2 * Copyright IBM Corporation 2001, 2005, 2006
3 * Copyright Dave Engebretsen & Todd Inglett 2001
4 * Copyright Linas Vepstas 2005, 2006
5 * Copyright 2001-2012 IBM Corporation.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 * Please address comments and feedback to Linas Vepstas <linas@austin.ibm.com>
24 #include <linux/delay.h>
25 #include <linux/sched.h>
26 #include <linux/init.h>
27 #include <linux/list.h>
28 #include <linux/pci.h>
29 #include <linux/iommu.h>
30 #include <linux/proc_fs.h>
31 #include <linux/rbtree.h>
32 #include <linux/reboot.h>
33 #include <linux/seq_file.h>
34 #include <linux/spinlock.h>
35 #include <linux/export.h>
38 #include <linux/atomic.h>
39 #include <asm/debugfs.h>
41 #include <asm/eeh_event.h>
43 #include <asm/iommu.h>
44 #include <asm/machdep.h>
45 #include <asm/ppc-pci.h>
47 #include <asm/pte-walk.h>
51 * EEH, or "Enhanced Error Handling" is a PCI bridge technology for
52 * dealing with PCI bus errors that can't be dealt with within the
53 * usual PCI framework, except by check-stopping the CPU. Systems
54 * that are designed for high-availability/reliability cannot afford
55 * to crash due to a "mere" PCI error, thus the need for EEH.
56 * An EEH-capable bridge operates by converting a detected error
57 * into a "slot freeze", taking the PCI adapter off-line, making
58 * the slot behave, from the OS'es point of view, as if the slot
59 * were "empty": all reads return 0xff's and all writes are silently
60 * ignored. EEH slot isolation events can be triggered by parity
61 * errors on the address or data busses (e.g. during posted writes),
62 * which in turn might be caused by low voltage on the bus, dust,
63 * vibration, humidity, radioactivity or plain-old failed hardware.
65 * Note, however, that one of the leading causes of EEH slot
66 * freeze events are buggy device drivers, buggy device microcode,
67 * or buggy device hardware. This is because any attempt by the
68 * device to bus-master data to a memory address that is not
69 * assigned to the device will trigger a slot freeze. (The idea
70 * is to prevent devices-gone-wild from corrupting system memory).
71 * Buggy hardware/drivers will have a miserable time co-existing
74 * Ideally, a PCI device driver, when suspecting that an isolation
75 * event has occurred (e.g. by reading 0xff's), will then ask EEH
76 * whether this is the case, and then take appropriate steps to
77 * reset the PCI slot, the PCI device, and then resume operations.
78 * However, until that day, the checking is done here, with the
79 * eeh_check_failure() routine embedded in the MMIO macros. If
80 * the slot is found to be isolated, an "EEH Event" is synthesized
81 * and sent out for processing.
84 /* If a device driver keeps reading an MMIO register in an interrupt
85 * handler after a slot isolation event, it might be broken.
86 * This sets the threshold for how many read attempts we allow
87 * before printing an error message.
89 #define EEH_MAX_FAILS 2100000
91 /* Time to wait for a PCI slot to report status, in milliseconds */
92 #define PCI_BUS_RESET_WAIT_MSEC (5*60*1000)
95 * EEH probe mode support, which is part of the flags,
96 * is to support multiple platforms for EEH. Some platforms
97 * like pSeries do PCI emunation based on device tree.
98 * However, other platforms like powernv probe PCI devices
99 * from hardware. The flag is used to distinguish that.
100 * In addition, struct eeh_ops::probe would be invoked for
101 * particular OF node or PCI device so that the corresponding
102 * PE would be created there.
104 int eeh_subsystem_flags;
105 EXPORT_SYMBOL(eeh_subsystem_flags);
108 * EEH allowed maximal frozen times. If one particular PE's
109 * frozen count in last hour exceeds this limit, the PE will
110 * be forced to be offline permanently.
112 int eeh_max_freezes = 5;
114 /* Platform dependent EEH operations */
115 struct eeh_ops *eeh_ops = NULL;
117 /* Lock to avoid races due to multiple reports of an error */
118 DEFINE_RAW_SPINLOCK(confirm_error_lock);
119 EXPORT_SYMBOL_GPL(confirm_error_lock);
121 /* Lock to protect passed flags */
122 static DEFINE_MUTEX(eeh_dev_mutex);
124 /* Buffer for reporting pci register dumps. Its here in BSS, and
125 * not dynamically alloced, so that it ends up in RMO where RTAS
128 #define EEH_PCI_REGS_LOG_LEN 8192
129 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
132 * The struct is used to maintain the EEH global statistic
133 * information. Besides, the EEH global statistics will be
134 * exported to user space through procfs
137 u64 no_device; /* PCI device not found */
138 u64 no_dn; /* OF node not found */
139 u64 no_cfg_addr; /* Config address not found */
140 u64 ignored_check; /* EEH check skipped */
141 u64 total_mmio_ffs; /* Total EEH checks */
142 u64 false_positives; /* Unnecessary EEH checks */
143 u64 slot_resets; /* PE reset */
146 static struct eeh_stats eeh_stats;
148 static int __init eeh_setup(char *str)
150 if (!strcmp(str, "off"))
151 eeh_add_flag(EEH_FORCE_DISABLED);
152 else if (!strcmp(str, "early_log"))
153 eeh_add_flag(EEH_EARLY_DUMP_LOG);
157 __setup("eeh=", eeh_setup);
160 * This routine captures assorted PCI configuration space data
161 * for the indicated PCI device, and puts them into a buffer
162 * for RTAS error logging.
