1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright IBM Corporation 2001, 2005, 2006
4 * Copyright Dave Engebretsen & Todd Inglett 2001
5 * Copyright Linas Vepstas 2005, 2006
6 * Copyright 2001-2012 IBM Corporation.
8 * Please address comments and feedback to Linas Vepstas <linas@austin.ibm.com>
11 #include <linux/delay.h>
12 #include <linux/sched.h>
13 #include <linux/init.h>
14 #include <linux/list.h>
15 #include <linux/pci.h>
16 #include <linux/iommu.h>
17 #include <linux/proc_fs.h>
18 #include <linux/rbtree.h>
19 #include <linux/reboot.h>
20 #include <linux/seq_file.h>
21 #include <linux/spinlock.h>
22 #include <linux/export.h>
25 #include <linux/atomic.h>
26 #include <asm/debugfs.h>
28 #include <asm/eeh_event.h>
30 #include <asm/iommu.h>
31 #include <asm/machdep.h>
32 #include <asm/ppc-pci.h>
34 #include <asm/pte-walk.h>
38 * EEH, or "Enhanced Error Handling" is a PCI bridge technology for
39 * dealing with PCI bus errors that can't be dealt with within the
40 * usual PCI framework, except by check-stopping the CPU. Systems
41 * that are designed for high-availability/reliability cannot afford
42 * to crash due to a "mere" PCI error, thus the need for EEH.
43 * An EEH-capable bridge operates by converting a detected error
44 * into a "slot freeze", taking the PCI adapter off-line, making
45 * the slot behave, from the OS'es point of view, as if the slot
46 * were "empty": all reads return 0xff's and all writes are silently
47 * ignored. EEH slot isolation events can be triggered by parity
48 * errors on the address or data busses (e.g. during posted writes),
49 * which in turn might be caused by low voltage on the bus, dust,
50 * vibration, humidity, radioactivity or plain-old failed hardware.
52 * Note, however, that one of the leading causes of EEH slot
53 * freeze events are buggy device drivers, buggy device microcode,
54 * or buggy device hardware. This is because any attempt by the
55 * device to bus-master data to a memory address that is not
56 * assigned to the device will trigger a slot freeze. (The idea
57 * is to prevent devices-gone-wild from corrupting system memory).
58 * Buggy hardware/drivers will have a miserable time co-existing
61 * Ideally, a PCI device driver, when suspecting that an isolation
62 * event has occurred (e.g. by reading 0xff's), will then ask EEH
63 * whether this is the case, and then take appropriate steps to
64 * reset the PCI slot, the PCI device, and then resume operations.
65 * However, until that day, the checking is done here, with the
66 * eeh_check_failure() routine embedded in the MMIO macros. If
67 * the slot is found to be isolated, an "EEH Event" is synthesized
68 * and sent out for processing.
71 /* If a device driver keeps reading an MMIO register in an interrupt
72 * handler after a slot isolation event, it might be broken.
73 * This sets the threshold for how many read attempts we allow
74 * before printing an error message.
76 #define EEH_MAX_FAILS 2100000
78 /* Time to wait for a PCI slot to report status, in milliseconds */
79 #define PCI_BUS_RESET_WAIT_MSEC (5*60*1000)
82 * EEH probe mode support, which is part of the flags,
83 * is to support multiple platforms for EEH. Some platforms
84 * like pSeries do PCI emunation based on device tree.
85 * However, other platforms like powernv probe PCI devices
86 * from hardware. The flag is used to distinguish that.
87 * In addition, struct eeh_ops::probe would be invoked for
88 * particular OF node or PCI device so that the corresponding
89 * PE would be created there.
91 int eeh_subsystem_flags;
92 EXPORT_SYMBOL(eeh_subsystem_flags);
95 * EEH allowed maximal frozen times. If one particular PE's
96 * frozen count in last hour exceeds this limit, the PE will
97 * be forced to be offline permanently.
99 u32 eeh_max_freezes = 5;
102 * Controls whether a recovery event should be scheduled when an
103 * isolated device is discovered. This is only really useful for
104 * debugging problems with the EEH core.
106 bool eeh_debugfs_no_recover;
108 /* Platform dependent EEH operations */
109 struct eeh_ops *eeh_ops = NULL;
111 /* Lock to avoid races due to multiple reports of an error */
112 DEFINE_RAW_SPINLOCK(confirm_error_lock);
113 EXPORT_SYMBOL_GPL(confirm_error_lock);
115 /* Lock to protect passed flags */
116 static DEFINE_MUTEX(eeh_dev_mutex);
118 /* Buffer for reporting pci register dumps. Its here in BSS, and
119 * not dynamically alloced, so that it ends up in RMO where RTAS
122 #define EEH_PCI_REGS_LOG_LEN 8192
123 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
126 * The struct is used to maintain the EEH global statistic
127 * information. Besides, the EEH global statistics will be
128 * exported to user space through procfs
131 u64 no_device; /* PCI device not found */
132 u64 no_dn; /* OF node not found */
133 u64 no_cfg_addr; /* Config address not found */
134 u64 ignored_check; /* EEH check skipped */
135 u64 total_mmio_ffs; /* Total EEH checks */
136 u64 false_positives; /* Unnecessary EEH checks */
137 u64 slot_resets; /* PE reset */
140 static struct eeh_stats eeh_stats;
142 static int __init eeh_setup(char *str)
144 if (!strcmp(str, "off"))
145 eeh_add_flag(EEH_FORCE_DISABLED);
146 else if (!strcmp(str, "early_log"))
147 eeh_add_flag(EEH_EARLY_DUMP_LOG);
151 __setup("eeh=", eeh_setup);
154 * This routine captures assorted PCI configuration space data
155 * for the indicated PCI device, and puts them into a buffer
156 * for RTAS error logging.
