xref: /linux/arch/powerpc/platforms/powernv/eeh-powernv.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * The file intends to implement the platform dependent EEH operations on
3  * powernv platform. Actually, the powernv was created in order to fully
4  * hypervisor support.
5  *
6  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  */
13 
14 #include <linux/atomic.h>
15 #include <linux/debugfs.h>
16 #include <linux/delay.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/interrupt.h>
20 #include <linux/list.h>
21 #include <linux/msi.h>
22 #include <linux/of.h>
23 #include <linux/pci.h>
24 #include <linux/proc_fs.h>
25 #include <linux/rbtree.h>
26 #include <linux/sched.h>
27 #include <linux/seq_file.h>
28 #include <linux/spinlock.h>
29 
30 #include <asm/eeh.h>
31 #include <asm/eeh_event.h>
32 #include <asm/firmware.h>
33 #include <asm/io.h>
34 #include <asm/iommu.h>
35 #include <asm/machdep.h>
36 #include <asm/msi_bitmap.h>
37 #include <asm/opal.h>
38 #include <asm/ppc-pci.h>
39 
40 #include "powernv.h"
41 #include "pci.h"
42 
43 static bool pnv_eeh_nb_init = false;
44 static int eeh_event_irq = -EINVAL;
45 
46 /**
47  * pnv_eeh_init - EEH platform dependent initialization
48  *
49  * EEH platform dependent initialization on powernv
50  */
51 static int pnv_eeh_init(void)
52 {
53 	struct pci_controller *hose;
54 	struct pnv_phb *phb;
55 
56 	/* We require OPALv3 */
57 	if (!firmware_has_feature(FW_FEATURE_OPALv3)) {
58 		pr_warn("%s: OPALv3 is required !\n",
59 			__func__);
60 		return -EINVAL;
61 	}
62 
63 	/* Set probe mode */
64 	eeh_add_flag(EEH_PROBE_MODE_DEV);
65 
66 	/*
67 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
68 	 * doesn't do that. So we have to selectively enable I/O
69 	 * prior to collecting error log.
70 	 */
71 	list_for_each_entry(hose, &hose_list, list_node) {
72 		phb = hose->private_data;
73 
74 		if (phb->model == PNV_PHB_MODEL_P7IOC)
75 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
76 
77 		/*
78 		 * PE#0 should be regarded as valid by EEH core
79 		 * if it's not the reserved one. Currently, we
80 		 * have the reserved PE#0 and PE#127 for PHB3
81 		 * and P7IOC separately. So we should regard
82 		 * PE#0 as valid for P7IOC.
83 		 */
84 		if (phb->ioda.reserved_pe != 0)
85 			eeh_add_flag(EEH_VALID_PE_ZERO);
86 
87 		break;
88 	}
89 
90 	return 0;
91 }
92 
93 static irqreturn_t pnv_eeh_event(int irq, void *data)
94 {
95 	/*
96 	 * We simply send a special EEH event if EEH has been
97 	 * enabled. We don't care about EEH events until we've
98 	 * finished processing the outstanding ones. Event processing
99 	 * gets unmasked in next_error() if EEH is enabled.
100 	 */
101 	disable_irq_nosync(irq);
102 
103 	if (eeh_enabled())
104 		eeh_send_failure_event(NULL);
105 
106 	return IRQ_HANDLED;
107 }
108 
109 #ifdef CONFIG_DEBUG_FS
110 static ssize_t pnv_eeh_ei_write(struct file *filp,
111 				const char __user *user_buf,
112 				size_t count, loff_t *ppos)
113 {
114 	struct pci_controller *hose = filp->private_data;
115 	struct eeh_dev *edev;
116 	struct eeh_pe *pe;
117 	int pe_no, type, func;
118 	unsigned long addr, mask;
119 	char buf[50];
120 	int ret;
121 
122 	if (!eeh_ops || !eeh_ops->err_inject)
123 		return -ENXIO;
124 
125 	/* Copy over argument buffer */
126 	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
127 	if (!ret)
128 		return -EFAULT;
129 
130 	/* Retrieve parameters */
131 	ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
132 		     &pe_no, &type, &func, &addr, &mask);
133 	if (ret != 5)
134 		return -EINVAL;
135 
136 	/* Retrieve PE */
137 	edev = kzalloc(sizeof(*edev), GFP_KERNEL);
138 	if (!edev)
139 		return -ENOMEM;
140 	edev->phb = hose;
141 	edev->pe_config_addr = pe_no;
142 	pe = eeh_pe_get(edev);
143 	kfree(edev);
144 	if (!pe)
145 		return -ENODEV;
146 
147 	/* Do error injection */
148 	ret = eeh_ops->err_inject(pe, type, func, addr, mask);
149 	return ret < 0 ? ret : count;
150 }
151 
152 static const struct file_operations pnv_eeh_ei_fops = {
153 	.open	= simple_open,
154 	.llseek	= no_llseek,
155 	.write	= pnv_eeh_ei_write,
156 };
157 
158 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val)
159 {
160 	struct pci_controller *hose = data;
161 	struct pnv_phb *phb = hose->private_data;
162 
163 	out_be64(phb->regs + offset, val);
164 	return 0;
165 }
166 
167 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val)
168 {
169 	struct pci_controller *hose = data;
170 	struct pnv_phb *phb = hose->private_data;
171 
172 	*val = in_be64(phb->regs + offset);
173 	return 0;
174 }
175 
176 static int pnv_eeh_outb_dbgfs_set(void *data, u64 val)
177 {
178 	return pnv_eeh_dbgfs_set(data, 0xD10, val);
179 }
180 
181 static int pnv_eeh_outb_dbgfs_get(void *data, u64 *val)
182 {
183 	return pnv_eeh_dbgfs_get(data, 0xD10, val);
184 }
185 
186 static int pnv_eeh_inbA_dbgfs_set(void *data, u64 val)
187 {
188 	return pnv_eeh_dbgfs_set(data, 0xD90, val);
189 }
190 
191 static int pnv_eeh_inbA_dbgfs_get(void *data, u64 *val)
192 {
193 	return pnv_eeh_dbgfs_get(data, 0xD90, val);
194 }
195 
196 static int pnv_eeh_inbB_dbgfs_set(void *data, u64 val)
197 {
198 	return pnv_eeh_dbgfs_set(data, 0xE10, val);
199 }
200 
201 static int pnv_eeh_inbB_dbgfs_get(void *data, u64 *val)
202 {
203 	return pnv_eeh_dbgfs_get(data, 0xE10, val);
204 }
205 
206 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_outb_dbgfs_ops, pnv_eeh_outb_dbgfs_get,
207 			pnv_eeh_outb_dbgfs_set, "0x%llx\n");
208 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_inbA_dbgfs_ops, pnv_eeh_inbA_dbgfs_get,
209 			pnv_eeh_inbA_dbgfs_set, "0x%llx\n");
210 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_inbB_dbgfs_ops, pnv_eeh_inbB_dbgfs_get,
211 			pnv_eeh_inbB_dbgfs_set, "0x%llx\n");
212 #endif /* CONFIG_DEBUG_FS */
213 
214 /**
215  * pnv_eeh_post_init - EEH platform dependent post initialization
216  *
217  * EEH platform dependent post initialization on powernv. When
218  * the function is called, the EEH PEs and devices should have
219  * been built. If the I/O cache staff has been built, EEH is
220  * ready to supply service.
