powerpc/kernel/eeh_driver.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * PCI Error Recovery Driver for RPA-compliant PPC64 platform.
0004  * Copyright IBM Corp. 2004 2005
0005  * Copyright Linas Vepstas <linas@linas.org> 2004, 2005
0006  *
0007  * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
0008  */
0009 #include <linux/delay.h>
0010 #include <linux/interrupt.h>
0011 #include <linux/irq.h>
0012 #include <linux/module.h>
0013 #include <linux/pci.h>
0014 #include <linux/pci_hotplug.h>
0015 #include <asm/eeh.h>
0016 #include <asm/eeh_event.h>
0017 #include <asm/ppc-pci.h>
0018 #include <asm/pci-bridge.h>
0019 #include <asm/rtas.h>
0020
0021 struct eeh_rmv_data {
0022     struct list_head removed_vf_list;
0023     int removed_dev_count;
0024 };
0025
0026 static int eeh_result_priority(enum pci_ers_result result)
0027 {
0028     switch (result) {
0029     case PCI_ERS_RESULT_NONE:
0030         return 1;
0031     case PCI_ERS_RESULT_NO_AER_DRIVER:
0032         return 2;
0033     case PCI_ERS_RESULT_RECOVERED:
0034         return 3;
0035     case PCI_ERS_RESULT_CAN_RECOVER:
0036         return 4;
0037     case PCI_ERS_RESULT_DISCONNECT:
0038         return 5;
0039     case PCI_ERS_RESULT_NEED_RESET:
0040         return 6;
0041     default:
0042         WARN_ONCE(1, "Unknown pci_ers_result value: %d\n", (int)result);
0043         return 0;
0044     }
0045 };
0046
0047 static const char *pci_ers_result_name(enum pci_ers_result result)
0048 {
0049     switch (result) {
0050     case PCI_ERS_RESULT_NONE:
0051         return "none";
0052     case PCI_ERS_RESULT_CAN_RECOVER:
0053         return "can recover";
0054     case PCI_ERS_RESULT_NEED_RESET:
0055         return "need reset";
0056     case PCI_ERS_RESULT_DISCONNECT:
0057         return "disconnect";
0058     case PCI_ERS_RESULT_RECOVERED:
0059         return "recovered";
0060     case PCI_ERS_RESULT_NO_AER_DRIVER:
0061         return "no AER driver";
0062     default:
0063         WARN_ONCE(1, "Unknown result type: %d\n", (int)result);
0064         return "unknown";
0065     }
0066 };
0067
0068 static enum pci_ers_result pci_ers_merge_result(enum pci_ers_result old,
0069                         enum pci_ers_result new)
0070 {
0071     if (eeh_result_priority(new) > eeh_result_priority(old))
0072         return new;
0073     return old;
0074 }
0075
0076 static bool eeh_dev_removed(struct eeh_dev *edev)
0077 {
0078     return !edev || (edev->mode & EEH_DEV_REMOVED);
0079 }
0080
0081 static bool eeh_edev_actionable(struct eeh_dev *edev)
0082 {
0083     if (!edev->pdev)
0084         return false;
0085     if (edev->pdev->error_state == pci_channel_io_perm_failure)
0086         return false;
0087     if (eeh_dev_removed(edev))
0088         return false;
0089     if (eeh_pe_passed(edev->pe))
0090         return false;
0091
0092     return true;
0093 }
0094
0095 /**
0096  * eeh_pcid_get - Get the PCI device driver
0097  * @pdev: PCI device
0098  *
0099  * The function is used to retrieve the PCI device driver for
0100  * the indicated PCI device. Besides, we will increase the reference
0101  * of the PCI device driver to prevent that being unloaded on
0102  * the fly. Otherwise, kernel crash would be seen.
0103  */
0104 static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
0105 {
0106     if (!pdev || !pdev->dev.driver)
0107         return NULL;
0108
0109     if (!try_module_get(pdev->dev.driver->owner))
0110         return NULL;
0111
0112     return to_pci_driver(pdev->dev.driver);
0113 }
0114
0115 /**
0116  * eeh_pcid_put - Dereference on the PCI device driver
0117  * @pdev: PCI device
0118  *
0119  * The function is called to do dereference on the PCI device
0120  * driver of the indicated PCI device.
0121  */
0122 static inline void eeh_pcid_put(struct pci_dev *pdev)
0123 {
0124     if (!pdev || !pdev->dev.driver)
0125         return;
0126
0127     module_put(pdev->dev.driver->owner);
0128 }
0129
0130 /**
0131  * eeh_disable_irq - Disable interrupt for the recovering device
0132  * @dev: PCI device
0133  *
0134  * This routine must be called when reporting temporary or permanent
0135  * error to the particular PCI device to disable interrupt of that
0136  * device. If the device has enabled MSI or MSI-X interrupt, we needn't
0137  * do real work because EEH should freeze DMA transfers for those PCI
0138  * devices encountering EEH errors, which includes MSI or MSI-X.
0139  */
0140 static void eeh_disable_irq(struct eeh_dev *edev)
0141 {
0142     /* Don't disable MSI and MSI-X interrupts. They are
0143      * effectively disabled by the DMA Stopped state
0144      * when an EEH error occurs.
0145      */
0146     if (edev->pdev->msi_enabled || edev->pdev->msix_enabled)
0147         return;
0148
0149     if (!irq_has_action(edev->pdev->irq))
0150         return;
0151
0152     edev->mode |= EEH_DEV_IRQ_DISABLED;
0153     disable_irq_nosync(edev->pdev->irq);
0154 }
0155
0156 /**
0157  * eeh_enable_irq - Enable interrupt for the recovering device
0158  * @dev: PCI device
0159  *
0160  * This routine must be called to enable interrupt while failed
0161  * device could be resumed.
