• 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/debugfs.h>
26 #include <linux/sched.h>
27 #include <linux/init.h>
28 #include <linux/list.h>
29 #include <linux/pci.h>
30 #include <linux/iommu.h>
31 #include <linux/proc_fs.h>
32 #include <linux/rbtree.h>
33 #include <linux/reboot.h>
34 #include <linux/seq_file.h>
35 #include <linux/spinlock.h>
36 #include <linux/export.h>
37 #include <linux/of.h>
38 
39 #include <linux/atomic.h>
40 #include <asm/debug.h>
41 #include <asm/eeh.h>
42 #include <asm/eeh_event.h>
43 #include <asm/io.h>
44 #include <asm/iommu.h>
45 #include <asm/machdep.h>
46 #include <asm/ppc-pci.h>
47 #include <asm/rtas.h>
48 
49 
50 /** Overview:
51  *  EEH, or "Extended Error Handling" is a PCI bridge technology for
52  *  dealing with PCI bus errors that can't be dealt with within the
53  *  usual PCI framework, except by check-stopping the CPU.  Systems
54  *  that are designed for high-availability/reliability cannot afford
55  *  to crash due to a "mere" PCI error, thus the need for EEH.
56  *  An EEH-capable bridge operates by converting a detected error
57  *  into a "slot freeze", taking the PCI adapter off-line, making
58  *  the slot behave, from the OS'es point of view, as if the slot
59  *  were "empty": all reads return 0xff's and all writes are silently
60  *  ignored.  EEH slot isolation events can be triggered by parity
61  *  errors on the address or data busses (e.g. during posted writes),
62  *  which in turn might be caused by low voltage on the bus, dust,
63  *  vibration, humidity, radioactivity or plain-old failed hardware.
64  *
65  *  Note, however, that one of the leading causes of EEH slot
66  *  freeze events are buggy device drivers, buggy device microcode,
67  *  or buggy device hardware.  This is because any attempt by the
68  *  device to bus-master data to a memory address that is not
69  *  assigned to the device will trigger a slot freeze.   (The idea
70  *  is to prevent devices-gone-wild from corrupting system memory).
71  *  Buggy hardware/drivers will have a miserable time co-existing
72  *  with EEH.
73  *
74  *  Ideally, a PCI device driver, when suspecting that an isolation
75  *  event has occurred (e.g. by reading 0xff's), will then ask EEH
76  *  whether this is the case, and then take appropriate steps to
77  *  reset the PCI slot, the PCI device, and then resume operations.
78  *  However, until that day,  the checking is done here, with the
79  *  eeh_check_failure() routine embedded in the MMIO macros.  If
80  *  the slot is found to be isolated, an "EEH Event" is synthesized
81  *  and sent out for processing.
82  */
83 
84 /* If a device driver keeps reading an MMIO register in an interrupt
85  * handler after a slot isolation event, it might be broken.
86  * This sets the threshold for how many read attempts we allow
87  * before printing an error message.
88  */
89 #define EEH_MAX_FAILS	2100000
90 
91 /* Time to wait for a PCI slot to report status, in milliseconds */
92 #define PCI_BUS_RESET_WAIT_MSEC (5*60*1000)
93 
94 /*
95  * EEH probe mode support, which is part of the flags,
96  * is to support multiple platforms for EEH. Some platforms
97  * like pSeries do PCI emunation based on device tree.
98  * However, other platforms like powernv probe PCI devices
99  * from hardware. The flag is used to distinguish that.
100  * In addition, struct eeh_ops::probe would be invoked for
101  * particular OF node or PCI device so that the corresponding
102  * PE would be created there.
103  */
104 int eeh_subsystem_flags;
105 EXPORT_SYMBOL(eeh_subsystem_flags);
106 
107 /*
108  * EEH allowed maximal frozen times. If one particular PE's
109  * frozen count in last hour exceeds this limit, the PE will
110  * be forced to be offline permanently.
111  */
112 int eeh_max_freezes = 5;
113 
114 /* Platform dependent EEH operations */
115 struct eeh_ops *eeh_ops = NULL;
116 
117 /* Lock to avoid races due to multiple reports of an error */
118 DEFINE_RAW_SPINLOCK(confirm_error_lock);
119 
120 /* Lock to protect passed flags */
121 static DEFINE_MUTEX(eeh_dev_mutex);
122 
123 /* Buffer for reporting pci register dumps. Its here in BSS, and
124  * not dynamically alloced, so that it ends up in RMO where RTAS
125  * can access it.
126  */
127 #define EEH_PCI_REGS_LOG_LEN 8192
128 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
129 
130 /*
131  * The struct is used to maintain the EEH global statistic
132  * information. Besides, the EEH global statistics will be
133  * exported to user space through procfs
134  */
135 struct eeh_stats {
136 	u64 no_device;		/* PCI device not found		*/
137 	u64 no_dn;		/* OF node not found		*/
138 	u64 no_cfg_addr;	/* Config address not found	*/
139 	u64 ignored_check;	/* EEH check skipped		*/
140 	u64 total_mmio_ffs;	/* Total EEH checks		*/
141 	u64 false_positives;	/* Unnecessary EEH checks	*/
142 	u64 slot_resets;	/* PE reset			*/
143 };
144 
145 static struct eeh_stats eeh_stats;
146 
eeh_setup(char * str)147 static int __init eeh_setup(char *str)
148 {
149 	if (!strcmp(str, "off"))
150 		eeh_add_flag(EEH_FORCE_DISABLED);
151 	else if (!strcmp(str, "early_log"))
152 		eeh_add_flag(EEH_EARLY_DUMP_LOG);
153 
154 	return 1;
155 }
156 __setup("eeh=", eeh_setup);
157 
158 /*
159  * This routine captures assorted PCI configuration space data
160  * for the indicated PCI device, and puts them into a buffer
161  * for RTAS error logging.
162  */
eeh_dump_dev_log(struct eeh_dev * edev,char * buf,size_t len)163 static size_t eeh_dump_dev_log(struct eeh_dev *edev, char *buf, size_t len)
164 {
165 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
166 	u32 cfg;
167 	int cap, i;
168 	int n = 0, l = 0;
169 	char buffer[128];
170 
171 	n += scnprintf(buf+n, len-n, "%04x:%02x:%02x:%01x\n",
172 		       edev->phb->global_number, pdn->busno,
173 		       PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
174 	pr_warn("EEH: of node=%04x:%02x:%02x:%01x\n",
175 		edev->phb->global_number, pdn->busno,
176 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
177 
178 	eeh_ops->read_config(pdn, PCI_VENDOR_ID, 4, &cfg);
179 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
180 	pr_warn("EEH: PCI device/vendor: %08x\n", cfg);
181 
182 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cfg);
183 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
184 	pr_warn("EEH: PCI cmd/status register: %08x\n", cfg);
185 
186 	/* Gather bridge-specific registers */
187 	if (edev->mode & EEH_DEV_BRIDGE) {
188 		eeh_ops->read_config(pdn, PCI_SEC_STATUS, 2, &cfg);
189 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
190 		pr_warn("EEH: Bridge secondary status: %04x\n", cfg);
191 
192 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &cfg);
193 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
194 		pr_warn("EEH: Bridge control: %04x\n", cfg);
195 	}
196 
197 	/* Dump out the PCI-X command and status regs */
198 	cap = edev->pcix_cap;
199 	if (cap) {
200 		eeh_ops->read_config(pdn, cap, 4, &cfg);
201 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
202 		pr_warn("EEH: PCI-X cmd: %08x\n", cfg);
203 
204 		eeh_ops->read_config(pdn, cap+4, 4, &cfg);
205 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
206 		pr_warn("EEH: PCI-X status: %08x\n", cfg);
207 	}
208 
209 	/* If PCI-E capable, dump PCI-E cap 10 */
210 	cap = edev->pcie_cap;
211 	if (cap) {
212 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
213 		pr_warn("EEH: PCI-E capabilities and status follow:\n");
214 
215 		for (i=0; i<=8; i++) {
216 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
217 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
218 
219 			if ((i % 4) == 0) {
220 				if (i != 0)
221 					pr_warn("%s\n", buffer);
222 
223 				l = scnprintf(buffer, sizeof(buffer),
224 					      "EEH: PCI-E %02x: %08x ",
225 					      4*i, cfg);
226 			} else {
227 				l += scnprintf(buffer+l, sizeof(buffer)-l,
228 					       "%08x ", cfg);
229 			}
230 
231 		}
232 
233 		pr_warn("%s\n", buffer);
234 	}
235 
236 	/* If AER capable, dump it */
237 	cap = edev->aer_cap;
238 	if (cap) {
239 		n += scnprintf(buf+n, len-n, "pci-e AER:\n");
240 		pr_warn("EEH: PCI-E AER capability register set follows:\n");
241 
242 		for (i=0; i<=13; i++) {
243 			eeh_ops->read_config(pdn, cap+4*i, 4, &cfg);
244 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
245 
246 			if ((i % 4) == 0) {
247 				if (i != 0)
248 					pr_warn("%s\n", buffer);
249 
250 				l = scnprintf(buffer, sizeof(buffer),
251 					      "EEH: PCI-E AER %02x: %08x ",
252 					      4*i, cfg);
253 			} else {
254 				l += scnprintf(buffer+l, sizeof(buffer)-l,
255 					       "%08x ", cfg);
256 			}
257 		}
258 
259 		pr_warn("%s\n", buffer);
260 	}
261 
262 	return n;
263 }
264 
eeh_dump_pe_log(void * data,void * flag)265 static void *eeh_dump_pe_log(void *data, void *flag)
266 {
267 	struct eeh_pe *pe = data;
268 	struct eeh_dev *edev, *tmp;
269 	size_t *plen = flag;
270 
271 	/* If the PE's config space is blocked, 0xFF's will be
272 	 * returned. It's pointless to collect the log in this
273 	 * case.
