xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision 12eb4683)
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 	if (pci_is_pcie(dev)) {
193 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
194 		printk(KERN_WARNING
195 		       "EEH: PCI-E capabilities and status follow:\n");
196 
197 		for (i=0; i<=8; i++) {
198 			eeh_ops->read_config(dn, dev->pcie_cap+4*i, 4, &cfg);
199 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
200 			printk(KERN_WARNING "EEH: PCI-E %02x: %08x\n", i, cfg);
201 		}
202 
203 		cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
204 		if (cap) {
205 			n += scnprintf(buf+n, len-n, "pci-e AER:\n");
206 			printk(KERN_WARNING
207 			       "EEH: PCI-E AER capability register set follows:\n");
208 
209 			for (i=0; i<14; i++) {
210 				eeh_ops->read_config(dn, cap+4*i, 4, &cfg);
211 				n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
212 				printk(KERN_WARNING "EEH: PCI-E AER %02x: %08x\n", i, cfg);
213 			}
214 		}
215 	}
216 
217 	return n;
218 }
219 
220 /**
221  * eeh_slot_error_detail - Generate combined log including driver log and error log
222  * @pe: EEH PE
223  * @severity: temporary or permanent error log
224  *
225  * This routine should be called to generate the combined log, which
226  * is comprised of driver log and error log. The driver log is figured
227  * out from the config space of the corresponding PCI device, while
228  * the error log is fetched through platform dependent function call.
229  */
230 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
231 {
232 	size_t loglen = 0;
233 	struct eeh_dev *edev, *tmp;
234 	bool valid_cfg_log = true;
235 
236 	/*
237 	 * When the PHB is fenced or dead, it's pointless to collect
238 	 * the data from PCI config space because it should return
239 	 * 0xFF's. For ER, we still retrieve the data from the PCI
240 	 * config space.
241 	 */
242 	if (eeh_probe_mode_dev() &&
243 	    (pe->type & EEH_PE_PHB) &&
244 	    (pe->state & (EEH_PE_ISOLATED | EEH_PE_PHB_DEAD)))
245 		valid_cfg_log = false;
246 
247 	if (valid_cfg_log) {
248 		eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
249 		eeh_ops->configure_bridge(pe);
250 		eeh_pe_restore_bars(pe);
251 
252 		pci_regs_buf[0] = 0;
253 		eeh_pe_for_each_dev(pe, edev, tmp) {
254 			loglen += eeh_gather_pci_data(edev, pci_regs_buf + loglen,
255 						      EEH_PCI_REGS_LOG_LEN - loglen);
256 		}
257 	}
258 
259 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
260 }
261 
262 /**
263  * eeh_token_to_phys - Convert EEH address token to phys address
264  * @token: I/O token, should be address in the form 0xA....
265  *
266  * This routine should be called to convert virtual I/O address
267  * to physical one.
268  */
269 static inline unsigned long eeh_token_to_phys(unsigned long token)
270 {
271 	pte_t *ptep;
272 	unsigned long pa;
273 	int hugepage_shift;
274 
275 	/*
276 	 * We won't find hugepages here, iomem
277 	 */
278 	ptep = find_linux_pte_or_hugepte(init_mm.pgd, token, &hugepage_shift);
279 	if (!ptep)
280 		return token;
281 	WARN_ON(hugepage_shift);
282 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
283 
284 	return pa | (token & (PAGE_SIZE-1));
285 }
286 
287 /*
288  * On PowerNV platform, we might already have fenced PHB there.
289  * For that case, it's meaningless to recover frozen PE. Intead,
290  * We have to handle fenced PHB firstly.
