xref: /linux/arch/powerpc/kernel/eeh.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 static int __init eeh_setup(char *str)
148 {
149 	if (!strcmp(str, "off"))
150 		eeh_add_flag(EEH_FORCE_DISABLED);
151 	else if (!strcmp(str, "early_log"))
152 		eeh_add_flag(EEH_EARLY_DUMP_LOG);
153 
154 	return 1;
155 }
156 __setup("eeh=", eeh_setup);
157 
158 /*
159  * This routine captures assorted PCI configuration space data
160  * for the indicated PCI device, and puts them into a buffer
161  * for RTAS error logging.
162  */
163 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
164 {
165 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
166 	u32 cfg;
167 	int cap, i;
168 	int n = 0, l = 0;
169 	char buffer[128];
170 
171 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x:%01x\n",
172 		       edev->phb->global_number, pdn->busno,
173 		       PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
174 	pr_warn("EEH: of node=%04x:%02x:%02x:%01x\n",
175 		edev->phb->global_number, pdn->busno,
176 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
177 
178 	eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
179 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
180 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
181 
182 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
183 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
184 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
185 
186 	/* Gather bridge-specific registers */
187 	if (edev->mode & EEH_DEV_BRIDGE) {
188 		eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
189 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
190 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
191 
192 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
193 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
194 		pr_warn("EEH: Bridge control: %04x\n", cfg);
195 	}
196 
197 	/* Dump out the PCI-X command and status regs */
198 	cap = edev->pcix_cap;
199 	if (cap) {
200 		eeh_ops->read_config(pdn, cap, 4, &cfg);
201 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
202 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
203 
204 		eeh_ops->read_config(pdn, cap+4, 4, &cfg);
205 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
206 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
207 	}
208 
209 	/* If PCI-E capable, dump PCI-E cap 10 */
210 	cap = edev->pcie_cap;
211 	if (cap) {
212 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
213 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
214 
215 		for (i=0; i<=8; i++) {
216 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
217 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
218 
219 			if ((i % 4) == 0) {
220 				if (i != 0)
221 					pr_warn("%s\n", buffer);
222 
223 				l = scnprintf(buffer, sizeof(buffer),
224 					      "EEH: PCI-E %02x: %08x ",
225 					      4*i, cfg);
226 			} else {
227 				l += scnprintf(buffer+l, sizeof(buffer)-l,
228 					       "%08x ", cfg);
229 			}
230 
231 		}
232 
233 		pr_warn("%s\n", buffer);
234 	}
235 
236 	/* If AER capable, dump it */
237 	cap = edev->aer_cap;
238 	if (cap) {
239 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
240 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
241 
242 		for (i=0; i<=13; i++) {
243 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
244 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
245 
246 			if ((i % 4) == 0) {
247 				if (i != 0)
248 					pr_warn("%s\n", buffer);
249 
250 				l = scnprintf(buffer, sizeof(buffer),
251 					      "EEH: PCI-E AER %02x: %08x ",
252 					      4*i, cfg);
253 			} else {
254 				l += scnprintf(buffer+l, sizeof(buffer)-l,
255 					       "%08x ", cfg);
256 			}
257 		}
258 
259 		pr_warn("%s\n", buffer);
260 	}
261 
262 	return n;
263 }
264 
265 static void *eeh_dump_pe_log(void *data, void *flag)
266 {
267 	struct eeh_pe *pe = data;
268 	struct eeh_dev *edev, *tmp;
269 	size_t *plen = flag;
270 
271 	/* If the PE's config space is blocked, 0xFF's will be
272 	 * returned. It's pointless to collect the log in this
273 	 * case.
274 	 */
275 	if (pe->state & EEH_PE_CFG_BLOCKED)
276 		return NULL;
277 
278 	eeh_pe_for_each_dev(pe, edev, tmp)
279 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
280 					  EEH_PCI_REGS_LOG_LEN - *plen);
281 
282 	return NULL;
283 }
284 
285 /**
286  * eeh_slot_error_detail - Generate combined log including driver log and error log
287  * @pe: EEH PE
288  * @severity: temporary or permanent error log
289  *
290  * This routine should be called to generate the combined log, which
291  * is comprised of driver log and error log. The driver log is figured
292  * out from the config space of the corresponding PCI device, while
293  * the error log is fetched through platform dependent function call.
294  */
295 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
296 {
297 	size_t loglen = 0;
298 
299 	/*
300 	 * When the PHB is fenced or dead, it's pointless to collect
301 	 * the data from PCI config space because it should return
302 	 * 0xFF's. For ER, we still retrieve the data from the PCI
303 	 * config space.
304 	 *
305 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
306 	 * 0xFF's is always returned from PCI config space.
307 	 */
308 	if (!(pe->type & EEH_PE_PHB)) {
309 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG))
310 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
311 
312 		/*
313 		 * The config space of some PCI devices can't be accessed
314 		 * when their PEs are in frozen state. Otherwise, fenced
315 		 * PHB might be seen. Those PEs are identified with flag
316 		 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
317 		 * is set automatically when the PE is put to EEH_PE_ISOLATED.
318 		 *
319 		 * Restoring BARs possibly triggers PCI config access in
320 		 * (OPAL) firmware and then causes fenced PHB. If the
321 		 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
322 		 * pointless to restore BARs and dump config space.
323 		 */
324 		eeh_ops->configure_bridge(pe);
325 		if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
326 			eeh_pe_restore_bars(pe);
327 
328 			pci_regs_buf[0] = 0;
329 			eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
330 		}
331 	}
332 
333 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
334 }
335 
336 /**
337  * eeh_token_to_phys - Convert EEH address token to phys address
338  * @token: I/O token, should be address in the form 0xA....
339  *
340  * This routine should be called to convert virtual I/O address
341  * to physical one.
342  */
343 static inline unsigned long eeh_token_to_phys(unsigned long token)
344 {
345 	pte_t *ptep;
346 	unsigned long pa;
347 	int hugepage_shift;
348 
349 	/*
350 	 * We won't find hugepages here(this is iomem). Hence we are not
351 	 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
352 	 * page table free, because of init_mm.
353 	 */
354 	ptep = __find_linux_pte_or_hugepte(init_mm.pgd, token, &hugepage_shift);
355 	if (!ptep)
356 		return token;
357 	WARN_ON(hugepage_shift);
358 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
359 
360 	return pa | (token & (PAGE_SIZE-1));
361 }
362 
363 /*
364  * On PowerNV platform, we might already have fenced PHB there.
