• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
eeh_setup(char * str)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  */
eeh_dump_dev_log(struct eeh_dev * edev,char * buf,size_t len)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 	if (!pdn) {
173 		pr_warn("EEH: Note: No error log for absent device.\n");
174 		return 0;
175 	}
176 
177 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x.%01x\n",
178 		       pdn->phb->global_number, pdn->busno,
179 		       PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
180 	pr_warn("EEH: of node=%04x:%02x:%02x.%01x\n",
181 		pdn->phb->global_number, pdn->busno,
182 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
183 
184 	eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
185 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
186 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
187 
188 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
189 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
190 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
191 
192 	/* Gather bridge-specific registers */
193 	if (edev->mode & EEH_DEV_BRIDGE) {
194 		eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
195 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
196 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
197 
198 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
199 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
200 		pr_warn("EEH: Bridge control: %04x\n", cfg);
201 	}
202 
203 	/* Dump out the PCI-X command and status regs */
204 	cap = edev->pcix_cap;
205 	if (cap) {
206 		eeh_ops->read_config(pdn, cap, 4, &cfg);
207 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
208 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
209 
210 		eeh_ops->read_config(pdn, cap+4, 4, &cfg);
211 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
212 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
213 	}
214 
215 	/* If PCI-E capable, dump PCI-E cap 10 */
216 	cap = edev->pcie_cap;
217 	if (cap) {
218 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
219 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
220 
221 		for (i=0; i<=8; i++) {
222 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
223 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
224 
225 			if ((i % 4) == 0) {
226 				if (i != 0)
227 					pr_warn("%s\n", buffer);
228 
229 				l = scnprintf(buffer, sizeof(buffer),
230 					      "EEH: PCI-E %02x: %08x ",
231 					      4*i, cfg);
232 			} else {
233 				l += scnprintf(buffer+l, sizeof(buffer)-l,
234 					       "%08x ", cfg);
235 			}
236 
237 		}
238 
239 		pr_warn("%s\n", buffer);
240 	}
241 
242 	/* If AER capable, dump it */
243 	cap = edev->aer_cap;
244 	if (cap) {
245 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
246 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
247 
248 		for (i=0; i<=13; i++) {
249 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
250 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
251 
252 			if ((i % 4) == 0) {
253 				if (i != 0)
254 					pr_warn("%s\n", buffer);
255 
256 				l = scnprintf(buffer, sizeof(buffer),
257 					      "EEH: PCI-E AER %02x: %08x ",
258 					      4*i, cfg);
259 			} else {
260 				l += scnprintf(buffer+l, sizeof(buffer)-l,
261 					       "%08x ", cfg);
262 			}
263 		}
264 
265 		pr_warn("%s\n", buffer);
266 	}
267 
268 	return n;
269 }
270 
eeh_dump_pe_log(void * data,void * flag)271 static void *eeh_dump_pe_log(void *data, void *flag)
272 {
273 	struct eeh_pe *pe = data;
274 	struct eeh_dev *edev, *tmp;
275 	size_t *plen = flag;
276 
277 	eeh_pe_for_each_dev(pe, edev, tmp)
278 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
279 					  EEH_PCI_REGS_LOG_LEN - *plen);
280 
281 	return NULL;
282 }
283 
284 /**
285  * eeh_slot_error_detail - Generate combined log including driver log and error log
286  * @pe: EEH PE
287  * @severity: temporary or permanent error log
288  *
289  * This routine should be called to generate the combined log, which
290  * is comprised of driver log and error log. The driver log is figured
291  * out from the config space of the corresponding PCI device, while
292  * the error log is fetched through platform dependent function call.
293  */
eeh_slot_error_detail(struct eeh_pe * pe,int severity)294 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
295 {
296 	size_t loglen = 0;
297 
298 	/*
299 	 * When the PHB is fenced or dead, it's pointless to collect
300 	 * the data from PCI config space because it should return
301 	 * 0xFF's. For ER, we still retrieve the data from the PCI
302 	 * config space.
303 	 *
304 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
305 	 * 0xFF's is always returned from PCI config space.
306 	 *
307 	 * When the @severity is EEH_LOG_PERM, the PE is going to be
308 	 * removed. Prior to that, the drivers for devices included in
309 	 * the PE will be closed. The drivers rely on working IO path
310 	 * to bring the devices to quiet state. Otherwise, PCI traffic
311 	 * from those devices after they are removed is like to cause
312 	 * another unexpected EEH error.
313 	 */
314 	if (!(pe->type & EEH_PE_PHB)) {
315 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG) ||
316 		    severity == EEH_LOG_PERM)
317 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
318 
319 		/*
320 		 * The config space of some PCI devices can't be accessed
321 		 * when their PEs are in frozen state. Otherwise, fenced
322 		 * PHB might be seen. Those PEs are identified with flag
323 		 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
324 		 * is set automatically when the PE is put to EEH_PE_ISOLATED.
325 		 *
326 		 * Restoring BARs possibly triggers PCI config access in
327 		 * (OPAL) firmware and then causes fenced PHB. If the
328 		 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
329 		 * pointless to restore BARs and dump config space.
330 		 */
331 		eeh_ops->configure_bridge(pe);
332 		if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
333 			eeh_pe_restore_bars(pe);
334 
335 			pci_regs_buf[0] = 0;
336 			eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
337 		}
338 	}
339 
340 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
341 }
342 
343 /**
344  * eeh_token_to_phys - Convert EEH address token to phys address
345  * @token: I/O token, should be address in the form 0xA....
346  *
347  * This routine should be called to convert virtual I/O address
348  * to physical one.
349  */
eeh_token_to_phys(unsigned long token)350 static inline unsigned long eeh_token_to_phys(unsigned long token)
351 {
352 	pte_t *ptep;
353 	unsigned long pa;
354 	int hugepage_shift;
355 
356 	/*
357 	 * We won't find hugepages here(this is iomem). Hence we are not
358 	 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
359 	 * page table free, because of init_mm.
360 	 */
361 	ptep = find_init_mm_pte(token, &hugepage_shift);
362 	if (!ptep)
363 		return token;
364 
365 	pa = pte_pfn(*ptep);
366 
367 	/* On radix we can do hugepage mappings for io, so handle that */
368 	if (hugepage_shift) {
369 		pa <<= hugepage_shift;
370 		pa |= token & ((1ul << hugepage_shift) - 1);
371 	} else {
372 		pa <<= PAGE_SHIFT;
373 		pa |= token & (PAGE_SIZE - 1);
374 	}
375 
376 	return pa;
377 }
378 
379 /*
380  * On PowerNV platform, we might already have fenced PHB there.
381  * For that case, it's meaningless to recover frozen PE. Intead,
382  * We have to handle fenced PHB firstly.
