1 /* SPDX-License-Identifier: GPL-2.0 2 * 3 * Copyright 2020-2022 HabanaLabs, Ltd. 4 * All Rights Reserved. 5 * 6 */ 7 8 #ifndef GAUDI2P_H_ 9 #define GAUDI2P_H_ 10 11 #include <uapi/drm/habanalabs_accel.h> 12 #include "../common/habanalabs.h" 13 #include "../include/common/hl_boot_if.h" 14 #include "../include/gaudi2/gaudi2.h" 15 #include "../include/gaudi2/gaudi2_packets.h" 16 #include "../include/gaudi2/gaudi2_fw_if.h" 17 #include "../include/gaudi2/gaudi2_async_events.h" 18 19 #define GAUDI2_LINUX_FW_FILE "habanalabs/gaudi2/gaudi2-fit.itb" 20 #define GAUDI2_BOOT_FIT_FILE "habanalabs/gaudi2/gaudi2-boot-fit.itb" 21 22 #define MMU_PAGE_TABLES_INITIAL_SIZE 0x10000000 /* 256MB */ 23 24 #define GAUDI2_CPU_TIMEOUT_USEC 30000000 /* 30s */ 25 26 #define NUMBER_OF_PDMA_QUEUES 2 27 #define NUMBER_OF_EDMA_QUEUES 8 28 #define NUMBER_OF_MME_QUEUES 4 29 #define NUMBER_OF_TPC_QUEUES 25 30 #define NUMBER_OF_NIC_QUEUES 24 31 #define NUMBER_OF_ROT_QUEUES 2 32 #define NUMBER_OF_CPU_QUEUES 1 33 34 #define NUMBER_OF_HW_QUEUES ((NUMBER_OF_PDMA_QUEUES + \ 35 NUMBER_OF_EDMA_QUEUES + \ 36 NUMBER_OF_MME_QUEUES + \ 37 NUMBER_OF_TPC_QUEUES + \ 38 NUMBER_OF_NIC_QUEUES + \ 39 NUMBER_OF_ROT_QUEUES + \ 40 NUMBER_OF_CPU_QUEUES) * \ 41 NUM_OF_PQ_PER_QMAN) 42 43 #define NUMBER_OF_QUEUES (NUMBER_OF_CPU_QUEUES + NUMBER_OF_HW_QUEUES) 44 45 #define DCORE_NUM_OF_SOB \ 46 (((mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_8191 - \ 47 mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0) + 4) >> 2) 48 49 #define DCORE_NUM_OF_MONITORS \ 50 (((mmDCORE0_SYNC_MNGR_OBJS_MON_STATUS_2047 - \ 51 mmDCORE0_SYNC_MNGR_OBJS_MON_STATUS_0) + 4) >> 2) 52 53 #define NUMBER_OF_DEC ((NUM_OF_DEC_PER_DCORE * NUM_OF_DCORES) + NUMBER_OF_PCIE_DEC) 54 55 /* Map all arcs dccm + arc schedulers acp blocks */ 56 #define NUM_OF_USER_ACP_BLOCKS (NUM_OF_SCHEDULER_ARC + 2) 57 #define NUM_OF_USER_NIC_UMR_BLOCKS 15 58 #define NUM_OF_EXPOSED_SM_BLOCKS ((NUM_OF_DCORES - 1) * 2) 59 #define NUM_USER_MAPPED_BLOCKS \ 60 (NUM_ARC_CPUS + NUM_OF_USER_ACP_BLOCKS + NUMBER_OF_DEC + \ 61 NUM_OF_EXPOSED_SM_BLOCKS + \ 62 (NIC_NUMBER_OF_ENGINES * NUM_OF_USER_NIC_UMR_BLOCKS)) 63 64 /* Within the user mapped array, decoder entries start post all the ARC related 65 * entries 66 */ 67 #define USR_MAPPED_BLK_DEC_START_IDX \ 68 (NUM_ARC_CPUS + NUM_OF_USER_ACP_BLOCKS + \ 69 (NIC_NUMBER_OF_ENGINES * NUM_OF_USER_NIC_UMR_BLOCKS)) 70 71 #define USR_MAPPED_BLK_SM_START_IDX \ 72 (NUM_ARC_CPUS + NUM_OF_USER_ACP_BLOCKS + NUMBER_OF_DEC + \ 73 (NIC_NUMBER_OF_ENGINES * NUM_OF_USER_NIC_UMR_BLOCKS)) 74 75 #define SM_OBJS_BLOCK_SIZE (mmDCORE0_SYNC_MNGR_OBJS_SM_SEC_0 - \ 76 mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0) 77 78 #define GAUDI2_MAX_PENDING_CS 64 79 80 #if !IS_MAX_PENDING_CS_VALID(GAUDI2_MAX_PENDING_CS) 81 #error "GAUDI2_MAX_PENDING_CS must be power of 2 and greater than 1" 82 #endif 83 84 #define CORESIGHT_TIMEOUT_USEC 100000 /* 100 ms */ 85 86 #define GAUDI2_PREBOOT_REQ_TIMEOUT_USEC 25000000 /* 25s */ 87 88 #define GAUDI2_BOOT_FIT_REQ_TIMEOUT_USEC 10000000 /* 10s */ 89 90 #define GAUDI2_NIC_CLK_FREQ 450000000ull /* 450 MHz */ 91 92 #define DC_POWER_DEFAULT 60000 /* 60W */ 93 94 #define GAUDI2_HBM_NUM 6 95 96 #define DMA_MAX_TRANSFER_SIZE U32_MAX 97 98 #define GAUDI2_DEFAULT_CARD_NAME "HL225" 99 100 #define QMAN_STREAMS 4 101 102 #define NUM_OF_MME_SBTE_PORTS 5 103 #define NUM_OF_MME_WB_PORTS 2 104 105 #define GAUDI2_ENGINE_ID_DCORE_OFFSET \ 106 (GAUDI2_DCORE1_ENGINE_ID_EDMA_0 - GAUDI2_DCORE0_ENGINE_ID_EDMA_0) 107 108 /* DRAM Memory Map */ 109 110 #define CPU_FW_IMAGE_SIZE 0x10000000 /* 256MB */ 111 112 /* This define should be used only when working in a debug mode without dram. 113 * When working with dram, the driver size will be calculated dynamically. 114 */ 115 #define NIC_DEFAULT_DRV_SIZE 0x20000000 /* 512MB */ 116 117 #define CPU_FW_IMAGE_ADDR DRAM_PHYS_BASE 118 119 #define NIC_NUMBER_OF_PORTS NIC_NUMBER_OF_ENGINES 120 121 #define NUMBER_OF_PCIE_DEC 2 122 #define PCIE_DEC_SHIFT 8 123 124 #define SRAM_USER_BASE_OFFSET 0 125 126 /* cluster binning */ 127 #define MAX_FAULTY_HBMS 1 128 #define GAUDI2_XBAR_EDGE_FULL_MASK 0xF 129 #define GAUDI2_EDMA_FULL_MASK 0xFF 130 #define GAUDI2_DRAM_FULL_MASK 0x3F 131 132 /* Host virtual address space. */ 133 134 #define VA_HOST_SPACE_PAGE_START 0xFFF0000000000000ull 135 #define VA_HOST_SPACE_PAGE_END 0xFFF0800000000000ull /* 140TB */ 136 137 #define VA_HOST_SPACE_HPAGE_START 0xFFF0800000000000ull 138 #define VA_HOST_SPACE_HPAGE_END 0xFFF1000000000000ull /* 140TB */ 139 140 /* 140TB */ 141 #define VA_HOST_SPACE_PAGE_SIZE (VA_HOST_SPACE_PAGE_END - VA_HOST_SPACE_PAGE_START) 142 143 /* 140TB */ 144 #define VA_HOST_SPACE_HPAGE_SIZE (VA_HOST_SPACE_HPAGE_END - VA_HOST_SPACE_HPAGE_START) 145 146 #define VA_HOST_SPACE_SIZE (VA_HOST_SPACE_PAGE_SIZE + VA_HOST_SPACE_HPAGE_SIZE) 147 148 #define HOST_SPACE_INTERNAL_CB_SZ SZ_2M 149 150 /* 151 * HBM virtual address space 152 * Gaudi2 has 6 HBM devices, each supporting 16GB total of 96GB at most. 153 * No core separation is supported so we can have one chunk of virtual address 154 * space just above the physical ones. 155 * The virtual address space starts immediately after the end of the physical 156 * address space which is determined at run-time. 157 */ 158 #define VA_HBM_SPACE_END 0x1002000000000000ull 159 160 #define HW_CAP_PLL BIT_ULL(0) 161 #define HW_CAP_DRAM BIT_ULL(1) 162 #define HW_CAP_PMMU BIT_ULL(2) 163 #define HW_CAP_CPU BIT_ULL(3) 164 #define HW_CAP_MSIX BIT_ULL(4) 165 166 #define HW_CAP_CPU_Q BIT_ULL(5) 167 #define HW_CAP_CPU_Q_SHIFT 5 168 169 #define HW_CAP_CLK_GATE BIT_ULL(6) 