xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision e8f6f3b4)
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 /* Platform dependent EEH operations */
108 struct eeh_ops *eeh_ops = NULL;
109 
110 /* Lock to avoid races due to multiple reports of an error */
111 DEFINE_RAW_SPINLOCK(confirm_error_lock);
112 
113 /* Lock to protect passed flags */
114 static DEFINE_MUTEX(eeh_dev_mutex);
115 
116 /* Buffer for reporting pci register dumps. Its here in BSS, and
117  * not dynamically alloced, so that it ends up in RMO where RTAS
118  * can access it.
119  */
120 #define EEH_PCI_REGS_LOG_LEN 8192
121 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
122 
123 /*
124  * The struct is used to maintain the EEH global statistic
125  * information. Besides, the EEH global statistics will be
126  * exported to user space through procfs
127  */
128 struct eeh_stats {
129 	u64 no_device;		/* PCI device not found		*/
130 	u64 no_dn;		/* OF node not found		*/
131 	u64 no_cfg_addr;	/* Config address not found	*/
132 	u64 ignored_check;	/* EEH check skipped		*/
133 	u64 total_mmio_ffs;	/* Total EEH checks		*/
134 	u64 false_positives;	/* Unnecessary EEH checks	*/
135 	u64 slot_resets;	/* PE reset			*/
136 };
137 
138 static struct eeh_stats eeh_stats;
139 
140 #define IS_BRIDGE(class_code) (((class_code)<<16) == PCI_BASE_CLASS_BRIDGE)
141 
142 static int __init eeh_setup(char *str)
143 {
144 	if (!strcmp(str, "off"))
145 		eeh_add_flag(EEH_FORCE_DISABLED);
146 	else if (!strcmp(str, "early_log"))
147 		eeh_add_flag(EEH_EARLY_DUMP_LOG);
148 
149 	return 1;
150 }
151 __setup("eeh=", eeh_setup);
152 
153 /*
154  * This routine captures assorted PCI configuration space data
155  * for the indicated PCI device, and puts them into a buffer
156  * for RTAS error logging.
157  */
158 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
159 {
160 	struct device_node *dn = eeh_dev_to_of_node(edev);
161 	u32 cfg;
162 	int cap, i;
163 	int n = 0, l = 0;
164 	char buffer[128];
165 
166 	n += scnprintf(buf+n, len-n, "%s\n", dn->full_name);
167 	pr_warn("EEH: of node=%s\n", dn->full_name);
168 
169 	eeh_ops->read_config(dn, PCI_VENDOR_ID, 4, &cfg);
170 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
171 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
172 
173 	eeh_ops->read_config(dn, PCI_COMMAND, 4, &cfg);
174 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
175 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
176 
177 	/* Gather bridge-specific registers */
178 	if (edev->mode & EEH_DEV_BRIDGE) {
179 		eeh_ops->read_config(dn, PCI_SEC_STATUS, 2, &cfg);
180 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
181 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
182 
183 		eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &cfg);
184 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
185 		pr_warn("EEH: Bridge control: %04x\n", cfg);
186 	}
187 
188 	/* Dump out the PCI-X command and status regs */
189 	cap = edev->pcix_cap;
190 	if (cap) {
191 		eeh_ops->read_config(dn, cap, 4, &cfg);
192 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
193 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
194 
195 		eeh_ops->read_config(dn, cap+4, 4, &cfg);
196 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
197 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
198 	}
199 
200 	/* If PCI-E capable, dump PCI-E cap 10 */
201 	cap = edev->pcie_cap;
202 	if (cap) {
203 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
204 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
205 
206 		for (i=0; i<=8; i++) {
207 			eeh_ops->read_config(dn, cap+4*i, 4, &cfg);
208 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
209 
210 			if ((i % 4) == 0) {
211 				if (i != 0)
212 					pr_warn("%s\n", buffer);
213 
214 				l = scnprintf(buffer, sizeof(buffer),
215 					      "EEH: PCI-E %02x: %08x ",
216 					      4*i, cfg);
217 			} else {
218 				l += scnprintf(buffer+l, sizeof(buffer)-l,
219 					       "%08x ", cfg);
220 			}
221 
222 		}
223 
224 		pr_warn("%s\n", buffer);
225 	}
226 
227 	/* If AER capable, dump it */
228 	cap = edev->aer_cap;
229 	if (cap) {
230 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
231 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
232 
233 		for (i=0; i<=13; i++) {
234 			eeh_ops->read_config(dn, cap+4*i, 4, &cfg);
235 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
236 
237 			if ((i % 4) == 0) {
238 				if (i != 0)
239 					pr_warn("%s\n", buffer);
240 
241 				l = scnprintf(buffer, sizeof(buffer),
242 					      "EEH: PCI-E AER %02x: %08x ",
243 					      4*i, cfg);
244 			} else {
245 				l += scnprintf(buffer+l, sizeof(buffer)-l,
246 					       "%08x ", cfg);
247 			}
248 		}
249 
250 		pr_warn("%s\n", buffer);
251 	}
252 
253 	return n;
254 }
255 
256 static void *eeh_dump_pe_log(void *data, void *flag)
257 {
258 	struct eeh_pe *pe = data;
259 	struct eeh_dev *edev, *tmp;
260 	size_t *plen = flag;
261 
262 	/* If the PE's config space is blocked, 0xFF's will be
263 	 * returned. It's pointless to collect the log in this
264 	 * case.
265 	 */
266 	if (pe->state & EEH_PE_CFG_BLOCKED)
267 		return NULL;
268 
269 	eeh_pe_for_each_dev(pe, edev, tmp)
270 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
271 					  EEH_PCI_REGS_LOG_LEN - *plen);
272 
273 	return NULL;
274 }
275 
276 /**
277  * eeh_slot_error_detail - Generate combined log including driver log and error log
278  * @pe: EEH PE
279  * @severity: temporary or permanent error log
280  *
281  * This routine should be called to generate the combined log, which
282  * is comprised of driver log and error log. The driver log is figured
283  * out from the config space of the corresponding PCI device, while
284  * the error log is fetched through platform dependent function call.
285  */
286 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
287 {
288 	size_t loglen = 0;
289 
290 	/*
291 	 * When the PHB is fenced or dead, it's pointless to collect
292 	 * the data from PCI config space because it should return
293 	 * 0xFF's. For ER, we still retrieve the data from the PCI
294 	 * config space.