383  */
eeh_phb_check_failure(struct eeh_pe * pe)384 static int eeh_phb_check_failure(struct eeh_pe *pe)
385 {
386 	struct eeh_pe *phb_pe;
387 	unsigned long flags;
388 	int ret;
389 
390 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
391 		return -EPERM;
392 
393 	/* Find the PHB PE */
394 	phb_pe = eeh_phb_pe_get(pe->phb);
395 	if (!phb_pe) {
396 		pr_warn("%s Can't find PE for PHB#%x\n",
397 			__func__, pe->phb->global_number);
398 		return -EEXIST;
399 	}
400 
401 	/* If the PHB has been in problematic state */
402 	eeh_serialize_lock(&flags);
403 	if (phb_pe->state & EEH_PE_ISOLATED) {
404 		ret = 0;
405 		goto out;
406 	}
407 
408 	/* Check PHB state */
409 	ret = eeh_ops->get_state(phb_pe, NULL);
410 	if ((ret < 0) ||
411 	    (ret == EEH_STATE_NOT_SUPPORT) ||
412 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
413 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
414 		ret = 0;
415 		goto out;
416 	}
417 
418 	/* Isolate the PHB and send event */
419 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
420 	eeh_serialize_unlock(flags);
421 
422 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
423 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
424 	dump_stack();
425 	eeh_send_failure_event(phb_pe);
426 
427 	return 1;
428 out:
429 	eeh_serialize_unlock(flags);
430 	return ret;
431 }
432 
433 /**
434  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
435  * @edev: eeh device
436  *
437  * Check for an EEH failure for the given device node.  Call this
438  * routine if the result of a read was all 0xff's and you want to
439  * find out if this is due to an EEH slot freeze.  This routine
440  * will query firmware for the EEH status.
441  *
442  * Returns 0 if there has not been an EEH error; otherwise returns
443  * a non-zero value and queues up a slot isolation event notification.
444  *
445  * It is safe to call this routine in an interrupt context.
446  */
eeh_dev_check_failure(struct eeh_dev * edev)447 int eeh_dev_check_failure(struct eeh_dev *edev)
448 {
449 	int ret;
450 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
451 	unsigned long flags;
452 	struct device_node *dn;
453 	struct pci_dev *dev;
454 	struct eeh_pe *pe, *parent_pe, *phb_pe;
455 	int rc = 0;
456 	const char *location = NULL;
457 
458 	eeh_stats.total_mmio_ffs++;
459 
460 	if (!eeh_enabled())
461 		return 0;
462 
463 	if (!edev) {
464 		eeh_stats.no_dn++;
465 		return 0;
466 	}
467 	dev = eeh_dev_to_pci_dev(edev);
468 	pe = eeh_dev_to_pe(edev);
469 
470 	/* Access to IO BARs might get this far and still not want checking. */
471 	if (!pe) {
472 		eeh_stats.ignored_check++;
473 		pr_debug("EEH: Ignored check for %s\n",
474 			eeh_pci_name(dev));
475 		return 0;
476 	}
477 
478 	if (!pe->addr && !pe->config_addr) {
479 		eeh_stats.no_cfg_addr++;
480 		return 0;
481 	}
482 
483 	/*
484 	 * On PowerNV platform, we might already have fenced PHB
485 	 * there and we need take care of that firstly.
486 	 */
487 	ret = eeh_phb_check_failure(pe);
488 	if (ret > 0)
489 		return ret;
490 
491 	/*
492 	 * If the PE isn't owned by us, we shouldn't check the
493 	 * state. Instead, let the owner handle it if the PE has
494 	 * been frozen.
495 	 */
496 	if (eeh_pe_passed(pe))
497 		return 0;
498 
499 	/* If we already have a pending isolation event for this
500 	 * slot, we know it's bad already, we don't need to check.
501 	 * Do this checking under a lock; as multiple PCI devices
502 	 * in one slot might report errors simultaneously, and we
503 	 * only want one error recovery routine running.
504 	 */
505 	eeh_serialize_lock(&flags);
506 	rc = 1;
507 	if (pe->state & EEH_PE_ISOLATED) {
508 		pe->check_count++;
509 		if (pe->check_count % EEH_MAX_FAILS == 0) {
510 			dn = pci_device_to_OF_node(dev);
511 			if (dn)
512 				location = of_get_property(dn, "ibm,loc-code",
513 						NULL);
514 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
515 				"location=%s driver=%s pci addr=%s\n",
516 				pe->check_count,
517 				location ? location : "unknown",
518 				eeh_driver_name(dev), eeh_pci_name(dev));
519 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
520 				eeh_driver_name(dev));
521 			dump_stack();
522 		}
523 		goto dn_unlock;
524 	}
525 
526 	/*
527 	 * Now test for an EEH failure.  This is VERY expensive.
528 	 * Note that the eeh_config_addr may be a parent device
529 	 * in the case of a device behind a bridge, or it may be
530 	 * function zero of a multi-function device.
531 	 * In any case they must share a common PHB.
532 	 */
533 	ret = eeh_ops->get_state(pe, NULL);
534 
535 	/* Note that config-io to empty slots may fail;
536 	 * they are empty when they don't have children.
537 	 * We will punt with the following conditions: Failure to get
538 	 * PE's state, EEH not support and Permanently unavailable
539 	 * state, PE is in good state.
540 	 */
541 	if ((ret < 0) ||
542 	    (ret == EEH_STATE_NOT_SUPPORT) ||
543 	    ((ret & active_flags) == active_flags)) {
544 		eeh_stats.false_positives++;
545 		pe->false_positives++;
546 		rc = 0;
547 		goto dn_unlock;
548 	}
549 
550 	/*
551 	 * It should be corner case that the parent PE has been
552 	 * put into frozen state as well. We should take care
553 	 * that at first.
554 	 */
555 	parent_pe = pe->parent;
556 	while (parent_pe) {
557 		/* Hit the ceiling ? */
558 		if (parent_pe->type & EEH_PE_PHB)
559 			break;
560 
561 		/* Frozen parent PE ? */
562 		ret = eeh_ops->get_state(parent_pe, NULL);
563 		if (ret > 0 &&
564 		    (ret & active_flags) != active_flags)
565 			pe = parent_pe;
566 
567 		/* Next parent level */
568 		parent_pe = parent_pe->parent;
569 	}
570 
571 	eeh_stats.slot_resets++;
572 
573 	/* Avoid repeated reports of this failure, including problems
574 	 * with other functions on this device, and functions under
575 	 * bridges.
576 	 */
577 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
578 	eeh_serialize_unlock(flags);
579 
580 	/* Most EEH events are due to device driver bugs.  Having
581 	 * a stack trace will help the device-driver authors figure
582 	 * out what happened.  So print that out.
