1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note 2 * 3 * Copyright 2016-2022 HabanaLabs, Ltd. 4 * All Rights Reserved. 5 * 6 */ 7 8 #ifndef HABANALABS_H_ 9 #define HABANALABS_H_ 10 11 #include <linux/types.h> 12 #include <linux/ioctl.h> 13 14 /* 15 * Defines that are asic-specific but constitutes as ABI between kernel driver 16 * and userspace 17 */ 18 #define GOYA_KMD_SRAM_RESERVED_SIZE_FROM_START 0x8000 /* 32KB */ 19 #define GAUDI_DRIVER_SRAM_RESERVED_SIZE_FROM_START 0x80 /* 128 bytes */ 20 21 /* 22 * 128 SOBs reserved for collective wait 23 * 16 SOBs reserved for sync stream 24 */ 25 #define GAUDI_FIRST_AVAILABLE_W_S_SYNC_OBJECT 144 26 27 /* 28 * 64 monitors reserved for collective wait 29 * 8 monitors reserved for sync stream 30 */ 31 #define GAUDI_FIRST_AVAILABLE_W_S_MONITOR 72 32 33 /* Max number of elements in timestamps registration buffers */ 34 #define TS_MAX_ELEMENTS_NUM (1 << 20) /* 1MB */ 35 36 /* 37 * Goya queue Numbering 38 * 39 * The external queues (PCI DMA channels) MUST be before the internal queues 40 * and each group (PCI DMA channels and internal) must be contiguous inside 41 * itself but there can be a gap between the two groups (although not 42 * recommended) 43 */ 44 45 enum goya_queue_id { 46 GOYA_QUEUE_ID_DMA_0 = 0, 47 GOYA_QUEUE_ID_DMA_1 = 1, 48 GOYA_QUEUE_ID_DMA_2 = 2, 49 GOYA_QUEUE_ID_DMA_3 = 3, 50 GOYA_QUEUE_ID_DMA_4 = 4, 51 GOYA_QUEUE_ID_CPU_PQ = 5, 52 GOYA_QUEUE_ID_MME = 6, /* Internal queues start here */ 53 GOYA_QUEUE_ID_TPC0 = 7, 54 GOYA_QUEUE_ID_TPC1 = 8, 55 GOYA_QUEUE_ID_TPC2 = 9, 56 GOYA_QUEUE_ID_TPC3 = 10, 57 GOYA_QUEUE_ID_TPC4 = 11, 58 GOYA_QUEUE_ID_TPC5 = 12, 59 GOYA_QUEUE_ID_TPC6 = 13, 60 GOYA_QUEUE_ID_TPC7 = 14, 61 GOYA_QUEUE_ID_SIZE 62 }; 63 64 /* 65 * Gaudi queue Numbering 66 * External queues (PCI DMA channels) are DMA_0_*, DMA_1_* and DMA_5_*. 67 * Except one CPU queue, all the rest are internal queues. 68 */ 69 70 enum gaudi_queue_id { 71 GAUDI_QUEUE_ID_DMA_0_0 = 0, /* external */ 72 GAUDI_QUEUE_ID_DMA_0_1 = 1, /* external */ 73 GAUDI_QUEUE_ID_DMA_0_2 = 2, /* external */ 74 GAUDI_QUEUE_ID_DMA_0_3 = 3, /* external */ 75 GAUDI_QUEUE_ID_DMA_1_0 = 4, /* external */ 76 GAUDI_QUEUE_ID_DMA_1_1 = 5, /* external */ 77 GAUDI_QUEUE_ID_DMA_1_2 = 6, /* external */ 78 GAUDI_QUEUE_ID_DMA_1_3 = 7, /* external */ 79 GAUDI_QUEUE_ID_CPU_PQ = 8, /* CPU */ 80 GAUDI_QUEUE_ID_DMA_2_0 = 9, /* internal */ 81 GAUDI_QUEUE_ID_DMA_2_1 = 10, /* internal */ 82 GAUDI_QUEUE_ID_DMA_2_2 = 11, /* internal */ 83 GAUDI_QUEUE_ID_DMA_2_3 = 12, /* internal */ 84 GAUDI_QUEUE_ID_DMA_3_0 = 13, /* internal */ 85 GAUDI_QUEUE_ID_DMA_3_1 = 14, /* internal */ 86 GAUDI_QUEUE_ID_DMA_3_2 = 15, /* internal */ 87 GAUDI_QUEUE_ID_DMA_3_3 = 16, /* internal */ 88 GAUDI_QUEUE_ID_DMA_4_0 = 17, /* internal */ 89 GAUDI_QUEUE_ID_DMA_4_1 = 18, /* internal */ 90 GAUDI_QUEUE_ID_DMA_4_2 = 19, /* internal */ 91 GAUDI_QUEUE_ID_DMA_4_3 = 20, /* internal */ 92 GAUDI_QUEUE_ID_DMA_5_0 = 21, /* internal */ 93 GAUDI_QUEUE_ID_DMA_5_1 = 22, /* internal */ 94 GAUDI_QUEUE_ID_DMA_5_2 = 23, /* internal */ 95 GAUDI_QUEUE_ID_DMA_5_3 = 24, /* internal */ 96 GAUDI_QUEUE_ID_DMA_6_0 = 25, /* internal */ 97 GAUDI_QUEUE_ID_DMA_6_1 = 26, /* internal */ 98 GAUDI_QUEUE_ID_DMA_6_2 = 27, /* internal */ 99 GAUDI_QUEUE_ID_DMA_6_3 = 28, /* internal */ 100 GAUDI_QUEUE_ID_DMA_7_0 = 29, /* internal */ 101 GAUDI_QUEUE_ID_DMA_7_1 = 30, /* internal */ 102 GAUDI_QUEUE_ID_DMA_7_2 = 31, /* internal */ 103 GAUDI_QUEUE_ID_DMA_7_3 = 32, /* internal */ 104 GAUDI_QUEUE_ID_MME_0_0 = 33, /* internal */ 105 GAUDI_QUEUE_ID_MME_0_1 = 34, /* internal */ 106 GAUDI_QUEUE_ID_MME_0_2 = 35, /* internal */ 107 GAUDI_QUEUE_ID_MME_0_3 = 36, /* internal */ 108 GAUDI_QUEUE_ID_MME_1_0 = 37, /* internal */ 109 GAUDI_QUEUE_ID_MME_1_1 = 38, /* internal */ 110 GAUDI_QUEUE_ID_MME_1_2 = 39, /* internal */ 111 GAUDI_QUEUE_ID_MME_1_3 = 40, /* internal */ 112 GAUDI_QUEUE_ID_TPC_0_0 = 41, /* internal */ 113 GAUDI_QUEUE_ID_TPC_0_1 = 42, /* internal */ 114 GAUDI_QUEUE_ID_TPC_0_2 = 43, /* internal */ 115 GAUDI_QUEUE_ID_TPC_0_3 = 44, /* internal */ 116 GAUDI_QUEUE_ID_TPC_1_0 = 45, /* internal */ 117 GAUDI_QUEUE_ID_TPC_1_1 = 46, /* internal */ 118 GAUDI_QUEUE_ID_TPC_1_2 = 47, /* internal */ 119 GAUDI_QUEUE_ID_TPC_1_3 = 48, /* internal */ 120 GAUDI_QUEUE_ID_TPC_2_0 = 49, /* internal */ 121 GAUDI_QUEUE_ID_TPC_2_1 = 50, /* internal */ 122 GAUDI_QUEUE_ID_TPC_2_2 = 51, /* internal */ 123 GAUDI_QUEUE_ID_TPC_2_3 = 52, /* internal */ 124 GAUDI_QUEUE_ID_TPC_3_0 = 53, /* internal */ 125 GAUDI_QUEUE_ID_TPC_3_1 = 54, /* internal */ 126 GAUDI_QUEUE_ID_TPC_3_2 = 55, /* internal */ 127 GAUDI_QUEUE_ID_TPC_3_3 = 56, /* internal */ 128 GAUDI_QUEUE_ID_TPC_4_0 = 57, /* internal */ 129 GAUDI_QUEUE_ID_TPC_4_1 = 58, /* internal */ 130 GAUDI_QUEUE_ID_TPC_4_2 = 59, /* internal */ 131 GAUDI_QUEUE_ID_TPC_4_3 = 60, /* internal */ 132 GAUDI_QUEUE_ID_TPC_5_0 = 61, /* internal */ 133 GAUDI_QUEUE_ID_TPC_5_1 = 62, /* internal */ 134 GAUDI_QUEUE_ID_TPC_5_2 = 63, /* internal */ 135 GAUDI_QUEUE_ID_TPC_5_3 = 64, /* internal */ 136 GAUDI_QUEUE_ID_TPC_6_0 = 65, /* internal */ 137 GAUDI_QUEUE_ID_TPC_6_1 = 66, /* internal */ 138 GAUDI_QUEUE_ID_TPC_6_2 = 67, /* internal */ 139 GAUDI_QUEUE_ID_TPC_6_3 = 68, /* internal */ 140 GAUDI_QUEUE_ID_TPC_7_0 = 69, /* internal */ 141 GAUDI_QUEUE_ID_TPC_7_1 = 70, /* internal */ 142 GAUDI_QUEUE_ID_TPC_7_2 = 71, /* internal */ 143 GAUDI_QUEUE_ID_TPC_7_3 = 72, /* internal */ 144 GAUDI_QUEUE_ID_NIC_0_0 = 73, /* internal */ 145 GAUDI_QUEUE_ID_NIC_0_1 = 74, /* internal */ 146 GAUDI_QUEUE_ID_NIC_0_2 = 75, /* internal */ 147 GAUDI_QUEUE_ID_NIC_0_3 = 76, /* internal */ 148 GAUDI_QUEUE_ID_NIC_1_0 = 77, /* internal */ 149 GAUDI_QUEUE_ID_NIC_1_1 = 78, /* internal */ 150 GAUDI_QUEUE_ID_NIC_1_2 = 79, /* internal */ 151 GAUDI_QUEUE_ID_NIC_1_3 = 80, /* internal */ 152 GAUDI_QUEUE_ID_NIC_2_0 = 81, /* internal */ 153 GAUDI_QUEUE_ID_NIC_2_1 = 82, /* internal */ 154 GAUDI_QUEUE_ID_NIC_2_2 = 83, /* internal */ 155 GAUDI_QUEUE_ID_NIC_2_3 = 84, /* internal */ 156 GAUDI_QUEUE_ID_NIC_3_0 = 85, /* internal */ 157 GAUDI_QUEUE_ID_NIC_3_1 = 86, /* internal */ 158 GAUDI_QUEUE_ID_NIC_3_2 = 87, /* internal */ 159 GAUDI_QUEUE_ID_NIC_3_3 = 88, /* internal */ 160 GAUDI_QUEUE_ID_NIC_4_0 = 89, /* internal */ 161 GAUDI_QUEUE_ID_NIC_4_1 = 90, /* internal */ 162 GAUDI_QUEUE_ID_NIC_4_2 = 91, /* internal */ 163 GAUDI_QUEUE_ID_NIC_4_3 = 92, /* internal */ 164 GAUDI_QUEUE_ID_NIC_5_0 = 93, /* internal */ 165 GAUDI_QUEUE_ID_NIC_5_1 = 94, /* internal */ 166 GAUDI_QUEUE_ID_NIC_5_2 = 95, /* internal */ 167 GAUDI_QUEUE_ID_NIC_5_3 = 96, /* internal */ 168 GAUDI_QUEUE_ID_NIC_6_0 = 97, /* internal */ 169 GAUDI_QUEUE_ID_NIC_6_1 = 98, /* internal */ 170 GAUDI_QUEUE_ID_NIC_6_2 = 99, /* internal */ 171 GAUDI_QUEUE_ID_NIC_6_3 = 100, /* internal */ 172 GAUDI_QUEUE_ID_NIC_7_0 = 101, /* internal */ 173 GAUDI_QUEUE_ID_NIC_7_1 = 102, /* internal */ 174 GAUDI_QUEUE_ID_NIC_7_2 = 103, /* internal */ 175 GAUDI_QUEUE_ID_NIC_7_3 = 104, /* internal */ 176 GAUDI_QUEUE_ID_NIC_8_0 = 105, /* internal */ 177 GAUDI_QUEUE_ID_NIC_8_1 = 106, /* internal */ 178 GAUDI_QUEUE_ID_NIC_8_2 = 107, /* internal */ 179 GAUDI_QUEUE_ID_NIC_8_3 = 108, /* internal */ 180 GAUDI_QUEUE_ID_NIC_9_0 = 109, /* internal */ 181 GAUDI_QUEUE_ID_NIC_9_1 = 110, /* internal */ 182 GAUDI_QUEUE_ID_NIC_9_2 = 111, /* internal */ 183 GAUDI_QUEUE_ID_NIC_9_3 = 112, /* internal */ 184 GAUDI_QUEUE_ID_SIZE 185 }; 186 187 /* 188 * In GAUDI2 we have two modes of operation in regard to queues: 189 * 1. Legacy mode, where each QMAN exposes 4 streams to the user 190 * 2. F/W mode, where we use F/W to schedule the JOBS to the different queues. 191 * 192 * When in legacy mode, the user sends the queue id per JOB according to 193 * enum gaudi2_queue_id below. 194 * 195 * When in F/W mode, the user sends a stream id per Command Submission. The 196 * stream id is a running number from 0 up to (N-1), where N is the number 197 * of streams the F/W exposes and is passed to the user in 198 * struct hl_info_hw_ip_info 199 */ 200 201 enum gaudi2_queue_id { 202 GAUDI2_QUEUE_ID_PDMA_0_0 = 0, 203 GAUDI2_QUEUE_ID_PDMA_0_1 = 1, 204 GAUDI2_QUEUE_ID_PDMA_0_2 = 2, 205 GAUDI2_QUEUE_ID_PDMA_0_3 = 3, 206 GAUDI2_QUEUE_ID_PDMA_1_0 = 4, 207 GAUDI2_QUEUE_ID_PDMA_1_1 = 5, 208 GAUDI2_QUEUE_ID_PDMA_1_2 = 6, 209 GAUDI2_QUEUE_ID_PDMA_1_3 = 7, 210 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0 = 8, 211 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_1 = 9, 212 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_2 = 10, 213 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_3 = 11, 214 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0 = 12, 215 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_1 = 13, 216 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_2 = 14, 217 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3 = 15, 218 GAUDI2_QUEUE_ID_DCORE0_MME_0_0 = 16, 219 GAUDI2_QUEUE_ID_DCORE0_MME_0_1 = 17, 220 GAUDI2_QUEUE_ID_DCORE0_MME_0_2 = 18, 221 GAUDI2_QUEUE_ID_DCORE0_MME_0_3 = 19, 222 GAUDI2_QUEUE_ID_DCORE0_TPC_0_0 = 20, 223 GAUDI2_QUEUE_ID_DCORE0_TPC_0_1 = 21, 224 GAUDI2_QUEUE_ID_DCORE0_TPC_0_2 = 22, 225 GAUDI2_QUEUE_ID_DCORE0_TPC_0_3 = 23, 226 GAUDI2_QUEUE_ID_DCORE0_TPC_1_0 = 24, 227 GAUDI2_QUEUE_ID_DCORE0_TPC_1_1 = 25, 228 GAUDI2_QUEUE_ID_DCORE0_TPC_1_2 = 26, 229 GAUDI2_QUEUE_ID_DCORE0_TPC_1_3 = 27, 230 GAUDI2_QUEUE_ID_DCORE0_TPC_2_0 = 28, 231 GAUDI2_QUEUE_ID_DCORE0_TPC_2_1 = 29, 232 GAUDI2_QUEUE_ID_DCORE0_TPC_2_2 = 30, 233 GAUDI2_QUEUE_ID_DCORE0_TPC_2_3 = 31, 234 GAUDI2_QUEUE_ID_DCORE0_TPC_3_0 = 32, 235 GAUDI2_QUEUE_ID_DCORE0_TPC_3_1 = 33, 236 GAUDI2_QUEUE_ID_DCORE0_TPC_3_2 = 34, 237 GAUDI2_QUEUE_ID_DCORE0_TPC_3_3 = 35, 238 GAUDI2_QUEUE_ID_DCORE0_TPC_4_0 = 36, 239 GAUDI2_QUEUE_ID_DCORE0_TPC_4_1 = 37, 240 GAUDI2_QUEUE_ID_DCORE0_TPC_4_2 = 38, 241 GAUDI2_QUEUE_ID_DCORE0_TPC_4_3 = 39, 242 GAUDI2_QUEUE_ID_DCORE0_TPC_5_0 = 40, 243 GAUDI2_QUEUE_ID_DCORE0_TPC_5_1 = 41, 244 GAUDI2_QUEUE_ID_DCORE0_TPC_5_2 = 42, 245 GAUDI2_QUEUE_ID_DCORE0_TPC_5_3 = 43, 246 GAUDI2_QUEUE_ID_DCORE0_TPC_6_0 = 44, 247 GAUDI2_QUEUE_ID_DCORE0_TPC_6_1 = 45, 248 GAUDI2_QUEUE_ID_DCORE0_TPC_6_2 = 46, 249 GAUDI2_QUEUE_ID_DCORE0_TPC_6_3 = 47, 250 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0 = 48, 251 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_1 = 49, 252 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_2 = 50, 253 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_3 = 51, 254 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0 = 52, 255 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_1 = 53, 256 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_2 = 54, 257 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3 = 55, 258 GAUDI2_QUEUE_ID_DCORE1_MME_0_0 = 56, 259 GAUDI2_QUEUE_ID_DCORE1_MME_0_1 = 57, 260 GAUDI2_QUEUE_ID_DCORE1_MME_0_2 = 58, 261 GAUDI2_QUEUE_ID_DCORE1_MME_0_3 = 59, 262 GAUDI2_QUEUE_ID_DCORE1_TPC_0_0 = 60, 263 GAUDI2_QUEUE_ID_DCORE1_TPC_0_1 = 61, 264 GAUDI2_QUEUE_ID_DCORE1_TPC_0_2 = 62, 265 GAUDI2_QUEUE_ID_DCORE1_TPC_0_3 = 63, 266 GAUDI2_QUEUE_ID_DCORE1_TPC_1_0 = 64, 267 GAUDI2_QUEUE_ID_DCORE1_TPC_1_1 = 65, 268 GAUDI2_QUEUE_ID_DCORE1_TPC_1_2 = 66, 269 GAUDI2_QUEUE_ID_DCORE1_TPC_1_3 = 67, 270 GAUDI2_QUEUE_ID_DCORE1_TPC_2_0 = 68, 271 GAUDI2_QUEUE_ID_DCORE1_TPC_2_1 = 69, 272 GAUDI2_QUEUE_ID_DCORE1_TPC_2_2 = 70, 273 GAUDI2_QUEUE_ID_DCORE1_TPC_2_3 = 71, 274 GAUDI2_QUEUE_ID_DCORE1_TPC_3_0 = 72, 275 GAUDI2_QUEUE_ID_DCORE1_TPC_3_1 = 73, 276 GAUDI2_QUEUE_ID_DCORE1_TPC_3_2 = 74, 277 GAUDI2_QUEUE_ID_DCORE1_TPC_3_3 = 75, 278 GAUDI2_QUEUE_ID_DCORE1_TPC_4_0 = 76, 279 GAUDI2_QUEUE_ID_DCORE1_TPC_4_1 = 77, 280 GAUDI2_QUEUE_ID_DCORE1_TPC_4_2 = 78, 281 GAUDI2_QUEUE_ID_DCORE1_TPC_4_3 = 79, 282 GAUDI2_QUEUE_ID_DCORE1_TPC_5_0 = 80, 283 GAUDI2_QUEUE_ID_DCORE1_TPC_5_1 = 81, 284 GAUDI2_QUEUE_ID_DCORE1_TPC_5_2 = 82, 285 GAUDI2_QUEUE_ID_DCORE1_TPC_5_3 = 83, 286 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0 = 84, 287 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_1 = 85, 288 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_2 = 86, 289 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_3 = 87, 290 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0 = 88, 291 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_1 = 89, 292 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_2 = 90, 293 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3 = 91, 294 GAUDI2_QUEUE_ID_DCORE2_MME_0_0 = 92, 295 GAUDI2_QUEUE_ID_DCORE2_MME_0_1 = 93, 296 GAUDI2_QUEUE_ID_DCORE2_MME_0_2 = 94, 297 GAUDI2_QUEUE_ID_DCORE2_MME_0_3 = 95, 298 GAUDI2_QUEUE_ID_DCORE2_TPC_0_0 = 96, 299 GAUDI2_QUEUE_ID_DCORE2_TPC_0_1 = 97, 300 GAUDI2_QUEUE_ID_DCORE2_TPC_0_2 = 98, 301 GAUDI2_QUEUE_ID_DCORE2_TPC_0_3 = 99, 302 GAUDI2_QUEUE_ID_DCORE2_TPC_1_0 = 100, 303 GAUDI2_QUEUE_ID_DCORE2_TPC_1_1 = 101, 304 GAUDI2_QUEUE_ID_DCORE2_TPC_1_2 = 102, 305 GAUDI2_QUEUE_ID_DCORE2_TPC_1_3 = 103, 306 GAUDI2_QUEUE_ID_DCORE2_TPC_2_0 = 104, 307 GAUDI2_QUEUE_ID_DCORE2_TPC_2_1 = 105, 308 GAUDI2_QUEUE_ID_DCORE2_TPC_2_2 = 106, 309 GAUDI2_QUEUE_ID_DCORE2_TPC_2_3 = 107, 310 GAUDI2_QUEUE_ID_DCORE2_TPC_3_0 = 108, 311 GAUDI2_QUEUE_ID_DCORE2_TPC_3_1 = 109, 312 GAUDI2_QUEUE_ID_DCORE2_TPC_3_2 = 110, 313 GAUDI2_QUEUE_ID_DCORE2_TPC_3_3 = 111, 314 GAUDI2_QUEUE_ID_DCORE2_TPC_4_0 = 112, 315 GAUDI2_QUEUE_ID_DCORE2_TPC_4_1 = 113, 316 GAUDI2_QUEUE_ID_DCORE2_TPC_4_2 = 114, 317 GAUDI2_QUEUE_ID_DCORE2_TPC_4_3 = 115, 318 GAUDI2_QUEUE_ID_DCORE2_TPC_5_0 = 116, 319 GAUDI2_QUEUE_ID_DCORE2_TPC_5_1 = 117, 320 GAUDI2_QUEUE_ID_DCORE2_TPC_5_2 = 118, 321 GAUDI2_QUEUE_ID_DCORE2_TPC_5_3 = 119, 322 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0 = 120, 323 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_1 = 121, 324 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_2 = 122, 325 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_3 = 123, 326 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0 = 124, 327 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_1 = 125, 328 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_2 = 126, 329 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3 = 127, 330 GAUDI2_QUEUE_ID_DCORE3_MME_0_0 = 128, 331 GAUDI2_QUEUE_ID_DCORE3_MME_0_1 = 129, 332 GAUDI2_QUEUE_ID_DCORE3_MME_0_2 = 130, 333 GAUDI2_QUEUE_ID_DCORE3_MME_0_3 = 131, 334 GAUDI2_QUEUE_ID_DCORE3_TPC_0_0 = 132, 335 GAUDI2_QUEUE_ID_DCORE3_TPC_0_1 = 133, 336 GAUDI2_QUEUE_ID_DCORE3_TPC_0_2 = 134, 337 GAUDI2_QUEUE_ID_DCORE3_TPC_0_3 = 135, 338 GAUDI2_QUEUE_ID_DCORE3_TPC_1_0 = 136, 339 GAUDI2_QUEUE_ID_DCORE3_TPC_1_1 = 137, 340 GAUDI2_QUEUE_ID_DCORE3_TPC_1_2 = 138, 341 GAUDI2_QUEUE_ID_DCORE3_TPC_1_3 = 139, 342 GAUDI2_QUEUE_ID_DCORE3_TPC_2_0 = 140, 343 GAUDI2_QUEUE_ID_DCORE3_TPC_2_1 = 141, 344 GAUDI2_QUEUE_ID_DCORE3_TPC_2_2 = 142, 345 GAUDI2_QUEUE_ID_DCORE3_TPC_2_3 = 143, 346 GAUDI2_QUEUE_ID_DCORE3_TPC_3_0 = 144, 347 GAUDI2_QUEUE_ID_DCORE3_TPC_3_1 = 145, 348 GAUDI2_QUEUE_ID_DCORE3_TPC_3_2 = 146, 349 GAUDI2_QUEUE_ID_DCORE3_TPC_3_3 = 147, 350 GAUDI2_QUEUE_ID_DCORE3_TPC_4_0 = 148, 351 GAUDI2_QUEUE_ID_DCORE3_TPC_4_1 = 149, 352 GAUDI2_QUEUE_ID_DCORE3_TPC_4_2 = 150, 353 GAUDI2_QUEUE_ID_DCORE3_TPC_4_3 = 151, 354 GAUDI2_QUEUE_ID_DCORE3_TPC_5_0 = 152, 355 GAUDI2_QUEUE_ID_DCORE3_TPC_5_1 = 153, 356 GAUDI2_QUEUE_ID_DCORE3_TPC_5_2 = 154, 357 GAUDI2_QUEUE_ID_DCORE3_TPC_5_3 = 155, 358 GAUDI2_QUEUE_ID_NIC_0_0 = 156, 359 GAUDI2_QUEUE_ID_NIC_0_1 = 157, 360 GAUDI2_QUEUE_ID_NIC_0_2 = 158, 361 GAUDI2_QUEUE_ID_NIC_0_3 = 159, 362 GAUDI2_QUEUE_ID_NIC_1_0 = 160, 363 GAUDI2_QUEUE_ID_NIC_1_1 = 161, 364 GAUDI2_QUEUE_ID_NIC_1_2 = 162, 365 GAUDI2_QUEUE_ID_NIC_1_3 = 163, 366 GAUDI2_QUEUE_ID_NIC_2_0 = 164, 367 GAUDI2_QUEUE_ID_NIC_2_1 = 165, 368 GAUDI2_QUEUE_ID_NIC_2_2 = 166, 369 GAUDI2_QUEUE_ID_NIC_2_3 = 167, 370 GAUDI2_QUEUE_ID_NIC_3_0 = 168, 371 GAUDI2_QUEUE_ID_NIC_3_1 = 169, 372 GAUDI2_QUEUE_ID_NIC_3_2 = 170, 373 GAUDI2_QUEUE_ID_NIC_3_3 = 171, 374 GAUDI2_QUEUE_ID_NIC_4_0 = 172, 375 GAUDI2_QUEUE_ID_NIC_4_1 = 173, 376 GAUDI2_QUEUE_ID_NIC_4_2 = 174, 377 GAUDI2_QUEUE_ID_NIC_4_3 = 175, 378 GAUDI2_QUEUE_ID_NIC_5_0 = 176, 379 GAUDI2_QUEUE_ID_NIC_5_1 = 177, 380 GAUDI2_QUEUE_ID_NIC_5_2 = 178, 381 GAUDI2_QUEUE_ID_NIC_5_3 = 179, 382 GAUDI2_QUEUE_ID_NIC_6_0 = 180, 383 GAUDI2_QUEUE_ID_NIC_6_1 = 181, 384 GAUDI2_QUEUE_ID_NIC_6_2 = 182, 385 GAUDI2_QUEUE_ID_NIC_6_3 = 183, 386 GAUDI2_QUEUE_ID_NIC_7_0 = 184, 387 GAUDI2_QUEUE_ID_NIC_7_1 = 185, 388 GAUDI2_QUEUE_ID_NIC_7_2 = 186, 389 GAUDI2_QUEUE_ID_NIC_7_3 = 187, 390 GAUDI2_QUEUE_ID_NIC_8_0 = 188, 391 GAUDI2_QUEUE_ID_NIC_8_1 = 189, 392 GAUDI2_QUEUE_ID_NIC_8_2 = 190, 393 GAUDI2_QUEUE_ID_NIC_8_3 = 191, 394 GAUDI2_QUEUE_ID_NIC_9_0 = 192, 395 GAUDI2_QUEUE_ID_NIC_9_1 = 193, 396 GAUDI2_QUEUE_ID_NIC_9_2 = 194, 397 GAUDI2_QUEUE_ID_NIC_9_3 = 195, 398 GAUDI2_QUEUE_ID_NIC_10_0 = 196, 399 GAUDI2_QUEUE_ID_NIC_10_1 = 197, 400 GAUDI2_QUEUE_ID_NIC_10_2 = 198, 401 GAUDI2_QUEUE_ID_NIC_10_3 = 199, 402 GAUDI2_QUEUE_ID_NIC_11_0 = 200, 403 GAUDI2_QUEUE_ID_NIC_11_1 = 201, 404 GAUDI2_QUEUE_ID_NIC_11_2 = 202, 405 GAUDI2_QUEUE_ID_NIC_11_3 = 203, 406 GAUDI2_QUEUE_ID_NIC_12_0 = 204, 407 GAUDI2_QUEUE_ID_NIC_12_1 = 205, 408 GAUDI2_QUEUE_ID_NIC_12_2 = 206, 409 GAUDI2_QUEUE_ID_NIC_12_3 = 207, 410 GAUDI2_QUEUE_ID_NIC_13_0 = 208, 411 GAUDI2_QUEUE_ID_NIC_13_1 = 209, 412 GAUDI2_QUEUE_ID_NIC_13_2 = 210, 413 GAUDI2_QUEUE_ID_NIC_13_3 = 211, 414 GAUDI2_QUEUE_ID_NIC_14_0 = 212, 415 GAUDI2_QUEUE_ID_NIC_14_1 = 213, 416 GAUDI2_QUEUE_ID_NIC_14_2 = 214, 417 GAUDI2_QUEUE_ID_NIC_14_3 = 215, 418 GAUDI2_QUEUE_ID_NIC_15_0 = 216, 419 GAUDI2_QUEUE_ID_NIC_15_1 = 217, 420 GAUDI2_QUEUE_ID_NIC_15_2 = 218, 421 GAUDI2_QUEUE_ID_NIC_15_3 = 219, 422 GAUDI2_QUEUE_ID_NIC_16_0 = 220, 423 GAUDI2_QUEUE_ID_NIC_16_1 = 221, 424 GAUDI2_QUEUE_ID_NIC_16_2 = 222, 425 GAUDI2_QUEUE_ID_NIC_16_3 = 223, 426 GAUDI2_QUEUE_ID_NIC_17_0 = 224, 427 GAUDI2_QUEUE_ID_NIC_17_1 = 225, 428 GAUDI2_QUEUE_ID_NIC_17_2 = 226, 429 GAUDI2_QUEUE_ID_NIC_17_3 = 227, 430 GAUDI2_QUEUE_ID_NIC_18_0 = 228, 431 GAUDI2_QUEUE_ID_NIC_18_1 = 229, 432 GAUDI2_QUEUE_ID_NIC_18_2 = 230, 433 GAUDI2_QUEUE_ID_NIC_18_3 = 231, 434 GAUDI2_QUEUE_ID_NIC_19_0 = 232, 435 GAUDI2_QUEUE_ID_NIC_19_1 = 233, 436 GAUDI2_QUEUE_ID_NIC_19_2 = 234, 437 GAUDI2_QUEUE_ID_NIC_19_3 = 235, 438 GAUDI2_QUEUE_ID_NIC_20_0 = 236, 439 GAUDI2_QUEUE_ID_NIC_20_1 = 237, 440 GAUDI2_QUEUE_ID_NIC_20_2 = 238, 441 GAUDI2_QUEUE_ID_NIC_20_3 = 239, 442 GAUDI2_QUEUE_ID_NIC_21_0 = 240, 443 GAUDI2_QUEUE_ID_NIC_21_1 = 241, 444 GAUDI2_QUEUE_ID_NIC_21_2 = 242, 445 GAUDI2_QUEUE_ID_NIC_21_3 = 243, 446 GAUDI2_QUEUE_ID_NIC_22_0 = 244, 447 GAUDI2_QUEUE_ID_NIC_22_1 = 245, 448 GAUDI2_QUEUE_ID_NIC_22_2 = 246, 449 GAUDI2_QUEUE_ID_NIC_22_3 = 247, 450 GAUDI2_QUEUE_ID_NIC_23_0 = 248, 451 GAUDI2_QUEUE_ID_NIC_23_1 = 249, 452 GAUDI2_QUEUE_ID_NIC_23_2 = 250, 453 GAUDI2_QUEUE_ID_NIC_23_3 = 251, 454 GAUDI2_QUEUE_ID_ROT_0_0 = 252, 455 GAUDI2_QUEUE_ID_ROT_0_1 = 253, 456 GAUDI2_QUEUE_ID_ROT_0_2 = 254, 457 GAUDI2_QUEUE_ID_ROT_0_3 = 255, 458 GAUDI2_QUEUE_ID_ROT_1_0 = 256, 459 GAUDI2_QUEUE_ID_ROT_1_1 = 257, 460 GAUDI2_QUEUE_ID_ROT_1_2 = 258, 461 GAUDI2_QUEUE_ID_ROT_1_3 = 259, 462 GAUDI2_QUEUE_ID_CPU_PQ = 260, 463 GAUDI2_QUEUE_ID_SIZE 464 }; 465 466 /* 467 * Engine Numbering 468 * 469 * Used in the "busy_engines_mask" field in `struct hl_info_hw_idle' 470 */ 471 472 enum goya_engine_id { 473 GOYA_ENGINE_ID_DMA_0 = 0, 474 GOYA_ENGINE_ID_DMA_1, 475 GOYA_ENGINE_ID_DMA_2, 476 GOYA_ENGINE_ID_DMA_3, 477 GOYA_ENGINE_ID_DMA_4, 478 GOYA_ENGINE_ID_MME_0, 479 GOYA_ENGINE_ID_TPC_0, 480 GOYA_ENGINE_ID_TPC_1, 481 GOYA_ENGINE_ID_TPC_2, 482 GOYA_ENGINE_ID_TPC_3, 483 GOYA_ENGINE_ID_TPC_4, 484 GOYA_ENGINE_ID_TPC_5, 485 GOYA_ENGINE_ID_TPC_6, 486 GOYA_ENGINE_ID_TPC_7, 487 GOYA_ENGINE_ID_SIZE 488 }; 489 490 enum gaudi_engine_id { 491 GAUDI_ENGINE_ID_DMA_0 = 0, 492 GAUDI_ENGINE_ID_DMA_1, 493 GAUDI_ENGINE_ID_DMA_2, 494 GAUDI_ENGINE_ID_DMA_3, 495 GAUDI_ENGINE_ID_DMA_4, 496 GAUDI_ENGINE_ID_DMA_5, 497 GAUDI_ENGINE_ID_DMA_6, 498 GAUDI_ENGINE_ID_DMA_7, 499 GAUDI_ENGINE_ID_MME_0, 500 GAUDI_ENGINE_ID_MME_1, 501 GAUDI_ENGINE_ID_MME_2, 502 GAUDI_ENGINE_ID_MME_3, 503 GAUDI_ENGINE_ID_TPC_0, 504 GAUDI_ENGINE_ID_TPC_1, 505 GAUDI_ENGINE_ID_TPC_2, 506 GAUDI_ENGINE_ID_TPC_3, 507 GAUDI_ENGINE_ID_TPC_4, 508 GAUDI_ENGINE_ID_TPC_5, 509 GAUDI_ENGINE_ID_TPC_6, 510 GAUDI_ENGINE_ID_TPC_7, 511 GAUDI_ENGINE_ID_NIC_0, 512 GAUDI_ENGINE_ID_NIC_1, 513 GAUDI_ENGINE_ID_NIC_2, 514 GAUDI_ENGINE_ID_NIC_3, 515 GAUDI_ENGINE_ID_NIC_4, 516 GAUDI_ENGINE_ID_NIC_5, 517 GAUDI_ENGINE_ID_NIC_6, 518 GAUDI_ENGINE_ID_NIC_7, 519 GAUDI_ENGINE_ID_NIC_8, 520 GAUDI_ENGINE_ID_NIC_9, 521 GAUDI_ENGINE_ID_SIZE 522 }; 523 524 enum gaudi2_engine_id { 525 GAUDI2_DCORE0_ENGINE_ID_EDMA_0 = 0, 526 GAUDI2_DCORE0_ENGINE_ID_EDMA_1, 527 GAUDI2_DCORE0_ENGINE_ID_MME, 528 GAUDI2_DCORE0_ENGINE_ID_TPC_0, 529 GAUDI2_DCORE0_ENGINE_ID_TPC_1, 530 GAUDI2_DCORE0_ENGINE_ID_TPC_2, 531 GAUDI2_DCORE0_ENGINE_ID_TPC_3, 532 GAUDI2_DCORE0_ENGINE_ID_TPC_4, 533 GAUDI2_DCORE0_ENGINE_ID_TPC_5, 534 GAUDI2_DCORE0_ENGINE_ID_DEC_0, 535 GAUDI2_DCORE0_ENGINE_ID_DEC_1, 536 GAUDI2_DCORE1_ENGINE_ID_EDMA_0, 537 GAUDI2_DCORE1_ENGINE_ID_EDMA_1, 538 GAUDI2_DCORE1_ENGINE_ID_MME, 539 GAUDI2_DCORE1_ENGINE_ID_TPC_0, 540 GAUDI2_DCORE1_ENGINE_ID_TPC_1, 541 GAUDI2_DCORE1_ENGINE_ID_TPC_2, 542 GAUDI2_DCORE1_ENGINE_ID_TPC_3, 543 GAUDI2_DCORE1_ENGINE_ID_TPC_4, 544 GAUDI2_DCORE1_ENGINE_ID_TPC_5, 545 GAUDI2_DCORE1_ENGINE_ID_DEC_0, 546 GAUDI2_DCORE1_ENGINE_ID_DEC_1, 547 GAUDI2_DCORE2_ENGINE_ID_EDMA_0, 548 GAUDI2_DCORE2_ENGINE_ID_EDMA_1, 549 GAUDI2_DCORE2_ENGINE_ID_MME, 550 GAUDI2_DCORE2_ENGINE_ID_TPC_0, 551 