295 	 *
296 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
297 	 * 0xFF's is always returned from PCI config space.
298 	 */
299 	if (!(pe->type & EEH_PE_PHB)) {
300 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG))
301 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
302 		eeh_ops->configure_bridge(pe);
303 		eeh_pe_restore_bars(pe);
304 
305 		pci_regs_buf[0] = 0;
306 		eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
307 	}
308 
309 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
310 }
311 
312 /**
313  * eeh_token_to_phys - Convert EEH address token to phys address
314  * @token: I/O token, should be address in the form 0xA....
315  *
316  * This routine should be called to convert virtual I/O address
317  * to physical one.
318  */
319 static inline unsigned long eeh_token_to_phys(unsigned long token)
320 {
321 	pte_t *ptep;
322 	unsigned long pa;
323 	int hugepage_shift;
324 
325 	/*
326 	 * We won't find hugepages here, iomem
327 	 */
328 	ptep = find_linux_pte_or_hugepte(init_mm.pgd, token, &hugepage_shift);
329 	if (!ptep)
330 		return token;
331 	WARN_ON(hugepage_shift);
332 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
333 
334 	return pa | (token & (PAGE_SIZE-1));
335 }
336 
337 /*
338  * On PowerNV platform, we might already have fenced PHB there.
339  * For that case, it's meaningless to recover frozen PE. Intead,
340  * We have to handle fenced PHB firstly.
341  */
342 static int eeh_phb_check_failure(struct eeh_pe *pe)
343 {
344 	struct eeh_pe *phb_pe;
345 	unsigned long flags;
346 	int ret;
347 
348 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
349 		return -EPERM;
350 
351 	/* Find the PHB PE */
352 	phb_pe = eeh_phb_pe_get(pe->phb);
353 	if (!phb_pe) {
354 		pr_warn("%s Can't find PE for PHB#%d\n",
355 			__func__, pe->phb->global_number);
356 		return -EEXIST;
357 	}
358 
359 	/* If the PHB has been in problematic state */
360 	eeh_serialize_lock(&flags);
361 	if (phb_pe->state & EEH_PE_ISOLATED) {
362 		ret = 0;
363 		goto out;
364 	}
365 
366 	/* Check PHB state */
367 	ret = eeh_ops->get_state(phb_pe, NULL);
368 	if ((ret < 0) ||
369 	    (ret == EEH_STATE_NOT_SUPPORT) ||
370 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
371 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
372 		ret = 0;
373 		goto out;
374 	}
375 
376 	/* Isolate the PHB and send event */
377 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
378 	eeh_serialize_unlock(flags);
379 
380 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
381 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
382 	dump_stack();
383 	eeh_send_failure_event(phb_pe);
384 
385 	return 1;
386 out:
387 	eeh_serialize_unlock(flags);
388 	return ret;
389 }
390 
391 /**
392  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
393  * @edev: eeh device
394  *
395  * Check for an EEH failure for the given device node.  Call this
396  * routine if the result of a read was all 0xff's and you want to
397  * find out if this is due to an EEH slot freeze.  This routine
398  * will query firmware for the EEH status.
399  *
400  * Returns 0 if there has not been an EEH error; otherwise returns
401  * a non-zero value and queues up a slot isolation event notification.
402  *
403  * It is safe to call this routine in an interrupt context.
404  */
405 int eeh_dev_check_failure(struct eeh_dev *edev)
406 {
407 	int ret;
408 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
409 	unsigned long flags;
410 	struct device_node *dn;
411 	struct pci_dev *dev;
412 	struct eeh_pe *pe, *parent_pe, *phb_pe;
413 	int rc = 0;
414 	const char *location;
415 
416 	eeh_stats.total_mmio_ffs++;
417 
418 	if (!eeh_enabled())
419 		return 0;
420 
421 	if (!edev) {
422 		eeh_stats.no_dn++;
423 		return 0;
424 	}
425 	dn = eeh_dev_to_of_node(edev);
426 	dev = eeh_dev_to_pci_dev(edev);
427 	pe = eeh_dev_to_pe(edev);
428 
429 	/* Access to IO BARs might get this far and still not want checking. */
430 	if (!pe) {
431 		eeh_stats.ignored_check++;
432 		pr_debug("EEH: Ignored check for %s %s\n",
433 			eeh_pci_name(dev), dn->full_name);
434 		return 0;
435 	}
436 
437 	if (!pe->addr && !pe->config_addr) {
438 		eeh_stats.no_cfg_addr++;
439 		return 0;
440 	}
441 
442 	/*
443 	 * On PowerNV platform, we might already have fenced PHB
444 	 * there and we need take care of that firstly.
445 	 */
446 	ret = eeh_phb_check_failure(pe);
447 	if (ret > 0)
448 		return ret;
449 
450 	/*
451 	 * If the PE isn't owned by us, we shouldn't check the
452 	 * state. Instead, let the owner handle it if the PE has
453 	 * been frozen.
454 	 */
455 	if (eeh_pe_passed(pe))
456 		return 0;
457 
458 	/* If we already have a pending isolation event for this
459 	 * slot, we know it's bad already, we don't need to check.
460 	 * Do this checking under a lock; as multiple PCI devices
461 	 * in one slot might report errors simultaneously, and we
462 	 * only want one error recovery routine running.
463 	 */
464 	eeh_serialize_lock(&flags);
465 	rc = 1;
466 	if (pe->state & EEH_PE_ISOLATED) {
467 		pe->check_count++;
468 		if (pe->check_count % EEH_MAX_FAILS == 0) {
469 			location = of_get_property(dn, "ibm,loc-code", NULL);
470 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
471 				"location=%s driver=%s pci addr=%s\n",
472 				pe->check_count, location,
473 				eeh_driver_name(dev), eeh_pci_name(dev));
474 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
475 				eeh_driver_name(dev));
476 			dump_stack();
477 		}
478 		goto dn_unlock;
479 	}
480 
481 	/*
482 	 * Now test for an EEH failure.  This is VERY expensive.
483 	 * Note that the eeh_config_addr may be a parent device
484 	 * in the case of a device behind a bridge, or it may be
485 	 * function zero of a multi-function device.
486 	 * In any case they must share a common PHB.
