Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * PCI Error Recovery Driver for RPA-compliant PPC64 platform.
   4 * Copyright IBM Corp. 2004 2005
   5 * Copyright Linas Vepstas <linas@linas.org> 2004, 2005
   6 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   7 * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
   8 */
   9#include <linux/delay.h>
  10#include <linux/interrupt.h>
  11#include <linux/irq.h>
  12#include <linux/module.h>
  13#include <linux/pci.h>
  14#include <linux/pci_hotplug.h>
  15#include <asm/eeh.h>
  16#include <asm/eeh_event.h>
  17#include <asm/ppc-pci.h>
  18#include <asm/pci-bridge.h>
  19#include <asm/prom.h>
  20#include <asm/rtas.h>
  21
  22struct eeh_rmv_data {
  23	struct list_head removed_vf_list;
  24	int removed_dev_count;
  25};
  26
  27static int eeh_result_priority(enum pci_ers_result result)
  28{
  29	switch (result) {
  30	case PCI_ERS_RESULT_NONE:
  31		return 1;
  32	case PCI_ERS_RESULT_NO_AER_DRIVER:
  33		return 2;
  34	case PCI_ERS_RESULT_RECOVERED:
  35		return 3;
  36	case PCI_ERS_RESULT_CAN_RECOVER:
  37		return 4;
  38	case PCI_ERS_RESULT_DISCONNECT:
  39		return 5;
  40	case PCI_ERS_RESULT_NEED_RESET:
  41		return 6;
  42	default:
  43		WARN_ONCE(1, "Unknown pci_ers_result value: %d\n", (int)result);
  44		return 0;
  45	}
  46};
  47
  48static const char *pci_ers_result_name(enum pci_ers_result result)
  49{
  50	switch (result) {
  51	case PCI_ERS_RESULT_NONE:
  52		return "none";
  53	case PCI_ERS_RESULT_CAN_RECOVER:
  54		return "can recover";
  55	case PCI_ERS_RESULT_NEED_RESET:
  56		return "need reset";
  57	case PCI_ERS_RESULT_DISCONNECT:
  58		return "disconnect";
  59	case PCI_ERS_RESULT_RECOVERED:
  60		return "recovered";
  61	case PCI_ERS_RESULT_NO_AER_DRIVER:
  62		return "no AER driver";
  63	default:
  64		WARN_ONCE(1, "Unknown result type: %d\n", (int)result);
  65		return "unknown";
  66	}
  67};
  68
  69static enum pci_ers_result pci_ers_merge_result(enum pci_ers_result old,
  70						enum pci_ers_result new)
  71{
  72	if (eeh_result_priority(new) > eeh_result_priority(old))
  73		return new;
  74	return old;
  75}
  76
  77static bool eeh_dev_removed(struct eeh_dev *edev)
  78{
  79	return !edev || (edev->mode & EEH_DEV_REMOVED);
  80}
  81
  82static bool eeh_edev_actionable(struct eeh_dev *edev)
  83{
  84	if (!edev->pdev)
  85		return false;
  86	if (edev->pdev->error_state == pci_channel_io_perm_failure)
  87		return false;
  88	if (eeh_dev_removed(edev))
  89		return false;
  90	if (eeh_pe_passed(edev->pe))
  91		return false;
  92
  93	return true;
  94}
  95
  96/**
  97 * eeh_pcid_get - Get the PCI device driver
  98 * @pdev: PCI device
  99 *
 100 * The function is used to retrieve the PCI device driver for
 101 * the indicated PCI device. Besides, we will increase the reference
 102 * of the PCI device driver to prevent that being unloaded on
 103 * the fly. Otherwise, kernel crash would be seen.
 104 */
 105static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
 106{
 107	if (!pdev || !pdev->driver)
 108		return NULL;
 109
 110	if (!try_module_get(pdev->driver->driver.owner))
 111		return NULL;
 112
 113	return pdev->driver;
 114}
 115
 116/**
 117 * eeh_pcid_put - Dereference on the PCI device driver
 118 * @pdev: PCI device
 119 *
 120 * The function is called to do dereference on the PCI device
 121 * driver of the indicated PCI device.
 122 */
 123static inline void eeh_pcid_put(struct pci_dev *pdev)
 124{
 125	if (!pdev || !pdev->driver)
 126		return;
 127
 128	module_put(pdev->driver->driver.owner);
 129}
 130
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 131/**
 132 * eeh_disable_irq - Disable interrupt for the recovering device
 133 * @dev: PCI device
 134 *
 135 * This routine must be called when reporting temporary or permanent
 136 * error to the particular PCI device to disable interrupt of that
 137 * device. If the device has enabled MSI or MSI-X interrupt, we needn't
 138 * do real work because EEH should freeze DMA transfers for those PCI
 139 * devices encountering EEH errors, which includes MSI or MSI-X.
 140 */
 141static void eeh_disable_irq(struct eeh_dev *edev)
 142{
 
 
 143	/* Don't disable MSI and MSI-X interrupts. They are
 144	 * effectively disabled by the DMA Stopped state
 145	 * when an EEH error occurs.
 146	 */
 147	if (edev->pdev->msi_enabled || edev->pdev->msix_enabled)
 148		return;
 149
 150	if (!irq_has_action(edev->pdev->irq))
 151		return;
 152
 153	edev->mode |= EEH_DEV_IRQ_DISABLED;
 154	disable_irq_nosync(edev->pdev->irq);
 155}
 156
 157/**
 158 * eeh_enable_irq - Enable interrupt for the recovering device
 159 * @dev: PCI device
 160 *
 161 * This routine must be called to enable interrupt while failed
 162 * device could be resumed.
 163 */
 164static void eeh_enable_irq(struct eeh_dev *edev)
 165{
 
 
 166	if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
 167		edev->mode &= ~EEH_DEV_IRQ_DISABLED;
 168		/*
 169		 * FIXME !!!!!
 170		 *
 171		 * This is just ass backwards. This maze has
 172		 * unbalanced irq_enable/disable calls. So instead of
 173		 * finding the root cause it works around the warning
 174		 * in the irq_enable code by conditionally calling
 175		 * into it.
 176		 *
 177		 * That's just wrong.The warning in the core code is
 178		 * there to tell people to fix their asymmetries in
 179		 * their own code, not by abusing the core information
 180		 * to avoid it.
 181		 *
 182		 * I so wish that the assymetry would be the other way
 183		 * round and a few more irq_disable calls render that
 184		 * shit unusable forever.
 185		 *
 186		 *	tglx
 187		 */
 188		if (irqd_irq_disabled(irq_get_irq_data(edev->pdev->irq)))
 189			enable_irq(edev->pdev->irq);
 190	}
 191}
 192
 193static void eeh_dev_save_state(struct eeh_dev *edev, void *userdata)
 
 
 
 
 
 
 
 
 
 194{
 195	struct pci_dev *pdev;
 196
 197	if (!edev)
 198		return;
 199
 200	/*
 201	 * We cannot access the config space on some adapters.
 202	 * Otherwise, it will cause fenced PHB. We don't save
 203	 * the content in their config space and will restore
 204	 * from the initial config space saved when the EEH
 205	 * device is created.
 206	 */
 207	if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED))
 208		return;
 209
 210	pdev = eeh_dev_to_pci_dev(edev);
 211	if (!pdev)
 212		return;
 213
 214	pci_save_state(pdev);
 215}
 216
 217static void eeh_set_channel_state(struct eeh_pe *root, pci_channel_state_t s)
 218{
 219	struct eeh_pe *pe;
 220	struct eeh_dev *edev, *tmp;
 221
 222	eeh_for_each_pe(root, pe)
 223		eeh_pe_for_each_dev(pe, edev, tmp)
 224			if (eeh_edev_actionable(edev))
 225				edev->pdev->error_state = s;
 226}
 227
 228static void eeh_set_irq_state(struct eeh_pe *root, bool enable)
 229{
 230	struct eeh_pe *pe;
 231	struct eeh_dev *edev, *tmp;
 232
 233	eeh_for_each_pe(root, pe) {
 234		eeh_pe_for_each_dev(pe, edev, tmp) {
 235			if (!eeh_edev_actionable(edev))
 236				continue;
 237
 238			if (!eeh_pcid_get(edev->pdev))
 239				continue;
 240
 241			if (enable)
 242				eeh_enable_irq(edev);
 243			else
 244				eeh_disable_irq(edev);
 245
 246			eeh_pcid_put(edev->pdev);
 247		}
 