170 #define HW_CAP_KDMA BIT_ULL(7) 171 #define HW_CAP_SRAM_SCRAMBLER BIT_ULL(8) 172 173 #define HW_CAP_DCORE0_DMMU0 BIT_ULL(9) 174 #define HW_CAP_DCORE0_DMMU1 BIT_ULL(10) 175 #define HW_CAP_DCORE0_DMMU2 BIT_ULL(11) 176 #define HW_CAP_DCORE0_DMMU3 BIT_ULL(12) 177 #define HW_CAP_DCORE1_DMMU0 BIT_ULL(13) 178 #define HW_CAP_DCORE1_DMMU1 BIT_ULL(14) 179 #define HW_CAP_DCORE1_DMMU2 BIT_ULL(15) 180 #define HW_CAP_DCORE1_DMMU3 BIT_ULL(16) 181 #define HW_CAP_DCORE2_DMMU0 BIT_ULL(17) 182 #define HW_CAP_DCORE2_DMMU1 BIT_ULL(18) 183 #define HW_CAP_DCORE2_DMMU2 BIT_ULL(19) 184 #define HW_CAP_DCORE2_DMMU3 BIT_ULL(20) 185 #define HW_CAP_DCORE3_DMMU0 BIT_ULL(21) 186 #define HW_CAP_DCORE3_DMMU1 BIT_ULL(22) 187 #define HW_CAP_DCORE3_DMMU2 BIT_ULL(23) 188 #define HW_CAP_DCORE3_DMMU3 BIT_ULL(24) 189 #define HW_CAP_DMMU_MASK GENMASK_ULL(24, 9) 190 #define HW_CAP_DMMU_SHIFT 9 191 #define HW_CAP_PDMA_MASK BIT_ULL(26) 192 #define HW_CAP_EDMA_MASK GENMASK_ULL(34, 27) 193 #define HW_CAP_EDMA_SHIFT 27 194 #define HW_CAP_MME_MASK GENMASK_ULL(38, 35) 195 #define HW_CAP_MME_SHIFT 35 196 #define HW_CAP_ROT_MASK GENMASK_ULL(40, 39) 197 #define HW_CAP_ROT_SHIFT 39 198 #define HW_CAP_HBM_SCRAMBLER_HW_RESET BIT_ULL(41) 199 #define HW_CAP_HBM_SCRAMBLER_SW_RESET BIT_ULL(42) 200 #define HW_CAP_HBM_SCRAMBLER_MASK (HW_CAP_HBM_SCRAMBLER_HW_RESET | \ 201 HW_CAP_HBM_SCRAMBLER_SW_RESET) 202 #define HW_CAP_HBM_SCRAMBLER_SHIFT 41 203 #define HW_CAP_RESERVED BIT(43) 204 #define HW_CAP_MMU_MASK (HW_CAP_PMMU | HW_CAP_DMMU_MASK) 205 206 /* Range Registers */ 207 #define RR_TYPE_SHORT 0 208 #define RR_TYPE_LONG 1 209 #define RR_TYPE_SHORT_PRIV 2 210 #define RR_TYPE_LONG_PRIV 3 211 #define NUM_SHORT_LBW_RR 14 212 #define NUM_LONG_LBW_RR 4 213 #define NUM_SHORT_HBW_RR 6 214 #define NUM_LONG_HBW_RR 4 215 216 /* RAZWI initiator coordinates- X- 5 bits, Y- 4 bits */ 217 #define RAZWI_INITIATOR_X_SHIFT 0 218 #define RAZWI_INITIATOR_X_MASK 0x1F 219 #define RAZWI_INITIATOR_Y_SHIFT 5 220 #define RAZWI_INITIATOR_Y_MASK 0xF 221 222 #define RTR_ID_X_Y(x, y) \ 223 ((((y) & RAZWI_INITIATOR_Y_MASK) << RAZWI_INITIATOR_Y_SHIFT) | \ 224 (((x) & RAZWI_INITIATOR_X_MASK) << RAZWI_INITIATOR_X_SHIFT)) 225 226 /* decoders have separate mask */ 227 #define HW_CAP_DEC_SHIFT 0 228 #define HW_CAP_DEC_MASK GENMASK_ULL(9, 0) 229 230 /* TPCs have separate mask */ 231 #define HW_CAP_TPC_SHIFT 0 232 #define HW_CAP_TPC_MASK GENMASK_ULL(24, 0) 233 234 /* nics have separate mask */ 235 #define HW_CAP_NIC_SHIFT 0 236 #define HW_CAP_NIC_MASK GENMASK_ULL(NIC_NUMBER_OF_ENGINES - 1, 0) 237 238 #define GAUDI2_ARC_PCI_MSB_ADDR(addr) (((addr) & GENMASK_ULL(49, 28)) >> 28) 239 240 #define GAUDI2_SOB_INCREMENT_BY_ONE (FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_VAL_MASK, 1) | \ 241 FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_INC_MASK, 