1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Implement the AER root port service driver. The driver registers an IRQ 4 * handler. When a root port triggers an AER interrupt, the IRQ handler 5 * collects root port status and schedules work. 6 * 7 * Copyright (C) 2006 Intel Corp. 8 * Tom Long Nguyen (tom.l.nguyen@intel.com) 9 * Zhang Yanmin (yanmin.zhang@intel.com) 10 * 11 * (C) Copyright 2009 Hewlett-Packard Development Company, L.P. 12 * Andrew Patterson <andrew.patterson@hp.com> 13 */ 14 15 #define pr_fmt(fmt) "AER: " fmt 16 #define dev_fmt pr_fmt 17 18 #include <linux/bitops.h> 19 #include <linux/cper.h> 20 #include <linux/pci.h> 21 #include <linux/pci-acpi.h> 22 #include <linux/sched.h> 23 #include <linux/kernel.h> 24 #include <linux/errno.h> 25 #include <linux/pm.h> 26 #include <linux/init.h> 27 #include <linux/interrupt.h> 28 #include <linux/delay.h> 29 #include <linux/kfifo.h> 30 #include <linux/slab.h> 31 #include <acpi/apei.h> 32 #include <acpi/ghes.h> 33 #include <ras/ras_event.h> 34 35 #include "../pci.h" 36 #include "portdrv.h" 37 38 #define AER_ERROR_SOURCES_MAX 128 39 40 #define AER_MAX_TYPEOF_COR_ERRS 16 /* as per PCI_ERR_COR_STATUS */ 41 #define AER_MAX_TYPEOF_UNCOR_ERRS 27 /* as per PCI_ERR_UNCOR_STATUS*/ 42 43 struct aer_err_source { 44 u32 status; /* PCI_ERR_ROOT_STATUS */ 45 u32 id; /* PCI_ERR_ROOT_ERR_SRC */ 46 }; 47 48 struct aer_rpc { 49 struct pci_dev *rpd; /* Root Port device */ 50 DECLARE_KFIFO(aer_fifo, struct aer_err_source, AER_ERROR_SOURCES_MAX); 51 }; 52 53 /* AER stats for the device */ 54 struct aer_stats { 55 56 /* 57 * Fields for all AER capable devices. They indicate the errors 58 * "as seen by this device". Note that this may mean that if an 59 * end point is causing problems, the AER counters may increment 60 * at its link partner (e.g. root port) because the errors will be 61 * "seen" by the link partner and not the problematic end point 62 * itself (which may report all counters as 0 as it never saw any 63 * problems). 64 */ 65 /* Counters for different type of correctable errors */ 66 u64 dev_cor_errs[AER_MAX_TYPEOF_COR_ERRS]; 67 /* Counters for different type of fatal uncorrectable errors */ 68 u64 dev_fatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS]; 69 /* Counters for different type of nonfatal uncorrectable errors */ 70 u64 dev_nonfatal_errs[AER_MAX_TYPEOF_UNCOR_ERRS]; 71 /* Total number of ERR_COR sent by this device */ 72 u64 dev_total_cor_errs; 73 /* Total number of ERR_FATAL sent by this device */ 74 u64 dev_total_fatal_errs; 75 /* Total number of ERR_NONFATAL sent by this device */ 76 u64 dev_total_nonfatal_errs; 77 78 /* 79 * Fields for Root ports & root complex event collectors only, these 80 * indicate the total number of ERR_COR, ERR_FATAL, and ERR_NONFATAL 81 * messages received by the root port / event collector, INCLUDING the 82 * ones that are generated internally (by the rootport itself) 83 */ 84 u64 rootport_total_cor_errs; 85 u64 rootport_total_fatal_errs; 86 u64 rootport_total_nonfatal_errs; 87 }; 88 89 #define AER_LOG_TLP_MASKS (PCI_ERR_UNC_POISON_TLP| \ 90 PCI_ERR_UNC_ECRC| \ 91 PCI_ERR_UNC_UNSUP| \ 92 PCI_ERR_UNC_COMP_ABORT| \ 93 PCI_ERR_UNC_UNX_COMP| \ 94 PCI_ERR_UNC_MALF_TLP) 95 96 #define SYSTEM_ERROR_INTR_ON_MESG_MASK (PCI_EXP_RTCTL_SECEE| \ 97 PCI_EXP_RTCTL_SENFEE| \ 98 PCI_EXP_RTCTL_SEFEE) 99 #define ROOT_PORT_INTR_ON_MESG_MASK (PCI_ERR_ROOT_CMD_COR_EN| \ 100 PCI_ERR_ROOT_CMD_NONFATAL_EN| \ 101 PCI_ERR_ROOT_CMD_FATAL_EN) 102 #define ERR_COR_ID(d) (d & 0xffff) 103 #define ERR_UNCOR_ID(d) (d >> 16) 104 105 #define AER_ERR_STATUS_MASK (PCI_ERR_ROOT_UNCOR_RCV | \ 106 PCI_ERR_ROOT_COR_RCV | \ 107 PCI_ERR_ROOT_MULTI_COR_RCV | \ 108 PCI_ERR_ROOT_MULTI_UNCOR_RCV) 109 110 static int pcie_aer_disable; 111 static pci_ers_result_t aer_root_reset(struct pci_dev *dev); 112 113 void pci_no_aer(void) 114 { 115 pcie_aer_disable = 1; 116 } 117 118 bool pci_aer_available(void) 119 { 120 return !pcie_aer_disable && pci_msi_enabled(); 121 } 122 123 #ifdef CONFIG_PCIE_ECRC 124 125 #define ECRC_POLICY_DEFAULT 0 /* ECRC set by BIOS */ 126 #define ECRC_POLICY_OFF 1 /* ECRC off for performance */ 127 #define ECRC_POLICY_ON 2 /* ECRC on for data integrity */ 128 129 static int ecrc_policy = ECRC_POLICY_DEFAULT; 130 131 static const char * const ecrc_policy_str[] = { 132 [ECRC_POLICY_DEFAULT] = "bios", 133 [ECRC_POLICY_OFF] = "off", 134 [ECRC_POLICY_ON] = "on" 135 }; 136 137 /** 138 * enable_ecrc_checking - enable PCIe ECRC checking for a device 139 * @dev: the PCI device 140 * 141 * Returns 0 on success, or negative on failure. 142 */ 143 static int enable_ecrc_checking(struct pci_dev *dev) 144 { 145 int aer = dev->aer_cap; 146 u32 reg32; 147 148 if (!aer) 149 return -ENODEV; 150 151 pci_read_config_dword(dev, aer + PCI_ERR_CAP, ®32); 152 if (reg32 & PCI_ERR_CAP_ECRC_GENC) 153 reg32 |= PCI_ERR_CAP_ECRC_GENE; 154 if (reg32 & PCI_ERR_CAP_ECRC_CHKC) 155 reg32 |= PCI_ERR_CAP_ECRC_CHKE; 156 pci_write_config_dword(dev, aer + PCI_ERR_CAP, reg32); 157 158 return 0; 159 } 160 161 /** 162 * disable_ecrc_checking - disables PCIe ECRC checking for a device 163 * @dev: the PCI device 164 * 165 * Returns 0 on success, or negative on failure. 