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 static int pnv_eeh_init(void)
47 {
48 	struct pci_controller *hose;
49 	struct pnv_phb *phb;
50 
51 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
52 		pr_warn("%s: OPAL is required !\n",
53 			__func__);
54 		return -EINVAL;
55 	}
56 
57 	/* Set probe mode */
58 	eeh_add_flag(EEH_PROBE_MODE_DEV);
59 
60 	/*
61 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
62 	 * doesn't do that. So we have to selectively enable I/O
63 	 * prior to collecting error log.
64 	 */
65 	list_for_each_entry(hose, &hose_list, list_node) {
66 		phb = hose->private_data;
67 
68 		if (phb->model == PNV_PHB_MODEL_P7IOC)
69 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
70 
71 		/*
72 		 * PE#0 should be regarded as valid by EEH core
73 		 * if it's not the reserved one. Currently, we
74 		 * have the reserved PE#255 and PE#127 for PHB3
75 		 * and P7IOC separately. So we should regard
76 		 * PE#0 as valid for PHB3 and P7IOC.
77 		 */
78 		if (phb->ioda.reserved_pe != 0)
79 			eeh_add_flag(EEH_VALID_PE_ZERO);
80 
81 		break;
82 	}
83 
84 	return 0;
85 }
86 
87 static irqreturn_t pnv_eeh_event(int irq, void *data)
88 {
89 	/*
90 	 * We simply send a special EEH event if EEH has been
91 	 * enabled. We don't care about EEH events until we've
92 	 * finished processing the outstanding ones. Event processing
93 	 * gets unmasked in next_error() if EEH is enabled.
94 	 */
95 	disable_irq_nosync(irq);
96 
97 	if (eeh_enabled())
98 		eeh_send_failure_event(NULL);
99 
100 	return IRQ_HANDLED;
101 }
102 
103 #ifdef CONFIG_DEBUG_FS
104 static ssize_t pnv_eeh_ei_write(struct file *filp,
105 				const char __user *user_buf,
106 				size_t count, loff_t *ppos)
107 {
108 	struct pci_controller *hose = filp->private_data;
109 	struct eeh_dev *edev;
110 	struct eeh_pe *pe;
111 	int pe_no, type, func;
112 	unsigned long addr, mask;
113 	char buf[50];
114 	int ret;
115 
116 	if (!eeh_ops || !eeh_ops->err_inject)
117 		return -ENXIO;
118 
119 	/* Copy over argument buffer */
120 	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
121 	if (!ret)
122 		return -EFAULT;
123 
124 	/* Retrieve parameters */
125 	ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
126 		     &pe_no, &type, &func, &addr, &mask);
127 	if (ret != 5)
128 		return -EINVAL;
129 
130 	/* Retrieve PE */
131 	edev = kzalloc(sizeof(*edev), GFP_KERNEL);
132 	if (!edev)
133 		return -ENOMEM;
134 	edev->phb = hose;
135 	edev->pe_config_addr = pe_no;
136 	pe = eeh_pe_get(edev);
137 	kfree(edev);
138 	if (!pe)
139 		return -ENODEV;
140 
141 	/* Do error injection */
142 	ret = eeh_ops->err_inject(pe, type, func, addr, mask);
143 	return ret < 0 ? ret : count;
144 }
145 
146 static const struct file_operations pnv_eeh_ei_fops = {
147 	.open	= simple_open,
148 	.llseek	= no_llseek,
149 	.write	= pnv_eeh_ei_write,
150 };
151 
152 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val)
153 {
154 	struct pci_controller *hose = data;
