xref: /openbmc/linux/arch/powerpc/kernel/eeh.c (revision c86085580d5f60d2d3cea9c60d50e284558d3de7)
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 /* Global EEH mutex */
107 DEFINE_MUTEX(eeh_mutex);
108 
109 /* Lock to avoid races due to multiple reports of an error */
110 static DEFINE_RAW_SPINLOCK(confirm_error_lock);
111 
112 /* Buffer for reporting pci register dumps. Its here in BSS, and
113  * not dynamically alloced, so that it ends up in RMO where RTAS
114  * can access it.
115  */
116 #define EEH_PCI_REGS_LOG_LEN 4096
117 static unsigned char pci_regs_buf[EEH_PCI_REGS_LOG_LEN];
118 
119 /*
120  * The struct is used to maintain the EEH global statistic
121  * information. Besides, the EEH global statistics will be
122  * exported to user space through procfs
123  */
124 struct eeh_stats {
125 	u64 no_device;		/* PCI device not found		*/
126 	u64 no_dn;		/* OF node not found		*/
127 	u64 no_cfg_addr;	/* Config address not found	*/
128 	u64 ignored_check;	/* EEH check skipped		*/
129 	u64 total_mmio_ffs;	/* Total EEH checks		*/
130 	u64 false_positives;	/* Unnecessary EEH checks	*/
131 	u64 slot_resets;	/* PE reset			*/
132 };
133 
134 static struct eeh_stats eeh_stats;
135 
136 #define IS_BRIDGE(class_code) (((class_code)<<16) == PCI_BASE_CLASS_BRIDGE)
137 
138 /**
139  * eeh_gather_pci_data - Copy assorted PCI config space registers to buff
140  * @edev: device to report data for
141  * @buf: point to buffer in which to log
142  * @len: amount of room in buffer
143  *
144  * This routine captures assorted PCI configuration space data,
145  * and puts them into a buffer for RTAS error logging.
146  */
147 static size_t eeh_gather_pci_data(struct eeh_dev *edev, char * buf, size_t len)
148 {
149 	struct device_node *dn = eeh_dev_to_of_node(edev);
150 	struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
151 	u32 cfg;
152 	int cap, i;
153 	int n = 0;
154 
155 	n += scnprintf(buf+n, len-n, "%s\n", dn->full_name);
156 	printk(KERN_WARNING "EEH: of node=%s\n", dn->full_name);
157 
158 	eeh_ops->read_config(dn, PCI_VENDOR_ID, 4, &cfg);
159 	n += scnprintf(buf+n, len-n, "dev/vend:%08x\n", cfg);
160 	printk(KERN_WARNING "EEH: PCI device/vendor: %08x\n", cfg);
161 
162 	eeh_ops->read_config(dn, PCI_COMMAND, 4, &cfg);
163 	n += scnprintf(buf+n, len-n, "cmd/stat:%x\n", cfg);
164 	printk(KERN_WARNING "EEH: PCI cmd/status register: %08x\n", cfg);
165 
166 	if (!dev) {
167 		printk(KERN_WARNING "EEH: no PCI device for this of node\n");
168 		return n;
169 	}
170 
171 	/* Gather bridge-specific registers */
172 	if (dev->class >> 16 == PCI_BASE_CLASS_BRIDGE) {
173 		eeh_ops->read_config(dn, PCI_SEC_STATUS, 2, &cfg);
174 		n += scnprintf(buf+n, len-n, "sec stat:%x\n", cfg);
175 		printk(KERN_WARNING "EEH: Bridge secondary status: %04x\n", cfg);
176 
177 		eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &cfg);
178 		n += scnprintf(buf+n, len-n, "brdg ctl:%x\n", cfg);
179 		printk(KERN_WARNING "EEH: Bridge control: %04x\n", cfg);
180 	}
181 
182 	/* Dump out the PCI-X command and status regs */
183 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
184 	if (cap) {
185 		eeh_ops->read_config(dn, cap, 4, &cfg);
186 		n += scnprintf(buf+n, len-n, "pcix-cmd:%x\n", cfg);
187 		printk(KERN_WARNING "EEH: PCI-X cmd: %08x\n", cfg);