221  */
222 static int pnv_eeh_post_init(void)
223 {
224 	struct pci_controller *hose;
225 	struct pnv_phb *phb;
226 	int ret = 0;
227 
228 	/* Register OPAL event notifier */
229 	if (!pnv_eeh_nb_init) {
230 		eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
231 		if (eeh_event_irq < 0) {
232 			pr_err("%s: Can't register OPAL event interrupt (%d)\n",
233 			       __func__, eeh_event_irq);
234 			return eeh_event_irq;
235 		}
236 
237 		ret = request_irq(eeh_event_irq, pnv_eeh_event,
238 				IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
239 		if (ret < 0) {
240 			irq_dispose_mapping(eeh_event_irq);
241 			pr_err("%s: Can't request OPAL event interrupt (%d)\n",
242 			       __func__, eeh_event_irq);
243 			return ret;
244 		}
245 
246 		pnv_eeh_nb_init = true;
247 	}
248 
249 	if (!eeh_enabled())
250 		disable_irq(eeh_event_irq);
251 
252 	list_for_each_entry(hose, &hose_list, list_node) {
253 		phb = hose->private_data;
254 
255 		/*
256 		 * If EEH is enabled, we're going to rely on that.
257 		 * Otherwise, we restore to conventional mechanism
258 		 * to clear frozen PE during PCI config access.
259 		 */
260 		if (eeh_enabled())
261 			phb->flags |= PNV_PHB_FLAG_EEH;
262 		else
263 			phb->flags &= ~PNV_PHB_FLAG_EEH;
264 
265 		/* Create debugfs entries */
266 #ifdef CONFIG_DEBUG_FS
267 		if (phb->has_dbgfs || !phb->dbgfs)
268 			continue;
269 
270 		phb->has_dbgfs = 1;
271 		debugfs_create_file("err_injct", 0200,
272 				    phb->dbgfs, hose,
273 				    &pnv_eeh_ei_fops);
274 
275 		debugfs_create_file("err_injct_outbound", 0600,
276 				    phb->dbgfs, hose,
277 				    &pnv_eeh_outb_dbgfs_ops);
278 		debugfs_create_file("err_injct_inboundA", 0600,
279 				    phb->dbgfs, hose,
280 				    &pnv_eeh_inbA_dbgfs_ops);
281 		debugfs_create_file("err_injct_inboundB", 0600,
282 				    phb->dbgfs, hose,
283 				    &pnv_eeh_inbB_dbgfs_ops);
284 #endif /* CONFIG_DEBUG_FS */
285 	}
286 
287 
288 	return ret;
289 }
290 
291 static int pnv_eeh_cap_start(struct pci_dn *pdn)
292 {
293 	u32 status;
294 
295 	if (!pdn)
296 		return 0;
297 
298 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
299 	if (!(status & PCI_STATUS_CAP_LIST))
300 		return 0;
301 
302 	return PCI_CAPABILITY_LIST;
303 }
304 
305 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
306 {
307 	int pos = pnv_eeh_cap_start(pdn);
308 	int cnt = 48;   /* Maximal number of capabilities */
309 	u32 id;
310 
311 	if (!pos)
312 		return 0;
313 
314 	while (cnt--) {
315 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
316 		if (pos < 0x40)
317 			break;
318 
319 		pos &= ~3;
320 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
321 		if (id == 0xff)
322 			break;
323 
324 		/* Found */
325 		if (id == cap)
326 			return pos;
327 
328 		/* Next one */
329 		pos += PCI_CAP_LIST_NEXT;
330 	}
331 
332 	return 0;
333 }
334 
335 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
336 {
337 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
338 	u32 header;
339 	int pos = 256, ttl = (4096 - 256) / 8;
340 
341 	if (!edev || !edev->pcie_cap)
342 		return 0;
343 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
344 		return 0;
345 	else if (!header)
346 		return 0;
347 
348 	while (ttl-- > 0) {
349 		if (PCI_EXT_CAP_ID(header) == cap && pos)
350 			return pos;
351 
352 		pos = PCI_EXT_CAP_NEXT(header);
353 		if (pos < 256)
354 			break;
355 
356 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
357 			break;
358 	}
359 
360 	return 0;
361 }
362 
363 /**
364  * pnv_eeh_probe - Do probe on PCI device
365  * @pdn: PCI device node
366  * @data: unused
367  *
368  * When EEH module is installed during system boot, all PCI devices
369  * are checked one by one to see if it supports EEH. The function
370  * is introduced for the purpose. By default, EEH has been enabled
371  * on all PCI devices. That's to say, we only need do necessary
372  * initialization on the corresponding eeh device and create PE
373  * accordingly.
374  *
375  * It's notable that's unsafe to retrieve the EEH device through
376  * the corresponding PCI device. During the PCI device hotplug, which
377  * was possiblly triggered by EEH core, the binding between EEH device
378  * and the PCI device isn't built yet.
379  */
380 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
381 {
382 	struct pci_controller *hose = pdn->phb;
383 	struct pnv_phb *phb = hose->private_data;
384 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
385 	uint32_t pcie_flags;
386 	int ret;
387 
388 	/*
389 	 * When probing the root bridge, which doesn't have any
390 	 * subordinate PCI devices. We don't have OF node for
391 	 * the root bridge. So it's not reasonable to continue
392 	 * the probing.