0162  */
0163 static void eeh_enable_irq(struct eeh_dev *edev)
0164 {
0165     if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
0166         edev->mode &= ~EEH_DEV_IRQ_DISABLED;
0167         /*
0168          * FIXME !!!!!
0169          *
0170          * This is just ass backwards. This maze has
0171          * unbalanced irq_enable/disable calls. So instead of
0172          * finding the root cause it works around the warning
0173          * in the irq_enable code by conditionally calling
0174          * into it.
0175          *
0176          * That's just wrong.The warning in the core code is
0177          * there to tell people to fix their asymmetries in
0178          * their own code, not by abusing the core information
0179          * to avoid it.
0180          *
0181          * I so wish that the assymetry would be the other way
0182          * round and a few more irq_disable calls render that
0183          * shit unusable forever.
0184          *
0185          *  tglx
0186          */
0187         if (irqd_irq_disabled(irq_get_irq_data(edev->pdev->irq)))
0188             enable_irq(edev->pdev->irq);
0189     }
0190 }
0191
0192 static void eeh_dev_save_state(struct eeh_dev *edev, void *userdata)
0193 {
0194     struct pci_dev *pdev;
0195
0196     if (!edev)
0197         return;
0198
0199     /*
0200      * We cannot access the config space on some adapters.
0201      * Otherwise, it will cause fenced PHB. We don't save
0202      * the content in their config space and will restore
0203      * from the initial config space saved when the EEH
0204      * device is created.
0205      */
0206     if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED))
0207         return;
0208
0209     pdev = eeh_dev_to_pci_dev(edev);
0210     if (!pdev)
0211         return;
0212
0213     pci_save_state(pdev);
0214 }
0215
0216 static void eeh_set_channel_state(struct eeh_pe *root, pci_channel_state_t s)
0217 {
0218     struct eeh_pe *pe;
0219     struct eeh_dev *edev, *tmp;
0220
0221     eeh_for_each_pe(root, pe)
0222         eeh_pe_for_each_dev(pe, edev, tmp)
0223             if (eeh_edev_actionable(edev))
0224                 edev->pdev->error_state = s;
0225 }
0226
0227 static void eeh_set_irq_state(struct eeh_pe *root, bool enable)
0228 {
0229     struct eeh_pe *pe;
0230     struct eeh_dev *edev, *tmp;
0231
0232     eeh_for_each_pe(root, pe) {
0233         eeh_pe_for_each_dev(pe, edev, tmp) {
0234             if (!eeh_edev_actionable(edev))
0235                 continue;
0236
0237             if (!eeh_pcid_get(edev->pdev))
0238                 continue;
0239
0240             if (enable)
0241                 eeh_enable_irq(edev);
0242             else
0243                 eeh_disable_irq(edev);
0244
0245             eeh_pcid_put(edev->pdev);
0246         }
0247     }
0248 }
0249
0250 typedef enum pci_ers_result (*eeh_report_fn)(struct eeh_dev *,
0251                          struct pci_dev *,
0252                          struct pci_driver *);
0253 static void eeh_pe_report_edev(struct eeh_dev *edev, eeh_report_fn fn,
0254                    enum pci_ers_result *result)
0255 {
0256     struct pci_dev *pdev;
0257     struct pci_driver *driver;
0258     enum pci_ers_result new_result;
0259
0260     pci_lock_rescan_remove();
0261     pdev = edev->pdev;
0262     if (pdev)
0263         get_device(&pdev->dev);
0264     pci_unlock_rescan_remove();
0265     if (!pdev) {
0266         eeh_edev_info(edev, "no device");
0267         return;
0268     }
0269     device_lock(&pdev->dev);
0270     if (eeh_edev_actionable(edev)) {
0271         driver = eeh_pcid_get(pdev);
0272
0273         if (!driver)
0274             eeh_edev_info(edev, "no driver");
0275         else if (!driver->err_handler)
0276             eeh_edev_info(edev, "driver not EEH aware");
0277         else if (edev->mode & EEH_DEV_NO_HANDLER)
0278             eeh_edev_info(edev, "driver bound too late");
0279         else {
0280             new_result = fn(edev, pdev, driver);
0281             eeh_edev_info(edev, "%s driver reports: '%s'",
0282                       driver->name,
0283                       pci_ers_result_name(new_result));
0284             if (result)
0285                 *result = pci_ers_merge_result(*result,
0286                                    new_result);
0287         }
0288         if (driver)
0289             eeh_pcid_put(pdev);
0290     } else {
0291         eeh_edev_info(edev, "not actionable (%d,%d,%d)", !!pdev,
0292                   !eeh_dev_removed(edev), !eeh_pe_passed(edev->pe));
0293     }
0294     device_unlock(&pdev->dev);
0295     if (edev->pdev != pdev)
0296         eeh_edev_warn(edev, "Device changed during processing!\n");
0297     put_device(&pdev->dev);
0298 }
0299
0300 static void eeh_pe_report(const char *name, struct eeh_pe *root,
0301               eeh_report_fn fn, enum pci_ers_result *result)
0302 {
0303     struct eeh_pe *pe;
0304     struct eeh_dev *edev, *tmp;
0305
0306     pr_info("EEH: Beginning: '%s'\n", name);
0307     eeh_for_each_pe(root, pe) eeh_pe_for_each_dev(pe, edev, tmp)
0308         eeh_pe_report_edev(edev, fn, result);
0309     if (result)
0310         pr_info("EEH: Finished:'%s' with aggregate recovery state:'%s'\n",
0311             name, pci_ers_result_name(*result));
0312     else
0313         pr_info("EEH: Finished:'%s'", name);
0314 }
0315
0316 /**
0317  * eeh_report_error - Report pci error to each device driver
0318  * @edev: eeh device
0319  * @driver: device's PCI driver
0320  *
0321  * Report an EEH error to each device driver.