1)) 242 243 #define GAUDI2_NUM_TESTED_QS (GAUDI2_QUEUE_ID_CPU_PQ - GAUDI2_QUEUE_ID_PDMA_0_0) 244 245 #define GAUDI2_NUM_OF_GLBL_ERR_CAUSE 8 246 247 enum gaudi2_reserved_sob_id { 248 GAUDI2_RESERVED_SOB_CS_COMPLETION_FIRST, 249 GAUDI2_RESERVED_SOB_CS_COMPLETION_LAST = 250 GAUDI2_RESERVED_SOB_CS_COMPLETION_FIRST + GAUDI2_MAX_PENDING_CS - 1, 251 GAUDI2_RESERVED_SOB_KDMA_COMPLETION, 252 GAUDI2_RESERVED_SOB_DEC_NRM_FIRST, 253 GAUDI2_RESERVED_SOB_DEC_NRM_LAST = 254 GAUDI2_RESERVED_SOB_DEC_NRM_FIRST + NUMBER_OF_DEC - 1, 255 GAUDI2_RESERVED_SOB_DEC_ABNRM_FIRST, 256 GAUDI2_RESERVED_SOB_DEC_ABNRM_LAST = 257 GAUDI2_RESERVED_SOB_DEC_ABNRM_FIRST + NUMBER_OF_DEC - 1, 258 GAUDI2_RESERVED_SOB_NUMBER 259 }; 260 261 enum gaudi2_reserved_mon_id { 262 GAUDI2_RESERVED_MON_CS_COMPLETION_FIRST, 263 GAUDI2_RESERVED_MON_CS_COMPLETION_LAST = 264 GAUDI2_RESERVED_MON_CS_COMPLETION_FIRST + GAUDI2_MAX_PENDING_CS - 1, 265 GAUDI2_RESERVED_MON_KDMA_COMPLETION, 266 GAUDI2_RESERVED_MON_DEC_NRM_FIRST, 267 GAUDI2_RESERVED_MON_DEC_NRM_LAST = 268 GAUDI2_RESERVED_MON_DEC_NRM_FIRST + 3 * NUMBER_OF_DEC - 1, 269 GAUDI2_RESERVED_MON_DEC_ABNRM_FIRST, 270 GAUDI2_RESERVED_MON_DEC_ABNRM_LAST = 271 GAUDI2_RESERVED_MON_DEC_ABNRM_FIRST + 3 * NUMBER_OF_DEC - 1, 272 GAUDI2_RESERVED_MON_NUMBER 273 }; 274 275 enum gaudi2_reserved_cq_id { 276 GAUDI2_RESERVED_CQ_CS_COMPLETION, 277 GAUDI2_RESERVED_CQ_KDMA_COMPLETION, 278 GAUDI2_RESERVED_CQ_NUMBER 279 }; 280 281 /* 282 * Gaudi2 subtitute TPCs Numbering 283 * At most- two faulty TPCs are allowed 284 * First replacement to a faulty TPC will be TPC24, second- TPC23 285 */ 286 enum substitude_tpc { 287 FAULTY_TPC_SUBTS_1_TPC_24, 288 FAULTY_TPC_SUBTS_2_TPC_23, 289 MAX_FAULTY_TPCS 290 }; 291 292 enum gaudi2_dma_core_id { 293 DMA_CORE_ID_PDMA0, /* Dcore 0 */ 294 DMA_CORE_ID_PDMA1, /* Dcore 0 */ 295 DMA_CORE_ID_EDMA0, /* Dcore 0 */ 296 DMA_CORE_ID_EDMA1, /* Dcore 0 */ 297 DMA_CORE_ID_EDMA2, /* Dcore 1 */ 298 DMA_CORE_ID_EDMA3, /* Dcore 1 */ 299 DMA_CORE_ID_EDMA4, /* Dcore 2 */ 300 DMA_CORE_ID_EDMA5, /* Dcore 2 */ 301 DMA_CORE_ID_EDMA6, /* Dcore 3 */ 302 DMA_CORE_ID_EDMA7, /* Dcore 3 */ 303 DMA_CORE_ID_KDMA, /* Dcore 0 */ 304 DMA_CORE_ID_SIZE 305 }; 306 307 enum gaudi2_rotator_id { 308 ROTATOR_ID_0, 309 ROTATOR_ID_1, 310 ROTATOR_ID_SIZE, 311 }; 312 313 enum gaudi2_mme_id { 314 MME_ID_DCORE0, 315 MME_ID_DCORE1, 316 MME_ID_DCORE2, 317 MME_ID_DCORE3, 318 MME_ID_SIZE, 319 }; 320 321 enum gaudi2_tpc_id { 322 TPC_ID_DCORE0_TPC0, 323 TPC_ID_DCORE0_TPC1, 324 TPC_ID_DCORE0_TPC2, 325 TPC_ID_DCORE0_TPC3, 326 TPC_ID_DCORE0_TPC4, 327 TPC_ID_DCORE0_TPC5, 328 TPC_ID_DCORE1_TPC0, 329 TPC_ID_DCORE1_TPC1, 330 TPC_ID_DCORE1_TPC2, 331 TPC_ID_DCORE1_TPC3, 332 TPC_ID_DCORE1_TPC4, 333 TPC_ID_DCORE1_TPC5, 334 TPC_ID_DCORE2_TPC0, 335 TPC_ID_DCORE2_TPC1, 336 