1/* 2 * Copyright 2014 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 */ 22 23#include <linux/types.h> 24#include <linux/kernel.h> 25#include <linux/pci.h> 26#include <linux/errno.h> 27#include <linux/acpi.h> 28#include <linux/hash.h> 29#include <linux/cpufreq.h> 30#include <linux/log2.h> 31#include <linux/dmi.h> 32#include <linux/atomic.h> 33 34#include "kfd_priv.h" 35#include "kfd_crat.h" 36#include "kfd_topology.h" 37#include "kfd_device_queue_manager.h" 38#include "kfd_iommu.h" 39#include "amdgpu_amdkfd.h" 40#include "amdgpu_ras.h" 41 42/* topology_device_list - Master list of all topology devices */ 43static struct list_head topology_device_list; 44static struct kfd_system_properties sys_props; 45 46static DECLARE_RWSEM(topology_lock); 47static atomic_t topology_crat_proximity_domain; 48 49struct kfd_topology_device *kfd_topology_device_by_proximity_domain( 50 uint32_t proximity_domain) 51{ 52 struct kfd_topology_device *top_dev; 53 struct kfd_topology_device *device = NULL; 54 55 down_read(&topology_lock); 56 57 list_for_each_entry(top_dev, &topology_device_list, list) 58 if (top_dev->proximity_domain == proximity_domain) { 59 device = top_dev; 60 break; 61 } 62 63 up_read(&topology_lock); 64 65 return device; 66} 67 68struct kfd_topology_device *kfd_topology_device_by_id(uint32_t gpu_id) 69{ 70 struct kfd_topology_device *top_dev = NULL; 71 struct kfd_topology_device *ret = NULL; 72 73 down_read(&topology_lock); 74 75 list_for_each_entry(top_dev, &topology_device_list, list) 76 if (top_dev->gpu_id == gpu_id) { 77 ret = top_dev; 78 break; 79 } 80 81 up_read(&topology_lock); 82 83 return ret; 84} 85 86struct kfd_dev *kfd_device_by_id(uint32_t gpu_id) 87{ 88 struct kfd_topology_device *top_dev; 89 90 top_dev = kfd_topology_device_by_id(gpu_id); 91 if (!top_dev) 92 return NULL; 93 94 return top_dev->gpu; 95} 96 97struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev) 98{ 99 struct kfd_topology_device *top_dev; 100 struct kfd_dev *device = NULL; 101 102 down_read(&topology_lock); 103 104 list_for_each_entry(top_dev, &topology_device_list, list) 105 if (top_dev->gpu && top_dev->gpu->pdev == pdev) { 106 device = top_dev->gpu; 107 break; 108 } 109 110 up_read(&topology_lock); 111 112 return device; 113} 114 115struct kfd_dev *kfd_device_by_kgd(const struct kgd_dev *kgd) 116{ 117 struct kfd_topology_device *top_dev; 118 struct kfd_dev *device = NULL; 119 120 down_read(&topology_lock); 121 122 list_for_each_entry(top_dev, &topology_device_list, list) 123 if (top_dev->gpu && top_dev->gpu->kgd == kgd) { 124 device = top_dev->gpu; 125 break; 126 } 127 128 up_read(&topology_lock); 129 130 return device; 131} 132 133/* Called with write topology_lock acquired */ 134static void kfd_release_topology_device(struct kfd_topology_device *dev) 135{ 136 struct kfd_mem_properties *mem; 137 struct kfd_cache_properties *cache; 138 struct kfd_iolink_properties *iolink; 139 struct kfd_perf_properties *perf; 140 141 list_del(&dev->list); 142 143 while (dev->mem_props.next != &dev->mem_props) { 144 mem = container_of(dev->mem_props.next, 145 struct kfd_mem_properties, list); 146 list_del(&mem->list); 147 kfree(mem); 148 } 149 150 while (dev->cache_props.next != &dev->cache_props) { 151 cache = container_of(dev->cache_props.next, 152 struct kfd_cache_properties, list); 153 list_del(&cache->list); 154 kfree(cache); 155 } 156 157 while (dev->io_link_props.next != &dev->io_link_props) { 158 iolink = container_of(dev->io_link_props.next, 159 struct kfd_iolink_properties, list); 160 list_del(&iolink->list); 161 kfree(iolink); 162 } 163 164 while (dev->perf_props.next != &dev->perf_props) { 165 perf = container_of(dev->perf_props.next, 166 struct kfd_perf_properties, list); 167 list_del(&perf->list); 168 kfree(perf); 169 } 170 171 kfree(dev); 172} 173 174void kfd_release_topology_device_list(struct list_head *device_list) 175{ 176 struct kfd_topology_device *dev; 177 178 while (!list_empty(device_list)) { 179 dev = list_first_entry(device_list, 180 struct kfd_topology_device, list); 181 kfd_release_topology_device(dev); 182 } 183} 184 185static void kfd_release_live_view(void) 186{ 187 kfd_release_topology_device_list(&topology_device_list); 188 memset(&sys_props, 0, sizeof(sys_props)); 189} 190 191struct kfd_topology_device *kfd_create_topology_device( 192 struct list_head *device_list) 193{ 194 struct kfd_topology_device *dev; 195 196 dev = kfd_alloc_struct(dev); 197 if (!dev) { 198 pr_err("No memory to allocate a topology device"); 199 return NULL; 200 } 201 202 INIT_LIST_HEAD(&dev->mem_props); 203 INIT_LIST_HEAD(&dev->cache_props); 204 INIT_LIST_HEAD(&dev->io_link_props); 205 INIT_LIST_HEAD(&dev->perf_props); 206 207 list_add_tail(&dev->list, device_list); 208 209 return dev; 210} 211 212 213#define sysfs_show_gen_prop(buffer, offs, fmt, ...) \ 214 (offs += snprintf(buffer+offs, PAGE_SIZE-offs, \ 215 fmt, __VA_ARGS__)) 216#define sysfs_show_32bit_prop(buffer, offs, name, value) \ 217 sysfs_show_gen_prop(buffer, offs, "%s %u\n", name, value) 218#define sysfs_show_64bit_prop(buffer, offs, name, value) \ 219 sysfs_show_gen_prop(buffer, offs, "%s %llu\n", name, value) 220#define sysfs_show_32bit_val(buffer, offs, value) \ 221 sysfs_show_gen_prop(buffer, offs, "%u\n", value) 222#define sysfs_show_str_val(buffer, offs, value) \ 223 sysfs_show_gen_prop(buffer, offs, "%s\n", value) 224 225static ssize_t sysprops_show(struct kobject *kobj, struct attribute *attr, 226 char *buffer) 227{ 228 int offs = 0; 229 230 /* Making sure that the buffer is an empty string */ 231 buffer[0] = 0; 232 233 if (attr == &sys_props.attr_genid) { 234 sysfs_show_32bit_val(buffer, offs, 235 sys_props.generation_count); 236 } else if (attr == &sys_props.attr_props) { 237 sysfs_show_64bit_prop(buffer, offs, "platform_oem", 238 sys_props.platform_oem); 239 sysfs_show_64bit_prop(buffer, offs, "platform_id", 240 sys_props.platform_id); 241 sysfs_show_64bit_prop(buffer, offs, "platform_rev", 