158 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
160 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
167 pr_warn("EEH: Note: No error log for absent device.\n");
171 n += scnprintf(buf+n, len-n, "%04x:%02x:%02x.%01x\n",
172 pdn->phb->global_number, pdn->busno,
173 PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
174 pr_warn("EEH: of node=%04x:%02x:%02x.%01x\n",
175 pdn->phb->global_number, pdn->busno,
176 PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
178 eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
179 n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
180 pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
182 eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
183 n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
184 pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
186 /* Gather bridge-specific registers */
187 if (edev->mode & EEH_DEV_BRIDGE) {
188 eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
189 n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
190 pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
192 eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
193 n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
194 pr_warn("EEH: Bridge control: %04x\n", cfg);
197 /* Dump out the PCI-X command and status regs */
198 cap = edev->pcix_cap;
200 eeh_ops->read_config(pdn, cap, 4, &cfg);
201 n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
202 pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
204 eeh_ops->read_config(pdn, cap+4, 4, &cfg);
205 n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
206 pr_warn("EEH: PCI-X status: %08x\n", cfg);
209 /* If PCI-E capable, dump PCI-E cap 10 */
210 cap = edev->pcie_cap;
212 n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
213 pr_warn("EEH: PCI-E capabilities and status follow:\n");
215 for (i=0; i<=8; i++) {
216 eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
217 n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
221 pr_warn("%s\n", buffer);
223 l = scnprintf(buffer, sizeof(buffer),
224 "EEH: PCI-E %02x: %08x ",
227 l += scnprintf(buffer+l, sizeof(buffer)-l,
233 pr_warn("%s\n", buffer);
236 /* If AER capable, dump it */
239 n += scnprintf(buf+n, len-n, "pci-e AER:\n");
240 pr_warn("EEH: PCI-E AER capability register set follows:\n");
242 for (i=0; i<=13; i++) {
243 eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
244 n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
248 pr_warn("%s\n", buffer);
250 l = scnprintf(buffer, sizeof(buffer),
251 "EEH: PCI-E AER %02x: %08x ",
254 l += scnprintf(buffer+l, sizeof(buffer)-l,
259 pr_warn("%s\n", buffer);
265 static void *eeh_dump_pe_log(struct eeh_pe *pe, void *flag)
267 struct eeh_dev *edev, *tmp;
270 eeh_pe_for_each_dev(pe, edev, tmp)
271 *plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
272 EEH_PCI_REGS_LOG_LEN - *plen);
278 * eeh_slot_error_detail - Generate combined log including driver log and error log
280 * @severity: temporary or permanent error log
282 * This routine should be called to generate the combined log, which
283 * is comprised of driver log and error log. The driver log is figured
284 * out from the config space of the corresponding PCI device, while
285 * the error log is fetched through platform dependent function call.
287 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
292 * When the PHB is fenced or dead, it's pointless to collect
293 * the data from PCI config space because it should return
294 * 0xFF's. For ER, we still retrieve the data from the PCI
297 * For pHyp, we have to enable IO for log retrieval. Otherwise,
298 * 0xFF's is always returned from PCI config space.
300 * When the @severity is EEH_LOG_PERM, the PE is going to be
301 * removed. Prior to that, the drivers for devices included in
302 * the PE will be closed. The drivers rely on working IO path
303 * to bring the devices to quiet state. Otherwise, PCI traffic
304 * from those devices after they are removed is like to cause
305 * another unexpected EEH error.
307 if (!(pe->type & EEH_PE_PHB)) {
308 if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG) ||
309 severity == EEH_LOG_PERM)
310 eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
313 * The config space of some PCI devices can't be accessed
314 * when their PEs are in frozen state. Otherwise, fenced
315 * PHB might be seen. Those PEs are identified with flag
316 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
317 * is set automatically when the PE is put to EEH_PE_ISOLATED.
319 * Restoring BARs possibly triggers PCI config access in
320 * (OPAL) firmware and then causes fenced PHB. If the
321 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
322 * pointless to restore BARs and dump config space.
324 eeh_ops->configure_bridge(pe);
325 if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
326 eeh_pe_restore_bars(pe);
329 eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
333 eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
337 * eeh_token_to_phys - Convert EEH address token to phys address
338 * @token: I/O token, should be address in the form 0xA....
340 * This routine should be called to convert virtual I/O address
343 static inline unsigned long eeh_token_to_phys(unsigned long token)
350 * We won't find hugepages here(this is iomem). Hence we are not
351 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
352 * page table free, because of init_mm.
354 ptep = find_init_mm_pte(token, &hugepage_shift);
360 /* On radix we can do hugepage mappings for io, so handle that */
361 if (hugepage_shift) {
362 pa <<= hugepage_shift;
363 pa |= token & ((1ul << hugepage_shift) - 1);
366 pa |= token & (PAGE_SIZE - 1);
373 * On PowerNV platform, we might already have fenced PHB there.
374 * For that case, it's meaningless to recover frozen PE. Intead,
375 * We have to handle fenced PHB firstly.
377 static int eeh_phb_check_failure(struct eeh_pe *pe)
379 struct eeh_pe *phb_pe;
383 if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
386 /* Find the PHB PE */
387 phb_pe = eeh_phb_pe_get(pe->phb);
389 pr_warn("%s Can't find PE for PHB#%x\n",
390 __func__, pe->phb->global_number);
394 /* If the PHB has been in problematic state */
395 eeh_serialize_lock(&flags);
396 if (phb_pe->state & EEH_PE_ISOLATED) {
401 /* Check PHB state */
402 ret = eeh_ops->get_state(phb_pe, NULL);
404 (ret == EEH_STATE_NOT_SUPPORT) || eeh_state_active(ret)) {
409 /* Isolate the PHB and send event */
410 eeh_pe_mark_isolated(phb_pe);
411 eeh_serialize_unlock(flags);
413 pr_err("EEH: PHB#%x failure detected, location: %s\n",
414 phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
416 eeh_send_failure_event(phb_pe);
420 eeh_serialize_unlock(flags);
425 * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
428 * Check for an EEH failure for the given device node. Call this
429 * routine if the result of a read was all 0xff's and you want to
430 * find out if this is due to an EEH slot freeze. This routine
431 * will query firmware for the EEH status.