274 	 */
275 	if (pe->state & EEH_PE_CFG_BLOCKED)
276 		return NULL;
277 
278 	eeh_pe_for_each_dev(pe, edev, tmp)
279 		*plen += eeh_dump_dev_log(edev, pci_regs_buf + *plen,
280 					  EEH_PCI_REGS_LOG_LEN - *plen);
281 
282 	return NULL;
283 }
284 
285 /**
286  * eeh_slot_error_detail - Generate combined log including driver log and error log
287  * @pe: EEH PE
288  * @severity: temporary or permanent error log
289  *
290  * This routine should be called to generate the combined log, which
291  * is comprised of driver log and error log. The driver log is figured
292  * out from the config space of the corresponding PCI device, while
293  * the error log is fetched through platform dependent function call.
294  */
eeh_slot_error_detail(struct eeh_pe * pe,int severity)295 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
296 {
297 	size_t loglen = 0;
298 
299 	/*
300 	 * When the PHB is fenced or dead, it's pointless to collect
301 	 * the data from PCI config space because it should return
302 	 * 0xFF's. For ER, we still retrieve the data from the PCI
303 	 * config space.
304 	 *
305 	 * For pHyp, we have to enable IO for log retrieval. Otherwise,
306 	 * 0xFF's is always returned from PCI config space.
307 	 *
308 	 * When the @severity is EEH_LOG_PERM, the PE is going to be
309 	 * removed. Prior to that, the drivers for devices included in
310 	 * the PE will be closed. The drivers rely on working IO path
311 	 * to bring the devices to quiet state. Otherwise, PCI traffic
312 	 * from those devices after they are removed is like to cause
313 	 * another unexpected EEH error.
314 	 */
315 	if (!(pe->type & EEH_PE_PHB)) {
316 		if (eeh_has_flag(EEH_ENABLE_IO_FOR_LOG) ||
317 		    severity == EEH_LOG_PERM)
318 			eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
319 
320 		/*
321 		 * The config space of some PCI devices can't be accessed
322 		 * when their PEs are in frozen state. Otherwise, fenced
323 		 * PHB might be seen. Those PEs are identified with flag
324 		 * EEH_PE_CFG_RESTRICTED, indicating EEH_PE_CFG_BLOCKED
325 		 * is set automatically when the PE is put to EEH_PE_ISOLATED.
326 		 *
327 		 * Restoring BARs possibly triggers PCI config access in
328 		 * (OPAL) firmware and then causes fenced PHB. If the
329 		 * PCI config is blocked with flag EEH_PE_CFG_BLOCKED, it's
330 		 * pointless to restore BARs and dump config space.
331 		 */
332 		eeh_ops->configure_bridge(pe);
333 		if (!(pe->state & EEH_PE_CFG_BLOCKED)) {
334 			eeh_pe_restore_bars(pe);
335 
336 			pci_regs_buf[0] = 0;
337 			eeh_pe_traverse(pe, eeh_dump_pe_log, &loglen);
338 		}
339 	}
340 
341 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
342 }
343 
344 /**
345  * eeh_token_to_phys - Convert EEH address token to phys address
346  * @token: I/O token, should be address in the form 0xA....
347  *
348  * This routine should be called to convert virtual I/O address
349  * to physical one.
350  */
eeh_token_to_phys(unsigned long token)351 static inline unsigned long eeh_token_to_phys(unsigned long token)
352 {
353 	pte_t *ptep;
354 	unsigned long pa;
355 	int hugepage_shift;
356 
357 	/*
358 	 * We won't find hugepages here(this is iomem). Hence we are not
359 	 * worried about _PAGE_SPLITTING/collapse. Also we will not hit
360 	 * page table free, because of init_mm.
361 	 */
362 	ptep = __find_linux_pte_or_hugepte(init_mm.pgd, token,
363 					   NULL, &hugepage_shift);
364 	if (!ptep)
365 		return token;
366 
367 	pa = pte_pfn(*ptep);
368 
369 	/* On radix we can do hugepage mappings for io, so handle that */
370 	if (!hugepage_shift)
371 		hugepage_shift = PAGE_SHIFT;
372 
373 	pa <<= PAGE_SHIFT;
374 	pa |= token & ((1ul << hugepage_shift) - 1);
375 	return pa;
376 }
377 
378 /*
379  * On PowerNV platform, we might already have fenced PHB there.
380  * For that case, it's meaningless to recover frozen PE. Intead,
381  * We have to handle fenced PHB firstly.
382  */
eeh_phb_check_failure(struct eeh_pe * pe)383 static int eeh_phb_check_failure(struct eeh_pe *pe)
384 {
385 	struct eeh_pe *phb_pe;
386 	unsigned long flags;
387 	int ret;
388 
389 	if (!eeh_has_flag(EEH_PROBE_MODE_DEV))
390 		return -EPERM;
391 
392 	/* Find the PHB PE */
393 	phb_pe = eeh_phb_pe_get(pe->phb);
394 	if (!phb_pe) {
395 		pr_warn("%s Can't find PE for PHB#%d\n",
396 			__func__, pe->phb->global_number);
397 		return -EEXIST;
398 	}
399 
400 	/* If the PHB has been in problematic state */
401 	eeh_serialize_lock(&flags);
402 	if (phb_pe->state & EEH_PE_ISOLATED) {
403 		ret = 0;
404 		goto out;
405 	}
406 
407 	/* Check PHB state */
408 	ret = eeh_ops->get_state(phb_pe, NULL);
409 	if ((ret < 0) ||
410 	    (ret == EEH_STATE_NOT_SUPPORT) ||
411 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
412 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
413 		ret = 0;
414 		goto out;
415 	}
416 
417 	/* Isolate the PHB and send event */
418 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
419 	eeh_serialize_unlock(flags);
420 
421 	pr_err("EEH: PHB#%x failure detected, location: %s\n",
422 		phb_pe->phb->global_number, eeh_pe_loc_get(phb_pe));
423 	dump_stack();
424 	eeh_send_failure_event(phb_pe);
425 
426 	return 1;
427 out:
428 	eeh_serialize_unlock(flags);
429 	return ret;
430 }
431 
432 /**
433  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
434  * @edev: eeh device
435  *
436  * Check for an EEH failure for the given device node.  Call this
437  * routine if the result of a read was all 0xff's and you want to
438  * find out if this is due to an EEH slot freeze.  This routine
439  * will query firmware for the EEH status.