TPC_ID_DCORE2_TPC2, 337 TPC_ID_DCORE2_TPC3, 338 TPC_ID_DCORE2_TPC4, 339 TPC_ID_DCORE2_TPC5, 340 TPC_ID_DCORE3_TPC0, 341 TPC_ID_DCORE3_TPC1, 342 TPC_ID_DCORE3_TPC2, 343 TPC_ID_DCORE3_TPC3, 344 TPC_ID_DCORE3_TPC4, 345 TPC_ID_DCORE3_TPC5, 346 /* the PCI TPC is placed last (mapped liked HW) */ 347 TPC_ID_DCORE0_TPC6, 348 TPC_ID_SIZE, 349 }; 350 351 enum gaudi2_dec_id { 352 DEC_ID_DCORE0_DEC0, 353 DEC_ID_DCORE0_DEC1, 354 DEC_ID_DCORE1_DEC0, 355 DEC_ID_DCORE1_DEC1, 356 DEC_ID_DCORE2_DEC0, 357 DEC_ID_DCORE2_DEC1, 358 DEC_ID_DCORE3_DEC0, 359 DEC_ID_DCORE3_DEC1, 360 DEC_ID_PCIE_VDEC0, 361 DEC_ID_PCIE_VDEC1, 362 DEC_ID_SIZE, 363 }; 364 365 enum gaudi2_hbm_id { 366 HBM_ID0, 367 HBM_ID1, 368 HBM_ID2, 369 HBM_ID3, 370 HBM_ID4, 371 HBM_ID5, 372 HBM_ID_SIZE, 373 }; 374 375 /* specific EDMA enumeration */ 376 enum gaudi2_edma_id { 377 EDMA_ID_DCORE0_INSTANCE0, 378 EDMA_ID_DCORE0_INSTANCE1, 379 EDMA_ID_DCORE1_INSTANCE0, 380 EDMA_ID_DCORE1_INSTANCE1, 381 EDMA_ID_DCORE2_INSTANCE0, 382 EDMA_ID_DCORE2_INSTANCE1, 383 EDMA_ID_DCORE3_INSTANCE0, 384 EDMA_ID_DCORE3_INSTANCE1, 385 EDMA_ID_SIZE, 386 }; 387 388 /* User interrupt count is aligned with HW CQ count. 389 * We have 64 CQ's per dcore, CQ0 in dcore 0 is reserved for legacy mode 390 */ 391 #define GAUDI2_NUM_USER_INTERRUPTS 255 392 #define GAUDI2_NUM_RESERVED_INTERRUPTS 1 393 #define GAUDI2_TOTAL_USER_INTERRUPTS (GAUDI2_NUM_USER_INTERRUPTS + GAUDI2_NUM_RESERVED_INTERRUPTS) 394 395 enum gaudi2_irq_num { 396 GAUDI2_IRQ_NUM_EVENT_QUEUE = GAUDI2_EVENT_QUEUE_MSIX_IDX, 397 GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM, 398 GAUDI2_IRQ_NUM_DCORE0_DEC0_ABNRM, 399 GAUDI2_IRQ_NUM_DCORE0_DEC1_NRM, 400 GAUDI2_IRQ_NUM_DCORE0_DEC1_ABNRM, 401 GAUDI2_IRQ_NUM_DCORE1_DEC0_NRM, 402 GAUDI2_IRQ_NUM_DCORE1_DEC0_ABNRM, 403 GAUDI2_IRQ_NUM_DCORE1_DEC1_NRM, 404 GAUDI2_IRQ_NUM_DCORE1_DEC1_ABNRM, 405 GAUDI2_IRQ_NUM_DCORE2_DEC0_NRM, 406 GAUDI2_IRQ_NUM_DCORE2_DEC0_ABNRM, 407 GAUDI2_IRQ_NUM_DCORE2_DEC1_NRM, 408 GAUDI2_IRQ_NUM_DCORE2_DEC1_ABNRM, 409 GAUDI2_IRQ_NUM_DCORE3_DEC0_NRM, 410 GAUDI2_IRQ_NUM_DCORE3_DEC0_ABNRM, 411 GAUDI2_IRQ_NUM_DCORE3_DEC1_NRM, 412 GAUDI2_IRQ_NUM_DCORE3_DEC1_ABNRM, 413 GAUDI2_IRQ_NUM_SHARED_DEC0_NRM, 414 GAUDI2_IRQ_NUM_SHARED_DEC0_ABNRM, 415 GAUDI2_IRQ_NUM_SHARED_DEC1_NRM, 416 GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM, 417 GAUDI2_IRQ_NUM_DEC_LAST = GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM, 418 GAUDI2_IRQ_NUM_COMPLETION, 419 GAUDI2_IRQ_NUM_NIC_PORT_FIRST, 420 GAUDI2_IRQ_NUM_NIC_PORT_LAST = (GAUDI2_IRQ_NUM_NIC_PORT_FIRST + NIC_NUMBER_OF_PORTS - 1), 421 GAUDI2_IRQ_NUM_TPC_ASSERT, 422 GAUDI2_IRQ_NUM_RESERVED_FIRST, 423 GAUDI2_IRQ_NUM_RESERVED_LAST = (GAUDI2_MSIX_ENTRIES - GAUDI2_TOTAL_USER_INTERRUPTS - 1), 424 GAUDI2_IRQ_NUM_UNEXPECTED_ERROR = RESERVED_MSIX_UNEXPECTED_USER_ERROR_INTERRUPT, 425 