xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision 8730046c)
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 "Enhanced Error Handling" is a PCI bridge technology for
52  *  dealing with PCI bus errors that can't be dealt with within the
53  *  usual PCI framework, except by check-stopping the CPU.  Systems
54  *  that are designed for high-availability/reliability cannot afford
55  *  to crash due to a "mere" PCI error, thus the need for EEH.
56  *  An EEH-capable bridge operates by converting a detected error
57  *  into a "slot freeze", taking the PCI adapter off-line, making
58  *  the slot behave, from the OS'es point of view, as if the slot
59  *  were "empty": all reads return 0xff's and all writes are silently
60  *  ignored.  EEH slot isolation events can be triggered by parity
61  *  errors on the address or data busses (e.g. during posted writes),
62  *  which in turn might be caused by low voltage on the bus, dust,
63  *  vibration, humidity, radioactivity or plain-old failed hardware.
64  *
65  *  Note, however, that one of the leading causes of EEH slot
66  *  freeze events are buggy device drivers, buggy device microcode,
67  *  or buggy device hardware.  This is because any attempt by the
68  *  device to bus-master data to a memory address that is not
69  *  assigned to the device will trigger a slot freeze.   (The idea
70  *  is to prevent devices-gone-wild from corrupting system memory).
71  *  Buggy hardware/drivers will have a miserable time co-existing
72  *  with EEH.
73  *
74  *  Ideally, a PCI device driver, when suspecting that an isolation
75  *  event has occurred (e.g. by reading 0xff's), will then ask EEH
76  *  whether this is the case, and then take appropriate steps to
77  *  reset the PCI slot, the PCI device, and then resume operations.
78  *  However, until that day,  the checking is done here, with the
79  *  eeh_check_failure() routine embedded in the MMIO macros.  If
80  *  the slot is found to be isolated, an "EEH Event" is synthesized
81  *  and sent out for processing.
82  */
83 
84 /* If a device driver keeps reading an MMIO register in an interrupt
85  * handler after a slot isolation event, it might be broken.
86  * This sets the threshold for how many read attempts we allow
87  * before printing an error message.
88  */
89 #define EEH_MAX_FAILS	2100000
90 
91 /* Time to wait for a PCI slot to report status, in milliseconds */
92 #define PCI_BUS_RESET_WAIT_MSEC (5*60*1000)
93 
94 /*
95  * EEH probe mode support, which is part of the flags,
96  * is to support multiple platforms for EEH. Some platforms
97  * like pSeries do PCI emunation based on device tree.
98  * However, other platforms like powernv probe PCI devices
99  * from hardware. The flag is used to distinguish that.
100  * In addition, struct eeh_ops::probe would be invoked for
101  * particular OF node or PCI device so that the corresponding
102  * PE would be created there.
103  */
104 int eeh_subsystem_flags;
105 EXPORT_SYMBOL(eeh_subsystem_flags);
106 
107 /*
108  * EEH allowed maximal frozen times. If one particular PE's
109  * frozen count in last hour exceeds this limit, the PE will
110  * be forced to be offline permanently.
111  */
112 int eeh_max_freezes = 5;
113 
114 /* Platform dependent EEH operations */
115 struct eeh_ops *eeh_ops = NULL;
116 
117 /* Lock to avoid races due to multiple reports of an error */
118 DEFINE_RAW_SPINLOCK(confirm_error_lock);
119 EXPORT_SYMBOL_GPL(confirm_error_lock);
120 
121 /* Lock to protect passed flags */
122 static DEFINE_MUTEX(eeh_dev_mutex);
123 
124 /* Buffer for reporting pci register dumps. Its here in BSS, and
125  * not dynamically alloced, so that it ends up in RMO where RTAS
126  * can access it.
127  */
128 #define EEH_PCI_REGS_LOG_LEN 8192
129 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
130 
131 /*
132  * The struct is used to maintain the EEH global statistic
133  * information. Besides, the EEH global statistics will be
134  * exported to user space through procfs
135  */
136 struct eeh_stats {
137 	u64 no_device;		/* PCI device not found		*/
138 	u64 no_dn;		/* OF node not found		*/
139 	u64 no_cfg_addr;	/* Config address not found	*/
140 	u64 ignored_check;	/* EEH check skipped		*/
141 	u64 total_mmio_ffs;	/* Total EEH checks		*/
142 	u64 false_positives;	/* Unnecessary EEH checks	*/
143 	u64 slot_resets;	/* PE reset			*/
144 };
145 
146 static struct eeh_stats eeh_stats;
147 
148 static int __init eeh_setup(char *str)
149 {
150 	if (!strcmp(str, "off"))
151 		eeh_add_flag(EEH_FORCE_DISABLED);
152 	else if (!strcmp(str, "early_log"))
153 		eeh_add_flag(EEH_EARLY_DUMP_LOG);
154 
155 	return 1;
156 }
157 __setup("eeh=", eeh_setup);
158 
159 /*
160  * This routine captures assorted PCI configuration space data
161  * for the indicated PCI device, and puts them into a buffer
162  * for RTAS error logging.
163  */
164 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
165 {
166 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
167 	u32 cfg;
168 	int cap, i;
169 	int n = 0, l = 0;
170 	char buffer[128];
171 
172 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x.%01x\n",
173 		       edev->phb->global_number, pdn->busno,
174 		       PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
175 	pr_warn("EEH: of node=%04x:%02x:%02x.%01x\n",
176 		edev->phb->global_number, pdn->busno,
177 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
178 
179 	eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
180 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
181 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
182 
183 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
184 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
185 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
186 
187 	/* Gather bridge-specific registers */
188 	if (edev->mode & EEH_DEV_BRIDGE) {
189 		eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
190 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
191 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
192 
193 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
194 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
195 		pr_warn("EEH: Bridge control: %04x\n", cfg);
196 	}
197 
198 	/* Dump out the PCI-X command and status regs */
199 	cap = edev->pcix_cap;
200 	if (cap) {
201 		eeh_ops->read_config(pdn, cap, 4, &cfg);
202 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
203 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
204 
205 		eeh_ops->read_config(pdn, cap+4, 4, &cfg);
206 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
207 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
208 	}
209 
210 	/* If PCI-E capable, dump PCI-E cap 10 */
211 	cap = edev->pcie_cap;
212 	if (cap) {
213 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
214 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
215 
216 		for (i=0; i<=8; i++) {
217 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
218 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
219 
220 			if ((i % 4) == 0) {
221 				if (i != 0)
222 					pr_warn("%s\n", buffer);
223 
224 				l = scnprintf(buffer, sizeof(buffer),
225 					      "EEH: PCI-E %02x: %08x ",
226 					      4*i, cfg);
227 			} else {
228 				l += scnprintf(buffer+l, sizeof(buffer)-l,
229 					       "%08x ", cfg);
230 			}
231 
232 		}
233 
234 		pr_warn("%s\n", buffer);
235 	}
236 
237 	/* If AER capable, dump it */
238 	cap = edev->aer_cap;
239 	if (cap) {
240 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
241 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
242 
243 		for (i=0; i<=13; i++) {
244 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
245 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
246 
247 			if ((i % 4) == 0) {
248 				if (i != 0)
249 					pr_warn("%s\n", buffer);
250 
251 				l = scnprintf(buffer, sizeof(buffer),
252 					      "EEH: PCI-E AER %02x: %08x ",
253 					      4*i, cfg);
254 			} else {
255 				l += scnprintf(buffer+l, sizeof(buffer)-l,
256 					       "%08x ", cfg);
257 			}
258 		}
259 
260 		pr_warn("%s\n", buffer);
261 	}
262 
263 	return n;
264 }
265 
266 static void *eeh_dump_pe_log(void *data, void *flag)
267 {
268 	struct eeh_pe *pe = data;
269 	struct eeh_dev *edev, *tmp;
270 	size_t *plen = flag;
271 
272 	eeh_pe_for_each_dev(pe, edev, tmp)
273 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
274 					  EEH_PCI_REGS_LOG_LEN - *plen);
275 
276 	return NULL;
277 }
278 
279 /**
280  * eeh_slot_error_detail - Generate combined log including driver log and error log
281  * @pe: EEH PE
282  * @severity: temporary or permanent error log
283  *
284  * This routine should be called to generate the combined log, which
285  * is comprised of driver log and error log. The driver log is figured
286  * out from the config space of the corresponding PCI device, while
287  * the error log is fetched through platform dependent function call.
