xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision bbecb07f)
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 #include <asm/pte-walk.h>
48 
49 
50 /** Overview:
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.
64  *
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
72  *  with EEH.
73  *
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.
82  */
83 
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.
88  */
89 #define EEH_MAX_FAILS	2100000
90 
91 /* Time to wait for a PCI slot to report status, in milliseconds */
92 #define PCI_BUS_RESET_WAIT_MSEC (5*60*1000)
93 
94 /*
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.
103  */
104 int eeh_subsystem_flags;
105 EXPORT_SYMBOL(eeh_subsystem_flags);
106 
107 /*
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.
111  */
112 int eeh_max_freezes = 5;
113 
114 /* Platform dependent EEH operations */
115 struct eeh_ops *eeh_ops = NULL;
116 
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);
120 
121 /* Lock to protect passed flags */
122 static DEFINE_MUTEX(eeh_dev_mutex);
123 
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
126  * can access it.
127  */
128 #define EEH_PCI_REGS_LOG_LEN 8192
129 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
130 
131 /*
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
135  */
136 struct eeh_stats {
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			*/
144 };
145 
146 static struct eeh_stats eeh_stats;
147 
148 static int __init eeh_setup(char *str)
149 {
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);
154 
155 	return 1;
156 }
157 __setup("eeh=", eeh_setup);
158 
159 /*
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.
163  */
164 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
165 {
166 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
167 	u32 cfg;
168 	int cap, i;
169 	int n = 0, l = 0;
170 	char buffer[128];
171 
172 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x.%01x\n",
173 		       pdn->phb->global_number, pdn->busno,
174 		       PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
175 	pr_warn("EEH: of node=%04x:%02x:%02x.%01x\n",
176 		pdn->phb->global_number, pdn->busno,
177 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
178 
179 	eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
180 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
181 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
182 
183 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
184 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
185 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
186 
187 	/* Gather bridge-specific registers */
188 	if (edev->mode & EEH_DEV_BRIDGE) {
189 		eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
190 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
191 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
192 
193 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
194 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
195 		pr_warn("EEH: Bridge control: %04x\n", cfg);
196 	}
197 
198 	/* Dump out the PCI-X command and status regs */
199 	cap = edev->pcix_cap;
200 	if (cap) {
201 		eeh_ops->read_config(pdn, cap, 4, &cfg);
202 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
203 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
204 
205 		eeh_ops->read_config(pdn, cap+4, 4, &cfg);
206 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
207 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
208 	}
209 
210 	/* If PCI-E capable, dump PCI-E cap 10 */
211 	cap = edev->pcie_cap;
212 	if (cap) {
213 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
214 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
215 
216 		for (i=0; i<=8; i++) {
217 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
218 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
219 
220 			if ((i % 4) == 0) {
221 				if (i != 0)
222 					pr_warn("%s\n", buffer);
223 
224 				l = scnprintf(buffer, sizeof(buffer),
225 					      "EEH: PCI-E %02x: %08x ",
226 					      4*i, cfg);
227 			} else {
228 				l += scnprintf(buffer+l, sizeof(buffer)-l,
229 					       "%08x ", cfg);
230 			}
231 
232 		}
233 
234 		pr_warn("%s\n", buffer);
235 	}
236 
237 	/* If AER capable, dump it */
238 	cap = edev->aer_cap;
239 	if (cap) {
240 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
241 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
242 
243 		for (i=0; i<=13; i++) {
244 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
245 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
246 
247 			if ((i % 4) == 0) {
248 				if (i != 0)
249 					pr_warn("%s\n", buffer);
250 
251 				l = scnprintf(buffer, sizeof(buffer),
252 					      "EEH: PCI-E AER %02x: %08x ",
253 					      4*i, cfg);
254 			} else {
255 				l += scnprintf(buffer+l, sizeof(buffer)-l,
256 					       "%08x ", cfg);
257 			}
258 		}
259 
260 		pr_warn("%s\n", buffer);
261 	}
262 
263 	return n;
264 }
265 
266 static void *eeh_dump_pe_log(void *data, void *flag)
267 {
268 	struct eeh_pe *pe = data;
269 	struct eeh_dev *edev, *tmp;
270 	size_t *plen = flag;
271 
272 	eeh_pe_for_each_dev(pe, edev, tmp)
273 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
274 					  EEH_PCI_REGS_LOG_LEN - *plen);
275 
276 	return NULL;
277 }
278 
279 /**
280  * eeh_slot_error_detail - Generate combined log including driver log and error log
281  * @pe: EEH PE
282  * @severity: temporary or permanent error log
283  *
284  * This routine should be called to generate the combined log, which
285  * is comprised of driver log and error log. The driver log is figured
286  * out from the config space of the corresponding PCI device, while
287  * the error log is fetched through platform dependent function call.
288  */
289 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
290 {
291 	size_t loglen = 0;
292 
293 	/*
294 	 * When the PHB is fenced or dead, it's pointless to collect
295 	 * the data from PCI config space because it should return
296 	 * 0xFF's. For ER, we still retrieve the data from the PCI
297 	 * config space.
298 	 *
299 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
300 	 * 0xFF's is always returned from PCI config space.
301 	 *
302 	 * When the @severity is EEH_LOG_PERM, the PE is going to be
303 	 * removed. Prior to that, the drivers for devices included in
304 	 * the PE will be closed. The drivers rely on working IO path
305 	 * to bring the devices to quiet state. Otherwise, PCI traffic
306 	 * from those devices after they are removed is like to cause
307 	 * another unexpected EEH error.
308 	 */
309 	if (!(pe->type & EEH_PE_PHB)) {
310 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG) ||
311 		    severity == EEH_LOG_PERM)
312 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
313 
314 		/*
315 		 * The config space of some PCI devices can't be accessed
316 		 * when their PEs are in frozen state. Otherwise, fenced
317 		 * PHB might be seen. Those PEs are identified with flag
318 		 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
319 		 * is set automatically when the PE is put to EEH_PE_ISOLATED.
