1 /* 2 * The file intends to implement the platform dependent EEH operations on 3 * powernv platform. Actually, the powernv was created in order to fully 4 * hypervisor support. 5 * 6 * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013. 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation; either version 2 of the License, or 11 * (at your option) any later version. 12 */ 13 14 #include <linux/atomic.h> 15 #include <linux/debugfs.h> 16 #include <linux/delay.h> 17 #include <linux/export.h> 18 #include <linux/init.h> 19 #include <linux/interrupt.h> 20 #include <linux/list.h> 21 #include <linux/msi.h> 22 #include <linux/of.h> 23 #include <linux/pci.h> 24 #include <linux/proc_fs.h> 25 #include <linux/rbtree.h> 26 #include <linux/sched.h> 27 #include <linux/seq_file.h> 28 #include <linux/spinlock.h> 29 30 #include <asm/eeh.h> 31 #include <asm/eeh_event.h> 32 #include <asm/firmware.h> 33 #include <asm/io.h> 34 #include <asm/iommu.h> 35 #include <asm/machdep.h> 36 #include <asm/msi_bitmap.h> 37 #include <asm/opal.h> 38 #include <asm/ppc-pci.h> 39 #include <asm/pnv-pci.h> 40 41 #include "powernv.h" 42 #include "pci.h" 43 44 static bool pnv_eeh_nb_init = false; 45 static int eeh_event_irq = -EINVAL; 46 47 static int pnv_eeh_init(void) 48 { 49 struct pci_controller *hose; 50 struct pnv_phb *phb; 51 52 if (!firmware_has_feature(FW_FEATURE_OPAL)) { 53 pr_warn("%s: OPAL is required !\n", 54 __func__); 55 return -EINVAL; 56 } 57 58 /* Set probe mode */ 59 eeh_add_flag(EEH_PROBE_MODE_DEV); 60 61 /* 62 * P7IOC blocks PCI config access to frozen PE, but PHB3 63 * doesn't do that. So we have to selectively enable I/O 64 * prior to collecting error log. 65 */ 66 list_for_each_entry(hose, &hose_list, list_node) { 67 phb = hose->private_data; 68 69 if (phb->model == PNV_PHB_MODEL_P7IOC) 70 eeh_add_flag(EEH_ENABLE_IO_FOR_LOG); 71 72 /* 73 * PE#0 should be regarded as valid by EEH core 74 * if it's not the reserved one. Currently, we 75 * have the reserved PE#255 and PE#127 for PHB3 76 * and P7IOC separately. So we should regard 77 * PE#0 as valid for PHB3 and P7IOC. 78 */ 79 if (phb->ioda.reserved_pe_idx != 0) 80 eeh_add_flag(EEH_VALID_PE_ZERO); 81 82 break; 83 } 84 85 return 0; 86 } 87 88 static irqreturn_t pnv_eeh_event(int irq, void *data) 89 { 90 /* 91 * We simply send a special EEH event if EEH has been 92 * enabled. We don't care about EEH events until we've 93 * finished processing the outstanding ones. Event processing 94 * gets unmasked in next_error() if EEH is enabled. 95 */ 96 disable_irq_nosync(irq); 97 98 if (eeh_enabled()) 99 eeh_send_failure_event(NULL); 100 101 return IRQ_HANDLED; 102 } 103 104 #ifdef CONFIG_DEBUG_FS 105 static ssize_t pnv_eeh_ei_write(struct file *filp, 106 const char __user *user_buf, 107 size_t count, loff_t *ppos) 108 { 109 struct pci_controller *hose = filp->private_data; 110 struct eeh_dev *edev; 111 struct eeh_pe *pe; 112 int pe_no, type, func; 113 unsigned long addr, mask; 114 char buf[50]; 115 int ret; 116 117 if (!eeh_ops || !eeh_ops->err_inject) 118 return -ENXIO; 119 120 /* Copy over argument buffer */ 121 ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count); 122 if (!ret) 123 return -EFAULT; 124 125 /* Retrieve parameters */ 126 ret = sscanf(buf, "%x:%x:%x:%lx:%lx", 127 &pe_no, &type, &func, &addr, &mask); 128 if (ret != 5) 129 return -EINVAL; 130 131 /* Retrieve PE */ 132 edev = kzalloc(sizeof(*edev), GFP_KERNEL); 133 if (!edev) 134 return -ENOMEM; 135 edev->phb = hose; 136 edev->pe_config_addr = pe_no; 137 pe = eeh_pe_get(edev); 138 kfree(edev); 139 if (!pe) 140 return -ENODEV; 141 142 /* Do error injection */ 143 ret = eeh_ops->err_inject(pe, type, func, addr, mask); 144 return ret < 0 ? ret : count; 145 } 146 147 static const struct file_operations pnv_eeh_ei_fops = { 148 .open = simple_open, 149 .llseek = no_llseek, 150 .write = pnv_eeh_ei_write, 151 }; 152 153 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val) 154 { 155 struct pci_controller *hose = data; 156 struct pnv_phb *phb = hose->private_data; 157 158 out_be64(phb->regs + offset, val); 159 return 0; 160 } 161 162 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val) 163 { 164 struct pci_controller *hose = data; 165 struct pnv_phb *phb = hose->private_data; 166 167 *val = in_be64(phb->regs + offset); 168 return 0; 169 } 170 171 #define PNV_EEH_DBGFS_ENTRY(name, reg) \ 172 static int pnv_eeh_dbgfs_set_##name(void *data, u64 val) \ 173 { \ 174 return pnv_eeh_dbgfs_set(data, reg, val); \ 175 } \ 176 \ 177 static int pnv_eeh_dbgfs_get_##name(void *data, u64 *val) \ 178 { \ 179 return pnv_eeh_dbgfs_get(data, reg, val); \ 180 } \ 181 \ 182 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_dbgfs_ops_##name, \ 183 pnv_eeh_dbgfs_get_##name, \ 184 pnv_eeh_dbgfs_set_##name, \ 185 "0x%llx\n") 186 187 PNV_EEH_DBGFS_ENTRY(outb, 0xD10); 188 PNV_EEH_DBGFS_ENTRY(inbA, 0xD90); 189 PNV_EEH_DBGFS_ENTRY(inbB, 0xE10); 190 191 #endif /* CONFIG_DEBUG_FS */ 192 193 /** 194 * pnv_eeh_post_init - EEH platform dependent post initialization 195 * 196 * EEH platform dependent post initialization on powernv. When 197 * the function is called, the EEH PEs and devices should have 198 * been built. If the I/O cache staff has been built, EEH is 199 * ready to supply service. 200 */ 201 static int pnv_eeh_post_init(void) 202 { 203 struct pci_controller *hose; 204 struct pnv_phb *phb; 205 int ret = 0; 206 207 /* Register OPAL event notifier */ 208 if (!pnv_eeh_nb_init) { 209 eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR)); 210 if (eeh_event_irq < 0) { 211 pr_err("%s: Can't register OPAL event interrupt (%d)\n", 212 __func__, eeh_event_irq); 213 return eeh_event_irq; 214 } 215 216 ret = request_irq(eeh_event_irq, pnv_eeh_event, 217 IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL); 218 if (ret < 0) { 219 irq_dispose_mapping(eeh_event_irq); 220 pr_err("%s: Can't request OPAL event interrupt (%d)\n", 221 __func__, eeh_event_irq); 222 return ret; 223 } 224 225 pnv_eeh_nb_init = true; 226 } 227 228 if (!eeh_enabled()) 229 disable_irq(eeh_event_irq); 230 231 list_for_each_entry(hose, &hose_list, list_node) { 232 phb = hose->private_data; 233 234 /* 235 * If EEH is enabled, we're going to rely on that. 236 * Otherwise, we restore to conventional mechanism 237 * to clear frozen PE during PCI config access. 