393 	 */
394 	if (!edev || edev->pe)
395 		return NULL;
396 
397 	/* Skip for PCI-ISA bridge */
398 	if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
399 		return NULL;
400 
401 	/* Initialize eeh device */
402 	edev->class_code = pdn->class_code;
403 	edev->mode	&= 0xFFFFFF00;
404 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
405 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
406 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
407 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
408 		edev->mode |= EEH_DEV_BRIDGE;
409 		if (edev->pcie_cap) {
410 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
411 					 2, &pcie_flags);
412 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
413 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
414 				edev->mode |= EEH_DEV_ROOT_PORT;
415 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
416 				edev->mode |= EEH_DEV_DS_PORT;
417 		}
418 	}
419 
420 	edev->config_addr    = (pdn->busno << 8) | (pdn->devfn);
421 	edev->pe_config_addr = phb->ioda.pe_rmap[edev->config_addr];
422 
423 	/* Create PE */
424 	ret = eeh_add_to_parent_pe(edev);
425 	if (ret) {
426 		pr_warn("%s: Can't add PCI dev %04x:%02x:%02x.%01x to parent PE (%d)\n",
427 			__func__, hose->global_number, pdn->busno,
428 			PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn), ret);
429 		return NULL;
430 	}
431 
432 	/*
433 	 * If the PE contains any one of following adapters, the
434 	 * PCI config space can't be accessed when dumping EEH log.
435 	 * Otherwise, we will run into fenced PHB caused by shortage
436 	 * of outbound credits in the adapter. The PCI config access
437 	 * should be blocked until PE reset. MMIO access is dropped
438 	 * by hardware certainly. In order to drop PCI config requests,
439 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
440 	 * will be checked in the backend for PE state retrival. If
441 	 * the PE becomes frozen for the first time and the flag has
442 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
443 	 * that PE to block its config space.
444 	 *
445 	 * Broadcom Austin 4-ports NICs (14e4:1657)
446 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
447 	 */
448 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
449 	     pdn->device_id == 0x1657) ||
450 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
451 	     pdn->device_id == 0x168e))
452 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
453 
454 	/*
455 	 * Cache the PE primary bus, which can't be fetched when
456 	 * full hotplug is in progress. In that case, all child
457 	 * PCI devices of the PE are expected to be removed prior
458 	 * to PE reset.
459 	 */
460 	if (!edev->pe->bus)
461 		edev->pe->bus = pci_find_bus(hose->global_number,
462 					     pdn->busno);
463 
464 	/*
465 	 * Enable EEH explicitly so that we will do EEH check
466 	 * while accessing I/O stuff
467 	 */
468 	eeh_add_flag(EEH_ENABLED);
469 
470 	/* Save memory bars */
471 	eeh_save_bars(edev);
472 
473 	return NULL;
474 }
475 
476 /**
477  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
478  * @pe: EEH PE
479  * @option: operation to be issued
480  *
481  * The function is used to control the EEH functionality globally.
482  * Currently, following options are support according to PAPR:
483  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
484  */
485 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
486 {
487 	struct pci_controller *hose = pe->phb;
488 	struct pnv_phb *phb = hose->private_data;
489 	bool freeze_pe = false;
490 	int opt, ret = 0;
491 	s64 rc;
492 
493 	/* Sanity check on option */
494 	switch (option) {
495 	case EEH_OPT_DISABLE:
496 		return -EPERM;
497 	case EEH_OPT_ENABLE:
498 		return 0;
499 	case EEH_OPT_THAW_MMIO:
500 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
501 		break;
502 	case EEH_OPT_THAW_DMA:
503 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
504 		break;
505 	case EEH_OPT_FREEZE_PE:
506 		freeze_pe = true;
507 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
508 		break;
509 	default:
510 		pr_warn("%s: Invalid option %d\n", __func__, option);
511 		return -EINVAL;
512 	}
513 
514 	/* If PHB supports compound PE, to handle it */
515 	if (freeze_pe) {
516 		if (phb->freeze_pe) {
517 			phb->freeze_pe(phb, pe->addr);
518 		} else {
519 			rc = opal_pci_eeh_freeze_set(phb->opal_id,
520 						     pe->addr, opt);
521 			if (rc != OPAL_SUCCESS) {
522 				pr_warn("%s: Failure %lld freezing "
523 					"PHB#%x-PE#%x\n",
524 					__func__, rc,
525 					phb->hose->global_number, pe->addr);
526 				ret = -EIO;
527 			}
528 		}
529 	} else {
530 		if (phb->unfreeze_pe) {
531 			ret = phb->unfreeze_pe(phb, pe->addr, opt);
532 		} else {
533 			rc = opal_pci_eeh_freeze_clear(phb->opal_id,
534 						       pe->addr, opt);
535 			if (rc != OPAL_SUCCESS) {
536 				pr_warn("%s: Failure %lld enable %d "
537 					"for PHB#%x-PE#%x\n",
538 					__func__, rc, option,
539 					phb->hose->global_number, pe->addr);
540 				ret = -EIO;
541 			}
542 		}
543 	}
544 
545 	return ret;
546 }
547 
548 /**
549  * pnv_eeh_get_pe_addr - Retrieve PE address
550  * @pe: EEH PE
551  *
552  * Retrieve the PE address according to the given tranditional
553  * PCI BDF (Bus/Device/Function) address.
554  */
555 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe)
556 {
557 	return pe->addr;
558 }
559 
560 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
561 {
562 	struct pnv_phb *phb = pe->phb->private_data;
563 	s64 rc;
564 
565 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
566 					 PNV_PCI_DIAG_BUF_SIZE);
567 	if (rc != OPAL_SUCCESS)
568 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
569 			__func__, rc, pe->phb->global_number);
570 }
571 
572 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
573 {
574 	struct pnv_phb *phb = pe->phb->private_data;
575 	u8 fstate;
576 	__be16 pcierr;
577 	s64 rc;
578 	int result = 0;
579 
580 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
581 					pe->addr,
582 					&fstate,
583 					&pcierr,
584 					NULL);
585 	if (rc != OPAL_SUCCESS) {
586 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
587 			__func__, rc, phb->hose->global_number);
588 		return EEH_STATE_NOT_SUPPORT;
589 	}
590 
591 	/*
592 	 * Check PHB state. If the PHB is frozen for the
593 	 * first time, to dump the PHB diag-data.
594 	 */
595 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
596 		result = (EEH_STATE_MMIO_ACTIVE  |
597 			  EEH_STATE_DMA_ACTIVE   |
598 			  EEH_STATE_MMIO_ENABLED |
599 			  EEH_STATE_DMA_ENABLED);
600 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
601 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
602 		pnv_eeh_get_phb_diag(pe);
603 
604 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
605 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
606 	}
607 
608 	return result;
609 }
610 
611 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
612 {
613 	struct pnv_phb *phb = pe->phb->private_data;
614 	u8 fstate;
615 	__be16 pcierr;
616 	s64 rc;
617 	int result;
618 
619 	/*
620 	 * We don't clobber hardware frozen state until PE
621 	 * reset is completed. In order to keep EEH core
622 	 * moving forward, we have to return operational
623 	 * state during PE reset.