583 	 */
584 	phb_pe = eeh_phb_pe_get(pe->phb);
585 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
586 	       pe->phb->global_number, pe->addr);
587 	pr_err("EEH: PE location: %s, PHB location: %s\n",
588 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
589 	dump_stack();
590 
591 	eeh_send_failure_event(pe);
592 
593 	return 1;
594 
595 dn_unlock:
596 	eeh_serialize_unlock(flags);
597 	return rc;
598 }
599 
600 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
601 
602 /**
603  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
604  * @token: I/O address
605  *
606  * Check for an EEH failure at the given I/O address. Call this
607  * routine if the result of a read was all 0xff's and you want to
608  * find out if this is due to an EEH slot freeze event. This routine
609  * will query firmware for the EEH status.
610  *
611  * Note this routine is safe to call in an interrupt context.
612  */
eeh_check_failure(const volatile void __iomem * token)613 int eeh_check_failure(const volatile void __iomem *token)
614 {
615 	unsigned long addr;
616 	struct eeh_dev *edev;
617 
618 	/* Finding the phys addr + pci device; this is pretty quick. */
619 	addr = eeh_token_to_phys((unsigned long __force) token);
620 	edev = eeh_addr_cache_get_dev(addr);
621 	if (!edev) {
622 		eeh_stats.no_device++;
623 		return 0;
624 	}
625 
626 	return eeh_dev_check_failure(edev);
627 }
628 EXPORT_SYMBOL(eeh_check_failure);
629 
630 
631 /**
632  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
633  * @pe: EEH PE
634  *
635  * This routine should be called to reenable frozen MMIO or DMA
636  * so that it would work correctly again. It's useful while doing
637  * recovery or log collection on the indicated device.
638  */
eeh_pci_enable(struct eeh_pe * pe,int function)639 int eeh_pci_enable(struct eeh_pe *pe, int function)
640 {
641 	int active_flag, rc;
642 
643 	/*
644 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
645 	 * Also, it's pointless to enable them on unfrozen PE. So
646 	 * we have to check before enabling IO or DMA.
647 	 */
648 	switch (function) {
649 	case EEH_OPT_THAW_MMIO:
650 		active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
651 		break;
652 	case EEH_OPT_THAW_DMA:
653 		active_flag = EEH_STATE_DMA_ACTIVE;
654 		break;
655 	case EEH_OPT_DISABLE:
656 	case EEH_OPT_ENABLE:
657 	case EEH_OPT_FREEZE_PE:
658 		active_flag = 0;
659 		break;
660 	default:
661 		pr_warn("%s: Invalid function %d\n",
662 			__func__, function);
663 		return -EINVAL;
664 	}
665 
666 	/*
667 	 * Check if IO or DMA has been enabled before
668 	 * enabling them.
669 	 */
670 	if (active_flag) {
671 		rc = eeh_ops->get_state(pe, NULL);
672 		if (rc < 0)
673 			return rc;
674 
675 		/* Needn't enable it at all */
676 		if (rc == EEH_STATE_NOT_SUPPORT)
677 			return 0;
678 
679 		/* It's already enabled */
680 		if (rc & active_flag)
681 			return 0;
682 	}
683 
684 
685 	/* Issue the request */
686 	rc = eeh_ops->set_option(pe, function);
687 	if (rc)
688 		pr_warn("%s: Unexpected state change %d on "
689 			"PHB#%x-PE#%x, err=%d\n",
690 			__func__, function, pe->phb->global_number,
691 			pe->addr, rc);
692 
693 	/* Check if the request is finished successfully */
694 	if (active_flag) {
695 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
696 		if (rc < 0)
697 			return rc;
698 
699 		if (rc & active_flag)
700 			return 0;
701 
702 		return -EIO;
703 	}
704 
705 	return rc;
706 }
707 
eeh_disable_and_save_dev_state(void * data,void * userdata)708 static void *eeh_disable_and_save_dev_state(void *data, void *userdata)
709 {
710 	struct eeh_dev *edev = data;
711 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
712 	struct pci_dev *dev = userdata;
713 
714 	/*
715 	 * The caller should have disabled and saved the
716 	 * state for the specified device
717 	 */
718 	if (!pdev || pdev == dev)
719 		return NULL;
720 
721 	/* Ensure we have D0 power state */
722 	pci_set_power_state(pdev, PCI_D0);
723 
724 	/* Save device state */
725 	pci_save_state(pdev);
726 
727 	/*
728 	 * Disable device to avoid any DMA traffic and
729 	 * interrupt from the device
730 	 */
731 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
732 
733 	return NULL;
734 }
735 
eeh_restore_dev_state(void * data,void * userdata)736 static void *eeh_restore_dev_state(void *data, void *userdata)
737 {
738 	struct eeh_dev *edev = data;
739 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
740 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
741 	struct pci_dev *dev = userdata;
742 
743 	if (!pdev)
744 		return NULL;
745 
746 	/* Apply customization from firmware */
747 	if (pdn && eeh_ops->restore_config)
748 		eeh_ops->restore_config(pdn);
749 
750 	/* The caller should restore state for the specified device */
751 	if (pdev != dev)
752 		pci_restore_state(pdev);
753 
754 	return NULL;
755 }
756 
757 /**
758  * pcibios_set_pcie_reset_state - Set PCI-E reset state
759  * @dev: pci device struct
760  * @state: reset state to enter
761  *
762  * Return value:
763  * 	0 if success
764  */
pcibios_set_pcie_reset_state(struct pci_dev * dev,enum pcie_reset_state state)765 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
766 {
767 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
768 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
769 
770 	if (!pe) {
771 		pr_err("%s: No PE found on PCI device %s\n",
772 			__func__, pci_name(dev));
773 		return -EINVAL;
774 	}
775 
776 	switch (state) {
777 	case pcie_deassert_reset:
778 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
779 		eeh_unfreeze_pe(pe, false);
780 		if (!(pe->type & EEH_PE_VF))
781 			eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
782 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
783 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
784 		break;
785 	case pcie_hot_reset:
786 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
787 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
788 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
789 		if (!(pe->type & EEH_PE_VF))
790 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
791 		eeh_ops->reset(pe, EEH_RESET_HOT);
792 		break;
793 	case pcie_warm_reset:
794 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
795 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
796 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
797 		if (!(pe->type & EEH_PE_VF))
798 			eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
799 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
800 		break;
801 	default:
802 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
803 		return -EINVAL;
804 	};
805 
806 	return 0;
807 }
808 
809 /**
810  * eeh_set_pe_freset - Check the required reset for the indicated device
811  * @data: EEH device
812  * @flag: return value
813  *
814  * Each device might have its preferred reset type: fundamental or
815  * hot reset. The routine is used to collected the information for
816  * the indicated device and its children so that the bunch of the
817  * devices could be reset properly.
