arch/powerpc/kernel/eeh_pe.c

   1 /*
   2  * The file intends to implement PE based on the information from
   3  * platforms. Basically, there have 3 types of PEs: PHB/Bus/Device.
   4  * All the PEs should be organized as hierarchy tree. The first level
   5  * of the tree will be associated to existing PHBs since the particular
   6  * PE is only meaningful in one PHB domain.
   7  *
   8  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2012.
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  23  */
  24
  25 #include <linux/export.h>
  26 #include <linux/gfp.h>
  27 #include <linux/init.h>
  28 #include <linux/kernel.h>
  29 #include <linux/pci.h>
  30 #include <linux/string.h>
  31
  32 #include <asm/pci-bridge.h>
  33 #include <asm/ppc-pci.h>
  34
  35 static LIST_HEAD(eeh_phb_pe);
  36
  37 /**
  38  * eeh_pe_alloc - Allocate PE
  39  * @phb: PCI controller
  40  * @type: PE type
  41  *
  42  * Allocate PE instance dynamically.
  43  */
  44 static struct eeh_pe *eeh_pe_alloc(struct pci_controller *phb, int type)
  45 {
  46         struct eeh_pe *pe;
  47
  48         /* Allocate PHB PE */
  49         pe = kzalloc(sizeof(struct eeh_pe), GFP_KERNEL);
  50         if (!pe) return NULL;
  51
  52         /* Initialize PHB PE */
  53         pe->type = type;
  54         pe->phb = phb;
  55         INIT_LIST_HEAD(&pe->child_list);
  56         INIT_LIST_HEAD(&pe->child);
  57         INIT_LIST_HEAD(&pe->edevs);
  58
  59         return pe;
  60 }
  61
  62 /**
  63  * eeh_phb_pe_create - Create PHB PE
  64  * @phb: PCI controller
  65  *
  66  * The function should be called while the PHB is detected during
  67  * system boot or PCI hotplug in order to create PHB PE.
  68  */
  69 int eeh_phb_pe_create(struct pci_controller *phb)
  70 {
  71         struct eeh_pe *pe;
  72
  73         /* Allocate PHB PE */
  74         pe = eeh_pe_alloc(phb, EEH_PE_PHB);
  75         if (!pe) {
  76                 pr_err("%s: out of memory!\n", __func__);
  77                 return -ENOMEM;
  78         }
  79
  80         /* Put it into the list */
  81         list_add_tail(&pe->child, &eeh_phb_pe);
  82
  83         pr_debug("EEH: Add PE for PHB#%d\n", phb->global_number);
  84
  85         return 0;
  86 }
  87
  88 /**
  89  * eeh_phb_pe_get - Retrieve PHB PE based on the given PHB
  90  * @phb: PCI controller
  91  *
  92  * The overall PEs form hierarchy tree. The first layer of the
  93  * hierarchy tree is composed of PHB PEs. The function is used
  94  * to retrieve the corresponding PHB PE according to the given PHB.
  95  */
  96 struct eeh_pe *eeh_phb_pe_get(struct pci_controller *phb)
  97 {
  98         struct eeh_pe *pe;
  99
 100         list_for_each_entry(pe, &eeh_phb_pe, child) {
 101                 /*
 102                  * Actually, we needn't check the type since
 103                  * the PE for PHB has been determined when that
 104                  * was created.
 105                  */
 106                 if ((pe->type & EEH_PE_PHB) && pe->phb == phb)
 107                         return pe;
 108         }
 109
 110         return NULL;
 111 }
 112
 113 /**
 114  * eeh_pe_next - Retrieve the next PE in the tree
 115  * @pe: current PE
 116  * @root: root PE
 117  *
 118  * The function is used to retrieve the next PE in the
 119  * hierarchy PE tree.