433 * Returns 0 if there has not been an EEH error; otherwise returns
434 * a non-zero value and queues up a slot isolation event notification.
436 * It is safe to call this routine in an interrupt context.
438 int eeh_dev_check_failure(struct eeh_dev *edev)
442 struct device_node *dn;
444 struct eeh_pe *pe, *parent_pe, *phb_pe;
446 const char *location = NULL;
448 eeh_stats.total_mmio_ffs++;
457 dev = eeh_dev_to_pci_dev(edev);
458 pe = eeh_dev_to_pe(edev);
460 /* Access to IO BARs might get this far and still not want checking. */
462 eeh_stats.ignored_check++;
463 pr_debug("EEH: Ignored check for %s\n",
468 if (!pe->addr && !pe->config_addr) {
469 eeh_stats.no_cfg_addr++;
474 * On PowerNV platform, we might already have fenced PHB
475 * there and we need take care of that firstly.
477 ret = eeh_phb_check_failure(pe);
482 * If the PE isn't owned by us, we shouldn't check the
483 * state. Instead, let the owner handle it if the PE has
486 if (eeh_pe_passed(pe))
489 /* If we already have a pending isolation event for this
490 * slot, we know it's bad already, we don't need to check.
491 * Do this checking under a lock; as multiple PCI devices
492 * in one slot might report errors simultaneously, and we
493 * only want one error recovery routine running.
495 eeh_serialize_lock(&flags);
497 if (pe->state & EEH_PE_ISOLATED) {
499 if (pe->check_count % EEH_MAX_FAILS == 0) {
500 dn = pci_device_to_OF_node(dev);
502 location = of_get_property(dn, "ibm,loc-code",
504 printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
505 "location=%s driver=%s pci addr=%s\n",
507 location ? location : "unknown",
508 eeh_driver_name(dev), eeh_pci_name(dev));
509 printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
510 eeh_driver_name(dev));
517 * Now test for an EEH failure. This is VERY expensive.
518 * Note that the eeh_config_addr may be a parent device
519 * in the case of a device behind a bridge, or it may be
520 * function zero of a multi-function device.
521 * In any case they must share a common PHB.
523 ret = eeh_ops->get_state(pe, NULL);
525 /* Note that config-io to empty slots may fail;
526 * they are empty when they don't have children.
527 * We will punt with the following conditions: Failure to get
528 * PE's state, EEH not support and Permanently unavailable
529 * state, PE is in good state.
532 (ret == EEH_STATE_NOT_SUPPORT) || eeh_state_active(ret)) {
533 eeh_stats.false_positives++;
534 pe->false_positives++;
540 * It should be corner case that the parent PE has been
541 * put into frozen state as well. We should take care
544 parent_pe = pe->parent;
546 /* Hit the ceiling ? */
547 if (parent_pe->type & EEH_PE_PHB)
550 /* Frozen parent PE ? */
551 ret = eeh_ops->get_state(parent_pe, NULL);
552 if (ret > 0 && !eeh_state_active(ret)) {
554 pr_err("EEH: Failure of PHB#%x-PE#%x will be handled at parent PHB#%x-PE#%x.\n",
555 pe->phb->global_number, pe->addr,
556 pe->phb->global_number, parent_pe->addr);
559 /* Next parent level */
560 parent_pe = parent_pe->parent;
563 eeh_stats.slot_resets++;
565 /* Avoid repeated reports of this failure, including problems
566 * with other functions on this device, and functions under
569 eeh_pe_mark_isolated(pe);
570 eeh_serialize_unlock(flags);
572 /* Most EEH events are due to device driver bugs. Having
573 * a stack trace will help the device-driver authors figure
574 * out what happened. So print that out.
576 phb_pe = eeh_phb_pe_get(pe->phb);
577 pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
578 pe->phb->global_number, pe->addr);
579 pr_err("EEH: PE location: %s, PHB location: %s\n",
580 eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
583 eeh_send_failure_event(pe);
588 eeh_serialize_unlock(flags);
592 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
595 * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
596 * @token: I/O address
598 * Check for an EEH failure at the given I/O address. Call this
599 * routine if the result of a read was all 0xff's and you want to
600 * find out if this is due to an EEH slot freeze event. This routine
601 * will query firmware for the EEH status.
603 * Note this routine is safe to call in an interrupt context.
605 int eeh_check_failure(const volatile void __iomem *token)
608 struct eeh_dev *edev;
610 /* Finding the phys addr + pci device; this is pretty quick. */
611 addr = eeh_token_to_phys((unsigned long __force) token);
612 edev = eeh_addr_cache_get_dev(addr);
614 eeh_stats.no_device++;
618 return eeh_dev_check_failure(edev);
620 EXPORT_SYMBOL(eeh_check_failure);
624 * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
627 * This routine should be called to reenable frozen MMIO or DMA
628 * so that it would work correctly again. It's useful while doing
629 * recovery or log collection on the indicated device.
631 int eeh_pci_enable(struct eeh_pe *pe, int function)
636 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
637 * Also, it's pointless to enable them on unfrozen PE. So
638 * we have to check before enabling IO or DMA.
