1 /* SPDX-License-Identifier: GPL-2.0 2 * 3 * Copyright 2019-2022 HabanaLabs, Ltd. 4 * All Rights Reserved. 5 * 6 */ 7 8 #ifndef GAUDIP_H_ 9 #define GAUDIP_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/gaudi/gaudi_packets.h" 15 #include "../include/gaudi/gaudi.h" 16 #include "../include/gaudi/gaudi_async_events.h" 17 #include "../include/gaudi/gaudi_fw_if.h" 18 19 #define NUMBER_OF_EXT_HW_QUEUES 8 20 #define NUMBER_OF_CMPLT_QUEUES NUMBER_OF_EXT_HW_QUEUES 21 #define NUMBER_OF_CPU_HW_QUEUES 1 22 #define NUMBER_OF_INT_HW_QUEUES 100 23 #define NUMBER_OF_HW_QUEUES (NUMBER_OF_EXT_HW_QUEUES + \ 24 NUMBER_OF_CPU_HW_QUEUES + \ 25 NUMBER_OF_INT_HW_QUEUES) 26 27 /* 10 NIC QMANs, DMA5 QMAN, TPC7 QMAN */ 28 #define NUMBER_OF_COLLECTIVE_QUEUES 12 29 #define NUMBER_OF_SOBS_IN_GRP 11 30 31 #define GAUDI_STREAM_MASTER_ARR_SIZE 8 32 33 #define CORESIGHT_TIMEOUT_USEC 100000 /* 100 ms */ 34 35 #define GAUDI_MAX_CLK_FREQ 2200000000ull /* 2200 MHz */ 36 37 #define MAX_POWER_DEFAULT_PCI 200000 /* 200W */ 38 #define MAX_POWER_DEFAULT_PMC 350000 /* 350W */ 39 40 #define DC_POWER_DEFAULT_PCI 60000 /* 60W */ 41 #define DC_POWER_DEFAULT_PMC 60000 /* 60W */ 42 43 #define DC_POWER_DEFAULT_PMC_SEC 97000 /* 97W */ 44 45 #define GAUDI_CPU_TIMEOUT_USEC 30000000 /* 30s */ 46 47 #define TPC_ENABLED_MASK 0xFF 48 49 #define GAUDI_HBM_SIZE_32GB 0x800000000ull 50 #define GAUDI_HBM_DEVICES 4 51 #define GAUDI_HBM_CHANNELS 8 52 #define GAUDI_HBM_CFG_BASE (mmHBM0_BASE - CFG_BASE) 53 #define GAUDI_HBM_CFG_OFFSET (mmHBM1_BASE - mmHBM0_BASE) 54 55 #define DMA_MAX_TRANSFER_SIZE U32_MAX 56 57 #define GAUDI_DEFAULT_CARD_NAME "HL205" 58 59 #define GAUDI_MAX_PENDING_CS SZ_16K 60 61 #if !IS_MAX_PENDING_CS_VALID(GAUDI_MAX_PENDING_CS) 62 #error "GAUDI_MAX_PENDING_CS must be power of 2 and greater than 1" 63 #endif 64 65 #define PCI_DMA_NUMBER_OF_CHNLS 2 66 #define HBM_DMA_NUMBER_OF_CHNLS 6 67 #define DMA_NUMBER_OF_CHNLS (PCI_DMA_NUMBER_OF_CHNLS + \ 68 HBM_DMA_NUMBER_OF_CHNLS) 69 70 #define MME_NUMBER_OF_SLAVE_ENGINES 2 71 #define MME_NUMBER_OF_ENGINES (MME_NUMBER_OF_MASTER_ENGINES + \ 72 MME_NUMBER_OF_SLAVE_ENGINES) 73 #define MME_NUMBER_OF_QMANS (MME_NUMBER_OF_MASTER_ENGINES * \ 74 QMAN_STREAMS) 75 76 #define QMAN_STREAMS 4 77 #define PQ_FETCHER_CACHE_SIZE 8 78 79 #define DMA_QMAN_OFFSET (mmDMA1_QM_BASE - mmDMA0_QM_BASE) 80 #define TPC_QMAN_OFFSET (mmTPC1_QM_BASE - mmTPC0_QM_BASE) 81 #define MME_QMAN_OFFSET (mmMME1_QM_BASE - mmMME0_QM_BASE) 82 #define NIC_MACRO_QMAN_OFFSET (mmNIC1_QM0_BASE - mmNIC0_QM0_BASE) 83 #define NIC_ENGINE_QMAN_OFFSET (mmNIC0_QM1_BASE - mmNIC0_QM0_BASE) 84 85 #define TPC_CFG_OFFSET (mmTPC1_CFG_BASE - mmTPC0_CFG_BASE) 86 87 #define DMA_CORE_OFFSET (mmDMA1_CORE_BASE - mmDMA0_CORE_BASE) 88 89 #define QMAN_LDMA_SRC_OFFSET (mmDMA0_CORE_SRC_BASE_LO - mmDMA0_CORE_CFG_0) 90 #define QMAN_LDMA_DST_OFFSET (mmDMA0_CORE_DST_BASE_LO - mmDMA0_CORE_CFG_0) 91 #define QMAN_LDMA_SIZE_OFFSET (mmDMA0_CORE_DST_TSIZE_0 - mmDMA0_CORE_CFG_0) 