 120  */
 121 static struct eeh_pe *eeh_pe_next(struct eeh_pe *pe,
 122                                   struct eeh_pe *root)
 123 {
 124         struct list_head *next = pe->child_list.next;
 125
 126         if (next == &pe->child_list) {
 127                 while (1) {
 128                         if (pe == root)
 129                                 return NULL;
 130                         next = pe->child.next;
 131                         if (next != &pe->parent->child_list)
 132                                 break;
 133                         pe = pe->parent;
 134                 }
 135         }
 136
 137         return list_entry(next, struct eeh_pe, child);
 138 }
 139
 140 /**
 141  * eeh_pe_traverse - Traverse PEs in the specified PHB
 142  * @root: root PE
 143  * @fn: callback
 144  * @flag: extra parameter to callback
 145  *
 146  * The function is used to traverse the specified PE and its
 147  * child PEs. The traversing is to be terminated once the
 148  * callback returns something other than NULL, or no more PEs
 149  * to be traversed.
 150  */
 151 static void *eeh_pe_traverse(struct eeh_pe *root,
 152                         eeh_traverse_func fn, void *flag)
 153 {
 154         struct eeh_pe *pe;
 155         void *ret;
 156
 157         for (pe = root; pe; pe = eeh_pe_next(pe, root)) {
 158                 ret = fn(pe, flag);
 159                 if (ret) return ret;
 160         }
 161
 162         return NULL;
 163 }
 164
 165 /**
 166  * eeh_pe_dev_traverse - Traverse the devices from the PE
 167  * @root: EEH PE
 168  * @fn: function callback
 169  * @flag: extra parameter to callback
 170  *
 171  * The function is used to traverse the devices of the specified
 172  * PE and its child PEs.
 173  */
 174 void *eeh_pe_dev_traverse(struct eeh_pe *root,
 175                 eeh_traverse_func fn, void *flag)
 176 {
 177         struct eeh_pe *pe;
 178         struct eeh_dev *edev;
 179         void *ret;
 180
 181         if (!root) {
 182                 pr_warning("%s: Invalid PE %p\n", __func__, root);
 183                 return NULL;
 184         }
 185
 186         /* Traverse root PE */
 187         for (pe = root; pe; pe = eeh_pe_next(pe, root)) {
 188                 eeh_pe_for_each_dev(pe, edev) {
 189                         ret = fn(edev, flag);
 190                         if (ret)
 191                                 return ret;
 192                 }
 193         }
 194
 195         return NULL;
 196 }
 197
 198 /**
 199  * __eeh_pe_get - Check the PE address
 200  * @data: EEH PE
 201  * @flag: EEH device
 202  *
 203  * For one particular PE, it can be identified by PE address
 204  * or tranditional BDF address. BDF address is composed of
 205  * Bus/Device/Function number. The extra data referred by flag
 206  * indicates which type of address should be used.
 207  */
 208 static void *__eeh_pe_get(void *data, void *flag)
 209 {
 210         struct eeh_pe *pe = (struct eeh_pe *)data;
 211         struct eeh_dev *edev = (struct eeh_dev *)flag;
 212
 213         /* Unexpected PHB PE */
 214         if (pe->type & EEH_PE_PHB)
 215                 return NULL;
 216
 217         /* We prefer PE address */
 218         if (edev->pe_config_addr &&
 219            (edev->pe_config_addr == pe->addr))
 220                 return pe;
 221
 222         /* Try BDF address */
 223         if (edev->config_addr &&
 224            (edev->config_addr == pe->config_addr))
 225                 return pe;
 226
 227         return NULL;
 228 }
 229
 230 /**
 231  * eeh_pe_get - Search PE based on the given address
 232  * @edev: EEH device
 233  *
 234  * Search the corresponding PE based on the specified address which
 235  * is included in the eeh device. The function is used to check if
 236  * the associated PE has been created against the PE address. It's
 237  * notable that the PE address has 2 format: traditional PE address
 238  * which is composed of PCI bus/device/function number, or unified
 239  * PE address.