238 */ 239 if (eeh_enabled()) 240 phb->flags |= PNV_PHB_FLAG_EEH; 241 else 242 phb->flags &= ~PNV_PHB_FLAG_EEH; 243 244 /* Create debugfs entries */ 245 #ifdef CONFIG_DEBUG_FS 246 if (phb->has_dbgfs || !phb->dbgfs) 247 continue; 248 249 phb->has_dbgfs = 1; 250 debugfs_create_file("err_injct", 0200, 251 phb->dbgfs, hose, 252 &pnv_eeh_ei_fops); 253 254 debugfs_create_file("err_injct_outbound", 0600, 255 phb->dbgfs, hose, 256 &pnv_eeh_dbgfs_ops_outb); 257 debugfs_create_file("err_injct_inboundA", 0600, 258 phb->dbgfs, hose, 259 &pnv_eeh_dbgfs_ops_inbA); 260 debugfs_create_file("err_injct_inboundB", 0600, 261 phb->dbgfs, hose, 262 &pnv_eeh_dbgfs_ops_inbB); 263 #endif /* CONFIG_DEBUG_FS */ 264 } 265 266 return ret; 267 } 268 269 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap) 270 { 271 int pos = PCI_CAPABILITY_LIST; 272 int cnt = 48; /* Maximal number of capabilities */ 273 u32 status, id; 274 275 if (!pdn) 276 return 0; 277 278 /* Check if the device supports capabilities */ 279 pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status); 280 if (!(status & PCI_STATUS_CAP_LIST)) 281 return 0; 282 283 while (cnt--) { 284 pnv_pci_cfg_read(pdn, pos, 1, &pos); 285 if (pos < 0x40) 286 break; 287 288 pos &= ~3; 289 pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id); 290 if (id == 0xff) 291 break; 292 293 /* Found */ 294 if (id == cap) 295 return pos; 296 297 /* Next one */ 298 pos += PCI_CAP_LIST_NEXT; 299 } 300 301 return 0; 302 } 303 304 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap) 305 { 306 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 307 u32 header; 308 int pos = 256, ttl = (4096 - 256) / 8; 309 310 if (!edev || !edev->pcie_cap) 311 return 0; 312 if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL) 313 return 0; 314 else if (!header) 315 return 0; 316 317 while (ttl-- > 0) { 318 if (PCI_EXT_CAP_ID(header) == cap && pos) 319 return pos; 320 321 pos = PCI_EXT_CAP_NEXT(header); 322 if (pos < 256) 323 break; 324 325 if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL) 326 break; 327 } 328 329 return 0; 330 } 331 332 /** 333 * pnv_eeh_probe - Do probe on PCI device 334 * @pdn: PCI device node 335 * @data: unused 336 * 337 * When EEH module is installed during system boot, all PCI devices 338 * are checked one by one to see if it supports EEH. The function 339 * is introduced for the purpose. By default, EEH has been enabled 340 * on all PCI devices. That's to say, we only need do necessary 341 * initialization on the corresponding eeh device and create PE 342 * accordingly. 343 * 344 * It's notable that's unsafe to retrieve the EEH device through 345 * the corresponding PCI device. During the PCI device hotplug, which 346 * was possiblly triggered by EEH core, the binding between EEH device 347 * and the PCI device isn't built yet. 348 */ 349 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data) 350 { 351 struct pci_controller *hose = pdn->phb; 352 struct pnv_phb *phb = hose->private_data; 353 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 354 uint32_t pcie_flags; 355 int ret; 356 357 /* 358 * When probing the root bridge, which doesn't have any 359 * subordinate PCI devices. We don't have OF node for 360 * the root bridge. So it's not reasonable to continue 361 * the probing. 362 */ 363 if (!edev || edev->pe) 364 return NULL; 365 366 /* Skip for PCI-ISA bridge */ 367 if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA) 368 return NULL; 369 370 /* Initialize eeh device */ 371 edev->class_code = pdn->class_code; 372 edev->mode &= 0xFFFFFF00; 373 edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX); 374 edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP); 375 edev->af_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF); 376 edev->aer_cap = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR); 377 if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) { 378 edev->mode |= EEH_DEV_BRIDGE; 379 if (edev->pcie_cap) { 380 pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS, 381 2, &pcie_flags); 382 pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4; 383 if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT) 384 edev->mode |= EEH_DEV_ROOT_PORT; 385 else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM) 386 edev->mode |= EEH_DEV_DS_PORT; 387 } 388 } 389 390 edev->config_addr = (pdn->busno << 8) | (pdn->devfn); 391 edev->pe_config_addr = phb->ioda.pe_rmap[edev->config_addr]; 392 393 /* Create PE */ 394 ret = eeh_add_to_parent_pe(edev); 395 if (ret) { 396 pr_warn("%s: Can't add PCI dev %04x:%02x:%02x.%01x to parent PE (%x)\n", 397 __func__, hose->global_number, pdn->busno, 398 PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn), ret); 399 return NULL; 400 } 401 402 /* 403 * If the PE contains any one of following adapters, the 404 * PCI config space can't be accessed when dumping EEH log. 405 * Otherwise, we will run into fenced PHB caused by shortage 406 * of outbound credits in the adapter. The PCI config access 407 * should be blocked until PE reset. MMIO access is dropped 408 * by hardware certainly. In order to drop PCI config requests, 409 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which 410 * will be checked in the backend for PE state retrival. If 411 * the PE becomes frozen for the first time and the flag has 412 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for 413 * that PE to block its config space. 414 * 415 * Broadcom Austin 4-ports NICs (14e4:1657) 416 * Broadcom Shiner 4-ports 1G NICs (14e4:168a) 417 * Broadcom Shiner 2-ports 10G NICs (14e4:168e) 418 */ 419 if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM && 420 pdn->device_id == 0x1657) || 421 (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM && 422 pdn->device_id == 0x168a) || 423 (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM && 424 pdn->device_id == 0x168e)) 425 edev->pe->state |= EEH_PE_CFG_RESTRICTED; 426 427 /* 428 * Cache the PE primary bus, which can't be fetched when 429 * full hotplug is in progress. In that case, all child 430 * PCI devices of the PE are expected to be removed prior 431 * to PE reset. 432 */ 433 if (!(edev->pe->state & EEH_PE_PRI_BUS)) { 434 edev->pe->bus = pci_find_bus(hose->global_number, 435 pdn->busno); 436 if (edev->pe->bus) 437 edev->pe->state |= EEH_PE_PRI_BUS; 438 } 439 440 /* 441 * Enable EEH explicitly so that we will do EEH check 442 * while accessing I/O stuff 443 */ 444 eeh_add_flag(EEH_ENABLED); 445 446 /* Save memory bars */ 447 eeh_save_bars(edev); 448 449 return NULL; 450 } 451 452 /** 453 * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable 454 * @pe: EEH PE 455 * @option: operation to be issued 456 * 457 * The function is used to control the EEH functionality globally. 