288  */
289 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
290 {
291 	size_t loglen = 0;
292 
293 	/*
294 	 * When the PHB is fenced or dead, it's pointless to collect
295 	 * the data from PCI config space because it should return
296 	 * 0xFF's. For ER, we still retrieve the data from the PCI
297 	 * config space.
298 	 *
299 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
300 	 * 0xFF's is always returned from PCI config space.
301 	 */
302 	if (!(pe->type & EEH_PE_PHB)) {
303 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG))
304 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
305 
306 		/*
307 		 * The config space of some PCI devices can't be accessed
308 		 * when their PEs are in frozen state. Otherwise, fenced
309 		 * PHB might be seen. Those PEs are identified with flag
310 		 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
311 		 * is set automatically when the PE is put to EEH_PE_ISOLATED.
312 		 *
313 		 * Restoring BARs possibly triggers PCI config access in
314 		 * (OPAL) firmware and then causes fenced PHB. If the
315 		 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
316 		 * pointless to restore BARs and dump config space.
317 		 */
318 		eeh_ops->configure_bridge(pe);
319 		if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
320 			eeh_pe_restore_bars(pe);
321 
322 			pci_regs_buf[0] = 0;
323 			eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
324 		}
325 	}
326 
327 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
328 }
329 
330 /**
331  * eeh_token_to_phys - Convert EEH address token to phys address
332  * @token: I/O token, should be address in the form 0xA....
333  *
334  * This routine should be called to convert virtual I/O address
335  * to physical one.
336  */
337 static inline unsigned long eeh_token_to_phys(unsigned long token)
338 {
339 	pte_t *ptep;
340 	unsigned long pa;
341 	int hugepage_shift;
342 
343 	/*
344 	 * We won't find hugepages here(this is iomem). Hence we are not
345 	 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
346 	 * page table free, because of init_mm.
347 	 */
348 	ptep = __find_linux_pte_or_hugepte(init_mm.pgd, token,
349 					   NULL, &hugepage_shift);
350 	if (!ptep)
351 		return token;
352 	WARN_ON(hugepage_shift);
353 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
354 
355 	return pa | (token & (PAGE_SIZE-1));
356 }
357 
358 /*
359  * On PowerNV platform, we might already have fenced PHB there.
360  * For that case, it's meaningless to recover frozen PE. Intead,
361  * We have to handle fenced PHB firstly.
362  */
363 static int eeh_phb_check_failure(struct eeh_pe *pe)
364 {
365 	struct eeh_pe *phb_pe;
366 	unsigned long flags;
367 	int ret;
368 
369 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
370 		return -EPERM;
371 
372 	/* Find the PHB PE */
373 	phb_pe = eeh_phb_pe_get(pe->phb);
374 	if (!phb_pe) {
375 		pr_warn("%s Can't find PE for PHB#%x\n",
376 			__func__, pe->phb->global_number);
377 		return -EEXIST;
378 	}
379 
380 	/* If the PHB has been in problematic state */
381 	eeh_serialize_lock(&flags);
382 	if (phb_pe->state & EEH_PE_ISOLATED) {
383 		ret = 0;
384 		goto out;
385 	}
386 
387 	/* Check PHB state */
388 	ret = eeh_ops->get_state(phb_pe, NULL);
389 	if ((ret < 0) ||
390 	    (ret == EEH_STATE_NOT_SUPPORT) ||
391 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
392 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
393 		ret = 0;
394 		goto out;
395 	}
396 
397 	/* Isolate the PHB and send event */
398 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
399 	eeh_serialize_unlock(flags);
400 
401 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
402 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
403 	dump_stack();
404 	eeh_send_failure_event(phb_pe);
405 
406 	return 1;
407 out:
408 	eeh_serialize_unlock(flags);
409 	return ret;
410 }
411 
412 /**
413  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
414  * @edev: eeh device
415  *
416  * Check for an EEH failure for the given device node.  Call this
417  * routine if the result of a read was all 0xff's and you want to
418  * find out if this is due to an EEH slot freeze.  This routine
419  * will query firmware for the EEH status.
420  *
421  * Returns 0 if there has not been an EEH error; otherwise returns
422  * a non-zero value and queues up a slot isolation event notification.
423  *
424  * It is safe to call this routine in an interrupt context.
425  */
426 int eeh_dev_check_failure(struct eeh_dev *edev)
427 {
428 	int ret;
429 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
430 	unsigned long flags;
431 	struct pci_dn *pdn;
432 	struct pci_dev *dev;
433 	struct eeh_pe *pe, *parent_pe, *phb_pe;
434 	int rc = 0;
435 	const char *location = NULL;
436 
437 	eeh_stats.total_mmio_ffs++;
438 
439 	if (!eeh_enabled())
440 		return 0;
441 
442 	if (!edev) {
443 		eeh_stats.no_dn++;
444 		return 0;
445 	}
446 	dev = eeh_dev_to_pci_dev(edev);
447 	pe = eeh_dev_to_pe(edev);
448 
449 	/* Access to IO BARs might get this far and still not want checking. */
450 	if (!pe) {
451 		eeh_stats.ignored_check++;
452 		pr_debug("EEH: Ignored check for %s\n",
453 			eeh_pci_name(dev));
454 		return 0;
455 	}
456 
457 	if (!pe->addr && !pe->config_addr) {
458 		eeh_stats.no_cfg_addr++;
459 		return 0;
460 	}
461 
462 	/*
463 	 * On PowerNV platform, we might already have fenced PHB
464 	 * there and we need take care of that firstly.
465 	 */
466 	ret = eeh_phb_check_failure(pe);
467 	if (ret > 0)
468 		return ret;
469 
470 	/*
471 	 * If the PE isn't owned by us, we shouldn't check the
472 	 * state. Instead, let the owner handle it if the PE has
473 	 * been frozen.
