arch/powerpc/platforms/pseries/eeh_driver.c

   1 /*
   2  * PCI Error Recovery Driver for RPA-compliant PPC64 platform.
   3  * Copyright IBM Corp. 2004 2005
   4  * Copyright Linas Vepstas <linas@linas.org> 2004, 2005
   5  *
   6  * All rights reserved.
   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 (at
  11  * your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful, but
  14  * WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  16  * NON INFRINGEMENT.  See the GNU General Public License for more
  17  * details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  *
  23  * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
  24  */
  25 #include <linux/delay.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/irq.h>
  28 #include <linux/pci.h>
  29 #include <asm/eeh.h>
  30 #include <asm/eeh_event.h>
  31 #include <asm/ppc-pci.h>
  32 #include <asm/pci-bridge.h>
  33 #include <asm/prom.h>
  34 #include <asm/rtas.h>
  35
  36
  37 static inline const char * pcid_name (struct pci_dev *pdev)
  38 {
  39         if (pdev && pdev->dev.driver)
  40                 return pdev->dev.driver->name;
  41         return "";
  42 }
  43
  44 #if 0
  45 static void print_device_node_tree(struct pci_dn *pdn, int dent)
  46 {
  47         int i;
  48         struct device_node *pc;
  49
  50         if (!pdn)
  51                 return;
  52         for (i = 0; i < dent; i++)
  53                 printk(" ");
  54         printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n",
  55                 pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr,
  56                 pdn->eeh_pe_config_addr, pdn->node->full_name);
  57         dent += 3;
  58         pc = pdn->node->child;
  59         while (pc) {
  60                 print_device_node_tree(PCI_DN(pc), dent);
  61                 pc = pc->sibling;
  62         }
  63 }
  64 #endif
  65
  66 /**
  67  * irq_in_use - return true if this irq is being used
  68  */
  69 static int irq_in_use(unsigned int irq)
  70 {
  71         int rc = 0;
  72         unsigned long flags;
  73    struct irq_desc *desc = irq_desc + irq;
  74
  75         spin_lock_irqsave(&desc->lock, flags);
  76         if (desc->action)
  77                 rc = 1;
  78         spin_unlock_irqrestore(&desc->lock, flags);
  79         return rc;
  80 }
  81
  82 /**
  83  * eeh_disable_irq - disable interrupt for the recovering device
  84  */
  85 static void eeh_disable_irq(struct pci_dev *dev)
  86 {
  87         struct device_node *dn = pci_device_to_OF_node(dev);
  88
  89         /* Don't disable MSI and MSI-X interrupts. They are
  90          * effectively disabled by the DMA Stopped state
  91          * when an EEH error occurs.
  92         */
  93         if (dev->msi_enabled || dev->msix_enabled)
  94                 return;
  95
  96         if (!irq_in_use(dev->irq))
  97                 return;
  98
  99         PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
 100         disable_irq_nosync(dev->irq);
 101 }
 102
 103 /**
 104  * eeh_enable_irq - enable interrupt for the recovering device
 105  */
 106 static void eeh_enable_irq(struct pci_dev *dev)
 107 {
 108         struct device_node *dn = pci_device_to_OF_node(dev);
 109
 110         if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
 111                 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
 112                 enable_irq(dev->irq);
 113         }
 114 }
 115
 116 /* ------------------------------------------------------- */
 117 /**
 118  * eeh_report_error - report pci error to each device driver
 119  *
 120  * Report an EEH error to each device driver, collect up and
 121  * merge the device driver responses. Cumulative response
 122  * passed back in "userdata".
 123  */
 124
 125 static void eeh_report_error(struct pci_dev *dev, void *userdata)
 126 {
 127         enum pci_ers_result rc, *res = userdata;
 128         struct pci_driver *driver = dev->driver;
 129
 130         dev->error_state = pci_channel_io_frozen;
 131
 132         if (!driver)
 133                 return;
 134
 135         eeh_disable_irq(dev);
 136
 137         if (!driver->err_handler ||
 138             !driver->err_handler->error_detected)
 139                 return;
 140
 141         rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
 142
 143         /* A driver that needs a reset trumps all others */
 144         if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 145         if (*res == PCI_ERS_RESULT_NONE) *res = rc;
 146 }
 147
 148 /**
 149  * eeh_report_mmio_enabled - tell drivers that MMIO has been enabled
 150  *
 151  * Tells each device driver that IO ports, MMIO and config space I/O
 152  * are now enabled. Collects up and merges the device driver responses.