440  *
441  * Returns 0 if there has not been an EEH error; otherwise returns
442  * a non-zero value and queues up a slot isolation event notification.
443  *
444  * It is safe to call this routine in an interrupt context.
445  */
eeh_dev_check_failure(struct eeh_dev * edev)446 int eeh_dev_check_failure(struct eeh_dev *edev)
447 {
448 	int ret;
449 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
450 	unsigned long flags;
451 	struct pci_dn *pdn;
452 	struct pci_dev *dev;
453 	struct eeh_pe *pe, *parent_pe, *phb_pe;
454 	int rc = 0;
455 	const char *location = NULL;
456 
457 	eeh_stats.total_mmio_ffs++;
458 
459 	if (!eeh_enabled())
460 		return 0;
461 
462 	if (!edev) {
463 		eeh_stats.no_dn++;
464 		return 0;
465 	}
466 	dev = eeh_dev_to_pci_dev(edev);
467 	pe = eeh_dev_to_pe(edev);
468 
469 	/* Access to IO BARs might get this far and still not want checking. */
470 	if (!pe) {
471 		eeh_stats.ignored_check++;
472 		pr_debug("EEH: Ignored check for %s\n",
473 			eeh_pci_name(dev));
474 		return 0;
475 	}
476 
477 	if (!pe->addr && !pe->config_addr) {
478 		eeh_stats.no_cfg_addr++;
479 		return 0;
480 	}
481 
482 	/*
483 	 * On PowerNV platform, we might already have fenced PHB
484 	 * there and we need take care of that firstly.
485 	 */
486 	ret = eeh_phb_check_failure(pe);
487 	if (ret > 0)
488 		return ret;
489 
490 	/*
491 	 * If the PE isn't owned by us, we shouldn't check the
492 	 * state. Instead, let the owner handle it if the PE has
493 	 * been frozen.
494 	 */
495 	if (eeh_pe_passed(pe))
496 		return 0;
497 
498 	/* If we already have a pending isolation event for this
499 	 * slot, we know it's bad already, we don't need to check.
500 	 * Do this checking under a lock; as multiple PCI devices
501 	 * in one slot might report errors simultaneously, and we
502 	 * only want one error recovery routine running.
503 	 */
504 	eeh_serialize_lock(&flags);
505 	rc = 1;
506 	if (pe->state & EEH_PE_ISOLATED) {
507 		pe->check_count++;
508 		if (pe->check_count % EEH_MAX_FAILS == 0) {
509 			pdn = eeh_dev_to_pdn(edev);
510 			if (pdn->node)
511 				location = of_get_property(pdn->node, "ibm,loc-code", NULL);
512 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
513 				"location=%s driver=%s pci addr=%s\n",
514 				pe->check_count,
515 				location ? location : "unknown",
516 				eeh_driver_name(dev), eeh_pci_name(dev));
517 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
518 				eeh_driver_name(dev));
519 			dump_stack();
520 		}
521 		goto dn_unlock;
522 	}
523 
524 	/*
525 	 * Now test for an EEH failure.  This is VERY expensive.
526 	 * Note that the eeh_config_addr may be a parent device
527 	 * in the case of a device behind a bridge, or it may be
528 	 * function zero of a multi-function device.
529 	 * In any case they must share a common PHB.
530 	 */
531 	ret = eeh_ops->get_state(pe, NULL);
532 
533 	/* Note that config-io to empty slots may fail;
534 	 * they are empty when they don't have children.
535 	 * We will punt with the following conditions: Failure to get
536 	 * PE's state, EEH not support and Permanently unavailable
537 	 * state, PE is in good state.
538 	 */
539 	if ((ret < 0) ||
540 	    (ret == EEH_STATE_NOT_SUPPORT) ||
541 	    ((ret & active_flags) == active_flags)) {
542 		eeh_stats.false_positives++;
543 		pe->false_positives++;
544 		rc = 0;
545 		goto dn_unlock;
546 	}
547 
548 	/*
549 	 * It should be corner case that the parent PE has been
550 	 * put into frozen state as well. We should take care
551 	 * that at first.
552 	 */
553 	parent_pe = pe->parent;
554 	while (parent_pe) {
555 		/* Hit the ceiling ? */
556 		if (parent_pe->type & EEH_PE_PHB)
557 			break;
558 
559 		/* Frozen parent PE ? */
560 		ret = eeh_ops->get_state(parent_pe, NULL);
561 		if (ret > 0 &&
562 		    (ret & active_flags) != active_flags)
563 			pe = parent_pe;
564 
565 		/* Next parent level */
566 		parent_pe = parent_pe->parent;
567 	}
568 
569 	eeh_stats.slot_resets++;
570 
571 	/* Avoid repeated reports of this failure, including problems
572 	 * with other functions on this device, and functions under
573 	 * bridges.
574 	 */
575 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
576 	eeh_serialize_unlock(flags);
577 
578 	/* Most EEH events are due to device driver bugs.  Having
579 	 * a stack trace will help the device-driver authors figure
580 	 * out what happened.  So print that out.
581 	 */
582 	phb_pe = eeh_phb_pe_get(pe->phb);
583 	pr_err("EEH: Frozen PHB#%x-PE#%x detected\n",
584 	       pe->phb->global_number, pe->addr);
585 	pr_err("EEH: PE location: %s, PHB location: %s\n",
586 	       eeh_pe_loc_get(pe), eeh_pe_loc_get(phb_pe));
587 	dump_stack();
588 
589 	eeh_send_failure_event(pe);
590 
591 	return 1;
592 
593 dn_unlock:
594 	eeh_serialize_unlock(flags);
595 	return rc;
596 }
597 
598 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
599 
600 /**
601  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
602  * @token: I/O address
603  *
604  * Check for an EEH failure at the given I/O address. Call this
605  * routine if the result of a read was all 0xff's and you want to
606  * find out if this is due to an EEH slot freeze event. This routine
607  * will query firmware for the EEH status.
608  *
609  * Note this routine is safe to call in an interrupt context.
610  */
eeh_check_failure(const volatile void __iomem * token)611 int eeh_check_failure(const volatile void __iomem *token)
612 {
613 	unsigned long addr;
614 	struct eeh_dev *edev;
615 
616 	/* Finding the phys addr + pci device; this is pretty quick. */
617 	addr = eeh_token_to_phys((unsigned long __force) token);
618 	edev = eeh_addr_cache_get_dev(addr);
619 	if (!edev) {
620 		eeh_stats.no_device++;
621 		return 0;
622 	}
623 
624 	return eeh_dev_check_failure(edev);
625 }
626 EXPORT_SYMBOL(eeh_check_failure);
627 
628 
629 /**
630  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
631  * @pe: EEH PE
632  *
633  * This routine should be called to reenable frozen MMIO or DMA
634  * so that it would work correctly again. It's useful while doing
635  * recovery or log collection on the indicated device.
636  */
eeh_pci_enable(struct eeh_pe * pe,int function)637 int eeh_pci_enable(struct eeh_pe *pe, int function)
638 {
639 	int active_flag, rc;
640 
641 	/*
642 	 * pHyp doesn't allow to enable IO or DMA on unfrozen PE.
643 	 * Also, it's pointless to enable them on unfrozen PE. So
644 	 * we have to check before enabling IO or DMA.
645 	 */
646 	switch (function) {
647 	case EEH_OPT_THAW_MMIO:
648 		active_flag = EEH_STATE_MMIO_ACTIVE | EEH_STATE_MMIO_ENABLED;
649 		break;
650 	case EEH_OPT_THAW_DMA:
651 		active_flag = EEH_STATE_DMA_ACTIVE;
652 		break;
653 	case EEH_OPT_DISABLE:
654 	case EEH_OPT_ENABLE:
655 	case EEH_OPT_FREEZE_PE:
656 		active_flag = 0;
657 		break;
658 	default:
659 		pr_warn("%s: Invalid function %d\n",
660 			__func__, function);
661 		return -EINVAL;
662 	}
663 
664 	/*
665 	 * Check if IO or DMA has been enabled before
666 	 * enabling them.
