xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision dea54fba)
1 /*
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.
6  *
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.
11  *
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.
16  *
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
20  *
21  * Please address comments and feedback to Linas Vepstas <linas@austin.ibm.com>
22  */
23 
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>
36 #include <linux/of.h>
37 
38 #include <linux/atomic.h>
39 #include <asm/debugfs.h>
40 #include <asm/eeh.h>
41 #include <asm/eeh_event.h>
42 #include <asm/io.h>
43 #include <asm/iommu.h>
44 #include <asm/machdep.h>
45 #include <asm/ppc-pci.h>
46 #include <asm/rtas.h>
47 
48 
49 /** Overview:
50  *  EEH, or "Enhanced Error Handling" is a PCI bridge technology for
51  *  dealing with PCI bus errors that can't be dealt with within the
52  *  usual PCI framework, except by check-stopping the CPU.  Systems
53  *  that are designed for high-availability/reliability cannot afford
54  *  to crash due to a "mere" PCI error, thus the need for EEH.
55  *  An EEH-capable bridge operates by converting a detected error
56  *  into a "slot freeze", taking the PCI adapter off-line, making
57  *  the slot behave, from the OS'es point of view, as if the slot
58  *  were "empty": all reads return 0xff's and all writes are silently
59  *  ignored.  EEH slot isolation events can be triggered by parity
60  *  errors on the address or data busses (e.g. during posted writes),
61  *  which in turn might be caused by low voltage on the bus, dust,
62  *  vibration, humidity, radioactivity or plain-old failed hardware.
63  *
64  *  Note, however, that one of the leading causes of EEH slot
65  *  freeze events are buggy device drivers, buggy device microcode,
66  *  or buggy device hardware.  This is because any attempt by the
67  *  device to bus-master data to a memory address that is not
68  *  assigned to the device will trigger a slot freeze.   (The idea
69  *  is to prevent devices-gone-wild from corrupting system memory).
70  *  Buggy hardware/drivers will have a miserable time co-existing
71  *  with EEH.
72  *
73  *  Ideally, a PCI device driver, when suspecting that an isolation
74  *  event has occurred (e.g. by reading 0xff's), will then ask EEH
75  *  whether this is the case, and then take appropriate steps to
76  *  reset the PCI slot, the PCI device, and then resume operations.
77  *  However, until that day,  the checking is done here, with the
78  *  eeh_check_failure() routine embedded in the MMIO macros.  If
79  *  the slot is found to be isolated, an "EEH Event" is synthesized
80  *  and sent out for processing.
81  */
82 
83 /* If a device driver keeps reading an MMIO register in an interrupt
84  * handler after a slot isolation event, it might be broken.
85  * This sets the threshold for how many read attempts we allow
86  * before printing an error message.
87  */
88 #define EEH_MAX_FAILS	2100000
89 
90 /* Time to wait for a PCI slot to report status, in milliseconds */
91 #define PCI_BUS_RESET_WAIT_MSEC (5*60*1000)
92 
93 /*
94  * EEH probe mode support, which is part of the flags,
95  * is to support multiple platforms for EEH. Some platforms
96  * like pSeries do PCI emunation based on device tree.
97  * However, other platforms like powernv probe PCI devices
98  * from hardware. The flag is used to distinguish that.
99  * In addition, struct eeh_ops::probe would be invoked for
100  * particular OF node or PCI device so that the corresponding
101  * PE would be created there.
102  */
103 int eeh_subsystem_flags;
104 EXPORT_SYMBOL(eeh_subsystem_flags);
105 
106 /*
107  * EEH allowed maximal frozen times. If one particular PE's
108  * frozen count in last hour exceeds this limit, the PE will
109  * be forced to be offline permanently.
110  */
111 int eeh_max_freezes = 5;
112 
113 /* Platform dependent EEH operations */
114 struct eeh_ops *eeh_ops = NULL;
115 
116 /* Lock to avoid races due to multiple reports of an error */
117 DEFINE_RAW_SPINLOCK(confirm_error_lock);
118 EXPORT_SYMBOL_GPL(confirm_error_lock);
119 
120 /* Lock to protect passed flags */
121 static DEFINE_MUTEX(eeh_dev_mutex);
122 
123 /* Buffer for reporting pci register dumps. Its here in BSS, and
124  * not dynamically alloced, so that it ends up in RMO where RTAS
125  * can access it.
126  */
127 #define EEH_PCI_REGS_LOG_LEN 8192
128 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
129 
130 /*
131  * The struct is used to maintain the EEH global statistic
132  * information. Besides, the EEH global statistics will be
133  * exported to user space through procfs
134  */
135 struct eeh_stats {
136 	u64 no_device;		/* PCI device not found		*/
137 	u64 no_dn;		/* OF node not found		*/
138 	u64 no_cfg_addr;	/* Config address not found	*/
139 	u64 ignored_check;	/* EEH check skipped		*/
140 	u64 total_mmio_ffs;	/* Total EEH checks		*/
141 	u64 false_positives;	/* Unnecessary EEH checks	*/
142 	u64 slot_resets;	/* PE reset			*/
143 };
144 
145 static struct eeh_stats eeh_stats;
146 
147 static int __init eeh_setup(char *str)
148 {
149 	if (!strcmp(str, "off"))
150 		eeh_add_flag(EEH_FORCE_DISABLED);
151 	else if (!strcmp(str, "early_log"))
152 		eeh_add_flag(EEH_EARLY_DUMP_LOG);
153 
154 	return 1;
155 }
156 __setup("eeh=", eeh_setup);
157 
158 /*
159  * This routine captures assorted PCI configuration space data
160  * for the indicated PCI device, and puts them into a buffer
161  * for RTAS error logging.
162  */
163 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
164 {
165 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
166 	u32 cfg;
167 	int cap, i;
168 	int n = 0, l = 0;
169 	char buffer[128];
170 
171 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x.%01x\n",
172 		       edev->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 		edev->phb->global_number, pdn->busno,
176 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
177 
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);
181 
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);
185 
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);
191 
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);
195 	}
196 
197 	/* Dump out the PCI-X command and status regs */
198 	cap = edev->pcix_cap;
199 	if (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);
203 
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);
207 	}
208 
209 	/* If PCI-E capable, dump PCI-E cap 10 */
210 	cap = edev->pcie_cap;
211 	if (cap) {
212 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
213 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
214 
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);
218 
219 			if ((i % 4) == 0) {
220 				if (i != 0)
221 					pr_warn("%s\n", buffer);
222 
223 				l = scnprintf(buffer, sizeof(buffer),
224 					      "EEH: PCI-E %02x: %08x ",
225 					      4*i, cfg);
226 			} else {
227 				l += scnprintf(buffer+l, sizeof(buffer)-l,
228 					       "%08x ", cfg);
229 			}
230 
231 		}
232 
233 		pr_warn("%s\n", buffer);
234 	}
235 
236 	/* If AER capable, dump it */
237 	cap = edev->aer_cap;
238 	if (cap) {
239 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
240 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
241 
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);
245 
246 			if ((i % 4) == 0) {
247 				if (i != 0)
248 					pr_warn("%s\n", buffer);
249 
250 				l = scnprintf(buffer, sizeof(buffer),
251 					      "EEH: PCI-E AER %02x: %08x ",
252 					      4*i, cfg);
253 			} else {
254 				l += scnprintf(buffer+l, sizeof(buffer)-l,
255 					       "%08x ", cfg);
256 			}
257 		}
258 
259 		pr_warn("%s\n", buffer);
260 	}
261 
262 	return n;
263 }
264 
265 static void *eeh_dump_pe_log(void *data, void *flag)
266 {
267 	struct eeh_pe *pe = data;
268 	struct eeh_dev *edev, *tmp;
269 	size_t *plen = flag;
270 
271 	eeh_pe_for_each_dev(pe, edev, tmp)
272 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
273 					  EEH_PCI_REGS_LOG_LEN - *plen);
274 
275 	return NULL;
276 }
277 
278 /**
279  * eeh_slot_error_detail - Generate combined log including driver log and error log
280  * @pe: EEH PE
281  * @severity: temporary or permanent error log
282  *
283  * This routine should be called to generate the combined log, which
284  * is comprised of driver log and error log. The driver log is figured
285  * out from the config space of the corresponding PCI device, while
286  * the error log is fetched through platform dependent function call.