818  */
eeh_set_dev_freset(void * data,void * flag)819 static void *eeh_set_dev_freset(void *data, void *flag)
820 {
821 	struct pci_dev *dev;
822 	unsigned int *freset = (unsigned int *)flag;
823 	struct eeh_dev *edev = (struct eeh_dev *)data;
824 
825 	dev = eeh_dev_to_pci_dev(edev);
826 	if (dev)
827 		*freset |= dev->needs_freset;
828 
829 	return NULL;
830 }
831 
832 /**
833  * eeh_pe_reset_full - Complete a full reset process on the indicated PE
834  * @pe: EEH PE
835  *
836  * This function executes a full reset procedure on a PE, including setting
837  * the appropriate flags, performing a fundamental or hot reset, and then
838  * deactivating the reset status.  It is designed to be used within the EEH
839  * subsystem, as opposed to eeh_pe_reset which is exported to drivers and
840  * only performs a single operation at a time.
841  *
842  * This function will attempt to reset a PE three times before failing.
843  */
eeh_pe_reset_full(struct eeh_pe * pe)844 int eeh_pe_reset_full(struct eeh_pe *pe)
845 {
846 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
847 	int reset_state = (EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
848 	int type = EEH_RESET_HOT;
849 	unsigned int freset = 0;
850 	int i, state, ret;
851 
852 	/*
853 	 * Determine the type of reset to perform - hot or fundamental.
854 	 * Hot reset is the default operation, unless any device under the
855 	 * PE requires a fundamental reset.
856 	 */
857 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
858 
859 	if (freset)
860 		type = EEH_RESET_FUNDAMENTAL;
861 
862 	/* Mark the PE as in reset state and block config space accesses */
863 	eeh_pe_state_mark(pe, reset_state);
864 
865 	/* Make three attempts at resetting the bus */
866 	for (i = 0; i < 3; i++) {
867 		ret = eeh_pe_reset(pe, type);
868 		if (ret)
869 			break;
870 
871 		ret = eeh_pe_reset(pe, EEH_RESET_DEACTIVATE);
872 		if (ret)
873 			break;
874 
875 		/* Wait until the PE is in a functioning state */
876 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
877 		if ((state & active_flags) == active_flags)
878 			break;
879 
880 		if (state < 0) {
881 			pr_warn("%s: Unrecoverable slot failure on PHB#%x-PE#%x",
882 				__func__, pe->phb->global_number, pe->addr);
883 			ret = -ENOTRECOVERABLE;
884 			break;
885 		}
886 
887 		/* Set error in case this is our last attempt */
888 		ret = -EIO;
889 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
890 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
891 	}
892 
893 	eeh_pe_state_clear(pe, reset_state);
894 	return ret;
895 }
896 
897 /**
898  * eeh_save_bars - Save device bars
899  * @edev: PCI device associated EEH device
900  *
901  * Save the values of the device bars. Unlike the restore
902  * routine, this routine is *not* recursive. This is because
903  * PCI devices are added individually; but, for the restore,
904  * an entire slot is reset at a time.
905  */
eeh_save_bars(struct eeh_dev * edev)906 void eeh_save_bars(struct eeh_dev *edev)
907 {
908 	struct pci_dn *pdn;
909 	int i;
910 
911 	pdn = eeh_dev_to_pdn(edev);
912 	if (!pdn)
913 		return;
914 
915 	for (i = 0; i < 16; i++)
916 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
917 
918 	/*
919 	 * For PCI bridges including root port, we need enable bus
920 	 * master explicitly. Otherwise, it can't fetch IODA table
921 	 * entries correctly. So we cache the bit in advance so that
922 	 * we can restore it after reset, either PHB range or PE range.
923 	 */
924 	if (edev->mode & EEH_DEV_BRIDGE)
925 		edev->config_space[1] |= PCI_COMMAND_MASTER;
926 }
927 
928 /**
929  * eeh_ops_register - Register platform dependent EEH operations
930  * @ops: platform dependent EEH operations
931  *
932  * Register the platform dependent EEH operation callback
933  * functions. The platform should call this function before
934  * any other EEH operations.
935  */
eeh_ops_register(struct eeh_ops * ops)936 int __init eeh_ops_register(struct eeh_ops *ops)
937 {
938 	if (!ops->name) {
939 		pr_warn("%s: Invalid EEH ops name for %p\n",
940 			__func__, ops);
941 		return -EINVAL;
942 	}
943 
944 	if (eeh_ops && eeh_ops != ops) {
945 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
946 			__func__, eeh_ops->name, ops->name);
947 		return -EEXIST;
948 	}
949 
950 	eeh_ops = ops;
951 
952 	return 0;
953 }
954 
955 /**
956  * eeh_ops_unregister - Unreigster platform dependent EEH operations
957  * @name: name of EEH platform operations
958  *
959  * Unregister the platform dependent EEH operation callback
960  * functions.
961  */
eeh_ops_unregister(const char * name)962 int __exit eeh_ops_unregister(const char *name)
963 {
964 	if (!name || !strlen(name)) {
965 		pr_warn("%s: Invalid EEH ops name\n",
966 			__func__);
967 		return -EINVAL;
968 	}
969 
970 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
971 		eeh_ops = NULL;
972 		return 0;
973 	}
974 
975 	return -EEXIST;
976 }
977 
eeh_reboot_notifier(struct notifier_block * nb,unsigned long action,void * unused)978 static int eeh_reboot_notifier(struct notifier_block *nb,
979 			       unsigned long action, void *unused)
980 {
981 	eeh_clear_flag(EEH_ENABLED);
982 	return NOTIFY_DONE;
983 }
984 
985 static struct notifier_block eeh_reboot_nb = {
986 	.notifier_call = eeh_reboot_notifier,
987 };
988 
989 /**
990  * eeh_init - EEH initialization
991  *
992  * Initialize EEH by trying to enable it for all of the adapters in the system.
993  * As a side effect we can determine here if eeh is supported at all.
994  * Note that we leave EEH on so failed config cycles won't cause a machine
995  * check.  If a user turns off EEH for a particular adapter they are really
996  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
997  * grant access to a slot if EEH isn't enabled, and so we always enable
998  * EEH for all slots/all devices.
999  *
1000  * The eeh-force-off option disables EEH checking globally, for all slots.
1001  * Even if force-off is set, the EEH hardware is still enabled, so that
1002  * newer systems can boot.