365  * For that case, it's meaningless to recover frozen PE. Intead,
366  * We have to handle fenced PHB firstly.
367  */
368 static int eeh_phb_check_failure(struct eeh_pe *pe)
369 {
370 	struct eeh_pe *phb_pe;
371 	unsigned long flags;
372 	int ret;
373 
374 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
375 		return -EPERM;
376 
377 	/* Find the PHB PE */
378 	phb_pe = eeh_phb_pe_get(pe->phb);
379 	if (!phb_pe) {
380 		pr_warn("%s Can't find PE for PHB#%d\n",
381 			__func__, pe->phb->global_number);
382 		return -EEXIST;
383 	}
384 
385 	/* If the PHB has been in problematic state */
386 	eeh_serialize_lock(&flags);
387 	if (phb_pe->state & EEH_PE_ISOLATED) {
388 		ret = 0;
389 		goto out;
390 	}
391 
392 	/* Check PHB state */
393 	ret = eeh_ops->get_state(phb_pe, NULL);
394 	if ((ret < 0) ||
395 	    (ret == EEH_STATE_NOT_SUPPORT) ||
396 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
397 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
398 		ret = 0;
399 		goto out;
400 	}
401 
402 	/* Isolate the PHB and send event */
403 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
404 	eeh_serialize_unlock(flags);
405 
406 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
407 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
408 	dump_stack();
409 	eeh_send_failure_event(phb_pe);
410 
411 	return 1;
412 out:
413 	eeh_serialize_unlock(flags);
414 	return ret;
415 }
416 
417 /**
418  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
419  * @edev: eeh device
420  *
421  * Check for an EEH failure for the given device node.  Call this
422  * routine if the result of a read was all 0xff's and you want to
423  * find out if this is due to an EEH slot freeze.  This routine
424  * will query firmware for the EEH status.
425  *
426  * Returns 0 if there has not been an EEH error; otherwise returns
427  * a non-zero value and queues up a slot isolation event notification.
428  *
429  * It is safe to call this routine in an interrupt context.
430  */
431 int eeh_dev_check_failure(struct eeh_dev *edev)
432 {
433 	int ret;
434 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
435 	unsigned long flags;
436 	struct pci_dn *pdn;
437 	struct pci_dev *dev;
438 	struct eeh_pe *pe, *parent_pe, *phb_pe;
439 	int rc = 0;
440 	const char *location = NULL;
441 
442 	eeh_stats.total_mmio_ffs++;
443 
444 	if (!eeh_enabled())
445 		return 0;
446 
447 	if (!edev) {
448 		eeh_stats.no_dn++;
449 		return 0;
450 	}
451 	dev = eeh_dev_to_pci_dev(edev);
452 	pe = eeh_dev_to_pe(edev);
453 
454 	/* Access to IO BARs might get this far and still not want checking. */
455 	if (!pe) {
456 		eeh_stats.ignored_check++;
457 		pr_debug("EEH: Ignored check for %s\n",
458 			eeh_pci_name(dev));
459 		return 0;
460 	}
461 
462 	if (!pe->addr && !pe->config_addr) {
463 		eeh_stats.no_cfg_addr++;
464 		return 0;
465 	}
466 
467 	/*
468 	 * On PowerNV platform, we might already have fenced PHB
469 	 * there and we need take care of that firstly.
470 	 */
471 	ret = eeh_phb_check_failure(pe);
472 	if (ret > 0)
473 		return ret;
474 
475 	/*
476 	 * If the PE isn't owned by us, we shouldn't check the
477 	 * state. Instead, let the owner handle it if the PE has
478 	 * been frozen.
479 	 */
480 	if (eeh_pe_passed(pe))
481 		return 0;
482 
483 	/* If we already have a pending isolation event for this
484 	 * slot, we know it's bad already, we don't need to check.
485 	 * Do this checking under a lock; as multiple PCI devices
486 	 * in one slot might report errors simultaneously, and we
487 	 * only want one error recovery routine running.
488 	 */
489 	eeh_serialize_lock(&flags);
490 	rc = 1;
491 	if (pe->state & EEH_PE_ISOLATED) {
492 		pe->check_count++;
493 		if (pe->check_count % EEH_MAX_FAILS == 0) {
494 			pdn = eeh_dev_to_pdn(edev);
495 			if (pdn->node)
496 				location = of_get_property(pdn->node, "ibm,loc-code", NULL);
497 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
498 				"location=%s driver=%s pci addr=%s\n",
499 				pe->check_count,
500 				location ? location : "unknown",
501 				eeh_driver_name(dev), eeh_pci_name(dev));
502 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
503 				eeh_driver_name(dev));
504 			dump_stack();
505 		}
506 		goto dn_unlock;
507 	}
508 
509 	/*
510 	 * Now test for an EEH failure.  This is VERY expensive.
511 	 * Note that the eeh_config_addr may be a parent device
512 	 * in the case of a device behind a bridge, or it may be
513 	 * function zero of a multi-function device.
514 	 * In any case they must share a common PHB.
515 	 */
516 	ret = eeh_ops->get_state(pe, NULL);
517 
518 	/* Note that config-io to empty slots may fail;
519 	 * they are empty when they don't have children.
520 	 * We will punt with the following conditions: Failure to get
521 	 * PE's state, EEH not support and Permanently unavailable
522 	 * state, PE is in good state.
523 	 */
524 	if ((ret < 0) ||
525 	    (ret == EEH_STATE_NOT_SUPPORT) ||
526 	    ((ret & active_flags) == active_flags)) {
527 		eeh_stats.false_positives++;
528 		pe->false_positives++;
529 		rc = 0;
530 		goto dn_unlock;
531 	}
532 
533 	/*
534 	 * It should be corner case that the parent PE has been
535 	 * put into frozen state as well. We should take care
536 	 * that at first.
537 	 */
538 	parent_pe = pe->parent;
539 	while (parent_pe) {
540 		/* Hit the ceiling ? */
541 		if (parent_pe->type & EEH_PE_PHB)
542 			break;
543 
544 		/* Frozen parent PE ? */
545 		ret = eeh_ops->get_state(parent_pe, NULL);
546 		if (ret > 0 &&
547 		    (ret & active_flags) != active_flags)
548 			pe = parent_pe;
549 
550 		/* Next parent level */
551 		parent_pe = parent_pe->parent;
552 	}
553 
554 	eeh_stats.slot_resets++;
555 
556 	/* Avoid repeated reports of this failure, including problems
557 	 * with other functions on this device, and functions under
558 	 * bridges.