 
 248	}
 249}
 250
 251typedef enum pci_ers_result (*eeh_report_fn)(struct eeh_dev *,
 252					     struct pci_dev *,
 253					     struct pci_driver *);
 254static void eeh_pe_report_edev(struct eeh_dev *edev, eeh_report_fn fn,
 255			       enum pci_ers_result *result)
 256{
 257	struct pci_dev *pdev;
 258	struct pci_driver *driver;
 259	enum pci_ers_result new_result;
 260
 261	pci_lock_rescan_remove();
 262	pdev = edev->pdev;
 263	if (pdev)
 264		get_device(&pdev->dev);
 265	pci_unlock_rescan_remove();
 266	if (!pdev) {
 267		eeh_edev_info(edev, "no device");
 268		return;
 269	}
 270	device_lock(&pdev->dev);
 271	if (eeh_edev_actionable(edev)) {
 272		driver = eeh_pcid_get(pdev);
 273
 274		if (!driver)
 275			eeh_edev_info(edev, "no driver");
 276		else if (!driver->err_handler)
 277			eeh_edev_info(edev, "driver not EEH aware");
 278		else if (edev->mode & EEH_DEV_NO_HANDLER)
 279			eeh_edev_info(edev, "driver bound too late");
 280		else {
 281			new_result = fn(edev, pdev, driver);
 282			eeh_edev_info(edev, "%s driver reports: '%s'",
 283				      driver->name,
 284				      pci_ers_result_name(new_result));
 285			if (result)
 286				*result = pci_ers_merge_result(*result,
 287							       new_result);
 288		}
 289		if (driver)
 290			eeh_pcid_put(pdev);
 291	} else {
 292		eeh_edev_info(edev, "not actionable (%d,%d,%d)", !!pdev,
 293			      !eeh_dev_removed(edev), !eeh_pe_passed(edev->pe));
 294	}
 295	device_unlock(&pdev->dev);
 296	if (edev->pdev != pdev)
 297		eeh_edev_warn(edev, "Device changed during processing!\n");
 298	put_device(&pdev->dev);
 299}
 300
 301static void eeh_pe_report(const char *name, struct eeh_pe *root,
 302			  eeh_report_fn fn, enum pci_ers_result *result)
 303{
 304	struct eeh_pe *pe;
 305	struct eeh_dev *edev, *tmp;
 306
 307	pr_info("EEH: Beginning: '%s'\n", name);
 308	eeh_for_each_pe(root, pe) eeh_pe_for_each_dev(pe, edev, tmp)
 309		eeh_pe_report_edev(edev, fn, result);
 310	if (result)
 311		pr_info("EEH: Finished:'%s' with aggregate recovery state:'%s'\n",
 312			name, pci_ers_result_name(*result));
 313	else
 314		pr_info("EEH: Finished:'%s'", name);
 315}
 316
 317/**
 318 * eeh_report_error - Report pci error to each device driver
 319 * @edev: eeh device
 320 * @driver: device's PCI driver
 321 *
 322 * Report an EEH error to each device driver.
 
 
 323 */
 324static enum pci_ers_result eeh_report_error(struct eeh_dev *edev,
 325					    struct pci_dev *pdev,
 326					    struct pci_driver *driver)
 327{
 328	enum pci_ers_result rc;
 
 
 
 329
 330	if (!driver->err_handler->error_detected)
 331		return PCI_ERS_RESULT_NONE;
 332
 333	eeh_edev_info(edev, "Invoking %s->error_detected(IO frozen)",
 334		      driver->name);
 335	rc = driver->err_handler->error_detected(pdev, pci_channel_io_frozen);
 336
 337	edev->in_error = true;
 338	pci_uevent_ers(pdev, PCI_ERS_RESULT_NONE);
 339	return rc;
 340}
 341
 342/**
 343 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
 344 * @edev: eeh device
 345 * @driver: device's PCI driver
 346 *
 347 * Tells each device driver that IO ports, MMIO and config space I/O
 348 * are now enabled.
 349 */
 350static enum pci_ers_result eeh_report_mmio_enabled(struct eeh_dev *edev,
 351						   struct pci_dev *pdev,
 352						   struct pci_driver *driver)
 353{
 354	if (!driver->err_handler->mmio_enabled)
 355		return PCI_ERS_RESULT_NONE;
 356	eeh_edev_info(edev, "Invoking %s->mmio_enabled()", driver->name);
 357	return driver->err_handler->mmio_enabled(pdev);
 358}
 359
 360/**
 361 * eeh_report_reset - Tell device that slot has been reset
 362 * @edev: eeh device
 363 * @driver: device's PCI driver
 364 *
 365 * This routine must be called while EEH tries to reset particular
 366 * PCI device so that the associated PCI device driver could take
 367 * some actions, usually to save data the driver needs so that the
 368 * driver can work again while the device is recovered.
 369 */
 370static enum pci_ers_result eeh_report_reset(struct eeh_dev *edev,
 371					    struct pci_dev *pdev,
 372					    struct pci_driver *driver)
 373{
 374	if (!driver->err_handler->slot_reset || !edev->in_error)
 375		return PCI_ERS_RESULT_NONE;
 376	eeh_edev_info(edev, "Invoking %s->slot_reset()", driver->name);
 377	return driver->err_handler->slot_reset(pdev);
 378}
 379
 380static void eeh_dev_restore_state(struct eeh_dev *edev, void *userdata)
 381{
 382	struct pci_dev *pdev;
 383
 384	if (!edev)
 385		return;
 386
 387	/*
 388	 * The content in the config space isn't saved because
 389	 * the blocked config space on some adapters. We have
 390	 * to restore the initial saved config space when the
 391	 * EEH device is created.
 392	 */
 393	if (edev->pe && (edev->pe->state & EEH_PE_CFG_RESTRICTED)) {
 394		if (list_is_last(&edev->entry, &edev->pe->edevs))
 395			eeh_pe_restore_bars(edev->pe);
 396
 397		return;
 
 
 
 
 398	}
 399
 400	pdev = eeh_dev_to_pci_dev(edev);
 401	if (!pdev)
 402		return;
 
 
 403
 404	pci_restore_state(pdev);
 
 405}
 406
 407/**
 408 * eeh_report_resume - Tell device to resume normal operations
 409 * @edev: eeh device
 410 * @driver: device's PCI driver
 411 *
 412 * This routine must be called to notify the device driver that it
 413 * could resume so that the device driver can do some initialization
 414 * to make the recovered device work again.
 415 */
 416static enum pci_ers_result eeh_report_resume(struct eeh_dev *edev,
 417					     struct pci_dev *pdev,
 418					     struct pci_driver *driver)
 419{
 420	if (!driver->err_handler->resume || !edev->in_error)
 421		return PCI_ERS_RESULT_NONE;
 
 422
 423	eeh_edev_info(edev, "Invoking %s->resume()", driver->name);
 424	driver->err_handler->resume(pdev);
 425
 426	pci_uevent_ers(edev->pdev, PCI_ERS_RESULT_RECOVERED);
 427#ifdef CONFIG_PCI_IOV
 428	if (eeh_ops->notify_resume)
 429		eeh_ops->notify_resume(edev);
 430#endif
 431	return PCI_ERS_RESULT_NONE;
 
 
 
 
 
 
 
 
 
 
 
 432}
 433
 434/**
 435 * eeh_report_failure - Tell device driver that device is dead.
 436 * @edev: eeh device
 437 * @driver: device's PCI driver
 438 *
 439 * This informs the device driver that the device is permanently
 440 * dead, and that no further recovery attempts will be made on it.
 441 */
 442static enum pci_ers_result eeh_report_failure(struct eeh_dev *edev,
 443					      struct pci_dev *pdev,
 444					      struct pci_driver *driver)
 445{
 446	enum pci_ers_result rc;
 447
 448	if (!driver->err_handler->error_detected)
 449		return PCI_ERS_RESULT_NONE;
 450
 451	eeh_edev_info(edev, "Invoking %s->error_detected(permanent failure)",
 452		      driver->name);
 453	rc = driver->err_handler->error_detected(pdev,
 454						 pci_channel_io_perm_failure);
 455
 456	pci_uevent_ers(pdev, PCI_ERS_RESULT_DISCONNECT);
 457	return rc;
 458}
 459
 460static void *eeh_add_virt_device(struct eeh_dev *edev)
 461{
 462	struct pci_driver *driver;
 463	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 464
 465	if (!(edev->physfn)) {
 466		eeh_edev_warn(edev, "Not for VF\n");
 
 467		return NULL;
 468	}
 469
 470	driver = eeh_pcid_get(dev);
 471	if (driver) {
 472		if (driver->err_handler) {
 473			eeh_pcid_put(dev);
 474			return NULL;
 475		}
 476		eeh_pcid_put(dev);
 477	}
 478
 479#ifdef CONFIG_PCI_IOV
 480	pci_iov_add_virtfn(edev->physfn, edev->vf_index);
 481#endif
 482	return NULL;
 483}
 484
 485static void eeh_rmv_device(struct eeh_dev *edev, void *userdata)
 486{
 487	struct pci_driver *driver;
 