474 	 */
475 	if (eeh_pe_passed(pe))
476 		return 0;
477 
478 	/* If we already have a pending isolation event for this
479 	 * slot, we know it's bad already, we don't need to check.
480 	 * Do this checking under a lock; as multiple PCI devices
481 	 * in one slot might report errors simultaneously, and we
482 	 * only want one error recovery routine running.
483 	 */
484 	eeh_serialize_lock(&flags);
485 	rc = 1;
486 	if (pe->state & EEH_PE_ISOLATED) {
487 		pe->check_count++;
488 		if (pe->check_count % EEH_MAX_FAILS == 0) {
489 			pdn = eeh_dev_to_pdn(edev);
490 			if (pdn->node)
491 				location = of_get_property(pdn->node, "ibm,loc-code", NULL);
492 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
493 				"location=%s driver=%s pci addr=%s\n",
494 				pe->check_count,
495 				location ? location : "unknown",
496 				eeh_driver_name(dev), eeh_pci_name(dev));
497 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
498 				eeh_driver_name(dev));
499 			dump_stack();
500 		}
501 		goto dn_unlock;
502 	}
503 
504 	/*
505 	 * Now test for an EEH failure.  This is VERY expensive.
506 	 * Note that the eeh_config_addr may be a parent device
507 	 * in the case of a device behind a bridge, or it may be
508 	 * function zero of a multi-function device.
509 	 * In any case they must share a common PHB.
510 	 */
511 	ret = eeh_ops->get_state(pe, NULL);
512 
513 	/* Note that config-io to empty slots may fail;
514 	 * they are empty when they don't have children.
515 	 * We will punt with the following conditions: Failure to get
516 	 * PE's state, EEH not support and Permanently unavailable
517 	 * state, PE is in good state.
518 	 */
519 	if ((ret < 0) ||
520 	    (ret == EEH_STATE_NOT_SUPPORT) ||
521 	    ((ret & active_flags) == active_flags)) {
522 		eeh_stats.false_positives++;
523 		pe->false_positives++;
524 		rc = 0;
525 		goto dn_unlock;
526 	}
527 
528 	/*
529 	 * It should be corner case that the parent PE has been
530 	 * put into frozen state as well. We should take care
531 	 * that at first.
532 	 */
533 	parent_pe = pe->parent;
534 	while (parent_pe) {
535 		/* Hit the ceiling ? */
536 		if (parent_pe->type & EEH_PE_PHB)
537 			break;
538 
539 		/* Frozen parent PE ? */
540 		ret = eeh_ops->get_state(parent_pe, NULL);
541 		if (ret > 0 &&
542 		    (ret & active_flags) != active_flags)
543 			pe = parent_pe;
544 
545 		/* Next parent level */
546 		parent_pe = parent_pe->parent;
547 	}
548 
549 	eeh_stats.slot_resets++;
550 
551 	/* Avoid repeated reports of this failure, including problems
552 	 * with other functions on this device, and functions under
553 	 * bridges.
554 	 */
555 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
556 	eeh_serialize_unlock(flags);
557 
558 	/* Most EEH events are due to device driver bugs.  Having
559 	 * a stack trace will help the device-driver authors figure
560 	 * out what happened.  So print that out.
561 	 */
562 	phb_pe = eeh_phb_pe_get(pe->phb);
563 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
564 	       pe->phb->global_number, pe->addr);
565 	pr_err("EEH: PE location: %s, PHB location: %s\n",
566 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
567 	dump_stack();
568 
569 	eeh_send_failure_event(pe);
570 
571 	return 1;
572 
573 dn_unlock:
574 	eeh_serialize_unlock(flags);
575 	return rc;
576 }
577 
578 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
579 
580 /**
581  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
582  * @token: I/O address
583  *
584  * Check for an EEH failure at the given I/O address. Call this
585  * routine if the result of a read was all 0xff's and you want to
586  * find out if this is due to an EEH slot freeze event. This routine
587  * will query firmware for the EEH status.
588  *
589  * Note this routine is safe to call in an interrupt context.
590  */
591 int eeh_check_failure(const volatile void __iomem *token)
592 {
593 	unsigned long addr;
594 	struct eeh_dev *edev;
595 
596 	/* Finding the phys addr + pci device; this is pretty quick. */
597 	addr = eeh_token_to_phys((unsigned long __force) token);
598 	edev = eeh_addr_cache_get_dev(addr);
599 	if (!edev) {
600 		eeh_stats.no_device++;
601 		return 0;
602 	}
603 
604 	return eeh_dev_check_failure(edev);
605 }
606 EXPORT_SYMBOL(eeh_check_failure);
607 
608 
609 /**
610  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
611  * @pe: EEH PE
612  *
613  * This routine should be called to reenable frozen MMIO or DMA
614  * so that it would work correctly again. It's useful while doing
615  * recovery or log collection on the indicated device.
616  */
617 int eeh_pci_enable(struct eeh_pe *pe, int function)
618 {
619 	int active_flag, rc;
620 
621 	/*
622 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
623 	 * Also, it's pointless to enable them on unfrozen PE. So
624 	 * we have to check before enabling IO or DMA.
625 	 */
626 	switch (function) {
627 	case EEH_OPT_THAW_MMIO:
628 		active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
629 		break;
630 	case EEH_OPT_THAW_DMA:
631 		active_flag = EEH_STATE_DMA_ACTIVE;
632 		break;
633 	case EEH_OPT_DISABLE:
634 	case EEH_OPT_ENABLE:
635 	case EEH_OPT_FREEZE_PE:
636 		active_flag = 0;
637 		break;
638 	default:
639 		pr_warn("%s: Invalid function %d\n",
640 			__func__, function);
641 		return -EINVAL;
642 	}
643 
644 	/*
645 	 * Check if IO or DMA has been enabled before
646 	 * enabling them.
