xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision 5bd8e16d)
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/proc_fs.h>
30 #include <linux/rbtree.h>
31 #include <linux/seq_file.h>
32 #include <linux/spinlock.h>
33 #include <linux/export.h>
34 #include <linux/of.h>
35 
36 #include <linux/atomic.h>
37 #include <asm/eeh.h>
38 #include <asm/eeh_event.h>
39 #include <asm/io.h>
40 #include <asm/machdep.h>
41 #include <asm/ppc-pci.h>
42 #include <asm/rtas.h>
43 
44 
45 /** Overview:
46  *  EEH, or "Extended Error Handling" is a PCI bridge technology for
47  *  dealing with PCI bus errors that can't be dealt with within the
48  *  usual PCI framework, except by check-stopping the CPU.  Systems
49  *  that are designed for high-availability/reliability cannot afford
50  *  to crash due to a "mere" PCI error, thus the need for EEH.
51  *  An EEH-capable bridge operates by converting a detected error
52  *  into a "slot freeze", taking the PCI adapter off-line, making
53  *  the slot behave, from the OS'es point of view, as if the slot
54  *  were "empty": all reads return 0xff's and all writes are silently
55  *  ignored.  EEH slot isolation events can be triggered by parity
56  *  errors on the address or data busses (e.g. during posted writes),
57  *  which in turn might be caused by low voltage on the bus, dust,
58  *  vibration, humidity, radioactivity or plain-old failed hardware.
59  *
60  *  Note, however, that one of the leading causes of EEH slot
61  *  freeze events are buggy device drivers, buggy device microcode,
62  *  or buggy device hardware.  This is because any attempt by the
63  *  device to bus-master data to a memory address that is not
64  *  assigned to the device will trigger a slot freeze.   (The idea
65  *  is to prevent devices-gone-wild from corrupting system memory).
66  *  Buggy hardware/drivers will have a miserable time co-existing
67  *  with EEH.
68  *
69  *  Ideally, a PCI device driver, when suspecting that an isolation
70  *  event has occurred (e.g. by reading 0xff's), will then ask EEH
71  *  whether this is the case, and then take appropriate steps to
72  *  reset the PCI slot, the PCI device, and then resume operations.
73  *  However, until that day,  the checking is done here, with the
74  *  eeh_check_failure() routine embedded in the MMIO macros.  If
75  *  the slot is found to be isolated, an "EEH Event" is synthesized
76  *  and sent out for processing.
77  */
78 
79 /* If a device driver keeps reading an MMIO register in an interrupt
80  * handler after a slot isolation event, it might be broken.
81  * This sets the threshold for how many read attempts we allow
82  * before printing an error message.
83  */
84 #define EEH_MAX_FAILS	2100000
85 
86 /* Time to wait for a PCI slot to report status, in milliseconds */
87 #define PCI_BUS_RESET_WAIT_MSEC (60*1000)
88 
89 /* Platform dependent EEH operations */
90 struct eeh_ops *eeh_ops = NULL;
91 
92 int eeh_subsystem_enabled;
93 EXPORT_SYMBOL(eeh_subsystem_enabled);
94 
95 /*
96  * EEH probe mode support. The intention is to support multiple
97  * platforms for EEH. Some platforms like pSeries do PCI emunation
98  * based on device tree. However, other platforms like powernv probe
99  * PCI devices from hardware. The flag is used to distinguish that.
100  * In addition, struct eeh_ops::probe would be invoked for particular
101  * OF node or PCI device so that the corresponding PE would be created
102  * there.
103  */
104 int eeh_probe_mode;
105 
106 /* Lock to avoid races due to multiple reports of an error */
107 DEFINE_RAW_SPINLOCK(confirm_error_lock);
108 
109 /* Buffer for reporting pci register dumps. Its here in BSS, and
110  * not dynamically alloced, so that it ends up in RMO where RTAS
111  * can access it.
112  */
113 #define EEH_PCI_REGS_LOG_LEN 4096
114 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
115 
116 /*
117  * The struct is used to maintain the EEH global statistic
118  * information. Besides, the EEH global statistics will be
119  * exported to user space through procfs
120  */
121 struct eeh_stats {
122 	u64 no_device;		/* PCI device not found		*/
123 	u64 no_dn;		/* OF node not found		*/
124 	u64 no_cfg_addr;	/* Config address not found	*/
125 	u64 ignored_check;	/* EEH check skipped		*/
126 	u64 total_mmio_ffs;	/* Total EEH checks		*/
127 	u64 false_positives;	/* Unnecessary EEH checks	*/
128 	u64 slot_resets;	/* PE reset			*/
129 };
130 
131 static struct eeh_stats eeh_stats;
132 
133 #define IS_BRIDGE(class_code) (((class_code)<<16) == PCI_BASE_CLASS_BRIDGE)
134 
135 /**
136  * eeh_gather_pci_data - Copy assorted PCI config space registers to buff
137  * @edev: device to report data for
138  * @buf: point to buffer in which to log
139  * @len: amount of room in buffer
140  *
141  * This routine captures assorted PCI configuration space data,
142  * and puts them into a buffer for RTAS error logging.
