1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * Support Intel uncore PerfMon discovery mechanism. 4 * Copyright(c) 2021 Intel Corporation. 5 */ 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 8 #include "uncore.h" 9 #include "uncore_discovery.h" 10 11 static struct rb_root discovery_tables = RB_ROOT; 12 static int num_discovered_types[UNCORE_ACCESS_MAX]; 13 14 static bool has_generic_discovery_table(void) 15 { 16 struct pci_dev *dev; 17 int dvsec; 18 19 dev = pci_get_device(PCI_VENDOR_ID_INTEL, UNCORE_DISCOVERY_TABLE_DEVICE, NULL); 20 if (!dev) 21 return false; 22 23 /* A discovery table device has the unique capability ID. */ 24 dvsec = pci_find_next_ext_capability(dev, 0, UNCORE_EXT_CAP_ID_DISCOVERY); 25 pci_dev_put(dev); 26 if (dvsec) 27 return true; 28 29 return false; 30 } 31 32 static int logical_die_id; 33 34 static int get_device_die_id(struct pci_dev *dev) 35 { 36 int cpu, node = pcibus_to_node(dev->bus); 37 38 /* 39 * If the NUMA info is not available, assume that the logical die id is 40 * continuous in the order in which the discovery table devices are 41 * detected. 42 */ 43 if (node < 0) 44 return logical_die_id++; 45 46 for_each_cpu(cpu, cpumask_of_node(node)) { 47 struct cpuinfo_x86 *c = &cpu_data(cpu); 48 49 if (c->initialized && cpu_to_node(cpu) == node) 50 return c->logical_die_id; 51 } 52 53 /* 54 * All CPUs of a node may be offlined. For this case, 55 * the PCI and MMIO type of uncore blocks which are 56 * enumerated by the device will be unavailable. 57 */ 58 return -1; 59 } 60 61 #define __node_2_type(cur) \ 62 rb_entry((cur), struct intel_uncore_discovery_type, node) 63 64 static inline int __type_cmp(const void *key, const struct rb_node *b) 65 { 66 struct intel_uncore_discovery_type *type_b = __node_2_type(b); 67 const u16 *type_id = key; 68 69 if (type_b->type > *type_id) 70 return -1; 71 else if (type_b->type < *type_id) 72 return 1; 73 74 return 0; 75 } 76 77 static inline struct intel_uncore_discovery_type * 78 search_uncore_discovery_type(u16 type_id) 79 { 80 struct rb_node *node = rb_find(&type_id, &discovery_tables, __type_cmp); 81 82 return (node) ? __node_2_type(node) : NULL; 83 } 84 85 static inline bool __type_less(struct rb_node *a, const struct rb_node *b) 86 { 87 return (__node_2_type(a)->type < __node_2_type(b)->type); 88 } 89 90 static struct intel_uncore_discovery_type * 91 add_uncore_discovery_type(struct uncore_unit_discovery *unit) 92 { 93 struct intel_uncore_discovery_type *type; 94 95 if (unit->access_type >= UNCORE_ACCESS_MAX) { 96 pr_warn("Unsupported access type %d\n", unit->access_type); 97 return NULL; 98 } 99 100 type = kzalloc(sizeof(struct intel_uncore_discovery_type), GFP_KERNEL); 101 if (!type) 102 return NULL; 103 104 type->box_ctrl_die = kcalloc(__uncore_max_dies, sizeof(u64), GFP_KERNEL); 105 if (!type->box_ctrl_die) 106 goto free_type; 107 108 type->access_type = unit->access_type; 109 num_discovered_types[type->access_type]++; 110 type->type = unit->box_type; 111 112 