arch/powerpc/platforms/powernv/eeh-ioda.c

   1 /*
   2  * The file intends to implement the functions needed by EEH, which is
   3  * built on IODA compliant chip. Actually, lots of functions related
   4  * to EEH would be built based on the OPAL APIs.
   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/bootmem.h>
  15 #include <linux/debugfs.h>
  16 #include <linux/delay.h>
  17 #include <linux/io.h>
  18 #include <linux/irq.h>
  19 #include <linux/kernel.h>
  20 #include <linux/msi.h>
  21 #include <linux/notifier.h>
  22 #include <linux/pci.h>
  23 #include <linux/string.h>
  24
  25 #include <asm/eeh.h>
  26 #include <asm/eeh_event.h>
  27 #include <asm/io.h>
  28 #include <asm/iommu.h>
  29 #include <asm/msi_bitmap.h>
  30 #include <asm/opal.h>
  31 #include <asm/pci-bridge.h>
  32 #include <asm/ppc-pci.h>
  33 #include <asm/tce.h>
  34
  35 #include "powernv.h"
  36 #include "pci.h"
  37
  38 static int ioda_eeh_nb_init = 0;
  39
  40 static int ioda_eeh_event(struct notifier_block *nb,
  41                           unsigned long events, void *change)
  42 {
  43         uint64_t changed_evts = (uint64_t)change;
  44
  45         /*
  46          * We simply send special EEH event if EEH has
  47          * been enabled, or clear pending events in
  48          * case that we enable EEH soon
  49          */
  50         if (!(changed_evts & OPAL_EVENT_PCI_ERROR) ||
  51             !(events & OPAL_EVENT_PCI_ERROR))
  52                 return 0;
  53
  54         if (eeh_enabled())
  55                 eeh_send_failure_event(NULL);
  56         else
  57                 opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul);
  58
  59         return 0;
  60 }
  61
  62 static struct notifier_block ioda_eeh_nb = {
  63         .notifier_call  = ioda_eeh_event,
  64         .next           = NULL,
  65         .priority       = 0
  66 };
  67
  68 #ifdef CONFIG_DEBUG_FS
  69 static int ioda_eeh_dbgfs_set(void *data, int offset, u64 val)
  70 {
  71         struct pci_controller *hose = data;
  72         struct pnv_phb *phb = hose->private_data;
  73
  74         out_be64(phb->regs + offset, val);
  75         return 0;
  76 }
  77
  78 static int ioda_eeh_dbgfs_get(void *data, int offset, u64 *val)
  79 {
  80         struct pci_controller *hose = data;
  81         struct pnv_phb *phb = hose->private_data;
  82
  83         *val = in_be64(phb->regs + offset);
  84         return 0;
  85 }
  86
  87 static int ioda_eeh_outb_dbgfs_set(void *data, u64 val)
  88 {
  89         return ioda_eeh_dbgfs_set(data, 0xD10, val);
  90 }
  91
  92 static int ioda_eeh_outb_dbgfs_get(void *data, u64 *val)
  93 {
  94         return ioda_eeh_dbgfs_get(data, 0xD10, val);
  95 }
  96
  97 static int ioda_eeh_inbA_dbgfs_set(void *data, u64 val)
  98 {
  99         return ioda_eeh_dbgfs_set(data, 0xD90, val);
 100 }
 101
 102 static int ioda_eeh_inbA_dbgfs_get(void *data, u64 *val)
 103 {
 104         return ioda_eeh_dbgfs_get(data, 0xD90, val);
 105 }
 106
 107 static int ioda_eeh_inbB_dbgfs_set(void *data, u64 val)
 108 {
 109         return ioda_eeh_dbgfs_set(data, 0xE10, val);
 110 }
 111
 112 static int ioda_eeh_inbB_dbgfs_get(void *data, u64 *val)
 113 {
 114         return ioda_eeh_dbgfs_get(data, 0xE10, val);
 115 }
 116
 117 DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_outb_dbgfs_ops, ioda_eeh_outb_dbgfs_get,
 118                         ioda_eeh_outb_dbgfs_set, "0x%llx\n");
 119 DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_inbA_dbgfs_ops, ioda_eeh_inbA_dbgfs_get,
 120                         ioda_eeh_inbA_dbgfs_set, "0x%llx\n");
 121 DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_inbB_dbgfs_ops, ioda_eeh_inbB_dbgfs_get,
 122                         ioda_eeh_inbB_dbgfs_set, "0x%llx\n");
 123 #endif /* CONFIG_DEBUG_FS */
 124
 125
 126 /**
 127  * ioda_eeh_post_init - Chip dependent post initialization
 128  * @hose: PCI controller
 129  *
 130  * The function will be called after eeh PEs and devices
 131  * have been built. That means the EEH is ready to supply
 132  * service with I/O cache.