287  */
288 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
289 {
290 	size_t loglen = 0;
291 
292 	/*
293 	 * When the PHB is fenced or dead, it's pointless to collect
294 	 * the data from PCI config space because it should return
295 	 * 0xFF's. For ER, we still retrieve the data from the PCI
296 	 * config space.
297 	 *
298 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
299 	 * 0xFF's is always returned from PCI config space.
300 	 *
301 	 * When the @severity is EEH_LOG_PERM, the PE is going to be
302 	 * removed. Prior to that, the drivers for devices included in
303 	 * the PE will be closed. The drivers rely on working IO path
304 	 * to bring the devices to quiet state. Otherwise, PCI traffic
305 	 * from those devices after they are removed is like to cause
306 	 * another unexpected EEH error.
307 	 */
308 	if (!(pe->type & EEH_PE_PHB)) {
309 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG) ||
310 		    severity == EEH_LOG_PERM)
311 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
312 
313 		/*
314 		 * The config space of some PCI devices can't be accessed
315 		 * when their PEs are in frozen state. Otherwise, fenced
316 		 * PHB might be seen. Those PEs are identified with flag
317 		 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
318 		 * is set automatically when the PE is put to EEH_PE_ISOLATED.
319 		 *
320 		 * Restoring BARs possibly triggers PCI config access in
321 		 * (OPAL) firmware and then causes fenced PHB. If the
322 		 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
323 		 * pointless to restore BARs and dump config space.
324 		 */
325 		eeh_ops->configure_bridge(pe);
326 		if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
327 			eeh_pe_restore_bars(pe);
328 
329 			pci_regs_buf[0] = 0;
330 			eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
331 		}
332 	}
333 
334 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
335 }
336 
337 /**
338  * eeh_token_to_phys - Convert EEH address token to phys address
339  * @token: I/O token, should be address in the form 0xA....
340  *
341  * This routine should be called to convert virtual I/O address
342  * to physical one.
343  */
344 static inline unsigned long eeh_token_to_phys(unsigned long token)
345 {
346 	pte_t *ptep;
347 	unsigned long pa;
348 	int hugepage_shift;
349 
350 	/*
351 	 * We won't find hugepages here(this is iomem). Hence we are not
352 	 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
353 	 * page table free, because of init_mm.
354 	 */
355 	ptep = __find_linux_pte_or_hugepte(init_mm.pgd, token,
356 					   NULL, &hugepage_shift);
357 	if (!ptep)
358 		return token;
359 	WARN_ON(hugepage_shift);
360 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
361 
362 	return pa | (token & (PAGE_SIZE-1));
363 }
364 
365 /*
366  * On PowerNV platform, we might already have fenced PHB there.
367  * For that case, it's meaningless to recover frozen PE. Intead,
368  * We have to handle fenced PHB firstly.
369  */
370 static int eeh_phb_check_failure(struct eeh_pe *pe)
371 {
372 	struct eeh_pe *phb_pe;
373 	unsigned long flags;
374 	int ret;
375 
376 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
377 		return -EPERM;
378 
379 	/* Find the PHB PE */
380 	phb_pe = eeh_phb_pe_get(pe->phb);
381 	if (!phb_pe) {
382 		pr_warn("%s Can't find PE for PHB#%x\n",
383 			__func__, pe->phb->global_number);
384 		return -EEXIST;
385 	}
386 
387 	/* If the PHB has been in problematic state */
388 	eeh_serialize_lock(&flags);
389 	if (phb_pe->state & EEH_PE_ISOLATED) {
390 		ret = 0;
391 		goto out;
392 	}
393 
394 	/* Check PHB state */
395 	ret = eeh_ops->get_state(phb_pe, NULL);
396 	if ((ret < 0) ||
397 	    (ret == EEH_STATE_NOT_SUPPORT) ||
398 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
399 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
400 		ret = 0;
401 		goto out;
402 	}
403 
404 	/* Isolate the PHB and send event */
405 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
406 	eeh_serialize_unlock(flags);
407 
408 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
409 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
410 	dump_stack();
411 	eeh_send_failure_event(phb_pe);
412 
413 	return 1;
414 out:
415 	eeh_serialize_unlock(flags);
416 	return ret;
417 }
418 
419 /**
420  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
421  * @edev: eeh device
422  *
423  * Check for an EEH failure for the given device node.  Call this
424  * routine if the result of a read was all 0xff's and you want to
425  * find out if this is due to an EEH slot freeze.  This routine
426  * will query firmware for the EEH status.
427  *
428  * Returns 0 if there has not been an EEH error; otherwise returns
429  * a non-zero value and queues up a slot isolation event notification.
430  *
431  * It is safe to call this routine in an interrupt context.
432  */
433 int eeh_dev_check_failure(struct eeh_dev *edev)
434 {
435 	int ret;
436 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
437 	unsigned long flags;
438 	struct pci_dn *pdn;
439 	struct pci_dev *dev;
440 	struct eeh_pe *pe, *parent_pe, *phb_pe;
441 	int rc = 0;
442 	const char *location = NULL;
443 
444 	eeh_stats.total_mmio_ffs++;
445 
446 	if (!eeh_enabled())
447 		return 0;
448 
449 	if (!edev) {
450 		eeh_stats.no_dn++;
451 		return 0;
452 	}
453 	dev = eeh_dev_to_pci_dev(edev);
454 	pe = eeh_dev_to_pe(edev);
455 
456 	/* Access to IO BARs might get this far and still not want checking. */
457 	if (!pe) {
458 		eeh_stats.ignored_check++;
459 		pr_debug("EEH: Ignored check for %s\n",
460 			eeh_pci_name(dev));
461 		return 0;
462 	}
463 
464 	if (!pe->addr && !pe->config_addr) {
465 		eeh_stats.no_cfg_addr++;
466 		return 0;
467 	}
468 
469 	/*
470 	 * On PowerNV platform, we might already have fenced PHB
471 	 * there and we need take care of that firstly.