0322  */
0323 static enum pci_ers_result eeh_report_error(struct eeh_dev *edev,
0324                         struct pci_dev *pdev,
0325                         struct pci_driver *driver)
0326 {
0327     enum pci_ers_result rc;
0328
0329     if (!driver->err_handler->error_detected)
0330         return PCI_ERS_RESULT_NONE;
0331
0332     eeh_edev_info(edev, "Invoking %s->error_detected(IO frozen)",
0333               driver->name);
0334     rc = driver->err_handler->error_detected(pdev, pci_channel_io_frozen);
0335
0336     edev->in_error = true;
0337     pci_uevent_ers(pdev, PCI_ERS_RESULT_NONE);
0338     return rc;
0339 }
0340
0341 /**
0342  * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
0343  * @edev: eeh device
0344  * @driver: device's PCI driver
0345  *
0346  * Tells each device driver that IO ports, MMIO and config space I/O
0347  * are now enabled.
0348  */
0349 static enum pci_ers_result eeh_report_mmio_enabled(struct eeh_dev *edev,
0350                            struct pci_dev *pdev,
0351                            struct pci_driver *driver)
0352 {
0353     if (!driver->err_handler->mmio_enabled)
0354         return PCI_ERS_RESULT_NONE;
0355     eeh_edev_info(edev, "Invoking %s->mmio_enabled()", driver->name);
0356     return driver->err_handler->mmio_enabled(pdev);
0357 }
0358
0359 /**
0360  * eeh_report_reset - Tell device that slot has been reset
0361  * @edev: eeh device
0362  * @driver: device's PCI driver
0363  *
0364  * This routine must be called while EEH tries to reset particular
0365  * PCI device so that the associated PCI device driver could take
0366  * some actions, usually to save data the driver needs so that the
0367  * driver can work again while the device is recovered.
0368  */
0369 static enum pci_ers_result eeh_report_reset(struct eeh_dev *edev,
0370                         struct pci_dev *pdev,
0371                         struct pci_driver *driver)
0372 {
0373     if (!driver->err_handler->slot_reset || !edev->in_error)
0374         return PCI_ERS_RESULT_NONE;
0375     eeh_edev_info(edev, "Invoking %s->slot_reset()", driver->name);
0376     return driver->err_handler->slot_reset(pdev);
0377 }
0378
0379 static void eeh_dev_restore_state(struct eeh_dev *edev, void *userdata)
0380 {
0381     struct pci_dev *pdev;
0382
0383     if (!edev)
0384         return;
0385
0386     /*
0387      * The content in the config space isn't saved because
0388      * the blocked config space on some adapters. We have
0389      * to restore the initial saved config space when the
0390      * EEH device is created.
0391      */
0392     if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) {
0393         if (list_is_last(&edev->entry, &edev->pe->edevs))
0394             eeh_pe_restore_bars(edev->pe);
0395
0396         return;
0397     }
0398
0399     pdev = eeh_dev_to_pci_dev(edev);
0400     if (!pdev)
0401         return;
0402
0403     pci_restore_state(pdev);
0404 }
0405
0406 /**
0407  * eeh_report_resume - Tell device to resume normal operations
0408  * @edev: eeh device
0409  * @driver: device's PCI driver
0410  *
0411  * This routine must be called to notify the device driver that it
0412  * could resume so that the device driver can do some initialization
0413  * to make the recovered device work again.
0414  */
0415 static enum pci_ers_result eeh_report_resume(struct eeh_dev *edev,
0416                          struct pci_dev *pdev,
0417                          struct pci_driver *driver)
0418 {
0419     if (!driver->err_handler->resume || !edev->in_error)
0420         return PCI_ERS_RESULT_NONE;
0421
0422     eeh_edev_info(edev, "Invoking %s->resume()", driver->name);
0423     driver->err_handler->resume(pdev);
0424
0425     pci_uevent_ers(edev->pdev, PCI_ERS_RESULT_RECOVERED);
0426 #ifdef CONFIG_PCI_IOV
0427     if (eeh_ops->notify_resume)
0428         eeh_ops->notify_resume(edev);
0429 #endif
0430     return PCI_ERS_RESULT_NONE;
0431 }
0432
0433 /**
0434  * eeh_report_failure - Tell device driver that device is dead.
0435  * @edev: eeh device
0436  * @driver: device's PCI driver
0437  *
0438  * This informs the device driver that the device is permanently
0439  * dead, and that no further recovery attempts will be made on it.
0440  */
0441 static enum pci_ers_result eeh_report_failure(struct eeh_dev *edev,
0442                           struct pci_dev *pdev,
0443                           struct pci_driver *driver)
0444 {
0445     enum pci_ers_result rc;
0446
0447     if (!driver->err_handler->error_detected)
0448         return PCI_ERS_RESULT_NONE;
0449
0450     eeh_edev_info(edev, "Invoking %s->error_detected(permanent failure)",
0451               driver->name);
0452     rc = driver->err_handler->error_detected(pdev,
0453                          pci_channel_io_perm_failure);
0454
0455     pci_uevent_ers(pdev, PCI_ERS_RESULT_DISCONNECT);
0456     return rc;
0457 }
0458
0459 static void *eeh_add_virt_device(struct eeh_dev *edev)
0460 {
0461     struct pci_driver *driver;
0462     struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
0463
0464     if (!(edev->physfn)) {
0465         eeh_edev_warn(edev, "Not for VF\n");
0466         return NULL;
0467     }
0468
0469     driver = eeh_pcid_get(dev);
0470     if (driver) {
0471         if (driver->err_handler) {
0472             eeh_pcid_put(dev);
0473             return NULL;
0474         }
0475         eeh_pcid_put(dev);
0476     }
0477
0478 #ifdef CONFIG_PCI_IOV
0479     pci_iov_add_virtfn(edev->physfn, edev->vf_index);
0480 #endif
0481     return NULL;
0482 }
0483
0484 static void eeh_rmv_device(struct eeh_dev *edev, void *userdata)
0485 {
0486     struct pci_driver *driver;
0487     struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
0488     struct eeh_rmv_data *rmv_data = (struct eeh_rmv_data *)userdata;
0489
0490     /*
0491      * Actually, we should remove the PCI bridges as well.