 488	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
 489	struct eeh_rmv_data *rmv_data = (struct eeh_rmv_data *)userdata;
 490
 491	/*
 492	 * Actually, we should remove the PCI bridges as well.
 493	 * However, that's lots of complexity to do that,
 494	 * particularly some of devices under the bridge might
 495	 * support EEH. So we just care about PCI devices for
 496	 * simplicity here.
 497	 */
 498	if (!eeh_edev_actionable(edev) ||
 499	    (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE))
 500		return;
 501
 502	if (rmv_data) {
 503		driver = eeh_pcid_get(dev);
 504		if (driver) {
 505			if (driver->err_handler &&
 506			    driver->err_handler->error_detected &&
 507			    driver->err_handler->slot_reset) {
 508				eeh_pcid_put(dev);
 509				return;
 510			}
 511			eeh_pcid_put(dev);
 512		}
 513	}
 514
 515	/* Remove it from PCI subsystem */
 516	pr_info("EEH: Removing %s without EEH sensitive driver\n",
 517		pci_name(dev));
 
 518	edev->mode |= EEH_DEV_DISCONNECTED;
 519	if (rmv_data)
 520		rmv_data->removed_dev_count++;
 521
 522	if (edev->physfn) {
 523#ifdef CONFIG_PCI_IOV
 524		pci_iov_remove_virtfn(edev->physfn, edev->vf_index);
 525		edev->pdev = NULL;
 526#endif
 527		if (rmv_data)
 528			list_add(&edev->rmv_entry, &rmv_data->removed_vf_list);
 529	} else {
 530		pci_lock_rescan_remove();
 531		pci_stop_and_remove_bus_device(dev);
 532		pci_unlock_rescan_remove();
 533	}
 534}
 535
 536static void *eeh_pe_detach_dev(struct eeh_pe *pe, void *userdata)
 537{
 
 538	struct eeh_dev *edev, *tmp;
 539
 540	eeh_pe_for_each_dev(pe, edev, tmp) {
 541		if (!(edev->mode & EEH_DEV_DISCONNECTED))
 542			continue;
 543
 544		edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
 545		eeh_pe_tree_remove(edev);
 546	}
 547
 548	return NULL;
 549}
 550
 551/*
 552 * Explicitly clear PE's frozen state for PowerNV where
 553 * we have frozen PE until BAR restore is completed. It's
 554 * harmless to clear it for pSeries. To be consistent with
 555 * PE reset (for 3 times), we try to clear the frozen state
 556 * for 3 times as well.
 557 */
 558static int eeh_clear_pe_frozen_state(struct eeh_pe *root, bool include_passed)
 559{
 560	struct eeh_pe *pe;
 561	int i;
 562
 563	eeh_for_each_pe(root, pe) {
 564		if (include_passed || !eeh_pe_passed(pe)) {
 565			for (i = 0; i < 3; i++)
 566				if (!eeh_unfreeze_pe(pe))
 567					break;
 568			if (i >= 3)
 569				return -EIO;
 570		}
 571	}
 572	eeh_pe_state_clear(root, EEH_PE_ISOLATED, include_passed);
 573	return 0;
 574}
 575
 576int eeh_pe_reset_and_recover(struct eeh_pe *pe)
 577{
 578	int ret;
 579
 580	/* Bail if the PE is being recovered */
 581	if (pe->state & EEH_PE_RECOVERING)
 582		return 0;
 583
 584	/* Put the PE into recovery mode */
 585	eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
 586
 587	/* Save states */
 588	eeh_pe_dev_traverse(pe, eeh_dev_save_state, NULL);
 589
 590	/* Issue reset */
 591	ret = eeh_pe_reset_full(pe, true);
 592	if (ret) {
 593		eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
 594		return ret;
 595	}
 596
 597	/* Unfreeze the PE */
 598	ret = eeh_clear_pe_frozen_state(pe, true);
 599	if (ret) {
 600		eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
 601		return ret;
 602	}
 603
 604	/* Restore device state */
 605	eeh_pe_dev_traverse(pe, eeh_dev_restore_state, NULL);
 606
 607	/* Clear recovery mode */
 608	eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
 609
 610	return 0;
 611}
 612
 613/**
 614 * eeh_reset_device - Perform actual reset of a pci slot
 615 * @driver_eeh_aware: Does the device's driver provide EEH support?
 616 * @pe: EEH PE
 617 * @bus: PCI bus corresponding to the isolcated slot
 618 * @rmv_data: Optional, list to record removed devices
 619 *
 620 * This routine must be called to do reset on the indicated PE.
 621 * During the reset, udev might be invoked because those affected
 622 * PCI devices will be removed and then added.
 623 */
 624static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus,
 625			    struct eeh_rmv_data *rmv_data,
 626			    bool driver_eeh_aware)
 627{
 628	time64_t tstamp;
 629	int cnt, rc;
 630	struct eeh_dev *edev;
 631	struct eeh_pe *tmp_pe;
 632	bool any_passed = false;
 633
 634	eeh_for_each_pe(pe, tmp_pe)
 635		any_passed |= eeh_pe_passed(tmp_pe);
 636
 637	/* pcibios will clear the counter; save the value */
 638	cnt = pe->freeze_count;
 639	tstamp = pe->tstamp;
 640
 641	/*
 642	 * We don't remove the corresponding PE instances because
 643	 * we need the information afterwords. The attached EEH
 644	 * devices are expected to be attached soon when calling
 645	 * into pci_hp_add_devices().
 646	 */
 647	eeh_pe_state_mark(pe, EEH_PE_KEEP);
 648	if (any_passed || driver_eeh_aware || (pe->type & EEH_PE_VF)) {
 649		eeh_pe_dev_traverse(pe, eeh_rmv_device, rmv_data);
 650	} else {
 651		pci_lock_rescan_remove();
 652		pci_hp_remove_devices(bus);
 653		pci_unlock_rescan_remove();
 