647 	 */
648 	if (active_flag) {
649 		rc = eeh_ops->get_state(pe, NULL);
650 		if (rc < 0)
651 			return rc;
652 
653 		/* Needn't enable it at all */
654 		if (rc == EEH_STATE_NOT_SUPPORT)
655 			return 0;
656 
657 		/* It's already enabled */
658 		if (rc & active_flag)
659 			return 0;
660 	}
661 
662 
663 	/* Issue the request */
664 	rc = eeh_ops->set_option(pe, function);
665 	if (rc)
666 		pr_warn("%s: Unexpected state change %d on "
667 			"PHB#%x-PE#%x, err=%d\n",
668 			__func__, function, pe->phb->global_number,
669 			pe->addr, rc);
670 
671 	/* Check if the request is finished successfully */
672 	if (active_flag) {
673 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
674 		if (rc < 0)
675 			return rc;
676 
677 		if (rc & active_flag)
678 			return 0;
679 
680 		return -EIO;
681 	}
682 
683 	return rc;
684 }
685 
686 static void *eeh_disable_and_save_dev_state(void *data, void *userdata)
687 {
688 	struct eeh_dev *edev = data;
689 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
690 	struct pci_dev *dev = userdata;
691 
692 	/*
693 	 * The caller should have disabled and saved the
694 	 * state for the specified device
695 	 */
696 	if (!pdev || pdev == dev)
697 		return NULL;
698 
699 	/* Ensure we have D0 power state */
700 	pci_set_power_state(pdev, PCI_D0);
701 
702 	/* Save device state */
703 	pci_save_state(pdev);
704 
705 	/*
706 	 * Disable device to avoid any DMA traffic and
707 	 * interrupt from the device
708 	 */
709 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
710 
711 	return NULL;
712 }
713 
714 static void *eeh_restore_dev_state(void *data, void *userdata)
715 {
716 	struct eeh_dev *edev = data;
717 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
718 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
719 	struct pci_dev *dev = userdata;
720 
721 	if (!pdev)
722 		return NULL;
723 
724 	/* Apply customization from firmware */
725 	if (pdn && eeh_ops->restore_config)
726 		eeh_ops->restore_config(pdn);
727 
728 	/* The caller should restore state for the specified device */
729 	if (pdev != dev)
730 		pci_restore_state(pdev);
731 
732 	return NULL;
733 }
734 
735 /**
736  * pcibios_set_pcie_reset_state - Set PCI-E reset state
737  * @dev: pci device struct
738  * @state: reset state to enter
739  *
740  * Return value:
741  * 	0 if success
742  */
743 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
744 {
745 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
746 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
747 
748 	if (!pe) {
749 		pr_err("%s: No PE found on PCI device %s\n",
750 			__func__, pci_name(dev));
751 		return -EINVAL;
752 	}
753 
754 	switch (state) {
755 	case pcie_deassert_reset:
756 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
757 		eeh_unfreeze_pe(pe, false);
758 		if (!(pe->type & EEH_PE_VF))
759 			eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
760 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
761 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
762 		break;
763 	case pcie_hot_reset:
764 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
765 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
766 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
767 		if (!(pe->type & EEH_PE_VF))
768 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
769 		eeh_ops->reset(pe, EEH_RESET_HOT);
770 		break;
771 	case pcie_warm_reset:
772 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
773 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
774 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
775 		if (!(pe->type & EEH_PE_VF))
776 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
777 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
778 		break;
779 	default:
780 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
781 		return -EINVAL;
782 	};
783 
784 	return 0;
785 }
786 
787 /**
788  * eeh_set_pe_freset - Check the required reset for the indicated device
789  * @data: EEH device
790  * @flag: return value
791  *
792  * Each device might have its preferred reset type: fundamental or
793  * hot reset. The routine is used to collected the information for
794  * the indicated device and its children so that the bunch of the
795  * devices could be reset properly.
796  */
797 static void *eeh_set_dev_freset(void *data, void *flag)
798 {
799 	struct pci_dev *dev;
800 	unsigned int *freset = (unsigned int *)flag;
801 	struct eeh_dev *edev = (struct eeh_dev *)data;
802 
803 	dev = eeh_dev_to_pci_dev(edev);
804 	if (dev)
805 		*freset |= dev->needs_freset;
806 
807 	return NULL;
808 }
809 
810 /**
811  * eeh_pe_reset_full - Complete a full reset process on the indicated PE
812  * @pe: EEH PE
813  *
814  * This function executes a full reset procedure on a PE, including setting
815  * the appropriate flags, performing a fundamental or hot reset, and then
816  * deactivating the reset status.  It is designed to be used within the EEH
817  * subsystem, as opposed to eeh_pe_reset which is exported to drivers and
818  * only performs a single operation at a time.
819  *
820  * This function will attempt to reset a PE three times before failing.
821  */
822 int eeh_pe_reset_full(struct eeh_pe *pe)
823 {
824 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
825 	int reset_state = (EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
826 	int type = EEH_RESET_HOT;
827 	unsigned int freset = 0;
828 	int i, state, ret;
829 
830 	/*
831 	 * Determine the type of reset to perform - hot or fundamental.
832 	 * Hot reset is the default operation, unless any device under the
833 	 * PE requires a fundamental reset.
834 	 */
835 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
836 
837 	if (freset)
838 		type = EEH_RESET_FUNDAMENTAL;
839 
840 	/* Mark the PE as in reset state and block config space accesses */
841 	eeh_pe_state_mark(pe, reset_state);
842 
843 	/* Make three attempts at resetting the bus */
844 	for (i = 0; i < 3; i++) {
845 		ret = eeh_pe_reset(pe, type);
846 		if (ret)
847 			break;
848 
849 		ret = eeh_pe_reset(pe, EEH_RESET_DEACTIVATE);
850 		if (ret)
851 			break;
852 
853 		/* Wait until the PE is in a functioning state */
854 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
855 		if ((state & active_flags) == active_flags)
856 			break;
857 
858 		if (state < 0) {
859 			pr_warn("%s: Unrecoverable slot failure on PHB#%x-PE#%x",
860 				__func__, pe->phb->global_number, pe->addr);
861 			ret = -ENOTRECOVERABLE;
862 			break;
863 		}
864 
865 		/* Set error in case this is our last attempt */
866 		ret = -EIO;
867 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
868 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
869 	}
870 
871 	eeh_pe_state_clear(pe, reset_state);
872 	return ret;
873 }
874 
875 /**
876  * eeh_save_bars - Save device bars
877  * @edev: PCI device associated EEH device
878  *
879  * Save the values of the device bars. Unlike the restore
880  * routine, this routine is *not* recursive. This is because
881  * PCI devices are added individually; but, for the restore,
882  * an entire slot is reset at a time.
883  */
884 void eeh_save_bars(struct eeh_dev *edev)
885 {
886 	struct pci_dn *pdn;
887 	int i;
888 
889 	pdn = eeh_dev_to_pdn(edev);
890 	if (!pdn)
891 		return;
892 
893 	for (i = 0; i < 16; i++)
894 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
895 
896 	/*
897 	 * For PCI bridges including root port, we need enable bus
898 	 * master explicitly. Otherwise, it can't fetch IODA table
899 	 * entries correctly. So we cache the bit in advance so that
900 	 * we can restore it after reset, either PHB range or PE range.
901 	 */
902 	if (edev->mode & EEH_DEV_BRIDGE)
903 		edev->config_space[1] |= PCI_COMMAND_MASTER;
904 }
905 
906 /**
907  * eeh_ops_register - Register platform dependent EEH operations
908  * @ops: platform dependent EEH operations
909  *
910  * Register the platform dependent EEH operation callback
911  * functions. The platform should call this function before
912  * any other EEH operations.
913  */
914 int __init eeh_ops_register(struct eeh_ops *ops)
915 {
916 	if (!ops->name) {
917 		pr_warn("%s: Invalid EEH ops name for %p\n",
918 			__func__, ops);
919 		return -EINVAL;
920 	}
921 
922 	if (eeh_ops && eeh_ops != ops) {
923 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
924 			__func__, eeh_ops->name, ops->name);
925 		return -EEXIST;
926 	}
927 
928 	eeh_ops = ops;
929 
930 	return 0;
931 }
932 
933 /**
934  * eeh_ops_unregister - Unreigster platform dependent EEH operations
935  * @name: name of EEH platform operations
936  *
937  * Unregister the platform dependent EEH operation callback
938  * functions.