1003  */
eeh_init(void)1004 int eeh_init(void)
1005 {
1006 	struct pci_controller *hose, *tmp;
1007 	struct pci_dn *pdn;
1008 	static int cnt = 0;
1009 	int ret = 0;
1010 
1011 	/*
1012 	 * We have to delay the initialization on PowerNV after
1013 	 * the PCI hierarchy tree has been built because the PEs
1014 	 * are figured out based on PCI devices instead of device
1015 	 * tree nodes
1016 	 */
1017 	if (machine_is(powernv) && cnt++ <= 0)
1018 		return ret;
1019 
1020 	/* Register reboot notifier */
1021 	ret = register_reboot_notifier(&eeh_reboot_nb);
1022 	if (ret) {
1023 		pr_warn("%s: Failed to register notifier (%d)\n",
1024 			__func__, ret);
1025 		return ret;
1026 	}
1027 
1028 	/* call platform initialization function */
1029 	if (!eeh_ops) {
1030 		pr_warn("%s: Platform EEH operation not found\n",
1031 			__func__);
1032 		return -EEXIST;
1033 	} else if ((ret = eeh_ops->init()))
1034 		return ret;
1035 
1036 	/* Initialize PHB PEs */
1037 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
1038 		eeh_dev_phb_init_dynamic(hose);
1039 
1040 	/* Initialize EEH event */
1041 	ret = eeh_event_init();
1042 	if (ret)
1043 		return ret;
1044 
1045 	/* Enable EEH for all adapters */
1046 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1047 		pdn = hose->pci_data;
1048 		traverse_pci_dn(pdn, eeh_ops->probe, NULL);
1049 	}
1050 
1051 	/*
1052 	 * Call platform post-initialization. Actually, It's good chance
1053 	 * to inform platform that EEH is ready to supply service if the
1054 	 * I/O cache stuff has been built up.
1055 	 */
1056 	if (eeh_ops->post_init) {
1057 		ret = eeh_ops->post_init();
1058 		if (ret)
1059 			return ret;
1060 	}
1061 
1062 	if (eeh_enabled())
1063 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
1064 	else
1065 		pr_info("EEH: No capable adapters found\n");
1066 
1067 	return ret;
1068 }
1069 
1070 core_initcall_sync(eeh_init);
1071 
1072 /**
1073  * eeh_add_device_early - Enable EEH for the indicated device node
1074  * @pdn: PCI device node for which to set up EEH
1075  *
1076  * This routine must be used to perform EEH initialization for PCI
1077  * devices that were added after system boot (e.g. hotplug, dlpar).
1078  * This routine must be called before any i/o is performed to the
1079  * adapter (inluding any config-space i/o).
1080  * Whether this actually enables EEH or not for this device depends
1081  * on the CEC architecture, type of the device, on earlier boot
1082  * command-line arguments & etc.
1083  */
eeh_add_device_early(struct pci_dn * pdn)1084 void eeh_add_device_early(struct pci_dn *pdn)
1085 {
1086 	struct pci_controller *phb = pdn ? pdn->phb : NULL;
1087 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1088 
1089 	if (!edev)
1090 		return;
1091 
1092 	if (!eeh_has_flag(EEH_PROBE_MODE_DEVTREE))
1093 		return;
1094 
1095 	/* USB Bus children of PCI devices will not have BUID's */
1096 	if (NULL == phb ||
1097 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1098 		return;
1099 
1100 	eeh_ops->probe(pdn, NULL);
1101 }
1102 
1103 /**
1104  * eeh_add_device_tree_early - Enable EEH for the indicated device
1105  * @pdn: PCI device node
1106  *
1107  * This routine must be used to perform EEH initialization for the
1108  * indicated PCI device that was added after system boot (e.g.
1109  * hotplug, dlpar).
1110  */
eeh_add_device_tree_early(struct pci_dn * pdn)1111 void eeh_add_device_tree_early(struct pci_dn *pdn)
1112 {
1113 	struct pci_dn *n;
1114 
1115 	if (!pdn)
1116 		return;
1117 
1118 	list_for_each_entry(n, &pdn->child_list, list)
1119 		eeh_add_device_tree_early(n);
1120 	eeh_add_device_early(pdn);
1121 }
1122 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1123 
1124 /**
1125  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1126  * @dev: pci device for which to set up EEH
1127  *
1128  * This routine must be used to complete EEH initialization for PCI
1129  * devices that were added after system boot (e.g. hotplug, dlpar).
1130  */
eeh_add_device_late(struct pci_dev * dev)1131 void eeh_add_device_late(struct pci_dev *dev)
1132 {
1133 	struct pci_dn *pdn;
1134 	struct eeh_dev *edev;
1135 
1136 	if (!dev || !eeh_enabled())
1137 		return;
1138 
1139 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1140 
1141 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1142 	edev = pdn_to_eeh_dev(pdn);
1143 	if (edev->pdev == dev) {
1144 		pr_debug("EEH: Already referenced !\n");
1145 		return;
1146 	}
1147 
1148 	/*
1149 	 * The EEH cache might not be removed correctly because of
1150 	 * unbalanced kref to the device during unplug time, which
1151 	 * relies on pcibios_release_device(). So we have to remove
1152 	 * that here explicitly.
1153 	 */
1154 	if (edev->pdev) {
1155 		eeh_rmv_from_parent_pe(edev);
1156 		eeh_addr_cache_rmv_dev(edev->pdev);
1157 		eeh_sysfs_remove_device(edev->pdev);
1158 		edev->mode &= ~EEH_DEV_SYSFS;
1159 
1160 		/*
1161 		 * We definitely should have the PCI device removed
1162 		 * though it wasn't correctly. So we needn't call
1163 		 * into error handler afterwards.
1164 		 */
1165 		edev->mode |= EEH_DEV_NO_HANDLER;
1166 
1167 		edev->pdev = NULL;
1168 		dev->dev.archdata.edev = NULL;
1169 	}
1170 
1171 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1172 		eeh_ops->probe(pdn, NULL);
1173 
1174 	edev->pdev = dev;
1175 	dev->dev.archdata.edev = edev;
1176 
1177 	eeh_addr_cache_insert_dev(dev);
1178 }
1179 
1180 /**
1181  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1182  * @bus: PCI bus
1183  *
1184  * This routine must be used to perform EEH initialization for PCI
1185  * devices which are attached to the indicated PCI bus. The PCI bus
1186  * is added after system boot through hotplug or dlpar.
1187  */
eeh_add_device_tree_late(struct pci_bus * bus)1188 void eeh_add_device_tree_late(struct pci_bus *bus)
1189 {
1190 	struct pci_dev *dev;
1191 
1192 	list_for_each_entry(dev, &bus->devices, bus_list) {
1193 		eeh_add_device_late(dev);
1194 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1195 			struct pci_bus *subbus = dev->subordinate;
1196 			if (subbus)
1197 				eeh_add_device_tree_late(subbus);
1198 		}
1199 	}
1200 }
1201 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1202 
1203 /**
1204  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1205  * @bus: PCI bus
1206  *
1207  * This routine must be used to add EEH sysfs files for PCI
1208  * devices which are attached to the indicated PCI bus. The PCI bus
1209  * is added after system boot through hotplug or dlpar.