164 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
166 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
173 pr_warn("EEH: Note: No error log for absent device.\n");
177 n += scnprintf(buf+n, len-n, "%04x:%02x:%02x.%01x\n",
178 pdn->phb->global_number, pdn->busno,
179 PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
180 pr_warn("EEH: of node=%04x:%02x:%02x.%01x\n",
181 pdn->phb->global_number, pdn->busno,
182 PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
184 eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
185 n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
186 pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
188 eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
189 n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
190 pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
192 /* Gather bridge-specific registers */
193 if (edev->mode & EEH_DEV_BRIDGE) {
194 eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
195 n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
196 pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
198 eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
199 n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
200 pr_warn("EEH: Bridge control: %04x\n", cfg);
203 /* Dump out the PCI-X command and status regs */
204 cap = edev->pcix_cap;
206 eeh_ops->read_config(pdn, cap, 4, &cfg);
207 n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
208 pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
210 eeh_ops->read_config(pdn, cap+4, 4, &cfg);
211 n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
212 pr_warn("EEH: PCI-X status: %08x\n", cfg);
215 /* If PCI-E capable, dump PCI-E cap 10 */
216 cap = edev->pcie_cap;
218 n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
219 pr_warn("EEH: PCI-E capabilities and status follow:\n");
221 for (i=0; i<=8; i++) {
222 eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
223 n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
227 pr_warn("%s\n", buffer);
229 l = scnprintf(buffer, sizeof(buffer),
230 "EEH: PCI-E %02x: %08x ",
233 l += scnprintf(buffer+l, sizeof(buffer)-l,
239 pr_warn("%s\n", buffer);
242 /* If AER capable, dump it */
245 n += scnprintf(buf+n, len-n, "pci-e AER:\n");
246 pr_warn("EEH: PCI-E AER capability register set follows:\n");
248 for (i=0; i<=13; i++) {
249 eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
250 n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
254 pr_warn("%s\n", buffer);
256 l = scnprintf(buffer, sizeof(buffer),
257 "EEH: PCI-E AER %02x: %08x ",
260 l += scnprintf(buffer+l, sizeof(buffer)-l,
265 pr_warn("%s\n", buffer);
271 static void *eeh_dump_pe_log(struct eeh_pe *pe, void *flag)
273 struct eeh_dev *edev, *tmp;
276 eeh_pe_for_each_dev(pe, edev, tmp)
277 *plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
278 EEH_PCI_REGS_LOG_LEN - *plen);
284 * eeh_slot_error_detail - Generate combined log including driver log and error log
286 * @severity: temporary or permanent error log
288 * This routine should be called to generate the combined log, which
289 * is comprised of driver log and error log. The driver log is figured
290 * out from the config space of the corresponding PCI device, while
291 * the error log is fetched through platform dependent function call.
293 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
298 * When the PHB is fenced or dead, it's pointless to collect
299 * the data from PCI config space because it should return
300 * 0xFF's. For ER, we still retrieve the data from the PCI
303 * For pHyp, we have to enable IO for log retrieval. Otherwise,
304 * 0xFF's is always returned from PCI config space.
306 * When the @severity is EEH_LOG_PERM, the PE is going to be
307 * removed. Prior to that, the drivers for devices included in
308 * the PE will be closed. The drivers rely on working IO path
309 * to bring the devices to quiet state. Otherwise, PCI traffic
310 * from those devices after they are removed is like to cause
311 * another unexpected EEH error.
313 if (!(pe->type & EEH_PE_PHB)) {
314 if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG) ||
315 severity == EEH_LOG_PERM)
316 eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
319 * The config space of some PCI devices can't be accessed
320 * when their PEs are in frozen state. Otherwise, fenced
321 * PHB might be seen. Those PEs are identified with flag
322 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
323 * is set automatically when the PE is put to EEH_PE_ISOLATED.
325 * Restoring BARs possibly triggers PCI config access in
326 * (OPAL) firmware and then causes fenced PHB. If the
327 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
328 * pointless to restore BARs and dump config space.
330 eeh_ops->configure_bridge(pe);
331 if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
332 eeh_pe_restore_bars(pe);
335 eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
339 eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
343 * eeh_token_to_phys - Convert EEH address token to phys address
344 * @token: I/O token, should be address in the form 0xA....
346 * This routine should be called to convert virtual I/O address
349 static inline unsigned long eeh_token_to_phys(unsigned long token)
356 * We won't find hugepages here(this is iomem). Hence we are not
357 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
358 * page table free, because of init_mm.
360 ptep = find_init_mm_pte(token, &hugepage_shift);
366 /* On radix we can do hugepage mappings for io, so handle that */
368 hugepage_shift = PAGE_SHIFT;
371 pa |= token & ((1ul << hugepage_shift) - 1);
376 * On PowerNV platform, we might already have fenced PHB there.
377 * For that case, it's meaningless to recover frozen PE. Intead,
378 * We have to handle fenced PHB firstly.
380 static int eeh_phb_check_failure(struct eeh_pe *pe)
382 struct eeh_pe *phb_pe;
386 if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
389 /* Find the PHB PE */
390 phb_pe = eeh_phb_pe_get(pe->phb);
392 pr_warn("%s Can't find PE for PHB#%x\n",
393 __func__, pe->phb->global_number);
397 /* If the PHB has been in problematic state */
398 eeh_serialize_lock(&flags);
399 if (phb_pe->state & EEH_PE_ISOLATED) {
404 /* Check PHB state */
405 ret = eeh_ops->get_state(phb_pe, NULL);
407 (ret == EEH_STATE_NOT_SUPPORT) || eeh_state_active(ret)) {
412 /* Isolate the PHB and send event */
413 eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
414 eeh_serialize_unlock(flags);
416 pr_err("EEH: PHB#%x failure detected, location: %s\n",
417 phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
419 eeh_send_failure_event(phb_pe);
423 eeh_serialize_unlock(flags);
428 * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
431 * Check for an EEH failure for the given device node. Call this
432 * routine if the result of a read was all 0xff's and you want to
433 * find out if this is due to an EEH slot freeze. This routine
434 * will query firmware for the EEH status.