320 		 *
321 		 * Restoring BARs possibly triggers PCI config access in
322 		 * (OPAL) firmware and then causes fenced PHB. If the
323 		 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
324 		 * pointless to restore BARs and dump config space.
325 		 */
326 		eeh_ops->configure_bridge(pe);
327 		if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
328 			eeh_pe_restore_bars(pe);
329 
330 			pci_regs_buf[0] = 0;
331 			eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
332 		}
333 	}
334 
335 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
336 }
337 
338 /**
339  * eeh_token_to_phys - Convert EEH address token to phys address
340  * @token: I/O token, should be address in the form 0xA....
341  *
342  * This routine should be called to convert virtual I/O address
343  * to physical one.
344  */
345 static inline unsigned long eeh_token_to_phys(unsigned long token)
346 {
347 	pte_t *ptep;
348 	unsigned long pa;
349 	int hugepage_shift;
350 
351 	/*
352 	 * We won't find hugepages here(this is iomem). Hence we are not
353 	 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
354 	 * page table free, because of init_mm.
355 	 */
356 	ptep = find_init_mm_pte(token, &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 device_node *dn;
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 			dn = pci_device_to_OF_node(dev);
497 			if (dn)
498 				location = of_get_property(dn, "ibm,loc-code",
499 						NULL);
500 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
501 				"location=%s driver=%s pci addr=%s\n",
502 				pe->check_count,
503 				location ? location : "unknown",
504 				eeh_driver_name(dev), eeh_pci_name(dev));
505 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
506 				eeh_driver_name(dev));
507 			dump_stack();
508 		}
509 		goto dn_unlock;
510 	}
511 
512 	/*
513 	 * Now test for an EEH failure.  This is VERY expensive.
514 	 * Note that the eeh_config_addr may be a parent device
515 	 * in the case of a device behind a bridge, or it may be
516 	 * function zero of a multi-function device.
517 	 * In any case they must share a common PHB.
518 	 */
519 	ret = eeh_ops->get_state(pe, NULL);
520 
521 	/* Note that config-io to empty slots may fail;
522 	 * they are empty when they don't have children.
523 	 * We will punt with the following conditions: Failure to get
524 	 * PE's state, EEH not support and Permanently unavailable
525 	 * state, PE is in good state.
526 	 */
527 	if ((ret < 0) ||
528 	    (ret == EEH_STATE_NOT_SUPPORT) ||
529 	    ((ret & active_flags) == active_flags)) {
530 		eeh_stats.false_positives++;
531 		pe->false_positives++;
532 		rc = 0;
533 		goto dn_unlock;
534 	}
535 
536 	/*
537 	 * It should be corner case that the parent PE has been
538 	 * put into frozen state as well. We should take care
539 	 * that at first.
540 	 */
541 	parent_pe = pe->parent;
542 	while (parent_pe) {
543 		/* Hit the ceiling ? */
544 		if (parent_pe->type & EEH_PE_PHB)
545 			break;
546 
547 		/* Frozen parent PE ? */
548 		ret = eeh_ops->get_state(parent_pe, NULL);
549 		if (ret > 0 &&
550 		    (ret & active_flags) != active_flags)
551 			pe = parent_pe;
552 
553 		/* Next parent level */
554 		parent_pe = parent_pe->parent;
555 	}
556 
557 	eeh_stats.slot_resets++;
558 
559 	/* Avoid repeated reports of this failure, including problems
560 	 * with other functions on this device, and functions under
561 	 * bridges.
562 	 */
563 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
564 	eeh_serialize_unlock(flags);
565 
566 	/* Most EEH events are due to device driver bugs.  Having
567 	 * a stack trace will help the device-driver authors figure
568 	 * out what happened.  So print that out.
569 	 */
570 	phb_pe = eeh_phb_pe_get(pe->phb);
571 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
572 	       pe->phb->global_number, pe->addr);
573 	pr_err("EEH: PE location: %s, PHB location: %s\n",
574 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
575 	dump_stack();
576 
577 	eeh_send_failure_event(pe);
578 
579 	return 1;
580 
581 dn_unlock:
582 	eeh_serialize_unlock(flags);
583 	return rc;
584 }
585 
586 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
587 
588 /**
589  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
590  * @token: I/O address
591  *
592  * Check for an EEH failure at the given I/O address. Call this
593  * routine if the result of a read was all 0xff's and you want to
594  * find out if this is due to an EEH slot freeze event. This routine
595  * will query firmware for the EEH status.
596  *
597  * Note this routine is safe to call in an interrupt context.
598  */
599 int eeh_check_failure(const volatile void __iomem *token)
600 {
601 	unsigned long addr;
602 	struct eeh_dev *edev;
603 
604 	/* Finding the phys addr + pci device; this is pretty quick. */
605 	addr = eeh_token_to_phys((unsigned long __force) token);
606 	edev = eeh_addr_cache_get_dev(addr);
607 	if (!edev) {
608 		eeh_stats.no_device++;
609 		return 0;
610 	}
611 
612 	return eeh_dev_check_failure(edev);
613 }
614 EXPORT_SYMBOL(eeh_check_failure);
615 
616 
617 /**
618  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
619  * @pe: EEH PE
620  *
621  * This routine should be called to reenable frozen MMIO or DMA
622  * so that it would work correctly again. It's useful while doing
623  * recovery or log collection on the indicated device.
624  */
625 int eeh_pci_enable(struct eeh_pe *pe, int function)
626 {
627 	int active_flag, rc;
628 
629 	/*
630 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
631 	 * Also, it's pointless to enable them on unfrozen PE. So
632 	 * we have to check before enabling IO or DMA.
633 	 */
634 	switch (function) {
635 	case EEH_OPT_THAW_MMIO:
636 		active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
637 		break;
638 	case EEH_OPT_THAW_DMA:
639 		active_flag = EEH_STATE_DMA_ACTIVE;
640 		break;
641 	case EEH_OPT_DISABLE:
642 	case EEH_OPT_ENABLE:
643 	case EEH_OPT_FREEZE_PE:
644 		active_flag = 0;
645 		break;
646 	default:
647 		pr_warn("%s: Invalid function %d\n",
648 			__func__, function);
649 		return -EINVAL;
650 	}
651 
652 	/*
653 	 * Check if IO or DMA has been enabled before
654 	 * enabling them.