 
 654	}
 655
 656	/*
 657	 * Reset the pci controller. (Asserts RST#; resets config space).
 658	 * Reconfigure bridges and devices. Don't try to bring the system
 659	 * up if the reset failed for some reason.
 660	 *
 661	 * During the reset, it's very dangerous to have uncontrolled PCI
 662	 * config accesses. So we prefer to block them. However, controlled
 663	 * PCI config accesses initiated from EEH itself are allowed.
 664	 */
 665	rc = eeh_pe_reset_full(pe, false);
 666	if (rc)
 667		return rc;
 668
 669	pci_lock_rescan_remove();
 670
 671	/* Restore PE */
 672	eeh_ops->configure_bridge(pe);
 673	eeh_pe_restore_bars(pe);
 674
 675	/* Clear frozen state */
 676	rc = eeh_clear_pe_frozen_state(pe, false);
 677	if (rc) {
 678		pci_unlock_rescan_remove();
 679		return rc;
 680	}
 681
 682	/* Give the system 5 seconds to finish running the user-space
 683	 * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes,
 684	 * this is a hack, but if we don't do this, and try to bring
 685	 * the device up before the scripts have taken it down,
 686	 * potentially weird things happen.
 687	 */
 688	if (!driver_eeh_aware || rmv_data->removed_dev_count) {
 689		pr_info("EEH: Sleep 5s ahead of %s hotplug\n",
 690			(driver_eeh_aware ? "partial" : "complete"));
 691		ssleep(5);
 692
 693		/*
 694		 * The EEH device is still connected with its parent
 695		 * PE. We should disconnect it so the binding can be
 696		 * rebuilt when adding PCI devices.
 697		 */
 698		edev = list_first_entry(&pe->edevs, struct eeh_dev, entry);
 699		eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
 700		if (pe->type & EEH_PE_VF) {
 701			eeh_add_virt_device(edev);
 702		} else {
 703			if (!driver_eeh_aware)
 704				eeh_pe_state_clear(pe, EEH_PE_PRI_BUS, true);
 705			pci_hp_add_devices(bus);
 706		}
 707	}
 708	eeh_pe_state_clear(pe, EEH_PE_KEEP, true);
 709
 710	pe->tstamp = tstamp;
 711	pe->freeze_count = cnt;
 712
 713	pci_unlock_rescan_remove();
 714	return 0;
 715}
 716
 717/* The longest amount of time to wait for a pci device
 718 * to come back on line, in seconds.
 719 */
 720#define MAX_WAIT_FOR_RECOVERY 300
 721
 722
 723/* Walks the PE tree after processing an event to remove any stale PEs.
 724 *
 725 * NB: This needs to be recursive to ensure the leaf PEs get removed
 726 * before their parents do. Although this is possible to do recursively
 727 * we don't since this is easier to read and we need to garantee
 728 * the leaf nodes will be handled first.
 729 */
 730static void eeh_pe_cleanup(struct eeh_pe *pe)
 731{
 732	struct eeh_pe *child_pe, *tmp;
 733
 734	list_for_each_entry_safe(child_pe, tmp, &pe->child_list, child)
 735		eeh_pe_cleanup(child_pe);
 736
 737	if (pe->state & EEH_PE_KEEP)
 738		return;
 739
 740	if (!(pe->state & EEH_PE_INVALID))
 741		return;
 742
 743	if (list_empty(&pe->edevs) && list_empty(&pe->child_list)) {
 744		list_del(&pe->child);
 745		kfree(pe);
 746	}
 747}
 748
 749/**
 750 * eeh_check_slot_presence - Check if a device is still present in a slot
 751 * @pdev: pci_dev to check
 752 *
 753 * This function may return a false positive if we can't determine the slot's
 754 * presence state. This might happen for for PCIe slots if the PE containing
 755 * the upstream bridge is also frozen, or the bridge is part of the same PE
 756 * as the device.
 757 *
 758 * This shouldn't happen often, but you might see it if you hotplug a PCIe
 759 * switch.
 760 */
 761static bool eeh_slot_presence_check(struct pci_dev *pdev)
 762{
 763	const struct hotplug_slot_ops *ops;
 764	struct pci_slot *slot;
 765	u8 state;
 766	int rc;
 767
 768	if (!pdev)
 769		return false;
 770
 771	if (pdev->error_state == pci_channel_io_perm_failure)
 772		return false;
 773
 774	slot = pdev->slot;
 775	if (!slot || !slot->hotplug)
 776		return true;
 777
 778	ops = slot->hotplug->ops;
 779	if (!ops || !ops->get_adapter_status)
 780		return true;
 781
 782	/* set the attention indicator while we've got the slot ops */
 783	if (ops->set_attention_status)
 784		ops->set_attention_status(slot->hotplug, 1);
 785
 786	rc = ops->get_adapter_status(slot->hotplug, &state);
 787	if (rc)
 788		return true;
 789
 790	return !!state;
 791}
 792
 793static void eeh_clear_slot_attention(struct pci_dev *pdev)
 794{
 795	const struct hotplug_slot_ops *ops;
 796	struct pci_slot *slot;
 797
 798	if (!pdev)
 799		return;
 800
 801	if (pdev->error_state == pci_channel_io_perm_failure)
 802		return;
 803
 804	slot = pdev->slot;
 805	if (!slot || !slot->hotplug)
 806		return;
 807
 808	ops = slot->hotplug->ops;
 809	if (!ops || !ops->set_attention_status)
 810		return;
 811
 812	ops->set_attention_status(slot->hotplug, 0);
 813}
 814
 815/**
 816 * eeh_handle_normal_event - Handle EEH events on a specific PE
 817 * @pe: EEH PE - which should not be used after we return, as it may
 818 * have been invalidated.
 819 *
 820 * Attempts to recover the given PE.  If recovery fails or the PE has failed
 821 * too many times, remove the PE.
 822 *
 823 * While PHB detects address or data parity errors on particular PCI
 824 * slot, the associated PE will be frozen. Besides, DMA's occurring
 825 * to wild addresses (which usually happen due to bugs in device
 826 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
 827 * #PERR or other misc PCI-related errors also can trigger EEH errors.
 828 *
 829 * Recovery process consists of unplugging the device driver (which
 830 * generated hotplug events to userspace), then issuing a PCI #RST to
 831 * the device, then reconfiguring the PCI config space for all bridges
 832 * & devices under this slot, and then finally restarting the device
 833 * drivers (which cause a second set of hotplug events to go out to
 834 * userspace).
 835 */
 836void eeh_handle_normal_event(struct eeh_pe *pe)
 837{
 838	struct pci_bus *bus;
 839	struct eeh_dev *edev, *tmp;
 840	struct eeh_pe *tmp_pe;
 841	int rc = 0;
 842	enum pci_ers_result result = PCI_ERS_RESULT_NONE;
 843	struct eeh_rmv_data rmv_data =
 844		{LIST_HEAD_INIT(rmv_data.removed_vf_list), 0};
 845	int devices = 0;
 846
 847	bus = eeh_pe_bus_get(pe);
 848	if (!bus) {
 849		pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
 850			__func__, pe->phb->global_number, pe->addr);
 851		return;
 852	}
 853
 854	/*
 855	 * When devices are hot-removed we might get an EEH due to
 856	 * a driver attempting to touch the MMIO space of a removed
 857	 * device. In this case we don't have a device to recover
 858	 * so suppress the event if we can't find any present devices.
 859	 *
 860	 * The hotplug driver should take care of tearing down the
 861	 * device itself.
 862	 */
 863	eeh_for_each_pe(pe, tmp_pe)
 864		eeh_pe_for_each_dev(tmp_pe, edev, tmp)
 865			if (eeh_slot_presence_check(edev->pdev))
 866				devices++;
 867
 868	if (!devices) {
 869		pr_debug("EEH: Frozen PHB#%x-PE#%x is empty!\n",
 870			pe->phb->global_number, pe->addr);
 871		goto out; /* nothing to recover */
 872	}
 873
 874	/* Log the event */
 875	if (pe->type & EEH_PE_PHB) {
 876		pr_err("EEH: Recovering PHB#%x, location: %s\n",
 877			pe->phb->global_number, eeh_pe_loc_get(pe));
 878	} else {
 879		struct eeh_pe *phb_pe = eeh_phb_pe_get(pe->phb);
 880
 881		pr_err("EEH: Recovering PHB#%x-PE#%x\n",
 882		       pe->phb->global_number, pe->addr);
 883		pr_err("EEH: PE location: %s, PHB location: %s\n",
 884		       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
 885	}
 886
 887#ifdef CONFIG_STACKTRACE
 888	/*
 889	 * Print the saved stack trace now that we've verified there's
 890	 * something to recover.
 891	 */
 892	if (pe->trace_entries) {
 893		void **ptrs = (void **) pe->stack_trace;
 894		int i;
 895
 896		pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
 897		       pe->phb->global_number, pe->addr);
 898
 899		/* FIXME: Use the same format as dump_stack() */
 900		pr_err("EEH: Call Trace:\n");
 901		for (i = 0; i < pe->trace_entries; i++)
 902			pr_err("EEH: [%pK] %pS\n", ptrs[i], ptrs[i]);
 903
 904		pe->trace_entries = 0;
 905	}
 906#endif /* CONFIG_STACKTRACE */
 907
 908	eeh_pe_update_time_stamp(pe);
 909	pe->freeze_count++;
 910	if (pe->freeze_count > eeh_max_freezes) {
 911		pr_err("EEH: PHB#%x-PE#%x has failed %d times in the last hour and has been permanently disabled.\n",
 912		       pe->phb->global_number, pe->addr,
 913		       pe->freeze_count);
 914		result = PCI_ERS_RESULT_DISCONNECT;
 915	}
 916
 917	eeh_for_each_pe(pe, tmp_pe)
 918		eeh_pe_for_each_dev(tmp_pe, edev, tmp)
 919			edev->mode &= ~EEH_DEV_NO_HANDLER;
 920
 921	/* Walk the various device drivers attached to this slot through
 922	 * a reset sequence, giving each an opportunity to do what it needs
 923	 * to accomplish the reset.  Each child gets a report of the
 924	 * status ... if any child can't handle the reset, then the entire
 925	 * slot is dlpar removed and added.
 926	 *
 927	 * When the PHB is fenced, we have to issue a reset to recover from
 928	 * the error. Override the result if necessary to have partially
 929	 * hotplug for this case.
 930	 */
 931	if (result != PCI_ERS_RESULT_DISCONNECT) {
 932		pr_warn("EEH: This PCI device has failed %d times in the last hour and will be permanently disabled after %d failures.\n",
 933			pe->freeze_count, eeh_max_freezes);
 934		pr_info("EEH: Notify device drivers to shutdown\n");
 935		eeh_set_channel_state(pe, pci_channel_io_frozen);
 936		eeh_set_irq_state(pe, false);
 937		eeh_pe_report("error_detected(IO frozen)", pe,
 938			      eeh_report_error, &result);
 939		if ((pe->type & EEH_PE_PHB) &&
 940		    result != PCI_ERS_RESULT_NONE &&
 941		    result != PCI_ERS_RESULT_NEED_RESET)
 942			result = PCI_ERS_RESULT_NEED_RESET;
 943	}
 944
 945	/* Get the current PCI slot state. This can take a long time,
 946	 * sometimes over 300 seconds for certain systems.
 947	 */
 948	if (result != PCI_ERS_RESULT_DISCONNECT) {
 949		rc = eeh_wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
 950		if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
 951			pr_warn("EEH: Permanent failure\n");
 952			result = PCI_ERS_RESULT_DISCONNECT;
 953		}
 954	}
 955
 956	/* Since rtas may enable MMIO when posting the error log,
 957	 * don't post the error log until after all dev drivers
 958	 * have been informed.
 959	 */
 960	if (result != PCI_ERS_RESULT_DISCONNECT) {
 961		pr_info("EEH: Collect temporary log\n");
 962		eeh_slot_error_detail(pe, EEH_LOG_TEMP);
 963	}
 964
 965	/* If all device drivers were EEH-unaware, then shut
 966	 * down all of the device drivers, and hope they
 967	 * go down willingly, without panicing the system.
 968	 */
 969	if (result == PCI_ERS_RESULT_NONE) {
 970		pr_info("EEH: Reset with hotplug activity\n");
 971		rc = eeh_reset_device(pe, bus, NULL, false);
 972		if (rc) {
 973			pr_warn("%s: Unable to reset, err=%d\n",
 974				__func__, rc);
 975			result = PCI_ERS_RESULT_DISCONNECT;
 976		}
 977	}
 978
 979	/* If all devices reported they can proceed, then re-enable MMIO */
 980	if (result == PCI_ERS_RESULT_CAN_RECOVER) {
 981		pr_info("EEH: Enable I/O for affected devices\n");
 982		rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
 983
 984		if (rc < 0) {
 985			result = PCI_ERS_RESULT_DISCONNECT;
 986		} else if (rc) {
 987			result = PCI_ERS_RESULT_NEED_RESET;
 988		} else {
 989			pr_info("EEH: Notify device drivers to resume I/O\n");
 990			eeh_pe_report("mmio_enabled", pe,
 991				      eeh_report_mmio_enabled, &result);
 992		}
 993	}
 994
 995	/* If all devices reported they can proceed, then re-enable DMA */
 996	if (result == PCI_ERS_RESULT_CAN_RECOVER) {
 997		pr_info("EEH: Enabled DMA for affected devices\n");
 998		rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
 999
1000		if (rc < 0) {
1001			result = PCI_ERS_RESULT_DISCONNECT;
1002		} else if (rc) {
1003			result = PCI_ERS_RESULT_NEED_RESET;
1004		} else {
1005			/*
1006			 * We didn't do PE reset for the case. The PE
1007			 * is still in frozen state. Clear it before
1008			 * resuming the PE.
1009			 */
1010			eeh_pe_state_clear(pe, EEH_PE_ISOLATED, true);
1011			result = PCI_ERS_RESULT_RECOVERED;
1012		}
 