166 */ 167 static int disable_ecrc_checking(struct pci_dev *dev) 168 { 169 int aer = dev->aer_cap; 170 u32 reg32; 171 172 if (!aer) 173 return -ENODEV; 174 175 pci_read_config_dword(dev, aer + PCI_ERR_CAP, ®32); 176 reg32 &= ~(PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE); 177 pci_write_config_dword(dev, aer + PCI_ERR_CAP, reg32); 178 179 return 0; 180 } 181 182 /** 183 * pcie_set_ecrc_checking - set/unset PCIe ECRC checking for a device based on global policy 184 * @dev: the PCI device 185 */ 186 void pcie_set_ecrc_checking(struct pci_dev *dev) 187 { 188 if (!pcie_aer_is_native(dev)) 189 return; 190 191 switch (ecrc_policy) { 192 case ECRC_POLICY_DEFAULT: 193 return; 194 case ECRC_POLICY_OFF: 195 disable_ecrc_checking(dev); 196 break; 197 case ECRC_POLICY_ON: 198 enable_ecrc_checking(dev); 199 break; 200 default: 201 return; 202 } 203 } 204 205 /** 206 * pcie_ecrc_get_policy - parse kernel command-line ecrc option 207 * @str: ECRC policy from kernel command line to use 208 */ 209 void pcie_ecrc_get_policy(char *str) 210 { 211 int i; 212 213 i = match_string(ecrc_policy_str, ARRAY_SIZE(ecrc_policy_str), str); 214 if (i < 0) 215 return; 216 217 ecrc_policy = i; 218 } 219 #endif /* CONFIG_PCIE_ECRC */ 220 221 #define PCI_EXP_AER_FLAGS (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \ 222 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE) 223 224 int pcie_aer_is_native(struct pci_dev *dev) 225 { 226 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus); 227 228 if (!dev->aer_cap) 229 return 0; 230 231 return pcie_ports_native || host->native_aer; 232 } 233 EXPORT_SYMBOL_NS_GPL(pcie_aer_is_native, CXL); 234 235 static int pci_enable_pcie_error_reporting(struct pci_dev *dev) 236 { 237 int rc; 238 239 if (!pcie_aer_is_native(dev)) 240 return -EIO; 241 242 rc = pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS); 243 return pcibios_err_to_errno(rc); 244 } 245 246 int pci_aer_clear_nonfatal_status(struct pci_dev *dev) 247 { 248 int aer = dev->aer_cap; 249 u32 status, sev; 250 251 if (!pcie_aer_is_native(dev)) 252 return -EIO; 253 254 /* Clear status bits for ERR_NONFATAL errors only */ 255 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status); 256 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, &sev); 257 status &= ~sev; 258 if (status) 259 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status); 260 261 return 0; 262 } 263 EXPORT_SYMBOL_GPL(pci_aer_clear_nonfatal_status); 264 265 void pci_aer_clear_fatal_status(struct pci_dev *dev) 266 { 267 int aer = dev->aer_cap; 268 u32 status, sev; 269 270 if (!pcie_aer_is_native(dev)) 271 return; 272 273 /* Clear status bits for ERR_FATAL errors only */ 274 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status); 275 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, &sev); 276 status &= sev; 277 if (status) 278 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status); 279 } 280 281 /** 282 * pci_aer_raw_clear_status - Clear AER error registers. 283 * @dev: the PCI device 284 * 285 * Clearing AER error status registers unconditionally, regardless of 286 * whether they're owned by firmware or the OS. 287 * 288 * Returns 0 on success, or negative on failure. 289 */ 290 int pci_aer_raw_clear_status(struct pci_dev *dev) 291 { 292 int aer = dev->aer_cap; 293 u32 status; 294 int port_type; 295 296 if (!aer) 297 return -EIO; 298 299 port_type = pci_pcie_type(dev); 300 if (port_type == PCI_EXP_TYPE_ROOT_PORT || 301 port_type == PCI_EXP_TYPE_RC_EC) { 302 pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &status); 303 pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, status); 304 } 305 306 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS, &status); 307 pci_write_config_dword(dev, aer + PCI_ERR_COR_STATUS, status); 308 309 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status); 310 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status); 311 312 return 0; 313 } 314 315 int pci_aer_clear_status(struct pci_dev *dev) 316 { 317 if (!pcie_aer_is_native(dev)) 318 return -EIO; 319 320 return pci_aer_raw_clear_status(dev); 321 } 322 323 void pci_save_aer_state(struct pci_dev *dev) 324 { 325 int aer = dev->aer_cap; 326 struct pci_cap_saved_state *save_state; 327 u32 *cap; 328 329 if (!aer) 330 return; 331 332 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR); 333 if (!save_state) 334 return; 335 336 cap = &save_state->cap.data[0]; 337 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, cap++); 338 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, cap++); 339 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, cap++); 340 pci_read_config_dword(dev, aer + PCI_ERR_CAP, cap++); 341 if (pcie_cap_has_rtctl(dev)) 342 pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, cap++); 343 } 344 345 void pci_restore_aer_state(struct pci_dev *dev) 346 { 347 int aer = dev->aer_cap; 348 struct pci_cap_saved_state *save_state; 349 u32 *cap; 350 351 if (!aer) 352 return; 353 354 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR); 355 if (!save_state) 356 return; 357 358 cap = &save_state->cap.data[0]; 359 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, *cap++); 360 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_SEVER, *cap++); 361 pci_write_config_dword(dev, aer + PCI_ERR_COR_MASK, *cap++); 362 pci_write_config_dword(dev, aer + PCI_ERR_CAP, *cap++); 363 if (pcie_cap_has_rtctl(dev)) 364 pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, *cap++); 365 } 366 367 void pci_aer_init(struct pci_dev *dev) 368 { 369 int n; 370 371 dev->aer_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR); 372 if (!dev->aer_cap) 373 return; 374 375 dev->aer_stats = kzalloc(sizeof(struct aer_stats), GFP_KERNEL); 376 377 /* 378 * We save/restore PCI_ERR_UNCOR_MASK, PCI_ERR_UNCOR_SEVER, 379 * PCI_ERR_COR_MASK, and PCI_ERR_CAP. Root and Root Complex Event 380 * Collectors also implement PCI_ERR_ROOT_COMMAND (PCIe r5.0, sec 381 * 7.8.4). 382 */ 383 n = pcie_cap_has_rtctl(dev) ? 