641 case EEH_OPT_THAW_MMIO:
642 active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
644 case EEH_OPT_THAW_DMA:
645 active_flag = EEH_STATE_DMA_ACTIVE;
647 case EEH_OPT_DISABLE:
649 case EEH_OPT_FREEZE_PE:
653 pr_warn("%s: Invalid function %d\n",
659 * Check if IO or DMA has been enabled before
663 rc = eeh_ops->get_state(pe, NULL);
667 /* Needn't enable it at all */
668 if (rc == EEH_STATE_NOT_SUPPORT)
671 /* It's already enabled */
672 if (rc & active_flag)
677 /* Issue the request */
678 rc = eeh_ops->set_option(pe, function);
680 pr_warn("%s: Unexpected state change %d on "
681 "PHB#%x-PE#%x, err=%d\n",
682 __func__, function, pe->phb->global_number,
685 /* Check if the request is finished successfully */
687 rc = eeh_wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
691 if (rc & active_flag)
700 static void *eeh_disable_and_save_dev_state(struct eeh_dev *edev,
703 struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
704 struct pci_dev *dev = userdata;
707 * The caller should have disabled and saved the
708 * state for the specified device
710 if (!pdev || pdev == dev)
713 /* Ensure we have D0 power state */
714 pci_set_power_state(pdev, PCI_D0);
716 /* Save device state */
717 pci_save_state(pdev);
720 * Disable device to avoid any DMA traffic and
721 * interrupt from the device
723 pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
728 static void *eeh_restore_dev_state(struct eeh_dev *edev, void *userdata)
730 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
731 struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
732 struct pci_dev *dev = userdata;
737 /* Apply customization from firmware */
738 if (pdn && eeh_ops->restore_config)
739 eeh_ops->restore_config(pdn);
741 /* The caller should restore state for the specified device */
743 pci_restore_state(pdev);
748 int eeh_restore_vf_config(struct pci_dn *pdn)
750 struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
751 u32 devctl, cmd, cap2, aer_capctl;
754 if (edev->pcie_cap) {
756 old_mps = (ffs(pdn->mps) - 8) << 5;
757 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
759 devctl &= ~PCI_EXP_DEVCTL_PAYLOAD;
761 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
764 /* Disable Completion Timeout if possible */
765 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP2,
767 if (cap2 & PCI_EXP_DEVCAP2_COMP_TMOUT_DIS) {
768 eeh_ops->read_config(pdn,
769 edev->pcie_cap + PCI_EXP_DEVCTL2,
771 cap2 |= PCI_EXP_DEVCTL2_COMP_TMOUT_DIS;
772 eeh_ops->write_config(pdn,
773 edev->pcie_cap + PCI_EXP_DEVCTL2,
778 /* Enable SERR and parity checking */
779 eeh_ops->read_config(pdn, PCI_COMMAND, 2, &cmd);
780 cmd |= (PCI_COMMAND_PARITY | PCI_COMMAND_SERR);
781 eeh_ops->write_config(pdn, PCI_COMMAND, 2, cmd);
783 /* Enable report various errors */
784 if (edev->pcie_cap) {
785 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
787 devctl &= ~PCI_EXP_DEVCTL_CERE;
788 devctl |= (PCI_EXP_DEVCTL_NFERE |
789 PCI_EXP_DEVCTL_FERE |
790 PCI_EXP_DEVCTL_URRE);
791 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
795 /* Enable ECRC generation and check */
796 if (edev->pcie_cap && edev->aer_cap) {
797 eeh_ops->read_config(pdn, edev->aer_cap + PCI_ERR_CAP,
799 aer_capctl |= (PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE);
800 eeh_ops->write_config(pdn, edev->aer_cap + PCI_ERR_CAP,
808 * pcibios_set_pcie_reset_state - Set PCI-E reset state
809 * @dev: pci device struct
810 * @state: reset state to enter
815 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
817 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
818 struct eeh_pe *pe = eeh_dev_to_pe(edev);
821 pr_err("%s: No PE found on PCI device %s\n",
822 __func__, pci_name(dev));
827 case pcie_deassert_reset:
828 eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
830 if (!(pe->type & EEH_PE_VF))
831 eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED, true);
832 eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
833 eeh_pe_state_clear(pe, EEH_PE_ISOLATED, true);
836 eeh_pe_mark_isolated(pe);
837 eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED, true);
838 eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
839 eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
840 if (!(pe->type & EEH_PE_VF))
841 eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
842 eeh_ops->reset(pe, EEH_RESET_HOT);
844 case pcie_warm_reset:
845 eeh_pe_mark_isolated(pe);
846 eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED, true);
847 eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
848 eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
849 if (!(pe->type & EEH_PE_VF))
850 eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
851 eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
854 eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED, true);
862 * eeh_set_pe_freset - Check the required reset for the indicated device
864 * @flag: return value
866 * Each device might have its preferred reset type: fundamental or
867 * hot reset. The routine is used to collected the information for
868 * the indicated device and its children so that the bunch of the
869 * devices could be reset properly.
871 static void *eeh_set_dev_freset(struct eeh_dev *edev, void *flag)
874 unsigned int *freset = (unsigned int *)flag;
876 dev = eeh_dev_to_pci_dev(edev);
878 *freset |= dev->needs_freset;
883 static void eeh_pe_refreeze_passed(struct eeh_pe *root)
888 eeh_for_each_pe(root, pe) {
889 if (eeh_pe_passed(pe)) {
890 state = eeh_ops->get_state(pe, NULL);
892 (EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED)) {
893 pr_info("EEH: Passed-through PE PHB#%x-PE#%x was thawed by reset, re-freezing for safety.\n",
894 pe->phb->global_number, pe->addr);
895 eeh_pe_set_option(pe, EEH_OPT_FREEZE_PE);
902 * eeh_pe_reset_full - Complete a full reset process on the indicated PE
905 * This function executes a full reset procedure on a PE, including setting
906 * the appropriate flags, performing a fundamental or hot reset, and then
907 * deactivating the reset status. It is designed to be used within the EEH
908 * subsystem, as opposed to eeh_pe_reset which is exported to drivers and
909 * only performs a single operation at a time.
911 * This function will attempt to reset a PE three times before failing.
913 int eeh_pe_reset_full(struct eeh_pe *pe, bool include_passed)
915 int reset_state = (EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
916 int type = EEH_RESET_HOT;
917 unsigned int freset = 0;
918 int i, state = 0, ret;
921 * Determine the type of reset to perform - hot or fundamental.
922 * Hot reset is the default operation, unless any device under the
923 * PE requires a fundamental reset.
