xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision 5c73cc4b6c83e88863a5de869cc5df3b913aef4a)
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/debugfs.h>
26 #include <linux/sched.h>
27 #include <linux/init.h>
28 #include <linux/list.h>
29 #include <linux/pci.h>
30 #include <linux/iommu.h>
31 #include <linux/proc_fs.h>
32 #include <linux/rbtree.h>
33 #include <linux/reboot.h>
34 #include <linux/seq_file.h>
35 #include <linux/spinlock.h>
36 #include <linux/export.h>
37 #include <linux/of.h>
38 
39 #include <linux/atomic.h>
40 #include <asm/debug.h>
41 #include <asm/eeh.h>
42 #include <asm/eeh_event.h>
43 #include <asm/io.h>
44 #include <asm/iommu.h>
45 #include <asm/machdep.h>
46 #include <asm/ppc-pci.h>
47 #include <asm/rtas.h>
48 
49 
50 /** Overview:
51  *  EEH, or "Extended 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 
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 #define IS_BRIDGE(class_code) (((class_code)<<16) == PCI_BASE_CLASS_BRIDGE)
148 
149 static int __init eeh_setup(char *str)
150 {
151 	if (!strcmp(str, "off"))
152 		eeh_add_flag(EEH_FORCE_DISABLED);
153 	else if (!strcmp(str, "early_log"))
154 		eeh_add_flag(EEH_EARLY_DUMP_LOG);
155 
156 	return 1;
157 }
158 __setup("eeh=", eeh_setup);
159 
160 /*
161  * This routine captures assorted PCI configuration space data
162  * for the indicated PCI device, and puts them into a buffer
163  * for RTAS error logging.
164  */
165 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
166 {
167 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
168 	u32 cfg;
169 	int cap, i;
170 	int n = 0, l = 0;
171 	char buffer[128];
172 
173 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x:%01x\n",
174 		       edev->phb->global_number, pdn->busno,
175 		       PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
176 	pr_warn("EEH: of node=%04x:%02x:%02x:%01x\n",
177 		edev->phb->global_number, pdn->busno,
178 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
179 
180 	eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
181 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
182 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
183 
184 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
185 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
186 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
187 
188 	/* Gather bridge-specific registers */
189 	if (edev->mode & EEH_DEV_BRIDGE) {
190 		eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
191 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
192 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
193 
194 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
195 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
196 		pr_warn("EEH: Bridge control: %04x\n", cfg);
197 	}
198 
199 	/* Dump out the PCI-X command and status regs */
200 	cap = edev->pcix_cap;
201 	if (cap) {
202 		eeh_ops->read_config(pdn, cap, 4, &cfg);
203 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
204 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
205 
206 		eeh_ops->read_config(pdn, cap+4, 4, &cfg);
207 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
208 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
209 	}
210 
211 	/* If PCI-E capable, dump PCI-E cap 10 */
212 	cap = edev->pcie_cap;
213 	if (cap) {
214 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
215 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
216 
217 		for (i=0; i<=8; i++) {
218 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
219 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
220 
221 			if ((i % 4) == 0) {
222 				if (i != 0)
223 					pr_warn("%s\n", buffer);
224 
225 				l = scnprintf(buffer, sizeof(buffer),
226 					      "EEH: PCI-E %02x: %08x ",
227 					      4*i, cfg);
228 			} else {
229 				l += scnprintf(buffer+l, sizeof(buffer)-l,
230 					       "%08x ", cfg);
231 			}
232 
233 		}
234 
235 		pr_warn("%s\n", buffer);
236 	}
237 
238 	/* If AER capable, dump it */
239 	cap = edev->aer_cap;
240 	if (cap) {
241 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
242 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
243 
244 		for (i=0; i<=13; i++) {
245 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
246 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
247 
248 			if ((i % 4) == 0) {
249 				if (i != 0)
250 					pr_warn("%s\n", buffer);
251 
252 				l = scnprintf(buffer, sizeof(buffer),
253 					      "EEH: PCI-E AER %02x: %08x ",
254 					      4*i, cfg);
255 			} else {
256 				l += scnprintf(buffer+l, sizeof(buffer)-l,
257 					       "%08x ", cfg);
258 			}
259 		}
260 
261 		pr_warn("%s\n", buffer);
262 	}
263 
264 	return n;
265 }
266 
267 static void *eeh_dump_pe_log(void *data, void *flag)
268 {
269 	struct eeh_pe *pe = data;
270 	struct eeh_dev *edev, *tmp;
271 	size_t *plen = flag;
272 
273 	/* If the PE's config space is blocked, 0xFF's will be
274 	 * returned. It's pointless to collect the log in this
275 	 * case.
276 	 */
277 	if (pe->state & EEH_PE_CFG_BLOCKED)
278 		return NULL;
279 
280 	eeh_pe_for_each_dev(pe, edev, tmp)
281 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
282 					  EEH_PCI_REGS_LOG_LEN - *plen);
283 
284 	return NULL;
285 }
286 
287 /**
288  * eeh_slot_error_detail - Generate combined log including driver log and error log
289  * @pe: EEH PE
290  * @severity: temporary or permanent error log
291  *
292  * This routine should be called to generate the combined log, which
293  * is comprised of driver log and error log. The driver log is figured
294  * out from the config space of the corresponding PCI device, while
295  * the error log is fetched through platform dependent function call.
296  */
297 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
298 {
299 	size_t loglen = 0;
300 
301 	/*
302 	 * When the PHB is fenced or dead, it's pointless to collect
303 	 * the data from PCI config space because it should return
304 	 * 0xFF's. For ER, we still retrieve the data from the PCI
305 	 * config space.
306 	 *
307 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
308 	 * 0xFF's is always returned from PCI config space.
309 	 */
310 	if (!(pe->type & EEH_PE_PHB)) {
311 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG))
312 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
313 		eeh_ops->configure_bridge(pe);
314 		eeh_pe_restore_bars(pe);
315 
316 		pci_regs_buf[0] = 0;
317 		eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
318 	}
319 
320 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
321 }
322 
323 /**
324  * eeh_token_to_phys - Convert EEH address token to phys address
325  * @token: I/O token, should be address in the form 0xA....
326  *
327  * This routine should be called to convert virtual I/O address
328  * to physical one.