143  */
144 static size_t eeh_gather_pci_data(struct eeh_dev *edev, char * buf, size_t len)
145 {
146 	struct device_node *dn = eeh_dev_to_of_node(edev);
147 	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
148 	u32 cfg;
149 	int cap, i;
150 	int n = 0;
151 
152 	n += scnprintf(buf+n, len-n, "%s\n", dn->full_name);
153 	printk(KERN_WARNING "EEH: of node=%s\n", dn->full_name);
154 
155 	eeh_ops->read_config(dn, PCI_VENDOR_ID, 4, &cfg);
156 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
157 	printk(KERN_WARNING "EEH: PCI device/vendor: %08x\n", cfg);
158 
159 	eeh_ops->read_config(dn, PCI_COMMAND, 4, &cfg);
160 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
161 	printk(KERN_WARNING "EEH: PCI cmd/status register: %08x\n", cfg);
162 
163 	if (!dev) {
164 		printk(KERN_WARNING "EEH: no PCI device for this of node\n");
165 		return n;
166 	}
167 
168 	/* Gather bridge-specific registers */
169 	if (dev->class >> 16 == PCI_BASE_CLASS_BRIDGE) {
170 		eeh_ops->read_config(dn, PCI_SEC_STATUS, 2, &cfg);
171 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
172 		printk(KERN_WARNING "EEH: Bridge secondary status: %04x\n", cfg);
173 
174 		eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &cfg);
175 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
176 		printk(KERN_WARNING "EEH: Bridge control: %04x\n", cfg);
177 	}
178 
179 	/* Dump out the PCI-X command and status regs */
180 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
181 	if (cap) {
182 		eeh_ops->read_config(dn, cap, 4, &cfg);
183 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
184 		printk(KERN_WARNING "EEH: PCI-X cmd: %08x\n", cfg);
185 
186 		eeh_ops->read_config(dn, cap+4, 4, &cfg);
187 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
188 		printk(KERN_WARNING "EEH: PCI-X status: %08x\n", cfg);
189 	}
190 
191 	/* If PCI-E capable, dump PCI-E cap 10, and the AER */
192 	cap = pci_find_capability(dev, PCI_CAP_ID_EXP);
193 	if (cap) {
194 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
195 		printk(KERN_WARNING
196 		       "EEH: PCI-E capabilities and status follow:\n");
197 
198 		for (i=0; i<=8; i++) {
199 			eeh_ops->read_config(dn, cap+4*i, 4, &cfg);
200 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
201 			printk(KERN_WARNING "EEH: PCI-E %02x: %08x\n", i, cfg);
202 		}
203 
204 		cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
205 		if (cap) {
206 			n += scnprintf(buf+n, len-n, "pci-e AER:\n");
207 			printk(KERN_WARNING
208 			       "EEH: PCI-E AER capability register set follows:\n");
209 
210 			for (i=0; i<14; i++) {
211 				eeh_ops->read_config(dn, cap+4*i, 4, &cfg);
212 				n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
213 				printk(KERN_WARNING "EEH: PCI-E AER %02x: %08x\n", i, cfg);
214 			}
215 		}
216 	}
217 
218 	return n;
219 }
220 
221 /**
222  * eeh_slot_error_detail - Generate combined log including driver log and error log
223  * @pe: EEH PE
224  * @severity: temporary or permanent error log
225  *
226  * This routine should be called to generate the combined log, which
227  * is comprised of driver log and error log. The driver log is figured
228  * out from the config space of the corresponding PCI device, while
229  * the error log is fetched through platform dependent function call.
230  */
231 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
232 {
233 	size_t loglen = 0;
234 	struct eeh_dev *edev, *tmp;
235 	bool valid_cfg_log = true;
236 
237 	/*
238 	 * When the PHB is fenced or dead, it's pointless to collect
239 	 * the data from PCI config space because it should return
240 	 * 0xFF's. For ER, we still retrieve the data from the PCI
241 	 * config space.
