xref: /linux/arch/powerpc/kernel/eeh.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1 /*
2  * Copyright IBM Corporation 2001, 2005, 2006
3  * Copyright Dave Engebretsen & Todd Inglett 2001
4  * Copyright Linas Vepstas 2005, 2006
5  * Copyright 2001-2012 IBM Corporation.
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 2 of the License, or
10  * (at your option) any later version.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program; if not, write to the Free Software
19  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
20  *
21  * Please address comments and feedback to Linas Vepstas <linas@austin.ibm.com>
22  */
23 
24 #include <linux/delay.h>
25 #include <linux/sched.h>
26 #include <linux/init.h>
27 #include <linux/list.h>
28 #include <linux/pci.h>
29 #include <linux/iommu.h>
30 #include <linux/proc_fs.h>
31 #include <linux/rbtree.h>
32 #include <linux/reboot.h>
33 #include <linux/seq_file.h>
34 #include <linux/spinlock.h>
35 #include <linux/export.h>
36 #include <linux/of.h>
37 
38 #include <linux/atomic.h>
39 #include <asm/debugfs.h>
40 #include <asm/eeh.h>
41 #include <asm/eeh_event.h>
42 #include <asm/io.h>
43 #include <asm/iommu.h>
44 #include <asm/machdep.h>
45 #include <asm/ppc-pci.h>
46 #include <asm/rtas.h>
47 #include <asm/pte-walk.h>
48 
49 
50 /** Overview:
51  *  EEH, or "Enhanced Error Handling" is a PCI bridge technology for
52  *  dealing with PCI bus errors that can't be dealt with within the
53  *  usual PCI framework, except by check-stopping the CPU.  Systems
54  *  that are designed for high-availability/reliability cannot afford
55  *  to crash due to a "mere" PCI error, thus the need for EEH.
56  *  An EEH-capable bridge operates by converting a detected error
57  *  into a "slot freeze", taking the PCI adapter off-line, making
58  *  the slot behave, from the OS'es point of view, as if the slot
59  *  were "empty": all reads return 0xff's and all writes are silently
60  *  ignored.  EEH slot isolation events can be triggered by parity
61  *  errors on the address or data busses (e.g. during posted writes),
62  *  which in turn might be caused by low voltage on the bus, dust,
63  *  vibration, humidity, radioactivity or plain-old failed hardware.
64  *
65  *  Note, however, that one of the leading causes of EEH slot
66  *  freeze events are buggy device drivers, buggy device microcode,
67  *  or buggy device hardware.  This is because any attempt by the
68  *  device to bus-master data to a memory address that is not
69  *  assigned to the device will trigger a slot freeze.   (The idea
70  *  is to prevent devices-gone-wild from corrupting system memory).
71  *  Buggy hardware/drivers will have a miserable time co-existing
72  *  with EEH.
73  *
74  *  Ideally, a PCI device driver, when suspecting that an isolation
75  *  event has occurred (e.g. by reading 0xff's), will then ask EEH
76  *  whether this is the case, and then take appropriate steps to
77  *  reset the PCI slot, the PCI device, and then resume operations.
78  *  However, until that day,  the checking is done here, with the
79  *  eeh_check_failure() routine embedded in the MMIO macros.  If
80  *  the slot is found to be isolated, an "EEH Event" is synthesized
81  *  and sent out for processing.
82  */
83 
84 /* If a device driver keeps reading an MMIO register in an interrupt
85  * handler after a slot isolation event, it might be broken.
86  * This sets the threshold for how many read attempts we allow
87  * before printing an error message.
88  */
89 #define EEH_MAX_FAILS	2100000
90 
91 /* Time to wait for a PCI slot to report status, in milliseconds */
92 #define PCI_BUS_RESET_WAIT_MSEC (5*60*1000)
93 
94 /*
95  * EEH probe mode support, which is part of the flags,
96  * is to support multiple platforms for EEH. Some platforms
97  * like pSeries do PCI emunation based on device tree.
98  * However, other platforms like powernv probe PCI devices
99  * from hardware. The flag is used to distinguish that.
100  * In addition, struct eeh_ops::probe would be invoked for
101  * particular OF node or PCI device so that the corresponding
102  * PE would be created there.
103  */
104 int eeh_subsystem_flags;
105 EXPORT_SYMBOL(eeh_subsystem_flags);
106 
107 /*
108  * EEH allowed maximal frozen times. If one particular PE's
109  * frozen count in last hour exceeds this limit, the PE will
110  * be forced to be offline permanently.
111  */
112 int eeh_max_freezes = 5;
113 
114 /* Platform dependent EEH operations */
115 struct eeh_ops *eeh_ops = NULL;
116 
117 /* Lock to avoid races due to multiple reports of an error */
118 DEFINE_RAW_SPINLOCK(confirm_error_lock);
119 EXPORT_SYMBOL_GPL(confirm_error_lock);
120 
121 /* Lock to protect passed flags */
122 static DEFINE_MUTEX(eeh_dev_mutex);
123 
124 /* Buffer for reporting pci register dumps. Its here in BSS, and
125  * not dynamically alloced, so that it ends up in RMO where RTAS
126  * can access it.
127  */
128 #define EEH_PCI_REGS_LOG_LEN 8192
129 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
130 
131 /*
132  * The struct is used to maintain the EEH global statistic
133  * information. Besides, the EEH global statistics will be
134  * exported to user space through procfs
135  */
136 struct eeh_stats {
137 	u64 no_device;		/* PCI device not found		*/
138 	u64 no_dn;		/* OF node not found		*/
139 	u64 no_cfg_addr;	/* Config address not found	*/
140 	u64 ignored_check;	/* EEH check skipped		*/
141 	u64 total_mmio_ffs;	/* Total EEH checks		*/
142 	u64 false_positives;	/* Unnecessary EEH checks	*/
143 	u64 slot_resets;	/* PE reset			*/
144 };
145 
146 static struct eeh_stats eeh_stats;
147 
148 static int __init eeh_setup(char *str)
149 {
150 	if (!strcmp(str, "off"))
151 		eeh_add_flag(EEH_FORCE_DISABLED);
152 	else if (!strcmp(str, "early_log"))
153 		eeh_add_flag(EEH_EARLY_DUMP_LOG);
154 
155 	return 1;
156 }
157 __setup("eeh=", eeh_setup);
158 
159 /*
160  * This routine captures assorted PCI configuration space data
161  * for the indicated PCI device, and puts them into a buffer
162  * for RTAS error logging.