329  */
330 static inline unsigned long eeh_token_to_phys(unsigned long token)
331 {
332 	pte_t *ptep;
333 	unsigned long pa;
334 	int hugepage_shift;
335 
336 	/*
337 	 * We won't find hugepages here, iomem
338 	 */
339 	ptep = find_linux_pte_or_hugepte(init_mm.pgd, token, &hugepage_shift);
340 	if (!ptep)
341 		return token;
342 	WARN_ON(hugepage_shift);
343 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
344 
345 	return pa | (token & (PAGE_SIZE-1));
346 }
347 
348 /*
349  * On PowerNV platform, we might already have fenced PHB there.
350  * For that case, it's meaningless to recover frozen PE. Intead,
351  * We have to handle fenced PHB firstly.
352  */
353 static int eeh_phb_check_failure(struct eeh_pe *pe)
354 {
355 	struct eeh_pe *phb_pe;
356 	unsigned long flags;
357 	int ret;
358 
359 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
360 		return -EPERM;
361 
362 	/* Find the PHB PE */
363 	phb_pe = eeh_phb_pe_get(pe->phb);
364 	if (!phb_pe) {
365 		pr_warn("%s Can't find PE for PHB#%d\n",
366 			__func__, pe->phb->global_number);
367 		return -EEXIST;
368 	}
369 
370 	/* If the PHB has been in problematic state */
371 	eeh_serialize_lock(&flags);
372 	if (phb_pe->state & EEH_PE_ISOLATED) {
373 		ret = 0;
374 		goto out;
375 	}
376 
377 	/* Check PHB state */
378 	ret = eeh_ops->get_state(phb_pe, NULL);
379 	if ((ret < 0) ||
380 	    (ret == EEH_STATE_NOT_SUPPORT) ||
381 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
382 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
383 		ret = 0;
384 		goto out;
385 	}
386 
387 	/* Isolate the PHB and send event */
388 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
389 	eeh_serialize_unlock(flags);
390 
391 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
392 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
393 	dump_stack();
394 	eeh_send_failure_event(phb_pe);
395 
396 	return 1;
397 out:
398 	eeh_serialize_unlock(flags);
399 	return ret;
400 }
401 
402 /**
403  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
404  * @edev: eeh device
405  *
406  * Check for an EEH failure for the given device node.  Call this
407  * routine if the result of a read was all 0xff's and you want to
408  * find out if this is due to an EEH slot freeze.  This routine
409  * will query firmware for the EEH status.
410  *
411  * Returns 0 if there has not been an EEH error; otherwise returns
412  * a non-zero value and queues up a slot isolation event notification.
413  *
414  * It is safe to call this routine in an interrupt context.
415  */
416 int eeh_dev_check_failure(struct eeh_dev *edev)
417 {
418 	int ret;
419 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
420 	unsigned long flags;
421 	struct pci_dn *pdn;
422 	struct pci_dev *dev;
423 	struct eeh_pe *pe, *parent_pe, *phb_pe;
424 	int rc = 0;
425 	const char *location = NULL;
426 
427 	eeh_stats.total_mmio_ffs++;
428 
429 	if (!eeh_enabled())
430 		return 0;
431 
432 	if (!edev) {
433 		eeh_stats.no_dn++;
434 		return 0;
435 	}
436 	dev = eeh_dev_to_pci_dev(edev);
437 	pe = eeh_dev_to_pe(edev);
438 
439 	/* Access to IO BARs might get this far and still not want checking. */
440 	if (!pe) {
441 		eeh_stats.ignored_check++;
442 		pr_debug("EEH: Ignored check for %s\n",
443 			eeh_pci_name(dev));
444 		return 0;
445 	}
446 
447 	if (!pe->addr && !pe->config_addr) {
448 		eeh_stats.no_cfg_addr++;
449 		return 0;
450 	}
451 
452 	/*
453 	 * On PowerNV platform, we might already have fenced PHB
454 	 * there and we need take care of that firstly.
455 	 */
456 	ret = eeh_phb_check_failure(pe);
457 	if (ret > 0)
458 		return ret;
459 
460 	/*
461 	 * If the PE isn't owned by us, we shouldn't check the
462 	 * state. Instead, let the owner handle it if the PE has
463 	 * been frozen.
464 	 */
465 	if (eeh_pe_passed(pe))
466 		return 0;
467 
468 	/* If we already have a pending isolation event for this
469 	 * slot, we know it's bad already, we don't need to check.
470 	 * Do this checking under a lock; as multiple PCI devices
471 	 * in one slot might report errors simultaneously, and we
472 	 * only want one error recovery routine running.
473 	 */
474 	eeh_serialize_lock(&flags);
475 	rc = 1;
476 	if (pe->state & EEH_PE_ISOLATED) {
477 		pe->check_count++;
478 		if (pe->check_count % EEH_MAX_FAILS == 0) {
479 			pdn = eeh_dev_to_pdn(edev);
480 			if (pdn->node)
481 				location = of_get_property(pdn->node, "ibm,loc-code", NULL);
482 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
483 				"location=%s driver=%s pci addr=%s\n",
484 				pe->check_count,
485 				location ? location : "unknown",
486 				eeh_driver_name(dev), eeh_pci_name(dev));
487 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
488 				eeh_driver_name(dev));
489 			dump_stack();
490 		}
491 		goto dn_unlock;
492 	}
493 
494 	/*
495 	 * Now test for an EEH failure.  This is VERY expensive.
496 	 * Note that the eeh_config_addr may be a parent device
497 	 * in the case of a device behind a bridge, or it may be
498 	 * function zero of a multi-function device.
499 	 * In any case they must share a common PHB.
500 	 */
501 	ret = eeh_ops->get_state(pe, NULL);
502 
503 	/* Note that config-io to empty slots may fail;
504 	 * they are empty when they don't have children.
505 	 * We will punt with the following conditions: Failure to get
506 	 * PE's state, EEH not support and Permanently unavailable
507 	 * state, PE is in good state.
508 	 */
509 	if ((ret < 0) ||
510 	    (ret == EEH_STATE_NOT_SUPPORT) ||
511 	    ((ret & active_flags) == active_flags)) {
512 		eeh_stats.false_positives++;
513 		pe->false_positives++;
514 		rc = 0;
515 		goto dn_unlock;
516 	}
517 
518 	/*
519 	 * It should be corner case that the parent PE has been
520 	 * put into frozen state as well. We should take care
521 	 * that at first.