291  */
292 static int eeh_phb_check_failure(struct eeh_pe *pe)
293 {
294 	struct eeh_pe *phb_pe;
295 	unsigned long flags;
296 	int ret;
297 
298 	if (!eeh_probe_mode_dev())
299 		return -EPERM;
300 
301 	/* Find the PHB PE */
302 	phb_pe = eeh_phb_pe_get(pe->phb);
303 	if (!phb_pe) {
304 		pr_warning("%s Can't find PE for PHB#%d\n",
305 			   __func__, pe->phb->global_number);
306 		return -EEXIST;
307 	}
308 
309 	/* If the PHB has been in problematic state */
310 	eeh_serialize_lock(&flags);
311 	if (phb_pe->state & (EEH_PE_ISOLATED | EEH_PE_PHB_DEAD)) {
312 		ret = 0;
313 		goto out;
314 	}
315 
316 	/* Check PHB state */
317 	ret = eeh_ops->get_state(phb_pe, NULL);
318 	if ((ret < 0) ||
319 	    (ret == EEH_STATE_NOT_SUPPORT) ||
320 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
321 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
322 		ret = 0;
323 		goto out;
324 	}
325 
326 	/* Isolate the PHB and send event */
327 	eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
328 	eeh_serialize_unlock(flags);
329 
330 	pr_err("EEH: PHB#%x failure detected\n",
331 		phb_pe->phb->global_number);
332 	dump_stack();
333 	eeh_send_failure_event(phb_pe);
334 
335 	return 1;
336 out:
337 	eeh_serialize_unlock(flags);
338 	return ret;
339 }
340 
341 /**
342  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
343  * @edev: eeh device
344  *
345  * Check for an EEH failure for the given device node.  Call this
346  * routine if the result of a read was all 0xff's and you want to
347  * find out if this is due to an EEH slot freeze.  This routine
348  * will query firmware for the EEH status.
349  *
350  * Returns 0 if there has not been an EEH error; otherwise returns
351  * a non-zero value and queues up a slot isolation event notification.
352  *
353  * It is safe to call this routine in an interrupt context.
354  */
355 int eeh_dev_check_failure(struct eeh_dev *edev)
356 {
357 	int ret;
358 	unsigned long flags;
359 	struct device_node *dn;
360 	struct pci_dev *dev;
361 	struct eeh_pe *pe;
362 	int rc = 0;
363 	const char *location;
364 
365 	eeh_stats.total_mmio_ffs++;
366 
367 	if (!eeh_subsystem_enabled)
368 		return 0;
369 
370 	if (!edev) {
371 		eeh_stats.no_dn++;
372 		return 0;
373 	}
374 	dn = eeh_dev_to_of_node(edev);
375 	dev = eeh_dev_to_pci_dev(edev);
376 	pe = edev->pe;
377 
378 	/* Access to IO BARs might get this far and still not want checking. */
379 	if (!pe) {
380 		eeh_stats.ignored_check++;
381 		pr_debug("EEH: Ignored check for %s %s\n",
382 			eeh_pci_name(dev), dn->full_name);
383 		return 0;
384 	}
385 
386 	if (!pe->addr && !pe->config_addr) {
387 		eeh_stats.no_cfg_addr++;
388 		return 0;
389 	}
390 
391 	/*
392 	 * On PowerNV platform, we might already have fenced PHB
393 	 * there and we need take care of that firstly.
394 	 */
395 	ret = eeh_phb_check_failure(pe);
396 	if (ret > 0)
397 		return ret;
398 
399 	/* If we already have a pending isolation event for this
400 	 * slot, we know it's bad already, we don't need to check.
401 	 * Do this checking under a lock; as multiple PCI devices
402 	 * in one slot might report errors simultaneously, and we
403 	 * only want one error recovery routine running.
404 	 */
405 	eeh_serialize_lock(&flags);
406 	rc = 1;
407 	if (pe->state & EEH_PE_ISOLATED) {
408 		pe->check_count++;
409 		if (pe->check_count % EEH_MAX_FAILS == 0) {
410 			location = of_get_property(dn, "ibm,loc-code", NULL);
411 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
412 				"location=%s driver=%s pci addr=%s\n",
413 				pe->check_count, location,
414 				eeh_driver_name(dev), eeh_pci_name(dev));
415 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
416 				eeh_driver_name(dev));
417 			dump_stack();
418 		}
419 		goto dn_unlock;
420 	}
421 
422 	/*
423 	 * Now test for an EEH failure.  This is VERY expensive.
424 	 * Note that the eeh_config_addr may be a parent device
425 	 * in the case of a device behind a bridge, or it may be
426 	 * function zero of a multi-function device.
427 	 * In any case they must share a common PHB.
428 	 */
429 	ret = eeh_ops->get_state(pe, NULL);
430 
431 	/* Note that config-io to empty slots may fail;
432 	 * they are empty when they don't have children.
433 	 * We will punt with the following conditions: Failure to get
434 	 * PE's state, EEH not support and Permanently unavailable
435 	 * state, PE is in good state.