472 	 */
473 	ret = eeh_phb_check_failure(pe);
474 	if (ret > 0)
475 		return ret;
476 
477 	/*
478 	 * If the PE isn't owned by us, we shouldn't check the
479 	 * state. Instead, let the owner handle it if the PE has
480 	 * been frozen.
481 	 */
482 	if (eeh_pe_passed(pe))
483 		return 0;
484 
485 	/* If we already have a pending isolation event for this
486 	 * slot, we know it's bad already, we don't need to check.
487 	 * Do this checking under a lock; as multiple PCI devices
488 	 * in one slot might report errors simultaneously, and we
489 	 * only want one error recovery routine running.
490 	 */
491 	eeh_serialize_lock(&flags);
492 	rc = 1;
493 	if (pe->state & EEH_PE_ISOLATED) {
494 		pe->check_count++;
495 		if (pe->check_count % EEH_MAX_FAILS == 0) {
496 			pdn = eeh_dev_to_pdn(edev);
497 			if (pdn->node)
498 				location = of_get_property(pdn->node, "ibm,loc-code", NULL);
499 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
500 				"location=%s driver=%s pci addr=%s\n",
501 				pe->check_count,
502 				location ? location : "unknown",
503 				eeh_driver_name(dev), eeh_pci_name(dev));
504 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
505 				eeh_driver_name(dev));
506 			dump_stack();
507 		}
508 		goto dn_unlock;
509 	}
510 
511 	/*
512 	 * Now test for an EEH failure.  This is VERY expensive.
513 	 * Note that the eeh_config_addr may be a parent device
514 	 * in the case of a device behind a bridge, or it may be
515 	 * function zero of a multi-function device.
516 	 * In any case they must share a common PHB.
517 	 */
518 	ret = eeh_ops->get_state(pe, NULL);
519 
520 	/* Note that config-io to empty slots may fail;
521 	 * they are empty when they don't have children.
522 	 * We will punt with the following conditions: Failure to get
523 	 * PE's state, EEH not support and Permanently unavailable
524 	 * state, PE is in good state.
525 	 */
526 	if ((ret < 0) ||
527 	    (ret == EEH_STATE_NOT_SUPPORT) ||
528 	    ((ret & active_flags) == active_flags)) {
529 		eeh_stats.false_positives++;
530 		pe->false_positives++;
531 		rc = 0;
532 		goto dn_unlock;
533 	}
534 
535 	/*
536 	 * It should be corner case that the parent PE has been
537 	 * put into frozen state as well. We should take care
538 	 * that at first.
539 	 */
540 	parent_pe = pe->parent;
541 	while (parent_pe) {
542 		/* Hit the ceiling ? */
543 		if (parent_pe->type & EEH_PE_PHB)
544 			break;
545 
546 		/* Frozen parent PE ? */
547 		ret = eeh_ops->get_state(parent_pe, NULL);
548 		if (ret > 0 &&
549 		    (ret & active_flags) != active_flags)
550 			pe = parent_pe;
551 
552 		/* Next parent level */
553 		parent_pe = parent_pe->parent;
554 	}
555 
556 	eeh_stats.slot_resets++;
557 
558 	/* Avoid repeated reports of this failure, including problems
559 	 * with other functions on this device, and functions under
560 	 * bridges.
561 	 */
562 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
563 	eeh_serialize_unlock(flags);
564 
565 	/* Most EEH events are due to device driver bugs.  Having
566 	 * a stack trace will help the device-driver authors figure
567 	 * out what happened.  So print that out.
568 	 */
569 	phb_pe = eeh_phb_pe_get(pe->phb);
570 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
571 	       pe->phb->global_number, pe->addr);
572 	pr_err("EEH: PE location: %s, PHB location: %s\n",
573 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
574 	dump_stack();
575 
576 	eeh_send_failure_event(pe);
577 
578 	return 1;
579 
580 dn_unlock:
581 	eeh_serialize_unlock(flags);
582 	return rc;
583 }
584 
585 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
586 
587 /**
588  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
589  * @token: I/O address
590  *
591  * Check for an EEH failure at the given I/O address. Call this
592  * routine if the result of a read was all 0xff's and you want to
593  * find out if this is due to an EEH slot freeze event. This routine
594  * will query firmware for the EEH status.
595  *
596  * Note this routine is safe to call in an interrupt context.
597  */
598 int eeh_check_failure(const volatile void __iomem *token)
599 {
600 	unsigned long addr;
601 	struct eeh_dev *edev;
602 
603 	/* Finding the phys addr + pci device; this is pretty quick. */
604 	addr = eeh_token_to_phys((unsigned long __force) token);
605 	edev = eeh_addr_cache_get_dev(addr);
606 	if (!edev) {
607 		eeh_stats.no_device++;
608 		return 0;
609 	}
610 
611 	return eeh_dev_check_failure(edev);
612 }
613 EXPORT_SYMBOL(eeh_check_failure);
614 
615 
616 /**
617  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
618  * @pe: EEH PE
619  *
620  * This routine should be called to reenable frozen MMIO or DMA
621  * so that it would work correctly again. It's useful while doing
622  * recovery or log collection on the indicated device.
623  */
624 int eeh_pci_enable(struct eeh_pe *pe, int function)
625 {
626 	int active_flag, rc;
627 
628 	/*
629 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
630 	 * Also, it's pointless to enable them on unfrozen PE. So
631 	 * we have to check before enabling IO or DMA.
632 	 */
633 	switch (function) {
634 	case EEH_OPT_THAW_MMIO:
635 		active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
636 		break;
637 	case EEH_OPT_THAW_DMA:
638 		active_flag = EEH_STATE_DMA_ACTIVE;
639 		break;
640 	case EEH_OPT_DISABLE:
641 	case EEH_OPT_ENABLE:
642 	case EEH_OPT_FREEZE_PE:
643 		active_flag = 0;
644 		break;
645 	default:
646 		pr_warn("%s: Invalid function %d\n",
647 			__func__, function);
648 		return -EINVAL;
649 	}
650 
651 	/*
652 	 * Check if IO or DMA has been enabled before
653 	 * enabling them.