 153  * Cumulative response passed back in "userdata".
 154  */
 155
 156 static void eeh_report_mmio_enabled(struct pci_dev *dev, void *userdata)
 157 {
 158         enum pci_ers_result rc, *res = userdata;
 159         struct pci_driver *driver = dev->driver;
 160
 161         if (!driver ||
 162             !driver->err_handler ||
 163             !driver->err_handler->mmio_enabled)
 164                 return;
 165
 166         rc = driver->err_handler->mmio_enabled (dev);
 167
 168         /* A driver that needs a reset trumps all others */
 169         if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 170         if (*res == PCI_ERS_RESULT_NONE) *res = rc;
 171 }
 172
 173 /**
 174  * eeh_report_reset - tell device that slot has been reset
 175  */
 176
 177 static void eeh_report_reset(struct pci_dev *dev, void *userdata)
 178 {
 179         enum pci_ers_result rc, *res = userdata;
 180         struct pci_driver *driver = dev->driver;
 181
 182         if (!driver)
 183                 return;
 184
 185         eeh_enable_irq(dev);
 186
 187         if (!driver->err_handler ||
 188             !driver->err_handler->slot_reset)
 189                 return;
 190
 191         rc = driver->err_handler->slot_reset(dev);
 192         if ((*res == PCI_ERS_RESULT_NONE) ||
 193             (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
 194         if (*res == PCI_ERS_RESULT_DISCONNECT &&
 195              rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 196 }
 197
 198 /**
 199  * eeh_report_resume - tell device to resume normal operations
 200  */
 201
 202 static void eeh_report_resume(struct pci_dev *dev, void *userdata)
 203 {
 204         struct pci_driver *driver = dev->driver;
 205
 206         dev->error_state = pci_channel_io_normal;
 207
 208         if (!driver)
 209                 return;
 210
 211         eeh_enable_irq(dev);
 212
 213         if (!driver->err_handler ||
 214             !driver->err_handler->resume)
 215                 return;
 216
 217         driver->err_handler->resume(dev);
 218 }
 219
 220 /**
 221  * eeh_report_failure - tell device driver that device is dead.
 222  *
 223  * This informs the device driver that the device is permanently
 224  * dead, and that no further recovery attempts will be made on it.
 225  */
 226
 227 static void eeh_report_failure(struct pci_dev *dev, void *userdata)
 228 {
 229         struct pci_driver *driver = dev->driver;
 230
 231         dev->error_state = pci_channel_io_perm_failure;
 232
 233         if (!driver)
 234                 return;
 235
 236         eeh_disable_irq(dev);
 237
 238         if (!driver->err_handler ||
 239             !driver->err_handler->error_detected)
 240                 return;
 241
 242         driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
 243 }
 244
 245 /* ------------------------------------------------------- */
 246 /**
 247  * handle_eeh_events -- reset a PCI device after hard lockup.
 248  *
 249  * pSeries systems will isolate a PCI slot if the PCI-Host
 250  * bridge detects address or data parity errors, DMA's
 251  * occurring to wild addresses (which usually happen due to
 252  * bugs in device drivers or in PCI adapter firmware).
 253  * Slot isolations also occur if #SERR, #PERR or other misc
 254  * PCI-related errors are detected.
 255  *
 256  * Recovery process consists of unplugging the device driver
 257  * (which generated hotplug events to userspace), then issuing
 258  * a PCI #RST to the device, then reconfiguring the PCI config
 259  * space for all bridges & devices under this slot, and then
 260  * finally restarting the device drivers (which cause a second
 261  * set of hotplug events to go out to userspace).
 262  */
 263
 264 /**
 265  * eeh_reset_device() -- perform actual reset of a pci slot
 266  * @bus: pointer to the pci bus structure corresponding
 267  *            to the isolated slot. A non-null value will
 268  *            cause all devices under the bus to be removed
 269  *            and then re-added.
 270  * @pe_dn: pointer to a "Partionable Endpoint" device node.
 271  *            This is the top-level structure on which pci
 272  *            bus resets can be performed.
 273  */
 274
 275 static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
 276 {
 277         struct device_node *dn;
 278         int cnt, rc;
 279
 280         /* pcibios will clear the counter; save the value */
 281         cnt = pe_dn->eeh_freeze_count;
 282
 283         if (bus)
 284                 pcibios_remove_pci_devices(bus);
 285
 286         /* Reset the pci controller. (Asserts RST#; resets config space).