242 sys_props.platform_rev); 243 } else { 244 offs = -EINVAL; 245 } 246 247 return offs; 248} 249 250static void kfd_topology_kobj_release(struct kobject *kobj) 251{ 252 kfree(kobj); 253} 254 255static const struct sysfs_ops sysprops_ops = { 256 .show = sysprops_show, 257}; 258 259static struct kobj_type sysprops_type = { 260 .release = kfd_topology_kobj_release, 261 .sysfs_ops = &sysprops_ops, 262}; 263 264static ssize_t iolink_show(struct kobject *kobj, struct attribute *attr, 265 char *buffer) 266{ 267 int offs = 0; 268 struct kfd_iolink_properties *iolink; 269 270 /* Making sure that the buffer is an empty string */ 271 buffer[0] = 0; 272 273 iolink = container_of(attr, struct kfd_iolink_properties, attr); 274 if (iolink->gpu && kfd_devcgroup_check_permission(iolink->gpu)) 275 return -EPERM; 276 sysfs_show_32bit_prop(buffer, offs, "type", iolink->iolink_type); 277 sysfs_show_32bit_prop(buffer, offs, "version_major", iolink->ver_maj); 278 sysfs_show_32bit_prop(buffer, offs, "version_minor", iolink->ver_min); 279 sysfs_show_32bit_prop(buffer, offs, "node_from", iolink->node_from); 280 sysfs_show_32bit_prop(buffer, offs, "node_to", iolink->node_to); 281 sysfs_show_32bit_prop(buffer, offs, "weight", iolink->weight); 282 sysfs_show_32bit_prop(buffer, offs, "min_latency", iolink->min_latency); 283 sysfs_show_32bit_prop(buffer, offs, "max_latency", iolink->max_latency); 284 sysfs_show_32bit_prop(buffer, offs, "min_bandwidth", 285 iolink->min_bandwidth); 286 sysfs_show_32bit_prop(buffer, offs, "max_bandwidth", 287 iolink->max_bandwidth); 288 sysfs_show_32bit_prop(buffer, offs, "recommended_transfer_size", 289 iolink->rec_transfer_size); 290 sysfs_show_32bit_prop(buffer, offs, "flags", iolink->flags); 291 292 return offs; 293} 294 295static const struct sysfs_ops iolink_ops = { 296 .show = iolink_show, 297}; 298 299static struct kobj_type iolink_type = { 300 .release = kfd_topology_kobj_release, 301 .sysfs_ops = &iolink_ops, 302}; 303 304static ssize_t mem_show(struct kobject *kobj, struct attribute *attr, 305 char *buffer) 306{ 307 int offs = 0; 308 struct kfd_mem_properties *mem; 309 310 /* Making sure that the buffer is an empty string */ 311 buffer[0] = 0; 312 313 mem = container_of(attr, struct kfd_mem_properties, attr); 314 if (mem->gpu && kfd_devcgroup_check_permission(mem->gpu)) 315 return -EPERM; 316 sysfs_show_32bit_prop(buffer, offs, "heap_type", mem->heap_type); 317 sysfs_show_64bit_prop(buffer, offs, "size_in_bytes", 318 mem->size_in_bytes); 319 sysfs_show_32bit_prop(buffer, offs, "flags", mem->flags); 320 sysfs_show_32bit_prop(buffer, offs, "width", mem->width); 321 sysfs_show_32bit_prop(buffer, offs, "mem_clk_max", 322 mem->mem_clk_max); 323 324 return offs; 325} 326 327static const struct sysfs_ops mem_ops = { 328 .show = mem_show, 329}; 330 331static struct kobj_type mem_type = { 332 .release = kfd_topology_kobj_release, 333 .sysfs_ops = &mem_ops, 334}; 335 336static ssize_t kfd_cache_show(struct kobject *kobj, struct attribute *attr, 337 char *buffer) 338{ 339 int offs = 0; 340 uint32_t i, j; 341 struct kfd_cache_properties *cache; 342 343 /* Making sure that the buffer is an empty string */ 344 buffer[0] = 0; 345 346 cache = container_of(attr, struct kfd_cache_properties, attr); 347 if (cache->gpu && kfd_devcgroup_check_permission(cache->gpu)) 348 return -EPERM; 349 sysfs_show_32bit_prop(buffer, offs, "processor_id_low", 350 cache->processor_id_low); 351 sysfs_show_32bit_prop(buffer, offs, "level", cache->cache_level); 352 sysfs_show_32bit_prop(buffer, offs, "size", cache->cache_size); 353 sysfs_show_32bit_prop(buffer, offs, "cache_line_size", 354 cache->cacheline_size); 355 sysfs_show_32bit_prop(buffer, offs, "cache_lines_per_tag", 356 cache->cachelines_per_tag); 357 sysfs_show_32bit_prop(buffer, offs, "association", cache->cache_assoc); 358 sysfs_show_32bit_prop(buffer, offs, "latency", cache->cache_latency); 359 sysfs_show_32bit_prop(buffer, offs, "type", cache->cache_type); 360 offs += snprintf(buffer+offs, PAGE_SIZE-offs, "sibling_map "); 361 for (i = 0; i < CRAT_SIBLINGMAP_SIZE; i++) 362 for (j = 0; j < sizeof(cache->sibling_map[0])*8; j++) 363 /* Check each bit */ 364 offs += snprintf(buffer+offs, PAGE_SIZE-offs, "%d,", 365 (cache->sibling_map[i] >> j) & 1); 366 367 /* Replace the last "," with end of line */ 368 buffer[offs-1] = '\n'; 369 return offs; 370} 371 372static const struct sysfs_ops cache_ops = { 373 .show = kfd_cache_show, 374}; 375 376static struct kobj_type cache_type = { 377 .release = kfd_topology_kobj_release, 378 .sysfs_ops = &cache_ops, 379}; 380 381/****** Sysfs of Performance Counters ******/ 382 383struct kfd_perf_attr { 384 struct kobj_attribute attr; 385 uint32_t data; 386}; 387 388static ssize_t perf_show(struct kobject *kobj, struct kobj_attribute *attrs, 389 char *buf) 390{ 391 int offs = 0; 392 struct kfd_perf_attr *attr; 393 394 buf[0] = 0; 395 attr = container_of(attrs, struct kfd_perf_attr, attr); 396 if (!attr->data) /* invalid data for PMC */ 397 return 0; 398 else 399 return sysfs_show_32bit_val(buf, offs, attr->data); 400} 401 402#define KFD_PERF_DESC(_name, _data) \ 403{ \ 404 .attr = __ATTR(_name, 0444, perf_show, NULL), \ 405 .data = _data, \ 406} 407 408static struct kfd_perf_attr perf_attr_iommu[] = { 409 KFD_PERF_DESC(max_concurrent, 0), 410 KFD_PERF_DESC(num_counters, 0), 411 KFD_PERF_DESC(counter_ids, 0), 412}; 413/****************************************/ 414 415static ssize_t node_show(struct kobject *kobj, struct attribute *attr, 416 char *buffer) 417{ 418 int offs = 0; 419 struct kfd_topology_device *dev; 420 uint32_t log_max_watch_addr; 421 422 /* Making sure that the buffer is an empty string */ 423 buffer[0] = 0; 424 425 if (strcmp(attr->name, "gpu_id") == 0) { 426 dev = container_of(attr, struct kfd_topology_device, 427 attr_gpuid); 428 if (dev->gpu && kfd_devcgroup_check_permission(dev->gpu)) 429 return -EPERM; 430 return sysfs_show_32bit_val(buffer, offs, dev->gpu_id); 