GAUDI2_IRQ_NUM_USER_FIRST = GAUDI2_IRQ_NUM_UNEXPECTED_ERROR + 1, 426 GAUDI2_IRQ_NUM_USER_LAST = (GAUDI2_IRQ_NUM_USER_FIRST + GAUDI2_NUM_USER_INTERRUPTS - 1), 427 GAUDI2_IRQ_NUM_LAST = (GAUDI2_MSIX_ENTRIES - 1) 428 }; 429 430 static_assert(GAUDI2_IRQ_NUM_USER_FIRST > GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM); 431 432 /** 433 * struct dup_block_ctx - context to initialize unit instances across multiple 434 * blocks where block can be either a dcore of duplicated 435 * common module. this code relies on constant offsets 436 * of blocks and unit instances in a block. 437 * @instance_cfg_fn: instance specific configuration function. 438 * @data: private configuration data. 439 * @base: base address of the first instance in the first block. 440 * @block_off: subsequent blocks address spacing. 441 * @instance_off: subsequent block's instances address spacing. 442 * @enabled_mask: mask of enabled instances (1- enabled, 0- disabled). 443 * @blocks: number of blocks. 444 * @instances: unit instances per block. 445 */ 446 struct dup_block_ctx { 447 void (*instance_cfg_fn)(struct hl_device *hdev, u64 base, void *data); 448 void *data; 449 u64 base; 450 u64 block_off; 451 u64 instance_off; 452 u64 enabled_mask; 453 unsigned int blocks; 454 unsigned int instances; 455 }; 456 457 /** 458 * struct gaudi2_queues_test_info - Holds the address of a the messages used for testing the 459 * device queues. 460 * @dma_addr: the address used by the HW for accessing the message. 461 * @kern_addr: The address used by the driver for accessing the message. 462 */ 463 struct gaudi2_queues_test_info { 464 dma_addr_t dma_addr; 465 void *kern_addr; 466 }; 467 468 /** 469 * struct gaudi2_device - ASIC specific manage structure. 470 * @cpucp_info_get: get information on device from CPU-CP 471 * @mapped_blocks: array that holds the base address and size of all blocks 472 * the user can map. 473 * @lfsr_rand_seeds: array of MME ACC random seeds to set. 474 * @hw_queues_lock: protects the H/W queues from concurrent access. 475 * @scratchpad_kernel_address: general purpose PAGE_SIZE contiguous memory, 476 * this memory region should be write-only. 477 * currently used for HBW QMAN writes which is 478 * redundant. 479 * @scratchpad_bus_address: scratchpad bus address 480 * @virt_msix_db_cpu_addr: host memory page for the virtual MSI-X doorbell. 481 * @virt_msix_db_dma_addr: bus address of the page for the virtual MSI-X doorbell. 482 * @dram_bar_cur_addr: current address of DRAM PCI bar. 483 * @hw_cap_initialized: This field contains a bit per H/W engine. When that 484 * engine is initialized, that bit is set by the driver to 485 * signal we can use this engine in later code paths. 486 * Each bit is cleared upon reset of its corresponding H/W 487 * engine. 