654 	 */
655 	if (active_flag) {
656 		rc = eeh_ops->get_state(pe, NULL);
657 		if (rc < 0)
658 			return rc;
659 
660 		/* Needn't enable it at all */
661 		if (rc == EEH_STATE_NOT_SUPPORT)
662 			return 0;
663 
664 		/* It's already enabled */
665 		if (rc & active_flag)
666 			return 0;
667 	}
668 
669 
670 	/* Issue the request */
671 	rc = eeh_ops->set_option(pe, function);
672 	if (rc)
673 		pr_warn("%s: Unexpected state change %d on "
674 			"PHB#%x-PE#%x, err=%d\n",
675 			__func__, function, pe->phb->global_number,
676 			pe->addr, rc);
677 
678 	/* Check if the request is finished successfully */
679 	if (active_flag) {
680 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
681 		if (rc < 0)
682 			return rc;
683 
684 		if (rc & active_flag)
685 			return 0;
686 
687 		return -EIO;
688 	}
689 
690 	return rc;
691 }
692 
693 static void *eeh_disable_and_save_dev_state(void *data, void *userdata)
694 {
695 	struct eeh_dev *edev = data;
696 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
697 	struct pci_dev *dev = userdata;
698 
699 	/*
700 	 * The caller should have disabled and saved the
701 	 * state for the specified device
702 	 */
703 	if (!pdev || pdev == dev)
704 		return NULL;
705 
706 	/* Ensure we have D0 power state */
707 	pci_set_power_state(pdev, PCI_D0);
708 
709 	/* Save device state */
710 	pci_save_state(pdev);
711 
712 	/*
713 	 * Disable device to avoid any DMA traffic and
714 	 * interrupt from the device
715 	 */
716 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
717 
718 	return NULL;
719 }
720 
721 static void *eeh_restore_dev_state(void *data, void *userdata)
722 {
723 	struct eeh_dev *edev = data;
724 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
725 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
726 	struct pci_dev *dev = userdata;
727 
728 	if (!pdev)
729 		return NULL;
730 
731 	/* Apply customization from firmware */
732 	if (pdn && eeh_ops->restore_config)
733 		eeh_ops->restore_config(pdn);
734 
735 	/* The caller should restore state for the specified device */
736 	if (pdev != dev)
737 		pci_restore_state(pdev);
738 
739 	return NULL;
740 }
741 
742 /**
743  * pcibios_set_pcie_reset_state - Set PCI-E reset state
744  * @dev: pci device struct
745  * @state: reset state to enter
746  *
747  * Return value:
748  * 	0 if success
749  */
750 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
751 {
752 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
753 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
754 
755 	if (!pe) {
756 		pr_err("%s: No PE found on PCI device %s\n",
757 			__func__, pci_name(dev));
758 		return -EINVAL;
759 	}
760 
761 	switch (state) {
762 	case pcie_deassert_reset:
763 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
764 		eeh_unfreeze_pe(pe, false);
765 		if (!(pe->type & EEH_PE_VF))
766 			eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
767 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
768 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
769 		break;
770 	case pcie_hot_reset:
771 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
772 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
773 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
774 		if (!(pe->type & EEH_PE_VF))
775 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
776 		eeh_ops->reset(pe, EEH_RESET_HOT);
777 		break;
778 	case pcie_warm_reset:
779 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
780 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
781 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
782 		if (!(pe->type & EEH_PE_VF))
783 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
784 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
785 		break;
786 	default:
787 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
788 		return -EINVAL;
789 	};
790 
791 	return 0;
792 }
793 
794 /**
795  * eeh_set_pe_freset - Check the required reset for the indicated device
796  * @data: EEH device
797  * @flag: return value
798  *
799  * Each device might have its preferred reset type: fundamental or
800  * hot reset. The routine is used to collected the information for
801  * the indicated device and its children so that the bunch of the
802  * devices could be reset properly.
803  */
804 static void *eeh_set_dev_freset(void *data, void *flag)
805 {
806 	struct pci_dev *dev;
807 	unsigned int *freset = (unsigned int *)flag;
808 	struct eeh_dev *edev = (struct eeh_dev *)data;
809 
810 	dev = eeh_dev_to_pci_dev(edev);
811 	if (dev)
812 		*freset |= dev->needs_freset;
813 
814 	return NULL;
815 }
816 
817 /**
818  * eeh_pe_reset_full - Complete a full reset process on the indicated PE
819  * @pe: EEH PE
820  *
821  * This function executes a full reset procedure on a PE, including setting
822  * the appropriate flags, performing a fundamental or hot reset, and then
823  * deactivating the reset status.  It is designed to be used within the EEH
824  * subsystem, as opposed to eeh_pe_reset which is exported to drivers and
825  * only performs a single operation at a time.
826  *
827  * This function will attempt to reset a PE three times before failing.
828  */
829 int eeh_pe_reset_full(struct eeh_pe *pe)
830 {
831 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
832 	int reset_state = (EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
833 	int type = EEH_RESET_HOT;
834 	unsigned int freset = 0;
835 	int i, state, ret;
836 
837 	/*
838 	 * Determine the type of reset to perform - hot or fundamental.
839 	 * Hot reset is the default operation, unless any device under the
840 	 * PE requires a fundamental reset.
841 	 */
842 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
843 
844 	if (freset)
845 		type = EEH_RESET_FUNDAMENTAL;
846 
847 	/* Mark the PE as in reset state and block config space accesses */
848 	eeh_pe_state_mark(pe, reset_state);
849 
850 	/* Make three attempts at resetting the bus */
851 	for (i = 0; i < 3; i++) {
852 		ret = eeh_pe_reset(pe, type);
853 		if (ret)
854 			break;
855 
856 		ret = eeh_pe_reset(pe, EEH_RESET_DEACTIVATE);
857 		if (ret)
858 			break;
859 
860 		/* Wait until the PE is in a functioning state */
861 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
862 		if ((state & active_flags) == active_flags)
863 			break;
864 
865 		if (state < 0) {
866 			pr_warn("%s: Unrecoverable slot failure on PHB#%x-PE#%x",
867 				__func__, pe->phb->global_number, pe->addr);
868 			ret = -ENOTRECOVERABLE;
869 			break;
870 		}
871 
872 		/* Set error in case this is our last attempt */
873 		ret = -EIO;
874 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
875 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
876 	}
877 
878 	eeh_pe_state_clear(pe, reset_state);
879 	return ret;
880 }
881 
882 /**
883  * eeh_save_bars - Save device bars
884  * @edev: PCI device associated EEH device
885  *
886  * Save the values of the device bars. Unlike the restore
887  * routine, this routine is *not* recursive. This is because
888  * PCI devices are added individually; but, for the restore,
889  * an entire slot is reset at a time.
890  */
891 void eeh_save_bars(struct eeh_dev *edev)
892 {
893 	struct pci_dn *pdn;
894 	int i;
895 
896 	pdn = eeh_dev_to_pdn(edev);
897 	if (!pdn)
898 		return;
899 
900 	for (i = 0; i < 16; i++)
901 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
902 
903 	/*
904 	 * For PCI bridges including root port, we need enable bus
905 	 * master explicitly. Otherwise, it can't fetch IODA table
906 	 * entries correctly. So we cache the bit in advance so that
907 	 * we can restore it after reset, either PHB range or PE range.
908 	 */
909 	if (edev->mode & EEH_DEV_BRIDGE)
910 		edev->config_space[1] |= PCI_COMMAND_MASTER;
911 }
912 
913 /**
914  * eeh_ops_register - Register platform dependent EEH operations
915  * @ops: platform dependent EEH operations
916  *
917  * Register the platform dependent EEH operation callback
918  * functions. The platform should call this function before
919  * any other EEH operations.