436 * Returns 0 if there has not been an EEH error; otherwise returns
437 * a non-zero value and queues up a slot isolation event notification.
439 * It is safe to call this routine in an interrupt context.
441 int eeh_dev_check_failure(struct eeh_dev *edev)
445 struct device_node *dn;
447 struct eeh_pe *pe, *parent_pe, *phb_pe;
449 const char *location = NULL;
451 eeh_stats.total_mmio_ffs++;
460 dev = eeh_dev_to_pci_dev(edev);
461 pe = eeh_dev_to_pe(edev);
463 /* Access to IO BARs might get this far and still not want checking. */
465 eeh_stats.ignored_check++;
466 pr_debug("EEH: Ignored check for %s\n",
471 if (!pe->addr && !pe->config_addr) {
472 eeh_stats.no_cfg_addr++;
477 * On PowerNV platform, we might already have fenced PHB
478 * there and we need take care of that firstly.
480 ret = eeh_phb_check_failure(pe);
485 * If the PE isn't owned by us, we shouldn't check the
486 * state. Instead, let the owner handle it if the PE has
489 if (eeh_pe_passed(pe))
492 /* If we already have a pending isolation event for this
493 * slot, we know it's bad already, we don't need to check.
494 * Do this checking under a lock; as multiple PCI devices
495 * in one slot might report errors simultaneously, and we
496 * only want one error recovery routine running.
498 eeh_serialize_lock(&flags);
500 if (pe->state & EEH_PE_ISOLATED) {
502 if (pe->check_count == EEH_MAX_FAILS) {
503 dn = pci_device_to_OF_node(dev);
505 location = of_get_property(dn, "ibm,loc-code",
507 printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
508 "location=%s driver=%s pci addr=%s\n",
510 location ? location : "unknown",
511 eeh_driver_name(dev), eeh_pci_name(dev));
512 printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
513 eeh_driver_name(dev));
520 * Now test for an EEH failure. This is VERY expensive.
521 * Note that the eeh_config_addr may be a parent device
522 * in the case of a device behind a bridge, or it may be
523 * function zero of a multi-function device.
524 * In any case they must share a common PHB.
526 ret = eeh_ops->get_state(pe, NULL);
528 /* Note that config-io to empty slots may fail;
529 * they are empty when they don't have children.
530 * We will punt with the following conditions: Failure to get
531 * PE's state, EEH not support and Permanently unavailable
532 * state, PE is in good state.
535 (ret == EEH_STATE_NOT_SUPPORT) || eeh_state_active(ret)) {
536 eeh_stats.false_positives++;
537 pe->false_positives++;
543 * It should be corner case that the parent PE has been
544 * put into frozen state as well. We should take care
547 parent_pe = pe->parent;
549 /* Hit the ceiling ? */
550 if (parent_pe->type & EEH_PE_PHB)
553 /* Frozen parent PE ? */
554 ret = eeh_ops->get_state(parent_pe, NULL);
555 if (ret > 0 && !eeh_state_active(ret)) {
557 pr_err("EEH: Failure of PHB#%x-PE#%x will be handled at parent PHB#%x-PE#%x.\n",
558 pe->phb->global_number, pe->addr,
559 pe->phb->global_number, parent_pe->addr);
562 /* Next parent level */
563 parent_pe = parent_pe->parent;
566 eeh_stats.slot_resets++;
568 /* Avoid repeated reports of this failure, including problems
569 * with other functions on this device, and functions under
572 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
573 eeh_serialize_unlock(flags);
575 /* Most EEH events are due to device driver bugs. Having
576 * a stack trace will help the device-driver authors figure
577 * out what happened. So print that out.
579 phb_pe = eeh_phb_pe_get(pe->phb);
580 pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
581 pe->phb->global_number, pe->addr);
582 pr_err("EEH: PE location: %s, PHB location: %s\n",
583 eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
586 eeh_send_failure_event(pe);
591 eeh_serialize_unlock(flags);
595 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
598 * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
599 * @token: I/O address
601 * Check for an EEH failure at the given I/O address. Call this
602 * routine if the result of a read was all 0xff's and you want to
603 * find out if this is due to an EEH slot freeze event. This routine
604 * will query firmware for the EEH status.
606 * Note this routine is safe to call in an interrupt context.
608 int eeh_check_failure(const volatile void __iomem *token)
611 struct eeh_dev *edev;
613 /* Finding the phys addr + pci device; this is pretty quick. */
614 addr = eeh_token_to_phys((unsigned long __force) token);
615 edev = eeh_addr_cache_get_dev(addr);
617 eeh_stats.no_device++;
621 return eeh_dev_check_failure(edev);
623 EXPORT_SYMBOL(eeh_check_failure);
627 * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
630 * This routine should be called to reenable frozen MMIO or DMA
631 * so that it would work correctly again. It's useful while doing
632 * recovery or log collection on the indicated device.
634 int eeh_pci_enable(struct eeh_pe *pe, int function)
639 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
640 * Also, it's pointless to enable them on unfrozen PE. So
641 * we have to check before enabling IO or DMA.