188 
189 		eeh_ops->read_config(dn, cap+4, 4, &cfg);
190 		n += scnprintf(buf+n, len-n, "pcix-stat:%x\n", cfg);
191 		printk(KERN_WARNING "EEH: PCI-X status: %08x\n", cfg);
192 	}
193 
194 	/* If PCI-E capable, dump PCI-E cap 10, and the AER */
195 	cap = pci_find_capability(dev, PCI_CAP_ID_EXP);
196 	if (cap) {
197 		n += scnprintf(buf+n, len-n, "pci-e cap10:\n");
198 		printk(KERN_WARNING
199 		       "EEH: PCI-E capabilities and status follow:\n");
200 
201 		for (i=0; i<=8; i++) {
202 			eeh_ops->read_config(dn, cap+4*i, 4, &cfg);
203 			n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
204 			printk(KERN_WARNING "EEH: PCI-E %02x: %08x\n", i, cfg);
205 		}
206 
207 		cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
208 		if (cap) {
209 			n += scnprintf(buf+n, len-n, "pci-e AER:\n");
210 			printk(KERN_WARNING
211 			       "EEH: PCI-E AER capability register set follows:\n");
212 
213 			for (i=0; i<14; i++) {
214 				eeh_ops->read_config(dn, cap+4*i, 4, &cfg);
215 				n += scnprintf(buf+n, len-n, "%02x:%x\n", 4*i, cfg);
216 				printk(KERN_WARNING "EEH: PCI-E AER %02x: %08x\n", i, cfg);
217 			}
218 		}
219 	}
220 
221 	return n;
222 }
223 
224 /**
225  * eeh_slot_error_detail - Generate combined log including driver log and error log
226  * @pe: EEH PE
227  * @severity: temporary or permanent error log
228  *
229  * This routine should be called to generate the combined log, which
230  * is comprised of driver log and error log. The driver log is figured
231  * out from the config space of the corresponding PCI device, while
232  * the error log is fetched through platform dependent function call.
233  */
234 void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
235 {
236 	size_t loglen = 0;
237 	struct eeh_dev *edev;
238 
239 	eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
240 	eeh_ops->configure_bridge(pe);
241 	eeh_pe_restore_bars(pe);
242 
243 	pci_regs_buf[0] = 0;
244 	eeh_pe_for_each_dev(pe, edev) {
245 		loglen += eeh_gather_pci_data(edev, pci_regs_buf,
246 				EEH_PCI_REGS_LOG_LEN);
247         }
248 
249 	eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
250 }
251 
252 /**
253  * eeh_token_to_phys - Convert EEH address token to phys address
254  * @token: I/O token, should be address in the form 0xA....
255  *
256  * This routine should be called to convert virtual I/O address
257  * to physical one.
258  */
259 static inline unsigned long eeh_token_to_phys(unsigned long token)
260 {
261 	pte_t *ptep;
262 	unsigned long pa;
263 
264 	ptep = find_linux_pte(init_mm.pgd, token);
265 	if (!ptep)
266 		return token;
267 	pa = pte_pfn(*ptep) << PAGE_SHIFT;
268 
269 	return pa | (token & (PAGE_SIZE-1));
270 }
271 
272 /**
273  * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
274  * @edev: eeh device
275  *
276  * Check for an EEH failure for the given device node.  Call this
277  * routine if the result of a read was all 0xff's and you want to
278  * find out if this is due to an EEH slot freeze.  This routine
279  * will query firmware for the EEH status.
280  *
281  * Returns 0 if there has not been an EEH error; otherwise returns
282  * a non-zero value and queues up a slot isolation event notification.
283  *
284  * It is safe to call this routine in an interrupt context.