655 	 */
656 	if (active_flag) {
657 		rc = eeh_ops->get_state(pe, NULL);
658 		if (rc < 0)
659 			return rc;
660 
661 		/* Needn't enable it at all */
662 		if (rc == EEH_STATE_NOT_SUPPORT)
663 			return 0;
664 
665 		/* It's already enabled */
666 		if (rc & active_flag)
667 			return 0;
668 	}
669 
670 
671 	/* Issue the request */
672 	rc = eeh_ops->set_option(pe, function);
673 	if (rc)
674 		pr_warn("%s: Unexpected state change %d on "
675 			"PHB#%x-PE#%x, err=%d\n",
676 			__func__, function, pe->phb->global_number,
677 			pe->addr, rc);
678 
679 	/* Check if the request is finished successfully */
680 	if (active_flag) {
681 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
682 		if (rc < 0)
683 			return rc;
684 
685 		if (rc & active_flag)
686 			return 0;
687 
688 		return -EIO;
689 	}
690 
691 	return rc;
692 }
693 
694 static void *eeh_disable_and_save_dev_state(void *data, void *userdata)
695 {
696 	struct eeh_dev *edev = data;
697 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
698 	struct pci_dev *dev = userdata;
699 
700 	/*
701 	 * The caller should have disabled and saved the
702 	 * state for the specified device
703 	 */
704 	if (!pdev || pdev == dev)
705 		return NULL;
706 
707 	/* Ensure we have D0 power state */
708 	pci_set_power_state(pdev, PCI_D0);
709 
710 	/* Save device state */
711 	pci_save_state(pdev);
712 
713 	/*
714 	 * Disable device to avoid any DMA traffic and
715 	 * interrupt from the device
716 	 */
717 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
718 
719 	return NULL;
720 }
721 
722 static void *eeh_restore_dev_state(void *data, void *userdata)
723 {
724 	struct eeh_dev *edev = data;
725 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
726 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
727 	struct pci_dev *dev = userdata;
728 
729 	if (!pdev)
730 		return NULL;
731 
732 	/* Apply customization from firmware */
733 	if (pdn && eeh_ops->restore_config)
734 		eeh_ops->restore_config(pdn);
735 
736 	/* The caller should restore state for the specified device */
737 	if (pdev != dev)
738 		pci_restore_state(pdev);
739 
740 	return NULL;
741 }
742 
743 /**
744  * pcibios_set_pcie_reset_state - Set PCI-E reset state
745  * @dev: pci device struct
746  * @state: reset state to enter
747  *
748  * Return value:
749  * 	0 if success
750  */
751 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
752 {
753 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
754 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
755 
756 	if (!pe) {
757 		pr_err("%s: No PE found on PCI device %s\n",
758 			__func__, pci_name(dev));
759 		return -EINVAL;
760 	}
761 
762 	switch (state) {
763 	case pcie_deassert_reset:
764 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
765 		eeh_unfreeze_pe(pe, false);
766 		if (!(pe->type & EEH_PE_VF))
767 			eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
768 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
769 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
770 		break;
771 	case pcie_hot_reset:
772 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
773 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
774 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
775 		if (!(pe->type & EEH_PE_VF))
776 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
777 		eeh_ops->reset(pe, EEH_RESET_HOT);
778 		break;
779 	case pcie_warm_reset:
780 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
781 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
782 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
783 		if (!(pe->type & EEH_PE_VF))
784 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
785 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
786 		break;
787 	default:
788 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
789 		return -EINVAL;
790 	};
791 
792 	return 0;
793 }
794 
795 /**
796  * eeh_set_pe_freset - Check the required reset for the indicated device
797  * @data: EEH device
798  * @flag: return value
799  *
800  * Each device might have its preferred reset type: fundamental or
801  * hot reset. The routine is used to collected the information for
802  * the indicated device and its children so that the bunch of the
803  * devices could be reset properly.
804  */
805 static void *eeh_set_dev_freset(void *data, void *flag)
806 {
807 	struct pci_dev *dev;
808 	unsigned int *freset = (unsigned int *)flag;
809 	struct eeh_dev *edev = (struct eeh_dev *)data;
810 
811 	dev = eeh_dev_to_pci_dev(edev);
812 	if (dev)
813 		*freset |= dev->needs_freset;
814 
815 	return NULL;
816 }
817 
818 /**
819  * eeh_pe_reset_full - Complete a full reset process on the indicated PE
820  * @pe: EEH PE
821  *
822  * This function executes a full reset procedure on a PE, including setting
823  * the appropriate flags, performing a fundamental or hot reset, and then
824  * deactivating the reset status.  It is designed to be used within the EEH
825  * subsystem, as opposed to eeh_pe_reset which is exported to drivers and
826  * only performs a single operation at a time.
827  *
828  * This function will attempt to reset a PE three times before failing.
829  */
830 int eeh_pe_reset_full(struct eeh_pe *pe)
831 {
832 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
833 	int reset_state = (EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
834 	int type = EEH_RESET_HOT;
835 	unsigned int freset = 0;
836 	int i, state, ret;
837 
838 	/*
839 	 * Determine the type of reset to perform - hot or fundamental.
840 	 * Hot reset is the default operation, unless any device under the
841 	 * PE requires a fundamental reset.