522 	 */
523 	parent_pe = pe->parent;
524 	while (parent_pe) {
525 		/* Hit the ceiling ? */
526 		if (parent_pe->type & EEH_PE_PHB)
527 			break;
528 
529 		/* Frozen parent PE ? */
530 		ret = eeh_ops->get_state(parent_pe, NULL);
531 		if (ret > 0 &&
532 		    (ret & active_flags) != active_flags)
533 			pe = parent_pe;
534 
535 		/* Next parent level */
536 		parent_pe = parent_pe->parent;
537 	}
538 
539 	eeh_stats.slot_resets++;
540 
541 	/* Avoid repeated reports of this failure, including problems
542 	 * with other functions on this device, and functions under
543 	 * bridges.
544 	 */
545 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
546 	eeh_serialize_unlock(flags);
547 
548 	/* Most EEH events are due to device driver bugs.  Having
549 	 * a stack trace will help the device-driver authors figure
550 	 * out what happened.  So print that out.
551 	 */
552 	phb_pe = eeh_phb_pe_get(pe->phb);
553 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
554 	       pe->phb->global_number, pe->addr);
555 	pr_err("EEH: PE location: %s, PHB location: %s\n",
556 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
557 	dump_stack();
558 
559 	eeh_send_failure_event(pe);
560 
561 	return 1;
562 
563 dn_unlock:
564 	eeh_serialize_unlock(flags);
565 	return rc;
566 }
567 
568 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
569 
570 /**
571  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
572  * @token: I/O address
573  *
574  * Check for an EEH failure at the given I/O address. Call this
575  * routine if the result of a read was all 0xff's and you want to
576  * find out if this is due to an EEH slot freeze event. This routine
577  * will query firmware for the EEH status.
578  *
579  * Note this routine is safe to call in an interrupt context.
580  */
581 int eeh_check_failure(const volatile void __iomem *token)
582 {
583 	unsigned long addr;
584 	struct eeh_dev *edev;
585 
586 	/* Finding the phys addr + pci device; this is pretty quick. */
587 	addr = eeh_token_to_phys((unsigned long __force) token);
588 	edev = eeh_addr_cache_get_dev(addr);
589 	if (!edev) {
590 		eeh_stats.no_device++;
591 		return 0;
592 	}
593 
594 	return eeh_dev_check_failure(edev);
595 }
596 EXPORT_SYMBOL(eeh_check_failure);
597 
598 
599 /**
600  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
601  * @pe: EEH PE
602  *
603  * This routine should be called to reenable frozen MMIO or DMA
604  * so that it would work correctly again. It's useful while doing
605  * recovery or log collection on the indicated device.
606  */
607 int eeh_pci_enable(struct eeh_pe *pe, int function)
608 {
609 	int active_flag, rc;
610 
611 	/*
612 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
613 	 * Also, it's pointless to enable them on unfrozen PE. So
614 	 * we have to check before enabling IO or DMA.
615 	 */
616 	switch (function) {
617 	case EEH_OPT_THAW_MMIO:
618 		active_flag = EEH_STATE_MMIO_ACTIVE;
619 		break;
620 	case EEH_OPT_THAW_DMA:
621 		active_flag = EEH_STATE_DMA_ACTIVE;
622 		break;
623 	case EEH_OPT_DISABLE:
624 	case EEH_OPT_ENABLE:
625 	case EEH_OPT_FREEZE_PE:
626 		active_flag = 0;
627 		break;
628 	default:
629 		pr_warn("%s: Invalid function %d\n",
630 			__func__, function);
631 		return -EINVAL;
632 	}
633 
634 	/*
635 	 * Check if IO or DMA has been enabled before
636 	 * enabling them.
637 	 */
638 	if (active_flag) {
639 		rc = eeh_ops->get_state(pe, NULL);
640 		if (rc < 0)
641 			return rc;
642 
643 		/* Needn't enable it at all */
644 		if (rc == EEH_STATE_NOT_SUPPORT)
645 			return 0;
646 
647 		/* It's already enabled */
648 		if (rc & active_flag)
649 			return 0;
650 	}
651 
652 
653 	/* Issue the request */
654 	rc = eeh_ops->set_option(pe, function);
655 	if (rc)
656 		pr_warn("%s: Unexpected state change %d on "
657 			"PHB#%d-PE#%x, err=%d\n",
658 			__func__, function, pe->phb->global_number,
659 			pe->addr, rc);
660 
661 	/* Check if the request is finished successfully */
662 	if (active_flag) {
663 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
664 		if (rc <= 0)
665 			return rc;
666 
667 		if (rc & active_flag)
668 			return 0;
669 
670 		return -EIO;
671 	}
672 
673 	return rc;
674 }
675 
676 static void *eeh_disable_and_save_dev_state(void *data, void *userdata)
677 {
678 	struct eeh_dev *edev = data;
679 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
680 	struct pci_dev *dev = userdata;
681 
682 	/*
683 	 * The caller should have disabled and saved the
684 	 * state for the specified device
685 	 */
686 	if (!pdev || pdev == dev)
687 		return NULL;
688 
689 	/* Ensure we have D0 power state */
690 	pci_set_power_state(pdev, PCI_D0);
691 
692 	/* Save device state */
693 	pci_save_state(pdev);
694 
695 	/*
696 	 * Disable device to avoid any DMA traffic and
697 	 * interrupt from the device
698 	 */
699 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
700 
701 	return NULL;
702 }
703 
704 static void *eeh_restore_dev_state(void *data, void *userdata)
705 {
706 	struct eeh_dev *edev = data;
707 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
708 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
709 	struct pci_dev *dev = userdata;
710 
711 	if (!pdev)
712 		return NULL;
713 
714 	/* Apply customization from firmware */
715 	if (pdn && eeh_ops->restore_config)
716 		eeh_ops->restore_config(pdn);
717 
718 	/* The caller should restore state for the specified device */
719 	if (pdev != dev)
720 		pci_save_state(pdev);
721 
722 	return NULL;
723 }
724 
725 /**
726  * pcibios_set_pcie_slot_reset - Set PCI-E reset state
727  * @dev: pci device struct
728  * @state: reset state to enter
729  *
730  * Return value:
731  * 	0 if success
732  */
733 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
734 {
735 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
736 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
737 
738 	if (!pe) {
739 		pr_err("%s: No PE found on PCI device %s\n",
740 			__func__, pci_name(dev));
741 		return -EINVAL;
742 	}
743 
744 	switch (state) {
745 	case pcie_deassert_reset:
746 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
747 		eeh_unfreeze_pe(pe, false);
748 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
749 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
750 		break;
751 	case pcie_hot_reset:
752 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
753 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
754 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
755 		eeh_ops->reset(pe, EEH_RESET_HOT);
756 		break;
757 	case pcie_warm_reset:
758 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
759 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
760 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
761 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
762 		break;
763 	default:
764 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
765 		return -EINVAL;
766 	};
767 
768 	return 0;
769 }
770 
771 /**
772  * eeh_set_pe_freset - Check the required reset for the indicated device
773  * @data: EEH device
774  * @flag: return value
775  *
776  * Each device might have its preferred reset type: fundamental or
777  * hot reset. The routine is used to collected the information for
778  * the indicated device and its children so that the bunch of the
779  * devices could be reset properly.