436 	 */
437 	if ((ret < 0) ||
438 	    (ret == EEH_STATE_NOT_SUPPORT) ||
439 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
440 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
441 		eeh_stats.false_positives++;
442 		pe->false_positives++;
443 		rc = 0;
444 		goto dn_unlock;
445 	}
446 
447 	eeh_stats.slot_resets++;
448 
449 	/* Avoid repeated reports of this failure, including problems
450 	 * with other functions on this device, and functions under
451 	 * bridges.
452 	 */
453 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
454 	eeh_serialize_unlock(flags);
455 
456 	/* Most EEH events are due to device driver bugs.  Having
457 	 * a stack trace will help the device-driver authors figure
458 	 * out what happened.  So print that out.
459 	 */
460 	pr_err("EEH: Frozen PE#%x detected on PHB#%x\n",
461 		pe->addr, pe->phb->global_number);
462 	dump_stack();
463 
464 	eeh_send_failure_event(pe);
465 
466 	return 1;
467 
468 dn_unlock:
469 	eeh_serialize_unlock(flags);
470 	return rc;
471 }
472 
473 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
474 
475 /**
476  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
477  * @token: I/O token, should be address in the form 0xA....
478  * @val: value, should be all 1's (XXX why do we need this arg??)
479  *
480  * Check for an EEH failure at the given token address.  Call this
481  * routine if the result of a read was all 0xff's and you want to
482  * find out if this is due to an EEH slot freeze event.  This routine
483  * will query firmware for the EEH status.
484  *
485  * Note this routine is safe to call in an interrupt context.
486  */
487 unsigned long eeh_check_failure(const volatile void __iomem *token, unsigned long val)
488 {
489 	unsigned long addr;
490 	struct eeh_dev *edev;
491 
492 	/* Finding the phys addr + pci device; this is pretty quick. */
493 	addr = eeh_token_to_phys((unsigned long __force) token);
494 	edev = eeh_addr_cache_get_dev(addr);
495 	if (!edev) {
496 		eeh_stats.no_device++;
497 		return val;
498 	}
499 
500 	eeh_dev_check_failure(edev);
501 	return val;
502 }
503 
504 EXPORT_SYMBOL(eeh_check_failure);
505 
506 
507 /**
508  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
509  * @pe: EEH PE
510  *
511  * This routine should be called to reenable frozen MMIO or DMA
512  * so that it would work correctly again. It's useful while doing
513  * recovery or log collection on the indicated device.
514  */
515 int eeh_pci_enable(struct eeh_pe *pe, int function)
516 {
517 	int rc;
518 
519 	rc = eeh_ops->set_option(pe, function);
520 	if (rc)
521 		pr_warning("%s: Unexpected state change %d on PHB#%d-PE#%x, err=%d\n",
522 			__func__, function, pe->phb->global_number, pe->addr, rc);
523 
524 	rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
525 	if (rc > 0 && (rc & EEH_STATE_MMIO_ENABLED) &&
526 	   (function == EEH_OPT_THAW_MMIO))
527 		return 0;
528 
529 	return rc;
530 }
531 
532 /**
533  * pcibios_set_pcie_slot_reset - Set PCI-E reset state
534  * @dev: pci device struct
535  * @state: reset state to enter
536  *
537  * Return value:
538  * 	0 if success
539  */
540 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
541 {
542 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
543 	struct eeh_pe *pe = edev->pe;
544 
545 	if (!pe) {
546 		pr_err("%s: No PE found on PCI device %s\n",
547 			__func__, pci_name(dev));
548 		return -EINVAL;
549 	}
550 
551 	switch (state) {
552 	case pcie_deassert_reset:
553 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
554 		break;
555 	case pcie_hot_reset:
556 		eeh_ops->reset(pe, EEH_RESET_HOT);
557 		break;
558 	case pcie_warm_reset:
559 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
560 		break;
561 	default:
562 		return -EINVAL;
563 	};
564 
565 	return 0;
566 }
567 
568 /**
569  * eeh_set_pe_freset - Check the required reset for the indicated device
570  * @data: EEH device
571  * @flag: return value
572  *
573  * Each device might have its preferred reset type: fundamental or
574  * hot reset. The routine is used to collected the information for
575  * the indicated device and its children so that the bunch of the
576  * devices could be reset properly.