 
 
 
 
1013	}
1014
1015	/* If any device called out for a reset, then reset the slot */
1016	if (result == PCI_ERS_RESULT_NEED_RESET) {
1017		pr_info("EEH: Reset without hotplug activity\n");
1018		rc = eeh_reset_device(pe, bus, &rmv_data, true);
1019		if (rc) {
1020			pr_warn("%s: Cannot reset, err=%d\n",
1021				__func__, rc);
1022			result = PCI_ERS_RESULT_DISCONNECT;
1023		} else {
1024			result = PCI_ERS_RESULT_NONE;
1025			eeh_set_channel_state(pe, pci_channel_io_normal);
1026			eeh_set_irq_state(pe, true);
1027			eeh_pe_report("slot_reset", pe, eeh_report_reset,
1028				      &result);
1029		}
1030	}
1031
1032	if ((result == PCI_ERS_RESULT_RECOVERED) ||
1033	    (result == PCI_ERS_RESULT_NONE)) {
1034		/*
1035		 * For those hot removed VFs, we should add back them after PF
1036		 * get recovered properly.
1037		 */
1038		list_for_each_entry_safe(edev, tmp, &rmv_data.removed_vf_list,
1039					 rmv_entry) {
1040			eeh_add_virt_device(edev);
1041			list_del(&edev->rmv_entry);
1042		}
1043
1044		/* Tell all device drivers that they can resume operations */
1045		pr_info("EEH: Notify device driver to resume\n");
1046		eeh_set_channel_state(pe, pci_channel_io_normal);
1047		eeh_set_irq_state(pe, true);
1048		eeh_pe_report("resume", pe, eeh_report_resume, NULL);
1049		eeh_for_each_pe(pe, tmp_pe) {
1050			eeh_pe_for_each_dev(tmp_pe, edev, tmp) {
1051				edev->mode &= ~EEH_DEV_NO_HANDLER;
1052				edev->in_error = false;
1053			}
1054		}
1055
1056		pr_info("EEH: Recovery successful.\n");
1057	} else  {
1058		/*
1059		 * About 90% of all real-life EEH failures in the field
1060		 * are due to poorly seated PCI cards. Only 10% or so are
1061		 * due to actual, failed cards.
1062		 */
1063		pr_err("EEH: Unable to recover from failure from PHB#%x-PE#%x.\n"
1064		       "Please try reseating or replacing it\n",
1065			pe->phb->global_number, pe->addr);
1066
1067		eeh_slot_error_detail(pe, EEH_LOG_PERM);
1068
1069		/* Notify all devices that they're about to go down. */
1070		eeh_set_channel_state(pe, pci_channel_io_perm_failure);
1071		eeh_set_irq_state(pe, false);
1072		eeh_pe_report("error_detected(permanent failure)", pe,
1073			      eeh_report_failure, NULL);
1074
1075		/* Mark the PE to be removed permanently */
1076		eeh_pe_state_mark(pe, EEH_PE_REMOVED);
 
 
 
 
1077
1078		/*
1079		 * Shut down the device drivers for good. We mark
1080		 * all removed devices correctly to avoid access
1081		 * the their PCI config any more.
1082		 */
1083		if (pe->type & EEH_PE_VF) {
1084			eeh_pe_dev_traverse(pe, eeh_rmv_device, NULL);
1085			eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
1086		} else {
1087			eeh_pe_state_clear(pe, EEH_PE_PRI_BUS, true);
1088			eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
1089
1090			pci_lock_rescan_remove();
1091			pci_hp_remove_devices(bus);
1092			pci_unlock_rescan_remove();
1093			/* The passed PE should no longer be used */
1094			return;
1095		}
1096	}
1097
1098out:
1099	/*
1100	 * Clean up any PEs without devices. While marked as EEH_PE_RECOVERYING
1101	 * we don't want to modify the PE tree structure so we do it here.
1102	 */
1103	eeh_pe_cleanup(pe);
 
 
 
 
 
 
 
 
 
 
 
1104
1105	/* clear the slot attention LED for all recovered devices */
1106	eeh_for_each_pe(pe, tmp_pe)
1107		eeh_pe_for_each_dev(tmp_pe, edev, tmp)
1108			eeh_clear_slot_attention(edev->pdev);
1109
1110	eeh_pe_state_clear(pe, EEH_PE_RECOVERING, true);
 
 
 
 
 
 
 
 
1111}
1112
1113/**
1114 * eeh_handle_special_event - Handle EEH events without a specific failing PE
1115 *
1116 * Called when an EEH event is detected but can't be narrowed down to a
1117 * specific PE.  Iterates through possible failures and handles them as
1118 * necessary.
1119 */
1120void eeh_handle_special_event(void)
1121{
1122	struct eeh_pe *pe, *phb_pe, *tmp_pe;
1123	struct eeh_dev *edev, *tmp_edev;
1124	struct pci_bus *bus;
1125	struct pci_controller *hose;
1126	unsigned long flags;
1127	int rc;
1128
1129
1130	do {
1131		rc = eeh_ops->next_error(&pe);
1132
1133		switch (rc) {
1134		case EEH_NEXT_ERR_DEAD_IOC:
1135			/* Mark all PHBs in dead state */
1136			eeh_serialize_lock(&flags);
1137
1138			/* Purge all events */
1139			eeh_remove_event(NULL, true);
1140
1141			list_for_each_entry(hose, &hose_list, list_node) {
1142				phb_pe = eeh_phb_pe_get(hose);
1143				if (!phb_pe) continue;
1144
1145				eeh_pe_mark_isolated(phb_pe);
 
1146			}
1147
1148			eeh_serialize_unlock(flags);
1149
1150			break;
1151		case EEH_NEXT_ERR_FROZEN_PE:
1152		case EEH_NEXT_ERR_FENCED_PHB:
1153		case EEH_NEXT_ERR_DEAD_PHB:
1154			/* Mark the PE in fenced state */
1155			eeh_serialize_lock(&flags);
1156
1157			/* Purge all events of the PHB */
1158			eeh_remove_event(pe, true);
1159
1160			if (rc != EEH_NEXT_ERR_DEAD_PHB)
1161				eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
1162			eeh_pe_mark_isolated(pe);
 
 
 
1163
1164			eeh_serialize_unlock(flags);
1165
1166			break;
1167		case EEH_NEXT_ERR_NONE:
1168			return;
1169		default:
1170			pr_warn("%s: Invalid value %d from next_error()\n",
1171				__func__, rc);
1172			return;
1173		}
1174
1175		/*
1176		 * For fenced PHB and frozen PE, it's handled as normal
1177		 * event. We have to remove the affected PHBs for dead
1178		 * PHB and IOC
1179		 */
1180		if (rc == EEH_NEXT_ERR_FROZEN_PE ||
1181		    rc == EEH_NEXT_ERR_FENCED_PHB) {
1182			eeh_pe_state_mark(pe, EEH_PE_RECOVERING);
1183			eeh_handle_normal_event(pe);
1184		} else {
1185			eeh_for_each_pe(pe, tmp_pe)
1186				eeh_pe_for_each_dev(tmp_pe, edev, tmp_edev)
1187					edev->mode &= ~EEH_DEV_NO_HANDLER;
1188
1189			/* Notify all devices to be down */
1190			eeh_pe_state_clear(pe, EEH_PE_PRI_BUS, true);
1191			eeh_set_channel_state(pe, pci_channel_io_perm_failure);
1192			eeh_pe_report(
1193				"error_detected(permanent failure)", pe,
1194				eeh_report_failure, NULL);
1195
1196			pci_lock_rescan_remove();
1197			list_for_each_entry(hose, &hose_list, list_node) {
1198				phb_pe = eeh_phb_pe_get(hose);
1199				if (!phb_pe ||
1200				    !(phb_pe->state & EEH_PE_ISOLATED) ||
1201				    (phb_pe->state & EEH_PE_RECOVERING))
1202					continue;
1203
 