920  */
921 int __init eeh_ops_register(struct eeh_ops *ops)
922 {
923 	if (!ops->name) {
924 		pr_warn("%s: Invalid EEH ops name for %p\n",
925 			__func__, ops);
926 		return -EINVAL;
927 	}
928 
929 	if (eeh_ops && eeh_ops != ops) {
930 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
931 			__func__, eeh_ops->name, ops->name);
932 		return -EEXIST;
933 	}
934 
935 	eeh_ops = ops;
936 
937 	return 0;
938 }
939 
940 /**
941  * eeh_ops_unregister - Unreigster platform dependent EEH operations
942  * @name: name of EEH platform operations
943  *
944  * Unregister the platform dependent EEH operation callback
945  * functions.
946  */
947 int __exit eeh_ops_unregister(const char *name)
948 {
949 	if (!name || !strlen(name)) {
950 		pr_warn("%s: Invalid EEH ops name\n",
951 			__func__);
952 		return -EINVAL;
953 	}
954 
955 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
956 		eeh_ops = NULL;
957 		return 0;
958 	}
959 
960 	return -EEXIST;
961 }
962 
963 static int eeh_reboot_notifier(struct notifier_block *nb,
964 			       unsigned long action, void *unused)
965 {
966 	eeh_clear_flag(EEH_ENABLED);
967 	return NOTIFY_DONE;
968 }
969 
970 static struct notifier_block eeh_reboot_nb = {
971 	.notifier_call = eeh_reboot_notifier,
972 };
973 
974 /**
975  * eeh_init - EEH initialization
976  *
977  * Initialize EEH by trying to enable it for all of the adapters in the system.
978  * As a side effect we can determine here if eeh is supported at all.
979  * Note that we leave EEH on so failed config cycles won't cause a machine
980  * check.  If a user turns off EEH for a particular adapter they are really
981  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
982  * grant access to a slot if EEH isn't enabled, and so we always enable
983  * EEH for all slots/all devices.
984  *
985  * The eeh-force-off option disables EEH checking globally, for all slots.
986  * Even if force-off is set, the EEH hardware is still enabled, so that
987  * newer systems can boot.
988  */
989 int eeh_init(void)
990 {
991 	struct pci_controller *hose, *tmp;
992 	struct pci_dn *pdn;
993 	static int cnt = 0;
994 	int ret = 0;
995 
996 	/*
997 	 * We have to delay the initialization on PowerNV after
998 	 * the PCI hierarchy tree has been built because the PEs
999 	 * are figured out based on PCI devices instead of device
1000 	 * tree nodes
1001 	 */
1002 	if (machine_is(powernv) && cnt++ <= 0)
1003 		return ret;
1004 
1005 	/* Register reboot notifier */
1006 	ret = register_reboot_notifier(&eeh_reboot_nb);
1007 	if (ret) {
1008 		pr_warn("%s: Failed to register notifier (%d)\n",
1009 			__func__, ret);
1010 		return ret;
1011 	}
1012 
1013 	/* call platform initialization function */
1014 	if (!eeh_ops) {
1015 		pr_warn("%s: Platform EEH operation not found\n",
1016 			__func__);
1017 		return -EEXIST;
1018 	} else if ((ret = eeh_ops->init()))
1019 		return ret;
1020 
1021 	/* Initialize EEH event */
1022 	ret = eeh_event_init();
1023 	if (ret)
1024 		return ret;
1025 
1026 	/* Enable EEH for all adapters */
1027 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1028 		pdn = hose->pci_data;
1029 		traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1030 	}
1031 
1032 	/*
1033 	 * Call platform post-initialization. Actually, It's good chance
1034 	 * to inform platform that EEH is ready to supply service if the
1035 	 * I/O cache stuff has been built up.
1036 	 */
1037 	if (eeh_ops->post_init) {
1038 		ret = eeh_ops->post_init();
1039 		if (ret)
1040 			return ret;
1041 	}
1042 
1043 	if (eeh_enabled())
1044 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1045 	else
1046 		pr_info("EEH: No capable adapters found\n");
1047 
1048 	return ret;
1049 }
1050 
1051 core_initcall_sync(eeh_init);
1052 
1053 /**
1054  * eeh_add_device_early - Enable EEH for the indicated device node
1055  * @pdn: PCI device node for which to set up EEH
1056  *
1057  * This routine must be used to perform EEH initialization for PCI
1058  * devices that were added after system boot (e.g. hotplug, dlpar).
1059  * This routine must be called before any i/o is performed to the
1060  * adapter (inluding any config-space i/o).
1061  * Whether this actually enables EEH or not for this device depends
1062  * on the CEC architecture, type of the device, on earlier boot
1063  * command-line arguments & etc.
1064  */
1065 void eeh_add_device_early(struct pci_dn *pdn)
1066 {
1067 	struct pci_controller *phb;
1068 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1069 
1070 	if (!edev)
1071 		return;
1072 
1073 	if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1074 		return;
1075 
1076 	/* USB Bus children of PCI devices will not have BUID's */
1077 	phb = edev->phb;
1078 	if (NULL == phb ||
1079 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1080 		return;
1081 
1082 	eeh_ops->probe(pdn, NULL);
1083 }
1084 
1085 /**
1086  * eeh_add_device_tree_early - Enable EEH for the indicated device
1087  * @pdn: PCI device node
1088  *
1089  * This routine must be used to perform EEH initialization for the
1090  * indicated PCI device that was added after system boot (e.g.
1091  * hotplug, dlpar).
1092  */
1093 void eeh_add_device_tree_early(struct pci_dn *pdn)
1094 {
1095 	struct pci_dn *n;
1096 
1097 	if (!pdn)
1098 		return;
1099 
1100 	list_for_each_entry(n, &pdn->child_list, list)
1101 		eeh_add_device_tree_early(n);
1102 	eeh_add_device_early(pdn);
1103 }
1104 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1105 
1106 /**
1107  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1108  * @dev: pci device for which to set up EEH
1109  *
1110  * This routine must be used to complete EEH initialization for PCI
1111  * devices that were added after system boot (e.g. hotplug, dlpar).
1112  */
1113 void eeh_add_device_late(struct pci_dev *dev)
1114 {
1115 	struct pci_dn *pdn;
1116 	struct eeh_dev *edev;
1117 
1118 	if (!dev || !eeh_enabled())
1119 		return;
1120 
1121 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1122 
1123 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1124 	edev = pdn_to_eeh_dev(pdn);
1125 	if (edev->pdev == dev) {
1126 		pr_debug("EEH: Already referenced !\n");
1127 		return;
1128 	}
1129 
1130 	/*
1131 	 * The EEH cache might not be removed correctly because of
1132 	 * unbalanced kref to the device during unplug time, which
1133 	 * relies on pcibios_release_device(). So we have to remove
1134 	 * that here explicitly.
1135 	 */
1136 	if (edev->pdev) {
1137 		eeh_rmv_from_parent_pe(edev);
1138 		eeh_addr_cache_rmv_dev(edev->pdev);
1139 		eeh_sysfs_remove_device(edev->pdev);
1140 		edev->mode &= ~EEH_DEV_SYSFS;
1141 
1142 		/*
1143 		 * We definitely should have the PCI device removed
1144 		 * though it wasn't correctly. So we needn't call
1145 		 * into error handler afterwards.