242 	 */
243 	if (eeh_probe_mode_dev() &&
244 	    (pe->type & EEH_PE_PHB) &&
245 	    (pe->state & (EEH_PE_ISOLATED | EEH_PE_PHB_DEAD)))
246 		valid_cfg_log = false;
247 
248 	if (valid_cfg_log) {
249 		eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
250 		eeh_ops->configure_bridge(pe);
251 		eeh_pe_restore_bars(pe);
252 
253 		pci_regs_buf[0] = 0;
254 		eeh_pe_for_each_dev(pe, edev, tmp) {
255 			loglen += eeh_gather_pci_data(edev, pci_regs_buf + loglen,
256 						      EEH_PCI_REGS_LOG_LEN - loglen);
257 		}
258 	}
259 
260 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
261 }
262 
263 /**
264  * eeh_token_to_phys - Convert EEH address token to phys address
265  * @token: I/O token, should be address in the form 0xA....
266  *
267  * This routine should be called to convert virtual I/O address
268  * to physical one.
269  */
270 static inline unsigned long eeh_token_to_phys(unsigned long token)
271 {
272 	pte_t *ptep;
273 	unsigned long pa;
274 	int hugepage_shift;
275 
276 	/*
277 	 * We won't find hugepages here, iomem
278 	 */
279 	ptep = find_linux_pte_or_hugepte(init_mm.pgd, token, &hugepage_shift);
280 	if (!ptep)
281 		return token;
282 	WARN_ON(hugepage_shift);
283 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
284 
285 	return pa | (token & (PAGE_SIZE-1));
286 }
287 
288 /*
289  * On PowerNV platform, we might already have fenced PHB there.
290  * For that case, it's meaningless to recover frozen PE. Intead,
291  * We have to handle fenced PHB firstly.
292  */
293 static int eeh_phb_check_failure(struct eeh_pe *pe)
294 {
295 	struct eeh_pe *phb_pe;
296 	unsigned long flags;
297 	int ret;
298 
299 	if (!eeh_probe_mode_dev())
300 		return -EPERM;
301 
302 	/* Find the PHB PE */
303 	phb_pe = eeh_phb_pe_get(pe->phb);
304 	if (!phb_pe) {
305 		pr_warning("%s Can't find PE for PHB#%d\n",
306 			   __func__, pe->phb->global_number);
307 		return -EEXIST;
308 	}
309 
310 	/* If the PHB has been in problematic state */
311 	eeh_serialize_lock(&flags);
312 	if (phb_pe->state & (EEH_PE_ISOLATED | EEH_PE_PHB_DEAD)) {
313 		ret = 0;
314 		goto out;
315 	}
316 
317 	/* Check PHB state */
318 	ret = eeh_ops->get_state(phb_pe, NULL);
319 	if ((ret < 0) ||
320 	    (ret == EEH_STATE_NOT_SUPPORT) ||
321 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
322 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
323 		ret = 0;
324 		goto out;
325 	}
326 
327 	/* Isolate the PHB and send event */
328 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
329 	eeh_serialize_unlock(flags);
330 	eeh_send_failure_event(phb_pe);
331 
332 	pr_err("EEH: PHB#%x failure detected\n",
333 		phb_pe->phb->global_number);
334 	dump_stack();
335 
336 	return 1;
337 out:
338 	eeh_serialize_unlock(flags);
339 	return ret;
340 }
341 
342 /**
343  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
344  * @edev: eeh device
345  *
346  * Check for an EEH failure for the given device node.  Call this
347  * routine if the result of a read was all 0xff's and you want to
348  * find out if this is due to an EEH slot freeze.  This routine
349  * will query firmware for the EEH status.
350  *
351  * Returns 0 if there has not been an EEH error; otherwise returns
352  * a non-zero value and queues up a slot isolation event notification.
353  *
354  * It is safe to call this routine in an interrupt context.
355  */
356 int eeh_dev_check_failure(struct eeh_dev *edev)
357 {
358 	int ret;
359 	unsigned long flags;
360 	struct device_node *dn;
361 	struct pci_dev *dev;
362 	struct eeh_pe *pe;
363 	int rc = 0;
364 	const char *location;
365 
366 	eeh_stats.total_mmio_ffs++;
367 
368 	if (!eeh_subsystem_enabled)
369 		return 0;
370 
371 	if (!edev) {
372 		eeh_stats.no_dn++;
373 		return 0;
374 	}
375 	dn = eeh_dev_to_of_node(edev);
376 	dev = eeh_dev_to_pci_dev(edev);
377 	pe = edev->pe;
378 
379 	/* Access to IO BARs might get this far and still not want checking. */
380 	if (!pe) {
381 		eeh_stats.ignored_check++;
382 		pr_debug("EEH: Ignored check for %s %s\n",
383 			eeh_pci_name(dev), dn->full_name);
384 		return 0;
385 	}
386 
387 	if (!pe->addr && !pe->config_addr) {
388 		eeh_stats.no_cfg_addr++;
389 		return 0;
390 	}
391 
392 	/*
393 	 * On PowerNV platform, we might already have fenced PHB
394 	 * there and we need take care of that firstly.