458 * Currently, following options are support according to PAPR: 459 * Enable EEH, Disable EEH, Enable MMIO and Enable DMA 460 */ 461 static int pnv_eeh_set_option(struct eeh_pe *pe, int option) 462 { 463 struct pci_controller *hose = pe->phb; 464 struct pnv_phb *phb = hose->private_data; 465 bool freeze_pe = false; 466 int opt; 467 s64 rc; 468 469 switch (option) { 470 case EEH_OPT_DISABLE: 471 return -EPERM; 472 case EEH_OPT_ENABLE: 473 return 0; 474 case EEH_OPT_THAW_MMIO: 475 opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO; 476 break; 477 case EEH_OPT_THAW_DMA: 478 opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA; 479 break; 480 case EEH_OPT_FREEZE_PE: 481 freeze_pe = true; 482 opt = OPAL_EEH_ACTION_SET_FREEZE_ALL; 483 break; 484 default: 485 pr_warn("%s: Invalid option %d\n", __func__, option); 486 return -EINVAL; 487 } 488 489 /* Freeze master and slave PEs if PHB supports compound PEs */ 490 if (freeze_pe) { 491 if (phb->freeze_pe) { 492 phb->freeze_pe(phb, pe->addr); 493 return 0; 494 } 495 496 rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt); 497 if (rc != OPAL_SUCCESS) { 498 pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n", 499 __func__, rc, phb->hose->global_number, 500 pe->addr); 501 return -EIO; 502 } 503 504 return 0; 505 } 506 507 /* Unfreeze master and slave PEs if PHB supports */ 508 if (phb->unfreeze_pe) 509 return phb->unfreeze_pe(phb, pe->addr, opt); 510 511 rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt); 512 if (rc != OPAL_SUCCESS) { 513 pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n", 514 __func__, rc, option, phb->hose->global_number, 515 pe->addr); 516 return -EIO; 517 } 518 519 return 0; 520 } 521 522 /** 523 * pnv_eeh_get_pe_addr - Retrieve PE address 524 * @pe: EEH PE 525 * 526 * Retrieve the PE address according to the given tranditional 527 * PCI BDF (Bus/Device/Function) address. 528 */ 529 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe) 530 { 531 return pe->addr; 532 } 533 534 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe) 535 { 536 struct pnv_phb *phb = pe->phb->private_data; 537 s64 rc; 538 539 rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data, 540 PNV_PCI_DIAG_BUF_SIZE); 541 if (rc != OPAL_SUCCESS) 542 pr_warn("%s: Failure %lld getting PHB#%x diag-data\n", 543 __func__, rc, pe->phb->global_number); 544 } 545 546 static int pnv_eeh_get_phb_state(struct eeh_pe *pe) 547 { 548 struct pnv_phb *phb = pe->phb->private_data; 549 u8 fstate; 550 __be16 pcierr; 551 s64 rc; 552 int result = 0; 553 554 rc = opal_pci_eeh_freeze_status(phb->opal_id, 555 pe->addr, 556 &fstate, 557 &pcierr, 558 NULL); 559 if (rc != OPAL_SUCCESS) { 560 pr_warn("%s: Failure %lld getting PHB#%x state\n", 561 __func__, rc, phb->hose->global_number); 562 return EEH_STATE_NOT_SUPPORT; 563 } 564 565 /* 566 * Check PHB state. If the PHB is frozen for the 567 * first time, to dump the PHB diag-data. 568 */ 569 if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) { 570 result = (EEH_STATE_MMIO_ACTIVE | 571 EEH_STATE_DMA_ACTIVE | 572 EEH_STATE_MMIO_ENABLED | 573 EEH_STATE_DMA_ENABLED); 574 } else if (!(pe->state & EEH_PE_ISOLATED)) { 575 eeh_pe_state_mark(pe, EEH_PE_ISOLATED); 576 pnv_eeh_get_phb_diag(pe); 577 578 if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) 579 pnv_pci_dump_phb_diag_data(pe->phb, pe->data); 580 } 581 582 return result; 583 } 584 585 static int pnv_eeh_get_pe_state(struct eeh_pe *pe) 586 { 587 struct pnv_phb *phb = pe->phb->private_data; 588 u8 fstate; 589 __be16 pcierr; 590 s64 rc; 591 int result; 592 593 /* 594 * We don't clobber hardware frozen state until PE 595 * reset is completed. In order to keep EEH core 596 * moving forward, we have to return operational 597 * state during PE reset. 598 */ 599 if (pe->state & EEH_PE_RESET) { 600 result = (EEH_STATE_MMIO_ACTIVE | 601 EEH_STATE_DMA_ACTIVE | 602 EEH_STATE_MMIO_ENABLED | 603 EEH_STATE_DMA_ENABLED); 604 return result; 605 } 606 607 /* 608 * Fetch PE state from hardware. If the PHB 609 * supports compound PE, let it handle that. 610 */ 611 if (phb->get_pe_state) { 612 fstate = phb->get_pe_state(phb, pe->addr); 613 } else { 614 rc = opal_pci_eeh_freeze_status(phb->opal_id, 615 pe->addr, 616 &fstate, 617 &pcierr, 618 NULL); 619 if (rc != OPAL_SUCCESS) { 620 pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n", 621 __func__, rc, phb->hose->global_number, 622 pe->addr); 623 return EEH_STATE_NOT_SUPPORT; 624 } 625 } 626 627 /* Figure out state */ 628 switch (fstate) { 629 case OPAL_EEH_STOPPED_NOT_FROZEN: 630 result = (EEH_STATE_MMIO_ACTIVE | 631 EEH_STATE_DMA_ACTIVE | 632 EEH_STATE_MMIO_ENABLED | 633 EEH_STATE_DMA_ENABLED); 634 break; 635 case OPAL_EEH_STOPPED_MMIO_FREEZE: 636 result = (EEH_STATE_DMA_ACTIVE | 637 EEH_STATE_DMA_ENABLED); 638 break; 639 case OPAL_EEH_STOPPED_DMA_FREEZE: 640 result = (EEH_STATE_MMIO_ACTIVE | 641 EEH_STATE_MMIO_ENABLED); 642 break; 643 case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE: 644 result = 0; 645 break; 646 case OPAL_EEH_STOPPED_RESET: 647 result = EEH_STATE_RESET_ACTIVE; 648 break; 649 case OPAL_EEH_STOPPED_TEMP_UNAVAIL: 650 result = EEH_STATE_UNAVAILABLE; 651 break; 652 case OPAL_EEH_STOPPED_PERM_UNAVAIL: 653 result = EEH_STATE_NOT_SUPPORT; 654 break; 655 default: 656 result = EEH_STATE_NOT_SUPPORT; 657 pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n", 658 __func__, phb->hose->global_number, 659 pe->addr, fstate); 660 } 661 662 /* 663 * If PHB supports compound PE, to freeze all 664 * slave PEs for consistency. 665 * 666 * If the PE is switching to frozen state for the 667 * first time, to dump the PHB diag-data. 668 */ 669 if (!(result & EEH_STATE_NOT_SUPPORT) && 670 !(result & EEH_STATE_UNAVAILABLE) && 671 !(result & EEH_STATE_MMIO_ACTIVE) && 672 !(result & EEH_STATE_DMA_ACTIVE) && 673 !