rb_add(&type->node, &discovery_tables, __type_less); 113 114 return type; 115 116 free_type: 117 kfree(type); 118 119 return NULL; 120 121 } 122 123 static struct intel_uncore_discovery_type * 124 get_uncore_discovery_type(struct uncore_unit_discovery *unit) 125 { 126 struct intel_uncore_discovery_type *type; 127 128 type = search_uncore_discovery_type(unit->box_type); 129 if (type) 130 return type; 131 132 return add_uncore_discovery_type(unit); 133 } 134 135 static void 136 uncore_insert_box_info(struct uncore_unit_discovery *unit, 137 int die, bool parsed) 138 { 139 struct intel_uncore_discovery_type *type; 140 unsigned int *box_offset, *ids; 141 int i; 142 143 if (WARN_ON_ONCE(!unit->ctl || !unit->ctl_offset || !unit->ctr_offset)) 144 return; 145 146 if (parsed) { 147 type = search_uncore_discovery_type(unit->box_type); 148 if (WARN_ON_ONCE(!type)) 149 return; 150 /* Store the first box of each die */ 151 if (!type->box_ctrl_die[die]) 152 type->box_ctrl_die[die] = unit->ctl; 153 return; 154 } 155 156 type = get_uncore_discovery_type(unit); 157 if (!type) 158 return; 159 160 box_offset = kcalloc(type->num_boxes + 1, sizeof(unsigned int), GFP_KERNEL); 161 if (!box_offset) 162 return; 163 164 ids = kcalloc(type->num_boxes + 1, sizeof(unsigned int), GFP_KERNEL); 165 if (!ids) 166 goto free_box_offset; 167 168 /* Store generic information for the first box */ 169 if (!type->num_boxes) { 170 type->box_ctrl = unit->ctl; 171 type->box_ctrl_die[die] = unit->ctl; 172 type->num_counters = unit->num_regs; 173 type->counter_width = unit->bit_width; 174 type->ctl_offset = unit->ctl_offset; 175 type->ctr_offset = unit->ctr_offset; 176 *ids = unit->box_id; 177 goto end; 178 } 179 180 for (i = 0; i < type->num_boxes; i++) { 181 ids[i] = type->ids[i]; 182 box_offset[i] = type->box_offset[i]; 183 184 if (WARN_ON_ONCE(unit->box_id == ids[i])) 185 goto free_ids; 186 } 187 ids[i] = unit->box_id; 188 box_offset[i] = unit->ctl - type->box_ctrl; 189 kfree(type->ids); 190 kfree(type->box_offset); 191 end: 192 type->ids = ids; 193 type->box_offset = box_offset; 194 type->num_boxes++; 195 return; 196 197 free_ids: 198 kfree(ids); 199 200 free_box_offset: 201 kfree(box_offset); 202 203 } 204 205 static int parse_discovery_table(struct pci_dev *dev, int die, 206 u32 bar_offset, bool *parsed) 207 { 208 struct uncore_global_discovery global; 209 struct uncore_unit_discovery unit; 210 void __iomem *io_addr; 211 resource_size_t addr; 212 unsigned long size; 213 u32 val; 214 int i; 215 216 pci_read_config_dword(dev, bar_offset, &val); 217 218 if (val & UNCORE_DISCOVERY_MASK) 219 return -EINVAL; 220 221 addr = (resource_size_t)(val & ~UNCORE_DISCOVERY_MASK); 222 size = UNCORE_DISCOVERY_GLOBAL_MAP_SIZE; 223 io_addr = ioremap(addr, size); 224 if (!io_addr) 225 return -ENOMEM; 226 227 /* Read Global Discovery State */ 228 memcpy_fromio(&global, io_addr, sizeof(struct uncore_global_discovery)); 229 if (uncore_discovery_invalid_unit(global)) { 230 pr_info("Invalid Global Discovery State: 0x%llx 0x%llx 0x%llx\n", 