395 	 */
396 	ret = eeh_phb_check_failure(pe);
397 	if (ret > 0)
398 		return ret;
399 
400 	/* If we already have a pending isolation event for this
401 	 * slot, we know it's bad already, we don't need to check.
402 	 * Do this checking under a lock; as multiple PCI devices
403 	 * in one slot might report errors simultaneously, and we
404 	 * only want one error recovery routine running.
405 	 */
406 	eeh_serialize_lock(&flags);
407 	rc = 1;
408 	if (pe->state & EEH_PE_ISOLATED) {
409 		pe->check_count++;
410 		if (pe->check_count % EEH_MAX_FAILS == 0) {
411 			location = of_get_property(dn, "ibm,loc-code", NULL);
412 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
413 				"location=%s driver=%s pci addr=%s\n",
414 				pe->check_count, location,
415 				eeh_driver_name(dev), eeh_pci_name(dev));
416 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
417 				eeh_driver_name(dev));
418 			dump_stack();
419 		}
420 		goto dn_unlock;
421 	}
422 
423 	/*
424 	 * Now test for an EEH failure.  This is VERY expensive.
425 	 * Note that the eeh_config_addr may be a parent device
426 	 * in the case of a device behind a bridge, or it may be
427 	 * function zero of a multi-function device.
428 	 * In any case they must share a common PHB.
429 	 */
430 	ret = eeh_ops->get_state(pe, NULL);
431 
432 	/* Note that config-io to empty slots may fail;
433 	 * they are empty when they don't have children.
434 	 * We will punt with the following conditions: Failure to get
435 	 * PE's state, EEH not support and Permanently unavailable
436 	 * state, PE is in good state.
437 	 */
438 	if ((ret < 0) ||
439 	    (ret == EEH_STATE_NOT_SUPPORT) ||
440 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
441 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
442 		eeh_stats.false_positives++;
443 		pe->false_positives++;
444 		rc = 0;
445 		goto dn_unlock;
446 	}
447 
448 	eeh_stats.slot_resets++;
449 
450 	/* Avoid repeated reports of this failure, including problems
451 	 * with other functions on this device, and functions under
452 	 * bridges.
453 	 */
454 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
455 	eeh_serialize_unlock(flags);
456 
457 	eeh_send_failure_event(pe);
458 
459 	/* Most EEH events are due to device driver bugs.  Having
460 	 * a stack trace will help the device-driver authors figure
461 	 * out what happened.  So print that out.
462 	 */
463 	pr_err("EEH: Frozen PE#%x detected on PHB#%x\n",
464 		pe->addr, pe->phb->global_number);
465 	dump_stack();
466 
467 	return 1;
468 
469 dn_unlock:
470 	eeh_serialize_unlock(flags);
471 	return rc;
472 }
473 
474 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
475 
476 /**
477  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
478  * @token: I/O token, should be address in the form 0xA....
479  * @val: value, should be all 1's (XXX why do we need this arg??)
480  *
481  * Check for an EEH failure at the given token address.  Call this
482  * routine if the result of a read was all 0xff's and you want to
483  * find out if this is due to an EEH slot freeze event.  This routine
484  * will query firmware for the EEH status.
485  *
486  * Note this routine is safe to call in an interrupt context.
487  */
488 unsigned long eeh_check_failure(const volatile void __iomem *token, unsigned long val)
489 {
490 	unsigned long addr;
491 	struct eeh_dev *edev;
492 
493 	/* Finding the phys addr + pci device; this is pretty quick. */
494 	addr = eeh_token_to_phys((unsigned long __force) token);
495 	edev = eeh_addr_cache_get_dev(addr);
496 	if (!edev) {
497 		eeh_stats.no_device++;
498 		return val;
499 	}
500 
501 	eeh_dev_check_failure(edev);
502 	return val;
503 }
504 
505 EXPORT_SYMBOL(eeh_check_failure);
506 
507 
508 /**
509  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
510  * @pe: EEH PE
511  *
512  * This routine should be called to reenable frozen MMIO or DMA
513  * so that it would work correctly again. It's useful while doing
514  * recovery or log collection on the indicated device.