155 	struct pnv_phb *phb = hose->private_data;
156 
157 	out_be64(phb->regs + offset, val);
158 	return 0;
159 }
160 
161 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val)
162 {
163 	struct pci_controller *hose = data;
164 	struct pnv_phb *phb = hose->private_data;
165 
166 	*val = in_be64(phb->regs + offset);
167 	return 0;
168 }
169 
170 static int pnv_eeh_outb_dbgfs_set(void *data, u64 val)
171 {
172 	return pnv_eeh_dbgfs_set(data, 0xD10, val);
173 }
174 
175 static int pnv_eeh_outb_dbgfs_get(void *data, u64 *val)
176 {
177 	return pnv_eeh_dbgfs_get(data, 0xD10, val);
178 }
179 
180 static int pnv_eeh_inbA_dbgfs_set(void *data, u64 val)
181 {
182 	return pnv_eeh_dbgfs_set(data, 0xD90, val);
183 }
184 
185 static int pnv_eeh_inbA_dbgfs_get(void *data, u64 *val)
186 {
187 	return pnv_eeh_dbgfs_get(data, 0xD90, val);
188 }
189 
190 static int pnv_eeh_inbB_dbgfs_set(void *data, u64 val)
191 {
192 	return pnv_eeh_dbgfs_set(data, 0xE10, val);
193 }
194 
195 static int pnv_eeh_inbB_dbgfs_get(void *data, u64 *val)
196 {
197 	return pnv_eeh_dbgfs_get(data, 0xE10, val);
198 }
199 
200 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_outb_dbgfs_ops, pnv_eeh_outb_dbgfs_get,
201 			pnv_eeh_outb_dbgfs_set, "0x%llx\n");
202 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_inbA_dbgfs_ops, pnv_eeh_inbA_dbgfs_get,
203 			pnv_eeh_inbA_dbgfs_set, "0x%llx\n");
204 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_inbB_dbgfs_ops, pnv_eeh_inbB_dbgfs_get,
205 			pnv_eeh_inbB_dbgfs_set, "0x%llx\n");
206 #endif /* CONFIG_DEBUG_FS */
207 
208 /**
209  * pnv_eeh_post_init - EEH platform dependent post initialization
210  *
211  * EEH platform dependent post initialization on powernv. When
212  * the function is called, the EEH PEs and devices should have
213  * been built. If the I/O cache staff has been built, EEH is
214  * ready to supply service.
215  */
216 static int pnv_eeh_post_init(void)
217 {
218 	struct pci_controller *hose;
219 	struct pnv_phb *phb;
220 	int ret = 0;
221 
222 	/* Register OPAL event notifier */
223 	if (!pnv_eeh_nb_init) {
224 		eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
225 		if (eeh_event_irq < 0) {
226 			pr_err("%s: Can't register OPAL event interrupt (%d)\n",
227 			       __func__, eeh_event_irq);
228 			return eeh_event_irq;
229 		}
230 
231 		ret = request_irq(eeh_event_irq, pnv_eeh_event,
232 				IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
233 		if (ret < 0) {
234 			irq_dispose_mapping(eeh_event_irq);
235 			pr_err("%s: Can't request OPAL event interrupt (%d)\n",
236 			       __func__, eeh_event_irq);
237 			return ret;
238 		}
239 
240 		pnv_eeh_nb_init = true;
241 	}
242 
243 	if (!eeh_enabled())
244 		disable_irq(eeh_event_irq);
245 
246 	list_for_each_entry(hose, &hose_list, list_node) {
247 		phb = hose->private_data;
248 
249 		/*
250 		 * If EEH is enabled, we're going to rely on that.
251 		 * Otherwise, we restore to conventional mechanism
252 		 * to clear frozen PE during PCI config access.