431 } 432 433 if (strcmp(attr->name, "name") == 0) { 434 dev = container_of(attr, struct kfd_topology_device, 435 attr_name); 436 437 if (dev->gpu && kfd_devcgroup_check_permission(dev->gpu)) 438 return -EPERM; 439 return sysfs_show_str_val(buffer, offs, dev->node_props.name); 440 } 441 442 dev = container_of(attr, struct kfd_topology_device, 443 attr_props); 444 if (dev->gpu && kfd_devcgroup_check_permission(dev->gpu)) 445 return -EPERM; 446 sysfs_show_32bit_prop(buffer, offs, "cpu_cores_count", 447 dev->node_props.cpu_cores_count); 448 sysfs_show_32bit_prop(buffer, offs, "simd_count", 449 dev->gpu ? dev->node_props.simd_count : 0); 450 sysfs_show_32bit_prop(buffer, offs, "mem_banks_count", 451 dev->node_props.mem_banks_count); 452 sysfs_show_32bit_prop(buffer, offs, "caches_count", 453 dev->node_props.caches_count); 454 sysfs_show_32bit_prop(buffer, offs, "io_links_count", 455 dev->node_props.io_links_count); 456 sysfs_show_32bit_prop(buffer, offs, "cpu_core_id_base", 457 dev->node_props.cpu_core_id_base); 458 sysfs_show_32bit_prop(buffer, offs, "simd_id_base", 459 dev->node_props.simd_id_base); 460 sysfs_show_32bit_prop(buffer, offs, "max_waves_per_simd", 461 dev->node_props.max_waves_per_simd); 462 sysfs_show_32bit_prop(buffer, offs, "lds_size_in_kb", 463 dev->node_props.lds_size_in_kb); 464 sysfs_show_32bit_prop(buffer, offs, "gds_size_in_kb", 465 dev->node_props.gds_size_in_kb); 466 sysfs_show_32bit_prop(buffer, offs, "num_gws", 467 dev->node_props.num_gws); 468 sysfs_show_32bit_prop(buffer, offs, "wave_front_size", 469 dev->node_props.wave_front_size); 470 sysfs_show_32bit_prop(buffer, offs, "array_count", 471 dev->node_props.array_count); 472 sysfs_show_32bit_prop(buffer, offs, "simd_arrays_per_engine", 473 dev->node_props.simd_arrays_per_engine); 474 sysfs_show_32bit_prop(buffer, offs, "cu_per_simd_array", 475 dev->node_props.cu_per_simd_array); 476 sysfs_show_32bit_prop(buffer, offs, "simd_per_cu", 477 dev->node_props.simd_per_cu); 478 sysfs_show_32bit_prop(buffer, offs, "max_slots_scratch_cu", 479 dev->node_props.max_slots_scratch_cu); 480 sysfs_show_32bit_prop(buffer, offs, "vendor_id", 481 dev->node_props.vendor_id); 482 sysfs_show_32bit_prop(buffer, offs, "device_id", 483 dev->node_props.device_id); 484 sysfs_show_32bit_prop(buffer, offs, "location_id", 485 dev->node_props.location_id); 486 sysfs_show_32bit_prop(buffer, offs, "domain", 487 dev->node_props.domain); 488 sysfs_show_32bit_prop(buffer, offs, "drm_render_minor", 489 dev->node_props.drm_render_minor); 490 sysfs_show_64bit_prop(buffer, offs, "hive_id", 491 dev->node_props.hive_id); 492 sysfs_show_32bit_prop(buffer, offs, "num_sdma_engines", 493 dev->node_props.num_sdma_engines); 494 sysfs_show_32bit_prop(buffer, offs, "num_sdma_xgmi_engines", 495 dev->node_props.num_sdma_xgmi_engines); 496 sysfs_show_32bit_prop(buffer, offs, "num_sdma_queues_per_engine", 497 dev->node_props.num_sdma_queues_per_engine); 498 sysfs_show_32bit_prop(buffer, offs, "num_cp_queues", 499 dev->node_props.num_cp_queues); 500 sysfs_show_64bit_prop(buffer, offs, "unique_id", 501 dev->node_props.unique_id); 502 503 if (dev->gpu) { 504 log_max_watch_addr = 505 __ilog2_u32(dev->gpu->device_info->num_of_watch_points); 506 507 if (log_max_watch_addr) { 508 dev->node_props.capability |= 509 HSA_CAP_WATCH_POINTS_SUPPORTED; 510 511 dev->node_props.capability |= 512 ((log_max_watch_addr << 513 HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT) & 514 HSA_CAP_WATCH_POINTS_TOTALBITS_MASK); 515 } 516 517 if (dev->gpu->device_info->asic_family == CHIP_TONGA) 518 dev->node_props.capability |= 519 HSA_CAP_AQL_QUEUE_DOUBLE_MAP; 520 521 sysfs_show_32bit_prop(buffer, offs, "max_engine_clk_fcompute", 522 dev->node_props.max_engine_clk_fcompute); 523 524 sysfs_show_64bit_prop(buffer, offs, "local_mem_size", 0ULL); 525 526 sysfs_show_32bit_prop(buffer, offs, "fw_version", 527 dev->gpu->mec_fw_version); 528 sysfs_show_32bit_prop(buffer, offs, "capability", 529 dev->node_props.capability); 530 sysfs_show_32bit_prop(buffer, offs, "sdma_fw_version", 531 dev->gpu->sdma_fw_version); 532 } 533 534 return sysfs_show_32bit_prop(buffer, offs, "max_engine_clk_ccompute", 535 cpufreq_quick_get_max(0)/1000); 536} 537 538static const struct sysfs_ops node_ops = { 539 .show = node_show, 540}; 541 542static struct kobj_type node_type = { 543 .release = kfd_topology_kobj_release, 544 .sysfs_ops = &node_ops, 545}; 546 547static void kfd_remove_sysfs_file(struct kobject *kobj, struct attribute *attr) 548{ 549 sysfs_remove_file(kobj, attr); 550 kobject_del(kobj); 551 kobject_put(kobj); 552} 553 554static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev) 555{ 556 struct kfd_iolink_properties *iolink; 557 struct kfd_cache_properties *cache; 558 struct kfd_mem_properties *mem; 559 struct kfd_perf_properties *perf; 560 561 if (dev->kobj_iolink) { 562 list_for_each_entry(iolink, &dev->io_link_props, list) 563 if (iolink->kobj) { 564 kfd_remove_sysfs_file(iolink->kobj, 565 &iolink->attr); 566 iolink->kobj = NULL; 567 } 568 kobject_del(dev->kobj_iolink); 569 kobject_put(dev->kobj_iolink); 570 dev->kobj_iolink = NULL; 571 } 572 573 if (dev->kobj_cache) { 574 list_for_each_entry(cache, &dev->cache_props, list) 575 if (cache->kobj) { 576 kfd_remove_sysfs_file(cache->kobj, 577 &cache->attr); 578 cache->kobj = NULL; 579 } 580 kobject_del(dev->kobj_cache); 581 kobject_put(dev->kobj_cache); 582 dev->kobj_cache = NULL; 583 } 584 585 if (dev->kobj_mem) { 586 list_for_each_entry(mem, &dev->mem_props, list) 587 if (mem->kobj) { 588 kfd_remove_sysfs_file(mem->kobj, &mem->attr); 589 mem->kobj = NULL; 590 } 591 kobject_del(dev->kobj_mem); 592 kobject_put(dev->kobj_mem); 593 dev->kobj_mem = NULL; 594 } 595 596 if (dev->kobj_perf) { 597 list_for_each_entry(perf, &dev->perf_props, list) { 598 kfree(perf->attr_group); 599 perf->attr_group = NULL; 600 } 601 kobject_del(dev->kobj_perf); 602 kobject_put(dev->kobj_perf); 603 dev->kobj_perf = NULL; 604 } 605 606 if (dev->kobj_node) { 607 sysfs_remove_file(dev->kobj_node, &dev->attr_gpuid); 608 sysfs_remove_file(dev->kobj_node, &dev->attr_name); 609 sysfs_remove_file(dev->kobj_node, &dev->attr_props); 610 kobject_del(dev->kobj_node); 611 kobject_put(dev->kobj_node); 612 dev->kobj_node = NULL; 613 } 614} 615 616static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev, 617 uint32_t id) 618{ 619 struct kfd_iolink_properties *iolink; 620 struct kfd_cache_properties *cache; 621 struct kfd_mem_properties *mem; 622 struct kfd_perf_properties *perf; 623 int ret; 624 uint32_t i, num_attrs; 625 struct attribute **attrs; 626 627 if (WARN_ON(dev->kobj_node)) 628 return -EEXIST; 629 630 /* 631 * Creating the sysfs folders 632 */ 633 dev->kobj_node = kfd_alloc_struct(dev->kobj_node); 634 if (!dev->kobj_node) 635 return -ENOMEM; 636 637 ret = kobject_init_and_add(dev->kobj_node, &node_type, 638 sys_props.kobj_nodes, "%d", id); 639 if (ret < 0) { 640 kobject_put(dev->kobj_node); 641 return ret; 642 } 643 644 dev->kobj_mem = kobject_create_and_add("mem_banks", dev->kobj_node); 645 if (!dev->kobj_mem) 646 return -ENOMEM; 647 648 dev->kobj_cache = kobject_create_and_add("caches", dev->kobj_node); 649 if (!dev->kobj_cache) 650 return -ENOMEM; 651 652 dev->kobj_iolink = kobject_create_and_add("io_links", dev->kobj_node); 653 if (!dev->kobj_iolink) 654 return -ENOMEM; 655 656 dev->kobj_perf = kobject_create_and_add("perf", dev->kobj_node); 657 if (!dev->kobj_perf) 658 return -ENOMEM; 659 660 /* 661 * Creating sysfs files for node properties 662 */ 663 dev->attr_gpuid.name = "gpu_id"; 664 dev->attr_gpuid.mode = KFD_SYSFS_FILE_MODE; 665 sysfs_attr_init(&dev->attr_gpuid); 666 dev->attr_name.name = "name"; 667 dev->attr_name.mode = KFD_SYSFS_FILE_MODE; 668 sysfs_attr_init(&dev->attr_name); 669 dev->attr_props.name = "properties"; 670 dev->attr_props.mode = KFD_SYSFS_FILE_MODE; 671 sysfs_attr_init(&dev->attr_props); 672 ret = sysfs_create_file(dev->kobj_node, &dev->attr_gpuid); 673 if (ret < 0) 674 return ret; 675 ret = sysfs_create_file(dev->kobj_node, &dev->attr_name); 676 if (ret < 0) 677 return ret; 678 ret = sysfs_create_file(dev->kobj_node, &dev->attr_props); 679 if (ret < 0) 680 return ret; 681 682 i = 0; 683 list_for_each_entry(mem, &dev->mem_props, list) { 684 mem->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); 685 if (!mem->kobj) 686 return -ENOMEM; 687 ret = kobject_init_and_add(mem->kobj, &mem_type, 688 dev->kobj_mem, "%d", i); 689 if (ret < 0) { 690 kobject_put(mem->kobj); 691 return ret; 692 } 693 694 mem->attr.name = "properties"; 695 mem->attr.mode = KFD_SYSFS_FILE_MODE; 696 sysfs_attr_init(&mem->attr); 697 ret = sysfs_create_file(mem->kobj, &mem->attr); 698 if (ret < 0) 699 return ret; 700 i++; 701 } 702 703 i = 0; 704 list_for_each_entry(cache, &dev->cache_props, list) { 705 cache->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); 706 if (!cache->kobj) 707 return -ENOMEM; 708 ret = kobject_init_and_add(cache->kobj, &cache_type, 709 dev->kobj_cache, "%d", i); 710 if (ret < 0) { 711 kobject_put(cache->kobj); 712 return ret; 713 } 714 715 cache->attr.name = "properties"; 716 cache->attr.mode = KFD_SYSFS_FILE_MODE; 717 sysfs_attr_init(&cache->attr); 718 ret = sysfs_create_file(cache->kobj, &cache->attr); 719 if (ret < 0) 720 return ret; 721 i++; 722 } 723 724 i = 0; 725 list_for_each_entry(iolink, &dev->io_link_props, list) { 726 iolink->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); 727 if (!iolink->kobj) 728 return -ENOMEM; 729 ret = kobject_init_and_add(iolink->kobj, &iolink_type, 730 dev->kobj_iolink, "%d", i); 731 if (ret < 0) { 732 kobject_put(iolink->kobj); 733 return ret; 734 } 735 736 iolink->attr.name = "properties"; 737 iolink->attr.mode = KFD_SYSFS_FILE_MODE; 738 sysfs_attr_init(&iolink->attr); 739 ret = sysfs_create_file(iolink->kobj, &iolink->attr); 740 if (ret < 0) 741 return ret; 742 i++; 743 } 744 745 /* All hardware blocks have the same number of attributes. */ 746 num_attrs = ARRAY_SIZE(perf_attr_iommu); 747 list_for_each_entry(perf, &dev->perf_props, list) { 748 perf->attr_group = kzalloc(sizeof(struct kfd_perf_attr) 749 * num_attrs + sizeof(struct attribute_group), 750 GFP_KERNEL); 751 if (!perf->attr_group) 752 return -ENOMEM; 753 754 attrs = (struct attribute **)(perf->attr_group + 1); 755 if (!strcmp(perf->block_name, "iommu")) { 756 /* Information of IOMMU's num_counters and counter_ids is shown 757 * under /sys/bus/event_source/devices/amd_iommu. We don't 758 * duplicate here. 759 */ 760 perf_attr_iommu[0].data = perf->max_concurrent; 761 for (i = 0; i < num_attrs; i++) 762 attrs[i] = &perf_attr_iommu[i].attr.attr; 763 } 764 perf->attr_group->name = perf->block_name; 765 perf->attr_group->attrs = attrs; 766 ret = sysfs_create_group(dev->kobj_perf, perf->attr_group); 767 if (ret < 0) 768 return ret; 769 } 770 771 return 0; 772} 773 774/* Called with write topology lock acquired */ 775static int kfd_build_sysfs_node_tree(void) 776{ 777 struct kfd_topology_device *dev; 778 int ret; 779 uint32_t i = 0; 780 781 list_for_each_entry(dev, &topology_device_list, list) { 782 ret = kfd_build_sysfs_node_entry(dev, i); 783 if (ret < 0) 784 return ret; 785 i++; 786 } 787 788 return 0; 789} 790 791/* Called with write topology lock acquired */ 792static void kfd_remove_sysfs_node_tree(void) 793{ 794 struct kfd_topology_device *dev; 795 796 list_for_each_entry(dev, &topology_device_list, list) 797 kfd_remove_sysfs_node_entry(dev); 798} 799 800static int kfd_topology_update_sysfs(void) 801{ 802 int ret; 803 804 if (!sys_props.kobj_topology) { 805 sys_props.kobj_topology = 806 kfd_alloc_struct(sys_props.kobj_topology); 807 if (!sys_props.kobj_topology) 808 return -ENOMEM; 809 810 ret = kobject_init_and_add(sys_props.kobj_topology, 811 &sysprops_type, &kfd_device->kobj, 812 "topology"); 813 if (ret < 0) { 814 kobject_put(sys_props.kobj_topology); 815 return ret; 816 } 817 818 sys_props.kobj_nodes = kobject_create_and_add("nodes", 819 sys_props.kobj_topology); 820 if (!sys_props.kobj_nodes) 821 return -ENOMEM; 822 823 sys_props.attr_genid.name = "generation_id"; 824 sys_props.attr_genid.mode = KFD_SYSFS_FILE_MODE; 825 sysfs_attr_init(&sys_props.attr_genid); 826 ret = sysfs_create_file(sys_props.kobj_topology, 827 &sys_props.attr_genid); 828 if (ret < 0) 829 return ret; 830 831 sys_props.attr_props.name = "system_properties"; 832 sys_props.attr_props.mode = KFD_SYSFS_FILE_MODE; 833 sysfs_attr_init(&sys_props.attr_props); 834 ret = sysfs_create_file(sys_props.kobj_topology, 835 &sys_props.attr_props); 836 if (ret < 0) 837 return ret; 838 } 839 840 kfd_remove_sysfs_node_tree(); 841 842 return kfd_build_sysfs_node_tree(); 843} 844 845static void kfd_topology_release_sysfs(void) 846{ 847 kfd_remove_sysfs_node_tree(); 848 if (sys_props.kobj_topology) { 849 sysfs_remove_file(sys_props.kobj_topology, 850 &sys_props.attr_genid); 851 sysfs_remove_file(sys_props.kobj_topology, 852 &sys_props.attr_props); 853 if (sys_props.kobj_nodes) { 854 kobject_del(sys_props.kobj_nodes); 855 kobject_put(sys_props.kobj_nodes); 856 sys_props.kobj_nodes = NULL; 857 } 858 kobject_del(sys_props.kobj_topology); 859 kobject_put(sys_props.kobj_topology); 860 sys_props.kobj_topology = NULL; 861 } 862} 863 864/* Called with write topology_lock acquired */ 865static void kfd_topology_update_device_list(struct list_head *temp_list, 866 struct list_head *master_list) 867{ 868 while (!list_empty(temp_list)) { 869 list_move_tail(temp_list->next, master_list); 870 sys_props.num_devices++; 871 } 872} 873 874static void kfd_debug_print_topology(void) 875{ 876 struct kfd_topology_device *dev; 877 878 down_read(&topology_lock); 879 880 dev = list_last_entry(&topology_device_list, 881 struct kfd_topology_device, list); 882 if (dev) { 883 if (dev->node_props.cpu_cores_count && 884 dev->node_props.simd_count) { 885 pr_info("Topology: Add APU node [0x%0x:0x%0x]\n", 886 dev->node_props.device_id, 887 dev->node_props.vendor_id); 888 } else if (dev->node_props.cpu_cores_count) 889 pr_info("Topology: Add CPU node\n"); 890 else if (dev->node_props.simd_count) 891 pr_info("Topology: Add dGPU node [0x%0x:0x%0x]\n", 892 dev->node_props.device_id, 893 dev->node_props.vendor_id); 894 } 895 up_read(&topology_lock); 896} 897 898/* Helper function for intializing platform_xx members of 899 * kfd_system_properties. Uses OEM info from the last CPU/APU node. 900 */ 901static void kfd_update_system_properties(void) 902{ 903 struct kfd_topology_device *dev; 904 905 down_read(&topology_lock); 906 dev = list_last_entry(&topology_device_list, 907 struct kfd_topology_device, list); 908 if (dev) { 909 sys_props.platform_id = 910 (*((uint64_t *)dev->oem_id)) & CRAT_OEMID_64BIT_MASK; 911 sys_props.platform_oem = *((uint64_t *)dev->oem_table_id); 912 sys_props.platform_rev = dev->oem_revision; 913 } 914 up_read(&topology_lock); 915} 916 917static void find_system_memory(const struct dmi_header *dm, 918 void *private) 919{ 920 struct kfd_mem_properties *mem; 921 u16 mem_width, mem_clock; 922 struct kfd_topology_device *kdev = 923 (struct kfd_topology_device *)private; 924 const u8 *dmi_data = (const u8 *)(dm + 1); 925 926 if (dm->type == DMI_ENTRY_MEM_DEVICE && dm->length >= 0x15) { 927 mem_width = (u16)(*(const u16 *)(dmi_data + 0x6)); 928 mem_clock = (u16)(*(const u16 *)(dmi_data + 0x11)); 929 list_for_each_entry(mem, &kdev->mem_props, list) { 930 if (mem_width != 0xFFFF && mem_width != 0) 931 mem->width = mem_width; 932 if (mem_clock != 0) 933 mem->mem_clk_max = mem_clock; 934 } 935 } 936} 937 938/* 939 * Performance counters information is not part of CRAT but we would like to 940 * put them in the sysfs under topology directory for Thunk to get the data. 941 * This function is called before updating the sysfs. 942 */ 943static int kfd_add_perf_to_topology(struct kfd_topology_device *kdev) 944{ 945 /* These are the only counters supported so far */ 946 return kfd_iommu_add_perf_counters(kdev); 947} 948 949/* kfd_add_non_crat_information - Add information that is not currently 950 * defined in CRAT but is necessary for KFD topology 951 * @dev - topology device to which addition info is added 952 */ 953static void kfd_add_non_crat_information(struct kfd_topology_device *kdev) 954{ 955 /* Check if CPU only node. */ 956 if (!kdev->gpu) { 957 /* Add system memory information */ 958 dmi_walk(find_system_memory, kdev); 959 } 960 /* TODO: For GPU node, rearrange code from kfd_topology_add_device */ 961} 962 963/* kfd_is_acpi_crat_invalid - CRAT from ACPI is valid only for AMD APU devices. 964 * Ignore CRAT for all other devices. AMD APU is identified if both CPU 965 * and GPU cores are present. 966 * @device_list - topology device list created by parsing ACPI CRAT table. 967 * @return - TRUE if invalid, FALSE is valid. 968 */ 969static bool kfd_is_acpi_crat_invalid(struct list_head *device_list) 970{ 971 struct kfd_topology_device *dev; 972 973 list_for_each_entry(dev, device_list, list) { 974 if (dev->node_props.cpu_cores_count && 975 dev->node_props.simd_count) 976 return false; 977 } 978 pr_info("Ignoring ACPI CRAT on non-APU system\n"); 979 return true; 980} 981 982int kfd_topology_init(void) 983{ 984 void *crat_image = NULL; 985 size_t image_size = 0; 986 int ret; 987 struct list_head temp_topology_device_list; 988 int cpu_only_node = 0; 989 struct kfd_topology_device *kdev; 990 int proximity_domain; 991 992 /* topology_device_list - Master list of all topology devices 993 * temp_topology_device_list - temporary list created while parsing CRAT 994 * or VCRAT. Once parsing is complete the contents of list is moved to 995 * topology_device_list 996 */ 997 998 /* Initialize the head for the both the lists */ 999 INIT_LIST_HEAD(&topology_device_list); 1000 INIT_LIST_HEAD(&temp_topology_device_list); 1001 init_rwsem(&topology_lock); 1002 1003 memset(&sys_props, 0, sizeof(sys_props)); 1004 1005 /* Proximity domains in ACPI CRAT tables start counting at 1006 * 0. The same should be true for virtual CRAT tables created 1007 * at this stage. GPUs added later in kfd_topology_add_device 1008 * use a counter. 