624 	 */
625 	if (pe->state & EEH_PE_RESET) {
626 		result = (EEH_STATE_MMIO_ACTIVE  |
627 			  EEH_STATE_DMA_ACTIVE   |
628 			  EEH_STATE_MMIO_ENABLED |
629 			  EEH_STATE_DMA_ENABLED);
630 		return result;
631 	}
632 
633 	/*
634 	 * Fetch PE state from hardware. If the PHB
635 	 * supports compound PE, let it handle that.
636 	 */
637 	if (phb->get_pe_state) {
638 		fstate = phb->get_pe_state(phb, pe->addr);
639 	} else {
640 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
641 						pe->addr,
642 						&fstate,
643 						&pcierr,
644 						NULL);
645 		if (rc != OPAL_SUCCESS) {
646 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
647 				__func__, rc, phb->hose->global_number,
648 				pe->addr);
649 			return EEH_STATE_NOT_SUPPORT;
650 		}
651 	}
652 
653 	/* Figure out state */
654 	switch (fstate) {
655 	case OPAL_EEH_STOPPED_NOT_FROZEN:
656 		result = (EEH_STATE_MMIO_ACTIVE  |
657 			  EEH_STATE_DMA_ACTIVE   |
658 			  EEH_STATE_MMIO_ENABLED |
659 			  EEH_STATE_DMA_ENABLED);
660 		break;
661 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
662 		result = (EEH_STATE_DMA_ACTIVE |
663 			  EEH_STATE_DMA_ENABLED);
664 		break;
665 	case OPAL_EEH_STOPPED_DMA_FREEZE:
666 		result = (EEH_STATE_MMIO_ACTIVE |
667 			  EEH_STATE_MMIO_ENABLED);
668 		break;
669 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
670 		result = 0;
671 		break;
672 	case OPAL_EEH_STOPPED_RESET:
673 		result = EEH_STATE_RESET_ACTIVE;
674 		break;
675 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
676 		result = EEH_STATE_UNAVAILABLE;
677 		break;
678 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
679 		result = EEH_STATE_NOT_SUPPORT;
680 		break;
681 	default:
682 		result = EEH_STATE_NOT_SUPPORT;
683 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
684 			__func__, phb->hose->global_number,
685 			pe->addr, fstate);
686 	}
687 
688 	/*
689 	 * If PHB supports compound PE, to freeze all
690 	 * slave PEs for consistency.
691 	 *
692 	 * If the PE is switching to frozen state for the
693 	 * first time, to dump the PHB diag-data.
694 	 */
695 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
696 	    !(result & EEH_STATE_UNAVAILABLE) &&
697 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
698 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
699 	    !(pe->state & EEH_PE_ISOLATED)) {
700 		if (phb->freeze_pe)
701 			phb->freeze_pe(phb, pe->addr);
702 
703 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
704 		pnv_eeh_get_phb_diag(pe);
705 
706 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
707 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
708 	}
709 
710 	return result;
711 }
712 
713 /**
714  * pnv_eeh_get_state - Retrieve PE state
715  * @pe: EEH PE
716  * @delay: delay while PE state is temporarily unavailable
717  *
718  * Retrieve the state of the specified PE. For IODA-compitable
719  * platform, it should be retrieved from IODA table. Therefore,
720  * we prefer passing down to hardware implementation to handle
721  * it.
722  */
723 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
724 {
725 	int ret;
726 
727 	if (pe->type & EEH_PE_PHB)
728 		ret = pnv_eeh_get_phb_state(pe);
729 	else
730 		ret = pnv_eeh_get_pe_state(pe);
731 
732 	if (!delay)
733 		return ret;
734 
735 	/*
736 	 * If the PE state is temporarily unavailable,
737 	 * to inform the EEH core delay for default
738 	 * period (1 second)
739 	 */
740 	*delay = 0;
741 	if (ret & EEH_STATE_UNAVAILABLE)
742 		*delay = 1000;
743 
744 	return ret;
745 }
746 
747 static s64 pnv_eeh_phb_poll(struct pnv_phb *phb)
748 {
749 	s64 rc = OPAL_HARDWARE;
750 
751 	while (1) {
752 		rc = opal_pci_poll(phb->opal_id);
753 		if (rc <= 0)
754 			break;
755 
756 		if (system_state < SYSTEM_RUNNING)
757 			udelay(1000 * rc);
758 		else
759 			msleep(rc);
760 	}
761 
762 	return rc;
763 }
764 
765 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
766 {
767 	struct pnv_phb *phb = hose->private_data;
768 	s64 rc = OPAL_HARDWARE;
769 
770 	pr_debug("%s: Reset PHB#%x, option=%d\n",
771 		 __func__, hose->global_number, option);
772 
773 	/* Issue PHB complete reset request */
774 	if (option == EEH_RESET_FUNDAMENTAL ||
775 	    option == EEH_RESET_HOT)
776 		rc = opal_pci_reset(phb->opal_id,
777 				    OPAL_RESET_PHB_COMPLETE,
778 				    OPAL_ASSERT_RESET);
779 	else if (option == EEH_RESET_DEACTIVATE)
780 		rc = opal_pci_reset(phb->opal_id,
781 				    OPAL_RESET_PHB_COMPLETE,
782 				    OPAL_DEASSERT_RESET);
783 	if (rc < 0)
784 		goto out;
785 
786 	/*
787 	 * Poll state of the PHB until the request is done
788 	 * successfully. The PHB reset is usually PHB complete
789 	 * reset followed by hot reset on root bus. So we also
790 	 * need the PCI bus settlement delay.
791 	 */
792 	rc = pnv_eeh_phb_poll(phb);
793 	if (option == EEH_RESET_DEACTIVATE) {
794 		if (system_state < SYSTEM_RUNNING)
795 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
796 		else
797 			msleep(EEH_PE_RST_SETTLE_TIME);
798 	}
799 out:
800 	if (rc != OPAL_SUCCESS)
801 		return -EIO;
802 
803 	return 0;
804 }
805 
806 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
807 {
808 	struct pnv_phb *phb = hose->private_data;
809 	s64 rc = OPAL_HARDWARE;
810 
811 	pr_debug("%s: Reset PHB#%x, option=%d\n",
812 		 __func__, hose->global_number, option);
813 
814 	/*
815 	 * During the reset deassert time, we needn't care
816 	 * the reset scope because the firmware does nothing
817 	 * for fundamental or hot reset during deassert phase.