780  */
781 static void *eeh_set_dev_freset(void *data, void *flag)
782 {
783 	struct pci_dev *dev;
784 	unsigned int *freset = (unsigned int *)flag;
785 	struct eeh_dev *edev = (struct eeh_dev *)data;
786 
787 	dev = eeh_dev_to_pci_dev(edev);
788 	if (dev)
789 		*freset |= dev->needs_freset;
790 
791 	return NULL;
792 }
793 
794 /**
795  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
796  * @pe: EEH PE
797  *
798  * Assert the PCI #RST line for 1/4 second.
799  */
800 static void eeh_reset_pe_once(struct eeh_pe *pe)
801 {
802 	unsigned int freset = 0;
803 
804 	/* Determine type of EEH reset required for
805 	 * Partitionable Endpoint, a hot-reset (1)
806 	 * or a fundamental reset (3).
807 	 * A fundamental reset required by any device under
808 	 * Partitionable Endpoint trumps hot-reset.
809 	 */
810 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
811 
812 	if (freset)
813 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
814 	else
815 		eeh_ops->reset(pe, EEH_RESET_HOT);
816 
817 	eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
818 }
819 
820 /**
821  * eeh_reset_pe - Reset the indicated PE
822  * @pe: EEH PE
823  *
824  * This routine should be called to reset indicated device, including
825  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
826  * might be involved as well.
827  */
828 int eeh_reset_pe(struct eeh_pe *pe)
829 {
830 	int flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
831 	int i, state, ret;
832 
833 	/* Mark as reset and block config space */
834 	eeh_pe_state_mark(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
835 
836 	/* Take three shots at resetting the bus */
837 	for (i = 0; i < 3; i++) {
838 		eeh_reset_pe_once(pe);
839 
840 		/*
841 		 * EEH_PE_ISOLATED is expected to be removed after
842 		 * BAR restore.
843 		 */
844 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
845 		if ((state & flags) == flags) {
846 			ret = 0;
847 			goto out;
848 		}
849 
850 		if (state < 0) {
851 			pr_warn("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
852 				__func__, pe->phb->global_number, pe->addr);
853 			ret = -ENOTRECOVERABLE;
854 			goto out;
855 		}
856 
857 		/* We might run out of credits */
858 		ret = -EIO;
859 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
860 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
861 	}
862 
863 out:
864 	eeh_pe_state_clear(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
865 	return ret;
866 }
867 
868 /**
869  * eeh_save_bars - Save device bars
870  * @edev: PCI device associated EEH device
871  *
872  * Save the values of the device bars. Unlike the restore
873  * routine, this routine is *not* recursive. This is because
874  * PCI devices are added individually; but, for the restore,
875  * an entire slot is reset at a time.
876  */
877 void eeh_save_bars(struct eeh_dev *edev)
878 {
879 	struct pci_dn *pdn;
880 	int i;
881 
882 	pdn = eeh_dev_to_pdn(edev);
883 	if (!pdn)
884 		return;
885 
886 	for (i = 0; i < 16; i++)
887 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
888 
889 	/*
890 	 * For PCI bridges including root port, we need enable bus
891 	 * master explicitly. Otherwise, it can't fetch IODA table
892 	 * entries correctly. So we cache the bit in advance so that
893 	 * we can restore it after reset, either PHB range or PE range.
894 	 */
895 	if (edev->mode & EEH_DEV_BRIDGE)
896 		edev->config_space[1] |= PCI_COMMAND_MASTER;
897 }
898 
899 /**
900  * eeh_ops_register - Register platform dependent EEH operations
901  * @ops: platform dependent EEH operations
902  *
903  * Register the platform dependent EEH operation callback
904  * functions. The platform should call this function before
905  * any other EEH operations.
906  */
907 int __init eeh_ops_register(struct eeh_ops *ops)
908 {
909 	if (!ops->name) {
910 		pr_warn("%s: Invalid EEH ops name for %p\n",
911 			__func__, ops);
912 		return -EINVAL;
913 	}
914 
915 	if (eeh_ops && eeh_ops != ops) {
916 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
917 			__func__, eeh_ops->name, ops->name);
918 		return -EEXIST;
919 	}
920 
921 	eeh_ops = ops;
922 
923 	return 0;
924 }
925 
926 /**
927  * eeh_ops_unregister - Unreigster platform dependent EEH operations
928  * @name: name of EEH platform operations
929  *
930  * Unregister the platform dependent EEH operation callback
931  * functions.
932  */
933 int __exit eeh_ops_unregister(const char *name)
934 {
935 	if (!name || !strlen(name)) {
936 		pr_warn("%s: Invalid EEH ops name\n",
937 			__func__);
938 		return -EINVAL;
939 	}
940 
941 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
942 		eeh_ops = NULL;
943 		return 0;
944 	}
945 
946 	return -EEXIST;
947 }
948 
949 static int eeh_reboot_notifier(struct notifier_block *nb,
950 			       unsigned long action, void *unused)
951 {
952 	eeh_clear_flag(EEH_ENABLED);
953 	return NOTIFY_DONE;
954 }
955 
956 static struct notifier_block eeh_reboot_nb = {
957 	.notifier_call = eeh_reboot_notifier,
958 };
959 
960 /**
961  * eeh_init - EEH initialization
962  *
963  * Initialize EEH by trying to enable it for all of the adapters in the system.