GAUDI2_DCORE2_ENGINE_ID_TPC_1, 552 GAUDI2_DCORE2_ENGINE_ID_TPC_2, 553 GAUDI2_DCORE2_ENGINE_ID_TPC_3, 554 GAUDI2_DCORE2_ENGINE_ID_TPC_4, 555 GAUDI2_DCORE2_ENGINE_ID_TPC_5, 556 GAUDI2_DCORE2_ENGINE_ID_DEC_0, 557 GAUDI2_DCORE2_ENGINE_ID_DEC_1, 558 GAUDI2_DCORE3_ENGINE_ID_EDMA_0, 559 GAUDI2_DCORE3_ENGINE_ID_EDMA_1, 560 GAUDI2_DCORE3_ENGINE_ID_MME, 561 GAUDI2_DCORE3_ENGINE_ID_TPC_0, 562 GAUDI2_DCORE3_ENGINE_ID_TPC_1, 563 GAUDI2_DCORE3_ENGINE_ID_TPC_2, 564 GAUDI2_DCORE3_ENGINE_ID_TPC_3, 565 GAUDI2_DCORE3_ENGINE_ID_TPC_4, 566 GAUDI2_DCORE3_ENGINE_ID_TPC_5, 567 GAUDI2_DCORE3_ENGINE_ID_DEC_0, 568 GAUDI2_DCORE3_ENGINE_ID_DEC_1, 569 GAUDI2_DCORE0_ENGINE_ID_TPC_6, 570 GAUDI2_ENGINE_ID_PDMA_0, 571 GAUDI2_ENGINE_ID_PDMA_1, 572 GAUDI2_ENGINE_ID_ROT_0, 573 GAUDI2_ENGINE_ID_ROT_1, 574 GAUDI2_PCIE_ENGINE_ID_DEC_0, 575 GAUDI2_PCIE_ENGINE_ID_DEC_1, 576 GAUDI2_ENGINE_ID_NIC0_0, 577 GAUDI2_ENGINE_ID_NIC0_1, 578 GAUDI2_ENGINE_ID_NIC1_0, 579 GAUDI2_ENGINE_ID_NIC1_1, 580 GAUDI2_ENGINE_ID_NIC2_0, 581 GAUDI2_ENGINE_ID_NIC2_1, 582 GAUDI2_ENGINE_ID_NIC3_0, 583 GAUDI2_ENGINE_ID_NIC3_1, 584 GAUDI2_ENGINE_ID_NIC4_0, 585 GAUDI2_ENGINE_ID_NIC4_1, 586 GAUDI2_ENGINE_ID_NIC5_0, 587 GAUDI2_ENGINE_ID_NIC5_1, 588 GAUDI2_ENGINE_ID_NIC6_0, 589 GAUDI2_ENGINE_ID_NIC6_1, 590 GAUDI2_ENGINE_ID_NIC7_0, 591 GAUDI2_ENGINE_ID_NIC7_1, 592 GAUDI2_ENGINE_ID_NIC8_0, 593 GAUDI2_ENGINE_ID_NIC8_1, 594 GAUDI2_ENGINE_ID_NIC9_0, 595 GAUDI2_ENGINE_ID_NIC9_1, 596 GAUDI2_ENGINE_ID_NIC10_0, 597 GAUDI2_ENGINE_ID_NIC10_1, 598 GAUDI2_ENGINE_ID_NIC11_0, 599 GAUDI2_ENGINE_ID_NIC11_1, 600 GAUDI2_ENGINE_ID_PCIE, 601 GAUDI2_ENGINE_ID_PSOC, 602 GAUDI2_ENGINE_ID_ARC_FARM, 603 GAUDI2_ENGINE_ID_KDMA, 604 GAUDI2_ENGINE_ID_SIZE 605 }; 606 607 /* 608 * ASIC specific PLL index 609 * 610 * Used to retrieve in frequency info of different IPs via 611 * HL_INFO_PLL_FREQUENCY under HL_IOCTL_INFO IOCTL. The enums need to be 612 * used as an index in struct hl_pll_frequency_info 613 */ 614 615 enum hl_goya_pll_index { 616 HL_GOYA_CPU_PLL = 0, 617 HL_GOYA_IC_PLL, 618 HL_GOYA_MC_PLL, 619 HL_GOYA_MME_PLL, 620 HL_GOYA_PCI_PLL, 621 HL_GOYA_EMMC_PLL, 622 HL_GOYA_TPC_PLL, 623 HL_GOYA_PLL_MAX 624 }; 625 626 enum hl_gaudi_pll_index { 627 HL_GAUDI_CPU_PLL = 0, 628 HL_GAUDI_PCI_PLL, 629 HL_GAUDI_SRAM_PLL, 630 HL_GAUDI_HBM_PLL, 631 HL_GAUDI_NIC_PLL, 632 HL_GAUDI_DMA_PLL, 633 HL_GAUDI_MESH_PLL, 634 HL_GAUDI_MME_PLL, 635 HL_GAUDI_TPC_PLL, 636 HL_GAUDI_IF_PLL, 637 HL_GAUDI_PLL_MAX 638 }; 639 640 enum hl_gaudi2_pll_index { 641 HL_GAUDI2_CPU_PLL = 0, 642 HL_GAUDI2_PCI_PLL, 643 HL_GAUDI2_SRAM_PLL, 644 HL_GAUDI2_HBM_PLL, 645 HL_GAUDI2_NIC_PLL, 646 HL_GAUDI2_DMA_PLL, 647 HL_GAUDI2_MESH_PLL, 648 HL_GAUDI2_MME_PLL, 649 HL_GAUDI2_TPC_PLL, 650 HL_GAUDI2_IF_PLL, 651 HL_GAUDI2_VID_PLL, 652 HL_GAUDI2_MSS_PLL, 653 HL_GAUDI2_PLL_MAX 654 }; 655 656 /** 657 * enum hl_goya_dma_direction - Direction of DMA operation inside a LIN_DMA packet that is 658 * submitted to the GOYA's DMA QMAN. This attribute is not relevant 659 * to the H/W but the kernel driver use it to parse the packet's 660 * addresses and patch/validate them. 661 * @HL_DMA_HOST_TO_DRAM: DMA operation from Host memory to GOYA's DDR. 662 * @HL_DMA_HOST_TO_SRAM: DMA operation from Host memory to GOYA's SRAM. 663 * @HL_DMA_DRAM_TO_SRAM: DMA operation from GOYA's DDR to GOYA's SRAM. 664 * @HL_DMA_SRAM_TO_DRAM: DMA operation from GOYA's SRAM to GOYA's DDR. 665 * @HL_DMA_SRAM_TO_HOST: DMA operation from GOYA's SRAM to Host memory. 666 * @HL_DMA_DRAM_TO_HOST: DMA operation from GOYA's DDR to Host memory. 667 * @HL_DMA_DRAM_TO_DRAM: DMA operation from GOYA's DDR to GOYA's DDR. 668 * @HL_DMA_SRAM_TO_SRAM: DMA operation from GOYA's SRAM to GOYA's SRAM. 669 * @HL_DMA_ENUM_MAX: number of values in enum 670 */ 671 enum hl_goya_dma_direction { 672 HL_DMA_HOST_TO_DRAM, 673 HL_DMA_HOST_TO_SRAM, 674 HL_DMA_DRAM_TO_SRAM, 675 HL_DMA_SRAM_TO_DRAM, 676 HL_DMA_SRAM_TO_HOST, 677 HL_DMA_DRAM_TO_HOST, 678 HL_DMA_DRAM_TO_DRAM, 679 HL_DMA_SRAM_TO_SRAM, 680 HL_DMA_ENUM_MAX 681 }; 682 683 /** 684 * enum hl_device_status - Device status information. 685 * @HL_DEVICE_STATUS_OPERATIONAL: Device is operational. 686 * @HL_DEVICE_STATUS_IN_RESET: Device is currently during reset. 687 * @HL_DEVICE_STATUS_MALFUNCTION: Device is unusable. 688 * @HL_DEVICE_STATUS_NEEDS_RESET: Device needs reset because auto reset was disabled. 689 * @HL_DEVICE_STATUS_IN_DEVICE_CREATION: Device is operational but its creation is still in 690 * progress. 691 * @HL_DEVICE_STATUS_IN_RESET_AFTER_DEVICE_RELEASE: Device is currently during reset that was 692 * triggered because the user released the device 693 * @HL_DEVICE_STATUS_LAST: Last status. 694 */ 695 enum hl_device_status { 696 HL_DEVICE_STATUS_OPERATIONAL, 697 HL_DEVICE_STATUS_IN_RESET, 698 HL_DEVICE_STATUS_MALFUNCTION, 699 HL_DEVICE_STATUS_NEEDS_RESET, 700 HL_DEVICE_STATUS_IN_DEVICE_CREATION, 701 HL_DEVICE_STATUS_IN_RESET_AFTER_DEVICE_RELEASE, 702 HL_DEVICE_STATUS_LAST = HL_DEVICE_STATUS_IN_RESET_AFTER_DEVICE_RELEASE 703 }; 704 705 enum hl_server_type { 706 HL_SERVER_TYPE_UNKNOWN = 0, 707 HL_SERVER_GAUDI_HLS1 = 1, 708 HL_SERVER_GAUDI_HLS1H = 2, 709 HL_SERVER_GAUDI_TYPE1 = 3, 710 HL_SERVER_GAUDI_TYPE2 = 4, 711 HL_SERVER_GAUDI2_HLS2 = 5, 712 HL_SERVER_GAUDI2_TYPE1 = 7 713 }; 714 715 /* 716 * Notifier event values - for the notification mechanism and the HL_INFO_GET_EVENTS command 717 * 718 * HL_NOTIFIER_EVENT_TPC_ASSERT - Indicates TPC assert event 719 * HL_NOTIFIER_EVENT_UNDEFINED_OPCODE - Indicates undefined operation code 720 * HL_NOTIFIER_EVENT_DEVICE_RESET - Indicates device requires a reset 721 * HL_NOTIFIER_EVENT_CS_TIMEOUT - Indicates CS timeout error 722 * HL_NOTIFIER_EVENT_DEVICE_UNAVAILABLE - Indicates device is unavailable 723 * HL_NOTIFIER_EVENT_USER_ENGINE_ERR - Indicates device engine in error state 724 * HL_NOTIFIER_EVENT_GENERAL_HW_ERR - Indicates device HW error 725 * HL_NOTIFIER_EVENT_RAZWI - Indicates razwi happened 726 * HL_NOTIFIER_EVENT_PAGE_FAULT - Indicates page fault happened 727 * HL_NOTIFIER_EVENT_CRITICAL_HW_ERR - Indicates a HW error that requires SW abort and 728 * HW reset 729 * HL_NOTIFIER_EVENT_CRITICAL_FW_ERR - Indicates a FW error that requires SW abort and 730 * HW reset 731 */ 732 #define HL_NOTIFIER_EVENT_TPC_ASSERT (1ULL << 0) 733 #define HL_NOTIFIER_EVENT_UNDEFINED_OPCODE (1ULL << 1) 734 #define HL_NOTIFIER_EVENT_DEVICE_RESET (1ULL << 2) 735 #define HL_NOTIFIER_EVENT_CS_TIMEOUT (1ULL << 3) 736 #define HL_NOTIFIER_EVENT_DEVICE_UNAVAILABLE (1ULL << 4) 737 #define HL_NOTIFIER_EVENT_USER_ENGINE_ERR (1ULL << 5) 738 #define HL_NOTIFIER_EVENT_GENERAL_HW_ERR (1ULL << 6) 739 #define HL_NOTIFIER_EVENT_RAZWI (1ULL << 7) 740 #define HL_NOTIFIER_EVENT_PAGE_FAULT (1ULL << 8) 741 #define HL_NOTIFIER_EVENT_CRITICL_HW_ERR (1ULL << 9) 742 #define HL_NOTIFIER_EVENT_CRITICL_FW_ERR (1ULL << 10) 743 744 /* Opcode for management ioctl 745 * 746 * HW_IP_INFO - Receive information about different IP blocks in the 747 * device. 748 * HL_INFO_HW_EVENTS - Receive an array describing how many times each event 749 * occurred since the last hard reset. 750 * HL_INFO_DRAM_USAGE - Retrieve the dram usage inside the device and of the 751 * specific context. This is relevant only for devices 752 * where the dram is managed by the kernel driver 753 * HL_INFO_HW_IDLE - Retrieve information about the idle status of each 754 * internal engine. 755 * HL_INFO_DEVICE_STATUS - Retrieve the device's status. This opcode doesn't 756 * require an open context. 757 * HL_INFO_DEVICE_UTILIZATION - Retrieve the total utilization of the device 758 * over the last period specified by the user. 759 * The period can be between 100ms to 1s, in 760 * resolution of 100ms. The return value is a 761 * percentage of the utilization rate. 762 * HL_INFO_HW_EVENTS_AGGREGATE - Receive an array describing how many times each 763 * event occurred since the driver was loaded. 764 * HL_INFO_CLK_RATE - Retrieve the current and maximum clock rate 765 * of the device in MHz. The maximum clock rate is 766 * configurable via sysfs parameter 767 * HL_INFO_RESET_COUNT - Retrieve the counts of the soft and hard reset 768 * operations performed on the device since the last 769 * time the driver was loaded. 770 * HL_INFO_TIME_SYNC - Retrieve the device's time alongside the host's time 771 * for synchronization. 772 * HL_INFO_CS_COUNTERS - Retrieve command submission counters 773 * HL_INFO_PCI_COUNTERS - Retrieve PCI counters 774 * HL_INFO_CLK_THROTTLE_REASON - Retrieve clock throttling reason 775 * HL_INFO_SYNC_MANAGER - Retrieve sync manager info per dcore 776 * HL_INFO_TOTAL_ENERGY - Retrieve total energy consumption 777 * HL_INFO_PLL_FREQUENCY - Retrieve PLL frequency 778 * HL_INFO_POWER - Retrieve power information 779 * HL_INFO_OPEN_STATS - Retrieve info regarding recent device open calls 780 * HL_INFO_DRAM_REPLACED_ROWS - Retrieve DRAM replaced rows info 781 * HL_INFO_DRAM_PENDING_ROWS - Retrieve DRAM pending rows num 782 * HL_INFO_LAST_ERR_OPEN_DEV_TIME - Retrieve timestamp of the last time the device was opened 783 * and CS timeout or razwi error occurred. 784 * HL_INFO_CS_TIMEOUT_EVENT - Retrieve CS timeout timestamp and its related CS sequence number. 785 * HL_INFO_RAZWI_EVENT - Retrieve parameters of razwi: 786 * Timestamp of razwi. 787 * The address which accessing it caused the razwi. 788 * Razwi initiator. 789 * Razwi cause, was it a page fault or MMU access error. 790 * HL_INFO_DEV_MEM_ALLOC_PAGE_SIZES - Retrieve valid page sizes for device memory allocation 791 * HL_INFO_SECURED_ATTESTATION - Retrieve attestation report of the boot. 792 * HL_INFO_REGISTER_EVENTFD - Register eventfd for event notifications. 793 * HL_INFO_UNREGISTER_EVENTFD - Unregister eventfd 794 * HL_INFO_GET_EVENTS - Retrieve the last occurred events 795 * HL_INFO_UNDEFINED_OPCODE_EVENT - Retrieve last undefined opcode error information. 796 * HL_INFO_ENGINE_STATUS - Retrieve the status of all the h/w engines in the asic. 797 * HL_INFO_PAGE_FAULT_EVENT - Retrieve parameters of captured page fault. 798 * HL_INFO_USER_MAPPINGS - Retrieve user mappings, captured after page fault event. 799 * HL_INFO_FW_GENERIC_REQ - Send generic request to FW. 800 * HL_INFO_HW_ERR_EVENT - Retrieve information on the reported HW error. 801 * HL_INFO_FW_ERR_EVENT - Retrieve information on the reported FW error. 