5 : 4; 384 pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_ERR, sizeof(u32) * n); 385 386 pci_aer_clear_status(dev); 387 388 if (pci_aer_available()) 389 pci_enable_pcie_error_reporting(dev); 390 391 pcie_set_ecrc_checking(dev); 392 } 393 394 void pci_aer_exit(struct pci_dev *dev) 395 { 396 kfree(dev->aer_stats); 397 dev->aer_stats = NULL; 398 } 399 400 #define AER_AGENT_RECEIVER 0 401 #define AER_AGENT_REQUESTER 1 402 #define AER_AGENT_COMPLETER 2 403 #define AER_AGENT_TRANSMITTER 3 404 405 #define AER_AGENT_REQUESTER_MASK(t) ((t == AER_CORRECTABLE) ? \ 406 0 : (PCI_ERR_UNC_COMP_TIME|PCI_ERR_UNC_UNSUP)) 407 #define AER_AGENT_COMPLETER_MASK(t) ((t == AER_CORRECTABLE) ? \ 408 0 : PCI_ERR_UNC_COMP_ABORT) 409 #define AER_AGENT_TRANSMITTER_MASK(t) ((t == AER_CORRECTABLE) ? \ 410 (PCI_ERR_COR_REP_ROLL|PCI_ERR_COR_REP_TIMER) : 0) 411 412 #define AER_GET_AGENT(t, e) \ 413 ((e & AER_AGENT_COMPLETER_MASK(t)) ? AER_AGENT_COMPLETER : \ 414 (e & AER_AGENT_REQUESTER_MASK(t)) ? AER_AGENT_REQUESTER : \ 415 (e & AER_AGENT_TRANSMITTER_MASK(t)) ? AER_AGENT_TRANSMITTER : \ 416 AER_AGENT_RECEIVER) 417 418 #define AER_PHYSICAL_LAYER_ERROR 0 419 #define AER_DATA_LINK_LAYER_ERROR 1 420 #define AER_TRANSACTION_LAYER_ERROR 2 421 422 #define AER_PHYSICAL_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \ 423 PCI_ERR_COR_RCVR : 0) 424 #define AER_DATA_LINK_LAYER_ERROR_MASK(t) ((t == AER_CORRECTABLE) ? \ 425 (PCI_ERR_COR_BAD_TLP| \ 426 PCI_ERR_COR_BAD_DLLP| \ 427 PCI_ERR_COR_REP_ROLL| \ 428 PCI_ERR_COR_REP_TIMER) : PCI_ERR_UNC_DLP) 429 430 #define AER_GET_LAYER_ERROR(t, e) \ 431 ((e & AER_PHYSICAL_LAYER_ERROR_MASK(t)) ? AER_PHYSICAL_LAYER_ERROR : \ 432 (e & AER_DATA_LINK_LAYER_ERROR_MASK(t)) ? AER_DATA_LINK_LAYER_ERROR : \ 433 AER_TRANSACTION_LAYER_ERROR) 434 435 /* 436 * AER error strings 437 */ 438 static const char * const aer_error_severity_string[] = { 439 "Uncorrectable (Non-Fatal)", 440 "Uncorrectable (Fatal)", 441 "Correctable" 442 }; 443 444 static const char *aer_error_layer[] = { 445 "Physical Layer", 446 "Data Link Layer", 447 "Transaction Layer" 448 }; 449 450 static const char *aer_correctable_error_string[] = { 451 "RxErr", /* Bit Position 0 */ 452 NULL, 453 NULL, 454 NULL, 455 NULL, 456 NULL, 457 "BadTLP", /* Bit Position 6 */ 458 "BadDLLP", /* Bit Position 7 */ 459 "Rollover", /* Bit Position 8 */ 460 NULL, 461 NULL, 462 NULL, 463 "Timeout", /* Bit Position 12 */ 464 "NonFatalErr", /* Bit Position 13 */ 465 "CorrIntErr", /* Bit Position 14 */ 466 "HeaderOF", /* Bit Position 15 */ 467 NULL, /* Bit Position 16 */ 468 NULL, /* Bit Position 17 */ 469 NULL, /* Bit Position 18 */ 470 NULL, /* Bit Position 19 */ 471 NULL, /* Bit Position 20 */ 472 NULL, /* Bit Position 21 */ 473 NULL, /* Bit Position 22 */ 474 NULL, /* Bit Position 23 */ 475 NULL, /* Bit Position 24 */ 476 NULL, /* Bit Position 25 */ 477 NULL, /* Bit Position 26 */ 478 NULL, /* Bit Position 27 */ 479 NULL, /* Bit Position 28 */ 480 NULL, /* Bit Position 29 */ 481 NULL, /* Bit Position 30 */ 482 NULL, /* Bit Position 31 */ 483 }; 484 485 static const char *aer_uncorrectable_error_string[] = { 486 "Undefined", /* Bit Position 0 */ 487 NULL, 488 NULL, 489 NULL, 490 "DLP", /* Bit Position 4 */ 491 "SDES", /* Bit Position 5 */ 492 NULL, 493 NULL, 494 NULL, 495 NULL, 496 NULL, 497 NULL, 498 "TLP", /* Bit Position 12 */ 499 "FCP", /* Bit Position 13 */ 500 "CmpltTO", /* Bit Position 14 */ 501 "CmpltAbrt", /* Bit Position 15 */ 502 "UnxCmplt", /* Bit Position 16 */ 503 "RxOF", /* Bit Position 17 */ 504 "MalfTLP", /* Bit Position 18 */ 505 "ECRC", /* Bit Position 19 */ 506 "UnsupReq", /* Bit Position 20 */ 507 "ACSViol", /* Bit Position 21 */ 508 "UncorrIntErr", /* Bit Position 22 */ 509 "BlockedTLP", /* Bit Position 23 */ 510 "AtomicOpBlocked", /* Bit Position 24 */ 511 "TLPBlockedErr", /* Bit Position 25 */ 512 "PoisonTLPBlocked", /* Bit Position 26 */ 513 NULL, /* Bit Position 27 */ 514 NULL, /* Bit Position 28 */ 515 NULL, /* Bit Position 29 */ 516 NULL, /* Bit Position 30 */ 517 NULL, /* Bit Position 31 */ 518 }; 519 520 static const char *aer_agent_string[] = { 521 "Receiver ID", 522 "Requester ID", 523 "Completer ID", 524 "Transmitter ID" 525 }; 526 527 #define aer_stats_dev_attr(name, stats_array, strings_array, \ 528 total_string, total_field) \ 529 static ssize_t \ 530 name##_show(struct device *dev, struct device_attribute *attr, \ 531 char *buf) \ 532 { \ 533 unsigned int i; \ 534 struct pci_dev *pdev = to_pci_dev(dev); \ 535 u64 *stats = pdev->aer_stats->stats_array; \ 536 size_t len = 0; \ 537 \ 538 for (i = 0; i < ARRAY_SIZE(pdev->aer_stats->stats_array); i++) {\ 539 if (strings_array[i]) \ 540 len += sysfs_emit_at(buf, len, "%s %llu\n", \ 541 strings_array[i], \ 542 stats[i]); \ 543 else if (stats[i]) \ 544 len += sysfs_emit_at(buf, len, \ 545 #stats_array "_bit[%d] %llu\n",\ 546 i, stats[i]); \ 547 } \ 548 len += sysfs_emit_at(buf, len, "TOTAL_%s %llu\n", total_string, \ 549 pdev->aer_stats->total_field); \ 550 return len; \ 551 } \ 552 static DEVICE_ATTR_RO(name) 553 554 aer_stats_dev_attr(aer_dev_correctable, dev_cor_errs, 555 aer_correctable_error_string, "ERR_COR", 556 dev_total_cor_errs); 557 aer_stats_dev_attr(aer_dev_fatal, dev_fatal_errs, 558 aer_uncorrectable_error_string, "ERR_FATAL", 559 dev_total_fatal_errs); 560 aer_stats_dev_attr(aer_dev_nonfatal, dev_nonfatal_errs, 561 aer_uncorrectable_error_string, "ERR_NONFATAL", 562 dev_total_nonfatal_errs); 563 564 #define aer_stats_rootport_attr(name, field) \ 565 static ssize_t \ 566 name##_show(struct device *dev, struct device_attribute *attr, \ 567 char *buf) \ 568 { \ 569 struct pci_dev *pdev = to_pci_dev(dev); \ 570 return sysfs_emit(buf, "%llu\n", pdev->aer_stats->field); \ 571 } \ 572 static DEVICE_ATTR_RO(name) 573 574 aer_stats_rootport_attr(aer_rootport_total_err_cor, 575 rootport_total_cor_errs); 576 aer_stats_rootport_attr(aer_rootport_total_err_fatal, 577 rootport_total_fatal_errs); 578 aer_stats_rootport_attr(aer_rootport_total_err_nonfatal, 579 rootport_total_nonfatal_errs); 580 581 static struct attribute *aer_stats_attrs[] __ro_after_init = { 582 &dev_attr_aer_dev_correctable.attr, 583 &dev_attr_aer_dev_fatal.attr, 584 &dev_attr_aer_dev_nonfatal.attr, 585 &dev_attr_aer_rootport_total_err_cor.attr, 586 &dev_attr_aer_rootport_total_err_fatal.attr, 587 &dev_attr_aer_rootport_total_err_nonfatal.attr, 588 NULL 589 }; 590 591 static umode_t aer_stats_attrs_are_visible(struct kobject *kobj, 592 struct attribute *a, int n) 593 { 594 struct device *dev = kobj_to_dev(kobj); 595 struct pci_dev *pdev = to_pci_dev(dev); 596 597 if (!pdev->aer_stats) 598 return 0; 599 600 if ((a == &dev_attr_aer_rootport_total_err_cor.attr || 601 a == &dev_attr_aer_rootport_total_err_fatal.attr || 602 a == &dev_attr_aer_rootport_total_err_nonfatal.attr) && 603 ((pci_pcie_type(pdev) != PCI_EXP_TYPE_ROOT_PORT) && 604 (pci_pcie_type(pdev) != PCI_EXP_TYPE_RC_EC))) 