0492      * However, that's lots of complexity to do that,
0493      * particularly some of devices under the bridge might
0494      * support EEH. So we just care about PCI devices for
0495      * simplicity here.
0496      */
0497     if (!eeh_edev_actionable(edev) ||
0498         (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE))
0499         return;
0500
0501     if (rmv_data) {
0502         driver = eeh_pcid_get(dev);
0503         if (driver) {
0504             if (driver->err_handler &&
0505                 driver->err_handler->error_detected &&
0506                 driver->err_handler->slot_reset) {
0507                 eeh_pcid_put(dev);
0508                 return;
0509             }
0510             eeh_pcid_put(dev);
0511         }
0512     }
0513
0514     /* Remove it from PCI subsystem */
0515     pr_info("EEH: Removing %s without EEH sensitive driver\n",
0516         pci_name(dev));
0517     edev->mode |= EEH_DEV_DISCONNECTED;
0518     if (rmv_data)
0519         rmv_data->removed_dev_count++;
0520
0521     if (edev->physfn) {
0522 #ifdef CONFIG_PCI_IOV
0523         pci_iov_remove_virtfn(edev->physfn, edev->vf_index);
0524         edev->pdev = NULL;
0525 #endif
0526         if (rmv_data)
0527             list_add(&edev->rmv_entry, &rmv_data->removed_vf_list);
0528     } else {
0529         pci_lock_rescan_remove();
0530         pci_stop_and_remove_bus_device(dev);
0531         pci_unlock_rescan_remove();
0532     }
0533 }
0534
0535 static void *eeh_pe_detach_dev(struct eeh_pe *pe, void *userdata)
0536 {
0537     struct eeh_dev *edev, *tmp;
0538
0539     eeh_pe_for_each_dev(pe, edev, tmp) {
0540         if (!(edev->mode & EEH_DEV_DISCONNECTED))
0541             continue;
0542
0543         edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
0544         eeh_pe_tree_remove(edev);
0545     }
0546
0547     return NULL;
0548 }
0549
0550 /*
0551  * Explicitly clear PE's frozen state for PowerNV where
0552  * we have frozen PE until BAR restore is completed. It's
0553  * harmless to clear it for pSeries. To be consistent with
0554  * PE reset (for 3 times), we try to clear the frozen state
0555  * for 3 times as well.
0556  */
0557 static int eeh_clear_pe_frozen_state(struct eeh_pe *root, bool include_passed)
0558 {
0559     struct eeh_pe *pe;
0560     int i;
0561
0562     eeh_for_each_pe(root, pe) {
0563         if (include_passed || !eeh_pe_passed(pe)) {
0564             for (i = 0; i < 3; i++)
0565                 if (!eeh_unfreeze_pe(pe))
0566                     break;
0567             if (i >= 3)
0568                 return -EIO;
0569         }
0570     }
0571     eeh_pe_state_clear(root, EEH_PE_ISOLATED, include_passed);
0572     return 0;
0573 }
0574
0575 int eeh_pe_reset_and_recover(struct eeh_pe *pe)
0576 {
0577     int ret;
0578
0579     /* Bail if the PE is being recovered */
0580     if (pe->state & EEH_PE_RECOVERING)
0581         return 0;
0582
0583     /* Put the PE into recovery mode */
0584     eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
0585
0586     /* Save states */
0587     eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL);
0588
0589     /* Issue reset */
0590     ret = eeh_pe_reset_full(pe, true);
0591     if (ret) {
0592         eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
0593         return ret;
0594     }
0595
0596     /* Unfreeze the PE */
0597     ret = eeh_clear_pe_frozen_state(pe, true);
0598     if (ret) {
0599         eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
0600         return ret;
0601     }
0602
0603     /* Restore device state */
0604     eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL);
0605
0606     /* Clear recovery mode */
0607     eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
0608
0609     return 0;
0610 }
0611
0612 /**
0613  * eeh_reset_device - Perform actual reset of a pci slot
0614  * @driver_eeh_aware: Does the device's driver provide EEH support?
0615  * @pe: EEH PE
0616  * @bus: PCI bus corresponding to the isolcated slot
0617  * @rmv_data: Optional, list to record removed devices
0618  *
0619  * This routine must be called to do reset on the indicated PE.
0620  * During the reset, udev might be invoked because those affected
0621  * PCI devices will be removed and then added.
0622  */
0623 static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus,
0624                 struct eeh_rmv_data *rmv_data,
0625                 bool driver_eeh_aware)
0626 {
0627     time64_t tstamp;
0628     int cnt, rc;
0629     struct eeh_dev *edev;
0630     struct eeh_pe *tmp_pe;
0631     bool any_passed = false;
0632
0633     eeh_for_each_pe(pe, tmp_pe)
0634         any_passed |= eeh_pe_passed(tmp_pe);
0635
0636     /* pcibios will clear the counter; save the value */
0637     cnt = pe->freeze_count;
0638     tstamp = pe->tstamp;
0639
0640     /*
0641      * We don't remove the corresponding PE instances because
0642      * we need the information afterwords. The attached EEH
0643      * devices are expected to be attached soon when calling
0644      * into pci_hp_add_devices().