1210  */
eeh_add_sysfs_files(struct pci_bus * bus)1211 void eeh_add_sysfs_files(struct pci_bus *bus)
1212 {
1213 	struct pci_dev *dev;
1214 
1215 	list_for_each_entry(dev, &bus->devices, bus_list) {
1216 		eeh_sysfs_add_device(dev);
1217 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1218 			struct pci_bus *subbus = dev->subordinate;
1219 			if (subbus)
1220 				eeh_add_sysfs_files(subbus);
1221 		}
1222 	}
1223 }
1224 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1225 
1226 /**
1227  * eeh_remove_device - Undo EEH setup for the indicated pci device
1228  * @dev: pci device to be removed
1229  *
1230  * This routine should be called when a device is removed from
1231  * a running system (e.g. by hotplug or dlpar).  It unregisters
1232  * the PCI device from the EEH subsystem.  I/O errors affecting
1233  * this device will no longer be detected after this call; thus,
1234  * i/o errors affecting this slot may leave this device unusable.
1235  */
eeh_remove_device(struct pci_dev * dev)1236 void eeh_remove_device(struct pci_dev *dev)
1237 {
1238 	struct eeh_dev *edev;
1239 
1240 	if (!dev || !eeh_enabled())
1241 		return;
1242 	edev = pci_dev_to_eeh_dev(dev);
1243 
1244 	/* Unregister the device with the EEH/PCI address search system */
1245 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1246 
1247 	if (!edev || !edev->pdev || !edev->pe) {
1248 		pr_debug("EEH: Not referenced !\n");
1249 		return;
1250 	}
1251 
1252 	/*
1253 	 * During the hotplug for EEH error recovery, we need the EEH
1254 	 * device attached to the parent PE in order for BAR restore
1255 	 * a bit later. So we keep it for BAR restore and remove it
1256 	 * from the parent PE during the BAR resotre.
1257 	 */
1258 	edev->pdev = NULL;
1259 
1260 	/*
1261 	 * The flag "in_error" is used to trace EEH devices for VFs
1262 	 * in error state or not. It's set in eeh_report_error(). If
1263 	 * it's not set, eeh_report_{reset,resume}() won't be called
1264 	 * for the VF EEH device.
1265 	 */
1266 	edev->in_error = false;
1267 	dev->dev.archdata.edev = NULL;
1268 	if (!(edev->pe->state & EEH_PE_KEEP))
1269 		eeh_rmv_from_parent_pe(edev);
1270 	else
1271 		edev->mode |= EEH_DEV_DISCONNECTED;
1272 
1273 	/*
1274 	 * We're removing from the PCI subsystem, that means
1275 	 * the PCI device driver can't support EEH or not
1276 	 * well. So we rely on hotplug completely to do recovery
1277 	 * for the specific PCI device.
1278 	 */
1279 	edev->mode |= EEH_DEV_NO_HANDLER;
1280 
1281 	eeh_addr_cache_rmv_dev(dev);
1282 	eeh_sysfs_remove_device(dev);
1283 	edev->mode &= ~EEH_DEV_SYSFS;
1284 }
1285 
eeh_unfreeze_pe(struct eeh_pe * pe,bool sw_state)1286 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1287 {
1288 	int ret;
1289 
1290 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1291 	if (ret) {
1292 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1293 			__func__, ret, pe->phb->global_number, pe->addr);
1294 		return ret;
1295 	}
1296 
1297 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1298 	if (ret) {
1299 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1300 			__func__, ret, pe->phb->global_number, pe->addr);
1301 		return ret;
1302 	}
1303 
1304 	/* Clear software isolated state */
1305 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1306 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1307 
1308 	return ret;
1309 }
1310 
1311 
1312 static struct pci_device_id eeh_reset_ids[] = {
1313 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1314 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1315 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1316 	{ 0 }
1317 };
1318 
eeh_pe_change_owner(struct eeh_pe * pe)1319 static int eeh_pe_change_owner(struct eeh_pe *pe)
1320 {
1321 	struct eeh_dev *edev, *tmp;
1322 	struct pci_dev *pdev;
1323 	struct pci_device_id *id;
1324 	int flags, ret;
1325 
1326 	/* Check PE state */
1327 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1328 	ret = eeh_ops->get_state(pe, NULL);
1329 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1330 		return 0;
1331 
1332 	/* Unfrozen PE, nothing to do */
1333 	if ((ret & flags) == flags)
1334 		return 0;
1335 
1336 	/* Frozen PE, check if it needs PE level reset */
1337 	eeh_pe_for_each_dev(pe, edev, tmp) {
1338 		pdev = eeh_dev_to_pci_dev(edev);
1339 		if (!pdev)
1340 			continue;
1341 
1342 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1343 			if (id->vendor != PCI_ANY_ID &&
1344 			    id->vendor != pdev->vendor)
1345 				continue;
1346 			if (id->device != PCI_ANY_ID &&
1347 			    id->device != pdev->device)
1348 				continue;
1349 			if (id->subvendor != PCI_ANY_ID &&
1350 			    id->subvendor != pdev->subsystem_vendor)
1351 				continue;
1352 			if (id->subdevice != PCI_ANY_ID &&
1353 			    id->subdevice != pdev->subsystem_device)
1354 				continue;
1355 
1356 			return eeh_pe_reset_and_recover(pe);
1357 		}
1358 	}
1359 
1360 	return eeh_unfreeze_pe(pe, true);
1361 }
1362 
1363 /**
1364  * eeh_dev_open - Increase count of pass through devices for PE
1365  * @pdev: PCI device
1366  *
1367  * Increase count of passed through devices for the indicated
1368  * PE. In the result, the EEH errors detected on the PE won't be
1369  * reported. The PE owner will be responsible for detection
1370  * and recovery.
1371  */
eeh_dev_open(struct pci_dev * pdev)1372 int eeh_dev_open(struct pci_dev *pdev)
1373 {
1374 	struct eeh_dev *edev;
1375 	int ret = -ENODEV;
1376 
1377 	mutex_lock(&eeh_dev_mutex);
1378 
1379 	/* No PCI device ? */
1380 	if (!pdev)
1381 		goto out;
1382 
1383 	/* No EEH device or PE ? */
1384 	edev = pci_dev_to_eeh_dev(pdev);
1385 	if (!edev || !edev->pe)
1386 		goto out;
1387 
1388 	/*
1389 	 * The PE might have been put into frozen state, but we
1390 	 * didn't detect that yet. The passed through PCI devices
1391 	 * in frozen PE won't work properly. Clear the frozen state
1392 	 * in advance.
1393 	 */
1394 	ret = eeh_pe_change_owner(edev->pe);
1395 	if (ret)
1396 		goto out;
1397 
1398 	/* Increase PE's pass through count */
1399 	atomic_inc(&edev->pe->pass_dev_cnt);
1400 	mutex_unlock(&eeh_dev_mutex);
1401 
1402 	return 0;
1403 out:
1404 	mutex_unlock(&eeh_dev_mutex);
1405 	return ret;
1406 }
1407 EXPORT_SYMBOL_GPL(eeh_dev_open);
1408 
1409 /**
1410  * eeh_dev_release - Decrease count of pass through devices for PE
1411  * @pdev: PCI device
1412  *
1413  * Decrease count of pass through devices for the indicated PE. If
1414  * there is no passed through device in PE, the EEH errors detected
1415  * on the PE will be reported and handled as usual.