559 	 */
560 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
561 	eeh_serialize_unlock(flags);
562 
563 	/* Most EEH events are due to device driver bugs.  Having
564 	 * a stack trace will help the device-driver authors figure
565 	 * out what happened.  So print that out.
566 	 */
567 	phb_pe = eeh_phb_pe_get(pe->phb);
568 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
569 	       pe->phb->global_number, pe->addr);
570 	pr_err("EEH: PE location: %s, PHB location: %s\n",
571 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
572 	dump_stack();
573 
574 	eeh_send_failure_event(pe);
575 
576 	return 1;
577 
578 dn_unlock:
579 	eeh_serialize_unlock(flags);
580 	return rc;
581 }
582 
583 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
584 
585 /**
586  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
587  * @token: I/O address
588  *
589  * Check for an EEH failure at the given I/O address. Call this
590  * routine if the result of a read was all 0xff's and you want to
591  * find out if this is due to an EEH slot freeze event. This routine
592  * will query firmware for the EEH status.
593  *
594  * Note this routine is safe to call in an interrupt context.
595  */
596 int eeh_check_failure(const volatile void __iomem *token)
597 {
598 	unsigned long addr;
599 	struct eeh_dev *edev;
600 
601 	/* Finding the phys addr + pci device; this is pretty quick. */
602 	addr = eeh_token_to_phys((unsigned long __force) token);
603 	edev = eeh_addr_cache_get_dev(addr);
604 	if (!edev) {
605 		eeh_stats.no_device++;
606 		return 0;
607 	}
608 
609 	return eeh_dev_check_failure(edev);
610 }
611 EXPORT_SYMBOL(eeh_check_failure);
612 
613 
614 /**
615  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
616  * @pe: EEH PE
617  *
618  * This routine should be called to reenable frozen MMIO or DMA
619  * so that it would work correctly again. It's useful while doing
620  * recovery or log collection on the indicated device.
621  */
622 int eeh_pci_enable(struct eeh_pe *pe, int function)
623 {
624 	int active_flag, rc;
625 
626 	/*
627 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
628 	 * Also, it's pointless to enable them on unfrozen PE. So
629 	 * we have to check before enabling IO or DMA.
630 	 */
631 	switch (function) {
632 	case EEH_OPT_THAW_MMIO:
633 		active_flag = EEH_STATE_MMIO_ACTIVE;
634 		break;
635 	case EEH_OPT_THAW_DMA:
636 		active_flag = EEH_STATE_DMA_ACTIVE;
637 		break;
638 	case EEH_OPT_DISABLE:
639 	case EEH_OPT_ENABLE:
640 	case EEH_OPT_FREEZE_PE:
641 		active_flag = 0;
642 		break;
643 	default:
644 		pr_warn("%s: Invalid function %d\n",
645 			__func__, function);
646 		return -EINVAL;
647 	}
648 
649 	/*
650 	 * Check if IO or DMA has been enabled before
651 	 * enabling them.
652 	 */
653 	if (active_flag) {
654 		rc = eeh_ops->get_state(pe, NULL);
655 		if (rc < 0)
656 			return rc;
657 
658 		/* Needn't enable it at all */
659 		if (rc == EEH_STATE_NOT_SUPPORT)
660 			return 0;
661 
662 		/* It's already enabled */
663 		if (rc & active_flag)
664 			return 0;
665 	}
666 
667 
668 	/* Issue the request */
669 	rc = eeh_ops->set_option(pe, function);
670 	if (rc)
671 		pr_warn("%s: Unexpected state change %d on "
672 			"PHB#%d-PE#%x, err=%d\n",
673 			__func__, function, pe->phb->global_number,
674 			pe->addr, rc);
675 
676 	/* Check if the request is finished successfully */
677 	if (active_flag) {
678 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
679 		if (rc <= 0)
680 			return rc;
681 
682 		if (rc & active_flag)
683 			return 0;
684 
685 		return -EIO;
686 	}
687 
688 	return rc;
689 }
690 
691 static void *eeh_disable_and_save_dev_state(void *data, void *userdata)
692 {
693 	struct eeh_dev *edev = data;
694 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
695 	struct pci_dev *dev = userdata;
696 
697 	/*
698 	 * The caller should have disabled and saved the
699 	 * state for the specified device
700 	 */
701 	if (!pdev || pdev == dev)
702 		return NULL;
703 
704 	/* Ensure we have D0 power state */
705 	pci_set_power_state(pdev, PCI_D0);
706 
707 	/* Save device state */
708 	pci_save_state(pdev);
709 
710 	/*
711 	 * Disable device to avoid any DMA traffic and
712 	 * interrupt from the device
713 	 */
714 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
715 
716 	return NULL;
717 }
718 
719 static void *eeh_restore_dev_state(void *data, void *userdata)
720 {
721 	struct eeh_dev *edev = data;
722 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
723 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
724 	struct pci_dev *dev = userdata;
725 
726 	if (!pdev)
727 		return NULL;
728 
729 	/* Apply customization from firmware */
730 	if (pdn && eeh_ops->restore_config)
731 		eeh_ops->restore_config(pdn);
732 
733 	/* The caller should restore state for the specified device */
734 	if (pdev != dev)
735 		pci_restore_state(pdev);
736 
737 	return NULL;
738 }
739 
740 /**
741  * pcibios_set_pcie_slot_reset - Set PCI-E reset state
742  * @dev: pci device struct
743  * @state: reset state to enter
744  *
745  * Return value:
746  * 	0 if success
747  */
748 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
749 {
750 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
751 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
752 
753 	if (!pe) {
754 		pr_err("%s: No PE found on PCI device %s\n",
755 			__func__, pci_name(dev));
756 		return -EINVAL;
757 	}
758 
759 	switch (state) {
760 	case pcie_deassert_reset:
761 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
762 		eeh_unfreeze_pe(pe, false);
763 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
764 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
765 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
766 		break;
767 	case pcie_hot_reset:
768 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
769 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
770 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
771 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
772 		eeh_ops->reset(pe, EEH_RESET_HOT);
773 		break;
774 	case pcie_warm_reset:
775 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
776 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
777 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
778 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
779 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
780 		break;
781 	default:
782 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
783 		return -EINVAL;
784 	};
785 
786 	return 0;
787 }
788 
789 /**
790  * eeh_set_pe_freset - Check the required reset for the indicated device
791  * @data: EEH device
792  * @flag: return value
793  *
794  * Each device might have its preferred reset type: fundamental or
795  * hot reset. The routine is used to collected the information for
796  * the indicated device and its children so that the bunch of the
797  * devices could be reset properly.