(pe->state & EEH_PE_ISOLATED)) { 674 if (phb->freeze_pe) 675 phb->freeze_pe(phb, pe->addr); 676 677 eeh_pe_state_mark(pe, EEH_PE_ISOLATED); 678 pnv_eeh_get_phb_diag(pe); 679 680 if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) 681 pnv_pci_dump_phb_diag_data(pe->phb, pe->data); 682 } 683 684 return result; 685 } 686 687 /** 688 * pnv_eeh_get_state - Retrieve PE state 689 * @pe: EEH PE 690 * @delay: delay while PE state is temporarily unavailable 691 * 692 * Retrieve the state of the specified PE. For IODA-compitable 693 * platform, it should be retrieved from IODA table. Therefore, 694 * we prefer passing down to hardware implementation to handle 695 * it. 696 */ 697 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay) 698 { 699 int ret; 700 701 if (pe->type & EEH_PE_PHB) 702 ret = pnv_eeh_get_phb_state(pe); 703 else 704 ret = pnv_eeh_get_pe_state(pe); 705 706 if (!delay) 707 return ret; 708 709 /* 710 * If the PE state is temporarily unavailable, 711 * to inform the EEH core delay for default 712 * period (1 second) 713 */ 714 *delay = 0; 715 if (ret & EEH_STATE_UNAVAILABLE) 716 *delay = 1000; 717 718 return ret; 719 } 720 721 static s64 pnv_eeh_poll(unsigned long id) 722 { 723 s64 rc = OPAL_HARDWARE; 724 725 while (1) { 726 rc = opal_pci_poll(id); 727 if (rc <= 0) 728 break; 729 730 if (system_state < SYSTEM_RUNNING) 731 udelay(1000 * rc); 732 else 733 msleep(rc); 734 } 735 736 return rc; 737 } 738 739 int pnv_eeh_phb_reset(struct pci_controller *hose, int option) 740 { 741 struct pnv_phb *phb = hose->private_data; 742 s64 rc = OPAL_HARDWARE; 743 744 pr_debug("%s: Reset PHB#%x, option=%d\n", 745 __func__, hose->global_number, option); 746 747 /* Issue PHB complete reset request */ 748 if (option == EEH_RESET_FUNDAMENTAL || 749 option == EEH_RESET_HOT) 750 rc = opal_pci_reset(phb->opal_id, 751 OPAL_RESET_PHB_COMPLETE, 752 OPAL_ASSERT_RESET); 753 else if (option == EEH_RESET_DEACTIVATE) 754 rc = opal_pci_reset(phb->opal_id, 755 OPAL_RESET_PHB_COMPLETE, 756 OPAL_DEASSERT_RESET); 757 if (rc < 0) 758 goto out; 759 760 /* 761 * Poll state of the PHB until the request is done 762 * successfully. The PHB reset is usually PHB complete 763 * reset followed by hot reset on root bus. So we also 764 * need the PCI bus settlement delay. 765 */ 766 if (rc > 0) 767 rc = pnv_eeh_poll(phb->opal_id); 768 if (option == EEH_RESET_DEACTIVATE) { 769 if (system_state < SYSTEM_RUNNING) 770 udelay(1000 * EEH_PE_RST_SETTLE_TIME); 771 else 772 msleep(EEH_PE_RST_SETTLE_TIME); 773 } 774 out: 775 if (rc != OPAL_SUCCESS) 776 return -EIO; 777 778 return 0; 779 } 780 781 static int pnv_eeh_root_reset(struct pci_controller *hose, int option) 782 { 783 struct pnv_phb *phb = hose->private_data; 784 s64 rc = OPAL_HARDWARE; 785 786 pr_debug("%s: Reset PHB#%x, option=%d\n", 787 __func__, hose->global_number, option); 788 789 /* 790 * During the reset deassert time, we needn't care 791 * the reset scope because the firmware does nothing 792 * for fundamental or hot reset during deassert phase. 793 */ 794 if (option == EEH_RESET_FUNDAMENTAL) 795 rc = opal_pci_reset(phb->opal_id, 796 OPAL_RESET_PCI_FUNDAMENTAL, 797 OPAL_ASSERT_RESET); 798 else if (option == EEH_RESET_HOT) 799 rc = opal_pci_reset(phb->opal_id, 800 OPAL_RESET_PCI_HOT, 801 OPAL_ASSERT_RESET); 802 else if (option == EEH_RESET_DEACTIVATE) 803 rc = opal_pci_reset(phb->opal_id, 804 OPAL_RESET_PCI_HOT, 805 OPAL_DEASSERT_RESET); 806 if (rc < 0) 807 goto out; 808 809 /* Poll state of the PHB until the request is done */ 810 if (rc > 0) 811 rc = pnv_eeh_poll(phb->opal_id); 812 if (option == EEH_RESET_DEACTIVATE) 813 msleep(EEH_PE_RST_SETTLE_TIME); 814 out: 815 if (rc != OPAL_SUCCESS) 816 return -EIO; 817 818 return 0; 819 } 820 821 static int __pnv_eeh_bridge_reset(struct pci_dev *dev, int option) 822 { 823 struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn); 824 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 825 int aer = edev ? edev->aer_cap : 0; 826 u32 ctrl; 827 828 pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n", 829 __func__, pci_domain_nr(dev->bus), 830 dev->bus->number, option); 831 832 switch (option) { 833 case EEH_RESET_FUNDAMENTAL: 834 case EEH_RESET_HOT: 835 /* Don't report linkDown event */ 836 if (aer) { 837 eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK, 838 4, &ctrl); 839 ctrl |= PCI_ERR_UNC_SURPDN; 840 eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK, 841 4, ctrl); 842 } 843 844 eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl); 845 ctrl |= PCI_BRIDGE_CTL_BUS_RESET; 846 eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl); 847 848 msleep(EEH_PE_RST_HOLD_TIME); 849 break; 850 case EEH_RESET_DEACTIVATE: 851 eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl); 852 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET; 853 eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl); 854 855 msleep(EEH_PE_RST_SETTLE_TIME); 856 857 /* Continue reporting linkDown event */ 858 if (aer) { 859 eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK, 860 4, &ctrl); 861 ctrl &= ~PCI_ERR_UNC_SURPDN; 862 eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK, 863 4, ctrl); 864 } 865 866 break; 867 } 868 869 return 0; 870 } 871 872 static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option) 873 { 874 struct pci_controller *hose = pci_bus_to_host(pdev->bus); 875 struct pnv_phb *phb = hose->private_data; 876 struct device_node *dn = pci_device_to_OF_node(pdev); 877 uint64_t id = PCI_SLOT_ID(phb->opal_id, 878 (pdev->bus->number << 8) | pdev->devfn); 879 uint8_t scope; 880 int64_t rc; 881 882 /* Hot reset to the bus if firmware cannot handle */ 883 if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL)) 884 return __pnv_eeh_bridge_reset(pdev, option); 885 886 switch (option) { 887 case EEH_RESET_FUNDAMENTAL: 888 scope = OPAL_RESET_PCI_FUNDAMENTAL; 889 break; 890 case EEH_RESET_HOT: 891 scope = OPAL_RESET_PCI_HOT; 892 break; 893 case EEH_RESET_DEACTIVATE: 894 return 0; 895 default: 896 dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n", 897 __func__, option); 898 return -EINVAL; 899 } 900 901 rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET); 902 if (rc <= OPAL_SUCCESS) 903 goto out; 904 905 rc = pnv_eeh_poll(id); 906 out: 907 return (rc == OPAL_SUCCESS) ? 0 : -EIO; 908 } 909 910 void pnv_pci_reset_secondary_bus(struct pci_dev *dev) 911 { 912 struct pci_controller *hose; 913 914 if (pci_is_root_bus(dev->bus)) { 915 hose = pci_bus_to_host(dev->bus); 916 pnv_eeh_root_reset(hose, EEH_RESET_HOT); 917 pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE); 918 } else { 919 pnv_eeh_bridge_reset(dev, EEH_RESET_HOT); 920 pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE); 921 } 922 } 923 924 static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type, 925 int pos, u16 mask) 926 { 927 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 928 int i, status = 0; 929 930 /* Wait for Transaction Pending bit to be cleared */ 931 for (i = 0; i < 4; i++) { 932 eeh_ops->read_config(pdn, pos, 2, &status); 933 if (!