515  */
516 int eeh_pci_enable(struct eeh_pe *pe, int function)
517 {
518 	int rc;
519 
520 	rc = eeh_ops->set_option(pe, function);
521 	if (rc)
522 		pr_warning("%s: Unexpected state change %d on PHB#%d-PE#%x, err=%d\n",
523 			__func__, function, pe->phb->global_number, pe->addr, rc);
524 
525 	rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
526 	if (rc > 0 && (rc & EEH_STATE_MMIO_ENABLED) &&
527 	   (function == EEH_OPT_THAW_MMIO))
528 		return 0;
529 
530 	return rc;
531 }
532 
533 /**
534  * pcibios_set_pcie_slot_reset - Set PCI-E reset state
535  * @dev: pci device struct
536  * @state: reset state to enter
537  *
538  * Return value:
539  * 	0 if success
540  */
541 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
542 {
543 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
544 	struct eeh_pe *pe = edev->pe;
545 
546 	if (!pe) {
547 		pr_err("%s: No PE found on PCI device %s\n",
548 			__func__, pci_name(dev));
549 		return -EINVAL;
550 	}
551 
552 	switch (state) {
553 	case pcie_deassert_reset:
554 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
555 		break;
556 	case pcie_hot_reset:
557 		eeh_ops->reset(pe, EEH_RESET_HOT);
558 		break;
559 	case pcie_warm_reset:
560 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
561 		break;
562 	default:
563 		return -EINVAL;
564 	};
565 
566 	return 0;
567 }
568 
569 /**
570  * eeh_set_pe_freset - Check the required reset for the indicated device
571  * @data: EEH device
572  * @flag: return value
573  *
574  * Each device might have its preferred reset type: fundamental or
575  * hot reset. The routine is used to collected the information for
576  * the indicated device and its children so that the bunch of the
577  * devices could be reset properly.
578  */
579 static void *eeh_set_dev_freset(void *data, void *flag)
580 {
581 	struct pci_dev *dev;
582 	unsigned int *freset = (unsigned int *)flag;
583 	struct eeh_dev *edev = (struct eeh_dev *)data;
584 
585 	dev = eeh_dev_to_pci_dev(edev);
586 	if (dev)
587 		*freset |= dev->needs_freset;
588 
589 	return NULL;
590 }
591 
592 /**
593  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
594  * @pe: EEH PE
595  *
596  * Assert the PCI #RST line for 1/4 second.
597  */
598 static void eeh_reset_pe_once(struct eeh_pe *pe)
599 {
600 	unsigned int freset = 0;
601 
602 	/* Determine type of EEH reset required for
603 	 * Partitionable Endpoint, a hot-reset (1)
604 	 * or a fundamental reset (3).
605 	 * A fundamental reset required by any device under
606 	 * Partitionable Endpoint trumps hot-reset.
607 	 */
608 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
609 
610 	if (freset)
611 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
612 	else
613 		eeh_ops->reset(pe, EEH_RESET_HOT);
614 
615 	/* The PCI bus requires that the reset be held high for at least
616 	 * a 100 milliseconds. We wait a bit longer 'just in case'.
617 	 */
618 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
619 	msleep(PCI_BUS_RST_HOLD_TIME_MSEC);
620 
621 	/* We might get hit with another EEH freeze as soon as the
622 	 * pci slot reset line is dropped. Make sure we don't miss
623 	 * these, and clear the flag now.
624 	 */
625 	eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
626 
627 	eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
628 
629 	/* After a PCI slot has been reset, the PCI Express spec requires
630 	 * a 1.5 second idle time for the bus to stabilize, before starting
631 	 * up traffic.
632 	 */
633 #define PCI_BUS_SETTLE_TIME_MSEC 1800
634 	msleep(PCI_BUS_SETTLE_TIME_MSEC);
635 }
636 
637 /**
638  * eeh_reset_pe - Reset the indicated PE
639  * @pe: EEH PE
640  *
641  * This routine should be called to reset indicated device, including
642  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
643  * might be involved as well.
644  */
645 int eeh_reset_pe(struct eeh_pe *pe)
646 {
647 	int flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
648 	int i, rc;
649 
650 	/* Take three shots at resetting the bus */
651 	for (i=0; i<3; i++) {
652 		eeh_reset_pe_once(pe);
653 
654 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
655 		if ((rc & flags) == flags)
656 			return 0;
657 
658 		if (rc < 0) {
659 			pr_err("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
660 				__func__, pe->phb->global_number, pe->addr);
661 			return -1;
662 		}
663 		pr_err("EEH: bus reset %d failed on PHB#%d-PE#%x, rc=%d\n",
664 			i+1, pe->phb->global_number, pe->addr, rc);
665 	}
666 
667 	return -1;
668 }
669 
670 /**
671  * eeh_save_bars - Save device bars
672  * @edev: PCI device associated EEH device
673  *
674  * Save the values of the device bars. Unlike the restore
675  * routine, this routine is *not* recursive. This is because
676  * PCI devices are added individually; but, for the restore,
677  * an entire slot is reset at a time.
