3165 | nic_offset += nic_delta_between_qmans; |
| 3166 | if (nic_id & 1) { |
| 3167 | nic_offset -= (nic_delta_between_qmans * 2); |
| 3168 | nic_offset += nic_delta_between_nics; |
| 3169 | } |
| 3170 | continue; |
| 3171 | } |
| 3172 | |
| 3173 | for (i = 0 ; i < QMAN_STREAMS ; i++) { |
| 3174 | internal_q_index = GAUDI_QUEUE_ID_NIC_0_0 + |
| 3175 | nic_id * QMAN_STREAMS + i; |
| 3176 | q = &gaudi->internal_qmans[internal_q_index]; |
| 3177 | qman_base_addr = (u64) q->pq_dma_addr; |
| 3178 | gaudi_init_nic_qman(hdev, nic_offset, (i & 0x3), |
| 3179 | qman_base_addr, nic_id); |
| 3180 | } |
| 3181 | |
| 3182 | /* Enable the QMAN */ |
| 3183 | WREG32(mmNIC0_QM0_GLBL_CFG0 + nic_offset, NIC_QMAN_ENABLE); |
| 3184 | |
| 3185 | nic_offset += nic_delta_between_qmans; |
| 3186 | if (nic_id & 1) { |
| 3187 | nic_offset -= (nic_delta_between_qmans * 2); |
| 3188 | nic_offset += nic_delta_between_nics; |
| 3189 | } |
| 3190 | |
| 3191 | gaudi->hw_cap_initialized |= 1 << (HW_CAP_NIC_SHIFT + nic_id); |
| 3192 | } |
| 3193 | } |
| 3194 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3195 | static void gaudi_disable_pci_dma_qmans(struct hl_device *hdev) |
| 3196 | { |
| 3197 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3198 | |
| 3199 | if (!(gaudi->hw_cap_initialized & HW_CAP_PCI_DMA)) |
| 3200 | return; |
| 3201 | |
| 3202 | WREG32(mmDMA0_QM_GLBL_CFG0, 0); |
| 3203 | WREG32(mmDMA1_QM_GLBL_CFG0, 0); |
| 3204 | WREG32(mmDMA5_QM_GLBL_CFG0, 0); |
| 3205 | } |
| 3206 | |
| 3207 | static void gaudi_disable_hbm_dma_qmans(struct hl_device *hdev) |
| 3208 | { |
| 3209 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3210 | |
| 3211 | if (!(gaudi->hw_cap_initialized & HW_CAP_HBM_DMA)) |
| 3212 | return; |
| 3213 | |
| 3214 | WREG32(mmDMA2_QM_GLBL_CFG0, 0); |
| 3215 | WREG32(mmDMA3_QM_GLBL_CFG0, 0); |
| 3216 | WREG32(mmDMA4_QM_GLBL_CFG0, 0); |
| 3217 | WREG32(mmDMA6_QM_GLBL_CFG0, 0); |
| 3218 | WREG32(mmDMA7_QM_GLBL_CFG0, 0); |
| 3219 | } |
| 3220 | |
| 3221 | static void gaudi_disable_mme_qmans(struct hl_device *hdev) |
| 3222 | { |
| 3223 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3224 | |
| 3225 | if (!(gaudi->hw_cap_initialized & HW_CAP_MME)) |
| 3226 | return; |
| 3227 | |
| 3228 | WREG32(mmMME2_QM_GLBL_CFG0, 0); |
| 3229 | WREG32(mmMME0_QM_GLBL_CFG0, 0); |
| 3230 | } |
| 3231 | |
| 3232 | static void gaudi_disable_tpc_qmans(struct hl_device *hdev) |
| 3233 | { |
| 3234 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3235 | u32 tpc_offset = 0; |
| 3236 | int tpc_id; |
| 3237 | |
| 3238 | if (!(gaudi->hw_cap_initialized & HW_CAP_TPC_MASK)) |
| 3239 | return; |
| 3240 | |
| 3241 | for (tpc_id = 0 ; tpc_id < TPC_NUMBER_OF_ENGINES ; tpc_id++) { |
| 3242 | WREG32(mmTPC0_QM_GLBL_CFG0 + tpc_offset, 0); |
| 3243 | tpc_offset += mmTPC1_QM_GLBL_CFG0 - mmTPC0_QM_GLBL_CFG0; |
| 3244 | } |
| 3245 | } |
| 3246 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3247 | static void gaudi_disable_nic_qmans(struct hl_device *hdev) |
| 3248 | { |
| 3249 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3250 | u32 nic_mask, nic_offset = 0; |
| 3251 | u32 nic_delta_between_qmans = |
| 3252 | mmNIC0_QM1_GLBL_CFG0 - mmNIC0_QM0_GLBL_CFG0; |
| 3253 | u32 nic_delta_between_nics = |
| 3254 | mmNIC1_QM0_GLBL_CFG0 - mmNIC0_QM0_GLBL_CFG0; |
| 3255 | int nic_id; |
| 3256 | |
| 3257 | for (nic_id = 0 ; nic_id < NIC_NUMBER_OF_ENGINES ; nic_id++) { |
| 3258 | nic_mask = 1 << (HW_CAP_NIC_SHIFT + nic_id); |
| 3259 | |
| 3260 | if (gaudi->hw_cap_initialized & nic_mask) |
| 3261 | WREG32(mmNIC0_QM0_GLBL_CFG0 + nic_offset, 0); |
| 3262 | |
| 3263 | nic_offset += nic_delta_between_qmans; |
| 3264 | if (nic_id & 1) { |
| 3265 | nic_offset -= (nic_delta_between_qmans * 2); |
| 3266 | nic_offset += nic_delta_between_nics; |
| 3267 | } |
| 3268 | } |
| 3269 | } |
| 3270 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3271 | static void gaudi_stop_pci_dma_qmans(struct hl_device *hdev) |
| 3272 | { |
| 3273 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3274 | |
| 3275 | if (!(gaudi->hw_cap_initialized & HW_CAP_PCI_DMA)) |
| 3276 | return; |
| 3277 | |
| 3278 | /* Stop upper CPs of QMANs 0.0 to 1.3 and 5.0 to 5.3 */ |
| 3279 | WREG32(mmDMA0_QM_GLBL_CFG1, 0xF << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3280 | WREG32(mmDMA1_QM_GLBL_CFG1, 0xF << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3281 | WREG32(mmDMA5_QM_GLBL_CFG1, 0xF << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3282 | } |
| 3283 | |
| 3284 | static void gaudi_stop_hbm_dma_qmans(struct hl_device *hdev) |
| 3285 | { |
| 3286 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3287 | |
| 3288 | if (!(gaudi->hw_cap_initialized & HW_CAP_HBM_DMA)) |
| 3289 | return; |
| 3290 | |
| 3291 | /* Stop CPs of HBM DMA QMANs */ |
| 3292 | |
| 3293 | WREG32(mmDMA2_QM_GLBL_CFG1, 0x1F << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3294 | WREG32(mmDMA3_QM_GLBL_CFG1, 0x1F << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3295 | WREG32(mmDMA4_QM_GLBL_CFG1, 0x1F << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3296 | WREG32(mmDMA6_QM_GLBL_CFG1, 0x1F << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3297 | WREG32(mmDMA7_QM_GLBL_CFG1, 0x1F << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3298 | } |
| 3299 | |
| 3300 | static void gaudi_stop_mme_qmans(struct hl_device *hdev) |
| 3301 | { |
| 3302 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3303 | |
| 3304 | if (!(gaudi->hw_cap_initialized & HW_CAP_MME)) |
| 3305 | return; |
| 3306 | |
| 3307 | /* Stop CPs of MME QMANs */ |
| 3308 | WREG32(mmMME2_QM_GLBL_CFG1, 0x1F << MME0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3309 | WREG32(mmMME0_QM_GLBL_CFG1, 0x1F << MME0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3310 | } |
| 3311 | |
| 3312 | static void gaudi_stop_tpc_qmans(struct hl_device *hdev) |
| 3313 | { |
| 3314 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3315 | |
| 3316 | if (!(gaudi->hw_cap_initialized & HW_CAP_TPC_MASK)) |
| 3317 | return; |
| 3318 | |
| 3319 | WREG32(mmTPC0_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3320 | WREG32(mmTPC1_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3321 | WREG32(mmTPC2_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3322 | WREG32(mmTPC3_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3323 | WREG32(mmTPC4_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3324 | WREG32(mmTPC5_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3325 | WREG32(mmTPC6_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3326 | WREG32(mmTPC7_QM_GLBL_CFG1, 0x1F << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 3327 | } |
| 3328 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3329 | static void gaudi_stop_nic_qmans(struct hl_device *hdev) |
| 3330 | { |
| 3331 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3332 | |
| 3333 | /* Stop upper CPs of QMANs */ |
| 3334 | |
| 3335 | if (gaudi->hw_cap_initialized & HW_CAP_NIC0) |
| 3336 | WREG32(mmNIC0_QM0_GLBL_CFG1, |
| 3337 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3338 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3339 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3340 | |
| 3341 | if (gaudi->hw_cap_initialized & HW_CAP_NIC1) |
| 3342 | WREG32(mmNIC0_QM1_GLBL_CFG1, |
| 3343 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3344 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3345 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3346 | |
| 3347 | if (gaudi->hw_cap_initialized & HW_CAP_NIC2) |
| 3348 | WREG32(mmNIC1_QM0_GLBL_CFG1, |
| 3349 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3350 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3351 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3352 | |
| 3353 | if (gaudi->hw_cap_initialized & HW_CAP_NIC3) |
| 3354 | WREG32(mmNIC1_QM1_GLBL_CFG1, |
| 3355 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3356 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3357 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3358 | |
| 3359 | if (gaudi->hw_cap_initialized & HW_CAP_NIC4) |
| 3360 | WREG32(mmNIC2_QM0_GLBL_CFG1, |
| 3361 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3362 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3363 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3364 | |
| 3365 | if (gaudi->hw_cap_initialized & HW_CAP_NIC5) |
| 3366 | WREG32(mmNIC2_QM1_GLBL_CFG1, |
| 3367 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3368 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3369 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3370 | |
| 3371 | if (gaudi->hw_cap_initialized & HW_CAP_NIC6) |
| 3372 | WREG32(mmNIC3_QM0_GLBL_CFG1, |
| 3373 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3374 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3375 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3376 | |
| 3377 | if (gaudi->hw_cap_initialized & HW_CAP_NIC7) |
| 3378 | WREG32(mmNIC3_QM1_GLBL_CFG1, |
| 3379 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3380 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3381 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3382 | |
| 3383 | if (gaudi->hw_cap_initialized & HW_CAP_NIC8) |
| 3384 | WREG32(mmNIC4_QM0_GLBL_CFG1, |
| 3385 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3386 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3387 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3388 | |
| 3389 | if (gaudi->hw_cap_initialized & HW_CAP_NIC9) |
| 3390 | WREG32(mmNIC4_QM1_GLBL_CFG1, |
| 3391 | NIC0_QM0_GLBL_CFG1_PQF_STOP_MASK | |
| 3392 | NIC0_QM0_GLBL_CFG1_CQF_STOP_MASK | |
| 3393 | NIC0_QM0_GLBL_CFG1_CP_STOP_MASK); |
| 3394 | } |
| 3395 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3396 | static void gaudi_pci_dma_stall(struct hl_device *hdev) |
| 3397 | { |
| 3398 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3399 | |
| 3400 | if (!(gaudi->hw_cap_initialized & HW_CAP_PCI_DMA)) |
| 3401 | return; |
| 3402 | |
| 3403 | WREG32(mmDMA0_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3404 | WREG32(mmDMA1_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3405 | WREG32(mmDMA5_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3406 | } |
| 3407 | |
| 3408 | static void gaudi_hbm_dma_stall(struct hl_device *hdev) |
| 3409 | { |
| 3410 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3411 | |
| 3412 | if (!(gaudi->hw_cap_initialized & HW_CAP_HBM_DMA)) |
| 3413 | return; |
| 3414 | |
| 3415 | WREG32(mmDMA2_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3416 | WREG32(mmDMA3_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3417 | WREG32(mmDMA4_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3418 | WREG32(mmDMA6_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3419 | WREG32(mmDMA7_CORE_CFG_1, 1 << DMA0_CORE_CFG_1_HALT_SHIFT); |
| 3420 | } |
| 3421 | |
| 3422 | static void gaudi_mme_stall(struct hl_device *hdev) |
| 3423 | { |
| 3424 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3425 | |
| 3426 | if (!(gaudi->hw_cap_initialized & HW_CAP_MME)) |
| 3427 | return; |
| 3428 | |
| 3429 | /* WA for H3-1800 bug: do ACC and SBAB writes twice */ |
| 3430 | WREG32(mmMME0_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3431 | WREG32(mmMME0_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3432 | WREG32(mmMME0_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3433 | WREG32(mmMME0_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3434 | WREG32(mmMME1_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3435 | WREG32(mmMME1_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3436 | WREG32(mmMME1_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3437 | WREG32(mmMME1_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3438 | WREG32(mmMME2_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3439 | WREG32(mmMME2_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3440 | WREG32(mmMME2_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3441 | WREG32(mmMME2_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3442 | WREG32(mmMME3_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3443 | WREG32(mmMME3_ACC_ACC_STALL, 1 << MME_ACC_ACC_STALL_R_SHIFT); |
| 3444 | WREG32(mmMME3_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3445 | WREG32(mmMME3_SBAB_SB_STALL, 1 << MME_SBAB_SB_STALL_R_SHIFT); |
| 3446 | } |
| 3447 | |
| 3448 | static void gaudi_tpc_stall(struct hl_device *hdev) |
| 3449 | { |
| 3450 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3451 | |
| 3452 | if (!(gaudi->hw_cap_initialized & HW_CAP_TPC_MASK)) |
| 3453 | return; |
| 3454 | |
| 3455 | WREG32(mmTPC0_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3456 | WREG32(mmTPC1_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3457 | WREG32(mmTPC2_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3458 | WREG32(mmTPC3_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3459 | WREG32(mmTPC4_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3460 | WREG32(mmTPC5_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3461 | WREG32(mmTPC6_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3462 | WREG32(mmTPC7_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); |
| 3463 | } |
| 3464 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3465 | static void gaudi_set_clock_gating(struct hl_device *hdev) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3466 | { |
| 3467 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3468 | u32 qman_offset; |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3469 | bool enable; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3470 | int i; |
| 3471 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3472 | /* In case we are during debug session, don't enable the clock gate |
| 3473 | * as it may interfere |
| 3474 | */ |
| 3475 | if (hdev->in_debug) |
| 3476 | return; |
| 3477 | |
Oded Gabbay | 0024c094 | 2020-12-05 22:55:09 +0200 | [diff] [blame] | 3478 | if (!hdev->asic_prop.fw_security_disabled) |
| 3479 | return; |
| 3480 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3481 | for (i = GAUDI_PCI_DMA_1, qman_offset = 0 ; i < GAUDI_HBM_DMA_1 ; i++) { |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3482 | enable = !!(hdev->clock_gating_mask & |
| 3483 | (BIT_ULL(gaudi_dma_assignment[i]))); |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3484 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3485 | qman_offset = gaudi_dma_assignment[i] * DMA_QMAN_OFFSET; |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3486 | WREG32(mmDMA0_QM_CGM_CFG1 + qman_offset, |
| 3487 | enable ? QMAN_CGM1_PWR_GATE_EN : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3488 | WREG32(mmDMA0_QM_CGM_CFG + qman_offset, |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3489 | enable ? QMAN_UPPER_CP_CGM_PWR_GATE_EN : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3490 | } |
| 3491 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3492 | for (i = GAUDI_HBM_DMA_1 ; i < GAUDI_DMA_MAX ; i++) { |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3493 | enable = !!(hdev->clock_gating_mask & |
| 3494 | (BIT_ULL(gaudi_dma_assignment[i]))); |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3495 | |
Oded Gabbay | da5dfbb | 2021-02-06 19:34:59 +0200 | [diff] [blame] | 3496 | /* GC sends work to DMA engine through Upper CP in DMA5 so |
| 3497 | * we need to not enable clock gating in that DMA |
| 3498 | */ |
| 3499 | if (i == GAUDI_HBM_DMA_4) |
| 3500 | enable = 0; |
| 3501 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3502 | qman_offset = gaudi_dma_assignment[i] * DMA_QMAN_OFFSET; |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3503 | WREG32(mmDMA0_QM_CGM_CFG1 + qman_offset, |
| 3504 | enable ? QMAN_CGM1_PWR_GATE_EN : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3505 | WREG32(mmDMA0_QM_CGM_CFG + qman_offset, |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3506 | enable ? QMAN_COMMON_CP_CGM_PWR_GATE_EN : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3507 | } |
| 3508 | |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3509 | enable = !!(hdev->clock_gating_mask & (BIT_ULL(GAUDI_ENGINE_ID_MME_0))); |
| 3510 | WREG32(mmMME0_QM_CGM_CFG1, enable ? QMAN_CGM1_PWR_GATE_EN : 0); |
| 3511 | WREG32(mmMME0_QM_CGM_CFG, enable ? QMAN_COMMON_CP_CGM_PWR_GATE_EN : 0); |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3512 | |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3513 | enable = !!(hdev->clock_gating_mask & (BIT_ULL(GAUDI_ENGINE_ID_MME_2))); |
| 3514 | WREG32(mmMME2_QM_CGM_CFG1, enable ? QMAN_CGM1_PWR_GATE_EN : 0); |
| 3515 | WREG32(mmMME2_QM_CGM_CFG, enable ? QMAN_COMMON_CP_CGM_PWR_GATE_EN : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3516 | |
| 3517 | for (i = 0, qman_offset = 0 ; i < TPC_NUMBER_OF_ENGINES ; i++) { |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3518 | enable = !!(hdev->clock_gating_mask & |
| 3519 | (BIT_ULL(GAUDI_ENGINE_ID_TPC_0 + i))); |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3520 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3521 | WREG32(mmTPC0_QM_CGM_CFG1 + qman_offset, |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3522 | enable ? QMAN_CGM1_PWR_GATE_EN : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3523 | WREG32(mmTPC0_QM_CGM_CFG + qman_offset, |
Ofir Bitton | f44d23b | 2020-08-04 13:38:43 +0300 | [diff] [blame] | 3524 | enable ? QMAN_COMMON_CP_CGM_PWR_GATE_EN : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3525 | |
| 3526 | qman_offset += TPC_QMAN_OFFSET; |
| 3527 | } |
| 3528 | |
| 3529 | gaudi->hw_cap_initialized |= HW_CAP_CLK_GATE; |
| 3530 | } |
| 3531 | |
| 3532 | static void gaudi_disable_clock_gating(struct hl_device *hdev) |
| 3533 | { |
| 3534 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3535 | u32 qman_offset; |
| 3536 | int i; |
| 3537 | |
Oded Gabbay | 0024c094 | 2020-12-05 22:55:09 +0200 | [diff] [blame] | 3538 | if (!hdev->asic_prop.fw_security_disabled) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3539 | return; |
| 3540 | |
| 3541 | for (i = 0, qman_offset = 0 ; i < DMA_NUMBER_OF_CHANNELS ; i++) { |
| 3542 | WREG32(mmDMA0_QM_CGM_CFG + qman_offset, 0); |
| 3543 | WREG32(mmDMA0_QM_CGM_CFG1 + qman_offset, 0); |
| 3544 | |
| 3545 | qman_offset += (mmDMA1_QM_CGM_CFG - mmDMA0_QM_CGM_CFG); |
| 3546 | } |
| 3547 | |
| 3548 | WREG32(mmMME0_QM_CGM_CFG, 0); |
| 3549 | WREG32(mmMME0_QM_CGM_CFG1, 0); |
| 3550 | WREG32(mmMME2_QM_CGM_CFG, 0); |
| 3551 | WREG32(mmMME2_QM_CGM_CFG1, 0); |
| 3552 | |
| 3553 | for (i = 0, qman_offset = 0 ; i < TPC_NUMBER_OF_ENGINES ; i++) { |
| 3554 | WREG32(mmTPC0_QM_CGM_CFG + qman_offset, 0); |
| 3555 | WREG32(mmTPC0_QM_CGM_CFG1 + qman_offset, 0); |
| 3556 | |
| 3557 | qman_offset += (mmTPC1_QM_CGM_CFG - mmTPC0_QM_CGM_CFG); |
| 3558 | } |
| 3559 | |
| 3560 | gaudi->hw_cap_initialized &= ~(HW_CAP_CLK_GATE); |
| 3561 | } |
| 3562 | |
| 3563 | static void gaudi_enable_timestamp(struct hl_device *hdev) |
| 3564 | { |
| 3565 | /* Disable the timestamp counter */ |
| 3566 | WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE, 0); |
| 3567 | |
| 3568 | /* Zero the lower/upper parts of the 64-bit counter */ |
| 3569 | WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE + 0xC, 0); |
| 3570 | WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE + 0x8, 0); |
| 3571 | |
| 3572 | /* Enable the counter */ |
| 3573 | WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE, 1); |
| 3574 | } |
| 3575 | |
| 3576 | static void gaudi_disable_timestamp(struct hl_device *hdev) |
| 3577 | { |
| 3578 | /* Disable the timestamp counter */ |
| 3579 | WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE, 0); |
| 3580 | } |
| 3581 | |
| 3582 | static void gaudi_halt_engines(struct hl_device *hdev, bool hard_reset) |
| 3583 | { |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3584 | u32 wait_timeout_ms; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3585 | |
| 3586 | dev_info(hdev->dev, |
| 3587 | "Halting compute engines and disabling interrupts\n"); |
| 3588 | |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3589 | if (hdev->pldm) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3590 | wait_timeout_ms = GAUDI_PLDM_RESET_WAIT_MSEC; |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3591 | else |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3592 | wait_timeout_ms = GAUDI_RESET_WAIT_MSEC; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3593 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3594 | gaudi_stop_nic_qmans(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3595 | gaudi_stop_mme_qmans(hdev); |
| 3596 | gaudi_stop_tpc_qmans(hdev); |
| 3597 | gaudi_stop_hbm_dma_qmans(hdev); |
| 3598 | gaudi_stop_pci_dma_qmans(hdev); |
| 3599 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3600 | hdev->asic_funcs->disable_clock_gating(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3601 | |
| 3602 | msleep(wait_timeout_ms); |
| 3603 | |
| 3604 | gaudi_pci_dma_stall(hdev); |
| 3605 | gaudi_hbm_dma_stall(hdev); |
| 3606 | gaudi_tpc_stall(hdev); |
| 3607 | gaudi_mme_stall(hdev); |
| 3608 | |
| 3609 | msleep(wait_timeout_ms); |
| 3610 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3611 | gaudi_disable_nic_qmans(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3612 | gaudi_disable_mme_qmans(hdev); |
| 3613 | gaudi_disable_tpc_qmans(hdev); |
| 3614 | gaudi_disable_hbm_dma_qmans(hdev); |
| 3615 | gaudi_disable_pci_dma_qmans(hdev); |
| 3616 | |
| 3617 | gaudi_disable_timestamp(hdev); |
| 3618 | |
Oded Gabbay | 12ae313 | 2020-07-03 20:58:23 +0300 | [diff] [blame] | 3619 | gaudi_disable_msi(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3620 | } |
| 3621 | |
| 3622 | static int gaudi_mmu_init(struct hl_device *hdev) |
| 3623 | { |
| 3624 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 3625 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3626 | u64 hop0_addr; |
| 3627 | int rc, i; |
| 3628 | |
| 3629 | if (!hdev->mmu_enable) |
| 3630 | return 0; |
| 3631 | |
| 3632 | if (gaudi->hw_cap_initialized & HW_CAP_MMU) |
| 3633 | return 0; |
| 3634 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3635 | for (i = 0 ; i < prop->max_asid ; i++) { |
| 3636 | hop0_addr = prop->mmu_pgt_addr + |
| 3637 | (i * prop->mmu_hop_table_size); |
| 3638 | |
| 3639 | rc = gaudi_mmu_update_asid_hop0_addr(hdev, i, hop0_addr); |
| 3640 | if (rc) { |
| 3641 | dev_err(hdev->dev, |
| 3642 | "failed to set hop0 addr for asid %d\n", i); |
| 3643 | goto err; |
| 3644 | } |
| 3645 | } |
| 3646 | |
| 3647 | /* init MMU cache manage page */ |
| 3648 | WREG32(mmSTLB_CACHE_INV_BASE_39_8, MMU_CACHE_MNG_ADDR >> 8); |
| 3649 | WREG32(mmSTLB_CACHE_INV_BASE_49_40, MMU_CACHE_MNG_ADDR >> 40); |
| 3650 | |
Tomer Tayar | 644883e | 2020-07-19 11:00:03 +0300 | [diff] [blame] | 3651 | hdev->asic_funcs->mmu_invalidate_cache(hdev, true, 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3652 | |
| 3653 | WREG32(mmMMU_UP_MMU_ENABLE, 1); |
| 3654 | WREG32(mmMMU_UP_SPI_MASK, 0xF); |
| 3655 | |
| 3656 | WREG32(mmSTLB_HOP_CONFIGURATION, |
| 3657 | hdev->mmu_huge_page_opt ? 0x30440 : 0x40440); |
| 3658 | |
Omer Shpigelman | cfd4176 | 2020-06-03 13:03:35 +0300 | [diff] [blame] | 3659 | /* |
| 3660 | * The H/W expects the first PI after init to be 1. After wraparound |
| 3661 | * we'll write 0. |
| 3662 | */ |
| 3663 | gaudi->mmu_cache_inv_pi = 1; |
| 3664 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3665 | gaudi->hw_cap_initialized |= HW_CAP_MMU; |
| 3666 | |
| 3667 | return 0; |
| 3668 | |
| 3669 | err: |
| 3670 | return rc; |
| 3671 | } |
| 3672 | |
| 3673 | static int gaudi_load_firmware_to_device(struct hl_device *hdev) |
| 3674 | { |
| 3675 | void __iomem *dst; |
| 3676 | |
| 3677 | /* HBM scrambler must be initialized before pushing F/W to HBM */ |
| 3678 | gaudi_init_scrambler_hbm(hdev); |
| 3679 | |
| 3680 | dst = hdev->pcie_bar[HBM_BAR_ID] + LINUX_FW_OFFSET; |
| 3681 | |
Ofir Bitton | 9bb86b6 | 2020-10-20 10:45:37 +0300 | [diff] [blame] | 3682 | return hl_fw_load_fw_to_device(hdev, GAUDI_LINUX_FW_FILE, dst, 0, 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3683 | } |
| 3684 | |
| 3685 | static int gaudi_load_boot_fit_to_device(struct hl_device *hdev) |
| 3686 | { |
| 3687 | void __iomem *dst; |
| 3688 | |
| 3689 | dst = hdev->pcie_bar[SRAM_BAR_ID] + BOOT_FIT_SRAM_OFFSET; |
| 3690 | |
Ofir Bitton | 9bb86b6 | 2020-10-20 10:45:37 +0300 | [diff] [blame] | 3691 | return hl_fw_load_fw_to_device(hdev, GAUDI_BOOT_FIT_FILE, dst, 0, 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3692 | } |
| 3693 | |
farah kassabri | eb10b89 | 2020-10-14 15:17:36 +0300 | [diff] [blame] | 3694 | static int gaudi_read_device_fw_version(struct hl_device *hdev, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3695 | enum hl_fw_component fwc) |
| 3696 | { |
| 3697 | const char *name; |
| 3698 | u32 ver_off; |
| 3699 | char *dest; |
| 3700 | |
| 3701 | switch (fwc) { |
| 3702 | case FW_COMP_UBOOT: |
| 3703 | ver_off = RREG32(mmUBOOT_VER_OFFSET); |
| 3704 | dest = hdev->asic_prop.uboot_ver; |
| 3705 | name = "U-Boot"; |
| 3706 | break; |
| 3707 | case FW_COMP_PREBOOT: |
| 3708 | ver_off = RREG32(mmPREBOOT_VER_OFFSET); |
| 3709 | dest = hdev->asic_prop.preboot_ver; |
| 3710 | name = "Preboot"; |
| 3711 | break; |
| 3712 | default: |
| 3713 | dev_warn(hdev->dev, "Undefined FW component: %d\n", fwc); |
farah kassabri | eb10b89 | 2020-10-14 15:17:36 +0300 | [diff] [blame] | 3714 | return -EIO; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3715 | } |
| 3716 | |
| 3717 | ver_off &= ~((u32)SRAM_BASE_ADDR); |
| 3718 | |
| 3719 | if (ver_off < SRAM_SIZE - VERSION_MAX_LEN) { |
| 3720 | memcpy_fromio(dest, hdev->pcie_bar[SRAM_BAR_ID] + ver_off, |
| 3721 | VERSION_MAX_LEN); |
| 3722 | } else { |
| 3723 | dev_err(hdev->dev, "%s version offset (0x%x) is above SRAM\n", |
| 3724 | name, ver_off); |
| 3725 | strcpy(dest, "unavailable"); |
farah kassabri | eb10b89 | 2020-10-14 15:17:36 +0300 | [diff] [blame] | 3726 | return -EIO; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3727 | } |
farah kassabri | eb10b89 | 2020-10-14 15:17:36 +0300 | [diff] [blame] | 3728 | |
| 3729 | return 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3730 | } |
| 3731 | |
Ohad Sharabi | a22f0ec | 2021-04-11 23:06:46 +0300 | [diff] [blame^] | 3732 | static void gaudi_init_firmware_loader(struct hl_device *hdev) |
| 3733 | { |
| 3734 | struct fw_load_mgr *fw_loader = &hdev->fw_loader; |
| 3735 | |
| 3736 | fw_loader->kmd_msg_to_cpu_reg = mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU; |
| 3737 | fw_loader->cpu_cmd_status_to_host_reg = mmCPU_CMD_STATUS_TO_HOST; |
| 3738 | fw_loader->cpu_timeout = GAUDI_CPU_TIMEOUT_USEC; |
| 3739 | fw_loader->boot_fit_timeout = GAUDI_BOOT_FIT_REQ_TIMEOUT_USEC; |
| 3740 | fw_loader->skip_bmc = !hdev->bmc_enable; |
| 3741 | fw_loader->cpu_boot_status_reg = mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS; |
| 3742 | fw_loader->cpu_boot_dev_status_reg = mmCPU_BOOT_DEV_STS0; |
| 3743 | fw_loader->boot_err0_reg = mmCPU_BOOT_ERR0; |
| 3744 | } |
| 3745 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3746 | static int gaudi_init_cpu(struct hl_device *hdev) |
| 3747 | { |
| 3748 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 3749 | int rc; |
| 3750 | |
Ofir Bitton | 6a2f5d7 | 2021-02-15 13:23:04 +0200 | [diff] [blame] | 3751 | if (!(hdev->fw_components & FW_TYPE_PREBOOT_CPU)) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3752 | return 0; |
| 3753 | |
| 3754 | if (gaudi->hw_cap_initialized & HW_CAP_CPU) |
| 3755 | return 0; |
| 3756 | |
| 3757 | /* |
| 3758 | * The device CPU works with 40 bits addresses. |
| 3759 | * This register sets the extension to 50 bits. |
| 3760 | */ |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 3761 | if (hdev->asic_prop.fw_security_disabled) |
| 3762 | WREG32(mmCPU_IF_CPU_MSB_ADDR, hdev->cpu_pci_msb_addr); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3763 | |
Ohad Sharabi | a22f0ec | 2021-04-11 23:06:46 +0300 | [diff] [blame^] | 3764 | rc = hl_fw_init_cpu(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3765 | |
| 3766 | if (rc) |
| 3767 | return rc; |
| 3768 | |
| 3769 | gaudi->hw_cap_initialized |= HW_CAP_CPU; |
| 3770 | |
| 3771 | return 0; |
| 3772 | } |
| 3773 | |
| 3774 | static int gaudi_init_cpu_queues(struct hl_device *hdev, u32 cpu_timeout) |
| 3775 | { |
| 3776 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ohad Sharabi | 5b6b780 | 2021-02-02 13:33:34 +0200 | [diff] [blame] | 3777 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3778 | struct hl_eq *eq; |
| 3779 | u32 status; |
| 3780 | struct hl_hw_queue *cpu_pq = |
| 3781 | &hdev->kernel_queues[GAUDI_QUEUE_ID_CPU_PQ]; |
| 3782 | int err; |
| 3783 | |
| 3784 | if (!hdev->cpu_queues_enable) |
| 3785 | return 0; |
| 3786 | |
| 3787 | if (gaudi->hw_cap_initialized & HW_CAP_CPU_Q) |
| 3788 | return 0; |
| 3789 | |
| 3790 | eq = &hdev->event_queue; |
| 3791 | |
| 3792 | WREG32(mmCPU_IF_PQ_BASE_ADDR_LOW, lower_32_bits(cpu_pq->bus_address)); |
| 3793 | WREG32(mmCPU_IF_PQ_BASE_ADDR_HIGH, upper_32_bits(cpu_pq->bus_address)); |
| 3794 | |
| 3795 | WREG32(mmCPU_IF_EQ_BASE_ADDR_LOW, lower_32_bits(eq->bus_address)); |
| 3796 | WREG32(mmCPU_IF_EQ_BASE_ADDR_HIGH, upper_32_bits(eq->bus_address)); |
| 3797 | |
| 3798 | WREG32(mmCPU_IF_CQ_BASE_ADDR_LOW, |
| 3799 | lower_32_bits(hdev->cpu_accessible_dma_address)); |
| 3800 | WREG32(mmCPU_IF_CQ_BASE_ADDR_HIGH, |
| 3801 | upper_32_bits(hdev->cpu_accessible_dma_address)); |
| 3802 | |
| 3803 | WREG32(mmCPU_IF_PQ_LENGTH, HL_QUEUE_SIZE_IN_BYTES); |
| 3804 | WREG32(mmCPU_IF_EQ_LENGTH, HL_EQ_SIZE_IN_BYTES); |
| 3805 | WREG32(mmCPU_IF_CQ_LENGTH, HL_CPU_ACCESSIBLE_MEM_SIZE); |
| 3806 | |
| 3807 | /* Used for EQ CI */ |
| 3808 | WREG32(mmCPU_IF_EQ_RD_OFFS, 0); |
| 3809 | |
| 3810 | WREG32(mmCPU_IF_PF_PQ_PI, 0); |
| 3811 | |
| 3812 | if (gaudi->multi_msi_mode) |
| 3813 | WREG32(mmCPU_IF_QUEUE_INIT, PQ_INIT_STATUS_READY_FOR_CP); |
| 3814 | else |
| 3815 | WREG32(mmCPU_IF_QUEUE_INIT, |
| 3816 | PQ_INIT_STATUS_READY_FOR_CP_SINGLE_MSI); |
| 3817 | |
| 3818 | WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR, GAUDI_EVENT_PI_UPDATE); |
| 3819 | |
| 3820 | err = hl_poll_timeout( |
| 3821 | hdev, |
| 3822 | mmCPU_IF_QUEUE_INIT, |
| 3823 | status, |
| 3824 | (status == PQ_INIT_STATUS_READY_FOR_HOST), |
| 3825 | 1000, |
| 3826 | cpu_timeout); |
| 3827 | |
| 3828 | if (err) { |
| 3829 | dev_err(hdev->dev, |
Oded Gabbay | 6138bbe | 2020-09-04 20:18:16 +0300 | [diff] [blame] | 3830 | "Failed to communicate with Device CPU (CPU-CP timeout)\n"); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3831 | return -EIO; |
| 3832 | } |
| 3833 | |
Ohad Sharabi | 5b6b780 | 2021-02-02 13:33:34 +0200 | [diff] [blame] | 3834 | /* update FW application security bits */ |
| 3835 | if (prop->fw_security_status_valid) |
| 3836 | prop->fw_app_security_map = RREG32(mmCPU_BOOT_DEV_STS0); |
| 3837 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3838 | gaudi->hw_cap_initialized |= HW_CAP_CPU_Q; |
| 3839 | return 0; |
| 3840 | } |
| 3841 | |
| 3842 | static void gaudi_pre_hw_init(struct hl_device *hdev) |
| 3843 | { |
| 3844 | /* Perform read from the device to make sure device is up */ |
Oded Gabbay | 377182a | 2020-12-09 19:50:46 +0200 | [diff] [blame] | 3845 | RREG32(mmHW_STATE); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3846 | |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 3847 | if (hdev->asic_prop.fw_security_disabled) { |
| 3848 | /* Set the access through PCI bars (Linux driver only) as |
| 3849 | * secured |
| 3850 | */ |
| 3851 | WREG32(mmPCIE_WRAP_LBW_PROT_OVR, |
| 3852 | (PCIE_WRAP_LBW_PROT_OVR_RD_EN_MASK | |
| 3853 | PCIE_WRAP_LBW_PROT_OVR_WR_EN_MASK)); |
Oded Gabbay | 57799ce | 2020-09-13 15:51:28 +0300 | [diff] [blame] | 3854 | |
Ofir Bitton | c692dec | 2020-10-04 17:34:37 +0300 | [diff] [blame] | 3855 | /* Perform read to flush the waiting writes to ensure |
| 3856 | * configuration was set in the device |
| 3857 | */ |