644 case EEH_OPT_THAW_MMIO:
645 active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
647 case EEH_OPT_THAW_DMA:
648 active_flag = EEH_STATE_DMA_ACTIVE;
650 case EEH_OPT_DISABLE:
652 case EEH_OPT_FREEZE_PE:
656 pr_warn("%s: Invalid function %d\n",
662 * Check if IO or DMA has been enabled before
666 rc = eeh_ops->get_state(pe, NULL);
670 /* Needn't enable it at all */
671 if (rc == EEH_STATE_NOT_SUPPORT)
674 /* It's already enabled */
675 if (rc & active_flag)
680 /* Issue the request */
681 rc = eeh_ops->set_option(pe, function);
683 pr_warn("%s: Unexpected state change %d on "
684 "PHB#%x-PE#%x, err=%d\n",
685 __func__, function, pe->phb->global_number,
688 /* Check if the request is finished successfully */
690 rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
694 if (rc & active_flag)
703 static void *eeh_disable_and_save_dev_state(struct eeh_dev *edev,
706 struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
707 struct pci_dev *dev = userdata;
710 * The caller should have disabled and saved the
711 * state for the specified device
713 if (!pdev || pdev == dev)
716 /* Ensure we have D0 power state */
717 pci_set_power_state(pdev, PCI_D0);
719 /* Save device state */
720 pci_save_state(pdev);
723 * Disable device to avoid any DMA traffic and
724 * interrupt from the device
726 pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
731 static void *eeh_restore_dev_state(struct eeh_dev *edev, void *userdata)
733 struct pci_dn *pdn = eeh_dev_to_pdn(edev);
734 struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
735 struct pci_dev *dev = userdata;
740 /* Apply customization from firmware */
741 if (pdn && eeh_ops->restore_config)
742 eeh_ops->restore_config(pdn);
744 /* The caller should restore state for the specified device */
746 pci_restore_state(pdev);
751 int eeh_restore_vf_config(struct pci_dn *pdn)
753 struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
754 u32 devctl, cmd, cap2, aer_capctl;
757 if (edev->pcie_cap) {
759 old_mps = (ffs(pdn->mps) - 8) << 5;
760 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
762 devctl &= ~PCI_EXP_DEVCTL_PAYLOAD;
764 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
767 /* Disable Completion Timeout if possible */
768 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP2,
770 if (cap2 & PCI_EXP_DEVCAP2_COMP_TMOUT_DIS) {
771 eeh_ops->read_config(pdn,
772 edev->pcie_cap + PCI_EXP_DEVCTL2,
774 cap2 |= PCI_EXP_DEVCTL2_COMP_TMOUT_DIS;
775 eeh_ops->write_config(pdn,
776 edev->pcie_cap + PCI_EXP_DEVCTL2,
781 /* Enable SERR and parity checking */
782 eeh_ops->read_config(pdn, PCI_COMMAND, 2, &cmd);
783 cmd |= (PCI_COMMAND_PARITY | PCI_COMMAND_SERR);
784 eeh_ops->write_config(pdn, PCI_COMMAND, 2, cmd);
786 /* Enable report various errors */
787 if (edev->pcie_cap) {
788 eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
790 devctl &= ~PCI_EXP_DEVCTL_CERE;
791 devctl |= (PCI_EXP_DEVCTL_NFERE |
792 PCI_EXP_DEVCTL_FERE |
793 PCI_EXP_DEVCTL_URRE);
794 eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
798 /* Enable ECRC generation and check */
799 if (edev->pcie_cap && edev->aer_cap) {
800 eeh_ops->read_config(pdn, edev->aer_cap + PCI_ERR_CAP,
802 aer_capctl |= (PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE);
803 eeh_ops->write_config(pdn, edev->aer_cap + PCI_ERR_CAP,
811 * pcibios_set_pcie_reset_state - Set PCI-E reset state
812 * @dev: pci device struct
813 * @state: reset state to enter
818 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
820 struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
821 struct eeh_pe *pe = eeh_dev_to_pe(edev);
824 pr_err("%s: No PE found on PCI device %s\n",
825 __func__, pci_name(dev));
830 case pcie_deassert_reset:
831 eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
832 eeh_unfreeze_pe(pe, false);
833 if (!(pe->type & EEH_PE_VF))
834 eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
835 eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
836 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
839 eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
840 eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
841 eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
842 if (!(pe->type & EEH_PE_VF))
843 eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
844 eeh_ops->reset(pe, EEH_RESET_HOT);
846 case pcie_warm_reset:
847 eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
848 eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
849 eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
850 if (!(pe->type & EEH_PE_VF))
851 eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
852 eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
855 eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
863 * eeh_set_pe_freset - Check the required reset for the indicated device
865 * @flag: return value
867 * Each device might have its preferred reset type: fundamental or
868 * hot reset. The routine is used to collected the information for
869 * the indicated device and its children so that the bunch of the
870 * devices could be reset properly.
872 static void *eeh_set_dev_freset(struct eeh_dev *edev, void *flag)
875 unsigned int *freset = (unsigned int *)flag;
877 dev = eeh_dev_to_pci_dev(edev);
879 *freset |= dev->needs_freset;
885 * eeh_pe_reset_full - Complete a full reset process on the indicated PE
888 * This function executes a full reset procedure on a PE, including setting
889 * the appropriate flags, performing a fundamental or hot reset, and then
890 * deactivating the reset status. It is designed to be used within the EEH
891 * subsystem, as opposed to eeh_pe_reset which is exported to drivers and
892 * only performs a single operation at a time.