253 		 */
254 		if (eeh_enabled())
255 			phb->flags |= PNV_PHB_FLAG_EEH;
256 		else
257 			phb->flags &= ~PNV_PHB_FLAG_EEH;
258 
259 		/* Create debugfs entries */
260 #ifdef CONFIG_DEBUG_FS
261 		if (phb->has_dbgfs || !phb->dbgfs)
262 			continue;
263 
264 		phb->has_dbgfs = 1;
265 		debugfs_create_file("err_injct", 0200,
266 				    phb->dbgfs, hose,
267 				    &pnv_eeh_ei_fops);
268 
269 		debugfs_create_file("err_injct_outbound", 0600,
270 				    phb->dbgfs, hose,
271 				    &pnv_eeh_outb_dbgfs_ops);
272 		debugfs_create_file("err_injct_inboundA", 0600,
273 				    phb->dbgfs, hose,
274 				    &pnv_eeh_inbA_dbgfs_ops);
275 		debugfs_create_file("err_injct_inboundB", 0600,
276 				    phb->dbgfs, hose,
277 				    &pnv_eeh_inbB_dbgfs_ops);
278 #endif /* CONFIG_DEBUG_FS */
279 	}
280 
281 	return ret;
282 }
283 
284 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
285 {
286 	int pos = PCI_CAPABILITY_LIST;
287 	int cnt = 48;   /* Maximal number of capabilities */
288 	u32 status, id;
289 
290 	if (!pdn)
291 		return 0;
292 
293 	/* Check if the device supports capabilities */
294 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
295 	if (!(status & PCI_STATUS_CAP_LIST))
296 		return 0;
297 
298 	while (cnt--) {
299 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
300 		if (pos < 0x40)
301 			break;
302 
303 		pos &= ~3;
304 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
305 		if (id == 0xff)
306 			break;
307 
308 		/* Found */
309 		if (id == cap)
310 			return pos;
311 
312 		/* Next one */
313 		pos += PCI_CAP_LIST_NEXT;
314 	}
315 
316 	return 0;
317 }
318 
319 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
320 {
321 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
322 	u32 header;
323 	int pos = 256, ttl = (4096 - 256) / 8;
324 
325 	if (!edev || !edev->pcie_cap)
326 		return 0;
327 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
328 		return 0;
329 	else if (!header)
330 		return 0;
331 
332 	while (ttl-- > 0) {
333 		if (PCI_EXT_CAP_ID(header) == cap && pos)
334 			return pos;
335 
336 		pos = PCI_EXT_CAP_NEXT(header);
337 		if (pos < 256)
338 			break;
339 
340 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
341 			break;
342 	}
343 
344 	return 0;
345 }
346 
347 /**
348  * pnv_eeh_probe - Do probe on PCI device
349  * @pdn: PCI device node
350  * @data: unused
351  *
352  * When EEH module is installed during system boot, all PCI devices
353  * are checked one by one to see if it supports EEH. The function
354  * is introduced for the purpose. By default, EEH has been enabled
355  * on all PCI devices. That's to say, we only need do necessary
356  * initialization on the corresponding eeh device and create PE
357  * accordingly.
358  *
359  * It's notable that's unsafe to retrieve the EEH device through
360  * the corresponding PCI device. During the PCI device hotplug, which
361  * was possiblly triggered by EEH core, the binding between EEH device
362  * and the PCI device isn't built yet.
363  */
364 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
365 {
366 	struct pci_controller *hose = pdn->phb;
367 	struct pnv_phb *phb = hose->private_data;
368 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
369 	uint32_t pcie_flags;
370 	int ret;
371 
372 	/*
373 	 * When probing the root bridge, which doesn't have any
374 	 * subordinate PCI devices. We don't have OF node for
375 	 * the root bridge. So it's not reasonable to continue
376 	 * the probing.
377 	 */
378 	if (!edev || edev->pe)
379 		return NULL;
380 
381 	/* Skip for PCI-ISA bridge */
382 	if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
383 		return NULL;
384 
385 	/* Initialize eeh device */
386 	edev->class_code = pdn->class_code;
387 	edev->mode	&= 0xFFFFFF00;
388 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
389 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
390 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
391 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
392 		edev->mode |= EEH_DEV_BRIDGE;
393 		if (edev->pcie_cap) {
394 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
395 					 2, &pcie_flags);
396 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
397 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
398 				edev->mode |= EEH_DEV_ROOT_PORT;
399 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
400 				edev->mode |= EEH_DEV_DS_PORT;
401 		}
402 	}
403 
404 	edev->config_addr    = (pdn->busno << 8) | (pdn->devfn);
405 	edev->pe_config_addr = phb->ioda.pe_rmap[edev->config_addr];
406 
407 	/* Create PE */
408 	ret = eeh_add_to_parent_pe(edev);
409 	if (ret) {
410 		pr_warn("%s: Can't add PCI dev %04x:%02x:%02x.%01x to parent PE (%d)\n",
411 			__func__, hose->global_number, pdn->busno,
412 			PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn), ret);
413 		return NULL;
414 	}
415 
416 	/*
417 	 * If the PE contains any one of following adapters, the
418 	 * PCI config space can't be accessed when dumping EEH log.