| 3858 | RREG32(mmPCIE_WRAP_LBW_PROT_OVR); |
| 3859 | } |
Oded Gabbay | 57799ce | 2020-09-13 15:51:28 +0300 | [diff] [blame] | 3860 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3861 | /* |
| 3862 | * Let's mark in the H/W that we have reached this point. We check |
| 3863 | * this value in the reset_before_init function to understand whether |
| 3864 | * we need to reset the chip before doing H/W init. This register is |
| 3865 | * cleared by the H/W upon H/W reset |
| 3866 | */ |
| 3867 | WREG32(mmHW_STATE, HL_DEVICE_HW_STATE_DIRTY); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3868 | } |
| 3869 | |
| 3870 | static int gaudi_hw_init(struct hl_device *hdev) |
| 3871 | { |
| 3872 | int rc; |
| 3873 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3874 | gaudi_pre_hw_init(hdev); |
| 3875 | |
| 3876 | gaudi_init_pci_dma_qmans(hdev); |
| 3877 | |
| 3878 | gaudi_init_hbm_dma_qmans(hdev); |
| 3879 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3880 | rc = gaudi_init_cpu(hdev); |
| 3881 | if (rc) { |
| 3882 | dev_err(hdev->dev, "failed to initialize CPU\n"); |
| 3883 | return rc; |
| 3884 | } |
| 3885 | |
Oded Gabbay | 0024c094 | 2020-12-05 22:55:09 +0200 | [diff] [blame] | 3886 | /* In case the clock gating was enabled in preboot we need to disable |
| 3887 | * it here before touching the MME/TPC registers. |
| 3888 | * There is no need to take clk gating mutex because when this function |
| 3889 | * runs, no other relevant code can run |
| 3890 | */ |
| 3891 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 3892 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3893 | /* SRAM scrambler must be initialized after CPU is running from HBM */ |
| 3894 | gaudi_init_scrambler_sram(hdev); |
| 3895 | |
| 3896 | /* This is here just in case we are working without CPU */ |
| 3897 | gaudi_init_scrambler_hbm(hdev); |
| 3898 | |
| 3899 | gaudi_init_golden_registers(hdev); |
| 3900 | |
| 3901 | rc = gaudi_mmu_init(hdev); |
| 3902 | if (rc) |
| 3903 | return rc; |
| 3904 | |
Omer Shpigelman | 3a3a5bf1 | 2020-05-11 10:45:12 +0300 | [diff] [blame] | 3905 | gaudi_init_security(hdev); |
| 3906 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3907 | gaudi_init_mme_qmans(hdev); |
| 3908 | |
| 3909 | gaudi_init_tpc_qmans(hdev); |
| 3910 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3911 | gaudi_init_nic_qmans(hdev); |
| 3912 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 3913 | hdev->asic_funcs->set_clock_gating(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3914 | |
| 3915 | gaudi_enable_timestamp(hdev); |
| 3916 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 3917 | /* MSI must be enabled before CPU queues and NIC are initialized */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3918 | rc = gaudi_enable_msi(hdev); |
| 3919 | if (rc) |
| 3920 | goto disable_queues; |
| 3921 | |
| 3922 | /* must be called after MSI was enabled */ |
| 3923 | rc = gaudi_init_cpu_queues(hdev, GAUDI_CPU_TIMEOUT_USEC); |
| 3924 | if (rc) { |
| 3925 | dev_err(hdev->dev, "failed to initialize CPU H/W queues %d\n", |
| 3926 | rc); |
| 3927 | goto disable_msi; |
| 3928 | } |
| 3929 | |
| 3930 | /* Perform read from the device to flush all configuration */ |
Oded Gabbay | 377182a | 2020-12-09 19:50:46 +0200 | [diff] [blame] | 3931 | RREG32(mmHW_STATE); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3932 | |
| 3933 | return 0; |
| 3934 | |
| 3935 | disable_msi: |
| 3936 | gaudi_disable_msi(hdev); |
| 3937 | disable_queues: |
| 3938 | gaudi_disable_mme_qmans(hdev); |
| 3939 | gaudi_disable_pci_dma_qmans(hdev); |
| 3940 | |
| 3941 | return rc; |
| 3942 | } |
| 3943 | |
| 3944 | static void gaudi_hw_fini(struct hl_device *hdev, bool hard_reset) |
| 3945 | { |
| 3946 | struct gaudi_device *gaudi = hdev->asic_specific; |
Igor Grinberg | b726a2f | 2020-10-29 14:06:54 +0200 | [diff] [blame] | 3947 | u32 status, reset_timeout_ms, cpu_timeout_ms; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3948 | |
Oded Gabbay | 12ae313 | 2020-07-03 20:58:23 +0300 | [diff] [blame] | 3949 | if (!hard_reset) { |
| 3950 | dev_err(hdev->dev, "GAUDI doesn't support soft-reset\n"); |
| 3951 | return; |
| 3952 | } |
| 3953 | |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3954 | if (hdev->pldm) { |
Oded Gabbay | 12ae313 | 2020-07-03 20:58:23 +0300 | [diff] [blame] | 3955 | reset_timeout_ms = GAUDI_PLDM_HRESET_TIMEOUT_MSEC; |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3956 | cpu_timeout_ms = GAUDI_PLDM_RESET_WAIT_MSEC; |
| 3957 | } else { |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 3958 | reset_timeout_ms = GAUDI_RESET_TIMEOUT_MSEC; |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3959 | cpu_timeout_ms = GAUDI_CPU_RESET_WAIT_MSEC; |
| 3960 | } |
| 3961 | |
| 3962 | /* Set device to handle FLR by H/W as we will put the device CPU to |
| 3963 | * halt mode |
| 3964 | */ |
Oded Gabbay | a63c3fb | 2020-11-26 18:11:05 +0200 | [diff] [blame] | 3965 | if (hdev->asic_prop.fw_security_disabled && |
| 3966 | !hdev->asic_prop.hard_reset_done_by_fw) |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 3967 | WREG32(mmPCIE_AUX_FLR_CTRL, (PCIE_AUX_FLR_CTRL_HW_CTRL_MASK | |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3968 | PCIE_AUX_FLR_CTRL_INT_MASK_MASK)); |
| 3969 | |
| 3970 | /* I don't know what is the state of the CPU so make sure it is |
| 3971 | * stopped in any means necessary |
| 3972 | */ |
Ofir Bitton | 9c9013c | 2020-12-01 10:39:54 +0200 | [diff] [blame] | 3973 | if (hdev->asic_prop.hard_reset_done_by_fw) |
| 3974 | WREG32(mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU, KMD_MSG_RST_DEV); |
| 3975 | else |
| 3976 | WREG32(mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU, KMD_MSG_GOTO_WFE); |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 3977 | |
Oded Gabbay | c83c417 | 2020-07-05 15:48:34 +0300 | [diff] [blame] | 3978 | WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR, GAUDI_EVENT_HALT_MACHINE); |
| 3979 | |
Oded Gabbay | a63c3fb | 2020-11-26 18:11:05 +0200 | [diff] [blame] | 3980 | if (hdev->asic_prop.fw_security_disabled && |
| 3981 | !hdev->asic_prop.hard_reset_done_by_fw) { |
| 3982 | |
| 3983 | /* Configure the reset registers. Must be done as early as |
| 3984 | * possible in case we fail during H/W initialization |
| 3985 | */ |
| 3986 | WREG32(mmPSOC_GLOBAL_CONF_SOFT_RST_CFG_H, |
| 3987 | (CFG_RST_H_DMA_MASK | |
| 3988 | CFG_RST_H_MME_MASK | |
| 3989 | CFG_RST_H_SM_MASK | |
| 3990 | CFG_RST_H_TPC_7_MASK)); |
| 3991 | |
| 3992 | WREG32(mmPSOC_GLOBAL_CONF_SOFT_RST_CFG_L, CFG_RST_L_TPC_MASK); |
| 3993 | |
| 3994 | WREG32(mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG_H, |
| 3995 | (CFG_RST_H_HBM_MASK | |
| 3996 | CFG_RST_H_TPC_7_MASK | |
| 3997 | CFG_RST_H_NIC_MASK | |
| 3998 | CFG_RST_H_SM_MASK | |
| 3999 | CFG_RST_H_DMA_MASK | |
| 4000 | CFG_RST_H_MME_MASK | |
| 4001 | CFG_RST_H_CPU_MASK | |
| 4002 | CFG_RST_H_MMU_MASK)); |
| 4003 | |
| 4004 | WREG32(mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG_L, |
| 4005 | (CFG_RST_L_IF_MASK | |
| 4006 | CFG_RST_L_PSOC_MASK | |
| 4007 | CFG_RST_L_TPC_MASK)); |
| 4008 | |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 4009 | msleep(cpu_timeout_ms); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4010 | |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 4011 | /* Tell ASIC not to re-initialize PCIe */ |
| 4012 | WREG32(mmPREBOOT_PCIE_EN, LKD_HARD_RESET_MAGIC); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4013 | |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 4014 | /* Restart BTL/BLR upon hard-reset */ |
| 4015 | if (hdev->asic_prop.fw_security_disabled) |
| 4016 | WREG32(mmPSOC_GLOBAL_CONF_BOOT_SEQ_RE_START, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4017 | |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 4018 | WREG32(mmPSOC_GLOBAL_CONF_SW_ALL_RST, |
Oded Gabbay | 12ae313 | 2020-07-03 20:58:23 +0300 | [diff] [blame] | 4019 | 1 << PSOC_GLOBAL_CONF_SW_ALL_RST_IND_SHIFT); |
Ofir Bitton | b90c894 | 2020-11-08 12:59:04 +0200 | [diff] [blame] | 4020 | |
Oded Gabbay | 13d0ee1 | 2020-12-06 23:48:45 +0200 | [diff] [blame] | 4021 | dev_info(hdev->dev, |
| 4022 | "Issued HARD reset command, going to wait %dms\n", |
| 4023 | reset_timeout_ms); |
| 4024 | } else { |
| 4025 | dev_info(hdev->dev, |
| 4026 | "Firmware performs HARD reset, going to wait %dms\n", |
| 4027 | reset_timeout_ms); |
| 4028 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4029 | |
| 4030 | /* |
| 4031 | * After hard reset, we can't poll the BTM_FSM register because the PSOC |
| 4032 | * itself is in reset. Need to wait until the reset is deasserted |
| 4033 | */ |
| 4034 | msleep(reset_timeout_ms); |
| 4035 | |
| 4036 | status = RREG32(mmPSOC_GLOBAL_CONF_BTM_FSM); |
| 4037 | if (status & PSOC_GLOBAL_CONF_BTM_FSM_STATE_MASK) |
| 4038 | dev_err(hdev->dev, |
| 4039 | "Timeout while waiting for device to reset 0x%x\n", |
| 4040 | status); |
| 4041 | |
farah kassabri | eb10b89 | 2020-10-14 15:17:36 +0300 | [diff] [blame] | 4042 | if (gaudi) { |
| 4043 | gaudi->hw_cap_initialized &= ~(HW_CAP_CPU | HW_CAP_CPU_Q | |
| 4044 | HW_CAP_HBM | HW_CAP_PCI_DMA | |
| 4045 | HW_CAP_MME | HW_CAP_TPC_MASK | |
| 4046 | HW_CAP_HBM_DMA | HW_CAP_PLL | |
| 4047 | HW_CAP_NIC_MASK | HW_CAP_MMU | |
| 4048 | HW_CAP_SRAM_SCRAMBLER | |
| 4049 | HW_CAP_HBM_SCRAMBLER | |
| 4050 | HW_CAP_CLK_GATE); |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 4051 | |
farah kassabri | eb10b89 | 2020-10-14 15:17:36 +0300 | [diff] [blame] | 4052 | memset(gaudi->events_stat, 0, sizeof(gaudi->events_stat)); |
| 4053 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4054 | } |
| 4055 | |
| 4056 | static int gaudi_suspend(struct hl_device *hdev) |
| 4057 | { |
| 4058 | int rc; |
| 4059 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 4060 | rc = hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_DISABLE_PCI_ACCESS); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4061 | if (rc) |
| 4062 | dev_err(hdev->dev, "Failed to disable PCI access from CPU\n"); |
| 4063 | |
| 4064 | return rc; |
| 4065 | } |
| 4066 | |
| 4067 | static int gaudi_resume(struct hl_device *hdev) |
| 4068 | { |
| 4069 | return gaudi_init_iatu(hdev); |
| 4070 | } |
| 4071 | |
| 4072 | static int gaudi_cb_mmap(struct hl_device *hdev, struct vm_area_struct *vma, |
Hillf Danton | 0db5753 | 2020-08-23 07:32:42 +0800 | [diff] [blame] | 4073 | void *cpu_addr, dma_addr_t dma_addr, size_t size) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4074 | { |
| 4075 | int rc; |
| 4076 | |
| 4077 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP | |
| 4078 | VM_DONTCOPY | VM_NORESERVE; |
| 4079 | |
Oded Gabbay | a9d4ef6 | 2021-01-11 13:49:38 +0200 | [diff] [blame] | 4080 | rc = dma_mmap_coherent(hdev->dev, vma, cpu_addr, |
| 4081 | (dma_addr - HOST_PHYS_BASE), size); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4082 | if (rc) |
Hillf Danton | 0db5753 | 2020-08-23 07:32:42 +0800 | [diff] [blame] | 4083 | dev_err(hdev->dev, "dma_mmap_coherent error %d", rc); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4084 | |
| 4085 | return rc; |
| 4086 | } |
| 4087 | |
| 4088 | static void gaudi_ring_doorbell(struct hl_device *hdev, u32 hw_queue_id, u32 pi) |
| 4089 | { |
| 4090 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 4091 | u32 db_reg_offset, db_value, dma_qm_offset, q_off; |
| 4092 | int dma_id; |
| 4093 | bool invalid_queue = false; |
| 4094 | |
| 4095 | switch (hw_queue_id) { |
| 4096 | case GAUDI_QUEUE_ID_DMA_0_0...GAUDI_QUEUE_ID_DMA_0_3: |
| 4097 | dma_id = gaudi_dma_assignment[GAUDI_PCI_DMA_1]; |
| 4098 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4099 | q_off = dma_qm_offset + (hw_queue_id & 0x3) * 4; |
| 4100 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4101 | break; |
| 4102 | |
| 4103 | case GAUDI_QUEUE_ID_DMA_1_0...GAUDI_QUEUE_ID_DMA_1_3: |
| 4104 | dma_id = gaudi_dma_assignment[GAUDI_PCI_DMA_2]; |
| 4105 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4106 | q_off = dma_qm_offset + (hw_queue_id & 0x3) * 4; |
| 4107 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4108 | break; |
| 4109 | |
| 4110 | case GAUDI_QUEUE_ID_DMA_2_0...GAUDI_QUEUE_ID_DMA_2_3: |
| 4111 | dma_id = gaudi_dma_assignment[GAUDI_HBM_DMA_1]; |
| 4112 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4113 | q_off = dma_qm_offset + ((hw_queue_id - 1) & 0x3) * 4; |
| 4114 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4115 | break; |
| 4116 | |
| 4117 | case GAUDI_QUEUE_ID_DMA_3_0...GAUDI_QUEUE_ID_DMA_3_3: |
| 4118 | dma_id = gaudi_dma_assignment[GAUDI_HBM_DMA_2]; |
| 4119 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4120 | q_off = dma_qm_offset + ((hw_queue_id - 1) & 0x3) * 4; |
| 4121 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4122 | break; |
| 4123 | |
| 4124 | case GAUDI_QUEUE_ID_DMA_4_0...GAUDI_QUEUE_ID_DMA_4_3: |
| 4125 | dma_id = gaudi_dma_assignment[GAUDI_HBM_DMA_3]; |
| 4126 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4127 | q_off = dma_qm_offset + ((hw_queue_id - 1) & 0x3) * 4; |
| 4128 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4129 | break; |
| 4130 | |
| 4131 | case GAUDI_QUEUE_ID_DMA_5_0...GAUDI_QUEUE_ID_DMA_5_3: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4132 | dma_id = gaudi_dma_assignment[GAUDI_HBM_DMA_4]; |
| 4133 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4134 | q_off = dma_qm_offset + ((hw_queue_id - 1) & 0x3) * 4; |
| 4135 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4136 | break; |
| 4137 | |
Ofir Bitton | 0940cab | 2020-08-31 08:52:56 +0300 | [diff] [blame] | 4138 | case GAUDI_QUEUE_ID_DMA_6_0...GAUDI_QUEUE_ID_DMA_6_3: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4139 | dma_id = gaudi_dma_assignment[GAUDI_HBM_DMA_5]; |
| 4140 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4141 | q_off = dma_qm_offset + ((hw_queue_id - 1) & 0x3) * 4; |
| 4142 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4143 | break; |
| 4144 | |
Ofir Bitton | 0940cab | 2020-08-31 08:52:56 +0300 | [diff] [blame] | 4145 | case GAUDI_QUEUE_ID_DMA_7_0...GAUDI_QUEUE_ID_DMA_7_3: |
| 4146 | dma_id = gaudi_dma_assignment[GAUDI_HBM_DMA_6]; |
| 4147 | dma_qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 4148 | q_off = dma_qm_offset + ((hw_queue_id - 1) & 0x3) * 4; |
| 4149 | db_reg_offset = mmDMA0_QM_PQ_PI_0 + q_off; |
| 4150 | break; |
| 4151 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4152 | case GAUDI_QUEUE_ID_CPU_PQ: |
| 4153 | if (gaudi->hw_cap_initialized & HW_CAP_CPU_Q) |
| 4154 | db_reg_offset = mmCPU_IF_PF_PQ_PI; |
| 4155 | else |
| 4156 | invalid_queue = true; |
| 4157 | break; |
| 4158 | |
| 4159 | case GAUDI_QUEUE_ID_MME_0_0: |
| 4160 | db_reg_offset = mmMME2_QM_PQ_PI_0; |
| 4161 | break; |
| 4162 | |
| 4163 | case GAUDI_QUEUE_ID_MME_0_1: |
| 4164 | db_reg_offset = mmMME2_QM_PQ_PI_1; |
| 4165 | break; |
| 4166 | |
| 4167 | case GAUDI_QUEUE_ID_MME_0_2: |
| 4168 | db_reg_offset = mmMME2_QM_PQ_PI_2; |
| 4169 | break; |
| 4170 | |
| 4171 | case GAUDI_QUEUE_ID_MME_0_3: |
| 4172 | db_reg_offset = mmMME2_QM_PQ_PI_3; |
| 4173 | break; |
| 4174 | |
| 4175 | case GAUDI_QUEUE_ID_MME_1_0: |
| 4176 | db_reg_offset = mmMME0_QM_PQ_PI_0; |
| 4177 | break; |
| 4178 | |
| 4179 | case GAUDI_QUEUE_ID_MME_1_1: |
| 4180 | db_reg_offset = mmMME0_QM_PQ_PI_1; |
| 4181 | break; |
| 4182 | |
| 4183 | case GAUDI_QUEUE_ID_MME_1_2: |
| 4184 | db_reg_offset = mmMME0_QM_PQ_PI_2; |
| 4185 | break; |
| 4186 | |
| 4187 | case GAUDI_QUEUE_ID_MME_1_3: |
| 4188 | db_reg_offset = mmMME0_QM_PQ_PI_3; |
| 4189 | break; |
| 4190 | |
| 4191 | case GAUDI_QUEUE_ID_TPC_0_0: |
| 4192 | db_reg_offset = mmTPC0_QM_PQ_PI_0; |
| 4193 | break; |
| 4194 | |
| 4195 | case GAUDI_QUEUE_ID_TPC_0_1: |
| 4196 | db_reg_offset = mmTPC0_QM_PQ_PI_1; |
| 4197 | break; |
| 4198 | |
| 4199 | case GAUDI_QUEUE_ID_TPC_0_2: |
| 4200 | db_reg_offset = mmTPC0_QM_PQ_PI_2; |
| 4201 | break; |
| 4202 | |
| 4203 | case GAUDI_QUEUE_ID_TPC_0_3: |
| 4204 | db_reg_offset = mmTPC0_QM_PQ_PI_3; |
| 4205 | break; |
| 4206 | |
| 4207 | case GAUDI_QUEUE_ID_TPC_1_0: |
| 4208 | db_reg_offset = mmTPC1_QM_PQ_PI_0; |
| 4209 | break; |
| 4210 | |
| 4211 | case GAUDI_QUEUE_ID_TPC_1_1: |
| 4212 | db_reg_offset = mmTPC1_QM_PQ_PI_1; |
| 4213 | break; |
| 4214 | |
| 4215 | case GAUDI_QUEUE_ID_TPC_1_2: |
| 4216 | db_reg_offset = mmTPC1_QM_PQ_PI_2; |
| 4217 | break; |
| 4218 | |
| 4219 | case GAUDI_QUEUE_ID_TPC_1_3: |
| 4220 | db_reg_offset = mmTPC1_QM_PQ_PI_3; |
| 4221 | break; |
| 4222 | |
| 4223 | case GAUDI_QUEUE_ID_TPC_2_0: |
| 4224 | db_reg_offset = mmTPC2_QM_PQ_PI_0; |
| 4225 | break; |
| 4226 | |
| 4227 | case GAUDI_QUEUE_ID_TPC_2_1: |
| 4228 | db_reg_offset = mmTPC2_QM_PQ_PI_1; |
| 4229 | break; |
| 4230 | |
| 4231 | case GAUDI_QUEUE_ID_TPC_2_2: |
| 4232 | db_reg_offset = mmTPC2_QM_PQ_PI_2; |
| 4233 | break; |
| 4234 | |
| 4235 | case GAUDI_QUEUE_ID_TPC_2_3: |
| 4236 | db_reg_offset = mmTPC2_QM_PQ_PI_3; |
| 4237 | break; |
| 4238 | |
| 4239 | case GAUDI_QUEUE_ID_TPC_3_0: |
| 4240 | db_reg_offset = mmTPC3_QM_PQ_PI_0; |
| 4241 | break; |
| 4242 | |
| 4243 | case GAUDI_QUEUE_ID_TPC_3_1: |
| 4244 | db_reg_offset = mmTPC3_QM_PQ_PI_1; |
| 4245 | break; |
| 4246 | |
| 4247 | case GAUDI_QUEUE_ID_TPC_3_2: |
| 4248 | db_reg_offset = mmTPC3_QM_PQ_PI_2; |
| 4249 | break; |
| 4250 | |
| 4251 | case GAUDI_QUEUE_ID_TPC_3_3: |
| 4252 | db_reg_offset = mmTPC3_QM_PQ_PI_3; |
| 4253 | break; |
| 4254 | |
| 4255 | case GAUDI_QUEUE_ID_TPC_4_0: |
| 4256 | db_reg_offset = mmTPC4_QM_PQ_PI_0; |
| 4257 | break; |
| 4258 | |
| 4259 | case GAUDI_QUEUE_ID_TPC_4_1: |
| 4260 | db_reg_offset = mmTPC4_QM_PQ_PI_1; |
| 4261 | break; |
| 4262 | |
| 4263 | case GAUDI_QUEUE_ID_TPC_4_2: |
| 4264 | db_reg_offset = mmTPC4_QM_PQ_PI_2; |
| 4265 | break; |
| 4266 | |
| 4267 | case GAUDI_QUEUE_ID_TPC_4_3: |
| 4268 | db_reg_offset = mmTPC4_QM_PQ_PI_3; |
| 4269 | break; |
| 4270 | |
| 4271 | case GAUDI_QUEUE_ID_TPC_5_0: |
| 4272 | db_reg_offset = mmTPC5_QM_PQ_PI_0; |
| 4273 | break; |
| 4274 | |
| 4275 | case GAUDI_QUEUE_ID_TPC_5_1: |
| 4276 | db_reg_offset = mmTPC5_QM_PQ_PI_1; |
| 4277 | break; |
| 4278 | |
| 4279 | case GAUDI_QUEUE_ID_TPC_5_2: |
| 4280 | db_reg_offset = mmTPC5_QM_PQ_PI_2; |
| 4281 | break; |
| 4282 | |
| 4283 | case GAUDI_QUEUE_ID_TPC_5_3: |
| 4284 | db_reg_offset = mmTPC5_QM_PQ_PI_3; |
| 4285 | break; |
| 4286 | |
| 4287 | case GAUDI_QUEUE_ID_TPC_6_0: |
| 4288 | db_reg_offset = mmTPC6_QM_PQ_PI_0; |
| 4289 | break; |
| 4290 | |
| 4291 | case GAUDI_QUEUE_ID_TPC_6_1: |
| 4292 | db_reg_offset = mmTPC6_QM_PQ_PI_1; |
| 4293 | break; |
| 4294 | |
| 4295 | case GAUDI_QUEUE_ID_TPC_6_2: |
| 4296 | db_reg_offset = mmTPC6_QM_PQ_PI_2; |
| 4297 | break; |
| 4298 | |
| 4299 | case GAUDI_QUEUE_ID_TPC_6_3: |
| 4300 | db_reg_offset = mmTPC6_QM_PQ_PI_3; |
| 4301 | break; |
| 4302 | |
| 4303 | case GAUDI_QUEUE_ID_TPC_7_0: |
| 4304 | db_reg_offset = mmTPC7_QM_PQ_PI_0; |
| 4305 | break; |
| 4306 | |
| 4307 | case GAUDI_QUEUE_ID_TPC_7_1: |
| 4308 | db_reg_offset = mmTPC7_QM_PQ_PI_1; |
| 4309 | break; |
| 4310 | |
| 4311 | case GAUDI_QUEUE_ID_TPC_7_2: |
| 4312 | db_reg_offset = mmTPC7_QM_PQ_PI_2; |
| 4313 | break; |
| 4314 | |
| 4315 | case GAUDI_QUEUE_ID_TPC_7_3: |
| 4316 | db_reg_offset = mmTPC7_QM_PQ_PI_3; |
| 4317 | break; |
| 4318 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 4319 | case GAUDI_QUEUE_ID_NIC_0_0: |
| 4320 | db_reg_offset = mmNIC0_QM0_PQ_PI_0; |
| 4321 | break; |
| 4322 | |
| 4323 | case GAUDI_QUEUE_ID_NIC_0_1: |
| 4324 | db_reg_offset = mmNIC0_QM0_PQ_PI_1; |
| 4325 | break; |
| 4326 | |
| 4327 | case GAUDI_QUEUE_ID_NIC_0_2: |
| 4328 | db_reg_offset = mmNIC0_QM0_PQ_PI_2; |
| 4329 | break; |
| 4330 | |
| 4331 | case GAUDI_QUEUE_ID_NIC_0_3: |
| 4332 | db_reg_offset = mmNIC0_QM0_PQ_PI_3; |
| 4333 | break; |
| 4334 | |
| 4335 | case GAUDI_QUEUE_ID_NIC_1_0: |
| 4336 | db_reg_offset = mmNIC0_QM1_PQ_PI_0; |
| 4337 | break; |
| 4338 | |
| 4339 | case GAUDI_QUEUE_ID_NIC_1_1: |
| 4340 | db_reg_offset = mmNIC0_QM1_PQ_PI_1; |
| 4341 | break; |
| 4342 | |
| 4343 | case GAUDI_QUEUE_ID_NIC_1_2: |
| 4344 | db_reg_offset = mmNIC0_QM1_PQ_PI_2; |
| 4345 | break; |
| 4346 | |
| 4347 | case GAUDI_QUEUE_ID_NIC_1_3: |
| 4348 | db_reg_offset = mmNIC0_QM1_PQ_PI_3; |
| 4349 | break; |
| 4350 | |
| 4351 | case GAUDI_QUEUE_ID_NIC_2_0: |
| 4352 | db_reg_offset = mmNIC1_QM0_PQ_PI_0; |
| 4353 | break; |
| 4354 | |
| 4355 | case GAUDI_QUEUE_ID_NIC_2_1: |
| 4356 | db_reg_offset = mmNIC1_QM0_PQ_PI_1; |
| 4357 | break; |
| 4358 | |
| 4359 | case GAUDI_QUEUE_ID_NIC_2_2: |
| 4360 | db_reg_offset = mmNIC1_QM0_PQ_PI_2; |
| 4361 | break; |
| 4362 | |
| 4363 | case GAUDI_QUEUE_ID_NIC_2_3: |
| 4364 | db_reg_offset = mmNIC1_QM0_PQ_PI_3; |
| 4365 | break; |
| 4366 | |
| 4367 | case GAUDI_QUEUE_ID_NIC_3_0: |
| 4368 | db_reg_offset = mmNIC1_QM1_PQ_PI_0; |
| 4369 | break; |
| 4370 | |
| 4371 | case GAUDI_QUEUE_ID_NIC_3_1: |
| 4372 | db_reg_offset = mmNIC1_QM1_PQ_PI_1; |
| 4373 | break; |
| 4374 | |
| 4375 | case GAUDI_QUEUE_ID_NIC_3_2: |
| 4376 | db_reg_offset = mmNIC1_QM1_PQ_PI_2; |
| 4377 | break; |
| 4378 | |
| 4379 | case GAUDI_QUEUE_ID_NIC_3_3: |
| 4380 | db_reg_offset = mmNIC1_QM1_PQ_PI_3; |
| 4381 | break; |
| 4382 | |
| 4383 | case GAUDI_QUEUE_ID_NIC_4_0: |
| 4384 | db_reg_offset = mmNIC2_QM0_PQ_PI_0; |
| 4385 | break; |
| 4386 | |
| 4387 | case GAUDI_QUEUE_ID_NIC_4_1: |
| 4388 | db_reg_offset = mmNIC2_QM0_PQ_PI_1; |
| 4389 | break; |
| 4390 | |
| 4391 | case GAUDI_QUEUE_ID_NIC_4_2: |
| 4392 | db_reg_offset = mmNIC2_QM0_PQ_PI_2; |
| 4393 | break; |
| 4394 | |
| 4395 | case GAUDI_QUEUE_ID_NIC_4_3: |
| 4396 | db_reg_offset = mmNIC2_QM0_PQ_PI_3; |
| 4397 | break; |
| 4398 | |
| 4399 | case GAUDI_QUEUE_ID_NIC_5_0: |
| 4400 | db_reg_offset = mmNIC2_QM1_PQ_PI_0; |
| 4401 | break; |
| 4402 | |
| 4403 | case GAUDI_QUEUE_ID_NIC_5_1: |
| 4404 | db_reg_offset = mmNIC2_QM1_PQ_PI_1; |
| 4405 | break; |
| 4406 | |
| 4407 | case GAUDI_QUEUE_ID_NIC_5_2: |
| 4408 | db_reg_offset = mmNIC2_QM1_PQ_PI_2; |
| 4409 | break; |
| 4410 | |
| 4411 | case GAUDI_QUEUE_ID_NIC_5_3: |
| 4412 | db_reg_offset = mmNIC2_QM1_PQ_PI_3; |
| 4413 | break; |
| 4414 | |
| 4415 | case GAUDI_QUEUE_ID_NIC_6_0: |
| 4416 | db_reg_offset = mmNIC3_QM0_PQ_PI_0; |
| 4417 | break; |
| 4418 | |
| 4419 | case GAUDI_QUEUE_ID_NIC_6_1: |
| 4420 | db_reg_offset = mmNIC3_QM0_PQ_PI_1; |
| 4421 | break; |
| 4422 | |
| 4423 | case GAUDI_QUEUE_ID_NIC_6_2: |
| 4424 | db_reg_offset = mmNIC3_QM0_PQ_PI_2; |
| 4425 | break; |
| 4426 | |
| 4427 | case GAUDI_QUEUE_ID_NIC_6_3: |
| 4428 | db_reg_offset = mmNIC3_QM0_PQ_PI_3; |
| 4429 | break; |
| 4430 | |
| 4431 | case GAUDI_QUEUE_ID_NIC_7_0: |
| 4432 | db_reg_offset = mmNIC3_QM1_PQ_PI_0; |
| 4433 | break; |
| 4434 | |
| 4435 | case GAUDI_QUEUE_ID_NIC_7_1: |
| 4436 | db_reg_offset = mmNIC3_QM1_PQ_PI_1; |
| 4437 | break; |
| 4438 | |
| 4439 | case GAUDI_QUEUE_ID_NIC_7_2: |
| 4440 | db_reg_offset = mmNIC3_QM1_PQ_PI_2; |
| 4441 | break; |
| 4442 | |
| 4443 | case GAUDI_QUEUE_ID_NIC_7_3: |
| 4444 | db_reg_offset = mmNIC3_QM1_PQ_PI_3; |
| 4445 | break; |
| 4446 | |
| 4447 | case GAUDI_QUEUE_ID_NIC_8_0: |
| 4448 | db_reg_offset = mmNIC4_QM0_PQ_PI_0; |
| 4449 | break; |
| 4450 | |
| 4451 | case GAUDI_QUEUE_ID_NIC_8_1: |
| 4452 | db_reg_offset = mmNIC4_QM0_PQ_PI_1; |
| 4453 | break; |
| 4454 | |
| 4455 | case GAUDI_QUEUE_ID_NIC_8_2: |
| 4456 | db_reg_offset = mmNIC4_QM0_PQ_PI_2; |
| 4457 | break; |
| 4458 | |
| 4459 | case GAUDI_QUEUE_ID_NIC_8_3: |
| 4460 | db_reg_offset = mmNIC4_QM0_PQ_PI_3; |
| 4461 | break; |
| 4462 | |
| 4463 | case GAUDI_QUEUE_ID_NIC_9_0: |
| 4464 | db_reg_offset = mmNIC4_QM1_PQ_PI_0; |
| 4465 | break; |
| 4466 | |
| 4467 | case GAUDI_QUEUE_ID_NIC_9_1: |
| 4468 | db_reg_offset = mmNIC4_QM1_PQ_PI_1; |
| 4469 | break; |
| 4470 | |
| 4471 | case GAUDI_QUEUE_ID_NIC_9_2: |
| 4472 | db_reg_offset = mmNIC4_QM1_PQ_PI_2; |
| 4473 | break; |
| 4474 | |
| 4475 | case GAUDI_QUEUE_ID_NIC_9_3: |
| 4476 | db_reg_offset = mmNIC4_QM1_PQ_PI_3; |
| 4477 | break; |
| 4478 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4479 | default: |
| 4480 | invalid_queue = true; |
| 4481 | } |
| 4482 | |
| 4483 | if (invalid_queue) { |
| 4484 | /* Should never get here */ |
| 4485 | dev_err(hdev->dev, "h/w queue %d is invalid. Can't set pi\n", |
| 4486 | hw_queue_id); |
| 4487 | return; |
| 4488 | } |
| 4489 | |
| 4490 | db_value = pi; |
| 4491 | |
| 4492 | /* ring the doorbell */ |
| 4493 | WREG32(db_reg_offset, db_value); |
| 4494 | |
Ofir Bitton | 5dbd7b4 | 2021-01-28 16:30:25 +0200 | [diff] [blame] | 4495 | if (hw_queue_id == GAUDI_QUEUE_ID_CPU_PQ) { |
| 4496 | /* make sure device CPU will read latest data from host */ |
| 4497 | mb(); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4498 | WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR, |
| 4499 | GAUDI_EVENT_PI_UPDATE); |
Ofir Bitton | 5dbd7b4 | 2021-01-28 16:30:25 +0200 | [diff] [blame] | 4500 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4501 | } |
| 4502 | |
| 4503 | static void gaudi_pqe_write(struct hl_device *hdev, __le64 *pqe, |
| 4504 | struct hl_bd *bd) |
| 4505 | { |
| 4506 | __le64 *pbd = (__le64 *) bd; |
| 4507 | |
| 4508 | /* The QMANs are on the host memory so a simple copy suffice */ |
| 4509 | pqe[0] = pbd[0]; |
| 4510 | pqe[1] = pbd[1]; |
| 4511 | } |
| 4512 | |
| 4513 | static void *gaudi_dma_alloc_coherent(struct hl_device *hdev, size_t size, |
| 4514 | dma_addr_t *dma_handle, gfp_t flags) |
| 4515 | { |
| 4516 | void *kernel_addr = dma_alloc_coherent(&hdev->pdev->dev, size, |
| 4517 | dma_handle, flags); |
| 4518 | |
| 4519 | /* Shift to the device's base physical address of host memory */ |
| 4520 | if (kernel_addr) |
| 4521 | *dma_handle += HOST_PHYS_BASE; |
| 4522 | |
| 4523 | return kernel_addr; |
| 4524 | } |
| 4525 | |
| 4526 | static void gaudi_dma_free_coherent(struct hl_device *hdev, size_t size, |
| 4527 | void *cpu_addr, dma_addr_t dma_handle) |
| 4528 | { |
| 4529 | /* Cancel the device's base physical address of host memory */ |
| 4530 | dma_addr_t fixed_dma_handle = dma_handle - HOST_PHYS_BASE; |
| 4531 | |
| 4532 | dma_free_coherent(&hdev->pdev->dev, size, cpu_addr, fixed_dma_handle); |
| 4533 | } |
| 4534 | |
farah kassabri | 03df136 | 2020-05-06 11:17:38 +0300 | [diff] [blame] | 4535 | static int gaudi_hbm_scrubbing(struct hl_device *hdev) |
| 4536 | { |
| 4537 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 4538 | u64 cur_addr = DRAM_BASE_ADDR_USER; |
| 4539 | u32 val; |
| 4540 | u32 chunk_size; |
| 4541 | int rc, dma_id; |
| 4542 | |
| 4543 | while (cur_addr < prop->dram_end_address) { |
| 4544 | for (dma_id = 0 ; dma_id < DMA_NUMBER_OF_CHANNELS ; dma_id++) { |
| 4545 | u32 dma_offset = dma_id * DMA_CORE_OFFSET; |
| 4546 | |
| 4547 | chunk_size = |
| 4548 | min((u64)SZ_2G, prop->dram_end_address - cur_addr); |
| 4549 | |
| 4550 | dev_dbg(hdev->dev, |
| 4551 | "Doing HBM scrubbing for 0x%09llx - 0x%09llx\n", |
| 4552 | cur_addr, cur_addr + chunk_size); |
| 4553 | |
| 4554 | WREG32(mmDMA0_CORE_SRC_BASE_LO + dma_offset, 0); |
| 4555 | WREG32(mmDMA0_CORE_SRC_BASE_HI + dma_offset, 0); |
| 4556 | WREG32(mmDMA0_CORE_DST_BASE_LO + dma_offset, |
| 4557 | lower_32_bits(cur_addr)); |
| 4558 | WREG32(mmDMA0_CORE_DST_BASE_HI + dma_offset, |
| 4559 | upper_32_bits(cur_addr)); |
| 4560 | WREG32(mmDMA0_CORE_DST_TSIZE_0 + dma_offset, |
| 4561 | chunk_size); |
| 4562 | WREG32(mmDMA0_CORE_COMMIT + dma_offset, |
| 4563 | ((1 << DMA0_CORE_COMMIT_LIN_SHIFT) | |
| 4564 | (1 << DMA0_CORE_COMMIT_MEM_SET_SHIFT))); |
| 4565 | |
| 4566 | cur_addr += chunk_size; |
| 4567 | |
| 4568 | if (cur_addr == prop->dram_end_address) |
| 4569 | break; |
| 4570 | } |
| 4571 | |
| 4572 | for (dma_id = 0 ; dma_id < DMA_NUMBER_OF_CHANNELS ; dma_id++) { |
| 4573 | u32 dma_offset = dma_id * DMA_CORE_OFFSET; |
| 4574 | |
| 4575 | rc = hl_poll_timeout( |
| 4576 | hdev, |
| 4577 | mmDMA0_CORE_STS0 + dma_offset, |
| 4578 | val, |
| 4579 | ((val & DMA0_CORE_STS0_BUSY_MASK) == 0), |
| 4580 | 1000, |
| 4581 | HBM_SCRUBBING_TIMEOUT_US); |
| 4582 | |
| 4583 | if (rc) { |
| 4584 | dev_err(hdev->dev, |
| 4585 | "DMA Timeout during HBM scrubbing of DMA #%d\n", |
| 4586 | dma_id); |
| 4587 | return -EIO; |
| 4588 | } |
| 4589 | } |
| 4590 | } |
| 4591 | |
| 4592 | return 0; |
| 4593 | } |
| 4594 | |
| 4595 | static int gaudi_scrub_device_mem(struct hl_device *hdev, u64 addr, u64 size) |
| 4596 | { |
| 4597 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 4598 | struct gaudi_device *gaudi = hdev->asic_specific; |
farah kassabri | 03df136 | 2020-05-06 11:17:38 +0300 | [diff] [blame] | 4599 | int rc = 0; |
| 4600 | u64 val = 0; |
| 4601 | |
| 4602 | if (!hdev->memory_scrub) |
| 4603 | return 0; |
| 4604 | |
| 4605 | if (!addr && !size) { |
| 4606 | /* Wait till device is idle */ |
| 4607 | rc = hl_poll_timeout( |
| 4608 | hdev, |
| 4609 | mmDMA0_CORE_STS0/* dummy */, |
| 4610 | val/* dummy */, |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 4611 | (hdev->asic_funcs->is_device_idle(hdev, NULL, |
| 4612 | 0, NULL)), |
farah kassabri | 03df136 | 2020-05-06 11:17:38 +0300 | [diff] [blame] | 4613 | 1000, |
| 4614 | HBM_SCRUBBING_TIMEOUT_US); |
| 4615 | if (rc) { |
| 4616 | dev_err(hdev->dev, "waiting for idle timeout\n"); |
| 4617 | return -EIO; |
| 4618 | } |
| 4619 | |
| 4620 | /* Scrub SRAM */ |
| 4621 | addr = prop->sram_user_base_address; |
| 4622 | size = hdev->pldm ? 0x10000 : |
| 4623 | (prop->sram_size - SRAM_USER_BASE_OFFSET); |
| 4624 | val = 0x7777777777777777ull; |
| 4625 | |
| 4626 | rc = gaudi_memset_device_memory(hdev, addr, size, val); |
| 4627 | if (rc) { |
| 4628 | dev_err(hdev->dev, |
| 4629 | "Failed to clear SRAM in mem scrub all\n"); |
| 4630 | return rc; |
| 4631 | } |
| 4632 | |
| 4633 | mutex_lock(&gaudi->clk_gate_mutex); |
| 4634 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 4635 | |
| 4636 | /* Scrub HBM using all DMA channels in parallel */ |
| 4637 | rc = gaudi_hbm_scrubbing(hdev); |
| 4638 | if (rc) |
| 4639 | dev_err(hdev->dev, |
| 4640 | "Failed to clear HBM in mem scrub all\n"); |
| 4641 | |
| 4642 | hdev->asic_funcs->set_clock_gating(hdev); |
| 4643 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 4644 | } |
| 4645 | |
| 4646 | return rc; |
| 4647 | } |
| 4648 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4649 | static void *gaudi_get_int_queue_base(struct hl_device *hdev, |
| 4650 | u32 queue_id, dma_addr_t *dma_handle, |
| 4651 | u16 *queue_len) |
| 4652 | { |
| 4653 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 4654 | struct gaudi_internal_qman_info *q; |
| 4655 | |
| 4656 | if (queue_id >= GAUDI_QUEUE_ID_SIZE || |
| 4657 | gaudi_queue_type[queue_id] != QUEUE_TYPE_INT) { |
| 4658 | dev_err(hdev->dev, "Got invalid queue id %d\n", queue_id); |
| 4659 | return NULL; |
| 4660 | } |
| 4661 | |
| 4662 | q = &gaudi->internal_qmans[queue_id]; |
| 4663 | *dma_handle = q->pq_dma_addr; |
| 4664 | *queue_len = q->pq_size / QMAN_PQ_ENTRY_SIZE; |
| 4665 | |
| 4666 | return q->pq_kernel_addr; |
| 4667 | } |
| 4668 | |
| 4669 | static int gaudi_send_cpu_message(struct hl_device *hdev, u32 *msg, |
Alon Mizrahi | 439bc47 | 2020-11-10 13:49:10 +0200 | [diff] [blame] | 4670 | u16 len, u32 timeout, u64 *result) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4671 | { |
| 4672 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 4673 | |
| 4674 | if (!(gaudi->hw_cap_initialized & HW_CAP_CPU_Q)) { |
| 4675 | if (result) |
| 4676 | *result = 0; |
| 4677 | return 0; |
| 4678 | } |
| 4679 | |
Oded Gabbay | 788cacf | 2020-07-07 17:30:13 +0300 | [diff] [blame] | 4680 | if (!timeout) |
| 4681 | timeout = GAUDI_MSG_TO_CPU_TIMEOUT_USEC; |
| 4682 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4683 | return hl_fw_send_cpu_message(hdev, GAUDI_QUEUE_ID_CPU_PQ, msg, len, |
| 4684 | timeout, result); |
| 4685 | } |
| 4686 | |
| 4687 | static int gaudi_test_queue(struct hl_device *hdev, u32 hw_queue_id) |
| 4688 | { |
| 4689 | struct packet_msg_prot *fence_pkt; |
| 4690 | dma_addr_t pkt_dma_addr; |
| 4691 | u32 fence_val, tmp, timeout_usec; |
| 4692 | dma_addr_t fence_dma_addr; |
| 4693 | u32 *fence_ptr; |