1146 		 */
1147 		edev->mode |= EEH_DEV_NO_HANDLER;
1148 
1149 		edev->pdev = NULL;
1150 		dev->dev.archdata.edev = NULL;
1151 	}
1152 
1153 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1154 		eeh_ops->probe(pdn, NULL);
1155 
1156 	edev->pdev = dev;
1157 	dev->dev.archdata.edev = edev;
1158 
1159 	eeh_addr_cache_insert_dev(dev);
1160 }
1161 
1162 /**
1163  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1164  * @bus: PCI bus
1165  *
1166  * This routine must be used to perform EEH initialization for PCI
1167  * devices which are attached to the indicated PCI bus. The PCI bus
1168  * is added after system boot through hotplug or dlpar.
1169  */
1170 void eeh_add_device_tree_late(struct pci_bus *bus)
1171 {
1172 	struct pci_dev *dev;
1173 
1174 	list_for_each_entry(dev, &bus->devices, bus_list) {
1175 		eeh_add_device_late(dev);
1176 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1177 			struct pci_bus *subbus = dev->subordinate;
1178 			if (subbus)
1179 				eeh_add_device_tree_late(subbus);
1180 		}
1181 	}
1182 }
1183 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1184 
1185 /**
1186  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1187  * @bus: PCI bus
1188  *
1189  * This routine must be used to add EEH sysfs files for PCI
1190  * devices which are attached to the indicated PCI bus. The PCI bus
1191  * is added after system boot through hotplug or dlpar.
1192  */
1193 void eeh_add_sysfs_files(struct pci_bus *bus)
1194 {
1195 	struct pci_dev *dev;
1196 
1197 	list_for_each_entry(dev, &bus->devices, bus_list) {
1198 		eeh_sysfs_add_device(dev);
1199 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1200 			struct pci_bus *subbus = dev->subordinate;
1201 			if (subbus)
1202 				eeh_add_sysfs_files(subbus);
1203 		}
1204 	}
1205 }
1206 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1207 
1208 /**
1209  * eeh_remove_device - Undo EEH setup for the indicated pci device
1210  * @dev: pci device to be removed
1211  *
1212  * This routine should be called when a device is removed from
1213  * a running system (e.g. by hotplug or dlpar).  It unregisters
1214  * the PCI device from the EEH subsystem.  I/O errors affecting
1215  * this device will no longer be detected after this call; thus,
1216  * i/o errors affecting this slot may leave this device unusable.
1217  */
1218 void eeh_remove_device(struct pci_dev *dev)
1219 {
1220 	struct eeh_dev *edev;
1221 
1222 	if (!dev || !eeh_enabled())
1223 		return;
1224 	edev = pci_dev_to_eeh_dev(dev);
1225 
1226 	/* Unregister the device with the EEH/PCI address search system */
1227 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1228 
1229 	if (!edev || !edev->pdev || !edev->pe) {
1230 		pr_debug("EEH: Not referenced !\n");
1231 		return;
1232 	}
1233 
1234 	/*
1235 	 * During the hotplug for EEH error recovery, we need the EEH
1236 	 * device attached to the parent PE in order for BAR restore
1237 	 * a bit later. So we keep it for BAR restore and remove it
1238 	 * from the parent PE during the BAR resotre.
1239 	 */
1240 	edev->pdev = NULL;
1241 
1242 	/*
1243 	 * The flag "in_error" is used to trace EEH devices for VFs
1244 	 * in error state or not. It's set in eeh_report_error(). If
1245 	 * it's not set, eeh_report_{reset,resume}() won't be called
1246 	 * for the VF EEH device.
1247 	 */
1248 	edev->in_error = false;
1249 	dev->dev.archdata.edev = NULL;
1250 	if (!(edev->pe->state & EEH_PE_KEEP))
1251 		eeh_rmv_from_parent_pe(edev);
1252 	else
1253 		edev->mode |= EEH_DEV_DISCONNECTED;
1254 
1255 	/*
1256 	 * We're removing from the PCI subsystem, that means
1257 	 * the PCI device driver can't support EEH or not
1258 	 * well. So we rely on hotplug completely to do recovery
1259 	 * for the specific PCI device.
1260 	 */
1261 	edev->mode |= EEH_DEV_NO_HANDLER;
1262 
1263 	eeh_addr_cache_rmv_dev(dev);
1264 	eeh_sysfs_remove_device(dev);
1265 	edev->mode &= ~EEH_DEV_SYSFS;
1266 }
1267 
1268 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1269 {
1270 	int ret;
1271 
1272 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1273 	if (ret) {
1274 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1275 			__func__, ret, pe->phb->global_number, pe->addr);
1276 		return ret;
1277 	}
1278 
1279 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1280 	if (ret) {
1281 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1282 			__func__, ret, pe->phb->global_number, pe->addr);
1283 		return ret;
1284 	}
1285 
1286 	/* Clear software isolated state */
1287 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1288 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1289 
1290 	return ret;
1291 }
1292 
1293 
1294 static struct pci_device_id eeh_reset_ids[] = {
1295 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1296 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1297 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1298 	{ 0 }
1299 };
1300 
1301 static int eeh_pe_change_owner(struct eeh_pe *pe)
1302 {
1303 	struct eeh_dev *edev, *tmp;
1304 	struct pci_dev *pdev;
1305 	struct pci_device_id *id;
1306 	int flags, ret;
1307 
1308 	/* Check PE state */
1309 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1310 	ret = eeh_ops->get_state(pe, NULL);
1311 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1312 		return 0;
1313 
1314 	/* Unfrozen PE, nothing to do */
1315 	if ((ret & flags) == flags)
1316 		return 0;
1317 
1318 	/* Frozen PE, check if it needs PE level reset */
1319 	eeh_pe_for_each_dev(pe, edev, tmp) {
1320 		pdev = eeh_dev_to_pci_dev(edev);
1321 		if (!pdev)
1322 			continue;
1323 
1324 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1325 			if (id->vendor != PCI_ANY_ID &&
1326 			    id->vendor != pdev->vendor)
1327 				continue;
1328 			if (id->device != PCI_ANY_ID &&
1329 			    id->device != pdev->device)
1330 				continue;
1331 			if (id->subvendor != PCI_ANY_ID &&
1332 			    id->subvendor != pdev->subsystem_vendor)
1333 				continue;
1334 			if (id->subdevice != PCI_ANY_ID &&
1335 			    id->subdevice != pdev->subsystem_device)
1336 				continue;
1337 
1338 			return eeh_pe_reset_and_recover(pe);
1339 		}
1340 	}
1341 
1342 	return eeh_unfreeze_pe(pe, true);
1343 }
1344 
1345 /**
1346  * eeh_dev_open - Increase count of pass through devices for PE
1347  * @pdev: PCI device
1348  *
1349  * Increase count of passed through devices for the indicated
1350  * PE. In the result, the EEH errors detected on the PE won't be
1351  * reported. The PE owner will be responsible for detection
1352  * and recovery.