488 * @active_hw_arc: This field contains a bit per ARC of an H/W engine with 489 * exception of TPC and NIC engines. Once an engine arc is 490 * initialized, its respective bit is set. Driver can uniquely 491 * identify each initialized ARC and use this information in 492 * later code paths. Each respective bit is cleared upon reset 493 * of its corresponding ARC of the H/W engine. 494 * @dec_hw_cap_initialized: This field contains a bit per decoder H/W engine. 495 * When that engine is initialized, that bit is set by 496 * the driver to signal we can use this engine in later 497 * code paths. 498 * Each bit is cleared upon reset of its corresponding H/W 499 * engine. 500 * @tpc_hw_cap_initialized: This field contains a bit per TPC H/W engine. 501 * When that engine is initialized, that bit is set by 502 * the driver to signal we can use this engine in later 503 * code paths. 504 * Each bit is cleared upon reset of its corresponding H/W 505 * engine. 506 * @active_tpc_arc: This field contains a bit per ARC of the TPC engines. 507 * Once an engine arc is initialized, its respective bit is 508 * set. Each respective bit is cleared upon reset of its 509 * corresponding ARC of the TPC engine. 510 * @nic_hw_cap_initialized: This field contains a bit per nic H/W engine. 511 * @active_nic_arc: This field contains a bit per ARC of the NIC engines. 512 * Once an engine arc is initialized, its respective bit is 513 * set. Each respective bit is cleared upon reset of its 514 * corresponding ARC of the NIC engine. 515 * @hw_events: array that holds all H/W events that are defined valid. 516 * @events_stat: array that holds histogram of all received events. 517 * @events_stat_aggregate: same as events_stat but doesn't get cleared on reset. 518 * @num_of_valid_hw_events: used to hold the number of valid H/W events. 519 * @nic_ports: array that holds all NIC ports manage structures. 520 * @nic_macros: array that holds all NIC macro manage structures. 521 * @core_info: core info to be used by the Ethernet driver. 522 * @aux_ops: functions for core <-> aux drivers communication. 523 * @flush_db_fifo: flag to force flush DB FIFO after a write. 524 * @hbm_cfg: HBM subsystem settings 525 * @hw_queues_lock_mutex: used by simulator instead of hw_queues_lock. 526 * @queues_test_info: information used by the driver when testing the HW queues. 527 */ 528 struct gaudi2_device { 529 int (*cpucp_info_get)(struct hl_device *hdev); 530 531 struct user_mapped_block mapped_blocks[NUM_USER_MAPPED_BLOCKS]; 532 int lfsr_rand_seeds[MME_NUM_OF_LFSR_SEEDS]; 533 534 spinlock_t hw_queues_lock; 535 536 void *scratchpad_kernel_address; 537 dma_addr_t scratchpad_bus_address; 538 539 void *virt_msix_db_cpu_addr; 540 dma_addr_t virt_msix_db_dma_addr; 541 542 u64 dram_bar_cur_addr; 543 u64 hw_cap_initialized; 