 240  */
 241 struct eeh_pe *eeh_pe_get(struct eeh_dev *edev)
 242 {
 243         struct eeh_pe *root = eeh_phb_pe_get(edev->phb);
 244         struct eeh_pe *pe;
 245
 246         pe = eeh_pe_traverse(root, __eeh_pe_get, edev);
 247
 248         return pe;
 249 }
 250
 251 /**
 252  * eeh_pe_get_parent - Retrieve the parent PE
 253  * @edev: EEH device
 254  *
 255  * The whole PEs existing in the system are organized as hierarchy
 256  * tree. The function is used to retrieve the parent PE according
 257  * to the parent EEH device.
 258  */
 259 static struct eeh_pe *eeh_pe_get_parent(struct eeh_dev *edev)
 260 {
 261         struct device_node *dn;
 262         struct eeh_dev *parent;
 263
 264         /*
 265          * It might have the case for the indirect parent
 266          * EEH device already having associated PE, but
 267          * the direct parent EEH device doesn't have yet.
 268          */
 269         dn = edev->dn->parent;
 270         while (dn) {
 271                 /* We're poking out of PCI territory */
 272                 if (!PCI_DN(dn)) return NULL;
 273
 274                 parent = of_node_to_eeh_dev(dn);
 275                 /* We're poking out of PCI territory */
 276                 if (!parent) return NULL;
 277
 278                 if (parent->pe)
 279                         return parent->pe;
 280
 281                 dn = dn->parent;
 282         }
 283
 284         return NULL;
 285 }
 286
 287 /**
 288  * eeh_add_to_parent_pe - Add EEH device to parent PE
 289  * @edev: EEH device
 290  *
 291  * Add EEH device to the parent PE. If the parent PE already
 292  * exists, the PE type will be changed to EEH_PE_BUS. Otherwise,
 293  * we have to create new PE to hold the EEH device and the new
 294  * PE will be linked to its parent PE as well.
 295  */
 296 int eeh_add_to_parent_pe(struct eeh_dev *edev)
 297 {
 298         struct eeh_pe *pe, *parent;
 299
 300         /*
 301          * Search the PE has been existing or not according
 302          * to the PE address. If that has been existing, the
 303          * PE should be composed of PCI bus and its subordinate
 304          * components.
 305          */
 306         pe = eeh_pe_get(edev);
 307         if (pe && !(pe->type & EEH_PE_INVALID)) {
 308                 if (!edev->pe_config_addr) {
 309                         pr_err("%s: PE with addr 0x%x already exists\n",
 310                                 __func__, edev->config_addr);
 311                         return -EEXIST;
 312                 }
 313
 314                 /* Mark the PE as type of PCI bus */
 315                 pe->type = EEH_PE_BUS;
 316                 edev->pe = pe;
 317
 318                 /* Put the edev to PE */
 319                 list_add_tail(&edev->list, &pe->edevs);
 320                 pr_debug("EEH: Add %s to Bus PE#%x\n",
 321                         edev->dn->full_name, pe->addr);
 322
 323                 return 0;
 324         } else if (pe && (pe->type & EEH_PE_INVALID)) {
 325                 list_add_tail(&edev->list, &pe->edevs);
 326                 edev->pe = pe;
 327                 /*
 328                  * We're running to here because of PCI hotplug caused by
 329                  * EEH recovery. We need clear EEH_PE_INVALID until the top.
 330                  */
 331                 parent = pe;
 332                 while (parent) {
 333                         if (!(parent->type & EEH_PE_INVALID))
 334                                 break;
 335                         parent->type &= ~EEH_PE_INVALID;
 336                         parent = parent->parent;
 337                 }
 338                 pr_debug("EEH: Add %s to Device PE#%x, Parent PE#%x\n",
 339                         edev->dn->full_name, pe->addr, pe->parent->addr);
 340
 341                 return 0;
 342         }
 343
 344         /* Create a new EEH PE */
 345         pe = eeh_pe_alloc(edev->phb, EEH_PE_DEVICE);
 346         if (!pe) {
 347                 pr_err("%s: out of memory!\n", __func__);
 348                 return -ENOMEM;
 349         }
 350         pe->addr        = edev->pe_config_addr;
 351         pe->config_addr = edev->config_addr;
 352
 353         /*
 354          * While doing PE reset, we probably hot-reset the
 355          * upstream bridge. However, the PCI devices including
 356          * the associated EEH devices might be removed when EEH
 357          * core is doing recovery. So that won't safe to retrieve
 358          * the bridge through downstream EEH device. We have to
 359          * trace the parent PCI bus, then the upstream bridge.