0645      */
0646     eeh_pe_state_mark(pe, EEH_PE_KEEP);
0647     if (any_passed || driver_eeh_aware || (pe->type & EEH_PE_VF)) {
0648         eeh_pe_dev_traverse(pe, eeh_rmv_device, rmv_data);
0649     } else {
0650         pci_lock_rescan_remove();
0651         pci_hp_remove_devices(bus);
0652         pci_unlock_rescan_remove();
0653     }
0654
0655     /*
0656      * Reset the pci controller. (Asserts RST#; resets config space).
0657      * Reconfigure bridges and devices. Don't try to bring the system
0658      * up if the reset failed for some reason.
0659      *
0660      * During the reset, it's very dangerous to have uncontrolled PCI
0661      * config accesses. So we prefer to block them. However, controlled
0662      * PCI config accesses initiated from EEH itself are allowed.
0663      */
0664     rc = eeh_pe_reset_full(pe, false);
0665     if (rc)
0666         return rc;
0667
0668     pci_lock_rescan_remove();
0669
0670     /* Restore PE */
0671     eeh_ops->configure_bridge(pe);
0672     eeh_pe_restore_bars(pe);
0673
0674     /* Clear frozen state */
0675     rc = eeh_clear_pe_frozen_state(pe, false);
0676     if (rc) {
0677         pci_unlock_rescan_remove();
0678         return rc;
0679     }
0680
0681     /* Give the system 5 seconds to finish running the user-space
0682      * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes,
0683      * this is a hack, but if we don't do this, and try to bring
0684      * the device up before the scripts have taken it down,
0685      * potentially weird things happen.
0686      */
0687     if (!driver_eeh_aware || rmv_data->removed_dev_count) {
0688         pr_info("EEH: Sleep 5s ahead of %s hotplug\n",
0689             (driver_eeh_aware ? "partial" : "complete"));
0690         ssleep(5);
0691
0692         /*
0693          * The EEH device is still connected with its parent
0694          * PE. We should disconnect it so the binding can be
0695          * rebuilt when adding PCI devices.
0696          */
0697         edev = list_first_entry(&pe->edevs, struct eeh_dev, entry);
0698         eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
0699         if (pe->type & EEH_PE_VF) {
0700             eeh_add_virt_device(edev);
0701         } else {
0702             if (!driver_eeh_aware)
0703                 eeh_pe_state_clear(pe, EEH_PE_PRI_BUS, true);
0704             pci_hp_add_devices(bus);
0705         }
0706     }
0707     eeh_pe_state_clear(pe, EEH_PE_KEEP, true);
0708
0709     pe->tstamp = tstamp;
0710     pe->freeze_count = cnt;
0711
0712     pci_unlock_rescan_remove();
0713     return 0;
0714 }
0715
0716 /* The longest amount of time to wait for a pci device
0717  * to come back on line, in seconds.
0718  */
0719 #define MAX_WAIT_FOR_RECOVERY 300
0720
0721
0722 /* Walks the PE tree after processing an event to remove any stale PEs.
0723  *
0724  * NB: This needs to be recursive to ensure the leaf PEs get removed
0725  * before their parents do. Although this is possible to do recursively
0726  * we don't since this is easier to read and we need to garantee
0727  * the leaf nodes will be handled first.
0728  */
0729 static void eeh_pe_cleanup(struct eeh_pe *pe)
0730 {
0731     struct eeh_pe *child_pe, *tmp;
0732
0733     list_for_each_entry_safe(child_pe, tmp, &pe->child_list, child)
0734         eeh_pe_cleanup(child_pe);
0735
0736     if (pe->state & EEH_PE_KEEP)
0737         return;
0738
0739     if (!(pe->state & EEH_PE_INVALID))
0740         return;
0741
0742     if (list_empty(&pe->edevs) && list_empty(&pe->child_list)) {
0743         list_del(&pe->child);
0744         kfree(pe);
0745     }
0746 }
0747
0748 /**
0749  * eeh_check_slot_presence - Check if a device is still present in a slot
0750  * @pdev: pci_dev to check
0751  *
0752  * This function may return a false positive if we can't determine the slot's
0753  * presence state. This might happen for PCIe slots if the PE containing
0754  * the upstream bridge is also frozen, or the bridge is part of the same PE
0755  * as the device.
0756  *
0757  * This shouldn't happen often, but you might see it if you hotplug a PCIe
0758  * switch.
0759  */
0760 static bool eeh_slot_presence_check(struct pci_dev *pdev)
0761 {
0762     const struct hotplug_slot_ops *ops;
0763     struct pci_slot *slot;
0764     u8 state;
0765     int rc;
0766
0767     if (!pdev)
0768         return false;
0769
0770     if (pdev->error_state == pci_channel_io_perm_failure)
0771         return false;
0772
0773     slot = pdev->slot;
0774     if (!slot || !slot->hotplug)
0775         return true;
0776
0777     ops = slot->hotplug->ops;
0778     if (!ops || !ops->get_adapter_status)
0779         return true;
0780
0781     /* set the attention indicator while we've got the slot ops */
0782     if (ops->set_attention_status)
0783         ops->set_attention_status(slot->hotplug, 1);
0784
0785     rc = ops->get_adapter_status(slot->hotplug, &state);
0786     if (rc)
0787         return true;
0788
0789     return !!state;
0790 }
0791
0792 static void eeh_clear_slot_attention(struct pci_dev *pdev)
0793 {
0794     const struct hotplug_slot_ops *ops;
0795     struct pci_slot *slot;
0796
0797     if (!pdev)
0798         return;
0799
0800     if (pdev->error_state == pci_channel_io_perm_failure)
0801         return;
0802
0803     slot = pdev->slot;
0804     if (!slot || !slot->hotplug)
0805         return;
0806
0807     ops = slot->hotplug->ops;
0808     if (!ops || !ops->set_attention_status)
0809         return;
0810
0811     ops->set_attention_status(slot->hotplug, 0);
0812 }
0813
0814 /**
0815  * eeh_handle_normal_event - Handle EEH events on a specific PE
0816  * @pe: EEH PE - which should not be used after we return, as it may
0817  * have been invalidated.