285  */
286 int eeh_dev_check_failure(struct eeh_dev *edev)
287 {
288 	int ret;
289 	unsigned long flags;
290 	struct device_node *dn;
291 	struct pci_dev *dev;
292 	struct eeh_pe *pe;
293 	int rc = 0;
294 	const char *location;
295 
296 	eeh_stats.total_mmio_ffs++;
297 
298 	if (!eeh_subsystem_enabled)
299 		return 0;
300 
301 	if (!edev) {
302 		eeh_stats.no_dn++;
303 		return 0;
304 	}
305 	dn = eeh_dev_to_of_node(edev);
306 	dev = eeh_dev_to_pci_dev(edev);
307 	pe = edev->pe;
308 
309 	/* Access to IO BARs might get this far and still not want checking. */
310 	if (!pe) {
311 		eeh_stats.ignored_check++;
312 		pr_debug("EEH: Ignored check for %s %s\n",
313 			eeh_pci_name(dev), dn->full_name);
314 		return 0;
315 	}
316 
317 	if (!pe->addr && !pe->config_addr) {
318 		eeh_stats.no_cfg_addr++;
319 		return 0;
320 	}
321 
322 	/* If we already have a pending isolation event for this
323 	 * slot, we know it's bad already, we don't need to check.
324 	 * Do this checking under a lock; as multiple PCI devices
325 	 * in one slot might report errors simultaneously, and we
326 	 * only want one error recovery routine running.
327 	 */
328 	raw_spin_lock_irqsave(&confirm_error_lock, flags);
329 	rc = 1;
330 	if (pe->state & EEH_PE_ISOLATED) {
331 		pe->check_count++;
332 		if (pe->check_count % EEH_MAX_FAILS == 0) {
333 			location = of_get_property(dn, "ibm,loc-code", NULL);
334 			printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
335 				"location=%s driver=%s pci addr=%s\n",
336 				pe->check_count, location,
337 				eeh_driver_name(dev), eeh_pci_name(dev));
338 			printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
339 				eeh_driver_name(dev));
340 			dump_stack();
341 		}
342 		goto dn_unlock;
343 	}
344 
345 	/*
346 	 * Now test for an EEH failure.  This is VERY expensive.
347 	 * Note that the eeh_config_addr may be a parent device
348 	 * in the case of a device behind a bridge, or it may be
349 	 * function zero of a multi-function device.
350 	 * In any case they must share a common PHB.
351 	 */
352 	ret = eeh_ops->get_state(pe, NULL);
353 
354 	/* Note that config-io to empty slots may fail;
355 	 * they are empty when they don't have children.
356 	 * We will punt with the following conditions: Failure to get
357 	 * PE's state, EEH not support and Permanently unavailable
358 	 * state, PE is in good state.
359 	 */
360 	if ((ret < 0) ||
361 	    (ret == EEH_STATE_NOT_SUPPORT) ||
362 	    (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
363 	    (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
364 		eeh_stats.false_positives++;
365 		pe->false_positives++;
366 		rc = 0;
367 		goto dn_unlock;
368 	}
369 
370 	eeh_stats.slot_resets++;
371 
372 	/* Avoid repeated reports of this failure, including problems
373 	 * with other functions on this device, and functions under
374 	 * bridges.
375 	 */
376 	eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
377 	raw_spin_unlock_irqrestore(&confirm_error_lock, flags);
378 
379 	eeh_send_failure_event(pe);
380 
381 	/* Most EEH events are due to device driver bugs.  Having
382 	 * a stack trace will help the device-driver authors figure
383 	 * out what happened.  So print that out.
384 	 */
385 	WARN(1, "EEH: failure detected\n");
386 	return 1;
387 
388 dn_unlock:
389 	raw_spin_unlock_irqrestore(&confirm_error_lock, flags);
390 	return rc;
391 }
392 
393 EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
394 
395 /**
396  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
397  * @token: I/O token, should be address in the form 0xA....
398  * @val: value, should be all 1's (XXX why do we need this arg??)
399  *
400  * Check for an EEH failure at the given token address.  Call this
401  * routine if the result of a read was all 0xff's and you want to
402  * find out if this is due to an EEH slot freeze event.  This routine
403  * will query firmware for the EEH status.
404  *
405  * Note this routine is safe to call in an interrupt context.