939  */
940 int __exit eeh_ops_unregister(const char *name)
941 {
942 	if (!name || !strlen(name)) {
943 		pr_warn("%s: Invalid EEH ops name\n",
944 			__func__);
945 		return -EINVAL;
946 	}
947 
948 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
949 		eeh_ops = NULL;
950 		return 0;
951 	}
952 
953 	return -EEXIST;
954 }
955 
956 static int eeh_reboot_notifier(struct notifier_block *nb,
957 			       unsigned long action, void *unused)
958 {
959 	eeh_clear_flag(EEH_ENABLED);
960 	return NOTIFY_DONE;
961 }
962 
963 static struct notifier_block eeh_reboot_nb = {
964 	.notifier_call = eeh_reboot_notifier,
965 };
966 
967 /**
968  * eeh_init - EEH initialization
969  *
970  * Initialize EEH by trying to enable it for all of the adapters in the system.
971  * As a side effect we can determine here if eeh is supported at all.
972  * Note that we leave EEH on so failed config cycles won't cause a machine
973  * check.  If a user turns off EEH for a particular adapter they are really
974  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
975  * grant access to a slot if EEH isn't enabled, and so we always enable
976  * EEH for all slots/all devices.
977  *
978  * The eeh-force-off option disables EEH checking globally, for all slots.
979  * Even if force-off is set, the EEH hardware is still enabled, so that
980  * newer systems can boot.
981  */
982 int eeh_init(void)
983 {
984 	struct pci_controller *hose, *tmp;
985 	struct pci_dn *pdn;
986 	static int cnt = 0;
987 	int ret = 0;
988 
989 	/*
990 	 * We have to delay the initialization on PowerNV after
991 	 * the PCI hierarchy tree has been built because the PEs
992 	 * are figured out based on PCI devices instead of device
993 	 * tree nodes
994 	 */
995 	if (machine_is(powernv) && cnt++ <= 0)
996 		return ret;
997 
998 	/* Register reboot notifier */
999 	ret = register_reboot_notifier(&eeh_reboot_nb);
1000 	if (ret) {
1001 		pr_warn("%s: Failed to register notifier (%d)\n",
1002 			__func__, ret);
1003 		return ret;
1004 	}
1005 
1006 	/* call platform initialization function */
1007 	if (!eeh_ops) {
1008 		pr_warn("%s: Platform EEH operation not found\n",
1009 			__func__);
1010 		return -EEXIST;
1011 	} else if ((ret = eeh_ops->init()))
1012 		return ret;
1013 
1014 	/* Initialize EEH event */
1015 	ret = eeh_event_init();
1016 	if (ret)
1017 		return ret;
1018 
1019 	/* Enable EEH for all adapters */
1020 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1021 		pdn = hose->pci_data;
1022 		traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1023 	}
1024 
1025 	/*
1026 	 * Call platform post-initialization. Actually, It's good chance
1027 	 * to inform platform that EEH is ready to supply service if the
1028 	 * I/O cache stuff has been built up.
1029 	 */
1030 	if (eeh_ops->post_init) {
1031 		ret = eeh_ops->post_init();
1032 		if (ret)
1033 			return ret;
1034 	}
1035 
1036 	if (eeh_enabled())
1037 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1038 	else
1039 		pr_info("EEH: No capable adapters found\n");
1040 
1041 	return ret;
1042 }
1043 
1044 core_initcall_sync(eeh_init);
1045 
1046 /**
1047  * eeh_add_device_early - Enable EEH for the indicated device node
1048  * @pdn: PCI device node for which to set up EEH
1049  *
1050  * This routine must be used to perform EEH initialization for PCI
1051  * devices that were added after system boot (e.g. hotplug, dlpar).
1052  * This routine must be called before any i/o is performed to the
1053  * adapter (inluding any config-space i/o).
1054  * Whether this actually enables EEH or not for this device depends
1055  * on the CEC architecture, type of the device, on earlier boot
1056  * command-line arguments & etc.
1057  */
1058 void eeh_add_device_early(struct pci_dn *pdn)
1059 {
1060 	struct pci_controller *phb;
1061 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1062 
1063 	if (!edev)
1064 		return;
1065 
1066 	if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1067 		return;
1068 
1069 	/* USB Bus children of PCI devices will not have BUID's */
1070 	phb = edev->phb;
1071 	if (NULL == phb ||
1072 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1073 		return;
1074 
1075 	eeh_ops->probe(pdn, NULL);
1076 }
1077 
1078 /**
1079  * eeh_add_device_tree_early - Enable EEH for the indicated device
1080  * @pdn: PCI device node
1081  *
1082  * This routine must be used to perform EEH initialization for the
1083  * indicated PCI device that was added after system boot (e.g.
1084  * hotplug, dlpar).
1085  */
1086 void eeh_add_device_tree_early(struct pci_dn *pdn)
1087 {
1088 	struct pci_dn *n;
1089 
1090 	if (!pdn)
1091 		return;
1092 
1093 	list_for_each_entry(n, &pdn->child_list, list)
1094 		eeh_add_device_tree_early(n);
1095 	eeh_add_device_early(pdn);
1096 }
1097 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1098 
1099 /**
1100  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1101  * @dev: pci device for which to set up EEH
1102  *
1103  * This routine must be used to complete EEH initialization for PCI
1104  * devices that were added after system boot (e.g. hotplug, dlpar).
1105  */
1106 void eeh_add_device_late(struct pci_dev *dev)
1107 {
1108 	struct pci_dn *pdn;
1109 	struct eeh_dev *edev;
1110 
1111 	if (!dev || !eeh_enabled())
1112 		return;
1113 
1114 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1115 
1116 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1117 	edev = pdn_to_eeh_dev(pdn);
1118 	if (edev->pdev == dev) {
1119 		pr_debug("EEH: Already referenced !\n");
1120 		return;
1121 	}
1122 
1123 	/*
1124 	 * The EEH cache might not be removed correctly because of
1125 	 * unbalanced kref to the device during unplug time, which
1126 	 * relies on pcibios_release_device(). So we have to remove
1127 	 * that here explicitly.
1128 	 */
1129 	if (edev->pdev) {
1130 		eeh_rmv_from_parent_pe(edev);
1131 		eeh_addr_cache_rmv_dev(edev->pdev);
1132 		eeh_sysfs_remove_device(edev->pdev);
1133 		edev->mode &= ~EEH_DEV_SYSFS;
1134 
1135 		/*
1136 		 * We definitely should have the PCI device removed
1137 		 * though it wasn't correctly. So we needn't call
1138 		 * into error handler afterwards.