577  */
578 static void *eeh_set_dev_freset(void *data, void *flag)
579 {
580 	struct pci_dev *dev;
581 	unsigned int *freset = (unsigned int *)flag;
582 	struct eeh_dev *edev = (struct eeh_dev *)data;
583 
584 	dev = eeh_dev_to_pci_dev(edev);
585 	if (dev)
586 		*freset |= dev->needs_freset;
587 
588 	return NULL;
589 }
590 
591 /**
592  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
593  * @pe: EEH PE
594  *
595  * Assert the PCI #RST line for 1/4 second.
596  */
597 static void eeh_reset_pe_once(struct eeh_pe *pe)
598 {
599 	unsigned int freset = 0;
600 
601 	/* Determine type of EEH reset required for
602 	 * Partitionable Endpoint, a hot-reset (1)
603 	 * or a fundamental reset (3).
604 	 * A fundamental reset required by any device under
605 	 * Partitionable Endpoint trumps hot-reset.
606 	 */
607 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
608 
609 	if (freset)
610 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
611 	else
612 		eeh_ops->reset(pe, EEH_RESET_HOT);
613 
614 	/* The PCI bus requires that the reset be held high for at least
615 	 * a 100 milliseconds. We wait a bit longer 'just in case'.
616 	 */
617 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
618 	msleep(PCI_BUS_RST_HOLD_TIME_MSEC);
619 
620 	/* We might get hit with another EEH freeze as soon as the
621 	 * pci slot reset line is dropped. Make sure we don't miss
622 	 * these, and clear the flag now.
623 	 */
624 	eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
625 
626 	eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
627 
628 	/* After a PCI slot has been reset, the PCI Express spec requires
629 	 * a 1.5 second idle time for the bus to stabilize, before starting
630 	 * up traffic.
631 	 */
632 #define PCI_BUS_SETTLE_TIME_MSEC 1800
633 	msleep(PCI_BUS_SETTLE_TIME_MSEC);
634 }
635 
636 /**
637  * eeh_reset_pe - Reset the indicated PE
638  * @pe: EEH PE
639  *
640  * This routine should be called to reset indicated device, including
641  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
642  * might be involved as well.
643  */
644 int eeh_reset_pe(struct eeh_pe *pe)
645 {
646 	int flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
647 	int i, rc;
648 
649 	/* Take three shots at resetting the bus */
650 	for (i=0; i<3; i++) {
651 		eeh_reset_pe_once(pe);
652 
653 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
654 		if ((rc & flags) == flags)
655 			return 0;
656 
657 		if (rc < 0) {
658 			pr_err("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
659 				__func__, pe->phb->global_number, pe->addr);
660 			return -1;
661 		}
662 		pr_err("EEH: bus reset %d failed on PHB#%d-PE#%x, rc=%d\n",
663 			i+1, pe->phb->global_number, pe->addr, rc);
664 	}
665 
666 	return -1;
667 }
668 
669 /**
670  * eeh_save_bars - Save device bars
671  * @edev: PCI device associated EEH device
672  *
673  * Save the values of the device bars. Unlike the restore
674  * routine, this routine is *not* recursive. This is because
675  * PCI devices are added individually; but, for the restore,
676  * an entire slot is reset at a time.
677  */
678 void eeh_save_bars(struct eeh_dev *edev)
679 {
680 	int i;
681 	struct device_node *dn;
682 
683 	if (!edev)
684 		return;
685 	dn = eeh_dev_to_of_node(edev);
686 
687 	for (i = 0; i < 16; i++)
688 		eeh_ops->read_config(dn, i * 4, 4, &edev->config_space[i]);
689 
690 	/*
691 	 * For PCI bridges including root port, we need enable bus
692 	 * master explicitly. Otherwise, it can't fetch IODA table
693 	 * entries correctly. So we cache the bit in advance so that
694 	 * we can restore it after reset, either PHB range or PE range.
695 	 */
696 	if (edev->mode & EEH_DEV_BRIDGE)
697 		edev->config_space[1] |= PCI_COMMAND_MASTER;
698 }
699 
700 /**
701  * eeh_ops_register - Register platform dependent EEH operations
702  * @ops: platform dependent EEH operations
703  *
704  * Register the platform dependent EEH operation callback
705  * functions. The platform should call this function before
706  * any other EEH operations.