605 return 0; 606 607 return a->mode; 608 } 609 610 const struct attribute_group aer_stats_attr_group = { 611 .attrs = aer_stats_attrs, 612 .is_visible = aer_stats_attrs_are_visible, 613 }; 614 615 static void pci_dev_aer_stats_incr(struct pci_dev *pdev, 616 struct aer_err_info *info) 617 { 618 unsigned long status = info->status & ~info->mask; 619 int i, max = -1; 620 u64 *counter = NULL; 621 struct aer_stats *aer_stats = pdev->aer_stats; 622 623 if (!aer_stats) 624 return; 625 626 switch (info->severity) { 627 case AER_CORRECTABLE: 628 aer_stats->dev_total_cor_errs++; 629 counter = &aer_stats->dev_cor_errs[0]; 630 max = AER_MAX_TYPEOF_COR_ERRS; 631 break; 632 case AER_NONFATAL: 633 aer_stats->dev_total_nonfatal_errs++; 634 counter = &aer_stats->dev_nonfatal_errs[0]; 635 max = AER_MAX_TYPEOF_UNCOR_ERRS; 636 break; 637 case AER_FATAL: 638 aer_stats->dev_total_fatal_errs++; 639 counter = &aer_stats->dev_fatal_errs[0]; 640 max = AER_MAX_TYPEOF_UNCOR_ERRS; 641 break; 642 } 643 644 for_each_set_bit(i, &status, max) 645 counter[i]++; 646 } 647 648 static void pci_rootport_aer_stats_incr(struct pci_dev *pdev, 649 struct aer_err_source *e_src) 650 { 651 struct aer_stats *aer_stats = pdev->aer_stats; 652 653 if (!aer_stats) 654 return; 655 656 if (e_src->status & PCI_ERR_ROOT_COR_RCV) 657 aer_stats->rootport_total_cor_errs++; 658 659 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) { 660 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV) 661 aer_stats->rootport_total_fatal_errs++; 662 else 663 aer_stats->rootport_total_nonfatal_errs++; 664 } 665 } 666 667 static void __print_tlp_header(struct pci_dev *dev, 668 struct aer_header_log_regs *t) 669 { 670 pci_err(dev, " TLP Header: %08x %08x %08x %08x\n", 671 t->dw0, t->dw1, t->dw2, t->dw3); 672 } 673 674 static void __aer_print_error(struct pci_dev *dev, 675 struct aer_err_info *info) 676 { 677 const char **strings; 678 unsigned long status = info->status & ~info->mask; 679 const char *level, *errmsg; 680 int i; 681 682 if (info->severity == AER_CORRECTABLE) { 683 strings = aer_correctable_error_string; 684 level = KERN_WARNING; 685 } else { 686 strings = aer_uncorrectable_error_string; 687 level = KERN_ERR; 688 } 689 690 for_each_set_bit(i, &status, 32) { 691 errmsg = strings[i]; 692 if (!errmsg) 693 errmsg = "Unknown Error Bit"; 694 695 pci_printk(level, dev, " [%2d] %-22s%s\n", i, errmsg, 696 info->first_error == i ? " (First)" : ""); 697 } 698 pci_dev_aer_stats_incr(dev, info); 699 } 700 701 void aer_print_error(struct pci_dev *dev, struct aer_err_info *info) 702 { 703 int layer, agent; 704 int id = pci_dev_id(dev); 705 const char *level; 706 707 if (!info->status) { 708 pci_err(dev, "PCIe Bus Error: severity=%s, type=Inaccessible, (Unregistered Agent ID)\n", 709 aer_error_severity_string[info->severity]); 710 goto out; 711 } 712 713 layer = AER_GET_LAYER_ERROR(info->severity, info->status); 714 agent = AER_GET_AGENT(info->severity, info->status); 715 716 level = (info->severity == AER_CORRECTABLE) ? KERN_WARNING : KERN_ERR; 717 718 pci_printk(level, dev, "PCIe Bus Error: severity=%s, type=%s, (%s)\n", 719 aer_error_severity_string[info->severity], 720 aer_error_layer[layer], aer_agent_string[agent]); 721 722 pci_printk(level, dev, " device [%04x:%04x] error status/mask=%08x/%08x\n", 723 dev->vendor, dev->device, info->status, info->mask); 724 725 __aer_print_error(dev, info); 726 727 if (info->tlp_header_valid) 728 __print_tlp_header(dev, &info->tlp); 729 730 out: 731 if (info->id && info->error_dev_num > 1 && info->id == id) 732 pci_err(dev, " Error of this Agent is reported first\n"); 733 734 trace_aer_event(dev_name(&dev->dev), (info->status & ~info->mask), 735 info->severity, info->tlp_header_valid, &info->tlp); 736 } 737 738 static void aer_print_port_info(struct pci_dev *dev, struct aer_err_info *info) 739 { 740 u8 bus = info->id >> 8; 741 u8 devfn = info->id & 0xff; 742 743 pci_info(dev, "%s%s error message received from %04x:%02x:%02x.%d\n", 744 info->multi_error_valid ? "Multiple " : "", 745 aer_error_severity_string[info->severity], 746 pci_domain_nr(dev->bus), bus, PCI_SLOT(devfn), 747 PCI_FUNC(devfn)); 748 } 749 750 #ifdef CONFIG_ACPI_APEI_PCIEAER 751 int cper_severity_to_aer(int cper_severity) 752 { 753 switch (cper_severity) { 754 case CPER_SEV_RECOVERABLE: 755 return AER_NONFATAL; 756 case CPER_SEV_FATAL: 757 return AER_FATAL; 758 default: 759 return AER_CORRECTABLE; 760 } 761 } 762 EXPORT_SYMBOL_GPL(cper_severity_to_aer); 763 #endif 764 765 void pci_print_aer(struct pci_dev *dev, int aer_severity, 766 struct aer_capability_regs *aer) 767 { 768 int layer, agent, tlp_header_valid = 0; 769 u32 status, mask; 770 struct aer_err_info info; 771 772 if (aer_severity == AER_CORRECTABLE) { 773 status = aer->cor_status; 774 mask = aer->cor_mask; 775 } else { 776 status = aer->uncor_status; 777 mask = aer->uncor_mask; 778 tlp_header_valid = status & AER_LOG_TLP_MASKS; 779 } 780 781 layer = AER_GET_LAYER_ERROR(aer_severity, status); 782 agent = AER_GET_AGENT(aer_severity, status); 783 784 memset(&info, 0, sizeof(info)); 785 info.severity = aer_severity; 786 info.status = status; 787 info.mask = mask; 788 info.first_error = PCI_ERR_CAP_FEP(aer->cap_control); 789 790 pci_err(dev, "aer_status: 0x%08x, aer_mask: 0x%08x\n", status, mask); 791 __aer_print_error(dev, &info); 792 pci_err(dev, "aer_layer=%s, aer_agent=%s\n", 793 aer_error_layer[layer], aer_agent_string[agent]); 794 795 if (aer_severity != AER_CORRECTABLE) 796 pci_err(dev, "aer_uncor_severity: 0x%08x\n", 797 aer->uncor_severity); 798 799 if (tlp_header_valid) 800 __print_tlp_header(dev, &aer->header_log); 801 802 trace_aer_event(dev_name(&dev->dev), (status & ~mask), 803 aer_severity, tlp_header_valid, &aer->header_log); 804 } 805 EXPORT_SYMBOL_NS_GPL(pci_print_aer, CXL); 806 807 /** 808 * add_error_device - list device to be handled 809 * @e_info: pointer to error info 810 * @dev: pointer to pci_dev to be added 811 */ 812 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev) 813 { 814 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) { 815 e_info->dev[e_info->error_dev_num] = pci_dev_get(dev); 816 e_info->error_dev_num++; 817 return 0; 818 } 819 return -ENOSPC; 820 } 821 822 /** 823 * is_error_source - check whether the device is source of reported error 824 * @dev: pointer to pci_dev to be checked 825 * @e_info: pointer to reported error info 826 */ 827 static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info) 828 { 829 int aer = dev->aer_cap; 830 u32 status, mask; 831 u16 reg16; 832 833 /* 834 * When bus id is equal to 0, it might be a bad id 835 * reported by root port. 836 */ 837 if ((PCI_BUS_NUM(e_info->id) != 0) && 838 !