| 4694 | int rc; |
| 4695 | |
| 4696 | if (hdev->pldm) |
| 4697 | timeout_usec = GAUDI_PLDM_TEST_QUEUE_WAIT_USEC; |
| 4698 | else |
| 4699 | timeout_usec = GAUDI_TEST_QUEUE_WAIT_USEC; |
| 4700 | |
| 4701 | fence_val = GAUDI_QMAN0_FENCE_VAL; |
| 4702 | |
| 4703 | fence_ptr = hdev->asic_funcs->asic_dma_pool_zalloc(hdev, 4, GFP_KERNEL, |
| 4704 | &fence_dma_addr); |
| 4705 | if (!fence_ptr) { |
| 4706 | dev_err(hdev->dev, |
Dotan Barak | d6b045c | 2020-08-06 09:20:49 +0300 | [diff] [blame] | 4707 | "Failed to allocate memory for H/W queue %d testing\n", |
| 4708 | hw_queue_id); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4709 | return -ENOMEM; |
| 4710 | } |
| 4711 | |
| 4712 | *fence_ptr = 0; |
| 4713 | |
| 4714 | fence_pkt = hdev->asic_funcs->asic_dma_pool_zalloc(hdev, |
| 4715 | sizeof(struct packet_msg_prot), |
| 4716 | GFP_KERNEL, &pkt_dma_addr); |
| 4717 | if (!fence_pkt) { |
| 4718 | dev_err(hdev->dev, |
Dotan Barak | d6b045c | 2020-08-06 09:20:49 +0300 | [diff] [blame] | 4719 | "Failed to allocate packet for H/W queue %d testing\n", |
| 4720 | hw_queue_id); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4721 | rc = -ENOMEM; |
| 4722 | goto free_fence_ptr; |
| 4723 | } |
| 4724 | |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 4725 | tmp = FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_PROT); |
| 4726 | tmp |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 1); |
| 4727 | tmp |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
| 4728 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4729 | fence_pkt->ctl = cpu_to_le32(tmp); |
| 4730 | fence_pkt->value = cpu_to_le32(fence_val); |
| 4731 | fence_pkt->addr = cpu_to_le64(fence_dma_addr); |
| 4732 | |
| 4733 | rc = hl_hw_queue_send_cb_no_cmpl(hdev, hw_queue_id, |
| 4734 | sizeof(struct packet_msg_prot), |
| 4735 | pkt_dma_addr); |
| 4736 | if (rc) { |
| 4737 | dev_err(hdev->dev, |
Dotan Barak | d6b045c | 2020-08-06 09:20:49 +0300 | [diff] [blame] | 4738 | "Failed to send fence packet to H/W queue %d\n", |
| 4739 | hw_queue_id); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4740 | goto free_pkt; |
| 4741 | } |
| 4742 | |
| 4743 | rc = hl_poll_timeout_memory(hdev, fence_ptr, tmp, (tmp == fence_val), |
| 4744 | 1000, timeout_usec, true); |
| 4745 | |
| 4746 | hl_hw_queue_inc_ci_kernel(hdev, hw_queue_id); |
| 4747 | |
| 4748 | if (rc == -ETIMEDOUT) { |
| 4749 | dev_err(hdev->dev, |
| 4750 | "H/W queue %d test failed (scratch(0x%08llX) == 0x%08X)\n", |
| 4751 | hw_queue_id, (unsigned long long) fence_dma_addr, tmp); |
| 4752 | rc = -EIO; |
| 4753 | } |
| 4754 | |
| 4755 | free_pkt: |
| 4756 | hdev->asic_funcs->asic_dma_pool_free(hdev, (void *) fence_pkt, |
| 4757 | pkt_dma_addr); |
| 4758 | free_fence_ptr: |
| 4759 | hdev->asic_funcs->asic_dma_pool_free(hdev, (void *) fence_ptr, |
| 4760 | fence_dma_addr); |
| 4761 | return rc; |
| 4762 | } |
| 4763 | |
| 4764 | static int gaudi_test_cpu_queue(struct hl_device *hdev) |
| 4765 | { |
| 4766 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 4767 | |
| 4768 | /* |
| 4769 | * check capability here as send_cpu_message() won't update the result |
| 4770 | * value if no capability |
| 4771 | */ |
| 4772 | if (!(gaudi->hw_cap_initialized & HW_CAP_CPU_Q)) |
| 4773 | return 0; |
| 4774 | |
| 4775 | return hl_fw_test_cpu_queue(hdev); |
| 4776 | } |
| 4777 | |
| 4778 | static int gaudi_test_queues(struct hl_device *hdev) |
| 4779 | { |
| 4780 | int i, rc, ret_val = 0; |
| 4781 | |
Ofir Bitton | 3abc99b | 2020-06-23 14:50:39 +0300 | [diff] [blame] | 4782 | for (i = 0 ; i < hdev->asic_prop.max_queues ; i++) { |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4783 | if (hdev->asic_prop.hw_queues_props[i].type == QUEUE_TYPE_EXT) { |
| 4784 | rc = gaudi_test_queue(hdev, i); |
| 4785 | if (rc) |
| 4786 | ret_val = -EINVAL; |
| 4787 | } |
| 4788 | } |
| 4789 | |
| 4790 | rc = gaudi_test_cpu_queue(hdev); |
| 4791 | if (rc) |
| 4792 | ret_val = -EINVAL; |
| 4793 | |
| 4794 | return ret_val; |
| 4795 | } |
| 4796 | |
| 4797 | static void *gaudi_dma_pool_zalloc(struct hl_device *hdev, size_t size, |
| 4798 | gfp_t mem_flags, dma_addr_t *dma_handle) |
| 4799 | { |
| 4800 | void *kernel_addr; |
| 4801 | |
| 4802 | if (size > GAUDI_DMA_POOL_BLK_SIZE) |
| 4803 | return NULL; |
| 4804 | |
| 4805 | kernel_addr = dma_pool_zalloc(hdev->dma_pool, mem_flags, dma_handle); |
| 4806 | |
| 4807 | /* Shift to the device's base physical address of host memory */ |
| 4808 | if (kernel_addr) |
| 4809 | *dma_handle += HOST_PHYS_BASE; |
| 4810 | |
| 4811 | return kernel_addr; |
| 4812 | } |
| 4813 | |
| 4814 | static void gaudi_dma_pool_free(struct hl_device *hdev, void *vaddr, |
| 4815 | dma_addr_t dma_addr) |
| 4816 | { |
| 4817 | /* Cancel the device's base physical address of host memory */ |
| 4818 | dma_addr_t fixed_dma_addr = dma_addr - HOST_PHYS_BASE; |
| 4819 | |
| 4820 | dma_pool_free(hdev->dma_pool, vaddr, fixed_dma_addr); |
| 4821 | } |
| 4822 | |
| 4823 | static void *gaudi_cpu_accessible_dma_pool_alloc(struct hl_device *hdev, |
| 4824 | size_t size, dma_addr_t *dma_handle) |
| 4825 | { |
| 4826 | return hl_fw_cpu_accessible_dma_pool_alloc(hdev, size, dma_handle); |
| 4827 | } |
| 4828 | |
| 4829 | static void gaudi_cpu_accessible_dma_pool_free(struct hl_device *hdev, |
| 4830 | size_t size, void *vaddr) |
| 4831 | { |
| 4832 | hl_fw_cpu_accessible_dma_pool_free(hdev, size, vaddr); |
| 4833 | } |
| 4834 | |
| 4835 | static int gaudi_dma_map_sg(struct hl_device *hdev, struct scatterlist *sgl, |
| 4836 | int nents, enum dma_data_direction dir) |
| 4837 | { |
| 4838 | struct scatterlist *sg; |
| 4839 | int i; |
| 4840 | |
| 4841 | if (!dma_map_sg(&hdev->pdev->dev, sgl, nents, dir)) |
| 4842 | return -ENOMEM; |
| 4843 | |
| 4844 | /* Shift to the device's base physical address of host memory */ |
| 4845 | for_each_sg(sgl, sg, nents, i) |
| 4846 | sg->dma_address += HOST_PHYS_BASE; |
| 4847 | |
| 4848 | return 0; |
| 4849 | } |
| 4850 | |
| 4851 | static void gaudi_dma_unmap_sg(struct hl_device *hdev, struct scatterlist *sgl, |
| 4852 | int nents, enum dma_data_direction dir) |
| 4853 | { |
| 4854 | struct scatterlist *sg; |
| 4855 | int i; |
| 4856 | |
| 4857 | /* Cancel the device's base physical address of host memory */ |
| 4858 | for_each_sg(sgl, sg, nents, i) |
| 4859 | sg->dma_address -= HOST_PHYS_BASE; |
| 4860 | |
| 4861 | dma_unmap_sg(&hdev->pdev->dev, sgl, nents, dir); |
| 4862 | } |
| 4863 | |
| 4864 | static u32 gaudi_get_dma_desc_list_size(struct hl_device *hdev, |
| 4865 | struct sg_table *sgt) |
| 4866 | { |
| 4867 | struct scatterlist *sg, *sg_next_iter; |
| 4868 | u32 count, dma_desc_cnt; |
| 4869 | u64 len, len_next; |
| 4870 | dma_addr_t addr, addr_next; |
| 4871 | |
| 4872 | dma_desc_cnt = 0; |
| 4873 | |
| 4874 | for_each_sg(sgt->sgl, sg, sgt->nents, count) { |
| 4875 | |
| 4876 | len = sg_dma_len(sg); |
| 4877 | addr = sg_dma_address(sg); |
| 4878 | |
| 4879 | if (len == 0) |
| 4880 | break; |
| 4881 | |
| 4882 | while ((count + 1) < sgt->nents) { |
| 4883 | sg_next_iter = sg_next(sg); |
| 4884 | len_next = sg_dma_len(sg_next_iter); |
| 4885 | addr_next = sg_dma_address(sg_next_iter); |
| 4886 | |
| 4887 | if (len_next == 0) |
| 4888 | break; |
| 4889 | |
| 4890 | if ((addr + len == addr_next) && |
| 4891 | (len + len_next <= DMA_MAX_TRANSFER_SIZE)) { |
| 4892 | len += len_next; |
| 4893 | count++; |
| 4894 | sg = sg_next_iter; |
| 4895 | } else { |
| 4896 | break; |
| 4897 | } |
| 4898 | } |
| 4899 | |
| 4900 | dma_desc_cnt++; |
| 4901 | } |
| 4902 | |
| 4903 | return dma_desc_cnt * sizeof(struct packet_lin_dma); |
| 4904 | } |
| 4905 | |
| 4906 | static int gaudi_pin_memory_before_cs(struct hl_device *hdev, |
| 4907 | struct hl_cs_parser *parser, |
| 4908 | struct packet_lin_dma *user_dma_pkt, |
| 4909 | u64 addr, enum dma_data_direction dir) |
| 4910 | { |
| 4911 | struct hl_userptr *userptr; |
| 4912 | int rc; |
| 4913 | |
| 4914 | if (hl_userptr_is_pinned(hdev, addr, le32_to_cpu(user_dma_pkt->tsize), |
| 4915 | parser->job_userptr_list, &userptr)) |
| 4916 | goto already_pinned; |
| 4917 | |
Ofir Bitton | d5eb837 | 2021-02-14 15:35:56 +0200 | [diff] [blame] | 4918 | userptr = kzalloc(sizeof(*userptr), GFP_KERNEL); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 4919 | if (!userptr) |
| 4920 | return -ENOMEM; |
| 4921 | |
| 4922 | rc = hl_pin_host_memory(hdev, addr, le32_to_cpu(user_dma_pkt->tsize), |
| 4923 | userptr); |
| 4924 | if (rc) |
| 4925 | goto free_userptr; |
| 4926 | |
| 4927 | list_add_tail(&userptr->job_node, parser->job_userptr_list); |
| 4928 | |
| 4929 | rc = hdev->asic_funcs->asic_dma_map_sg(hdev, userptr->sgt->sgl, |
| 4930 | userptr->sgt->nents, dir); |
| 4931 | if (rc) { |
| 4932 | dev_err(hdev->dev, "failed to map sgt with DMA region\n"); |
| 4933 | goto unpin_memory; |
| 4934 | } |
| 4935 | |
| 4936 | userptr->dma_mapped = true; |
| 4937 | userptr->dir = dir; |
| 4938 | |
| 4939 | already_pinned: |
| 4940 | parser->patched_cb_size += |
| 4941 | gaudi_get_dma_desc_list_size(hdev, userptr->sgt); |
| 4942 | |
| 4943 | return 0; |
| 4944 | |
| 4945 | unpin_memory: |
| 4946 | hl_unpin_host_memory(hdev, userptr); |
| 4947 | free_userptr: |
| 4948 | kfree(userptr); |
| 4949 | return rc; |
| 4950 | } |
| 4951 | |
| 4952 | static int gaudi_validate_dma_pkt_host(struct hl_device *hdev, |
| 4953 | struct hl_cs_parser *parser, |
| 4954 | struct packet_lin_dma *user_dma_pkt, |
| 4955 | bool src_in_host) |
| 4956 | { |
| 4957 | enum dma_data_direction dir; |
| 4958 | bool skip_host_mem_pin = false, user_memset; |
| 4959 | u64 addr; |
| 4960 | int rc = 0; |
| 4961 | |
| 4962 | user_memset = (le32_to_cpu(user_dma_pkt->ctl) & |
| 4963 | GAUDI_PKT_LIN_DMA_CTL_MEMSET_MASK) >> |
| 4964 | GAUDI_PKT_LIN_DMA_CTL_MEMSET_SHIFT; |
| 4965 | |
| 4966 | if (src_in_host) { |
| 4967 | if (user_memset) |
| 4968 | skip_host_mem_pin = true; |
| 4969 | |
| 4970 | dev_dbg(hdev->dev, "DMA direction is HOST --> DEVICE\n"); |
| 4971 | dir = DMA_TO_DEVICE; |
| 4972 | addr = le64_to_cpu(user_dma_pkt->src_addr); |
| 4973 | } else { |
| 4974 | dev_dbg(hdev->dev, "DMA direction is DEVICE --> HOST\n"); |
| 4975 | dir = DMA_FROM_DEVICE; |
| 4976 | addr = (le64_to_cpu(user_dma_pkt->dst_addr) & |
| 4977 | GAUDI_PKT_LIN_DMA_DST_ADDR_MASK) >> |
| 4978 | GAUDI_PKT_LIN_DMA_DST_ADDR_SHIFT; |
| 4979 | } |
| 4980 | |
| 4981 | if (skip_host_mem_pin) |
| 4982 | parser->patched_cb_size += sizeof(*user_dma_pkt); |
| 4983 | else |
| 4984 | rc = gaudi_pin_memory_before_cs(hdev, parser, user_dma_pkt, |
| 4985 | addr, dir); |
| 4986 | |
| 4987 | return rc; |
| 4988 | } |
| 4989 | |
| 4990 | static int gaudi_validate_dma_pkt_no_mmu(struct hl_device *hdev, |
| 4991 | struct hl_cs_parser *parser, |
| 4992 | struct packet_lin_dma *user_dma_pkt) |
| 4993 | { |
| 4994 | bool src_in_host = false; |
| 4995 | u64 dst_addr = (le64_to_cpu(user_dma_pkt->dst_addr) & |
| 4996 | GAUDI_PKT_LIN_DMA_DST_ADDR_MASK) >> |
| 4997 | GAUDI_PKT_LIN_DMA_DST_ADDR_SHIFT; |
| 4998 | |
| 4999 | dev_dbg(hdev->dev, "DMA packet details:\n"); |
| 5000 | dev_dbg(hdev->dev, "source == 0x%llx\n", |
| 5001 | le64_to_cpu(user_dma_pkt->src_addr)); |
| 5002 | dev_dbg(hdev->dev, "destination == 0x%llx\n", dst_addr); |
| 5003 | dev_dbg(hdev->dev, "size == %u\n", le32_to_cpu(user_dma_pkt->tsize)); |
| 5004 | |
| 5005 | /* |
| 5006 | * Special handling for DMA with size 0. Bypass all validations |
| 5007 | * because no transactions will be done except for WR_COMP, which |
| 5008 | * is not a security issue |
| 5009 | */ |
| 5010 | if (!le32_to_cpu(user_dma_pkt->tsize)) { |
| 5011 | parser->patched_cb_size += sizeof(*user_dma_pkt); |
| 5012 | return 0; |
| 5013 | } |
| 5014 | |
| 5015 | if (parser->hw_queue_id <= GAUDI_QUEUE_ID_DMA_0_3) |
| 5016 | src_in_host = true; |
| 5017 | |
| 5018 | return gaudi_validate_dma_pkt_host(hdev, parser, user_dma_pkt, |
| 5019 | src_in_host); |
| 5020 | } |
| 5021 | |
Oded Gabbay | 64536ab | 2020-05-27 12:38:16 +0300 | [diff] [blame] | 5022 | static int gaudi_validate_load_and_exe_pkt(struct hl_device *hdev, |
| 5023 | struct hl_cs_parser *parser, |
| 5024 | struct packet_load_and_exe *user_pkt) |
| 5025 | { |
| 5026 | u32 cfg; |
| 5027 | |
| 5028 | cfg = le32_to_cpu(user_pkt->cfg); |
| 5029 | |
| 5030 | if (cfg & GAUDI_PKT_LOAD_AND_EXE_CFG_DST_MASK) { |
| 5031 | dev_err(hdev->dev, |
| 5032 | "User not allowed to use Load and Execute\n"); |
| 5033 | return -EPERM; |
| 5034 | } |
| 5035 | |
| 5036 | parser->patched_cb_size += sizeof(struct packet_load_and_exe); |
| 5037 | |
| 5038 | return 0; |
| 5039 | } |
| 5040 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5041 | static int gaudi_validate_cb(struct hl_device *hdev, |
| 5042 | struct hl_cs_parser *parser, bool is_mmu) |
| 5043 | { |
| 5044 | u32 cb_parsed_length = 0; |
| 5045 | int rc = 0; |
| 5046 | |
| 5047 | parser->patched_cb_size = 0; |
| 5048 | |
| 5049 | /* cb_user_size is more than 0 so loop will always be executed */ |
| 5050 | while (cb_parsed_length < parser->user_cb_size) { |
| 5051 | enum packet_id pkt_id; |
| 5052 | u16 pkt_size; |
| 5053 | struct gaudi_packet *user_pkt; |
| 5054 | |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 5055 | user_pkt = parser->user_cb->kernel_address + cb_parsed_length; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5056 | |
| 5057 | pkt_id = (enum packet_id) ( |
| 5058 | (le64_to_cpu(user_pkt->header) & |
| 5059 | PACKET_HEADER_PACKET_ID_MASK) >> |
| 5060 | PACKET_HEADER_PACKET_ID_SHIFT); |
| 5061 | |
Ofir Bitton | bc75be2 | 2020-07-30 14:56:38 +0300 | [diff] [blame] | 5062 | if (!validate_packet_id(pkt_id)) { |
| 5063 | dev_err(hdev->dev, "Invalid packet id %u\n", pkt_id); |
| 5064 | rc = -EINVAL; |
| 5065 | break; |
| 5066 | } |
| 5067 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5068 | pkt_size = gaudi_packet_sizes[pkt_id]; |
| 5069 | cb_parsed_length += pkt_size; |
| 5070 | if (cb_parsed_length > parser->user_cb_size) { |
| 5071 | dev_err(hdev->dev, |
| 5072 | "packet 0x%x is out of CB boundary\n", pkt_id); |
| 5073 | rc = -EINVAL; |
| 5074 | break; |
| 5075 | } |
| 5076 | |
| 5077 | switch (pkt_id) { |
| 5078 | case PACKET_MSG_PROT: |
| 5079 | dev_err(hdev->dev, |
| 5080 | "User not allowed to use MSG_PROT\n"); |
| 5081 | rc = -EPERM; |
| 5082 | break; |
| 5083 | |
| 5084 | case PACKET_CP_DMA: |
| 5085 | dev_err(hdev->dev, "User not allowed to use CP_DMA\n"); |
| 5086 | rc = -EPERM; |
| 5087 | break; |
| 5088 | |
| 5089 | case PACKET_STOP: |
| 5090 | dev_err(hdev->dev, "User not allowed to use STOP\n"); |
| 5091 | rc = -EPERM; |
| 5092 | break; |
| 5093 | |
Oded Gabbay | 2edc66e | 2020-07-03 19:28:54 +0300 | [diff] [blame] | 5094 | case PACKET_WREG_BULK: |
| 5095 | dev_err(hdev->dev, |
| 5096 | "User not allowed to use WREG_BULK\n"); |
| 5097 | rc = -EPERM; |
| 5098 | break; |
| 5099 | |
Oded Gabbay | 64536ab | 2020-05-27 12:38:16 +0300 | [diff] [blame] | 5100 | case PACKET_LOAD_AND_EXE: |
| 5101 | rc = gaudi_validate_load_and_exe_pkt(hdev, parser, |
| 5102 | (struct packet_load_and_exe *) user_pkt); |
| 5103 | break; |
| 5104 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5105 | case PACKET_LIN_DMA: |
| 5106 | parser->contains_dma_pkt = true; |
| 5107 | if (is_mmu) |
| 5108 | parser->patched_cb_size += pkt_size; |
| 5109 | else |
| 5110 | rc = gaudi_validate_dma_pkt_no_mmu(hdev, parser, |
| 5111 | (struct packet_lin_dma *) user_pkt); |
| 5112 | break; |
| 5113 | |
| 5114 | case PACKET_WREG_32: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5115 | case PACKET_MSG_LONG: |
| 5116 | case PACKET_MSG_SHORT: |
| 5117 | case PACKET_REPEAT: |
| 5118 | case PACKET_FENCE: |
| 5119 | case PACKET_NOP: |
| 5120 | case PACKET_ARB_POINT: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5121 | parser->patched_cb_size += pkt_size; |
| 5122 | break; |
| 5123 | |
| 5124 | default: |
| 5125 | dev_err(hdev->dev, "Invalid packet header 0x%x\n", |
| 5126 | pkt_id); |
| 5127 | rc = -EINVAL; |
| 5128 | break; |
| 5129 | } |
| 5130 | |
| 5131 | if (rc) |
| 5132 | break; |
| 5133 | } |
| 5134 | |
| 5135 | /* |
| 5136 | * The new CB should have space at the end for two MSG_PROT packets: |
| 5137 | * 1. A packet that will act as a completion packet |
| 5138 | * 2. A packet that will generate MSI-X interrupt |
| 5139 | */ |
Ofir Bitton | ac6fdbf | 2020-12-03 16:59:28 +0200 | [diff] [blame] | 5140 | if (parser->completion) |
| 5141 | parser->patched_cb_size += sizeof(struct packet_msg_prot) * 2; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5142 | |
| 5143 | return rc; |
| 5144 | } |
| 5145 | |
| 5146 | static int gaudi_patch_dma_packet(struct hl_device *hdev, |
| 5147 | struct hl_cs_parser *parser, |
| 5148 | struct packet_lin_dma *user_dma_pkt, |
| 5149 | struct packet_lin_dma *new_dma_pkt, |
| 5150 | u32 *new_dma_pkt_size) |
| 5151 | { |
| 5152 | struct hl_userptr *userptr; |
| 5153 | struct scatterlist *sg, *sg_next_iter; |
| 5154 | u32 count, dma_desc_cnt, user_wrcomp_en_mask, ctl; |
| 5155 | u64 len, len_next; |
| 5156 | dma_addr_t dma_addr, dma_addr_next; |
| 5157 | u64 device_memory_addr, addr; |
| 5158 | enum dma_data_direction dir; |
| 5159 | struct sg_table *sgt; |
| 5160 | bool src_in_host = false; |
| 5161 | bool skip_host_mem_pin = false; |
| 5162 | bool user_memset; |
| 5163 | |
| 5164 | ctl = le32_to_cpu(user_dma_pkt->ctl); |
| 5165 | |
| 5166 | if (parser->hw_queue_id <= GAUDI_QUEUE_ID_DMA_0_3) |
| 5167 | src_in_host = true; |
| 5168 | |
| 5169 | user_memset = (ctl & GAUDI_PKT_LIN_DMA_CTL_MEMSET_MASK) >> |
| 5170 | GAUDI_PKT_LIN_DMA_CTL_MEMSET_SHIFT; |
| 5171 | |
| 5172 | if (src_in_host) { |
| 5173 | addr = le64_to_cpu(user_dma_pkt->src_addr); |
| 5174 | device_memory_addr = le64_to_cpu(user_dma_pkt->dst_addr); |
| 5175 | dir = DMA_TO_DEVICE; |
| 5176 | if (user_memset) |
| 5177 | skip_host_mem_pin = true; |
| 5178 | } else { |
| 5179 | addr = le64_to_cpu(user_dma_pkt->dst_addr); |
| 5180 | device_memory_addr = le64_to_cpu(user_dma_pkt->src_addr); |
| 5181 | dir = DMA_FROM_DEVICE; |
| 5182 | } |
| 5183 | |
| 5184 | if ((!skip_host_mem_pin) && |
| 5185 | (!hl_userptr_is_pinned(hdev, addr, |
| 5186 | le32_to_cpu(user_dma_pkt->tsize), |
| 5187 | parser->job_userptr_list, &userptr))) { |
| 5188 | dev_err(hdev->dev, "Userptr 0x%llx + 0x%x NOT mapped\n", |
| 5189 | addr, user_dma_pkt->tsize); |
| 5190 | return -EFAULT; |
| 5191 | } |
| 5192 | |
| 5193 | if ((user_memset) && (dir == DMA_TO_DEVICE)) { |
| 5194 | memcpy(new_dma_pkt, user_dma_pkt, sizeof(*user_dma_pkt)); |
| 5195 | *new_dma_pkt_size = sizeof(*user_dma_pkt); |
| 5196 | return 0; |
| 5197 | } |
| 5198 | |
| 5199 | user_wrcomp_en_mask = ctl & GAUDI_PKT_LIN_DMA_CTL_WRCOMP_EN_MASK; |
| 5200 | |
| 5201 | sgt = userptr->sgt; |
| 5202 | dma_desc_cnt = 0; |
| 5203 | |
| 5204 | for_each_sg(sgt->sgl, sg, sgt->nents, count) { |
| 5205 | len = sg_dma_len(sg); |
| 5206 | dma_addr = sg_dma_address(sg); |
| 5207 | |
| 5208 | if (len == 0) |
| 5209 | break; |
| 5210 | |
| 5211 | while ((count + 1) < sgt->nents) { |
| 5212 | sg_next_iter = sg_next(sg); |
| 5213 | len_next = sg_dma_len(sg_next_iter); |
| 5214 | dma_addr_next = sg_dma_address(sg_next_iter); |
| 5215 | |
| 5216 | if (len_next == 0) |
| 5217 | break; |
| 5218 | |
| 5219 | if ((dma_addr + len == dma_addr_next) && |
| 5220 | (len + len_next <= DMA_MAX_TRANSFER_SIZE)) { |
| 5221 | len += len_next; |
| 5222 | count++; |
| 5223 | sg = sg_next_iter; |
| 5224 | } else { |
| 5225 | break; |
| 5226 | } |
| 5227 | } |
| 5228 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5229 | ctl = le32_to_cpu(user_dma_pkt->ctl); |
| 5230 | if (likely(dma_desc_cnt)) |
| 5231 | ctl &= ~GAUDI_PKT_CTL_EB_MASK; |
| 5232 | ctl &= ~GAUDI_PKT_LIN_DMA_CTL_WRCOMP_EN_MASK; |
| 5233 | new_dma_pkt->ctl = cpu_to_le32(ctl); |
| 5234 | new_dma_pkt->tsize = cpu_to_le32(len); |
| 5235 | |
| 5236 | if (dir == DMA_TO_DEVICE) { |
| 5237 | new_dma_pkt->src_addr = cpu_to_le64(dma_addr); |
| 5238 | new_dma_pkt->dst_addr = cpu_to_le64(device_memory_addr); |
| 5239 | } else { |
| 5240 | new_dma_pkt->src_addr = cpu_to_le64(device_memory_addr); |
| 5241 | new_dma_pkt->dst_addr = cpu_to_le64(dma_addr); |
| 5242 | } |
| 5243 | |
| 5244 | if (!user_memset) |
| 5245 | device_memory_addr += len; |
| 5246 | dma_desc_cnt++; |
| 5247 | new_dma_pkt++; |
| 5248 | } |
| 5249 | |
| 5250 | if (!dma_desc_cnt) { |
| 5251 | dev_err(hdev->dev, |
| 5252 | "Error of 0 SG entries when patching DMA packet\n"); |
| 5253 | return -EFAULT; |
| 5254 | } |
| 5255 | |
| 5256 | /* Fix the last dma packet - wrcomp must be as user set it */ |
| 5257 | new_dma_pkt--; |
| 5258 | new_dma_pkt->ctl |= cpu_to_le32(user_wrcomp_en_mask); |
| 5259 | |
| 5260 | *new_dma_pkt_size = dma_desc_cnt * sizeof(struct packet_lin_dma); |
| 5261 | |
| 5262 | return 0; |
| 5263 | } |
| 5264 | |
| 5265 | static int gaudi_patch_cb(struct hl_device *hdev, |
| 5266 | struct hl_cs_parser *parser) |
| 5267 | { |
| 5268 | u32 cb_parsed_length = 0; |
| 5269 | u32 cb_patched_cur_length = 0; |
| 5270 | int rc = 0; |
| 5271 | |
| 5272 | /* cb_user_size is more than 0 so loop will always be executed */ |
| 5273 | while (cb_parsed_length < parser->user_cb_size) { |
| 5274 | enum packet_id pkt_id; |
| 5275 | u16 pkt_size; |
| 5276 | u32 new_pkt_size = 0; |
| 5277 | struct gaudi_packet *user_pkt, *kernel_pkt; |
| 5278 | |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 5279 | user_pkt = parser->user_cb->kernel_address + cb_parsed_length; |
| 5280 | kernel_pkt = parser->patched_cb->kernel_address + |
| 5281 | cb_patched_cur_length; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5282 | |
| 5283 | pkt_id = (enum packet_id) ( |
| 5284 | (le64_to_cpu(user_pkt->header) & |
| 5285 | PACKET_HEADER_PACKET_ID_MASK) >> |
| 5286 | PACKET_HEADER_PACKET_ID_SHIFT); |
| 5287 | |
Ofir Bitton | bc75be2 | 2020-07-30 14:56:38 +0300 | [diff] [blame] | 5288 | if (!validate_packet_id(pkt_id)) { |
| 5289 | dev_err(hdev->dev, "Invalid packet id %u\n", pkt_id); |
| 5290 | rc = -EINVAL; |
| 5291 | break; |
| 5292 | } |
| 5293 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5294 | pkt_size = gaudi_packet_sizes[pkt_id]; |
| 5295 | cb_parsed_length += pkt_size; |
| 5296 | if (cb_parsed_length > parser->user_cb_size) { |
| 5297 | dev_err(hdev->dev, |
| 5298 | "packet 0x%x is out of CB boundary\n", pkt_id); |
| 5299 | rc = -EINVAL; |
| 5300 | break; |
| 5301 | } |
| 5302 | |
| 5303 | switch (pkt_id) { |
| 5304 | case PACKET_LIN_DMA: |
| 5305 | rc = gaudi_patch_dma_packet(hdev, parser, |
| 5306 | (struct packet_lin_dma *) user_pkt, |
| 5307 | (struct packet_lin_dma *) kernel_pkt, |
| 5308 | &new_pkt_size); |
| 5309 | cb_patched_cur_length += new_pkt_size; |
| 5310 | break; |
| 5311 | |
| 5312 | case PACKET_MSG_PROT: |
| 5313 | dev_err(hdev->dev, |
| 5314 | "User not allowed to use MSG_PROT\n"); |
| 5315 | rc = -EPERM; |
| 5316 | break; |
| 5317 | |
| 5318 | case PACKET_CP_DMA: |
| 5319 | dev_err(hdev->dev, "User not allowed to use CP_DMA\n"); |
| 5320 | rc = -EPERM; |
| 5321 | break; |
| 5322 | |
| 5323 | case PACKET_STOP: |
| 5324 | dev_err(hdev->dev, "User not allowed to use STOP\n"); |
| 5325 | rc = -EPERM; |
| 5326 | break; |
| 5327 | |
| 5328 | case PACKET_WREG_32: |
| 5329 | case PACKET_WREG_BULK: |
| 5330 | case PACKET_MSG_LONG: |
| 5331 | case PACKET_MSG_SHORT: |
| 5332 | case PACKET_REPEAT: |
| 5333 | case PACKET_FENCE: |
| 5334 | case PACKET_NOP: |
| 5335 | case PACKET_ARB_POINT: |
| 5336 | case PACKET_LOAD_AND_EXE: |
| 5337 | memcpy(kernel_pkt, user_pkt, pkt_size); |
| 5338 | cb_patched_cur_length += pkt_size; |
| 5339 | break; |
| 5340 | |
| 5341 | default: |
| 5342 | dev_err(hdev->dev, "Invalid packet header 0x%x\n", |
| 5343 | pkt_id); |
| 5344 | rc = -EINVAL; |
| 5345 | break; |
| 5346 | } |
| 5347 | |
| 5348 | if (rc) |
| 5349 | break; |
| 5350 | } |
| 5351 | |
| 5352 | return rc; |
| 5353 | } |
| 5354 | |
| 5355 | static int gaudi_parse_cb_mmu(struct hl_device *hdev, |
| 5356 | struct hl_cs_parser *parser) |
| 5357 | { |
| 5358 | u64 patched_cb_handle; |
| 5359 | u32 patched_cb_size; |
| 5360 | struct hl_cb *user_cb; |
| 5361 | int rc; |
| 5362 | |
| 5363 | /* |
| 5364 | * The new CB should have space at the end for two MSG_PROT pkt: |
| 5365 | * 1. A packet that will act as a completion packet |
| 5366 | * 2. A packet that will generate MSI interrupt |
| 5367 | */ |
Ofir Bitton | ac6fdbf | 2020-12-03 16:59:28 +0200 | [diff] [blame] | 5368 | if (parser->completion) |
| 5369 | parser->patched_cb_size = parser->user_cb_size + |
| 5370 | sizeof(struct packet_msg_prot) * 2; |
| 5371 | else |
| 5372 | parser->patched_cb_size = parser->user_cb_size; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5373 | |
Tomer Tayar | fa8641a1 | 2020-09-07 17:36:41 +0300 | [diff] [blame] | 5374 | rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr, hdev->kernel_ctx, |
Tomer Tayar | ef6a0f6 | 2020-07-09 16:17:48 +0300 | [diff] [blame] | 5375 | parser->patched_cb_size, false, false, |
Tomer Tayar | fa8641a1 | 2020-09-07 17:36:41 +0300 | [diff] [blame] | 5376 | &patched_cb_handle); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5377 | |
| 5378 | if (rc) { |
| 5379 | dev_err(hdev->dev, |
| 5380 | "Failed to allocate patched CB for DMA CS %d\n", |
| 5381 | rc); |
| 5382 | return rc; |
| 5383 | } |
| 5384 | |
| 5385 | patched_cb_handle >>= PAGE_SHIFT; |
| 5386 | parser->patched_cb = hl_cb_get(hdev, &hdev->kernel_cb_mgr, |
| 5387 | (u32) patched_cb_handle); |
Alon Mizrahi | 75d9a2a | 2020-12-03 17:32:19 +0200 | [diff] [blame] | 5388 | /* hl_cb_get should never fail */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5389 | if (!parser->patched_cb) { |
Alon Mizrahi | 75d9a2a | 2020-12-03 17:32:19 +0200 | [diff] [blame] | 5390 | dev_crit(hdev->dev, "DMA CB handle invalid 0x%x\n", |
| 5391 | (u32) patched_cb_handle); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5392 | rc = -EFAULT; |
| 5393 | goto out; |
| 5394 | } |
| 5395 | |
| 5396 | /* |
| 5397 | * The check that parser->user_cb_size <= parser->user_cb->size was done |
| 5398 | * in validate_queue_index(). |
| 5399 | */ |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 5400 | memcpy(parser->patched_cb->kernel_address, |
| 5401 | parser->user_cb->kernel_address, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5402 | parser->user_cb_size); |
| 5403 | |
| 5404 | patched_cb_size = parser->patched_cb_size; |
| 5405 | |
| 5406 | /* Validate patched CB instead of user CB */ |
| 5407 | user_cb = parser->user_cb; |
| 5408 | parser->user_cb = parser->patched_cb; |
| 5409 | rc = gaudi_validate_cb(hdev, parser, true); |
| 5410 | parser->user_cb = user_cb; |
| 5411 | |
| 5412 | if (rc) { |
| 5413 | hl_cb_put(parser->patched_cb); |
| 5414 | goto out; |
| 5415 | } |
| 5416 | |
| 5417 | if (patched_cb_size != parser->patched_cb_size) { |
| 5418 | dev_err(hdev->dev, "user CB size mismatch\n"); |
| 5419 | hl_cb_put(parser->patched_cb); |
| 5420 | rc = -EINVAL; |
| 5421 | goto out; |
| 5422 | } |
| 5423 | |
| 5424 | out: |
| 5425 | /* |
| 5426 | * Always call cb destroy here because we still have 1 reference |
| 5427 | * to it by calling cb_get earlier. After the job will be completed, |
| 5428 | * cb_put will release it, but here we want to remove it from the |
| 5429 | * idr |
| 5430 | */ |
| 5431 | hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, |
| 5432 | patched_cb_handle << PAGE_SHIFT); |
| 5433 | |
| 5434 | return rc; |
| 5435 | } |
| 5436 | |
| 5437 | static int gaudi_parse_cb_no_mmu(struct hl_device *hdev, |
| 5438 | struct hl_cs_parser *parser) |
| 5439 | { |
| 5440 | u64 patched_cb_handle; |
| 5441 | int rc; |
| 5442 | |
| 5443 | rc = gaudi_validate_cb(hdev, parser, false); |
| 5444 | |
| 5445 | if (rc) |
| 5446 | goto free_userptr; |
| 5447 | |
Tomer Tayar | fa8641a1 | 2020-09-07 17:36:41 +0300 | [diff] [blame] | 5448 | rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr, hdev->kernel_ctx, |
Tomer Tayar | ef6a0f6 | 2020-07-09 16:17:48 +0300 | [diff] [blame] | 5449 | parser->patched_cb_size, false, false, |
Tomer Tayar | fa8641a1 | 2020-09-07 17:36:41 +0300 | [diff] [blame] | 5450 | &patched_cb_handle); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5451 | if (rc) { |
| 5452 | dev_err(hdev->dev, |
| 5453 | "Failed to allocate patched CB for DMA CS %d\n", rc); |
| 5454 | goto free_userptr; |
| 5455 | } |
| 5456 | |
| 5457 | patched_cb_handle >>= PAGE_SHIFT; |
| 5458 | parser->patched_cb = hl_cb_get(hdev, &hdev->kernel_cb_mgr, |
| 5459 | (u32) patched_cb_handle); |
Alon Mizrahi | 75d9a2a | 2020-12-03 17:32:19 +0200 | [diff] [blame] | 5460 | /* hl_cb_get should never fail here */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5461 | if (!parser->patched_cb) { |
Alon Mizrahi | 75d9a2a | 2020-12-03 17:32:19 +0200 | [diff] [blame] | 5462 | dev_crit(hdev->dev, "DMA CB handle invalid 0x%x\n", |
| 5463 | (u32) patched_cb_handle); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5464 | rc = -EFAULT; |
| 5465 | goto out; |
| 5466 | } |
| 5467 | |
| 5468 | rc = gaudi_patch_cb(hdev, parser); |
| 5469 | |
| 5470 | if (rc) |
| 5471 | hl_cb_put(parser->patched_cb); |
| 5472 | |
| 5473 | out: |
| 5474 | /* |
| 5475 | * Always call cb destroy here because we still have 1 reference |
| 5476 | * to it by calling cb_get earlier. After the job will be completed, |
| 5477 | * cb_put will release it, but here we want to remove it from the |
| 5478 | * idr |
| 5479 | */ |
| 5480 | hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, |
| 5481 | patched_cb_handle << PAGE_SHIFT); |
| 5482 | |
| 5483 | free_userptr: |
| 5484 | if (rc) |
| 5485 | hl_userptr_delete_list(hdev, parser->job_userptr_list); |
| 5486 | return rc; |
| 5487 | } |
| 5488 | |
| 5489 | static int gaudi_parse_cb_no_ext_queue(struct hl_device *hdev, |
| 5490 | struct hl_cs_parser *parser) |
| 5491 | { |
| 5492 | struct asic_fixed_properties *asic_prop = &hdev->asic_prop; |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 5493 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 5494 | u32 nic_mask_q_id = 1 << (HW_CAP_NIC_SHIFT + |
| 5495 | ((parser->hw_queue_id - GAUDI_QUEUE_ID_NIC_0_0) >> 2)); |
| 5496 | |
| 5497 | if ((parser->hw_queue_id >= GAUDI_QUEUE_ID_NIC_0_0) && |
| 5498 | (parser->hw_queue_id <= GAUDI_QUEUE_ID_NIC_9_3) && |
| 5499 | (!(gaudi->hw_cap_initialized & nic_mask_q_id))) { |
| 5500 | dev_err(hdev->dev, "h/w queue %d is disabled\n", |
| 5501 | parser->hw_queue_id); |
| 5502 | return -EINVAL; |
| 5503 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5504 | |
| 5505 | /* For internal queue jobs just check if CB address is valid */ |
| 5506 | if (hl_mem_area_inside_range((u64) (uintptr_t) parser->user_cb, |
| 5507 | parser->user_cb_size, |
| 5508 | asic_prop->sram_user_base_address, |
| 5509 | asic_prop->sram_end_address)) |
| 5510 | return 0; |
| 5511 | |
| 5512 | if (hl_mem_area_inside_range((u64) (uintptr_t) parser->user_cb, |
| 5513 | parser->user_cb_size, |
| 5514 | asic_prop->dram_user_base_address, |
| 5515 | asic_prop->dram_end_address)) |
| 5516 | return 0; |
| 5517 | |
| 5518 | /* PMMU and HPMMU addresses are equal, check only one of them */ |
| 5519 | if (hl_mem_area_inside_range((u64) (uintptr_t) parser->user_cb, |
| 5520 | parser->user_cb_size, |
| 5521 | asic_prop->pmmu.start_addr, |
| 5522 | asic_prop->pmmu.end_addr)) |
| 5523 | return 0; |
| 5524 | |
| 5525 | dev_err(hdev->dev, |
| 5526 | "CB address 0x%px + 0x%x for internal QMAN is not valid\n", |
| 5527 | parser->user_cb, parser->user_cb_size); |
| 5528 | |
| 5529 | return -EFAULT; |
| 5530 | } |
| 5531 | |
| 5532 | static int gaudi_cs_parser(struct hl_device *hdev, struct hl_cs_parser *parser) |
| 5533 | { |
| 5534 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 5535 | |
| 5536 | if (parser->queue_type == QUEUE_TYPE_INT) |
| 5537 | return gaudi_parse_cb_no_ext_queue(hdev, parser); |
| 5538 | |
| 5539 | if (gaudi->hw_cap_initialized & HW_CAP_MMU) |
| 5540 | return gaudi_parse_cb_mmu(hdev, parser); |
| 5541 | else |
| 5542 | return gaudi_parse_cb_no_mmu(hdev, parser); |
| 5543 | } |
| 5544 | |