802 */ 803 #define HL_INFO_HW_IP_INFO 0 804 #define HL_INFO_HW_EVENTS 1 805 #define HL_INFO_DRAM_USAGE 2 806 #define HL_INFO_HW_IDLE 3 807 #define HL_INFO_DEVICE_STATUS 4 808 #define HL_INFO_DEVICE_UTILIZATION 6 809 #define HL_INFO_HW_EVENTS_AGGREGATE 7 810 #define HL_INFO_CLK_RATE 8 811 #define HL_INFO_RESET_COUNT 9 812 #define HL_INFO_TIME_SYNC 10 813 #define HL_INFO_CS_COUNTERS 11 814 #define HL_INFO_PCI_COUNTERS 12 815 #define HL_INFO_CLK_THROTTLE_REASON 13 816 #define HL_INFO_SYNC_MANAGER 14 817 #define HL_INFO_TOTAL_ENERGY 15 818 #define HL_INFO_PLL_FREQUENCY 16 819 #define HL_INFO_POWER 17 820 #define HL_INFO_OPEN_STATS 18 821 #define HL_INFO_DRAM_REPLACED_ROWS 21 822 #define HL_INFO_DRAM_PENDING_ROWS 22 823 #define HL_INFO_LAST_ERR_OPEN_DEV_TIME 23 824 #define HL_INFO_CS_TIMEOUT_EVENT 24 825 #define HL_INFO_RAZWI_EVENT 25 826 #define HL_INFO_DEV_MEM_ALLOC_PAGE_SIZES 26 827 #define HL_INFO_SECURED_ATTESTATION 27 828 #define HL_INFO_REGISTER_EVENTFD 28 829 #define HL_INFO_UNREGISTER_EVENTFD 29 830 #define HL_INFO_GET_EVENTS 30 831 #define HL_INFO_UNDEFINED_OPCODE_EVENT 31 832 #define HL_INFO_ENGINE_STATUS 32 833 #define HL_INFO_PAGE_FAULT_EVENT 33 834 #define HL_INFO_USER_MAPPINGS 34 835 #define HL_INFO_FW_GENERIC_REQ 35 836 #define HL_INFO_HW_ERR_EVENT 36 837 #define HL_INFO_FW_ERR_EVENT 37 838 839 #define HL_INFO_VERSION_MAX_LEN 128 840 #define HL_INFO_CARD_NAME_MAX_LEN 16 841 842 /* Maximum buffer size for retrieving engines status */ 843 #define HL_ENGINES_DATA_MAX_SIZE SZ_1M 844 845 /** 846 * struct hl_info_hw_ip_info - hardware information on various IPs in the ASIC 847 * @sram_base_address: The first SRAM physical base address that is free to be 848 * used by the user. 849 * @dram_base_address: The first DRAM virtual or physical base address that is 850 * free to be used by the user. 851 * @dram_size: The DRAM size that is available to the user. 852 * @sram_size: The SRAM size that is available to the user. 853 * @num_of_events: The number of events that can be received from the f/w. This 854 * is needed so the user can what is the size of the h/w events 855 * array he needs to pass to the kernel when he wants to fetch 856 * the event counters. 857 * @device_id: PCI device ID of the ASIC. 858 * @module_id: Module ID of the ASIC for mezzanine cards in servers 859 * (From OCP spec). 860 * @decoder_enabled_mask: Bit-mask that represents which decoders are enabled. 861 * @first_available_interrupt_id: The first available interrupt ID for the user 862 * to be used when it works with user interrupts. 863 * Relevant for Gaudi2 and later. 864 * @server_type: Server type that the Gaudi ASIC is currently installed in. 865 * The value is according to enum hl_server_type 866 * @cpld_version: CPLD version on the board. 867 * @psoc_pci_pll_nr: PCI PLL NR value. Needed by the profiler in some ASICs. 868 * @psoc_pci_pll_nf: PCI PLL NF value. Needed by the profiler in some ASICs. 869 * @psoc_pci_pll_od: PCI PLL OD value. Needed by the profiler in some ASICs. 870 * @psoc_pci_pll_div_factor: PCI PLL DIV factor value. Needed by the profiler 871 * in some ASICs. 872 * @tpc_enabled_mask: Bit-mask that represents which TPCs are enabled. Relevant 873 * for Goya/Gaudi only. 874 * @dram_enabled: Whether the DRAM is enabled. 875 * @security_enabled: Whether security is enabled on device. 876 * @mme_master_slave_mode: Indicate whether the MME is working in master/slave 877 * configuration. Relevant for Greco and later. 878 * @cpucp_version: The CPUCP f/w version. 879 * @card_name: The card name as passed by the f/w. 880 * @tpc_enabled_mask_ext: Bit-mask that represents which TPCs are enabled. 881 * Relevant for Greco and later. 882 * @dram_page_size: The DRAM physical page size. 883 * @edma_enabled_mask: Bit-mask that represents which EDMAs are enabled. 884 * Relevant for Gaudi2 and later. 885 * @number_of_user_interrupts: The number of interrupts that are available to the userspace 886 * application to use. Relevant for Gaudi2 and later. 887 * @device_mem_alloc_default_page_size: default page size used in device memory allocation. 888 * @revision_id: PCI revision ID of the ASIC. 889 * @tpc_interrupt_id: interrupt id for TPC to use in order to raise events towards the host. 890 * @rotator_enabled_mask: Bit-mask that represents which rotators are enabled. 891 * Relevant for Gaudi3 and later. 892 * @engine_core_interrupt_reg_addr: interrupt register address for engine core to use 893 * in order to raise events toward FW. 894 * @reserved_dram_size: DRAM size reserved for driver and firmware. 895 */ 896 struct hl_info_hw_ip_info { 897 __u64 sram_base_address; 898 __u64 dram_base_address; 899 __u64 dram_size; 900 __u32 sram_size; 901 __u32 num_of_events; 902 __u32 device_id; 903 __u32 module_id; 904 __u32 decoder_enabled_mask; 905 __u16 first_available_interrupt_id; 906 __u16 server_type; 907 __u32 cpld_version; 908 __u32 psoc_pci_pll_nr; 909 __u32 psoc_pci_pll_nf; 910 __u32 psoc_pci_pll_od; 911 __u32 psoc_pci_pll_div_factor; 912 __u8 tpc_enabled_mask; 913 __u8 dram_enabled; 914 __u8 security_enabled; 915 __u8 mme_master_slave_mode; 916 __u8 cpucp_version[HL_INFO_VERSION_MAX_LEN]; 917 __u8 card_name[HL_INFO_CARD_NAME_MAX_LEN]; 918 __u64 tpc_enabled_mask_ext; 919 __u64 dram_page_size; 920 __u32 edma_enabled_mask; 921 __u16 number_of_user_interrupts; 922 __u8 reserved1; 923 __u8 reserved2; 924 __u64 reserved3; 925 __u64 device_mem_alloc_default_page_size; 926 __u64 reserved4; 927 __u64 reserved5; 928 __u32 reserved6; 929 __u8 reserved7; 930 __u8 revision_id; 931 __u16 tpc_interrupt_id; 932 __u32 rotator_enabled_mask; 933 __u32 reserved9; 934 __u64 engine_core_interrupt_reg_addr; 935 __u64 reserved_dram_size; 936 }; 937 938 struct hl_info_dram_usage { 939 __u64 dram_free_mem; 940 __u64 ctx_dram_mem; 941 }; 942 943 #define HL_BUSY_ENGINES_MASK_EXT_SIZE 4 944 945 struct hl_info_hw_idle { 946 __u32 is_idle; 947 /* 948 * Bitmask of busy engines. 949 * Bits definition is according to `enum <chip>_engine_id'. 950 */ 951 __u32 busy_engines_mask; 952 953 /* 954 * Extended Bitmask of busy engines. 955 * Bits definition is according to `enum <chip>_engine_id'. 956 */ 957 __u64 busy_engines_mask_ext[HL_BUSY_ENGINES_MASK_EXT_SIZE]; 958 }; 959 960 struct hl_info_device_status { 961 __u32 status; 962 __u32 pad; 963 }; 964 965 struct hl_info_device_utilization { 966 __u32 utilization; 967 __u32 pad; 968 }; 969 970 struct hl_info_clk_rate { 971 __u32 cur_clk_rate_mhz; 972 __u32 max_clk_rate_mhz; 973 }; 974 975 struct hl_info_reset_count { 976 __u32 hard_reset_cnt; 977 __u32 soft_reset_cnt; 978 }; 979 980 struct hl_info_time_sync { 981 __u64 device_time; 982 __u64 host_time; 983 }; 984 985 /** 986 * struct hl_info_pci_counters - pci counters 987 * @rx_throughput: PCI rx throughput KBps 988 * @tx_throughput: PCI tx throughput KBps 989 * @replay_cnt: PCI replay counter 990 */ 991 struct hl_info_pci_counters { 992 __u64 rx_throughput; 993 __u64 tx_throughput; 994 __u64 replay_cnt; 995 }; 996 997 enum hl_clk_throttling_type { 998 HL_CLK_THROTTLE_TYPE_POWER, 999 HL_CLK_THROTTLE_TYPE_THERMAL, 1000 HL_CLK_THROTTLE_TYPE_MAX 1001 }; 1002 1003 /* clk_throttling_reason masks */ 1004 #define HL_CLK_THROTTLE_POWER (1 << HL_CLK_THROTTLE_TYPE_POWER) 1005 #define HL_CLK_THROTTLE_THERMAL (1 << HL_CLK_THROTTLE_TYPE_THERMAL) 1006 1007 /** 1008 * struct hl_info_clk_throttle - clock throttling reason 1009 * @clk_throttling_reason: each bit represents a clk throttling reason 1010 * @clk_throttling_timestamp_us: represents CPU timestamp in microseconds of the start-event 1011 * @clk_throttling_duration_ns: the clock throttle time in nanosec 1012 */ 1013 struct hl_info_clk_throttle { 1014 __u32 clk_throttling_reason; 1015 __u32 pad; 1016 __u64 clk_throttling_timestamp_us[HL_CLK_THROTTLE_TYPE_MAX]; 1017 __u64 clk_throttling_duration_ns[HL_CLK_THROTTLE_TYPE_MAX]; 1018 }; 1019 1020 /** 1021 * struct hl_info_energy - device energy information 1022 * @total_energy_consumption: total device energy consumption 1023 */ 1024 struct hl_info_energy { 1025 __u64 total_energy_consumption; 1026 }; 1027 1028 #define HL_PLL_NUM_OUTPUTS 4 1029 1030 struct hl_pll_frequency_info { 1031 __u16 output[HL_PLL_NUM_OUTPUTS]; 1032 }; 1033 1034 /** 1035 * struct hl_open_stats_info - device open statistics information 1036 * @open_counter: ever growing counter, increased on each successful dev open 1037 * @last_open_period_ms: duration (ms) device was open last time 1038 * @is_compute_ctx_active: Whether there is an active compute context executing 1039 * @compute_ctx_in_release: true if the current compute context is being released 1040 */ 1041 struct hl_open_stats_info { 1042 __u64 open_counter; 1043 __u64 last_open_period_ms; 1044 __u8 is_compute_ctx_active; 1045 __u8 compute_ctx_in_release; 1046 __u8 pad[6]; 1047 }; 1048 1049 /** 1050 * struct hl_power_info - power information 1051 * @power: power consumption 1052 */ 1053 struct hl_power_info { 1054 __u64 power; 1055 }; 1056 1057 /** 1058 * struct hl_info_sync_manager - sync manager information 1059 * @first_available_sync_object: first available sob 1060 * @first_available_monitor: first available monitor 1061 * @first_available_cq: first available cq 1062 */ 1063 struct hl_info_sync_manager { 1064 __u32 first_available_sync_object; 1065 __u32 first_available_monitor; 1066 __u32 first_available_cq; 1067 __u32 reserved; 1068 }; 1069 1070 /** 1071 * struct hl_info_cs_counters - command submission counters 1072 * @total_out_of_mem_drop_cnt: total dropped due to memory allocation issue 1073 * @ctx_out_of_mem_drop_cnt: context dropped due to memory allocation issue 1074 * @total_parsing_drop_cnt: total dropped due to error in packet parsing 1075 * @ctx_parsing_drop_cnt: context dropped due to error in packet parsing 1076 * @total_queue_full_drop_cnt: total dropped due to queue full 1077 * @ctx_queue_full_drop_cnt: context dropped due to queue full 1078 * @total_device_in_reset_drop_cnt: total dropped due to device in reset 1079 * @ctx_device_in_reset_drop_cnt: context dropped due to device in reset 1080 * @total_max_cs_in_flight_drop_cnt: total dropped due to maximum CS in-flight 1081 * @ctx_max_cs_in_flight_drop_cnt: context dropped due to maximum CS in-flight 1082 * @total_validation_drop_cnt: total dropped due to validation error 1083 * @ctx_validation_drop_cnt: context dropped due to validation error 1084 */ 1085 struct hl_info_cs_counters { 1086 __u64 total_out_of_mem_drop_cnt; 1087 __u64 ctx_out_of_mem_drop_cnt; 1088 __u64 total_parsing_drop_cnt; 1089 __u64 ctx_parsing_drop_cnt; 1090 __u64 total_queue_full_drop_cnt; 1091 __u64 ctx_queue_full_drop_cnt; 1092 __u64 total_device_in_reset_drop_cnt; 1093 __u64 ctx_device_in_reset_drop_cnt; 1094 __u64 total_max_cs_in_flight_drop_cnt; 1095 __u64 ctx_max_cs_in_flight_drop_cnt; 1096 __u64 total_validation_drop_cnt; 1097 __u64 ctx_validation_drop_cnt; 1098 }; 1099 1100 /** 1101 * struct hl_info_last_err_open_dev_time - last error boot information. 1102 * @timestamp: timestamp of last time the device was opened and error occurred. 