487 	 */
488 	ret = eeh_ops->get_state(pe, NULL);
489 
490 	/* Note that config-io to empty slots may fail;
491 	 * they are empty when they don't have children.
492 	 * We will punt with the following conditions: Failure to get
493 	 * PE's state, EEH not support and Permanently unavailable
494 	 * state, PE is in good state.
495 	 */
496 	if ((ret < 0) ||
497 	    (ret == EEH_STATE_NOT_SUPPORT) ||
498 	    ((ret & active_flags) == active_flags)) {
499 		eeh_stats.false_positives++;
500 		pe->false_positives++;
501 		rc = 0;
502 		goto dn_unlock;
503 	}
504 
505 	/*
506 	 * It should be corner case that the parent PE has been
507 	 * put into frozen state as well. We should take care
508 	 * that at first.
509 	 */
510 	parent_pe = pe->parent;
511 	while (parent_pe) {
512 		/* Hit the ceiling ? */
513 		if (parent_pe->type & EEH_PE_PHB)
514 			break;
515 
516 		/* Frozen parent PE ? */
517 		ret = eeh_ops->get_state(parent_pe, NULL);
518 		if (ret > 0 &&
519 		    (ret & active_flags) != active_flags)
520 			pe = parent_pe;
521 
522 		/* Next parent level */
523 		parent_pe = parent_pe->parent;
524 	}
525 
526 	eeh_stats.slot_resets++;
527 
528 	/* Avoid repeated reports of this failure, including problems
529 	 * with other functions on this device, and functions under
530 	 * bridges.
531 	 */
532 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
533 	eeh_serialize_unlock(flags);
534 
535 	/* Most EEH events are due to device driver bugs.  Having
536 	 * a stack trace will help the device-driver authors figure
537 	 * out what happened.  So print that out.
538 	 */
539 	phb_pe = eeh_phb_pe_get(pe->phb);
540 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
541 	       pe->phb->global_number, pe->addr);
542 	pr_err("EEH: PE location: %s, PHB location: %s\n",
543 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
544 	dump_stack();
545 
546 	eeh_send_failure_event(pe);
547 
548 	return 1;
549 
550 dn_unlock:
551 	eeh_serialize_unlock(flags);
552 	return rc;
553 }
554 
555 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
556 
557 /**
558  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
559  * @token: I/O address
560  *
561  * Check for an EEH failure at the given I/O address. Call this
562  * routine if the result of a read was all 0xff's and you want to
563  * find out if this is due to an EEH slot freeze event. This routine
564  * will query firmware for the EEH status.
565  *
566  * Note this routine is safe to call in an interrupt context.
567  */
568 int eeh_check_failure(const volatile void __iomem *token)
569 {
570 	unsigned long addr;
571 	struct eeh_dev *edev;
572 
573 	/* Finding the phys addr + pci device; this is pretty quick. */
574 	addr = eeh_token_to_phys((unsigned long __force) token);
575 	edev = eeh_addr_cache_get_dev(addr);
576 	if (!edev) {
577 		eeh_stats.no_device++;
578 		return 0;
579 	}
580 
581 	return eeh_dev_check_failure(edev);
582 }
583 EXPORT_SYMBOL(eeh_check_failure);
584 
585 
586 /**
587  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
588  * @pe: EEH PE
589  *
590  * This routine should be called to reenable frozen MMIO or DMA
591  * so that it would work correctly again. It's useful while doing
592  * recovery or log collection on the indicated device.
593  */
594 int eeh_pci_enable(struct eeh_pe *pe, int function)
595 {
596 	int active_flag, rc;
597 
598 	/*
599 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
600 	 * Also, it's pointless to enable them on unfrozen PE. So
601 	 * we have to check before enabling IO or DMA.
602 	 */
603 	switch (function) {
604 	case EEH_OPT_THAW_MMIO:
605 		active_flag = EEH_STATE_MMIO_ACTIVE;
606 		break;
607 	case EEH_OPT_THAW_DMA:
608 		active_flag = EEH_STATE_DMA_ACTIVE;
609 		break;
610 	case EEH_OPT_DISABLE:
611 	case EEH_OPT_ENABLE:
612 	case EEH_OPT_FREEZE_PE:
613 		active_flag = 0;
614 		break;
615 	default:
616 		pr_warn("%s: Invalid function %d\n",
617 			__func__, function);
618 		return -EINVAL;
619 	}
620 
621 	/*
622 	 * Check if IO or DMA has been enabled before
623 	 * enabling them.
624 	 */
625 	if (active_flag) {
626 		rc = eeh_ops->get_state(pe, NULL);
627 		if (rc < 0)
628 			return rc;
629 
630 		/* Needn't enable it at all */
631 		if (rc == EEH_STATE_NOT_SUPPORT)
632 			return 0;
633 
634 		/* It's already enabled */
635 		if (rc & active_flag)
636 			return 0;
637 	}
638 
639 
640 	/* Issue the request */
641 	rc = eeh_ops->set_option(pe, function);
642 	if (rc)
643 		pr_warn("%s: Unexpected state change %d on "
644 			"PHB#%d-PE#%x, err=%d\n",
645 			__func__, function, pe->phb->global_number,
646 			pe->addr, rc);
647 
648 	/* Check if the request is finished successfully */
649 	if (active_flag) {
650 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
651 		if (rc <= 0)
652 			return rc;
653 
654 		if (rc & active_flag)
655 			return 0;
656 
657 		return -EIO;
658 	}
659 
660 	return rc;
661 }
662 
663 /**
664  * pcibios_set_pcie_slot_reset - Set PCI-E reset state
665  * @dev: pci device struct
666  * @state: reset state to enter
667  *
668  * Return value:
669  * 	0 if success
670  */
671 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
672 {
673 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
674 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
675 
676 	if (!pe) {
677 		pr_err("%s: No PE found on PCI device %s\n",
678 			__func__, pci_name(dev));
679 		return -EINVAL;
680 	}
681 
682 	switch (state) {
683 	case pcie_deassert_reset:
684 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
685 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
686 		break;
687 	case pcie_hot_reset:
688 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
689 		eeh_ops->reset(pe, EEH_RESET_HOT);
690 		break;
691 	case pcie_warm_reset:
692 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
693 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
694 		break;
695 	default:
696 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
697 		return -EINVAL;
698 	};
699 
700 	return 0;
701 }
702 
703 /**
704  * eeh_set_pe_freset - Check the required reset for the indicated device
705  * @data: EEH device
706  * @flag: return value
707  *
708  * Each device might have its preferred reset type: fundamental or
709  * hot reset. The routine is used to collected the information for
710  * the indicated device and its children so that the bunch of the
711  * devices could be reset properly.