842 	 */
843 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
844 
845 	if (freset)
846 		type = EEH_RESET_FUNDAMENTAL;
847 
848 	/* Mark the PE as in reset state and block config space accesses */
849 	eeh_pe_state_mark(pe, reset_state);
850 
851 	/* Make three attempts at resetting the bus */
852 	for (i = 0; i < 3; i++) {
853 		ret = eeh_pe_reset(pe, type);
854 		if (ret)
855 			break;
856 
857 		ret = eeh_pe_reset(pe, EEH_RESET_DEACTIVATE);
858 		if (ret)
859 			break;
860 
861 		/* Wait until the PE is in a functioning state */
862 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
863 		if ((state & active_flags) == active_flags)
864 			break;
865 
866 		if (state < 0) {
867 			pr_warn("%s: Unrecoverable slot failure on PHB#%x-PE#%x",
868 				__func__, pe->phb->global_number, pe->addr);
869 			ret = -ENOTRECOVERABLE;
870 			break;
871 		}
872 
873 		/* Set error in case this is our last attempt */
874 		ret = -EIO;
875 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
876 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
877 	}
878 
879 	eeh_pe_state_clear(pe, reset_state);
880 	return ret;
881 }
882 
883 /**
884  * eeh_save_bars - Save device bars
885  * @edev: PCI device associated EEH device
886  *
887  * Save the values of the device bars. Unlike the restore
888  * routine, this routine is *not* recursive. This is because
889  * PCI devices are added individually; but, for the restore,
890  * an entire slot is reset at a time.
891  */
892 void eeh_save_bars(struct eeh_dev *edev)
893 {
894 	struct pci_dn *pdn;
895 	int i;
896 
897 	pdn = eeh_dev_to_pdn(edev);
898 	if (!pdn)
899 		return;
900 
901 	for (i = 0; i < 16; i++)
902 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
903 
904 	/*
905 	 * For PCI bridges including root port, we need enable bus
906 	 * master explicitly. Otherwise, it can't fetch IODA table
907 	 * entries correctly. So we cache the bit in advance so that
908 	 * we can restore it after reset, either PHB range or PE range.
909 	 */
910 	if (edev->mode & EEH_DEV_BRIDGE)
911 		edev->config_space[1] |= PCI_COMMAND_MASTER;
912 }
913 
914 /**
915  * eeh_ops_register - Register platform dependent EEH operations
916  * @ops: platform dependent EEH operations
917  *
918  * Register the platform dependent EEH operation callback
919  * functions. The platform should call this function before
920  * any other EEH operations.
921  */
922 int __init eeh_ops_register(struct eeh_ops *ops)
923 {
924 	if (!ops->name) {
925 		pr_warn("%s: Invalid EEH ops name for %p\n",
926 			__func__, ops);
927 		return -EINVAL;
928 	}
929 
930 	if (eeh_ops && eeh_ops != ops) {
931 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
932 			__func__, eeh_ops->name, ops->name);
933 		return -EEXIST;
934 	}
935 
936 	eeh_ops = ops;
937 
938 	return 0;
939 }
940 
941 /**
942  * eeh_ops_unregister - Unreigster platform dependent EEH operations
943  * @name: name of EEH platform operations
944  *
945  * Unregister the platform dependent EEH operation callback
946  * functions.
947  */
948 int __exit eeh_ops_unregister(const char *name)
949 {
950 	if (!name || !strlen(name)) {
951 		pr_warn("%s: Invalid EEH ops name\n",
952 			__func__);
953 		return -EINVAL;
954 	}
955 
956 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
957 		eeh_ops = NULL;
958 		return 0;
959 	}
960 
961 	return -EEXIST;
962 }
963 
964 static int eeh_reboot_notifier(struct notifier_block *nb,
965 			       unsigned long action, void *unused)
966 {
967 	eeh_clear_flag(EEH_ENABLED);
968 	return NOTIFY_DONE;
969 }
970 
971 static struct notifier_block eeh_reboot_nb = {
972 	.notifier_call = eeh_reboot_notifier,
973 };
974 
975 void eeh_probe_devices(void)
976 {
977 	struct pci_controller *hose, *tmp;
978 	struct pci_dn *pdn;
979 
980 	/* Enable EEH for all adapters */
981 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
982 		pdn = hose->pci_data;
983 		traverse_pci_dn(pdn, eeh_ops->probe, NULL);
984 	}
985 }
986 
987 /**
988  * eeh_init - EEH initialization
989  *
990  * Initialize EEH by trying to enable it for all of the adapters in the system.
991  * As a side effect we can determine here if eeh is supported at all.
992  * Note that we leave EEH on so failed config cycles won't cause a machine
993  * check.  If a user turns off EEH for a particular adapter they are really
994  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
995  * grant access to a slot if EEH isn't enabled, and so we always enable
996  * EEH for all slots/all devices.
997  *
998  * The eeh-force-off option disables EEH checking globally, for all slots.
999  * Even if force-off is set, the EEH hardware is still enabled, so that
1000  * newer systems can boot.
1001  */
1002 static int eeh_init(void)
1003 {
1004 	struct pci_controller *hose, *tmp;
1005 	int ret = 0;
1006 
1007 	/* Register reboot notifier */
1008 	ret = register_reboot_notifier(&eeh_reboot_nb);
1009 	if (ret) {
1010 		pr_warn("%s: Failed to register notifier (%d)\n",
1011 			__func__, ret);
1012 		return ret;
1013 	}
1014 
1015 	/* call platform initialization function */
1016 	if (!eeh_ops) {
1017 		pr_warn("%s: Platform EEH operation not found\n",
1018 			__func__);
1019 		return -EEXIST;
1020 	} else if ((ret = eeh_ops->init()))
1021 		return ret;
1022 
1023 	/* Initialize PHB PEs */
1024 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
1025 		eeh_dev_phb_init_dynamic(hose);
1026 
1027 	/* Initialize EEH event */
1028 	ret = eeh_event_init();
1029 	if (ret)
1030 		return ret;
1031 
1032 	eeh_probe_devices();
1033 
1034 	if (eeh_enabled())
1035 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1036 	else
1037 		pr_info("EEH: No capable adapters found\n");
1038 
1039 	return ret;
1040 }
1041 
1042 core_initcall_sync(eeh_init);
1043 
1044 /**
1045  * eeh_add_device_early - Enable EEH for the indicated device node
1046  * @pdn: PCI device node for which to set up EEH
1047  *
1048  * This routine must be used to perform EEH initialization for PCI
1049  * devices that were added after system boot (e.g. hotplug, dlpar).