925 eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
928 type = EEH_RESET_FUNDAMENTAL;
930 /* Mark the PE as in reset state and block config space accesses */
931 eeh_pe_state_mark(pe, reset_state);
933 /* Make three attempts at resetting the bus */
934 for (i = 0; i < 3; i++) {
935 ret = eeh_pe_reset(pe, type, include_passed);
937 ret = eeh_pe_reset(pe, EEH_RESET_DEACTIVATE,
941 pr_warn("EEH: Failure %d resetting PHB#%x-PE#%x (attempt %d)\n\n",
942 state, pe->phb->global_number, pe->addr, i + 1);
946 pr_warn("EEH: PHB#%x-PE#%x: Successful reset (attempt %d)\n",
947 pe->phb->global_number, pe->addr, i + 1);
949 /* Wait until the PE is in a functioning state */
950 state = eeh_wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
952 pr_warn("EEH: Unrecoverable slot failure on PHB#%x-PE#%x",
953 pe->phb->global_number, pe->addr);
954 ret = -ENOTRECOVERABLE;
957 if (eeh_state_active(state))
960 pr_warn("EEH: PHB#%x-PE#%x: Slot inactive after reset: 0x%x (attempt %d)\n",
961 pe->phb->global_number, pe->addr, state, i + 1);
964 /* Resetting the PE may have unfrozen child PEs. If those PEs have been
965 * (potentially) passed through to a guest, re-freeze them:
968 eeh_pe_refreeze_passed(pe);
970 eeh_pe_state_clear(pe, reset_state, true);
975 * eeh_save_bars - Save device bars
976 * @edev: PCI device associated EEH device
978 * Save the values of the device bars. Unlike the restore
979 * routine, this routine is *not* recursive. This is because
980 * PCI devices are added individually; but, for the restore,
981 * an entire slot is reset at a time.
983 void eeh_save_bars(struct eeh_dev *edev)
988 pdn = eeh_dev_to_pdn(edev);
992 for (i = 0; i < 16; i++)
993 eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
996 * For PCI bridges including root port, we need enable bus
997 * master explicitly. Otherwise, it can't fetch IODA table
998 * entries correctly. So we cache the bit in advance so that
999 * we can restore it after reset, either PHB range or PE range.
1001 if (edev->mode & EEH_DEV_BRIDGE)
1002 edev->config_space[1] |= PCI_COMMAND_MASTER;
1006 * eeh_ops_register - Register platform dependent EEH operations
1007 * @ops: platform dependent EEH operations
1009 * Register the platform dependent EEH operation callback
1010 * functions. The platform should call this function before
1011 * any other EEH operations.
1013 int __init eeh_ops_register(struct eeh_ops *ops)
1016 pr_warn("%s: Invalid EEH ops name for %p\n",
1021 if (eeh_ops && eeh_ops != ops) {
1022 pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
1023 __func__, eeh_ops->name, ops->name);
1033 * eeh_ops_unregister - Unreigster platform dependent EEH operations
1034 * @name: name of EEH platform operations
1036 * Unregister the platform dependent EEH operation callback
1039 int __exit eeh_ops_unregister(const char *name)
1041 if (!name || !strlen(name)) {
1042 pr_warn("%s: Invalid EEH ops name\n",
1047 if (eeh_ops && !strcmp(eeh_ops->name, name)) {
1055 static int eeh_reboot_notifier(struct notifier_block *nb,
1056 unsigned long action, void *unused)
1058 eeh_clear_flag(EEH_ENABLED);
1062 static struct notifier_block eeh_reboot_nb = {
1063 .notifier_call = eeh_reboot_notifier,
1066 void eeh_probe_devices(void)
1068 struct pci_controller *hose, *tmp;
1071 /* Enable EEH for all adapters */
1072 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1073 pdn = hose->pci_data;
1074 traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1077 pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1079 pr_info("EEH: No capable adapters found\n");
1084 * eeh_init - EEH initialization
1086 * Initialize EEH by trying to enable it for all of the adapters in the system.
1087 * As a side effect we can determine here if eeh is supported at all.
1088 * Note that we leave EEH on so failed config cycles won't cause a machine
1089 * check. If a user turns off EEH for a particular adapter they are really
1090 * telling Linux to ignore errors. Some hardware (e.g. POWER5) won't
1091 * grant access to a slot if EEH isn't enabled, and so we always enable
1092 * EEH for all slots/all devices.
1094 * The eeh-force-off option disables EEH checking globally, for all slots.
1095 * Even if force-off is set, the EEH hardware is still enabled, so that
1096 * newer systems can boot.
1098 static int eeh_init(void)
1100 struct pci_controller *hose, *tmp;
1103 /* Register reboot notifier */
1104 ret = register_reboot_notifier(&eeh_reboot_nb);
1106 pr_warn("%s: Failed to register notifier (%d)\n",
1111 /* call platform initialization function */
1113 pr_warn("%s: Platform EEH operation not found\n",
1116 } else if ((ret = eeh_ops->init()))
1119 /* Initialize PHB PEs */
1120 list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
1121 eeh_dev_phb_init_dynamic(hose);
1123 /* Initialize EEH event */
1124 return eeh_event_init();
1127 core_initcall_sync(eeh_init);
1130 * eeh_add_device_early - Enable EEH for the indicated device node
1131 * @pdn: PCI device node for which to set up EEH
1133 * This routine must be used to perform EEH initialization for PCI
1134 * devices that were added after system boot (e.g. hotplug, dlpar).
1135 * This routine must be called before any i/o is performed to the
1136 * adapter (inluding any config-space i/o).
1137 * Whether this actually enables EEH or not for this device depends
1138 * on the CEC architecture, type of the device, on earlier boot
1139 * command-line arguments & etc.
1141 void eeh_add_device_early(struct pci_dn *pdn)
1143 struct pci_controller *phb = pdn ? pdn->phb : NULL;
1144 struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1149 if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1152 /* USB Bus children of PCI devices will not have BUID's */
1154 (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1157 eeh_ops->probe(pdn, NULL);
1161 * eeh_add_device_tree_early - Enable EEH for the indicated device
1162 * @pdn: PCI device node
1164 * This routine must be used to perform EEH initialization for the
1165 * indicated PCI device that was added after system boot (e.g.