712  */
713 static void *eeh_set_dev_freset(void *data, void *flag)
714 {
715 	struct pci_dev *dev;
716 	unsigned int *freset = (unsigned int *)flag;
717 	struct eeh_dev *edev = (struct eeh_dev *)data;
718 
719 	dev = eeh_dev_to_pci_dev(edev);
720 	if (dev)
721 		*freset |= dev->needs_freset;
722 
723 	return NULL;
724 }
725 
726 /**
727  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
728  * @pe: EEH PE
729  *
730  * Assert the PCI #RST line for 1/4 second.
731  */
732 static void eeh_reset_pe_once(struct eeh_pe *pe)
733 {
734 	unsigned int freset = 0;
735 
736 	/* Determine type of EEH reset required for
737 	 * Partitionable Endpoint, a hot-reset (1)
738 	 * or a fundamental reset (3).
739 	 * A fundamental reset required by any device under
740 	 * Partitionable Endpoint trumps hot-reset.
741 	 */
742 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
743 
744 	if (freset)
745 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
746 	else
747 		eeh_ops->reset(pe, EEH_RESET_HOT);
748 
749 	eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
750 }
751 
752 /**
753  * eeh_reset_pe - Reset the indicated PE
754  * @pe: EEH PE
755  *
756  * This routine should be called to reset indicated device, including
757  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
758  * might be involved as well.
759  */
760 int eeh_reset_pe(struct eeh_pe *pe)
761 {
762 	int flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
763 	int i, state, ret;
764 
765 	/* Mark as reset and block config space */
766 	eeh_pe_state_mark(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
767 
768 	/* Take three shots at resetting the bus */
769 	for (i = 0; i < 3; i++) {
770 		eeh_reset_pe_once(pe);
771 
772 		/*
773 		 * EEH_PE_ISOLATED is expected to be removed after
774 		 * BAR restore.
775 		 */
776 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
777 		if ((state & flags) == flags) {
778 			ret = 0;
779 			goto out;
780 		}
781 
782 		if (state < 0) {
783 			pr_warn("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
784 				__func__, pe->phb->global_number, pe->addr);
785 			ret = -ENOTRECOVERABLE;
786 			goto out;
787 		}
788 
789 		/* We might run out of credits */
790 		ret = -EIO;
791 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
792 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
793 	}
794 
795 out:
796 	eeh_pe_state_clear(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
797 	return ret;
798 }
799 
800 /**
801  * eeh_save_bars - Save device bars
802  * @edev: PCI device associated EEH device
803  *
804  * Save the values of the device bars. Unlike the restore
805  * routine, this routine is *not* recursive. This is because
806  * PCI devices are added individually; but, for the restore,
807  * an entire slot is reset at a time.
808  */
809 void eeh_save_bars(struct eeh_dev *edev)
810 {
811 	int i;
812 	struct device_node *dn;
813 
814 	if (!edev)
815 		return;
816 	dn = eeh_dev_to_of_node(edev);
817 
818 	for (i = 0; i < 16; i++)
819 		eeh_ops->read_config(dn, i * 4, 4, &edev->config_space[i]);
820 
821 	/*
822 	 * For PCI bridges including root port, we need enable bus
823 	 * master explicitly. Otherwise, it can't fetch IODA table
824 	 * entries correctly. So we cache the bit in advance so that
825 	 * we can restore it after reset, either PHB range or PE range.
826 	 */
827 	if (edev->mode & EEH_DEV_BRIDGE)
828 		edev->config_space[1] |= PCI_COMMAND_MASTER;
829 }
830 
831 /**
832  * eeh_ops_register - Register platform dependent EEH operations
833  * @ops: platform dependent EEH operations
834  *
835  * Register the platform dependent EEH operation callback
836  * functions. The platform should call this function before
837  * any other EEH operations.
838  */
839 int __init eeh_ops_register(struct eeh_ops *ops)
840 {
841 	if (!ops->name) {
842 		pr_warn("%s: Invalid EEH ops name for %p\n",
843 			__func__, ops);
844 		return -EINVAL;
845 	}
846 
847 	if (eeh_ops && eeh_ops != ops) {
848 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
849 			__func__, eeh_ops->name, ops->name);
850 		return -EEXIST;
851 	}
852 
853 	eeh_ops = ops;
854 
855 	return 0;
856 }
857 
858 /**
859  * eeh_ops_unregister - Unreigster platform dependent EEH operations
860  * @name: name of EEH platform operations
861  *
862  * Unregister the platform dependent EEH operation callback
863  * functions.
864  */
865 int __exit eeh_ops_unregister(const char *name)
866 {
867 	if (!name || !strlen(name)) {
868 		pr_warn("%s: Invalid EEH ops name\n",
869 			__func__);
870 		return -EINVAL;
871 	}
872 
873 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
874 		eeh_ops = NULL;
875 		return 0;
876 	}
877 
878 	return -EEXIST;
879 }
880 
881 static int eeh_reboot_notifier(struct notifier_block *nb,
882 			       unsigned long action, void *unused)
883 {
884 	eeh_clear_flag(EEH_ENABLED);
885 	return NOTIFY_DONE;
886 }
887 
888 static struct notifier_block eeh_reboot_nb = {
889 	.notifier_call = eeh_reboot_notifier,
890 };
891 
892 /**
893  * eeh_init - EEH initialization
894  *
895  * Initialize EEH by trying to enable it for all of the adapters in the system.
896  * As a side effect we can determine here if eeh is supported at all.
897  * Note that we leave EEH on so failed config cycles won't cause a machine
898  * check.  If a user turns off EEH for a particular adapter they are really
899  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
900  * grant access to a slot if EEH isn't enabled, and so we always enable
901  * EEH for all slots/all devices.
902  *
903  * The eeh-force-off option disables EEH checking globally, for all slots.
904  * Even if force-off is set, the EEH hardware is still enabled, so that
905  * newer systems can boot.