231 global.table1, global.ctl, global.table3); 232 iounmap(io_addr); 233 return -EINVAL; 234 } 235 iounmap(io_addr); 236 237 size = (1 + global.max_units) * global.stride * 8; 238 io_addr = ioremap(addr, size); 239 if (!io_addr) 240 return -ENOMEM; 241 242 /* Parsing Unit Discovery State */ 243 for (i = 0; i < global.max_units; i++) { 244 memcpy_fromio(&unit, io_addr + (i + 1) * (global.stride * 8), 245 sizeof(struct uncore_unit_discovery)); 246 247 if (uncore_discovery_invalid_unit(unit)) 248 continue; 249 250 if (unit.access_type >= UNCORE_ACCESS_MAX) 251 continue; 252 253 uncore_insert_box_info(&unit, die, *parsed); 254 } 255 256 *parsed = true; 257 iounmap(io_addr); 258 return 0; 259 } 260 261 bool intel_uncore_has_discovery_tables(void) 262 { 263 u32 device, val, entry_id, bar_offset; 264 int die, dvsec = 0, ret = true; 265 struct pci_dev *dev = NULL; 266 bool parsed = false; 267 268 if (has_generic_discovery_table()) 269 device = UNCORE_DISCOVERY_TABLE_DEVICE; 270 else 271 device = PCI_ANY_ID; 272 273 /* 274 * Start a new search and iterates through the list of 275 * the discovery table devices. 276 */ 277 while ((dev = pci_get_device(PCI_VENDOR_ID_INTEL, device, dev)) != NULL) { 278 while ((dvsec = pci_find_next_ext_capability(dev, dvsec, UNCORE_EXT_CAP_ID_DISCOVERY))) { 279 pci_read_config_dword(dev, dvsec + UNCORE_DISCOVERY_DVSEC_OFFSET, &val); 280 entry_id = val & UNCORE_DISCOVERY_DVSEC_ID_MASK; 281 if (entry_id != UNCORE_DISCOVERY_DVSEC_ID_PMON) 282 continue; 283 284 pci_read_config_dword(dev, dvsec + UNCORE_DISCOVERY_DVSEC2_OFFSET, &val); 285 286 if (val & ~UNCORE_DISCOVERY_DVSEC2_BIR_MASK) { 287 ret = false; 288 goto err; 289 } 290 bar_offset = UNCORE_DISCOVERY_BIR_BASE + 291 (val & UNCORE_DISCOVERY_DVSEC2_BIR_MASK) * UNCORE_DISCOVERY_BIR_STEP; 292 293 die = get_device_die_id(dev); 294 if (die < 0) 295 continue; 296 297 parse_discovery_table(dev, die, bar_offset, &parsed); 298 } 299 } 300 301 /* None of the discovery tables are available */ 302 if (!parsed) 303 ret = false; 304 err: 305 pci_dev_put(dev); 306 307 return ret; 308 } 309 310 void intel_uncore_clear_discovery_tables(void) 311 { 312 struct intel_uncore_discovery_type *type, *next; 313 314 rbtree_postorder_for_each_entry_safe(type, next, &discovery_tables, node) { 315 kfree(type->box_ctrl_die); 316 kfree(type); 317 } 318 } 319 320 DEFINE_UNCORE_FORMAT_ATTR(event, event, "config:0-7"); 321 DEFINE_UNCORE_FORMAT_ATTR(umask, umask, "config:8-15"); 322 DEFINE_UNCORE_FORMAT_ATTR(edge, edge, "config:18"); 323 DEFINE_UNCORE_FORMAT_ATTR(inv, inv, "config:23"); 324 DEFINE_UNCORE_FORMAT_ATTR(thresh, thresh, "config:24-31"); 325 326 static struct attribute *generic_uncore_formats_attr[] = { 327 &format_attr_event.attr, 328 &format_attr_umask.attr, 329 &format_attr_edge.attr, 330 &format_attr_inv.attr, 331 &format_attr_thresh.attr, 332 NULL, 333 }; 334 335 static const struct attribute_group generic_uncore_format_group = { 336 .name = "format", 337 .attrs = generic_uncore_formats_attr, 338 }; 339 