1009 */ 1010 proximity_domain = 0; 1011 1012 /* 1013 * Get the CRAT image from the ACPI. If ACPI doesn't have one 1014 * or if ACPI CRAT is invalid create a virtual CRAT. 1015 * NOTE: The current implementation expects all AMD APUs to have 1016 * CRAT. If no CRAT is available, it is assumed to be a CPU 1017 */ 1018 ret = kfd_create_crat_image_acpi(&crat_image, &image_size); 1019 if (!ret) { 1020 ret = kfd_parse_crat_table(crat_image, 1021 &temp_topology_device_list, 1022 proximity_domain); 1023 if (ret || 1024 kfd_is_acpi_crat_invalid(&temp_topology_device_list)) { 1025 kfd_release_topology_device_list( 1026 &temp_topology_device_list); 1027 kfd_destroy_crat_image(crat_image); 1028 crat_image = NULL; 1029 } 1030 } 1031 1032 if (!crat_image) { 1033 ret = kfd_create_crat_image_virtual(&crat_image, &image_size, 1034 COMPUTE_UNIT_CPU, NULL, 1035 proximity_domain); 1036 cpu_only_node = 1; 1037 if (ret) { 1038 pr_err("Error creating VCRAT table for CPU\n"); 1039 return ret; 1040 } 1041 1042 ret = kfd_parse_crat_table(crat_image, 1043 &temp_topology_device_list, 1044 proximity_domain); 1045 if (ret) { 1046 pr_err("Error parsing VCRAT table for CPU\n"); 1047 goto err; 1048 } 1049 } 1050 1051 kdev = list_first_entry(&temp_topology_device_list, 1052 struct kfd_topology_device, list); 1053 kfd_add_perf_to_topology(kdev); 1054 1055 down_write(&topology_lock); 1056 kfd_topology_update_device_list(&temp_topology_device_list, 1057 &topology_device_list); 1058 atomic_set(&topology_crat_proximity_domain, sys_props.num_devices-1); 1059 ret = kfd_topology_update_sysfs(); 1060 up_write(&topology_lock); 1061 1062 if (!ret) { 1063 sys_props.generation_count++; 1064 kfd_update_system_properties(); 1065 kfd_debug_print_topology(); 1066 } else 1067 pr_err("Failed to update topology in sysfs ret=%d\n", ret); 1068 1069 /* For nodes with GPU, this information gets added 1070 * when GPU is detected (kfd_topology_add_device). 1071 */ 1072 if (cpu_only_node) { 1073 /* Add additional information to CPU only node created above */ 1074 down_write(&topology_lock); 1075 kdev = list_first_entry(&topology_device_list, 1076 struct kfd_topology_device, list); 1077 up_write(&topology_lock); 1078 kfd_add_non_crat_information(kdev); 1079 } 1080 1081err: 1082 kfd_destroy_crat_image(crat_image); 1083 return ret; 1084} 1085 1086void kfd_topology_shutdown(void) 1087{ 1088 down_write(&topology_lock); 1089 kfd_topology_release_sysfs(); 1090 kfd_release_live_view(); 1091 up_write(&topology_lock); 1092} 1093 1094static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu) 1095{ 1096 uint32_t hashout; 1097 uint32_t buf[7]; 1098 uint64_t local_mem_size; 1099 int i; 1100 struct kfd_local_mem_info local_mem_info; 1101 1102 if (!gpu) 1103 return 0; 1104 1105 amdgpu_amdkfd_get_local_mem_info(gpu->kgd, &local_mem_info); 1106 1107 local_mem_size = local_mem_info.local_mem_size_private + 1108 local_mem_info.local_mem_size_public; 1109 1110 buf[0] = gpu->pdev->devfn; 1111 buf[1] = gpu->pdev->subsystem_vendor | 1112 (gpu->pdev->subsystem_device << 16); 1113 buf[2] = pci_domain_nr(gpu->pdev->bus); 1114 buf[3] = gpu->pdev->device; 1115 buf[4] = gpu->pdev->bus->number; 1116 buf[5] = lower_32_bits(local_mem_size); 1117 buf[6] = upper_32_bits(local_mem_size); 1118 1119 for (i = 0, hashout = 0; i < 7; i++) 1120 hashout ^= hash_32(buf[i], KFD_GPU_ID_HASH_WIDTH); 1121 1122 return hashout; 1123} 1124/* kfd_assign_gpu - Attach @gpu to the correct kfd topology device. If 1125 * the GPU device is not already present in the topology device 1126 * list then return NULL. This means a new topology device has to 1127 * be created for this GPU. 1128 */ 1129static struct kfd_topology_device *kfd_assign_gpu(struct kfd_dev *gpu) 1130{ 1131 struct kfd_topology_device *dev; 1132 struct kfd_topology_device *out_dev = NULL; 1133 struct kfd_mem_properties *mem; 1134 struct kfd_cache_properties *cache; 1135 struct kfd_iolink_properties *iolink; 1136 1137 down_write(&topology_lock); 1138 list_for_each_entry(dev, &topology_device_list, list) { 1139 /* Discrete GPUs need their own topology device list 1140 * entries. Don't assign them to CPU/APU nodes. 1141 */ 1142 if (!gpu->use_iommu_v2 && 1143 dev->node_props.cpu_cores_count) 1144 continue; 1145 1146 if (!dev->gpu && (dev->node_props.simd_count > 0)) { 1147 dev->gpu = gpu; 1148 out_dev = dev; 1149 1150 list_for_each_entry(mem, &dev->mem_props, list) 1151 mem->gpu = dev->gpu; 1152 list_for_each_entry(cache, &dev->cache_props, list) 1153 cache->gpu = dev->gpu; 1154 list_for_each_entry(iolink, &dev->io_link_props, list) 1155 iolink->gpu = dev->gpu; 1156 break; 1157 } 1158 } 1159 up_write(&topology_lock); 1160 return out_dev; 1161} 1162 1163static void kfd_notify_gpu_change(uint32_t gpu_id, int arrival) 1164{ 1165 /* 1166 * TODO: Generate an event for thunk about the arrival/removal 1167 * of the GPU 1168 */ 1169} 1170 1171/* kfd_fill_mem_clk_max_info - Since CRAT doesn't have memory clock info, 1172 * patch this after CRAT parsing. 1173 */ 1174static void kfd_fill_mem_clk_max_info(struct kfd_topology_device *dev) 1175{ 1176 struct kfd_mem_properties *mem; 1177 struct kfd_local_mem_info local_mem_info; 1178 1179 if (!dev) 1180 return; 1181 1182 /* Currently, amdgpu driver (amdgpu_mc) deals only with GPUs with 1183 * single bank of VRAM local memory. 1184 * for dGPUs - VCRAT reports only one bank of Local Memory 1185 * for APUs - If CRAT from ACPI reports more than one bank, then 1186 * all the banks will report the same mem_clk_max information 1187 */ 1188 amdgpu_amdkfd_get_local_mem_info(dev->gpu->kgd, &local_mem_info); 1189 1190 list_for_each_entry(mem, &dev->mem_props, list) 1191 mem->mem_clk_max = local_mem_info.mem_clk_max; 1192} 1193 1194static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev) 1195{ 1196 struct kfd_iolink_properties *link, *cpu_link; 1197 struct kfd_topology_device *cpu_dev; 1198 uint32_t cap; 1199 uint32_t cpu_flag = CRAT_IOLINK_FLAGS_ENABLED; 1200 uint32_t flag = CRAT_IOLINK_FLAGS_ENABLED; 1201 1202 if (!dev || !dev->gpu) 1203 return; 1204 1205 pcie_capability_read_dword(dev->gpu->pdev, 1206 PCI_EXP_DEVCAP2, &cap); 1207 1208 if (!