906  */
907 int eeh_init(void)
908 {
909 	struct pci_controller *hose, *tmp;
910 	struct device_node *phb;
911 	static int cnt = 0;
912 	int ret = 0;
913 
914 	/*
915 	 * We have to delay the initialization on PowerNV after
916 	 * the PCI hierarchy tree has been built because the PEs
917 	 * are figured out based on PCI devices instead of device
918 	 * tree nodes
919 	 */
920 	if (machine_is(powernv) && cnt++ <= 0)
921 		return ret;
922 
923 	/* Register reboot notifier */
924 	ret = register_reboot_notifier(&eeh_reboot_nb);
925 	if (ret) {
926 		pr_warn("%s: Failed to register notifier (%d)\n",
927 			__func__, ret);
928 		return ret;
929 	}
930 
931 	/* call platform initialization function */
932 	if (!eeh_ops) {
933 		pr_warn("%s: Platform EEH operation not found\n",
934 			__func__);
935 		return -EEXIST;
936 	} else if ((ret = eeh_ops->init()))
937 		return ret;
938 
939 	/* Initialize EEH event */
940 	ret = eeh_event_init();
941 	if (ret)
942 		return ret;
943 
944 	/* Enable EEH for all adapters */
945 	if (eeh_has_flag(EEH_PROBE_MODE_DEVTREE)) {
946 		list_for_each_entry_safe(hose, tmp,
947 			&hose_list, list_node) {
948 			phb = hose->dn;
949 			traverse_pci_devices(phb, eeh_ops->of_probe, NULL);
950 		}
951 	} else if (eeh_has_flag(EEH_PROBE_MODE_DEV)) {
952 		list_for_each_entry_safe(hose, tmp,
953 			&hose_list, list_node)
954 			pci_walk_bus(hose->bus, eeh_ops->dev_probe, NULL);
955 	} else {
956 		pr_warn("%s: Invalid probe mode %x",
957 			__func__, eeh_subsystem_flags);
958 		return -EINVAL;
959 	}
960 
961 	/*
962 	 * Call platform post-initialization. Actually, It's good chance
963 	 * to inform platform that EEH is ready to supply service if the
964 	 * I/O cache stuff has been built up.
965 	 */
966 	if (eeh_ops->post_init) {
967 		ret = eeh_ops->post_init();
968 		if (ret)
969 			return ret;
970 	}
971 
972 	if (eeh_enabled())
973 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
974 	else
975 		pr_warn("EEH: No capable adapters found\n");
976 
977 	return ret;
978 }
979 
980 core_initcall_sync(eeh_init);
981 
982 /**
983  * eeh_add_device_early - Enable EEH for the indicated device_node
984  * @dn: device node for which to set up EEH
985  *
986  * This routine must be used to perform EEH initialization for PCI
987  * devices that were added after system boot (e.g. hotplug, dlpar).
988  * This routine must be called before any i/o is performed to the
989  * adapter (inluding any config-space i/o).
990  * Whether this actually enables EEH or not for this device depends
991  * on the CEC architecture, type of the device, on earlier boot
992  * command-line arguments & etc.
993  */
994 void eeh_add_device_early(struct device_node *dn)
995 {
996 	struct pci_controller *phb;
997 
998 	/*
999 	 * If we're doing EEH probe based on PCI device, we
1000 	 * would delay the probe until late stage because
1001 	 * the PCI device isn't available this moment.
1002 	 */
1003 	if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1004 		return;
1005 
1006 	if (!of_node_to_eeh_dev(dn))
1007 		return;
1008 	phb = of_node_to_eeh_dev(dn)->phb;
1009 
1010 	/* USB Bus children of PCI devices will not have BUID's */
1011 	if (NULL == phb || 0 == phb->buid)
1012 		return;
1013 
1014 	eeh_ops->of_probe(dn, NULL);
1015 }
1016 
1017 /**
1018  * eeh_add_device_tree_early - Enable EEH for the indicated device
1019  * @dn: device node
1020  *
1021  * This routine must be used to perform EEH initialization for the
1022  * indicated PCI device that was added after system boot (e.g.
1023  * hotplug, dlpar).
1024  */
1025 void eeh_add_device_tree_early(struct device_node *dn)
1026 {
1027 	struct device_node *sib;
1028 
1029 	for_each_child_of_node(dn, sib)
1030 		eeh_add_device_tree_early(sib);
1031 	eeh_add_device_early(dn);
1032 }
1033 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1034 
1035 /**
1036  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1037  * @dev: pci device for which to set up EEH
1038  *
1039  * This routine must be used to complete EEH initialization for PCI
1040  * devices that were added after system boot (e.g. hotplug, dlpar).
1041  */
1042 void eeh_add_device_late(struct pci_dev *dev)
1043 {
1044 	struct device_node *dn;
1045 	struct eeh_dev *edev;
1046 
1047 	if (!dev || !eeh_enabled())
1048 		return;
1049 
1050 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1051 
1052 	dn = pci_device_to_OF_node(dev);
1053 	edev = of_node_to_eeh_dev(dn);
1054 	if (edev->pdev == dev) {
1055 		pr_debug("EEH: Already referenced !\n");
1056 		return;
1057 	}
1058 
1059 	/*
1060 	 * The EEH cache might not be removed correctly because of
1061 	 * unbalanced kref to the device during unplug time, which
1062 	 * relies on pcibios_release_device(). So we have to remove
1063 	 * that here explicitly.
1064 	 */
1065 	if (edev->pdev) {
1066 		eeh_rmv_from_parent_pe(edev);
1067 		eeh_addr_cache_rmv_dev(edev->pdev);
1068 		eeh_sysfs_remove_device(edev->pdev);
1069 		edev->mode &= ~EEH_DEV_SYSFS;
1070 
1071 		/*
1072 		 * We definitely should have the PCI device removed
1073 		 * though it wasn't correctly. So we needn't call
1074 		 * into error handler afterwards.
1075 		 */
1076 		edev->mode |= EEH_DEV_NO_HANDLER;
1077 
1078 		edev->pdev = NULL;
1079 		dev->dev.archdata.edev = NULL;
1080 	}
1081 
1082 	edev->pdev = dev;
1083 	dev->dev.archdata.edev = edev;
1084 
1085 	/*
1086 	 * We have to do the EEH probe here because the PCI device
1087 	 * hasn't been created yet in the early stage.