92 93 #define QMAN_CPDMA_SRC_OFFSET (mmDMA0_QM_CQ_PTR_LO_4 - mmDMA0_CORE_CFG_0) 94 #define QMAN_CPDMA_DST_OFFSET (mmDMA0_CORE_DST_BASE_LO - mmDMA0_CORE_CFG_0) 95 #define QMAN_CPDMA_SIZE_OFFSET (mmDMA0_QM_CQ_TSIZE_4 - mmDMA0_CORE_CFG_0) 96 97 #define SIF_RTR_CTRL_OFFSET (mmSIF_RTR_CTRL_1_BASE - mmSIF_RTR_CTRL_0_BASE) 98 99 #define NIF_RTR_CTRL_OFFSET (mmNIF_RTR_CTRL_1_BASE - mmNIF_RTR_CTRL_0_BASE) 100 101 #define MME_ACC_OFFSET (mmMME1_ACC_BASE - mmMME0_ACC_BASE) 102 #define SRAM_BANK_OFFSET (mmSRAM_Y0_X1_RTR_BASE - mmSRAM_Y0_X0_RTR_BASE) 103 104 #define NUM_OF_SOB_IN_BLOCK \ 105 (((mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_2047 - \ 106 mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0) + 4) >> 2) 107 108 #define NUM_OF_MONITORS_IN_BLOCK \ 109 (((mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_STATUS_511 - \ 110 mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_MON_STATUS_0) + 4) >> 2) 111 112 #define MONITOR_MAX_SOBS 8 113 114 /* DRAM Memory Map */ 115 116 #define CPU_FW_IMAGE_SIZE 0x10000000 /* 256MB */ 117 #define MMU_PAGE_TABLES_SIZE 0x0BF00000 /* 191MB */ 118 #define MMU_CACHE_MNG_SIZE 0x00100000 /* 1MB */ 119 #define RESERVED 0x04000000 /* 64MB */ 120 121 #define CPU_FW_IMAGE_ADDR DRAM_PHYS_BASE 122 #define MMU_PAGE_TABLES_ADDR (CPU_FW_IMAGE_ADDR + CPU_FW_IMAGE_SIZE) 123 #define MMU_CACHE_MNG_ADDR (MMU_PAGE_TABLES_ADDR + MMU_PAGE_TABLES_SIZE) 124 125 #define DRAM_DRIVER_END_ADDR (MMU_CACHE_MNG_ADDR + MMU_CACHE_MNG_SIZE +\ 126 RESERVED) 127 128 #define DRAM_BASE_ADDR_USER 0x20000000 129 130 #if (DRAM_DRIVER_END_ADDR > DRAM_BASE_ADDR_USER) 131 #error "Driver must reserve no more than 512MB" 132 #endif 133 134 /* Internal QMANs PQ sizes */ 135 136 #define MME_QMAN_LENGTH 1024 137 #define MME_QMAN_SIZE_IN_BYTES (MME_QMAN_LENGTH * QMAN_PQ_ENTRY_SIZE) 138 139 #define HBM_DMA_QMAN_LENGTH 4096 140 #define HBM_DMA_QMAN_SIZE_IN_BYTES \ 141 (HBM_DMA_QMAN_LENGTH * QMAN_PQ_ENTRY_SIZE) 142 143 #define TPC_QMAN_LENGTH 1024 144 #define TPC_QMAN_SIZE_IN_BYTES (TPC_QMAN_LENGTH * QMAN_PQ_ENTRY_SIZE) 145 146 #define NIC_QMAN_LENGTH 4096 147 #define NIC_QMAN_SIZE_IN_BYTES (NIC_QMAN_LENGTH * QMAN_PQ_ENTRY_SIZE) 148 149 150 #define SRAM_USER_BASE_OFFSET GAUDI_DRIVER_SRAM_RESERVED_SIZE_FROM_START 151 152 /* Virtual address space */ 153 #define VA_HOST_SPACE_START 0x1000000000000ull /* 256TB */ 154 #define VA_HOST_SPACE_END 0x3FF8000000000ull /* 1PB - 512GB */ 155 #define VA_HOST_SPACE_SIZE (VA_HOST_SPACE_END - \ 156 VA_HOST_SPACE_START) /* 767TB */ 157 #define HOST_SPACE_INTERNAL_CB_SZ SZ_2M 158 159 #define HW_CAP_PLL BIT(0) 160 #define HW_CAP_HBM BIT(1) 161 #define HW_CAP_MMU BIT(2) 162 #define HW_CAP_MME BIT(3) 163 #define HW_CAP_CPU BIT(4) 164 #define HW_CAP_PCI_DMA BIT(5) 165 #define HW_CAP_MSI BIT(6) 166 #define HW_CAP_CPU_Q BIT(7) 167 #define HW_CAP_HBM_DMA BIT(8) 168 #define HW_CAP_SRAM_SCRAMBLER BIT(10) 169 #define HW_CAP_HBM_SCRAMBLER BIT(11) 170 171 #define