818 	 */
819 	if (option == EEH_RESET_FUNDAMENTAL)
820 		rc = opal_pci_reset(phb->opal_id,
821 				    OPAL_RESET_PCI_FUNDAMENTAL,
822 				    OPAL_ASSERT_RESET);
823 	else if (option == EEH_RESET_HOT)
824 		rc = opal_pci_reset(phb->opal_id,
825 				    OPAL_RESET_PCI_HOT,
826 				    OPAL_ASSERT_RESET);
827 	else if (option == EEH_RESET_DEACTIVATE)
828 		rc = opal_pci_reset(phb->opal_id,
829 				    OPAL_RESET_PCI_HOT,
830 				    OPAL_DEASSERT_RESET);
831 	if (rc < 0)
832 		goto out;
833 
834 	/* Poll state of the PHB until the request is done */
835 	rc = pnv_eeh_phb_poll(phb);
836 	if (option == EEH_RESET_DEACTIVATE)
837 		msleep(EEH_PE_RST_SETTLE_TIME);
838 out:
839 	if (rc != OPAL_SUCCESS)
840 		return -EIO;
841 
842 	return 0;
843 }
844 
845 static int pnv_eeh_bridge_reset(struct pci_dev *dev, int option)
846 {
847 	struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
848 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
849 	int aer = edev ? edev->aer_cap : 0;
850 	u32 ctrl;
851 
852 	pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
853 		 __func__, pci_domain_nr(dev->bus),
854 		 dev->bus->number, option);
855 
856 	switch (option) {
857 	case EEH_RESET_FUNDAMENTAL:
858 	case EEH_RESET_HOT:
859 		/* Don't report linkDown event */
860 		if (aer) {
861 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
862 					     4, &ctrl);
863 			ctrl |= PCI_ERR_UNC_SURPDN;
864 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
865 					      4, ctrl);
866 		}
867 
868 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
869 		ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
870 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
871 
872 		msleep(EEH_PE_RST_HOLD_TIME);
873 		break;
874 	case EEH_RESET_DEACTIVATE:
875 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
876 		ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
877 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
878 
879 		msleep(EEH_PE_RST_SETTLE_TIME);
880 
881 		/* Continue reporting linkDown event */
882 		if (aer) {
883 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
884 					     4, &ctrl);
885 			ctrl &= ~PCI_ERR_UNC_SURPDN;
886 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
887 					      4, ctrl);
888 		}
889 
890 		break;
891 	}
892 
893 	return 0;
894 }
895 
896 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
897 {
898 	struct pci_controller *hose;
899 
900 	if (pci_is_root_bus(dev->bus)) {
901 		hose = pci_bus_to_host(dev->bus);
902 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
903 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
904 	} else {
905 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
906 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
907 	}
908 }
909 
910 /**
911  * pnv_eeh_reset - Reset the specified PE
912  * @pe: EEH PE
913  * @option: reset option
914  *
915  * Do reset on the indicated PE. For PCI bus sensitive PE,
916  * we need to reset the parent p2p bridge. The PHB has to
917  * be reinitialized if the p2p bridge is root bridge. For
918  * PCI device sensitive PE, we will try to reset the device
919  * through FLR. For now, we don't have OPAL APIs to do HARD
920  * reset yet, so all reset would be SOFT (HOT) reset.
921  */
922 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
923 {
924 	struct pci_controller *hose = pe->phb;
925 	struct pci_bus *bus;
926 	int ret;
927 
928 	/*
929 	 * For PHB reset, we always have complete reset. For those PEs whose
930 	 * primary bus derived from root complex (root bus) or root port
931 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
932 	 * For other PEs, we always have hot reset on the PE primary bus.
933 	 *
934 	 * Here, we have different design to pHyp, which always clear the
935 	 * frozen state during PE reset. However, the good idea here from
936 	 * benh is to keep frozen state before we get PE reset done completely
937 	 * (until BAR restore). With the frozen state, HW drops illegal IO
938 	 * or MMIO access, which can incur recrusive frozen PE during PE
939 	 * reset. The side effect is that EEH core has to clear the frozen
940 	 * state explicitly after BAR restore.
941 	 */
942 	if (pe->type & EEH_PE_PHB) {
943 		ret = pnv_eeh_phb_reset(hose, option);
944 	} else {
945 		struct pnv_phb *phb;
946 		s64 rc;
947 
948 		/*
949 		 * The frozen PE might be caused by PAPR error injection
950 		 * registers, which are expected to be cleared after hitting
951 		 * frozen PE as stated in the hardware spec. Unfortunately,
952 		 * that's not true on P7IOC. So we have to clear it manually
953 		 * to avoid recursive EEH errors during recovery.
954 		 */
955 		phb = hose->private_data;
956 		if (phb->model == PNV_PHB_MODEL_P7IOC &&
957 		    (option == EEH_RESET_HOT ||
958 		    option == EEH_RESET_FUNDAMENTAL)) {
959 			rc = opal_pci_reset(phb->opal_id,
960 					    OPAL_RESET_PHB_ERROR,
961 					    OPAL_ASSERT_RESET);
962 			if (rc != OPAL_SUCCESS) {
963 				pr_warn("%s: Failure %lld clearing "
964 					"error injection registers\n",
965 					__func__, rc);
966 				return -EIO;
967 			}
968 		}
969 
970 		bus = eeh_pe_bus_get(pe);
971 		if (pci_is_root_bus(bus) ||
972 			pci_is_root_bus(bus->parent))
973 			ret = pnv_eeh_root_reset(hose, option);
974 		else
975 			ret = pnv_eeh_bridge_reset(bus->self, option);
976 	}
977 
978 	return ret;
979 }
980 
981 /**
982  * pnv_eeh_wait_state - Wait for PE state
983  * @pe: EEH PE
984  * @max_wait: maximal period in millisecond
985  *
986  * Wait for the state of associated PE. It might take some time
987  * to retrieve the PE's state.
988  */
989 static int pnv_eeh_wait_state(struct eeh_pe *pe, int max_wait)
990 {
991 	int ret;
992 	int mwait;
993 
994 	while (1) {
995 		ret = pnv_eeh_get_state(pe, &mwait);
996 
997 		/*
998 		 * If the PE's state is temporarily unavailable,
999 		 * we have to wait for the specified time. Otherwise,
1000 		 * the PE's state will be returned immediately.