0818  *
0819  * Attempts to recover the given PE.  If recovery fails or the PE has failed
0820  * too many times, remove the PE.
0821  *
0822  * While PHB detects address or data parity errors on particular PCI
0823  * slot, the associated PE will be frozen. Besides, DMA's occurring
0824  * to wild addresses (which usually happen due to bugs in device
0825  * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
0826  * #PERR or other misc PCI-related errors also can trigger EEH errors.
0827  *
0828  * Recovery process consists of unplugging the device driver (which
0829  * generated hotplug events to userspace), then issuing a PCI #RST to
0830  * the device, then reconfiguring the PCI config space for all bridges
0831  * & devices under this slot, and then finally restarting the device
0832  * drivers (which cause a second set of hotplug events to go out to
0833  * userspace).
0834  */
0835 void eeh_handle_normal_event(struct eeh_pe *pe)
0836 {
0837     struct pci_bus *bus;
0838     struct eeh_dev *edev, *tmp;
0839     struct eeh_pe *tmp_pe;
0840     int rc = 0;
0841     enum pci_ers_result result = PCI_ERS_RESULT_NONE;
0842     struct eeh_rmv_data rmv_data =
0843         {LIST_HEAD_INIT(rmv_data.removed_vf_list), 0};
0844     int devices = 0;
0845
0846     bus = eeh_pe_bus_get(pe);
0847     if (!bus) {
0848         pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
0849             __func__, pe->phb->global_number, pe->addr);
0850         return;
0851     }
0852
0853     /*
0854      * When devices are hot-removed we might get an EEH due to
0855      * a driver attempting to touch the MMIO space of a removed
0856      * device. In this case we don't have a device to recover
0857      * so suppress the event if we can't find any present devices.
0858      *
0859      * The hotplug driver should take care of tearing down the
0860      * device itself.
0861      */
0862     eeh_for_each_pe(pe, tmp_pe)
0863         eeh_pe_for_each_dev(tmp_pe, edev, tmp)
0864             if (eeh_slot_presence_check(edev->pdev))
0865                 devices++;
0866
0867     if (!devices) {
0868         pr_debug("EEH: Frozen PHB#%x-PE#%x is empty!\n",
0869             pe->phb->global_number, pe->addr);
0870         goto out; /* nothing to recover */
0871     }
0872
0873     /* Log the event */
0874     if (pe->type & EEH_PE_PHB) {
0875         pr_err("EEH: Recovering PHB#%x, location: %s\n",
0876             pe->phb->global_number, eeh_pe_loc_get(pe));
0877     } else {
0878         struct eeh_pe *phb_pe = eeh_phb_pe_get(pe->phb);
0879
0880         pr_err("EEH: Recovering PHB#%x-PE#%x\n",
0881                pe->phb->global_number, pe->addr);
0882         pr_err("EEH: PE location: %s, PHB location: %s\n",
0883                eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
0884     }
0885
0886 #ifdef CONFIG_STACKTRACE
0887     /*
0888      * Print the saved stack trace now that we've verified there's
0889      * something to recover.
0890      */
0891     if (pe->trace_entries) {
0892         void **ptrs = (void **) pe->stack_trace;
0893         int i;
0894
0895         pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
0896                pe->phb->global_number, pe->addr);
0897
0898         /* FIXME: Use the same format as dump_stack() */
0899         pr_err("EEH: Call Trace:\n");
0900         for (i = 0; i < pe->trace_entries; i++)
0901             pr_err("EEH: [%pK] %pS\n", ptrs[i], ptrs[i]);
0902
0903         pe->trace_entries = 0;
0904     }
0905 #endif /* CONFIG_STACKTRACE */
0906
0907     eeh_for_each_pe(pe, tmp_pe)
0908         eeh_pe_for_each_dev(tmp_pe, edev, tmp)
0909             edev->mode &= ~EEH_DEV_NO_HANDLER;
0910
0911     eeh_pe_update_time_stamp(pe);
0912     pe->freeze_count++;
0913     if (pe->freeze_count > eeh_max_freezes) {
0914         pr_err("EEH: PHB#%x-PE#%x has failed %d times in the last hour and has been permanently disabled.\n",
0915                pe->phb->global_number, pe->addr,
0916                pe->freeze_count);
0917
0918         goto recover_failed;
0919     }
0920
0921     /* Walk the various device drivers attached to this slot through
0922      * a reset sequence, giving each an opportunity to do what it needs
0923      * to accomplish the reset.  Each child gets a report of the
0924      * status ... if any child can't handle the reset, then the entire
0925      * slot is dlpar removed and added.
0926      *
0927      * When the PHB is fenced, we have to issue a reset to recover from
0928      * the error. Override the result if necessary to have partially
0929      * hotplug for this case.
0930      */
0931     pr_warn("EEH: This PCI device has failed %d times in the last hour and will be permanently disabled after %d failures.\n",
0932         pe->freeze_count, eeh_max_freezes);
0933     pr_info("EEH: Notify device drivers to shutdown\n");
0934     eeh_set_channel_state(pe, pci_channel_io_frozen);
0935     eeh_set_irq_state(pe, false);
0936     eeh_pe_report("error_detected(IO frozen)", pe,
0937               eeh_report_error, &result);
0938     if (result == PCI_ERS_RESULT_DISCONNECT)
0939         goto recover_failed;
0940
0941     /*
0942      * Error logged on a PHB are always fences which need a full
0943      * PHB reset to clear so force that to happen.