894 * This function will attempt to reset a PE three times before failing.
896 int eeh_pe_reset_full(struct eeh_pe *pe)
898 int reset_state = (EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
899 int type = EEH_RESET_HOT;
900 unsigned int freset = 0;
904 * Determine the type of reset to perform - hot or fundamental.
905 * Hot reset is the default operation, unless any device under the
906 * PE requires a fundamental reset.
908 eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
911 type = EEH_RESET_FUNDAMENTAL;
913 /* Mark the PE as in reset state and block config space accesses */
914 eeh_pe_state_mark(pe, reset_state);
916 /* Make three attempts at resetting the bus */
917 for (i = 0; i < 3; i++) {
918 ret = eeh_pe_reset(pe, type);
922 ret = eeh_pe_reset(pe, EEH_RESET_DEACTIVATE);
926 /* Wait until the PE is in a functioning state */
927 state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
928 if (eeh_state_active(state))
932 pr_warn("%s: Unrecoverable slot failure on PHB#%x-PE#%x",
933 __func__, pe->phb->global_number, pe->addr);
934 ret = -ENOTRECOVERABLE;
938 /* Set error in case this is our last attempt */
940 pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
941 __func__, state, pe->phb->global_number, pe->addr, (i + 1));
944 eeh_pe_state_clear(pe, reset_state);
949 * eeh_save_bars - Save device bars
950 * @edev: PCI device associated EEH device
952 * Save the values of the device bars. Unlike the restore
953 * routine, this routine is *not* recursive. This is because
954 * PCI devices are added individually; but, for the restore,
955 * an entire slot is reset at a time.
957 void eeh_save_bars(struct eeh_dev *edev)
962 pdn = eeh_dev_to_pdn(edev);
966 for (i = 0; i < 16; i++)
967 eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
970 * For PCI bridges including root port, we need enable bus
971 * master explicitly. Otherwise, it can't fetch IODA table
972 * entries correctly. So we cache the bit in advance so that
973 * we can restore it after reset, either PHB range or PE range.
975 if (edev->mode & EEH_DEV_BRIDGE)
976 edev->config_space[1] |= PCI_COMMAND_MASTER;
980 * eeh_ops_register - Register platform dependent EEH operations
981 * @ops: platform dependent EEH operations
983 * Register the platform dependent EEH operation callback
984 * functions. The platform should call this function before
985 * any other EEH operations.
987 int __init eeh_ops_register(struct eeh_ops *ops)
990 pr_warn("%s: Invalid EEH ops name for %p\n",
995 if (eeh_ops && eeh_ops != ops) {
996 pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
997 __func__, eeh_ops->name, ops->name);
1007 * eeh_ops_unregister - Unreigster platform dependent EEH operations
1008 * @name: name of EEH platform operations
1010 * Unregister the platform dependent EEH operation callback
1013 int __exit eeh_ops_unregister(const char *name)
1015 if (!name || !strlen(name)) {
1016 pr_warn("%s: Invalid EEH ops name\n",
1021 if (eeh_ops && !strcmp(eeh_ops->name, name)) {
1029 static int eeh_reboot_notifier(struct notifier_block *nb,
1030 unsigned long action, void *unused)
1032 eeh_clear_flag(EEH_ENABLED);
1036 static struct notifier_block eeh_reboot_nb = {
1037 .notifier_call = eeh_reboot_notifier,
1040 void eeh_probe_devices(void)
1042 struct pci_controller *hose, *tmp;
1045 /* Enable EEH for all adapters */
1046 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1047 pdn = hose->pci_data;
1048 traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1053 * eeh_init - EEH initialization
1055 * Initialize EEH by trying to enable it for all of the adapters in the system.
1056 * As a side effect we can determine here if eeh is supported at all.
1057 * Note that we leave EEH on so failed config cycles won't cause a machine
1058 * check. If a user turns off EEH for a particular adapter they are really
1059 * telling Linux to ignore errors. Some hardware (e.g. POWER5) won't
1060 * grant access to a slot if EEH isn't enabled, and so we always enable
1061 * EEH for all slots/all devices.
1063 * The eeh-force-off option disables EEH checking globally, for all slots.
1064 * Even if force-off is set, the EEH hardware is still enabled, so that
1065 * newer systems can boot.
1067 static int eeh_init(void)
1069 struct pci_controller *hose, *tmp;
1072 /* Register reboot notifier */
1073 ret = register_reboot_notifier(&eeh_reboot_nb);
1075 pr_warn("%s: Failed to register notifier (%d)\n",
1080 /* call platform initialization function */
1082 pr_warn("%s: Platform EEH operation not found\n",
1085 } else if ((ret = eeh_ops->init()))
1088 /* Initialize PHB PEs */
1089 list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
1090 eeh_dev_phb_init_dynamic(hose);
1092 /* Initialize EEH event */
1093 ret = eeh_event_init();
1097 eeh_probe_devices();
1100 pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1101 else if (!eeh_has_flag(EEH_POSTPONED_PROBE))
1102 pr_info("EEH: No capable adapters found\n");
1107 core_initcall_sync(eeh_init);
1110 * eeh_add_device_early - Enable EEH for the indicated device node
1111 * @pdn: PCI device node for which to set up EEH
1113 * This routine must be used to perform EEH initialization for PCI
1114 * devices that were added after system boot (e.g. hotplug, dlpar).
1115 * This routine must be called before any i/o is performed to the
1116 * adapter (inluding any config-space i/o).
1117 * Whether this actually enables EEH or not for this device depends
1118 * on the CEC architecture, type of the device, on earlier boot
1119 * command-line arguments & etc.