(status & mask)) 934 return; 935 936 msleep((1 << i) * 100); 937 } 938 939 pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n", 940 __func__, type, 941 edev->phb->global_number, pdn->busno, 942 PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn)); 943 } 944 945 static int pnv_eeh_do_flr(struct pci_dn *pdn, int option) 946 { 947 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 948 u32 reg = 0; 949 950 if (WARN_ON(!edev->pcie_cap)) 951 return -ENOTTY; 952 953 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP, 4, ®); 954 if (!(reg & PCI_EXP_DEVCAP_FLR)) 955 return -ENOTTY; 956 957 switch (option) { 958 case EEH_RESET_HOT: 959 case EEH_RESET_FUNDAMENTAL: 960 pnv_eeh_wait_for_pending(pdn, "", 961 edev->pcie_cap + PCI_EXP_DEVSTA, 962 PCI_EXP_DEVSTA_TRPND); 963 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 964 4, ®); 965 reg |= PCI_EXP_DEVCTL_BCR_FLR; 966 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 967 4, reg); 968 msleep(EEH_PE_RST_HOLD_TIME); 969 break; 970 case EEH_RESET_DEACTIVATE: 971 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 972 4, ®); 973 reg &= ~PCI_EXP_DEVCTL_BCR_FLR; 974 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 975 4, reg); 976 msleep(EEH_PE_RST_SETTLE_TIME); 977 break; 978 } 979 980 return 0; 981 } 982 983 static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option) 984 { 985 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 986 u32 cap = 0; 987 988 if (WARN_ON(!edev->af_cap)) 989 return -ENOTTY; 990 991 eeh_ops->read_config(pdn, edev->af_cap + PCI_AF_CAP, 1, &cap); 992 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR)) 993 return -ENOTTY; 994 995 switch (option) { 996 case EEH_RESET_HOT: 997 case EEH_RESET_FUNDAMENTAL: 998 /* 999 * Wait for Transaction Pending bit to clear. A word-aligned 1000 * test is used, so we use the conrol offset rather than status 1001 * and shift the test bit to match. 1002 */ 1003 pnv_eeh_wait_for_pending(pdn, "AF", 1004 edev->af_cap + PCI_AF_CTRL, 1005 PCI_AF_STATUS_TP << 8); 1006 eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL, 1007 1, PCI_AF_CTRL_FLR); 1008 msleep(EEH_PE_RST_HOLD_TIME); 1009 break; 1010 case EEH_RESET_DEACTIVATE: 1011 eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL, 1, 0); 1012 msleep(EEH_PE_RST_SETTLE_TIME); 1013 break; 1014 } 1015 1016 return 0; 1017 } 1018 1019 static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option) 1020 { 1021 struct eeh_dev *edev; 1022 struct pci_dn *pdn; 1023 int ret; 1024 1025 /* The VF PE should have only one child device */ 1026 edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, list); 1027 pdn = eeh_dev_to_pdn(edev); 1028 if (!pdn) 1029 return -ENXIO; 1030 1031 ret = pnv_eeh_do_flr(pdn, option); 1032 if (!ret) 1033 return ret; 1034 1035 return pnv_eeh_do_af_flr(pdn, option); 1036 } 1037 1038 /** 1039 * pnv_eeh_reset - Reset the specified PE 1040 * @pe: EEH PE 1041 * @option: reset option 1042 * 1043 * Do reset on the indicated PE. For PCI bus sensitive PE, 1044 * we need to reset the parent p2p bridge. The PHB has to 1045 * be reinitialized if the p2p bridge is root bridge. For 1046 * PCI device sensitive PE, we will try to reset the device 1047 * through FLR. For now, we don't have OPAL APIs to do HARD 1048 * reset yet, so all reset would be SOFT (HOT) reset. 1049 */ 1050 static int pnv_eeh_reset(struct eeh_pe *pe, int option) 1051 { 1052 struct pci_controller *hose = pe->phb; 1053 struct pnv_phb *phb; 1054 struct pci_bus *bus; 1055 int64_t rc; 1056 1057 /* 1058 * For PHB reset, we always have complete reset. For those PEs whose 1059 * primary bus derived from root complex (root bus) or root port 1060 * (usually bus#1), we apply hot or fundamental reset on the root port. 1061 * For other PEs, we always have hot reset on the PE primary bus. 1062 * 1063 * Here, we have different design to pHyp, which always clear the 1064 * frozen state during PE reset. However, the good idea here from 1065 * benh is to keep frozen state before we get PE reset done completely 1066 * (until BAR restore). With the frozen state, HW drops illegal IO 1067 * or MMIO access, which can incur recrusive frozen PE during PE 1068 * reset. The side effect is that EEH core has to clear the frozen 1069 * state explicitly after BAR restore. 1070 */ 1071 if (pe->type & EEH_PE_PHB) 1072 return pnv_eeh_phb_reset(hose, option); 1073 1074 /* 1075 * The frozen PE might be caused by PAPR error injection 1076 * registers, which are expected to be cleared after hitting 1077 * frozen PE as stated in the hardware spec. Unfortunately, 1078 * that's not true on P7IOC. So we have to clear it manually 1079 * to avoid recursive EEH errors during recovery. 1080 */ 1081 phb = hose->private_data; 1082 if (phb->model == PNV_PHB_MODEL_P7IOC && 1083 (option == EEH_RESET_HOT || 1084 option == EEH_RESET_FUNDAMENTAL)) { 1085 rc = opal_pci_reset(phb->opal_id, 1086 OPAL_RESET_PHB_ERROR, 1087 OPAL_ASSERT_RESET); 1088 if (rc != OPAL_SUCCESS) { 1089 pr_warn("%s: Failure %lld clearing error injection registers\n", 1090 __func__, rc); 1091 return -EIO; 1092 } 1093 } 1094 1095 if (pe->type & EEH_PE_VF) 1096 return pnv_eeh_reset_vf_pe(pe, option); 1097 1098 bus = eeh_pe_bus_get(pe); 1099 if (!bus) { 1100 pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n", 1101 __func__, pe->phb->global_number, pe->addr); 1102 return -EIO; 1103 } 1104 1105 if (pci_is_root_bus(bus) || 1106 pci_is_root_bus(bus->parent)) 1107 return pnv_eeh_root_reset(hose, option); 1108 1109 return pnv_eeh_bridge_reset(bus->self, option); 1110 } 1111 1112 /** 1113 * pnv_eeh_wait_state - Wait for PE state 1114 * @pe: EEH PE 1115 * @max_wait: maximal period in millisecond 1116 * 1117 * Wait for the state of associated PE. It might take some time 1118 * to retrieve the PE's state. 1119 */ 1120 static int pnv_eeh_wait_state(struct eeh_pe *pe, int max_wait) 1121 { 1122 int ret; 1123 int mwait; 1124 1125 while (1) { 1126 ret = pnv_eeh_get_state(pe, &mwait); 1127 1128 /* 1129 * If the PE's state is temporarily unavailable, 1130 * we have to wait for the specified time. Otherwise, 1131 * the PE's state will be returned immediately. 