0944      */
0945     if ((pe->type & EEH_PE_PHB) && result != PCI_ERS_RESULT_NONE)
0946         result = PCI_ERS_RESULT_NEED_RESET;
0947
0948     /* Get the current PCI slot state. This can take a long time,
0949      * sometimes over 300 seconds for certain systems.
0950      */
0951     rc = eeh_wait_state(pe, MAX_WAIT_FOR_RECOVERY * 1000);
0952     if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
0953         pr_warn("EEH: Permanent failure\n");
0954         goto recover_failed;
0955     }
0956
0957     /* Since rtas may enable MMIO when posting the error log,
0958      * don't post the error log until after all dev drivers
0959      * have been informed.
0960      */
0961     pr_info("EEH: Collect temporary log\n");
0962     eeh_slot_error_detail(pe, EEH_LOG_TEMP);
0963
0964     /* If all device drivers were EEH-unaware, then shut
0965      * down all of the device drivers, and hope they
0966      * go down willingly, without panicing the system.
0967      */
0968     if (result == PCI_ERS_RESULT_NONE) {
0969         pr_info("EEH: Reset with hotplug activity\n");
0970         rc = eeh_reset_device(pe, bus, NULL, false);
0971         if (rc) {
0972             pr_warn("%s: Unable to reset, err=%d\n", __func__, rc);
0973             goto recover_failed;
0974         }
0975     }
0976
0977     /* If all devices reported they can proceed, then re-enable MMIO */
0978     if (result == PCI_ERS_RESULT_CAN_RECOVER) {
0979         pr_info("EEH: Enable I/O for affected devices\n");
0980         rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
0981         if (rc < 0)
0982             goto recover_failed;
0983
0984         if (rc) {
0985             result = PCI_ERS_RESULT_NEED_RESET;
0986         } else {
0987             pr_info("EEH: Notify device drivers to resume I/O\n");
0988             eeh_pe_report("mmio_enabled", pe,
0989                       eeh_report_mmio_enabled, &result);
0990         }
0991     }
0992     if (result == PCI_ERS_RESULT_CAN_RECOVER) {
0993         pr_info("EEH: Enabled DMA for affected devices\n");
0994         rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
0995         if (rc < 0)
0996             goto recover_failed;
0997
0998         if (rc) {
0999             result = PCI_ERS_RESULT_NEED_RESET;
1000         } else {
1001             /*
1002              * We didn't do PE reset for the case. The PE
1003              * is still in frozen state. Clear it before
1004              * resuming the PE.
1005              */
1006             eeh_pe_state_clear(pe, EEH_PE_ISOLATED, true);
1007             result = PCI_ERS_RESULT_RECOVERED;
1008         }
1009     }
1010
1011     /* If any device called out for a reset, then reset the slot */
1012     if (result == PCI_ERS_RESULT_NEED_RESET) {
1013         pr_info("EEH: Reset without hotplug activity\n");
1014         rc = eeh_reset_device(pe, bus, &rmv_data, true);
1015         if (rc) {
1016             pr_warn("%s: Cannot reset, err=%d\n", __func__, rc);
1017             goto recover_failed;
1018         }
1019
1020         result = PCI_ERS_RESULT_NONE;
1021         eeh_set_channel_state(pe, pci_channel_io_normal);
1022         eeh_set_irq_state(pe, true);
1023         eeh_pe_report("slot_reset", pe, eeh_report_reset,
1024                   &result);
1025     }
1026
1027     if ((result == PCI_ERS_RESULT_RECOVERED) ||
1028         (result == PCI_ERS_RESULT_NONE)) {
1029         /*
1030          * For those hot removed VFs, we should add back them after PF
1031          * get recovered properly.
1032          */
1033         list_for_each_entry_safe(edev, tmp, &rmv_data.removed_vf_list,
1034                      rmv_entry) {
1035             eeh_add_virt_device(edev);
1036             list_del(&edev->rmv_entry);
1037         }
1038
1039         /* Tell all device drivers that they can resume operations */
1040         pr_info("EEH: Notify device driver to resume\n");
1041         eeh_set_channel_state(pe, pci_channel_io_normal);
1042         eeh_set_irq_state(pe, true);
1043         eeh_pe_report("resume", pe, eeh_report_resume, NULL);
1044         eeh_for_each_pe(pe, tmp_pe) {
1045             eeh_pe_for_each_dev(tmp_pe, edev, tmp) {
1046                 edev->mode &= ~EEH_DEV_NO_HANDLER;
1047                 edev->in_error = false;
1048             }
1049         }
1050
1051         pr_info("EEH: Recovery successful.\n");
1052         goto out;
1053     }
1054
1055 recover_failed:
1056     /*
1057      * About 90% of all real-life EEH failures in the field
1058      * are due to poorly seated PCI cards. Only 10% or so are
1059      * due to actual, failed cards.
1060      */
1061     pr_err("EEH: Unable to recover from failure from PHB#%x-PE#%x.\n"
1062         "Please try reseating or replacing it\n",
1063         pe->phb->global_number, pe->addr);
1064
1065     eeh_slot_error_detail(pe, EEH_LOG_PERM);
1066
1067     /* Notify all devices that they're about to go down. */
1068     eeh_set_channel_state(pe, pci_channel_io_perm_failure);
1069     eeh_set_irq_state(pe, false);
1070     eeh_pe_report("error_detected(permanent failure)", pe,
1071               eeh_report_failure, NULL);
1072
1073     /* Mark the PE to be removed permanently */
1074     eeh_pe_state_mark(pe, EEH_PE_REMOVED);
1075
1076     /*
1077      * Shut down the device drivers for good. We mark
1078      * all removed devices correctly to avoid access
1079      * the their PCI config any more.