 287          * Reconfigure bridges and devices. Don't try to bring the system
 288          * up if the reset failed for some reason. */
 289         rc = rtas_set_slot_reset(pe_dn);
 290         if (rc)
 291                 return rc;
 292
 293         /* Walk over all functions on this device.  */
 294         dn = pe_dn->node;
 295         if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
 296                 dn = dn->parent->child;
 297
 298         while (dn) {
 299                 struct pci_dn *ppe = PCI_DN(dn);
 300                 /* On Power4, always true because eeh_pe_config_addr=0 */
 301                 if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) {
 302                         rtas_configure_bridge(ppe);
 303                         eeh_restore_bars(ppe);
 304                 }
 305                 dn = dn->sibling;
 306         }
 307
 308         /* Give the system 5 seconds to finish running the user-space
 309          * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes,
 310          * this is a hack, but if we don't do this, and try to bring
 311          * the device up before the scripts have taken it down,
 312          * potentially weird things happen.
 313          */
 314         if (bus) {
 315                 ssleep (5);
 316                 pcibios_add_pci_devices(bus);
 317         }
 318         pe_dn->eeh_freeze_count = cnt;
 319
 320         return 0;
 321 }
 322
 323 /* The longest amount of time to wait for a pci device
 324  * to come back on line, in seconds.
 325  */
 326 #define MAX_WAIT_FOR_RECOVERY 150
 327
 328 struct pci_dn * handle_eeh_events (struct eeh_event *event)
 329 {
 330         struct device_node *frozen_dn;
 331         struct pci_dn *frozen_pdn;
 332         struct pci_bus *frozen_bus;
 333         int rc = 0;
 334         enum pci_ers_result result = PCI_ERS_RESULT_NONE;
 335         const char *location, *pci_str, *drv_str;
 336
 337         frozen_dn = find_device_pe(event->dn);
 338         if (!frozen_dn) {
 339
 340                 location = of_get_property(event->dn, "ibm,loc-code", NULL);
 341                 location = location ? location : "unknown";
 342                 printk(KERN_ERR "EEH: Error: Cannot find partition endpoint "
 343                                 "for location=%s pci addr=%s\n",
 344                         location, pci_name(event->dev));
 345                 return NULL;
 346         }
 347
 348         frozen_bus = pcibios_find_pci_bus(frozen_dn);
 349         location = of_get_property(frozen_dn, "ibm,loc-code", NULL);
 350         location = location ? location : "unknown";
 351
 352         /* There are two different styles for coming up with the PE.
 353          * In the old style, it was the highest EEH-capable device
 354          * which was always an EADS pci bridge.  In the new style,
 355          * there might not be any EADS bridges, and even when there are,
 356          * the firmware marks them as "EEH incapable". So another
 357          * two-step is needed to find the pci bus.. */
 358         if (!frozen_bus)
 359                 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
 360
 361         if (!frozen_bus) {
 362                 printk(KERN_ERR "EEH: Cannot find PCI bus "
 363                         "for location=%s dn=%s\n",
 364                         location, frozen_dn->full_name);
 365                 return NULL;
 366         }
 367
 368         frozen_pdn = PCI_DN(frozen_dn);
 369         frozen_pdn->eeh_freeze_count++;
 370
 371         if (frozen_pdn->pcidev) {
 372                 pci_str = pci_name (frozen_pdn->pcidev);
 373                 drv_str = pcid_name (frozen_pdn->pcidev);
 374         } else {
 375                 pci_str = pci_name (event->dev);
 376                 drv_str = pcid_name (event->dev);
 377         }
 378
 379         if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
 380                 goto excess_failures;
 381
 382         printk(KERN_WARNING
 383            "EEH: This PCI device has failed %d times in the last hour:\n",
 384                 frozen_pdn->eeh_freeze_count);
 385         printk(KERN_WARNING
 386                 "EEH: location=%s driver=%s pci addr=%s\n",
 387                 location, drv_str, pci_str);
 388
 389         /* Walk the various device drivers attached to this slot through
 390          * a reset sequence, giving each an opportunity to do what it needs
 391          * to accomplish the reset.  Each child gets a report of the
 392          * status ... if any child can't handle the reset, then the entire
 393          * slot is dlpar removed and added.
 394          */
 395         pci_walk_bus(frozen_bus, eeh_report_error, &result);
 396
 397         /* Get the current PCI slot state. This can take a long time,
 398          * sometimes over 3 seconds for certain systems. */
 399         rc = eeh_wait_for_slot_status (frozen_pdn, MAX_WAIT_FOR_RECOVERY*1000);
 400         if (rc < 0) {
 401                 printk(KERN_WARNING "EEH: Permanent failure\n");
 402                 goto hard_fail;
 403         }
 404
 405         /* Since rtas may enable MMIO when posting the error log,
 406          * don't post the error log until after all dev drivers
 407          * have been informed.