1416  */
eeh_dev_release(struct pci_dev * pdev)1417 void eeh_dev_release(struct pci_dev *pdev)
1418 {
1419 	struct eeh_dev *edev;
1420 
1421 	mutex_lock(&eeh_dev_mutex);
1422 
1423 	/* No PCI device ? */
1424 	if (!pdev)
1425 		goto out;
1426 
1427 	/* No EEH device ? */
1428 	edev = pci_dev_to_eeh_dev(pdev);
1429 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1430 		goto out;
1431 
1432 	/* Decrease PE's pass through count */
1433 	WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1434 	eeh_pe_change_owner(edev->pe);
1435 out:
1436 	mutex_unlock(&eeh_dev_mutex);
1437 }
1438 EXPORT_SYMBOL(eeh_dev_release);
1439 
1440 #ifdef CONFIG_IOMMU_API
1441 
dev_has_iommu_table(struct device * dev,void * data)1442 static int dev_has_iommu_table(struct device *dev, void *data)
1443 {
1444 	struct pci_dev *pdev = to_pci_dev(dev);
1445 	struct pci_dev **ppdev = data;
1446 
1447 	if (!dev)
1448 		return 0;
1449 
1450 	if (dev->iommu_group) {
1451 		*ppdev = pdev;
1452 		return 1;
1453 	}
1454 
1455 	return 0;
1456 }
1457 
1458 /**
1459  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1460  * @group: IOMMU group
1461  *
1462  * The routine is called to convert IOMMU group to EEH PE.
1463  */
eeh_iommu_group_to_pe(struct iommu_group * group)1464 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1465 {
1466 	struct pci_dev *pdev = NULL;
1467 	struct eeh_dev *edev;
1468 	int ret;
1469 
1470 	/* No IOMMU group ? */
1471 	if (!group)
1472 		return NULL;
1473 
1474 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1475 	if (!ret || !pdev)
1476 		return NULL;
1477 
1478 	/* No EEH device or PE ? */
1479 	edev = pci_dev_to_eeh_dev(pdev);
1480 	if (!edev || !edev->pe)
1481 		return NULL;
1482 
1483 	return edev->pe;
1484 }
1485 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1486 
1487 #endif /* CONFIG_IOMMU_API */
1488 
1489 /**
1490  * eeh_pe_set_option - Set options for the indicated PE
1491  * @pe: EEH PE
1492  * @option: requested option
1493  *
1494  * The routine is called to enable or disable EEH functionality
1495  * on the indicated PE, to enable IO or DMA for the frozen PE.
1496  */
eeh_pe_set_option(struct eeh_pe * pe,int option)1497 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1498 {
1499 	int ret = 0;
1500 
1501 	/* Invalid PE ? */
1502 	if (!pe)
1503 		return -ENODEV;
1504 
1505 	/*
1506 	 * EEH functionality could possibly be disabled, just
1507 	 * return error for the case. And the EEH functinality
1508 	 * isn't expected to be disabled on one specific PE.
1509 	 */
1510 	switch (option) {
1511 	case EEH_OPT_ENABLE:
1512 		if (eeh_enabled()) {
1513 			ret = eeh_pe_change_owner(pe);
1514 			break;
1515 		}
1516 		ret = -EIO;
1517 		break;
1518 	case EEH_OPT_DISABLE:
1519 		break;
1520 	case EEH_OPT_THAW_MMIO:
1521 	case EEH_OPT_THAW_DMA:
1522 	case EEH_OPT_FREEZE_PE:
1523 		if (!eeh_ops || !eeh_ops->set_option) {
1524 			ret = -ENOENT;
1525 			break;
1526 		}
1527 
1528 		ret = eeh_pci_enable(pe, option);
1529 		break;
1530 	default:
1531 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1532 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1533 		ret = -EINVAL;
1534 	}
1535 
1536 	return ret;
1537 }
1538 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1539 
1540 /**
1541  * eeh_pe_get_state - Retrieve PE's state
1542  * @pe: EEH PE
1543  *
1544  * Retrieve the PE's state, which includes 3 aspects: enabled
1545  * DMA, enabled IO and asserted reset.
1546  */
eeh_pe_get_state(struct eeh_pe * pe)1547 int eeh_pe_get_state(struct eeh_pe *pe)
1548 {
1549 	int result, ret = 0;
1550 	bool rst_active, dma_en, mmio_en;
1551 
1552 	/* Existing PE ? */
1553 	if (!pe)
1554 		return -ENODEV;
1555 
1556 	if (!eeh_ops || !eeh_ops->get_state)
1557 		return -ENOENT;
1558 
1559 	/*
1560 	 * If the parent PE is owned by the host kernel and is undergoing
1561 	 * error recovery, we should return the PE state as temporarily
1562 	 * unavailable so that the error recovery on the guest is suspended
1563 	 * until the recovery completes on the host.
1564 	 */
1565 	if (pe->parent &&
1566 	    !(pe->state & EEH_PE_REMOVED) &&
1567 	    (pe->parent->state & (EEH_PE_ISOLATED | EEH_PE_RECOVERING)))
1568 		return EEH_PE_STATE_UNAVAIL;
1569 
1570 	result = eeh_ops->get_state(pe, NULL);
1571 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1572 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1573 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1574 
1575 	if (rst_active)
1576 		ret = EEH_PE_STATE_RESET;
1577 	else if (dma_en && mmio_en)
1578 		ret = EEH_PE_STATE_NORMAL;
1579 	else if (!dma_en && !mmio_en)
1580 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1581 	else if (!dma_en && mmio_en)
1582 		ret = EEH_PE_STATE_STOPPED_DMA;
1583 	else
1584 		ret = EEH_PE_STATE_UNAVAIL;
1585 
1586 	return ret;
1587 }
1588 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1589 
eeh_pe_reenable_devices(struct eeh_pe * pe)1590 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1591 {
1592 	struct eeh_dev *edev, *tmp;
1593 	struct pci_dev *pdev;
1594 	int ret = 0;
1595 
1596 	/* Restore config space */
1597 	eeh_pe_restore_bars(pe);
1598 
1599 	/*
1600 	 * Reenable PCI devices as the devices passed
1601 	 * through are always enabled before the reset.
1602 	 */
1603 	eeh_pe_for_each_dev(pe, edev, tmp) {
1604 		pdev = eeh_dev_to_pci_dev(edev);
1605 		if (!pdev)
1606 			continue;
1607 
1608 		ret = pci_reenable_device(pdev);
1609 		if (ret) {
1610 			pr_warn("%s: Failure %d reenabling %s\n",
1611 				__func__, ret, pci_name(pdev));
1612 			return ret;
1613 		}
1614 	}
1615 
1616 	/* The PE is still in frozen state */
1617 	return eeh_unfreeze_pe(pe, true);
1618 }
1619 
1620 
1621 /**
1622  * eeh_pe_reset - Issue PE reset according to specified type
1623  * @pe: EEH PE
1624  * @option: reset type
1625  *
1626  * The routine is called to reset the specified PE with the
1627  * indicated type, either fundamental reset or hot reset.