340 static void intel_generic_uncore_msr_init_box(struct intel_uncore_box *box) 341 { 342 wrmsrl(uncore_msr_box_ctl(box), GENERIC_PMON_BOX_CTL_INT); 343 } 344 345 static void intel_generic_uncore_msr_disable_box(struct intel_uncore_box *box) 346 { 347 wrmsrl(uncore_msr_box_ctl(box), GENERIC_PMON_BOX_CTL_FRZ); 348 } 349 350 static void intel_generic_uncore_msr_enable_box(struct intel_uncore_box *box) 351 { 352 wrmsrl(uncore_msr_box_ctl(box), 0); 353 } 354 355 static void intel_generic_uncore_msr_enable_event(struct intel_uncore_box *box, 356 struct perf_event *event) 357 { 358 struct hw_perf_event *hwc = &event->hw; 359 360 wrmsrl(hwc->config_base, hwc->config); 361 } 362 363 static void intel_generic_uncore_msr_disable_event(struct intel_uncore_box *box, 364 struct perf_event *event) 365 { 366 struct hw_perf_event *hwc = &event->hw; 367 368 wrmsrl(hwc->config_base, 0); 369 } 370 371 static struct intel_uncore_ops generic_uncore_msr_ops = { 372 .init_box = intel_generic_uncore_msr_init_box, 373 .disable_box = intel_generic_uncore_msr_disable_box, 374 .enable_box = intel_generic_uncore_msr_enable_box, 375 .disable_event = intel_generic_uncore_msr_disable_event, 376 .enable_event = intel_generic_uncore_msr_enable_event, 377 .read_counter = uncore_msr_read_counter, 378 }; 379 380 static void intel_generic_uncore_pci_init_box(struct intel_uncore_box *box) 381 { 382 struct pci_dev *pdev = box->pci_dev; 383 int box_ctl = uncore_pci_box_ctl(box); 384 385 __set_bit(UNCORE_BOX_FLAG_CTL_OFFS8, &box->flags); 386 pci_write_config_dword(pdev, box_ctl, GENERIC_PMON_BOX_CTL_INT); 387 } 388 389 static void intel_generic_uncore_pci_disable_box(struct intel_uncore_box *box) 390 { 391 struct pci_dev *pdev = box->pci_dev; 392 int box_ctl = uncore_pci_box_ctl(box); 393 394 pci_write_config_dword(pdev, box_ctl, GENERIC_PMON_BOX_CTL_FRZ); 395 } 396 397 static void intel_generic_uncore_pci_enable_box(struct intel_uncore_box *box) 398 { 399 struct pci_dev *pdev = box->pci_dev; 400 int box_ctl = uncore_pci_box_ctl(box); 401 402 pci_write_config_dword(pdev, box_ctl, 0); 403 } 404 405 static void intel_generic_uncore_pci_enable_event(struct intel_uncore_box *box, 406 struct perf_event *event) 407 { 408 struct pci_dev *pdev = box->pci_dev; 409 struct hw_perf_event *hwc = &event->hw; 410 411 pci_write_config_dword(pdev, hwc->config_base, hwc->config); 412 } 413 414 static void intel_generic_uncore_pci_disable_event(struct intel_uncore_box *box, 415 struct perf_event *event) 416 { 417 struct pci_dev *pdev = box->pci_dev; 418 struct hw_perf_event *hwc = &event->hw; 419 420 pci_write_config_dword(pdev, hwc->config_base, 0); 421 } 422 423 static u64 intel_generic_uncore_pci_read_counter(struct intel_uncore_box *box, 424 struct perf_event *event) 425 { 426 struct pci_dev *pdev = box->pci_dev; 427 struct hw_perf_event *hwc = &event->hw; 428 u64 count = 0; 429 430 pci_read_config_dword(pdev, hwc->event_base, (u32 *)&count); 431 pci_read_config_dword(pdev, hwc->event_base + 4, (u32 *)&count + 1); 