419 	 * Otherwise, we will run into fenced PHB caused by shortage
420 	 * of outbound credits in the adapter. The PCI config access
421 	 * should be blocked until PE reset. MMIO access is dropped
422 	 * by hardware certainly. In order to drop PCI config requests,
423 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
424 	 * will be checked in the backend for PE state retrival. If
425 	 * the PE becomes frozen for the first time and the flag has
426 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
427 	 * that PE to block its config space.
428 	 *
429 	 * Broadcom Austin 4-ports NICs (14e4:1657)
430 	 * Broadcom Shiner 4-ports 1G NICs (14e4:168a)
431 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
432 	 */
433 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
434 	     pdn->device_id == 0x1657) ||
435 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
436 	     pdn->device_id == 0x168a) ||
437 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
438 	     pdn->device_id == 0x168e))
439 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
440 
441 	/*
442 	 * Cache the PE primary bus, which can't be fetched when
443 	 * full hotplug is in progress. In that case, all child
444 	 * PCI devices of the PE are expected to be removed prior
445 	 * to PE reset.
446 	 */
447 	if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
448 		edev->pe->bus = pci_find_bus(hose->global_number,
449 					     pdn->busno);
450 		if (edev->pe->bus)
451 			edev->pe->state |= EEH_PE_PRI_BUS;
452 	}
453 
454 	/*
455 	 * Enable EEH explicitly so that we will do EEH check
456 	 * while accessing I/O stuff
457 	 */
458 	eeh_add_flag(EEH_ENABLED);
459 
460 	/* Save memory bars */
461 	eeh_save_bars(edev);
462 
463 	return NULL;
464 }
465 
466 /**
467  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
468  * @pe: EEH PE
469  * @option: operation to be issued
470  *
471  * The function is used to control the EEH functionality globally.
472  * Currently, following options are support according to PAPR:
473  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
474  */
475 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
476 {
477 	struct pci_controller *hose = pe->phb;
478 	struct pnv_phb *phb = hose->private_data;
479 	bool freeze_pe = false;
480 	int opt;
481 	s64 rc;
482 
483 	switch (option) {
484 	case EEH_OPT_DISABLE:
485 		return -EPERM;
486 	case EEH_OPT_ENABLE:
487 		return 0;
488 	case EEH_OPT_THAW_MMIO:
489 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
490 		break;
491 	case EEH_OPT_THAW_DMA:
492 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
493 		break;
494 	case EEH_OPT_FREEZE_PE:
495 		freeze_pe = true;
496 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
497 		break;
498 	default:
499 		pr_warn("%s: Invalid option %d\n", __func__, option);
500 		return -EINVAL;
501 	}
502 
503 	/* Freeze master and slave PEs if PHB supports compound PEs */
504 	if (freeze_pe) {
505 		if (phb->freeze_pe) {
506 			phb->freeze_pe(phb, pe->addr);
507 			return 0;
508 		}
509 
510 		rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt);
511 		if (rc != OPAL_SUCCESS) {
512 			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
513 				__func__, rc, phb->hose->global_number,
514 				pe->addr);
515 			return -EIO;
516 		}
517 
518 		return 0;
519 	}
520 
521 	/* Unfreeze master and slave PEs if PHB supports */
522 	if (phb->unfreeze_pe)
523 		return phb->unfreeze_pe(phb, pe->addr, opt);
524 
525 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt);
526 	if (rc != OPAL_SUCCESS) {
527 		pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n",
528 			__func__, rc, option, phb->hose->global_number,
529 			pe->addr);
530 		return -EIO;
531 	}
532 
533 	return 0;
534 }
535 
536 /**
537  * pnv_eeh_get_pe_addr - Retrieve PE address
538  * @pe: EEH PE
539  *
540  * Retrieve the PE address according to the given tranditional
541  * PCI BDF (Bus/Device/Function) address.