1080      */
1081     if (pe->type & EEH_PE_VF) {
1082         eeh_pe_dev_traverse(pe, eeh_rmv_device, NULL);
1083         eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
1084     } else {
1085         eeh_pe_state_clear(pe, EEH_PE_PRI_BUS, true);
1086         eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
1087
1088         pci_lock_rescan_remove();
1089         pci_hp_remove_devices(bus);
1090         pci_unlock_rescan_remove();
1091         /* The passed PE should no longer be used */
1092         return;
1093     }
1094
1095 out:
1096     /*
1097      * Clean up any PEs without devices. While marked as EEH_PE_RECOVERYING
1098      * we don't want to modify the PE tree structure so we do it here.
1099      */
1100     eeh_pe_cleanup(pe);
1101
1102     /* clear the slot attention LED for all recovered devices */
1103     eeh_for_each_pe(pe, tmp_pe)
1104         eeh_pe_for_each_dev(tmp_pe, edev, tmp)
1105             eeh_clear_slot_attention(edev->pdev);
1106
1107     eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
1108 }
1109
1110 /**
1111  * eeh_handle_special_event - Handle EEH events without a specific failing PE
1112  *
1113  * Called when an EEH event is detected but can't be narrowed down to a
1114  * specific PE.  Iterates through possible failures and handles them as
1115  * necessary.
1116  */
1117 void eeh_handle_special_event(void)
1118 {
1119     struct eeh_pe *pe, *phb_pe, *tmp_pe;
1120     struct eeh_dev *edev, *tmp_edev;
1121     struct pci_bus *bus;
1122     struct pci_controller *hose;
1123     unsigned long flags;
1124     int rc;
1125
1126
1127     do {
1128         rc = eeh_ops->next_error(&pe);
1129
1130         switch (rc) {
1131         case EEH_NEXT_ERR_DEAD_IOC:
1132             /* Mark all PHBs in dead state */
1133             eeh_serialize_lock(&flags);
1134
1135             /* Purge all events */
1136             eeh_remove_event(NULL, true);
1137
1138             list_for_each_entry(hose, &hose_list, list_node) {
1139                 phb_pe = eeh_phb_pe_get(hose);
1140                 if (!phb_pe) continue;
1141
1142                 eeh_pe_mark_isolated(phb_pe);
1143             }
1144
1145             eeh_serialize_unlock(flags);
1146
1147             break;
1148         case EEH_NEXT_ERR_FROZEN_PE:
1149         case EEH_NEXT_ERR_FENCED_PHB:
1150         case EEH_NEXT_ERR_DEAD_PHB:
1151             /* Mark the PE in fenced state */
1152             eeh_serialize_lock(&flags);
1153
1154             /* Purge all events of the PHB */
1155             eeh_remove_event(pe, true);
1156
1157             if (rc != EEH_NEXT_ERR_DEAD_PHB)
1158                 eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
1159             eeh_pe_mark_isolated(pe);
1160
1161             eeh_serialize_unlock(flags);
1162
1163             break;
1164         case EEH_NEXT_ERR_NONE:
1165             return;
1166         default:
1167             pr_warn("%s: Invalid value %d from next_error()\n",
1168                 __func__, rc);
1169             return;
1170         }
1171
1172         /*
1173          * For fenced PHB and frozen PE, it's handled as normal
1174          * event. We have to remove the affected PHBs for dead
1175          * PHB and IOC
1176          */
1177         if (rc == EEH_NEXT_ERR_FROZEN_PE ||
1178             rc == EEH_NEXT_ERR_FENCED_PHB) {
1179             eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
1180             eeh_handle_normal_event(pe);
1181         } else {
1182             eeh_for_each_pe(pe, tmp_pe)
1183                 eeh_pe_for_each_dev(tmp_pe, edev, tmp_edev)
1184                     edev->mode &= ~EEH_DEV_NO_HANDLER;
1185
1186             /* Notify all devices to be down */
1187             eeh_pe_state_clear(pe, EEH_PE_PRI_BUS, true);
1188             eeh_set_channel_state(pe, pci_channel_io_perm_failure);
1189             eeh_pe_report(
1190                 "error_detected(permanent failure)", pe,
1191                 eeh_report_failure, NULL);
1192
1193             pci_lock_rescan_remove();
1194             list_for_each_entry(hose, &hose_list, list_node) {
1195                 phb_pe = eeh_phb_pe_get(hose);
1196                 if (!phb_pe ||
1197                     !(phb_pe->state & EEH_PE_ISOLATED) ||
1198                     (phb_pe->state & EEH_PE_RECOVERING))
1199                     continue;
1200
1201                 bus = eeh_pe_bus_get(phb_pe);
1202                 if (!bus) {
1203                     pr_err("%s: Cannot find PCI bus for "
1204                            "PHB#%x-PE#%x\n",
1205                            __func__,
1206                            pe->phb->global_number,
1207                            pe->addr);
1208                     break;
1209                 }
1210                 pci_hp_remove_devices(bus);
1211             }
1212             pci_unlock_rescan_remove();
1213         }
1214
1215         /*
1216          * If we have detected dead IOC, we needn't proceed
1217          * any more since all PHBs would have been removed
1218          */
1219         if (rc == EEH_NEXT_ERR_DEAD_IOC)
1220             break;
1221     } while (rc != EEH_NEXT_ERR_NONE);
1222 }