HW_CAP_NIC0 BIT(14) 172 #define HW_CAP_NIC1 BIT(15) 173 #define HW_CAP_NIC2 BIT(16) 174 #define HW_CAP_NIC3 BIT(17) 175 #define HW_CAP_NIC4 BIT(18) 176 #define HW_CAP_NIC5 BIT(19) 177 #define HW_CAP_NIC6 BIT(20) 178 #define HW_CAP_NIC7 BIT(21) 179 #define HW_CAP_NIC8 BIT(22) 180 #define HW_CAP_NIC9 BIT(23) 181 #define HW_CAP_NIC_MASK GENMASK(23, 14) 182 #define HW_CAP_NIC_SHIFT 14 183 184 #define HW_CAP_TPC0 BIT(24) 185 #define HW_CAP_TPC1 BIT(25) 186 #define HW_CAP_TPC2 BIT(26) 187 #define HW_CAP_TPC3 BIT(27) 188 #define HW_CAP_TPC4 BIT(28) 189 #define HW_CAP_TPC5 BIT(29) 190 #define HW_CAP_TPC6 BIT(30) 191 #define HW_CAP_TPC7 BIT(31) 192 #define HW_CAP_TPC_MASK GENMASK(31, 24) 193 #define HW_CAP_TPC_SHIFT 24 194 195 #define NEXT_SYNC_OBJ_ADDR_INTERVAL \ 196 (mmSYNC_MNGR_W_N_SYNC_MNGR_OBJS_SOB_OBJ_0 - \ 197 mmSYNC_MNGR_E_N_SYNC_MNGR_OBJS_SOB_OBJ_0) 198 #define NUM_OF_MME_ENGINES 2 199 #define NUM_OF_MME_SUB_ENGINES 2 200 #define NUM_OF_TPC_ENGINES 8 201 #define NUM_OF_DMA_ENGINES 8 202 #define NUM_OF_QUEUES 5 203 #define NUM_OF_STREAMS 4 204 #define NUM_OF_FENCES 4 205 206 207 #define GAUDI_CPU_PCI_MSB_ADDR(addr) (((addr) & GENMASK_ULL(49, 39)) >> 39) 208 #define GAUDI_PCI_TO_CPU_ADDR(addr) \ 209 do { \ 210 (addr) &= ~GENMASK_ULL(49, 39); \ 211 (addr) |= BIT_ULL(39); \ 212 } while (0) 213 #define GAUDI_CPU_TO_PCI_ADDR(addr, extension) \ 214 do { \ 215 (addr) &= ~GENMASK_ULL(49, 39); \ 216 (addr) |= (u64) (extension) << 39; \ 217 } while (0) 218 219 enum gaudi_dma_channels { 220 GAUDI_PCI_DMA_1, 221 GAUDI_PCI_DMA_2, 222 GAUDI_HBM_DMA_1, 223 GAUDI_HBM_DMA_2, 224 GAUDI_HBM_DMA_3, 225 GAUDI_HBM_DMA_4, 226 GAUDI_HBM_DMA_5, 227 GAUDI_HBM_DMA_6, 228 GAUDI_DMA_MAX 229 }; 230 231 enum gaudi_tpc_mask { 232 GAUDI_TPC_MASK_TPC0 = 0x01, 233 GAUDI_TPC_MASK_TPC1 = 0x02, 234 GAUDI_TPC_MASK_TPC2 = 0x04, 235 GAUDI_TPC_MASK_TPC3 = 0x08, 236 GAUDI_TPC_MASK_TPC4 = 0x10, 237 GAUDI_TPC_MASK_TPC5 = 0x20, 238 GAUDI_TPC_MASK_TPC6 = 0x40, 239 GAUDI_TPC_MASK_TPC7 = 0x80, 240 GAUDI_TPC_MASK_ALL = 0xFF 241 }; 242 243 enum gaudi_nic_mask { 244 GAUDI_NIC_MASK_NIC0 = 0x01, 245 GAUDI_NIC_MASK_NIC1 = 0x02, 246 GAUDI_NIC_MASK_NIC2 = 0x04, 247 GAUDI_NIC_MASK_NIC3 = 0x08, 248 GAUDI_NIC_MASK_NIC4 = 0x10, 249 GAUDI_NIC_MASK_NIC5 = 0x20, 250 GAUDI_NIC_MASK_NIC6 = 0x40, 251 GAUDI_NIC_MASK_NIC7 = 0x80, 252 GAUDI_NIC_MASK_NIC8 = 0x100, 253 GAUDI_NIC_MASK_NIC9 = 0x200, 254 GAUDI_NIC_MASK_ALL = 0x3FF 255 }; 256 257 /* 258 * struct gaudi_hw_sob_group - H/W SOB group info. 259 * @hdev: habanalabs device structure. 260 * @kref: refcount of this SOB group. group will reset once refcount is zero. 261 * @base_sob_id: base sob id of this SOB group. 262 * @queue_id: id of the queue that waits on this sob group 263 */ 264 struct gaudi_hw_sob_group { 265 struct hl_device *hdev; 266 struct kref kref; 267 u32 base_sob_id; 268 u32 queue_id; 269 }; 270 271 #define NUM_SOB_GROUPS (HL_RSVD_SOBS * QMAN_STREAMS) 272 /** 273 * struct gaudi_collective_properties - 274 * holds all SOB groups and queues info reserved for the collective 275 * @hw_sob_group: H/W SOB groups. 