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