1088 	 */
1089 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1090 		eeh_ops->dev_probe(dev, NULL);
1091 
1092 	eeh_addr_cache_insert_dev(dev);
1093 }
1094 
1095 /**
1096  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1097  * @bus: PCI bus
1098  *
1099  * This routine must be used to perform EEH initialization for PCI
1100  * devices which are attached to the indicated PCI bus. The PCI bus
1101  * is added after system boot through hotplug or dlpar.
1102  */
1103 void eeh_add_device_tree_late(struct pci_bus *bus)
1104 {
1105 	struct pci_dev *dev;
1106 
1107 	list_for_each_entry(dev, &bus->devices, bus_list) {
1108 		eeh_add_device_late(dev);
1109 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1110 			struct pci_bus *subbus = dev->subordinate;
1111 			if (subbus)
1112 				eeh_add_device_tree_late(subbus);
1113 		}
1114 	}
1115 }
1116 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1117 
1118 /**
1119  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1120  * @bus: PCI bus
1121  *
1122  * This routine must be used to add EEH sysfs files for PCI
1123  * devices which are attached to the indicated PCI bus. The PCI bus
1124  * is added after system boot through hotplug or dlpar.
1125  */
1126 void eeh_add_sysfs_files(struct pci_bus *bus)
1127 {
1128 	struct pci_dev *dev;
1129 
1130 	list_for_each_entry(dev, &bus->devices, bus_list) {
1131 		eeh_sysfs_add_device(dev);
1132 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1133 			struct pci_bus *subbus = dev->subordinate;
1134 			if (subbus)
1135 				eeh_add_sysfs_files(subbus);
1136 		}
1137 	}
1138 }
1139 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1140 
1141 /**
1142  * eeh_remove_device - Undo EEH setup for the indicated pci device
1143  * @dev: pci device to be removed
1144  *
1145  * This routine should be called when a device is removed from
1146  * a running system (e.g. by hotplug or dlpar).  It unregisters
1147  * the PCI device from the EEH subsystem.  I/O errors affecting
1148  * this device will no longer be detected after this call; thus,
1149  * i/o errors affecting this slot may leave this device unusable.
1150  */
1151 void eeh_remove_device(struct pci_dev *dev)
1152 {
1153 	struct eeh_dev *edev;
1154 
1155 	if (!dev || !eeh_enabled())
1156 		return;
1157 	edev = pci_dev_to_eeh_dev(dev);
1158 
1159 	/* Unregister the device with the EEH/PCI address search system */
1160 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1161 
1162 	if (!edev || !edev->pdev || !edev->pe) {
1163 		pr_debug("EEH: Not referenced !\n");
1164 		return;
1165 	}
1166 
1167 	/*
1168 	 * During the hotplug for EEH error recovery, we need the EEH
1169 	 * device attached to the parent PE in order for BAR restore
1170 	 * a bit later. So we keep it for BAR restore and remove it
1171 	 * from the parent PE during the BAR resotre.
1172 	 */
1173 	edev->pdev = NULL;
1174 	dev->dev.archdata.edev = NULL;
1175 	if (!(edev->pe->state & EEH_PE_KEEP))
1176 		eeh_rmv_from_parent_pe(edev);
1177 	else
1178 		edev->mode |= EEH_DEV_DISCONNECTED;
1179 
1180 	/*
1181 	 * We're removing from the PCI subsystem, that means
1182 	 * the PCI device driver can't support EEH or not
1183 	 * well. So we rely on hotplug completely to do recovery
1184 	 * for the specific PCI device.
1185 	 */
1186 	edev->mode |= EEH_DEV_NO_HANDLER;
1187 
1188 	eeh_addr_cache_rmv_dev(dev);
1189 	eeh_sysfs_remove_device(dev);
1190 	edev->mode &= ~EEH_DEV_SYSFS;
1191 }
1192 
1193 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1194 {
1195 	int ret;
1196 
1197 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1198 	if (ret) {
1199 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1200 			__func__, ret, pe->phb->global_number, pe->addr);
1201 		return ret;
1202 	}
1203 
1204 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1205 	if (ret) {
1206 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1207 			__func__, ret, pe->phb->global_number, pe->addr);
1208 		return ret;
1209 	}
1210 
1211 	/* Clear software isolated state */
1212 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1213 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1214 
1215 	return ret;
1216 }
1217 
1218 
1219 static struct pci_device_id eeh_reset_ids[] = {
1220 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1221 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1222 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1223 	{ 0 }
1224 };
1225 
1226 static int eeh_pe_change_owner(struct eeh_pe *pe)
1227 {
1228 	struct eeh_dev *edev, *tmp;
1229 	struct pci_dev *pdev;
1230 	struct pci_device_id *id;
1231 	int flags, ret;
1232 
1233 	/* Check PE state */
1234 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1235 	ret = eeh_ops->get_state(pe, NULL);
1236 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1237 		return 0;
1238 
1239 	/* Unfrozen PE, nothing to do */
1240 	if ((ret & flags) == flags)
1241 		return 0;
1242 
1243 	/* Frozen PE, check if it needs PE level reset */
1244 	eeh_pe_for_each_dev(pe, edev, tmp) {
1245 		pdev = eeh_dev_to_pci_dev(edev);
1246 		if (!pdev)
1247 			continue;
1248 
1249 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1250 			if (id->vendor != PCI_ANY_ID &&
1251 			    id->vendor != pdev->vendor)
1252 				continue;
1253 			if (id->device != PCI_ANY_ID &&
1254 			    id->device != pdev->device)
1255 				continue;
1256 			if (id->subvendor != PCI_ANY_ID &&
1257 			    id->subvendor != pdev->subsystem_vendor)
1258 				continue;
1259 			if (id->subdevice != PCI_ANY_ID &&
1260 			    id->subdevice != pdev->subsystem_device)
1261 				continue;
1262 
1263 			goto reset;
1264 		}
1265 	}
1266 
1267 	return eeh_unfreeze_pe(pe, true);
1268 
1269 reset:
1270 	return eeh_pe_reset_and_recover(pe);
1271 }
1272 
1273 /**
1274  * eeh_dev_open - Increase count of pass through devices for PE
1275  * @pdev: PCI device
1276  *
1277  * Increase count of passed through devices for the indicated
1278  * PE. In the result, the EEH errors detected on the PE won't be
1279  * reported. The PE owner will be responsible for detection
1280  * and recovery.