667 	 */
668 	if (active_flag) {
669 		rc = eeh_ops->get_state(pe, NULL);
670 		if (rc < 0)
671 			return rc;
672 
673 		/* Needn't enable it at all */
674 		if (rc == EEH_STATE_NOT_SUPPORT)
675 			return 0;
676 
677 		/* It's already enabled */
678 		if (rc & active_flag)
679 			return 0;
680 	}
681 
682 
683 	/* Issue the request */
684 	rc = eeh_ops->set_option(pe, function);
685 	if (rc)
686 		pr_warn("%s: Unexpected state change %d on "
687 			"PHB#%d-PE#%x, err=%d\n",
688 			__func__, function, pe->phb->global_number,
689 			pe->addr, rc);
690 
691 	/* Check if the request is finished successfully */
692 	if (active_flag) {
693 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
694 		if (rc < 0)
695 			return rc;
696 
697 		if (rc & active_flag)
698 			return 0;
699 
700 		return -EIO;
701 	}
702 
703 	return rc;
704 }
705 
eeh_disable_and_save_dev_state(void * data,void * userdata)706 static void *eeh_disable_and_save_dev_state(void *data, void *userdata)
707 {
708 	struct eeh_dev *edev = data;
709 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
710 	struct pci_dev *dev = userdata;
711 
712 	/*
713 	 * The caller should have disabled and saved the
714 	 * state for the specified device
715 	 */
716 	if (!pdev || pdev == dev)
717 		return NULL;
718 
719 	/* Ensure we have D0 power state */
720 	pci_set_power_state(pdev, PCI_D0);
721 
722 	/* Save device state */
723 	pci_save_state(pdev);
724 
725 	/*
726 	 * Disable device to avoid any DMA traffic and
727 	 * interrupt from the device
728 	 */
729 	pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
730 
731 	return NULL;
732 }
733 
eeh_restore_dev_state(void * data,void * userdata)734 static void *eeh_restore_dev_state(void *data, void *userdata)
735 {
736 	struct eeh_dev *edev = data;
737 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
738 	struct pci_dev *pdev = eeh_dev_to_pci_dev(edev);
739 	struct pci_dev *dev = userdata;
740 
741 	if (!pdev)
742 		return NULL;
743 
744 	/* Apply customization from firmware */
745 	if (pdn && eeh_ops->restore_config)
746 		eeh_ops->restore_config(pdn);
747 
748 	/* The caller should restore state for the specified device */
749 	if (pdev != dev)
750 		pci_restore_state(pdev);
751 
752 	return NULL;
753 }
754 
755 /**
756  * pcibios_set_pcie_slot_reset - Set PCI-E reset state
757  * @dev: pci device struct
758  * @state: reset state to enter
759  *
760  * Return value:
761  * 	0 if success
762  */
pcibios_set_pcie_reset_state(struct pci_dev * dev,enum pcie_reset_state state)763 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
764 {
765 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
766 	struct eeh_pe *pe = eeh_dev_to_pe(edev);
767 
768 	if (!pe) {
769 		pr_err("%s: No PE found on PCI device %s\n",
770 			__func__, pci_name(dev));
771 		return -EINVAL;
772 	}
773 
774 	switch (state) {
775 	case pcie_deassert_reset:
776 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
777 		eeh_unfreeze_pe(pe, false);
778 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
779 		eeh_pe_dev_traverse(pe, eeh_restore_dev_state, dev);
780 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
781 		break;
782 	case pcie_hot_reset:
783 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
784 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
785 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
786 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
787 		eeh_ops->reset(pe, EEH_RESET_HOT);
788 		break;
789 	case pcie_warm_reset:
790 		eeh_pe_state_mark_with_cfg(pe, EEH_PE_ISOLATED);
791 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
792 		eeh_pe_dev_traverse(pe, eeh_disable_and_save_dev_state, dev);
793 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
794 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
795 		break;
796 	default:
797 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED | EEH_PE_CFG_BLOCKED);
798 		return -EINVAL;
799 	};
800 
801 	return 0;
802 }
803 
804 /**
805  * eeh_set_pe_freset - Check the required reset for the indicated device
806  * @data: EEH device
807  * @flag: return value
808  *
809  * Each device might have its preferred reset type: fundamental or
810  * hot reset. The routine is used to collected the information for
811  * the indicated device and its children so that the bunch of the
812  * devices could be reset properly.
813  */
eeh_set_dev_freset(void * data,void * flag)814 static void *eeh_set_dev_freset(void *data, void *flag)
815 {
816 	struct pci_dev *dev;
817 	unsigned int *freset = (unsigned int *)flag;
818 	struct eeh_dev *edev = (struct eeh_dev *)data;
819 
820 	dev = eeh_dev_to_pci_dev(edev);
821 	if (dev)
822 		*freset |= dev->needs_freset;
823 
824 	return NULL;
825 }
826 
827 /**
828  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
829  * @pe: EEH PE
830  *
831  * Assert the PCI #RST line for 1/4 second.
832  */
eeh_reset_pe_once(struct eeh_pe * pe)833 static void eeh_reset_pe_once(struct eeh_pe *pe)
834 {
835 	unsigned int freset = 0;
836 
837 	/* Determine type of EEH reset required for
838 	 * Partitionable Endpoint, a hot-reset (1)
839 	 * or a fundamental reset (3).
840 	 * A fundamental reset required by any device under
841 	 * Partitionable Endpoint trumps hot-reset.
842 	 */
843 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
844 
845 	if (freset)
846 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
847 	else
848 		eeh_ops->reset(pe, EEH_RESET_HOT);
849 
850 	eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
851 }
852 
853 /**
854  * eeh_reset_pe - Reset the indicated PE
855  * @pe: EEH PE
856  *
857  * This routine should be called to reset indicated device, including
858  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
859  * might be involved as well.
860  */
eeh_reset_pe(struct eeh_pe * pe)861 int eeh_reset_pe(struct eeh_pe *pe)
862 {
863 	int flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
864 	int i, state, ret;
865 
866 	/* Mark as reset and block config space */
867 	eeh_pe_state_mark(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
868 
869 	/* Take three shots at resetting the bus */
870 	for (i = 0; i < 3; i++) {
871 		eeh_reset_pe_once(pe);
872 
873 		/*
874 		 * EEH_PE_ISOLATED is expected to be removed after
875 		 * BAR restore.
876 		 */
877 		state = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
878 		if ((state & flags) == flags) {
879 			ret = 0;
880 			goto out;
881 		}
882 
883 		if (state < 0) {
884 			pr_warn("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
885 				__func__, pe->phb->global_number, pe->addr);
886 			ret = -ENOTRECOVERABLE;
887 			goto out;
888 		}
889 
890 		/* We might run out of credits */
891 		ret = -EIO;
892 		pr_warn("%s: Failure %d resetting PHB#%x-PE#%x\n (%d)\n",
893 			__func__, state, pe->phb->global_number, pe->addr, (i + 1));
894 	}
895 
896 out:
897 	eeh_pe_state_clear(pe, EEH_PE_RESET | EEH_PE_CFG_BLOCKED);
898 	return ret;
899 }
900 
901 /**
902  * eeh_save_bars - Save device bars
903  * @edev: PCI device associated EEH device
904  *
905  * Save the values of the device bars. Unlike the restore
906  * routine, this routine is *not* recursive. This is because
907  * PCI devices are added individually; but, for the restore,
908  * an entire slot is reset at a time.