678  */
679 void eeh_save_bars(struct eeh_dev *edev)
680 {
681 	int i;
682 	struct device_node *dn;
683 
684 	if (!edev)
685 		return;
686 	dn = eeh_dev_to_of_node(edev);
687 
688 	for (i = 0; i < 16; i++)
689 		eeh_ops->read_config(dn, i * 4, 4, &edev->config_space[i]);
690 }
691 
692 /**
693  * eeh_ops_register - Register platform dependent EEH operations
694  * @ops: platform dependent EEH operations
695  *
696  * Register the platform dependent EEH operation callback
697  * functions. The platform should call this function before
698  * any other EEH operations.
699  */
700 int __init eeh_ops_register(struct eeh_ops *ops)
701 {
702 	if (!ops->name) {
703 		pr_warning("%s: Invalid EEH ops name for %p\n",
704 			__func__, ops);
705 		return -EINVAL;
706 	}
707 
708 	if (eeh_ops && eeh_ops != ops) {
709 		pr_warning("%s: EEH ops of platform %s already existing (%s)\n",
710 			__func__, eeh_ops->name, ops->name);
711 		return -EEXIST;
712 	}
713 
714 	eeh_ops = ops;
715 
716 	return 0;
717 }
718 
719 /**
720  * eeh_ops_unregister - Unreigster platform dependent EEH operations
721  * @name: name of EEH platform operations
722  *
723  * Unregister the platform dependent EEH operation callback
724  * functions.
725  */
726 int __exit eeh_ops_unregister(const char *name)
727 {
728 	if (!name || !strlen(name)) {
729 		pr_warning("%s: Invalid EEH ops name\n",
730 			__func__);
731 		return -EINVAL;
732 	}
733 
734 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
735 		eeh_ops = NULL;
736 		return 0;
737 	}
738 
739 	return -EEXIST;
740 }
741 
742 /**
743  * eeh_init - EEH initialization
744  *
745  * Initialize EEH by trying to enable it for all of the adapters in the system.
746  * As a side effect we can determine here if eeh is supported at all.
747  * Note that we leave EEH on so failed config cycles won't cause a machine
748  * check.  If a user turns off EEH for a particular adapter they are really
749  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
750  * grant access to a slot if EEH isn't enabled, and so we always enable
751  * EEH for all slots/all devices.
752  *
753  * The eeh-force-off option disables EEH checking globally, for all slots.
754  * Even if force-off is set, the EEH hardware is still enabled, so that
755  * newer systems can boot.
756  */
757 int eeh_init(void)
758 {
759 	struct pci_controller *hose, *tmp;
760 	struct device_node *phb;
761 	static int cnt = 0;
762 	int ret = 0;
763 
764 	/*
765 	 * We have to delay the initialization on PowerNV after
766 	 * the PCI hierarchy tree has been built because the PEs
767 	 * are figured out based on PCI devices instead of device
768 	 * tree nodes
769 	 */
770 	if (machine_is(powernv) && cnt++ <= 0)
771 		return ret;
772 
773 	/* call platform initialization function */
774 	if (!eeh_ops) {
775 		pr_warning("%s: Platform EEH operation not found\n",
776 			__func__);
777 		return -EEXIST;
778 	} else if ((ret = eeh_ops->init())) {
779 		pr_warning("%s: Failed to call platform init function (%d)\n",
780 			__func__, ret);
781 		return ret;
782 	}
783 
784 	/* Initialize EEH event */
785 	ret = eeh_event_init();
786 	if (ret)
787 		return ret;
788 
789 	/* Enable EEH for all adapters */
790 	if (eeh_probe_mode_devtree()) {
791 		list_for_each_entry_safe(hose, tmp,
792 			&hose_list, list_node) {
793 			phb = hose->dn;
794 			traverse_pci_devices(phb, eeh_ops->of_probe, NULL);
795 		}
796 	} else if (eeh_probe_mode_dev()) {
797 		list_for_each_entry_safe(hose, tmp,
798 			&hose_list, list_node)
799 			pci_walk_bus(hose->bus, eeh_ops->dev_probe, NULL);
800 	} else {
801 		pr_warning("%s: Invalid probe mode %d\n",
802 			   __func__, eeh_probe_mode);
803 		return -EINVAL;
804 	}
805 
806 	/*
807 	 * Call platform post-initialization. Actually, It's good chance
808 	 * to inform platform that EEH is ready to supply service if the
809 	 * I/O cache stuff has been built up.