1168 void eeh_add_device_tree_early(struct pci_dn *pdn)
1175 list_for_each_entry(n, &pdn->child_list, list)
1176 eeh_add_device_tree_early(n);
1177 eeh_add_device_early(pdn);
1179 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1182 * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1183 * @dev: pci device for which to set up EEH
1185 * This routine must be used to complete EEH initialization for PCI
1186 * devices that were added after system boot (e.g. hotplug, dlpar).
1188 void eeh_add_device_late(struct pci_dev *dev)
1191 struct eeh_dev *edev;
1193 if (!dev || !eeh_enabled())
1196 pr_debug("EEH: Adding device %s\n", pci_name(dev));
1198 pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1199 edev = pdn_to_eeh_dev(pdn);
1200 if (edev->pdev == dev) {
1201 pr_debug("EEH: Already referenced !\n");
1206 * The EEH cache might not be removed correctly because of
1207 * unbalanced kref to the device during unplug time, which
1208 * relies on pcibios_release_device(). So we have to remove
1209 * that here explicitly.
1212 eeh_rmv_from_parent_pe(edev);
1213 eeh_addr_cache_rmv_dev(edev->pdev);
1214 eeh_sysfs_remove_device(edev->pdev);
1215 edev->mode &= ~EEH_DEV_SYSFS;
1218 * We definitely should have the PCI device removed
1219 * though it wasn't correctly. So we needn't call
1220 * into error handler afterwards.
1222 edev->mode |= EEH_DEV_NO_HANDLER;
1225 dev->dev.archdata.edev = NULL;
1228 if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1229 eeh_ops->probe(pdn, NULL);
1232 dev->dev.archdata.edev = edev;
1234 eeh_addr_cache_insert_dev(dev);
1238 * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1241 * This routine must be used to perform EEH initialization for PCI
1242 * devices which are attached to the indicated PCI bus. The PCI bus
1243 * is added after system boot through hotplug or dlpar.
1245 void eeh_add_device_tree_late(struct pci_bus *bus)
1247 struct pci_dev *dev;
1249 list_for_each_entry(dev, &bus->devices, bus_list) {
1250 eeh_add_device_late(dev);
1251 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1252 struct pci_bus *subbus = dev->subordinate;
1254 eeh_add_device_tree_late(subbus);
1258 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1261 * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1264 * This routine must be used to add EEH sysfs files for PCI
1265 * devices which are attached to the indicated PCI bus. The PCI bus
1266 * is added after system boot through hotplug or dlpar.
1268 void eeh_add_sysfs_files(struct pci_bus *bus)
1270 struct pci_dev *dev;
1272 list_for_each_entry(dev, &bus->devices, bus_list) {
1273 eeh_sysfs_add_device(dev);
1274 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1275 struct pci_bus *subbus = dev->subordinate;
1277 eeh_add_sysfs_files(subbus);
1281 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1284 * eeh_remove_device - Undo EEH setup for the indicated pci device
1285 * @dev: pci device to be removed
1287 * This routine should be called when a device is removed from
1288 * a running system (e.g. by hotplug or dlpar). It unregisters
1289 * the PCI device from the EEH subsystem. I/O errors affecting
1290 * this device will no longer be detected after this call; thus,
1291 * i/o errors affecting this slot may leave this device unusable.
1293 void eeh_remove_device(struct pci_dev *dev)
1295 struct eeh_dev *edev;
1297 if (!dev || !eeh_enabled())
1299 edev = pci_dev_to_eeh_dev(dev);
1301 /* Unregister the device with the EEH/PCI address search system */
1302 pr_debug("EEH: Removing device %s\n", pci_name(dev));
1304 if (!edev || !edev->pdev || !edev->pe) {
1305 pr_debug("EEH: Not referenced !\n");
1310 * During the hotplug for EEH error recovery, we need the EEH
1311 * device attached to the parent PE in order for BAR restore
1312 * a bit later. So we keep it for BAR restore and remove it
1313 * from the parent PE during the BAR resotre.
1318 * The flag "in_error" is used to trace EEH devices for VFs
1319 * in error state or not. It's set in eeh_report_error(). If
1320 * it's not set, eeh_report_{reset,resume}() won't be called
1321 * for the VF EEH device.
1323 edev->in_error = false;
1324 dev->dev.archdata.edev = NULL;
1325 if (!(edev->pe->state & EEH_PE_KEEP))
1326 eeh_rmv_from_parent_pe(edev);
1328 edev->mode |= EEH_DEV_DISCONNECTED;
1331 * We're removing from the PCI subsystem, that means
1332 * the PCI device driver can't support EEH or not
1333 * well. So we rely on hotplug completely to do recovery
1334 * for the specific PCI device.