(dev->bus->bus_flags & PCI_BUS_FLAGS_NO_AERSID)) { 839 /* Device ID match? */ 840 if (e_info->id == pci_dev_id(dev)) 841 return true; 842 843 /* Continue id comparing if there is no multiple error */ 844 if (!e_info->multi_error_valid) 845 return false; 846 } 847 848 /* 849 * When either 850 * 1) bus id is equal to 0. Some ports might lose the bus 851 * id of error source id; 852 * 2) bus flag PCI_BUS_FLAGS_NO_AERSID is set 853 * 3) There are multiple errors and prior ID comparing fails; 854 * We check AER status registers to find possible reporter. 855 */ 856 if (atomic_read(&dev->enable_cnt) == 0) 857 return false; 858 859 /* Check if AER is enabled */ 860 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, ®16); 861 if (!(reg16 & PCI_EXP_AER_FLAGS)) 862 return false; 863 864 if (!aer) 865 return false; 866 867 /* Check if error is recorded */ 868 if (e_info->severity == AER_CORRECTABLE) { 869 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS, &status); 870 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, &mask); 871 } else { 872 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, &status); 873 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &mask); 874 } 875 if (status & ~mask) 876 return true; 877 878 return false; 879 } 880 881 static int find_device_iter(struct pci_dev *dev, void *data) 882 { 883 struct aer_err_info *e_info = (struct aer_err_info *)data; 884 885 if (is_error_source(dev, e_info)) { 886 /* List this device */ 887 if (add_error_device(e_info, dev)) { 888 /* We cannot handle more... Stop iteration */ 889 /* TODO: Should print error message here? */ 890 return 1; 891 } 892 893 /* If there is only a single error, stop iteration */ 894 if (!e_info->multi_error_valid) 895 return 1; 896 } 897 return 0; 898 } 899 900 /** 901 * find_source_device - search through device hierarchy for source device 902 * @parent: pointer to Root Port pci_dev data structure 903 * @e_info: including detailed error information such like id 904 * 905 * Return true if found. 906 * 907 * Invoked by DPC when error is detected at the Root Port. 908 * Caller of this function must set id, severity, and multi_error_valid of 909 * struct aer_err_info pointed by @e_info properly. This function must fill 910 * e_info->error_dev_num and e_info->dev[], based on the given information. 911 */ 912 static bool find_source_device(struct pci_dev *parent, 913 struct aer_err_info *e_info) 914 { 915 struct pci_dev *dev = parent; 916 int result; 917 918 /* Must reset in this function */ 919 e_info->error_dev_num = 0; 920 921 /* Is Root Port an agent that sends error message? */ 922 result = find_device_iter(dev, e_info); 923 if (result) 924 return true; 925 926 if (pci_pcie_type(parent) == PCI_EXP_TYPE_RC_EC) 927 pcie_walk_rcec(parent, find_device_iter, e_info); 928 else 929 pci_walk_bus(parent->subordinate, find_device_iter, e_info); 930 931 if (!e_info->error_dev_num) { 932 u8 bus = e_info->id >> 8; 933 u8 devfn = e_info->id & 0xff; 934 935 pci_info(parent, "found no error details for %04x:%02x:%02x.%d\n", 936 pci_domain_nr(parent->bus), bus, PCI_SLOT(devfn), 937 PCI_FUNC(devfn)); 938 return false; 939 } 940 return true; 941 } 942 943 #ifdef CONFIG_PCIEAER_CXL 944 945 /** 946 * pci_aer_unmask_internal_errors - unmask internal errors 947 * @dev: pointer to the pcie_dev data structure 948 * 949 * Unmasks internal errors in the Uncorrectable and Correctable Error 950 * Mask registers. 951 * 952 * Note: AER must be enabled and supported by the device which must be 953 * checked in advance, e.g. with pcie_aer_is_native(). 954 */ 955 static void pci_aer_unmask_internal_errors(struct pci_dev *dev) 956 { 957 int aer = dev->aer_cap; 958 u32 mask; 959 960 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, &mask); 961 mask &= ~PCI_ERR_UNC_INTN; 962 pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, mask); 963 964 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, &mask); 965 mask &= ~PCI_ERR_COR_INTERNAL; 966 pci_write_config_dword(dev, aer + PCI_ERR_COR_MASK, mask); 967 } 968 969 static bool is_cxl_mem_dev(struct pci_dev *dev) 970 { 971 /* 972 * The capability, status, and control fields in Device 0, 973 * Function 0 DVSEC control the CXL functionality of the 974 * entire device (CXL 3.0, 8.1.3). 975 */ 976 if (dev->devfn != PCI_DEVFN(0, 0)) 977 return false; 978 979 /* 980 * CXL Memory Devices must have the 502h class code set (CXL 981 * 3.0, 8.1.12.1). 982 */ 983 if ((dev->class >> 8) != PCI_CLASS_MEMORY_CXL) 984 return false; 985 986 return true; 987 } 988 989 static bool cxl_error_is_native(struct pci_dev *dev) 990 { 991 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus); 992 993 return (pcie_ports_native || host->native_aer); 994 } 995 996 static bool is_internal_error(struct aer_err_info *info) 997 { 998 if (info->severity == AER_CORRECTABLE) 999 return info->status & PCI_ERR_COR_INTERNAL; 1000 1001 return info->status & PCI_ERR_UNC_INTN; 1002 } 1003 1004 static int cxl_rch_handle_error_iter(struct pci_dev *dev, void *data) 1005 { 1006 struct aer_err_info *info = (struct aer_err_info *)data; 1007 const struct pci_error_handlers *err_handler; 1008 1009 if (!is_cxl_mem_dev(dev) || !cxl_error_is_native(dev)) 1010 return 0; 1011 1012 /* protect dev->driver */ 1013 device_lock(&dev->dev); 1014 1015 err_handler = dev->driver ? dev->driver->err_handler : NULL; 1016 if (!err_handler) 1017 goto out; 1018 1019 if (info->severity == AER_CORRECTABLE) { 1020 if (err_handler->cor_error_detected) 1021 err_handler->cor_error_detected(dev); 1022 } else if (err_handler->error_detected) { 1023 if (info->severity == AER_NONFATAL) 1024 err_handler->error_detected(dev, pci_channel_io_normal); 1025 else if (info->severity == AER_FATAL) 1026 err_handler->error_detected(dev, pci_channel_io_frozen); 1027 } 1028 out: 1029 device_unlock(&dev->dev); 1030 return 0; 1031 } 1032 1033 static void cxl_rch_handle_error(struct pci_dev *dev, struct aer_err_info *info) 1034 { 1035 /* 1036 * Internal errors of an RCEC indicate an AER error in an 1037 * RCH's downstream port. Check and handle them in the CXL.mem 1038 * device driver. 