707  */
708 int __init eeh_ops_register(struct eeh_ops *ops)
709 {
710 	if (!ops->name) {
711 		pr_warning("%s: Invalid EEH ops name for %p\n",
712 			__func__, ops);
713 		return -EINVAL;
714 	}
715 
716 	if (eeh_ops && eeh_ops != ops) {
717 		pr_warning("%s: EEH ops of platform %s already existing (%s)\n",
718 			__func__, eeh_ops->name, ops->name);
719 		return -EEXIST;
720 	}
721 
722 	eeh_ops = ops;
723 
724 	return 0;
725 }
726 
727 /**
728  * eeh_ops_unregister - Unreigster platform dependent EEH operations
729  * @name: name of EEH platform operations
730  *
731  * Unregister the platform dependent EEH operation callback
732  * functions.
733  */
734 int __exit eeh_ops_unregister(const char *name)
735 {
736 	if (!name || !strlen(name)) {
737 		pr_warning("%s: Invalid EEH ops name\n",
738 			__func__);
739 		return -EINVAL;
740 	}
741 
742 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
743 		eeh_ops = NULL;
744 		return 0;
745 	}
746 
747 	return -EEXIST;
748 }
749 
750 /**
751  * eeh_init - EEH initialization
752  *
753  * Initialize EEH by trying to enable it for all of the adapters in the system.
754  * As a side effect we can determine here if eeh is supported at all.
755  * Note that we leave EEH on so failed config cycles won't cause a machine
756  * check.  If a user turns off EEH for a particular adapter they are really
757  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
758  * grant access to a slot if EEH isn't enabled, and so we always enable
759  * EEH for all slots/all devices.
760  *
761  * The eeh-force-off option disables EEH checking globally, for all slots.
762  * Even if force-off is set, the EEH hardware is still enabled, so that
763  * newer systems can boot.
764  */
765 int eeh_init(void)
766 {
767 	struct pci_controller *hose, *tmp;
768 	struct device_node *phb;
769 	static int cnt = 0;
770 	int ret = 0;
771 
772 	/*
773 	 * We have to delay the initialization on PowerNV after
774 	 * the PCI hierarchy tree has been built because the PEs
775 	 * are figured out based on PCI devices instead of device
776 	 * tree nodes
777 	 */
778 	if (machine_is(powernv) && cnt++ <= 0)
779 		return ret;
780 
781 	/* call platform initialization function */
782 	if (!eeh_ops) {
783 		pr_warning("%s: Platform EEH operation not found\n",
784 			__func__);
785 		return -EEXIST;
786 	} else if ((ret = eeh_ops->init())) {
787 		pr_warning("%s: Failed to call platform init function (%d)\n",
788 			__func__, ret);
789 		return ret;
790 	}
791 
792 	/* Initialize EEH event */
793 	ret = eeh_event_init();
794 	if (ret)
795 		return ret;
796 
797 	/* Enable EEH for all adapters */
798 	if (eeh_probe_mode_devtree()) {
799 		list_for_each_entry_safe(hose, tmp,
800 			&hose_list, list_node) {
801 			phb = hose->dn;
802 			traverse_pci_devices(phb, eeh_ops->of_probe, NULL);
803 		}
804 	} else if (eeh_probe_mode_dev()) {
805 		list_for_each_entry_safe(hose, tmp,
806 			&hose_list, list_node)
807 			pci_walk_bus(hose->bus, eeh_ops->dev_probe, NULL);
808 	} else {
809 		pr_warning("%s: Invalid probe mode %d\n",
810 			   __func__, eeh_probe_mode);
811 		return -EINVAL;
812 	}
813 
814 	/*
815 	 * Call platform post-initialization. Actually, It's good chance
816 	 * to inform platform that EEH is ready to supply service if the
817 	 * I/O cache stuff has been built up.
818 	 */
819 	if (eeh_ops->post_init) {
820 		ret = eeh_ops->post_init();
821 		if (ret)
822 			return ret;
823 	}
824 
825 	if (eeh_subsystem_enabled)
826 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
827 	else
828 		pr_warning("EEH: No capable adapters found\n");
829 
830 	return ret;
831 }
832 
833 core_initcall_sync(eeh_init);
834 
835 /**
836  * eeh_add_device_early - Enable EEH for the indicated device_node
837  * @dn: device node for which to set up EEH
838  *
839  * This routine must be used to perform EEH initialization for PCI
840  * devices that were added after system boot (e.g. hotplug, dlpar).