1050  * This routine must be called before any i/o is performed to the
1051  * adapter (inluding any config-space i/o).
1052  * Whether this actually enables EEH or not for this device depends
1053  * on the CEC architecture, type of the device, on earlier boot
1054  * command-line arguments & etc.
1055  */
1056 void eeh_add_device_early(struct pci_dn *pdn)
1057 {
1058 	struct pci_controller *phb = pdn ? pdn->phb : NULL;
1059 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1060 
1061 	if (!edev)
1062 		return;
1063 
1064 	if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1065 		return;
1066 
1067 	/* USB Bus children of PCI devices will not have BUID's */
1068 	if (NULL == phb ||
1069 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1070 		return;
1071 
1072 	eeh_ops->probe(pdn, NULL);
1073 }
1074 
1075 /**
1076  * eeh_add_device_tree_early - Enable EEH for the indicated device
1077  * @pdn: PCI device node
1078  *
1079  * This routine must be used to perform EEH initialization for the
1080  * indicated PCI device that was added after system boot (e.g.
1081  * hotplug, dlpar).
1082  */
1083 void eeh_add_device_tree_early(struct pci_dn *pdn)
1084 {
1085 	struct pci_dn *n;
1086 
1087 	if (!pdn)
1088 		return;
1089 
1090 	list_for_each_entry(n, &pdn->child_list, list)
1091 		eeh_add_device_tree_early(n);
1092 	eeh_add_device_early(pdn);
1093 }
1094 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1095 
1096 /**
1097  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1098  * @dev: pci device for which to set up EEH
1099  *
1100  * This routine must be used to complete EEH initialization for PCI
1101  * devices that were added after system boot (e.g. hotplug, dlpar).
1102  */
1103 void eeh_add_device_late(struct pci_dev *dev)
1104 {
1105 	struct pci_dn *pdn;
1106 	struct eeh_dev *edev;
1107 
1108 	if (!dev || !eeh_enabled())
1109 		return;
1110 
1111 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1112 
1113 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1114 	edev = pdn_to_eeh_dev(pdn);
1115 	if (edev->pdev == dev) {
1116 		pr_debug("EEH: Already referenced !\n");
1117 		return;
1118 	}
1119 
1120 	/*
1121 	 * The EEH cache might not be removed correctly because of
1122 	 * unbalanced kref to the device during unplug time, which
1123 	 * relies on pcibios_release_device(). So we have to remove
1124 	 * that here explicitly.
1125 	 */
1126 	if (edev->pdev) {
1127 		eeh_rmv_from_parent_pe(edev);
1128 		eeh_addr_cache_rmv_dev(edev->pdev);
1129 		eeh_sysfs_remove_device(edev->pdev);
1130 		edev->mode &= ~EEH_DEV_SYSFS;
1131 
1132 		/*
1133 		 * We definitely should have the PCI device removed
1134 		 * though it wasn't correctly. So we needn't call
1135 		 * into error handler afterwards.
1136 		 */
1137 		edev->mode |= EEH_DEV_NO_HANDLER;
1138 
1139 		edev->pdev = NULL;
1140 		dev->dev.archdata.edev = NULL;
1141 	}
1142 
1143 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1144 		eeh_ops->probe(pdn, NULL);
1145 
1146 	edev->pdev = dev;
1147 	dev->dev.archdata.edev = edev;
1148 
1149 	eeh_addr_cache_insert_dev(dev);
1150 }
1151 
1152 /**
1153  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1154  * @bus: PCI bus
1155  *
1156  * This routine must be used to perform EEH initialization for PCI
1157  * devices which are attached to the indicated PCI bus. The PCI bus
1158  * is added after system boot through hotplug or dlpar.
1159  */
1160 void eeh_add_device_tree_late(struct pci_bus *bus)
1161 {
1162 	struct pci_dev *dev;
1163 
1164 	list_for_each_entry(dev, &bus->devices, bus_list) {
1165 		eeh_add_device_late(dev);
1166 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1167 			struct pci_bus *subbus = dev->subordinate;
1168 			if (subbus)
1169 				eeh_add_device_tree_late(subbus);
1170 		}
1171 	}
1172 }
1173 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1174 
1175 /**
1176  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1177  * @bus: PCI bus
1178  *
1179  * This routine must be used to add EEH sysfs files for PCI
1180  * devices which are attached to the indicated PCI bus. The PCI bus
1181  * is added after system boot through hotplug or dlpar.
1182  */
1183 void eeh_add_sysfs_files(struct pci_bus *bus)
1184 {
1185 	struct pci_dev *dev;
1186 
1187 	list_for_each_entry(dev, &bus->devices, bus_list) {
1188 		eeh_sysfs_add_device(dev);
1189 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1190 			struct pci_bus *subbus = dev->subordinate;
1191 			if (subbus)
1192 				eeh_add_sysfs_files(subbus);
1193 		}
1194 	}
1195 }
1196 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1197 
1198 /**
1199  * eeh_remove_device - Undo EEH setup for the indicated pci device
1200  * @dev: pci device to be removed
1201  *
1202  * This routine should be called when a device is removed from
1203  * a running system (e.g. by hotplug or dlpar).  It unregisters
1204  * the PCI device from the EEH subsystem.  I/O errors affecting
1205  * this device will no longer be detected after this call; thus,
1206  * i/o errors affecting this slot may leave this device unusable.
1207  */
1208 void eeh_remove_device(struct pci_dev *dev)
1209 {
1210 	struct eeh_dev *edev;
1211 
1212 	if (!dev || !eeh_enabled())
1213 		return;
1214 	edev = pci_dev_to_eeh_dev(dev);
1215 
1216 	/* Unregister the device with the EEH/PCI address search system */
1217 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1218 
1219 	if (!edev || !edev->pdev || !edev->pe) {
1220 		pr_debug("EEH: Not referenced !\n");
1221 		return;
1222 	}
1223 
1224 	/*
1225 	 * During the hotplug for EEH error recovery, we need the EEH
1226 	 * device attached to the parent PE in order for BAR restore
1227 	 * a bit later. So we keep it for BAR restore and remove it
1228 	 * from the parent PE during the BAR resotre.