798  */
799 static void *eeh_set_dev_freset(void *data, void *flag)
800 {
801 	struct pci_dev *dev;
802 	unsigned int *freset = (unsigned int *)flag;
803 	struct eeh_dev *edev = (struct eeh_dev *)data;
804 
805 	dev = eeh_dev_to_pci_dev(edev);
806 	if (dev)
807 		*freset |= dev->needs_freset;
808 
809 	return NULL;
810 }
811 
812 /**
813  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
814  * @pe: EEH PE
815  *
816  * Assert the PCI #RST line for 1/4 second.
817  */
818 static void eeh_reset_pe_once(struct eeh_pe *pe)
819 {
820 	unsigned int freset = 0;
821 
822 	/* Determine type of EEH reset required for
823 	 * Partitionable Endpoint, a hot-reset (1)
824 	 * or a fundamental reset (3).
825 	 * A fundamental reset required by any device under
826 	 * Partitionable Endpoint trumps hot-reset.
827 	 */
828 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
829 
830 	if (freset)
831 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
832 	else
833 		eeh_ops->reset(pe, EEH_RESET_HOT);
834 
835 	eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
836 }
837 
838 /**
839  * eeh_reset_pe - Reset the indicated PE
840  * @pe: EEH PE
841  *
842  * This routine should be called to reset indicated device, including
843  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
844  * might be involved as well.
845  */
846 int eeh_reset_pe(struct eeh_pe *pe)
847 {
848 	int flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
849 	int i, state, ret;
850 
851 	/* Mark as reset and block config space */
852 	eeh_pe_state_mark(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
853 
854 	/* Take three shots at resetting the bus */
855 	for (i = 0; i < 3; i++) {
856 		eeh_reset_pe_once(pe);
857 
858 		/*
859 		 * EEH_PE_ISOLATED is expected to be removed after
860 		 * BAR restore.
861 		 */
862 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
863 		if ((state & flags) == flags) {
864 			ret = 0;
865 			goto out;
866 		}
867 
868 		if (state < 0) {
869 			pr_warn("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
870 				__func__, pe->phb->global_number, pe->addr);
871 			ret = -ENOTRECOVERABLE;
872 			goto out;
873 		}
874 
875 		/* We might run out of credits */
876 		ret = -EIO;
877 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
878 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
879 	}
880 
881 out:
882 	eeh_pe_state_clear(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
883 	return ret;
884 }
885 
886 /**
887  * eeh_save_bars - Save device bars
888  * @edev: PCI device associated EEH device
889  *
890  * Save the values of the device bars. Unlike the restore
891  * routine, this routine is *not* recursive. This is because
892  * PCI devices are added individually; but, for the restore,
893  * an entire slot is reset at a time.
894  */
895 void eeh_save_bars(struct eeh_dev *edev)
896 {
897 	struct pci_dn *pdn;
898 	int i;
899 
900 	pdn = eeh_dev_to_pdn(edev);
901 	if (!pdn)
902 		return;
903 
904 	for (i = 0; i < 16; i++)
905 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
906 
907 	/*
908 	 * For PCI bridges including root port, we need enable bus
909 	 * master explicitly. Otherwise, it can't fetch IODA table
910 	 * entries correctly. So we cache the bit in advance so that
911 	 * we can restore it after reset, either PHB range or PE range.
912 	 */
913 	if (edev->mode & EEH_DEV_BRIDGE)
914 		edev->config_space[1] |= PCI_COMMAND_MASTER;
915 }
916 
917 /**
918  * eeh_ops_register - Register platform dependent EEH operations
919  * @ops: platform dependent EEH operations
920  *
921  * Register the platform dependent EEH operation callback
922  * functions. The platform should call this function before
923  * any other EEH operations.
924  */
925 int __init eeh_ops_register(struct eeh_ops *ops)
926 {
927 	if (!ops->name) {
928 		pr_warn("%s: Invalid EEH ops name for %p\n",
929 			__func__, ops);
930 		return -EINVAL;
931 	}
932 
933 	if (eeh_ops && eeh_ops != ops) {
934 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
935 			__func__, eeh_ops->name, ops->name);
936 		return -EEXIST;
937 	}
938 
939 	eeh_ops = ops;
940 
941 	return 0;
942 }
943 
944 /**
945  * eeh_ops_unregister - Unreigster platform dependent EEH operations
946  * @name: name of EEH platform operations
947  *
948  * Unregister the platform dependent EEH operation callback
949  * functions.
950  */
951 int __exit eeh_ops_unregister(const char *name)
952 {
953 	if (!name || !strlen(name)) {
954 		pr_warn("%s: Invalid EEH ops name\n",
955 			__func__);
956 		return -EINVAL;
957 	}
958 
959 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
960 		eeh_ops = NULL;
961 		return 0;
962 	}
963 
964 	return -EEXIST;
965 }
966 
967 static int eeh_reboot_notifier(struct notifier_block *nb,
968 			       unsigned long action, void *unused)
969 {
970 	eeh_clear_flag(EEH_ENABLED);
971 	return NOTIFY_DONE;
972 }
973 
974 static struct notifier_block eeh_reboot_nb = {
975 	.notifier_call = eeh_reboot_notifier,
976 };
977 
978 /**
979  * eeh_init - EEH initialization
980  *
981  * Initialize EEH by trying to enable it for all of the adapters in the system.
982  * As a side effect we can determine here if eeh is supported at all.
983  * Note that we leave EEH on so failed config cycles won't cause a machine
984  * check.  If a user turns off EEH for a particular adapter they are really
985  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
986  * grant access to a slot if EEH isn't enabled, and so we always enable
987  * EEH for all slots/all devices.
988  *
989  * The eeh-force-off option disables EEH checking globally, for all slots.
990  * Even if force-off is set, the EEH hardware is still enabled, so that
991  * newer systems can boot.