544 u64 active_hw_arc; 545 u64 dec_hw_cap_initialized; 546 u64 tpc_hw_cap_initialized; 547 u64 active_tpc_arc; 548 u64 nic_hw_cap_initialized; 549 u64 active_nic_arc; 550 u32 hw_events[GAUDI2_EVENT_SIZE]; 551 u32 events_stat[GAUDI2_EVENT_SIZE]; 552 u32 events_stat_aggregate[GAUDI2_EVENT_SIZE]; 553 u32 num_of_valid_hw_events; 554 555 /* Queue testing */ 556 struct gaudi2_queues_test_info queues_test_info[GAUDI2_NUM_TESTED_QS]; 557 }; 558 559 /* 560 * Types of the Gaudi2 IP blocks, used by special blocks iterator. 561 * Required for scenarios where only particular block types can be 562 * addressed (e.g., special PLDM images). 563 */ 564 enum gaudi2_block_types { 565 GAUDI2_BLOCK_TYPE_PLL, 566 GAUDI2_BLOCK_TYPE_RTR, 567 GAUDI2_BLOCK_TYPE_CPU, 568 GAUDI2_BLOCK_TYPE_HIF, 569 GAUDI2_BLOCK_TYPE_HBM, 570 GAUDI2_BLOCK_TYPE_NIC, 571 GAUDI2_BLOCK_TYPE_PCIE, 572 GAUDI2_BLOCK_TYPE_PCIE_PMA, 573 GAUDI2_BLOCK_TYPE_PDMA, 574 GAUDI2_BLOCK_TYPE_EDMA, 575 GAUDI2_BLOCK_TYPE_PMMU, 576 GAUDI2_BLOCK_TYPE_PSOC, 577 GAUDI2_BLOCK_TYPE_ROT, 578 GAUDI2_BLOCK_TYPE_ARC_FARM, 579 GAUDI2_BLOCK_TYPE_DEC, 580 GAUDI2_BLOCK_TYPE_MME, 581 GAUDI2_BLOCK_TYPE_EU_BIST, 582 GAUDI2_BLOCK_TYPE_SYNC_MNGR, 583 GAUDI2_BLOCK_TYPE_STLB, 584 GAUDI2_BLOCK_TYPE_TPC, 585 GAUDI2_BLOCK_TYPE_HMMU, 586 GAUDI2_BLOCK_TYPE_SRAM, 587 GAUDI2_BLOCK_TYPE_XBAR, 588 GAUDI2_BLOCK_TYPE_KDMA, 589 GAUDI2_BLOCK_TYPE_XDMA, 590 GAUDI2_BLOCK_TYPE_XFT, 591 GAUDI2_BLOCK_TYPE_MAX 592 }; 593 594 extern const u32 gaudi2_dma_core_blocks_bases[DMA_CORE_ID_SIZE]; 595 extern const u32 gaudi2_qm_blocks_bases[GAUDI2_QUEUE_ID_SIZE]; 596 extern const u32 gaudi2_mme_acc_blocks_bases[MME_ID_SIZE]; 597 extern const u32 gaudi2_mme_ctrl_lo_blocks_bases[MME_ID_SIZE]; 598 extern const u32 edma_stream_base[NUM_OF_EDMA_PER_DCORE * NUM_OF_DCORES]; 599 extern const u32 gaudi2_rot_blocks_bases[ROTATOR_ID_SIZE]; 600 601 void gaudi2_iterate_tpcs(struct hl_device *hdev, struct iterate_module_ctx *ctx); 602 int gaudi2_coresight_init(struct hl_device *hdev); 603 int gaudi2_debug_coresight(struct hl_device *hdev, struct hl_ctx *ctx, void *data); 604 void gaudi2_halt_coresight(struct hl_device *hdev, struct hl_ctx *ctx); 605 void gaudi2_init_blocks(struct hl_device *hdev, struct dup_block_ctx *cfg_ctx); 606 bool gaudi2_is_hmmu_enabled(struct hl_device *hdev, int dcore_id, int hmmu_id); 607 void gaudi2_write_rr_to_all_lbw_rtrs(struct hl_device *hdev, u8 rr_type, u32 rr_index, u64 min_val, 608 u64 max_val); 609 void gaudi2_pb_print_security_errors(struct hl_device *hdev, u32 block_addr, u32 cause, 610 u32 offended_addr); 611 int gaudi2_init_security(struct hl_device *hdev); 612 void gaudi2_ack_protection_bits_errors(struct hl_device *hdev); 613 int gaudi2_send_device_activity(struct hl_device *hdev, bool open); 614 615 #endif /* GAUDI2P_H_ */ 616