909  */
eeh_save_bars(struct eeh_dev * edev)910 void eeh_save_bars(struct eeh_dev *edev)
911 {
912 	struct pci_dn *pdn;
913 	int i;
914 
915 	pdn = eeh_dev_to_pdn(edev);
916 	if (!pdn)
917 		return;
918 
919 	for (i = 0; i < 16; i++)
920 		eeh_ops->read_config(pdn, i * 4, 4, &edev->config_space[i]);
921 
922 	/*
923 	 * For PCI bridges including root port, we need enable bus
924 	 * master explicitly. Otherwise, it can't fetch IODA table
925 	 * entries correctly. So we cache the bit in advance so that
926 	 * we can restore it after reset, either PHB range or PE range.
927 	 */
928 	if (edev->mode & EEH_DEV_BRIDGE)
929 		edev->config_space[1] |= PCI_COMMAND_MASTER;
930 }
931 
932 /**
933  * eeh_ops_register - Register platform dependent EEH operations
934  * @ops: platform dependent EEH operations
935  *
936  * Register the platform dependent EEH operation callback
937  * functions. The platform should call this function before
938  * any other EEH operations.
939  */
eeh_ops_register(struct eeh_ops * ops)940 int __init eeh_ops_register(struct eeh_ops *ops)
941 {
942 	if (!ops->name) {
943 		pr_warn("%s: Invalid EEH ops name for %p\n",
944 			__func__, ops);
945 		return -EINVAL;
946 	}
947 
948 	if (eeh_ops && eeh_ops != ops) {
949 		pr_warn("%s: EEH ops of platform %s already existing (%s)\n",
950 			__func__, eeh_ops->name, ops->name);
951 		return -EEXIST;
952 	}
953 
954 	eeh_ops = ops;
955 
956 	return 0;
957 }
958 
959 /**
960  * eeh_ops_unregister - Unreigster platform dependent EEH operations
961  * @name: name of EEH platform operations
962  *
963  * Unregister the platform dependent EEH operation callback
964  * functions.
965  */
eeh_ops_unregister(const char * name)966 int __exit eeh_ops_unregister(const char *name)
967 {
968 	if (!name || !strlen(name)) {
969 		pr_warn("%s: Invalid EEH ops name\n",
970 			__func__);
971 		return -EINVAL;
972 	}
973 
974 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
975 		eeh_ops = NULL;
976 		return 0;
977 	}
978 
979 	return -EEXIST;
980 }
981 
eeh_reboot_notifier(struct notifier_block * nb,unsigned long action,void * unused)982 static int eeh_reboot_notifier(struct notifier_block *nb,
983 			       unsigned long action, void *unused)
984 {
985 	eeh_clear_flag(EEH_ENABLED);
986 	return NOTIFY_DONE;
987 }
988 
989 static struct notifier_block eeh_reboot_nb = {
990 	.notifier_call = eeh_reboot_notifier,
991 };
992 
993 /**
994  * eeh_init - EEH initialization
995  *
996  * Initialize EEH by trying to enable it for all of the adapters in the system.
997  * As a side effect we can determine here if eeh is supported at all.
998  * Note that we leave EEH on so failed config cycles won't cause a machine
999  * check.  If a user turns off EEH for a particular adapter they are really
1000  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
1001  * grant access to a slot if EEH isn't enabled, and so we always enable
1002  * EEH for all slots/all devices.
1003  *
1004  * The eeh-force-off option disables EEH checking globally, for all slots.
1005  * Even if force-off is set, the EEH hardware is still enabled, so that
1006  * newer systems can boot.
1007  */
eeh_init(void)1008 int eeh_init(void)
1009 {
1010 	struct pci_controller *hose, *tmp;
1011 	struct pci_dn *pdn;
1012 	static int cnt = 0;
1013 	int ret = 0;
1014 
1015 	/*
1016 	 * We have to delay the initialization on PowerNV after
1017 	 * the PCI hierarchy tree has been built because the PEs
1018 	 * are figured out based on PCI devices instead of device
1019 	 * tree nodes
1020 	 */
1021 	if (machine_is(powernv) && cnt++ <= 0)
1022 		return ret;
1023 
1024 	/* Register reboot notifier */
1025 	ret = register_reboot_notifier(&eeh_reboot_nb);
1026 	if (ret) {
1027 		pr_warn("%s: Failed to register notifier (%d)\n",
1028 			__func__, ret);
1029 		return ret;
1030 	}
1031 
1032 	/* call platform initialization function */
1033 	if (!eeh_ops) {
1034 		pr_warn("%s: Platform EEH operation not found\n",
1035 			__func__);
1036 		return -EEXIST;
1037 	} else if ((ret = eeh_ops->init()))
1038 		return ret;
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_warn("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;
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 	phb = edev->phb;
1097 	if (NULL == phb ||
1098 	    (eeh_has_flag(EEH_PROBE_MODE_DEVTREE) && 0 == phb->buid))
1099 		return;
1100 
1101 	eeh_ops->probe(pdn, NULL);
1102 }
1103 
1104 /**
1105  * eeh_add_device_tree_early - Enable EEH for the indicated device
1106  * @pdn: PCI device node
1107  *
1108  * This routine must be used to perform EEH initialization for the
1109  * indicated PCI device that was added after system boot (e.g.
1110  * hotplug, dlpar).
1111  */
eeh_add_device_tree_early(struct pci_dn * pdn)1112 void eeh_add_device_tree_early(struct pci_dn *pdn)
1113 {
1114 	struct pci_dn *n;
1115 
1116 	if (!pdn)
1117 		return;
1118 
1119 	list_for_each_entry(n, &pdn->child_list, list)
1120 		eeh_add_device_tree_early(n);
1121 	eeh_add_device_early(pdn);
1122 }
1123 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
1124 
1125 /**
1126  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
1127  * @dev: pci device for which to set up EEH
1128  *
1129  * This routine must be used to complete EEH initialization for PCI
1130  * devices that were added after system boot (e.g. hotplug, dlpar).
1131  */
eeh_add_device_late(struct pci_dev * dev)1132 void eeh_add_device_late(struct pci_dev *dev)
1133 {
1134 	struct pci_dn *pdn;
1135 	struct eeh_dev *edev;
1136 
1137 	if (!dev || !eeh_enabled())
1138 		return;
1139 
1140 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
1141 
1142 	pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
1143 	edev = pdn_to_eeh_dev(pdn);
1144 	if (edev->pdev == dev) {
1145 		pr_debug("EEH: Already referenced !\n");
1146 		return;
1147 	}
1148 
1149 	/*
1150 	 * The EEH cache might not be removed correctly because of
1151 	 * unbalanced kref to the device during unplug time, which
1152 	 * relies on pcibios_release_device(). So we have to remove
1153 	 * that here explicitly.
1154 	 */
1155 	if (edev->pdev) {
1156 		eeh_rmv_from_parent_pe(edev);
1157 		eeh_addr_cache_rmv_dev(edev->pdev);
1158 		eeh_sysfs_remove_device(edev->pdev);
1159 		edev->mode &= ~EEH_DEV_SYSFS;
1160 
1161 		/*
1162 		 * We definitely should have the PCI device removed
1163 		 * though it wasn't correctly. So we needn't call
1164 		 * into error handler afterwards.
1165 		 */
1166 		edev->mode |= EEH_DEV_NO_HANDLER;
1167 
1168 		edev->pdev = NULL;
1169 		dev->dev.archdata.edev = NULL;
1170 	}
1171 
1172 	if (eeh_has_flag(EEH_PROBE_MODE_DEV))
1173 		eeh_ops->probe(pdn, NULL);
1174 
1175 	edev->pdev = dev;
1176 	dev->dev.archdata.edev = edev;
1177 
1178 	eeh_addr_cache_insert_dev(dev);
1179 }
1180 
1181 /**
1182  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
1183  * @bus: PCI bus
1184  *
1185  * This routine must be used to perform EEH initialization for PCI
1186  * devices which are attached to the indicated PCI bus. The PCI bus
1187  * is added after system boot through hotplug or dlpar.