542  */
543 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe)
544 {
545 	return pe->addr;
546 }
547 
548 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
549 {
550 	struct pnv_phb *phb = pe->phb->private_data;
551 	s64 rc;
552 
553 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
554 					 PNV_PCI_DIAG_BUF_SIZE);
555 	if (rc != OPAL_SUCCESS)
556 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
557 			__func__, rc, pe->phb->global_number);
558 }
559 
560 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
561 {
562 	struct pnv_phb *phb = pe->phb->private_data;
563 	u8 fstate;
564 	__be16 pcierr;
565 	s64 rc;
566 	int result = 0;
567 
568 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
569 					pe->addr,
570 					&fstate,
571 					&pcierr,
572 					NULL);
573 	if (rc != OPAL_SUCCESS) {
574 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
575 			__func__, rc, phb->hose->global_number);
576 		return EEH_STATE_NOT_SUPPORT;
577 	}
578 
579 	/*
580 	 * Check PHB state. If the PHB is frozen for the
581 	 * first time, to dump the PHB diag-data.
582 	 */
583 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
584 		result = (EEH_STATE_MMIO_ACTIVE  |
585 			  EEH_STATE_DMA_ACTIVE   |
586 			  EEH_STATE_MMIO_ENABLED |
587 			  EEH_STATE_DMA_ENABLED);
588 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
589 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
590 		pnv_eeh_get_phb_diag(pe);
591 
592 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
593 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
594 	}
595 
596 	return result;
597 }
598 
599 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
600 {
601 	struct pnv_phb *phb = pe->phb->private_data;
602 	u8 fstate;
603 	__be16 pcierr;
604 	s64 rc;
605 	int result;
606 
607 	/*
608 	 * We don't clobber hardware frozen state until PE
609 	 * reset is completed. In order to keep EEH core
610 	 * moving forward, we have to return operational
611 	 * state during PE reset.
612 	 */
613 	if (pe->state & EEH_PE_RESET) {
614 		result = (EEH_STATE_MMIO_ACTIVE  |
615 			  EEH_STATE_DMA_ACTIVE   |
616 			  EEH_STATE_MMIO_ENABLED |
617 			  EEH_STATE_DMA_ENABLED);
618 		return result;
619 	}
620 
621 	/*
622 	 * Fetch PE state from hardware. If the PHB
623 	 * supports compound PE, let it handle that.
624 	 */
625 	if (phb->get_pe_state) {
626 		fstate = phb->get_pe_state(phb, pe->addr);
627 	} else {
628 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
629 						pe->addr,
630 						&fstate,
631 						&pcierr,
632 						NULL);
633 		if (rc != OPAL_SUCCESS) {
634 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
635 				__func__, rc, phb->hose->global_number,
636 				pe->addr);
637 			return EEH_STATE_NOT_SUPPORT;
638 		}
639 	}
640 
641 	/* Figure out state */
642 	switch (fstate) {
643 	case OPAL_EEH_STOPPED_NOT_FROZEN:
644 		result = (EEH_STATE_MMIO_ACTIVE  |
645 			  EEH_STATE_DMA_ACTIVE   |
646 			  EEH_STATE_MMIO_ENABLED |
647 			  EEH_STATE_DMA_ENABLED);
648 		break;
649 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
650 		result = (EEH_STATE_DMA_ACTIVE |
651 			  EEH_STATE_DMA_ENABLED);
652 		break;
653 	case OPAL_EEH_STOPPED_DMA_FREEZE:
654 		result = (EEH_STATE_MMIO_ACTIVE |
655 			  EEH_STATE_MMIO_ENABLED);
656 		break;
657 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
658 		result = 0;
659 		break;
660 	case OPAL_EEH_STOPPED_RESET:
661 		result = EEH_STATE_RESET_ACTIVE;
662 		break;
663 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
664 		result = EEH_STATE_UNAVAILABLE;
665 		break;
666 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
667 		result = EEH_STATE_NOT_SUPPORT;
668 		break;
669 	default:
670 		result = EEH_STATE_NOT_SUPPORT;
671 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
672 			__func__, phb->hose->global_number,
673 			pe->addr, fstate);
674 	}
675 
676 	/*
677 	 * If PHB supports compound PE, to freeze all
678 	 * slave PEs for consistency.
679 	 *
680 	 * If the PE is switching to frozen state for the
681 	 * first time, to dump the PHB diag-data.