 133  */
 134 static int ioda_eeh_post_init(struct pci_controller *hose)
 135 {
 136         struct pnv_phb *phb = hose->private_data;
 137         int ret;
 138
 139         /* Register OPAL event notifier */
 140         if (!ioda_eeh_nb_init) {
 141                 ret = opal_notifier_register(&ioda_eeh_nb);
 142                 if (ret) {
 143                         pr_err("%s: Can't register OPAL event notifier (%d)\n",
 144                                __func__, ret);
 145                         return ret;
 146                 }
 147
 148                 ioda_eeh_nb_init = 1;
 149         }
 150
 151 #ifdef CONFIG_DEBUG_FS
 152         if (!phb->has_dbgfs && phb->dbgfs) {
 153                 phb->has_dbgfs = 1;
 154
 155                 debugfs_create_file("err_injct_outbound", 0600,
 156                                     phb->dbgfs, hose,
 157                                     &ioda_eeh_outb_dbgfs_ops);
 158                 debugfs_create_file("err_injct_inboundA", 0600,
 159                                     phb->dbgfs, hose,
 160                                     &ioda_eeh_inbA_dbgfs_ops);
 161                 debugfs_create_file("err_injct_inboundB", 0600,
 162                                     phb->dbgfs, hose,
 163                                     &ioda_eeh_inbB_dbgfs_ops);
 164         }
 165 #endif
 166
 167         /* If EEH is enabled, we're going to rely on that.
 168          * Otherwise, we restore to conventional mechanism
 169          * to clear frozen PE during PCI config access.
 170          */
 171         if (eeh_enabled())
 172                 phb->flags |= PNV_PHB_FLAG_EEH;
 173         else
 174                 phb->flags &= ~PNV_PHB_FLAG_EEH;
 175
 176         return 0;
 177 }
 178
 179 /**
 180  * ioda_eeh_set_option - Set EEH operation or I/O setting
 181  * @pe: EEH PE
 182  * @option: options
 183  *
 184  * Enable or disable EEH option for the indicated PE. The
 185  * function also can be used to enable I/O or DMA for the
 186  * PE.
 187  */
 188 static int ioda_eeh_set_option(struct eeh_pe *pe, int option)
 189 {
 190         s64 ret;
 191         u32 pe_no;
 192         struct pci_controller *hose = pe->phb;
 193         struct pnv_phb *phb = hose->private_data;
 194
 195         /* Check on PE number */
 196         if (pe->addr < 0 || pe->addr >= phb->ioda.total_pe) {
 197                 pr_err("%s: PE address %x out of range [0, %x] "
 198                        "on PHB#%x\n",
 199                         __func__, pe->addr, phb->ioda.total_pe,
 200                         hose->global_number);
 201                 return -EINVAL;
 202         }
 203
 204         pe_no = pe->addr;
 205         switch (option) {
 206         case EEH_OPT_DISABLE:
 207                 ret = -EEXIST;
 208                 break;
 209         case EEH_OPT_ENABLE:
 210                 ret = 0;
 211                 break;
 212         case EEH_OPT_THAW_MMIO:
 213                 ret = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
 214                                 OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO);
 215                 if (ret) {
 216                         pr_warning("%s: Failed to enable MMIO for "
 217                                    "PHB#%x-PE#%x, err=%lld\n",
 218                                 __func__, hose->global_number, pe_no, ret);
 219                         return -EIO;
 220                 }
 221
 222                 break;
 223         case EEH_OPT_THAW_DMA:
 224                 ret = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
 225                                 OPAL_EEH_ACTION_CLEAR_FREEZE_DMA);
 226                 if (ret) {
 227                         pr_warning("%s: Failed to enable DMA for "