 408          */
 409         eeh_slot_error_detail(frozen_pdn, EEH_LOG_TEMP_FAILURE);
 410
 411         /* If all device drivers were EEH-unaware, then shut
 412          * down all of the device drivers, and hope they
 413          * go down willingly, without panicing the system.
 414          */
 415         if (result == PCI_ERS_RESULT_NONE) {
 416                 rc = eeh_reset_device(frozen_pdn, frozen_bus);
 417                 if (rc) {
 418                         printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc);
 419                         goto hard_fail;
 420                 }
 421         }
 422
 423         /* If all devices reported they can proceed, then re-enable MMIO */
 424         if (result == PCI_ERS_RESULT_CAN_RECOVER) {
 425                 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_MMIO);
 426
 427                 if (rc < 0)
 428                         goto hard_fail;
 429                 if (rc) {
 430                         result = PCI_ERS_RESULT_NEED_RESET;
 431                 } else {
 432                         result = PCI_ERS_RESULT_NONE;
 433                         pci_walk_bus(frozen_bus, eeh_report_mmio_enabled, &result);
 434                 }
 435         }
 436
 437         /* If all devices reported they can proceed, then re-enable DMA */
 438         if (result == PCI_ERS_RESULT_CAN_RECOVER) {
 439                 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_DMA);
 440
 441                 if (rc < 0)
 442                         goto hard_fail;
 443                 if (rc)
 444                         result = PCI_ERS_RESULT_NEED_RESET;
 445                 else
 446                         result = PCI_ERS_RESULT_RECOVERED;
 447         }
 448
 449         /* If any device has a hard failure, then shut off everything. */
 450         if (result == PCI_ERS_RESULT_DISCONNECT) {
 451                 printk(KERN_WARNING "EEH: Device driver gave up\n");
 452                 goto hard_fail;
 453         }
 454
 455         /* If any device called out for a reset, then reset the slot */
 456         if (result == PCI_ERS_RESULT_NEED_RESET) {
 457                 rc = eeh_reset_device(frozen_pdn, NULL);
 458                 if (rc) {
 459                         printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc);
 460                         goto hard_fail;
 461                 }
 462                 result = PCI_ERS_RESULT_NONE;
 463                 pci_walk_bus(frozen_bus, eeh_report_reset, &result);
 464         }
 465
 466         /* All devices should claim they have recovered by now. */
 467         if ((result != PCI_ERS_RESULT_RECOVERED) &&
 468             (result != PCI_ERS_RESULT_NONE)) {
 469                 printk(KERN_WARNING "EEH: Not recovered\n");
 470                 goto hard_fail;
 471         }
 472
 473         /* Tell all device drivers that they can resume operations */
 474         pci_walk_bus(frozen_bus, eeh_report_resume, NULL);
 475
 476         return frozen_pdn;
 477
 478 excess_failures:
 479         /*
 480          * About 90% of all real-life EEH failures in the field
 481          * are due to poorly seated PCI cards. Only 10% or so are
 482          * due to actual, failed cards.
 483          */
 484         printk(KERN_ERR
 485            "EEH: PCI device at location=%s driver=%s pci addr=%s \n"
 486                 "has failed %d times in the last hour "
 487                 "and has been permanently disabled. \n"
 488                 "Please try reseating this device or replacing it.\n",
 489                 location, drv_str, pci_str, frozen_pdn->eeh_freeze_count);
 490         goto perm_error;
 491
 492 hard_fail:
 493         printk(KERN_ERR
 494            "EEH: Unable to recover from failure of PCI device "
 495            "at location=%s driver=%s pci addr=%s \n"
 496            "Please try reseating this device or replacing it.\n",
 497                 location, drv_str, pci_str);
 498
 499 perm_error:
 500         eeh_slot_error_detail(frozen_pdn, EEH_LOG_PERM_FAILURE);
 501
 502         /* Notify all devices that they're about to go down. */
 503         pci_walk_bus(frozen_bus, eeh_report_failure, NULL);
 504
 505         /* Shut down the device drivers for good. */
 506         pcibios_remove_pci_devices(frozen_bus);
 507
 508         return NULL;
 509 }
 510
 511 /* ---------- end of file ---------- */