1103 */ 1104 struct hl_info_last_err_open_dev_time { 1105 __s64 timestamp; 1106 }; 1107 1108 /** 1109 * struct hl_info_cs_timeout_event - last CS timeout information. 1110 * @timestamp: timestamp when last CS timeout event occurred. 1111 * @seq: sequence number of last CS timeout event. 1112 */ 1113 struct hl_info_cs_timeout_event { 1114 __s64 timestamp; 1115 __u64 seq; 1116 }; 1117 1118 #define HL_RAZWI_NA_ENG_ID U16_MAX 1119 #define HL_RAZWI_MAX_NUM_OF_ENGINES_PER_RTR 128 1120 #define HL_RAZWI_READ BIT(0) 1121 #define HL_RAZWI_WRITE BIT(1) 1122 #define HL_RAZWI_LBW BIT(2) 1123 #define HL_RAZWI_HBW BIT(3) 1124 #define HL_RAZWI_RR BIT(4) 1125 #define HL_RAZWI_ADDR_DEC BIT(5) 1126 1127 /** 1128 * struct hl_info_razwi_event - razwi information. 1129 * @timestamp: timestamp of razwi. 1130 * @addr: address which accessing it caused razwi. 1131 * @engine_id: engine id of the razwi initiator, if it was initiated by engine that does not 1132 * have engine id it will be set to HL_RAZWI_NA_ENG_ID. If there are several possible 1133 * engines which caused the razwi, it will hold all of them. 1134 * @num_of_possible_engines: contains number of possible engine ids. In some asics, razwi indication 1135 * might be common for several engines and there is no way to get the 1136 * exact engine. In this way, engine_id array will be filled with all 1137 * possible engines caused this razwi. Also, there might be possibility 1138 * in gaudi, where we don't indication on specific engine, in that case 1139 * the value of this parameter will be zero. 1140 * @flags: bitmask for additional data: HL_RAZWI_READ - razwi caused by read operation 1141 * HL_RAZWI_WRITE - razwi caused by write operation 1142 * HL_RAZWI_LBW - razwi caused by lbw fabric transaction 1143 * HL_RAZWI_HBW - razwi caused by hbw fabric transaction 1144 * HL_RAZWI_RR - razwi caused by range register 1145 * HL_RAZWI_ADDR_DEC - razwi caused by address decode error 1146 * Note: this data is not supported by all asics, in that case the relevant bits will not 1147 * be set. 1148 */ 1149 struct hl_info_razwi_event { 1150 __s64 timestamp; 1151 __u64 addr; 1152 __u16 engine_id[HL_RAZWI_MAX_NUM_OF_ENGINES_PER_RTR]; 1153 __u16 num_of_possible_engines; 1154 __u8 flags; 1155 __u8 pad[5]; 1156 }; 1157 1158 #define MAX_QMAN_STREAMS_INFO 4 1159 #define OPCODE_INFO_MAX_ADDR_SIZE 8 1160 /** 1161 * struct hl_info_undefined_opcode_event - info about last undefined opcode error 1162 * @timestamp: timestamp of the undefined opcode error 1163 * @cb_addr_streams: CB addresses (per stream) that are currently exists in the PQ 1164 * entries. In case all streams array entries are 1165 * filled with values, it means the execution was in Lower-CP. 1166 * @cq_addr: the address of the current handled command buffer 1167 * @cq_size: the size of the current handled command buffer 1168 * @cb_addr_streams_len: num of streams - actual len of cb_addr_streams array. 1169 * should be equal to 1 in case of undefined opcode 1170 * in Upper-CP (specific stream) and equal to 4 incase 1171 * of undefined opcode in Lower-CP. 1172 * @engine_id: engine-id that the error occurred on 1173 * @stream_id: the stream id the error occurred on. In case the stream equals to 1174 * MAX_QMAN_STREAMS_INFO it means the error occurred on a Lower-CP. 1175 */ 1176 struct hl_info_undefined_opcode_event { 1177 __s64 timestamp; 1178 __u64 cb_addr_streams[MAX_QMAN_STREAMS_INFO][OPCODE_INFO_MAX_ADDR_SIZE]; 1179 __u64 cq_addr; 1180 __u32 cq_size; 1181 __u32 cb_addr_streams_len; 1182 __u32 engine_id; 1183 __u32 stream_id; 1184 }; 1185 1186 /** 1187 * struct hl_info_hw_err_event - info about HW error 1188 * @timestamp: timestamp of error occurrence 1189 * @event_id: The async event ID (specific to each device type). 1190 * @pad: size padding for u64 granularity. 1191 */ 1192 struct hl_info_hw_err_event { 1193 __s64 timestamp; 1194 __u16 event_id; 1195 __u16 pad[3]; 1196 }; 1197 1198 /* FW error definition for event_type in struct hl_info_fw_err_event */ 1199 enum hl_info_fw_err_type { 1200 HL_INFO_FW_HEARTBEAT_ERR, 1201 HL_INFO_FW_REPORTED_ERR, 1202 }; 1203 1204 /** 1205 * struct hl_info_fw_err_event - info about FW error 1206 * @timestamp: time-stamp of error occurrence 1207 * @err_type: The type of event as defined in hl_info_fw_err_type. 1208 * @event_id: The async event ID (specific to each device type, applicable only when event type is 1209 * HL_INFO_FW_REPORTED_ERR). 1210 * @pad: size padding for u64 granularity. 1211 */ 1212 struct hl_info_fw_err_event { 1213 __s64 timestamp; 1214 __u16 err_type; 1215 __u16 event_id; 1216 __u32 pad; 1217 }; 1218 1219 /** 1220 * struct hl_info_dev_memalloc_page_sizes - valid page sizes in device mem alloc information. 1221 * @page_order_bitmask: bitmap in which a set bit represents the order of the supported page size 1222 * (e.g. 0x2100000 means that 1MB and 32MB pages are supported). 1223 */ 1224 struct hl_info_dev_memalloc_page_sizes { 1225 __u64 page_order_bitmask; 1226 }; 1227 1228 #define SEC_PCR_DATA_BUF_SZ 256 1229 #define SEC_PCR_QUOTE_BUF_SZ 510 /* (512 - 2) 2 bytes used for size */ 1230 #define SEC_SIGNATURE_BUF_SZ 255 /* (256 - 1) 1 byte used for size */ 1231 #define SEC_PUB_DATA_BUF_SZ 510 /* (512 - 2) 2 bytes used for size */ 1232 #define SEC_CERTIFICATE_BUF_SZ 2046 /* (2048 - 2) 2 bytes used for size */ 1233 1234 /* 1235 * struct hl_info_sec_attest - attestation report of the boot 1236 * @nonce: number only used once. random number provided by host. this also passed to the quote 1237 * command as a qualifying data. 1238 * @pcr_quote_len: length of the attestation quote data (bytes) 1239 * @pub_data_len: length of the public data (bytes) 1240 * @certificate_len: length of the certificate (bytes) 1241 * @pcr_num_reg: number of PCR registers in the pcr_data array 1242 * @pcr_reg_len: length of each PCR register in the pcr_data array (bytes) 1243 * @quote_sig_len: length of the attestation report signature (bytes) 1244 * @pcr_data: raw values of the PCR registers 1245 * @pcr_quote: attestation report data structure 1246 * @quote_sig: signature structure of the attestation report 1247 * @public_data: public key for the signed attestation 1248 * (outPublic + name + qualifiedName) 1249 * @certificate: certificate for the attestation signing key 1250 */ 1251 struct hl_info_sec_attest { 1252 __u32 nonce; 1253 __u16 pcr_quote_len; 1254 __u16 pub_data_len; 1255 __u16 certificate_len; 1256 __u8 pcr_num_reg; 1257 __u8 pcr_reg_len; 1258 __u8 quote_sig_len; 1259 __u8 pcr_data[SEC_PCR_DATA_BUF_SZ]; 1260 __u8 pcr_quote[SEC_PCR_QUOTE_BUF_SZ]; 1261 __u8 quote_sig[SEC_SIGNATURE_BUF_SZ]; 1262 __u8 public_data[SEC_PUB_DATA_BUF_SZ]; 1263 __u8 certificate[SEC_CERTIFICATE_BUF_SZ]; 1264 __u8 pad0[2]; 1265 }; 1266 1267 /** 1268 * struct hl_page_fault_info - page fault information. 1269 * @timestamp: timestamp of page fault. 1270 * @addr: address which accessing it caused page fault. 1271 * @engine_id: engine id which caused the page fault, supported only in gaudi3. 1272 */ 1273 struct hl_page_fault_info { 1274 __s64 timestamp; 1275 __u64 addr; 1276 __u16 engine_id; 1277 __u8 pad[6]; 1278 }; 1279 1280 /** 1281 * struct hl_user_mapping - user mapping information. 1282 * @dev_va: device virtual address. 1283 * @size: virtual address mapping size. 1284 */ 1285 struct hl_user_mapping { 1286 __u64 dev_va; 1287 __u64 size; 1288 }; 1289 1290 enum gaudi_dcores { 1291 HL_GAUDI_WS_DCORE, 1292 HL_GAUDI_WN_DCORE, 1293 HL_GAUDI_EN_DCORE, 1294 HL_GAUDI_ES_DCORE 1295 }; 1296 1297 /** 1298 * struct hl_info_args - Main structure to retrieve device related information. 1299 * @return_pointer: User space address of the relevant structure related to HL_INFO_* operation 1300 * mentioned in @op. 1301 * @return_size: Size of the structure used in @return_pointer, just like "size" in "snprintf", it 1302 * limits how many bytes the kernel can write. For hw_events array, the size should be 1303 * hl_info_hw_ip_info.num_of_events * sizeof(__u32). 1304 * @op: Defines which type of information to be retrieved. Refer HL_INFO_* for details. 1305 * @dcore_id: DCORE id for which the information is relevant (for Gaudi refer to enum gaudi_dcores). 1306 * @ctx_id: Context ID of the user. Currently not in use. 1307 * @period_ms: Period value, in milliseconds, for utilization rate in range 100ms - 1000ms in 100 ms 1308 * resolution. Currently not in use. 1309 * @pll_index: Index as defined in hl_<asic type>_pll_index enumeration. 1310 * @eventfd: event file descriptor for event notifications. 1311 * @user_buffer_actual_size: Actual data size which was copied to user allocated buffer by the 1312 * driver. It is possible for the user to allocate buffer larger than 1313 * needed, hence updating this variable so user will know the exact amount 1314 * of bytes copied by the kernel to the buffer. 1315 * @sec_attest_nonce: Nonce number used for attestation report. 1316 * @array_size: Number of array members copied to user buffer. 1317 * Relevant for HL_INFO_USER_MAPPINGS info ioctl. 1318 * @fw_sub_opcode: generic requests sub opcodes. 1319 * @pad: Padding to 64 bit. 1320 */ 1321 struct hl_info_args { 1322 __u64 return_pointer; 1323 __u32 return_size; 1324 __u32 op; 1325 1326 union { 1327 __u32 dcore_id; 1328 __u32 ctx_id; 1329 __u32 period_ms; 1330 __u32 pll_index; 1331 __u32 eventfd; 1332 __u32 user_buffer_actual_size; 1333 __u32 sec_attest_nonce; 1334 __u32 array_size; 1335 __u32 fw_sub_opcode; 1336 }; 1337 1338 __u32 pad; 1339 }; 1340 1341 /* Opcode to create a new command buffer */ 1342 #define HL_CB_OP_CREATE 0 1343 /* Opcode to destroy previously created command buffer */ 1344 #define HL_CB_OP_DESTROY 1 1345 /* Opcode to retrieve information about a command buffer */ 1346 #define HL_CB_OP_INFO 2 1347 1348 /* 2MB minus 32 bytes for 2xMSG_PROT */ 1349 #define HL_MAX_CB_SIZE (0x200000 - 32) 1350 1351 /* Indicates whether the command buffer should be mapped to the device's MMU */ 1352 #define HL_CB_FLAGS_MAP 0x1 1353 1354 /* Used with HL_CB_OP_INFO opcode to get the device va address for kernel mapped CB */ 1355 #define HL_CB_FLAGS_GET_DEVICE_VA 0x2 1356 1357 struct hl_cb_in { 1358 /* Handle of CB or 0 if we want to create one */ 1359 __u64 cb_handle; 1360 /* HL_CB_OP_* */ 1361 __u32 op; 1362 1363 /* Size of CB. Maximum size is HL_MAX_CB_SIZE. The minimum size that 1364 * will be allocated, regardless of this parameter's value, is PAGE_SIZE 1365 */ 1366 __u32 cb_size; 1367 1368 /* Context ID - Currently not in use */ 1369 __u32 ctx_id; 1370 /* HL_CB_FLAGS_* */ 1371 __u32 flags; 1372 }; 1373 1374 struct hl_cb_out { 1375 union { 1376 /* Handle of CB */ 1377 __u64 cb_handle; 1378 1379 union { 1380 /* Information about CB */ 1381 struct { 1382 /* Usage count of CB */ 1383 __u32 usage_cnt; 1384 __u32 pad; 1385 }; 1386 1387 /* CB mapped address to device MMU */ 1388 __u64 device_va; 1389 }; 1390 }; 1391 }; 1392 1393 union hl_cb_args { 1394 struct hl_cb_in in; 1395 struct hl_cb_out out; 1396 }; 1397 1398 /* HL_CS_CHUNK_FLAGS_ values 1399 * 1400 * HL_CS_CHUNK_FLAGS_USER_ALLOC_CB: 1401 * Indicates if the CB was allocated and mapped by userspace 1402 * (relevant to greco and above). User allocated CB is a command buffer, 1403 * allocated by the user, via malloc (or similar). After allocating the 1404 * CB, the user invokes - “memory ioctl” to map the user memory into a 1405 * device virtual address. The user provides this address via the 1406 * cb_handle field. The interface provides the ability to create a 1407 * large CBs, Which aren’t limited to “HL_MAX_CB_SIZE”. Therefore, it 1408 * increases the PCI-DMA queues throughput. This CB allocation method 1409 * also reduces the use of Linux DMA-able memory pool. Which are limited 1410 * and used by other Linux sub-systems. 