| 5545 | static void gaudi_add_end_of_cb_packets(struct hl_device *hdev, |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 5546 | void *kernel_address, u32 len, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5547 | u64 cq_addr, u32 cq_val, u32 msi_vec, |
| 5548 | bool eb) |
| 5549 | { |
| 5550 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 5551 | struct packet_msg_prot *cq_pkt; |
| 5552 | u32 tmp; |
| 5553 | |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 5554 | cq_pkt = kernel_address + len - (sizeof(struct packet_msg_prot) * 2); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5555 | |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 5556 | tmp = FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_PROT); |
| 5557 | tmp |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5558 | |
| 5559 | if (eb) |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 5560 | tmp |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5561 | |
| 5562 | cq_pkt->ctl = cpu_to_le32(tmp); |
| 5563 | cq_pkt->value = cpu_to_le32(cq_val); |
| 5564 | cq_pkt->addr = cpu_to_le64(cq_addr); |
| 5565 | |
| 5566 | cq_pkt++; |
| 5567 | |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 5568 | tmp = FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_PROT); |
| 5569 | tmp |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5570 | cq_pkt->ctl = cpu_to_le32(tmp); |
| 5571 | cq_pkt->value = cpu_to_le32(1); |
| 5572 | |
| 5573 | if (!gaudi->multi_msi_mode) |
| 5574 | msi_vec = 0; |
| 5575 | |
| 5576 | cq_pkt->addr = cpu_to_le64(CFG_BASE + mmPCIE_MSI_INTR_0 + msi_vec * 4); |
| 5577 | } |
| 5578 | |
| 5579 | static void gaudi_update_eq_ci(struct hl_device *hdev, u32 val) |
| 5580 | { |
| 5581 | WREG32(mmCPU_IF_EQ_RD_OFFS, val); |
| 5582 | } |
| 5583 | |
| 5584 | static int gaudi_memset_device_memory(struct hl_device *hdev, u64 addr, |
| 5585 | u32 size, u64 val) |
| 5586 | { |
| 5587 | struct packet_lin_dma *lin_dma_pkt; |
| 5588 | struct hl_cs_job *job; |
Moti Haimovski | a9855a2d9 | 2020-06-24 19:40:57 +0300 | [diff] [blame] | 5589 | u32 cb_size, ctl, err_cause; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5590 | struct hl_cb *cb; |
Lv Yunlong | 115726c | 2021-04-26 06:43:46 -0700 | [diff] [blame] | 5591 | u64 id; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5592 | int rc; |
| 5593 | |
Ofir Bitton | a04b7cd | 2020-07-13 13:36:55 +0300 | [diff] [blame] | 5594 | cb = hl_cb_kernel_create(hdev, PAGE_SIZE, false); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5595 | if (!cb) |
| 5596 | return -EFAULT; |
| 5597 | |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 5598 | lin_dma_pkt = cb->kernel_address; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5599 | memset(lin_dma_pkt, 0, sizeof(*lin_dma_pkt)); |
| 5600 | cb_size = sizeof(*lin_dma_pkt); |
| 5601 | |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 5602 | ctl = FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_LIN_DMA); |
| 5603 | ctl |= FIELD_PREP(GAUDI_PKT_LIN_DMA_CTL_MEMSET_MASK, 1); |
| 5604 | ctl |= FIELD_PREP(GAUDI_PKT_LIN_DMA_CTL_LIN_MASK, 1); |
| 5605 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
| 5606 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 5607 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5608 | lin_dma_pkt->ctl = cpu_to_le32(ctl); |
| 5609 | lin_dma_pkt->src_addr = cpu_to_le64(val); |
| 5610 | lin_dma_pkt->dst_addr |= cpu_to_le64(addr); |
| 5611 | lin_dma_pkt->tsize = cpu_to_le32(size); |
| 5612 | |
| 5613 | job = hl_cs_allocate_job(hdev, QUEUE_TYPE_EXT, true); |
| 5614 | if (!job) { |
| 5615 | dev_err(hdev->dev, "Failed to allocate a new job\n"); |
| 5616 | rc = -ENOMEM; |
| 5617 | goto release_cb; |
| 5618 | } |
| 5619 | |
Moti Haimovski | a9855a2d9 | 2020-06-24 19:40:57 +0300 | [diff] [blame] | 5620 | /* Verify DMA is OK */ |
| 5621 | err_cause = RREG32(mmDMA0_CORE_ERR_CAUSE); |
| 5622 | if (err_cause && !hdev->init_done) { |
| 5623 | dev_dbg(hdev->dev, |
| 5624 | "Clearing DMA0 engine from errors (cause 0x%x)\n", |
| 5625 | err_cause); |
| 5626 | WREG32(mmDMA0_CORE_ERR_CAUSE, err_cause); |
| 5627 | } |
| 5628 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5629 | job->id = 0; |
| 5630 | job->user_cb = cb; |
Tomer Tayar | f0748674 | 2020-08-02 22:51:31 +0300 | [diff] [blame] | 5631 | atomic_inc(&job->user_cb->cs_cnt); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5632 | job->user_cb_size = cb_size; |
| 5633 | job->hw_queue_id = GAUDI_QUEUE_ID_DMA_0_0; |
| 5634 | job->patched_cb = job->user_cb; |
| 5635 | job->job_cb_size = job->user_cb_size + sizeof(struct packet_msg_prot); |
| 5636 | |
| 5637 | hl_debugfs_add_job(hdev, job); |
| 5638 | |
| 5639 | rc = gaudi_send_job_on_qman0(hdev, job); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5640 | hl_debugfs_remove_job(hdev, job); |
| 5641 | kfree(job); |
Tomer Tayar | f0748674 | 2020-08-02 22:51:31 +0300 | [diff] [blame] | 5642 | atomic_dec(&cb->cs_cnt); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5643 | |
Moti Haimovski | a9855a2d9 | 2020-06-24 19:40:57 +0300 | [diff] [blame] | 5644 | /* Verify DMA is OK */ |
| 5645 | err_cause = RREG32(mmDMA0_CORE_ERR_CAUSE); |
| 5646 | if (err_cause) { |
| 5647 | dev_err(hdev->dev, "DMA Failed, cause 0x%x\n", err_cause); |
| 5648 | rc = -EIO; |
| 5649 | if (!hdev->init_done) { |
| 5650 | dev_dbg(hdev->dev, |
| 5651 | "Clearing DMA0 engine from errors (cause 0x%x)\n", |
| 5652 | err_cause); |
| 5653 | WREG32(mmDMA0_CORE_ERR_CAUSE, err_cause); |
| 5654 | } |
| 5655 | } |
| 5656 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5657 | release_cb: |
Lv Yunlong | 115726c | 2021-04-26 06:43:46 -0700 | [diff] [blame] | 5658 | id = cb->id; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5659 | hl_cb_put(cb); |
Lv Yunlong | 115726c | 2021-04-26 06:43:46 -0700 | [diff] [blame] | 5660 | hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, id << PAGE_SHIFT); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5661 | |
| 5662 | return rc; |
| 5663 | } |
| 5664 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5665 | static int gaudi_memset_registers(struct hl_device *hdev, u64 reg_base, |
| 5666 | u32 num_regs, u32 val) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5667 | { |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5668 | struct packet_msg_long *pkt; |
| 5669 | struct hl_cs_job *job; |
| 5670 | u32 cb_size, ctl; |
| 5671 | struct hl_cb *cb; |
| 5672 | int i, rc; |
| 5673 | |
| 5674 | cb_size = (sizeof(*pkt) * num_regs) + sizeof(struct packet_msg_prot); |
| 5675 | |
| 5676 | if (cb_size > SZ_2M) { |
| 5677 | dev_err(hdev->dev, "CB size must be smaller than %uMB", SZ_2M); |
| 5678 | return -ENOMEM; |
| 5679 | } |
| 5680 | |
| 5681 | cb = hl_cb_kernel_create(hdev, cb_size, false); |
| 5682 | if (!cb) |
| 5683 | return -EFAULT; |
| 5684 | |
| 5685 | pkt = cb->kernel_address; |
| 5686 | |
| 5687 | ctl = FIELD_PREP(GAUDI_PKT_LONG_CTL_OP_MASK, 0); /* write the value */ |
| 5688 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_LONG); |
| 5689 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 1); |
| 5690 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 5691 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
| 5692 | |
| 5693 | for (i = 0; i < num_regs ; i++, pkt++) { |
| 5694 | pkt->ctl = cpu_to_le32(ctl); |
| 5695 | pkt->value = cpu_to_le32(val); |
| 5696 | pkt->addr = cpu_to_le64(reg_base + (i * 4)); |
| 5697 | } |
| 5698 | |
| 5699 | job = hl_cs_allocate_job(hdev, QUEUE_TYPE_EXT, true); |
| 5700 | if (!job) { |
| 5701 | dev_err(hdev->dev, "Failed to allocate a new job\n"); |
| 5702 | rc = -ENOMEM; |
| 5703 | goto release_cb; |
| 5704 | } |
| 5705 | |
| 5706 | job->id = 0; |
| 5707 | job->user_cb = cb; |
| 5708 | atomic_inc(&job->user_cb->cs_cnt); |
| 5709 | job->user_cb_size = cb_size; |
| 5710 | job->hw_queue_id = GAUDI_QUEUE_ID_DMA_0_0; |
| 5711 | job->patched_cb = job->user_cb; |
| 5712 | job->job_cb_size = cb_size; |
| 5713 | |
| 5714 | hl_debugfs_add_job(hdev, job); |
| 5715 | |
| 5716 | rc = gaudi_send_job_on_qman0(hdev, job); |
| 5717 | hl_debugfs_remove_job(hdev, job); |
| 5718 | kfree(job); |
| 5719 | atomic_dec(&cb->cs_cnt); |
| 5720 | |
| 5721 | release_cb: |
| 5722 | hl_cb_put(cb); |
| 5723 | hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, cb->id << PAGE_SHIFT); |
| 5724 | |
| 5725 | return rc; |
| 5726 | } |
| 5727 | |
| 5728 | static int gaudi_schedule_register_memset(struct hl_device *hdev, |
| 5729 | u32 hw_queue_id, u64 reg_base, u32 num_regs, u32 val) |
| 5730 | { |
farah kassabri | e65448f | 2021-03-30 16:38:02 +0300 | [diff] [blame] | 5731 | struct hl_ctx *ctx; |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5732 | struct hl_pending_cb *pending_cb; |
| 5733 | struct packet_msg_long *pkt; |
| 5734 | u32 cb_size, ctl; |
| 5735 | struct hl_cb *cb; |
farah kassabri | e65448f | 2021-03-30 16:38:02 +0300 | [diff] [blame] | 5736 | int i, rc; |
| 5737 | |
| 5738 | mutex_lock(&hdev->fpriv_list_lock); |
| 5739 | ctx = hdev->compute_ctx; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5740 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5741 | /* If no compute context available or context is going down |
| 5742 | * memset registers directly |
| 5743 | */ |
farah kassabri | e65448f | 2021-03-30 16:38:02 +0300 | [diff] [blame] | 5744 | if (!ctx || kref_read(&ctx->refcount) == 0) { |
| 5745 | rc = gaudi_memset_registers(hdev, reg_base, num_regs, val); |
| 5746 | mutex_unlock(&hdev->fpriv_list_lock); |
| 5747 | return rc; |
| 5748 | } |
| 5749 | |
| 5750 | mutex_unlock(&hdev->fpriv_list_lock); |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5751 | |
| 5752 | cb_size = (sizeof(*pkt) * num_regs) + |
| 5753 | sizeof(struct packet_msg_prot) * 2; |
| 5754 | |
| 5755 | if (cb_size > SZ_2M) { |
| 5756 | dev_err(hdev->dev, "CB size must be smaller than %uMB", SZ_2M); |
| 5757 | return -ENOMEM; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5758 | } |
| 5759 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5760 | pending_cb = kzalloc(sizeof(*pending_cb), GFP_KERNEL); |
| 5761 | if (!pending_cb) |
| 5762 | return -ENOMEM; |
| 5763 | |
| 5764 | cb = hl_cb_kernel_create(hdev, cb_size, false); |
| 5765 | if (!cb) { |
| 5766 | kfree(pending_cb); |
| 5767 | return -EFAULT; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5768 | } |
| 5769 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5770 | pkt = cb->kernel_address; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5771 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5772 | ctl = FIELD_PREP(GAUDI_PKT_LONG_CTL_OP_MASK, 0); /* write the value */ |
| 5773 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_LONG); |
| 5774 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 1); |
| 5775 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 5776 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5777 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5778 | for (i = 0; i < num_regs ; i++, pkt++) { |
| 5779 | pkt->ctl = cpu_to_le32(ctl); |
| 5780 | pkt->value = cpu_to_le32(val); |
| 5781 | pkt->addr = cpu_to_le64(reg_base + (i * 4)); |
| 5782 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5783 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5784 | hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, cb->id << PAGE_SHIFT); |
| 5785 | |
| 5786 | pending_cb->cb = cb; |
| 5787 | pending_cb->cb_size = cb_size; |
| 5788 | /* The queue ID MUST be an external queue ID. Otherwise, we will |
| 5789 | * have undefined behavior |
| 5790 | */ |
| 5791 | pending_cb->hw_queue_id = hw_queue_id; |
| 5792 | |
| 5793 | spin_lock(&ctx->pending_cb_lock); |
| 5794 | list_add_tail(&pending_cb->cb_node, &ctx->pending_cb_list); |
| 5795 | spin_unlock(&ctx->pending_cb_lock); |
| 5796 | |
| 5797 | return 0; |
| 5798 | } |
| 5799 | |
| 5800 | static int gaudi_restore_sm_registers(struct hl_device *hdev) |
| 5801 | { |
| 5802 | u64 base_addr; |
| 5803 | u32 num_regs; |
| 5804 | int rc; |
| 5805 | |
| 5806 | base_addr = CFG_BASE + mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0; |
| 5807 | num_regs = NUM_OF_SOB_IN_BLOCK; |
| 5808 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5809 | if (rc) { |
| 5810 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5811 | return -ENOMEM; |
| 5812 | } |
| 5813 | |
| 5814 | base_addr = CFG_BASE + mmSYNC_MNGR_E_S_SYNC_MNGR_OBJS_SOB_OBJ_0; |
| 5815 | num_regs = NUM_OF_SOB_IN_BLOCK; |
| 5816 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5817 | if (rc) { |
| 5818 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5819 | return -ENOMEM; |
| 5820 | } |
| 5821 | |
| 5822 | base_addr = CFG_BASE + mmSYNC_MNGR_W_N_SYNC_MNGR_OBJS_SOB_OBJ_0; |
| 5823 | num_regs = NUM_OF_SOB_IN_BLOCK; |
| 5824 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5825 | if (rc) { |
| 5826 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5827 | return -ENOMEM; |
| 5828 | } |
| 5829 | |
| 5830 | base_addr = CFG_BASE + mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_STATUS_0; |
| 5831 | num_regs = NUM_OF_MONITORS_IN_BLOCK; |
| 5832 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5833 | if (rc) { |
| 5834 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5835 | return -ENOMEM; |
| 5836 | } |
| 5837 | |
| 5838 | base_addr = CFG_BASE + mmSYNC_MNGR_E_S_SYNC_MNGR_OBJS_MON_STATUS_0; |
| 5839 | num_regs = NUM_OF_MONITORS_IN_BLOCK; |
| 5840 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5841 | if (rc) { |
| 5842 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5843 | return -ENOMEM; |
| 5844 | } |
| 5845 | |
| 5846 | base_addr = CFG_BASE + mmSYNC_MNGR_W_N_SYNC_MNGR_OBJS_MON_STATUS_0; |
| 5847 | num_regs = NUM_OF_MONITORS_IN_BLOCK; |
| 5848 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5849 | if (rc) { |
| 5850 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5851 | return -ENOMEM; |
| 5852 | } |
| 5853 | |
| 5854 | base_addr = CFG_BASE + mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0 + |
| 5855 | (GAUDI_FIRST_AVAILABLE_W_S_SYNC_OBJECT * 4); |
| 5856 | num_regs = NUM_OF_SOB_IN_BLOCK - GAUDI_FIRST_AVAILABLE_W_S_SYNC_OBJECT; |
| 5857 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5858 | if (rc) { |
| 5859 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5860 | return -ENOMEM; |
| 5861 | } |
| 5862 | |
| 5863 | base_addr = CFG_BASE + mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_STATUS_0 + |
| 5864 | (GAUDI_FIRST_AVAILABLE_W_S_MONITOR * 4); |
| 5865 | num_regs = NUM_OF_MONITORS_IN_BLOCK - GAUDI_FIRST_AVAILABLE_W_S_MONITOR; |
| 5866 | rc = gaudi_memset_registers(hdev, base_addr, num_regs, 0); |
| 5867 | if (rc) { |
| 5868 | dev_err(hdev->dev, "failed resetting SM registers"); |
| 5869 | return -ENOMEM; |
| 5870 | } |
| 5871 | |
| 5872 | return 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5873 | } |
| 5874 | |
| 5875 | static void gaudi_restore_dma_registers(struct hl_device *hdev) |
| 5876 | { |
| 5877 | u32 sob_delta = mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_1 - |
| 5878 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0; |
| 5879 | int i; |
| 5880 | |
| 5881 | for (i = 0 ; i < DMA_NUMBER_OF_CHANNELS ; i++) { |
| 5882 | u64 sob_addr = CFG_BASE + |
| 5883 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0 + |
| 5884 | (i * sob_delta); |
| 5885 | u32 dma_offset = i * DMA_CORE_OFFSET; |
| 5886 | |
| 5887 | WREG32(mmDMA0_CORE_WR_COMP_ADDR_LO + dma_offset, |
| 5888 | lower_32_bits(sob_addr)); |
| 5889 | WREG32(mmDMA0_CORE_WR_COMP_ADDR_HI + dma_offset, |
| 5890 | upper_32_bits(sob_addr)); |
| 5891 | WREG32(mmDMA0_CORE_WR_COMP_WDATA + dma_offset, 0x80000001); |
| 5892 | |
| 5893 | /* For DMAs 2-7, need to restore WR_AWUSER_31_11 as it can be |
| 5894 | * modified by the user for SRAM reduction |
| 5895 | */ |
| 5896 | if (i > 1) |
| 5897 | WREG32(mmDMA0_CORE_WR_AWUSER_31_11 + dma_offset, |
| 5898 | 0x00000001); |
| 5899 | } |
| 5900 | } |
| 5901 | |
| 5902 | static void gaudi_restore_qm_registers(struct hl_device *hdev) |
| 5903 | { |
| 5904 | u32 qman_offset; |
| 5905 | int i; |
| 5906 | |
| 5907 | for (i = 0 ; i < DMA_NUMBER_OF_CHANNELS ; i++) { |
| 5908 | qman_offset = i * DMA_QMAN_OFFSET; |
| 5909 | WREG32(mmDMA0_QM_ARB_CFG_0 + qman_offset, 0); |
| 5910 | } |
| 5911 | |
| 5912 | for (i = 0 ; i < MME_NUMBER_OF_MASTER_ENGINES ; i++) { |
| 5913 | qman_offset = i * (mmMME2_QM_BASE - mmMME0_QM_BASE); |
| 5914 | WREG32(mmMME0_QM_ARB_CFG_0 + qman_offset, 0); |
| 5915 | } |
| 5916 | |
| 5917 | for (i = 0 ; i < TPC_NUMBER_OF_ENGINES ; i++) { |
| 5918 | qman_offset = i * TPC_QMAN_OFFSET; |
| 5919 | WREG32(mmTPC0_QM_ARB_CFG_0 + qman_offset, 0); |
| 5920 | } |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 5921 | |
| 5922 | for (i = 0 ; i < NIC_NUMBER_OF_ENGINES ; i++) { |
| 5923 | qman_offset = (i >> 1) * NIC_MACRO_QMAN_OFFSET + |
| 5924 | (i & 0x1) * NIC_ENGINE_QMAN_OFFSET; |
| 5925 | WREG32(mmNIC0_QM0_ARB_CFG_0 + qman_offset, 0); |
| 5926 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5927 | } |
| 5928 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5929 | static int gaudi_restore_user_registers(struct hl_device *hdev) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5930 | { |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5931 | int rc; |
| 5932 | |
| 5933 | rc = gaudi_restore_sm_registers(hdev); |
| 5934 | if (rc) |
| 5935 | return rc; |
| 5936 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5937 | gaudi_restore_dma_registers(hdev); |
| 5938 | gaudi_restore_qm_registers(hdev); |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5939 | |
| 5940 | return 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5941 | } |
| 5942 | |
| 5943 | static int gaudi_context_switch(struct hl_device *hdev, u32 asid) |
| 5944 | { |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 5945 | return gaudi_restore_user_registers(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5946 | } |
| 5947 | |
| 5948 | static int gaudi_mmu_clear_pgt_range(struct hl_device *hdev) |
| 5949 | { |
| 5950 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 5951 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 5952 | u64 addr = prop->mmu_pgt_addr; |
| 5953 | u32 size = prop->mmu_pgt_size + MMU_CACHE_MNG_SIZE; |
| 5954 | |
| 5955 | if (!(gaudi->hw_cap_initialized & HW_CAP_MMU)) |
| 5956 | return 0; |
| 5957 | |
| 5958 | return gaudi_memset_device_memory(hdev, addr, size, 0); |
| 5959 | } |
| 5960 | |
| 5961 | static void gaudi_restore_phase_topology(struct hl_device *hdev) |
| 5962 | { |
| 5963 | |
| 5964 | } |
| 5965 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 5966 | static int gaudi_debugfs_read32(struct hl_device *hdev, u64 addr, |
| 5967 | bool user_address, u32 *val) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5968 | { |
| 5969 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 5970 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 5971 | u64 hbm_bar_addr, host_phys_end; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5972 | int rc = 0; |
| 5973 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 5974 | host_phys_end = HOST_PHYS_BASE + HOST_PHYS_SIZE; |
| 5975 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5976 | if ((addr >= CFG_BASE) && (addr < CFG_BASE + CFG_SIZE)) { |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 5977 | |
| 5978 | if ((gaudi->hw_cap_initialized & HW_CAP_CLK_GATE) && |
| 5979 | (hdev->clock_gating_mask & |
| 5980 | GAUDI_CLK_GATE_DEBUGFS_MASK)) { |
| 5981 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5982 | dev_err_ratelimited(hdev->dev, |
| 5983 | "Can't read register - clock gating is enabled!\n"); |
| 5984 | rc = -EFAULT; |
| 5985 | } else { |
| 5986 | *val = RREG32(addr - CFG_BASE); |
| 5987 | } |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 5988 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 5989 | } else if ((addr >= SRAM_BASE_ADDR) && |
| 5990 | (addr < SRAM_BASE_ADDR + SRAM_BAR_SIZE)) { |
| 5991 | *val = readl(hdev->pcie_bar[SRAM_BAR_ID] + |
| 5992 | (addr - SRAM_BASE_ADDR)); |
| 5993 | } else if (addr < DRAM_PHYS_BASE + hdev->asic_prop.dram_size) { |
| 5994 | u64 bar_base_addr = DRAM_PHYS_BASE + |
| 5995 | (addr & ~(prop->dram_pci_bar_size - 0x1ull)); |
| 5996 | |
| 5997 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, bar_base_addr); |
| 5998 | if (hbm_bar_addr != U64_MAX) { |
| 5999 | *val = readl(hdev->pcie_bar[HBM_BAR_ID] + |
| 6000 | (addr - bar_base_addr)); |
| 6001 | |
| 6002 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, |
| 6003 | hbm_bar_addr); |
| 6004 | } |
| 6005 | if (hbm_bar_addr == U64_MAX) |
| 6006 | rc = -EIO; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6007 | } else if (addr >= HOST_PHYS_BASE && addr < host_phys_end && |
| 6008 | user_address && !iommu_present(&pci_bus_type)) { |
| 6009 | *val = *(u32 *) phys_to_virt(addr - HOST_PHYS_BASE); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6010 | } else { |
| 6011 | rc = -EFAULT; |
| 6012 | } |
| 6013 | |
| 6014 | return rc; |
| 6015 | } |
| 6016 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6017 | static int gaudi_debugfs_write32(struct hl_device *hdev, u64 addr, |
| 6018 | bool user_address, u32 val) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6019 | { |
| 6020 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 6021 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6022 | u64 hbm_bar_addr, host_phys_end; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6023 | int rc = 0; |
| 6024 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6025 | host_phys_end = HOST_PHYS_BASE + HOST_PHYS_SIZE; |
| 6026 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6027 | if ((addr >= CFG_BASE) && (addr < CFG_BASE + CFG_SIZE)) { |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 6028 | |
| 6029 | if ((gaudi->hw_cap_initialized & HW_CAP_CLK_GATE) && |
| 6030 | (hdev->clock_gating_mask & |
| 6031 | GAUDI_CLK_GATE_DEBUGFS_MASK)) { |
| 6032 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6033 | dev_err_ratelimited(hdev->dev, |
| 6034 | "Can't write register - clock gating is enabled!\n"); |
| 6035 | rc = -EFAULT; |
| 6036 | } else { |
| 6037 | WREG32(addr - CFG_BASE, val); |
| 6038 | } |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 6039 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6040 | } else if ((addr >= SRAM_BASE_ADDR) && |
| 6041 | (addr < SRAM_BASE_ADDR + SRAM_BAR_SIZE)) { |
| 6042 | writel(val, hdev->pcie_bar[SRAM_BAR_ID] + |
| 6043 | (addr - SRAM_BASE_ADDR)); |
| 6044 | } else if (addr < DRAM_PHYS_BASE + hdev->asic_prop.dram_size) { |
| 6045 | u64 bar_base_addr = DRAM_PHYS_BASE + |
| 6046 | (addr & ~(prop->dram_pci_bar_size - 0x1ull)); |
| 6047 | |
| 6048 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, bar_base_addr); |
| 6049 | if (hbm_bar_addr != U64_MAX) { |
| 6050 | writel(val, hdev->pcie_bar[HBM_BAR_ID] + |
| 6051 | (addr - bar_base_addr)); |
| 6052 | |
| 6053 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, |
| 6054 | hbm_bar_addr); |
| 6055 | } |
| 6056 | if (hbm_bar_addr == U64_MAX) |
| 6057 | rc = -EIO; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6058 | } else if (addr >= HOST_PHYS_BASE && addr < host_phys_end && |
| 6059 | user_address && !iommu_present(&pci_bus_type)) { |
| 6060 | *(u32 *) phys_to_virt(addr - HOST_PHYS_BASE) = val; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6061 | } else { |
| 6062 | rc = -EFAULT; |
| 6063 | } |
| 6064 | |
| 6065 | return rc; |
| 6066 | } |
| 6067 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6068 | static int gaudi_debugfs_read64(struct hl_device *hdev, u64 addr, |
| 6069 | bool user_address, u64 *val) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6070 | { |
| 6071 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 6072 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6073 | u64 hbm_bar_addr, host_phys_end; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6074 | int rc = 0; |
| 6075 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6076 | host_phys_end = HOST_PHYS_BASE + HOST_PHYS_SIZE; |
| 6077 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6078 | if ((addr >= CFG_BASE) && (addr <= CFG_BASE + CFG_SIZE - sizeof(u64))) { |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 6079 | |
| 6080 | if ((gaudi->hw_cap_initialized & HW_CAP_CLK_GATE) && |
| 6081 | (hdev->clock_gating_mask & |
| 6082 | GAUDI_CLK_GATE_DEBUGFS_MASK)) { |
| 6083 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6084 | dev_err_ratelimited(hdev->dev, |
| 6085 | "Can't read register - clock gating is enabled!\n"); |
| 6086 | rc = -EFAULT; |
| 6087 | } else { |
| 6088 | u32 val_l = RREG32(addr - CFG_BASE); |
| 6089 | u32 val_h = RREG32(addr + sizeof(u32) - CFG_BASE); |
| 6090 | |
| 6091 | *val = (((u64) val_h) << 32) | val_l; |
| 6092 | } |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 6093 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6094 | } else if ((addr >= SRAM_BASE_ADDR) && |
| 6095 | (addr <= SRAM_BASE_ADDR + SRAM_BAR_SIZE - sizeof(u64))) { |
| 6096 | *val = readq(hdev->pcie_bar[SRAM_BAR_ID] + |
| 6097 | (addr - SRAM_BASE_ADDR)); |
| 6098 | } else if (addr <= |
| 6099 | DRAM_PHYS_BASE + hdev->asic_prop.dram_size - sizeof(u64)) { |
| 6100 | u64 bar_base_addr = DRAM_PHYS_BASE + |
| 6101 | (addr & ~(prop->dram_pci_bar_size - 0x1ull)); |
| 6102 | |
| 6103 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, bar_base_addr); |
| 6104 | if (hbm_bar_addr != U64_MAX) { |
| 6105 | *val = readq(hdev->pcie_bar[HBM_BAR_ID] + |
| 6106 | (addr - bar_base_addr)); |
| 6107 | |
| 6108 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, |
| 6109 | hbm_bar_addr); |
| 6110 | } |
| 6111 | if (hbm_bar_addr == U64_MAX) |
| 6112 | rc = -EIO; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6113 | } else if (addr >= HOST_PHYS_BASE && addr < host_phys_end && |
| 6114 | user_address && !iommu_present(&pci_bus_type)) { |
| 6115 | *val = *(u64 *) phys_to_virt(addr - HOST_PHYS_BASE); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6116 | } else { |
| 6117 | rc = -EFAULT; |
| 6118 | } |
| 6119 | |
| 6120 | return rc; |
| 6121 | } |
| 6122 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6123 | static int gaudi_debugfs_write64(struct hl_device *hdev, u64 addr, |
| 6124 | bool user_address, u64 val) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6125 | { |
| 6126 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 6127 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6128 | u64 hbm_bar_addr, host_phys_end; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6129 | int rc = 0; |
| 6130 | |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6131 | host_phys_end = HOST_PHYS_BASE + HOST_PHYS_SIZE; |
| 6132 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6133 | if ((addr >= CFG_BASE) && (addr <= CFG_BASE + CFG_SIZE - sizeof(u64))) { |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 6134 | |
| 6135 | if ((gaudi->hw_cap_initialized & HW_CAP_CLK_GATE) && |
| 6136 | (hdev->clock_gating_mask & |
| 6137 | GAUDI_CLK_GATE_DEBUGFS_MASK)) { |
| 6138 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6139 | dev_err_ratelimited(hdev->dev, |
| 6140 | "Can't write register - clock gating is enabled!\n"); |
| 6141 | rc = -EFAULT; |
| 6142 | } else { |
| 6143 | WREG32(addr - CFG_BASE, lower_32_bits(val)); |
| 6144 | WREG32(addr + sizeof(u32) - CFG_BASE, |
| 6145 | upper_32_bits(val)); |
| 6146 | } |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 6147 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6148 | } else if ((addr >= SRAM_BASE_ADDR) && |
| 6149 | (addr <= SRAM_BASE_ADDR + SRAM_BAR_SIZE - sizeof(u64))) { |
| 6150 | writeq(val, hdev->pcie_bar[SRAM_BAR_ID] + |
| 6151 | (addr - SRAM_BASE_ADDR)); |
| 6152 | } else if (addr <= |
| 6153 | DRAM_PHYS_BASE + hdev->asic_prop.dram_size - sizeof(u64)) { |
| 6154 | u64 bar_base_addr = DRAM_PHYS_BASE + |
| 6155 | (addr & ~(prop->dram_pci_bar_size - 0x1ull)); |
| 6156 | |
| 6157 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, bar_base_addr); |
| 6158 | if (hbm_bar_addr != U64_MAX) { |
| 6159 | writeq(val, hdev->pcie_bar[HBM_BAR_ID] + |
| 6160 | (addr - bar_base_addr)); |
| 6161 | |
| 6162 | hbm_bar_addr = gaudi_set_hbm_bar_base(hdev, |
| 6163 | hbm_bar_addr); |
| 6164 | } |
| 6165 | if (hbm_bar_addr == U64_MAX) |
| 6166 | rc = -EIO; |
Ofir Bitton | a5778d1 | 2021-02-24 11:51:40 +0200 | [diff] [blame] | 6167 | } else if (addr >= HOST_PHYS_BASE && addr < host_phys_end && |
| 6168 | user_address && !iommu_present(&pci_bus_type)) { |
| 6169 | *(u64 *) phys_to_virt(addr - HOST_PHYS_BASE) = val; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6170 | } else { |
| 6171 | rc = -EFAULT; |
| 6172 | } |
| 6173 | |
| 6174 | return rc; |
| 6175 | } |
| 6176 | |
Oded Gabbay | 639781d | 2021-04-02 01:43:18 +0300 | [diff] [blame] | 6177 | static int gaudi_dma_core_transfer(struct hl_device *hdev, int dma_id, u64 addr, |
| 6178 | u32 size_to_dma, dma_addr_t dma_addr) |
| 6179 | { |
| 6180 | u32 err_cause, val; |
| 6181 | u64 dma_offset; |
| 6182 | int rc; |
| 6183 | |
| 6184 | dma_offset = dma_id * DMA_CORE_OFFSET; |
| 6185 | |
| 6186 | WREG32(mmDMA0_CORE_SRC_BASE_LO + dma_offset, lower_32_bits(addr)); |
| 6187 | WREG32(mmDMA0_CORE_SRC_BASE_HI + dma_offset, upper_32_bits(addr)); |
| 6188 | WREG32(mmDMA0_CORE_DST_BASE_LO + dma_offset, lower_32_bits(dma_addr)); |
| 6189 | WREG32(mmDMA0_CORE_DST_BASE_HI + dma_offset, upper_32_bits(dma_addr)); |
| 6190 | WREG32(mmDMA0_CORE_DST_TSIZE_0 + dma_offset, size_to_dma); |
| 6191 | WREG32(mmDMA0_CORE_COMMIT + dma_offset, |
| 6192 | (1 << DMA0_CORE_COMMIT_LIN_SHIFT)); |
| 6193 | |
| 6194 | rc = hl_poll_timeout( |
| 6195 | hdev, |
| 6196 | mmDMA0_CORE_STS0 + dma_offset, |
| 6197 | val, |
| 6198 | ((val & DMA0_CORE_STS0_BUSY_MASK) == 0), |
| 6199 | 0, |
| 6200 | 1000000); |
| 6201 | |
| 6202 | if (rc) { |
| 6203 | dev_err(hdev->dev, |
| 6204 | "DMA %d timed-out during reading of 0x%llx\n", |
| 6205 | dma_id, addr); |
| 6206 | return -EIO; |
| 6207 | } |
| 6208 | |
| 6209 | /* Verify DMA is OK */ |
| 6210 | err_cause = RREG32(mmDMA0_CORE_ERR_CAUSE + dma_offset); |
| 6211 | if (err_cause) { |
| 6212 | dev_err(hdev->dev, "DMA Failed, cause 0x%x\n", err_cause); |
| 6213 | dev_dbg(hdev->dev, |
| 6214 | "Clearing DMA0 engine from errors (cause 0x%x)\n", |
| 6215 | err_cause); |
| 6216 | WREG32(mmDMA0_CORE_ERR_CAUSE + dma_offset, err_cause); |
| 6217 | |
| 6218 | return -EIO; |
| 6219 | } |
| 6220 | |
| 6221 | return 0; |
| 6222 | } |
| 6223 | |
| 6224 | static int gaudi_debugfs_read_dma(struct hl_device *hdev, u64 addr, u32 size, |
| 6225 | void *blob_addr) |
| 6226 | { |
| 6227 | u32 dma_core_sts0, err_cause, cfg1, size_left, pos, size_to_dma; |
| 6228 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 6229 | u64 dma_offset, qm_offset; |
| 6230 | dma_addr_t dma_addr; |
| 6231 | void *kernel_addr; |
| 6232 | bool is_eng_idle; |
Colin Ian King | b4e964f | 2021-04-12 17:10:12 +0100 | [diff] [blame] | 6233 | int rc = 0, dma_id; |
Oded Gabbay | 639781d | 2021-04-02 01:43:18 +0300 | [diff] [blame] | 6234 | |
| 6235 | kernel_addr = hdev->asic_funcs->asic_dma_alloc_coherent( |
| 6236 | hdev, SZ_2M, |
| 6237 | &dma_addr, |
| 6238 | GFP_KERNEL | __GFP_ZERO); |
| 6239 | |
| 6240 | if (!kernel_addr) |
| 6241 | return -ENOMEM; |
| 6242 | |
| 6243 | mutex_lock(&gaudi->clk_gate_mutex); |
| 6244 | |
| 6245 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 6246 | |
| 6247 | hdev->asic_funcs->hw_queues_lock(hdev); |
| 6248 | |
| 6249 | dma_id = gaudi_dma_assignment[GAUDI_PCI_DMA_1]; |
| 6250 | dma_offset = dma_id * DMA_CORE_OFFSET; |
| 6251 | qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 6252 | dma_core_sts0 = RREG32(mmDMA0_CORE_STS0 + dma_offset); |
| 6253 | is_eng_idle = IS_DMA_IDLE(dma_core_sts0); |
| 6254 | |
| 6255 | if (!is_eng_idle) { |
| 6256 | dma_id = gaudi_dma_assignment[GAUDI_PCI_DMA_2]; |
| 6257 | dma_offset = dma_id * DMA_CORE_OFFSET; |
| 6258 | qm_offset = dma_id * DMA_QMAN_OFFSET; |
| 6259 | dma_core_sts0 = RREG32(mmDMA0_CORE_STS0 + dma_offset); |
| 6260 | is_eng_idle = IS_DMA_IDLE(dma_core_sts0); |