 228                                    "PHB#%x-PE#%x, err=%lld\n",
 229                                 __func__, hose->global_number, pe_no, ret);
 230                         return -EIO;
 231                 }
 232
 233                 break;
 234         default:
 235                 pr_warning("%s: Invalid option %d\n", __func__, option);
 236                 return -EINVAL;
 237         }
 238
 239         return ret;
 240 }
 241
 242 static void ioda_eeh_phb_diag(struct pci_controller *hose)
 243 {
 244         struct pnv_phb *phb = hose->private_data;
 245         long rc;
 246
 247         rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
 248                                          PNV_PCI_DIAG_BUF_SIZE);
 249         if (rc != OPAL_SUCCESS) {
 250                 pr_warning("%s: Failed to get diag-data for PHB#%x (%ld)\n",
 251                             __func__, hose->global_number, rc);
 252                 return;
 253         }
 254
 255         pnv_pci_dump_phb_diag_data(hose, phb->diag.blob);
 256 }
 257
 258 /**
 259  * ioda_eeh_get_state - Retrieve the state of PE
 260  * @pe: EEH PE
 261  *
 262  * The PE's state should be retrieved from the PEEV, PEST
 263  * IODA tables. Since the OPAL has exported the function
 264  * to do it, it'd better to use that.
 265  */
 266 static int ioda_eeh_get_state(struct eeh_pe *pe)
 267 {
 268         s64 ret = 0;
 269         u8 fstate;
 270         u16 pcierr;
 271         u32 pe_no;
 272         int result;
 273         struct pci_controller *hose = pe->phb;
 274         struct pnv_phb *phb = hose->private_data;
 275
 276         /*
 277          * Sanity check on PE address. The PHB PE address should
 278          * be zero.
 279          */
 280         if (pe->addr < 0 || pe->addr >= phb->ioda.total_pe) {
 281                 pr_err("%s: PE address %x out of range [0, %x] "
 282                        "on PHB#%x\n",
 283                        __func__, pe->addr, phb->ioda.total_pe,
 284                        hose->global_number);
 285                 return EEH_STATE_NOT_SUPPORT;
 286         }
 287
 288         /*
 289          * If we're in middle of PE reset, return normal
 290          * state to keep EEH core going. For PHB reset, we
 291          * still expect to have fenced PHB cleared with
 292          * PHB reset.
 293          */
 294         if (!(pe->type & EEH_PE_PHB) &&
 295             (pe->state & EEH_PE_RESET)) {
 296                 result = (EEH_STATE_MMIO_ACTIVE |
 297                           EEH_STATE_DMA_ACTIVE |
 298                           EEH_STATE_MMIO_ENABLED |
 299                           EEH_STATE_DMA_ENABLED);
 300                 return result;
 301         }
 302
 303         /* Retrieve PE status through OPAL */
 304         pe_no = pe->addr;
 305         ret = opal_pci_eeh_freeze_status(phb->opal_id, pe_no,
 306                         &fstate, &pcierr, NULL);
 307         if (ret) {
 308                 pr_err("%s: Failed to get EEH status on "
 309                        "PHB#%x-PE#%x\n, err=%lld\n",
 310                        __func__, hose->global_number, pe_no, ret);
 311                 return EEH_STATE_NOT_SUPPORT;
 312         }
 313
 314         /* Check PHB status */
 315         if (pe->type & EEH_PE_PHB) {
 316                 result = 0;
 317                 result &= ~EEH_STATE_RESET_ACTIVE;
 318
 319                 if (pcierr != OPAL_EEH_PHB_ERROR) {
 320                         result |= EEH_STATE_MMIO_ACTIVE;
 321                         result |= EEH_STATE_DMA_ACTIVE;
 322                         result |= EEH_STATE_MMIO_ENABLED;
 323                         result |= EEH_STATE_DMA_ENABLED;
 324                 } else if (!(pe->state & EEH_PE_ISOLATED)) {