1229 	 */
1230 	edev->pdev = NULL;
1231 
1232 	/*
1233 	 * The flag "in_error" is used to trace EEH devices for VFs
1234 	 * in error state or not. It's set in eeh_report_error(). If
1235 	 * it's not set, eeh_report_{reset,resume}() won't be called
1236 	 * for the VF EEH device.
1237 	 */
1238 	edev->in_error = false;
1239 	dev->dev.archdata.edev = NULL;
1240 	if (!(edev->pe->state & EEH_PE_KEEP))
1241 		eeh_rmv_from_parent_pe(edev);
1242 	else
1243 		edev->mode |= EEH_DEV_DISCONNECTED;
1244 
1245 	/*
1246 	 * We're removing from the PCI subsystem, that means
1247 	 * the PCI device driver can't support EEH or not
1248 	 * well. So we rely on hotplug completely to do recovery
1249 	 * for the specific PCI device.
1250 	 */
1251 	edev->mode |= EEH_DEV_NO_HANDLER;
1252 
1253 	eeh_addr_cache_rmv_dev(dev);
1254 	eeh_sysfs_remove_device(dev);
1255 	edev->mode &= ~EEH_DEV_SYSFS;
1256 }
1257 
1258 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1259 {
1260 	int ret;
1261 
1262 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1263 	if (ret) {
1264 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1265 			__func__, ret, pe->phb->global_number, pe->addr);
1266 		return ret;
1267 	}
1268 
1269 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1270 	if (ret) {
1271 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1272 			__func__, ret, pe->phb->global_number, pe->addr);
1273 		return ret;
1274 	}
1275 
1276 	/* Clear software isolated state */
1277 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1278 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1279 
1280 	return ret;
1281 }
1282 
1283 
1284 static struct pci_device_id eeh_reset_ids[] = {
1285 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1286 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1287 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1288 	{ 0 }
1289 };
1290 
1291 static int eeh_pe_change_owner(struct eeh_pe *pe)
1292 {
1293 	struct eeh_dev *edev, *tmp;
1294 	struct pci_dev *pdev;
1295 	struct pci_device_id *id;
1296 	int flags, ret;
1297 
1298 	/* Check PE state */
1299 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1300 	ret = eeh_ops->get_state(pe, NULL);
1301 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1302 		return 0;
1303 
1304 	/* Unfrozen PE, nothing to do */
1305 	if ((ret & flags) == flags)
1306 		return 0;
1307 
1308 	/* Frozen PE, check if it needs PE level reset */
1309 	eeh_pe_for_each_dev(pe, edev, tmp) {
1310 		pdev = eeh_dev_to_pci_dev(edev);
1311 		if (!pdev)
1312 			continue;
1313 
1314 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1315 			if (id->vendor != PCI_ANY_ID &&
1316 			    id->vendor != pdev->vendor)
1317 				continue;
1318 			if (id->device != PCI_ANY_ID &&
1319 			    id->device != pdev->device)
1320 				continue;
1321 			if (id->subvendor != PCI_ANY_ID &&
1322 			    id->subvendor != pdev->subsystem_vendor)
1323 				continue;
1324 			if (id->subdevice != PCI_ANY_ID &&
1325 			    id->subdevice != pdev->subsystem_device)
1326 				continue;
1327 
1328 			return eeh_pe_reset_and_recover(pe);
1329 		}
1330 	}
1331 
1332 	return eeh_unfreeze_pe(pe, true);
1333 }
1334 
1335 /**
1336  * eeh_dev_open - Increase count of pass through devices for PE
1337  * @pdev: PCI device
1338  *
1339  * Increase count of passed through devices for the indicated
1340  * PE. In the result, the EEH errors detected on the PE won't be
1341  * reported. The PE owner will be responsible for detection
1342  * and recovery.
1343  */
1344 int eeh_dev_open(struct pci_dev *pdev)
1345 {
1346 	struct eeh_dev *edev;
1347 	int ret = -ENODEV;
1348 
1349 	mutex_lock(&eeh_dev_mutex);
1350 
1351 	/* No PCI device ? */
1352 	if (!pdev)
1353 		goto out;
1354 
1355 	/* No EEH device or PE ? */
1356 	edev = pci_dev_to_eeh_dev(pdev);
1357 	if (!edev || !edev->pe)
1358 		goto out;
1359 
1360 	/*
1361 	 * The PE might have been put into frozen state, but we
1362 	 * didn't detect that yet. The passed through PCI devices
1363 	 * in frozen PE won't work properly. Clear the frozen state
1364 	 * in advance.
1365 	 */
1366 	ret = eeh_pe_change_owner(edev->pe);
1367 	if (ret)
1368 		goto out;
1369 
1370 	/* Increase PE's pass through count */
1371 	atomic_inc(&edev->pe->pass_dev_cnt);
1372 	mutex_unlock(&eeh_dev_mutex);
1373 
1374 	return 0;
1375 out:
1376 	mutex_unlock(&eeh_dev_mutex);
1377 	return ret;
1378 }
1379 EXPORT_SYMBOL_GPL(eeh_dev_open);
1380 
1381 /**
1382  * eeh_dev_release - Decrease count of pass through devices for PE
1383  * @pdev: PCI device
1384  *
1385  * Decrease count of pass through devices for the indicated PE. If
1386  * there is no passed through device in PE, the EEH errors detected
1387  * on the PE will be reported and handled as usual.