964  * As a side effect we can determine here if eeh is supported at all.
965  * Note that we leave EEH on so failed config cycles won't cause a machine
966  * check.  If a user turns off EEH for a particular adapter they are really
967  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
968  * grant access to a slot if EEH isn't enabled, and so we always enable
969  * EEH for all slots/all devices.
970  *
971  * The eeh-force-off option disables EEH checking globally, for all slots.
972  * Even if force-off is set, the EEH hardware is still enabled, so that
973  * newer systems can boot.
974  */
975 int eeh_init(void)
976 {
977 	struct pci_controller *hose, *tmp;
978 	struct pci_dn *pdn;
979 	static int cnt = 0;
980 	int ret = 0;
981 
982 	/*
983 	 * We have to delay the initialization on PowerNV after
984 	 * the PCI hierarchy tree has been built because the PEs
985 	 * are figured out based on PCI devices instead of device
986 	 * tree nodes
987 	 */
988 	if (machine_is(powernv) && cnt++ <= 0)
989 		return ret;
990 
991 	/* Register reboot notifier */
992 	ret = register_reboot_notifier(&eeh_reboot_nb);
993 	if (ret) {
994 		pr_warn("%s: Failed to register notifier (%d)\n",
995 			__func__, ret);
996 		return ret;
997 	}
998 
999 	/* call platform initialization function */
1000 	if (!eeh_ops) {
1001 		pr_warn("%s: Platform EEH operation not found\n",
1002 			__func__);
1003 		return -EEXIST;
1004 	} else if ((ret = eeh_ops->init()))
1005 		return ret;
1006 
1007 	/* Initialize EEH event */
1008 	ret = eeh_event_init();
1009 	if (ret)
1010 		return ret;
1011 
1012 	/* Enable EEH for all adapters */
1013 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1014 		pdn = hose->pci_data;
1015 		traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1016 	}
1017 
1018 	/*
1019 	 * Call platform post-initialization. Actually, It's good chance
1020 	 * to inform platform that EEH is ready to supply service if the
1021 	 * I/O cache stuff has been built up.
1022 	 */
1023 	if (eeh_ops->post_init) {
1024 		ret = eeh_ops->post_init();
1025 		if (ret)
1026 			return ret;
1027 	}
1028 
1029 	if (eeh_enabled())
1030 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1031 	else
1032 		pr_warn("EEH: No capable adapters found\n");
1033 
1034 	return ret;
1035 }
1036 
1037 core_initcall_sync(eeh_init);
1038 
1039 /**
1040  * eeh_add_device_early - Enable EEH for the indicated device node
1041  * @pdn: PCI device node for which to set up EEH
1042  *
1043  * This routine must be used to perform EEH initialization for PCI
1044  * devices that were added after system boot (e.g. hotplug, dlpar).
1045  * This routine must be called before any i/o is performed to the
1046  * adapter (inluding any config-space i/o).
1047  * Whether this actually enables EEH or not for this device depends
1048  * on the CEC architecture, type of the device, on earlier boot
1049  * command-line arguments & etc.
1050  */
1051 void eeh_add_device_early(struct pci_dn *pdn)
1052 {
1053 	struct pci_controller *phb;
1054 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1055 
1056 	if (!edev || !eeh_enabled())
1057 		return;
1058 
1059 	/* USB Bus children of PCI devices will not have BUID's */
1060 	phb = edev->phb;
1061 	if (NULL == phb ||
1062 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1063 		return;
1064 
1065 	eeh_ops->probe(pdn, NULL);
1066 }
1067 
1068 /**
1069  * eeh_add_device_tree_early - Enable EEH for the indicated device
1070  * @pdn: PCI device node
1071  *
1072  * This routine must be used to perform EEH initialization for the
1073  * indicated PCI device that was added after system boot (e.g.
1074  * hotplug, dlpar).
1075  */
1076 void eeh_add_device_tree_early(struct pci_dn *pdn)
1077 {
1078 	struct pci_dn *n;
1079 
1080 	if (!pdn)
1081 		return;
1082 
1083 	list_for_each_entry(n, &pdn->child_list, list)
1084 		eeh_add_device_tree_early(n);
1085 	eeh_add_device_early(pdn);
1086 }
1087 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1088 
1089 /**
1090  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1091  * @dev: pci device for which to set up EEH
1092  *
1093  * This routine must be used to complete EEH initialization for PCI
1094  * devices that were added after system boot (e.g. hotplug, dlpar).
1095  */
1096 void eeh_add_device_late(struct pci_dev *dev)
1097 {
1098 	struct pci_dn *pdn;
1099 	struct eeh_dev *edev;
1100 
1101 	if (!dev || !eeh_enabled())
1102 		return;
1103 
1104 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1105 
1106 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1107 	edev = pdn_to_eeh_dev(pdn);
1108 	if (edev->pdev == dev) {
1109 		pr_debug("EEH: Already referenced !\n");
1110 		return;
1111 	}
1112 
1113 	/*
1114 	 * The EEH cache might not be removed correctly because of
1115 	 * unbalanced kref to the device during unplug time, which
1116 	 * relies on pcibios_release_device(). So we have to remove
1117 	 * that here explicitly.
1118 	 */
1119 	if (edev->pdev) {
1120 		eeh_rmv_from_parent_pe(edev);
1121 		eeh_addr_cache_rmv_dev(edev->pdev);
1122 		eeh_sysfs_remove_device(edev->pdev);
1123 		edev->mode &= ~EEH_DEV_SYSFS;
1124 
1125 		/*
1126 		 * We definitely should have the PCI device removed
1127 		 * though it wasn't correctly. So we needn't call
1128 		 * into error handler afterwards.
1129 		 */
1130 		edev->mode |= EEH_DEV_NO_HANDLER;
1131 
1132 		edev->pdev = NULL;
1133 		dev->dev.archdata.edev = NULL;
1134 	}
1135 
1136 	edev->pdev = dev;
1137 	dev->dev.archdata.edev = edev;
1138 
1139 	eeh_addr_cache_insert_dev(dev);
1140 }
1141 
1142 /**
1143  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1144  * @bus: PCI bus
1145  *
1146  * This routine must be used to perform EEH initialization for PCI
1147  * devices which are attached to the indicated PCI bus. The PCI bus
1148  * is added after system boot through hotplug or dlpar.