1411 */ 1412 #define HL_CS_CHUNK_FLAGS_USER_ALLOC_CB 0x1 1413 1414 /* 1415 * This structure size must always be fixed to 64-bytes for backward 1416 * compatibility 1417 */ 1418 struct hl_cs_chunk { 1419 union { 1420 /* Goya/Gaudi: 1421 * For external queue, this represents a Handle of CB on the 1422 * Host. 1423 * For internal queue in Goya, this represents an SRAM or 1424 * a DRAM address of the internal CB. In Gaudi, this might also 1425 * represent a mapped host address of the CB. 1426 * 1427 * Greco onwards: 1428 * For H/W queue, this represents either a Handle of CB on the 1429 * Host, or an SRAM, a DRAM, or a mapped host address of the CB. 1430 * 1431 * A mapped host address is in the device address space, after 1432 * a host address was mapped by the device MMU. 1433 */ 1434 __u64 cb_handle; 1435 1436 /* Relevant only when HL_CS_FLAGS_WAIT or 1437 * HL_CS_FLAGS_COLLECTIVE_WAIT is set 1438 * This holds address of array of u64 values that contain 1439 * signal CS sequence numbers. The wait described by 1440 * this job will listen on all those signals 1441 * (wait event per signal) 1442 */ 1443 __u64 signal_seq_arr; 1444 1445 /* 1446 * Relevant only when HL_CS_FLAGS_WAIT or 1447 * HL_CS_FLAGS_COLLECTIVE_WAIT is set 1448 * along with HL_CS_FLAGS_ENCAP_SIGNALS. 1449 * This is the CS sequence which has the encapsulated signals. 1450 */ 1451 __u64 encaps_signal_seq; 1452 }; 1453 1454 /* Index of queue to put the CB on */ 1455 __u32 queue_index; 1456 1457 union { 1458 /* 1459 * Size of command buffer with valid packets 1460 * Can be smaller then actual CB size 1461 */ 1462 __u32 cb_size; 1463 1464 /* Relevant only when HL_CS_FLAGS_WAIT or 1465 * HL_CS_FLAGS_COLLECTIVE_WAIT is set. 1466 * Number of entries in signal_seq_arr 1467 */ 1468 __u32 num_signal_seq_arr; 1469 1470 /* Relevant only when HL_CS_FLAGS_WAIT or 1471 * HL_CS_FLAGS_COLLECTIVE_WAIT is set along 1472 * with HL_CS_FLAGS_ENCAP_SIGNALS 1473 * This set the signals range that the user want to wait for 1474 * out of the whole reserved signals range. 1475 * e.g if the signals range is 20, and user don't want 1476 * to wait for signal 8, so he set this offset to 7, then 1477 * he call the API again with 9 and so on till 20. 1478 */ 1479 __u32 encaps_signal_offset; 1480 }; 1481 1482 /* HL_CS_CHUNK_FLAGS_* */ 1483 __u32 cs_chunk_flags; 1484 1485 /* Relevant only when HL_CS_FLAGS_COLLECTIVE_WAIT is set. 1486 * This holds the collective engine ID. The wait described by this job 1487 * will sync with this engine and with all NICs before completion. 1488 */ 1489 __u32 collective_engine_id; 1490 1491 /* Align structure to 64 bytes */ 1492 __u32 pad[10]; 1493 }; 1494 1495 /* SIGNAL/WAIT/COLLECTIVE_WAIT flags are mutually exclusive */ 1496 #define HL_CS_FLAGS_FORCE_RESTORE 0x1 1497 #define HL_CS_FLAGS_SIGNAL 0x2 1498 #define HL_CS_FLAGS_WAIT 0x4 1499 #define HL_CS_FLAGS_COLLECTIVE_WAIT 0x8 1500 1501 #define HL_CS_FLAGS_TIMESTAMP 0x20 1502 #define HL_CS_FLAGS_STAGED_SUBMISSION 0x40 1503 #define HL_CS_FLAGS_STAGED_SUBMISSION_FIRST 0x80 1504 #define HL_CS_FLAGS_STAGED_SUBMISSION_LAST 0x100 1505 #define HL_CS_FLAGS_CUSTOM_TIMEOUT 0x200 1506 #define HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT 0x400 1507 1508 /* 1509 * The encapsulated signals CS is merged into the existing CS ioctls. 1510 * In order to use this feature need to follow the below procedure: 1511 * 1. Reserve signals, set the CS type to HL_CS_FLAGS_RESERVE_SIGNALS_ONLY 1512 * the output of this API will be the SOB offset from CFG_BASE. 1513 * this address will be used to patch CB cmds to do the signaling for this 1514 * SOB by incrementing it's value. 1515 * for reverting the reservation use HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY 1516 * CS type, note that this might fail if out-of-sync happened to the SOB 1517 * value, in case other signaling request to the same SOB occurred between 1518 * reserve-unreserve calls. 1519 * 2. Use the staged CS to do the encapsulated signaling jobs. 1520 * use HL_CS_FLAGS_STAGED_SUBMISSION and HL_CS_FLAGS_STAGED_SUBMISSION_FIRST 1521 * along with HL_CS_FLAGS_ENCAP_SIGNALS flag, and set encaps_signal_offset 1522 * field. This offset allows app to wait on part of the reserved signals. 1523 * 3. Use WAIT/COLLECTIVE WAIT CS along with HL_CS_FLAGS_ENCAP_SIGNALS flag 1524 * to wait for the encapsulated signals. 1525 */ 1526 #define HL_CS_FLAGS_ENCAP_SIGNALS 0x800 1527 #define HL_CS_FLAGS_RESERVE_SIGNALS_ONLY 0x1000 1528 #define HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY 0x2000 1529 1530 /* 1531 * The engine cores CS is merged into the existing CS ioctls. 1532 * Use it to control the engine cores mode. 1533 */ 1534 #define HL_CS_FLAGS_ENGINE_CORE_COMMAND 0x4000 1535 1536 /* 1537 * The flush HBW PCI writes is merged into the existing CS ioctls. 1538 * Used to flush all HBW PCI writes. 1539 * This is a blocking operation and for this reason the user shall not use 1540 * the return sequence number (which will be invalid anyway) 1541 */ 1542 #define HL_CS_FLAGS_FLUSH_PCI_HBW_WRITES 0x8000 1543 1544 /* 1545 * The engines CS is merged into the existing CS ioctls. 1546 * Use it to control engines modes. 1547 */ 1548 #define HL_CS_FLAGS_ENGINES_COMMAND 0x10000 1549 1550 #define HL_CS_STATUS_SUCCESS 0 1551 1552 #define HL_MAX_JOBS_PER_CS 512 1553 1554 /* 1555 * enum hl_engine_command - engine command 1556 * 1557 * @HL_ENGINE_CORE_HALT: engine core halt 1558 * @HL_ENGINE_CORE_RUN: engine core run 1559 * @HL_ENGINE_STALL: user engine/s stall 1560 * @HL_ENGINE_RESUME: user engine/s resume 1561 */ 1562 enum hl_engine_command { 1563 HL_ENGINE_CORE_HALT = 1, 1564 HL_ENGINE_CORE_RUN = 2, 1565 HL_ENGINE_STALL = 3, 1566 HL_ENGINE_RESUME = 4, 1567 HL_ENGINE_COMMAND_MAX 1568 }; 1569 1570 struct hl_cs_in { 1571 1572 union { 1573 struct { 1574 /* this holds address of array of hl_cs_chunk for restore phase */ 1575 __u64 chunks_restore; 1576 1577 /* holds address of array of hl_cs_chunk for execution phase */ 1578 __u64 chunks_execute; 1579 }; 1580 1581 /* Valid only when HL_CS_FLAGS_ENGINE_CORE_COMMAND is set */ 1582 struct { 1583 /* this holds address of array of uint32 for engine_cores */ 1584 __u64 engine_cores; 1585 1586 /* number of engine cores in engine_cores array */ 1587 __u32 num_engine_cores; 1588 1589 /* the core command to be sent towards engine cores */ 1590 __u32 core_command; 1591 }; 1592 1593 /* Valid only when HL_CS_FLAGS_ENGINES_COMMAND is set */ 1594 struct { 1595 /* this holds address of array of uint32 for engines */ 1596 __u64 engines; 1597 1598 /* number of engines in engines array */ 1599 __u32 num_engines; 1600 1601 /* the engine command to be sent towards engines */ 1602 __u32 engine_command; 1603 }; 1604 }; 1605 1606 union { 1607 /* 1608 * Sequence number of a staged submission CS 1609 * valid only if HL_CS_FLAGS_STAGED_SUBMISSION is set and 1610 * HL_CS_FLAGS_STAGED_SUBMISSION_FIRST is unset. 1611 */ 1612 __u64 seq; 1613 1614 /* 1615 * Encapsulated signals handle id 1616 * Valid for two flows: 1617 * 1. CS with encapsulated signals: 1618 * when HL_CS_FLAGS_STAGED_SUBMISSION and 1619 * HL_CS_FLAGS_STAGED_SUBMISSION_FIRST 1620 * and HL_CS_FLAGS_ENCAP_SIGNALS are set. 1621 * 2. unreserve signals: 1622 * valid when HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY is set. 1623 */ 1624 __u32 encaps_sig_handle_id; 1625 1626 /* Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY is set */ 1627 struct { 1628 /* Encapsulated signals number */ 1629 __u32 encaps_signals_count; 1630 1631 /* Encapsulated signals queue index (stream) */ 1632 __u32 encaps_signals_q_idx; 1633 }; 1634 }; 1635 1636 /* Number of chunks in restore phase array. Maximum number is 1637 * HL_MAX_JOBS_PER_CS 1638 */ 1639 __u32 num_chunks_restore; 1640 1641 /* Number of chunks in execution array. Maximum number is 1642 * HL_MAX_JOBS_PER_CS 1643 */ 1644 __u32 num_chunks_execute; 1645 1646 /* timeout in seconds - valid only if HL_CS_FLAGS_CUSTOM_TIMEOUT 1647 * is set 1648 */ 1649 __u32 timeout; 1650 1651 /* HL_CS_FLAGS_* */ 1652 __u32 cs_flags; 1653 1654 /* Context ID - Currently not in use */ 1655 __u32 ctx_id; 1656 __u8 pad[4]; 1657 }; 1658 1659 struct hl_cs_out { 1660 union { 1661 /* 1662 * seq holds the sequence number of the CS to pass to wait 1663 * ioctl. All values are valid except for 0 and ULLONG_MAX 1664 */ 1665 __u64 seq; 1666 1667 /* Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY is set */ 1668 struct { 1669 /* This is the reserved signal handle id */ 1670 __u32 handle_id; 1671 1672 /* This is the signals count */ 1673 __u32 count; 1674 }; 1675 }; 1676 1677 /* HL_CS_STATUS */ 1678 __u32 status; 1679 1680 /* 1681 * SOB base address offset 1682 * Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY or HL_CS_FLAGS_SIGNAL is set 1683 */ 1684 __u32 sob_base_addr_offset; 1685 1686 /* 1687 * Count of completed signals in SOB before current signal submission. 1688 * Valid only when (HL_CS_FLAGS_ENCAP_SIGNALS & HL_CS_FLAGS_STAGED_SUBMISSION) 1689 * or HL_CS_FLAGS_SIGNAL is set 1690 */ 1691 __u16 sob_count_before_submission; 1692 __u16 pad[3]; 1693 }; 1694 1695 union hl_cs_args { 1696 struct hl_cs_in in; 1697 struct hl_cs_out out; 1698 }; 1699 1700 #define HL_WAIT_CS_FLAGS_INTERRUPT 0x2 1701 #define HL_WAIT_CS_FLAGS_INTERRUPT_MASK 0xFFF00000 1702 #define HL_WAIT_CS_FLAGS_ANY_CQ_INTERRUPT 0xFFF00000 1703 #define HL_WAIT_CS_FLAGS_ANY_DEC_INTERRUPT 0xFFE00000 1704 #define HL_WAIT_CS_FLAGS_MULTI_CS 0x4 1705 #define HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ 0x10 1706 #define HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT 0x20 1707 1708 #define HL_WAIT_MULTI_CS_LIST_MAX_LEN 32 1709 1710 struct hl_wait_cs_in { 1711 union { 1712 struct { 1713 /* 1714 * In case of wait_cs holds the CS sequence number. 