1001 		 */
1002 		if (ret != EEH_STATE_UNAVAILABLE)
1003 			return ret;
1004 
1005 		if (max_wait <= 0) {
1006 			pr_warn("%s: Timeout getting PE#%x's state (%d)\n",
1007 				__func__, pe->addr, max_wait);
1008 			return EEH_STATE_NOT_SUPPORT;
1009 		}
1010 
1011 		max_wait -= mwait;
1012 		msleep(mwait);
1013 	}
1014 
1015 	return EEH_STATE_NOT_SUPPORT;
1016 }
1017 
1018 /**
1019  * pnv_eeh_get_log - Retrieve error log
1020  * @pe: EEH PE
1021  * @severity: temporary or permanent error log
1022  * @drv_log: driver log to be combined with retrieved error log
1023  * @len: length of driver log
1024  *
1025  * Retrieve the temporary or permanent error from the PE.
1026  */
1027 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1028 			   char *drv_log, unsigned long len)
1029 {
1030 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1031 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1032 
1033 	return 0;
1034 }
1035 
1036 /**
1037  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1038  * @pe: EEH PE
1039  *
1040  * The function will be called to reconfigure the bridges included
1041  * in the specified PE so that the mulfunctional PE would be recovered
1042  * again.
1043  */
1044 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1045 {
1046 	return 0;
1047 }
1048 
1049 /**
1050  * pnv_pe_err_inject - Inject specified error to the indicated PE
1051  * @pe: the indicated PE
1052  * @type: error type
1053  * @func: specific error type
1054  * @addr: address
1055  * @mask: address mask
1056  *
1057  * The routine is called to inject specified error, which is
1058  * determined by @type and @func, to the indicated PE for
1059  * testing purpose.
1060  */
1061 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1062 			      unsigned long addr, unsigned long mask)
1063 {
1064 	struct pci_controller *hose = pe->phb;
1065 	struct pnv_phb *phb = hose->private_data;
1066 	s64 rc;
1067 
1068 	/* Sanity check on error type */
1069 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1070 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1071 		pr_warn("%s: Invalid error type %d\n",
1072 			__func__, type);
1073 		return -ERANGE;
1074 	}
1075 
1076 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1077 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1078 		pr_warn("%s: Invalid error function %d\n",
1079 			__func__, func);
1080 		return -ERANGE;
1081 	}
1082 
1083 	/* Firmware supports error injection ? */
1084 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1085 		pr_warn("%s: Firmware doesn't support error injection\n",
1086 			__func__);
1087 		return -ENXIO;
1088 	}
1089 
1090 	/* Do error injection */
1091 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1092 				 type, func, addr, mask);
1093 	if (rc != OPAL_SUCCESS) {
1094 		pr_warn("%s: Failure %lld injecting error "
1095 			"%d-%d to PHB#%x-PE#%x\n",
1096 			__func__, rc, type, func,
1097 			hose->global_number, pe->addr);
1098 		return -EIO;
1099 	}
1100 
1101 	return 0;
1102 }
1103 
1104 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1105 {
1106 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1107 
1108 	if (!edev || !edev->pe)
1109 		return false;
1110 
1111 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1112 		return true;
1113 
1114 	return false;
1115 }
1116 
1117 static int pnv_eeh_read_config(struct pci_dn *pdn,
1118 			       int where, int size, u32 *val)
1119 {
1120 	if (!pdn)
1121 		return PCIBIOS_DEVICE_NOT_FOUND;
1122 
1123 	if (pnv_eeh_cfg_blocked(pdn)) {
1124 		*val = 0xFFFFFFFF;
1125 		return PCIBIOS_SET_FAILED;
1126 	}
1127 
1128 	return pnv_pci_cfg_read(pdn, where, size, val);
1129 }
1130 
1131 static int pnv_eeh_write_config(struct pci_dn *pdn,
1132 				int where, int size, u32 val)
1133 {
1134 	if (!pdn)
1135 		return PCIBIOS_DEVICE_NOT_FOUND;
1136 
1137 	if (pnv_eeh_cfg_blocked(pdn))
1138 		return PCIBIOS_SET_FAILED;
1139 
1140 	return pnv_pci_cfg_write(pdn, where, size, val);
1141 }
1142 
1143 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1144 {
1145 	/* GEM */
1146 	if (data->gemXfir || data->gemRfir ||
1147 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1148 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1149 			be64_to_cpu(data->gemXfir),
1150 			be64_to_cpu(data->gemRfir),
1151 			be64_to_cpu(data->gemRirqfir),
1152 			be64_to_cpu(data->gemMask),
1153 			be64_to_cpu(data->gemRwof));
1154 
1155 	/* LEM */
1156 	if (data->lemFir || data->lemErrMask ||
1157 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1158 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1159 			be64_to_cpu(data->lemFir),
1160 			be64_to_cpu(data->lemErrMask),
1161 			be64_to_cpu(data->lemAction0),
1162 			be64_to_cpu(data->lemAction1),
1163 			be64_to_cpu(data->lemWof));
1164 }
1165 
1166 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1167 {
1168 	struct pnv_phb *phb = hose->private_data;
1169 	struct OpalIoP7IOCErrorData *data = &phb->diag.hub_diag;
1170 	long rc;
1171 
1172 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1173 	if (rc != OPAL_SUCCESS) {
1174 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1175 			__func__, phb->hub_id, rc);
1176 		return;
1177 	}
1178 
1179 	switch (data->type) {
1180 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1181 		pr_info("P7IOC diag-data for RGC\n\n");
1182 		pnv_eeh_dump_hub_diag_common(data);
1183 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1184 			pr_info("  RGC: %016llx %016llx\n",
1185 				be64_to_cpu(data->rgc.rgcStatus),
1186 				be64_to_cpu(data->rgc.rgcLdcp));
1187 		break;
1188 	case OPAL_P7IOC_DIAG_TYPE_BI:
1189 		pr_info("P7IOC diag-data for BI %s\n\n",
1190 			data->bi.biDownbound ? "Downbound" : "Upbound");
1191 		pnv_eeh_dump_hub_diag_common(data);
1192 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1193 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1194 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1195 				be64_to_cpu(data->bi.biLdcp0),
1196 				be64_to_cpu(data->bi.biLdcp1),
1197 				be64_to_cpu(data->bi.biLdcp2),
1198 				be64_to_cpu(data->bi.biFenceStatus));
1199 		break;
1200 	case OPAL_P7IOC_DIAG_TYPE_CI:
1201 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1202 			data->ci.ciPort);
1203 		pnv_eeh_dump_hub_diag_common(data);
1204 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1205 			pr_info("  CI:  %016llx %016llx\n",
1206 				be64_to_cpu(data->ci.ciPortStatus),
1207 				be64_to_cpu(data->ci.ciPortLdcp));
1208 		break;
1209 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1210 		pr_info("P7IOC diag-data for MISC\n\n");
1211 		pnv_eeh_dump_hub_diag_common(data);
1212 		break;
1213 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1214 		pr_info("P7IOC diag-data for I2C\n\n");
1215 		pnv_eeh_dump_hub_diag_common(data);
1216 		break;
1217 	default:
1218 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1219 			__func__, phb->hub_id, data->type);
1220 	}
1221 }
1222 
1223 static int pnv_eeh_get_pe(struct pci_controller *hose,
1224 			  u16 pe_no, struct eeh_pe **pe)
1225 {
1226 	struct pnv_phb *phb = hose->private_data;
1227 	struct pnv_ioda_pe *pnv_pe;
1228 	struct eeh_pe *dev_pe;
1229 	struct eeh_dev edev;
1230 
1231 	/*
1232 	 * If PHB supports compound PE, to fetch
1233 	 * the master PE because slave PE is invisible
1234 	 * to EEH core.