163  */
164 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
165 {
166 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
167 	u32 cfg;
168 	int cap, i;
169 	int n = 0, l = 0;
170 	char buffer[128];
171 
172 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x.%01x\n",
173 		       pdn->phb->global_number, pdn->busno,
174 		       PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
175 	pr_warn("EEH: of node=%04x:%02x:%02x.%01x\n",
176 		pdn->phb->global_number, pdn->busno,
177 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
178 
179 	eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
180 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
181 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
182 
183 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
184 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
185 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
186 
187 	/* Gather bridge-specific registers */
188 	if (edev->mode & EEH_DEV_BRIDGE) {
189 		eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
190 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
191 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
192 
193 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
194 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
195 		pr_warn("EEH: Bridge control: %04x\n", cfg);
196 	}
197 
198 	/* Dump out the PCI-X command and status regs */
199 	cap = edev->pcix_cap;
200 	if (cap) {
201 		eeh_ops->read_config(pdn, cap, 4, &cfg);
202 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
203 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
204 
205 		eeh_ops->read_config(pdn, cap+4, 4, &cfg);
206 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
207 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
208 	}
209 
210 	/* If PCI-E capable, dump PCI-E cap 10 */
211 	cap = edev->pcie_cap;
212 	if (cap) {
213 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
214 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
215 
216 		for (i=0; i<=8; i++) {
217 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
218 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
219 
220 			if ((i % 4) == 0) {
221 				if (i != 0)
222 					pr_warn("%s\n", buffer);
223 
224 				l = scnprintf(buffer, sizeof(buffer),
225 					      "EEH: PCI-E %02x: %08x ",
226 					      4*i, cfg);
227 			} else {
228 				l += scnprintf(buffer+l, sizeof(buffer)-l,
229 					       "%08x ", cfg);
230 			}
231 
232 		}
233 
234 		pr_warn("%s\n", buffer);
235 	}
236 
237 	/* If AER capable, dump it */
238 	cap = edev->aer_cap;
239 	if (cap) {
240 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
241 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
242 
243 		for (i=0; i<=13; i++) {
244 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
245 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
246 
247 			if ((i % 4) == 0) {
248 				if (i != 0)
249 					pr_warn("%s\n", buffer);
250 
251 				l = scnprintf(buffer, sizeof(buffer),
252 					      "EEH: PCI-E AER %02x: %08x ",
253 					      4*i, cfg);
254 			} else {
255 				l += scnprintf(buffer+l, sizeof(buffer)-l,
256 					       "%08x ", cfg);
257 			}
258 		}
259 
260 		pr_warn("%s\n", buffer);
261 	}
262 
263 	return n;
264 }
265 
266 static void *eeh_dump_pe_log(struct eeh_pe *pe, void *flag)
267 {
268 	struct eeh_dev *edev, *tmp;
269 	size_t *plen = flag;
270 
271 	eeh_pe_for_each_dev(pe, edev, tmp)
272 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
273 					  EEH_PCI_REGS_LOG_LEN - *plen);
274 
275 	return NULL;
276 }
277 
278 /**
279  * eeh_slot_error_detail - Generate combined log including driver log and error log
280  * @pe: EEH PE
281  * @severity: temporary or permanent error log
282  *
283  * This routine should be called to generate the combined log, which
284  * is comprised of driver log and error log. The driver log is figured
285  * out from the config space of the corresponding PCI device, while
286  * the error log is fetched through platform dependent function call.
287  */
288 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
289 {
290 	size_t loglen = 0;
291 
292 	/*
293 	 * When the PHB is fenced or dead, it's pointless to collect
294 	 * the data from PCI config space because it should return
295 	 * 0xFF's. For ER, we still retrieve the data from the PCI
296 	 * config space.
297 	 *
298 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
299 	 * 0xFF's is always returned from PCI config space.
300 	 *
301 	 * When the @severity is EEH_LOG_PERM, the PE is going to be
302 	 * removed. Prior to that, the drivers for devices included in
303 	 * the PE will be closed. The drivers rely on working IO path
304 	 * to bring the devices to quiet state. Otherwise, PCI traffic
305 	 * from those devices after they are removed is like to cause
306 	 * another unexpected EEH error.
307 	 */
308 	if (!(pe->type & EEH_PE_PHB)) {
309 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG) ||
310 		    severity == EEH_LOG_PERM)
311 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
312 
313 		/*
314 		 * The config space of some PCI devices can't be accessed
315 		 * when their PEs are in frozen state. Otherwise, fenced
316 		 * PHB might be seen. Those PEs are identified with flag
317 		 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
318 		 * is set automatically when the PE is put to EEH_PE_ISOLATED.
319 		 *
320 		 * Restoring BARs possibly triggers PCI config access in
321 		 * (OPAL) firmware and then causes fenced PHB. If the
322 		 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
323 		 * pointless to restore BARs and dump config space.
324 		 */
325 		eeh_ops->configure_bridge(pe);
326 		if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
327 			eeh_pe_restore_bars(pe);
328 
329 			pci_regs_buf[0] = 0;
330 			eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
331 		}
332 	}
333 
334 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
335 }
336 
337 /**
338  * eeh_token_to_phys - Convert EEH address token to phys address
339  * @token: I/O token, should be address in the form 0xA....
340  *
341  * This routine should be called to convert virtual I/O address
342  * to physical one.
343  */
344 static inline unsigned long eeh_token_to_phys(unsigned long token)
345 {
346 	pte_t *ptep;
347 	unsigned long pa;
348 	int hugepage_shift;
349 
350 	/*
351 	 * We won't find hugepages here(this is iomem). Hence we are not
352 	 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
353 	 * page table free, because of init_mm.
354 	 */
355 	ptep = find_init_mm_pte(token, &hugepage_shift);
356 	if (!ptep)
357 		return token;
358 	WARN_ON(hugepage_shift);
359 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
360 
361 	return pa | (token & (PAGE_SIZE-1));
362 }
363 
364 /*
365  * On PowerNV platform, we might already have fenced PHB there.
366  * For that case, it's meaningless to recover frozen PE. Intead,
367  * We have to handle fenced PHB firstly.
368  */
369 static int eeh_phb_check_failure(struct eeh_pe *pe)
370 {
371 	struct eeh_pe *phb_pe;
372 	unsigned long flags;
373 	int ret;
374 
375 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
376 		return -EPERM;
377 
378 	/* Find the PHB PE */
379 	phb_pe = eeh_phb_pe_get(pe->phb);
380 	if (!phb_pe) {
381 		pr_warn("%s Can't find PE for PHB#%x\n",
382 			__func__, pe->phb->global_number);
383 		return -EEXIST;
384 	}
385 
386 	/* If the PHB has been in problematic state */
387 	eeh_serialize_lock(&flags);
388 	if (phb_pe->state & EEH_PE_ISOLATED) {
389 		ret = 0;
390 		goto out;
391 	}
392 
393 	/* Check PHB state */
394 	ret = eeh_ops->get_state(phb_pe, NULL);
395 	if ((ret < 0) ||
396 	    (ret == EEH_STATE_NOT_SUPPORT) || eeh_state_active(ret)) {
397 		ret = 0;
398 		goto out;
399 	}
400 
401 	/* Isolate the PHB and send event */
402 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
403 	eeh_serialize_unlock(flags);
404 
405 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
406 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
407 	dump_stack();
408 	eeh_send_failure_event(phb_pe);
409 
410 	return 1;
411 out:
412 	eeh_serialize_unlock(flags);
413 	return ret;
414 }
415 
416 /**
417  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
418  * @edev: eeh device
419  *
420  * Check for an EEH failure for the given device node.  Call this
421  * routine if the result of a read was all 0xff's and you want to
422  * find out if this is due to an EEH slot freeze.  This routine
423  * will query firmware for the EEH status.
424  *
425  * Returns 0 if there has not been an EEH error; otherwise returns
426  * a non-zero value and queues up a slot isolation event notification.
427  *
428  * It is safe to call this routine in an interrupt context.