 325                         eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
 326                         ioda_eeh_phb_diag(hose);
 327                 }
 328
 329                 return result;
 330         }
 331
 332         /* Parse result out */
 333         result = 0;
 334         switch (fstate) {
 335         case OPAL_EEH_STOPPED_NOT_FROZEN:
 336                 result &= ~EEH_STATE_RESET_ACTIVE;
 337                 result |= EEH_STATE_MMIO_ACTIVE;
 338                 result |= EEH_STATE_DMA_ACTIVE;
 339                 result |= EEH_STATE_MMIO_ENABLED;
 340                 result |= EEH_STATE_DMA_ENABLED;
 341                 break;
 342         case OPAL_EEH_STOPPED_MMIO_FREEZE:
 343                 result &= ~EEH_STATE_RESET_ACTIVE;
 344                 result |= EEH_STATE_DMA_ACTIVE;
 345                 result |= EEH_STATE_DMA_ENABLED;
 346                 break;
 347         case OPAL_EEH_STOPPED_DMA_FREEZE:
 348                 result &= ~EEH_STATE_RESET_ACTIVE;
 349                 result |= EEH_STATE_MMIO_ACTIVE;
 350                 result |= EEH_STATE_MMIO_ENABLED;
 351                 break;
 352         case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
 353                 result &= ~EEH_STATE_RESET_ACTIVE;
 354                 break;
 355         case OPAL_EEH_STOPPED_RESET:
 356                 result |= EEH_STATE_RESET_ACTIVE;
 357                 break;
 358         case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
 359                 result |= EEH_STATE_UNAVAILABLE;
 360                 break;
 361         case OPAL_EEH_STOPPED_PERM_UNAVAIL:
 362                 result |= EEH_STATE_NOT_SUPPORT;
 363                 break;
 364         default:
 365                 pr_warning("%s: Unexpected EEH status 0x%x "
 366                            "on PHB#%x-PE#%x\n",
 367                            __func__, fstate, hose->global_number, pe_no);
 368         }
 369
 370         /* Dump PHB diag-data for frozen PE */
 371         if (result != EEH_STATE_NOT_SUPPORT &&
 372             (result & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) !=
 373             (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE) &&
 374             !(pe->state & EEH_PE_ISOLATED)) {
 375                 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
 376                 ioda_eeh_phb_diag(hose);
 377         }
 378
 379         return result;
 380 }
 381
 382 static s64 ioda_eeh_phb_poll(struct pnv_phb *phb)
 383 {
 384         s64 rc = OPAL_HARDWARE;
 385
 386         while (1) {
 387                 rc = opal_pci_poll(phb->opal_id);
 388                 if (rc <= 0)
 389                         break;
 390
 391                 msleep(rc);
 392         }
 393
 394         return rc;
 395 }
 396
 397 static int ioda_eeh_phb_reset(struct pci_controller *hose, int option)
 398 {
 399         struct pnv_phb *phb = hose->private_data;
 400         s64 rc = OPAL_HARDWARE;
 401
 402         pr_debug("%s: Reset PHB#%x, option=%d\n",
 403                  __func__, hose->global_number, option);
 404
 405         /* Issue PHB complete reset request */
 406         if (option == EEH_RESET_FUNDAMENTAL ||
 407             option == EEH_RESET_HOT)
 408                 rc = opal_pci_reset(phb->opal_id,
 409                                 OPAL_PHB_COMPLETE,
 410                                 OPAL_ASSERT_RESET);
 411         else if (option == EEH_RESET_DEACTIVATE)
 412                 rc = opal_pci_reset(phb->opal_id,
 413                                 OPAL_PHB_COMPLETE,
 414                                 OPAL_DEASSERT_RESET);
 415         if (rc < 0)
 416                 goto out;
 417
 418         /*
 419          * Poll state of the PHB until the request is done
 420          * successfully.