1235 	 */
1236 	pnv_pe = &phb->ioda.pe_array[pe_no];
1237 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1238 		pnv_pe = pnv_pe->master;
1239 		WARN_ON(!pnv_pe ||
1240 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1241 		pe_no = pnv_pe->pe_number;
1242 	}
1243 
1244 	/* Find the PE according to PE# */
1245 	memset(&edev, 0, sizeof(struct eeh_dev));
1246 	edev.phb = hose;
1247 	edev.pe_config_addr = pe_no;
1248 	dev_pe = eeh_pe_get(&edev);
1249 	if (!dev_pe)
1250 		return -EEXIST;
1251 
1252 	/* Freeze the (compound) PE */
1253 	*pe = dev_pe;
1254 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1255 		phb->freeze_pe(phb, pe_no);
1256 
1257 	/*
1258 	 * At this point, we're sure the (compound) PE should
1259 	 * have been frozen. However, we still need poke until
1260 	 * hitting the frozen PE on top level.
1261 	 */
1262 	dev_pe = dev_pe->parent;
1263 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1264 		int ret;
1265 		int active_flags = (EEH_STATE_MMIO_ACTIVE |
1266 				    EEH_STATE_DMA_ACTIVE);
1267 
1268 		ret = eeh_ops->get_state(dev_pe, NULL);
1269 		if (ret <= 0 || (ret & active_flags) == active_flags) {
1270 			dev_pe = dev_pe->parent;
1271 			continue;
1272 		}
1273 
1274 		/* Frozen parent PE */
1275 		*pe = dev_pe;
1276 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1277 			phb->freeze_pe(phb, dev_pe->addr);
1278 
1279 		/* Next one */
1280 		dev_pe = dev_pe->parent;
1281 	}
1282 
1283 	return 0;
1284 }
1285 
1286 /**
1287  * pnv_eeh_next_error - Retrieve next EEH error to handle
1288  * @pe: Affected PE
1289  *
1290  * The function is expected to be called by EEH core while it gets
1291  * special EEH event (without binding PE). The function calls to
1292  * OPAL APIs for next error to handle. The informational error is
1293  * handled internally by platform. However, the dead IOC, dead PHB,
1294  * fenced PHB and frozen PE should be handled by EEH core eventually.
1295  */
1296 static int pnv_eeh_next_error(struct eeh_pe **pe)
1297 {
1298 	struct pci_controller *hose;
1299 	struct pnv_phb *phb;
1300 	struct eeh_pe *phb_pe, *parent_pe;
1301 	__be64 frozen_pe_no;
1302 	__be16 err_type, severity;
1303 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1304 	long rc;
1305 	int state, ret = EEH_NEXT_ERR_NONE;
1306 
1307 	/*
1308 	 * While running here, it's safe to purge the event queue. The
1309 	 * event should still be masked.
1310 	 */
1311 	eeh_remove_event(NULL, false);
1312 
1313 	list_for_each_entry(hose, &hose_list, list_node) {
1314 		/*
1315 		 * If the subordinate PCI buses of the PHB has been
1316 		 * removed or is exactly under error recovery, we
1317 		 * needn't take care of it any more.
1318 		 */
1319 		phb = hose->private_data;
1320 		phb_pe = eeh_phb_pe_get(hose);
1321 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1322 			continue;
1323 
1324 		rc = opal_pci_next_error(phb->opal_id,
1325 					 &frozen_pe_no, &err_type, &severity);
1326 		if (rc != OPAL_SUCCESS) {
1327 			pr_devel("%s: Invalid return value on "
1328 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1329 				 __func__, hose->global_number, rc);
1330 			continue;
1331 		}
1332 
1333 		/* If the PHB doesn't have error, stop processing */
1334 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1335 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1336 			pr_devel("%s: No error found on PHB#%x\n",
1337 				 __func__, hose->global_number);
1338 			continue;
1339 		}
1340 
1341 		/*
1342 		 * Processing the error. We're expecting the error with
1343 		 * highest priority reported upon multiple errors on the
1344 		 * specific PHB.
1345 		 */
1346 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1347 			__func__, be16_to_cpu(err_type),
1348 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1349 			hose->global_number);
1350 		switch (be16_to_cpu(err_type)) {
1351 		case OPAL_EEH_IOC_ERROR:
1352 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1353 				pr_err("EEH: dead IOC detected\n");
1354 				ret = EEH_NEXT_ERR_DEAD_IOC;
1355 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1356 				pr_info("EEH: IOC informative error "
1357 					"detected\n");
1358 				pnv_eeh_get_and_dump_hub_diag(hose);
1359 				ret = EEH_NEXT_ERR_NONE;
1360 			}
1361 
1362 			break;
1363 		case OPAL_EEH_PHB_ERROR:
1364 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1365 				*pe = phb_pe;
1366 				pr_err("EEH: dead PHB#%x detected, "
1367 				       "location: %s\n",
1368 					hose->global_number,
1369 					eeh_pe_loc_get(phb_pe));
1370 				ret = EEH_NEXT_ERR_DEAD_PHB;
1371 			} else if (be16_to_cpu(severity) ==
1372 				   OPAL_EEH_SEV_PHB_FENCED) {
1373 				*pe = phb_pe;
1374 				pr_err("EEH: Fenced PHB#%x detected, "
1375 				       "location: %s\n",
1376 					hose->global_number,
1377 					eeh_pe_loc_get(phb_pe));
1378 				ret = EEH_NEXT_ERR_FENCED_PHB;
1379 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1380 				pr_info("EEH: PHB#%x informative error "
1381 					"detected, location: %s\n",
1382 					hose->global_number,
1383 					eeh_pe_loc_get(phb_pe));
1384 				pnv_eeh_get_phb_diag(phb_pe);
1385 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1386 				ret = EEH_NEXT_ERR_NONE;
1387 			}
1388 
1389 			break;
1390 		case OPAL_EEH_PE_ERROR:
1391 			/*
1392 			 * If we can't find the corresponding PE, we
1393 			 * just try to unfreeze.