1188  */
eeh_add_device_tree_late(struct pci_bus * bus)1189 void eeh_add_device_tree_late(struct pci_bus *bus)
1190 {
1191 	struct pci_dev *dev;
1192 
1193 	list_for_each_entry(dev, &bus->devices, bus_list) {
1194 		eeh_add_device_late(dev);
1195 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1196 			struct pci_bus *subbus = dev->subordinate;
1197 			if (subbus)
1198 				eeh_add_device_tree_late(subbus);
1199 		}
1200 	}
1201 }
1202 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
1203 
1204 /**
1205  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
1206  * @bus: PCI bus
1207  *
1208  * This routine must be used to add EEH sysfs files for PCI
1209  * devices which are attached to the indicated PCI bus. The PCI bus
1210  * is added after system boot through hotplug or dlpar.
1211  */
eeh_add_sysfs_files(struct pci_bus * bus)1212 void eeh_add_sysfs_files(struct pci_bus *bus)
1213 {
1214 	struct pci_dev *dev;
1215 
1216 	list_for_each_entry(dev, &bus->devices, bus_list) {
1217 		eeh_sysfs_add_device(dev);
1218 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1219 			struct pci_bus *subbus = dev->subordinate;
1220 			if (subbus)
1221 				eeh_add_sysfs_files(subbus);
1222 		}
1223 	}
1224 }
1225 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
1226 
1227 /**
1228  * eeh_remove_device - Undo EEH setup for the indicated pci device
1229  * @dev: pci device to be removed
1230  *
1231  * This routine should be called when a device is removed from
1232  * a running system (e.g. by hotplug or dlpar).  It unregisters
1233  * the PCI device from the EEH subsystem.  I/O errors affecting
1234  * this device will no longer be detected after this call; thus,
1235  * i/o errors affecting this slot may leave this device unusable.
1236  */
eeh_remove_device(struct pci_dev * dev)1237 void eeh_remove_device(struct pci_dev *dev)
1238 {
1239 	struct eeh_dev *edev;
1240 
1241 	if (!dev || !eeh_enabled())
1242 		return;
1243 	edev = pci_dev_to_eeh_dev(dev);
1244 
1245 	/* Unregister the device with the EEH/PCI address search system */
1246 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1247 
1248 	if (!edev || !edev->pdev || !edev->pe) {
1249 		pr_debug("EEH: Not referenced !\n");
1250 		return;
1251 	}
1252 
1253 	/*
1254 	 * During the hotplug for EEH error recovery, we need the EEH
1255 	 * device attached to the parent PE in order for BAR restore
1256 	 * a bit later. So we keep it for BAR restore and remove it
1257 	 * from the parent PE during the BAR resotre.
1258 	 */
1259 	edev->pdev = NULL;
1260 	dev->dev.archdata.edev = NULL;
1261 	if (!(edev->pe->state & EEH_PE_KEEP))
1262 		eeh_rmv_from_parent_pe(edev);
1263 	else
1264 		edev->mode |= EEH_DEV_DISCONNECTED;
1265 
1266 	/*
1267 	 * We're removing from the PCI subsystem, that means
1268 	 * the PCI device driver can't support EEH or not
1269 	 * well. So we rely on hotplug completely to do recovery
1270 	 * for the specific PCI device.
1271 	 */
1272 	edev->mode |= EEH_DEV_NO_HANDLER;
1273 
1274 	eeh_addr_cache_rmv_dev(dev);
1275 	eeh_sysfs_remove_device(dev);
1276 	edev->mode &= ~EEH_DEV_SYSFS;
1277 }
1278 
eeh_unfreeze_pe(struct eeh_pe * pe,bool sw_state)1279 int eeh_unfreeze_pe(struct eeh_pe *pe, bool sw_state)
1280 {
1281 	int ret;
1282 
1283 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
1284 	if (ret) {
1285 		pr_warn("%s: Failure %d enabling IO on PHB#%x-PE#%x\n",
1286 			__func__, ret, pe->phb->global_number, pe->addr);
1287 		return ret;
1288 	}
1289 
1290 	ret = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
1291 	if (ret) {
1292 		pr_warn("%s: Failure %d enabling DMA on PHB#%x-PE#%x\n",
1293 			__func__, ret, pe->phb->global_number, pe->addr);
1294 		return ret;
1295 	}
1296 
1297 	/* Clear software isolated state */
1298 	if (sw_state && (pe->state & EEH_PE_ISOLATED))
1299 		eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
1300 
1301 	return ret;
1302 }
1303 
1304 
1305 static struct pci_device_id eeh_reset_ids[] = {
1306 	{ PCI_DEVICE(0x19a2, 0x0710) },	/* Emulex, BE     */
1307 	{ PCI_DEVICE(0x10df, 0xe220) },	/* Emulex, Lancer */
1308 	{ PCI_DEVICE(0x14e4, 0x1657) }, /* Broadcom BCM5719 */
1309 	{ 0 }
1310 };
1311 
eeh_pe_change_owner(struct eeh_pe * pe)1312 static int eeh_pe_change_owner(struct eeh_pe *pe)
1313 {
1314 	struct eeh_dev *edev, *tmp;
1315 	struct pci_dev *pdev;
1316 	struct pci_device_id *id;
1317 	int flags, ret;
1318 
1319 	/* Check PE state */
1320 	flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1321 	ret = eeh_ops->get_state(pe, NULL);
1322 	if (ret < 0 || ret == EEH_STATE_NOT_SUPPORT)
1323 		return 0;
1324 
1325 	/* Unfrozen PE, nothing to do */
1326 	if ((ret & flags) == flags)
1327 		return 0;
1328 
1329 	/* Frozen PE, check if it needs PE level reset */
1330 	eeh_pe_for_each_dev(pe, edev, tmp) {
1331 		pdev = eeh_dev_to_pci_dev(edev);
1332 		if (!pdev)
1333 			continue;
1334 
1335 		for (id = &eeh_reset_ids[0]; id->vendor != 0; id++) {
1336 			if (id->vendor != PCI_ANY_ID &&
1337 			    id->vendor != pdev->vendor)
1338 				continue;
1339 			if (id->device != PCI_ANY_ID &&
1340 			    id->device != pdev->device)
1341 				continue;
1342 			if (id->subvendor != PCI_ANY_ID &&
1343 			    id->subvendor != pdev->subsystem_vendor)
1344 				continue;
1345 			if (id->subdevice != PCI_ANY_ID &&
1346 			    id->subdevice != pdev->subsystem_device)
1347 				continue;
1348 
1349 			goto reset;
1350 		}
1351 	}
1352 
1353 	return eeh_unfreeze_pe(pe, true);
1354 
1355 reset:
1356 	return eeh_pe_reset_and_recover(pe);
1357 }
1358 
1359 /**
1360  * eeh_dev_open - Increase count of pass through devices for PE
1361  * @pdev: PCI device
1362  *
1363  * Increase count of passed through devices for the indicated
1364  * PE. In the result, the EEH errors detected on the PE won't be
1365  * reported. The PE owner will be responsible for detection
1366  * and recovery.
1367  */
eeh_dev_open(struct pci_dev * pdev)1368 int eeh_dev_open(struct pci_dev *pdev)
1369 {
1370 	struct eeh_dev *edev;
1371 	int ret = -ENODEV;
1372 
1373 	mutex_lock(&eeh_dev_mutex);
1374 
1375 	/* No PCI device ? */
1376 	if (!pdev)
1377 		goto out;
1378 
1379 	/* No EEH device or PE ? */
1380 	edev = pci_dev_to_eeh_dev(pdev);
1381 	if (!edev || !edev->pe)
1382 		goto out;
1383 
1384 	/*
1385 	 * The PE might have been put into frozen state, but we
1386 	 * didn't detect that yet. The passed through PCI devices
1387 	 * in frozen PE won't work properly. Clear the frozen state
1388 	 * in advance.