1388  */
1389 void eeh_dev_release(struct pci_dev *pdev)
1390 {
1391 	struct eeh_dev *edev;
1392 
1393 	mutex_lock(&eeh_dev_mutex);
1394 
1395 	/* No PCI device ? */
1396 	if (!pdev)
1397 		goto out;
1398 
1399 	/* No EEH device ? */
1400 	edev = pci_dev_to_eeh_dev(pdev);
1401 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1402 		goto out;
1403 
1404 	/* Decrease PE's pass through count */
1405 	WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1406 	eeh_pe_change_owner(edev->pe);
1407 out:
1408 	mutex_unlock(&eeh_dev_mutex);
1409 }
1410 EXPORT_SYMBOL(eeh_dev_release);
1411 
1412 #ifdef CONFIG_IOMMU_API
1413 
1414 static int dev_has_iommu_table(struct device *dev, void *data)
1415 {
1416 	struct pci_dev *pdev = to_pci_dev(dev);
1417 	struct pci_dev **ppdev = data;
1418 
1419 	if (!dev)
1420 		return 0;
1421 
1422 	if (dev->iommu_group) {
1423 		*ppdev = pdev;
1424 		return 1;
1425 	}
1426 
1427 	return 0;
1428 }
1429 
1430 /**
1431  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1432  * @group: IOMMU group
1433  *
1434  * The routine is called to convert IOMMU group to EEH PE.
1435  */
1436 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1437 {
1438 	struct pci_dev *pdev = NULL;
1439 	struct eeh_dev *edev;
1440 	int ret;
1441 
1442 	/* No IOMMU group ? */
1443 	if (!group)
1444 		return NULL;
1445 
1446 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1447 	if (!ret || !pdev)
1448 		return NULL;
1449 
1450 	/* No EEH device or PE ? */
1451 	edev = pci_dev_to_eeh_dev(pdev);
1452 	if (!edev || !edev->pe)
1453 		return NULL;
1454 
1455 	return edev->pe;
1456 }
1457 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1458 
1459 #endif /* CONFIG_IOMMU_API */
1460 
1461 /**
1462  * eeh_pe_set_option - Set options for the indicated PE
1463  * @pe: EEH PE
1464  * @option: requested option
1465  *
1466  * The routine is called to enable or disable EEH functionality
1467  * on the indicated PE, to enable IO or DMA for the frozen PE.
1468  */
1469 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1470 {
1471 	int ret = 0;
1472 
1473 	/* Invalid PE ? */
1474 	if (!pe)
1475 		return -ENODEV;
1476 
1477 	/*
1478 	 * EEH functionality could possibly be disabled, just
1479 	 * return error for the case. And the EEH functinality
1480 	 * isn't expected to be disabled on one specific PE.
1481 	 */
1482 	switch (option) {
1483 	case EEH_OPT_ENABLE:
1484 		if (eeh_enabled()) {
1485 			ret = eeh_pe_change_owner(pe);
1486 			break;
1487 		}
1488 		ret = -EIO;
1489 		break;
1490 	case EEH_OPT_DISABLE:
1491 		break;
1492 	case EEH_OPT_THAW_MMIO:
1493 	case EEH_OPT_THAW_DMA:
1494 	case EEH_OPT_FREEZE_PE:
1495 		if (!eeh_ops || !eeh_ops->set_option) {
1496 			ret = -ENOENT;
1497 			break;
1498 		}
1499 
1500 		ret = eeh_pci_enable(pe, option);
1501 		break;
1502 	default:
1503 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1504 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1505 		ret = -EINVAL;
1506 	}
1507 
1508 	return ret;
1509 }
1510 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1511 
1512 /**
1513  * eeh_pe_get_state - Retrieve PE's state
1514  * @pe: EEH PE
1515  *
1516  * Retrieve the PE's state, which includes 3 aspects: enabled
1517  * DMA, enabled IO and asserted reset.
1518  */
1519 int eeh_pe_get_state(struct eeh_pe *pe)
1520 {
1521 	int result, ret = 0;
1522 	bool rst_active, dma_en, mmio_en;
1523 
1524 	/* Existing PE ? */
1525 	if (!pe)
1526 		return -ENODEV;
1527 
1528 	if (!eeh_ops || !eeh_ops->get_state)
1529 		return -ENOENT;
1530 
1531 	/*
1532 	 * If the parent PE is owned by the host kernel and is undergoing
1533 	 * error recovery, we should return the PE state as temporarily
1534 	 * unavailable so that the error recovery on the guest is suspended
1535 	 * until the recovery completes on the host.
1536 	 */
1537 	if (pe->parent &&
1538 	    !(pe->state & EEH_PE_REMOVED) &&
1539 	    (pe->parent->state & (EEH_PE_ISOLATED | EEH_PE_RECOVERING)))
1540 		return EEH_PE_STATE_UNAVAIL;
1541 
1542 	result = eeh_ops->get_state(pe, NULL);
1543 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1544 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1545 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1546 
1547 	if (rst_active)
1548 		ret = EEH_PE_STATE_RESET;
1549 	else if (dma_en && mmio_en)
1550 		ret = EEH_PE_STATE_NORMAL;
1551 	else if (!dma_en && !mmio_en)
1552 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1553 	else if (!dma_en && mmio_en)
1554 		ret = EEH_PE_STATE_STOPPED_DMA;
1555 	else
1556 		ret = EEH_PE_STATE_UNAVAIL;
1557 
1558 	return ret;
1559 }
1560 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1561 
1562 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1563 {
1564 	struct eeh_dev *edev, *tmp;
1565 	struct pci_dev *pdev;
1566 	int ret = 0;
1567 
1568 	/* Restore config space */
1569 	eeh_pe_restore_bars(pe);
1570 
1571 	/*
1572 	 * Reenable PCI devices as the devices passed
1573 	 * through are always enabled before the reset.
1574 	 */
1575 	eeh_pe_for_each_dev(pe, edev, tmp) {
1576 		pdev = eeh_dev_to_pci_dev(edev);
1577 		if (!pdev)
1578 			continue;
1579 
1580 		ret = pci_reenable_device(pdev);
1581 		if (ret) {
1582 			pr_warn("%s: Failure %d reenabling %s\n",
1583 				__func__, ret, pci_name(pdev));
1584 			return ret;
1585 		}
1586 	}
1587 
1588 	/* The PE is still in frozen state */
1589 	return eeh_unfreeze_pe(pe, true);
1590 }
1591 
1592 
1593 /**
1594  * eeh_pe_reset - Issue PE reset according to specified type
1595  * @pe: EEH PE
1596  * @option: reset type
1597  *
1598  * The routine is called to reset the specified PE with the
1599  * indicated type, either fundamental reset or hot reset.