(cap & (PCI_EXP_DEVCAP2_ATOMIC_COMP32 | 1209 PCI_EXP_DEVCAP2_ATOMIC_COMP64))) 1210 cpu_flag |= CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT | 1211 CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT; 1212 1213 if (!dev->gpu->pci_atomic_requested || 1214 dev->gpu->device_info->asic_family == CHIP_HAWAII) 1215 flag |= CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT | 1216 CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT; 1217 1218 /* GPU only creates direct links so apply flags setting to all */ 1219 list_for_each_entry(link, &dev->io_link_props, list) { 1220 link->flags = flag; 1221 cpu_dev = kfd_topology_device_by_proximity_domain( 1222 link->node_to); 1223 if (cpu_dev) { 1224 list_for_each_entry(cpu_link, 1225 &cpu_dev->io_link_props, list) 1226 if (cpu_link->node_to == link->node_from) 1227 cpu_link->flags = cpu_flag; 1228 } 1229 } 1230} 1231 1232int kfd_topology_add_device(struct kfd_dev *gpu) 1233{ 1234 uint32_t gpu_id; 1235 struct kfd_topology_device *dev; 1236 struct kfd_cu_info cu_info; 1237 int res = 0; 1238 struct list_head temp_topology_device_list; 1239 void *crat_image = NULL; 1240 size_t image_size = 0; 1241 int proximity_domain; 1242 struct amdgpu_device *adev; 1243 1244 INIT_LIST_HEAD(&temp_topology_device_list); 1245 1246 gpu_id = kfd_generate_gpu_id(gpu); 1247 1248 pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id); 1249 1250 proximity_domain = atomic_inc_return(&topology_crat_proximity_domain); 1251 1252 /* Check to see if this gpu device exists in the topology_device_list. 1253 * If so, assign the gpu to that device, 1254 * else create a Virtual CRAT for this gpu device and then parse that 1255 * CRAT to create a new topology device. Once created assign the gpu to 1256 * that topology device 1257 */ 1258 dev = kfd_assign_gpu(gpu); 1259 if (!dev) { 1260 res = kfd_create_crat_image_virtual(&crat_image, &image_size, 1261 COMPUTE_UNIT_GPU, gpu, 1262 proximity_domain); 1263 if (res) { 1264 pr_err("Error creating VCRAT for GPU (ID: 0x%x)\n", 1265 gpu_id); 1266 return res; 1267 } 1268 res = kfd_parse_crat_table(crat_image, 1269 &temp_topology_device_list, 1270 proximity_domain); 1271 if (res) { 1272 pr_err("Error parsing VCRAT for GPU (ID: 0x%x)\n", 1273 gpu_id); 1274 goto err; 1275 } 1276 1277 down_write(&topology_lock); 1278 kfd_topology_update_device_list(&temp_topology_device_list, 1279 &topology_device_list); 1280 1281 /* Update the SYSFS tree, since we added another topology 1282 * device 1283 */ 1284 res = kfd_topology_update_sysfs(); 1285 up_write(&topology_lock); 1286 1287 if (!res) 1288 sys_props.generation_count++; 1289 else 1290 pr_err("Failed to update GPU (ID: 0x%x) to sysfs topology. res=%d\n", 1291 gpu_id, res); 1292 dev = kfd_assign_gpu(gpu); 1293 if (WARN_ON(!dev)) { 1294 res = -ENODEV; 1295 goto err; 1296 } 1297 } 1298 1299 dev->gpu_id = gpu_id; 1300 gpu->id = gpu_id; 1301 1302 /* TODO: Move the following lines to function 1303 * kfd_add_non_crat_information 1304 */ 1305 1306 /* Fill-in additional information that is not available in CRAT but 1307 * needed for the topology 1308 */ 1309 1310 amdgpu_amdkfd_get_cu_info(dev->gpu->kgd, &cu_info); 1311 1312 strncpy(dev->node_props.name, gpu->device_info->asic_name, 1313 KFD_TOPOLOGY_PUBLIC_NAME_SIZE); 1314 1315 dev->node_props.simd_arrays_per_engine = 1316 cu_info.num_shader_arrays_per_engine; 1317 1318 dev->node_props.vendor_id = gpu->pdev->vendor; 1319 dev->node_props.device_id = gpu->pdev->device; 1320 dev->node_props.capability |= 1321 ((amdgpu_amdkfd_get_asic_rev_id(dev->gpu->kgd) << 1322 HSA_CAP_ASIC_REVISION_SHIFT) & 1323 HSA_CAP_ASIC_REVISION_MASK); 1324 dev->node_props.location_id = pci_dev_id(gpu->pdev); 1325 dev->node_props.domain = pci_domain_nr(gpu->pdev->bus); 1326 dev->node_props.max_engine_clk_fcompute = 1327 amdgpu_amdkfd_get_max_engine_clock_in_mhz(dev->gpu->kgd); 1328 dev->node_props.max_engine_clk_ccompute = 1329 cpufreq_quick_get_max(0) / 1000; 1330 dev->node_props.drm_render_minor = 1331 gpu->shared_resources.drm_render_minor; 1332 1333 dev->node_props.hive_id = gpu->hive_id; 1334 dev->node_props.num_sdma_engines = gpu->device_info->num_sdma_engines; 1335 dev->node_props.num_sdma_xgmi_engines = 1336 gpu->device_info->num_xgmi_sdma_engines; 1337 dev->node_props.num_sdma_queues_per_engine = 1338 gpu->device_info->num_sdma_queues_per_engine; 1339 dev->node_props.num_gws = (dev->gpu->gws && 1340 dev->gpu->dqm->sched_policy != KFD_SCHED_POLICY_NO_HWS) ? 1341 amdgpu_amdkfd_get_num_gws(dev->gpu->kgd) : 0; 1342 dev->node_props.num_cp_queues = get_cp_queues_num(dev->gpu->dqm); 1343 dev->node_props.unique_id = gpu->unique_id; 1344 1345 kfd_fill_mem_clk_max_info(dev); 1346 kfd_fill_iolink_non_crat_info(dev); 1347 1348 switch (dev->gpu->device_info->asic_family) { 1349 case CHIP_KAVERI: 1350 case CHIP_HAWAII: 1351 case CHIP_TONGA: 1352 dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_PRE_1_0 << 1353 HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & 1354 HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); 1355 break; 1356 case CHIP_CARRIZO: 1357 case CHIP_FIJI: 1358 case CHIP_POLARIS10: 1359 case CHIP_POLARIS11: 1360 case CHIP_POLARIS12: 1361 case CHIP_VEGAM: 1362 pr_debug("Adding doorbell packet type capability\n"); 1363 dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_1_0 << 1364 HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & 1365 HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); 1366 break; 1367 case CHIP_VEGA10: 1368 case CHIP_VEGA12: 1369 case CHIP_VEGA20: 1370 case CHIP_RAVEN: 1371 case CHIP_RENOIR: 