| 6261 | |
| 6262 | if (!is_eng_idle) { |
| 6263 | dev_err_ratelimited(hdev->dev, |
| 6264 | "Can't read via DMA because it is BUSY\n"); |
| 6265 | rc = -EAGAIN; |
| 6266 | goto out; |
| 6267 | } |
| 6268 | } |
| 6269 | |
| 6270 | cfg1 = RREG32(mmDMA0_QM_GLBL_CFG1 + qm_offset); |
| 6271 | WREG32(mmDMA0_QM_GLBL_CFG1 + qm_offset, |
| 6272 | 0xF << DMA0_QM_GLBL_CFG1_CP_STOP_SHIFT); |
| 6273 | |
| 6274 | /* TODO: remove this by mapping the DMA temporary buffer to the MMU |
| 6275 | * using the compute ctx ASID, if exists. If not, use the kernel ctx |
| 6276 | * ASID |
| 6277 | */ |
| 6278 | WREG32_OR(mmDMA0_CORE_PROT + dma_offset, BIT(DMA0_CORE_PROT_VAL_SHIFT)); |
| 6279 | |
| 6280 | /* Verify DMA is OK */ |
| 6281 | err_cause = RREG32(mmDMA0_CORE_ERR_CAUSE + dma_offset); |
| 6282 | if (err_cause) { |
| 6283 | dev_dbg(hdev->dev, |
| 6284 | "Clearing DMA0 engine from errors (cause 0x%x)\n", |
| 6285 | err_cause); |
| 6286 | WREG32(mmDMA0_CORE_ERR_CAUSE + dma_offset, err_cause); |
| 6287 | } |
| 6288 | |
| 6289 | pos = 0; |
| 6290 | size_left = size; |
| 6291 | size_to_dma = SZ_2M; |
| 6292 | |
| 6293 | while (size_left > 0) { |
| 6294 | |
| 6295 | if (size_left < SZ_2M) |
| 6296 | size_to_dma = size_left; |
| 6297 | |
| 6298 | rc = gaudi_dma_core_transfer(hdev, dma_id, addr, size_to_dma, |
| 6299 | dma_addr); |
| 6300 | if (rc) |
| 6301 | break; |
| 6302 | |
| 6303 | memcpy(blob_addr + pos, kernel_addr, size_to_dma); |
| 6304 | |
| 6305 | if (size_left <= SZ_2M) |
| 6306 | break; |
| 6307 | |
| 6308 | pos += SZ_2M; |
| 6309 | addr += SZ_2M; |
| 6310 | size_left -= SZ_2M; |
| 6311 | } |
| 6312 | |
| 6313 | /* TODO: remove this by mapping the DMA temporary buffer to the MMU |
| 6314 | * using the compute ctx ASID, if exists. If not, use the kernel ctx |
| 6315 | * ASID |
| 6316 | */ |
| 6317 | WREG32_AND(mmDMA0_CORE_PROT + dma_offset, |
| 6318 | ~BIT(DMA0_CORE_PROT_VAL_SHIFT)); |
| 6319 | |
| 6320 | WREG32(mmDMA0_QM_GLBL_CFG1 + qm_offset, cfg1); |
| 6321 | |
| 6322 | out: |
| 6323 | hdev->asic_funcs->hw_queues_unlock(hdev); |
| 6324 | |
| 6325 | hdev->asic_funcs->set_clock_gating(hdev); |
| 6326 | |
| 6327 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 6328 | |
| 6329 | hdev->asic_funcs->asic_dma_free_coherent(hdev, SZ_2M, kernel_addr, |
| 6330 | dma_addr); |
| 6331 | |
| 6332 | return rc; |
| 6333 | } |
| 6334 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6335 | static u64 gaudi_read_pte(struct hl_device *hdev, u64 addr) |
| 6336 | { |
| 6337 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 6338 | |
| 6339 | if (hdev->hard_reset_pending) |
| 6340 | return U64_MAX; |
| 6341 | |
| 6342 | return readq(hdev->pcie_bar[HBM_BAR_ID] + |
| 6343 | (addr - gaudi->hbm_bar_cur_addr)); |
| 6344 | } |
| 6345 | |
| 6346 | static void gaudi_write_pte(struct hl_device *hdev, u64 addr, u64 val) |
| 6347 | { |
| 6348 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 6349 | |
| 6350 | if (hdev->hard_reset_pending) |
| 6351 | return; |
| 6352 | |
| 6353 | writeq(val, hdev->pcie_bar[HBM_BAR_ID] + |
| 6354 | (addr - gaudi->hbm_bar_cur_addr)); |
| 6355 | } |
| 6356 | |
Ofir Bitton | 1137e1e | 2020-09-30 18:43:52 +0300 | [diff] [blame] | 6357 | void gaudi_mmu_prepare_reg(struct hl_device *hdev, u64 reg, u32 asid) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6358 | { |
| 6359 | /* mask to zero the MMBP and ASID bits */ |
| 6360 | WREG32_AND(reg, ~0x7FF); |
| 6361 | WREG32_OR(reg, asid); |
| 6362 | } |
| 6363 | |
| 6364 | static void gaudi_mmu_prepare(struct hl_device *hdev, u32 asid) |
| 6365 | { |
| 6366 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 6367 | |
| 6368 | if (!(gaudi->hw_cap_initialized & HW_CAP_MMU)) |
| 6369 | return; |
| 6370 | |
| 6371 | if (asid & ~DMA0_QM_GLBL_NON_SECURE_PROPS_0_ASID_MASK) { |
Alon Mizrahi | 75d9a2a | 2020-12-03 17:32:19 +0200 | [diff] [blame] | 6372 | dev_crit(hdev->dev, "asid %u is too big\n", asid); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6373 | return; |
| 6374 | } |
| 6375 | |
| 6376 | mutex_lock(&gaudi->clk_gate_mutex); |
| 6377 | |
| 6378 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 6379 | |
| 6380 | gaudi_mmu_prepare_reg(hdev, mmDMA0_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6381 | gaudi_mmu_prepare_reg(hdev, mmDMA0_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6382 | gaudi_mmu_prepare_reg(hdev, mmDMA0_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6383 | gaudi_mmu_prepare_reg(hdev, mmDMA0_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6384 | gaudi_mmu_prepare_reg(hdev, mmDMA0_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6385 | |
| 6386 | gaudi_mmu_prepare_reg(hdev, mmDMA1_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6387 | gaudi_mmu_prepare_reg(hdev, mmDMA1_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6388 | gaudi_mmu_prepare_reg(hdev, mmDMA1_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6389 | gaudi_mmu_prepare_reg(hdev, mmDMA1_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6390 | gaudi_mmu_prepare_reg(hdev, mmDMA1_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6391 | |
| 6392 | gaudi_mmu_prepare_reg(hdev, mmDMA2_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6393 | gaudi_mmu_prepare_reg(hdev, mmDMA2_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6394 | gaudi_mmu_prepare_reg(hdev, mmDMA2_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6395 | gaudi_mmu_prepare_reg(hdev, mmDMA2_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6396 | gaudi_mmu_prepare_reg(hdev, mmDMA2_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6397 | |
| 6398 | gaudi_mmu_prepare_reg(hdev, mmDMA3_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6399 | gaudi_mmu_prepare_reg(hdev, mmDMA3_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6400 | gaudi_mmu_prepare_reg(hdev, mmDMA3_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6401 | gaudi_mmu_prepare_reg(hdev, mmDMA3_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6402 | gaudi_mmu_prepare_reg(hdev, mmDMA3_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6403 | |
| 6404 | gaudi_mmu_prepare_reg(hdev, mmDMA4_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6405 | gaudi_mmu_prepare_reg(hdev, mmDMA4_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6406 | gaudi_mmu_prepare_reg(hdev, mmDMA4_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6407 | gaudi_mmu_prepare_reg(hdev, mmDMA4_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6408 | gaudi_mmu_prepare_reg(hdev, mmDMA4_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6409 | |
| 6410 | gaudi_mmu_prepare_reg(hdev, mmDMA5_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6411 | gaudi_mmu_prepare_reg(hdev, mmDMA5_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6412 | gaudi_mmu_prepare_reg(hdev, mmDMA5_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6413 | gaudi_mmu_prepare_reg(hdev, mmDMA5_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6414 | gaudi_mmu_prepare_reg(hdev, mmDMA5_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6415 | |
| 6416 | gaudi_mmu_prepare_reg(hdev, mmDMA6_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6417 | gaudi_mmu_prepare_reg(hdev, mmDMA6_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6418 | gaudi_mmu_prepare_reg(hdev, mmDMA6_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6419 | gaudi_mmu_prepare_reg(hdev, mmDMA6_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6420 | gaudi_mmu_prepare_reg(hdev, mmDMA6_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6421 | |
| 6422 | gaudi_mmu_prepare_reg(hdev, mmDMA7_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6423 | gaudi_mmu_prepare_reg(hdev, mmDMA7_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6424 | gaudi_mmu_prepare_reg(hdev, mmDMA7_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6425 | gaudi_mmu_prepare_reg(hdev, mmDMA7_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6426 | gaudi_mmu_prepare_reg(hdev, mmDMA7_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6427 | |
| 6428 | gaudi_mmu_prepare_reg(hdev, mmDMA0_CORE_NON_SECURE_PROPS, asid); |
| 6429 | gaudi_mmu_prepare_reg(hdev, mmDMA1_CORE_NON_SECURE_PROPS, asid); |
| 6430 | gaudi_mmu_prepare_reg(hdev, mmDMA2_CORE_NON_SECURE_PROPS, asid); |
| 6431 | gaudi_mmu_prepare_reg(hdev, mmDMA3_CORE_NON_SECURE_PROPS, asid); |
| 6432 | gaudi_mmu_prepare_reg(hdev, mmDMA4_CORE_NON_SECURE_PROPS, asid); |
| 6433 | gaudi_mmu_prepare_reg(hdev, mmDMA5_CORE_NON_SECURE_PROPS, asid); |
| 6434 | gaudi_mmu_prepare_reg(hdev, mmDMA6_CORE_NON_SECURE_PROPS, asid); |
| 6435 | gaudi_mmu_prepare_reg(hdev, mmDMA7_CORE_NON_SECURE_PROPS, asid); |
| 6436 | |
| 6437 | gaudi_mmu_prepare_reg(hdev, mmTPC0_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6438 | gaudi_mmu_prepare_reg(hdev, mmTPC0_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6439 | gaudi_mmu_prepare_reg(hdev, mmTPC0_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6440 | gaudi_mmu_prepare_reg(hdev, mmTPC0_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6441 | gaudi_mmu_prepare_reg(hdev, mmTPC0_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6442 | gaudi_mmu_prepare_reg(hdev, mmTPC0_CFG_ARUSER_LO, asid); |
| 6443 | gaudi_mmu_prepare_reg(hdev, mmTPC0_CFG_AWUSER_LO, asid); |
| 6444 | |
| 6445 | gaudi_mmu_prepare_reg(hdev, mmTPC1_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6446 | gaudi_mmu_prepare_reg(hdev, mmTPC1_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6447 | gaudi_mmu_prepare_reg(hdev, mmTPC1_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6448 | gaudi_mmu_prepare_reg(hdev, mmTPC1_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6449 | gaudi_mmu_prepare_reg(hdev, mmTPC1_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6450 | gaudi_mmu_prepare_reg(hdev, mmTPC1_CFG_ARUSER_LO, asid); |
| 6451 | gaudi_mmu_prepare_reg(hdev, mmTPC1_CFG_AWUSER_LO, asid); |
| 6452 | |
| 6453 | gaudi_mmu_prepare_reg(hdev, mmTPC2_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6454 | gaudi_mmu_prepare_reg(hdev, mmTPC2_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6455 | gaudi_mmu_prepare_reg(hdev, mmTPC2_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6456 | gaudi_mmu_prepare_reg(hdev, mmTPC2_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6457 | gaudi_mmu_prepare_reg(hdev, mmTPC2_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6458 | gaudi_mmu_prepare_reg(hdev, mmTPC2_CFG_ARUSER_LO, asid); |
| 6459 | gaudi_mmu_prepare_reg(hdev, mmTPC2_CFG_AWUSER_LO, asid); |
| 6460 | |
| 6461 | gaudi_mmu_prepare_reg(hdev, mmTPC3_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6462 | gaudi_mmu_prepare_reg(hdev, mmTPC3_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6463 | gaudi_mmu_prepare_reg(hdev, mmTPC3_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6464 | gaudi_mmu_prepare_reg(hdev, mmTPC3_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6465 | gaudi_mmu_prepare_reg(hdev, mmTPC3_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6466 | gaudi_mmu_prepare_reg(hdev, mmTPC3_CFG_ARUSER_LO, asid); |
| 6467 | gaudi_mmu_prepare_reg(hdev, mmTPC3_CFG_AWUSER_LO, asid); |
| 6468 | |
| 6469 | gaudi_mmu_prepare_reg(hdev, mmTPC4_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6470 | gaudi_mmu_prepare_reg(hdev, mmTPC4_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6471 | gaudi_mmu_prepare_reg(hdev, mmTPC4_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6472 | gaudi_mmu_prepare_reg(hdev, mmTPC4_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6473 | gaudi_mmu_prepare_reg(hdev, mmTPC4_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6474 | gaudi_mmu_prepare_reg(hdev, mmTPC4_CFG_ARUSER_LO, asid); |
| 6475 | gaudi_mmu_prepare_reg(hdev, mmTPC4_CFG_AWUSER_LO, asid); |
| 6476 | |
| 6477 | gaudi_mmu_prepare_reg(hdev, mmTPC5_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6478 | gaudi_mmu_prepare_reg(hdev, mmTPC5_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6479 | gaudi_mmu_prepare_reg(hdev, mmTPC5_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6480 | gaudi_mmu_prepare_reg(hdev, mmTPC5_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6481 | gaudi_mmu_prepare_reg(hdev, mmTPC5_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6482 | gaudi_mmu_prepare_reg(hdev, mmTPC5_CFG_ARUSER_LO, asid); |
| 6483 | gaudi_mmu_prepare_reg(hdev, mmTPC5_CFG_AWUSER_LO, asid); |
| 6484 | |
| 6485 | gaudi_mmu_prepare_reg(hdev, mmTPC6_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6486 | gaudi_mmu_prepare_reg(hdev, mmTPC6_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6487 | gaudi_mmu_prepare_reg(hdev, mmTPC6_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6488 | gaudi_mmu_prepare_reg(hdev, mmTPC6_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6489 | gaudi_mmu_prepare_reg(hdev, mmTPC6_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6490 | gaudi_mmu_prepare_reg(hdev, mmTPC6_CFG_ARUSER_LO, asid); |
| 6491 | gaudi_mmu_prepare_reg(hdev, mmTPC6_CFG_AWUSER_LO, asid); |
| 6492 | |
| 6493 | gaudi_mmu_prepare_reg(hdev, mmTPC7_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6494 | gaudi_mmu_prepare_reg(hdev, mmTPC7_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6495 | gaudi_mmu_prepare_reg(hdev, mmTPC7_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6496 | gaudi_mmu_prepare_reg(hdev, mmTPC7_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6497 | gaudi_mmu_prepare_reg(hdev, mmTPC7_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6498 | gaudi_mmu_prepare_reg(hdev, mmTPC7_CFG_ARUSER_LO, asid); |
| 6499 | gaudi_mmu_prepare_reg(hdev, mmTPC7_CFG_AWUSER_LO, asid); |
| 6500 | |
| 6501 | gaudi_mmu_prepare_reg(hdev, mmMME0_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6502 | gaudi_mmu_prepare_reg(hdev, mmMME0_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6503 | gaudi_mmu_prepare_reg(hdev, mmMME0_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6504 | gaudi_mmu_prepare_reg(hdev, mmMME0_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6505 | gaudi_mmu_prepare_reg(hdev, mmMME0_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6506 | gaudi_mmu_prepare_reg(hdev, mmMME2_QM_GLBL_NON_SECURE_PROPS_0, asid); |
| 6507 | gaudi_mmu_prepare_reg(hdev, mmMME2_QM_GLBL_NON_SECURE_PROPS_1, asid); |
| 6508 | gaudi_mmu_prepare_reg(hdev, mmMME2_QM_GLBL_NON_SECURE_PROPS_2, asid); |
| 6509 | gaudi_mmu_prepare_reg(hdev, mmMME2_QM_GLBL_NON_SECURE_PROPS_3, asid); |
| 6510 | gaudi_mmu_prepare_reg(hdev, mmMME2_QM_GLBL_NON_SECURE_PROPS_4, asid); |
| 6511 | |
| 6512 | gaudi_mmu_prepare_reg(hdev, mmMME0_SBAB_ARUSER0, asid); |
| 6513 | gaudi_mmu_prepare_reg(hdev, mmMME0_SBAB_ARUSER1, asid); |
| 6514 | gaudi_mmu_prepare_reg(hdev, mmMME1_SBAB_ARUSER0, asid); |
| 6515 | gaudi_mmu_prepare_reg(hdev, mmMME1_SBAB_ARUSER1, asid); |
| 6516 | gaudi_mmu_prepare_reg(hdev, mmMME2_SBAB_ARUSER0, asid); |
| 6517 | gaudi_mmu_prepare_reg(hdev, mmMME2_SBAB_ARUSER1, asid); |
| 6518 | gaudi_mmu_prepare_reg(hdev, mmMME3_SBAB_ARUSER0, asid); |
| 6519 | gaudi_mmu_prepare_reg(hdev, mmMME3_SBAB_ARUSER1, asid); |
| 6520 | gaudi_mmu_prepare_reg(hdev, mmMME0_ACC_WBC, asid); |
| 6521 | gaudi_mmu_prepare_reg(hdev, mmMME1_ACC_WBC, asid); |
| 6522 | gaudi_mmu_prepare_reg(hdev, mmMME2_ACC_WBC, asid); |
| 6523 | gaudi_mmu_prepare_reg(hdev, mmMME3_ACC_WBC, asid); |
| 6524 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 6525 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC0) { |
| 6526 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM0_GLBL_NON_SECURE_PROPS_0, |
| 6527 | asid); |
| 6528 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM0_GLBL_NON_SECURE_PROPS_1, |
| 6529 | asid); |
| 6530 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM0_GLBL_NON_SECURE_PROPS_2, |
| 6531 | asid); |
| 6532 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM0_GLBL_NON_SECURE_PROPS_3, |
| 6533 | asid); |
| 6534 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM0_GLBL_NON_SECURE_PROPS_4, |
| 6535 | asid); |
| 6536 | } |
| 6537 | |
| 6538 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC1) { |
| 6539 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM1_GLBL_NON_SECURE_PROPS_0, |
| 6540 | asid); |
| 6541 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM1_GLBL_NON_SECURE_PROPS_1, |
| 6542 | asid); |
| 6543 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM1_GLBL_NON_SECURE_PROPS_2, |
| 6544 | asid); |
| 6545 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM1_GLBL_NON_SECURE_PROPS_3, |
| 6546 | asid); |
| 6547 | gaudi_mmu_prepare_reg(hdev, mmNIC0_QM1_GLBL_NON_SECURE_PROPS_4, |
| 6548 | asid); |
| 6549 | } |
| 6550 | |
| 6551 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC2) { |
| 6552 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM0_GLBL_NON_SECURE_PROPS_0, |
| 6553 | asid); |
| 6554 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM0_GLBL_NON_SECURE_PROPS_1, |
| 6555 | asid); |
| 6556 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM0_GLBL_NON_SECURE_PROPS_2, |
| 6557 | asid); |
| 6558 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM0_GLBL_NON_SECURE_PROPS_3, |
| 6559 | asid); |
| 6560 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM0_GLBL_NON_SECURE_PROPS_4, |
| 6561 | asid); |
| 6562 | } |
| 6563 | |
| 6564 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC3) { |
| 6565 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM1_GLBL_NON_SECURE_PROPS_0, |
| 6566 | asid); |
| 6567 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM1_GLBL_NON_SECURE_PROPS_1, |
| 6568 | asid); |
| 6569 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM1_GLBL_NON_SECURE_PROPS_2, |
| 6570 | asid); |
| 6571 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM1_GLBL_NON_SECURE_PROPS_3, |
| 6572 | asid); |
| 6573 | gaudi_mmu_prepare_reg(hdev, mmNIC1_QM1_GLBL_NON_SECURE_PROPS_4, |
| 6574 | asid); |
| 6575 | } |
| 6576 | |
| 6577 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC4) { |
| 6578 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM0_GLBL_NON_SECURE_PROPS_0, |
| 6579 | asid); |
| 6580 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM0_GLBL_NON_SECURE_PROPS_1, |
| 6581 | asid); |
| 6582 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM0_GLBL_NON_SECURE_PROPS_2, |
| 6583 | asid); |
| 6584 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM0_GLBL_NON_SECURE_PROPS_3, |
| 6585 | asid); |
| 6586 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM0_GLBL_NON_SECURE_PROPS_4, |
| 6587 | asid); |
| 6588 | } |
| 6589 | |
| 6590 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC5) { |
| 6591 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM1_GLBL_NON_SECURE_PROPS_0, |
| 6592 | asid); |
| 6593 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM1_GLBL_NON_SECURE_PROPS_1, |
| 6594 | asid); |
| 6595 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM1_GLBL_NON_SECURE_PROPS_2, |
| 6596 | asid); |
| 6597 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM1_GLBL_NON_SECURE_PROPS_3, |
| 6598 | asid); |
| 6599 | gaudi_mmu_prepare_reg(hdev, mmNIC2_QM1_GLBL_NON_SECURE_PROPS_4, |
| 6600 | asid); |
| 6601 | } |
| 6602 | |
| 6603 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC6) { |
| 6604 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM0_GLBL_NON_SECURE_PROPS_0, |
| 6605 | asid); |
| 6606 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM0_GLBL_NON_SECURE_PROPS_1, |
| 6607 | asid); |
| 6608 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM0_GLBL_NON_SECURE_PROPS_2, |
| 6609 | asid); |
| 6610 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM0_GLBL_NON_SECURE_PROPS_3, |
| 6611 | asid); |
| 6612 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM0_GLBL_NON_SECURE_PROPS_4, |
| 6613 | asid); |
| 6614 | } |
| 6615 | |
| 6616 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC7) { |
| 6617 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM1_GLBL_NON_SECURE_PROPS_0, |
| 6618 | asid); |
| 6619 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM1_GLBL_NON_SECURE_PROPS_1, |
| 6620 | asid); |
| 6621 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM1_GLBL_NON_SECURE_PROPS_2, |
| 6622 | asid); |
| 6623 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM1_GLBL_NON_SECURE_PROPS_3, |
| 6624 | asid); |
| 6625 | gaudi_mmu_prepare_reg(hdev, mmNIC3_QM1_GLBL_NON_SECURE_PROPS_4, |
| 6626 | asid); |
| 6627 | } |
| 6628 | |
| 6629 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC8) { |
| 6630 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM0_GLBL_NON_SECURE_PROPS_0, |
| 6631 | asid); |
| 6632 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM0_GLBL_NON_SECURE_PROPS_1, |
| 6633 | asid); |
| 6634 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM0_GLBL_NON_SECURE_PROPS_2, |
| 6635 | asid); |
| 6636 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM0_GLBL_NON_SECURE_PROPS_3, |
| 6637 | asid); |
| 6638 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM0_GLBL_NON_SECURE_PROPS_4, |
| 6639 | asid); |
| 6640 | } |
| 6641 | |
| 6642 | if (hdev->nic_ports_mask & GAUDI_NIC_MASK_NIC9) { |
| 6643 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM1_GLBL_NON_SECURE_PROPS_0, |
| 6644 | asid); |
| 6645 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM1_GLBL_NON_SECURE_PROPS_1, |
| 6646 | asid); |
| 6647 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM1_GLBL_NON_SECURE_PROPS_2, |
| 6648 | asid); |
| 6649 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM1_GLBL_NON_SECURE_PROPS_3, |
| 6650 | asid); |
| 6651 | gaudi_mmu_prepare_reg(hdev, mmNIC4_QM1_GLBL_NON_SECURE_PROPS_4, |
| 6652 | asid); |
| 6653 | } |
| 6654 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 6655 | hdev->asic_funcs->set_clock_gating(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6656 | |
| 6657 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 6658 | } |
| 6659 | |
| 6660 | static int gaudi_send_job_on_qman0(struct hl_device *hdev, |
| 6661 | struct hl_cs_job *job) |
| 6662 | { |
| 6663 | struct packet_msg_prot *fence_pkt; |
| 6664 | u32 *fence_ptr; |
| 6665 | dma_addr_t fence_dma_addr; |
| 6666 | struct hl_cb *cb; |
| 6667 | u32 tmp, timeout, dma_offset; |
| 6668 | int rc; |
| 6669 | |
| 6670 | if (hdev->pldm) |
| 6671 | timeout = GAUDI_PLDM_QMAN0_TIMEOUT_USEC; |
| 6672 | else |
| 6673 | timeout = HL_DEVICE_TIMEOUT_USEC; |
| 6674 | |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 6675 | if (!hdev->asic_funcs->is_device_idle(hdev, NULL, 0, NULL)) { |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6676 | dev_err_ratelimited(hdev->dev, |
| 6677 | "Can't send driver job on QMAN0 because the device is not idle\n"); |
| 6678 | return -EBUSY; |
| 6679 | } |
| 6680 | |
| 6681 | fence_ptr = hdev->asic_funcs->asic_dma_pool_zalloc(hdev, 4, GFP_KERNEL, |
| 6682 | &fence_dma_addr); |
| 6683 | if (!fence_ptr) { |
| 6684 | dev_err(hdev->dev, |
| 6685 | "Failed to allocate fence memory for QMAN0\n"); |
| 6686 | return -ENOMEM; |
| 6687 | } |
| 6688 | |
| 6689 | cb = job->patched_cb; |
| 6690 | |
Arnd Bergmann | 82948e6 | 2020-10-26 17:08:06 +0100 | [diff] [blame] | 6691 | fence_pkt = cb->kernel_address + |
| 6692 | job->job_cb_size - sizeof(struct packet_msg_prot); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6693 | |
Oded Gabbay | 6588729 | 2020-08-12 11:21:01 +0300 | [diff] [blame] | 6694 | tmp = FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_PROT); |
| 6695 | tmp |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 1); |
| 6696 | tmp |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
| 6697 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6698 | fence_pkt->ctl = cpu_to_le32(tmp); |
| 6699 | fence_pkt->value = cpu_to_le32(GAUDI_QMAN0_FENCE_VAL); |
| 6700 | fence_pkt->addr = cpu_to_le64(fence_dma_addr); |
| 6701 | |
| 6702 | dma_offset = gaudi_dma_assignment[GAUDI_PCI_DMA_1] * DMA_CORE_OFFSET; |
| 6703 | |
| 6704 | WREG32_OR(mmDMA0_CORE_PROT + dma_offset, BIT(DMA0_CORE_PROT_VAL_SHIFT)); |
| 6705 | |
| 6706 | rc = hl_hw_queue_send_cb_no_cmpl(hdev, GAUDI_QUEUE_ID_DMA_0_0, |
| 6707 | job->job_cb_size, cb->bus_address); |
| 6708 | if (rc) { |
| 6709 | dev_err(hdev->dev, "Failed to send CB on QMAN0, %d\n", rc); |
| 6710 | goto free_fence_ptr; |
| 6711 | } |
| 6712 | |
| 6713 | rc = hl_poll_timeout_memory(hdev, fence_ptr, tmp, |
| 6714 | (tmp == GAUDI_QMAN0_FENCE_VAL), 1000, |
| 6715 | timeout, true); |
| 6716 | |
| 6717 | hl_hw_queue_inc_ci_kernel(hdev, GAUDI_QUEUE_ID_DMA_0_0); |
| 6718 | |
| 6719 | if (rc == -ETIMEDOUT) { |
| 6720 | dev_err(hdev->dev, "QMAN0 Job timeout (0x%x)\n", tmp); |
| 6721 | goto free_fence_ptr; |
| 6722 | } |
| 6723 | |
| 6724 | free_fence_ptr: |
| 6725 | WREG32_AND(mmDMA0_CORE_PROT + dma_offset, |
| 6726 | ~BIT(DMA0_CORE_PROT_VAL_SHIFT)); |
| 6727 | |
| 6728 | hdev->asic_funcs->asic_dma_pool_free(hdev, (void *) fence_ptr, |
| 6729 | fence_dma_addr); |
| 6730 | return rc; |
| 6731 | } |
| 6732 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6733 | static void gaudi_get_event_desc(u16 event_type, char *desc, size_t size) |
| 6734 | { |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 6735 | if (event_type >= GAUDI_EVENT_SIZE) |
| 6736 | goto event_not_supported; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6737 | |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 6738 | if (!gaudi_irq_map_table[event_type].valid) |
| 6739 | goto event_not_supported; |
| 6740 | |
| 6741 | snprintf(desc, size, gaudi_irq_map_table[event_type].name); |
| 6742 | |
| 6743 | return; |
| 6744 | |
| 6745 | event_not_supported: |
| 6746 | snprintf(desc, size, "N/A"); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6747 | } |
| 6748 | |
| 6749 | static const char *gaudi_get_razwi_initiator_dma_name(struct hl_device *hdev, |
| 6750 | u32 x_y, bool is_write) |
| 6751 | { |
| 6752 | u32 dma_id[2], dma_offset, err_cause[2], mask, i; |
| 6753 | |
| 6754 | mask = is_write ? DMA0_CORE_ERR_CAUSE_HBW_WR_ERR_MASK : |
| 6755 | DMA0_CORE_ERR_CAUSE_HBW_RD_ERR_MASK; |
| 6756 | |
| 6757 | switch (x_y) { |
| 6758 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_S_0: |
| 6759 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_S_1: |
| 6760 | dma_id[0] = 0; |
| 6761 | dma_id[1] = 2; |
| 6762 | break; |
| 6763 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_S_0: |
| 6764 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_S_1: |
| 6765 | dma_id[0] = 1; |
| 6766 | dma_id[1] = 3; |
| 6767 | break; |
| 6768 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_N_0: |
| 6769 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_N_1: |
| 6770 | dma_id[0] = 4; |
| 6771 | dma_id[1] = 6; |
| 6772 | break; |
| 6773 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_N_0: |
| 6774 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_N_1: |
| 6775 | dma_id[0] = 5; |
| 6776 | dma_id[1] = 7; |
| 6777 | break; |
| 6778 | default: |
| 6779 | goto unknown_initiator; |
| 6780 | } |
| 6781 | |
| 6782 | for (i = 0 ; i < 2 ; i++) { |
| 6783 | dma_offset = dma_id[i] * DMA_CORE_OFFSET; |
| 6784 | err_cause[i] = RREG32(mmDMA0_CORE_ERR_CAUSE + dma_offset); |
| 6785 | } |
| 6786 | |
| 6787 | switch (x_y) { |
| 6788 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_S_0: |
| 6789 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_S_1: |
| 6790 | if ((err_cause[0] & mask) && !(err_cause[1] & mask)) |
| 6791 | return "DMA0"; |
| 6792 | else if (!(err_cause[0] & mask) && (err_cause[1] & mask)) |
| 6793 | return "DMA2"; |
| 6794 | else |
| 6795 | return "DMA0 or DMA2"; |
| 6796 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_S_0: |
| 6797 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_S_1: |
| 6798 | if ((err_cause[0] & mask) && !(err_cause[1] & mask)) |
| 6799 | return "DMA1"; |
| 6800 | else if (!(err_cause[0] & mask) && (err_cause[1] & mask)) |
| 6801 | return "DMA3"; |
| 6802 | else |
| 6803 | return "DMA1 or DMA3"; |
| 6804 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_N_0: |
| 6805 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_N_1: |
| 6806 | if ((err_cause[0] & mask) && !(err_cause[1] & mask)) |
| 6807 | return "DMA4"; |
| 6808 | else if (!(err_cause[0] & mask) && (err_cause[1] & mask)) |
| 6809 | return "DMA6"; |
| 6810 | else |
| 6811 | return "DMA4 or DMA6"; |
| 6812 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_N_0: |
| 6813 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_N_1: |
| 6814 | if ((err_cause[0] & mask) && !(err_cause[1] & mask)) |
| 6815 | return "DMA5"; |
| 6816 | else if (!(err_cause[0] & mask) && (err_cause[1] & mask)) |
| 6817 | return "DMA7"; |
| 6818 | else |
| 6819 | return "DMA5 or DMA7"; |
| 6820 | } |
| 6821 | |
| 6822 | unknown_initiator: |
| 6823 | return "unknown initiator"; |
| 6824 | } |
| 6825 | |
| 6826 | static const char *gaudi_get_razwi_initiator_name(struct hl_device *hdev, |
| 6827 | bool is_write) |
| 6828 | { |
| 6829 | u32 val, x_y, axi_id; |
| 6830 | |
| 6831 | val = is_write ? RREG32(mmMMU_UP_RAZWI_WRITE_ID) : |
| 6832 | RREG32(mmMMU_UP_RAZWI_READ_ID); |
| 6833 | x_y = val & ((RAZWI_INITIATOR_Y_MASK << RAZWI_INITIATOR_Y_SHIFT) | |
| 6834 | (RAZWI_INITIATOR_X_MASK << RAZWI_INITIATOR_X_SHIFT)); |
| 6835 | axi_id = val & (RAZWI_INITIATOR_AXI_ID_MASK << |
| 6836 | RAZWI_INITIATOR_AXI_ID_SHIFT); |
| 6837 | |
| 6838 | switch (x_y) { |
| 6839 | case RAZWI_INITIATOR_ID_X_Y_TPC0_NIC0: |
| 6840 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_TPC)) |
| 6841 | return "TPC0"; |
| 6842 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_NIC)) |
| 6843 | return "NIC0"; |
| 6844 | break; |
| 6845 | case RAZWI_INITIATOR_ID_X_Y_TPC1: |
| 6846 | return "TPC1"; |
| 6847 | case RAZWI_INITIATOR_ID_X_Y_MME0_0: |
| 6848 | case RAZWI_INITIATOR_ID_X_Y_MME0_1: |
| 6849 | return "MME0"; |
| 6850 | case RAZWI_INITIATOR_ID_X_Y_MME1_0: |
| 6851 | case RAZWI_INITIATOR_ID_X_Y_MME1_1: |
| 6852 | return "MME1"; |
| 6853 | case RAZWI_INITIATOR_ID_X_Y_TPC2: |
| 6854 | return "TPC2"; |
| 6855 | case RAZWI_INITIATOR_ID_X_Y_TPC3_PCI_CPU_PSOC: |
| 6856 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_TPC)) |
| 6857 | return "TPC3"; |
| 6858 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_PCI)) |
| 6859 | return "PCI"; |
| 6860 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_CPU)) |
| 6861 | return "CPU"; |
| 6862 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_PSOC)) |
| 6863 | return "PSOC"; |
| 6864 | break; |
| 6865 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_S_0: |
| 6866 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_S_1: |
| 6867 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_S_0: |
| 6868 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_S_1: |
| 6869 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_N_0: |
| 6870 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_W_N_1: |
| 6871 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_N_0: |
| 6872 | case RAZWI_INITIATOR_ID_X_Y_DMA_IF_E_N_1: |
| 6873 | return gaudi_get_razwi_initiator_dma_name(hdev, x_y, is_write); |
| 6874 | case RAZWI_INITIATOR_ID_X_Y_TPC4_NIC1_NIC2: |
| 6875 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_TPC)) |
| 6876 | return "TPC4"; |
| 6877 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_NIC)) |
| 6878 | return "NIC1"; |
| 6879 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_NIC_FT)) |
| 6880 | return "NIC2"; |
| 6881 | break; |
| 6882 | case RAZWI_INITIATOR_ID_X_Y_TPC5: |
| 6883 | return "TPC5"; |
| 6884 | case RAZWI_INITIATOR_ID_X_Y_MME2_0: |
| 6885 | case RAZWI_INITIATOR_ID_X_Y_MME2_1: |
| 6886 | return "MME2"; |
| 6887 | case RAZWI_INITIATOR_ID_X_Y_MME3_0: |
| 6888 | case RAZWI_INITIATOR_ID_X_Y_MME3_1: |
| 6889 | return "MME3"; |
| 6890 | case RAZWI_INITIATOR_ID_X_Y_TPC6: |
| 6891 | return "TPC6"; |
| 6892 | case RAZWI_INITIATOR_ID_X_Y_TPC7_NIC4_NIC5: |
| 6893 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_TPC)) |
| 6894 | return "TPC7"; |
| 6895 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_NIC)) |
| 6896 | return "NIC4"; |
| 6897 | if (axi_id == RAZWI_INITIATOR_ID_AXI_ID(AXI_ID_NIC_FT)) |