841  * This routine must be called before any i/o is performed to the
842  * adapter (inluding any config-space i/o).
843  * Whether this actually enables EEH or not for this device depends
844  * on the CEC architecture, type of the device, on earlier boot
845  * command-line arguments & etc.
846  */
847 void eeh_add_device_early(struct device_node *dn)
848 {
849 	struct pci_controller *phb;
850 
851 	/*
852 	 * If we're doing EEH probe based on PCI device, we
853 	 * would delay the probe until late stage because
854 	 * the PCI device isn't available this moment.
855 	 */
856 	if (!eeh_probe_mode_devtree())
857 		return;
858 
859 	if (!of_node_to_eeh_dev(dn))
860 		return;
861 	phb = of_node_to_eeh_dev(dn)->phb;
862 
863 	/* USB Bus children of PCI devices will not have BUID's */
864 	if (NULL == phb || 0 == phb->buid)
865 		return;
866 
867 	eeh_ops->of_probe(dn, NULL);
868 }
869 
870 /**
871  * eeh_add_device_tree_early - Enable EEH for the indicated device
872  * @dn: device node
873  *
874  * This routine must be used to perform EEH initialization for the
875  * indicated PCI device that was added after system boot (e.g.
876  * hotplug, dlpar).
877  */
878 void eeh_add_device_tree_early(struct device_node *dn)
879 {
880 	struct device_node *sib;
881 
882 	for_each_child_of_node(dn, sib)
883 		eeh_add_device_tree_early(sib);
884 	eeh_add_device_early(dn);
885 }
886 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
887 
888 /**
889  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
890  * @dev: pci device for which to set up EEH
891  *
892  * This routine must be used to complete EEH initialization for PCI
893  * devices that were added after system boot (e.g. hotplug, dlpar).
894  */
895 void eeh_add_device_late(struct pci_dev *dev)
896 {
897 	struct device_node *dn;
898 	struct eeh_dev *edev;
899 
900 	if (!dev || !eeh_subsystem_enabled)
901 		return;
902 
903 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
904 
905 	dn = pci_device_to_OF_node(dev);
906 	edev = of_node_to_eeh_dev(dn);
907 	if (edev->pdev == dev) {
908 		pr_debug("EEH: Already referenced !\n");
909 		return;
910 	}
911 
912 	/*
913 	 * The EEH cache might not be removed correctly because of
914 	 * unbalanced kref to the device during unplug time, which
915 	 * relies on pcibios_release_device(). So we have to remove
916 	 * that here explicitly.
917 	 */
918 	if (edev->pdev) {
919 		eeh_rmv_from_parent_pe(edev);
920 		eeh_addr_cache_rmv_dev(edev->pdev);
921 		eeh_sysfs_remove_device(edev->pdev);
922 		edev->mode &= ~EEH_DEV_SYSFS;
923 
924 		edev->pdev = NULL;
925 		dev->dev.archdata.edev = NULL;
926 	}
927 
928 	edev->pdev = dev;
929 	dev->dev.archdata.edev = edev;
930 
931 	/*
932 	 * We have to do the EEH probe here because the PCI device
933 	 * hasn't been created yet in the early stage.
934 	 */
935 	if (eeh_probe_mode_dev())
936 		eeh_ops->dev_probe(dev, NULL);
937 
938 	eeh_addr_cache_insert_dev(dev);
939 }
940 
941 /**
942  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
943  * @bus: PCI bus
944  *
945  * This routine must be used to perform EEH initialization for PCI
946  * devices which are attached to the indicated PCI bus. The PCI bus
947  * is added after system boot through hotplug or dlpar.
948  */
949 void eeh_add_device_tree_late(struct pci_bus *bus)
950 {
951 	struct pci_dev *dev;
952 
953 	list_for_each_entry(dev, &bus->devices, bus_list) {
954 		eeh_add_device_late(dev);
955 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
956 			struct pci_bus *subbus = dev->subordinate;
957 			if (subbus)
958 				eeh_add_device_tree_late(subbus);
959 		}
960 	}
961 }
962 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
963 
964 /**
965  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
966  * @bus: PCI bus
967  *
968  * This routine must be used to add EEH sysfs files for PCI
969  * devices which are attached to the indicated PCI bus. The PCI bus
970  * is added after system boot through hotplug or dlpar.