1281  */
1282 int eeh_dev_open(struct pci_dev *pdev)
1283 {
1284 	struct eeh_dev *edev;
1285 	int ret = -ENODEV;
1286 
1287 	mutex_lock(&eeh_dev_mutex);
1288 
1289 	/* No PCI device ? */
1290 	if (!pdev)
1291 		goto out;
1292 
1293 	/* No EEH device or PE ? */
1294 	edev = pci_dev_to_eeh_dev(pdev);
1295 	if (!edev || !edev->pe)
1296 		goto out;
1297 
1298 	/*
1299 	 * The PE might have been put into frozen state, but we
1300 	 * didn't detect that yet. The passed through PCI devices
1301 	 * in frozen PE won't work properly. Clear the frozen state
1302 	 * in advance.
1303 	 */
1304 	ret = eeh_pe_change_owner(edev->pe);
1305 	if (ret)
1306 		goto out;
1307 
1308 	/* Increase PE's pass through count */
1309 	atomic_inc(&edev->pe->pass_dev_cnt);
1310 	mutex_unlock(&eeh_dev_mutex);
1311 
1312 	return 0;
1313 out:
1314 	mutex_unlock(&eeh_dev_mutex);
1315 	return ret;
1316 }
1317 EXPORT_SYMBOL_GPL(eeh_dev_open);
1318 
1319 /**
1320  * eeh_dev_release - Decrease count of pass through devices for PE
1321  * @pdev: PCI device
1322  *
1323  * Decrease count of pass through devices for the indicated PE. If
1324  * there is no passed through device in PE, the EEH errors detected
1325  * on the PE will be reported and handled as usual.
1326  */
1327 void eeh_dev_release(struct pci_dev *pdev)
1328 {
1329 	struct eeh_dev *edev;
1330 
1331 	mutex_lock(&eeh_dev_mutex);
1332 
1333 	/* No PCI device ? */
1334 	if (!pdev)
1335 		goto out;
1336 
1337 	/* No EEH device ? */
1338 	edev = pci_dev_to_eeh_dev(pdev);
1339 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1340 		goto out;
1341 
1342 	/* Decrease PE's pass through count */
1343 	atomic_dec(&edev->pe->pass_dev_cnt);
1344 	WARN_ON(atomic_read(&edev->pe->pass_dev_cnt) < 0);
1345 	eeh_pe_change_owner(edev->pe);
1346 out:
1347 	mutex_unlock(&eeh_dev_mutex);
1348 }
1349 EXPORT_SYMBOL(eeh_dev_release);
1350 
1351 #ifdef CONFIG_IOMMU_API
1352 
1353 static int dev_has_iommu_table(struct device *dev, void *data)
1354 {
1355 	struct pci_dev *pdev = to_pci_dev(dev);
1356 	struct pci_dev **ppdev = data;
1357 	struct iommu_table *tbl;
1358 
1359 	if (!dev)
1360 		return 0;
1361 
1362 	tbl = get_iommu_table_base(dev);
1363 	if (tbl && tbl->it_group) {
1364 		*ppdev = pdev;
1365 		return 1;
1366 	}
1367 
1368 	return 0;
1369 }
1370 
1371 /**
1372  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1373  * @group: IOMMU group
1374  *
1375  * The routine is called to convert IOMMU group to EEH PE.
1376  */
1377 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1378 {
1379 	struct pci_dev *pdev = NULL;
1380 	struct eeh_dev *edev;
1381 	int ret;
1382 
1383 	/* No IOMMU group ? */
1384 	if (!group)
1385 		return NULL;
1386 
1387 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1388 	if (!ret || !pdev)
1389 		return NULL;
1390 
1391 	/* No EEH device or PE ? */
1392 	edev = pci_dev_to_eeh_dev(pdev);
1393 	if (!edev || !edev->pe)
1394 		return NULL;
1395 
1396 	return edev->pe;
1397 }
1398 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1399 
1400 #endif /* CONFIG_IOMMU_API */
1401 
1402 /**
1403  * eeh_pe_set_option - Set options for the indicated PE
1404  * @pe: EEH PE
1405  * @option: requested option
1406  *
1407  * The routine is called to enable or disable EEH functionality
1408  * on the indicated PE, to enable IO or DMA for the frozen PE.
1409  */
1410 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1411 {
1412 	int ret = 0;
1413 
1414 	/* Invalid PE ? */
1415 	if (!pe)
1416 		return -ENODEV;
1417 
1418 	/*
1419 	 * EEH functionality could possibly be disabled, just
1420 	 * return error for the case. And the EEH functinality
1421 	 * isn't expected to be disabled on one specific PE.
1422 	 */
1423 	switch (option) {
1424 	case EEH_OPT_ENABLE:
1425 		if (eeh_enabled()) {
1426 			ret = eeh_pe_change_owner(pe);
1427 			break;
1428 		}
1429 		ret = -EIO;
1430 		break;
1431 	case EEH_OPT_DISABLE:
1432 		break;
1433 	case EEH_OPT_THAW_MMIO:
1434 	case EEH_OPT_THAW_DMA:
1435 		if (!eeh_ops || !eeh_ops->set_option) {
1436 			ret = -ENOENT;
1437 			break;
1438 		}
1439 
1440 		ret = eeh_pci_enable(pe, option);
1441 		break;
1442 	default:
1443 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1444 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1445 		ret = -EINVAL;
1446 	}
1447 
1448 	return ret;
1449 }
1450 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1451 
1452 /**
1453  * eeh_pe_get_state - Retrieve PE's state
1454  * @pe: EEH PE
1455  *
1456  * Retrieve the PE's state, which includes 3 aspects: enabled
1457  * DMA, enabled IO and asserted reset.