1204				bus = eeh_pe_bus_get(phb_pe);
1205				if (!bus) {
1206					pr_err("%s: Cannot find PCI bus for "
1207					       "PHB#%x-PE#%x\n",
1208					       __func__,
1209					       pe->phb->global_number,
1210					       pe->addr);
1211					break;
1212				}
1213				pci_hp_remove_devices(bus);
1214			}
1215			pci_unlock_rescan_remove();
1216		}
1217
1218		/*
1219		 * If we have detected dead IOC, we needn't proceed
1220		 * any more since all PHBs would have been removed
1221		 */
1222		if (rc == EEH_NEXT_ERR_DEAD_IOC)
1223			break;
1224	} while (rc != EEH_NEXT_ERR_NONE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1225}
v3.15
 
  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/module.h>
 29#include <linux/pci.h>
 
 30#include <asm/eeh.h>
 31#include <asm/eeh_event.h>
 32#include <asm/ppc-pci.h>
 33#include <asm/pci-bridge.h>
 34#include <asm/prom.h>
 35#include <asm/rtas.h>
 36
 37/**
 38 * eeh_pcid_name - Retrieve name of PCI device driver
 39 * @pdev: PCI device
 40 *
 41 * This routine is used to retrieve the name of PCI device driver
 42 * if that's valid.
 43 */
 44static inline const char *eeh_pcid_name(struct pci_dev *pdev)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 45{
 46	if (pdev && pdev->dev.driver)
 47		return pdev->dev.driver->name;
 48	return "";
 
 
 
 
 
 
 
 49}
 50
 51/**
 52 * eeh_pcid_get - Get the PCI device driver
 53 * @pdev: PCI device
 54 *
 55 * The function is used to retrieve the PCI device driver for
 56 * the indicated PCI device. Besides, we will increase the reference
 57 * of the PCI device driver to prevent that being unloaded on
 58 * the fly. Otherwise, kernel crash would be seen.
 59 */
 60static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev)
 61{
 62	if (!pdev || !pdev->driver)
 63		return NULL;
 64
 65	if (!try_module_get(pdev->driver->driver.owner))
 66		return NULL;
 67
 68	return pdev->driver;
 69}
 70
 71/**
 72 * eeh_pcid_put - Dereference on the PCI device driver
 73 * @pdev: PCI device
 74 *
 75 * The function is called to do dereference on the PCI device
 76 * driver of the indicated PCI device.
 77 */
 78static inline void eeh_pcid_put(struct pci_dev *pdev)
 79{
 80	if (!pdev || !pdev->driver)
 81		return;
 82
 83	module_put(pdev->driver->driver.owner);
 84}
 85
 86#if 0
 87static void print_device_node_tree(struct pci_dn *pdn, int dent)
 88{
 89	int i;
 90	struct device_node *pc;
 91
 92	if (!pdn)
 93		return;
 94	for (i = 0; i < dent; i++)
 95		printk(" ");
 96	printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n",
 97		pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr,
 98		pdn->eeh_pe_config_addr, pdn->node->full_name);
 99	dent += 3;
100	pc = pdn->node->child;
101	while (pc) {
102		print_device_node_tree(PCI_DN(pc), dent);
103		pc = pc->sibling;
104	}
105}
106#endif
107
108/**
109 * eeh_disable_irq - Disable interrupt for the recovering device
110 * @dev: PCI device
111 *
112 * This routine must be called when reporting temporary or permanent
113 * error to the particular PCI device to disable interrupt of that
114 * device. If the device has enabled MSI or MSI-X interrupt, we needn't
115 * do real work because EEH should freeze DMA transfers for those PCI
116 * devices encountering EEH errors, which includes MSI or MSI-X.
117 */
118static void eeh_disable_irq(struct pci_dev *dev)
119{
120	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
121
122	/* Don't disable MSI and MSI-X interrupts. They are
123	 * effectively disabled by the DMA Stopped state
124	 * when an EEH error occurs.
125	 */
126	if (dev->msi_enabled || dev->msix_enabled)
127		return;
128
129	if (!irq_has_action(dev->irq))
130		return;
131
132	edev->mode |= EEH_DEV_IRQ_DISABLED;
133	disable_irq_nosync(dev->irq);
134}
135
136/**
137 * eeh_enable_irq - Enable interrupt for the recovering device
138 * @dev: PCI device
139 *
140 * This routine must be called to enable interrupt while failed
141 * device could be resumed.
142 */
143static void eeh_enable_irq(struct pci_dev *dev)
144{
145	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
146
147	if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
148		edev->mode &= ~EEH_DEV_IRQ_DISABLED;
149		/*
150		 * FIXME !!!!!
151		 *
152		 * This is just ass backwards. This maze has
153		 * unbalanced irq_enable/disable calls. So instead of
154		 * finding the root cause it works around the warning
155		 * in the irq_enable code by conditionally calling
156		 * into it.
157		 *
158		 * That's just wrong.The warning in the core code is
159		 * there to tell people to fix their assymetries in
160		 * their own code, not by abusing the core information
161		 * to avoid it.
162		 *
163		 * I so wish that the assymetry would be the other way
164		 * round and a few more irq_disable calls render that
165		 * shit unusable forever.
166		 *
167		 *	tglx
168		 */
169		if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
170			enable_irq(dev->irq);
171	}
172}
173
174/**
175 * eeh_report_error - Report pci error to each device driver
176 * @data: eeh device
177 * @userdata: return value
178 *
179 * Report an EEH error to each device driver, collect up and
180 * merge the device driver responses. Cumulative response
181 * passed back in "userdata".
182 */
183static void *eeh_report_error(void *data, void *userdata)
184{
185	struct eeh_dev *edev = (struct eeh_dev *)data;
186	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
187	enum pci_ers_result rc, *res = userdata;
188	struct pci_driver *driver;
189
190	/* We might not have the associated PCI device,
191	 * then we should continue for next one.
 
 
 
 
192	 */
193	if (!dev) return NULL;
194	dev->error_state = pci_channel_io_frozen;
195
196	driver = eeh_pcid_get(dev);
197	if (!driver) return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
198
199	eeh_disable_irq(dev);
 
 
 
200
201	if (!driver->err_handler ||
202	    !driver->err_handler->error_detected) {
203		eeh_pcid_put(dev);
204		return NULL;
205	}
 
 
 
 
 
 
 
 
 
 
 
206
207	rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
208
209	/* A driver that needs a reset trumps all others */
210	if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
211	if (*res == PCI_ERS_RESULT_NONE) *res = rc;
 
 
212
213	eeh_pcid_put(dev);
214	return NULL;
 
 
 
 
 
 
215}
216
217/**
218 * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled
219 * @data: eeh device
220 * @userdata: return value
221 *
222 * Tells each device driver that IO ports, MMIO and config space I/O
223 * are now enabled. Collects up and merges the device driver responses.
224 * Cumulative response passed back in "userdata".
225 */
226static void *eeh_report_mmio_enabled(void *data, void *userdata)
 
 
227{
228	struct eeh_dev *edev = (struct eeh_dev *)data;
229	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
230	enum pci_ers_result rc, *res = userdata;
231	struct pci_driver *driver;
232
233	driver = eeh_pcid_get(dev);
234	if (!driver) return NULL;
235
236	if (!driver->err_handler ||
237	    !driver->err_handler->mmio_enabled ||
238	    (edev->mode & EEH_DEV_NO_HANDLER)) {
239		eeh_pcid_put(dev);
240		return NULL;
241	}
 
 
242
243	rc = driver->err_handler->mmio_enabled(dev);
244
245	/* A driver that needs a reset trumps all others */
246	if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
247	if (*res == PCI_ERS_RESULT_NONE) *res = rc;
248
249	eeh_pcid_put(dev);
250	return NULL;
 
 
 
 
 
 
 
 
251}
252
253/**
254 * eeh_report_reset - Tell device that slot has been reset
255 * @data: eeh device
256 * @userdata: return value
257 *
258 * This routine must be called while EEH tries to reset particular
259 * PCI device so that the associated PCI device driver could take
260 * some actions, usually to save data the driver needs so that the
261 * driver can work again while the device is recovered.
262 */
263static void *eeh_report_reset(void *data, void *userdata)
 