 360          */
 361         if (eeh_probe_mode_dev())
 362                 pe->bus = eeh_dev_to_pci_dev(edev)->bus;
 363
 364         /*
 365          * Put the new EEH PE into hierarchy tree. If the parent
 366          * can't be found, the newly created PE will be attached
 367          * to PHB directly. Otherwise, we have to associate the
 368          * PE with its parent.
 369          */
 370         parent = eeh_pe_get_parent(edev);
 371         if (!parent) {
 372                 parent = eeh_phb_pe_get(edev->phb);
 373                 if (!parent) {
 374                         pr_err("%s: No PHB PE is found (PHB Domain=%d)\n",
 375                                 __func__, edev->phb->global_number);
 376                         edev->pe = NULL;
 377                         kfree(pe);
 378                         return -EEXIST;
 379                 }
 380         }
 381         pe->parent = parent;
 382
 383         /*
 384          * Put the newly created PE into the child list and
 385          * link the EEH device accordingly.
 386          */
 387         list_add_tail(&pe->child, &parent->child_list);
 388         list_add_tail(&edev->list, &pe->edevs);
 389         edev->pe = pe;
 390         pr_debug("EEH: Add %s to Device PE#%x, Parent PE#%x\n",
 391                 edev->dn->full_name, pe->addr, pe->parent->addr);
 392
 393         return 0;
 394 }
 395
 396 /**
 397  * eeh_rmv_from_parent_pe - Remove one EEH device from the associated PE
 398  * @edev: EEH device
 399  * @purge_pe: remove PE or not
 400  *
 401  * The PE hierarchy tree might be changed when doing PCI hotplug.
 402  * Also, the PCI devices or buses could be removed from the system
 403  * during EEH recovery. So we have to call the function remove the
 404  * corresponding PE accordingly if necessary.
 405  */
 406 int eeh_rmv_from_parent_pe(struct eeh_dev *edev, int purge_pe)
 407 {
 408         struct eeh_pe *pe, *parent, *child;
 409         int cnt;
 410
 411         if (!edev->pe) {
 412                 pr_warning("%s: No PE found for EEH device %s\n",
 413                         __func__, edev->dn->full_name);
 414                 return -EEXIST;
 415         }
 416
 417         /* Remove the EEH device */
 418         pe = edev->pe;
 419         edev->pe = NULL;
 420         list_del(&edev->list);
 421
 422         /*
 423          * Check if the parent PE includes any EEH devices.
 424          * If not, we should delete that. Also, we should
 425          * delete the parent PE if it doesn't have associated
 426          * child PEs and EEH devices.
 427          */
 428         while (1) {
 429                 parent = pe->parent;
 430                 if (pe->type & EEH_PE_PHB)
 431                         break;
 432
 433                 if (purge_pe) {
 434                         if (list_empty(&pe->edevs) &&
 435                             list_empty(&pe->child_list)) {
 436                                 list_del(&pe->child);
 437                                 kfree(pe);
 438                         } else {
 439                                 break;
 440                         }
 441                 } else {
 442                         if (list_empty(&pe->edevs)) {
 443                                 cnt = 0;
 444                                 list_for_each_entry(child, &pe->child_list, child) {
 445                                         if (!(child->type & EEH_PE_INVALID)) {
 446                                                 cnt++;
 447                                                 break;
 448                                         }
 449                                 }
 450
 451                                 if (!cnt)
 452                                         pe->type |= EEH_PE_INVALID;
 453                                 else
 454                                         break;
 455                         }
 456                 }
 457
 458                 pe = parent;
 459         }
 460
 461         return 0;
 462 }
 463
 464 /**
 465  * eeh_pe_update_time_stamp - Update PE's frozen time stamp
 466  * @pe: EEH PE
 467  *
 468  * We have time stamp for each PE to trace its time of getting
 469  * frozen in last hour. The function should be called to update
 470  * the time stamp on first error of the specific PE. On the other
 471  * handle, we needn't account for errors happened in last hour.