406  */
407 unsigned long eeh_check_failure(const volatile void __iomem *token, unsigned long val)
408 {
409 	unsigned long addr;
410 	struct eeh_dev *edev;
411 
412 	/* Finding the phys addr + pci device; this is pretty quick. */
413 	addr = eeh_token_to_phys((unsigned long __force) token);
414 	edev = eeh_addr_cache_get_dev(addr);
415 	if (!edev) {
416 		eeh_stats.no_device++;
417 		return val;
418 	}
419 
420 	eeh_dev_check_failure(edev);
421 
422 	pci_dev_put(eeh_dev_to_pci_dev(edev));
423 	return val;
424 }
425 
426 EXPORT_SYMBOL(eeh_check_failure);
427 
428 
429 /**
430  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
431  * @pe: EEH PE
432  *
433  * This routine should be called to reenable frozen MMIO or DMA
434  * so that it would work correctly again. It's useful while doing
435  * recovery or log collection on the indicated device.
436  */
437 int eeh_pci_enable(struct eeh_pe *pe, int function)
438 {
439 	int rc;
440 
441 	rc = eeh_ops->set_option(pe, function);
442 	if (rc)
443 		pr_warning("%s: Unexpected state change %d on PHB#%d-PE#%x, err=%d\n",
444 			__func__, function, pe->phb->global_number, pe->addr, rc);
445 
446 	rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
447 	if (rc > 0 && (rc & EEH_STATE_MMIO_ENABLED) &&
448 	   (function == EEH_OPT_THAW_MMIO))
449 		return 0;
450 
451 	return rc;
452 }
453 
454 /**
455  * pcibios_set_pcie_slot_reset - Set PCI-E reset state
456  * @dev: pci device struct
457  * @state: reset state to enter
458  *
459  * Return value:
460  * 	0 if success
461  */
462 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
463 {
464 	struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
465 	struct eeh_pe *pe = edev->pe;
466 
467 	if (!pe) {
468 		pr_err("%s: No PE found on PCI device %s\n",
469 			__func__, pci_name(dev));
470 		return -EINVAL;
471 	}
472 
473 	switch (state) {
474 	case pcie_deassert_reset:
475 		eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
476 		break;
477 	case pcie_hot_reset:
478 		eeh_ops->reset(pe, EEH_RESET_HOT);
479 		break;
480 	case pcie_warm_reset:
481 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
482 		break;
483 	default:
484 		return -EINVAL;
485 	};
486 
487 	return 0;
488 }
489 
490 /**
491  * eeh_set_pe_freset - Check the required reset for the indicated device
492  * @data: EEH device
493  * @flag: return value
494  *
495  * Each device might have its preferred reset type: fundamental or
496  * hot reset. The routine is used to collected the information for
497  * the indicated device and its children so that the bunch of the
498  * devices could be reset properly.
499  */
500 static void *eeh_set_dev_freset(void *data, void *flag)
501 {
502 	struct pci_dev *dev;
503 	unsigned int *freset = (unsigned int *)flag;
504 	struct eeh_dev *edev = (struct eeh_dev *)data;
505 
506 	dev = eeh_dev_to_pci_dev(edev);
507 	if (dev)
508 		*freset |= dev->needs_freset;
509 
510 	return NULL;
511 }
512 
513 /**
514  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
515  * @pe: EEH PE
516  *
517  * Assert the PCI #RST line for 1/4 second.
518  */
519 static void eeh_reset_pe_once(struct eeh_pe *pe)
520 {
521 	unsigned int freset = 0;
522 
523 	/* Determine type of EEH reset required for
524 	 * Partitionable Endpoint, a hot-reset (1)
525 	 * or a fundamental reset (3).
526 	 * A fundamental reset required by any device under
527 	 * Partitionable Endpoint trumps hot-reset.
528 	 */
529 	eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
530 
531 	if (freset)
532 		eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
533 	else
534 		eeh_ops->reset(pe, EEH_RESET_HOT);
535 
536 	/* The PCI bus requires that the reset be held high for at least
537 	 * a 100 milliseconds. We wait a bit longer 'just in case'.