1139 		 */
1140 		edev->mode |= EEH_DEV_NO_HANDLER;
1141 
1142 		edev->pdev = NULL;
1143 		dev->dev.archdata.edev = NULL;
1144 	}
1145 
1146 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1147 		eeh_ops->probe(pdn, NULL);
1148 
1149 	edev->pdev = dev;
1150 	dev->dev.archdata.edev = edev;
1151 
1152 	eeh_addr_cache_insert_dev(dev);
1153 }
1154 
1155 /**
1156  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1157  * @bus: PCI bus
1158  *
1159  * This routine must be used to perform EEH initialization for PCI
1160  * devices which are attached to the indicated PCI bus. The PCI bus
1161  * is added after system boot through hotplug or dlpar.
1162  */
1163 void eeh_add_device_tree_late(struct pci_bus *bus)
1164 {
1165 	struct pci_dev *dev;
1166 
1167 	list_for_each_entry(dev, &bus->devices, bus_list) {
1168 		eeh_add_device_late(dev);
1169 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1170 			struct pci_bus *subbus = dev->subordinate;
1171 			if (subbus)
1172 				eeh_add_device_tree_late(subbus);
1173 		}
1174 	}
1175 }
1176 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1177 
1178 /**
1179  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1180  * @bus: PCI bus
1181  *
1182  * This routine must be used to add EEH sysfs files for PCI
1183  * devices which are attached to the indicated PCI bus. The PCI bus
1184  * is added after system boot through hotplug or dlpar.
1185  */
1186 void eeh_add_sysfs_files(struct pci_bus *bus)
1187 {
1188 	struct pci_dev *dev;
1189 
1190 	list_for_each_entry(dev, &bus->devices, bus_list) {
1191 		eeh_sysfs_add_device(dev);
1192 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1193 			struct pci_bus *subbus = dev->subordinate;
1194 			if (subbus)
1195 				eeh_add_sysfs_files(subbus);
1196 		}
1197 	}
1198 }
1199 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1200 
1201 /**
1202  * eeh_remove_device - Undo EEH setup for the indicated pci device
1203  * @dev: pci device to be removed
1204  *
1205  * This routine should be called when a device is removed from
1206  * a running system (e.g. by hotplug or dlpar).  It unregisters
1207  * the PCI device from the EEH subsystem.  I/O errors affecting
1208  * this device will no longer be detected after this call; thus,
1209  * i/o errors affecting this slot may leave this device unusable.
1210  */
1211 void eeh_remove_device(struct pci_dev *dev)
1212 {
1213 	struct eeh_dev *edev;
1214 
1215 	if (!dev || !eeh_enabled())
1216 		return;
1217 	edev = pci_dev_to_eeh_dev(dev);
1218 
1219 	/* Unregister the device with the EEH/PCI address search system */
1220 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1221 
1222 	if (!edev || !edev->pdev || !edev->pe) {
1223 		pr_debug("EEH: Not referenced !\n");
1224 		return;
1225 	}
1226 
1227 	/*
1228 	 * During the hotplug for EEH error recovery, we need the EEH
1229 	 * device attached to the parent PE in order for BAR restore
1230 	 * a bit later. So we keep it for BAR restore and remove it
1231 	 * from the parent PE during the BAR resotre.
1232 	 */
1233 	edev->pdev = NULL;
1234 
1235 	/*
1236 	 * The flag "in_error" is used to trace EEH devices for VFs
1237 	 * in error state or not. It's set in eeh_report_error(). If
1238 	 * it's not set, eeh_report_{reset,resume}() won't be called
1239 	 * for the VF EEH device.
1240 	 */
1241 	edev->in_error = false;
1242 	dev->dev.archdata.edev = NULL;
1243 	if (!(edev->pe->state & EEH_PE_KEEP))
1244 		eeh_rmv_from_parent_pe(edev);
1245 	else
1246 		edev->mode |= EEH_DEV_DISCONNECTED;
1247 
1248 	/*
1249 	 * We're removing from the PCI subsystem, that means
1250 	 * the PCI device driver can't support EEH or not
1251 	 * well. So we rely on hotplug completely to do recovery
1252 	 * for the specific PCI device.
1253 	 */
1254 	edev->mode |= EEH_DEV_NO_HANDLER;
1255 
1256 	eeh_addr_cache_rmv_dev(dev);
1257 	eeh_sysfs_remove_device(dev);
1258 	edev->mode &= ~EEH_DEV_SYSFS;
1259 }
1260 
1261 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1262 {
1263 	int ret;
1264 
1265 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1266 	if (ret) {
1267 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1268 			__func__, ret, pe->phb->global_number, pe->addr);
1269 		return ret;
1270 	}
1271 
1272 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1273 	if (ret) {
1274 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1275 			__func__, ret, pe->phb->global_number, pe->addr);
1276 		return ret;
1277 	}
1278 
1279 	/* Clear software isolated state */
1280 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1281 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1282 
1283 	return ret;
1284 }
1285 
1286 
1287 static struct pci_device_id eeh_reset_ids[] = {
1288 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1289 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1290 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1291 	{ 0 }
1292 };
1293 
1294 static int eeh_pe_change_owner(struct eeh_pe *pe)
1295 {
1296 	struct eeh_dev *edev, *tmp;
1297 	struct pci_dev *pdev;
1298 	struct pci_device_id *id;
1299 	int flags, ret;
1300 
1301 	/* Check PE state */
1302 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1303 	ret = eeh_ops->get_state(pe, NULL);
1304 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1305 		return 0;
1306 
1307 	/* Unfrozen PE, nothing to do */
1308 	if ((ret & flags) == flags)
1309 		return 0;
1310 
1311 	/* Frozen PE, check if it needs PE level reset */
1312 	eeh_pe_for_each_dev(pe, edev, tmp) {
1313 		pdev = eeh_dev_to_pci_dev(edev);
1314 		if (!pdev)
1315 			continue;
1316 
1317 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1318 			if (id->vendor != PCI_ANY_ID &&
1319 			    id->vendor != pdev->vendor)
1320 				continue;
1321 			if (id->device != PCI_ANY_ID &&
1322 			    id->device != pdev->device)
1323 				continue;
1324 			if (id->subvendor != PCI_ANY_ID &&
1325 			    id->subvendor != pdev->subsystem_vendor)
1326 				continue;
1327 			if (id->subdevice != PCI_ANY_ID &&
1328 			    id->subdevice != pdev->subsystem_device)
1329 				continue;
1330 
1331 			return eeh_pe_reset_and_recover(pe);
1332 		}
1333 	}
1334 
1335 	return eeh_unfreeze_pe(pe, true);
1336 }
1337 
1338 /**
1339  * eeh_dev_open - Increase count of pass through devices for PE
1340  * @pdev: PCI device
1341  *
1342  * Increase count of passed through devices for the indicated
1343  * PE. In the result, the EEH errors detected on the PE won't be
1344  * reported. The PE owner will be responsible for detection
1345  * and recovery.