 472  */
 473 void eeh_pe_update_time_stamp(struct eeh_pe *pe)
 474 {
 475         struct timeval tstamp;
 476
 477         if (!pe) return;
 478
 479         if (pe->freeze_count <= 0) {
 480                 pe->freeze_count = 0;
 481                 do_gettimeofday(&pe->tstamp);
 482         } else {
 483                 do_gettimeofday(&tstamp);
 484                 if (tstamp.tv_sec - pe->tstamp.tv_sec > 3600) {
 485                         pe->tstamp = tstamp;
 486                         pe->freeze_count = 0;
 487                 }
 488         }
 489 }
 490
 491 /**
 492  * __eeh_pe_state_mark - Mark the state for the PE
 493  * @data: EEH PE
 494  * @flag: state
 495  *
 496  * The function is used to mark the indicated state for the given
 497  * PE. Also, the associated PCI devices will be put into IO frozen
 498  * state as well.
 499  */
 500 static void *__eeh_pe_state_mark(void *data, void *flag)
 501 {
 502         struct eeh_pe *pe = (struct eeh_pe *)data;
 503         int state = *((int *)flag);
 504         struct eeh_dev *tmp;
 505         struct pci_dev *pdev;
 506
 507         /*
 508          * Mark the PE with the indicated state. Also,
 509          * the associated PCI device will be put into
 510          * I/O frozen state to avoid I/O accesses from
 511          * the PCI device driver.
 512          */
 513         pe->state |= state;
 514         eeh_pe_for_each_dev(pe, tmp) {
 515                 pdev = eeh_dev_to_pci_dev(tmp);
 516                 if (pdev)
 517                         pdev->error_state = pci_channel_io_frozen;
 518         }
 519
 520         return NULL;
 521 }
 522
 523 /**
 524  * eeh_pe_state_mark - Mark specified state for PE and its associated device
 525  * @pe: EEH PE
 526  *
 527  * EEH error affects the current PE and its child PEs. The function
 528  * is used to mark appropriate state for the affected PEs and the
 529  * associated devices.
 530  */
 531 void eeh_pe_state_mark(struct eeh_pe *pe, int state)
 532 {
 533         eeh_pe_traverse(pe, __eeh_pe_state_mark, &state);
 534 }
 535
 536 /**
 537  * __eeh_pe_state_clear - Clear state for the PE
 538  * @data: EEH PE
 539  * @flag: state
 540  *
 541  * The function is used to clear the indicated state from the
 542  * given PE. Besides, we also clear the check count of the PE
 543  * as well.
 544  */
 545 static void *__eeh_pe_state_clear(void *data, void *flag)
 546 {
 547         struct eeh_pe *pe = (struct eeh_pe *)data;
 548         int state = *((int *)flag);
 549
 550         pe->state &= ~state;
 551         pe->check_count = 0;
 552
 553         return NULL;
 554 }
 555
 556 /**
 557  * eeh_pe_state_clear - Clear state for the PE and its children
 558  * @pe: PE
 559  * @state: state to be cleared
 560  *
 561  * When the PE and its children has been recovered from error,
 562  * we need clear the error state for that. The function is used
 563  * for the purpose.
 564  */
 565 void eeh_pe_state_clear(struct eeh_pe *pe, int state)
 566 {
 567         eeh_pe_traverse(pe, __eeh_pe_state_clear, &state);
 568 }
 569
 570 /**
 571  * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
 572  * @data: EEH device
 573  * @flag: Unused
 574  *
 575  * Loads the PCI configuration space base address registers,
 576  * the expansion ROM base address, the latency timer, and etc.