1353  */
1354 int eeh_dev_open(struct pci_dev *pdev)
1355 {
1356 	struct eeh_dev *edev;
1357 	int ret = -ENODEV;
1358 
1359 	mutex_lock(&eeh_dev_mutex);
1360 
1361 	/* No PCI device ? */
1362 	if (!pdev)
1363 		goto out;
1364 
1365 	/* No EEH device or PE ? */
1366 	edev = pci_dev_to_eeh_dev(pdev);
1367 	if (!edev || !edev->pe)
1368 		goto out;
1369 
1370 	/*
1371 	 * The PE might have been put into frozen state, but we
1372 	 * didn't detect that yet. The passed through PCI devices
1373 	 * in frozen PE won't work properly. Clear the frozen state
1374 	 * in advance.
1375 	 */
1376 	ret = eeh_pe_change_owner(edev->pe);
1377 	if (ret)
1378 		goto out;
1379 
1380 	/* Increase PE's pass through count */
1381 	atomic_inc(&edev->pe->pass_dev_cnt);
1382 	mutex_unlock(&eeh_dev_mutex);
1383 
1384 	return 0;
1385 out:
1386 	mutex_unlock(&eeh_dev_mutex);
1387 	return ret;
1388 }
1389 EXPORT_SYMBOL_GPL(eeh_dev_open);
1390 
1391 /**
1392  * eeh_dev_release - Decrease count of pass through devices for PE
1393  * @pdev: PCI device
1394  *
1395  * Decrease count of pass through devices for the indicated PE. If
1396  * there is no passed through device in PE, the EEH errors detected
1397  * on the PE will be reported and handled as usual.
1398  */
1399 void eeh_dev_release(struct pci_dev *pdev)
1400 {
1401 	struct eeh_dev *edev;
1402 
1403 	mutex_lock(&eeh_dev_mutex);
1404 
1405 	/* No PCI device ? */
1406 	if (!pdev)
1407 		goto out;
1408 
1409 	/* No EEH device ? */
1410 	edev = pci_dev_to_eeh_dev(pdev);
1411 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1412 		goto out;
1413 
1414 	/* Decrease PE's pass through count */
1415 	WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1416 	eeh_pe_change_owner(edev->pe);
1417 out:
1418 	mutex_unlock(&eeh_dev_mutex);
1419 }
1420 EXPORT_SYMBOL(eeh_dev_release);
1421 
1422 #ifdef CONFIG_IOMMU_API
1423 
1424 static int dev_has_iommu_table(struct device *dev, void *data)
1425 {
1426 	struct pci_dev *pdev = to_pci_dev(dev);
1427 	struct pci_dev **ppdev = data;
1428 
1429 	if (!dev)
1430 		return 0;
1431 
1432 	if (dev->iommu_group) {
1433 		*ppdev = pdev;
1434 		return 1;
1435 	}
1436 
1437 	return 0;
1438 }
1439 
1440 /**
1441  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1442  * @group: IOMMU group
1443  *
1444  * The routine is called to convert IOMMU group to EEH PE.
1445  */
1446 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1447 {
1448 	struct pci_dev *pdev = NULL;
1449 	struct eeh_dev *edev;
1450 	int ret;
1451 
1452 	/* No IOMMU group ? */
1453 	if (!group)
1454 		return NULL;
1455 
1456 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1457 	if (!ret || !pdev)
1458 		return NULL;
1459 
1460 	/* No EEH device or PE ? */
1461 	edev = pci_dev_to_eeh_dev(pdev);
1462 	if (!edev || !edev->pe)
1463 		return NULL;
1464 
1465 	return edev->pe;
1466 }
1467 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1468 
1469 #endif /* CONFIG_IOMMU_API */
1470 
1471 /**
1472  * eeh_pe_set_option - Set options for the indicated PE
1473  * @pe: EEH PE
1474  * @option: requested option
1475  *
1476  * The routine is called to enable or disable EEH functionality
1477  * on the indicated PE, to enable IO or DMA for the frozen PE.
1478  */
1479 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1480 {
1481 	int ret = 0;
1482 
1483 	/* Invalid PE ? */
1484 	if (!pe)
1485 		return -ENODEV;
1486 
1487 	/*
1488 	 * EEH functionality could possibly be disabled, just
1489 	 * return error for the case. And the EEH functinality
1490 	 * isn't expected to be disabled on one specific PE.
1491 	 */
1492 	switch (option) {
1493 	case EEH_OPT_ENABLE:
1494 		if (eeh_enabled()) {
1495 			ret = eeh_pe_change_owner(pe);
1496 			break;
1497 		}
1498 		ret = -EIO;
1499 		break;
1500 	case EEH_OPT_DISABLE:
1501 		break;
1502 	case EEH_OPT_THAW_MMIO:
1503 	case EEH_OPT_THAW_DMA:
1504 	case EEH_OPT_FREEZE_PE:
1505 		if (!eeh_ops || !eeh_ops->set_option) {
1506 			ret = -ENOENT;
1507 			break;
1508 		}
1509 
1510 		ret = eeh_pci_enable(pe, option);
1511 		break;
1512 	default:
1513 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1514 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1515 		ret = -EINVAL;
1516 	}
1517 
1518 	return ret;
1519 }
1520 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1521 
1522 /**
1523  * eeh_pe_get_state - Retrieve PE's state
1524  * @pe: EEH PE
1525  *
1526  * Retrieve the PE's state, which includes 3 aspects: enabled
1527  * DMA, enabled IO and asserted reset.
1528  */
1529 int eeh_pe_get_state(struct eeh_pe *pe)
1530 {
1531 	int result, ret = 0;
1532 	bool rst_active, dma_en, mmio_en;
1533 
1534 	/* Existing PE ? */
1535 	if (!pe)
1536 		return -ENODEV;
1537 
1538 	if (!eeh_ops || !eeh_ops->get_state)
1539 		return -ENOENT;
1540 
1541 	/*
1542 	 * If the parent PE is owned by the host kernel and is undergoing
1543 	 * error recovery, we should return the PE state as temporarily
1544 	 * unavailable so that the error recovery on the guest is suspended
1545 	 * until the recovery completes on the host.
1546 	 */
1547 	if (pe->parent &&
1548 	    !(pe->state & EEH_PE_REMOVED) &&
1549 	    (pe->parent->state & (EEH_PE_ISOLATED | EEH_PE_RECOVERING)))
1550 		return EEH_PE_STATE_UNAVAIL;
1551 
1552 	result = eeh_ops->get_state(pe, NULL);
1553 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1554 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1555 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1556 
1557 	if (rst_active)
1558 		ret = EEH_PE_STATE_RESET;
1559 	else if (dma_en && mmio_en)
1560 		ret = EEH_PE_STATE_NORMAL;
1561 	else if (!dma_en && !mmio_en)
1562 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1563 	else if (!dma_en && mmio_en)
1564 		ret = EEH_PE_STATE_STOPPED_DMA;
1565 	else
1566 		ret = EEH_PE_STATE_UNAVAIL;
1567 
1568 	return ret;
1569 }
1570 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1571 
1572 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1573 {
1574 	struct eeh_dev *edev, *tmp;
1575 	struct pci_dev *pdev;
1576 	int ret = 0;
1577 
1578 	/* Restore config space */
1579 	eeh_pe_restore_bars(pe);
1580 
1581 	/*
1582 	 * Reenable PCI devices as the devices passed
1583 	 * through are always enabled before the reset.