 421          */
 422         rc = ioda_eeh_phb_poll(phb);
 423 out:
 424         if (rc != OPAL_SUCCESS)
 425                 return -EIO;
 426
 427         return 0;
 428 }
 429
 430 static int ioda_eeh_root_reset(struct pci_controller *hose, int option)
 431 {
 432         struct pnv_phb *phb = hose->private_data;
 433         s64 rc = OPAL_SUCCESS;
 434
 435         pr_debug("%s: Reset PHB#%x, option=%d\n",
 436                  __func__, hose->global_number, option);
 437
 438         /*
 439          * During the reset deassert time, we needn't care
 440          * the reset scope because the firmware does nothing
 441          * for fundamental or hot reset during deassert phase.
 442          */
 443         if (option == EEH_RESET_FUNDAMENTAL)
 444                 rc = opal_pci_reset(phb->opal_id,
 445                                 OPAL_PCI_FUNDAMENTAL_RESET,
 446                                 OPAL_ASSERT_RESET);
 447         else if (option == EEH_RESET_HOT)
 448                 rc = opal_pci_reset(phb->opal_id,
 449                                 OPAL_PCI_HOT_RESET,
 450                                 OPAL_ASSERT_RESET);
 451         else if (option == EEH_RESET_DEACTIVATE)
 452                 rc = opal_pci_reset(phb->opal_id,
 453                                 OPAL_PCI_HOT_RESET,
 454                                 OPAL_DEASSERT_RESET);
 455         if (rc < 0)
 456                 goto out;
 457
 458         /* Poll state of the PHB until the request is done */
 459         rc = ioda_eeh_phb_poll(phb);
 460 out:
 461         if (rc != OPAL_SUCCESS)
 462                 return -EIO;
 463
 464         return 0;
 465 }
 466
 467 static int ioda_eeh_bridge_reset(struct pci_dev *dev, int option)
 468
 469 {
 470         struct device_node *dn = pci_device_to_OF_node(dev);
 471         u32 ctrl;
 472
 473         pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
 474                  __func__, pci_domain_nr(dev->bus),
 475                  dev->bus->number, option);
 476
 477         switch (option) {
 478         case EEH_RESET_FUNDAMENTAL:
 479         case EEH_RESET_HOT:
 480                 eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &ctrl);
 481                 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
 482                 eeh_ops->write_config(dn, PCI_BRIDGE_CONTROL, 2, ctrl);
 483                 break;
 484         case EEH_RESET_DEACTIVATE:
 485                 eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &ctrl);
 486                 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
 487                 eeh_ops->write_config(dn, PCI_BRIDGE_CONTROL, 2, ctrl);
 488                 break;
 489         }
 490
 491         return 0;
 492 }
 493
 494 /**
 495  * ioda_eeh_reset - Reset the indicated PE
 496  * @pe: EEH PE
 497  * @option: reset option
 498  *
 499  * Do reset on the indicated PE. For PCI bus sensitive PE,
 500  * we need to reset the parent p2p bridge. The PHB has to
 501  * be reinitialized if the p2p bridge is root bridge. For
 502  * PCI device sensitive PE, we will try to reset the device
 503  * through FLR. For now, we don't have OPAL APIs to do HARD
 504  * reset yet, so all reset would be SOFT (HOT) reset.
 505  */
 506 static int ioda_eeh_reset(struct eeh_pe *pe, int option)
 507 {
 508         struct pci_controller *hose = pe->phb;
 509         struct pci_bus *bus;
 510         int ret;
 511
 512         /*
 513          * The rules applied to reset, either fundamental or hot reset:
 514          *
 515          * We always reset the direct upstream bridge of the PE. If the
 516          * direct upstream bridge isn't root bridge, we always take hot
 517          * reset no matter what option (fundamental or hot) is. Otherwise,
 518          * we should do the reset according to the required option.
 519          *
 520          * Here, we have different design to pHyp, which always clear the
 521          * frozen state during PE reset. However, the good idea here from
 522          * benh is to keep frozen state before we get PE reset done completely
 523          * (until BAR restore). With the frozen state, HW drops illegal IO
 524          * or MMIO access, which can incur recrusive frozen PE during PE
 525          * reset. The side effect is that EEH core has to clear the frozen
 526          * state explicitly after BAR restore.