992  */
993 int eeh_init(void)
994 {
995 	struct pci_controller *hose, *tmp;
996 	struct pci_dn *pdn;
997 	static int cnt = 0;
998 	int ret = 0;
999 
1000 	/*
1001 	 * We have to delay the initialization on PowerNV after
1002 	 * the PCI hierarchy tree has been built because the PEs
1003 	 * are figured out based on PCI devices instead of device
1004 	 * tree nodes
1005 	 */
1006 	if (machine_is(powernv) && cnt++ <= 0)
1007 		return ret;
1008 
1009 	/* Register reboot notifier */
1010 	ret = register_reboot_notifier(&eeh_reboot_nb);
1011 	if (ret) {
1012 		pr_warn("%s: Failed to register notifier (%d)\n",
1013 			__func__, ret);
1014 		return ret;
1015 	}
1016 
1017 	/* call platform initialization function */
1018 	if (!eeh_ops) {
1019 		pr_warn("%s: Platform EEH operation not found\n",
1020 			__func__);
1021 		return -EEXIST;
1022 	} else if ((ret = eeh_ops->init()))
1023 		return ret;
1024 
1025 	/* Initialize EEH event */
1026 	ret = eeh_event_init();
1027 	if (ret)
1028 		return ret;
1029 
1030 	/* Enable EEH for all adapters */
1031 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1032 		pdn = hose->pci_data;
1033 		traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1034 	}
1035 
1036 	/*
1037 	 * Call platform post-initialization. Actually, It's good chance
1038 	 * to inform platform that EEH is ready to supply service if the
1039 	 * I/O cache stuff has been built up.
1040 	 */
1041 	if (eeh_ops->post_init) {
1042 		ret = eeh_ops->post_init();
1043 		if (ret)
1044 			return ret;
1045 	}
1046 
1047 	if (eeh_enabled())
1048 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1049 	else
1050 		pr_warn("EEH: No capable adapters found\n");
1051 
1052 	return ret;
1053 }
1054 
1055 core_initcall_sync(eeh_init);
1056 
1057 /**
1058  * eeh_add_device_early - Enable EEH for the indicated device node
1059  * @pdn: PCI device node for which to set up EEH
1060  *
1061  * This routine must be used to perform EEH initialization for PCI
1062  * devices that were added after system boot (e.g. hotplug, dlpar).
1063  * This routine must be called before any i/o is performed to the
1064  * adapter (inluding any config-space i/o).
1065  * Whether this actually enables EEH or not for this device depends
1066  * on the CEC architecture, type of the device, on earlier boot
1067  * command-line arguments & etc.
1068  */
1069 void eeh_add_device_early(struct pci_dn *pdn)
1070 {
1071 	struct pci_controller *phb;
1072 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1073 
1074 	if (!edev || !eeh_enabled())
1075 		return;
1076 
1077 	if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1078 		return;
1079 
1080 	/* USB Bus children of PCI devices will not have BUID's */
1081 	phb = edev->phb;
1082 	if (NULL == phb ||
1083 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1084 		return;
1085 
1086 	eeh_ops->probe(pdn, NULL);
1087 }
1088 
1089 /**
1090  * eeh_add_device_tree_early - Enable EEH for the indicated device
1091  * @pdn: PCI device node
1092  *
1093  * This routine must be used to perform EEH initialization for the
1094  * indicated PCI device that was added after system boot (e.g.
1095  * hotplug, dlpar).
1096  */
1097 void eeh_add_device_tree_early(struct pci_dn *pdn)
1098 {
1099 	struct pci_dn *n;
1100 
1101 	if (!pdn)
1102 		return;
1103 
1104 	list_for_each_entry(n, &pdn->child_list, list)
1105 		eeh_add_device_tree_early(n);
1106 	eeh_add_device_early(pdn);
1107 }
1108 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1109 
1110 /**
1111  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1112  * @dev: pci device for which to set up EEH
1113  *
1114  * This routine must be used to complete EEH initialization for PCI
1115  * devices that were added after system boot (e.g. hotplug, dlpar).
1116  */
1117 void eeh_add_device_late(struct pci_dev *dev)
1118 {
1119 	struct pci_dn *pdn;
1120 	struct eeh_dev *edev;
1121 
1122 	if (!dev || !eeh_enabled())
1123 		return;
1124 
1125 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1126 
1127 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1128 	edev = pdn_to_eeh_dev(pdn);
1129 	if (edev->pdev == dev) {
1130 		pr_debug("EEH: Already referenced !\n");
1131 		return;
1132 	}
1133 
1134 	/*
1135 	 * The EEH cache might not be removed correctly because of
1136 	 * unbalanced kref to the device during unplug time, which
1137 	 * relies on pcibios_release_device(). So we have to remove
1138 	 * that here explicitly.
1139 	 */
1140 	if (edev->pdev) {
1141 		eeh_rmv_from_parent_pe(edev);
1142 		eeh_addr_cache_rmv_dev(edev->pdev);
1143 		eeh_sysfs_remove_device(edev->pdev);
1144 		edev->mode &= ~EEH_DEV_SYSFS;
1145 
1146 		/*
1147 		 * We definitely should have the PCI device removed
1148 		 * though it wasn't correctly. So we needn't call
1149 		 * into error handler afterwards.
1150 		 */
1151 		edev->mode |= EEH_DEV_NO_HANDLER;
1152 
1153 		edev->pdev = NULL;
1154 		dev->dev.archdata.edev = NULL;
1155 	}
1156 
1157 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1158 		eeh_ops->probe(pdn, NULL);
1159 
1160 	edev->pdev = dev;
1161 	dev->dev.archdata.edev = edev;
1162 
1163 	eeh_addr_cache_insert_dev(dev);
1164 }
1165 
1166 /**
1167  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1168  * @bus: PCI bus
1169  *
1170  * This routine must be used to perform EEH initialization for PCI
1171  * devices which are attached to the indicated PCI bus. The PCI bus
1172  * is added after system boot through hotplug or dlpar.
1173  */
1174 void eeh_add_device_tree_late(struct pci_bus *bus)
1175 {
1176 	struct pci_dev *dev;
1177 
1178 	list_for_each_entry(dev, &bus->devices, bus_list) {
1179 		eeh_add_device_late(dev);
1180 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1181 			struct pci_bus *subbus = dev->subordinate;
1182 			if (subbus)
1183 				eeh_add_device_tree_late(subbus);
1184 		}
1185 	}
1186 }
1187 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1188 
1189 /**
1190  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1191  * @bus: PCI bus
1192  *
1193  * This routine must be used to add EEH sysfs files for PCI
1194  * devices which are attached to the indicated PCI bus. The PCI bus
1195  * is added after system boot through hotplug or dlpar.