1372 case CHIP_ARCTURUS: 1373 case CHIP_NAVI10: 1374 case CHIP_NAVI12: 1375 case CHIP_NAVI14: 1376 case CHIP_SIENNA_CICHLID: 1377 case CHIP_NAVY_FLOUNDER: 1378 dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_2_0 << 1379 HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & 1380 HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); 1381 break; 1382 default: 1383 WARN(1, "Unexpected ASIC family %u", 1384 dev->gpu->device_info->asic_family); 1385 } 1386 1387 /* 1388 * Overwrite ATS capability according to needs_iommu_device to fix 1389 * potential missing corresponding bit in CRAT of BIOS. 1390 */ 1391 if (dev->gpu->use_iommu_v2) 1392 dev->node_props.capability |= HSA_CAP_ATS_PRESENT; 1393 else 1394 dev->node_props.capability &= ~HSA_CAP_ATS_PRESENT; 1395 1396 /* Fix errors in CZ CRAT. 1397 * simd_count: Carrizo CRAT reports wrong simd_count, probably 1398 * because it doesn't consider masked out CUs 1399 * max_waves_per_simd: Carrizo reports wrong max_waves_per_simd 1400 */ 1401 if (dev->gpu->device_info->asic_family == CHIP_CARRIZO) { 1402 dev->node_props.simd_count = 1403 cu_info.simd_per_cu * cu_info.cu_active_number; 1404 dev->node_props.max_waves_per_simd = 10; 1405 } 1406 1407 adev = (struct amdgpu_device *)(dev->gpu->kgd); 1408 /* kfd only concerns sram ecc on GFX and HBM ecc on UMC */ 1409 dev->node_props.capability |= 1410 ((adev->ras_features & BIT(AMDGPU_RAS_BLOCK__GFX)) != 0) ? 1411 HSA_CAP_SRAM_EDCSUPPORTED : 0; 1412 dev->node_props.capability |= ((adev->ras_features & BIT(AMDGPU_RAS_BLOCK__UMC)) != 0) ? 1413 HSA_CAP_MEM_EDCSUPPORTED : 0; 1414 1415 if (adev->asic_type != CHIP_VEGA10) 1416 dev->node_props.capability |= (adev->ras_features != 0) ? 1417 HSA_CAP_RASEVENTNOTIFY : 0; 1418 1419 kfd_debug_print_topology(); 1420 1421 if (!res) 1422 kfd_notify_gpu_change(gpu_id, 1); 1423err: 1424 kfd_destroy_crat_image(crat_image); 1425 return res; 1426} 1427 1428int kfd_topology_remove_device(struct kfd_dev *gpu) 1429{ 1430 struct kfd_topology_device *dev, *tmp; 1431 uint32_t gpu_id; 1432 int res = -ENODEV; 1433 1434 down_write(&topology_lock); 1435 1436 list_for_each_entry_safe(dev, tmp, &topology_device_list, list) 1437 if (dev->gpu == gpu) { 1438 gpu_id = dev->gpu_id; 1439 kfd_remove_sysfs_node_entry(dev); 1440 kfd_release_topology_device(dev); 1441 sys_props.num_devices--; 1442 res = 0; 1443 if (kfd_topology_update_sysfs() < 0) 1444 kfd_topology_release_sysfs(); 1445 break; 1446 } 1447 1448 up_write(&topology_lock); 1449 1450 if (!res) 1451 kfd_notify_gpu_change(gpu_id, 0); 1452 1453 return res; 1454} 1455 1456/* kfd_topology_enum_kfd_devices - Enumerate through all devices in KFD 1457 * topology. If GPU device is found @idx, then valid kfd_dev pointer is 1458 * returned through @kdev 1459 * Return - 0: On success (@kdev will be NULL for non GPU nodes) 1460 * -1: If end of list 1461 */ 1462int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev) 1463{ 1464 1465 struct kfd_topology_device *top_dev; 1466 uint8_t device_idx = 0; 1467 1468 *kdev = NULL; 1469 down_read(&topology_lock); 1470 1471 list_for_each_entry(top_dev, &topology_device_list, list) { 1472 if (device_idx == idx) { 1473 *kdev = top_dev->gpu; 1474 up_read(&topology_lock); 1475 return 0; 1476 } 1477 1478 device_idx++; 1479 } 1480 1481 up_read(&topology_lock); 1482 1483 return -1; 1484 1485} 1486 1487static int kfd_cpumask_to_apic_id(const struct cpumask *cpumask) 1488{ 1489 int first_cpu_of_numa_node; 1490 1491 if (!cpumask || cpumask == cpu_none_mask) 1492 return -1; 1493 first_cpu_of_numa_node = cpumask_first(cpumask); 1494 if (first_cpu_of_numa_node >= nr_cpu_ids) 1495 return -1; 1496#ifdef CONFIG_X86_64 1497 return cpu_data(first_cpu_of_numa_node).apicid; 1498#else 1499 return first_cpu_of_numa_node; 1500#endif 1501} 1502 1503/* kfd_numa_node_to_apic_id - Returns the APIC ID of the first logical processor 1504 * of the given NUMA node (numa_node_id) 1505 * Return -1 on failure 1506 */ 1507int kfd_numa_node_to_apic_id(int numa_node_id) 1508{ 1509 if (numa_node_id == -1) { 1510 pr_warn("Invalid NUMA Node. Use online CPU mask\n"); 1511 return kfd_cpumask_to_apic_id(cpu_online_mask); 1512 } 1513 return kfd_cpumask_to_apic_id(cpumask_of_node(numa_node_id)); 1514} 1515 1516void kfd_double_confirm_iommu_support(struct kfd_dev *gpu) 1517{ 1518 struct kfd_topology_device *dev; 1519 1520 gpu->use_iommu_v2 = false; 1521 1522 if (!gpu->device_info->needs_iommu_device) 1523 return; 1524 1525 down_read(&topology_lock); 1526 1527 /* Only use IOMMUv2 if there is an APU topology node with no GPU 1528 * assigned yet. This GPU will be assigned to it. 1529 */ 1530 list_for_each_entry(dev, &topology_device_list, list) 1531 if (dev->node_props.cpu_cores_count && 1532 dev->node_props.simd_count && 1533 !dev->gpu) 1534 gpu->use_iommu_v2 = true; 1535 1536 up_read(&topology_lock); 1537} 1538 1539#if defined(CONFIG_DEBUG_FS) 1540 1541int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data) 1542{ 1543 struct kfd_topology_device *dev; 1544 unsigned int i = 0; 1545 int r = 0; 1546 1547 down_read(&topology_lock); 1548 1549 list_for_each_entry(dev, &topology_device_list, list) { 1550 if (!dev->gpu) { 1551 i++; 1552 continue; 1553 } 1554 1555 seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id); 1556 r = dqm_debugfs_hqds(m, dev->gpu->dqm); 1557 if (r) 1558 break; 1559 } 1560 1561 up_read(&topology_lock); 1562 1563 return r; 1564} 1565 1566int kfd_debugfs_rls_by_device(struct seq_file *m, void *data) 1567{ 1568 struct kfd_topology_device *dev; 1569 unsigned int i = 0; 1570 int r = 0; 1571 1572 down_read(&topology_lock); 1573 1574 list_for_each_entry(dev, &topology_device_list, list) { 1575 if (!dev->gpu) { 1576 i++; 1577 continue; 1578 } 1579 1580 seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id); 1581 r = pm_debugfs_runlist(m, &dev->gpu->dqm->packets); 1582 if (r) 1583 break; 1584 } 1585 1586 up_read(&topology_lock); 1587 1588 return r; 1589} 1590 1591#endif 1592