429  */
430 int eeh_dev_check_failure(struct eeh_dev *edev)
431 {
432 	int ret;
433 	unsigned long flags;
434 	struct device_node *dn;
435 	struct pci_dev *dev;
436 	struct eeh_pe *pe, *parent_pe, *phb_pe;
437 	int rc = 0;
438 	const char *location = NULL;
439 
440 	eeh_stats.total_mmio_ffs++;
441 
442 	if (!eeh_enabled())
443 		return 0;
444 
445 	if (!edev) {
446 		eeh_stats.no_dn++;
447 		return 0;
448 	}
449 	dev = eeh_dev_to_pci_dev(edev);
450 	pe = eeh_dev_to_pe(edev);
451 
452 	/* Access to IO BARs might get this far and still not want checking. */
453 	if (!pe) {
454 		eeh_stats.ignored_check++;
455 		pr_debug("EEH: Ignored check for %s\n",
456 			eeh_pci_name(dev));
457 		return 0;
458 	}
459 
460 	if (!pe->addr && !pe->config_addr) {
461 		eeh_stats.no_cfg_addr++;
462 		return 0;
463 	}
464 
465 	/*
466 	 * On PowerNV platform, we might already have fenced PHB
467 	 * there and we need take care of that firstly.
468 	 */
469 	ret = eeh_phb_check_failure(pe);
470 	if (ret > 0)
471 		return ret;
472 
473 	/*
474 	 * If the PE isn't owned by us, we shouldn't check the
475 	 * state. Instead, let the owner handle it if the PE has
476 	 * been frozen.
477 	 */
478 	if (eeh_pe_passed(pe))
479 		return 0;
480 
481 	/* If we already have a pending isolation event for this
482 	 * slot, we know it's bad already, we don't need to check.
483 	 * Do this checking under a lock; as multiple PCI devices
484 	 * in one slot might report errors simultaneously, and we
485 	 * only want one error recovery routine running.
486 	 */
487 	eeh_serialize_lock(&flags);
488 	rc = 1;
489 	if (pe->state & EEH_PE_ISOLATED) {
490 		pe->check_count++;
491 		if (pe->check_count % EEH_MAX_FAILS == 0) {
492 			dn = pci_device_to_OF_node(dev);
493 			if (dn)
494 				location = of_get_property(dn, "ibm,loc-code",
495 						NULL);
496 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
497 				"location=%s driver=%s pci addr=%s\n",
498 				pe->check_count,
499 				location ? location : "unknown",
500 				eeh_driver_name(dev), eeh_pci_name(dev));
501 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
502 				eeh_driver_name(dev));
503 			dump_stack();
504 		}
505 		goto dn_unlock;
506 	}
507 
508 	/*
509 	 * Now test for an EEH failure.  This is VERY expensive.
510 	 * Note that the eeh_config_addr may be a parent device
511 	 * in the case of a device behind a bridge, or it may be
512 	 * function zero of a multi-function device.
513 	 * In any case they must share a common PHB.
514 	 */
515 	ret = eeh_ops->get_state(pe, NULL);
516 
517 	/* Note that config-io to empty slots may fail;
518 	 * they are empty when they don't have children.
519 	 * We will punt with the following conditions: Failure to get
520 	 * PE's state, EEH not support and Permanently unavailable
521 	 * state, PE is in good state.
522 	 */
523 	if ((ret < 0) ||
524 	    (ret == EEH_STATE_NOT_SUPPORT) || eeh_state_active(ret)) {
525 		eeh_stats.false_positives++;
526 		pe->false_positives++;
527 		rc = 0;
528 		goto dn_unlock;
529 	}
530 
531 	/*
532 	 * It should be corner case that the parent PE has been
533 	 * put into frozen state as well. We should take care
534 	 * that at first.
535 	 */
536 	parent_pe = pe->parent;
537 	while (parent_pe) {
538 		/* Hit the ceiling ? */
539 		if (parent_pe->type & EEH_PE_PHB)
540 			break;
541 
542 		/* Frozen parent PE ? */
543 		ret = eeh_ops->get_state(parent_pe, NULL);
544 		if (ret > 0 && !eeh_state_active(ret)) {
545 			pe = parent_pe;
546 			pr_err("EEH: Failure of PHB#%x-PE#%x will be handled at parent PHB#%x-PE#%x.\n",
547 			       pe->phb->global_number, pe->addr,
548 			       pe->phb->global_number, parent_pe->addr);
549 		}
550 
551 		/* Next parent level */
552 		parent_pe = parent_pe->parent;
553 	}
554 
555 	eeh_stats.slot_resets++;
556 
557 	/* Avoid repeated reports of this failure, including problems
558 	 * with other functions on this device, and functions under
559 	 * bridges.
560 	 */
561 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
562 	eeh_serialize_unlock(flags);
563 
564 	/* Most EEH events are due to device driver bugs.  Having
565 	 * a stack trace will help the device-driver authors figure
566 	 * out what happened.  So print that out.
567 	 */
568 	phb_pe = eeh_phb_pe_get(pe->phb);
569 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
570 	       pe->phb->global_number, pe->addr);
571 	pr_err("EEH: PE location: %s, PHB location: %s\n",
572 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
573 	dump_stack();
574 
575 	eeh_send_failure_event(pe);
576 
577 	return 1;
578 
579 dn_unlock:
580 	eeh_serialize_unlock(flags);
581 	return rc;
582 }
583 
584 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
585 
586 /**
587  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
588  * @token: I/O address
589  *
590  * Check for an EEH failure at the given I/O address. Call this
591  * routine if the result of a read was all 0xff's and you want to
592  * find out if this is due to an EEH slot freeze event. This routine
593  * will query firmware for the EEH status.
594  *
595  * Note this routine is safe to call in an interrupt context.
596  */
597 int eeh_check_failure(const volatile void __iomem *token)
598 {
599 	unsigned long addr;
600 	struct eeh_dev *edev;
601 
602 	/* Finding the phys addr + pci device; this is pretty quick. */
603 	addr = eeh_token_to_phys((unsigned long __force) token);
604 	edev = eeh_addr_cache_get_dev(addr);
605 	if (!edev) {
606 		eeh_stats.no_device++;
607 		return 0;
608 	}
609 
610 	return eeh_dev_check_failure(edev);
611 }
612 EXPORT_SYMBOL(eeh_check_failure);
613 
614 
615 /**
616  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
617  * @pe: EEH PE
618  *
619  * This routine should be called to reenable frozen MMIO or DMA
620  * so that it would work correctly again. It's useful while doing
621  * recovery or log collection on the indicated device.
622  */
623 int eeh_pci_enable(struct eeh_pe *pe, int function)
624 {
625 	int active_flag, rc;
626 
627 	/*
628 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
629 	 * Also, it's pointless to enable them on unfrozen PE. So
630 	 * we have to check before enabling IO or DMA.
631 	 */
632 	switch (function) {
633 	case EEH_OPT_THAW_MMIO:
634 		active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
635 		break;
636 	case EEH_OPT_THAW_DMA:
637 		active_flag = EEH_STATE_DMA_ACTIVE;
638 		break;
639 	case EEH_OPT_DISABLE:
640 	case EEH_OPT_ENABLE:
641 	case EEH_OPT_FREEZE_PE:
642 		active_flag = 0;
643 		break;
644 	default:
645 		pr_warn("%s: Invalid function %d\n",
646 			__func__, function);
647 		return -EINVAL;
648 	}
649 
650 	/*
651 	 * Check if IO or DMA has been enabled before
652 	 * enabling them.