1196  */
1197 void eeh_add_sysfs_files(struct pci_bus *bus)
1198 {
1199 	struct pci_dev *dev;
1200 
1201 	list_for_each_entry(dev, &bus->devices, bus_list) {
1202 		eeh_sysfs_add_device(dev);
1203 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1204 			struct pci_bus *subbus = dev->subordinate;
1205 			if (subbus)
1206 				eeh_add_sysfs_files(subbus);
1207 		}
1208 	}
1209 }
1210 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1211 
1212 /**
1213  * eeh_remove_device - Undo EEH setup for the indicated pci device
1214  * @dev: pci device to be removed
1215  *
1216  * This routine should be called when a device is removed from
1217  * a running system (e.g. by hotplug or dlpar).  It unregisters
1218  * the PCI device from the EEH subsystem.  I/O errors affecting
1219  * this device will no longer be detected after this call; thus,
1220  * i/o errors affecting this slot may leave this device unusable.
1221  */
1222 void eeh_remove_device(struct pci_dev *dev)
1223 {
1224 	struct eeh_dev *edev;
1225 
1226 	if (!dev || !eeh_enabled())
1227 		return;
1228 	edev = pci_dev_to_eeh_dev(dev);
1229 
1230 	/* Unregister the device with the EEH/PCI address search system */
1231 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1232 
1233 	if (!edev || !edev->pdev || !edev->pe) {
1234 		pr_debug("EEH: Not referenced !\n");
1235 		return;
1236 	}
1237 
1238 	/*
1239 	 * During the hotplug for EEH error recovery, we need the EEH
1240 	 * device attached to the parent PE in order for BAR restore
1241 	 * a bit later. So we keep it for BAR restore and remove it
1242 	 * from the parent PE during the BAR resotre.
1243 	 */
1244 	edev->pdev = NULL;
1245 	dev->dev.archdata.edev = NULL;
1246 	if (!(edev->pe->state & EEH_PE_KEEP))
1247 		eeh_rmv_from_parent_pe(edev);
1248 	else
1249 		edev->mode |= EEH_DEV_DISCONNECTED;
1250 
1251 	/*
1252 	 * We're removing from the PCI subsystem, that means
1253 	 * the PCI device driver can't support EEH or not
1254 	 * well. So we rely on hotplug completely to do recovery
1255 	 * for the specific PCI device.
1256 	 */
1257 	edev->mode |= EEH_DEV_NO_HANDLER;
1258 
1259 	eeh_addr_cache_rmv_dev(dev);
1260 	eeh_sysfs_remove_device(dev);
1261 	edev->mode &= ~EEH_DEV_SYSFS;
1262 }
1263 
1264 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1265 {
1266 	int ret;
1267 
1268 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1269 	if (ret) {
1270 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1271 			__func__, ret, pe->phb->global_number, pe->addr);
1272 		return ret;
1273 	}
1274 
1275 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1276 	if (ret) {
1277 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1278 			__func__, ret, pe->phb->global_number, pe->addr);
1279 		return ret;
1280 	}
1281 
1282 	/* Clear software isolated state */
1283 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1284 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1285 
1286 	return ret;
1287 }
1288 
1289 
1290 static struct pci_device_id eeh_reset_ids[] = {
1291 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1292 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1293 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1294 	{ 0 }
1295 };
1296 
1297 static int eeh_pe_change_owner(struct eeh_pe *pe)
1298 {
1299 	struct eeh_dev *edev, *tmp;
1300 	struct pci_dev *pdev;
1301 	struct pci_device_id *id;
1302 	int flags, ret;
1303 
1304 	/* Check PE state */
1305 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1306 	ret = eeh_ops->get_state(pe, NULL);
1307 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1308 		return 0;
1309 
1310 	/* Unfrozen PE, nothing to do */
1311 	if ((ret & flags) == flags)
1312 		return 0;
1313 
1314 	/* Frozen PE, check if it needs PE level reset */
1315 	eeh_pe_for_each_dev(pe, edev, tmp) {
1316 		pdev = eeh_dev_to_pci_dev(edev);
1317 		if (!pdev)
1318 			continue;
1319 
1320 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1321 			if (id->vendor != PCI_ANY_ID &&
1322 			    id->vendor != pdev->vendor)
1323 				continue;
1324 			if (id->device != PCI_ANY_ID &&
1325 			    id->device != pdev->device)
1326 				continue;
1327 			if (id->subvendor != PCI_ANY_ID &&
1328 			    id->subvendor != pdev->subsystem_vendor)
1329 				continue;
1330 			if (id->subdevice != PCI_ANY_ID &&
1331 			    id->subdevice != pdev->subsystem_device)
1332 				continue;
1333 
1334 			goto reset;
1335 		}
1336 	}
1337 
1338 	return eeh_unfreeze_pe(pe, true);
1339 
1340 reset:
1341 	return eeh_pe_reset_and_recover(pe);
1342 }
1343 
1344 /**
1345  * eeh_dev_open - Increase count of pass through devices for PE
1346  * @pdev: PCI device
1347  *
1348  * Increase count of passed through devices for the indicated
1349  * PE. In the result, the EEH errors detected on the PE won't be
1350  * reported. The PE owner will be responsible for detection
1351  * and recovery.
1352  */
1353 int eeh_dev_open(struct pci_dev *pdev)
1354 {
1355 	struct eeh_dev *edev;
1356 	int ret = -ENODEV;
1357 
1358 	mutex_lock(&eeh_dev_mutex);
1359 
1360 	/* No PCI device ? */
1361 	if (!pdev)
1362 		goto out;
1363 
1364 	/* No EEH device or PE ? */
1365 	edev = pci_dev_to_eeh_dev(pdev);
1366 	if (!edev || !edev->pe)
1367 		goto out;
1368 
1369 	/*
1370 	 * The PE might have been put into frozen state, but we
1371 	 * didn't detect that yet. The passed through PCI devices
1372 	 * in frozen PE won't work properly. Clear the frozen state
1373 	 * in advance.