1600  * PE reset is the most important part for error recovery.
1601  */
1602 int eeh_pe_reset(struct eeh_pe *pe, int option)
1603 {
1604 	int ret = 0;
1605 
1606 	/* Invalid PE ? */
1607 	if (!pe)
1608 		return -ENODEV;
1609 
1610 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1611 		return -ENOENT;
1612 
1613 	switch (option) {
1614 	case EEH_RESET_DEACTIVATE:
1615 		ret = eeh_ops->reset(pe, option);
1616 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1617 		if (ret)
1618 			break;
1619 
1620 		ret = eeh_pe_reenable_devices(pe);
1621 		break;
1622 	case EEH_RESET_HOT:
1623 	case EEH_RESET_FUNDAMENTAL:
1624 		/*
1625 		 * Proactively freeze the PE to drop all MMIO access
1626 		 * during reset, which should be banned as it's always
1627 		 * cause recursive EEH error.
1628 		 */
1629 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1630 
1631 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1632 		ret = eeh_ops->reset(pe, option);
1633 		break;
1634 	default:
1635 		pr_debug("%s: Unsupported option %d\n",
1636 			__func__, option);
1637 		ret = -EINVAL;
1638 	}
1639 
1640 	return ret;
1641 }
1642 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1643 
1644 /**
1645  * eeh_pe_configure - Configure PCI bridges after PE reset
1646  * @pe: EEH PE
1647  *
1648  * The routine is called to restore the PCI config space for
1649  * those PCI devices, especially PCI bridges affected by PE
1650  * reset issued previously.
1651  */
1652 int eeh_pe_configure(struct eeh_pe *pe)
1653 {
1654 	int ret = 0;
1655 
1656 	/* Invalid PE ? */
1657 	if (!pe)
1658 		return -ENODEV;
1659 
1660 	return ret;
1661 }
1662 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1663 
1664 /**
1665  * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1666  * @pe: the indicated PE
1667  * @type: error type
1668  * @function: error function
1669  * @addr: address
1670  * @mask: address mask
1671  *
1672  * The routine is called to inject the specified PCI error, which
1673  * is determined by @type and @function, to the indicated PE for
1674  * testing purpose.
1675  */
1676 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1677 		      unsigned long addr, unsigned long mask)
1678 {
1679 	/* Invalid PE ? */
1680 	if (!pe)
1681 		return -ENODEV;
1682 
1683 	/* Unsupported operation ? */
1684 	if (!eeh_ops || !eeh_ops->err_inject)
1685 		return -ENOENT;
1686 
1687 	/* Check on PCI error type */
1688 	if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1689 		return -EINVAL;
1690 
1691 	/* Check on PCI error function */
1692 	if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1693 		return -EINVAL;
1694 
1695 	return eeh_ops->err_inject(pe, type, func, addr, mask);
1696 }
1697 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1698 
1699 static int proc_eeh_show(struct seq_file *m, void *v)
1700 {
1701 	if (!eeh_enabled()) {
1702 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1703 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1704 	} else {
1705 		seq_printf(m, "EEH Subsystem is enabled\n");
1706 		seq_printf(m,
1707 				"no device=%llu\n"
1708 				"no device node=%llu\n"
1709 				"no config address=%llu\n"
1710 				"check not wanted=%llu\n"
1711 				"eeh_total_mmio_ffs=%llu\n"
1712 				"eeh_false_positives=%llu\n"
1713 				"eeh_slot_resets=%llu\n",
1714 				eeh_stats.no_device,
1715 				eeh_stats.no_dn,
1716 				eeh_stats.no_cfg_addr,
1717 				eeh_stats.ignored_check,
1718 				eeh_stats.total_mmio_ffs,
1719 				eeh_stats.false_positives,
1720 				eeh_stats.slot_resets);
1721 	}
1722 
1723 	return 0;
1724 }
1725 
1726 static int proc_eeh_open(struct inode *inode, struct file *file)
1727 {
1728 	return single_open(file, proc_eeh_show, NULL);
1729 }
1730 
1731 static const struct file_operations proc_eeh_operations = {
1732 	.open      = proc_eeh_open,
1733 	.read      = seq_read,
1734 	.llseek    = seq_lseek,
1735 	.release   = single_release,
1736 };
1737 
1738 #ifdef CONFIG_DEBUG_FS
1739 static int eeh_enable_dbgfs_set(void *data, u64 val)
1740 {
1741 	if (val)
1742 		eeh_clear_flag(EEH_FORCE_DISABLED);
1743 	else
1744 		eeh_add_flag(EEH_FORCE_DISABLED);
1745 
1746 	return 0;
1747 }
1748 
1749 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1750 {
1751 	if (eeh_enabled())
1752 		*val = 0x1ul;
1753 	else
1754 		*val = 0x0ul;
1755 	return 0;
1756 }
1757 
1758 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1759 {
1760 	eeh_max_freezes = val;
1761 	return 0;
1762 }
1763 
1764 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1765 {
1766 	*val = eeh_max_freezes;
1767 	return 0;
1768 }
1769 
1770 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1771 			eeh_enable_dbgfs_set, "0x%llx\n");
1772 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1773 			eeh_freeze_dbgfs_set, "0x%llx\n");
1774 #endif
1775 
1776 static int __init eeh_init_proc(void)
1777 {
1778 	if (machine_is(pseries) || machine_is(powernv)) {
1779 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1780 #ifdef CONFIG_DEBUG_FS
1781 		debugfs_create_file("eeh_enable", 0600,
1782                                     powerpc_debugfs_root, NULL,
1783                                     &eeh_enable_dbgfs_ops);
1784 		debugfs_create_file("eeh_max_freezes", 0600,
1785 				    powerpc_debugfs_root, NULL,
1786 				    &eeh_freeze_dbgfs_ops);
1787 #endif
1788 	}
1789 
1790 	return 0;
1791 }
1792 __initcall(eeh_init_proc);
1793