1121 void eeh_add_device_early(struct pci_dn *pdn)
1123 struct pci_controller *phb = pdn ? pdn->phb : NULL;
1124 struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1129 if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1132 /* USB Bus children of PCI devices will not have BUID's */
1134 (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1137 eeh_ops->probe(pdn, NULL);
1141 * eeh_add_device_tree_early - Enable EEH for the indicated device
1142 * @pdn: PCI device node
1144 * This routine must be used to perform EEH initialization for the
1145 * indicated PCI device that was added after system boot (e.g.
1148 void eeh_add_device_tree_early(struct pci_dn *pdn)
1155 list_for_each_entry(n, &pdn->child_list, list)
1156 eeh_add_device_tree_early(n);
1157 eeh_add_device_early(pdn);
1159 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1162 * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1163 * @dev: pci device for which to set up EEH
1165 * This routine must be used to complete EEH initialization for PCI
1166 * devices that were added after system boot (e.g. hotplug, dlpar).
1168 void eeh_add_device_late(struct pci_dev *dev)
1171 struct eeh_dev *edev;
1173 if (!dev || !eeh_enabled())
1176 pr_debug("EEH: Adding device %s\n", pci_name(dev));
1178 pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1179 edev = pdn_to_eeh_dev(pdn);
1180 if (edev->pdev == dev) {
1181 pr_debug("EEH: Already referenced !\n");
1186 * The EEH cache might not be removed correctly because of
1187 * unbalanced kref to the device during unplug time, which
1188 * relies on pcibios_release_device(). So we have to remove
1189 * that here explicitly.
1192 eeh_rmv_from_parent_pe(edev);
1193 eeh_addr_cache_rmv_dev(edev->pdev);
1194 eeh_sysfs_remove_device(edev->pdev);
1195 edev->mode &= ~EEH_DEV_SYSFS;
1198 * We definitely should have the PCI device removed
1199 * though it wasn't correctly. So we needn't call
1200 * into error handler afterwards.
1202 edev->mode |= EEH_DEV_NO_HANDLER;
1205 dev->dev.archdata.edev = NULL;
1208 if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1209 eeh_ops->probe(pdn, NULL);
1212 dev->dev.archdata.edev = edev;
1214 eeh_addr_cache_insert_dev(dev);
1218 * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1221 * This routine must be used to perform EEH initialization for PCI
1222 * devices which are attached to the indicated PCI bus. The PCI bus
1223 * is added after system boot through hotplug or dlpar.
1225 void eeh_add_device_tree_late(struct pci_bus *bus)
1227 struct pci_dev *dev;
1229 list_for_each_entry(dev, &bus->devices, bus_list) {
1230 eeh_add_device_late(dev);
1231 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1232 struct pci_bus *subbus = dev->subordinate;
1234 eeh_add_device_tree_late(subbus);
1238 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1241 * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1244 * This routine must be used to add EEH sysfs files for PCI
1245 * devices which are attached to the indicated PCI bus. The PCI bus
1246 * is added after system boot through hotplug or dlpar.
1248 void eeh_add_sysfs_files(struct pci_bus *bus)
1250 struct pci_dev *dev;
1252 list_for_each_entry(dev, &bus->devices, bus_list) {
1253 eeh_sysfs_add_device(dev);
1254 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1255 struct pci_bus *subbus = dev->subordinate;
1257 eeh_add_sysfs_files(subbus);
1261 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1264 * eeh_remove_device - Undo EEH setup for the indicated pci device
1265 * @dev: pci device to be removed
1267 * This routine should be called when a device is removed from
1268 * a running system (e.g. by hotplug or dlpar). It unregisters
1269 * the PCI device from the EEH subsystem. I/O errors affecting
1270 * this device will no longer be detected after this call; thus,
1271 * i/o errors affecting this slot may leave this device unusable.
1273 void eeh_remove_device(struct pci_dev *dev)
1275 struct eeh_dev *edev;
1277 if (!dev || !eeh_enabled())
1279 edev = pci_dev_to_eeh_dev(dev);
1281 /* Unregister the device with the EEH/PCI address search system */
1282 pr_debug("EEH: Removing device %s\n", pci_name(dev));
1284 if (!edev || !edev->pdev || !edev->pe) {
1285 pr_debug("EEH: Not referenced !\n");
1290 * During the hotplug for EEH error recovery, we need the EEH
1291 * device attached to the parent PE in order for BAR restore
1292 * a bit later. So we keep it for BAR restore and remove it
1293 * from the parent PE during the BAR resotre.
1298 * The flag "in_error" is used to trace EEH devices for VFs
1299 * in error state or not. It's set in eeh_report_error(). If
1300 * it's not set, eeh_report_{reset,resume}() won't be called
1301 * for the VF EEH device.
1303 edev->in_error = false;
1304 dev->dev.archdata.edev = NULL;
1305 if (!(edev->pe->state & EEH_PE_KEEP))
1306 eeh_rmv_from_parent_pe(edev);
1308 edev->mode |= EEH_DEV_DISCONNECTED;
1311 * We're removing from the PCI subsystem, that means
1312 * the PCI device driver can't support EEH or not
1313 * well. So we rely on hotplug completely to do recovery
1314 * for the specific PCI device.