1132 */ 1133 if (ret != EEH_STATE_UNAVAILABLE) 1134 return ret; 1135 1136 if (max_wait <= 0) { 1137 pr_warn("%s: Timeout getting PE#%x's state (%d)\n", 1138 __func__, pe->addr, max_wait); 1139 return EEH_STATE_NOT_SUPPORT; 1140 } 1141 1142 max_wait -= mwait; 1143 msleep(mwait); 1144 } 1145 1146 return EEH_STATE_NOT_SUPPORT; 1147 } 1148 1149 /** 1150 * pnv_eeh_get_log - Retrieve error log 1151 * @pe: EEH PE 1152 * @severity: temporary or permanent error log 1153 * @drv_log: driver log to be combined with retrieved error log 1154 * @len: length of driver log 1155 * 1156 * Retrieve the temporary or permanent error from the PE. 1157 */ 1158 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity, 1159 char *drv_log, unsigned long len) 1160 { 1161 if (!eeh_has_flag(EEH_EARLY_DUMP_LOG)) 1162 pnv_pci_dump_phb_diag_data(pe->phb, pe->data); 1163 1164 return 0; 1165 } 1166 1167 /** 1168 * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE 1169 * @pe: EEH PE 1170 * 1171 * The function will be called to reconfigure the bridges included 1172 * in the specified PE so that the mulfunctional PE would be recovered 1173 * again. 1174 */ 1175 static int pnv_eeh_configure_bridge(struct eeh_pe *pe) 1176 { 1177 return 0; 1178 } 1179 1180 /** 1181 * pnv_pe_err_inject - Inject specified error to the indicated PE 1182 * @pe: the indicated PE 1183 * @type: error type 1184 * @func: specific error type 1185 * @addr: address 1186 * @mask: address mask 1187 * 1188 * The routine is called to inject specified error, which is 1189 * determined by @type and @func, to the indicated PE for 1190 * testing purpose. 1191 */ 1192 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func, 1193 unsigned long addr, unsigned long mask) 1194 { 1195 struct pci_controller *hose = pe->phb; 1196 struct pnv_phb *phb = hose->private_data; 1197 s64 rc; 1198 1199 if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR && 1200 type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) { 1201 pr_warn("%s: Invalid error type %d\n", 1202 __func__, type); 1203 return -ERANGE; 1204 } 1205 1206 if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR || 1207 func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) { 1208 pr_warn("%s: Invalid error function %d\n", 1209 __func__, func); 1210 return -ERANGE; 1211 } 1212 1213 /* Firmware supports error injection ? */ 1214 if (!opal_check_token(OPAL_PCI_ERR_INJECT)) { 1215 pr_warn("%s: Firmware doesn't support error injection\n", 1216 __func__); 1217 return -ENXIO; 1218 } 1219 1220 /* Do error injection */ 1221 rc = opal_pci_err_inject(phb->opal_id, pe->addr, 1222 type, func, addr, mask); 1223 if (rc != OPAL_SUCCESS) { 1224 pr_warn("%s: Failure %lld injecting error " 1225 "%d-%d to PHB#%x-PE#%x\n", 1226 __func__, rc, type, func, 1227 hose->global_number, pe->addr); 1228 return -EIO; 1229 } 1230 1231 return 0; 1232 } 1233 1234 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn) 1235 { 1236 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 1237 1238 if (!edev || !edev->pe) 1239 return false; 1240 1241 /* 1242 * We will issue FLR or AF FLR to all VFs, which are contained 1243 * in VF PE. It relies on the EEH PCI config accessors. So we 1244 * can't block them during the window. 1245 */ 1246 if (edev->physfn && (edev->pe->state & EEH_PE_RESET)) 1247 return false; 1248 1249 if (edev->pe->state & EEH_PE_CFG_BLOCKED) 1250 return true; 1251 1252 return false; 1253 } 1254 1255 static int pnv_eeh_read_config(struct pci_dn *pdn, 1256 int where, int size, u32 *val) 1257 { 1258 if (!pdn) 1259 return PCIBIOS_DEVICE_NOT_FOUND; 1260 1261 if (pnv_eeh_cfg_blocked(pdn)) { 1262 *val = 0xFFFFFFFF; 1263 return PCIBIOS_SET_FAILED; 1264 } 1265 1266 return pnv_pci_cfg_read(pdn, where, size, val); 1267 } 1268 1269 static int pnv_eeh_write_config(struct pci_dn *pdn, 1270 int where, int size, u32 val) 1271 { 1272 if (!pdn) 1273 return PCIBIOS_DEVICE_NOT_FOUND; 1274 1275 if (pnv_eeh_cfg_blocked(pdn)) 1276 return PCIBIOS_SET_FAILED; 1277 1278 return pnv_pci_cfg_write(pdn, where, size, val); 1279 } 1280 1281 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data) 1282 { 1283 /* GEM */ 1284 if (data->gemXfir || data->gemRfir || 1285 data->gemRirqfir || data->gemMask || data->gemRwof) 1286 pr_info(" GEM: %016llx %016llx %016llx %016llx %016llx\n", 1287 be64_to_cpu(data->gemXfir), 1288 be64_to_cpu(data->gemRfir), 1289 be64_to_cpu(data->gemRirqfir), 1290 be64_to_cpu(data->gemMask), 1291 be64_to_cpu(data->gemRwof)); 1292 1293 /* LEM */ 1294 if (data->lemFir || data->lemErrMask || 1295 data->lemAction0 || data->lemAction1 || data->lemWof) 1296 pr_info(" LEM: %016llx %016llx %016llx %016llx %016llx\n", 1297 be64_to_cpu(data->lemFir), 1298 be64_to_cpu(data->lemErrMask), 1299 be64_to_cpu(data->lemAction0), 1300 be64_to_cpu(data->lemAction1), 1301 be64_to_cpu(data->lemWof)); 1302 } 1303 1304 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose) 1305 { 1306 struct pnv_phb *phb = hose->private_data; 1307 struct OpalIoP7IOCErrorData *data = &phb->diag.hub_diag; 1308 long rc; 1309 1310 rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data)); 1311 if (rc != OPAL_SUCCESS) { 1312 pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n", 1313 __func__, phb->hub_id, rc); 1314 return; 1315 } 1316 1317 switch (be16_to_cpu(data->type)) { 1318 case OPAL_P7IOC_DIAG_TYPE_RGC: 1319 pr_info("P7IOC diag-data for RGC\n\n"); 1320 pnv_eeh_dump_hub_diag_common(data); 1321 if (data->rgc.rgcStatus || data->rgc.rgcLdcp) 1322 pr_info(" RGC: %016llx %016llx\n", 1323 be64_to_cpu(data->rgc.rgcStatus), 1324 be64_to_cpu(data->rgc.rgcLdcp)); 1325 break; 1326 case OPAL_P7IOC_DIAG_TYPE_BI: 1327 pr_info("P7IOC diag-data for BI %s\n\n", 1328 data->bi.biDownbound ? "Downbound" : "Upbound"); 1329 pnv_eeh_dump_hub_diag_common(data); 1330 if (data->bi.biLdcp0 || data->bi.biLdcp1 || 1331 data->bi.biLdcp2 || data->bi.biFenceStatus) 1332 pr_info(" BI: %016llx %016llx %016llx %016llx\n", 1333 be64_to_cpu(data->bi.biLdcp0), 1334 be64_to_cpu(data->bi.biLdcp1), 1335 be64_to_cpu(data->bi.biLdcp2), 1336 be64_to_cpu(data->bi.biFenceStatus)); 1337 break; 1338 case OPAL_P7IOC_DIAG_TYPE_CI: 1339 pr_info("P7IOC diag-data for CI Port %d\n\n", 1340 data->ci.ciPort); 1341 pnv_eeh_dump_hub_diag_common(data); 1342 if (data->ci.ciPortStatus || data->ci.ciPortLdcp) 1343 pr_info(" CI: %016llx %016llx\n", 1344 be64_to_cpu(data->ci.ciPortStatus), 1345 be64_to_cpu(data->ci.ciPortLdcp)); 1346 break; 1347 case OPAL_P7IOC_DIAG_TYPE_MISC: 1348 pr_info("P7IOC diag-data for MISC\n\n"); 1349 pnv_eeh_dump_hub_diag_common(data); 1350 break; 1351 case OPAL_P7IOC_DIAG_TYPE_I2C: 1352 pr_info("P7IOC diag-data for I2C\n\n"); 1353 pnv_eeh_dump_hub_diag_common(data); 1354 break; 1355 default: 1356 pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n", 1357 __func__, phb->hub_id, data->type); 1358 } 1359 } 1360 1361 static int pnv_eeh_get_pe(struct pci_controller *hose, 1362 u16 pe_no, struct eeh_pe **pe) 1363 { 1364 struct pnv_phb *phb = hose->private_data; 1365 struct pnv_ioda_pe *pnv_pe; 1366 struct eeh_pe *dev_pe; 1367 struct eeh_dev edev; 1368 1369 /* 1370 * If PHB supports compound PE, to fetch 1371 * the master PE because slave PE is invisible 1372 * to EEH core. 1373 */ 1374 pnv_pe = &phb->ioda.pe_array[pe_no]; 1375 if (pnv_pe->flags & PNV_IODA_PE_SLAVE) { 1376 pnv_pe = pnv_pe->master; 1377 WARN_ON(!pnv_pe || 1378 !