1389 	 */
1390 	ret = eeh_pe_change_owner(edev->pe);
1391 	if (ret)
1392 		goto out;
1393 
1394 	/* Increase PE's pass through count */
1395 	atomic_inc(&edev->pe->pass_dev_cnt);
1396 	mutex_unlock(&eeh_dev_mutex);
1397 
1398 	return 0;
1399 out:
1400 	mutex_unlock(&eeh_dev_mutex);
1401 	return ret;
1402 }
1403 EXPORT_SYMBOL_GPL(eeh_dev_open);
1404 
1405 /**
1406  * eeh_dev_release - Decrease count of pass through devices for PE
1407  * @pdev: PCI device
1408  *
1409  * Decrease count of pass through devices for the indicated PE. If
1410  * there is no passed through device in PE, the EEH errors detected
1411  * on the PE will be reported and handled as usual.
1412  */
eeh_dev_release(struct pci_dev * pdev)1413 void eeh_dev_release(struct pci_dev *pdev)
1414 {
1415 	struct eeh_dev *edev;
1416 
1417 	mutex_lock(&eeh_dev_mutex);
1418 
1419 	/* No PCI device ? */
1420 	if (!pdev)
1421 		goto out;
1422 
1423 	/* No EEH device ? */
1424 	edev = pci_dev_to_eeh_dev(pdev);
1425 	if (!edev || !edev->pe || !eeh_pe_passed(edev->pe))
1426 		goto out;
1427 
1428 	/* Decrease PE's pass through count */
1429 	WARN_ON(atomic_dec_if_positive(&edev->pe->pass_dev_cnt) < 0);
1430 	eeh_pe_change_owner(edev->pe);
1431 out:
1432 	mutex_unlock(&eeh_dev_mutex);
1433 }
1434 EXPORT_SYMBOL(eeh_dev_release);
1435 
1436 #ifdef CONFIG_IOMMU_API
1437 
dev_has_iommu_table(struct device * dev,void * data)1438 static int dev_has_iommu_table(struct device *dev, void *data)
1439 {
1440 	struct pci_dev *pdev = to_pci_dev(dev);
1441 	struct pci_dev **ppdev = data;
1442 
1443 	if (!dev)
1444 		return 0;
1445 
1446 	if (dev->iommu_group) {
1447 		*ppdev = pdev;
1448 		return 1;
1449 	}
1450 
1451 	return 0;
1452 }
1453 
1454 /**
1455  * eeh_iommu_group_to_pe - Convert IOMMU group to EEH PE
1456  * @group: IOMMU group
1457  *
1458  * The routine is called to convert IOMMU group to EEH PE.
1459  */
eeh_iommu_group_to_pe(struct iommu_group * group)1460 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group)
1461 {
1462 	struct pci_dev *pdev = NULL;
1463 	struct eeh_dev *edev;
1464 	int ret;
1465 
1466 	/* No IOMMU group ? */
1467 	if (!group)
1468 		return NULL;
1469 
1470 	ret = iommu_group_for_each_dev(group, &pdev, dev_has_iommu_table);
1471 	if (!ret || !pdev)
1472 		return NULL;
1473 
1474 	/* No EEH device or PE ? */
1475 	edev = pci_dev_to_eeh_dev(pdev);
1476 	if (!edev || !edev->pe)
1477 		return NULL;
1478 
1479 	return edev->pe;
1480 }
1481 EXPORT_SYMBOL_GPL(eeh_iommu_group_to_pe);
1482 
1483 #endif /* CONFIG_IOMMU_API */
1484 
1485 /**
1486  * eeh_pe_set_option - Set options for the indicated PE
1487  * @pe: EEH PE
1488  * @option: requested option
1489  *
1490  * The routine is called to enable or disable EEH functionality
1491  * on the indicated PE, to enable IO or DMA for the frozen PE.
1492  */
eeh_pe_set_option(struct eeh_pe * pe,int option)1493 int eeh_pe_set_option(struct eeh_pe *pe, int option)
1494 {
1495 	int ret = 0;
1496 
1497 	/* Invalid PE ? */
1498 	if (!pe)
1499 		return -ENODEV;
1500 
1501 	/*
1502 	 * EEH functionality could possibly be disabled, just
1503 	 * return error for the case. And the EEH functinality
1504 	 * isn't expected to be disabled on one specific PE.
1505 	 */
1506 	switch (option) {
1507 	case EEH_OPT_ENABLE:
1508 		if (eeh_enabled()) {
1509 			ret = eeh_pe_change_owner(pe);
1510 			break;
1511 		}
1512 		ret = -EIO;
1513 		break;
1514 	case EEH_OPT_DISABLE:
1515 		break;
1516 	case EEH_OPT_THAW_MMIO:
1517 	case EEH_OPT_THAW_DMA:
1518 		if (!eeh_ops || !eeh_ops->set_option) {
1519 			ret = -ENOENT;
1520 			break;
1521 		}
1522 
1523 		ret = eeh_pci_enable(pe, option);
1524 		break;
1525 	default:
1526 		pr_debug("%s: Option %d out of range (%d, %d)\n",
1527 			__func__, option, EEH_OPT_DISABLE, EEH_OPT_THAW_DMA);
1528 		ret = -EINVAL;
1529 	}
1530 
1531 	return ret;
1532 }
1533 EXPORT_SYMBOL_GPL(eeh_pe_set_option);
1534 
1535 /**
1536  * eeh_pe_get_state - Retrieve PE's state
1537  * @pe: EEH PE
1538  *
1539  * Retrieve the PE's state, which includes 3 aspects: enabled
1540  * DMA, enabled IO and asserted reset.
1541  */
eeh_pe_get_state(struct eeh_pe * pe)1542 int eeh_pe_get_state(struct eeh_pe *pe)
1543 {
1544 	int result, ret = 0;
1545 	bool rst_active, dma_en, mmio_en;
1546 
1547 	/* Existing PE ? */
1548 	if (!pe)
1549 		return -ENODEV;
1550 
1551 	if (!eeh_ops || !eeh_ops->get_state)
1552 		return -ENOENT;
1553 
1554 	result = eeh_ops->get_state(pe, NULL);
1555 	rst_active = !!(result & EEH_STATE_RESET_ACTIVE);
1556 	dma_en = !!(result & EEH_STATE_DMA_ENABLED);
1557 	mmio_en = !!(result & EEH_STATE_MMIO_ENABLED);
1558 
1559 	if (rst_active)
1560 		ret = EEH_PE_STATE_RESET;
1561 	else if (dma_en && mmio_en)
1562 		ret = EEH_PE_STATE_NORMAL;
1563 	else if (!dma_en && !mmio_en)
1564 		ret = EEH_PE_STATE_STOPPED_IO_DMA;
1565 	else if (!dma_en && mmio_en)
1566 		ret = EEH_PE_STATE_STOPPED_DMA;
1567 	else
1568 		ret = EEH_PE_STATE_UNAVAIL;
1569 
1570 	return ret;
1571 }
1572 EXPORT_SYMBOL_GPL(eeh_pe_get_state);
1573 
eeh_pe_reenable_devices(struct eeh_pe * pe)1574 static int eeh_pe_reenable_devices(struct eeh_pe *pe)
1575 {
1576 	struct eeh_dev *edev, *tmp;
1577 	struct pci_dev *pdev;
1578 	int ret = 0;
1579 
1580 	/* Restore config space */
1581 	eeh_pe_restore_bars(pe);
1582 
1583 	/*
1584 	 * Reenable PCI devices as the devices passed
1585 	 * through are always enabled before the reset.
1586 	 */
1587 	eeh_pe_for_each_dev(pe, edev, tmp) {
1588 		pdev = eeh_dev_to_pci_dev(edev);
1589 		if (!pdev)
1590 			continue;
1591 
1592 		ret = pci_reenable_device(pdev);
1593 		if (ret) {
1594 			pr_warn("%s: Failure %d reenabling %s\n",
1595 				__func__, ret, pci_name(pdev));
1596 			return ret;
1597 		}
1598 	}
1599 
1600 	/* The PE is still in frozen state */
1601 	return eeh_unfreeze_pe(pe, true);
1602 }
1603 
1604 /**
1605  * eeh_pe_reset - Issue PE reset according to specified type
1606  * @pe: EEH PE
1607  * @option: reset type
1608  *
1609  * The routine is called to reset the specified PE with the
1610  * indicated type, either fundamental reset or hot reset.