538 	 */
539 #define PCI_BUS_RST_HOLD_TIME_MSEC 250
540 	msleep(PCI_BUS_RST_HOLD_TIME_MSEC);
541 
542 	/* We might get hit with another EEH freeze as soon as the
543 	 * pci slot reset line is dropped. Make sure we don't miss
544 	 * these, and clear the flag now.
545 	 */
546 	eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
547 
548 	eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
549 
550 	/* After a PCI slot has been reset, the PCI Express spec requires
551 	 * a 1.5 second idle time for the bus to stabilize, before starting
552 	 * up traffic.
553 	 */
554 #define PCI_BUS_SETTLE_TIME_MSEC 1800
555 	msleep(PCI_BUS_SETTLE_TIME_MSEC);
556 }
557 
558 /**
559  * eeh_reset_pe - Reset the indicated PE
560  * @pe: EEH PE
561  *
562  * This routine should be called to reset indicated device, including
563  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
564  * might be involved as well.
565  */
566 int eeh_reset_pe(struct eeh_pe *pe)
567 {
568 	int flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
569 	int i, rc;
570 
571 	/* Take three shots at resetting the bus */
572 	for (i=0; i<3; i++) {
573 		eeh_reset_pe_once(pe);
574 
575 		rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
576 		if ((rc & flags) == flags)
577 			return 0;
578 
579 		if (rc < 0) {
580 			pr_err("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
581 				__func__, pe->phb->global_number, pe->addr);
582 			return -1;
583 		}
584 		pr_err("EEH: bus reset %d failed on PHB#%d-PE#%x, rc=%d\n",
585 			i+1, pe->phb->global_number, pe->addr, rc);
586 	}
587 
588 	return -1;
589 }
590 
591 /**
592  * eeh_save_bars - Save device bars
593  * @edev: PCI device associated EEH device
594  *
595  * Save the values of the device bars. Unlike the restore
596  * routine, this routine is *not* recursive. This is because
597  * PCI devices are added individually; but, for the restore,
598  * an entire slot is reset at a time.
599  */
600 void eeh_save_bars(struct eeh_dev *edev)
601 {
602 	int i;
603 	struct device_node *dn;
604 
605 	if (!edev)
606 		return;
607 	dn = eeh_dev_to_of_node(edev);
608 
609 	for (i = 0; i < 16; i++)
610 		eeh_ops->read_config(dn, i * 4, 4, &edev->config_space[i]);
611 }
612 
613 /**
614  * eeh_ops_register - Register platform dependent EEH operations
615  * @ops: platform dependent EEH operations
616  *
617  * Register the platform dependent EEH operation callback
618  * functions. The platform should call this function before
619  * any other EEH operations.
620  */
621 int __init eeh_ops_register(struct eeh_ops *ops)
622 {
623 	if (!ops->name) {
624 		pr_warning("%s: Invalid EEH ops name for %p\n",
625 			__func__, ops);
626 		return -EINVAL;
627 	}
628 
629 	if (eeh_ops && eeh_ops != ops) {
630 		pr_warning("%s: EEH ops of platform %s already existing (%s)\n",
631 			__func__, eeh_ops->name, ops->name);
632 		return -EEXIST;
633 	}
634 
635 	eeh_ops = ops;
636 
637 	return 0;
638 }
639 
640 /**
641  * eeh_ops_unregister - Unreigster platform dependent EEH operations
642  * @name: name of EEH platform operations
643  *
644  * Unregister the platform dependent EEH operation callback
645  * functions.
646  */
647 int __exit eeh_ops_unregister(const char *name)
648 {
649 	if (!name || !strlen(name)) {
650 		pr_warning("%s: Invalid EEH ops name\n",
651 			__func__);
652 		return -EINVAL;
653 	}
654 
655 	if (eeh_ops && !strcmp(eeh_ops->name, name)) {
656 		eeh_ops = NULL;
657 		return 0;
658 	}
659 
660 	return -EEXIST;
661 }
662 
663 /**
664  * eeh_init - EEH initialization
665  *
666  * Initialize EEH by trying to enable it for all of the adapters in the system.