1316 edev->mode |= EEH_DEV_NO_HANDLER;
1318 eeh_addr_cache_rmv_dev(dev);
1319 eeh_sysfs_remove_device(dev);
1320 edev->mode &= ~EEH_DEV_SYSFS;
1323 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1327 ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1329 pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1330 __func__, ret, pe->phb->global_number, pe->addr);
1334 ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1336 pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1337 __func__, ret, pe->phb->global_number, pe->addr);
1341 /* Clear software isolated state */
1342 if (sw_state && (pe->state & EEH_PE_ISOLATED))
1343 eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1349 static struct pci_device_id eeh_reset_ids[] = {
1350 { PCI_DEVICE(0x19a2, 0x0710) }, /* Emulex, BE */
1351 { PCI_DEVICE(0x10df, 0xe220) }, /* Emulex, Lancer */
1352 { PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1356 static int eeh_pe_change_owner(struct eeh_pe *pe)
1358 struct eeh_dev *edev, *tmp;
1359 struct pci_dev *pdev;
1360 struct pci_device_id *id;
1363 /* Check PE state */
1364 ret = eeh_ops->get_state(pe, NULL);
1365 if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1368 /* Unfrozen PE, nothing to do */
1369 if (eeh_state_active(ret))
1372 /* Frozen PE, check if it needs PE level reset */
1373 eeh_pe_for_each_dev(pe, edev, tmp) {
1374 pdev = eeh_dev_to_pci_dev(edev);
1378 for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1379 if (id->vendor != PCI_ANY_ID &&
1380 id->vendor != pdev->vendor)
1382 if (id->device != PCI_ANY_ID &&
1383 id->device != pdev->device)
1385 if (id->subvendor != PCI_ANY_ID &&
1386 id->subvendor != pdev->subsystem_vendor)
1388 if (id->subdevice != PCI_ANY_ID &&
1389 id->subdevice != pdev->subsystem_device)
1392 return eeh_pe_reset_and_recover(pe);
1396 return eeh_unfreeze_pe(pe, true);
1400 * eeh_dev_open - Increase count of pass through devices for PE
1403 * Increase count of passed through devices for the indicated
1404 * PE. In the result, the EEH errors detected on the PE won't be
1405 * reported. The PE owner will be responsible for detection
1408 int eeh_dev_open(struct pci_dev *pdev)
1410 struct eeh_dev *edev;
1413 mutex_lock(&eeh_dev_mutex);
1415 /* No PCI device ? */
1419 /* No EEH device or PE ? */
1420 edev = pci_dev_to_eeh_dev(pdev);
1421 if (!edev || !edev->pe)
1425 * The PE might have been put into frozen state, but we
1426 * didn't detect that yet. The passed through PCI devices
1427 * in frozen PE won't work properly. Clear the frozen state
1430 ret = eeh_pe_change_owner(edev->pe);
1434 /* Increase PE's pass through count */
1435 atomic_inc(&edev->pe->pass_dev_cnt);
1436 mutex_unlock(&eeh_dev_mutex);
1440 mutex_unlock(&eeh_dev_mutex);
1443 EXPORT_SYMBOL_GPL(eeh_dev_open);
1446 * eeh_dev_release - Decrease count of pass through devices for PE
1449 * Decrease count of pass through devices for the indicated PE. If
1450 * there is no passed through device in PE, the EEH errors detected
1451 * on the PE will be reported and handled as usual.
1453 void eeh_dev_release(struct pci_dev *pdev)
1455 struct eeh_dev *edev;
1457 mutex_lock(&eeh_dev_mutex);
1459 /* No PCI device ? */
1463 /* No EEH device ? */
1464 edev = pci_dev_to_eeh_dev(pdev);
1465 if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1468 /* Decrease PE's pass through count */
1469 WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1470 eeh_pe_change_owner(edev->pe);
1472 mutex_unlock(&eeh_dev_mutex);
1474 EXPORT_SYMBOL(eeh_dev_release);
1476 #ifdef CONFIG_IOMMU_API
1478 static int dev_has_iommu_table(struct device *dev, void *data)
1480 struct pci_dev *pdev = to_pci_dev(dev);
1481 struct pci_dev **ppdev = data;
1486 if (dev->iommu_group) {
1495 * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1496 * @group: IOMMU group
1498 * The routine is called to convert IOMMU group to EEH PE.
1500 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1502 struct pci_dev *pdev = NULL;
1503 struct eeh_dev *edev;
1506 /* No IOMMU group ? */
1510 ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1514 /* No EEH device or PE ? */
1515 edev = pci_dev_to_eeh_dev(pdev);
1516 if (!edev || !edev->pe)
1521 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1523 #endif /* CONFIG_IOMMU_API */
1526 * eeh_pe_set_option - Set options for the indicated PE
1528 * @option: requested option
1530 * The routine is called to enable or disable EEH functionality
1531 * on the indicated PE, to enable IO or DMA for the frozen PE.
1533 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1542 * EEH functionality could possibly be disabled, just
1543 * return error for the case. And the EEH functinality
1544 * isn't expected to be disabled on one specific PE.
1547 case EEH_OPT_ENABLE:
1548 if (eeh_enabled()) {
1549 ret = eeh_pe_change_owner(pe);
1554 case EEH_OPT_DISABLE:
1556 case EEH_OPT_THAW_MMIO:
1557 case EEH_OPT_THAW_DMA:
1558 case EEH_OPT_FREEZE_PE:
1559 if (!eeh_ops || !eeh_ops->set_option) {
1564 ret = eeh_pci_enable(pe, option);
1567 pr_debug("%s: Option %d out of range (%d, %d)\n",
1568 __func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1574 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1577 * eeh_pe_get_state - Retrieve PE's state
1580 * Retrieve the PE's state, which includes 3 aspects: enabled
1581 * DMA, enabled IO and asserted reset.