1336 edev->mode |= EEH_DEV_NO_HANDLER;
1338 eeh_addr_cache_rmv_dev(dev);
1339 eeh_sysfs_remove_device(dev);
1340 edev->mode &= ~EEH_DEV_SYSFS;
1343 int eeh_unfreeze_pe(struct eeh_pe *pe)
1347 ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1349 pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1350 __func__, ret, pe->phb->global_number, pe->addr);
1354 ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1356 pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1357 __func__, ret, pe->phb->global_number, pe->addr);
1365 static struct pci_device_id eeh_reset_ids[] = {
1366 { PCI_DEVICE(0x19a2, 0x0710) }, /* Emulex, BE */
1367 { PCI_DEVICE(0x10df, 0xe220) }, /* Emulex, Lancer */
1368 { PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1372 static int eeh_pe_change_owner(struct eeh_pe *pe)
1374 struct eeh_dev *edev, *tmp;
1375 struct pci_dev *pdev;
1376 struct pci_device_id *id;
1379 /* Check PE state */
1380 ret = eeh_ops->get_state(pe, NULL);
1381 if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1384 /* Unfrozen PE, nothing to do */
1385 if (eeh_state_active(ret))
1388 /* Frozen PE, check if it needs PE level reset */
1389 eeh_pe_for_each_dev(pe, edev, tmp) {
1390 pdev = eeh_dev_to_pci_dev(edev);
1394 for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1395 if (id->vendor != PCI_ANY_ID &&
1396 id->vendor != pdev->vendor)
1398 if (id->device != PCI_ANY_ID &&
1399 id->device != pdev->device)
1401 if (id->subvendor != PCI_ANY_ID &&
1402 id->subvendor != pdev->subsystem_vendor)
1404 if (id->subdevice != PCI_ANY_ID &&
1405 id->subdevice != pdev->subsystem_device)
1408 return eeh_pe_reset_and_recover(pe);
1412 ret = eeh_unfreeze_pe(pe);
1414 eeh_pe_state_clear(pe, EEH_PE_ISOLATED, true);
1419 * eeh_dev_open - Increase count of pass through devices for PE
1422 * Increase count of passed through devices for the indicated
1423 * PE. In the result, the EEH errors detected on the PE won't be
1424 * reported. The PE owner will be responsible for detection
1427 int eeh_dev_open(struct pci_dev *pdev)
1429 struct eeh_dev *edev;
1432 mutex_lock(&eeh_dev_mutex);
1434 /* No PCI device ? */
1438 /* No EEH device or PE ? */
1439 edev = pci_dev_to_eeh_dev(pdev);
1440 if (!edev || !edev->pe)
1444 * The PE might have been put into frozen state, but we
1445 * didn't detect that yet. The passed through PCI devices
1446 * in frozen PE won't work properly. Clear the frozen state
1449 ret = eeh_pe_change_owner(edev->pe);
1453 /* Increase PE's pass through count */
1454 atomic_inc(&edev->pe->pass_dev_cnt);
1455 mutex_unlock(&eeh_dev_mutex);
1459 mutex_unlock(&eeh_dev_mutex);
1462 EXPORT_SYMBOL_GPL(eeh_dev_open);
1465 * eeh_dev_release - Decrease count of pass through devices for PE
1468 * Decrease count of pass through devices for the indicated PE. If
1469 * there is no passed through device in PE, the EEH errors detected
1470 * on the PE will be reported and handled as usual.
1472 void eeh_dev_release(struct pci_dev *pdev)
1474 struct eeh_dev *edev;
1476 mutex_lock(&eeh_dev_mutex);
1478 /* No PCI device ? */
1482 /* No EEH device ? */
1483 edev = pci_dev_to_eeh_dev(pdev);
1484 if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1487 /* Decrease PE's pass through count */
1488 WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1489 eeh_pe_change_owner(edev->pe);
1491 mutex_unlock(&eeh_dev_mutex);
1493 EXPORT_SYMBOL(eeh_dev_release);
1495 #ifdef CONFIG_IOMMU_API
1497 static int dev_has_iommu_table(struct device *dev, void *data)
1499 struct pci_dev *pdev = to_pci_dev(dev);
1500 struct pci_dev **ppdev = data;
1505 if (device_iommu_mapped(dev)) {
1514 * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1515 * @group: IOMMU group
1517 * The routine is called to convert IOMMU group to EEH PE.
1519 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1521 struct pci_dev *pdev = NULL;
1522 struct eeh_dev *edev;
1525 /* No IOMMU group ? */
1529 ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1533 /* No EEH device or PE ? */
1534 edev = pci_dev_to_eeh_dev(pdev);
1535 if (!edev || !edev->pe)
1540 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1542 #endif /* CONFIG_IOMMU_API */
1545 * eeh_pe_set_option - Set options for the indicated PE
1547 * @option: requested option
1549 * The routine is called to enable or disable EEH functionality
1550 * on the indicated PE, to enable IO or DMA for the frozen PE.
1552 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1561 * EEH functionality could possibly be disabled, just
1562 * return error for the case. And the EEH functinality
1563 * isn't expected to be disabled on one specific PE.
1566 case EEH_OPT_ENABLE:
1567 if (eeh_enabled()) {
1568 ret = eeh_pe_change_owner(pe);
1573 case EEH_OPT_DISABLE:
1575 case EEH_OPT_THAW_MMIO:
1576 case EEH_OPT_THAW_DMA:
1577 case EEH_OPT_FREEZE_PE:
1578 if (!eeh_ops || !eeh_ops->set_option) {
1583 ret = eeh_pci_enable(pe, option);
1586 pr_debug("%s: Option %d out of range (%d, %d)\n",
1587 __func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1593 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1596 * eeh_pe_get_state - Retrieve PE's state
1599 * Retrieve the PE's state, which includes 3 aspects: enabled
1600 * DMA, enabled IO and asserted reset.
1602 int eeh_pe_get_state(struct eeh_pe *pe)
1604 int result, ret = 0;
1605 bool rst_active, dma_en, mmio_en;
1611 if (!eeh_ops || !eeh_ops->get_state)
1615 * If the parent PE is owned by the host kernel and is undergoing
1616 * error recovery, we should return the PE state as temporarily
1617 * unavailable so that the error recovery on the guest is suspended
1618 * until the recovery completes on the host.
1621 !(pe->state & EEH_PE_REMOVED) &&
1622 (pe->parent->state & (EEH_PE_ISOLATED | EEH_PE_RECOVERING)))
1623 return EEH_PE_STATE_UNAVAIL;
1625 result = eeh_ops->get_state(pe, NULL);
1626 rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1627 dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1628 mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1631 ret = EEH_PE_STATE_RESET;
1632 else if (dma_en && mmio_en)
1633 ret = EEH_PE_STATE_NORMAL;
1634 else if (!dma_en && !mmio_en)
1635 ret = EEH_PE_STATE_STOPPED_IO_DMA;
1636 else if (!dma_en && mmio_en)
1637 ret = EEH_PE_STATE_STOPPED_DMA;
1639 ret = EEH_PE_STATE_UNAVAIL;
1643 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1645 static int eeh_pe_reenable_devices(struct eeh_pe *pe, bool include_passed)
1647 struct eeh_dev *edev, *tmp;
1648 struct pci_dev *pdev;
1651 eeh_pe_restore_bars(pe);
1654 * Reenable PCI devices as the devices passed
1655 * through are always enabled before the reset.