1039 */ 1040 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_EC && 1041 is_internal_error(info)) 1042 pcie_walk_rcec(dev, cxl_rch_handle_error_iter, info); 1043 } 1044 1045 static int handles_cxl_error_iter(struct pci_dev *dev, void *data) 1046 { 1047 bool *handles_cxl = data; 1048 1049 if (!*handles_cxl) 1050 *handles_cxl = is_cxl_mem_dev(dev) && cxl_error_is_native(dev); 1051 1052 /* Non-zero terminates iteration */ 1053 return *handles_cxl; 1054 } 1055 1056 static bool handles_cxl_errors(struct pci_dev *rcec) 1057 { 1058 bool handles_cxl = false; 1059 1060 if (pci_pcie_type(rcec) == PCI_EXP_TYPE_RC_EC && 1061 pcie_aer_is_native(rcec)) 1062 pcie_walk_rcec(rcec, handles_cxl_error_iter, &handles_cxl); 1063 1064 return handles_cxl; 1065 } 1066 1067 static void cxl_rch_enable_rcec(struct pci_dev *rcec) 1068 { 1069 if (!handles_cxl_errors(rcec)) 1070 return; 1071 1072 pci_aer_unmask_internal_errors(rcec); 1073 pci_info(rcec, "CXL: Internal errors unmasked"); 1074 } 1075 1076 #else 1077 static inline void cxl_rch_enable_rcec(struct pci_dev *dev) { } 1078 static inline void cxl_rch_handle_error(struct pci_dev *dev, 1079 struct aer_err_info *info) { } 1080 #endif 1081 1082 /** 1083 * pci_aer_handle_error - handle logging error into an event log 1084 * @dev: pointer to pci_dev data structure of error source device 1085 * @info: comprehensive error information 1086 * 1087 * Invoked when an error being detected by Root Port. 1088 */ 1089 static void pci_aer_handle_error(struct pci_dev *dev, struct aer_err_info *info) 1090 { 1091 int aer = dev->aer_cap; 1092 1093 if (info->severity == AER_CORRECTABLE) { 1094 /* 1095 * Correctable error does not need software intervention. 1096 * No need to go through error recovery process. 1097 */ 1098 if (aer) 1099 pci_write_config_dword(dev, aer + PCI_ERR_COR_STATUS, 1100 info->status); 1101 if (pcie_aer_is_native(dev)) { 1102 struct pci_driver *pdrv = dev->driver; 1103 1104 if (pdrv && pdrv->err_handler && 1105 pdrv->err_handler->cor_error_detected) 1106 pdrv->err_handler->cor_error_detected(dev); 1107 pcie_clear_device_status(dev); 1108 } 1109 } else if (info->severity == AER_NONFATAL) 1110 pcie_do_recovery(dev, pci_channel_io_normal, aer_root_reset); 1111 else if (info->severity == AER_FATAL) 1112 pcie_do_recovery(dev, pci_channel_io_frozen, aer_root_reset); 1113 } 1114 1115 static void handle_error_source(struct pci_dev *dev, struct aer_err_info *info) 1116 { 1117 cxl_rch_handle_error(dev, info); 1118 pci_aer_handle_error(dev, info); 1119 pci_dev_put(dev); 1120 } 1121 1122 #ifdef CONFIG_ACPI_APEI_PCIEAER 1123 1124 #define AER_RECOVER_RING_SIZE 16 1125 1126 struct aer_recover_entry { 1127 u8 bus; 1128 u8 devfn; 1129 u16 domain; 1130 int severity; 1131 struct aer_capability_regs *regs; 1132 }; 1133 1134 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry, 1135 AER_RECOVER_RING_SIZE); 1136 1137 static void aer_recover_work_func(struct work_struct *work) 1138 { 1139 struct aer_recover_entry entry; 1140 struct pci_dev *pdev; 1141 1142 while (kfifo_get(&aer_recover_ring, &entry)) { 1143 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus, 1144 entry.devfn); 1145 if (!pdev) { 1146 pr_err("no pci_dev for %04x:%02x:%02x.%x\n", 1147 entry.domain, entry.bus, 1148 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn)); 1149 continue; 1150 } 1151 pci_print_aer(pdev, entry.severity, entry.regs); 1152 /* 1153 * Memory for aer_capability_regs(entry.regs) is being allocated from the 1154 * ghes_estatus_pool to protect it from overwriting when multiple sections 1155 * are present in the error status. Thus free the same after processing 1156 * the data. 1157 */ 1158 ghes_estatus_pool_region_free((unsigned long)entry.regs, 1159 sizeof(struct aer_capability_regs)); 1160 1161 if (entry.severity == AER_NONFATAL) 1162 pcie_do_recovery(pdev, pci_channel_io_normal, 1163 aer_root_reset); 1164 else if (entry.severity == AER_FATAL) 1165 pcie_do_recovery(pdev, pci_channel_io_frozen, 1166 aer_root_reset); 1167 pci_dev_put(pdev); 1168 } 1169 } 1170 1171 /* 1172 * Mutual exclusion for writers of aer_recover_ring, reader side don't 1173 * need lock, because there is only one reader and lock is not needed 1174 * between reader and writer. 1175 */ 1176 static DEFINE_SPINLOCK(aer_recover_ring_lock); 1177 static DECLARE_WORK(aer_recover_work, aer_recover_work_func); 1178 1179 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn, 1180 int severity, struct aer_capability_regs *aer_regs) 1181 { 1182 struct aer_recover_entry entry = { 1183 .bus = bus, 1184 .devfn = devfn, 1185 .domain = domain, 1186 .severity = severity, 1187 .regs = aer_regs, 1188 }; 1189 1190 if (kfifo_in_spinlocked(&aer_recover_ring, &entry, 1, 1191 &aer_recover_ring_lock)) 1192 schedule_work(&aer_recover_work); 1193 else 1194 pr_err("buffer overflow in recovery for %04x:%02x:%02x.%x\n", 1195 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn)); 1196 } 1197 EXPORT_SYMBOL_GPL(aer_recover_queue); 1198 #endif 1199 1200 /** 1201 * aer_get_device_error_info - read error status from dev and store it to info 1202 * @dev: pointer to the device expected to have a error record 1203 * @info: pointer to structure to store the error record 1204 * 1205 * Return 1 on success, 0 on error. 