1374 	 */
1375 	ret = eeh_pe_change_owner(edev->pe);
1376 	if (ret)
1377 		goto out;
1378 
1379 	/* Increase PE's pass through count */
1380 	atomic_inc(&edev->pe->pass_dev_cnt);
1381 	mutex_unlock(&eeh_dev_mutex);
1382 
1383 	return 0;
1384 out:
1385 	mutex_unlock(&eeh_dev_mutex);
1386 	return ret;
1387 }
1388 EXPORT_SYMBOL_GPL(eeh_dev_open);
1389 
1390 /**
1391  * eeh_dev_release - Decrease count of pass through devices for PE
1392  * @pdev: PCI device
1393  *
1394  * Decrease count of pass through devices for the indicated PE. If
1395  * there is no passed through device in PE, the EEH errors detected
1396  * on the PE will be reported and handled as usual.
1397  */
1398 void eeh_dev_release(struct pci_dev *pdev)
1399 {
1400 	struct eeh_dev *edev;
1401 
1402 	mutex_lock(&eeh_dev_mutex);
1403 
1404 	/* No PCI device ? */
1405 	if (!pdev)
1406 		goto out;
1407 
1408 	/* No EEH device ? */
1409 	edev = pci_dev_to_eeh_dev(pdev);
1410 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1411 		goto out;
1412 
1413 	/* Decrease PE's pass through count */
1414 	atomic_dec(&edev->pe->pass_dev_cnt);
1415 	WARN_ON(atomic_read(&edev->pe->pass_dev_cnt) < 0);
1416 	eeh_pe_change_owner(edev->pe);
1417 out:
1418 	mutex_unlock(&eeh_dev_mutex);
1419 }
1420 EXPORT_SYMBOL(eeh_dev_release);
1421 
1422 #ifdef CONFIG_IOMMU_API
1423 
1424 static int dev_has_iommu_table(struct device *dev, void *data)
1425 {
1426 	struct pci_dev *pdev = to_pci_dev(dev);
1427 	struct pci_dev **ppdev = data;
1428 
1429 	if (!dev)
1430 		return 0;
1431 
1432 	if (dev->iommu_group) {
1433 		*ppdev = pdev;
1434 		return 1;
1435 	}
1436 
1437 	return 0;
1438 }
1439 
1440 /**
1441  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1442  * @group: IOMMU group
1443  *
1444  * The routine is called to convert IOMMU group to EEH PE.
1445  */
1446 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1447 {
1448 	struct pci_dev *pdev = NULL;
1449 	struct eeh_dev *edev;
1450 	int ret;
1451 
1452 	/* No IOMMU group ? */
1453 	if (!group)
1454 		return NULL;
1455 
1456 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1457 	if (!ret || !pdev)
1458 		return NULL;
1459 
1460 	/* No EEH device or PE ? */
1461 	edev = pci_dev_to_eeh_dev(pdev);
1462 	if (!edev || !edev->pe)
1463 		return NULL;
1464 
1465 	return edev->pe;
1466 }
1467 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1468 
1469 #endif /* CONFIG_IOMMU_API */
1470 
1471 /**
1472  * eeh_pe_set_option - Set options for the indicated PE
1473  * @pe: EEH PE
1474  * @option: requested option
1475  *
1476  * The routine is called to enable or disable EEH functionality
1477  * on the indicated PE, to enable IO or DMA for the frozen PE.
1478  */
1479 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1480 {
1481 	int ret = 0;
1482 
1483 	/* Invalid PE ? */
1484 	if (!pe)
1485 		return -ENODEV;
1486 
1487 	/*
1488 	 * EEH functionality could possibly be disabled, just
1489 	 * return error for the case. And the EEH functinality
1490 	 * isn't expected to be disabled on one specific PE.
1491 	 */
1492 	switch (option) {
1493 	case EEH_OPT_ENABLE:
1494 		if (eeh_enabled()) {
1495 			ret = eeh_pe_change_owner(pe);
1496 			break;
1497 		}
1498 		ret = -EIO;
1499 		break;
1500 	case EEH_OPT_DISABLE:
1501 		break;
1502 	case EEH_OPT_THAW_MMIO:
1503 	case EEH_OPT_THAW_DMA:
1504 		if (!eeh_ops || !eeh_ops->set_option) {
1505 			ret = -ENOENT;
1506 			break;
1507 		}
1508 
1509 		ret = eeh_pci_enable(pe, option);
1510 		break;
1511 	default:
1512 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1513 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1514 		ret = -EINVAL;
1515 	}
1516 
1517 	return ret;
1518 }
1519 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1520 
1521 /**
1522  * eeh_pe_get_state - Retrieve PE's state
1523  * @pe: EEH PE
1524  *
1525  * Retrieve the PE's state, which includes 3 aspects: enabled
1526  * DMA, enabled IO and asserted reset.
1527  */
1528 int eeh_pe_get_state(struct eeh_pe *pe)
1529 {
1530 	int result, ret = 0;
1531 	bool rst_active, dma_en, mmio_en;
1532 
1533 	/* Existing PE ? */
1534 	if (!pe)
1535 		return -ENODEV;
1536 
1537 	if (!eeh_ops || !eeh_ops->get_state)
1538 		return -ENOENT;
1539 
1540 	result = eeh_ops->get_state(pe, NULL);
1541 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1542 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1543 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1544 
1545 	if (rst_active)
1546 		ret = EEH_PE_STATE_RESET;
1547 	else if (dma_en && mmio_en)
1548 		ret = EEH_PE_STATE_NORMAL;
1549 	else if (!dma_en && !mmio_en)
1550 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1551 	else if (!dma_en && mmio_en)
1552 		ret = EEH_PE_STATE_STOPPED_DMA;
1553 	else
1554 		ret = EEH_PE_STATE_UNAVAIL;
1555 
1556 	return ret;
1557 }
1558 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1559 
1560 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1561 {
1562 	struct eeh_dev *edev, *tmp;
1563 	struct pci_dev *pdev;
1564 	int ret = 0;
1565 
1566 	/* Restore config space */
1567 	eeh_pe_restore_bars(pe);
1568 
1569 	/*
1570 	 * Reenable PCI devices as the devices passed
1571 	 * through are always enabled before the reset.
1572 	 */
1573 	eeh_pe_for_each_dev(pe, edev, tmp) {
1574 		pdev = eeh_dev_to_pci_dev(edev);
1575 		if (!pdev)
1576 			continue;
1577 
1578 		ret = pci_reenable_device(pdev);
1579 		if (ret) {
1580 			pr_warn("%s: Failure %d reenabling %s\n",
1581 				__func__, ret, pci_name(pdev));
1582 			return ret;
1583 		}
1584 	}
1585 
1586 	/* The PE is still in frozen state */
1587 	return eeh_unfreeze_pe(pe, true);
1588 }
1589 
1590 /**
1591  * eeh_pe_reset - Issue PE reset according to specified type
1592  * @pe: EEH PE
1593  * @option: reset type
1594  *
1595  * The routine is called to reset the specified PE with the
1596  * indicated type, either fundamental reset or hot reset.