667  * As a side effect we can determine here if eeh is supported at all.
668  * Note that we leave EEH on so failed config cycles won't cause a machine
669  * check.  If a user turns off EEH for a particular adapter they are really
670  * telling Linux to ignore errors.  Some hardware (e.g. POWER5) won't
671  * grant access to a slot if EEH isn't enabled, and so we always enable
672  * EEH for all slots/all devices.
673  *
674  * The eeh-force-off option disables EEH checking globally, for all slots.
675  * Even if force-off is set, the EEH hardware is still enabled, so that
676  * newer systems can boot.
677  */
678 int __init eeh_init(void)
679 {
680 	struct pci_controller *hose, *tmp;
681 	struct device_node *phb;
682 	static int cnt = 0;
683 	int ret = 0;
684 
685 	/*
686 	 * We have to delay the initialization on PowerNV after
687 	 * the PCI hierarchy tree has been built because the PEs
688 	 * are figured out based on PCI devices instead of device
689 	 * tree nodes
690 	 */
691 	if (machine_is(powernv) && cnt++ <= 0)
692 		return ret;
693 
694 	/* call platform initialization function */
695 	if (!eeh_ops) {
696 		pr_warning("%s: Platform EEH operation not found\n",
697 			__func__);
698 		return -EEXIST;
699 	} else if ((ret = eeh_ops->init())) {
700 		pr_warning("%s: Failed to call platform init function (%d)\n",
701 			__func__, ret);
702 		return ret;
703 	}
704 
705 	raw_spin_lock_init(&confirm_error_lock);
706 
707 	/* Initialize EEH event */
708 	ret = eeh_event_init();
709 	if (ret)
710 		return ret;
711 
712 	/* Enable EEH for all adapters */
713 	if (eeh_probe_mode_devtree()) {
714 		list_for_each_entry_safe(hose, tmp,
715 			&hose_list, list_node) {
716 			phb = hose->dn;
717 			traverse_pci_devices(phb, eeh_ops->of_probe, NULL);
718 		}
719 	} else if (eeh_probe_mode_dev()) {
720 		list_for_each_entry_safe(hose, tmp,
721 			&hose_list, list_node)
722 			pci_walk_bus(hose->bus, eeh_ops->dev_probe, NULL);
723 	} else {
724 		pr_warning("%s: Invalid probe mode %d\n",
725 			   __func__, eeh_probe_mode);
726 		return -EINVAL;
727 	}
728 
729 	/*
730 	 * Call platform post-initialization. Actually, It's good chance
731 	 * to inform platform that EEH is ready to supply service if the
732 	 * I/O cache stuff has been built up.
733 	 */
734 	if (eeh_ops->post_init) {
735 		ret = eeh_ops->post_init();
736 		if (ret)
737 			return ret;
738 	}
739 
740 	if (eeh_subsystem_enabled)
741 		pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
742 	else
743 		pr_warning("EEH: No capable adapters found\n");
744 
745 	return ret;
746 }
747 
748 core_initcall_sync(eeh_init);
749 
750 /**
751  * eeh_add_device_early - Enable EEH for the indicated device_node
752  * @dn: device node for which to set up EEH
753  *
754  * This routine must be used to perform EEH initialization for PCI
755  * devices that were added after system boot (e.g. hotplug, dlpar).
756  * This routine must be called before any i/o is performed to the
757  * adapter (inluding any config-space i/o).
758  * Whether this actually enables EEH or not for this device depends
759  * on the CEC architecture, type of the device, on earlier boot
760  * command-line arguments & etc.
761  */
762 static void eeh_add_device_early(struct device_node *dn)
763 {
764 	struct pci_controller *phb;
765 
766 	/*
767 	 * If we're doing EEH probe based on PCI device, we
768 	 * would delay the probe until late stage because
769 	 * the PCI device isn't available this moment.
