linux/arch/powerpc/platforms/pseries/eeh_driver.c
<<
>>
Prefs
   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
  37static 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
  45static 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 * eeh_disable_irq - disable interrupt for the recovering device
  68 */
  69static void eeh_disable_irq(struct pci_dev *dev)
  70{
  71        struct device_node *dn = pci_device_to_OF_node(dev);
  72
  73        /* Don't disable MSI and MSI-X interrupts. They are
  74         * effectively disabled by the DMA Stopped state
  75         * when an EEH error occurs.
  76        */
  77        if (dev->msi_enabled || dev->msix_enabled)
  78                return;
  79
  80        if (!irq_has_action(dev->irq))
  81                return;
  82
  83        PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
  84        disable_irq_nosync(dev->irq);
  85}
  86
  87/**
  88 * eeh_enable_irq - enable interrupt for the recovering device
  89 */
  90static void eeh_enable_irq(struct pci_dev *dev)
  91{
  92        struct device_node *dn = pci_device_to_OF_node(dev);
  93
  94        if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
  95                PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
  96                enable_irq(dev->irq);
  97        }
  98}
  99
 100/* ------------------------------------------------------- */
 101/**
 102 * eeh_report_error - report pci error to each device driver
 103 * 
 104 * Report an EEH error to each device driver, collect up and 
 105 * merge the device driver responses. Cumulative response 
 106 * passed back in "userdata".
 107 */
 108
 109static int eeh_report_error(struct pci_dev *dev, void *userdata)
 110{
 111        enum pci_ers_result rc, *res = userdata;
 112        struct pci_driver *driver = dev->driver;
 113
 114        dev->error_state = pci_channel_io_frozen;
 115
 116        if (!driver)
 117                return 0;
 118
 119        eeh_disable_irq(dev);
 120
 121        if (!driver->err_handler ||
 122            !driver->err_handler->error_detected)
 123                return 0;
 124
 125        rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
 126
 127        /* A driver that needs a reset trumps all others */
 128        if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 129        if (*res == PCI_ERS_RESULT_NONE) *res = rc;
 130
 131        return 0;
 132}
 133
 134/**
 135 * eeh_report_mmio_enabled - tell drivers that MMIO has been enabled
 136 *
 137 * Tells each device driver that IO ports, MMIO and config space I/O
 138 * are now enabled. Collects up and merges the device driver responses.
 139 * Cumulative response passed back in "userdata".
 140 */
 141
 142static int eeh_report_mmio_enabled(struct pci_dev *dev, void *userdata)
 143{
 144        enum pci_ers_result rc, *res = userdata;
 145        struct pci_driver *driver = dev->driver;
 146
 147        if (!driver ||
 148            !driver->err_handler ||
 149            !driver->err_handler->mmio_enabled)
 150                return 0;
 151
 152        rc = driver->err_handler->mmio_enabled (dev);
 153
 154        /* A driver that needs a reset trumps all others */
 155        if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 156        if (*res == PCI_ERS_RESULT_NONE) *res = rc;
 157
 158        return 0;
 159}
 160
 161/**
 162 * eeh_report_reset - tell device that slot has been reset
 163 */
 164
 165static int eeh_report_reset(struct pci_dev *dev, void *userdata)
 166{
 167        enum pci_ers_result rc, *res = userdata;
 168        struct pci_driver *driver = dev->driver;
 169
 170        if (!driver)
 171                return 0;
 172
 173        dev->error_state = pci_channel_io_normal;
 174
 175        eeh_enable_irq(dev);
 176
 177        if (!driver->err_handler ||
 178            !driver->err_handler->slot_reset)
 179                return 0;
 180
 181        rc = driver->err_handler->slot_reset(dev);
 182        if ((*res == PCI_ERS_RESULT_NONE) ||
 183            (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
 184        if (*res == PCI_ERS_RESULT_DISCONNECT &&
 185             rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
 186
 187        return 0;
 188}
 189
 190/**
 191 * eeh_report_resume - tell device to resume normal operations
 192 */
 193
 194static int eeh_report_resume(struct pci_dev *dev, void *userdata)
 195{
 196        struct pci_driver *driver = dev->driver;
 197
 198        dev->error_state = pci_channel_io_normal;
 199
 200        if (!driver)
 201                return 0;
 202
 203        eeh_enable_irq(dev);
 204
 205        if (!driver->err_handler ||
 206            !driver->err_handler->resume)
 207                return 0;
 208
 209        driver->err_handler->resume(dev);
 210
 211        return 0;
 212}
 213
 214/**
 215 * eeh_report_failure - tell device driver that device is dead.
 216 *
 217 * This informs the device driver that the device is permanently
 218 * dead, and that no further recovery attempts will be made on it.
 219 */
 220
 221static int eeh_report_failure(struct pci_dev *dev, void *userdata)
 222{
 223        struct pci_driver *driver = dev->driver;
 224
 225        dev->error_state = pci_channel_io_perm_failure;
 226
 227        if (!driver)
 228                return 0;
 229
 230        eeh_disable_irq(dev);
 231
 232        if (!driver->err_handler ||
 233            !driver->err_handler->error_detected)
 234                return 0;
 235
 236        driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
 237
 238        return 0;
 239}
 240
 241/* ------------------------------------------------------- */
 242/**
 243 * handle_eeh_events -- reset a PCI device after hard lockup.
 244 *
 245 * pSeries systems will isolate a PCI slot if the PCI-Host
 246 * bridge detects address or data parity errors, DMA's
 247 * occurring to wild addresses (which usually happen due to
 248 * bugs in device drivers or in PCI adapter firmware).