(pnv_pe->flags & PNV_IODA_PE_MASTER)); 1379 pe_no = pnv_pe->pe_number; 1380 } 1381 1382 /* Find the PE according to PE# */ 1383 memset(&edev, 0, sizeof(struct eeh_dev)); 1384 edev.phb = hose; 1385 edev.pe_config_addr = pe_no; 1386 dev_pe = eeh_pe_get(&edev); 1387 if (!dev_pe) 1388 return -EEXIST; 1389 1390 /* Freeze the (compound) PE */ 1391 *pe = dev_pe; 1392 if (!(dev_pe->state & EEH_PE_ISOLATED)) 1393 phb->freeze_pe(phb, pe_no); 1394 1395 /* 1396 * At this point, we're sure the (compound) PE should 1397 * have been frozen. However, we still need poke until 1398 * hitting the frozen PE on top level. 1399 */ 1400 dev_pe = dev_pe->parent; 1401 while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) { 1402 int ret; 1403 int active_flags = (EEH_STATE_MMIO_ACTIVE | 1404 EEH_STATE_DMA_ACTIVE); 1405 1406 ret = eeh_ops->get_state(dev_pe, NULL); 1407 if (ret <= 0 || (ret & active_flags) == active_flags) { 1408 dev_pe = dev_pe->parent; 1409 continue; 1410 } 1411 1412 /* Frozen parent PE */ 1413 *pe = dev_pe; 1414 if (!(dev_pe->state & EEH_PE_ISOLATED)) 1415 phb->freeze_pe(phb, dev_pe->addr); 1416 1417 /* Next one */ 1418 dev_pe = dev_pe->parent; 1419 } 1420 1421 return 0; 1422 } 1423 1424 /** 1425 * pnv_eeh_next_error - Retrieve next EEH error to handle 1426 * @pe: Affected PE 1427 * 1428 * The function is expected to be called by EEH core while it gets 1429 * special EEH event (without binding PE). The function calls to 1430 * OPAL APIs for next error to handle. The informational error is 1431 * handled internally by platform. However, the dead IOC, dead PHB, 1432 * fenced PHB and frozen PE should be handled by EEH core eventually. 1433 */ 1434 static int pnv_eeh_next_error(struct eeh_pe **pe) 1435 { 1436 struct pci_controller *hose; 1437 struct pnv_phb *phb; 1438 struct eeh_pe *phb_pe, *parent_pe; 1439 __be64 frozen_pe_no; 1440 __be16 err_type, severity; 1441 int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE); 1442 long rc; 1443 int state, ret = EEH_NEXT_ERR_NONE; 1444 1445 /* 1446 * While running here, it's safe to purge the event queue. The 1447 * event should still be masked. 1448 */ 1449 eeh_remove_event(NULL, false); 1450 1451 list_for_each_entry(hose, &hose_list, list_node) { 1452 /* 1453 * If the subordinate PCI buses of the PHB has been 1454 * removed or is exactly under error recovery, we 1455 * needn't take care of it any more. 1456 */ 1457 phb = hose->private_data; 1458 phb_pe = eeh_phb_pe_get(hose); 1459 if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED)) 1460 continue; 1461 1462 rc = opal_pci_next_error(phb->opal_id, 1463 &frozen_pe_no, &err_type, &severity); 1464 if (rc != OPAL_SUCCESS) { 1465 pr_devel("%s: Invalid return value on " 1466 "PHB#%x (0x%lx) from opal_pci_next_error", 1467 __func__, hose->global_number, rc); 1468 continue; 1469 } 1470 1471 /* If the PHB doesn't have error, stop processing */ 1472 if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR || 1473 be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) { 1474 pr_devel("%s: No error found on PHB#%x\n", 1475 __func__, hose->global_number); 1476 continue; 1477 } 1478 1479 /* 1480 * Processing the error. We're expecting the error with 1481 * highest priority reported upon multiple errors on the 1482 * specific PHB. 1483 */ 1484 pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n", 1485 __func__, be16_to_cpu(err_type), 1486 be16_to_cpu(severity), be64_to_cpu(frozen_pe_no), 1487 hose->global_number); 1488 switch (be16_to_cpu(err_type)) { 1489 case OPAL_EEH_IOC_ERROR: 1490 if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) { 1491 pr_err("EEH: dead IOC detected\n"); 1492 ret = EEH_NEXT_ERR_DEAD_IOC; 1493 } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) { 1494 pr_info("EEH: IOC informative error " 1495 "detected\n"); 1496 pnv_eeh_get_and_dump_hub_diag(hose); 1497 ret = EEH_NEXT_ERR_NONE; 1498 } 1499 1500 break; 1501 case OPAL_EEH_PHB_ERROR: 1502 if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) { 1503 *pe = phb_pe; 1504 pr_err("EEH: dead PHB#%x detected, " 1505 "location: %s\n", 1506 hose->global_number, 1507 eeh_pe_loc_get(phb_pe)); 1508 ret = EEH_NEXT_ERR_DEAD_PHB; 1509 } else if (be16_to_cpu(severity) == 1510 OPAL_EEH_SEV_PHB_FENCED) { 1511 *pe = phb_pe; 1512 pr_err("EEH: Fenced PHB#%x detected, " 1513 "location: %s\n", 1514 hose->global_number, 1515 eeh_pe_loc_get(phb_pe)); 1516 ret = EEH_NEXT_ERR_FENCED_PHB; 1517 } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) { 1518 pr_info("EEH: PHB#%x informative error " 1519 "detected, location: %s\n", 1520 hose->global_number, 1521 eeh_pe_loc_get(phb_pe)); 1522 pnv_eeh_get_phb_diag(phb_pe); 1523 pnv_pci_dump_phb_diag_data(hose, phb_pe->data); 1524 ret = EEH_NEXT_ERR_NONE; 1525 } 1526 1527 break; 1528 case OPAL_EEH_PE_ERROR: 1529 /* 1530 * If we can't find the corresponding PE, we 1531 * just try to unfreeze. 1532 */ 1533 if (pnv_eeh_get_pe(hose, 1534 be64_to_cpu(frozen_pe_no), pe)) { 1535 pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n", 1536 hose->global_number, be64_to_cpu(frozen_pe_no)); 1537 pr_info("EEH: PHB location: %s\n", 1538 eeh_pe_loc_get(phb_pe)); 1539 1540 /* Dump PHB diag-data */ 1541 rc = opal_pci_get_phb_diag_data2(phb->opal_id, 1542 phb->diag.blob, PNV_PCI_DIAG_BUF_SIZE); 1543 if (rc == OPAL_SUCCESS) 1544 pnv_pci_dump_phb_diag_data(hose, 1545 phb->diag.blob); 1546 1547 /* Try best to clear it */ 1548 opal_pci_eeh_freeze_clear(phb->opal_id, 1549 be64_to_cpu(frozen_pe_no), 1550 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL); 1551 ret = EEH_NEXT_ERR_NONE; 1552 } else if ((*pe)->state & EEH_PE_ISOLATED || 1553 eeh_pe_passed(*pe)) { 1554 ret = EEH_NEXT_ERR_NONE; 1555 } else { 1556 pr_err("EEH: Frozen PE#%x " 1557 "on PHB#%x detected\n", 1558 (*pe)->addr, 1559 (*pe)->phb->global_number); 1560 pr_err("EEH: PE location: %s, " 1561 "PHB location: %s\n", 1562 eeh_pe_loc_get(*pe), 1563 eeh_pe_loc_get(phb_pe)); 1564 ret = EEH_NEXT_ERR_FROZEN_PE; 1565 } 1566 1567 break; 1568 default: 1569 pr_warn("%s: Unexpected error type %d\n", 1570 __func__, be16_to_cpu(err_type)); 1571 } 1572 1573 /* 1574 * EEH core will try recover from fenced PHB or 1575 * frozen PE. In the time for frozen PE, EEH core 1576 * enable IO path for that before collecting logs, 1577 * but it ruins the site. So we have to dump the 1578 * log in advance here. 1579 */ 1580 if ((ret == EEH_NEXT_ERR_FROZEN_PE || 1581 ret == EEH_NEXT_ERR_FENCED_PHB) && 1582 !