1611  * PE reset is the most important part for error recovery.
1612  */
eeh_pe_reset(struct eeh_pe * pe,int option)1613 int eeh_pe_reset(struct eeh_pe *pe, int option)
1614 {
1615 	int ret = 0;
1616 
1617 	/* Invalid PE ? */
1618 	if (!pe)
1619 		return -ENODEV;
1620 
1621 	if (!eeh_ops || !eeh_ops->set_option || !eeh_ops->reset)
1622 		return -ENOENT;
1623 
1624 	switch (option) {
1625 	case EEH_RESET_DEACTIVATE:
1626 		ret = eeh_ops->reset(pe, option);
1627 		eeh_pe_state_clear(pe, EEH_PE_CFG_BLOCKED);
1628 		if (ret)
1629 			break;
1630 
1631 		ret = eeh_pe_reenable_devices(pe);
1632 		break;
1633 	case EEH_RESET_HOT:
1634 	case EEH_RESET_FUNDAMENTAL:
1635 		/*
1636 		 * Proactively freeze the PE to drop all MMIO access
1637 		 * during reset, which should be banned as it's always
1638 		 * cause recursive EEH error.
1639 		 */
1640 		eeh_ops->set_option(pe, EEH_OPT_FREEZE_PE);
1641 
1642 		eeh_pe_state_mark(pe, EEH_PE_CFG_BLOCKED);
1643 		ret = eeh_ops->reset(pe, option);
1644 		break;
1645 	default:
1646 		pr_debug("%s: Unsupported option %d\n",
1647 			__func__, option);
1648 		ret = -EINVAL;
1649 	}
1650 
1651 	return ret;
1652 }
1653 EXPORT_SYMBOL_GPL(eeh_pe_reset);
1654 
1655 /**
1656  * eeh_pe_configure - Configure PCI bridges after PE reset
1657  * @pe: EEH PE
1658  *
1659  * The routine is called to restore the PCI config space for
1660  * those PCI devices, especially PCI bridges affected by PE
1661  * reset issued previously.
1662  */
eeh_pe_configure(struct eeh_pe * pe)1663 int eeh_pe_configure(struct eeh_pe *pe)
1664 {
1665 	int ret = 0;
1666 
1667 	/* Invalid PE ? */
1668 	if (!pe)
1669 		return -ENODEV;
1670 
1671 	return ret;
1672 }
1673 EXPORT_SYMBOL_GPL(eeh_pe_configure);
1674 
1675 /**
1676  * eeh_pe_inject_err - Injecting the specified PCI error to the indicated PE
1677  * @pe: the indicated PE
1678  * @type: error type
1679  * @function: error function
1680  * @addr: address
1681  * @mask: address mask
1682  *
1683  * The routine is called to inject the specified PCI error, which
1684  * is determined by @type and @function, to the indicated PE for
1685  * testing purpose.
1686  */
eeh_pe_inject_err(struct eeh_pe * pe,int type,int func,unsigned long addr,unsigned long mask)1687 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
1688 		      unsigned long addr, unsigned long mask)
1689 {
1690 	/* Invalid PE ? */
1691 	if (!pe)
1692 		return -ENODEV;
1693 
1694 	/* Unsupported operation ? */
1695 	if (!eeh_ops || !eeh_ops->err_inject)
1696 		return -ENOENT;
1697 
1698 	/* Check on PCI error type */
1699 	if (type != EEH_ERR_TYPE_32 && type != EEH_ERR_TYPE_64)
1700 		return -EINVAL;
1701 
1702 	/* Check on PCI error function */
1703 	if (func < EEH_ERR_FUNC_MIN || func > EEH_ERR_FUNC_MAX)
1704 		return -EINVAL;
1705 
1706 	return eeh_ops->err_inject(pe, type, func, addr, mask);
1707 }
1708 EXPORT_SYMBOL_GPL(eeh_pe_inject_err);
1709 
proc_eeh_show(struct seq_file * m,void * v)1710 static int proc_eeh_show(struct seq_file *m, void *v)
1711 {
1712 	if (!eeh_enabled()) {
1713 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1714 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1715 	} else {
1716 		seq_printf(m, "EEH Subsystem is enabled\n");
1717 		seq_printf(m,
1718 				"no device=%llu\n"
1719 				"no device node=%llu\n"
1720 				"no config address=%llu\n"
1721 				"check not wanted=%llu\n"
1722 				"eeh_total_mmio_ffs=%llu\n"
1723 				"eeh_false_positives=%llu\n"
1724 				"eeh_slot_resets=%llu\n",
1725 				eeh_stats.no_device,
1726 				eeh_stats.no_dn,
1727 				eeh_stats.no_cfg_addr,
1728 				eeh_stats.ignored_check,
1729 				eeh_stats.total_mmio_ffs,
1730 				eeh_stats.false_positives,
1731 				eeh_stats.slot_resets);
1732 	}
1733 
1734 	return 0;
1735 }
1736 
proc_eeh_open(struct inode * inode,struct file * file)1737 static int proc_eeh_open(struct inode *inode, struct file *file)
1738 {
1739 	return single_open(file, proc_eeh_show, NULL);
1740 }
1741 
1742 static const struct file_operations proc_eeh_operations = {
1743 	.open      = proc_eeh_open,
1744 	.read      = seq_read,
1745 	.llseek    = seq_lseek,
1746 	.release   = single_release,
1747 };
1748 
1749 #ifdef CONFIG_DEBUG_FS
eeh_enable_dbgfs_set(void * data,u64 val)1750 static int eeh_enable_dbgfs_set(void *data, u64 val)
1751 {
1752 	if (val)
1753 		eeh_clear_flag(EEH_FORCE_DISABLED);
1754 	else
1755 		eeh_add_flag(EEH_FORCE_DISABLED);
1756 
1757 	/* Notify the backend */
1758 	if (eeh_ops->post_init)
1759 		eeh_ops->post_init();
1760 
1761 	return 0;
1762 }
1763 
eeh_enable_dbgfs_get(void * data,u64 * val)1764 static int eeh_enable_dbgfs_get(void *data, u64 *val)
1765 {
1766 	if (eeh_enabled())
1767 		*val = 0x1ul;
1768 	else
1769 		*val = 0x0ul;
1770 	return 0;
1771 }
1772 
eeh_freeze_dbgfs_set(void * data,u64 val)1773 static int eeh_freeze_dbgfs_set(void *data, u64 val)
1774 {
1775 	eeh_max_freezes = val;
1776 	return 0;
1777 }
1778 
eeh_freeze_dbgfs_get(void * data,u64 * val)1779 static int eeh_freeze_dbgfs_get(void *data, u64 *val)
1780 {
1781 	*val = eeh_max_freezes;
1782 	return 0;
1783 }
1784 
1785 DEFINE_SIMPLE_ATTRIBUTE(eeh_enable_dbgfs_ops, eeh_enable_dbgfs_get,
1786 			eeh_enable_dbgfs_set, "0x%llx\n");
1787 DEFINE_SIMPLE_ATTRIBUTE(eeh_freeze_dbgfs_ops, eeh_freeze_dbgfs_get,
1788 			eeh_freeze_dbgfs_set, "0x%llx\n");
1789 #endif
1790 
eeh_init_proc(void)1791 static int __init eeh_init_proc(void)
1792 {
1793 	if (machine_is(pseries) || machine_is(powernv)) {
1794 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1795 #ifdef CONFIG_DEBUG_FS
1796 		debugfs_create_file("eeh_enable", 0600,
1797                                     powerpc_debugfs_root, NULL,
1798                                     &eeh_enable_dbgfs_ops);
1799 		debugfs_create_file("eeh_max_freezes", 0600,
1800 				    powerpc_debugfs_root, NULL,
1801 				    &eeh_freeze_dbgfs_ops);
1802 #endif
1803 	}
1804 
1805 	return 0;
1806 }
1807 __initcall(eeh_init_proc);
1808