1584 	 */
1585 	eeh_pe_for_each_dev(pe, edev, tmp) {
1586 		pdev = eeh_dev_to_pci_dev(edev);
1587 		if (!pdev)
1588 			continue;
1589 
1590 		ret = pci_reenable_device(pdev);
1591 		if (ret) {
1592 			pr_warn("%s: Failure %d reenabling %s\n",
1593 				__func__, ret, pci_name(pdev));
1594 			return ret;
1595 		}
1596 	}
1597 
1598 	/* The PE is still in frozen state */
1599 	return eeh_unfreeze_pe(pe, true);
1600 }
1601 
1602 
1603 /**
1604  * eeh_pe_reset - Issue PE reset according to specified type
1605  * @pe: EEH PE
1606  * @option: reset type
1607  *
1608  * The routine is called to reset the specified PE with the
1609  * indicated type, either fundamental reset or hot reset.
1610  * PE reset is the most important part for error recovery.
1611  */
1612 int eeh_pe_reset(struct eeh_pe *pe, int option)
1613 {
1614 	int ret = 0;
1615 
1616 	/* Invalid PE ? */
1617 	if (!pe)
1618 		return -ENODEV;
1619 
1620 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1621 		return -ENOENT;
1622 
1623 	switch (option) {
1624 	case EEH_RESET_DEACTIVATE:
1625 		ret = eeh_ops->reset(pe, option);
1626 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1627 		if (ret)
1628 			break;
1629 
1630 		ret = eeh_pe_reenable_devices(pe);
1631 		break;
1632 	case EEH_RESET_HOT:
1633 	case EEH_RESET_FUNDAMENTAL:
1634 		/*
1635 		 * Proactively freeze the PE to drop all MMIO access
1636 		 * during reset, which should be banned as it's always
1637 		 * cause recursive EEH error.
1638 		 */
1639 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1640 
1641 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1642 		ret = eeh_ops->reset(pe, option);
1643 		break;
1644 	default:
1645 		pr_debug("%s: Unsupported option %d\n",
1646 			__func__, option);
1647 		ret = -EINVAL;
1648 	}
1649 
1650 	return ret;
1651 }
1652 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1653 
1654 /**
1655  * eeh_pe_configure - Configure PCI bridges after PE reset
1656  * @pe: EEH PE
1657  *
1658  * The routine is called to restore the PCI config space for
1659  * those PCI devices, especially PCI bridges affected by PE
1660  * reset issued previously.
1661  */
1662 int eeh_pe_configure(struct eeh_pe *pe)
1663 {
1664 	int ret = 0;
1665 
1666 	/* Invalid PE ? */
1667 	if (!pe)
1668 		return -ENODEV;
1669 
1670 	return ret;
1671 }
1672 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1673 
1674 /**
1675  * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1676  * @pe: the indicated PE
1677  * @type: error type
1678  * @function: error function
1679  * @addr: address
1680  * @mask: address mask
1681  *
1682  * The routine is called to inject the specified PCI error, which
1683  * is determined by @type and @function, to the indicated PE for
1684  * testing purpose.
1685  */
1686 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1687 		      unsigned long addr, unsigned long mask)
1688 {
1689 	/* Invalid PE ? */
1690 	if (!pe)
1691 		return -ENODEV;
1692 
1693 	/* Unsupported operation ? */
1694 	if (!eeh_ops || !eeh_ops->err_inject)
1695 		return -ENOENT;
1696 
1697 	/* Check on PCI error type */
1698 	if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1699 		return -EINVAL;
1700 
1701 	/* Check on PCI error function */
1702 	if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1703 		return -EINVAL;
1704 
1705 	return eeh_ops->err_inject(pe, type, func, addr, mask);
1706 }
1707 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1708 
1709 static int proc_eeh_show(struct seq_file *m, void *v)
1710 {
1711 	if (!eeh_enabled()) {
1712 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1713 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1714 	} else {
1715 		seq_printf(m, "EEH Subsystem is enabled\n");
1716 		seq_printf(m,
1717 				"no device=%llu\n"
1718 				"no device node=%llu\n"
1719 				"no config address=%llu\n"
1720 				"check not wanted=%llu\n"
1721 				"eeh_total_mmio_ffs=%llu\n"
1722 				"eeh_false_positives=%llu\n"
1723 				"eeh_slot_resets=%llu\n",
1724 				eeh_stats.no_device,
1725 				eeh_stats.no_dn,
1726 				eeh_stats.no_cfg_addr,
1727 				eeh_stats.ignored_check,
1728 				eeh_stats.total_mmio_ffs,
1729 				eeh_stats.false_positives,
1730 				eeh_stats.slot_resets);
1731 	}
1732 
1733 	return 0;
1734 }
1735 
1736 static int proc_eeh_open(struct inode *inode, struct file *file)
1737 {
1738 	return single_open(file, proc_eeh_show, NULL);
1739 }
1740 
1741 static const struct file_operations proc_eeh_operations = {
1742 	.open      = proc_eeh_open,
1743 	.read      = seq_read,
1744 	.llseek    = seq_lseek,
1745 	.release   = single_release,
1746 };
1747 
1748 #ifdef CONFIG_DEBUG_FS
1749 static int eeh_enable_dbgfs_set(void *data, u64 val)
1750 {
1751 	if (val)
1752 		eeh_clear_flag(EEH_FORCE_DISABLED);
1753 	else
1754 		eeh_add_flag(EEH_FORCE_DISABLED);
1755 
1756 	/* Notify the backend */
1757 	if (eeh_ops->post_init)
1758 		eeh_ops->post_init();
1759 
1760 	return 0;
1761 }
1762 
1763 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1764 {
1765 	if (eeh_enabled())
1766 		*val = 0x1ul;
1767 	else
1768 		*val = 0x0ul;
1769 	return 0;
1770 }
1771 
1772 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1773 {
1774 	eeh_max_freezes = val;
1775 	return 0;
1776 }
1777 
1778 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1779 {
1780 	*val = eeh_max_freezes;
1781 	return 0;
1782 }
1783 
1784 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1785 			eeh_enable_dbgfs_set, "0x%llx\n");
1786 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1787 			eeh_freeze_dbgfs_set, "0x%llx\n");
1788 #endif
1789 
1790 static int __init eeh_init_proc(void)
1791 {
1792 	if (machine_is(pseries) || machine_is(powernv)) {
1793 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1794 #ifdef CONFIG_DEBUG_FS
1795 		debugfs_create_file("eeh_enable", 0600,
1796                                     powerpc_debugfs_root, NULL,
1797                                     &eeh_enable_dbgfs_ops);
1798 		debugfs_create_file("eeh_max_freezes", 0600,
1799 				    powerpc_debugfs_root, NULL,
1800 				    &eeh_freeze_dbgfs_ops);
1801 #endif
1802 	}
1803 
1804 	return 0;
1805 }
1806 __initcall(eeh_init_proc);
1807