276 * @next_sob_group_val: the next value to use for the currently used SOB group. 277 * @curr_sob_group_idx: the index of the currently used SOB group. 278 * @mstr_sob_mask: pre-defined masks for collective master monitors 279 */ 280 struct gaudi_collective_properties { 281 struct gaudi_hw_sob_group hw_sob_group[NUM_SOB_GROUPS]; 282 u16 next_sob_group_val[QMAN_STREAMS]; 283 u8 curr_sob_group_idx[QMAN_STREAMS]; 284 u8 mstr_sob_mask[HL_COLLECTIVE_RSVD_MSTR_MONS]; 285 }; 286 287 /** 288 * struct gaudi_internal_qman_info - Internal QMAN information. 289 * @pq_kernel_addr: Kernel address of the PQ memory area in the host. 290 * @pq_dma_addr: DMA address of the PQ memory area in the host. 291 * @pq_size: Size of allocated host memory for PQ. 292 */ 293 struct gaudi_internal_qman_info { 294 void *pq_kernel_addr; 295 dma_addr_t pq_dma_addr; 296 size_t pq_size; 297 }; 298 299 /** 300 * struct gaudi_device - ASIC specific manage structure. 301 * @cpucp_info_get: get information on device from CPU-CP 302 * @hw_queues_lock: protects the H/W queues from concurrent access. 303 * @internal_qmans: Internal QMANs information. The array size is larger than 304 * the actual number of internal queues because they are not in 305 * consecutive order. 306 * @hbm_bar_cur_addr: current address of HBM PCI bar. 307 * @events: array that holds all event id's 308 * @events_stat: array that holds histogram of all received events. 309 * @events_stat_aggregate: same as events_stat but doesn't get cleared on reset 310 * @hw_cap_initialized: This field contains a bit per H/W engine. When that 311 * engine is initialized, that bit is set by the driver to 312 * signal we can use this engine in later code paths. 313 * Each bit is cleared upon reset of its corresponding H/W 314 * engine. 315 * @mmu_cache_inv_pi: PI for MMU cache invalidation flow. The H/W expects an 316 * 8-bit value so use u8. 317 */ 318 struct gaudi_device { 319 int (*cpucp_info_get)(struct hl_device *hdev); 320 321 /* TODO: remove hw_queues_lock after moving to scheduler code */ 322 spinlock_t hw_queues_lock; 323 324 struct gaudi_internal_qman_info internal_qmans[GAUDI_QUEUE_ID_SIZE]; 325 326 struct gaudi_collective_properties collective_props; 327 328 u64 hbm_bar_cur_addr; 329 330 u32 events[GAUDI_EVENT_SIZE]; 331 u32 events_stat[GAUDI_EVENT_SIZE]; 332 u32 events_stat_aggregate[GAUDI_EVENT_SIZE]; 333 u32 hw_cap_initialized; 334 u8 mmu_cache_inv_pi; 335 }; 336 337 void gaudi_init_security(struct hl_device *hdev); 338 void gaudi_ack_protection_bits_errors(struct hl_device *hdev); 339 int gaudi_debug_coresight(struct hl_device *hdev, struct hl_ctx *ctx, void *data); 340 void gaudi_halt_coresight(struct hl_device *hdev, struct hl_ctx *ctx); 341 void gaudi_mmu_prepare_reg(struct hl_device *hdev, u64 reg, u32 asid); 342 343 #endif /* GAUDIP_H_ */ 344