682 	 */
683 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
684 	    !(result & EEH_STATE_UNAVAILABLE) &&
685 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
686 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
687 	    !(pe->state & EEH_PE_ISOLATED)) {
688 		if (phb->freeze_pe)
689 			phb->freeze_pe(phb, pe->addr);
690 
691 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
692 		pnv_eeh_get_phb_diag(pe);
693 
694 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
695 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
696 	}
697 
698 	return result;
699 }
700 
701 /**
702  * pnv_eeh_get_state - Retrieve PE state
703  * @pe: EEH PE
704  * @delay: delay while PE state is temporarily unavailable
705  *
706  * Retrieve the state of the specified PE. For IODA-compitable
707  * platform, it should be retrieved from IODA table. Therefore,
708  * we prefer passing down to hardware implementation to handle
709  * it.
710  */
711 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
712 {
713 	int ret;
714 
715 	if (pe->type & EEH_PE_PHB)
716 		ret = pnv_eeh_get_phb_state(pe);
717 	else
718 		ret = pnv_eeh_get_pe_state(pe);
719 
720 	if (!delay)
721 		return ret;
722 
723 	/*
724 	 * If the PE state is temporarily unavailable,
725 	 * to inform the EEH core delay for default
726 	 * period (1 second)
727 	 */
728 	*delay = 0;
729 	if (ret & EEH_STATE_UNAVAILABLE)
730 		*delay = 1000;
731 
732 	return ret;
733 }
734 
735 static s64 pnv_eeh_phb_poll(struct pnv_phb *phb)
736 {
737 	s64 rc = OPAL_HARDWARE;
738 
739 	while (1) {
740 		rc = opal_pci_poll(phb->opal_id);
741 		if (rc <= 0)
742 			break;
743 
744 		if (system_state < SYSTEM_RUNNING)
745 			udelay(1000 * rc);
746 		else
747 			msleep(rc);
748 	}
749 
750 	return rc;
751 }
752 
753 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
754 {
755 	struct pnv_phb *phb = hose->private_data;
756 	s64 rc = OPAL_HARDWARE;
757 
758 	pr_debug("%s: Reset PHB#%x, option=%d\n",
759 		 __func__, hose->global_number, option);
760 
761 	/* Issue PHB complete reset request */
762 	if (option == EEH_RESET_FUNDAMENTAL ||
763 	    option == EEH_RESET_HOT)
764 		rc = opal_pci_reset(phb->opal_id,
765 				    OPAL_RESET_PHB_COMPLETE,
766 				    OPAL_ASSERT_RESET);
767 	else if (option == EEH_RESET_DEACTIVATE)
768 		rc = opal_pci_reset(phb->opal_id,
769 				    OPAL_RESET_PHB_COMPLETE,
770 				    OPAL_DEASSERT_RESET);
771 	if (rc < 0)
772 		goto out;
773 
774 	/*
775 	 * Poll state of the PHB until the request is done
776 	 * successfully. The PHB reset is usually PHB complete
777 	 * reset followed by hot reset on root bus. So we also
778 	 * need the PCI bus settlement delay.
779 	 */
780 	rc = pnv_eeh_phb_poll(phb);
781 	if (option == EEH_RESET_DEACTIVATE) {
782 		if (system_state < SYSTEM_RUNNING)
783 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
784 		else
785 			msleep(EEH_PE_RST_SETTLE_TIME);
786 	}
787 out:
788 	if (rc != OPAL_SUCCESS)
789 		return -EIO;
790 
791 	return 0;
792 }
793 
794 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
795 {
796 	struct pnv_phb *phb = hose->private_data;
797 	s64 rc = OPAL_HARDWARE;
798 
799 	pr_debug("%s: Reset PHB#%x, option=%d\n",
800 		 __func__, hose->global_number, option);
801 
802 	/*
803 	 * During the reset deassert time, we needn't care
804 	 * the reset scope because the firmware does nothing
805 	 * for fundamental or hot reset during deassert phase.