653 	 */
654 	if (active_flag) {
655 		rc = eeh_ops->get_state(pe, NULL);
656 		if (rc < 0)
657 			return rc;
658 
659 		/* Needn't enable it at all */
660 		if (rc == EEH_STATE_NOT_SUPPORT)
661 			return 0;
662 
663 		/* It's already enabled */
664 		if (rc & active_flag)
665 			return 0;
666 	}
667 
668 
669 	/* Issue the request */
670 	rc = eeh_ops->set_option(pe, function);
671 	if (rc)
672 		pr_warn("%s: Unexpected state change %d on "
673 			"PHB#%x-PE#%x, err=%d\n",
674 			__func__, function, pe->phb->global_number,
675 			pe->addr, rc);
676 
677 	/* Check if the request is finished successfully */
678 	if (active_flag) {
679 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
680 		if (rc < 0)
681 			return rc;
682 
683 		if (rc & active_flag)
684 			return 0;
685 
686 		return -EIO;
687 	}
688 
689 	return rc;
690 }
691 
692 static void *eeh_disable_and_save_dev_state(struct eeh_dev *edev,
693 					    void *userdata)
694 {
695 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
696 	struct pci_dev *dev = userdata;
697 
698 	/*
699 	 * The caller should have disabled and saved the
700 	 * state for the specified device
701 	 */
702 	if (!pdev || pdev == dev)
703 		return NULL;
704 
705 	/* Ensure we have D0 power state */
706 	pci_set_power_state(pdev, PCI_D0);
707 
708 	/* Save device state */
709 	pci_save_state(pdev);
710 
711 	/*
712 	 * Disable device to avoid any DMA traffic and
713 	 * interrupt from the device
714 	 */
715 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
716 
717 	return NULL;
718 }
719 
720 static void *eeh_restore_dev_state(struct eeh_dev *edev, void *userdata)
721 {
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 int eeh_restore_vf_config(struct pci_dn *pdn)
741 {
742 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
743 	u32 devctl, cmd, cap2, aer_capctl;
744 	int old_mps;
745 
746 	if (edev->pcie_cap) {
747 		/* Restore MPS */
748 		old_mps = (ffs(pdn->mps) - 8) << 5;
749 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
750 				     2, &devctl);
751 		devctl &= ~PCI_EXP_DEVCTL_PAYLOAD;
752 		devctl |= old_mps;
753 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
754 				      2, devctl);
755 
756 		/* Disable Completion Timeout if possible */
757 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP2,
758 				     4, &cap2);
759 		if (cap2 & PCI_EXP_DEVCAP2_COMP_TMOUT_DIS) {
760 			eeh_ops->read_config(pdn,
761 					     edev->pcie_cap + PCI_EXP_DEVCTL2,
762 					     4, &cap2);
763 			cap2 |= PCI_EXP_DEVCTL2_COMP_TMOUT_DIS;
764 			eeh_ops->write_config(pdn,
765 					      edev->pcie_cap + PCI_EXP_DEVCTL2,
766 					      4, cap2);
767 		}
768 	}
769 
770 	/* Enable SERR and parity checking */
771 	eeh_ops->read_config(pdn, PCI_COMMAND, 2, &cmd);
772 	cmd |= (PCI_COMMAND_PARITY | PCI_COMMAND_SERR);
773 	eeh_ops->write_config(pdn, PCI_COMMAND, 2, cmd);
774 
775 	/* Enable report various errors */
776 	if (edev->pcie_cap) {
777 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
778 				     2, &devctl);
779 		devctl &= ~PCI_EXP_DEVCTL_CERE;
780 		devctl |= (PCI_EXP_DEVCTL_NFERE |
781 			   PCI_EXP_DEVCTL_FERE |
782 			   PCI_EXP_DEVCTL_URRE);
783 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
784 				      2, devctl);
785 	}
786 
787 	/* Enable ECRC generation and check */
788 	if (edev->pcie_cap && edev->aer_cap) {
789 		eeh_ops->read_config(pdn, edev->aer_cap + PCI_ERR_CAP,
790 				     4, &aer_capctl);
791 		aer_capctl |= (PCI_ERR_CAP_ECRC_GENE | PCI_ERR_CAP_ECRC_CHKE);
792 		eeh_ops->write_config(pdn, edev->aer_cap + PCI_ERR_CAP,
793 				      4, aer_capctl);
794 	}
795 
796 	return 0;
797 }
798 
799 /**
800  * pcibios_set_pcie_reset_state - Set PCI-E reset state
801  * @dev: pci device struct
802  * @state: reset state to enter
803  *
804  * Return value:
805  * 	0 if success
806  */
807 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
808 {
809 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
810 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
811 
812 	if (!pe) {
813 		pr_err("%s: No PE found on PCI device %s\n",
814 			__func__, pci_name(dev));
815 		return -EINVAL;
816 	}
817 
818 	switch (state) {
819 	case pcie_deassert_reset:
820 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
821 		eeh_unfreeze_pe(pe, false);
822 		if (!(pe->type & EEH_PE_VF))
823 			eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
824 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
825 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
826 		break;
827 	case pcie_hot_reset:
828 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
829 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
830 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
831 		if (!(pe->type & EEH_PE_VF))
832 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
833 		eeh_ops->reset(pe, EEH_RESET_HOT);
834 		break;
835 	case pcie_warm_reset:
836 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
837 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
838 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
839 		if (!(pe->type & EEH_PE_VF))
840 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
841 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
842 		break;
843 	default:
844 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
845 		return -EINVAL;
846 	};
847 
848 	return 0;
849 }
850 
851 /**
852  * eeh_set_pe_freset - Check the required reset for the indicated device
853  * @data: EEH device
854  * @flag: return value
855  *
856  * Each device might have its preferred reset type: fundamental or
857  * hot reset. The routine is used to collected the information for
858  * the indicated device and its children so that the bunch of the
859  * devices could be reset properly.
860  */
861 static void *eeh_set_dev_freset(struct eeh_dev *edev, void *flag)
862 {
863 	struct pci_dev *dev;
864 	unsigned int *freset = (unsigned int *)flag;
865 
866 	dev = eeh_dev_to_pci_dev(edev);
867 	if (dev)
868 		*freset |= dev->needs_freset;
869 
870 	return NULL;
871 }
872 
873 /**
874  * eeh_pe_reset_full - Complete a full reset process on the indicated PE
875  * @pe: EEH PE
876  *
877  * This function executes a full reset procedure on a PE, including setting
878  * the appropriate flags, performing a fundamental or hot reset, and then
879  * deactivating the reset status.  It is designed to be used within the EEH
880  * subsystem, as opposed to eeh_pe_reset which is exported to drivers and
881  * only performs a single operation at a time.
882  *
883  * This function will attempt to reset a PE three times before failing.
884  */
885 int eeh_pe_reset_full(struct eeh_pe *pe)
886 {
887 	int reset_state = (EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
888 	int type = EEH_RESET_HOT;
889 	unsigned int freset = 0;
890 	int i, state, ret;
891 
892 	/*
893 	 * Determine the type of reset to perform - hot or fundamental.
894 	 * Hot reset is the default operation, unless any device under the
895 	 * PE requires a fundamental reset.