| 6898 | return "NIC5"; |
| 6899 | break; |
| 6900 | default: |
| 6901 | break; |
| 6902 | } |
| 6903 | |
| 6904 | dev_err(hdev->dev, |
| 6905 | "Unknown RAZWI initiator ID 0x%x [Y=%d, X=%d, AXI_ID=%d]\n", |
| 6906 | val, |
| 6907 | (val >> RAZWI_INITIATOR_Y_SHIFT) & RAZWI_INITIATOR_Y_MASK, |
| 6908 | (val >> RAZWI_INITIATOR_X_SHIFT) & RAZWI_INITIATOR_X_MASK, |
| 6909 | (val >> RAZWI_INITIATOR_AXI_ID_SHIFT) & |
| 6910 | RAZWI_INITIATOR_AXI_ID_MASK); |
| 6911 | |
| 6912 | return "unknown initiator"; |
| 6913 | } |
| 6914 | |
| 6915 | static void gaudi_print_razwi_info(struct hl_device *hdev) |
| 6916 | { |
| 6917 | if (RREG32(mmMMU_UP_RAZWI_WRITE_VLD)) { |
| 6918 | dev_err_ratelimited(hdev->dev, |
| 6919 | "RAZWI event caused by illegal write of %s\n", |
| 6920 | gaudi_get_razwi_initiator_name(hdev, true)); |
| 6921 | WREG32(mmMMU_UP_RAZWI_WRITE_VLD, 0); |
| 6922 | } |
| 6923 | |
| 6924 | if (RREG32(mmMMU_UP_RAZWI_READ_VLD)) { |
| 6925 | dev_err_ratelimited(hdev->dev, |
| 6926 | "RAZWI event caused by illegal read of %s\n", |
| 6927 | gaudi_get_razwi_initiator_name(hdev, false)); |
| 6928 | WREG32(mmMMU_UP_RAZWI_READ_VLD, 0); |
| 6929 | } |
| 6930 | } |
| 6931 | |
| 6932 | static void gaudi_print_mmu_error_info(struct hl_device *hdev) |
| 6933 | { |
| 6934 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 6935 | u64 addr; |
| 6936 | u32 val; |
| 6937 | |
| 6938 | if (!(gaudi->hw_cap_initialized & HW_CAP_MMU)) |
| 6939 | return; |
| 6940 | |
| 6941 | val = RREG32(mmMMU_UP_PAGE_ERROR_CAPTURE); |
| 6942 | if (val & MMU_UP_PAGE_ERROR_CAPTURE_ENTRY_VALID_MASK) { |
| 6943 | addr = val & MMU_UP_PAGE_ERROR_CAPTURE_VA_49_32_MASK; |
| 6944 | addr <<= 32; |
| 6945 | addr |= RREG32(mmMMU_UP_PAGE_ERROR_CAPTURE_VA); |
| 6946 | |
| 6947 | dev_err_ratelimited(hdev->dev, "MMU page fault on va 0x%llx\n", |
| 6948 | addr); |
| 6949 | |
| 6950 | WREG32(mmMMU_UP_PAGE_ERROR_CAPTURE, 0); |
| 6951 | } |
| 6952 | |
| 6953 | val = RREG32(mmMMU_UP_ACCESS_ERROR_CAPTURE); |
| 6954 | if (val & MMU_UP_ACCESS_ERROR_CAPTURE_ENTRY_VALID_MASK) { |
| 6955 | addr = val & MMU_UP_ACCESS_ERROR_CAPTURE_VA_49_32_MASK; |
| 6956 | addr <<= 32; |
| 6957 | addr |= RREG32(mmMMU_UP_ACCESS_ERROR_CAPTURE_VA); |
| 6958 | |
| 6959 | dev_err_ratelimited(hdev->dev, |
| 6960 | "MMU access error on va 0x%llx\n", addr); |
| 6961 | |
| 6962 | WREG32(mmMMU_UP_ACCESS_ERROR_CAPTURE, 0); |
| 6963 | } |
| 6964 | } |
| 6965 | |
| 6966 | /* |
| 6967 | * +-------------------+------------------------------------------------------+ |
| 6968 | * | Configuration Reg | Description | |
| 6969 | * | Address | | |
| 6970 | * +-------------------+------------------------------------------------------+ |
| 6971 | * | 0xF30 - 0xF3F |ECC single error indication (1 bit per memory wrapper)| |
| 6972 | * | |0xF30 memory wrappers 31:0 (MSB to LSB) | |
| 6973 | * | |0xF34 memory wrappers 63:32 | |
| 6974 | * | |0xF38 memory wrappers 95:64 | |
| 6975 | * | |0xF3C memory wrappers 127:96 | |
| 6976 | * +-------------------+------------------------------------------------------+ |
| 6977 | * | 0xF40 - 0xF4F |ECC double error indication (1 bit per memory wrapper)| |
| 6978 | * | |0xF40 memory wrappers 31:0 (MSB to LSB) | |
| 6979 | * | |0xF44 memory wrappers 63:32 | |
| 6980 | * | |0xF48 memory wrappers 95:64 | |
| 6981 | * | |0xF4C memory wrappers 127:96 | |
| 6982 | * +-------------------+------------------------------------------------------+ |
| 6983 | */ |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 6984 | static int gaudi_extract_ecc_info(struct hl_device *hdev, |
| 6985 | struct ecc_info_extract_params *params, u64 *ecc_address, |
| 6986 | u64 *ecc_syndrom, u8 *memory_wrapper_idx) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6987 | { |
| 6988 | struct gaudi_device *gaudi = hdev->asic_specific; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 6989 | u32 i, num_mem_regs, reg, err_bit; |
| 6990 | u64 err_addr, err_word = 0; |
| 6991 | int rc = 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6992 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 6993 | num_mem_regs = params->num_memories / 32 + |
| 6994 | ((params->num_memories % 32) ? 1 : 0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 6995 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 6996 | if (params->block_address >= CFG_BASE) |
| 6997 | params->block_address -= CFG_BASE; |
| 6998 | |
| 6999 | if (params->derr) |
| 7000 | err_addr = params->block_address + GAUDI_ECC_DERR0_OFFSET; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7001 | else |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7002 | err_addr = params->block_address + GAUDI_ECC_SERR0_OFFSET; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7003 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7004 | if (params->disable_clock_gating) { |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7005 | mutex_lock(&gaudi->clk_gate_mutex); |
| 7006 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 7007 | } |
| 7008 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7009 | /* Set invalid wrapper index */ |
| 7010 | *memory_wrapper_idx = 0xFF; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7011 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7012 | /* Iterate through memory wrappers, a single bit must be set */ |
Dan Carpenter | b035354 | 2020-08-05 12:51:05 +0300 | [diff] [blame] | 7013 | for (i = 0 ; i < num_mem_regs ; i++) { |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7014 | err_addr += i * 4; |
| 7015 | err_word = RREG32(err_addr); |
| 7016 | if (err_word) { |
| 7017 | err_bit = __ffs(err_word); |
| 7018 | *memory_wrapper_idx = err_bit + (32 * i); |
| 7019 | break; |
| 7020 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7021 | } |
| 7022 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7023 | if (*memory_wrapper_idx == 0xFF) { |
| 7024 | dev_err(hdev->dev, "ECC error information cannot be found\n"); |
| 7025 | rc = -EINVAL; |
| 7026 | goto enable_clk_gate; |
| 7027 | } |
| 7028 | |
| 7029 | WREG32(params->block_address + GAUDI_ECC_MEM_SEL_OFFSET, |
| 7030 | *memory_wrapper_idx); |
| 7031 | |
| 7032 | *ecc_address = |
| 7033 | RREG32(params->block_address + GAUDI_ECC_ADDRESS_OFFSET); |
| 7034 | *ecc_syndrom = |
| 7035 | RREG32(params->block_address + GAUDI_ECC_SYNDROME_OFFSET); |
| 7036 | |
| 7037 | /* Clear error indication */ |
| 7038 | reg = RREG32(params->block_address + GAUDI_ECC_MEM_INFO_CLR_OFFSET); |
| 7039 | if (params->derr) |
| 7040 | reg |= FIELD_PREP(GAUDI_ECC_MEM_INFO_CLR_DERR_MASK, 1); |
| 7041 | else |
| 7042 | reg |= FIELD_PREP(GAUDI_ECC_MEM_INFO_CLR_SERR_MASK, 1); |
| 7043 | |
| 7044 | WREG32(params->block_address + GAUDI_ECC_MEM_INFO_CLR_OFFSET, reg); |
| 7045 | |
| 7046 | enable_clk_gate: |
| 7047 | if (params->disable_clock_gating) { |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 7048 | hdev->asic_funcs->set_clock_gating(hdev); |
Greg Kroah-Hartman | 65a9bde6 | 2020-07-27 11:49:37 +0200 | [diff] [blame] | 7049 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7050 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 7051 | } |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7052 | |
| 7053 | return rc; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7054 | } |
| 7055 | |
| 7056 | static void gaudi_handle_qman_err_generic(struct hl_device *hdev, |
| 7057 | const char *qm_name, |
| 7058 | u64 glbl_sts_addr, |
| 7059 | u64 arb_err_addr) |
| 7060 | { |
| 7061 | u32 i, j, glbl_sts_val, arb_err_val, glbl_sts_clr_val; |
| 7062 | char reg_desc[32]; |
| 7063 | |
| 7064 | /* Iterate through all stream GLBL_STS1 registers + Lower CP */ |
| 7065 | for (i = 0 ; i < QMAN_STREAMS + 1 ; i++) { |
| 7066 | glbl_sts_clr_val = 0; |
| 7067 | glbl_sts_val = RREG32(glbl_sts_addr + 4 * i); |
| 7068 | |
| 7069 | if (!glbl_sts_val) |
| 7070 | continue; |
| 7071 | |
| 7072 | if (i == QMAN_STREAMS) |
| 7073 | snprintf(reg_desc, ARRAY_SIZE(reg_desc), "LowerCP"); |
| 7074 | else |
| 7075 | snprintf(reg_desc, ARRAY_SIZE(reg_desc), "stream%u", i); |
| 7076 | |
| 7077 | for (j = 0 ; j < GAUDI_NUM_OF_QM_ERR_CAUSE ; j++) { |
| 7078 | if (glbl_sts_val & BIT(j)) { |
| 7079 | dev_err_ratelimited(hdev->dev, |
| 7080 | "%s %s. err cause: %s\n", |
| 7081 | qm_name, reg_desc, |
| 7082 | gaudi_qman_error_cause[j]); |
| 7083 | glbl_sts_clr_val |= BIT(j); |
| 7084 | } |
| 7085 | } |
| 7086 | |
| 7087 | /* Write 1 clear errors */ |
Tomer Tayar | 1b49715 | 2021-04-06 13:32:20 +0300 | [diff] [blame] | 7088 | if (!hdev->stop_on_err) |
| 7089 | WREG32(glbl_sts_addr + 4 * i, glbl_sts_clr_val); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7090 | } |
| 7091 | |
| 7092 | arb_err_val = RREG32(arb_err_addr); |
| 7093 | |
| 7094 | if (!arb_err_val) |
| 7095 | return; |
| 7096 | |
| 7097 | for (j = 0 ; j < GAUDI_NUM_OF_QM_ARB_ERR_CAUSE ; j++) { |
| 7098 | if (arb_err_val & BIT(j)) { |
| 7099 | dev_err_ratelimited(hdev->dev, |
| 7100 | "%s ARB_ERR. err cause: %s\n", |
| 7101 | qm_name, |
| 7102 | gaudi_qman_arb_error_cause[j]); |
| 7103 | } |
| 7104 | } |
| 7105 | } |
| 7106 | |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7107 | static void gaudi_print_sm_sei_info(struct hl_device *hdev, u16 event_type, |
| 7108 | struct hl_eq_sm_sei_data *sei_data) |
| 7109 | { |
| 7110 | u32 index = event_type - GAUDI_EVENT_DMA_IF_SEI_0; |
| 7111 | |
| 7112 | switch (sei_data->sei_cause) { |
Oded Gabbay | 7838504 | 2021-01-26 22:56:56 +0200 | [diff] [blame] | 7113 | case SM_SEI_SO_OVERFLOW: |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7114 | dev_err(hdev->dev, |
| 7115 | "SM %u SEI Error: SO %u overflow/underflow", |
Oded Gabbay | 7838504 | 2021-01-26 22:56:56 +0200 | [diff] [blame] | 7116 | index, le32_to_cpu(sei_data->sei_log)); |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7117 | break; |
Oded Gabbay | 7838504 | 2021-01-26 22:56:56 +0200 | [diff] [blame] | 7118 | case SM_SEI_LBW_4B_UNALIGNED: |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7119 | dev_err(hdev->dev, |
| 7120 | "SM %u SEI Error: Unaligned 4B LBW access, monitor agent address low - %#x", |
Oded Gabbay | 7838504 | 2021-01-26 22:56:56 +0200 | [diff] [blame] | 7121 | index, le32_to_cpu(sei_data->sei_log)); |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7122 | break; |
Oded Gabbay | 7838504 | 2021-01-26 22:56:56 +0200 | [diff] [blame] | 7123 | case SM_SEI_AXI_RESPONSE_ERR: |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7124 | dev_err(hdev->dev, |
| 7125 | "SM %u SEI Error: AXI ID %u response error", |
Oded Gabbay | 7838504 | 2021-01-26 22:56:56 +0200 | [diff] [blame] | 7126 | index, le32_to_cpu(sei_data->sei_log)); |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7127 | break; |
| 7128 | default: |
| 7129 | dev_err(hdev->dev, "Unknown SM SEI cause %u", |
Oded Gabbay | 7838504 | 2021-01-26 22:56:56 +0200 | [diff] [blame] | 7130 | le32_to_cpu(sei_data->sei_log)); |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7131 | break; |
| 7132 | } |
| 7133 | } |
| 7134 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7135 | static void gaudi_handle_ecc_event(struct hl_device *hdev, u16 event_type, |
| 7136 | struct hl_eq_ecc_data *ecc_data) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7137 | { |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7138 | struct ecc_info_extract_params params; |
| 7139 | u64 ecc_address = 0, ecc_syndrom = 0; |
| 7140 | u8 index, memory_wrapper_idx = 0; |
| 7141 | bool extract_info_from_fw; |
| 7142 | int rc; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7143 | |
| 7144 | switch (event_type) { |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7145 | case GAUDI_EVENT_PCIE_CORE_SERR ... GAUDI_EVENT_PCIE_PHY_DERR: |
| 7146 | case GAUDI_EVENT_DMA0_SERR_ECC ... GAUDI_EVENT_MMU_DERR: |
| 7147 | extract_info_from_fw = true; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7148 | break; |
| 7149 | case GAUDI_EVENT_TPC0_SERR ... GAUDI_EVENT_TPC7_SERR: |
| 7150 | index = event_type - GAUDI_EVENT_TPC0_SERR; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7151 | params.block_address = mmTPC0_CFG_BASE + index * TPC_CFG_OFFSET; |
| 7152 | params.num_memories = 90; |
| 7153 | params.derr = false; |
| 7154 | params.disable_clock_gating = true; |
| 7155 | extract_info_from_fw = false; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7156 | break; |
| 7157 | case GAUDI_EVENT_TPC0_DERR ... GAUDI_EVENT_TPC7_DERR: |
| 7158 | index = event_type - GAUDI_EVENT_TPC0_DERR; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7159 | params.block_address = |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7160 | mmTPC0_CFG_BASE + index * TPC_CFG_OFFSET; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7161 | params.num_memories = 90; |
| 7162 | params.derr = true; |
| 7163 | params.disable_clock_gating = true; |
| 7164 | extract_info_from_fw = false; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7165 | break; |
| 7166 | case GAUDI_EVENT_MME0_ACC_SERR: |
| 7167 | case GAUDI_EVENT_MME1_ACC_SERR: |
| 7168 | case GAUDI_EVENT_MME2_ACC_SERR: |
| 7169 | case GAUDI_EVENT_MME3_ACC_SERR: |
| 7170 | index = (event_type - GAUDI_EVENT_MME0_ACC_SERR) / 4; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7171 | params.block_address = mmMME0_ACC_BASE + index * MME_ACC_OFFSET; |
| 7172 | params.num_memories = 128; |
| 7173 | params.derr = false; |
| 7174 | params.disable_clock_gating = true; |
| 7175 | extract_info_from_fw = false; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7176 | break; |
| 7177 | case GAUDI_EVENT_MME0_ACC_DERR: |
| 7178 | case GAUDI_EVENT_MME1_ACC_DERR: |
| 7179 | case GAUDI_EVENT_MME2_ACC_DERR: |
| 7180 | case GAUDI_EVENT_MME3_ACC_DERR: |
| 7181 | index = (event_type - GAUDI_EVENT_MME0_ACC_DERR) / 4; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7182 | params.block_address = mmMME0_ACC_BASE + index * MME_ACC_OFFSET; |
| 7183 | params.num_memories = 128; |
| 7184 | params.derr = true; |
| 7185 | params.disable_clock_gating = true; |
| 7186 | extract_info_from_fw = false; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7187 | break; |
| 7188 | case GAUDI_EVENT_MME0_SBAB_SERR: |
| 7189 | case GAUDI_EVENT_MME1_SBAB_SERR: |
| 7190 | case GAUDI_EVENT_MME2_SBAB_SERR: |
| 7191 | case GAUDI_EVENT_MME3_SBAB_SERR: |
| 7192 | index = (event_type - GAUDI_EVENT_MME0_SBAB_SERR) / 4; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7193 | params.block_address = |
| 7194 | mmMME0_SBAB_BASE + index * MME_ACC_OFFSET; |
| 7195 | params.num_memories = 33; |
| 7196 | params.derr = false; |
| 7197 | params.disable_clock_gating = true; |
| 7198 | extract_info_from_fw = false; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7199 | break; |
| 7200 | case GAUDI_EVENT_MME0_SBAB_DERR: |
| 7201 | case GAUDI_EVENT_MME1_SBAB_DERR: |
| 7202 | case GAUDI_EVENT_MME2_SBAB_DERR: |
| 7203 | case GAUDI_EVENT_MME3_SBAB_DERR: |
| 7204 | index = (event_type - GAUDI_EVENT_MME0_SBAB_DERR) / 4; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7205 | params.block_address = |
| 7206 | mmMME0_SBAB_BASE + index * MME_ACC_OFFSET; |
| 7207 | params.num_memories = 33; |
| 7208 | params.derr = true; |
| 7209 | params.disable_clock_gating = true; |
Oded Gabbay | 652b444 | 2020-11-21 14:35:35 +0200 | [diff] [blame] | 7210 | extract_info_from_fw = false; |
| 7211 | break; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7212 | default: |
| 7213 | return; |
| 7214 | } |
| 7215 | |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7216 | if (extract_info_from_fw) { |
| 7217 | ecc_address = le64_to_cpu(ecc_data->ecc_address); |
| 7218 | ecc_syndrom = le64_to_cpu(ecc_data->ecc_syndrom); |
| 7219 | memory_wrapper_idx = ecc_data->memory_wrapper_idx; |
| 7220 | } else { |
| 7221 | rc = gaudi_extract_ecc_info(hdev, ¶ms, &ecc_address, |
| 7222 | &ecc_syndrom, &memory_wrapper_idx); |
| 7223 | if (rc) |
| 7224 | return; |
| 7225 | } |
| 7226 | |
| 7227 | dev_err(hdev->dev, |
| 7228 | "ECC error detected. address: %#llx. Syndrom: %#llx. block id %u\n", |
| 7229 | ecc_address, ecc_syndrom, memory_wrapper_idx); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7230 | } |
| 7231 | |
| 7232 | static void gaudi_handle_qman_err(struct hl_device *hdev, u16 event_type) |
| 7233 | { |
| 7234 | u64 glbl_sts_addr, arb_err_addr; |
| 7235 | u8 index; |
| 7236 | char desc[32]; |
| 7237 | |
| 7238 | switch (event_type) { |
| 7239 | case GAUDI_EVENT_TPC0_QM ... GAUDI_EVENT_TPC7_QM: |
| 7240 | index = event_type - GAUDI_EVENT_TPC0_QM; |
| 7241 | glbl_sts_addr = |
| 7242 | mmTPC0_QM_GLBL_STS1_0 + index * TPC_QMAN_OFFSET; |
| 7243 | arb_err_addr = |
| 7244 | mmTPC0_QM_ARB_ERR_CAUSE + index * TPC_QMAN_OFFSET; |
| 7245 | snprintf(desc, ARRAY_SIZE(desc), "%s%d", "TPC_QM", index); |
| 7246 | break; |
| 7247 | case GAUDI_EVENT_MME0_QM ... GAUDI_EVENT_MME2_QM: |
| 7248 | index = event_type - GAUDI_EVENT_MME0_QM; |
| 7249 | glbl_sts_addr = |
| 7250 | mmMME0_QM_GLBL_STS1_0 + index * MME_QMAN_OFFSET; |
| 7251 | arb_err_addr = |
| 7252 | mmMME0_QM_ARB_ERR_CAUSE + index * MME_QMAN_OFFSET; |
| 7253 | snprintf(desc, ARRAY_SIZE(desc), "%s%d", "MME_QM", index); |
| 7254 | break; |
| 7255 | case GAUDI_EVENT_DMA0_QM ... GAUDI_EVENT_DMA7_QM: |
| 7256 | index = event_type - GAUDI_EVENT_DMA0_QM; |
| 7257 | glbl_sts_addr = |
| 7258 | mmDMA0_QM_GLBL_STS1_0 + index * DMA_QMAN_OFFSET; |
| 7259 | arb_err_addr = |
| 7260 | mmDMA0_QM_ARB_ERR_CAUSE + index * DMA_QMAN_OFFSET; |
| 7261 | snprintf(desc, ARRAY_SIZE(desc), "%s%d", "DMA_QM", index); |
| 7262 | break; |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 7263 | case GAUDI_EVENT_NIC0_QM0: |
| 7264 | glbl_sts_addr = mmNIC0_QM0_GLBL_STS1_0; |
| 7265 | arb_err_addr = mmNIC0_QM0_ARB_ERR_CAUSE; |
| 7266 | snprintf(desc, ARRAY_SIZE(desc), "NIC0_QM0"); |
| 7267 | break; |
| 7268 | case GAUDI_EVENT_NIC0_QM1: |
| 7269 | glbl_sts_addr = mmNIC0_QM1_GLBL_STS1_0; |
| 7270 | arb_err_addr = mmNIC0_QM1_ARB_ERR_CAUSE; |
| 7271 | snprintf(desc, ARRAY_SIZE(desc), "NIC0_QM1"); |
| 7272 | break; |
| 7273 | case GAUDI_EVENT_NIC1_QM0: |
| 7274 | glbl_sts_addr = mmNIC1_QM0_GLBL_STS1_0; |
| 7275 | arb_err_addr = mmNIC1_QM0_ARB_ERR_CAUSE; |
| 7276 | snprintf(desc, ARRAY_SIZE(desc), "NIC1_QM0"); |
| 7277 | break; |
| 7278 | case GAUDI_EVENT_NIC1_QM1: |
| 7279 | glbl_sts_addr = mmNIC1_QM1_GLBL_STS1_0; |
| 7280 | arb_err_addr = mmNIC1_QM1_ARB_ERR_CAUSE; |
| 7281 | snprintf(desc, ARRAY_SIZE(desc), "NIC1_QM1"); |
| 7282 | break; |
| 7283 | case GAUDI_EVENT_NIC2_QM0: |
| 7284 | glbl_sts_addr = mmNIC2_QM0_GLBL_STS1_0; |
| 7285 | arb_err_addr = mmNIC2_QM0_ARB_ERR_CAUSE; |
| 7286 | snprintf(desc, ARRAY_SIZE(desc), "NIC2_QM0"); |
| 7287 | break; |
| 7288 | case GAUDI_EVENT_NIC2_QM1: |
| 7289 | glbl_sts_addr = mmNIC2_QM1_GLBL_STS1_0; |
| 7290 | arb_err_addr = mmNIC2_QM1_ARB_ERR_CAUSE; |
| 7291 | snprintf(desc, ARRAY_SIZE(desc), "NIC2_QM1"); |
| 7292 | break; |
| 7293 | case GAUDI_EVENT_NIC3_QM0: |
| 7294 | glbl_sts_addr = mmNIC3_QM0_GLBL_STS1_0; |
| 7295 | arb_err_addr = mmNIC3_QM0_ARB_ERR_CAUSE; |
| 7296 | snprintf(desc, ARRAY_SIZE(desc), "NIC3_QM0"); |
| 7297 | break; |
| 7298 | case GAUDI_EVENT_NIC3_QM1: |
| 7299 | glbl_sts_addr = mmNIC3_QM1_GLBL_STS1_0; |
| 7300 | arb_err_addr = mmNIC3_QM1_ARB_ERR_CAUSE; |
| 7301 | snprintf(desc, ARRAY_SIZE(desc), "NIC3_QM1"); |
| 7302 | break; |
| 7303 | case GAUDI_EVENT_NIC4_QM0: |
| 7304 | glbl_sts_addr = mmNIC4_QM0_GLBL_STS1_0; |
| 7305 | arb_err_addr = mmNIC4_QM0_ARB_ERR_CAUSE; |
| 7306 | snprintf(desc, ARRAY_SIZE(desc), "NIC4_QM0"); |
| 7307 | break; |
| 7308 | case GAUDI_EVENT_NIC4_QM1: |
| 7309 | glbl_sts_addr = mmNIC4_QM1_GLBL_STS1_0; |
| 7310 | arb_err_addr = mmNIC4_QM1_ARB_ERR_CAUSE; |
| 7311 | snprintf(desc, ARRAY_SIZE(desc), "NIC4_QM1"); |
| 7312 | break; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7313 | default: |
| 7314 | return; |
| 7315 | } |
| 7316 | |
| 7317 | gaudi_handle_qman_err_generic(hdev, desc, glbl_sts_addr, arb_err_addr); |
| 7318 | } |
| 7319 | |
| 7320 | static void gaudi_print_irq_info(struct hl_device *hdev, u16 event_type, |
| 7321 | bool razwi) |
| 7322 | { |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 7323 | char desc[64] = ""; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7324 | |
| 7325 | gaudi_get_event_desc(event_type, desc, sizeof(desc)); |
| 7326 | dev_err_ratelimited(hdev->dev, "Received H/W interrupt %d [\"%s\"]\n", |
| 7327 | event_type, desc); |
| 7328 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7329 | if (razwi) { |
| 7330 | gaudi_print_razwi_info(hdev); |
| 7331 | gaudi_print_mmu_error_info(hdev); |
| 7332 | } |
| 7333 | } |
| 7334 | |
Ohad Sharabi | 5d6a198 | 2021-02-08 14:53:56 +0200 | [diff] [blame] | 7335 | static void gaudi_print_out_of_sync_info(struct hl_device *hdev, |
| 7336 | struct cpucp_pkt_sync_err *sync_err) |
| 7337 | { |
| 7338 | struct hl_hw_queue *q = &hdev->kernel_queues[GAUDI_QUEUE_ID_CPU_PQ]; |
| 7339 | |
| 7340 | dev_err(hdev->dev, "Out of sync with FW, FW: pi=%u, ci=%u, LKD: pi=%u, ci=%u\n", |
| 7341 | sync_err->pi, sync_err->ci, q->pi, atomic_read(&q->ci)); |
| 7342 | } |
| 7343 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7344 | static int gaudi_soft_reset_late_init(struct hl_device *hdev) |
| 7345 | { |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 7346 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7347 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7348 | /* Unmask all IRQs since some could have been received |
| 7349 | * during the soft reset |
| 7350 | */ |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 7351 | return hl_fw_unmask_irq_arr(hdev, gaudi->events, sizeof(gaudi->events)); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7352 | } |
| 7353 | |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7354 | static int gaudi_hbm_read_interrupts(struct hl_device *hdev, int device, |
| 7355 | struct hl_eq_hbm_ecc_data *hbm_ecc_data) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7356 | { |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7357 | u32 base, val, val2, wr_par, rd_par, ca_par, derr, serr, type, ch; |
| 7358 | int err = 0; |
| 7359 | |
Ohad Sharabi | b520ca5 | 2021-01-27 15:42:53 +0200 | [diff] [blame] | 7360 | if (hdev->asic_prop.fw_security_status_valid && |
| 7361 | (hdev->asic_prop.fw_app_security_map & |
| 7362 | CPU_BOOT_DEV_STS0_HBM_ECC_EN)) { |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7363 | if (!hbm_ecc_data) { |
| 7364 | dev_err(hdev->dev, "No FW ECC data"); |
| 7365 | return 0; |
| 7366 | } |
| 7367 | |
| 7368 | wr_par = FIELD_GET(CPUCP_PKT_HBM_ECC_INFO_WR_PAR_MASK, |
| 7369 | le32_to_cpu(hbm_ecc_data->hbm_ecc_info)); |
| 7370 | rd_par = FIELD_GET(CPUCP_PKT_HBM_ECC_INFO_RD_PAR_MASK, |
| 7371 | le32_to_cpu(hbm_ecc_data->hbm_ecc_info)); |
| 7372 | ca_par = FIELD_GET(CPUCP_PKT_HBM_ECC_INFO_CA_PAR_MASK, |
| 7373 | le32_to_cpu(hbm_ecc_data->hbm_ecc_info)); |
| 7374 | derr = FIELD_GET(CPUCP_PKT_HBM_ECC_INFO_DERR_MASK, |
| 7375 | le32_to_cpu(hbm_ecc_data->hbm_ecc_info)); |
| 7376 | serr = FIELD_GET(CPUCP_PKT_HBM_ECC_INFO_SERR_MASK, |
| 7377 | le32_to_cpu(hbm_ecc_data->hbm_ecc_info)); |
| 7378 | type = FIELD_GET(CPUCP_PKT_HBM_ECC_INFO_TYPE_MASK, |
| 7379 | le32_to_cpu(hbm_ecc_data->hbm_ecc_info)); |
| 7380 | ch = FIELD_GET(CPUCP_PKT_HBM_ECC_INFO_HBM_CH_MASK, |
| 7381 | le32_to_cpu(hbm_ecc_data->hbm_ecc_info)); |
| 7382 | |
| 7383 | dev_err(hdev->dev, |
Ohad Sharabi | b520ca5 | 2021-01-27 15:42:53 +0200 | [diff] [blame] | 7384 | "HBM%d pc%d interrupts info: WR_PAR=%d, RD_PAR=%d, CA_PAR=%d, SERR=%d, DERR=%d\n", |
| 7385 | device, ch, wr_par, rd_par, ca_par, serr, derr); |
| 7386 | dev_err(hdev->dev, |
| 7387 | "HBM%d pc%d ECC info: 1ST_ERR_ADDR=0x%x, 1ST_ERR_TYPE=%d, SEC_CONT_CNT=%u, SEC_CNT=%d, DEC_CNT=%d\n", |
| 7388 | device, ch, hbm_ecc_data->first_addr, type, |
| 7389 | hbm_ecc_data->sec_cont_cnt, hbm_ecc_data->sec_cnt, |
| 7390 | hbm_ecc_data->dec_cnt); |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7391 | return 0; |
| 7392 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7393 | |
Ohad Sharabi | b520ca5 | 2021-01-27 15:42:53 +0200 | [diff] [blame] | 7394 | if (!hdev->asic_prop.fw_security_disabled) { |
| 7395 | dev_info(hdev->dev, "Cannot access MC regs for ECC data while security is enabled\n"); |
| 7396 | return 0; |
| 7397 | } |
| 7398 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7399 | base = GAUDI_HBM_CFG_BASE + device * GAUDI_HBM_CFG_OFFSET; |
| 7400 | for (ch = 0 ; ch < GAUDI_HBM_CHANNELS ; ch++) { |
| 7401 | val = RREG32_MASK(base + ch * 0x1000 + 0x06C, 0x0000FFFF); |
| 7402 | val = (val & 0xFF) | ((val >> 8) & 0xFF); |
| 7403 | if (val) { |
| 7404 | err = 1; |
| 7405 | dev_err(hdev->dev, |
| 7406 | "HBM%d pc%d interrupts info: WR_PAR=%d, RD_PAR=%d, CA_PAR=%d, SERR=%d, DERR=%d\n", |
| 7407 | device, ch * 2, val & 0x1, (val >> 1) & 0x1, |
| 7408 | (val >> 2) & 0x1, (val >> 3) & 0x1, |
| 7409 | (val >> 4) & 0x1); |
| 7410 | |
| 7411 | val2 = RREG32(base + ch * 0x1000 + 0x060); |
| 7412 | dev_err(hdev->dev, |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7413 | "HBM%d pc%d ECC info: 1ST_ERR_ADDR=0x%x, 1ST_ERR_TYPE=%d, SEC_CONT_CNT=%d, SEC_CNT=%d, DEC_CNT=%d\n", |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7414 | device, ch * 2, |
| 7415 | RREG32(base + ch * 0x1000 + 0x064), |
| 7416 | (val2 & 0x200) >> 9, (val2 & 0xFC00) >> 10, |
| 7417 | (val2 & 0xFF0000) >> 16, |
| 7418 | (val2 & 0xFF000000) >> 24); |
| 7419 | } |
| 7420 | |
| 7421 | val = RREG32_MASK(base + ch * 0x1000 + 0x07C, 0x0000FFFF); |
| 7422 | val = (val & 0xFF) | ((val >> 8) & 0xFF); |
| 7423 | if (val) { |
| 7424 | err = 1; |
| 7425 | dev_err(hdev->dev, |
| 7426 | "HBM%d pc%d interrupts info: WR_PAR=%d, RD_PAR=%d, CA_PAR=%d, SERR=%d, DERR=%d\n", |
| 7427 | device, ch * 2 + 1, val & 0x1, (val >> 1) & 0x1, |
| 7428 | (val >> 2) & 0x1, (val >> 3) & 0x1, |
| 7429 | (val >> 4) & 0x1); |
| 7430 | |
| 7431 | val2 = RREG32(base + ch * 0x1000 + 0x070); |
| 7432 | dev_err(hdev->dev, |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7433 | "HBM%d pc%d ECC info: 1ST_ERR_ADDR=0x%x, 1ST_ERR_TYPE=%d, SEC_CONT_CNT=%d, SEC_CNT=%d, DEC_CNT=%d\n", |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7434 | device, ch * 2 + 1, |
| 7435 | RREG32(base + ch * 0x1000 + 0x074), |
| 7436 | (val2 & 0x200) >> 9, (val2 & 0xFC00) >> 10, |
| 7437 | (val2 & 0xFF0000) >> 16, |
| 7438 | (val2 & 0xFF000000) >> 24); |
| 7439 | } |
| 7440 | |
| 7441 | /* Clear interrupts */ |
| 7442 | RMWREG32(base + (ch * 0x1000) + 0x060, 0x1C8, 0x1FF); |
| 7443 | RMWREG32(base + (ch * 0x1000) + 0x070, 0x1C8, 0x1FF); |
| 7444 | WREG32(base + (ch * 0x1000) + 0x06C, 0x1F1F); |
| 7445 | WREG32(base + (ch * 0x1000) + 0x07C, 0x1F1F); |
| 7446 | RMWREG32(base + (ch * 0x1000) + 0x060, 0x0, 0xF); |
| 7447 | RMWREG32(base + (ch * 0x1000) + 0x070, 0x0, 0xF); |
| 7448 | } |
| 7449 | |
| 7450 | val = RREG32(base + 0x8F30); |
| 7451 | val2 = RREG32(base + 0x8F34); |
| 7452 | if (val | val2) { |
| 7453 | err = 1; |
| 7454 | dev_err(hdev->dev, |
| 7455 | "HBM %d MC SRAM SERR info: Reg 0x8F30=0x%x, Reg 0x8F34=0x%x\n", |
| 7456 | device, val, val2); |
| 7457 | } |
| 7458 | val = RREG32(base + 0x8F40); |
| 7459 | val2 = RREG32(base + 0x8F44); |
| 7460 | if (val | val2) { |
| 7461 | err = 1; |
| 7462 | dev_err(hdev->dev, |
| 7463 | "HBM %d MC SRAM DERR info: Reg 0x8F40=0x%x, Reg 0x8F44=0x%x\n", |
| 7464 | device, val, val2); |
| 7465 | } |
| 7466 | |
| 7467 | return err; |
| 7468 | } |
| 7469 | |
| 7470 | static int gaudi_hbm_event_to_dev(u16 hbm_event_type) |
| 7471 | { |
| 7472 | switch (hbm_event_type) { |
| 7473 | case GAUDI_EVENT_HBM0_SPI_0: |
| 7474 | case GAUDI_EVENT_HBM0_SPI_1: |
| 7475 | return 0; |
| 7476 | case GAUDI_EVENT_HBM1_SPI_0: |
| 7477 | case GAUDI_EVENT_HBM1_SPI_1: |
| 7478 | return 1; |
| 7479 | case GAUDI_EVENT_HBM2_SPI_0: |
| 7480 | case GAUDI_EVENT_HBM2_SPI_1: |
| 7481 | return 2; |
| 7482 | case GAUDI_EVENT_HBM3_SPI_0: |
| 7483 | case GAUDI_EVENT_HBM3_SPI_1: |
| 7484 | return 3; |
| 7485 | default: |
| 7486 | break; |
| 7487 | } |
| 7488 | |
| 7489 | /* Should never happen */ |
| 7490 | return 0; |
| 7491 | } |
| 7492 | |
| 7493 | static bool gaudi_tpc_read_interrupts(struct hl_device *hdev, u8 tpc_id, |
| 7494 | char *interrupt_name) |
| 7495 | { |
| 7496 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7497 | u32 tpc_offset = tpc_id * TPC_CFG_OFFSET, tpc_interrupts_cause, i; |
| 7498 | bool soft_reset_required = false; |
| 7499 | |
| 7500 | /* Accessing the TPC_INTR_CAUSE registers requires disabling the clock |
Oded Gabbay | 6138bbe | 2020-09-04 20:18:16 +0300 | [diff] [blame] | 7501 | * gating, and thus cannot be done in CPU-CP and should be done instead |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7502 | * by the driver. |
| 7503 | */ |
| 7504 | |
| 7505 | mutex_lock(&gaudi->clk_gate_mutex); |
| 7506 | |
| 7507 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 7508 | |
| 7509 | tpc_interrupts_cause = RREG32(mmTPC0_CFG_TPC_INTR_CAUSE + tpc_offset) & |
| 7510 | TPC0_CFG_TPC_INTR_CAUSE_CAUSE_MASK; |
| 7511 | |
| 7512 | for (i = 0 ; i < GAUDI_NUM_OF_TPC_INTR_CAUSE ; i++) |
| 7513 | if (tpc_interrupts_cause & BIT(i)) { |
| 7514 | dev_err_ratelimited(hdev->dev, |
| 7515 | "TPC%d_%s interrupt cause: %s\n", |
| 7516 | tpc_id, interrupt_name, |
| 7517 | gaudi_tpc_interrupts_cause[i]); |
| 7518 | /* If this is QM error, we need to soft-reset */ |
| 7519 | if (i == 15) |
| 7520 | soft_reset_required = true; |
| 7521 | } |
| 7522 | |
| 7523 | /* Clear interrupts */ |
| 7524 | WREG32(mmTPC0_CFG_TPC_INTR_CAUSE + tpc_offset, 0); |
| 7525 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 7526 | hdev->asic_funcs->set_clock_gating(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7527 | |
| 7528 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 7529 | |
| 7530 | return soft_reset_required; |
| 7531 | } |
| 7532 | |
| 7533 | static int tpc_dec_event_to_tpc_id(u16 tpc_dec_event_type) |
| 7534 | { |
| 7535 | return (tpc_dec_event_type - GAUDI_EVENT_TPC0_DEC) >> 1; |
| 7536 | } |
| 7537 | |
| 7538 | static int tpc_krn_event_to_tpc_id(u16 tpc_dec_event_type) |
| 7539 | { |
| 7540 | return (tpc_dec_event_type - GAUDI_EVENT_TPC0_KRN_ERR) / 6; |
| 7541 | } |
| 7542 | |
| 7543 | static void gaudi_print_clk_change_info(struct hl_device *hdev, |
| 7544 | u16 event_type) |
| 7545 | { |
| 7546 | switch (event_type) { |
| 7547 | case GAUDI_EVENT_FIX_POWER_ENV_S: |
Ofir Bitton | 0a068ad | 2020-07-21 10:49:51 +0300 | [diff] [blame] | 7548 | hdev->clk_throttling_reason |= HL_CLK_THROTTLE_POWER; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7549 | dev_info_ratelimited(hdev->dev, |
| 7550 | "Clock throttling due to power consumption\n"); |
| 7551 | break; |
| 7552 | |
| 7553 | case GAUDI_EVENT_FIX_POWER_ENV_E: |
Ofir Bitton | 0a068ad | 2020-07-21 10:49:51 +0300 | [diff] [blame] | 7554 | hdev->clk_throttling_reason &= ~HL_CLK_THROTTLE_POWER; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7555 | dev_info_ratelimited(hdev->dev, |
| 7556 | "Power envelop is safe, back to optimal clock\n"); |
| 7557 | break; |
| 7558 | |
| 7559 | case GAUDI_EVENT_FIX_THERMAL_ENV_S: |
Ofir Bitton | 0a068ad | 2020-07-21 10:49:51 +0300 | [diff] [blame] | 7560 | hdev->clk_throttling_reason |= HL_CLK_THROTTLE_THERMAL; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7561 | dev_info_ratelimited(hdev->dev, |
| 7562 | "Clock throttling due to overheating\n"); |
| 7563 | break; |
| 7564 | |
| 7565 | case GAUDI_EVENT_FIX_THERMAL_ENV_E: |
Ofir Bitton | 0a068ad | 2020-07-21 10:49:51 +0300 | [diff] [blame] | 7566 | hdev->clk_throttling_reason &= ~HL_CLK_THROTTLE_THERMAL; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7567 | dev_info_ratelimited(hdev->dev, |
| 7568 | "Thermal envelop is safe, back to optimal clock\n"); |
| 7569 | break; |
| 7570 | |
| 7571 | default: |
| 7572 | dev_err(hdev->dev, "Received invalid clock change event %d\n", |
| 7573 | event_type); |
| 7574 | break; |
| 7575 | } |
| 7576 | } |
| 7577 | |
| 7578 | static void gaudi_handle_eqe(struct hl_device *hdev, |
| 7579 | struct hl_eq_entry *eq_entry) |