810 	 */
811 	if (eeh_ops->post_init) {
812 		ret = eeh_ops->post_init();
813 		if (ret)
814 			return ret;
815 	}
816 
817 	if (eeh_subsystem_enabled)
818 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
819 	else
820 		pr_warning("EEH: No capable adapters found\n");
821 
822 	return ret;
823 }
824 
825 core_initcall_sync(eeh_init);
826 
827 /**
828  * eeh_add_device_early - Enable EEH for the indicated device_node
829  * @dn: device node for which to set up EEH
830  *
831  * This routine must be used to perform EEH initialization for PCI
832  * devices that were added after system boot (e.g. hotplug, dlpar).
833  * This routine must be called before any i/o is performed to the
834  * adapter (inluding any config-space i/o).
835  * Whether this actually enables EEH or not for this device depends
836  * on the CEC architecture, type of the device, on earlier boot
837  * command-line arguments & etc.
838  */
839 void eeh_add_device_early(struct device_node *dn)
840 {
841 	struct pci_controller *phb;
842 
843 	/*
844 	 * If we're doing EEH probe based on PCI device, we
845 	 * would delay the probe until late stage because
846 	 * the PCI device isn't available this moment.
847 	 */
848 	if (!eeh_probe_mode_devtree())
849 		return;
850 
851 	if (!of_node_to_eeh_dev(dn))
852 		return;
853 	phb = of_node_to_eeh_dev(dn)->phb;
854 
855 	/* USB Bus children of PCI devices will not have BUID's */
856 	if (NULL == phb || 0 == phb->buid)
857 		return;
858 
859 	eeh_ops->of_probe(dn, NULL);
860 }
861 
862 /**
863  * eeh_add_device_tree_early - Enable EEH for the indicated device
864  * @dn: device node
865  *
866  * This routine must be used to perform EEH initialization for the
867  * indicated PCI device that was added after system boot (e.g.
868  * hotplug, dlpar).
869  */
870 void eeh_add_device_tree_early(struct device_node *dn)
871 {
872 	struct device_node *sib;
873 
874 	for_each_child_of_node(dn, sib)
875 		eeh_add_device_tree_early(sib);
876 	eeh_add_device_early(dn);
877 }
878 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
879 
880 /**
881  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
882  * @dev: pci device for which to set up EEH
883  *
884  * This routine must be used to complete EEH initialization for PCI
885  * devices that were added after system boot (e.g. hotplug, dlpar).
886  */
887 void eeh_add_device_late(struct pci_dev *dev)
888 {
889 	struct device_node *dn;
890 	struct eeh_dev *edev;
891 
892 	if (!dev || !eeh_subsystem_enabled)
893 		return;
894 
895 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
896 
897 	dn = pci_device_to_OF_node(dev);
898 	edev = of_node_to_eeh_dev(dn);
899 	if (edev->pdev == dev) {
900 		pr_debug("EEH: Already referenced !\n");
901 		return;
902 	}
903 
904 	/*
905 	 * The EEH cache might not be removed correctly because of
906 	 * unbalanced kref to the device during unplug time, which
907 	 * relies on pcibios_release_device(). So we have to remove
908 	 * that here explicitly.
909 	 */
910 	if (edev->pdev) {
911 		eeh_rmv_from_parent_pe(edev);
912 		eeh_addr_cache_rmv_dev(edev->pdev);
913 		eeh_sysfs_remove_device(edev->pdev);
914 		edev->mode &= ~EEH_DEV_SYSFS;
915 
916 		edev->pdev = NULL;
917 		dev->dev.archdata.edev = NULL;
918 	}
919 
920 	edev->pdev = dev;
921 	dev->dev.archdata.edev = edev;
922 
923 	/*
924 	 * We have to do the EEH probe here because the PCI device
925 	 * hasn't been created yet in the early stage.
926 	 */
927 	if (eeh_probe_mode_dev())
928 		eeh_ops->dev_probe(dev, NULL);
929 
930 	eeh_addr_cache_insert_dev(dev);
931 }
932 
933 /**
934  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
935  * @bus: PCI bus
936  *
937  * This routine must be used to perform EEH initialization for PCI
938  * devices which are attached to the indicated PCI bus. The PCI bus
939  * is added after system boot through hotplug or dlpar.