1715 * In case of wait for multi CS hold a user pointer to 1716 * an array of CS sequence numbers 1717 */ 1718 __u64 seq; 1719 /* Absolute timeout to wait for command submission 1720 * in microseconds 1721 */ 1722 __u64 timeout_us; 1723 }; 1724 1725 struct { 1726 union { 1727 /* User address for completion comparison. 1728 * upon interrupt, driver will compare the value pointed 1729 * by this address with the supplied target value. 1730 * in order not to perform any comparison, set address 1731 * to all 1s. 1732 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT is set 1733 */ 1734 __u64 addr; 1735 1736 /* cq_counters_handle to a kernel mapped cb which contains 1737 * cq counters. 1738 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ is set 1739 */ 1740 __u64 cq_counters_handle; 1741 }; 1742 1743 /* Target value for completion comparison */ 1744 __u64 target; 1745 }; 1746 }; 1747 1748 /* Context ID - Currently not in use */ 1749 __u32 ctx_id; 1750 1751 /* HL_WAIT_CS_FLAGS_* 1752 * If HL_WAIT_CS_FLAGS_INTERRUPT is set, this field should include 1753 * interrupt id according to HL_WAIT_CS_FLAGS_INTERRUPT_MASK 1754 * 1755 * in order to wait for any CQ interrupt, set interrupt value to 1756 * HL_WAIT_CS_FLAGS_ANY_CQ_INTERRUPT. 1757 * 1758 * in order to wait for any decoder interrupt, set interrupt value to 1759 * HL_WAIT_CS_FLAGS_ANY_DEC_INTERRUPT. 1760 */ 1761 __u32 flags; 1762 1763 union { 1764 struct { 1765 /* Multi CS API info- valid entries in multi-CS array */ 1766 __u8 seq_arr_len; 1767 __u8 pad[7]; 1768 }; 1769 1770 /* Absolute timeout to wait for an interrupt in microseconds. 1771 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT is set 1772 */ 1773 __u64 interrupt_timeout_us; 1774 }; 1775 1776 /* 1777 * cq counter offset inside the counters cb pointed by cq_counters_handle above. 1778 * upon interrupt, driver will compare the value pointed 1779 * by this address (cq_counters_handle + cq_counters_offset) 1780 * with the supplied target value. 1781 * relevant only when HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ is set 1782 */ 1783 __u64 cq_counters_offset; 1784 1785 /* 1786 * Timestamp_handle timestamps buffer handle. 1787 * relevant only when HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT is set 1788 */ 1789 __u64 timestamp_handle; 1790 1791 /* 1792 * Timestamp_offset is offset inside the timestamp buffer pointed by timestamp_handle above. 1793 * upon interrupt, if the cq reached the target value then driver will write 1794 * timestamp to this offset. 1795 * relevant only when HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT is set 1796 */ 1797 __u64 timestamp_offset; 1798 }; 1799 1800 #define HL_WAIT_CS_STATUS_COMPLETED 0 1801 #define HL_WAIT_CS_STATUS_BUSY 1 1802 #define HL_WAIT_CS_STATUS_TIMEDOUT 2 1803 #define HL_WAIT_CS_STATUS_ABORTED 3 1804 1805 #define HL_WAIT_CS_STATUS_FLAG_GONE 0x1 1806 #define HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD 0x2 1807 1808 struct hl_wait_cs_out { 1809 /* HL_WAIT_CS_STATUS_* */ 1810 __u32 status; 1811 /* HL_WAIT_CS_STATUS_FLAG* */ 1812 __u32 flags; 1813 /* 1814 * valid only if HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD is set 1815 * for wait_cs: timestamp of CS completion 1816 * for wait_multi_cs: timestamp of FIRST CS completion 1817 */ 1818 __s64 timestamp_nsec; 1819 /* multi CS completion bitmap */ 1820 __u32 cs_completion_map; 1821 __u32 pad; 1822 }; 1823 1824 union hl_wait_cs_args { 1825 struct hl_wait_cs_in in; 1826 struct hl_wait_cs_out out; 1827 }; 1828 1829 /* Opcode to allocate device memory */ 1830 #define HL_MEM_OP_ALLOC 0 1831 1832 /* Opcode to free previously allocated device memory */ 1833 #define HL_MEM_OP_FREE 1 1834 1835 /* Opcode to map host and device memory */ 1836 #define HL_MEM_OP_MAP 2 1837 1838 /* Opcode to unmap previously mapped host and device memory */ 1839 #define HL_MEM_OP_UNMAP 3 1840 1841 /* Opcode to map a hw block */ 1842 #define HL_MEM_OP_MAP_BLOCK 4 1843 1844 /* Opcode to create DMA-BUF object for an existing device memory allocation 1845 * and to export an FD of that DMA-BUF back to the caller 1846 */ 1847 #define HL_MEM_OP_EXPORT_DMABUF_FD 5 1848 1849 /* Opcode to create timestamps pool for user interrupts registration support 1850 * The memory will be allocated by the kernel driver, A timestamp buffer which the user 1851 * will get handle to it for mmap, and another internal buffer used by the 1852 * driver for registration management 1853 * The memory will be freed when the user closes the file descriptor(ctx close) 1854 */ 1855 #define HL_MEM_OP_TS_ALLOC 6 1856 1857 /* Memory flags */ 1858 #define HL_MEM_CONTIGUOUS 0x1 1859 #define HL_MEM_SHARED 0x2 1860 #define HL_MEM_USERPTR 0x4 1861 #define HL_MEM_FORCE_HINT 0x8 1862 #define HL_MEM_PREFETCH 0x40 1863 1864 /** 1865 * structure hl_mem_in - structure that handle input args for memory IOCTL 1866 * @union arg: union of structures to be used based on the input operation 1867 * @op: specify the requested memory operation (one of the HL_MEM_OP_* definitions). 1868 * @flags: flags for the memory operation (one of the HL_MEM_* definitions). 1869 * For the HL_MEM_OP_EXPORT_DMABUF_FD opcode, this field holds the DMA-BUF file/FD flags. 1870 * @ctx_id: context ID - currently not in use. 1871 * @num_of_elements: number of timestamp elements used only with HL_MEM_OP_TS_ALLOC opcode. 1872 */ 1873 struct hl_mem_in { 1874 union { 1875 /** 1876 * structure for device memory allocation (used with the HL_MEM_OP_ALLOC op) 1877 * @mem_size: memory size to allocate 1878 * @page_size: page size to use on allocation. when the value is 0 the default page 1879 * size will be taken. 1880 */ 1881 struct { 1882 __u64 mem_size; 1883 __u64 page_size; 1884 } alloc; 1885 1886 /** 1887 * structure for free-ing device memory (used with the HL_MEM_OP_FREE op) 1888 * @handle: handle returned from HL_MEM_OP_ALLOC 1889 */ 1890 struct { 1891 __u64 handle; 1892 } free; 1893 1894 /** 1895 * structure for mapping device memory (used with the HL_MEM_OP_MAP op) 1896 * @hint_addr: requested virtual address of mapped memory. 1897 * the driver will try to map the requested region to this hint 1898 * address, as long as the address is valid and not already mapped. 1899 * the user should check the returned address of the IOCTL to make 1900 * sure he got the hint address. 1901 * passing 0 here means that the driver will choose the address itself. 1902 * @handle: handle returned from HL_MEM_OP_ALLOC. 1903 */ 1904 struct { 1905 __u64 hint_addr; 1906 __u64 handle; 1907 } map_device; 1908 1909 /** 1910 * structure for mapping host memory (used with the HL_MEM_OP_MAP op) 1911 * @host_virt_addr: address of allocated host memory. 1912 * @hint_addr: requested virtual address of mapped memory. 1913 * the driver will try to map the requested region to this hint 1914 * address, as long as the address is valid and not already mapped. 1915 * the user should check the returned address of the IOCTL to make 1916 * sure he got the hint address. 1917 * passing 0 here means that the driver will choose the address itself. 1918 * @size: size of allocated host memory. 1919 */ 1920 struct { 1921 __u64 host_virt_addr; 1922 __u64 hint_addr; 1923 __u64 mem_size; 1924 } map_host; 1925 1926 /** 1927 * structure for mapping hw block (used with the HL_MEM_OP_MAP_BLOCK op) 1928 * @block_addr:HW block address to map, a handle and size will be returned 1929 * to the user and will be used to mmap the relevant block. 1930 * only addresses from configuration space are allowed. 1931 */ 1932 struct { 1933 __u64 block_addr; 1934 } map_block; 1935 1936 /** 1937 * structure for unmapping host memory (used with the HL_MEM_OP_UNMAP op) 1938 * @device_virt_addr: virtual address returned from HL_MEM_OP_MAP 1939 */ 1940 struct { 1941 __u64 device_virt_addr; 1942 } unmap; 1943 1944 /** 1945 * structure for exporting DMABUF object (used with 1946 * the HL_MEM_OP_EXPORT_DMABUF_FD op) 1947 * @addr: for Gaudi1, the driver expects a physical address 1948 * inside the device's DRAM. this is because in Gaudi1 1949 * we don't have MMU that covers the device's DRAM. 1950 * for all other ASICs, the driver expects a device 1951 * virtual address that represents the start address of 1952 * a mapped DRAM memory area inside the device. 1953 * the address must be the same as was received from the 1954 * driver during a previous HL_MEM_OP_MAP operation. 1955 * @mem_size: size of memory to export. 1956 * @offset: for Gaudi1, this value must be 0. For all other ASICs, 1957 * the driver expects an offset inside of the memory area 1958 * describe by addr. the offset represents the start 1959 * address of that the exported dma-buf object describes. 1960 */ 1961 struct { 1962 __u64 addr; 1963 __u64 mem_size; 1964 __u64 offset; 1965 } export_dmabuf_fd; 1966 }; 1967 1968 __u32 op; 1969 __u32 flags; 1970 __u32 ctx_id; 1971 __u32 num_of_elements; 1972 }; 1973 1974 struct hl_mem_out { 1975 union { 1976 /* 1977 * Used for HL_MEM_OP_MAP as the virtual address that was 1978 * assigned in the device VA space. 1979 * A value of 0 means the requested operation failed. 1980 */ 1981 __u64 device_virt_addr; 1982 1983 /* 1984 * Used in HL_MEM_OP_ALLOC 1985 * This is the assigned handle for the allocated memory 1986 */ 1987 __u64 handle; 1988 1989 struct { 1990 /* 1991 * Used in HL_MEM_OP_MAP_BLOCK. 