 577  * from the saved values in the device node.
 578  */
 579 static void *eeh_restore_one_device_bars(void *data, void *flag)
 580 {
 581         int i;
 582         u32 cmd;
 583         struct eeh_dev *edev = (struct eeh_dev *)data;
 584         struct device_node *dn = eeh_dev_to_of_node(edev);
 585
 586         for (i = 4; i < 10; i++)
 587                 eeh_ops->write_config(dn, i*4, 4, edev->config_space[i]);
 588         /* 12 == Expansion ROM Address */
 589         eeh_ops->write_config(dn, 12*4, 4, edev->config_space[12]);
 590
 591 #define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
 592 #define SAVED_BYTE(OFF) (((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
 593
 594         eeh_ops->write_config(dn, PCI_CACHE_LINE_SIZE, 1,
 595                 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
 596         eeh_ops->write_config(dn, PCI_LATENCY_TIMER, 1,
 597                 SAVED_BYTE(PCI_LATENCY_TIMER));
 598
 599         /* max latency, min grant, interrupt pin and line */
 600         eeh_ops->write_config(dn, 15*4, 4, edev->config_space[15]);
 601
 602         /*
 603          * Restore PERR & SERR bits, some devices require it,
 604          * don't touch the other command bits
 605          */
 606         eeh_ops->read_config(dn, PCI_COMMAND, 4, &cmd);
 607         if (edev->config_space[1] & PCI_COMMAND_PARITY)
 608                 cmd |= PCI_COMMAND_PARITY;
 609         else
 610                 cmd &= ~PCI_COMMAND_PARITY;
 611         if (edev->config_space[1] & PCI_COMMAND_SERR)
 612                 cmd |= PCI_COMMAND_SERR;
 613         else
 614                 cmd &= ~PCI_COMMAND_SERR;
 615         eeh_ops->write_config(dn, PCI_COMMAND, 4, cmd);
 616
 617         return NULL;
 618 }
 619
 620 /**
 621  * eeh_pe_restore_bars - Restore the PCI config space info
 622  * @pe: EEH PE
 623  *
 624  * This routine performs a recursive walk to the children
 625  * of this device as well.
 626  */
 627 void eeh_pe_restore_bars(struct eeh_pe *pe)
 628 {
 629         /*
 630          * We needn't take the EEH lock since eeh_pe_dev_traverse()
 631          * will take that.
 632          */
 633         eeh_pe_dev_traverse(pe, eeh_restore_one_device_bars, NULL);
 634 }
 635
 636 /**
 637  * eeh_pe_bus_get - Retrieve PCI bus according to the given PE
 638  * @pe: EEH PE
 639  *
 640  * Retrieve the PCI bus according to the given PE. Basically,
 641  * there're 3 types of PEs: PHB/Bus/Device. For PHB PE, the
 642  * primary PCI bus will be retrieved. The parent bus will be
 643  * returned for BUS PE. However, we don't have associated PCI
 644  * bus for DEVICE PE.
 645  */
 646 struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe)
 647 {
 648         struct pci_bus *bus = NULL;
 649         struct eeh_dev *edev;
 650         struct pci_dev *pdev;
 651
 652         if (pe->type & EEH_PE_PHB) {
 653                 bus = pe->phb->bus;
 654         } else if (pe->type & EEH_PE_BUS ||
 655                    pe->type & EEH_PE_DEVICE) {
 656                 if (pe->bus) {
 657                         bus = pe->bus;
 658                         goto out;
 659                 }
 660
 661                 edev = list_first_entry(&pe->edevs, struct eeh_dev, list);
 662                 pdev = eeh_dev_to_pci_dev(edev);
 663                 if (pdev)
 664                         bus = pdev->bus;
 665         }
 666
 667 out:
 668         return bus;
 669 }