1206 * 1207 * Note that @info is reused among all error devices. Clear fields properly. 1208 */ 1209 int aer_get_device_error_info(struct pci_dev *dev, struct aer_err_info *info) 1210 { 1211 int type = pci_pcie_type(dev); 1212 int aer = dev->aer_cap; 1213 int temp; 1214 1215 /* Must reset in this function */ 1216 info->status = 0; 1217 info->tlp_header_valid = 0; 1218 1219 /* The device might not support AER */ 1220 if (!aer) 1221 return 0; 1222 1223 if (info->severity == AER_CORRECTABLE) { 1224 pci_read_config_dword(dev, aer + PCI_ERR_COR_STATUS, 1225 &info->status); 1226 pci_read_config_dword(dev, aer + PCI_ERR_COR_MASK, 1227 &info->mask); 1228 if (!(info->status & ~info->mask)) 1229 return 0; 1230 } else if (type == PCI_EXP_TYPE_ROOT_PORT || 1231 type == PCI_EXP_TYPE_RC_EC || 1232 type == PCI_EXP_TYPE_DOWNSTREAM || 1233 info->severity == AER_NONFATAL) { 1234 1235 /* Link is still healthy for IO reads */ 1236 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, 1237 &info->status); 1238 pci_read_config_dword(dev, aer + PCI_ERR_UNCOR_MASK, 1239 &info->mask); 1240 if (!(info->status & ~info->mask)) 1241 return 0; 1242 1243 /* Get First Error Pointer */ 1244 pci_read_config_dword(dev, aer + PCI_ERR_CAP, &temp); 1245 info->first_error = PCI_ERR_CAP_FEP(temp); 1246 1247 if (info->status & AER_LOG_TLP_MASKS) { 1248 info->tlp_header_valid = 1; 1249 pci_read_config_dword(dev, 1250 aer + PCI_ERR_HEADER_LOG, &info->tlp.dw0); 1251 pci_read_config_dword(dev, 1252 aer + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1); 1253 pci_read_config_dword(dev, 1254 aer + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2); 1255 pci_read_config_dword(dev, 1256 aer + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3); 1257 } 1258 } 1259 1260 return 1; 1261 } 1262 1263 static inline void aer_process_err_devices(struct aer_err_info *e_info) 1264 { 1265 int i; 1266 1267 /* Report all before handle them, not to lost records by reset etc. */ 1268 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) { 1269 if (aer_get_device_error_info(e_info->dev[i], e_info)) 1270 aer_print_error(e_info->dev[i], e_info); 1271 } 1272 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) { 1273 if (aer_get_device_error_info(e_info->dev[i], e_info)) 1274 handle_error_source(e_info->dev[i], e_info); 1275 } 1276 } 1277 1278 /** 1279 * aer_isr_one_error - consume an error detected by root port 1280 * @rpc: pointer to the root port which holds an error 1281 * @e_src: pointer to an error source 1282 */ 1283 static void aer_isr_one_error(struct aer_rpc *rpc, 1284 struct aer_err_source *e_src) 1285 { 1286 struct pci_dev *pdev = rpc->rpd; 1287 struct aer_err_info e_info; 1288 1289 pci_rootport_aer_stats_incr(pdev, e_src); 1290 1291 /* 1292 * There is a possibility that both correctable error and 1293 * uncorrectable error being logged. Report correctable error first. 1294 */ 1295 if (e_src->status & PCI_ERR_ROOT_COR_RCV) { 1296 e_info.id = ERR_COR_ID(e_src->id); 1297 e_info.severity = AER_CORRECTABLE; 1298 1299 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV) 1300 e_info.multi_error_valid = 1; 1301 else 1302 e_info.multi_error_valid = 0; 1303 aer_print_port_info(pdev, &e_info); 1304 1305 if (find_source_device(pdev, &e_info)) 1306 aer_process_err_devices(&e_info); 1307 } 1308 1309 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) { 1310 e_info.id = ERR_UNCOR_ID(e_src->id); 1311 1312 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV) 1313 e_info.severity = AER_FATAL; 1314 else 1315 e_info.severity = AER_NONFATAL; 1316 1317 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV) 1318 e_info.multi_error_valid = 1; 1319 else 1320 e_info.multi_error_valid = 0; 1321 1322 aer_print_port_info(pdev, &e_info); 1323 1324 if (find_source_device(pdev, &e_info)) 1325 aer_process_err_devices(&e_info); 1326 } 1327 } 1328 1329 /** 1330 * aer_isr - consume errors detected by root port 1331 * @irq: IRQ assigned to Root Port 1332 * @context: pointer to Root Port data structure 1333 * 1334 * Invoked, as DPC, when root port records new detected error 1335 */ 1336 static irqreturn_t aer_isr(int irq, void *context) 1337 { 1338 struct pcie_device *dev = (struct pcie_device *)context; 1339 struct aer_rpc *rpc = get_service_data(dev); 1340 struct aer_err_source e_src; 1341 1342 if (kfifo_is_empty(&rpc->aer_fifo)) 1343 return IRQ_NONE; 1344 1345 while (kfifo_get(&rpc->aer_fifo, &e_src)) 1346 aer_isr_one_error(rpc, &e_src); 1347 return IRQ_HANDLED; 1348 } 1349 1350 /** 1351 * aer_irq - Root Port's ISR 1352 * @irq: IRQ assigned to Root Port 1353 * @context: pointer to Root Port data structure 1354 * 1355 * Invoked when Root Port detects AER messages. 1356 */ 1357 static irqreturn_t aer_irq(int irq, void *context) 1358 { 1359 struct pcie_device *pdev = (struct pcie_device *)context; 1360 struct aer_rpc *rpc = get_service_data(pdev); 1361 struct pci_dev *rp = rpc->rpd; 1362 int aer = rp->aer_cap; 1363 struct aer_err_source e_src = {}; 1364 1365 pci_read_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, &e_src.status); 1366 if (!(e_src.status & AER_ERR_STATUS_MASK)) 1367 return IRQ_NONE; 1368 1369 pci_read_config_dword(rp, aer + PCI_ERR_ROOT_ERR_SRC, &e_src.id); 1370 pci_write_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, e_src.status); 1371 1372 if (!kfifo_put(&rpc->aer_fifo, e_src)) 1373 return IRQ_HANDLED; 1374 1375 return IRQ_WAKE_THREAD; 1376 } 1377 1378 static void aer_enable_irq(struct pci_dev *pdev) 1379 { 1380 int aer = pdev->aer_cap; 1381 u32 reg32; 1382 1383 /* Enable Root Port's interrupt in response to error messages */ 1384 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, ®32); 1385 reg32 |= ROOT_PORT_INTR_ON_MESG_MASK; 1386 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, reg32); 1387 } 1388 1389 static void aer_disable_irq(struct pci_dev *pdev) 1390 { 1391 int aer = pdev->aer_cap; 1392 u32 reg32; 1393 1394 /* Disable Root's interrupt in response to error messages */ 1395 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, ®32); 1396 reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK; 1397 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_COMMAND, reg32); 1398 } 1399 1400 /** 1401 * aer_enable_rootport - enable Root Port's interrupts when receiving messages 1402 * @rpc: pointer to a Root Port data structure 1403 * 1404 * Invoked when PCIe bus loads AER service driver. 