806 	 */
807 	if (option == EEH_RESET_FUNDAMENTAL)
808 		rc = opal_pci_reset(phb->opal_id,
809 				    OPAL_RESET_PCI_FUNDAMENTAL,
810 				    OPAL_ASSERT_RESET);
811 	else if (option == EEH_RESET_HOT)
812 		rc = opal_pci_reset(phb->opal_id,
813 				    OPAL_RESET_PCI_HOT,
814 				    OPAL_ASSERT_RESET);
815 	else if (option == EEH_RESET_DEACTIVATE)
816 		rc = opal_pci_reset(phb->opal_id,
817 				    OPAL_RESET_PCI_HOT,
818 				    OPAL_DEASSERT_RESET);
819 	if (rc < 0)
820 		goto out;
821 
822 	/* Poll state of the PHB until the request is done */
823 	rc = pnv_eeh_phb_poll(phb);
824 	if (option == EEH_RESET_DEACTIVATE)
825 		msleep(EEH_PE_RST_SETTLE_TIME);
826 out:
827 	if (rc != OPAL_SUCCESS)
828 		return -EIO;
829 
830 	return 0;
831 }
832 
833 static int pnv_eeh_bridge_reset(struct pci_dev *dev, int option)
834 {
835 	struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
836 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
837 	int aer = edev ? edev->aer_cap : 0;
838 	u32 ctrl;
839 
840 	pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
841 		 __func__, pci_domain_nr(dev->bus),
842 		 dev->bus->number, option);
843 
844 	switch (option) {
845 	case EEH_RESET_FUNDAMENTAL:
846 	case EEH_RESET_HOT:
847 		/* Don't report linkDown event */
848 		if (aer) {
849 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
850 					     4, &ctrl);
851 			ctrl |= PCI_ERR_UNC_SURPDN;
852 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
853 					      4, ctrl);
854 		}
855 
856 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
857 		ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
858 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
859 
860 		msleep(EEH_PE_RST_HOLD_TIME);
861 		break;
862 	case EEH_RESET_DEACTIVATE:
863 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
864 		ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
865 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
866 
867 		msleep(EEH_PE_RST_SETTLE_TIME);
868 
869 		/* Continue reporting linkDown event */
870 		if (aer) {
871 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
872 					     4, &ctrl);
873 			ctrl &= ~PCI_ERR_UNC_SURPDN;
874 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
875 					      4, ctrl);
876 		}
877 
878 		break;
879 	}
880 
881 	return 0;
882 }
883 
884 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
885 {
886 	struct pci_controller *hose;
887 
888 	if (pci_is_root_bus(dev->bus)) {
889 		hose = pci_bus_to_host(dev->bus);
890 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
891 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
892 	} else {
893 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
894 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
895 	}
896 }
897 
898 /**
899  * pnv_eeh_reset - Reset the specified PE
900  * @pe: EEH PE
901  * @option: reset option
902  *
903  * Do reset on the indicated PE. For PCI bus sensitive PE,
904  * we need to reset the parent p2p bridge. The PHB has to
905  * be reinitialized if the p2p bridge is root bridge. For
906  * PCI device sensitive PE, we will try to reset the device
907  * through FLR. For now, we don't have OPAL APIs to do HARD
908  * reset yet, so all reset would be SOFT (HOT) reset.
909  */
910 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
911 {
912 	struct pci_controller *hose = pe->phb;
913 	struct pci_bus *bus;
914 	int ret;
915 
916 	/*
917 	 * For PHB reset, we always have complete reset. For those PEs whose
918 	 * primary bus derived from root complex (root bus) or root port
919 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
920 	 * For other PEs, we always have hot reset on the PE primary bus.
921 	 *
922 	 * Here, we have different design to pHyp, which always clear the
923 	 * frozen state during PE reset. However, the good idea here from
924 	 * benh is to keep frozen state before we get PE reset done completely
925 	 * (until BAR restore). With the frozen state, HW drops illegal IO
926 	 * or MMIO access, which can incur recrusive frozen PE during PE
927 	 * reset. The side effect is that EEH core has to clear the frozen
928 	 * state explicitly after BAR restore.
929 	 */
930 	if (pe->type & EEH_PE_PHB) {
931 		ret = pnv_eeh_phb_reset(hose, option);
932 	} else {
933 		struct pnv_phb *phb;
934 		s64 rc;
935 
936 		/*
937 		 * The frozen PE might be caused by PAPR error injection
938 		 * registers, which are expected to be cleared after hitting
939 		 * frozen PE as stated in the hardware spec. Unfortunately,
940 		 * that's not true on P7IOC. So we have to clear it manually
941 		 * to avoid recursive EEH errors during recovery.