 249 * Slot isolations also occur if #SERR, #PERR or other misc
 250 * PCI-related errors are detected.
 251 *
 252 * Recovery process consists of unplugging the device driver
 253 * (which generated hotplug events to userspace), then issuing
 254 * a PCI #RST to the device, then reconfiguring the PCI config
 255 * space for all bridges & devices under this slot, and then
 256 * finally restarting the device drivers (which cause a second
 257 * set of hotplug events to go out to userspace).
 258 */
 259
 260/**
 261 * eeh_reset_device() -- perform actual reset of a pci slot
 262 * @bus: pointer to the pci bus structure corresponding
 263 *            to the isolated slot. A non-null value will
 264 *            cause all devices under the bus to be removed
 265 *            and then re-added.
 266 * @pe_dn: pointer to a "Partionable Endpoint" device node.
 267 *            This is the top-level structure on which pci
 268 *            bus resets can be performed.
 269 */
 270
 271static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
 272{
 273        struct device_node *dn;
 274        int cnt, rc;
 275
 276        /* pcibios will clear the counter; save the value */
 277        cnt = pe_dn->eeh_freeze_count;
 278
 279        if (bus)
 280                pcibios_remove_pci_devices(bus);
 281
 282        /* Reset the pci controller. (Asserts RST#; resets config space).
 283         * Reconfigure bridges and devices. Don't try to bring the system
 284         * up if the reset failed for some reason. */
 285        rc = rtas_set_slot_reset(pe_dn);
 286        if (rc)
 287                return rc;
 288
 289        /* Walk over all functions on this device.  */
 290        dn = pe_dn->node;
 291        if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
 292                dn = dn->parent->child;
 293
 294        while (dn) {
 295                struct pci_dn *ppe = PCI_DN(dn);
 296                /* On Power4, always true because eeh_pe_config_addr=0 */
 297                if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) {
 298                        rtas_configure_bridge(ppe);
 299                        eeh_restore_bars(ppe);
 300                }
 301                dn = dn->sibling;
 302        }
 303
 304        /* Give the system 5 seconds to finish running the user-space
 305         * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes, 
 306         * this is a hack, but if we don't do this, and try to bring 
 307         * the device up before the scripts have taken it down, 
 308         * potentially weird things happen.
 309         */
 310        if (bus) {
 311                ssleep (5);
 312                pcibios_add_pci_devices(bus);
 313        }
 314        pe_dn->eeh_freeze_count = cnt;
 315
 316        return 0;
 317}
 318
 319/* The longest amount of time to wait for a pci device
 320 * to come back on line, in seconds.
 321 */
 322#define MAX_WAIT_FOR_RECOVERY 150
 323
 324struct pci_dn * handle_eeh_events (struct eeh_event *event)
 325{
 326        struct device_node *frozen_dn;
 327        struct pci_dn *frozen_pdn;
 328        struct pci_bus *frozen_bus;
 329        int rc = 0;
 330        enum pci_ers_result result = PCI_ERS_RESULT_NONE;
 331        const char *location, *pci_str, *drv_str, *bus_pci_str, *bus_drv_str;
 332
 333        frozen_dn = find_device_pe(event->dn);
 334        if (!frozen_dn) {
 335
 336                location = of_get_property(event->dn, "ibm,loc-code", NULL);
 337                location = location ? location : "unknown";
 338                printk(KERN_ERR "EEH: Error: Cannot find partition endpoint "
 339                                "for location=%s pci addr=%s\n",
 340                        location, eeh_pci_name(event->dev));
 341                return NULL;
 342        }
 343
 344        frozen_bus = pcibios_find_pci_bus(frozen_dn);
 345        location = of_get_property(frozen_dn, "ibm,loc-code", NULL);
 346        location = location ? location : "unknown";
 347
 348        /* There are two different styles for coming up with the PE.
 349         * In the old style, it was the highest EEH-capable device
 350         * which was always an EADS pci bridge.  In the new style,
 351         * there might not be any EADS bridges, and even when there are,
 352         * the firmware marks them as "EEH incapable". So another
 353         * two-step is needed to find the pci bus.. */
 354        if (!frozen_bus)
 355                frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
 356
 357        if (!frozen_bus) {
 358                printk(KERN_ERR "EEH: Cannot find PCI bus "
 359                        "for location=%s dn=%s\n",
 360                        location, frozen_dn->full_name);
 361                return NULL;
 362        }
 363
 364        frozen_pdn = PCI_DN(frozen_dn);
 365        frozen_pdn->eeh_freeze_count++;
 366
 367        pci_str = eeh_pci_name(event->dev);
 368        drv_str = pcid_name(event->dev);
 369        
 370        if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
 371                goto excess_failures;
 372
 373        printk(KERN_WARNING
 374           "EEH: This PCI device has failed %d times in the last hour:\n",
 375                frozen_pdn->eeh_freeze_count);
 376
 377        if (frozen_pdn->pcidev) {
 378                bus_pci_str = pci_name(frozen_pdn->pcidev);
 379                bus_drv_str = pcid_name(frozen_pdn->pcidev);
 380                printk(KERN_WARNING
 381                        "EEH: Bus location=%s driver=%s pci addr=%s\n",
 382                        location, bus_drv_str, bus_pci_str);
 383        }
 384
 385        printk(KERN_WARNING
 386                "EEH: Device 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        
 478excess_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
 492hard_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
 499perm_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 ---------- */
 512