1458  */
1459 int eeh_pe_get_state(struct eeh_pe *pe)
1460 {
1461 	int result, ret = 0;
1462 	bool rst_active, dma_en, mmio_en;
1463 
1464 	/* Existing PE ? */
1465 	if (!pe)
1466 		return -ENODEV;
1467 
1468 	if (!eeh_ops || !eeh_ops->get_state)
1469 		return -ENOENT;
1470 
1471 	result = eeh_ops->get_state(pe, NULL);
1472 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1473 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1474 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1475 
1476 	if (rst_active)
1477 		ret = EEH_PE_STATE_RESET;
1478 	else if (dma_en && mmio_en)
1479 		ret = EEH_PE_STATE_NORMAL;
1480 	else if (!dma_en && !mmio_en)
1481 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1482 	else if (!dma_en && mmio_en)
1483 		ret = EEH_PE_STATE_STOPPED_DMA;
1484 	else
1485 		ret = EEH_PE_STATE_UNAVAIL;
1486 
1487 	return ret;
1488 }
1489 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1490 
1491 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1492 {
1493 	struct eeh_dev *edev, *tmp;
1494 	struct pci_dev *pdev;
1495 	int ret = 0;
1496 
1497 	/* Restore config space */
1498 	eeh_pe_restore_bars(pe);
1499 
1500 	/*
1501 	 * Reenable PCI devices as the devices passed
1502 	 * through are always enabled before the reset.
1503 	 */
1504 	eeh_pe_for_each_dev(pe, edev, tmp) {
1505 		pdev = eeh_dev_to_pci_dev(edev);
1506 		if (!pdev)
1507 			continue;
1508 
1509 		ret = pci_reenable_device(pdev);
1510 		if (ret) {
1511 			pr_warn("%s: Failure %d reenabling %s\n",
1512 				__func__, ret, pci_name(pdev));
1513 			return ret;
1514 		}
1515 	}
1516 
1517 	/* The PE is still in frozen state */
1518 	return eeh_unfreeze_pe(pe, true);
1519 }
1520 
1521 /**
1522  * eeh_pe_reset - Issue PE reset according to specified type
1523  * @pe: EEH PE
1524  * @option: reset type
1525  *
1526  * The routine is called to reset the specified PE with the
1527  * indicated type, either fundamental reset or hot reset.
1528  * PE reset is the most important part for error recovery.
1529  */
1530 int eeh_pe_reset(struct eeh_pe *pe, int option)
1531 {
1532 	int ret = 0;
1533 
1534 	/* Invalid PE ? */
1535 	if (!pe)
1536 		return -ENODEV;
1537 
1538 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1539 		return -ENOENT;
1540 
1541 	switch (option) {
1542 	case EEH_RESET_DEACTIVATE:
1543 		ret = eeh_ops->reset(pe, option);
1544 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1545 		if (ret)
1546 			break;
1547 
1548 		ret = eeh_pe_reenable_devices(pe);
1549 		break;
1550 	case EEH_RESET_HOT:
1551 	case EEH_RESET_FUNDAMENTAL:
1552 		/*
1553 		 * Proactively freeze the PE to drop all MMIO access
1554 		 * during reset, which should be banned as it's always
1555 		 * cause recursive EEH error.
1556 		 */
1557 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1558 
1559 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1560 		ret = eeh_ops->reset(pe, option);
1561 		break;
1562 	default:
1563 		pr_debug("%s: Unsupported option %d\n",
1564 			__func__, option);
1565 		ret = -EINVAL;
1566 	}
1567 
1568 	return ret;
1569 }
1570 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1571 
1572 /**
1573  * eeh_pe_configure - Configure PCI bridges after PE reset
1574  * @pe: EEH PE
1575  *
1576  * The routine is called to restore the PCI config space for
1577  * those PCI devices, especially PCI bridges affected by PE
1578  * reset issued previously.
1579  */
1580 int eeh_pe_configure(struct eeh_pe *pe)
1581 {
1582 	int ret = 0;
1583 
1584 	/* Invalid PE ? */
1585 	if (!pe)
1586 		return -ENODEV;
1587 
1588 	return ret;
1589 }
1590 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1591 
1592 static int proc_eeh_show(struct seq_file *m, void *v)
1593 {
1594 	if (!eeh_enabled()) {
1595 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1596 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1597 	} else {
1598 		seq_printf(m, "EEH Subsystem is enabled\n");
1599 		seq_printf(m,
1600 				"no device=%llu\n"
1601 				"no device node=%llu\n"
1602 				"no config address=%llu\n"
1603 				"check not wanted=%llu\n"
1604 				"eeh_total_mmio_ffs=%llu\n"
1605 				"eeh_false_positives=%llu\n"
1606 				"eeh_slot_resets=%llu\n",
1607 				eeh_stats.no_device,
1608 				eeh_stats.no_dn,
1609 				eeh_stats.no_cfg_addr,
1610 				eeh_stats.ignored_check,
1611 				eeh_stats.total_mmio_ffs,
1612 				eeh_stats.false_positives,
1613 				eeh_stats.slot_resets);
1614 	}
1615 
1616 	return 0;
1617 }
1618 
1619 static int proc_eeh_open(struct inode *inode, struct file *file)
1620 {
1621 	return single_open(file, proc_eeh_show, NULL);
1622 }
1623 
1624 static const struct file_operations proc_eeh_operations = {
1625 	.open      = proc_eeh_open,
1626 	.read      = seq_read,
1627 	.llseek    = seq_lseek,
1628 	.release   = single_release,
1629 };
1630 
1631 #ifdef CONFIG_DEBUG_FS
1632 static int eeh_enable_dbgfs_set(void *data, u64 val)
1633 {
1634 	if (val)
1635 		eeh_clear_flag(EEH_FORCE_DISABLED);
1636 	else
1637 		eeh_add_flag(EEH_FORCE_DISABLED);
1638 
1639 	/* Notify the backend */
1640 	if (eeh_ops->post_init)
1641 		eeh_ops->post_init();
1642 
1643 	return 0;
1644 }
1645 
1646 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1647 {
1648 	if (eeh_enabled())
1649 		*val = 0x1ul;
1650 	else
1651 		*val = 0x0ul;
1652 	return 0;
1653 }
1654 
1655 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1656 			eeh_enable_dbgfs_set, "0x%llx\n");
1657 #endif
1658 
1659 static int __init eeh_init_proc(void)
1660 {
1661 	if (machine_is(pseries) || machine_is(powernv)) {
1662 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1663 #ifdef CONFIG_DEBUG_FS
1664 		debugfs_create_file("eeh_enable", 0600,
1665                                     powerpc_debugfs_root, NULL,
1666                                     &eeh_enable_dbgfs_ops);
1667 #endif
1668 	}
1669 
1670 	return 0;
1671 }
1672 __initcall(eeh_init_proc);
1673