1394 			 */
1395 			if (pnv_eeh_get_pe(hose,
1396 				be64_to_cpu(frozen_pe_no), pe)) {
1397 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1398 					hose->global_number, be64_to_cpu(frozen_pe_no));
1399 				pr_info("EEH: PHB location: %s\n",
1400 					eeh_pe_loc_get(phb_pe));
1401 
1402 				/* Dump PHB diag-data */
1403 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1404 					phb->diag.blob, PNV_PCI_DIAG_BUF_SIZE);
1405 				if (rc == OPAL_SUCCESS)
1406 					pnv_pci_dump_phb_diag_data(hose,
1407 							phb->diag.blob);
1408 
1409 				/* Try best to clear it */
1410 				opal_pci_eeh_freeze_clear(phb->opal_id,
1411 					frozen_pe_no,
1412 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1413 				ret = EEH_NEXT_ERR_NONE;
1414 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1415 				   eeh_pe_passed(*pe)) {
1416 				ret = EEH_NEXT_ERR_NONE;
1417 			} else {
1418 				pr_err("EEH: Frozen PE#%x "
1419 				       "on PHB#%x detected\n",
1420 				       (*pe)->addr,
1421 					(*pe)->phb->global_number);
1422 				pr_err("EEH: PE location: %s, "
1423 				       "PHB location: %s\n",
1424 				       eeh_pe_loc_get(*pe),
1425 				       eeh_pe_loc_get(phb_pe));
1426 				ret = EEH_NEXT_ERR_FROZEN_PE;
1427 			}
1428 
1429 			break;
1430 		default:
1431 			pr_warn("%s: Unexpected error type %d\n",
1432 				__func__, be16_to_cpu(err_type));
1433 		}
1434 
1435 		/*
1436 		 * EEH core will try recover from fenced PHB or
1437 		 * frozen PE. In the time for frozen PE, EEH core
1438 		 * enable IO path for that before collecting logs,
1439 		 * but it ruins the site. So we have to dump the
1440 		 * log in advance here.
1441 		 */
1442 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1443 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1444 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1445 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1446 			pnv_eeh_get_phb_diag(*pe);
1447 
1448 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1449 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1450 							   (*pe)->data);
1451 		}
1452 
1453 		/*
1454 		 * We probably have the frozen parent PE out there and
1455 		 * we need have to handle frozen parent PE firstly.
1456 		 */
1457 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1458 			parent_pe = (*pe)->parent;
1459 			while (parent_pe) {
1460 				/* Hit the ceiling ? */
1461 				if (parent_pe->type & EEH_PE_PHB)
1462 					break;
1463 
1464 				/* Frozen parent PE ? */
1465 				state = eeh_ops->get_state(parent_pe, NULL);
1466 				if (state > 0 &&
1467 				    (state & active_flags) != active_flags)
1468 					*pe = parent_pe;
1469 
1470 				/* Next parent level */
1471 				parent_pe = parent_pe->parent;
1472 			}
1473 
1474 			/* We possibly migrate to another PE */
1475 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1476 		}
1477 
1478 		/*
1479 		 * If we have no errors on the specific PHB or only
1480 		 * informative error there, we continue poking it.
1481 		 * Otherwise, we need actions to be taken by upper
1482 		 * layer.
1483 		 */
1484 		if (ret > EEH_NEXT_ERR_INF)
1485 			break;
1486 	}
1487 
1488 	/* Unmask the event */
1489 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1490 		enable_irq(eeh_event_irq);
1491 
1492 	return ret;
1493 }
1494 
1495 static int pnv_eeh_restore_config(struct pci_dn *pdn)
1496 {
1497 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1498 	struct pnv_phb *phb;
1499 	s64 ret;
1500 
1501 	if (!edev)
1502 		return -EEXIST;
1503 
1504 	phb = edev->phb->private_data;
1505 	ret = opal_pci_reinit(phb->opal_id,
1506 			      OPAL_REINIT_PCI_DEV, edev->config_addr);
1507 	if (ret) {
1508 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1509 			__func__, edev->config_addr, ret);
1510 		return -EIO;
1511 	}
1512 
1513 	return 0;
1514 }
1515 
1516 static struct eeh_ops pnv_eeh_ops = {
1517 	.name                   = "powernv",
1518 	.init                   = pnv_eeh_init,
1519 	.post_init              = pnv_eeh_post_init,
1520 	.probe			= pnv_eeh_probe,
1521 	.set_option             = pnv_eeh_set_option,
1522 	.get_pe_addr            = pnv_eeh_get_pe_addr,
1523 	.get_state              = pnv_eeh_get_state,
1524 	.reset                  = pnv_eeh_reset,
1525 	.wait_state             = pnv_eeh_wait_state,
1526 	.get_log                = pnv_eeh_get_log,
1527 	.configure_bridge       = pnv_eeh_configure_bridge,
1528 	.err_inject		= pnv_eeh_err_inject,
1529 	.read_config            = pnv_eeh_read_config,
1530 	.write_config           = pnv_eeh_write_config,
1531 	.next_error		= pnv_eeh_next_error,
1532 	.restore_config		= pnv_eeh_restore_config
1533 };
1534 
1535 /**
1536  * eeh_powernv_init - Register platform dependent EEH operations
1537  *
1538  * EEH initialization on powernv platform. This function should be
1539  * called before any EEH related functions.
1540  */
1541 static int __init eeh_powernv_init(void)
1542 {
1543 	int ret = -EINVAL;
1544 
1545 	eeh_set_pe_aux_size(PNV_PCI_DIAG_BUF_SIZE);
1546 	ret = eeh_ops_register(&pnv_eeh_ops);
1547 	if (!ret)
1548 		pr_info("EEH: PowerNV platform initialized\n");
1549 	else
1550 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
1551 
1552 	return ret;
1553 }
1554 machine_early_initcall(powernv, eeh_powernv_init);
1555