1149  */
1150 void eeh_add_device_tree_late(struct pci_bus *bus)
1151 {
1152 	struct pci_dev *dev;
1153 
1154 	list_for_each_entry(dev, &bus->devices, bus_list) {
1155 		eeh_add_device_late(dev);
1156 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1157 			struct pci_bus *subbus = dev->subordinate;
1158 			if (subbus)
1159 				eeh_add_device_tree_late(subbus);
1160 		}
1161 	}
1162 }
1163 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1164 
1165 /**
1166  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1167  * @bus: PCI bus
1168  *
1169  * This routine must be used to add EEH sysfs files for PCI
1170  * devices which are attached to the indicated PCI bus. The PCI bus
1171  * is added after system boot through hotplug or dlpar.
1172  */
1173 void eeh_add_sysfs_files(struct pci_bus *bus)
1174 {
1175 	struct pci_dev *dev;
1176 
1177 	list_for_each_entry(dev, &bus->devices, bus_list) {
1178 		eeh_sysfs_add_device(dev);
1179 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1180 			struct pci_bus *subbus = dev->subordinate;
1181 			if (subbus)
1182 				eeh_add_sysfs_files(subbus);
1183 		}
1184 	}
1185 }
1186 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1187 
1188 /**
1189  * eeh_remove_device - Undo EEH setup for the indicated pci device
1190  * @dev: pci device to be removed
1191  *
1192  * This routine should be called when a device is removed from
1193  * a running system (e.g. by hotplug or dlpar).  It unregisters
1194  * the PCI device from the EEH subsystem.  I/O errors affecting
1195  * this device will no longer be detected after this call; thus,
1196  * i/o errors affecting this slot may leave this device unusable.
1197  */
1198 void eeh_remove_device(struct pci_dev *dev)
1199 {
1200 	struct eeh_dev *edev;
1201 
1202 	if (!dev || !eeh_enabled())
1203 		return;
1204 	edev = pci_dev_to_eeh_dev(dev);
1205 
1206 	/* Unregister the device with the EEH/PCI address search system */
1207 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1208 
1209 	if (!edev || !edev->pdev || !edev->pe) {
1210 		pr_debug("EEH: Not referenced !\n");
1211 		return;
1212 	}
1213 
1214 	/*
1215 	 * During the hotplug for EEH error recovery, we need the EEH
1216 	 * device attached to the parent PE in order for BAR restore
1217 	 * a bit later. So we keep it for BAR restore and remove it
1218 	 * from the parent PE during the BAR resotre.
1219 	 */
1220 	edev->pdev = NULL;
1221 	dev->dev.archdata.edev = NULL;
1222 	if (!(edev->pe->state & EEH_PE_KEEP))
1223 		eeh_rmv_from_parent_pe(edev);
1224 	else
1225 		edev->mode |= EEH_DEV_DISCONNECTED;
1226 
1227 	/*
1228 	 * We're removing from the PCI subsystem, that means
1229 	 * the PCI device driver can't support EEH or not
1230 	 * well. So we rely on hotplug completely to do recovery
1231 	 * for the specific PCI device.
1232 	 */
1233 	edev->mode |= EEH_DEV_NO_HANDLER;
1234 
1235 	eeh_addr_cache_rmv_dev(dev);
1236 	eeh_sysfs_remove_device(dev);
1237 	edev->mode &= ~EEH_DEV_SYSFS;
1238 }
1239 
1240 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1241 {
1242 	int ret;
1243 
1244 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1245 	if (ret) {
1246 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1247 			__func__, ret, pe->phb->global_number, pe->addr);
1248 		return ret;
1249 	}
1250 
1251 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1252 	if (ret) {
1253 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1254 			__func__, ret, pe->phb->global_number, pe->addr);
1255 		return ret;
1256 	}
1257 
1258 	/* Clear software isolated state */
1259 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1260 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1261 
1262 	return ret;
1263 }
1264 
1265 
1266 static struct pci_device_id eeh_reset_ids[] = {
1267 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1268 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1269 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1270 	{ 0 }
1271 };
1272 
1273 static int eeh_pe_change_owner(struct eeh_pe *pe)
1274 {
1275 	struct eeh_dev *edev, *tmp;
1276 	struct pci_dev *pdev;
1277 	struct pci_device_id *id;
1278 	int flags, ret;
1279 
1280 	/* Check PE state */
1281 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1282 	ret = eeh_ops->get_state(pe, NULL);
1283 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1284 		return 0;
1285 
1286 	/* Unfrozen PE, nothing to do */
1287 	if ((ret & flags) == flags)
1288 		return 0;
1289 
1290 	/* Frozen PE, check if it needs PE level reset */
1291 	eeh_pe_for_each_dev(pe, edev, tmp) {
1292 		pdev = eeh_dev_to_pci_dev(edev);
1293 		if (!pdev)
1294 			continue;
1295 
1296 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1297 			if (id->vendor != PCI_ANY_ID &&
1298 			    id->vendor != pdev->vendor)
1299 				continue;
1300 			if (id->device != PCI_ANY_ID &&
1301 			    id->device != pdev->device)
1302 				continue;
1303 			if (id->subvendor != PCI_ANY_ID &&
1304 			    id->subvendor != pdev->subsystem_vendor)
1305 				continue;
1306 			if (id->subdevice != PCI_ANY_ID &&
1307 			    id->subdevice != pdev->subsystem_device)
1308 				continue;
1309 
1310 			goto reset;
1311 		}
1312 	}
1313 
1314 	return eeh_unfreeze_pe(pe, true);
1315 
1316 reset:
1317 	return eeh_pe_reset_and_recover(pe);
1318 }
1319 
1320 /**
1321  * eeh_dev_open - Increase count of pass through devices for PE
1322  * @pdev: PCI device
1323  *
1324  * Increase count of passed through devices for the indicated
1325  * PE. In the result, the EEH errors detected on the PE won't be
1326  * reported. The PE owner will be responsible for detection
1327  * and recovery.