| 7580 | { |
| 7581 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7582 | u32 ctl = le32_to_cpu(eq_entry->hdr.ctl); |
| 7583 | u16 event_type = ((ctl & EQ_CTL_EVENT_TYPE_MASK) |
| 7584 | >> EQ_CTL_EVENT_TYPE_SHIFT); |
| 7585 | u8 cause; |
Oded Gabbay | 6644682 | 2020-05-18 16:48:01 +0300 | [diff] [blame] | 7586 | bool reset_required; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7587 | |
| 7588 | gaudi->events_stat[event_type]++; |
| 7589 | gaudi->events_stat_aggregate[event_type]++; |
| 7590 | |
| 7591 | switch (event_type) { |
| 7592 | case GAUDI_EVENT_PCIE_CORE_DERR: |
| 7593 | case GAUDI_EVENT_PCIE_IF_DERR: |
| 7594 | case GAUDI_EVENT_PCIE_PHY_DERR: |
| 7595 | case GAUDI_EVENT_TPC0_DERR ... GAUDI_EVENT_TPC7_DERR: |
| 7596 | case GAUDI_EVENT_MME0_ACC_DERR: |
| 7597 | case GAUDI_EVENT_MME0_SBAB_DERR: |
| 7598 | case GAUDI_EVENT_MME1_ACC_DERR: |
| 7599 | case GAUDI_EVENT_MME1_SBAB_DERR: |
| 7600 | case GAUDI_EVENT_MME2_ACC_DERR: |
| 7601 | case GAUDI_EVENT_MME2_SBAB_DERR: |
| 7602 | case GAUDI_EVENT_MME3_ACC_DERR: |
| 7603 | case GAUDI_EVENT_MME3_SBAB_DERR: |
| 7604 | case GAUDI_EVENT_DMA0_DERR_ECC ... GAUDI_EVENT_DMA7_DERR_ECC: |
| 7605 | fallthrough; |
| 7606 | case GAUDI_EVENT_CPU_IF_ECC_DERR: |
| 7607 | case GAUDI_EVENT_PSOC_MEM_DERR: |
| 7608 | case GAUDI_EVENT_PSOC_CORESIGHT_DERR: |
| 7609 | case GAUDI_EVENT_SRAM0_DERR ... GAUDI_EVENT_SRAM28_DERR: |
| 7610 | case GAUDI_EVENT_DMA_IF0_DERR ... GAUDI_EVENT_DMA_IF3_DERR: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7611 | case GAUDI_EVENT_HBM_0_DERR ... GAUDI_EVENT_HBM_3_DERR: |
| 7612 | case GAUDI_EVENT_MMU_DERR: |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7613 | gaudi_print_irq_info(hdev, event_type, true); |
| 7614 | gaudi_handle_ecc_event(hdev, event_type, &eq_entry->ecc_data); |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7615 | goto reset_device; |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7616 | |
| 7617 | case GAUDI_EVENT_GIC500: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7618 | case GAUDI_EVENT_AXI_ECC: |
| 7619 | case GAUDI_EVENT_L2_RAM_ECC: |
| 7620 | case GAUDI_EVENT_PLL0 ... GAUDI_EVENT_PLL17: |
| 7621 | gaudi_print_irq_info(hdev, event_type, false); |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7622 | goto reset_device; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7623 | |
| 7624 | case GAUDI_EVENT_HBM0_SPI_0: |
| 7625 | case GAUDI_EVENT_HBM1_SPI_0: |
| 7626 | case GAUDI_EVENT_HBM2_SPI_0: |
| 7627 | case GAUDI_EVENT_HBM3_SPI_0: |
| 7628 | gaudi_print_irq_info(hdev, event_type, false); |
| 7629 | gaudi_hbm_read_interrupts(hdev, |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7630 | gaudi_hbm_event_to_dev(event_type), |
| 7631 | &eq_entry->hbm_ecc_data); |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7632 | goto reset_device; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7633 | |
| 7634 | case GAUDI_EVENT_HBM0_SPI_1: |
| 7635 | case GAUDI_EVENT_HBM1_SPI_1: |
| 7636 | case GAUDI_EVENT_HBM2_SPI_1: |
| 7637 | case GAUDI_EVENT_HBM3_SPI_1: |
| 7638 | gaudi_print_irq_info(hdev, event_type, false); |
| 7639 | gaudi_hbm_read_interrupts(hdev, |
Ofir Bitton | 5a2998f | 2020-10-05 13:44:59 +0300 | [diff] [blame] | 7640 | gaudi_hbm_event_to_dev(event_type), |
| 7641 | &eq_entry->hbm_ecc_data); |
Oded Gabbay | 230cd89 | 2021-01-26 22:58:13 +0200 | [diff] [blame] | 7642 | hl_fw_unmask_irq(hdev, event_type); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7643 | break; |
| 7644 | |
| 7645 | case GAUDI_EVENT_TPC0_DEC: |
| 7646 | case GAUDI_EVENT_TPC1_DEC: |
| 7647 | case GAUDI_EVENT_TPC2_DEC: |
| 7648 | case GAUDI_EVENT_TPC3_DEC: |
| 7649 | case GAUDI_EVENT_TPC4_DEC: |
| 7650 | case GAUDI_EVENT_TPC5_DEC: |
| 7651 | case GAUDI_EVENT_TPC6_DEC: |
| 7652 | case GAUDI_EVENT_TPC7_DEC: |
| 7653 | gaudi_print_irq_info(hdev, event_type, true); |
Oded Gabbay | 6644682 | 2020-05-18 16:48:01 +0300 | [diff] [blame] | 7654 | reset_required = gaudi_tpc_read_interrupts(hdev, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7655 | tpc_dec_event_to_tpc_id(event_type), |
| 7656 | "AXI_SLV_DEC_Error"); |
Oded Gabbay | 6644682 | 2020-05-18 16:48:01 +0300 | [diff] [blame] | 7657 | if (reset_required) { |
| 7658 | dev_err(hdev->dev, "hard reset required due to %s\n", |
| 7659 | gaudi_irq_map_table[event_type].name); |
| 7660 | |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7661 | goto reset_device; |
Oded Gabbay | 6644682 | 2020-05-18 16:48:01 +0300 | [diff] [blame] | 7662 | } else { |
| 7663 | hl_fw_unmask_irq(hdev, event_type); |
Omer Shpigelman | d798507 | 2020-05-17 23:01:22 +0300 | [diff] [blame] | 7664 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7665 | break; |
| 7666 | |
| 7667 | case GAUDI_EVENT_TPC0_KRN_ERR: |
| 7668 | case GAUDI_EVENT_TPC1_KRN_ERR: |
| 7669 | case GAUDI_EVENT_TPC2_KRN_ERR: |
| 7670 | case GAUDI_EVENT_TPC3_KRN_ERR: |
| 7671 | case GAUDI_EVENT_TPC4_KRN_ERR: |
| 7672 | case GAUDI_EVENT_TPC5_KRN_ERR: |
| 7673 | case GAUDI_EVENT_TPC6_KRN_ERR: |
| 7674 | case GAUDI_EVENT_TPC7_KRN_ERR: |
| 7675 | gaudi_print_irq_info(hdev, event_type, true); |
Oded Gabbay | 6644682 | 2020-05-18 16:48:01 +0300 | [diff] [blame] | 7676 | reset_required = gaudi_tpc_read_interrupts(hdev, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7677 | tpc_krn_event_to_tpc_id(event_type), |
| 7678 | "KRN_ERR"); |
Oded Gabbay | 6644682 | 2020-05-18 16:48:01 +0300 | [diff] [blame] | 7679 | if (reset_required) { |
| 7680 | dev_err(hdev->dev, "hard reset required due to %s\n", |
| 7681 | gaudi_irq_map_table[event_type].name); |
| 7682 | |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7683 | goto reset_device; |
Oded Gabbay | 6644682 | 2020-05-18 16:48:01 +0300 | [diff] [blame] | 7684 | } else { |
| 7685 | hl_fw_unmask_irq(hdev, event_type); |
Omer Shpigelman | d798507 | 2020-05-17 23:01:22 +0300 | [diff] [blame] | 7686 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7687 | break; |
| 7688 | |
| 7689 | case GAUDI_EVENT_PCIE_CORE_SERR: |
| 7690 | case GAUDI_EVENT_PCIE_IF_SERR: |
| 7691 | case GAUDI_EVENT_PCIE_PHY_SERR: |
| 7692 | case GAUDI_EVENT_TPC0_SERR ... GAUDI_EVENT_TPC7_SERR: |
| 7693 | case GAUDI_EVENT_MME0_ACC_SERR: |
| 7694 | case GAUDI_EVENT_MME0_SBAB_SERR: |
| 7695 | case GAUDI_EVENT_MME1_ACC_SERR: |
| 7696 | case GAUDI_EVENT_MME1_SBAB_SERR: |
| 7697 | case GAUDI_EVENT_MME2_ACC_SERR: |
| 7698 | case GAUDI_EVENT_MME2_SBAB_SERR: |
| 7699 | case GAUDI_EVENT_MME3_ACC_SERR: |
| 7700 | case GAUDI_EVENT_MME3_SBAB_SERR: |
| 7701 | case GAUDI_EVENT_DMA0_SERR_ECC ... GAUDI_EVENT_DMA7_SERR_ECC: |
| 7702 | case GAUDI_EVENT_CPU_IF_ECC_SERR: |
| 7703 | case GAUDI_EVENT_PSOC_MEM_SERR: |
| 7704 | case GAUDI_EVENT_PSOC_CORESIGHT_SERR: |
| 7705 | case GAUDI_EVENT_SRAM0_SERR ... GAUDI_EVENT_SRAM28_SERR: |
| 7706 | case GAUDI_EVENT_DMA_IF0_SERR ... GAUDI_EVENT_DMA_IF3_SERR: |
| 7707 | case GAUDI_EVENT_HBM_0_SERR ... GAUDI_EVENT_HBM_3_SERR: |
| 7708 | fallthrough; |
| 7709 | case GAUDI_EVENT_MMU_SERR: |
Oded Gabbay | fcc6a4e | 2020-05-17 08:20:35 +0300 | [diff] [blame] | 7710 | gaudi_print_irq_info(hdev, event_type, true); |
| 7711 | gaudi_handle_ecc_event(hdev, event_type, &eq_entry->ecc_data); |
| 7712 | hl_fw_unmask_irq(hdev, event_type); |
| 7713 | break; |
| 7714 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7715 | case GAUDI_EVENT_PCIE_DEC: |
| 7716 | case GAUDI_EVENT_MME0_WBC_RSP: |
| 7717 | case GAUDI_EVENT_MME0_SBAB0_RSP: |
| 7718 | case GAUDI_EVENT_MME1_WBC_RSP: |
| 7719 | case GAUDI_EVENT_MME1_SBAB0_RSP: |
| 7720 | case GAUDI_EVENT_MME2_WBC_RSP: |
| 7721 | case GAUDI_EVENT_MME2_SBAB0_RSP: |
| 7722 | case GAUDI_EVENT_MME3_WBC_RSP: |
| 7723 | case GAUDI_EVENT_MME3_SBAB0_RSP: |
| 7724 | case GAUDI_EVENT_CPU_AXI_SPLITTER: |
| 7725 | case GAUDI_EVENT_PSOC_AXI_DEC: |
| 7726 | case GAUDI_EVENT_PSOC_PRSTN_FALL: |
| 7727 | case GAUDI_EVENT_MMU_PAGE_FAULT: |
| 7728 | case GAUDI_EVENT_MMU_WR_PERM: |
| 7729 | case GAUDI_EVENT_RAZWI_OR_ADC: |
| 7730 | case GAUDI_EVENT_TPC0_QM ... GAUDI_EVENT_TPC7_QM: |
| 7731 | case GAUDI_EVENT_MME0_QM ... GAUDI_EVENT_MME2_QM: |
| 7732 | case GAUDI_EVENT_DMA0_QM ... GAUDI_EVENT_DMA7_QM: |
| 7733 | fallthrough; |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 7734 | case GAUDI_EVENT_NIC0_QM0: |
| 7735 | case GAUDI_EVENT_NIC0_QM1: |
| 7736 | case GAUDI_EVENT_NIC1_QM0: |
| 7737 | case GAUDI_EVENT_NIC1_QM1: |
| 7738 | case GAUDI_EVENT_NIC2_QM0: |
| 7739 | case GAUDI_EVENT_NIC2_QM1: |
| 7740 | case GAUDI_EVENT_NIC3_QM0: |
| 7741 | case GAUDI_EVENT_NIC3_QM1: |
| 7742 | case GAUDI_EVENT_NIC4_QM0: |
| 7743 | case GAUDI_EVENT_NIC4_QM1: |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7744 | case GAUDI_EVENT_DMA0_CORE ... GAUDI_EVENT_DMA7_CORE: |
| 7745 | gaudi_print_irq_info(hdev, event_type, true); |
| 7746 | gaudi_handle_qman_err(hdev, event_type); |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 7747 | hl_fw_unmask_irq(hdev, event_type); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7748 | break; |
| 7749 | |
| 7750 | case GAUDI_EVENT_RAZWI_OR_ADC_SW: |
| 7751 | gaudi_print_irq_info(hdev, event_type, true); |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7752 | goto reset_device; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7753 | |
| 7754 | case GAUDI_EVENT_TPC0_BMON_SPMU: |
| 7755 | case GAUDI_EVENT_TPC1_BMON_SPMU: |
| 7756 | case GAUDI_EVENT_TPC2_BMON_SPMU: |
| 7757 | case GAUDI_EVENT_TPC3_BMON_SPMU: |
| 7758 | case GAUDI_EVENT_TPC4_BMON_SPMU: |
| 7759 | case GAUDI_EVENT_TPC5_BMON_SPMU: |
| 7760 | case GAUDI_EVENT_TPC6_BMON_SPMU: |
| 7761 | case GAUDI_EVENT_TPC7_BMON_SPMU: |
| 7762 | case GAUDI_EVENT_DMA_BM_CH0 ... GAUDI_EVENT_DMA_BM_CH7: |
| 7763 | gaudi_print_irq_info(hdev, event_type, false); |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 7764 | hl_fw_unmask_irq(hdev, event_type); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7765 | break; |
| 7766 | |
Ofir Bitton | f8bc7f0 | 2021-01-03 20:52:40 +0200 | [diff] [blame] | 7767 | case GAUDI_EVENT_DMA_IF_SEI_0 ... GAUDI_EVENT_DMA_IF_SEI_3: |
| 7768 | gaudi_print_irq_info(hdev, event_type, false); |
| 7769 | gaudi_print_sm_sei_info(hdev, event_type, |
| 7770 | &eq_entry->sm_sei_data); |
| 7771 | hl_fw_unmask_irq(hdev, event_type); |
| 7772 | break; |
| 7773 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7774 | case GAUDI_EVENT_FIX_POWER_ENV_S ... GAUDI_EVENT_FIX_THERMAL_ENV_E: |
| 7775 | gaudi_print_clk_change_info(hdev, event_type); |
Ofir Bitton | ebd8d12 | 2020-05-10 13:41:28 +0300 | [diff] [blame] | 7776 | hl_fw_unmask_irq(hdev, event_type); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7777 | break; |
| 7778 | |
| 7779 | case GAUDI_EVENT_PSOC_GPIO_U16_0: |
| 7780 | cause = le64_to_cpu(eq_entry->data[0]) & 0xFF; |
| 7781 | dev_err(hdev->dev, |
| 7782 | "Received high temp H/W interrupt %d (cause %d)\n", |
| 7783 | event_type, cause); |
| 7784 | break; |
| 7785 | |
Ofir Bitton | d661d79 | 2021-03-09 14:45:04 +0200 | [diff] [blame] | 7786 | case GAUDI_EVENT_DEV_RESET_REQ: |
Ofir Bitton | 2ea0953 | 2021-03-03 13:23:47 +0200 | [diff] [blame] | 7787 | gaudi_print_irq_info(hdev, event_type, false); |
| 7788 | goto reset_device; |
| 7789 | |
Ohad Sharabi | 5d6a198 | 2021-02-08 14:53:56 +0200 | [diff] [blame] | 7790 | case GAUDI_EVENT_PKT_QUEUE_OUT_SYNC: |
| 7791 | gaudi_print_irq_info(hdev, event_type, false); |
| 7792 | gaudi_print_out_of_sync_info(hdev, &eq_entry->pkt_sync_err); |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7793 | goto reset_device; |
Ohad Sharabi | 5d6a198 | 2021-02-08 14:53:56 +0200 | [diff] [blame] | 7794 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7795 | default: |
| 7796 | dev_err(hdev->dev, "Received invalid H/W interrupt %d\n", |
| 7797 | event_type); |
| 7798 | break; |
| 7799 | } |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7800 | |
| 7801 | return; |
| 7802 | |
| 7803 | reset_device: |
| 7804 | if (hdev->hard_reset_on_fw_events) |
| 7805 | hl_device_reset(hdev, HL_RESET_HARD); |
| 7806 | else |
| 7807 | hl_fw_unmask_irq(hdev, event_type); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7808 | } |
| 7809 | |
| 7810 | static void *gaudi_get_events_stat(struct hl_device *hdev, bool aggregate, |
| 7811 | u32 *size) |
| 7812 | { |
| 7813 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7814 | |
| 7815 | if (aggregate) { |
| 7816 | *size = (u32) sizeof(gaudi->events_stat_aggregate); |
| 7817 | return gaudi->events_stat_aggregate; |
| 7818 | } |
| 7819 | |
| 7820 | *size = (u32) sizeof(gaudi->events_stat); |
| 7821 | return gaudi->events_stat; |
| 7822 | } |
| 7823 | |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7824 | static int gaudi_mmu_invalidate_cache(struct hl_device *hdev, bool is_hard, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7825 | u32 flags) |
| 7826 | { |
| 7827 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7828 | u32 status, timeout_usec; |
| 7829 | int rc; |
| 7830 | |
| 7831 | if (!(gaudi->hw_cap_initialized & HW_CAP_MMU) || |
| 7832 | hdev->hard_reset_pending) |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7833 | return 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7834 | |
| 7835 | if (hdev->pldm) |
| 7836 | timeout_usec = GAUDI_PLDM_MMU_TIMEOUT_USEC; |
| 7837 | else |
| 7838 | timeout_usec = MMU_CONFIG_TIMEOUT_USEC; |
| 7839 | |
| 7840 | /* L0 & L1 invalidation */ |
Omer Shpigelman | cfd4176 | 2020-06-03 13:03:35 +0300 | [diff] [blame] | 7841 | WREG32(mmSTLB_INV_PS, 3); |
| 7842 | WREG32(mmSTLB_CACHE_INV, gaudi->mmu_cache_inv_pi++); |
Omer Shpigelman | 42d0b0b | 2020-05-17 17:35:39 +0300 | [diff] [blame] | 7843 | WREG32(mmSTLB_INV_PS, 2); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7844 | |
| 7845 | rc = hl_poll_timeout( |
| 7846 | hdev, |
Omer Shpigelman | 42d0b0b | 2020-05-17 17:35:39 +0300 | [diff] [blame] | 7847 | mmSTLB_INV_PS, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7848 | status, |
| 7849 | !status, |
| 7850 | 1000, |
| 7851 | timeout_usec); |
| 7852 | |
Omer Shpigelman | 42d0b0b | 2020-05-17 17:35:39 +0300 | [diff] [blame] | 7853 | WREG32(mmSTLB_INV_SET, 0); |
| 7854 | |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7855 | if (rc) { |
| 7856 | dev_err_ratelimited(hdev->dev, |
| 7857 | "MMU cache invalidation timeout\n"); |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7858 | hl_device_reset(hdev, HL_RESET_HARD); |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7859 | } |
| 7860 | |
| 7861 | return rc; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7862 | } |
| 7863 | |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7864 | static int gaudi_mmu_invalidate_cache_range(struct hl_device *hdev, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7865 | bool is_hard, u32 asid, u64 va, u64 size) |
| 7866 | { |
| 7867 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7868 | u32 status, timeout_usec; |
| 7869 | u32 inv_data; |
| 7870 | u32 pi; |
| 7871 | int rc; |
| 7872 | |
| 7873 | if (!(gaudi->hw_cap_initialized & HW_CAP_MMU) || |
| 7874 | hdev->hard_reset_pending) |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7875 | return 0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7876 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7877 | if (hdev->pldm) |
| 7878 | timeout_usec = GAUDI_PLDM_MMU_TIMEOUT_USEC; |
| 7879 | else |
| 7880 | timeout_usec = MMU_CONFIG_TIMEOUT_USEC; |
| 7881 | |
| 7882 | /* |
| 7883 | * TODO: currently invalidate entire L0 & L1 as in regular hard |
| 7884 | * invalidation. Need to apply invalidation of specific cache |
| 7885 | * lines with mask of ASID & VA & size. |
| 7886 | * Note that L1 with be flushed entirely in any case. |
| 7887 | */ |
| 7888 | |
| 7889 | /* L0 & L1 invalidation */ |
| 7890 | inv_data = RREG32(mmSTLB_CACHE_INV); |
| 7891 | /* PI is 8 bit */ |
| 7892 | pi = ((inv_data & STLB_CACHE_INV_PRODUCER_INDEX_MASK) + 1) & 0xFF; |
| 7893 | WREG32(mmSTLB_CACHE_INV, |
| 7894 | (inv_data & STLB_CACHE_INV_INDEX_MASK_MASK) | pi); |
| 7895 | |
| 7896 | rc = hl_poll_timeout( |
| 7897 | hdev, |
| 7898 | mmSTLB_INV_CONSUMER_INDEX, |
| 7899 | status, |
| 7900 | status == pi, |
| 7901 | 1000, |
| 7902 | timeout_usec); |
| 7903 | |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7904 | if (rc) { |
| 7905 | dev_err_ratelimited(hdev->dev, |
| 7906 | "MMU cache invalidation timeout\n"); |
Ohad Sharabi | e42a6400f | 2021-02-17 20:42:48 +0200 | [diff] [blame] | 7907 | hl_device_reset(hdev, HL_RESET_HARD); |
Omer Shpigelman | 8ff5f4f | 2020-05-24 23:06:59 +0300 | [diff] [blame] | 7908 | } |
| 7909 | |
| 7910 | return rc; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7911 | } |
| 7912 | |
| 7913 | static int gaudi_mmu_update_asid_hop0_addr(struct hl_device *hdev, |
| 7914 | u32 asid, u64 phys_addr) |
| 7915 | { |
| 7916 | u32 status, timeout_usec; |
| 7917 | int rc; |
| 7918 | |
| 7919 | if (hdev->pldm) |
| 7920 | timeout_usec = GAUDI_PLDM_MMU_TIMEOUT_USEC; |
| 7921 | else |
| 7922 | timeout_usec = MMU_CONFIG_TIMEOUT_USEC; |
| 7923 | |
| 7924 | WREG32(MMU_ASID, asid); |
| 7925 | WREG32(MMU_HOP0_PA43_12, phys_addr >> MMU_HOP0_PA43_12_SHIFT); |
| 7926 | WREG32(MMU_HOP0_PA49_44, phys_addr >> MMU_HOP0_PA49_44_SHIFT); |
| 7927 | WREG32(MMU_BUSY, 0x80000000); |
| 7928 | |
| 7929 | rc = hl_poll_timeout( |
| 7930 | hdev, |
| 7931 | MMU_BUSY, |
| 7932 | status, |
| 7933 | !(status & 0x80000000), |
| 7934 | 1000, |
| 7935 | timeout_usec); |
| 7936 | |
| 7937 | if (rc) { |
| 7938 | dev_err(hdev->dev, |
| 7939 | "Timeout during MMU hop0 config of asid %d\n", asid); |
| 7940 | return rc; |
| 7941 | } |
| 7942 | |
| 7943 | return 0; |
| 7944 | } |
| 7945 | |
| 7946 | static int gaudi_send_heartbeat(struct hl_device *hdev) |
| 7947 | { |
| 7948 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7949 | |
| 7950 | if (!(gaudi->hw_cap_initialized & HW_CAP_CPU_Q)) |
| 7951 | return 0; |
| 7952 | |
| 7953 | return hl_fw_send_heartbeat(hdev); |
| 7954 | } |
| 7955 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 7956 | static int gaudi_cpucp_info_get(struct hl_device *hdev) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7957 | { |
| 7958 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7959 | struct asic_fixed_properties *prop = &hdev->asic_prop; |
| 7960 | int rc; |
| 7961 | |
| 7962 | if (!(gaudi->hw_cap_initialized & HW_CAP_CPU_Q)) |
| 7963 | return 0; |
| 7964 | |
Ohad Sharabi | e9c2003 | 2021-03-22 14:30:52 +0200 | [diff] [blame] | 7965 | rc = hl_fw_cpucp_handshake(hdev, mmCPU_BOOT_DEV_STS0, mmCPU_BOOT_ERR0); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7966 | if (rc) |
| 7967 | return rc; |
| 7968 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 7969 | if (!strlen(prop->cpucp_info.card_name)) |
| 7970 | strncpy(prop->cpucp_info.card_name, GAUDI_DEFAULT_CARD_NAME, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7971 | CARD_NAME_MAX_LEN); |
| 7972 | |
Oded Gabbay | 2f55342 | 2020-08-15 16:28:10 +0300 | [diff] [blame] | 7973 | hdev->card_type = le32_to_cpu(hdev->asic_prop.cpucp_info.card_type); |
Oded Gabbay | 58361aa | 2020-08-08 23:34:47 +0300 | [diff] [blame] | 7974 | |
Koby Elbaz | cd5def8 | 2021-02-23 21:31:27 +0200 | [diff] [blame] | 7975 | set_default_power_values(hdev); |
Oded Gabbay | 58361aa | 2020-08-08 23:34:47 +0300 | [diff] [blame] | 7976 | |
| 7977 | hdev->max_power = prop->max_power_default; |
| 7978 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7979 | return 0; |
| 7980 | } |
| 7981 | |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 7982 | static bool gaudi_is_device_idle(struct hl_device *hdev, u64 *mask_arr, |
| 7983 | u8 mask_len, struct seq_file *s) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7984 | { |
| 7985 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 7986 | const char *fmt = "%-5d%-9s%#-14x%#-12x%#x\n"; |
| 7987 | const char *mme_slave_fmt = "%-5d%-9s%-14s%-12s%#x\n"; |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 7988 | const char *nic_fmt = "%-5d%-9s%#-14x%#x\n"; |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 7989 | unsigned long *mask = (unsigned long *)mask_arr; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7990 | u32 qm_glbl_sts0, qm_cgm_sts, dma_core_sts0, tpc_cfg_sts, mme_arch_sts; |
| 7991 | bool is_idle = true, is_eng_idle, is_slave; |
| 7992 | u64 offset; |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 7993 | int i, dma_id, port; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 7994 | |
| 7995 | mutex_lock(&gaudi->clk_gate_mutex); |
| 7996 | |
| 7997 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 7998 | |
| 7999 | if (s) |
| 8000 | seq_puts(s, |
| 8001 | "\nDMA is_idle QM_GLBL_STS0 QM_CGM_STS DMA_CORE_STS0\n" |
| 8002 | "--- ------- ------------ ---------- -------------\n"); |
| 8003 | |
| 8004 | for (i = 0 ; i < DMA_NUMBER_OF_CHNLS ; i++) { |
| 8005 | dma_id = gaudi_dma_assignment[i]; |
| 8006 | offset = dma_id * DMA_QMAN_OFFSET; |
| 8007 | |
| 8008 | qm_glbl_sts0 = RREG32(mmDMA0_QM_GLBL_STS0 + offset); |
| 8009 | qm_cgm_sts = RREG32(mmDMA0_QM_CGM_STS + offset); |
| 8010 | dma_core_sts0 = RREG32(mmDMA0_CORE_STS0 + offset); |
| 8011 | is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_cgm_sts) && |
| 8012 | IS_DMA_IDLE(dma_core_sts0); |
| 8013 | is_idle &= is_eng_idle; |
| 8014 | |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 8015 | if (mask && !is_eng_idle) |
| 8016 | set_bit(GAUDI_ENGINE_ID_DMA_0 + dma_id, mask); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8017 | if (s) |
| 8018 | seq_printf(s, fmt, dma_id, |
| 8019 | is_eng_idle ? "Y" : "N", qm_glbl_sts0, |
| 8020 | qm_cgm_sts, dma_core_sts0); |
| 8021 | } |
| 8022 | |
| 8023 | if (s) |
| 8024 | seq_puts(s, |
| 8025 | "\nTPC is_idle QM_GLBL_STS0 QM_CGM_STS CFG_STATUS\n" |
| 8026 | "--- ------- ------------ ---------- ----------\n"); |
| 8027 | |
| 8028 | for (i = 0 ; i < TPC_NUMBER_OF_ENGINES ; i++) { |
| 8029 | offset = i * TPC_QMAN_OFFSET; |
| 8030 | qm_glbl_sts0 = RREG32(mmTPC0_QM_GLBL_STS0 + offset); |
| 8031 | qm_cgm_sts = RREG32(mmTPC0_QM_CGM_STS + offset); |
| 8032 | tpc_cfg_sts = RREG32(mmTPC0_CFG_STATUS + offset); |
| 8033 | is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_cgm_sts) && |
| 8034 | IS_TPC_IDLE(tpc_cfg_sts); |
| 8035 | is_idle &= is_eng_idle; |
| 8036 | |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 8037 | if (mask && !is_eng_idle) |
| 8038 | set_bit(GAUDI_ENGINE_ID_TPC_0 + i, mask); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8039 | if (s) |
| 8040 | seq_printf(s, fmt, i, |
| 8041 | is_eng_idle ? "Y" : "N", |
| 8042 | qm_glbl_sts0, qm_cgm_sts, tpc_cfg_sts); |
| 8043 | } |
| 8044 | |
| 8045 | if (s) |
| 8046 | seq_puts(s, |
| 8047 | "\nMME is_idle QM_GLBL_STS0 QM_CGM_STS ARCH_STATUS\n" |
| 8048 | "--- ------- ------------ ---------- -----------\n"); |
| 8049 | |
| 8050 | for (i = 0 ; i < MME_NUMBER_OF_ENGINES ; i++) { |
| 8051 | offset = i * MME_QMAN_OFFSET; |
| 8052 | mme_arch_sts = RREG32(mmMME0_CTRL_ARCH_STATUS + offset); |
| 8053 | is_eng_idle = IS_MME_IDLE(mme_arch_sts); |
| 8054 | |
| 8055 | /* MME 1 & 3 are slaves, no need to check their QMANs */ |
| 8056 | is_slave = i % 2; |
| 8057 | if (!is_slave) { |
| 8058 | qm_glbl_sts0 = RREG32(mmMME0_QM_GLBL_STS0 + offset); |
| 8059 | qm_cgm_sts = RREG32(mmMME0_QM_CGM_STS + offset); |
| 8060 | is_eng_idle &= IS_QM_IDLE(qm_glbl_sts0, qm_cgm_sts); |
| 8061 | } |
| 8062 | |
| 8063 | is_idle &= is_eng_idle; |
| 8064 | |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 8065 | if (mask && !is_eng_idle) |
| 8066 | set_bit(GAUDI_ENGINE_ID_MME_0 + i, mask); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8067 | if (s) { |
| 8068 | if (!is_slave) |
| 8069 | seq_printf(s, fmt, i, |
| 8070 | is_eng_idle ? "Y" : "N", |
| 8071 | qm_glbl_sts0, qm_cgm_sts, mme_arch_sts); |
| 8072 | else |
| 8073 | seq_printf(s, mme_slave_fmt, i, |
| 8074 | is_eng_idle ? "Y" : "N", "-", |
| 8075 | "-", mme_arch_sts); |
| 8076 | } |
| 8077 | } |
| 8078 | |
| 8079 | if (s) |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 8080 | seq_puts(s, "\nNIC is_idle QM_GLBL_STS0 QM_CGM_STS\n" |
| 8081 | "--- ------- ------------ ----------\n"); |
| 8082 | |
| 8083 | for (i = 0 ; i < (NIC_NUMBER_OF_ENGINES / 2) ; i++) { |
| 8084 | offset = i * NIC_MACRO_QMAN_OFFSET; |
| 8085 | port = 2 * i; |
| 8086 | if (hdev->nic_ports_mask & BIT(port)) { |
| 8087 | qm_glbl_sts0 = RREG32(mmNIC0_QM0_GLBL_STS0 + offset); |
| 8088 | qm_cgm_sts = RREG32(mmNIC0_QM0_CGM_STS + offset); |
| 8089 | is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_cgm_sts); |
| 8090 | is_idle &= is_eng_idle; |
| 8091 | |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 8092 | if (mask && !is_eng_idle) |
| 8093 | set_bit(GAUDI_ENGINE_ID_NIC_0 + port, mask); |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 8094 | if (s) |
| 8095 | seq_printf(s, nic_fmt, port, |
| 8096 | is_eng_idle ? "Y" : "N", |
| 8097 | qm_glbl_sts0, qm_cgm_sts); |
| 8098 | } |
| 8099 | |
| 8100 | port = 2 * i + 1; |
| 8101 | if (hdev->nic_ports_mask & BIT(port)) { |
| 8102 | qm_glbl_sts0 = RREG32(mmNIC0_QM1_GLBL_STS0 + offset); |
| 8103 | qm_cgm_sts = RREG32(mmNIC0_QM1_CGM_STS + offset); |
| 8104 | is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_cgm_sts); |
| 8105 | is_idle &= is_eng_idle; |
| 8106 | |
Ohad Sharabi | cf30339 | 2021-01-17 16:01:56 +0200 | [diff] [blame] | 8107 | if (mask && !is_eng_idle) |
| 8108 | set_bit(GAUDI_ENGINE_ID_NIC_0 + port, mask); |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 8109 | if (s) |
| 8110 | seq_printf(s, nic_fmt, port, |
| 8111 | is_eng_idle ? "Y" : "N", |
| 8112 | qm_glbl_sts0, qm_cgm_sts); |
| 8113 | } |
| 8114 | } |
| 8115 | |
| 8116 | if (s) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8117 | seq_puts(s, "\n"); |
| 8118 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 8119 | hdev->asic_funcs->set_clock_gating(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8120 | |
| 8121 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 8122 | |
| 8123 | return is_idle; |
| 8124 | } |
| 8125 | |
| 8126 | static void gaudi_hw_queues_lock(struct hl_device *hdev) |
| 8127 | __acquires(&gaudi->hw_queues_lock) |
| 8128 | { |
| 8129 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 8130 | |
| 8131 | spin_lock(&gaudi->hw_queues_lock); |
| 8132 | } |
| 8133 | |
| 8134 | static void gaudi_hw_queues_unlock(struct hl_device *hdev) |
| 8135 | __releases(&gaudi->hw_queues_lock) |
| 8136 | { |
| 8137 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 8138 | |
| 8139 | spin_unlock(&gaudi->hw_queues_lock); |
| 8140 | } |
| 8141 | |
| 8142 | static u32 gaudi_get_pci_id(struct hl_device *hdev) |
| 8143 | { |
| 8144 | return hdev->pdev->device; |
| 8145 | } |
| 8146 | |
| 8147 | static int gaudi_get_eeprom_data(struct hl_device *hdev, void *data, |
| 8148 | size_t max_size) |
| 8149 | { |
| 8150 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 8151 | |
| 8152 | if (!(gaudi->hw_cap_initialized & HW_CAP_CPU_Q)) |
| 8153 | return 0; |
| 8154 | |
| 8155 | return hl_fw_get_eeprom_data(hdev, data, max_size); |
| 8156 | } |
| 8157 | |
| 8158 | /* |
| 8159 | * this function should be used only during initialization and/or after reset, |
| 8160 | * when there are no active users. |
| 8161 | */ |
| 8162 | static int gaudi_run_tpc_kernel(struct hl_device *hdev, u64 tpc_kernel, |
| 8163 | u32 tpc_id) |
| 8164 | { |
| 8165 | struct gaudi_device *gaudi = hdev->asic_specific; |
| 8166 | u64 kernel_timeout; |
| 8167 | u32 status, offset; |
| 8168 | int rc; |
| 8169 | |
| 8170 | offset = tpc_id * (mmTPC1_CFG_STATUS - mmTPC0_CFG_STATUS); |
| 8171 | |
| 8172 | if (hdev->pldm) |
| 8173 | kernel_timeout = GAUDI_PLDM_TPC_KERNEL_WAIT_USEC; |
| 8174 | else |
| 8175 | kernel_timeout = HL_DEVICE_TIMEOUT_USEC; |
| 8176 | |
| 8177 | mutex_lock(&gaudi->clk_gate_mutex); |
| 8178 | |
| 8179 | hdev->asic_funcs->disable_clock_gating(hdev); |
| 8180 | |
| 8181 | WREG32(mmTPC0_CFG_QM_KERNEL_BASE_ADDRESS_LOW + offset, |
| 8182 | lower_32_bits(tpc_kernel)); |
| 8183 | WREG32(mmTPC0_CFG_QM_KERNEL_BASE_ADDRESS_HIGH + offset, |
| 8184 | upper_32_bits(tpc_kernel)); |
| 8185 | |
| 8186 | WREG32(mmTPC0_CFG_ICACHE_BASE_ADDERESS_LOW + offset, |
| 8187 | lower_32_bits(tpc_kernel)); |
| 8188 | WREG32(mmTPC0_CFG_ICACHE_BASE_ADDERESS_HIGH + offset, |
| 8189 | upper_32_bits(tpc_kernel)); |
| 8190 | /* set a valid LUT pointer, content is of no significance */ |
| 8191 | WREG32(mmTPC0_CFG_LUT_FUNC256_BASE_ADDR_LO + offset, |
| 8192 | lower_32_bits(tpc_kernel)); |
| 8193 | WREG32(mmTPC0_CFG_LUT_FUNC256_BASE_ADDR_HI + offset, |
| 8194 | upper_32_bits(tpc_kernel)); |
| 8195 | |
| 8196 | WREG32(mmTPC0_CFG_QM_SYNC_OBJECT_ADDR + offset, |
| 8197 | lower_32_bits(CFG_BASE + |
| 8198 | mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0)); |
| 8199 | |
| 8200 | WREG32(mmTPC0_CFG_TPC_CMD + offset, |
| 8201 | (1 << TPC0_CFG_TPC_CMD_ICACHE_INVALIDATE_SHIFT | |
| 8202 | 1 << TPC0_CFG_TPC_CMD_ICACHE_PREFETCH_64KB_SHIFT)); |
| 8203 | /* wait a bit for the engine to start executing */ |
| 8204 | usleep_range(1000, 1500); |
| 8205 | |
| 8206 | /* wait until engine has finished executing */ |
| 8207 | rc = hl_poll_timeout( |
| 8208 | hdev, |
| 8209 | mmTPC0_CFG_STATUS + offset, |
| 8210 | status, |
| 8211 | (status & TPC0_CFG_STATUS_VECTOR_PIPE_EMPTY_MASK) == |
| 8212 | TPC0_CFG_STATUS_VECTOR_PIPE_EMPTY_MASK, |
| 8213 | 1000, |
| 8214 | kernel_timeout); |
| 8215 | |
| 8216 | if (rc) { |
| 8217 | dev_err(hdev->dev, |
| 8218 | "Timeout while waiting for TPC%d icache prefetch\n", |
| 8219 | tpc_id); |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 8220 | hdev->asic_funcs->set_clock_gating(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8221 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 8222 | return -EIO; |
| 8223 | } |
| 8224 | |
| 8225 | WREG32(mmTPC0_CFG_TPC_EXECUTE + offset, |
| 8226 | 1 << TPC0_CFG_TPC_EXECUTE_V_SHIFT); |
| 8227 | |
| 8228 | /* wait a bit for the engine to start executing */ |
| 8229 | usleep_range(1000, 1500); |
| 8230 | |
| 8231 | /* wait until engine has finished executing */ |
| 8232 | rc = hl_poll_timeout( |
| 8233 | hdev, |
| 8234 | mmTPC0_CFG_STATUS + offset, |
| 8235 | status, |
| 8236 | (status & TPC0_CFG_STATUS_VECTOR_PIPE_EMPTY_MASK) == |
| 8237 | TPC0_CFG_STATUS_VECTOR_PIPE_EMPTY_MASK, |
| 8238 | 1000, |
| 8239 | kernel_timeout); |
| 8240 | |
Oded Gabbay | 31ac1f1 | 2020-08-12 11:28:13 +0300 | [diff] [blame] | 8241 | if (rc) { |
| 8242 | dev_err(hdev->dev, |
| 8243 | "Timeout while waiting for TPC%d vector pipe\n", |
| 8244 | tpc_id); |
| 8245 | hdev->asic_funcs->set_clock_gating(hdev); |
| 8246 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 8247 | return -EIO; |
| 8248 | } |
| 8249 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8250 | rc = hl_poll_timeout( |
| 8251 | hdev, |
| 8252 | mmTPC0_CFG_WQ_INFLIGHT_CNTR + offset, |
| 8253 | status, |
| 8254 | (status == 0), |
| 8255 | 1000, |
| 8256 | kernel_timeout); |
| 8257 | |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 8258 | hdev->asic_funcs->set_clock_gating(hdev); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8259 | mutex_unlock(&gaudi->clk_gate_mutex); |
| 8260 | |
| 8261 | if (rc) { |
| 8262 | dev_err(hdev->dev, |
| 8263 | "Timeout while waiting for TPC%d kernel to execute\n", |
| 8264 | tpc_id); |
| 8265 | return -EIO; |
| 8266 | } |
| 8267 | |
| 8268 | return 0; |
| 8269 | } |
| 8270 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8271 | static int gaudi_internal_cb_pool_init(struct hl_device *hdev, |
| 8272 | struct hl_ctx *ctx) |
| 8273 | { |
| 8274 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8275 | int min_alloc_order, rc, collective_cb_size; |
| 8276 | |
| 8277 | if (!(gaudi->hw_cap_initialized & HW_CAP_MMU)) |
| 8278 | return 0; |
| 8279 | |
| 8280 | hdev->internal_cb_pool_virt_addr = |
| 8281 | hdev->asic_funcs->asic_dma_alloc_coherent(hdev, |
| 8282 | HOST_SPACE_INTERNAL_CB_SZ, |
| 8283 | &hdev->internal_cb_pool_dma_addr, |
| 8284 | GFP_KERNEL | __GFP_ZERO); |
| 8285 | |
| 8286 | if (!hdev->internal_cb_pool_virt_addr) |
| 8287 | return -ENOMEM; |
| 8288 | |