1583 int eeh_pe_get_state(struct eeh_pe *pe)
1585 int result, ret = 0;
1586 bool rst_active, dma_en, mmio_en;
1592 if (!eeh_ops || !eeh_ops->get_state)
1596 * If the parent PE is owned by the host kernel and is undergoing
1597 * error recovery, we should return the PE state as temporarily
1598 * unavailable so that the error recovery on the guest is suspended
1599 * until the recovery completes on the host.
1602 !(pe->state & EEH_PE_REMOVED) &&
1603 (pe->parent->state & (EEH_PE_ISOLATED | EEH_PE_RECOVERING)))
1604 return EEH_PE_STATE_UNAVAIL;
1606 result = eeh_ops->get_state(pe, NULL);
1607 rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1608 dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1609 mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1612 ret = EEH_PE_STATE_RESET;
1613 else if (dma_en && mmio_en)
1614 ret = EEH_PE_STATE_NORMAL;
1615 else if (!dma_en && !mmio_en)
1616 ret = EEH_PE_STATE_STOPPED_IO_DMA;
1617 else if (!dma_en && mmio_en)
1618 ret = EEH_PE_STATE_STOPPED_DMA;
1620 ret = EEH_PE_STATE_UNAVAIL;
1624 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1626 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1628 struct eeh_dev *edev, *tmp;
1629 struct pci_dev *pdev;
1632 /* Restore config space */
1633 eeh_pe_restore_bars(pe);
1636 * Reenable PCI devices as the devices passed
1637 * through are always enabled before the reset.
1639 eeh_pe_for_each_dev(pe, edev, tmp) {
1640 pdev = eeh_dev_to_pci_dev(edev);
1644 ret = pci_reenable_device(pdev);
1646 pr_warn("%s: Failure %d reenabling %s\n",
1647 __func__, ret, pci_name(pdev));
1652 /* The PE is still in frozen state */
1653 return eeh_unfreeze_pe(pe, true);
1658 * eeh_pe_reset - Issue PE reset according to specified type
1660 * @option: reset type
1662 * The routine is called to reset the specified PE with the
1663 * indicated type, either fundamental reset or hot reset.
1664 * PE reset is the most important part for error recovery.
1666 int eeh_pe_reset(struct eeh_pe *pe, int option)
1674 if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1678 case EEH_RESET_DEACTIVATE:
1679 ret = eeh_ops->reset(pe, option);
1680 eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1684 ret = eeh_pe_reenable_devices(pe);
1687 case EEH_RESET_FUNDAMENTAL:
1689 * Proactively freeze the PE to drop all MMIO access
1690 * during reset, which should be banned as it's always
1691 * cause recursive EEH error.
1693 eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1695 eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1696 ret = eeh_ops->reset(pe, option);
1699 pr_debug("%s: Unsupported option %d\n",
1706 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1709 * eeh_pe_configure - Configure PCI bridges after PE reset
1712 * The routine is called to restore the PCI config space for
1713 * those PCI devices, especially PCI bridges affected by PE
1714 * reset issued previously.
1716 int eeh_pe_configure(struct eeh_pe *pe)
1726 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1729 * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1730 * @pe: the indicated PE
1732 * @function: error function
1734 * @mask: address mask
1736 * The routine is called to inject the specified PCI error, which
1737 * is determined by @type and @function, to the indicated PE for
1740 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1741 unsigned long addr, unsigned long mask)
1747 /* Unsupported operation ? */
1748 if (!eeh_ops || !eeh_ops->err_inject)
1751 /* Check on PCI error type */
1752 if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1755 /* Check on PCI error function */
1756 if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1759 return eeh_ops->err_inject(pe, type, func, addr, mask);
1761 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1763 static int proc_eeh_show(struct seq_file *m, void *v)
1765 if (!eeh_enabled()) {
1766 seq_printf(m, "EEH Subsystem is globally disabled\n");
1767 seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1769 seq_printf(m, "EEH Subsystem is enabled\n");
1772 "no device node=%llu\n"
1773 "no config address=%llu\n"
1774 "check not wanted=%llu\n"
1775 "eeh_total_mmio_ffs=%llu\n"
1776 "eeh_false_positives=%llu\n"
1777 "eeh_slot_resets=%llu\n",
1778 eeh_stats.no_device,
1780 eeh_stats.no_cfg_addr,
1781 eeh_stats.ignored_check,
1782 eeh_stats.total_mmio_ffs,
1783 eeh_stats.false_positives,
1784 eeh_stats.slot_resets);
1790 #ifdef CONFIG_DEBUG_FS
1791 static int eeh_enable_dbgfs_set(void *data, u64 val)
1794 eeh_clear_flag(EEH_FORCE_DISABLED);
1796 eeh_add_flag(EEH_FORCE_DISABLED);
1801 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1810 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1812 eeh_max_freezes = val;
1816 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1818 *val = eeh_max_freezes;
1822 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1823 eeh_enable_dbgfs_set, "0x%llx\n");
1824 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1825 eeh_freeze_dbgfs_set, "0x%llx\n");
1828 static int __init eeh_init_proc(void)
1830 if (machine_is(pseries) || machine_is(powernv)) {
1831 proc_create_single("powerpc/eeh", 0, NULL, proc_eeh_show);
1832 #ifdef CONFIG_DEBUG_FS
1833 debugfs_create_file("eeh_enable", 0600,
1834 powerpc_debugfs_root, NULL,
1835 &eeh_enable_dbgfs_ops);
1836 debugfs_create_file("eeh_max_freezes", 0600,
1837 powerpc_debugfs_root, NULL,
1838 &eeh_freeze_dbgfs_ops);
1844 __initcall(eeh_init_proc);