971  */
972 void eeh_add_sysfs_files(struct pci_bus *bus)
973 {
974 	struct pci_dev *dev;
975 
976 	list_for_each_entry(dev, &bus->devices, bus_list) {
977 		eeh_sysfs_add_device(dev);
978 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
979 			struct pci_bus *subbus = dev->subordinate;
980 			if (subbus)
981 				eeh_add_sysfs_files(subbus);
982 		}
983 	}
984 }
985 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
986 
987 /**
988  * eeh_remove_device - Undo EEH setup for the indicated pci device
989  * @dev: pci device to be removed
990  *
991  * This routine should be called when a device is removed from
992  * a running system (e.g. by hotplug or dlpar).  It unregisters
993  * the PCI device from the EEH subsystem.  I/O errors affecting
994  * this device will no longer be detected after this call; thus,
995  * i/o errors affecting this slot may leave this device unusable.
996  */
997 void eeh_remove_device(struct pci_dev *dev)
998 {
999 	struct eeh_dev *edev;
1000 
1001 	if (!dev || !eeh_subsystem_enabled)
1002 		return;
1003 	edev = pci_dev_to_eeh_dev(dev);
1004 
1005 	/* Unregister the device with the EEH/PCI address search system */
1006 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
1007 
1008 	if (!edev || !edev->pdev || !edev->pe) {
1009 		pr_debug("EEH: Not referenced !\n");
1010 		return;
1011 	}
1012 
1013 	/*
1014 	 * During the hotplug for EEH error recovery, we need the EEH
1015 	 * device attached to the parent PE in order for BAR restore
1016 	 * a bit later. So we keep it for BAR restore and remove it
1017 	 * from the parent PE during the BAR resotre.
1018 	 */
1019 	edev->pdev = NULL;
1020 	dev->dev.archdata.edev = NULL;
1021 	if (!(edev->pe->state & EEH_PE_KEEP))
1022 		eeh_rmv_from_parent_pe(edev);
1023 	else
1024 		edev->mode |= EEH_DEV_DISCONNECTED;
1025 
1026 	eeh_addr_cache_rmv_dev(dev);
1027 	eeh_sysfs_remove_device(dev);
1028 	edev->mode &= ~EEH_DEV_SYSFS;
1029 }
1030 
1031 static int proc_eeh_show(struct seq_file *m, void *v)
1032 {
1033 	if (0 == eeh_subsystem_enabled) {
1034 		seq_printf(m, "EEH Subsystem is globally disabled\n");
1035 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
1036 	} else {
1037 		seq_printf(m, "EEH Subsystem is enabled\n");
1038 		seq_printf(m,
1039 				"no device=%llu\n"
1040 				"no device node=%llu\n"
1041 				"no config address=%llu\n"
1042 				"check not wanted=%llu\n"
1043 				"eeh_total_mmio_ffs=%llu\n"
1044 				"eeh_false_positives=%llu\n"
1045 				"eeh_slot_resets=%llu\n",
1046 				eeh_stats.no_device,
1047 				eeh_stats.no_dn,
1048 				eeh_stats.no_cfg_addr,
1049 				eeh_stats.ignored_check,
1050 				eeh_stats.total_mmio_ffs,
1051 				eeh_stats.false_positives,
1052 				eeh_stats.slot_resets);
1053 	}
1054 
1055 	return 0;
1056 }
1057 
1058 static int proc_eeh_open(struct inode *inode, struct file *file)
1059 {
1060 	return single_open(file, proc_eeh_show, NULL);
1061 }
1062 
1063 static const struct file_operations proc_eeh_operations = {
1064 	.open      = proc_eeh_open,
1065 	.read      = seq_read,
1066 	.llseek    = seq_lseek,
1067 	.release   = single_release,
1068 };
1069 
1070 static int __init eeh_init_proc(void)
1071 {
1072 	if (machine_is(pseries) || machine_is(powernv))
1073 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
1074 	return 0;
1075 }
1076 __initcall(eeh_init_proc);
1077