1346  */
1347 int eeh_dev_open(struct pci_dev *pdev)
1348 {
1349 	struct eeh_dev *edev;
1350 	int ret = -ENODEV;
1351 
1352 	mutex_lock(&eeh_dev_mutex);
1353 
1354 	/* No PCI device ? */
1355 	if (!pdev)
1356 		goto out;
1357 
1358 	/* No EEH device or PE ? */
1359 	edev = pci_dev_to_eeh_dev(pdev);
1360 	if (!edev || !edev->pe)
1361 		goto out;
1362 
1363 	/*
1364 	 * The PE might have been put into frozen state, but we
1365 	 * didn't detect that yet. The passed through PCI devices
1366 	 * in frozen PE won't work properly. Clear the frozen state
1367 	 * in advance.
1368 	 */
1369 	ret = eeh_pe_change_owner(edev->pe);
1370 	if (ret)
1371 		goto out;
1372 
1373 	/* Increase PE's pass through count */
1374 	atomic_inc(&edev->pe->pass_dev_cnt);
1375 	mutex_unlock(&eeh_dev_mutex);
1376 
1377 	return 0;
1378 out:
1379 	mutex_unlock(&eeh_dev_mutex);
1380 	return ret;
1381 }
1382 EXPORT_SYMBOL_GPL(eeh_dev_open);
1383 
1384 /**
1385  * eeh_dev_release - Decrease count of pass through devices for PE
1386  * @pdev: PCI device
1387  *
1388  * Decrease count of pass through devices for the indicated PE. If
1389  * there is no passed through device in PE, the EEH errors detected
1390  * on the PE will be reported and handled as usual.
1391  */
1392 void eeh_dev_release(struct pci_dev *pdev)
1393 {
1394 	struct eeh_dev *edev;
1395 
1396 	mutex_lock(&eeh_dev_mutex);
1397 
1398 	/* No PCI device ? */
1399 	if (!pdev)
1400 		goto out;
1401 
1402 	/* No EEH device ? */
1403 	edev = pci_dev_to_eeh_dev(pdev);
1404 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1405 		goto out;
1406 
1407 	/* Decrease PE's pass through count */
1408 	WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1409 	eeh_pe_change_owner(edev->pe);
1410 out:
1411 	mutex_unlock(&eeh_dev_mutex);
1412 }
1413 EXPORT_SYMBOL(eeh_dev_release);
1414 
1415 #ifdef CONFIG_IOMMU_API
1416 
1417 static int dev_has_iommu_table(struct device *dev, void *data)
1418 {
1419 	struct pci_dev *pdev = to_pci_dev(dev);
1420 	struct pci_dev **ppdev = data;
1421 
1422 	if (!dev)
1423 		return 0;
1424 
1425 	if (dev->iommu_group) {
1426 		*ppdev = pdev;
1427 		return 1;
1428 	}
1429 
1430 	return 0;
1431 }
1432 
1433 /**
1434  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1435  * @group: IOMMU group
1436  *
1437  * The routine is called to convert IOMMU group to EEH PE.
1438  */
1439 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1440 {
1441 	struct pci_dev *pdev = NULL;
1442 	struct eeh_dev *edev;
1443 	int ret;
1444 
1445 	/* No IOMMU group ? */
1446 	if (!group)
1447 		return NULL;
1448 
1449 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1450 	if (!ret || !pdev)
1451 		return NULL;
1452 
1453 	/* No EEH device or PE ? */
1454 	edev = pci_dev_to_eeh_dev(pdev);
1455 	if (!edev || !edev->pe)
1456 		return NULL;
1457 
1458 	return edev->pe;
1459 }
1460 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1461 
1462 #endif /* CONFIG_IOMMU_API */
1463 
1464 /**
1465  * eeh_pe_set_option - Set options for the indicated PE
1466  * @pe: EEH PE
1467  * @option: requested option
1468  *
1469  * The routine is called to enable or disable EEH functionality
1470  * on the indicated PE, to enable IO or DMA for the frozen PE.
1471  */
1472 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1473 {
1474 	int ret = 0;
1475 
1476 	/* Invalid PE ? */
1477 	if (!pe)
1478 		return -ENODEV;
1479 
1480 	/*
1481 	 * EEH functionality could possibly be disabled, just
1482 	 * return error for the case. And the EEH functinality
1483 	 * isn't expected to be disabled on one specific PE.
1484 	 */
1485 	switch (option) {
1486 	case EEH_OPT_ENABLE:
1487 		if (eeh_enabled()) {
1488 			ret = eeh_pe_change_owner(pe);
1489 			break;
1490 		}
1491 		ret = -EIO;
1492 		break;
1493 	case EEH_OPT_DISABLE:
1494 		break;
1495 	case EEH_OPT_THAW_MMIO:
1496 	case EEH_OPT_THAW_DMA:
1497 	case EEH_OPT_FREEZE_PE:
1498 		if (!eeh_ops || !eeh_ops->set_option) {
1499 			ret = -ENOENT;
1500 			break;
1501 		}
1502 
1503 		ret = eeh_pci_enable(pe, option);
1504 		break;
1505 	default:
1506 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1507 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1508 		ret = -EINVAL;
1509 	}
1510 
1511 	return ret;
1512 }
1513 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1514 
1515 /**
1516  * eeh_pe_get_state - Retrieve PE's state
1517  * @pe: EEH PE
1518  *
1519  * Retrieve the PE's state, which includes 3 aspects: enabled
1520  * DMA, enabled IO and asserted reset.
1521  */
1522 int eeh_pe_get_state(struct eeh_pe *pe)
1523 {
1524 	int result, ret = 0;
1525 	bool rst_active, dma_en, mmio_en;
1526 
1527 	/* Existing PE ? */
1528 	if (!pe)
1529 		return -ENODEV;
1530 
1531 	if (!eeh_ops || !eeh_ops->get_state)
1532 		return -ENOENT;
1533 
1534 	/*
1535 	 * If the parent PE is owned by the host kernel and is undergoing
1536 	 * error recovery, we should return the PE state as temporarily
1537 	 * unavailable so that the error recovery on the guest is suspended
1538 	 * until the recovery completes on the host.
1539 	 */
1540 	if (pe->parent &&
1541 	    !(pe->state & EEH_PE_REMOVED) &&
1542 	    (pe->parent->state & (EEH_PE_ISOLATED | EEH_PE_RECOVERING)))
1543 		return EEH_PE_STATE_UNAVAIL;
1544 
1545 	result = eeh_ops->get_state(pe, NULL);
1546 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1547 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1548 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1549 
1550 	if (rst_active)
1551 		ret = EEH_PE_STATE_RESET;
1552 	else if (dma_en && mmio_en)
1553 		ret = EEH_PE_STATE_NORMAL;
1554 	else if (!dma_en && !mmio_en)
1555 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1556 	else if (!dma_en && mmio_en)
1557 		ret = EEH_PE_STATE_STOPPED_DMA;
1558 	else
1559 		ret = EEH_PE_STATE_UNAVAIL;
1560 
1561 	return ret;
1562 }
1563 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1564 
1565 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1566 {
1567 	struct eeh_dev *edev, *tmp;
1568 	struct pci_dev *pdev;
1569 	int ret = 0;
1570 
1571 	/* Restore config space */
1572 	eeh_pe_restore_bars(pe);
1573 
1574 	/*
1575 	 * Reenable PCI devices as the devices passed
1576 	 * through are always enabled before the reset.