 527          */
 528         if (pe->type & EEH_PE_PHB) {
 529                 ret = ioda_eeh_phb_reset(hose, option);
 530         } else {
 531                 bus = eeh_pe_bus_get(pe);
 532                 if (pci_is_root_bus(bus))
 533                         ret = ioda_eeh_root_reset(hose, option);
 534                 else
 535                         ret = ioda_eeh_bridge_reset(bus->self, option);
 536         }
 537
 538         return ret;
 539 }
 540
 541 /**
 542  * ioda_eeh_configure_bridge - Configure the PCI bridges for the indicated PE
 543  * @pe: EEH PE
 544  *
 545  * For particular PE, it might have included PCI bridges. In order
 546  * to make the PE work properly, those PCI bridges should be configured
 547  * correctly. However, we need do nothing on P7IOC since the reset
 548  * function will do everything that should be covered by the function.
 549  */
 550 static int ioda_eeh_configure_bridge(struct eeh_pe *pe)
 551 {
 552         return 0;
 553 }
 554
 555 static void ioda_eeh_hub_diag_common(struct OpalIoP7IOCErrorData *data)
 556 {
 557         /* GEM */
 558         pr_info("  GEM XFIR:        %016llx\n", data->gemXfir);
 559         pr_info("  GEM RFIR:        %016llx\n", data->gemRfir);
 560         pr_info("  GEM RIRQFIR:     %016llx\n", data->gemRirqfir);
 561         pr_info("  GEM Mask:        %016llx\n", data->gemMask);
 562         pr_info("  GEM RWOF:        %016llx\n", data->gemRwof);
 563
 564         /* LEM */
 565         pr_info("  LEM FIR:         %016llx\n", data->lemFir);
 566         pr_info("  LEM Error Mask:  %016llx\n", data->lemErrMask);
 567         pr_info("  LEM Action 0:    %016llx\n", data->lemAction0);
 568         pr_info("  LEM Action 1:    %016llx\n", data->lemAction1);
 569         pr_info("  LEM WOF:         %016llx\n", data->lemWof);
 570 }
 571
 572 static void ioda_eeh_hub_diag(struct pci_controller *hose)
 573 {
 574         struct pnv_phb *phb = hose->private_data;
 575         struct OpalIoP7IOCErrorData *data = &phb->diag.hub_diag;
 576         long rc;
 577
 578         rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
 579         if (rc != OPAL_SUCCESS) {
 580                 pr_warning("%s: Failed to get HUB#%llx diag-data (%ld)\n",
 581                            __func__, phb->hub_id, rc);
 582                 return;
 583         }
 584
 585         switch (data->type) {
 586         case OPAL_P7IOC_DIAG_TYPE_RGC:
 587                 pr_info("P7IOC diag-data for RGC\n\n");
 588                 ioda_eeh_hub_diag_common(data);
 589                 pr_info("  RGC Status:      %016llx\n", data->rgc.rgcStatus);
 590                 pr_info("  RGC LDCP:        %016llx\n", data->rgc.rgcLdcp);
 591                 break;
 592         case OPAL_P7IOC_DIAG_TYPE_BI:
 593                 pr_info("P7IOC diag-data for BI %s\n\n",
 594                         data->bi.biDownbound ? "Downbound" : "Upbound");
 595                 ioda_eeh_hub_diag_common(data);
 596                 pr_info("  BI LDCP 0:       %016llx\n", data->bi.biLdcp0);
 597                 pr_info("  BI LDCP 1:       %016llx\n", data->bi.biLdcp1);
 598                 pr_info("  BI LDCP 2:       %016llx\n", data->bi.biLdcp2);
 599                 pr_info("  BI Fence Status: %016llx\n", data->bi.biFenceStatus);
 600                 break;
 601         case OPAL_P7IOC_DIAG_TYPE_CI:
 602                 pr_info("P7IOC diag-data for CI Port %d\\nn",
 603                         data->ci.ciPort);
 604                 ioda_eeh_hub_diag_common(data);
 605                 pr_info("  CI Port Status:  %016llx\n", data->ci.ciPortStatus);
 606                 pr_info("  CI Port LDCP:    %016llx\n", data->ci.ciPortLdcp);
 607                 break;
 608         case OPAL_P7IOC_DIAG_TYPE_MISC:
 609                 pr_info("P7IOC diag-data for MISC\n\n");
 610                 ioda_eeh_hub_diag_common(data);
 611                 break;