 
264{
265	struct eeh_dev *edev = (struct eeh_dev *)data;
266	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
267	enum pci_ers_result rc, *res = userdata;
268	struct pci_driver *driver;
 
269
270	if (!dev) return NULL;
271	dev->error_state = pci_channel_io_normal;
 
272
273	driver = eeh_pcid_get(dev);
274	if (!driver) return NULL;
275
276	eeh_enable_irq(dev);
 
 
 
 
 
 
 
 
277
278	if (!driver->err_handler ||
279	    !driver->err_handler->slot_reset ||
280	    (edev->mode & EEH_DEV_NO_HANDLER)) {
281		eeh_pcid_put(dev);
282		return NULL;
283	}
284
285	rc = driver->err_handler->slot_reset(dev);
286	if ((*res == PCI_ERS_RESULT_NONE) ||
287	    (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
288	if (*res == PCI_ERS_RESULT_DISCONNECT &&
289	     rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
290
291	eeh_pcid_put(dev);
292	return NULL;
293}
294
295/**
296 * eeh_report_resume - Tell device to resume normal operations
297 * @data: eeh device
298 * @userdata: return value
299 *
300 * This routine must be called to notify the device driver that it
301 * could resume so that the device driver can do some initialization
302 * to make the recovered device work again.
303 */
304static void *eeh_report_resume(void *data, void *userdata)
 
 
305{
306	struct eeh_dev *edev = (struct eeh_dev *)data;
307	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
308	struct pci_driver *driver;
309
310	if (!dev) return NULL;
311	dev->error_state = pci_channel_io_normal;
312
313	driver = eeh_pcid_get(dev);
314	if (!driver) return NULL;
315
316	eeh_enable_irq(dev);
317
318	if (!driver->err_handler ||
319	    !driver->err_handler->resume ||
320	    (edev->mode & EEH_DEV_NO_HANDLER)) {
321		edev->mode &= ~EEH_DEV_NO_HANDLER;
322		eeh_pcid_put(dev);
323		return NULL;
324	}
325
326	driver->err_handler->resume(dev);
327
328	eeh_pcid_put(dev);
329	return NULL;
330}
331
332/**
333 * eeh_report_failure - Tell device driver that device is dead.
334 * @data: eeh device
335 * @userdata: return value
336 *
337 * This informs the device driver that the device is permanently
338 * dead, and that no further recovery attempts will be made on it.
339 */
340static void *eeh_report_failure(void *data, void *userdata)
 
 
341{
342	struct eeh_dev *edev = (struct eeh_dev *)data;
343	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
344	struct pci_driver *driver;
 
345
346	if (!dev) return NULL;
347	dev->error_state = pci_channel_io_perm_failure;
 
 
348
349	driver = eeh_pcid_get(dev);
350	if (!driver) return NULL;
 
351
352	eeh_disable_irq(dev);
 
 
 
353
354	if (!driver->err_handler ||
355	    !driver->err_handler->error_detected) {
356		eeh_pcid_put(dev);
357		return NULL;
358	}
359
360	driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
 
 
 
 
 
 
 
361
362	eeh_pcid_put(dev);
 
 
363	return NULL;
364}
365
366static void *eeh_rmv_device(void *data, void *userdata)
367{
368	struct pci_driver *driver;
369	struct eeh_dev *edev = (struct eeh_dev *)data;
370	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
371	int *removed = (int *)userdata;
372
373	/*
374	 * Actually, we should remove the PCI bridges as well.
375	 * However, that's lots of complexity to do that,
376	 * particularly some of devices under the bridge might
377	 * support EEH. So we just care about PCI devices for
378	 * simplicity here.
379	 */
380	if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
381		return NULL;
 
382
383	driver = eeh_pcid_get(dev);
384	if (driver) {
385		eeh_pcid_put(dev);
386		if (driver->err_handler)
387			return NULL;
 
 
 
 
 
 
388	}
389
390	/* Remove it from PCI subsystem */
391	pr_debug("EEH: Removing %s without EEH sensitive driver\n",
392		 pci_name(dev));
393	edev->bus = dev->bus;
394	edev->mode |= EEH_DEV_DISCONNECTED;
395	(*removed)++;
 
396
397	pci_lock_rescan_remove();
398	pci_stop_and_remove_bus_device(dev);
399	pci_unlock_rescan_remove();
400
401	return NULL;
 
 
 
 
 
 
 
402}
403
404static void *eeh_pe_detach_dev(void *data, void *userdata)
405{
406	struct eeh_pe *pe = (struct eeh_pe *)data;
407	struct eeh_dev *edev, *tmp;
408
409	eeh_pe_for_each_dev(pe, edev, tmp) {
410		if (!(edev->mode & EEH_DEV_DISCONNECTED))
411			continue;
412
413		edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED);
414		eeh_rmv_from_parent_pe(edev);
415	}
416
417	return NULL;
418}
419
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
420/**
421 * eeh_reset_device - Perform actual reset of a pci slot
 
422 * @pe: EEH PE
423 * @bus: PCI bus corresponding to the isolcated slot
 
424 *
425 * This routine must be called to do reset on the indicated PE.
426 * During the reset, udev might be invoked because those affected
427 * PCI devices will be removed and then added.
428 */
429static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
 
 
430{
431	struct pci_bus *frozen_bus = eeh_pe_bus_get(pe);
432	struct timeval tstamp;
433	int cnt, rc, removed = 0;
 
 
 
 
 
434
435	/* pcibios will clear the counter; save the value */
436	cnt = pe->freeze_count;
437	tstamp = pe->tstamp;
438
439	/*
440	 * We don't remove the corresponding PE instances because
441	 * we need the information afterwords. The attached EEH
442	 * devices are expected to be attached soon when calling
443	 * into pcibios_add_pci_devices().
444	 */
445	eeh_pe_state_mark(pe, EEH_PE_KEEP);
446	if (bus) {
 
 
447		pci_lock_rescan_remove();
448		pcibios_remove_pci_devices(bus);
449		pci_unlock_rescan_remove();
450	} else if (frozen_bus) {
451		eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
452	}
453
454	/* Reset the pci controller. (Asserts RST#; resets config space).
 
455	 * Reconfigure bridges and devices. Don't try to bring the system
456	 * up if the reset failed for some reason.
 
 
 
 
457	 */
458	rc = eeh_reset_pe(pe);
459	if (rc)
460		return rc;
461
462	pci_lock_rescan_remove();
463
464	/* Restore PE */
465	eeh_ops->configure_bridge(pe);
466	eeh_pe_restore_bars(pe);
467
 
 
 
 
 
 
 
468	/* Give the system 5 seconds to finish running the user-space
469	 * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes,
470	 * this is a hack, but if we don't do this, and try to bring
471	 * the device up before the scripts have taken it down,
472	 * potentially weird things happen.
473	 */
474	if (bus) {
475		pr_info("EEH: Sleep 5s ahead of complete hotplug\n");
 
476		ssleep(5);
477
478		/*
479		 * The EEH device is still connected with its parent
480		 * PE. We should disconnect it so the binding can be
481		 * rebuilt when adding PCI devices.
482		 */
 
483		eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
484		pcibios_add_pci_devices(bus);
485	} else if (frozen_bus && removed) {
486		pr_info("EEH: Sleep 5s ahead of partial hotplug\n");
487		ssleep(5);
488
489		eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL);
490		pcibios_add_pci_devices(frozen_bus);
491	}
492	eeh_pe_state_clear(pe, EEH_PE_KEEP);
493
494	pe->tstamp = tstamp;
495	pe->freeze_count = cnt;
496
497	pci_unlock_rescan_remove();
498	return 0;
499}
500
501/* The longest amount of time to wait for a pci device
502 * to come back on line, in seconds.
503 */
504#define MAX_WAIT_FOR_RECOVERY 300
505
506static void eeh_handle_normal_event(struct eeh_pe *pe)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
507{
508	struct pci_bus *frozen_bus;
 
 
509	int rc = 0;
510	enum pci_ers_result result = PCI_ERS_RESULT_NONE;
511
512	frozen_bus = eeh_pe_bus_get(pe);
513	if (!frozen_bus) {
514		pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n",
 
 
 
515			__func__, pe->phb->global_number, pe->addr);
516		return;
517	}
518
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
519	eeh_pe_update_time_stamp(pe);
520	pe->freeze_count++;
521	if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES)
522		goto excess_failures;
523	pr_warning("EEH: This PCI device has failed %d times in the last hour\n",
524		pe->freeze_count);
 
 
 
 
 
 
525
526	/* Walk the various device drivers attached to this slot through
527	 * a reset sequence, giving each an opportunity to do what it needs
528	 * to accomplish the reset.  Each child gets a report of the
529	 * status ... if any child can't handle the reset, then the entire
530	 * slot is dlpar removed and added.
 