1657 eeh_pe_for_each_dev(pe, edev, tmp) {
1658 pdev = eeh_dev_to_pci_dev(edev);
1662 ret = pci_reenable_device(pdev);
1664 pr_warn("%s: Failure %d reenabling %s\n",
1665 __func__, ret, pci_name(pdev));
1670 /* The PE is still in frozen state */
1671 if (include_passed || !eeh_pe_passed(pe)) {
1672 ret = eeh_unfreeze_pe(pe);
1674 pr_info("EEH: Note: Leaving passthrough PHB#%x-PE#%x frozen.\n",
1675 pe->phb->global_number, pe->addr);
1677 eeh_pe_state_clear(pe, EEH_PE_ISOLATED, include_passed);
1683 * eeh_pe_reset - Issue PE reset according to specified type
1685 * @option: reset type
1687 * The routine is called to reset the specified PE with the
1688 * indicated type, either fundamental reset or hot reset.
1689 * PE reset is the most important part for error recovery.
1691 int eeh_pe_reset(struct eeh_pe *pe, int option, bool include_passed)
1699 if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1703 case EEH_RESET_DEACTIVATE:
1704 ret = eeh_ops->reset(pe, option);
1705 eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED, include_passed);
1709 ret = eeh_pe_reenable_devices(pe, include_passed);
1712 case EEH_RESET_FUNDAMENTAL:
1714 * Proactively freeze the PE to drop all MMIO access
1715 * during reset, which should be banned as it's always
1716 * cause recursive EEH error.
1718 eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1720 eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1721 ret = eeh_ops->reset(pe, option);
1724 pr_debug("%s: Unsupported option %d\n",
1731 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1734 * eeh_pe_configure - Configure PCI bridges after PE reset
1737 * The routine is called to restore the PCI config space for
1738 * those PCI devices, especially PCI bridges affected by PE
1739 * reset issued previously.
1741 int eeh_pe_configure(struct eeh_pe *pe)
1751 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1754 * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1755 * @pe: the indicated PE
1757 * @function: error function
1759 * @mask: address mask
1761 * The routine is called to inject the specified PCI error, which
1762 * is determined by @type and @function, to the indicated PE for
1765 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1766 unsigned long addr, unsigned long mask)
1772 /* Unsupported operation ? */
1773 if (!eeh_ops || !eeh_ops->err_inject)
1776 /* Check on PCI error type */
1777 if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1780 /* Check on PCI error function */
1781 if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1784 return eeh_ops->err_inject(pe, type, func, addr, mask);
1786 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1788 static int proc_eeh_show(struct seq_file *m, void *v)
1790 if (!eeh_enabled()) {
1791 seq_printf(m, "EEH Subsystem is globally disabled\n");
1792 seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1794 seq_printf(m, "EEH Subsystem is enabled\n");
1797 "no device node=%llu\n"
1798 "no config address=%llu\n"
1799 "check not wanted=%llu\n"
1800 "eeh_total_mmio_ffs=%llu\n"
1801 "eeh_false_positives=%llu\n"
1802 "eeh_slot_resets=%llu\n",
1803 eeh_stats.no_device,
1805 eeh_stats.no_cfg_addr,
1806 eeh_stats.ignored_check,
1807 eeh_stats.total_mmio_ffs,
1808 eeh_stats.false_positives,
1809 eeh_stats.slot_resets);
1815 #ifdef CONFIG_DEBUG_FS
1816 static int eeh_enable_dbgfs_set(void *data, u64 val)
1819 eeh_clear_flag(EEH_FORCE_DISABLED);
1821 eeh_add_flag(EEH_FORCE_DISABLED);
1826 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1835 DEFINE_DEBUGFS_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1836 eeh_enable_dbgfs_set, "0x%llx\n");
1838 static ssize_t eeh_force_recover_write(struct file *filp,
1839 const char __user *user_buf,
1840 size_t count, loff_t *ppos)
1842 struct pci_controller *hose;
1843 uint32_t phbid, pe_no;
1848 ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
1853 * When PE is NULL the event is a "special" event. Rather than
1854 * recovering a specific PE it forces the EEH core to scan for failed
1855 * PHBs and recovers each. This needs to be done before any device
1856 * recoveries can occur.
1858 if (!strncmp(buf, "hwcheck", 7)) {
1859 __eeh_send_failure_event(NULL);
1863 ret = sscanf(buf, "%x:%x", &phbid, &pe_no);
1867 hose = pci_find_controller_for_domain(phbid);
1872 pe = eeh_pe_get(hose, pe_no, 0);
1877 * We don't do any state checking here since the detection
1878 * process is async to the recovery process. The recovery
1879 * thread *should* not break even if we schedule a recovery
1880 * from an odd state (e.g. PE removed, or recovery of a
1883 __eeh_send_failure_event(pe);
1885 return ret < 0 ? ret : count;
1888 static const struct file_operations eeh_force_recover_fops = {
1889 .open = simple_open,
1890 .llseek = no_llseek,
1891 .write = eeh_force_recover_write,
1895 static int __init eeh_init_proc(void)
1897 if (machine_is(pseries) || machine_is(powernv)) {
1898 proc_create_single("powerpc/eeh", 0, NULL, proc_eeh_show);
1899 #ifdef CONFIG_DEBUG_FS
1900 debugfs_create_file_unsafe("eeh_enable", 0600,
1901 powerpc_debugfs_root, NULL,
1902 &eeh_enable_dbgfs_ops);
1903 debugfs_create_u32("eeh_max_freezes", 0600,
1904 powerpc_debugfs_root, &eeh_max_freezes);
1905 debugfs_create_bool("eeh_disable_recovery", 0600,
1906 powerpc_debugfs_root,
1907 &eeh_debugfs_no_recover);
1908 debugfs_create_file_unsafe("eeh_force_recover", 0600,
1909 powerpc_debugfs_root, NULL,
1910 &eeh_force_recover_fops);
1911 eeh_cache_debugfs_init();
1917 __initcall(eeh_init_proc);