 612         case OPAL_P7IOC_DIAG_TYPE_I2C:
 613                 pr_info("P7IOC diag-data for I2C\n\n");
 614                 ioda_eeh_hub_diag_common(data);
 615                 break;
 616         default:
 617                 pr_warning("%s: Invalid type of HUB#%llx diag-data (%d)\n",
 618                            __func__, phb->hub_id, data->type);
 619         }
 620 }
 621
 622 static int ioda_eeh_get_pe(struct pci_controller *hose,
 623                            u16 pe_no, struct eeh_pe **pe)
 624 {
 625         struct eeh_pe *phb_pe, *dev_pe;
 626         struct eeh_dev dev;
 627
 628         /* Find the PHB PE */
 629         phb_pe = eeh_phb_pe_get(hose);
 630         if (!phb_pe)
 631                 return -EEXIST;
 632
 633         /* Find the PE according to PE# */
 634         memset(&dev, 0, sizeof(struct eeh_dev));
 635         dev.phb = hose;
 636         dev.pe_config_addr = pe_no;
 637         dev_pe = eeh_pe_get(&dev);
 638         if (!dev_pe) return -EEXIST;
 639
 640         *pe = dev_pe;
 641         return 0;
 642 }
 643
 644 /**
 645  * ioda_eeh_next_error - Retrieve next error for EEH core to handle
 646  * @pe: The affected PE
 647  *
 648  * The function is expected to be called by EEH core while it gets
 649  * special EEH event (without binding PE). The function calls to
 650  * OPAL APIs for next error to handle. The informational error is
 651  * handled internally by platform. However, the dead IOC, dead PHB,
 652  * fenced PHB and frozen PE should be handled by EEH core eventually.
 653  */
 654 static int ioda_eeh_next_error(struct eeh_pe **pe)
 655 {
 656         struct pci_controller *hose;
 657         struct pnv_phb *phb;
 658         struct eeh_pe *phb_pe;
 659         u64 frozen_pe_no;
 660         u16 err_type, severity;
 661         long rc;
 662         int ret = EEH_NEXT_ERR_NONE;
 663
 664         /*
 665          * While running here, it's safe to purge the event queue.
 666          * And we should keep the cached OPAL notifier event sychronized
 667          * between the kernel and firmware.
 668          */
 669         eeh_remove_event(NULL);
 670         opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul);
 671
 672         list_for_each_entry(hose, &hose_list, list_node) {
 673                 /*
 674                  * If the subordinate PCI buses of the PHB has been
 675                  * removed or is exactly under error recovery, we
 676                  * needn't take care of it any more.
 677                  */
 678                 phb = hose->private_data;
 679                 phb_pe = eeh_phb_pe_get(hose);
 680                 if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
 681                         continue;
 682
 683                 rc = opal_pci_next_error(phb->opal_id,
 684                                 &frozen_pe_no, &err_type, &severity);
 685
 686                 /* If OPAL API returns error, we needn't proceed */
 687                 if (rc != OPAL_SUCCESS) {
 688                         pr_devel("%s: Invalid return value on "
 689                                  "PHB#%x (0x%lx) from opal_pci_next_error",
 690                                  __func__, hose->global_number, rc);
 691                         continue;
 692                 }
 693
 694                 /* If the PHB doesn't have error, stop processing */
 695                 if (err_type == OPAL_EEH_NO_ERROR ||
 696                     severity == OPAL_EEH_SEV_NO_ERROR) {
 697                         pr_devel("%s: No error found on PHB#%x\n",
 698                                  __func__, hose->global_number);
 699                         continue;
 700                 }
 701
 702                 /*
 703                  * Processing the error. We're expecting the error with
 704                  * highest priority reported upon multiple errors on the
 705                  * specific PHB.