((*pe)->state & EEH_PE_ISOLATED)) { 1583 eeh_pe_state_mark(*pe, EEH_PE_ISOLATED); 1584 pnv_eeh_get_phb_diag(*pe); 1585 1586 if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) 1587 pnv_pci_dump_phb_diag_data((*pe)->phb, 1588 (*pe)->data); 1589 } 1590 1591 /* 1592 * We probably have the frozen parent PE out there and 1593 * we need have to handle frozen parent PE firstly. 1594 */ 1595 if (ret == EEH_NEXT_ERR_FROZEN_PE) { 1596 parent_pe = (*pe)->parent; 1597 while (parent_pe) { 1598 /* Hit the ceiling ? */ 1599 if (parent_pe->type & EEH_PE_PHB) 1600 break; 1601 1602 /* Frozen parent PE ? */ 1603 state = eeh_ops->get_state(parent_pe, NULL); 1604 if (state > 0 && 1605 (state & active_flags) != active_flags) 1606 *pe = parent_pe; 1607 1608 /* Next parent level */ 1609 parent_pe = parent_pe->parent; 1610 } 1611 1612 /* We possibly migrate to another PE */ 1613 eeh_pe_state_mark(*pe, EEH_PE_ISOLATED); 1614 } 1615 1616 /* 1617 * If we have no errors on the specific PHB or only 1618 * informative error there, we continue poking it. 1619 * Otherwise, we need actions to be taken by upper 1620 * layer. 1621 */ 1622 if (ret > EEH_NEXT_ERR_INF) 1623 break; 1624 } 1625 1626 /* Unmask the event */ 1627 if (ret == EEH_NEXT_ERR_NONE && eeh_enabled()) 1628 enable_irq(eeh_event_irq); 1629 1630 return ret; 1631 } 1632 1633 static int pnv_eeh_restore_vf_config(struct pci_dn *pdn) 1634 { 1635 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 1636 u32 devctl, cmd, cap2, aer_capctl; 1637 int old_mps; 1638 1639 if (edev->pcie_cap) { 1640 /* Restore MPS */ 1641 old_mps = (ffs(pdn->mps) - 8) << 5; 1642 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 1643 2, &devctl); 1644 devctl &= ~PCI_EXP_DEVCTL_PAYLOAD; 1645 devctl |= old_mps; 1646 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 1647 2, devctl); 1648 1649 /* Disable Completion Timeout */ 1650 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP2, 1651 4, &cap2); 1652 if (cap2 & 0x10) { 1653 eeh_ops->read_config(pdn, 1654 edev->pcie_cap + PCI_EXP_DEVCTL2, 1655 4, &cap2); 1656 cap2 |= 0x10; 1657 eeh_ops->write_config(pdn, 1658 edev->pcie_cap + PCI_EXP_DEVCTL2, 1659 4, cap2); 1660 } 1661 } 1662 1663 /* Enable SERR and parity checking */ 1664 eeh_ops->read_config(pdn, PCI_COMMAND, 2, &cmd); 1665 cmd |= (PCI_COMMAND_PARITY | PCI_COMMAND_SERR); 1666 eeh_ops->write_config(pdn, PCI_COMMAND, 2, cmd); 1667 1668 /* Enable report various errors */ 1669 if (edev->pcie_cap) { 1670 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 1671 2, &devctl); 1672 devctl &= ~PCI_EXP_DEVCTL_CERE; 1673 devctl |= (PCI_EXP_DEVCTL_NFERE | 1674 PCI_EXP_DEVCTL_FERE | 1675 PCI_EXP_DEVCTL_URRE); 1676 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL, 1677 2, devctl); 1678 } 1679 1680 /* Enable ECRC generation and check */ 1681 if (edev->pcie_cap && edev->aer_cap) { 1682 eeh_ops->read_config(pdn, edev->aer_cap + PCI_ERR_CAP, 1683 4, &aer_capctl); 1684 aer_capctl |= (PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE); 1685 eeh_ops->write_config(pdn, edev->aer_cap + PCI_ERR_CAP, 1686 4, aer_capctl); 1687 } 1688 1689 return 0; 1690 } 1691 1692 static int pnv_eeh_restore_config(struct pci_dn *pdn) 1693 { 1694 struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 1695 struct pnv_phb *phb; 1696 s64 ret; 1697 1698 if (!edev) 1699 return -EEXIST; 1700 1701 /* 1702 * We have to restore the PCI config space after reset since the 1703 * firmware can't see SRIOV VFs. 1704 * 1705 * FIXME: The MPS, error routing rules, timeout setting are worthy 1706 * to be exported by firmware in extendible way. 1707 */ 1708 if (edev->physfn) { 1709 ret = pnv_eeh_restore_vf_config(pdn); 1710 } else { 1711 phb = edev->phb->private_data; 1712 ret = opal_pci_reinit(phb->opal_id, 1713 OPAL_REINIT_PCI_DEV, edev->config_addr); 1714 } 1715 1716 if (ret) { 1717 pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n", 1718 __func__, edev->config_addr, ret); 1719 return -EIO; 1720 } 1721 1722 return 0; 1723 } 1724 1725 static struct eeh_ops pnv_eeh_ops = { 1726 .name = "powernv", 1727 .init = pnv_eeh_init, 1728 .post_init = pnv_eeh_post_init, 1729 .probe = pnv_eeh_probe, 1730 .set_option = pnv_eeh_set_option, 1731 .get_pe_addr = pnv_eeh_get_pe_addr, 1732 .get_state = pnv_eeh_get_state, 1733 .reset = pnv_eeh_reset, 1734 .wait_state = pnv_eeh_wait_state, 1735 .get_log = pnv_eeh_get_log, 1736 .configure_bridge = pnv_eeh_configure_bridge, 1737 .err_inject = pnv_eeh_err_inject, 1738 .read_config = pnv_eeh_read_config, 1739 .write_config = pnv_eeh_write_config, 1740 .next_error = pnv_eeh_next_error, 1741 .restore_config = pnv_eeh_restore_config 1742 }; 1743 1744 void pcibios_bus_add_device(struct pci_dev *pdev) 1745 { 1746 struct pci_dn *pdn = pci_get_pdn(pdev); 1747 1748 if (!pdev->is_virtfn) 1749 return; 1750 1751 /* 1752 * The following operations will fail if VF's sysfs files 1753 * aren't created or its resources aren't finalized. 1754 */ 1755 eeh_add_device_early(pdn); 1756 eeh_add_device_late(pdev); 1757 eeh_sysfs_add_device(pdev); 1758 } 1759 1760 #ifdef CONFIG_PCI_IOV 1761 static void pnv_pci_fixup_vf_mps(struct pci_dev *pdev) 1762 { 1763 struct pci_dn *pdn = pci_get_pdn(pdev); 1764 int parent_mps; 1765 1766 if (!pdev->is_virtfn) 1767 return; 1768 1769 /* Synchronize MPS for VF and PF */ 1770 parent_mps = pcie_get_mps(pdev->physfn); 1771 if ((128 << pdev->pcie_mpss) >= parent_mps) 1772 pcie_set_mps(pdev, parent_mps); 1773 pdn->mps = pcie_get_mps(pdev); 1774 } 1775 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pnv_pci_fixup_vf_mps); 1776 #endif /* CONFIG_PCI_IOV */ 1777 1778 /** 1779 * eeh_powernv_init - Register platform dependent EEH operations 1780 * 1781 * EEH initialization on powernv platform. This function should be 1782 * called before any EEH related functions. 1783 */ 1784 static int __init eeh_powernv_init(void) 1785 { 1786 int ret = -EINVAL; 1787 1788 eeh_set_pe_aux_size(PNV_PCI_DIAG_BUF_SIZE); 1789 ret = eeh_ops_register(&pnv_eeh_ops); 1790 if (!ret) 1791 pr_info("EEH: PowerNV platform initialized\n"); 1792 else 1793 pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret); 1794 1795 return ret; 1796 } 1797 machine_early_initcall(powernv, eeh_powernv_init); 1798