1597  * PE reset is the most important part for error recovery.
1598  */
1599 int eeh_pe_reset(struct eeh_pe *pe, int option)
1600 {
1601 	int ret = 0;
1602 
1603 	/* Invalid PE ? */
1604 	if (!pe)
1605 		return -ENODEV;
1606 
1607 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1608 		return -ENOENT;
1609 
1610 	switch (option) {
1611 	case EEH_RESET_DEACTIVATE:
1612 		ret = eeh_ops->reset(pe, option);
1613 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1614 		if (ret)
1615 			break;
1616 
1617 		ret = eeh_pe_reenable_devices(pe);
1618 		break;
1619 	case EEH_RESET_HOT:
1620 	case EEH_RESET_FUNDAMENTAL:
1621 		/*
1622 		 * Proactively freeze the PE to drop all MMIO access
1623 		 * during reset, which should be banned as it's always
1624 		 * cause recursive EEH error.
1625 		 */
1626 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1627 
1628 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1629 		ret = eeh_ops->reset(pe, option);
1630 		break;
1631 	default:
1632 		pr_debug("%s: Unsupported option %d\n",
1633 			__func__, option);
1634 		ret = -EINVAL;
1635 	}
1636 
1637 	return ret;
1638 }
1639 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1640 
1641 /**
1642  * eeh_pe_configure - Configure PCI bridges after PE reset
1643  * @pe: EEH PE
1644  *
1645  * The routine is called to restore the PCI config space for
1646  * those PCI devices, especially PCI bridges affected by PE
1647  * reset issued previously.
1648  */
1649 int eeh_pe_configure(struct eeh_pe *pe)
1650 {
1651 	int ret = 0;
1652 
1653 	/* Invalid PE ? */
1654 	if (!pe)
1655 		return -ENODEV;
1656 
1657 	return ret;
1658 }
1659 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1660 
1661 /**
1662  * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1663  * @pe: the indicated PE
1664  * @type: error type
1665  * @function: error function
1666  * @addr: address
1667  * @mask: address mask
1668  *
1669  * The routine is called to inject the specified PCI error, which
1670  * is determined by @type and @function, to the indicated PE for
1671  * testing purpose.
1672  */
1673 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1674 		      unsigned long addr, unsigned long mask)
1675 {
1676 	/* Invalid PE ? */
1677 	if (!pe)
1678 		return -ENODEV;
1679 
1680 	/* Unsupported operation ? */
1681 	if (!eeh_ops || !eeh_ops->err_inject)
1682 		return -ENOENT;
1683 
1684 	/* Check on PCI error type */
1685 	if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1686 		return -EINVAL;
1687 
1688 	/* Check on PCI error function */
1689 	if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1690 		return -EINVAL;
1691 
1692 	return eeh_ops->err_inject(pe, type, func, addr, mask);
1693 }
1694 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1695 
1696 static int proc_eeh_show(struct seq_file *m, void *v)
1697 {
1698 	if (!eeh_enabled()) {
1699 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1700 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1701 	} else {
1702 		seq_printf(m, "EEH Subsystem is enabled\n");
1703 		seq_printf(m,
1704 				"no device=%llu\n"
1705 				"no device node=%llu\n"
1706 				"no config address=%llu\n"
1707 				"check not wanted=%llu\n"
1708 				"eeh_total_mmio_ffs=%llu\n"
1709 				"eeh_false_positives=%llu\n"
1710 				"eeh_slot_resets=%llu\n",
1711 				eeh_stats.no_device,
1712 				eeh_stats.no_dn,
1713 				eeh_stats.no_cfg_addr,
1714 				eeh_stats.ignored_check,
1715 				eeh_stats.total_mmio_ffs,
1716 				eeh_stats.false_positives,
1717 				eeh_stats.slot_resets);
1718 	}
1719 
1720 	return 0;
1721 }
1722 
1723 static int proc_eeh_open(struct inode *inode, struct file *file)
1724 {
1725 	return single_open(file, proc_eeh_show, NULL);
1726 }
1727 
1728 static const struct file_operations proc_eeh_operations = {
1729 	.open      = proc_eeh_open,
1730 	.read      = seq_read,
1731 	.llseek    = seq_lseek,
1732 	.release   = single_release,
1733 };
1734 
1735 #ifdef CONFIG_DEBUG_FS
1736 static int eeh_enable_dbgfs_set(void *data, u64 val)
1737 {
1738 	if (val)
1739 		eeh_clear_flag(EEH_FORCE_DISABLED);
1740 	else
1741 		eeh_add_flag(EEH_FORCE_DISABLED);
1742 
1743 	/* Notify the backend */
1744 	if (eeh_ops->post_init)
1745 		eeh_ops->post_init();
1746 
1747 	return 0;
1748 }
1749 
1750 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1751 {
1752 	if (eeh_enabled())
1753 		*val = 0x1ul;
1754 	else
1755 		*val = 0x0ul;
1756 	return 0;
1757 }
1758 
1759 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1760 {
1761 	eeh_max_freezes = val;
1762 	return 0;
1763 }
1764 
1765 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1766 {
1767 	*val = eeh_max_freezes;
1768 	return 0;
1769 }
1770 
1771 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1772 			eeh_enable_dbgfs_set, "0x%llx\n");
1773 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1774 			eeh_freeze_dbgfs_set, "0x%llx\n");
1775 #endif
1776 
1777 static int __init eeh_init_proc(void)
1778 {
1779 	if (machine_is(pseries) || machine_is(powernv)) {
1780 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1781 #ifdef CONFIG_DEBUG_FS
1782 		debugfs_create_file("eeh_enable", 0600,
1783                                     powerpc_debugfs_root, NULL,
1784                                     &eeh_enable_dbgfs_ops);
1785 		debugfs_create_file("eeh_max_freezes", 0600,
1786 				    powerpc_debugfs_root, NULL,
1787 				    &eeh_freeze_dbgfs_ops);
1788 #endif
1789 	}
1790 
1791 	return 0;
1792 }
1793 __initcall(eeh_init_proc);
1794