1328  */
1329 int eeh_dev_open(struct pci_dev *pdev)
1330 {
1331 	struct eeh_dev *edev;
1332 	int ret = -ENODEV;
1333 
1334 	mutex_lock(&eeh_dev_mutex);
1335 
1336 	/* No PCI device ? */
1337 	if (!pdev)
1338 		goto out;
1339 
1340 	/* No EEH device or PE ? */
1341 	edev = pci_dev_to_eeh_dev(pdev);
1342 	if (!edev || !edev->pe)
1343 		goto out;
1344 
1345 	/*
1346 	 * The PE might have been put into frozen state, but we
1347 	 * didn't detect that yet. The passed through PCI devices
1348 	 * in frozen PE won't work properly. Clear the frozen state
1349 	 * in advance.
1350 	 */
1351 	ret = eeh_pe_change_owner(edev->pe);
1352 	if (ret)
1353 		goto out;
1354 
1355 	/* Increase PE's pass through count */
1356 	atomic_inc(&edev->pe->pass_dev_cnt);
1357 	mutex_unlock(&eeh_dev_mutex);
1358 
1359 	return 0;
1360 out:
1361 	mutex_unlock(&eeh_dev_mutex);
1362 	return ret;
1363 }
1364 EXPORT_SYMBOL_GPL(eeh_dev_open);
1365 
1366 /**
1367  * eeh_dev_release - Decrease count of pass through devices for PE
1368  * @pdev: PCI device
1369  *
1370  * Decrease count of pass through devices for the indicated PE. If
1371  * there is no passed through device in PE, the EEH errors detected
1372  * on the PE will be reported and handled as usual.
1373  */
1374 void eeh_dev_release(struct pci_dev *pdev)
1375 {
1376 	struct eeh_dev *edev;
1377 
1378 	mutex_lock(&eeh_dev_mutex);
1379 
1380 	/* No PCI device ? */
1381 	if (!pdev)
1382 		goto out;
1383 
1384 	/* No EEH device ? */
1385 	edev = pci_dev_to_eeh_dev(pdev);
1386 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1387 		goto out;
1388 
1389 	/* Decrease PE's pass through count */
1390 	atomic_dec(&edev->pe->pass_dev_cnt);
1391 	WARN_ON(atomic_read(&edev->pe->pass_dev_cnt) < 0);
1392 	eeh_pe_change_owner(edev->pe);
1393 out:
1394 	mutex_unlock(&eeh_dev_mutex);
1395 }
1396 EXPORT_SYMBOL(eeh_dev_release);
1397 
1398 #ifdef CONFIG_IOMMU_API
1399 
1400 static int dev_has_iommu_table(struct device *dev, void *data)
1401 {
1402 	struct pci_dev *pdev = to_pci_dev(dev);
1403 	struct pci_dev **ppdev = data;
1404 	struct iommu_table *tbl;
1405 
1406 	if (!dev)
1407 		return 0;
1408 
1409 	tbl = get_iommu_table_base(dev);
1410 	if (tbl && tbl->it_group) {
1411 		*ppdev = pdev;
1412 		return 1;
1413 	}
1414 
1415 	return 0;
1416 }
1417 
1418 /**
1419  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1420  * @group: IOMMU group
1421  *
1422  * The routine is called to convert IOMMU group to EEH PE.
1423  */
1424 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1425 {
1426 	struct pci_dev *pdev = NULL;
1427 	struct eeh_dev *edev;
1428 	int ret;
1429 
1430 	/* No IOMMU group ? */
1431 	if (!group)
1432 		return NULL;
1433 
1434 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1435 	if (!ret || !pdev)
1436 		return NULL;
1437 
1438 	/* No EEH device or PE ? */
1439 	edev = pci_dev_to_eeh_dev(pdev);
1440 	if (!edev || !edev->pe)
1441 		return NULL;
1442 
1443 	return edev->pe;
1444 }
1445 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1446 
1447 #endif /* CONFIG_IOMMU_API */
1448 
1449 /**
1450  * eeh_pe_set_option - Set options for the indicated PE
1451  * @pe: EEH PE
1452  * @option: requested option
1453  *
1454  * The routine is called to enable or disable EEH functionality
1455  * on the indicated PE, to enable IO or DMA for the frozen PE.
1456  */
1457 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1458 {
1459 	int ret = 0;
1460 
1461 	/* Invalid PE ? */
1462 	if (!pe)
1463 		return -ENODEV;
1464 
1465 	/*
1466 	 * EEH functionality could possibly be disabled, just
1467 	 * return error for the case. And the EEH functinality
1468 	 * isn't expected to be disabled on one specific PE.
1469 	 */
1470 	switch (option) {
1471 	case EEH_OPT_ENABLE:
1472 		if (eeh_enabled()) {
1473 			ret = eeh_pe_change_owner(pe);
1474 			break;
1475 		}
1476 		ret = -EIO;
1477 		break;
1478 	case EEH_OPT_DISABLE:
1479 		break;
1480 	case EEH_OPT_THAW_MMIO:
1481 	case EEH_OPT_THAW_DMA:
1482 		if (!eeh_ops || !eeh_ops->set_option) {
1483 			ret = -ENOENT;
1484 			break;
1485 		}
1486 
1487 		ret = eeh_pci_enable(pe, option);
1488 		break;
1489 	default:
1490 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1491 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1492 		ret = -EINVAL;
1493 	}
1494 
1495 	return ret;
1496 }
1497 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1498 
1499 /**
1500  * eeh_pe_get_state - Retrieve PE's state
1501  * @pe: EEH PE
1502  *
1503  * Retrieve the PE's state, which includes 3 aspects: enabled
1504  * DMA, enabled IO and asserted reset.
1505  */
1506 int eeh_pe_get_state(struct eeh_pe *pe)
1507 {
1508 	int result, ret = 0;
1509 	bool rst_active, dma_en, mmio_en;
1510 
1511 	/* Existing PE ? */
1512 	if (!pe)
1513 		return -ENODEV;
1514 
1515 	if (!eeh_ops || !eeh_ops->get_state)
1516 		return -ENOENT;
1517 
1518 	result = eeh_ops->get_state(pe, NULL);
1519 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1520 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1521 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1522 
1523 	if (rst_active)
1524 		ret = EEH_PE_STATE_RESET;
1525 	else if (dma_en && mmio_en)
1526 		ret = EEH_PE_STATE_NORMAL;
1527 	else if (!dma_en && !mmio_en)
1528 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1529 	else if (!dma_en && mmio_en)
1530 		ret = EEH_PE_STATE_STOPPED_DMA;
1531 	else
1532 		ret = EEH_PE_STATE_UNAVAIL;
1533 
1534 	return ret;
1535 }
1536 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1537 
1538 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1539 {
1540 	struct eeh_dev *edev, *tmp;
1541 	struct pci_dev *pdev;
1542 	int ret = 0;
1543 
1544 	/* Restore config space */
1545 	eeh_pe_restore_bars(pe);
1546 
1547 	/*
1548 	 * Reenable PCI devices as the devices passed
1549 	 * through are always enabled before the reset.