770 	 */
771 	if (!eeh_probe_mode_devtree())
772 		return;
773 
774 	if (!of_node_to_eeh_dev(dn))
775 		return;
776 	phb = of_node_to_eeh_dev(dn)->phb;
777 
778 	/* USB Bus children of PCI devices will not have BUID's */
779 	if (NULL == phb || 0 == phb->buid)
780 		return;
781 
782 	eeh_ops->of_probe(dn, NULL);
783 }
784 
785 /**
786  * eeh_add_device_tree_early - Enable EEH for the indicated device
787  * @dn: device node
788  *
789  * This routine must be used to perform EEH initialization for the
790  * indicated PCI device that was added after system boot (e.g.
791  * hotplug, dlpar).
792  */
793 void eeh_add_device_tree_early(struct device_node *dn)
794 {
795 	struct device_node *sib;
796 
797 	for_each_child_of_node(dn, sib)
798 		eeh_add_device_tree_early(sib);
799 	eeh_add_device_early(dn);
800 }
801 EXPORT_SYMBOL_GPL(eeh_add_device_tree_early);
802 
803 /**
804  * eeh_add_device_late - Perform EEH initialization for the indicated pci device
805  * @dev: pci device for which to set up EEH
806  *
807  * This routine must be used to complete EEH initialization for PCI
808  * devices that were added after system boot (e.g. hotplug, dlpar).
809  */
810 static void eeh_add_device_late(struct pci_dev *dev)
811 {
812 	struct device_node *dn;
813 	struct eeh_dev *edev;
814 
815 	if (!dev || !eeh_subsystem_enabled)
816 		return;
817 
818 	pr_debug("EEH: Adding device %s\n", pci_name(dev));
819 
820 	dn = pci_device_to_OF_node(dev);
821 	edev = of_node_to_eeh_dev(dn);
822 	if (edev->pdev == dev) {
823 		pr_debug("EEH: Already referenced !\n");
824 		return;
825 	}
826 	WARN_ON(edev->pdev);
827 
828 	pci_dev_get(dev);
829 	edev->pdev = dev;
830 	dev->dev.archdata.edev = edev;
831 
832 	/*
833 	 * We have to do the EEH probe here because the PCI device
834 	 * hasn't been created yet in the early stage.
835 	 */
836 	if (eeh_probe_mode_dev())
837 		eeh_ops->dev_probe(dev, NULL);
838 
839 	eeh_addr_cache_insert_dev(dev);
840 }
841 
842 /**
843  * eeh_add_device_tree_late - Perform EEH initialization for the indicated PCI bus
844  * @bus: PCI bus
845  *
846  * This routine must be used to perform EEH initialization for PCI
847  * devices which are attached to the indicated PCI bus. The PCI bus
848  * is added after system boot through hotplug or dlpar.
849  */
850 void eeh_add_device_tree_late(struct pci_bus *bus)
851 {
852 	struct pci_dev *dev;
853 
854 	list_for_each_entry(dev, &bus->devices, bus_list) {
855 		eeh_add_device_late(dev);
856 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
857 			struct pci_bus *subbus = dev->subordinate;
858 			if (subbus)
859 				eeh_add_device_tree_late(subbus);
860 		}
861 	}
862 }
863 EXPORT_SYMBOL_GPL(eeh_add_device_tree_late);
864 
865 /**
866  * eeh_add_sysfs_files - Add EEH sysfs files for the indicated PCI bus
867  * @bus: PCI bus
868  *
869  * This routine must be used to add EEH sysfs files for PCI
870  * devices which are attached to the indicated PCI bus. The PCI bus
871  * is added after system boot through hotplug or dlpar.
872  */
873 void eeh_add_sysfs_files(struct pci_bus *bus)
874 {
875 	struct pci_dev *dev;
876 
877 	list_for_each_entry(dev, &bus->devices, bus_list) {
878 		eeh_sysfs_add_device(dev);
879 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
880 			struct pci_bus *subbus = dev->subordinate;
881 			if (subbus)
882 				eeh_add_sysfs_files(subbus);
883 		}
884 	}
885 }
886 EXPORT_SYMBOL_GPL(eeh_add_sysfs_files);
887 
888 /**
889  * eeh_remove_device - Undo EEH setup for the indicated pci device
890  * @dev: pci device to be removed
891  * @purge_pe: remove the PE or not
892  *
893  * This routine should be called when a device is removed from
894  * a running system (e.g. by hotplug or dlpar).  It unregisters
895  * the PCI device from the EEH subsystem.  I/O errors affecting
896  * this device will no longer be detected after this call; thus,
897  * i/o errors affecting this slot may leave this device unusable.