940  */
941 void eeh_add_device_tree_late(struct pci_bus *bus)
942 {
943 	struct pci_dev *dev;
944 
945 	list_for_each_entry(dev, &bus->devices, bus_list) {
946 		eeh_add_device_late(dev);
947 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
948 			struct pci_bus *subbus = dev->subordinate;
949 			if (subbus)
950 				eeh_add_device_tree_late(subbus);
951 		}
952 	}
953 }
954 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
955 
956 /**
957  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
958  * @bus: PCI bus
959  *
960  * This routine must be used to add EEH sysfs files for PCI
961  * devices which are attached to the indicated PCI bus. The PCI bus
962  * is added after system boot through hotplug or dlpar.
963  */
964 void eeh_add_sysfs_files(struct pci_bus *bus)
965 {
966 	struct pci_dev *dev;
967 
968 	list_for_each_entry(dev, &bus->devices, bus_list) {
969 		eeh_sysfs_add_device(dev);
970 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
971 			struct pci_bus *subbus = dev->subordinate;
972 			if (subbus)
973 				eeh_add_sysfs_files(subbus);
974 		}
975 	}
976 }
977 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
978 
979 /**
980  * eeh_remove_device - Undo EEH setup for the indicated pci device
981  * @dev: pci device to be removed
982  *
983  * This routine should be called when a device is removed from
984  * a running system (e.g. by hotplug or dlpar).  It unregisters
985  * the PCI device from the EEH subsystem.  I/O errors affecting
986  * this device will no longer be detected after this call; thus,
987  * i/o errors affecting this slot may leave this device unusable.
988  */
989 void eeh_remove_device(struct pci_dev *dev)
990 {
991 	struct eeh_dev *edev;
992 
993 	if (!dev || !eeh_subsystem_enabled)
994 		return;
995 	edev = pci_dev_to_eeh_dev(dev);
996 
997 	/* Unregister the device with the EEH/PCI address search system */
998 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
999 
1000 	if (!edev || !edev->pdev || !edev->pe) {
1001 		pr_debug("EEH: Not referenced !\n");
1002 		return;
1003 	}
1004 
1005 	/*
1006 	 * During the hotplug for EEH error recovery, we need the EEH
1007 	 * device attached to the parent PE in order for BAR restore
1008 	 * a bit later. So we keep it for BAR restore and remove it
1009 	 * from the parent PE during the BAR resotre.
1010 	 */
1011 	edev->pdev = NULL;
1012 	dev->dev.archdata.edev = NULL;
1013 	if (!(edev->pe->state & EEH_PE_KEEP))
1014 		eeh_rmv_from_parent_pe(edev);
1015 	else
1016 		edev->mode |= EEH_DEV_DISCONNECTED;
1017 
1018 	eeh_addr_cache_rmv_dev(dev);
1019 	eeh_sysfs_remove_device(dev);
1020 	edev->mode &= ~EEH_DEV_SYSFS;
1021 }
1022 
1023 static int proc_eeh_show(struct seq_file *m, void *v)
1024 {
1025 	if (0 == eeh_subsystem_enabled) {
1026 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1027 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1028 	} else {
1029 		seq_printf(m, "EEH Subsystem is enabled\n");
1030 		seq_printf(m,
1031 				"no device=%llu\n"
1032 				"no device node=%llu\n"
1033 				"no config address=%llu\n"
1034 				"check not wanted=%llu\n"
1035 				"eeh_total_mmio_ffs=%llu\n"
1036 				"eeh_false_positives=%llu\n"
1037 				"eeh_slot_resets=%llu\n",
1038 				eeh_stats.no_device,
1039 				eeh_stats.no_dn,
1040 				eeh_stats.no_cfg_addr,
1041 				eeh_stats.ignored_check,
1042 				eeh_stats.total_mmio_ffs,
1043 				eeh_stats.false_positives,
1044 				eeh_stats.slot_resets);
1045 	}
1046 
1047 	return 0;
1048 }
1049 
1050 static int proc_eeh_open(struct inode *inode, struct file *file)
1051 {
1052 	return single_open(file, proc_eeh_show, NULL);
1053 }
1054 
1055 static const struct file_operations proc_eeh_operations = {
1056 	.open      = proc_eeh_open,
1057 	.read      = seq_read,
1058 	.llseek    = seq_lseek,
1059 	.release   = single_release,
1060 };
1061 
1062 static int __init eeh_init_proc(void)
1063 {
1064 	if (machine_is(pseries) || machine_is(powernv))
1065 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1066 	return 0;
1067 }
1068 __initcall(eeh_init_proc);
1069