1992 * This is the assigned handle for the mapped block 1993 */ 1994 __u64 block_handle; 1995 1996 /* 1997 * Used in HL_MEM_OP_MAP_BLOCK 1998 * This is the size of the mapped block 1999 */ 2000 __u32 block_size; 2001 2002 __u32 pad; 2003 }; 2004 2005 /* Returned in HL_MEM_OP_EXPORT_DMABUF_FD. Represents the 2006 * DMA-BUF object that was created to describe a memory 2007 * allocation on the device's memory space. The FD should be 2008 * passed to the importer driver 2009 */ 2010 __s32 fd; 2011 }; 2012 }; 2013 2014 union hl_mem_args { 2015 struct hl_mem_in in; 2016 struct hl_mem_out out; 2017 }; 2018 2019 #define HL_DEBUG_MAX_AUX_VALUES 10 2020 2021 struct hl_debug_params_etr { 2022 /* Address in memory to allocate buffer */ 2023 __u64 buffer_address; 2024 2025 /* Size of buffer to allocate */ 2026 __u64 buffer_size; 2027 2028 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */ 2029 __u32 sink_mode; 2030 __u32 pad; 2031 }; 2032 2033 struct hl_debug_params_etf { 2034 /* Address in memory to allocate buffer */ 2035 __u64 buffer_address; 2036 2037 /* Size of buffer to allocate */ 2038 __u64 buffer_size; 2039 2040 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */ 2041 __u32 sink_mode; 2042 __u32 pad; 2043 }; 2044 2045 struct hl_debug_params_stm { 2046 /* Two bit masks for HW event and Stimulus Port */ 2047 __u64 he_mask; 2048 __u64 sp_mask; 2049 2050 /* Trace source ID */ 2051 __u32 id; 2052 2053 /* Frequency for the timestamp register */ 2054 __u32 frequency; 2055 }; 2056 2057 struct hl_debug_params_bmon { 2058 /* Two address ranges that the user can request to filter */ 2059 __u64 start_addr0; 2060 __u64 addr_mask0; 2061 2062 __u64 start_addr1; 2063 __u64 addr_mask1; 2064 2065 /* Capture window configuration */ 2066 __u32 bw_win; 2067 __u32 win_capture; 2068 2069 /* Trace source ID */ 2070 __u32 id; 2071 2072 /* Control register */ 2073 __u32 control; 2074 2075 /* Two more address ranges that the user can request to filter */ 2076 __u64 start_addr2; 2077 __u64 end_addr2; 2078 2079 __u64 start_addr3; 2080 __u64 end_addr3; 2081 }; 2082 2083 struct hl_debug_params_spmu { 2084 /* Event types selection */ 2085 __u64 event_types[HL_DEBUG_MAX_AUX_VALUES]; 2086 2087 /* Number of event types selection */ 2088 __u32 event_types_num; 2089 2090 /* TRC configuration register values */ 2091 __u32 pmtrc_val; 2092 __u32 trc_ctrl_host_val; 2093 __u32 trc_en_host_val; 2094 }; 2095 2096 /* Opcode for ETR component */ 2097 #define HL_DEBUG_OP_ETR 0 2098 /* Opcode for ETF component */ 2099 #define HL_DEBUG_OP_ETF 1 2100 /* Opcode for STM component */ 2101 #define HL_DEBUG_OP_STM 2 2102 /* Opcode for FUNNEL component */ 2103 #define HL_DEBUG_OP_FUNNEL 3 2104 /* Opcode for BMON component */ 2105 #define HL_DEBUG_OP_BMON 4 2106 /* Opcode for SPMU component */ 2107 #define HL_DEBUG_OP_SPMU 5 2108 /* Opcode for timestamp (deprecated) */ 2109 #define HL_DEBUG_OP_TIMESTAMP 6 2110 /* Opcode for setting the device into or out of debug mode. The enable 2111 * variable should be 1 for enabling debug mode and 0 for disabling it 2112 */ 2113 #define HL_DEBUG_OP_SET_MODE 7 2114 2115 struct hl_debug_args { 2116 /* 2117 * Pointer to user input structure. 2118 * This field is relevant to specific opcodes. 2119 */ 2120 __u64 input_ptr; 2121 /* Pointer to user output structure */ 2122 __u64 output_ptr; 2123 /* Size of user input structure */ 2124 __u32 input_size; 2125 /* Size of user output structure */ 2126 __u32 output_size; 2127 /* HL_DEBUG_OP_* */ 2128 __u32 op; 2129 /* 2130 * Register index in the component, taken from the debug_regs_index enum 2131 * in the various ASIC header files 2132 */ 2133 __u32 reg_idx; 2134 /* Enable/disable */ 2135 __u32 enable; 2136 /* Context ID - Currently not in use */ 2137 __u32 ctx_id; 2138 }; 2139 2140 /* 2141 * Various information operations such as: 2142 * - H/W IP information 2143 * - Current dram usage 2144 * 2145 * The user calls this IOCTL with an opcode that describes the required 2146 * information. The user should supply a pointer to a user-allocated memory 2147 * chunk, which will be filled by the driver with the requested information. 2148 * 2149 * The user supplies the maximum amount of size to copy into the user's memory, 2150 * in order to prevent data corruption in case of differences between the 2151 * definitions of structures in kernel and userspace, e.g. in case of old 2152 * userspace and new kernel driver 2153 */ 2154 #define HL_IOCTL_INFO \ 2155 _IOWR('H', 0x01, struct hl_info_args) 2156 2157 /* 2158 * Command Buffer 2159 * - Request a Command Buffer 2160 * - Destroy a Command Buffer 2161 * 2162 * The command buffers are memory blocks that reside in DMA-able address 2163 * space and are physically contiguous so they can be accessed by the device 2164 * directly. They are allocated using the coherent DMA API. 2165 * 2166 * When creating a new CB, the IOCTL returns a handle of it, and the user-space 2167 * process needs to use that handle to mmap the buffer so it can access them. 2168 * 2169 * In some instances, the device must access the command buffer through the 2170 * device's MMU, and thus its memory should be mapped. In these cases, user can 2171 * indicate the driver that such a mapping is required. 2172 * The resulting device virtual address will be used internally by the driver, 2173 * and won't be returned to user. 2174 * 2175 */ 2176 #define HL_IOCTL_CB \ 2177 _IOWR('H', 0x02, union hl_cb_args) 2178 2179 /* 2180 * Command Submission 2181 * 2182 * To submit work to the device, the user need to call this IOCTL with a set 2183 * of JOBS. That set of JOBS constitutes a CS object. 2184 * Each JOB will be enqueued on a specific queue, according to the user's input. 2185 * There can be more then one JOB per queue. 2186 * 2187 * The CS IOCTL will receive two sets of JOBS. One set is for "restore" phase 2188 * and a second set is for "execution" phase. 2189 * The JOBS on the "restore" phase are enqueued only after context-switch 2190 * (or if its the first CS for this context). The user can also order the 2191 * driver to run the "restore" phase explicitly 2192 * 2193 * Goya/Gaudi: 2194 * There are two types of queues - external and internal. External queues 2195 * are DMA queues which transfer data from/to the Host. All other queues are 2196 * internal. The driver will get completion notifications from the device only 2197 * on JOBS which are enqueued in the external queues. 2198 * 2199 * Greco onwards: 2200 * There is a single type of queue for all types of engines, either DMA engines 2201 * for transfers from/to the host or inside the device, or compute engines. 2202 * The driver will get completion notifications from the device for all queues. 2203 * 2204 * For jobs on external queues, the user needs to create command buffers 2205 * through the CB ioctl and give the CB's handle to the CS ioctl. For jobs on 2206 * internal queues, the user needs to prepare a "command buffer" with packets 2207 * on either the device SRAM/DRAM or the host, and give the device address of 2208 * that buffer to the CS ioctl. 2209 * For jobs on H/W queues both options of command buffers are valid. 2210 * 2211 * This IOCTL is asynchronous in regard to the actual execution of the CS. This 2212 * means it returns immediately after ALL the JOBS were enqueued on their 2213 * relevant queues. Therefore, the user mustn't assume the CS has been completed 2214 * or has even started to execute. 2215 * 2216 * Upon successful enqueue, the IOCTL returns a sequence number which the user 2217 * can use with the "Wait for CS" IOCTL to check whether the handle's CS 2218 * non-internal JOBS have been completed. Note that if the CS has internal JOBS 2219 * which can execute AFTER the external JOBS have finished, the driver might 2220 * report that the CS has finished executing BEFORE the internal JOBS have 2221 * actually finished executing. 2222 * 2223 * Even though the sequence number increments per CS, the user can NOT 2224 * automatically assume that if CS with sequence number N finished, then CS 2225 * with sequence number N-1 also finished. The user can make this assumption if 2226 * and only if CS N and CS N-1 are exactly the same (same CBs for the same 2227 * queues). 2228 */ 2229 #define HL_IOCTL_CS \ 2230 _IOWR('H', 0x03, union hl_cs_args) 2231 2232 /* 2233 * Wait for Command Submission 2234 * 2235 * The user can call this IOCTL with a handle it received from the CS IOCTL 2236 * to wait until the handle's CS has finished executing. The user will wait 2237 * inside the kernel until the CS has finished or until the user-requested 2238 * timeout has expired. 2239 * 2240 * If the timeout value is 0, the driver won't sleep at all. It will check 2241 * the status of the CS and return immediately 2242 * 2243 * The return value of the IOCTL is a standard Linux error code. The possible 2244 * values are: 2245 * 2246 * EINTR - Kernel waiting has been interrupted, e.g. due to OS signal 2247 * that the user process received 2248 * ETIMEDOUT - The CS has caused a timeout on the device 2249 * EIO - The CS was aborted (usually because the device was reset) 2250 * ENODEV - The device wants to do hard-reset (so user need to close FD) 2251 * 2252 * The driver also returns a custom define in case the IOCTL call returned 0. 2253 * The define can be one of the following: 2254 * 2255 * HL_WAIT_CS_STATUS_COMPLETED - The CS has been completed successfully (0) 2256 * HL_WAIT_CS_STATUS_BUSY - The CS is still executing (0) 2257 * HL_WAIT_CS_STATUS_TIMEDOUT - The CS has caused a timeout on the device 2258 * (ETIMEDOUT) 2259 * HL_WAIT_CS_STATUS_ABORTED - The CS was aborted, usually because the 2260 * device was reset (EIO) 2261 */ 2262 2263 #define HL_IOCTL_WAIT_CS \ 2264 _IOWR('H', 0x04, union hl_wait_cs_args) 2265 2266 /* 2267 * Memory 2268 * - Map host memory to device MMU 2269 * - Unmap host memory from device MMU 2270 * 2271 * This IOCTL allows the user to map host memory to the device MMU 2272 * 2273 * For host memory, the IOCTL doesn't allocate memory. The user is supposed 2274 * to allocate the memory in user-space (malloc/new). The driver pins the 2275 * physical pages (up to the allowed limit by the OS), assigns a virtual 2276 * address in the device VA space and initializes the device MMU. 2277 * 2278 * There is an option for the user to specify the requested virtual address. 2279 * 2280 */ 2281 #define HL_IOCTL_MEMORY \ 2282 _IOWR('H', 0x05, union hl_mem_args) 2283 2284 /* 2285 * Debug 2286 * - Enable/disable the ETR/ETF/FUNNEL/STM/BMON/SPMU debug traces 2287 * 2288 * This IOCTL allows the user to get debug traces from the chip. 2289 * 2290 * Before the user can send configuration requests of the various 2291 * debug/profile engines, it needs to set the device into debug mode. 2292 * This is because the debug/profile infrastructure is shared component in the 2293 * device and we can't allow multiple users to access it at the same time. 2294 * 2295 * Once a user set the device into debug mode, the driver won't allow other 2296 * users to "work" with the device, i.e. open a FD. If there are multiple users 2297 * opened on the device, the driver won't allow any user to debug the device. 2298 * 2299 * For each configuration request, the user needs to provide the register index 2300 * and essential data such as buffer address and size. 2301 * 2302 * Once the user has finished using the debug/profile engines, he should 2303 * set the device into non-debug mode, i.e. disable debug mode. 2304 * 2305 * The driver can decide to "kick out" the user if he abuses this interface. 2306 * 2307 */ 2308 #define HL_IOCTL_DEBUG \ 2309 _IOWR('H', 0x06, struct hl_debug_args) 2310 2311 #define HL_COMMAND_START 0x01 2312 #define HL_COMMAND_END 0x07 2313 2314 #endif /* HABANALABS_H_ */ 2315