1577 	 */
1578 	eeh_pe_for_each_dev(pe, edev, tmp) {
1579 		pdev = eeh_dev_to_pci_dev(edev);
1580 		if (!pdev)
1581 			continue;
1582 
1583 		ret = pci_reenable_device(pdev);
1584 		if (ret) {
1585 			pr_warn("%s: Failure %d reenabling %s\n",
1586 				__func__, ret, pci_name(pdev));
1587 			return ret;
1588 		}
1589 	}
1590 
1591 	/* The PE is still in frozen state */
1592 	return eeh_unfreeze_pe(pe, true);
1593 }
1594 
1595 
1596 /**
1597  * eeh_pe_reset - Issue PE reset according to specified type
1598  * @pe: EEH PE
1599  * @option: reset type
1600  *
1601  * The routine is called to reset the specified PE with the
1602  * indicated type, either fundamental reset or hot reset.
1603  * PE reset is the most important part for error recovery.
1604  */
1605 int eeh_pe_reset(struct eeh_pe *pe, int option)
1606 {
1607 	int ret = 0;
1608 
1609 	/* Invalid PE ? */
1610 	if (!pe)
1611 		return -ENODEV;
1612 
1613 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1614 		return -ENOENT;
1615 
1616 	switch (option) {
1617 	case EEH_RESET_DEACTIVATE:
1618 		ret = eeh_ops->reset(pe, option);
1619 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1620 		if (ret)
1621 			break;
1622 
1623 		ret = eeh_pe_reenable_devices(pe);
1624 		break;
1625 	case EEH_RESET_HOT:
1626 	case EEH_RESET_FUNDAMENTAL:
1627 		/*
1628 		 * Proactively freeze the PE to drop all MMIO access
1629 		 * during reset, which should be banned as it's always
1630 		 * cause recursive EEH error.
1631 		 */
1632 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1633 
1634 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1635 		ret = eeh_ops->reset(pe, option);
1636 		break;
1637 	default:
1638 		pr_debug("%s: Unsupported option %d\n",
1639 			__func__, option);
1640 		ret = -EINVAL;
1641 	}
1642 
1643 	return ret;
1644 }
1645 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1646 
1647 /**
1648  * eeh_pe_configure - Configure PCI bridges after PE reset
1649  * @pe: EEH PE
1650  *
1651  * The routine is called to restore the PCI config space for
1652  * those PCI devices, especially PCI bridges affected by PE
1653  * reset issued previously.
1654  */
1655 int eeh_pe_configure(struct eeh_pe *pe)
1656 {
1657 	int ret = 0;
1658 
1659 	/* Invalid PE ? */
1660 	if (!pe)
1661 		return -ENODEV;
1662 
1663 	return ret;
1664 }
1665 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1666 
1667 /**
1668  * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1669  * @pe: the indicated PE
1670  * @type: error type
1671  * @function: error function
1672  * @addr: address
1673  * @mask: address mask
1674  *
1675  * The routine is called to inject the specified PCI error, which
1676  * is determined by @type and @function, to the indicated PE for
1677  * testing purpose.
1678  */
1679 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1680 		      unsigned long addr, unsigned long mask)
1681 {
1682 	/* Invalid PE ? */
1683 	if (!pe)
1684 		return -ENODEV;
1685 
1686 	/* Unsupported operation ? */
1687 	if (!eeh_ops || !eeh_ops->err_inject)
1688 		return -ENOENT;
1689 
1690 	/* Check on PCI error type */
1691 	if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1692 		return -EINVAL;
1693 
1694 	/* Check on PCI error function */
1695 	if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1696 		return -EINVAL;
1697 
1698 	return eeh_ops->err_inject(pe, type, func, addr, mask);
1699 }
1700 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1701 
1702 static int proc_eeh_show(struct seq_file *m, void *v)
1703 {
1704 	if (!eeh_enabled()) {
1705 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1706 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1707 	} else {
1708 		seq_printf(m, "EEH Subsystem is enabled\n");
1709 		seq_printf(m,
1710 				"no device=%llu\n"
1711 				"no device node=%llu\n"
1712 				"no config address=%llu\n"
1713 				"check not wanted=%llu\n"
1714 				"eeh_total_mmio_ffs=%llu\n"
1715 				"eeh_false_positives=%llu\n"
1716 				"eeh_slot_resets=%llu\n",
1717 				eeh_stats.no_device,
1718 				eeh_stats.no_dn,
1719 				eeh_stats.no_cfg_addr,
1720 				eeh_stats.ignored_check,
1721 				eeh_stats.total_mmio_ffs,
1722 				eeh_stats.false_positives,
1723 				eeh_stats.slot_resets);
1724 	}
1725 
1726 	return 0;
1727 }
1728 
1729 static int proc_eeh_open(struct inode *inode, struct file *file)
1730 {
1731 	return single_open(file, proc_eeh_show, NULL);
1732 }
1733 
1734 static const struct file_operations proc_eeh_operations = {
1735 	.open      = proc_eeh_open,
1736 	.read      = seq_read,
1737 	.llseek    = seq_lseek,
1738 	.release   = single_release,
1739 };
1740 
1741 #ifdef CONFIG_DEBUG_FS
1742 static int eeh_enable_dbgfs_set(void *data, u64 val)
1743 {
1744 	if (val)
1745 		eeh_clear_flag(EEH_FORCE_DISABLED);
1746 	else
1747 		eeh_add_flag(EEH_FORCE_DISABLED);
1748 
1749 	/* Notify the backend */
1750 	if (eeh_ops->post_init)
1751 		eeh_ops->post_init();
1752 
1753 	return 0;
1754 }
1755 
1756 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1757 {
1758 	if (eeh_enabled())
1759 		*val = 0x1ul;
1760 	else
1761 		*val = 0x0ul;
1762 	return 0;
1763 }
1764 
1765 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1766 {
1767 	eeh_max_freezes = val;
1768 	return 0;
1769 }
1770 
1771 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1772 {
1773 	*val = eeh_max_freezes;
1774 	return 0;
1775 }
1776 
1777 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1778 			eeh_enable_dbgfs_set, "0x%llx\n");
1779 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1780 			eeh_freeze_dbgfs_set, "0x%llx\n");
1781 #endif
1782 
1783 static int __init eeh_init_proc(void)
1784 {
1785 	if (machine_is(pseries) || machine_is(powernv)) {
1786 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1787 #ifdef CONFIG_DEBUG_FS
1788 		debugfs_create_file("eeh_enable", 0600,
1789                                     powerpc_debugfs_root, NULL,
1790                                     &eeh_enable_dbgfs_ops);
1791 		debugfs_create_file("eeh_max_freezes", 0600,
1792 				    powerpc_debugfs_root, NULL,
1793 				    &eeh_freeze_dbgfs_ops);
1794 #endif
1795 	}
1796 
1797 	return 0;
1798 }
1799 __initcall(eeh_init_proc);
1800