432 433 return count; 434 } 435 436 static struct intel_uncore_ops generic_uncore_pci_ops = { 437 .init_box = intel_generic_uncore_pci_init_box, 438 .disable_box = intel_generic_uncore_pci_disable_box, 439 .enable_box = intel_generic_uncore_pci_enable_box, 440 .disable_event = intel_generic_uncore_pci_disable_event, 441 .enable_event = intel_generic_uncore_pci_enable_event, 442 .read_counter = intel_generic_uncore_pci_read_counter, 443 }; 444 445 #define UNCORE_GENERIC_MMIO_SIZE 0x4000 446 447 static unsigned int generic_uncore_mmio_box_ctl(struct intel_uncore_box *box) 448 { 449 struct intel_uncore_type *type = box->pmu->type; 450 451 if (!type->box_ctls || !type->box_ctls[box->dieid] || !type->mmio_offsets) 452 return 0; 453 454 return type->box_ctls[box->dieid] + type->mmio_offsets[box->pmu->pmu_idx]; 455 } 456 457 static void intel_generic_uncore_mmio_init_box(struct intel_uncore_box *box) 458 { 459 unsigned int box_ctl = generic_uncore_mmio_box_ctl(box); 460 struct intel_uncore_type *type = box->pmu->type; 461 resource_size_t addr; 462 463 if (!box_ctl) { 464 pr_warn("Uncore type %d box %d: Invalid box control address.\n", 465 type->type_id, type->box_ids[box->pmu->pmu_idx]); 466 return; 467 } 468 469 addr = box_ctl; 470 box->io_addr = ioremap(addr, UNCORE_GENERIC_MMIO_SIZE); 471 if (!box->io_addr) { 472 pr_warn("Uncore type %d box %d: ioremap error for 0x%llx.\n", 473 type->type_id, type->box_ids[box->pmu->pmu_idx], 474 (unsigned long long)addr); 475 return; 476 } 477 478 writel(GENERIC_PMON_BOX_CTL_INT, box->io_addr); 479 } 480 481 static void intel_generic_uncore_mmio_disable_box(struct intel_uncore_box *box) 482 { 483 if (!box->io_addr) 484 return; 485 486 writel(GENERIC_PMON_BOX_CTL_FRZ, box->io_addr); 487 } 488 489 static void intel_generic_uncore_mmio_enable_box(struct intel_uncore_box *box) 490 { 491 if (!box->io_addr) 492 return; 493 494 writel(0, box->io_addr); 495 } 496 497 static void intel_generic_uncore_mmio_enable_event(struct intel_uncore_box *box, 498 struct perf_event *event) 499 { 500 struct hw_perf_event *hwc = &event->hw; 501 502 if (!box->io_addr) 503 return; 504 505 writel(hwc->config, box->io_addr + hwc->config_base); 506 } 507 508 static void intel_generic_uncore_mmio_disable_event(struct intel_uncore_box *box, 509 struct perf_event *event) 510 { 511 struct hw_perf_event *hwc = &event->hw; 512 513 if (!box->io_addr) 514 return; 515 516 writel(0, box->io_addr + hwc->config_base); 517 } 518 519 static struct intel_uncore_ops generic_uncore_mmio_ops = { 520 .init_box = intel_generic_uncore_mmio_init_box, 521 .exit_box = uncore_mmio_exit_box, 522 .disable_box = intel_generic_uncore_mmio_disable_box, 523 .enable_box = intel_generic_uncore_mmio_enable_box, 524 .disable_event = intel_generic_uncore_mmio_disable_event, 525 .enable_event = intel_generic_uncore_mmio_enable_event, 526 .read_counter = uncore_mmio_read_counter, 527 }; 528 529 static bool uncore_update_uncore_type(enum uncore_access_type type_id, 530 struct intel_uncore_type *uncore, 531 struct intel_uncore_discovery_type *type) 532 { 533 uncore->type_id = type->type; 534 uncore->num_boxes = type->num_boxes; 535 uncore->num_counters = type->num_counters; 536 uncore->perf_ctr_bits = type->counter_width; 537 uncore->box_ids = type->ids; 538 539 switch (type_id) { 540 case UNCORE_ACCESS_MSR: 541 uncore->ops = &generic_uncore_msr_ops; 542 uncore->perf_ctr = (unsigned int)type->box_ctrl + type->ctr_offset; 543 uncore->event_ctl = (unsigned int)type->box_ctrl + type->ctl_offset; 544 uncore->box_ctl = (unsigned int)type->box_ctrl; 545 uncore->msr_offsets = type->box_offset; 546 break; 547 case UNCORE_ACCESS_PCI: 548 uncore->ops = &generic_uncore_pci_ops; 549 uncore->perf_ctr = (unsigned int)UNCORE_DISCOVERY_PCI_BOX_CTRL(type->box_ctrl) + type->ctr_offset; 550 uncore->event_ctl = (unsigned int)UNCORE_DISCOVERY_PCI_BOX_CTRL(type->box_ctrl) + type->ctl_offset; 551 uncore->box_ctl = (unsigned int)UNCORE_DISCOVERY_PCI_BOX_CTRL(type->box_ctrl); 552 uncore->box_ctls = type->box_ctrl_die; 553 uncore->pci_offsets = type->box_offset; 554 break; 555 case UNCORE_ACCESS_MMIO: 556 uncore->ops = &generic_uncore_mmio_ops; 557 uncore->perf_ctr = (unsigned int)type->ctr_offset; 558 uncore->event_ctl = (unsigned int)type->ctl_offset; 559 uncore->box_ctl = (unsigned int)type->box_ctrl; 560 uncore->box_ctls = type->box_ctrl_die; 561 uncore->mmio_offsets = type->box_offset; 562 uncore->mmio_map_size = UNCORE_GENERIC_MMIO_SIZE; 563 break; 564 default: 565 return false; 566 } 567 568 return true; 569 } 570 571 static struct intel_uncore_type ** 572 intel_uncore_generic_init_uncores(enum uncore_access_type type_id) 573 { 574 struct intel_uncore_discovery_type *type; 575 struct intel_uncore_type **uncores; 576 struct intel_uncore_type *uncore; 577 struct rb_node *node; 578 int i = 0; 579 580 uncores = kcalloc(num_discovered_types[type_id] + 1, 581 sizeof(struct intel_uncore_type *), GFP_KERNEL); 582 if (!uncores) 583 return empty_uncore; 584 585 for (node = rb_first(&discovery_tables); node; node = rb_next(node)) { 586 type = rb_entry(node, struct intel_uncore_discovery_type, node); 587 if (type->access_type != type_id) 588 continue; 589 590 uncore = kzalloc(sizeof(struct intel_uncore_type), GFP_KERNEL); 591 if (!uncore) 592 break; 593 594 uncore->event_mask = GENERIC_PMON_RAW_EVENT_MASK; 595 uncore->format_group = &generic_uncore_format_group; 596 597 if (!uncore_update_uncore_type(type_id, uncore, type)) { 598 kfree(uncore); 599 continue; 600 } 601 uncores[i++] = uncore; 602 } 603 604 return uncores; 605 } 606 607 void intel_uncore_generic_uncore_cpu_init(void) 608 { 609 uncore_msr_uncores = intel_uncore_generic_init_uncores(UNCORE_ACCESS_MSR); 610 } 611 612 int intel_uncore_generic_uncore_pci_init(void) 613 { 614 uncore_pci_uncores = intel_uncore_generic_init_uncores(UNCORE_ACCESS_PCI); 615 616 return 0; 617 } 618 619 void intel_uncore_generic_uncore_mmio_init(void) 620 { 621 uncore_mmio_uncores = intel_uncore_generic_init_uncores(UNCORE_ACCESS_MMIO); 622 } 623