896 	 */
897 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
898 
899 	if (freset)
900 		type = EEH_RESET_FUNDAMENTAL;
901 
902 	/* Mark the PE as in reset state and block config space accesses */
903 	eeh_pe_state_mark(pe, reset_state);
904 
905 	/* Make three attempts at resetting the bus */
906 	for (i = 0; i < 3; i++) {
907 		ret = eeh_pe_reset(pe, type);
908 		if (ret)
909 			break;
910 
911 		ret = eeh_pe_reset(pe, EEH_RESET_DEACTIVATE);
912 		if (ret)
913 			break;
914 
915 		/* Wait until the PE is in a functioning state */
916 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
917 		if (eeh_state_active(state))
918 			break;
919 
920 		if (state < 0) {
921 			pr_warn("%s: Unrecoverable slot failure on PHB#%x-PE#%x",
922 				__func__, pe->phb->global_number, pe->addr);
923 			ret = -ENOTRECOVERABLE;
924 			break;
925 		}
926 
927 		/* Set error in case this is our last attempt */
928 		ret = -EIO;
929 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
930 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
931 	}
932 
933 	eeh_pe_state_clear(pe, reset_state);
934 	return ret;
935 }
936 
937 /**
938  * eeh_save_bars - Save device bars
939  * @edev: PCI device associated EEH device
940  *
941  * Save the values of the device bars. Unlike the restore
942  * routine, this routine is *not* recursive. This is because
943  * PCI devices are added individually; but, for the restore,
944  * an entire slot is reset at a time.
945  */
946 void eeh_save_bars(struct eeh_dev *edev)
947 {
948 	struct pci_dn *pdn;
949 	int i;
950 
951 	pdn = eeh_dev_to_pdn(edev);
952 	if (!pdn)
953 		return;
954 
955 	for (i = 0; i < 16; i++)
956 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
957 
958 	/*
959 	 * For PCI bridges including root port, we need enable bus
960 	 * master explicitly. Otherwise, it can't fetch IODA table
961 	 * entries correctly. So we cache the bit in advance so that
962 	 * we can restore it after reset, either PHB range or PE range.
963 	 */
964 	if (edev->mode & EEH_DEV_BRIDGE)
965 		edev->config_space[1] |= PCI_COMMAND_MASTER;
966 }
967 
968 /**
969  * eeh_ops_register - Register platform dependent EEH operations
970  * @ops: platform dependent EEH operations
971  *
972  * Register the platform dependent EEH operation callback
973  * functions. The platform should call this function before
974  * any other EEH operations.
975  */
976 int __init eeh_ops_register(struct eeh_ops *ops)
977 {
978 	if (!ops->name) {
979 		pr_warn("%s: Invalid EEH ops name for %p\n",
980 			__func__, ops);
981 		return -EINVAL;
982 	}
983 
984 	if (eeh_ops && eeh_ops != ops) {
985 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
986 			__func__, eeh_ops->name, ops->name);
987 		return -EEXIST;
988 	}
989 
990 	eeh_ops = ops;
991 
992 	return 0;
993 }
994 
995 /**
996  * eeh_ops_unregister - Unreigster platform dependent EEH operations
997  * @name: name of EEH platform operations
998  *
999  * Unregister the platform dependent EEH operation callback
1000  * functions.
1001  */
1002 int __exit eeh_ops_unregister(const char *name)
1003 {
1004 	if (!name || !strlen(name)) {
1005 		pr_warn("%s: Invalid EEH ops name\n",
1006 			__func__);
1007 		return -EINVAL;
1008 	}
1009 
1010 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
1011 		eeh_ops = NULL;
1012 		return 0;
1013 	}
1014 
1015 	return -EEXIST;
1016 }
1017 
1018 static int eeh_reboot_notifier(struct notifier_block *nb,
1019 			       unsigned long action, void *unused)
1020 {
1021 	eeh_clear_flag(EEH_ENABLED);
1022 	return NOTIFY_DONE;
1023 }
1024 
1025 static struct notifier_block eeh_reboot_nb = {
1026 	.notifier_call = eeh_reboot_notifier,
1027 };
1028 
1029 void eeh_probe_devices(void)
1030 {
1031 	struct pci_controller *hose, *tmp;
1032 	struct pci_dn *pdn;
1033 
1034 	/* Enable EEH for all adapters */
1035 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1036 		pdn = hose->pci_data;
1037 		traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1038 	}
1039 }
1040 
1041 /**
1042  * eeh_init - EEH initialization
1043  *
1044  * Initialize EEH by trying to enable it for all of the adapters in the system.
1045  * As a side effect we can determine here if eeh is supported at all.
1046  * Note that we leave EEH on so failed config cycles won't cause a machine
1047  * check.  If a user turns off EEH for a particular adapter they are really
1048  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
1049  * grant access to a slot if EEH isn't enabled, and so we always enable
1050  * EEH for all slots/all devices.
1051  *
1052  * The eeh-force-off option disables EEH checking globally, for all slots.
1053  * Even if force-off is set, the EEH hardware is still enabled, so that
1054  * newer systems can boot.
1055  */
1056 static int eeh_init(void)
1057 {
1058 	struct pci_controller *hose, *tmp;
1059 	int ret = 0;
1060 
1061 	/* Register reboot notifier */
1062 	ret = register_reboot_notifier(&eeh_reboot_nb);
1063 	if (ret) {
1064 		pr_warn("%s: Failed to register notifier (%d)\n",
1065 			__func__, ret);
1066 		return ret;
1067 	}
1068 
1069 	/* call platform initialization function */
1070 	if (!eeh_ops) {
1071 		pr_warn("%s: Platform EEH operation not found\n",
1072 			__func__);
1073 		return -EEXIST;
1074 	} else if ((ret = eeh_ops->init()))
1075 		return ret;
1076 
1077 	/* Initialize PHB PEs */
1078 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
1079 		eeh_dev_phb_init_dynamic(hose);
1080 
1081 	/* Initialize EEH event */
1082 	ret = eeh_event_init();
1083 	if (ret)
1084 		return ret;
1085 
1086 	eeh_probe_devices();
1087 
1088 	if (eeh_enabled())
1089 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1090 	else if (!eeh_has_flag(EEH_POSTPONED_PROBE))
1091 		pr_info("EEH: No capable adapters found\n");
1092 
1093 	return ret;
1094 }
1095 
1096 core_initcall_sync(eeh_init);
1097 
1098 /**
1099  * eeh_add_device_early - Enable EEH for the indicated device node
1100  * @pdn: PCI device node for which to set up EEH
1101  *
1102  * This routine must be used to perform EEH initialization for PCI
1103  * devices that were added after system boot (e.g. hotplug, dlpar).
1104  * This routine must be called before any i/o is performed to the
1105  * adapter (inluding any config-space i/o).
1106  * Whether this actually enables EEH or not for this device depends
1107  * on the CEC architecture, type of the device, on earlier boot
1108  * command-line arguments & etc.
1109  */
1110 void eeh_add_device_early(struct pci_dn *pdn)
1111 {
1112 	struct pci_controller *phb = pdn ? pdn->phb : NULL;
1113 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1114 
1115 	if (!edev)
1116 		return;
1117 
1118 	if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1119 		return;
1120 
1121 	/* USB Bus children of PCI devices will not have BUID's */
1122 	if (NULL == phb ||
1123 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1124 		return;
1125 
1126 	eeh_ops->probe(pdn, NULL);
1127 }
1128 
1129 /**
1130  * eeh_add_device_tree_early - Enable EEH for the indicated device
1131  * @pdn: PCI device node
1132  *
1133  * This routine must be used to perform EEH initialization for the
1134  * indicated PCI device that was added after system boot (e.g.
1135  * hotplug, dlpar).
1136  */
1137 void eeh_add_device_tree_early(struct pci_dn *pdn)
1138 {
1139 	struct pci_dn *n;
1140 
1141 	if (!pdn)
1142 		return;
1143 
1144 	list_for_each_entry(n, &pdn->child_list, list)
1145 		eeh_add_device_tree_early(n);
1146 	eeh_add_device_early(pdn);
1147 }
1148 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1149 
1150 /**
1151  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1152  * @dev: pci device for which to set up EEH
1153  *
1154  * This routine must be used to complete EEH initialization for PCI
1155  * devices that were added after system boot (e.g. hotplug, dlpar).