| 8289 | collective_cb_size = sizeof(struct packet_msg_short) * 5 + |
| 8290 | sizeof(struct packet_fence); |
| 8291 | min_alloc_order = ilog2(collective_cb_size); |
| 8292 | |
| 8293 | hdev->internal_cb_pool = gen_pool_create(min_alloc_order, -1); |
| 8294 | if (!hdev->internal_cb_pool) { |
| 8295 | dev_err(hdev->dev, |
| 8296 | "Failed to create internal CB pool\n"); |
| 8297 | rc = -ENOMEM; |
| 8298 | goto free_internal_cb_pool; |
| 8299 | } |
| 8300 | |
| 8301 | rc = gen_pool_add(hdev->internal_cb_pool, |
| 8302 | (uintptr_t) hdev->internal_cb_pool_virt_addr, |
| 8303 | HOST_SPACE_INTERNAL_CB_SZ, -1); |
| 8304 | if (rc) { |
| 8305 | dev_err(hdev->dev, |
| 8306 | "Failed to add memory to internal CB pool\n"); |
| 8307 | rc = -EFAULT; |
| 8308 | goto destroy_internal_cb_pool; |
| 8309 | } |
| 8310 | |
Ofir Bitton | be91b91 | 2020-10-22 15:04:10 +0300 | [diff] [blame] | 8311 | hdev->internal_cb_va_base = hl_reserve_va_block(hdev, ctx, |
Ofir Bitton | 412c41f | 2020-11-04 15:18:55 +0200 | [diff] [blame] | 8312 | HL_VA_RANGE_TYPE_HOST, HOST_SPACE_INTERNAL_CB_SZ, |
| 8313 | HL_MMU_VA_ALIGNMENT_NOT_NEEDED); |
Ofir Bitton | be91b91 | 2020-10-22 15:04:10 +0300 | [diff] [blame] | 8314 | |
| 8315 | if (!hdev->internal_cb_va_base) |
| 8316 | goto destroy_internal_cb_pool; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8317 | |
| 8318 | mutex_lock(&ctx->mmu_lock); |
Ofir Bitton | 5c05487 | 2020-10-22 15:13:10 +0300 | [diff] [blame] | 8319 | rc = hl_mmu_map_contiguous(ctx, hdev->internal_cb_va_base, |
| 8320 | hdev->internal_cb_pool_dma_addr, |
| 8321 | HOST_SPACE_INTERNAL_CB_SZ); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8322 | |
| 8323 | hdev->asic_funcs->mmu_invalidate_cache(hdev, false, VM_TYPE_USERPTR); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8324 | mutex_unlock(&ctx->mmu_lock); |
| 8325 | |
Ofir Bitton | 5c05487 | 2020-10-22 15:13:10 +0300 | [diff] [blame] | 8326 | if (rc) |
| 8327 | goto unreserve_internal_cb_pool; |
| 8328 | |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8329 | return 0; |
| 8330 | |
Ofir Bitton | 5c05487 | 2020-10-22 15:13:10 +0300 | [diff] [blame] | 8331 | unreserve_internal_cb_pool: |
Ofir Bitton | be91b91 | 2020-10-22 15:04:10 +0300 | [diff] [blame] | 8332 | hl_unreserve_va_block(hdev, ctx, hdev->internal_cb_va_base, |
| 8333 | HOST_SPACE_INTERNAL_CB_SZ); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8334 | destroy_internal_cb_pool: |
| 8335 | gen_pool_destroy(hdev->internal_cb_pool); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8336 | free_internal_cb_pool: |
| 8337 | hdev->asic_funcs->asic_dma_free_coherent(hdev, |
| 8338 | HOST_SPACE_INTERNAL_CB_SZ, |
| 8339 | hdev->internal_cb_pool_virt_addr, |
| 8340 | hdev->internal_cb_pool_dma_addr); |
| 8341 | |
| 8342 | return rc; |
| 8343 | } |
| 8344 | |
| 8345 | static void gaudi_internal_cb_pool_fini(struct hl_device *hdev, |
| 8346 | struct hl_ctx *ctx) |
| 8347 | { |
| 8348 | struct gaudi_device *gaudi = hdev->asic_specific; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8349 | |
| 8350 | if (!(gaudi->hw_cap_initialized & HW_CAP_MMU)) |
| 8351 | return; |
| 8352 | |
| 8353 | mutex_lock(&ctx->mmu_lock); |
Ofir Bitton | 5c05487 | 2020-10-22 15:13:10 +0300 | [diff] [blame] | 8354 | hl_mmu_unmap_contiguous(ctx, hdev->internal_cb_va_base, |
| 8355 | HOST_SPACE_INTERNAL_CB_SZ); |
Ofir Bitton | be91b91 | 2020-10-22 15:04:10 +0300 | [diff] [blame] | 8356 | hl_unreserve_va_block(hdev, ctx, hdev->internal_cb_va_base, |
| 8357 | HOST_SPACE_INTERNAL_CB_SZ); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8358 | hdev->asic_funcs->mmu_invalidate_cache(hdev, true, VM_TYPE_USERPTR); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8359 | mutex_unlock(&ctx->mmu_lock); |
| 8360 | |
| 8361 | gen_pool_destroy(hdev->internal_cb_pool); |
| 8362 | |
| 8363 | hdev->asic_funcs->asic_dma_free_coherent(hdev, |
| 8364 | HOST_SPACE_INTERNAL_CB_SZ, |
| 8365 | hdev->internal_cb_pool_virt_addr, |
| 8366 | hdev->internal_cb_pool_dma_addr); |
| 8367 | } |
| 8368 | |
kernel test robot | bb34bf7 | 2020-07-29 08:03:13 +0800 | [diff] [blame] | 8369 | static int gaudi_ctx_init(struct hl_ctx *ctx) |
Ofir Bitton | a04b7cd | 2020-07-13 13:36:55 +0300 | [diff] [blame] | 8370 | { |
Ofir Bitton | 8e39e75 | 2020-11-12 11:03:32 +0200 | [diff] [blame] | 8371 | if (ctx->asid == HL_KERNEL_ASID_ID) |
| 8372 | return 0; |
| 8373 | |
Ofir Bitton | 20b7525 | 2020-09-30 15:51:10 +0300 | [diff] [blame] | 8374 | gaudi_mmu_prepare(ctx->hdev, ctx->asid); |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8375 | return gaudi_internal_cb_pool_init(ctx->hdev, ctx); |
| 8376 | } |
Ofir Bitton | 20b7525 | 2020-09-30 15:51:10 +0300 | [diff] [blame] | 8377 | |
kernel test robot | 293744d | 2020-11-19 12:25:43 +0800 | [diff] [blame] | 8378 | static void gaudi_ctx_fini(struct hl_ctx *ctx) |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8379 | { |
Ofir Bitton | 8e39e75 | 2020-11-12 11:03:32 +0200 | [diff] [blame] | 8380 | if (ctx->asid == HL_KERNEL_ASID_ID) |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8381 | return; |
| 8382 | |
| 8383 | gaudi_internal_cb_pool_fini(ctx->hdev, ctx); |
Ofir Bitton | a04b7cd | 2020-07-13 13:36:55 +0300 | [diff] [blame] | 8384 | } |
| 8385 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8386 | static u32 gaudi_get_queue_id_for_cq(struct hl_device *hdev, u32 cq_idx) |
| 8387 | { |
| 8388 | return gaudi_cq_assignment[cq_idx]; |
| 8389 | } |
| 8390 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8391 | static u32 gaudi_get_signal_cb_size(struct hl_device *hdev) |
| 8392 | { |
| 8393 | return sizeof(struct packet_msg_short) + |
| 8394 | sizeof(struct packet_msg_prot) * 2; |
| 8395 | } |
| 8396 | |
| 8397 | static u32 gaudi_get_wait_cb_size(struct hl_device *hdev) |
| 8398 | { |
| 8399 | return sizeof(struct packet_msg_short) * 4 + |
| 8400 | sizeof(struct packet_fence) + |
| 8401 | sizeof(struct packet_msg_prot) * 2; |
| 8402 | } |
| 8403 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8404 | static u32 gaudi_gen_signal_cb(struct hl_device *hdev, void *data, u16 sob_id, |
Alon Mizrahi | 72ab9ca5 | 2020-12-02 19:55:30 +0200 | [diff] [blame] | 8405 | u32 size, bool eb) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8406 | { |
| 8407 | struct hl_cb *cb = (struct hl_cb *) data; |
| 8408 | struct packet_msg_short *pkt; |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8409 | u32 value, ctl, pkt_size = sizeof(*pkt); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8410 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8411 | pkt = cb->kernel_address + size; |
| 8412 | memset(pkt, 0, pkt_size); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8413 | |
Ofir Bitton | 6c07bab | 2020-06-01 10:38:46 +0300 | [diff] [blame] | 8414 | /* Inc by 1, Mode ADD */ |
| 8415 | value = FIELD_PREP(GAUDI_PKT_SHORT_VAL_SOB_SYNC_VAL_MASK, 1); |
| 8416 | value |= FIELD_PREP(GAUDI_PKT_SHORT_VAL_SOB_MOD_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8417 | |
Ofir Bitton | 6c07bab | 2020-06-01 10:38:46 +0300 | [diff] [blame] | 8418 | ctl = FIELD_PREP(GAUDI_PKT_SHORT_CTL_ADDR_MASK, sob_id * 4); |
| 8419 | ctl |= FIELD_PREP(GAUDI_PKT_SHORT_CTL_OP_MASK, 0); /* write the value */ |
| 8420 | ctl |= FIELD_PREP(GAUDI_PKT_SHORT_CTL_BASE_MASK, 3); /* W_S SOB base */ |
Ofir Bitton | f8b0f2e | 2020-12-06 10:22:32 +0200 | [diff] [blame] | 8421 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_SHORT); |
| 8422 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, eb); |
| 8423 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 8424 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8425 | |
| 8426 | pkt->value = cpu_to_le32(value); |
| 8427 | pkt->ctl = cpu_to_le32(ctl); |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8428 | |
| 8429 | return size + pkt_size; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8430 | } |
| 8431 | |
| 8432 | static u32 gaudi_add_mon_msg_short(struct packet_msg_short *pkt, u32 value, |
| 8433 | u16 addr) |
| 8434 | { |
| 8435 | u32 ctl, pkt_size = sizeof(*pkt); |
| 8436 | |
| 8437 | memset(pkt, 0, pkt_size); |
| 8438 | |
Ofir Bitton | 6c07bab | 2020-06-01 10:38:46 +0300 | [diff] [blame] | 8439 | ctl = FIELD_PREP(GAUDI_PKT_SHORT_CTL_ADDR_MASK, addr); |
| 8440 | ctl |= FIELD_PREP(GAUDI_PKT_SHORT_CTL_BASE_MASK, 2); /* W_S MON base */ |
Ofir Bitton | f8b0f2e | 2020-12-06 10:22:32 +0200 | [diff] [blame] | 8441 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_SHORT); |
| 8442 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 0); |
| 8443 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 8444 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 0); /* last pkt MB */ |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8445 | |
| 8446 | pkt->value = cpu_to_le32(value); |
| 8447 | pkt->ctl = cpu_to_le32(ctl); |
| 8448 | |
| 8449 | return pkt_size; |
| 8450 | } |
| 8451 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8452 | static u32 gaudi_add_arm_monitor_pkt(struct hl_device *hdev, |
| 8453 | struct packet_msg_short *pkt, u16 sob_base, u8 sob_mask, |
| 8454 | u16 sob_val, u16 mon_id) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8455 | { |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8456 | u64 monitor_base; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8457 | u32 ctl, value, pkt_size = sizeof(*pkt); |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8458 | u16 msg_addr_offset; |
| 8459 | u8 mask; |
| 8460 | |
| 8461 | if (hl_gen_sob_mask(sob_base, sob_mask, &mask)) { |
| 8462 | dev_err(hdev->dev, |
| 8463 | "sob_base %u (mask %#x) is not valid\n", |
| 8464 | sob_base, sob_mask); |
| 8465 | return 0; |
| 8466 | } |
| 8467 | |
| 8468 | /* |
| 8469 | * monitor_base should be the content of the base0 address registers, |
| 8470 | * so it will be added to the msg short offsets |
| 8471 | */ |
| 8472 | monitor_base = mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0; |
| 8473 | |
| 8474 | msg_addr_offset = |
| 8475 | (mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_ARM_0 + mon_id * 4) - |
| 8476 | monitor_base; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8477 | |
| 8478 | memset(pkt, 0, pkt_size); |
| 8479 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8480 | /* Monitor config packet: bind the monitor to a sync object */ |
| 8481 | value = FIELD_PREP(GAUDI_PKT_SHORT_VAL_MON_SYNC_GID_MASK, sob_base / 8); |
Ofir Bitton | 6c07bab | 2020-06-01 10:38:46 +0300 | [diff] [blame] | 8482 | value |= FIELD_PREP(GAUDI_PKT_SHORT_VAL_MON_SYNC_VAL_MASK, sob_val); |
| 8483 | value |= FIELD_PREP(GAUDI_PKT_SHORT_VAL_MON_MODE_MASK, |
| 8484 | 0); /* GREATER OR EQUAL*/ |
| 8485 | value |= FIELD_PREP(GAUDI_PKT_SHORT_VAL_MON_MASK_MASK, mask); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8486 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8487 | ctl = FIELD_PREP(GAUDI_PKT_SHORT_CTL_ADDR_MASK, msg_addr_offset); |
Ofir Bitton | 6c07bab | 2020-06-01 10:38:46 +0300 | [diff] [blame] | 8488 | ctl |= FIELD_PREP(GAUDI_PKT_SHORT_CTL_OP_MASK, 0); /* write the value */ |
| 8489 | ctl |= FIELD_PREP(GAUDI_PKT_SHORT_CTL_BASE_MASK, 2); /* W_S MON base */ |
Ofir Bitton | f8b0f2e | 2020-12-06 10:22:32 +0200 | [diff] [blame] | 8490 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_MSG_SHORT); |
| 8491 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 0); |
| 8492 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 8493 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8494 | |
| 8495 | pkt->value = cpu_to_le32(value); |
| 8496 | pkt->ctl = cpu_to_le32(ctl); |
| 8497 | |
| 8498 | return pkt_size; |
| 8499 | } |
| 8500 | |
| 8501 | static u32 gaudi_add_fence_pkt(struct packet_fence *pkt) |
| 8502 | { |
| 8503 | u32 ctl, cfg, pkt_size = sizeof(*pkt); |
| 8504 | |
| 8505 | memset(pkt, 0, pkt_size); |
| 8506 | |
Ofir Bitton | 6c07bab | 2020-06-01 10:38:46 +0300 | [diff] [blame] | 8507 | cfg = FIELD_PREP(GAUDI_PKT_FENCE_CFG_DEC_VAL_MASK, 1); |
| 8508 | cfg |= FIELD_PREP(GAUDI_PKT_FENCE_CFG_TARGET_VAL_MASK, 1); |
| 8509 | cfg |= FIELD_PREP(GAUDI_PKT_FENCE_CFG_ID_MASK, 2); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8510 | |
Ofir Bitton | f8b0f2e | 2020-12-06 10:22:32 +0200 | [diff] [blame] | 8511 | ctl = FIELD_PREP(GAUDI_PKT_CTL_OPCODE_MASK, PACKET_FENCE); |
| 8512 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_EB_MASK, 0); |
| 8513 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_RB_MASK, 1); |
| 8514 | ctl |= FIELD_PREP(GAUDI_PKT_CTL_MB_MASK, 1); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8515 | |
| 8516 | pkt->cfg = cpu_to_le32(cfg); |
| 8517 | pkt->ctl = cpu_to_le32(ctl); |
| 8518 | |
| 8519 | return pkt_size; |
| 8520 | } |
| 8521 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8522 | static int gaudi_get_fence_addr(struct hl_device *hdev, u32 queue_id, u64 *addr) |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8523 | { |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8524 | u32 offset, nic_index; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8525 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8526 | switch (queue_id) { |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8527 | case GAUDI_QUEUE_ID_DMA_0_0: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8528 | offset = mmDMA0_QM_CP_FENCE2_RDATA_0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8529 | break; |
| 8530 | case GAUDI_QUEUE_ID_DMA_0_1: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8531 | offset = mmDMA0_QM_CP_FENCE2_RDATA_1; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8532 | break; |
| 8533 | case GAUDI_QUEUE_ID_DMA_0_2: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8534 | offset = mmDMA0_QM_CP_FENCE2_RDATA_2; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8535 | break; |
| 8536 | case GAUDI_QUEUE_ID_DMA_0_3: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8537 | offset = mmDMA0_QM_CP_FENCE2_RDATA_3; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8538 | break; |
| 8539 | case GAUDI_QUEUE_ID_DMA_1_0: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8540 | offset = mmDMA1_QM_CP_FENCE2_RDATA_0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8541 | break; |
| 8542 | case GAUDI_QUEUE_ID_DMA_1_1: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8543 | offset = mmDMA1_QM_CP_FENCE2_RDATA_1; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8544 | break; |
| 8545 | case GAUDI_QUEUE_ID_DMA_1_2: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8546 | offset = mmDMA1_QM_CP_FENCE2_RDATA_2; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8547 | break; |
| 8548 | case GAUDI_QUEUE_ID_DMA_1_3: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8549 | offset = mmDMA1_QM_CP_FENCE2_RDATA_3; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8550 | break; |
| 8551 | case GAUDI_QUEUE_ID_DMA_5_0: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8552 | offset = mmDMA5_QM_CP_FENCE2_RDATA_0; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8553 | break; |
| 8554 | case GAUDI_QUEUE_ID_DMA_5_1: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8555 | offset = mmDMA5_QM_CP_FENCE2_RDATA_1; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8556 | break; |
| 8557 | case GAUDI_QUEUE_ID_DMA_5_2: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8558 | offset = mmDMA5_QM_CP_FENCE2_RDATA_2; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8559 | break; |
| 8560 | case GAUDI_QUEUE_ID_DMA_5_3: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8561 | offset = mmDMA5_QM_CP_FENCE2_RDATA_3; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8562 | break; |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8563 | case GAUDI_QUEUE_ID_TPC_7_0: |
| 8564 | offset = mmTPC7_QM_CP_FENCE2_RDATA_0; |
| 8565 | break; |
| 8566 | case GAUDI_QUEUE_ID_TPC_7_1: |
| 8567 | offset = mmTPC7_QM_CP_FENCE2_RDATA_1; |
| 8568 | break; |
| 8569 | case GAUDI_QUEUE_ID_TPC_7_2: |
| 8570 | offset = mmTPC7_QM_CP_FENCE2_RDATA_2; |
| 8571 | break; |
| 8572 | case GAUDI_QUEUE_ID_TPC_7_3: |
| 8573 | offset = mmTPC7_QM_CP_FENCE2_RDATA_3; |
| 8574 | break; |
| 8575 | case GAUDI_QUEUE_ID_NIC_0_0: |
| 8576 | case GAUDI_QUEUE_ID_NIC_1_0: |
| 8577 | case GAUDI_QUEUE_ID_NIC_2_0: |
| 8578 | case GAUDI_QUEUE_ID_NIC_3_0: |
| 8579 | case GAUDI_QUEUE_ID_NIC_4_0: |
| 8580 | case GAUDI_QUEUE_ID_NIC_5_0: |
| 8581 | case GAUDI_QUEUE_ID_NIC_6_0: |
| 8582 | case GAUDI_QUEUE_ID_NIC_7_0: |
| 8583 | case GAUDI_QUEUE_ID_NIC_8_0: |
| 8584 | case GAUDI_QUEUE_ID_NIC_9_0: |
| 8585 | nic_index = (queue_id - GAUDI_QUEUE_ID_NIC_0_0) >> 2; |
| 8586 | offset = mmNIC0_QM0_CP_FENCE2_RDATA_0 + |
| 8587 | (nic_index >> 1) * NIC_MACRO_QMAN_OFFSET + |
| 8588 | (nic_index & 0x1) * NIC_ENGINE_QMAN_OFFSET; |
| 8589 | break; |
| 8590 | case GAUDI_QUEUE_ID_NIC_0_1: |
| 8591 | case GAUDI_QUEUE_ID_NIC_1_1: |
| 8592 | case GAUDI_QUEUE_ID_NIC_2_1: |
| 8593 | case GAUDI_QUEUE_ID_NIC_3_1: |
| 8594 | case GAUDI_QUEUE_ID_NIC_4_1: |
| 8595 | case GAUDI_QUEUE_ID_NIC_5_1: |
| 8596 | case GAUDI_QUEUE_ID_NIC_6_1: |
| 8597 | case GAUDI_QUEUE_ID_NIC_7_1: |
| 8598 | case GAUDI_QUEUE_ID_NIC_8_1: |
| 8599 | case GAUDI_QUEUE_ID_NIC_9_1: |
| 8600 | nic_index = (queue_id - GAUDI_QUEUE_ID_NIC_0_1) >> 2; |
| 8601 | offset = mmNIC0_QM0_CP_FENCE2_RDATA_1 + |
| 8602 | (nic_index >> 1) * NIC_MACRO_QMAN_OFFSET + |
| 8603 | (nic_index & 0x1) * NIC_ENGINE_QMAN_OFFSET; |
| 8604 | break; |
| 8605 | case GAUDI_QUEUE_ID_NIC_0_2: |
| 8606 | case GAUDI_QUEUE_ID_NIC_1_2: |
| 8607 | case GAUDI_QUEUE_ID_NIC_2_2: |
| 8608 | case GAUDI_QUEUE_ID_NIC_3_2: |
| 8609 | case GAUDI_QUEUE_ID_NIC_4_2: |
| 8610 | case GAUDI_QUEUE_ID_NIC_5_2: |
| 8611 | case GAUDI_QUEUE_ID_NIC_6_2: |
| 8612 | case GAUDI_QUEUE_ID_NIC_7_2: |
| 8613 | case GAUDI_QUEUE_ID_NIC_8_2: |
| 8614 | case GAUDI_QUEUE_ID_NIC_9_2: |
| 8615 | nic_index = (queue_id - GAUDI_QUEUE_ID_NIC_0_2) >> 2; |
| 8616 | offset = mmNIC0_QM0_CP_FENCE2_RDATA_2 + |
| 8617 | (nic_index >> 1) * NIC_MACRO_QMAN_OFFSET + |
| 8618 | (nic_index & 0x1) * NIC_ENGINE_QMAN_OFFSET; |
| 8619 | break; |
| 8620 | case GAUDI_QUEUE_ID_NIC_0_3: |
| 8621 | case GAUDI_QUEUE_ID_NIC_1_3: |
| 8622 | case GAUDI_QUEUE_ID_NIC_2_3: |
| 8623 | case GAUDI_QUEUE_ID_NIC_3_3: |
| 8624 | case GAUDI_QUEUE_ID_NIC_4_3: |
| 8625 | case GAUDI_QUEUE_ID_NIC_5_3: |
| 8626 | case GAUDI_QUEUE_ID_NIC_6_3: |
| 8627 | case GAUDI_QUEUE_ID_NIC_7_3: |
| 8628 | case GAUDI_QUEUE_ID_NIC_8_3: |
| 8629 | case GAUDI_QUEUE_ID_NIC_9_3: |
| 8630 | nic_index = (queue_id - GAUDI_QUEUE_ID_NIC_0_3) >> 2; |
| 8631 | offset = mmNIC0_QM0_CP_FENCE2_RDATA_3 + |
| 8632 | (nic_index >> 1) * NIC_MACRO_QMAN_OFFSET + |
| 8633 | (nic_index & 0x1) * NIC_ENGINE_QMAN_OFFSET; |
| 8634 | break; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8635 | default: |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8636 | return -EINVAL; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8637 | } |
| 8638 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8639 | *addr = CFG_BASE + offset; |
| 8640 | |
| 8641 | return 0; |
| 8642 | } |
| 8643 | |
| 8644 | static u32 gaudi_add_mon_pkts(void *buf, u16 mon_id, u64 fence_addr) |
| 8645 | { |
| 8646 | u64 monitor_base; |
| 8647 | u32 size = 0; |
| 8648 | u16 msg_addr_offset; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8649 | |
| 8650 | /* |
| 8651 | * monitor_base should be the content of the base0 address registers, |
| 8652 | * so it will be added to the msg short offsets |
| 8653 | */ |
| 8654 | monitor_base = mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0; |
| 8655 | |
| 8656 | /* First monitor config packet: low address of the sync */ |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8657 | msg_addr_offset = |
| 8658 | (mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 + mon_id * 4) - |
| 8659 | monitor_base; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8660 | |
| 8661 | size += gaudi_add_mon_msg_short(buf + size, (u32) fence_addr, |
| 8662 | msg_addr_offset); |
| 8663 | |
| 8664 | /* Second monitor config packet: high address of the sync */ |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8665 | msg_addr_offset = |
| 8666 | (mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRH_0 + mon_id * 4) - |
| 8667 | monitor_base; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8668 | |
| 8669 | size += gaudi_add_mon_msg_short(buf + size, (u32) (fence_addr >> 32), |
| 8670 | msg_addr_offset); |
| 8671 | |
| 8672 | /* |
| 8673 | * Third monitor config packet: the payload, i.e. what to write when the |
| 8674 | * sync triggers |
| 8675 | */ |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8676 | msg_addr_offset = |
| 8677 | (mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_DATA_0 + mon_id * 4) - |
| 8678 | monitor_base; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8679 | |
| 8680 | size += gaudi_add_mon_msg_short(buf + size, 1, msg_addr_offset); |
| 8681 | |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8682 | return size; |
| 8683 | } |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8684 | |
Oded Gabbay | 3c68157 | 2020-11-02 21:10:39 +0200 | [diff] [blame] | 8685 | static u32 gaudi_gen_wait_cb(struct hl_device *hdev, |
| 8686 | struct hl_gen_wait_properties *prop) |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8687 | { |
| 8688 | struct hl_cb *cb = (struct hl_cb *) prop->data; |
| 8689 | void *buf = cb->kernel_address; |
| 8690 | u64 fence_addr = 0; |
| 8691 | u32 size = prop->size; |
| 8692 | |
| 8693 | if (gaudi_get_fence_addr(hdev, prop->q_idx, &fence_addr)) { |
| 8694 | dev_crit(hdev->dev, "wrong queue id %d for wait packet\n", |
| 8695 | prop->q_idx); |
| 8696 | return 0; |
| 8697 | } |
| 8698 | |
| 8699 | size += gaudi_add_mon_pkts(buf + size, prop->mon_id, fence_addr); |
| 8700 | size += gaudi_add_arm_monitor_pkt(hdev, buf + size, prop->sob_base, |
| 8701 | prop->sob_mask, prop->sob_val, prop->mon_id); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8702 | size += gaudi_add_fence_pkt(buf + size); |
Ofir Bitton | 2992c1d | 2020-09-10 09:40:35 +0300 | [diff] [blame] | 8703 | |
| 8704 | return size; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8705 | } |
| 8706 | |
| 8707 | static void gaudi_reset_sob(struct hl_device *hdev, void *data) |
| 8708 | { |
| 8709 | struct hl_hw_sob *hw_sob = (struct hl_hw_sob *) data; |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 8710 | int rc; |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8711 | |
| 8712 | dev_dbg(hdev->dev, "reset SOB, q_idx: %d, sob_id: %d\n", hw_sob->q_idx, |
| 8713 | hw_sob->sob_id); |
| 8714 | |
Ofir Bitton | 423815b | 2021-01-05 09:04:07 +0200 | [diff] [blame] | 8715 | rc = gaudi_schedule_register_memset(hdev, hw_sob->q_idx, |
| 8716 | CFG_BASE + mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0 + |
| 8717 | hw_sob->sob_id * 4, 1, 0); |
| 8718 | if (rc) |
| 8719 | dev_err(hdev->dev, "failed resetting sob %u", hw_sob->sob_id); |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8720 | |
| 8721 | kref_init(&hw_sob->kref); |
| 8722 | } |
| 8723 | |
| 8724 | static void gaudi_set_dma_mask_from_fw(struct hl_device *hdev) |
| 8725 | { |
| 8726 | if (RREG32(mmPSOC_GLOBAL_CONF_NON_RST_FLOPS_0) == |
| 8727 | HL_POWER9_HOST_MAGIC) { |
| 8728 | hdev->power9_64bit_dma_enable = 1; |
| 8729 | hdev->dma_mask = 64; |
| 8730 | } else { |
| 8731 | hdev->power9_64bit_dma_enable = 0; |
| 8732 | hdev->dma_mask = 48; |
| 8733 | } |
| 8734 | } |
| 8735 | |
| 8736 | static u64 gaudi_get_device_time(struct hl_device *hdev) |
| 8737 | { |
| 8738 | u64 device_time = ((u64) RREG32(mmPSOC_TIMESTAMP_CNTCVU)) << 32; |
| 8739 | |
| 8740 | return device_time | RREG32(mmPSOC_TIMESTAMP_CNTCVL); |
| 8741 | } |
| 8742 | |
Ofir Bitton | d00697f | 2021-01-05 12:55:06 +0200 | [diff] [blame] | 8743 | static int gaudi_get_hw_block_id(struct hl_device *hdev, u64 block_addr, |
Oded Gabbay | 6df50d2 | 2021-02-05 16:04:34 +0200 | [diff] [blame] | 8744 | u32 *block_size, u32 *block_id) |
Ofir Bitton | d00697f | 2021-01-05 12:55:06 +0200 | [diff] [blame] | 8745 | { |
| 8746 | return -EPERM; |
| 8747 | } |
| 8748 | |
| 8749 | static int gaudi_block_mmap(struct hl_device *hdev, |
| 8750 | struct vm_area_struct *vma, |
| 8751 | u32 block_id, u32 block_size) |
| 8752 | { |
| 8753 | return -EPERM; |
| 8754 | } |
| 8755 | |
Oded Gabbay | 28bcf1f | 2021-02-01 21:23:43 +0200 | [diff] [blame] | 8756 | static void gaudi_enable_events_from_fw(struct hl_device *hdev) |
| 8757 | { |
| 8758 | WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR, GAUDI_EVENT_INTS_REGISTER); |
| 8759 | } |
| 8760 | |
Bharat Jauhari | 285c0fa | 2021-03-25 18:15:40 +0200 | [diff] [blame] | 8761 | static int gaudi_map_pll_idx_to_fw_idx(u32 pll_idx) |
| 8762 | { |
| 8763 | switch (pll_idx) { |
| 8764 | case HL_GAUDI_CPU_PLL: return CPU_PLL; |
| 8765 | case HL_GAUDI_PCI_PLL: return PCI_PLL; |
| 8766 | case HL_GAUDI_NIC_PLL: return NIC_PLL; |
| 8767 | case HL_GAUDI_DMA_PLL: return DMA_PLL; |
| 8768 | case HL_GAUDI_MESH_PLL: return MESH_PLL; |
| 8769 | case HL_GAUDI_MME_PLL: return MME_PLL; |
| 8770 | case HL_GAUDI_TPC_PLL: return TPC_PLL; |
| 8771 | case HL_GAUDI_IF_PLL: return IF_PLL; |
| 8772 | case HL_GAUDI_SRAM_PLL: return SRAM_PLL; |
| 8773 | case HL_GAUDI_HBM_PLL: return HBM_PLL; |
| 8774 | default: return -EINVAL; |
| 8775 | } |
| 8776 | } |
| 8777 | |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8778 | static const struct hl_asic_funcs gaudi_funcs = { |
| 8779 | .early_init = gaudi_early_init, |
| 8780 | .early_fini = gaudi_early_fini, |
| 8781 | .late_init = gaudi_late_init, |
| 8782 | .late_fini = gaudi_late_fini, |
| 8783 | .sw_init = gaudi_sw_init, |
| 8784 | .sw_fini = gaudi_sw_fini, |
| 8785 | .hw_init = gaudi_hw_init, |
| 8786 | .hw_fini = gaudi_hw_fini, |
| 8787 | .halt_engines = gaudi_halt_engines, |
| 8788 | .suspend = gaudi_suspend, |
| 8789 | .resume = gaudi_resume, |
| 8790 | .cb_mmap = gaudi_cb_mmap, |
| 8791 | .ring_doorbell = gaudi_ring_doorbell, |
| 8792 | .pqe_write = gaudi_pqe_write, |
| 8793 | .asic_dma_alloc_coherent = gaudi_dma_alloc_coherent, |
| 8794 | .asic_dma_free_coherent = gaudi_dma_free_coherent, |
farah kassabri | 03df136 | 2020-05-06 11:17:38 +0300 | [diff] [blame] | 8795 | .scrub_device_mem = gaudi_scrub_device_mem, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8796 | .get_int_queue_base = gaudi_get_int_queue_base, |
| 8797 | .test_queues = gaudi_test_queues, |
| 8798 | .asic_dma_pool_zalloc = gaudi_dma_pool_zalloc, |
| 8799 | .asic_dma_pool_free = gaudi_dma_pool_free, |
| 8800 | .cpu_accessible_dma_pool_alloc = gaudi_cpu_accessible_dma_pool_alloc, |
| 8801 | .cpu_accessible_dma_pool_free = gaudi_cpu_accessible_dma_pool_free, |
| 8802 | .hl_dma_unmap_sg = gaudi_dma_unmap_sg, |
| 8803 | .cs_parser = gaudi_cs_parser, |
| 8804 | .asic_dma_map_sg = gaudi_dma_map_sg, |
| 8805 | .get_dma_desc_list_size = gaudi_get_dma_desc_list_size, |
| 8806 | .add_end_of_cb_packets = gaudi_add_end_of_cb_packets, |
| 8807 | .update_eq_ci = gaudi_update_eq_ci, |
| 8808 | .context_switch = gaudi_context_switch, |
| 8809 | .restore_phase_topology = gaudi_restore_phase_topology, |
| 8810 | .debugfs_read32 = gaudi_debugfs_read32, |
| 8811 | .debugfs_write32 = gaudi_debugfs_write32, |
| 8812 | .debugfs_read64 = gaudi_debugfs_read64, |
| 8813 | .debugfs_write64 = gaudi_debugfs_write64, |
Oded Gabbay | 639781d | 2021-04-02 01:43:18 +0300 | [diff] [blame] | 8814 | .debugfs_read_dma = gaudi_debugfs_read_dma, |
Oded Gabbay | bcaf415 | 2020-05-11 10:41:37 +0300 | [diff] [blame] | 8815 | .add_device_attr = gaudi_add_device_attr, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8816 | .handle_eqe = gaudi_handle_eqe, |
Oded Gabbay | bcaf415 | 2020-05-11 10:41:37 +0300 | [diff] [blame] | 8817 | .set_pll_profile = gaudi_set_pll_profile, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8818 | .get_events_stat = gaudi_get_events_stat, |
| 8819 | .read_pte = gaudi_read_pte, |
| 8820 | .write_pte = gaudi_write_pte, |
| 8821 | .mmu_invalidate_cache = gaudi_mmu_invalidate_cache, |
| 8822 | .mmu_invalidate_cache_range = gaudi_mmu_invalidate_cache_range, |
| 8823 | .send_heartbeat = gaudi_send_heartbeat, |
Oded Gabbay | e38bfd3 | 2020-07-03 20:46:12 +0300 | [diff] [blame] | 8824 | .set_clock_gating = gaudi_set_clock_gating, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8825 | .disable_clock_gating = gaudi_disable_clock_gating, |
Omer Shpigelman | 79fc7a9 | 2020-05-11 10:46:29 +0300 | [diff] [blame] | 8826 | .debug_coresight = gaudi_debug_coresight, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8827 | .is_device_idle = gaudi_is_device_idle, |
| 8828 | .soft_reset_late_init = gaudi_soft_reset_late_init, |
| 8829 | .hw_queues_lock = gaudi_hw_queues_lock, |
| 8830 | .hw_queues_unlock = gaudi_hw_queues_unlock, |
| 8831 | .get_pci_id = gaudi_get_pci_id, |
| 8832 | .get_eeprom_data = gaudi_get_eeprom_data, |
| 8833 | .send_cpu_message = gaudi_send_cpu_message, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8834 | .pci_bars_map = gaudi_pci_bars_map, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8835 | .init_iatu = gaudi_init_iatu, |
| 8836 | .rreg = hl_rreg, |
| 8837 | .wreg = hl_wreg, |
Omer Shpigelman | 79fc7a9 | 2020-05-11 10:46:29 +0300 | [diff] [blame] | 8838 | .halt_coresight = gaudi_halt_coresight, |
Ofir Bitton | a04b7cd | 2020-07-13 13:36:55 +0300 | [diff] [blame] | 8839 | .ctx_init = gaudi_ctx_init, |
Ofir Bitton | 5de406c | 2020-09-10 10:56:26 +0300 | [diff] [blame] | 8840 | .ctx_fini = gaudi_ctx_fini, |
Oded Gabbay | bcaf415 | 2020-05-11 10:41:37 +0300 | [diff] [blame] | 8841 | .get_clk_rate = gaudi_get_clk_rate, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8842 | .get_queue_id_for_cq = gaudi_get_queue_id_for_cq, |
| 8843 | .read_device_fw_version = gaudi_read_device_fw_version, |
| 8844 | .load_firmware_to_device = gaudi_load_firmware_to_device, |
| 8845 | .load_boot_fit_to_device = gaudi_load_boot_fit_to_device, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8846 | .get_signal_cb_size = gaudi_get_signal_cb_size, |
| 8847 | .get_wait_cb_size = gaudi_get_wait_cb_size, |
| 8848 | .gen_signal_cb = gaudi_gen_signal_cb, |
| 8849 | .gen_wait_cb = gaudi_gen_wait_cb, |
| 8850 | .reset_sob = gaudi_reset_sob, |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 8851 | .reset_sob_group = gaudi_reset_sob_group, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8852 | .set_dma_mask_from_fw = gaudi_set_dma_mask_from_fw, |
Ofir Bitton | 5fe1c17 | 2020-09-10 10:10:55 +0300 | [diff] [blame] | 8853 | .get_device_time = gaudi_get_device_time, |
| 8854 | .collective_wait_init_cs = gaudi_collective_wait_init_cs, |
Moti Haimovski | b19dc67 | 2020-11-18 20:15:29 +0200 | [diff] [blame] | 8855 | .collective_wait_create_jobs = gaudi_collective_wait_create_jobs, |
farah kassabri | 89473a1 | 2021-01-12 17:24:00 +0200 | [diff] [blame] | 8856 | .scramble_addr = hl_mmu_scramble_addr, |
| 8857 | .descramble_addr = hl_mmu_descramble_addr, |
Ofir Bitton | d00697f | 2021-01-05 12:55:06 +0200 | [diff] [blame] | 8858 | .ack_protection_bits_errors = gaudi_ack_protection_bits_errors, |
| 8859 | .get_hw_block_id = gaudi_get_hw_block_id, |
Oded Gabbay | 28bcf1f | 2021-02-01 21:23:43 +0200 | [diff] [blame] | 8860 | .hw_block_mmap = gaudi_block_mmap, |
Bharat Jauhari | 285c0fa | 2021-03-25 18:15:40 +0200 | [diff] [blame] | 8861 | .enable_events_from_fw = gaudi_enable_events_from_fw, |
Ohad Sharabi | a22f0ec | 2021-04-11 23:06:46 +0300 | [diff] [blame^] | 8862 | .map_pll_idx_to_fw_idx = gaudi_map_pll_idx_to_fw_idx, |
| 8863 | .init_firmware_loader = gaudi_init_firmware_loader, |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8864 | }; |
| 8865 | |
| 8866 | /** |
| 8867 | * gaudi_set_asic_funcs - set GAUDI function pointers |
| 8868 | * |
Lee Jones | f7d227c | 2020-07-01 09:58:42 +0100 | [diff] [blame] | 8869 | * @hdev: pointer to hl_device structure |
Oded Gabbay | ac0ae6a | 2020-05-11 10:29:27 +0300 | [diff] [blame] | 8870 | * |
| 8871 | */ |
| 8872 | void gaudi_set_asic_funcs(struct hl_device *hdev) |
| 8873 | { |
| 8874 | hdev->asic_funcs = &gaudi_funcs; |
| 8875 | } |