1550 	 */
1551 	eeh_pe_for_each_dev(pe, edev, tmp) {
1552 		pdev = eeh_dev_to_pci_dev(edev);
1553 		if (!pdev)
1554 			continue;
1555 
1556 		ret = pci_reenable_device(pdev);
1557 		if (ret) {
1558 			pr_warn("%s: Failure %d reenabling %s\n",
1559 				__func__, ret, pci_name(pdev));
1560 			return ret;
1561 		}
1562 	}
1563 
1564 	/* The PE is still in frozen state */
1565 	return eeh_unfreeze_pe(pe, true);
1566 }
1567 
1568 /**
1569  * eeh_pe_reset - Issue PE reset according to specified type
1570  * @pe: EEH PE
1571  * @option: reset type
1572  *
1573  * The routine is called to reset the specified PE with the
1574  * indicated type, either fundamental reset or hot reset.
1575  * PE reset is the most important part for error recovery.
1576  */
1577 int eeh_pe_reset(struct eeh_pe *pe, int option)
1578 {
1579 	int ret = 0;
1580 
1581 	/* Invalid PE ? */
1582 	if (!pe)
1583 		return -ENODEV;
1584 
1585 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1586 		return -ENOENT;
1587 
1588 	switch (option) {
1589 	case EEH_RESET_DEACTIVATE:
1590 		ret = eeh_ops->reset(pe, option);
1591 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1592 		if (ret)
1593 			break;
1594 
1595 		ret = eeh_pe_reenable_devices(pe);
1596 		break;
1597 	case EEH_RESET_HOT:
1598 	case EEH_RESET_FUNDAMENTAL:
1599 		/*
1600 		 * Proactively freeze the PE to drop all MMIO access
1601 		 * during reset, which should be banned as it's always
1602 		 * cause recursive EEH error.
1603 		 */
1604 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1605 
1606 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1607 		ret = eeh_ops->reset(pe, option);
1608 		break;
1609 	default:
1610 		pr_debug("%s: Unsupported option %d\n",
1611 			__func__, option);
1612 		ret = -EINVAL;
1613 	}
1614 
1615 	return ret;
1616 }
1617 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1618 
1619 /**
1620  * eeh_pe_configure - Configure PCI bridges after PE reset
1621  * @pe: EEH PE
1622  *
1623  * The routine is called to restore the PCI config space for
1624  * those PCI devices, especially PCI bridges affected by PE
1625  * reset issued previously.
1626  */
1627 int eeh_pe_configure(struct eeh_pe *pe)
1628 {
1629 	int ret = 0;
1630 
1631 	/* Invalid PE ? */
1632 	if (!pe)
1633 		return -ENODEV;
1634 
1635 	return ret;
1636 }
1637 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1638 
1639 static int proc_eeh_show(struct seq_file *m, void *v)
1640 {
1641 	if (!eeh_enabled()) {
1642 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1643 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1644 	} else {
1645 		seq_printf(m, "EEH Subsystem is enabled\n");
1646 		seq_printf(m,
1647 				"no device=%llu\n"
1648 				"no device node=%llu\n"
1649 				"no config address=%llu\n"
1650 				"check not wanted=%llu\n"
1651 				"eeh_total_mmio_ffs=%llu\n"
1652 				"eeh_false_positives=%llu\n"
1653 				"eeh_slot_resets=%llu\n",
1654 				eeh_stats.no_device,
1655 				eeh_stats.no_dn,
1656 				eeh_stats.no_cfg_addr,
1657 				eeh_stats.ignored_check,
1658 				eeh_stats.total_mmio_ffs,
1659 				eeh_stats.false_positives,
1660 				eeh_stats.slot_resets);
1661 	}
1662 
1663 	return 0;
1664 }
1665 
1666 static int proc_eeh_open(struct inode *inode, struct file *file)
1667 {
1668 	return single_open(file, proc_eeh_show, NULL);
1669 }
1670 
1671 static const struct file_operations proc_eeh_operations = {
1672 	.open      = proc_eeh_open,
1673 	.read      = seq_read,
1674 	.llseek    = seq_lseek,
1675 	.release   = single_release,
1676 };
1677 
1678 #ifdef CONFIG_DEBUG_FS
1679 static int eeh_enable_dbgfs_set(void *data, u64 val)
1680 {
1681 	if (val)
1682 		eeh_clear_flag(EEH_FORCE_DISABLED);
1683 	else
1684 		eeh_add_flag(EEH_FORCE_DISABLED);
1685 
1686 	/* Notify the backend */
1687 	if (eeh_ops->post_init)
1688 		eeh_ops->post_init();
1689 
1690 	return 0;
1691 }
1692 
1693 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1694 {
1695 	if (eeh_enabled())
1696 		*val = 0x1ul;
1697 	else
1698 		*val = 0x0ul;
1699 	return 0;
1700 }
1701 
1702 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1703 {
1704 	eeh_max_freezes = val;
1705 	return 0;
1706 }
1707 
1708 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1709 {
1710 	*val = eeh_max_freezes;
1711 	return 0;
1712 }
1713 
1714 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1715 			eeh_enable_dbgfs_set, "0x%llx\n");
1716 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1717 			eeh_freeze_dbgfs_set, "0x%llx\n");
1718 #endif
1719 
1720 static int __init eeh_init_proc(void)
1721 {
1722 	if (machine_is(pseries) || machine_is(powernv)) {
1723 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1724 #ifdef CONFIG_DEBUG_FS
1725 		debugfs_create_file("eeh_enable", 0600,
1726                                     powerpc_debugfs_root, NULL,
1727                                     &eeh_enable_dbgfs_ops);
1728 		debugfs_create_file("eeh_max_freezes", 0600,
1729 				    powerpc_debugfs_root, NULL,
1730 				    &eeh_freeze_dbgfs_ops);
1731 #endif
1732 	}
1733 
1734 	return 0;
1735 }
1736 __initcall(eeh_init_proc);
1737