898  */
899 static void eeh_remove_device(struct pci_dev *dev, int purge_pe)
900 {
901 	struct eeh_dev *edev;
902 
903 	if (!dev || !eeh_subsystem_enabled)
904 		return;
905 	edev = pci_dev_to_eeh_dev(dev);
906 
907 	/* Unregister the device with the EEH/PCI address search system */
908 	pr_debug("EEH: Removing device %s\n", pci_name(dev));
909 
910 	if (!edev || !edev->pdev) {
911 		pr_debug("EEH: Not referenced !\n");
912 		return;
913 	}
914 	edev->pdev = NULL;
915 	dev->dev.archdata.edev = NULL;
916 	pci_dev_put(dev);
917 
918 	eeh_rmv_from_parent_pe(edev, purge_pe);
919 	eeh_addr_cache_rmv_dev(dev);
920 	eeh_sysfs_remove_device(dev);
921 }
922 
923 /**
924  * eeh_remove_bus_device - Undo EEH setup for the indicated PCI device
925  * @dev: PCI device
926  * @purge_pe: remove the corresponding PE or not
927  *
928  * This routine must be called when a device is removed from the
929  * running system through hotplug or dlpar. The corresponding
930  * PCI address cache will be removed.
931  */
932 void eeh_remove_bus_device(struct pci_dev *dev, int purge_pe)
933 {
934 	struct pci_bus *bus = dev->subordinate;
935 	struct pci_dev *child, *tmp;
936 
937 	eeh_remove_device(dev, purge_pe);
938 
939 	if (bus && dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
940 		list_for_each_entry_safe(child, tmp, &bus->devices, bus_list)
941 			 eeh_remove_bus_device(child, purge_pe);
942 	}
943 }
944 EXPORT_SYMBOL_GPL(eeh_remove_bus_device);
945 
946 static int proc_eeh_show(struct seq_file *m, void *v)
947 {
948 	if (0 == eeh_subsystem_enabled) {
949 		seq_printf(m, "EEH Subsystem is globally disabled\n");
950 		seq_printf(m, "eeh_total_mmio_ffs=%llu\n", eeh_stats.total_mmio_ffs);
951 	} else {
952 		seq_printf(m, "EEH Subsystem is enabled\n");
953 		seq_printf(m,
954 				"no device=%llu\n"
955 				"no device node=%llu\n"
956 				"no config address=%llu\n"
957 				"check not wanted=%llu\n"
958 				"eeh_total_mmio_ffs=%llu\n"
959 				"eeh_false_positives=%llu\n"
960 				"eeh_slot_resets=%llu\n",
961 				eeh_stats.no_device,
962 				eeh_stats.no_dn,
963 				eeh_stats.no_cfg_addr,
964 				eeh_stats.ignored_check,
965 				eeh_stats.total_mmio_ffs,
966 				eeh_stats.false_positives,
967 				eeh_stats.slot_resets);
968 	}
969 
970 	return 0;
971 }
972 
973 static int proc_eeh_open(struct inode *inode, struct file *file)
974 {
975 	return single_open(file, proc_eeh_show, NULL);
976 }
977 
978 static const struct file_operations proc_eeh_operations = {
979 	.open      = proc_eeh_open,
980 	.read      = seq_read,
981 	.llseek    = seq_lseek,
982 	.release   = single_release,
983 };
984 
985 static int __init eeh_init_proc(void)
986 {
987 	if (machine_is(pseries))
988 		proc_create("powerpc/eeh", 0, NULL, &proc_eeh_operations);
989 	return 0;
990 }
991 __initcall(eeh_init_proc);
992