1156  */
1157 void eeh_add_device_late(struct pci_dev *dev)
1158 {
1159 	struct pci_dn *pdn;
1160 	struct eeh_dev *edev;
1161 
1162 	if (!dev || !eeh_enabled())
1163 		return;
1164 
1165 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1166 
1167 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1168 	edev = pdn_to_eeh_dev(pdn);
1169 	if (edev->pdev == dev) {
1170 		pr_debug("EEH: Already referenced !\n");
1171 		return;
1172 	}
1173 
1174 	/*
1175 	 * The EEH cache might not be removed correctly because of
1176 	 * unbalanced kref to the device during unplug time, which
1177 	 * relies on pcibios_release_device(). So we have to remove
1178 	 * that here explicitly.
1179 	 */
1180 	if (edev->pdev) {
1181 		eeh_rmv_from_parent_pe(edev);
1182 		eeh_addr_cache_rmv_dev(edev->pdev);
1183 		eeh_sysfs_remove_device(edev->pdev);
1184 		edev->mode &= ~EEH_DEV_SYSFS;
1185 
1186 		/*
1187 		 * We definitely should have the PCI device removed
1188 		 * though it wasn't correctly. So we needn't call
1189 		 * into error handler afterwards.
1190 		 */
1191 		edev->mode |= EEH_DEV_NO_HANDLER;
1192 
1193 		edev->pdev = NULL;
1194 		dev->dev.archdata.edev = NULL;
1195 	}
1196 
1197 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1198 		eeh_ops->probe(pdn, NULL);
1199 
1200 	edev->pdev = dev;
1201 	dev->dev.archdata.edev = edev;
1202 
1203 	eeh_addr_cache_insert_dev(dev);
1204 }
1205 
1206 /**
1207  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1208  * @bus: PCI bus
1209  *
1210  * This routine must be used to perform EEH initialization for PCI
1211  * devices which are attached to the indicated PCI bus. The PCI bus
1212  * is added after system boot through hotplug or dlpar.
1213  */
1214 void eeh_add_device_tree_late(struct pci_bus *bus)
1215 {
1216 	struct pci_dev *dev;
1217 
1218 	list_for_each_entry(dev, &bus->devices, bus_list) {
1219 		eeh_add_device_late(dev);
1220 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1221 			struct pci_bus *subbus = dev->subordinate;
1222 			if (subbus)
1223 				eeh_add_device_tree_late(subbus);
1224 		}
1225 	}
1226 }
1227 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1228 
1229 /**
1230  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1231  * @bus: PCI bus
1232  *
1233  * This routine must be used to add EEH sysfs files for PCI
1234  * devices which are attached to the indicated PCI bus. The PCI bus
1235  * is added after system boot through hotplug or dlpar.
1236  */
1237 void eeh_add_sysfs_files(struct pci_bus *bus)
1238 {
1239 	struct pci_dev *dev;
1240 
1241 	list_for_each_entry(dev, &bus->devices, bus_list) {
1242 		eeh_sysfs_add_device(dev);
1243 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1244 			struct pci_bus *subbus = dev->subordinate;
1245 			if (subbus)
1246 				eeh_add_sysfs_files(subbus);
1247 		}
1248 	}
1249 }
1250 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1251 
1252 /**
1253  * eeh_remove_device - Undo EEH setup for the indicated pci device
1254  * @dev: pci device to be removed
1255  *
1256  * This routine should be called when a device is removed from
1257  * a running system (e.g. by hotplug or dlpar).  It unregisters
1258  * the PCI device from the EEH subsystem.  I/O errors affecting
1259  * this device will no longer be detected after this call; thus,
1260  * i/o errors affecting this slot may leave this device unusable.
1261  */
1262 void eeh_remove_device(struct pci_dev *dev)
1263 {
1264 	struct eeh_dev *edev;
1265 
1266 	if (!dev || !eeh_enabled())
1267 		return;
1268 	edev = pci_dev_to_eeh_dev(dev);
1269 
1270 	/* Unregister the device with the EEH/PCI address search system */
1271 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1272 
1273 	if (!edev || !edev->pdev || !edev->pe) {
1274 		pr_debug("EEH: Not referenced !\n");
1275 		return;
1276 	}
1277 
1278 	/*
1279 	 * During the hotplug for EEH error recovery, we need the EEH
1280 	 * device attached to the parent PE in order for BAR restore
1281 	 * a bit later. So we keep it for BAR restore and remove it
1282 	 * from the parent PE during the BAR resotre.
1283 	 */
1284 	edev->pdev = NULL;
1285 
1286 	/*
1287 	 * The flag "in_error" is used to trace EEH devices for VFs
1288 	 * in error state or not. It's set in eeh_report_error(). If
1289 	 * it's not set, eeh_report_{reset,resume}() won't be called
1290 	 * for the VF EEH device.
1291 	 */
1292 	edev->in_error = false;
1293 	dev->dev.archdata.edev = NULL;
1294 	if (!(edev->pe->state & EEH_PE_KEEP))
1295 		eeh_rmv_from_parent_pe(edev);
1296 	else
1297 		edev->mode |= EEH_DEV_DISCONNECTED;
1298 
1299 	/*
1300 	 * We're removing from the PCI subsystem, that means
1301 	 * the PCI device driver can't support EEH or not
1302 	 * well. So we rely on hotplug completely to do recovery
1303 	 * for the specific PCI device.
1304 	 */
1305 	edev->mode |= EEH_DEV_NO_HANDLER;
1306 
1307 	eeh_addr_cache_rmv_dev(dev);
1308 	eeh_sysfs_remove_device(dev);
1309 	edev->mode &= ~EEH_DEV_SYSFS;
1310 }
1311 
1312 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1313 {
1314 	int ret;
1315 
1316 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1317 	if (ret) {
1318 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1319 			__func__, ret, pe->phb->global_number, pe->addr);
1320 		return ret;
1321 	}
1322 
1323 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1324 	if (ret) {
1325 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1326 			__func__, ret, pe->phb->global_number, pe->addr);
1327 		return ret;
1328 	}
1329 
1330 	/* Clear software isolated state */
1331 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1332 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1333 
1334 	return ret;
1335 }
1336 
1337 
1338 static struct pci_device_id eeh_reset_ids[] = {
1339 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1340 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1341 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1342 	{ 0 }
1343 };
1344 
1345 static int eeh_pe_change_owner(struct eeh_pe *pe)
1346 {
1347 	struct eeh_dev *edev, *tmp;
1348 	struct pci_dev *pdev;
1349 	struct pci_device_id *id;
1350 	int ret;
1351 
1352 	/* Check PE state */
1353 	ret = eeh_ops->get_state(pe, NULL);
1354 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1355 		return 0;
1356 
1357 	/* Unfrozen PE, nothing to do */
1358 	if (eeh_state_active(ret))
1359 		return 0;
1360 
1361 	/* Frozen PE, check if it needs PE level reset */
1362 	eeh_pe_for_each_dev(pe, edev, tmp) {
1363 		pdev = eeh_dev_to_pci_dev(edev);
1364 		if (!pdev)
1365 			continue;
1366 
1367 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1368 			if (id->vendor != PCI_ANY_ID &&
1369 			    id->vendor != pdev->vendor)
1370 				continue;
1371 			if (id->device != PCI_ANY_ID &&
1372 			    id->device != pdev->device)
1373 				continue;
1374 			if (id->subvendor != PCI_ANY_ID &&
1375 			    id->subvendor != pdev->subsystem_vendor)
1376 				continue;
1377 			if (id->subdevice != PCI_ANY_ID &&
1378 			    id->subdevice != pdev->subsystem_device)
1379 				continue;
1380 
1381 			return eeh_pe_reset_and_recover(pe);
1382 		}
1383 	}
1384 
1385 	return eeh_unfreeze_pe(pe, true);
1386 }
1387 
1388 /**
1389  * eeh_dev_open - Increase count of pass through devices for PE
1390  * @pdev: PCI device
1391  *
1392  * Increase count of passed through devices for the indicated
1393  * PE. In the result, the EEH errors detected on the PE won't be
1394  * reported. The PE owner will be responsible for detection
1395  * and recovery.