1628  * PE reset is the most important part for error recovery.
1629  */
eeh_pe_reset(struct eeh_pe * pe,int option)1630 int eeh_pe_reset(struct eeh_pe *pe, int option)
1631 {
1632 	int ret = 0;
1633 
1634 	/* Invalid PE ? */
1635 	if (!pe)
1636 		return -ENODEV;
1637 
1638 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1639 		return -ENOENT;
1640 
1641 	switch (option) {
1642 	case EEH_RESET_DEACTIVATE:
1643 		ret = eeh_ops->reset(pe, option);
1644 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1645 		if (ret)
1646 			break;
1647 
1648 		ret = eeh_pe_reenable_devices(pe);
1649 		break;
1650 	case EEH_RESET_HOT:
1651 	case EEH_RESET_FUNDAMENTAL:
1652 		/*
1653 		 * Proactively freeze the PE to drop all MMIO access
1654 		 * during reset, which should be banned as it's always
1655 		 * cause recursive EEH error.
1656 		 */
1657 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1658 
1659 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1660 		ret = eeh_ops->reset(pe, option);
1661 		break;
1662 	default:
1663 		pr_debug("%s: Unsupported option %d\n",
1664 			__func__, option);
1665 		ret = -EINVAL;
1666 	}
1667 
1668 	return ret;
1669 }
1670 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1671 
1672 /**
1673  * eeh_pe_configure - Configure PCI bridges after PE reset
1674  * @pe: EEH PE
1675  *
1676  * The routine is called to restore the PCI config space for
1677  * those PCI devices, especially PCI bridges affected by PE
1678  * reset issued previously.
1679  */
eeh_pe_configure(struct eeh_pe * pe)1680 int eeh_pe_configure(struct eeh_pe *pe)
1681 {
1682 	int ret = 0;
1683 
1684 	/* Invalid PE ? */
1685 	if (!pe)
1686 		return -ENODEV;
1687 
1688 	return ret;
1689 }
1690 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1691 
1692 /**
1693  * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1694  * @pe: the indicated PE
1695  * @type: error type
1696  * @function: error function
1697  * @addr: address
1698  * @mask: address mask
1699  *
1700  * The routine is called to inject the specified PCI error, which
1701  * is determined by @type and @function, to the indicated PE for
1702  * testing purpose.
1703  */
eeh_pe_inject_err(struct eeh_pe * pe,int type,int func,unsigned long addr,unsigned long mask)1704 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1705 		      unsigned long addr, unsigned long mask)
1706 {
1707 	/* Invalid PE ? */
1708 	if (!pe)
1709 		return -ENODEV;
1710 
1711 	/* Unsupported operation ? */
1712 	if (!eeh_ops || !eeh_ops->err_inject)
1713 		return -ENOENT;
1714 
1715 	/* Check on PCI error type */
1716 	if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1717 		return -EINVAL;
1718 
1719 	/* Check on PCI error function */
1720 	if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1721 		return -EINVAL;
1722 
1723 	return eeh_ops->err_inject(pe, type, func, addr, mask);
1724 }
1725 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1726 
proc_eeh_show(struct seq_file * m,void * v)1727 static int proc_eeh_show(struct seq_file *m, void *v)
1728 {
1729 	if (!eeh_enabled()) {
1730 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1731 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1732 	} else {
1733 		seq_printf(m, "EEH Subsystem is enabled\n");
1734 		seq_printf(m,
1735 				"no device=%llu\n"
1736 				"no device node=%llu\n"
1737 				"no config address=%llu\n"
1738 				"check not wanted=%llu\n"
1739 				"eeh_total_mmio_ffs=%llu\n"
1740 				"eeh_false_positives=%llu\n"
1741 				"eeh_slot_resets=%llu\n",
1742 				eeh_stats.no_device,
1743 				eeh_stats.no_dn,
1744 				eeh_stats.no_cfg_addr,
1745 				eeh_stats.ignored_check,
1746 				eeh_stats.total_mmio_ffs,
1747 				eeh_stats.false_positives,
1748 				eeh_stats.slot_resets);
1749 	}
1750 
1751 	return 0;
1752 }
1753 
proc_eeh_open(struct inode * inode,struct file * file)1754 static int proc_eeh_open(struct inode *inode, struct file *file)
1755 {
1756 	return single_open(file, proc_eeh_show, NULL);
1757 }
1758 
1759 static const struct file_operations proc_eeh_operations = {
1760 	.open      = proc_eeh_open,
1761 	.read      = seq_read,
1762 	.llseek    = seq_lseek,
1763 	.release   = single_release,
1764 };
1765 
1766 #ifdef CONFIG_DEBUG_FS
eeh_enable_dbgfs_set(void * data,u64 val)1767 static int eeh_enable_dbgfs_set(void *data, u64 val)
1768 {
1769 	if (val)
1770 		eeh_clear_flag(EEH_FORCE_DISABLED);
1771 	else
1772 		eeh_add_flag(EEH_FORCE_DISABLED);
1773 
1774 	/* Notify the backend */
1775 	if (eeh_ops->post_init)
1776 		eeh_ops->post_init();
1777 
1778 	return 0;
1779 }
1780 
eeh_enable_dbgfs_get(void * data,u64 * val)1781 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1782 {
1783 	if (eeh_enabled())
1784 		*val = 0x1ul;
1785 	else
1786 		*val = 0x0ul;
1787 	return 0;
1788 }
1789 
eeh_freeze_dbgfs_set(void * data,u64 val)1790 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1791 {
1792 	eeh_max_freezes = val;
1793 	return 0;
1794 }
1795 
eeh_freeze_dbgfs_get(void * data,u64 * val)1796 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1797 {
1798 	*val = eeh_max_freezes;
1799 	return 0;
1800 }
1801 
1802 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1803 			eeh_enable_dbgfs_set, "0x%llx\n");
1804 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1805 			eeh_freeze_dbgfs_set, "0x%llx\n");
1806 #endif
1807 
eeh_init_proc(void)1808 static int __init eeh_init_proc(void)
1809 {
1810 	if (machine_is(pseries) || machine_is(powernv)) {
1811 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1812 #ifdef CONFIG_DEBUG_FS
1813 		debugfs_create_file("eeh_enable", 0600,
1814                                     powerpc_debugfs_root, NULL,
1815                                     &eeh_enable_dbgfs_ops);
1816 		debugfs_create_file("eeh_max_freezes", 0600,
1817 				    powerpc_debugfs_root, NULL,
1818 				    &eeh_freeze_dbgfs_ops);
1819 #endif
1820 	}
1821 
1822 	return 0;
1823 }
1824 __initcall(eeh_init_proc);
1825