1405 */ 1406 static void aer_enable_rootport(struct aer_rpc *rpc) 1407 { 1408 struct pci_dev *pdev = rpc->rpd; 1409 int aer = pdev->aer_cap; 1410 u16 reg16; 1411 u32 reg32; 1412 1413 /* Clear PCIe Capability's Device Status */ 1414 pcie_capability_read_word(pdev, PCI_EXP_DEVSTA, ®16); 1415 pcie_capability_write_word(pdev, PCI_EXP_DEVSTA, reg16); 1416 1417 /* Disable system error generation in response to error messages */ 1418 pcie_capability_clear_word(pdev, PCI_EXP_RTCTL, 1419 SYSTEM_ERROR_INTR_ON_MESG_MASK); 1420 1421 /* Clear error status */ 1422 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, ®32); 1423 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, reg32); 1424 pci_read_config_dword(pdev, aer + PCI_ERR_COR_STATUS, ®32); 1425 pci_write_config_dword(pdev, aer + PCI_ERR_COR_STATUS, reg32); 1426 pci_read_config_dword(pdev, aer + PCI_ERR_UNCOR_STATUS, ®32); 1427 pci_write_config_dword(pdev, aer + PCI_ERR_UNCOR_STATUS, reg32); 1428 1429 aer_enable_irq(pdev); 1430 } 1431 1432 /** 1433 * aer_disable_rootport - disable Root Port's interrupts when receiving messages 1434 * @rpc: pointer to a Root Port data structure 1435 * 1436 * Invoked when PCIe bus unloads AER service driver. 1437 */ 1438 static void aer_disable_rootport(struct aer_rpc *rpc) 1439 { 1440 struct pci_dev *pdev = rpc->rpd; 1441 int aer = pdev->aer_cap; 1442 u32 reg32; 1443 1444 aer_disable_irq(pdev); 1445 1446 /* Clear Root's error status reg */ 1447 pci_read_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, ®32); 1448 pci_write_config_dword(pdev, aer + PCI_ERR_ROOT_STATUS, reg32); 1449 } 1450 1451 /** 1452 * aer_remove - clean up resources 1453 * @dev: pointer to the pcie_dev data structure 1454 * 1455 * Invoked when PCI Express bus unloads or AER probe fails. 1456 */ 1457 static void aer_remove(struct pcie_device *dev) 1458 { 1459 struct aer_rpc *rpc = get_service_data(dev); 1460 1461 aer_disable_rootport(rpc); 1462 } 1463 1464 /** 1465 * aer_probe - initialize resources 1466 * @dev: pointer to the pcie_dev data structure 1467 * 1468 * Invoked when PCI Express bus loads AER service driver. 1469 */ 1470 static int aer_probe(struct pcie_device *dev) 1471 { 1472 int status; 1473 struct aer_rpc *rpc; 1474 struct device *device = &dev->device; 1475 struct pci_dev *port = dev->port; 1476 1477 BUILD_BUG_ON(ARRAY_SIZE(aer_correctable_error_string) < 1478 AER_MAX_TYPEOF_COR_ERRS); 1479 BUILD_BUG_ON(ARRAY_SIZE(aer_uncorrectable_error_string) < 1480 AER_MAX_TYPEOF_UNCOR_ERRS); 1481 1482 /* Limit to Root Ports or Root Complex Event Collectors */ 1483 if ((pci_pcie_type(port) != PCI_EXP_TYPE_RC_EC) && 1484 (pci_pcie_type(port) != PCI_EXP_TYPE_ROOT_PORT)) 1485 return -ENODEV; 1486 1487 rpc = devm_kzalloc(device, sizeof(struct aer_rpc), GFP_KERNEL); 1488 if (!rpc) 1489 return -ENOMEM; 1490 1491 rpc->rpd = port; 1492 INIT_KFIFO(rpc->aer_fifo); 1493 set_service_data(dev, rpc); 1494 1495 status = devm_request_threaded_irq(device, dev->irq, aer_irq, aer_isr, 1496 IRQF_SHARED, "aerdrv", dev); 1497 if (status) { 1498 pci_err(port, "request AER IRQ %d failed\n", dev->irq); 1499 return status; 1500 } 1501 1502 cxl_rch_enable_rcec(port); 1503 aer_enable_rootport(rpc); 1504 pci_info(port, "enabled with IRQ %d\n", dev->irq); 1505 return 0; 1506 } 1507 1508 /** 1509 * aer_root_reset - reset Root Port hierarchy, RCEC, or RCiEP 1510 * @dev: pointer to Root Port, RCEC, or RCiEP 1511 * 1512 * Invoked by Port Bus driver when performing reset. 1513 */ 1514 static pci_ers_result_t aer_root_reset(struct pci_dev *dev) 1515 { 1516 int type = pci_pcie_type(dev); 1517 struct pci_dev *root; 1518 int aer; 1519 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus); 1520 u32 reg32; 1521 int rc; 1522 1523 /* 1524 * Only Root Ports and RCECs have AER Root Command and Root Status 1525 * registers. If "dev" is an RCiEP, the relevant registers are in 1526 * the RCEC. 1527 */ 1528 if (type == PCI_EXP_TYPE_RC_END) 1529 root = dev->rcec; 1530 else 1531 root = pcie_find_root_port(dev); 1532 1533 /* 1534 * If the platform retained control of AER, an RCiEP may not have 1535 * an RCEC visible to us, so dev->rcec ("root") may be NULL. In 1536 * that case, firmware is responsible for these registers. 1537 */ 1538 aer = root ? root->aer_cap : 0; 1539 1540 if ((host->native_aer || pcie_ports_native) && aer) 1541 aer_disable_irq(root); 1542 1543 if (type == PCI_EXP_TYPE_RC_EC || type == PCI_EXP_TYPE_RC_END) { 1544 rc = pcie_reset_flr(dev, PCI_RESET_DO_RESET); 1545 if (!rc) 1546 pci_info(dev, "has been reset\n"); 1547 else 1548 pci_info(dev, "not reset (no FLR support: %d)\n", rc); 1549 } else { 1550 rc = pci_bus_error_reset(dev); 1551 pci_info(dev, "%s Port link has been reset (%d)\n", 1552 pci_is_root_bus(dev->bus) ? "Root" : "Downstream", rc); 1553 } 1554 1555 if ((host->native_aer || pcie_ports_native) && aer) { 1556 /* Clear Root Error Status */ 1557 pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, ®32); 1558 pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32); 1559 1560 aer_enable_irq(root); 1561 } 1562 1563 return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; 1564 } 1565 1566 static struct pcie_port_service_driver aerdriver = { 1567 .name = "aer", 1568 .port_type = PCIE_ANY_PORT, 1569 .service = PCIE_PORT_SERVICE_AER, 1570 1571 .probe = aer_probe, 1572 .remove = aer_remove, 1573 }; 1574 1575 /** 1576 * pcie_aer_init - register AER root service driver 1577 * 1578 * Invoked when AER root service driver is loaded. 1579 */ 1580 int __init pcie_aer_init(void) 1581 { 1582 if (!pci_aer_available()) 1583 return -ENXIO; 1584 return pcie_port_service_register(&aerdriver); 1585 } 1586