1396  */
1397 int eeh_dev_open(struct pci_dev *pdev)
1398 {
1399 	struct eeh_dev *edev;
1400 	int ret = -ENODEV;
1401 
1402 	mutex_lock(&eeh_dev_mutex);
1403 
1404 	/* No PCI device ? */
1405 	if (!pdev)
1406 		goto out;
1407 
1408 	/* No EEH device or PE ? */
1409 	edev = pci_dev_to_eeh_dev(pdev);
1410 	if (!edev || !edev->pe)
1411 		goto out;
1412 
1413 	/*
1414 	 * The PE might have been put into frozen state, but we
1415 	 * didn't detect that yet. The passed through PCI devices
1416 	 * in frozen PE won't work properly. Clear the frozen state
1417 	 * in advance.
1418 	 */
1419 	ret = eeh_pe_change_owner(edev->pe);
1420 	if (ret)
1421 		goto out;
1422 
1423 	/* Increase PE's pass through count */
1424 	atomic_inc(&edev->pe->pass_dev_cnt);
1425 	mutex_unlock(&eeh_dev_mutex);
1426 
1427 	return 0;
1428 out:
1429 	mutex_unlock(&eeh_dev_mutex);
1430 	return ret;
1431 }
1432 EXPORT_SYMBOL_GPL(eeh_dev_open);
1433 
1434 /**
1435  * eeh_dev_release - Decrease count of pass through devices for PE
1436  * @pdev: PCI device
1437  *
1438  * Decrease count of pass through devices for the indicated PE. If
1439  * there is no passed through device in PE, the EEH errors detected
1440  * on the PE will be reported and handled as usual.
1441  */
1442 void eeh_dev_release(struct pci_dev *pdev)
1443 {
1444 	struct eeh_dev *edev;
1445 
1446 	mutex_lock(&eeh_dev_mutex);
1447 
1448 	/* No PCI device ? */
1449 	if (!pdev)
1450 		goto out;
1451 
1452 	/* No EEH device ? */
1453 	edev = pci_dev_to_eeh_dev(pdev);
1454 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1455 		goto out;
1456 
1457 	/* Decrease PE's pass through count */
1458 	WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1459 	eeh_pe_change_owner(edev->pe);
1460 out:
1461 	mutex_unlock(&eeh_dev_mutex);
1462 }
1463 EXPORT_SYMBOL(eeh_dev_release);
1464 
1465 #ifdef CONFIG_IOMMU_API
1466 
1467 static int dev_has_iommu_table(struct device *dev, void *data)
1468 {
1469 	struct pci_dev *pdev = to_pci_dev(dev);
1470 	struct pci_dev **ppdev = data;
1471 
1472 	if (!dev)
1473 		return 0;
1474 
1475 	if (dev->iommu_group) {
1476 		*ppdev = pdev;
1477 		return 1;
1478 	}
1479 
1480 	return 0;
1481 }
1482 
1483 /**
1484  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1485  * @group: IOMMU group
1486  *
1487  * The routine is called to convert IOMMU group to EEH PE.
1488  */
1489 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1490 {
1491 	struct pci_dev *pdev = NULL;
1492 	struct eeh_dev *edev;
1493 	int ret;
1494 
1495 	/* No IOMMU group ? */
1496 	if (!group)
1497 		return NULL;
1498 
1499 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1500 	if (!ret || !pdev)
1501 		return NULL;
1502 
1503 	/* No EEH device or PE ? */
1504 	edev = pci_dev_to_eeh_dev(pdev);
1505 	if (!edev || !edev->pe)
1506 		return NULL;
1507 
1508 	return edev->pe;
1509 }
1510 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1511 
1512 #endif /* CONFIG_IOMMU_API */
1513 
1514 /**
1515  * eeh_pe_set_option - Set options for the indicated PE
1516  * @pe: EEH PE
1517  * @option: requested option
1518  *
1519  * The routine is called to enable or disable EEH functionality
1520  * on the indicated PE, to enable IO or DMA for the frozen PE.
1521  */
1522 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1523 {
1524 	int ret = 0;
1525 
1526 	/* Invalid PE ? */
1527 	if (!pe)
1528 		return -ENODEV;
1529 
1530 	/*
1531 	 * EEH functionality could possibly be disabled, just
1532 	 * return error for the case. And the EEH functinality
1533 	 * isn't expected to be disabled on one specific PE.
1534 	 */
1535 	switch (option) {
1536 	case EEH_OPT_ENABLE:
1537 		if (eeh_enabled()) {
1538 			ret = eeh_pe_change_owner(pe);
1539 			break;
1540 		}
1541 		ret = -EIO;
1542 		break;
1543 	case EEH_OPT_DISABLE:
1544 		break;
1545 	case EEH_OPT_THAW_MMIO:
1546 	case EEH_OPT_THAW_DMA:
1547 	case EEH_OPT_FREEZE_PE:
1548 		if (!eeh_ops || !eeh_ops->set_option) {
1549 			ret = -ENOENT;
1550 			break;
1551 		}
1552 
1553 		ret = eeh_pci_enable(pe, option);
1554 		break;
1555 	default:
1556 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1557 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1558 		ret = -EINVAL;
1559 	}
1560 
1561 	return ret;
1562 }
1563 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1564 
1565 /**
1566  * eeh_pe_get_state - Retrieve PE's state
1567  * @pe: EEH PE
1568  *
1569  * Retrieve the PE's state, which includes 3 aspects: enabled
1570  * DMA, enabled IO and asserted reset.
1571  */
1572 int eeh_pe_get_state(struct eeh_pe *pe)
1573 {
1574 	int result, ret = 0;
1575 	bool rst_active, dma_en, mmio_en;
1576 
1577 	/* Existing PE ? */
1578 	if (!pe)
1579 		return -ENODEV;
1580 
1581 	if (!eeh_ops || !eeh_ops->get_state)
1582 		return -ENOENT;
1583 
1584 	/*
1585 	 * If the parent PE is owned by the host kernel and is undergoing
1586 	 * error recovery, we should return the PE state as temporarily
1587 	 * unavailable so that the error recovery on the guest is suspended
1588 	 * until the recovery completes on the host.