 706                  */
 707                 pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
 708                          __func__, err_type, severity,
 709                          frozen_pe_no, hose->global_number);
 710                 switch (err_type) {
 711                 case OPAL_EEH_IOC_ERROR:
 712                         if (severity == OPAL_EEH_SEV_IOC_DEAD) {
 713                                 pr_err("EEH: dead IOC detected\n");
 714                                 ret = EEH_NEXT_ERR_DEAD_IOC;
 715                         } else if (severity == OPAL_EEH_SEV_INF) {
 716                                 pr_info("EEH: IOC informative error "
 717                                         "detected\n");
 718                                 ioda_eeh_hub_diag(hose);
 719                                 ret = EEH_NEXT_ERR_NONE;
 720                         }
 721
 722                         break;
 723                 case OPAL_EEH_PHB_ERROR:
 724                         if (severity == OPAL_EEH_SEV_PHB_DEAD) {
 725                                 *pe = phb_pe;
 726                                 pr_err("EEH: dead PHB#%x detected\n",
 727                                         hose->global_number);
 728                                 ret = EEH_NEXT_ERR_DEAD_PHB;
 729                         } else if (severity == OPAL_EEH_SEV_PHB_FENCED) {
 730                                 *pe = phb_pe;
 731                                 pr_err("EEH: fenced PHB#%x detected\n",
 732                                         hose->global_number);
 733                                 ret = EEH_NEXT_ERR_FENCED_PHB;
 734                         } else if (severity == OPAL_EEH_SEV_INF) {
 735                                 pr_info("EEH: PHB#%x informative error "
 736                                         "detected\n",
 737                                         hose->global_number);
 738                                 ioda_eeh_phb_diag(hose);
 739                                 ret = EEH_NEXT_ERR_NONE;
 740                         }
 741
 742                         break;
 743                 case OPAL_EEH_PE_ERROR:
 744                         /*
 745                          * If we can't find the corresponding PE, the
 746                          * PEEV / PEST would be messy. So we force an
 747                          * fenced PHB so that it can be recovered.
 748                          */
 749                         if (ioda_eeh_get_pe(hose, frozen_pe_no, pe)) {
 750                                 *pe = phb_pe;
 751                                 pr_err("EEH: Escalated fenced PHB#%x "
 752                                        "detected for PE#%llx\n",
 753                                         hose->global_number,
 754                                         frozen_pe_no);
 755                                 ret = EEH_NEXT_ERR_FENCED_PHB;
 756                         } else {
 757                                 pr_err("EEH: Frozen PE#%x on PHB#%x detected\n",
 758                                         (*pe)->addr, (*pe)->phb->global_number);
 759                                 ret = EEH_NEXT_ERR_FROZEN_PE;
 760                         }
 761
 762                         break;
 763                 default:
 764                         pr_warn("%s: Unexpected error type %d\n",
 765                                 __func__, err_type);
 766                 }
 767
 768                 /*
 769                  * EEH core will try recover from fenced PHB or
 770                  * frozen PE. In the time for frozen PE, EEH core
 771                  * enable IO path for that before collecting logs,
 772                  * but it ruins the site. So we have to dump the
 773                  * log in advance here.
 774                  */
 775                 if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
 776                     ret == EEH_NEXT_ERR_FENCED_PHB) &&
 777                     !((*pe)->state & EEH_PE_ISOLATED)) {
 778                         eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
 779                         ioda_eeh_phb_diag(hose);
 780                 }
 781
 782                 /*
 783                  * If we have no errors on the specific PHB or only
 784                  * informative error there, we continue poking it.
 785                  * Otherwise, we need actions to be taken by upper
 786                  * layer.
 787                  */
 788                 if (ret > EEH_NEXT_ERR_INF)
 789                         break;
 790         }
 791
 792         return ret;
 793 }
 794
 795 struct pnv_eeh_ops ioda_eeh_ops = {
 796         .post_init              = ioda_eeh_post_init,
 797         .set_option             = ioda_eeh_set_option,
 798         .get_state              = ioda_eeh_get_state,
 799         .reset                  = ioda_eeh_reset,
 800         .configure_bridge       = ioda_eeh_configure_bridge,
 801         .next_error             = ioda_eeh_next_error
 802 };