 
 
 
531	 */
532	pr_info("EEH: Notify device drivers to shutdown\n");
533	eeh_pe_dev_traverse(pe, eeh_report_error, &result);
 
 
 
 
 
 
 
 
 
 
 
534
535	/* Get the current PCI slot state. This can take a long time,
536	 * sometimes over 3 seconds for certain systems.
537	 */
538	rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000);
539	if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
540		pr_warning("EEH: Permanent failure\n");
541		goto hard_fail;
 
 
542	}
543
544	/* Since rtas may enable MMIO when posting the error log,
545	 * don't post the error log until after all dev drivers
546	 * have been informed.
547	 */
548	pr_info("EEH: Collect temporary log\n");
549	eeh_slot_error_detail(pe, EEH_LOG_TEMP);
 
 
550
551	/* If all device drivers were EEH-unaware, then shut
552	 * down all of the device drivers, and hope they
553	 * go down willingly, without panicing the system.
554	 */
555	if (result == PCI_ERS_RESULT_NONE) {
556		pr_info("EEH: Reset with hotplug activity\n");
557		rc = eeh_reset_device(pe, frozen_bus);
558		if (rc) {
559			pr_warning("%s: Unable to reset, err=%d\n",
560				   __func__, rc);
561			goto hard_fail;
562		}
563	}
564
565	/* If all devices reported they can proceed, then re-enable MMIO */
566	if (result == PCI_ERS_RESULT_CAN_RECOVER) {
567		pr_info("EEH: Enable I/O for affected devices\n");
568		rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
569
570		if (rc < 0)
571			goto hard_fail;
572		if (rc) {
573			result = PCI_ERS_RESULT_NEED_RESET;
574		} else {
575			pr_info("EEH: Notify device drivers to resume I/O\n");
576			result = PCI_ERS_RESULT_NONE;
577			eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
578		}
579	}
580
581	/* If all devices reported they can proceed, then re-enable DMA */
582	if (result == PCI_ERS_RESULT_CAN_RECOVER) {
583		pr_info("EEH: Enabled DMA for affected devices\n");
584		rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
585
586		if (rc < 0)
587			goto hard_fail;
588		if (rc)
589			result = PCI_ERS_RESULT_NEED_RESET;
590		else
 
 
 
 
 
 
591			result = PCI_ERS_RESULT_RECOVERED;
592	}
593
594	/* If any device has a hard failure, then shut off everything. */
595	if (result == PCI_ERS_RESULT_DISCONNECT) {
596		pr_warning("EEH: Device driver gave up\n");
597		goto hard_fail;
598	}
599
600	/* If any device called out for a reset, then reset the slot */
601	if (result == PCI_ERS_RESULT_NEED_RESET) {
602		pr_info("EEH: Reset without hotplug activity\n");
603		rc = eeh_reset_device(pe, NULL);
604		if (rc) {
605			pr_warning("%s: Cannot reset, err=%d\n",
606				   __func__, rc);
607			goto hard_fail;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
608		}
609
610		pr_info("EEH: Notify device drivers "
611			"the completion of reset\n");
612		result = PCI_ERS_RESULT_NONE;
613		eeh_pe_dev_traverse(pe, eeh_report_reset, &result);
614	}
 
 
 
 
 
 
 
 
 
 
 
 
 
615
616	/* All devices should claim they have recovered by now. */
617	if ((result != PCI_ERS_RESULT_RECOVERED) &&
618	    (result != PCI_ERS_RESULT_NONE)) {
619		pr_warning("EEH: Not recovered\n");
620		goto hard_fail;
621	}
622
623	/* Tell all device drivers that they can resume operations */
624	pr_info("EEH: Notify device driver to resume\n");
625	eeh_pe_dev_traverse(pe, eeh_report_resume, NULL);
 
 
 
 
 
 
 
 
626
627	return;
 
 
 
 
 
 
628
629excess_failures:
630	/*
631	 * About 90% of all real-life EEH failures in the field
632	 * are due to poorly seated PCI cards. Only 10% or so are
633	 * due to actual, failed cards.
634	 */
635	pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n"
636	       "last hour and has been permanently disabled.\n"
637	       "Please try reseating or replacing it.\n",
638		pe->phb->global_number, pe->addr,
639		pe->freeze_count);
640	goto perm_error;
641
642hard_fail:
643	pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n"
644	       "Please try reseating or replacing it\n",
645		pe->phb->global_number, pe->addr);
646
647perm_error:
648	eeh_slot_error_detail(pe, EEH_LOG_PERM);
 
 
649
650	/* Notify all devices that they're about to go down. */
651	eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
652
653	/* Shut down the device drivers for good. */
654	if (frozen_bus) {
655		pci_lock_rescan_remove();
656		pcibios_remove_pci_devices(frozen_bus);
657		pci_unlock_rescan_remove();
658	}
659}
660
661static void eeh_handle_special_event(void)
 
 
 
 
 
 
 
662{
663	struct eeh_pe *pe, *phb_pe;
 
664	struct pci_bus *bus;
665	struct pci_controller *hose;
666	unsigned long flags;
667	int rc;
668
669
670	do {
671		rc = eeh_ops->next_error(&pe);
672
673		switch (rc) {
674		case EEH_NEXT_ERR_DEAD_IOC:
675			/* Mark all PHBs in dead state */
676			eeh_serialize_lock(&flags);
677
678			/* Purge all events */
679			eeh_remove_event(NULL);
680
681			list_for_each_entry(hose, &hose_list, list_node) {
682				phb_pe = eeh_phb_pe_get(hose);
683				if (!phb_pe) continue;
684
685				eeh_pe_state_mark(phb_pe,
686					EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
687			}
688
689			eeh_serialize_unlock(flags);
690
691			break;
692		case EEH_NEXT_ERR_FROZEN_PE:
693		case EEH_NEXT_ERR_FENCED_PHB:
694		case EEH_NEXT_ERR_DEAD_PHB:
695			/* Mark the PE in fenced state */
696			eeh_serialize_lock(&flags);
697
698			/* Purge all events of the PHB */
699			eeh_remove_event(pe);
700
701			if (rc == EEH_NEXT_ERR_DEAD_PHB)
702				eeh_pe_state_mark(pe,
703					EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
704			else
705				eeh_pe_state_mark(pe,
706					EEH_PE_ISOLATED | EEH_PE_RECOVERING);
707
708			eeh_serialize_unlock(flags);
709
710			break;
711		case EEH_NEXT_ERR_NONE:
712			return;
713		default:
714			pr_warn("%s: Invalid value %d from next_error()\n",
715				__func__, rc);
716			return;
717		}
718
719		/*
720		 * For fenced PHB and frozen PE, it's handled as normal
721		 * event. We have to remove the affected PHBs for dead
722		 * PHB and IOC
723		 */
724		if (rc == EEH_NEXT_ERR_FROZEN_PE ||
725		    rc == EEH_NEXT_ERR_FENCED_PHB) {
 
726			eeh_handle_normal_event(pe);
727		} else {
 
 
 
 
 
 
 
 
 
 
 
728			pci_lock_rescan_remove();
729			list_for_each_entry(hose, &hose_list, list_node) {
730				phb_pe = eeh_phb_pe_get(hose);
731				if (!phb_pe ||
732				    !(phb_pe->state & EEH_PE_PHB_DEAD))
 
733					continue;
734
735				/* Notify all devices to be down */
736				bus = eeh_pe_bus_get(phb_pe);
737				eeh_pe_dev_traverse(pe,
738					eeh_report_failure, NULL);
739				pcibios_remove_pci_devices(bus);
 
 
 
 
 
 
740			}
741			pci_unlock_rescan_remove();
742		}
743
744		/*
745		 * If we have detected dead IOC, we needn't proceed
746		 * any more since all PHBs would have been removed
747		 */
748		if (rc == EEH_NEXT_ERR_DEAD_IOC)
749			break;
750	} while (rc != EEH_NEXT_ERR_NONE);
751}
752
753/**
754 * eeh_handle_event - Reset a PCI device after hard lockup.
755 * @pe: EEH PE
756 *
757 * While PHB detects address or data parity errors on particular PCI
758 * slot, the associated PE will be frozen. Besides, DMA's occurring
759 * to wild addresses (which usually happen due to bugs in device
760 * drivers or in PCI adapter firmware) can cause EEH error. #SERR,
761 * #PERR or other misc PCI-related errors also can trigger EEH errors.
762 *
763 * Recovery process consists of unplugging the device driver (which
764 * generated hotplug events to userspace), then issuing a PCI #RST to
765 * the device, then reconfiguring the PCI config space for all bridges
766 * & devices under this slot, and then finally restarting the device
767 * drivers (which cause a second set of hotplug events to go out to
768 * userspace).
769 */
770void eeh_handle_event(struct eeh_pe *pe)
771{
772	if (pe)
773		eeh_handle_normal_event(pe);
774	else
775		eeh_handle_special_event();
776}