1589 	 */
1590 	if (pe->parent &&
1591 	    !(pe->state & EEH_PE_REMOVED) &&
1592 	    (pe->parent->state & (EEH_PE_ISOLATED | EEH_PE_RECOVERING)))
1593 		return EEH_PE_STATE_UNAVAIL;
1594 
1595 	result = eeh_ops->get_state(pe, NULL);
1596 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1597 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1598 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1599 
1600 	if (rst_active)
1601 		ret = EEH_PE_STATE_RESET;
1602 	else if (dma_en && mmio_en)
1603 		ret = EEH_PE_STATE_NORMAL;
1604 	else if (!dma_en && !mmio_en)
1605 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1606 	else if (!dma_en && mmio_en)
1607 		ret = EEH_PE_STATE_STOPPED_DMA;
1608 	else
1609 		ret = EEH_PE_STATE_UNAVAIL;
1610 
1611 	return ret;
1612 }
1613 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1614 
1615 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1616 {
1617 	struct eeh_dev *edev, *tmp;
1618 	struct pci_dev *pdev;
1619 	int ret = 0;
1620 
1621 	/* Restore config space */
1622 	eeh_pe_restore_bars(pe);
1623 
1624 	/*
1625 	 * Reenable PCI devices as the devices passed
1626 	 * through are always enabled before the reset.
1627 	 */
1628 	eeh_pe_for_each_dev(pe, edev, tmp) {
1629 		pdev = eeh_dev_to_pci_dev(edev);
1630 		if (!pdev)
1631 			continue;
1632 
1633 		ret = pci_reenable_device(pdev);
1634 		if (ret) {
1635 			pr_warn("%s: Failure %d reenabling %s\n",
1636 				__func__, ret, pci_name(pdev));
1637 			return ret;
1638 		}
1639 	}
1640 
1641 	/* The PE is still in frozen state */
1642 	return eeh_unfreeze_pe(pe, true);
1643 }
1644 
1645 
1646 /**
1647  * eeh_pe_reset - Issue PE reset according to specified type
1648  * @pe: EEH PE
1649  * @option: reset type
1650  *
1651  * The routine is called to reset the specified PE with the
1652  * indicated type, either fundamental reset or hot reset.
1653  * PE reset is the most important part for error recovery.
1654  */
1655 int eeh_pe_reset(struct eeh_pe *pe, int option)
1656 {
1657 	int ret = 0;
1658 
1659 	/* Invalid PE ? */
1660 	if (!pe)
1661 		return -ENODEV;
1662 
1663 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1664 		return -ENOENT;
1665 
1666 	switch (option) {
1667 	case EEH_RESET_DEACTIVATE:
1668 		ret = eeh_ops->reset(pe, option);
1669 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1670 		if (ret)
1671 			break;
1672 
1673 		ret = eeh_pe_reenable_devices(pe);
1674 		break;
1675 	case EEH_RESET_HOT:
1676 	case EEH_RESET_FUNDAMENTAL:
1677 		/*
1678 		 * Proactively freeze the PE to drop all MMIO access
1679 		 * during reset, which should be banned as it's always
1680 		 * cause recursive EEH error.
1681 		 */
1682 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1683 
1684 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1685 		ret = eeh_ops->reset(pe, option);
1686 		break;
1687 	default:
1688 		pr_debug("%s: Unsupported option %d\n",
1689 			__func__, option);
1690 		ret = -EINVAL;
1691 	}
1692 
1693 	return ret;
1694 }
1695 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1696 
1697 /**
1698  * eeh_pe_configure - Configure PCI bridges after PE reset
1699  * @pe: EEH PE
1700  *
1701  * The routine is called to restore the PCI config space for
1702  * those PCI devices, especially PCI bridges affected by PE
1703  * reset issued previously.
1704  */
1705 int eeh_pe_configure(struct eeh_pe *pe)
1706 {
1707 	int ret = 0;
1708 
1709 	/* Invalid PE ? */
1710 	if (!pe)
1711 		return -ENODEV;
1712 
1713 	return ret;
1714 }
1715 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1716 
1717 /**
1718  * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1719  * @pe: the indicated PE
1720  * @type: error type
1721  * @function: error function
1722  * @addr: address
1723  * @mask: address mask
1724  *
1725  * The routine is called to inject the specified PCI error, which
1726  * is determined by @type and @function, to the indicated PE for
1727  * testing purpose.
1728  */
1729 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1730 		      unsigned long addr, unsigned long mask)
1731 {
1732 	/* Invalid PE ? */
1733 	if (!pe)
1734 		return -ENODEV;
1735 
1736 	/* Unsupported operation ? */
1737 	if (!eeh_ops || !eeh_ops->err_inject)
1738 		return -ENOENT;
1739 
1740 	/* Check on PCI error type */
1741 	if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1742 		return -EINVAL;
1743 
1744 	/* Check on PCI error function */
1745 	if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1746 		return -EINVAL;
1747 
1748 	return eeh_ops->err_inject(pe, type, func, addr, mask);
1749 }
1750 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1751 
1752 static int proc_eeh_show(struct seq_file *m, void *v)
1753 {
1754 	if (!eeh_enabled()) {
1755 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1756 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1757 	} else {
1758 		seq_printf(m, "EEH Subsystem is enabled\n");
1759 		seq_printf(m,
1760 				"no device=%llu\n"
1761 				"no device node=%llu\n"
1762 				"no config address=%llu\n"
1763 				"check not wanted=%llu\n"
1764 				"eeh_total_mmio_ffs=%llu\n"
1765 				"eeh_false_positives=%llu\n"
1766 				"eeh_slot_resets=%llu\n",
1767 				eeh_stats.no_device,
1768 				eeh_stats.no_dn,
1769 				eeh_stats.no_cfg_addr,
1770 				eeh_stats.ignored_check,
1771 				eeh_stats.total_mmio_ffs,
1772 				eeh_stats.false_positives,
1773 				eeh_stats.slot_resets);
1774 	}
1775 
1776 	return 0;
1777 }
1778 
1779 #ifdef CONFIG_DEBUG_FS
1780 static int eeh_enable_dbgfs_set(void *data, u64 val)
1781 {
1782 	if (val)
1783 		eeh_clear_flag(EEH_FORCE_DISABLED);
1784 	else
1785 		eeh_add_flag(EEH_FORCE_DISABLED);
1786 
1787 	return 0;
1788 }
1789 
1790 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1791 {
1792 	if (eeh_enabled())
1793 		*val = 0x1ul;
1794 	else
1795 		*val = 0x0ul;
1796 	return 0;
1797 }
1798 
1799 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1800 {
1801 	eeh_max_freezes = val;
1802 	return 0;
1803 }
1804 
1805 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1806 {
1807 	*val = eeh_max_freezes;
1808 	return 0;
1809 }
1810 
1811 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1812 			eeh_enable_dbgfs_set, "0x%llx\n");
1813 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1814 			eeh_freeze_dbgfs_set, "0x%llx\n");
1815 #endif
1816 
1817 static int __init eeh_init_proc(void)
1818 {
1819 	if (machine_is(pseries) || machine_is(powernv)) {
1820 		proc_create_single("powerpc/eeh", 0, NULL, proc_eeh_show);
1821 #ifdef CONFIG_DEBUG_FS
1822 		debugfs_create_file("eeh_enable", 0600,
1823                                     powerpc_debugfs_root, NULL,
1824                                     &eeh_enable_dbgfs_ops);
1825 		debugfs_create_file("eeh_max_freezes", 0600,
1826 				    powerpc_debugfs_root, NULL,
1827 				    &eeh_freeze_dbgfs_ops);
1828 #endif
1829 	}
1830 
1831 	return 0;
1832 }
1833 __initcall(eeh_init_proc);
1834