942 		 */
943 		phb = hose->private_data;
944 		if (phb->model == PNV_PHB_MODEL_P7IOC &&
945 		    (option == EEH_RESET_HOT ||
946 		    option == EEH_RESET_FUNDAMENTAL)) {
947 			rc = opal_pci_reset(phb->opal_id,
948 					    OPAL_RESET_PHB_ERROR,
949 					    OPAL_ASSERT_RESET);
950 			if (rc != OPAL_SUCCESS) {
951 				pr_warn("%s: Failure %lld clearing "
952 					"error injection registers\n",
953 					__func__, rc);
954 				return -EIO;
955 			}
956 		}
957 
958 		bus = eeh_pe_bus_get(pe);
959 		if (pci_is_root_bus(bus) ||
960 			pci_is_root_bus(bus->parent))
961 			ret = pnv_eeh_root_reset(hose, option);
962 		else
963 			ret = pnv_eeh_bridge_reset(bus->self, option);
964 	}
965 
966 	return ret;
967 }
968 
969 /**
970  * pnv_eeh_wait_state - Wait for PE state
971  * @pe: EEH PE
972  * @max_wait: maximal period in millisecond
973  *
974  * Wait for the state of associated PE. It might take some time
975  * to retrieve the PE's state.
976  */
977 static int pnv_eeh_wait_state(struct eeh_pe *pe, int max_wait)
978 {
979 	int ret;
980 	int mwait;
981 
982 	while (1) {
983 		ret = pnv_eeh_get_state(pe, &mwait);
984 
985 		/*
986 		 * If the PE's state is temporarily unavailable,
987 		 * we have to wait for the specified time. Otherwise,
988 		 * the PE's state will be returned immediately.
989 		 */
990 		if (ret != EEH_STATE_UNAVAILABLE)
991 			return ret;
992 
993 		if (max_wait <= 0) {
994 			pr_warn("%s: Timeout getting PE#%x's state (%d)\n",
995 				__func__, pe->addr, max_wait);
996 			return EEH_STATE_NOT_SUPPORT;
997 		}
998 
999 		max_wait -= mwait;
1000 		msleep(mwait);
1001 	}
1002 
1003 	return EEH_STATE_NOT_SUPPORT;
1004 }
1005 
1006 /**
1007  * pnv_eeh_get_log - Retrieve error log
1008  * @pe: EEH PE
1009  * @severity: temporary or permanent error log
1010  * @drv_log: driver log to be combined with retrieved error log
1011  * @len: length of driver log
1012  *
1013  * Retrieve the temporary or permanent error from the PE.
1014  */
1015 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1016 			   char *drv_log, unsigned long len)
1017 {
1018 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1019 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1020 
1021 	return 0;
1022 }
1023 
1024 /**
1025  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1026  * @pe: EEH PE
1027  *
1028  * The function will be called to reconfigure the bridges included
1029  * in the specified PE so that the mulfunctional PE would be recovered
1030  * again.
1031  */
1032 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1033 {
1034 	return 0;
1035 }
1036 
1037 /**
1038  * pnv_pe_err_inject - Inject specified error to the indicated PE
1039  * @pe: the indicated PE
1040  * @type: error type
1041  * @func: specific error type
1042  * @addr: address
1043  * @mask: address mask
1044  *
1045  * The routine is called to inject specified error, which is
1046  * determined by @type and @func, to the indicated PE for
1047  * testing purpose.
1048  */
1049 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1050 			      unsigned long addr, unsigned long mask)
1051 {
1052 	struct pci_controller *hose = pe->phb;
1053 	struct pnv_phb *phb = hose->private_data;
1054 	s64 rc;
1055 
1056 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1057 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1058 		pr_warn("%s: Invalid error type %d\n",
1059 			__func__, type);
1060 		return -ERANGE;
1061 	}
1062 
1063 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1064 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1065 		pr_warn("%s: Invalid error function %d\n",
1066 			__func__, func);
1067 		return -ERANGE;
1068 	}
1069 
1070 	/* Firmware supports error injection ? */
1071 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1072 		pr_warn("%s: Firmware doesn't support error injection\n",
1073 			__func__);
1074 		return -ENXIO;
1075 	}
1076 
1077 	/* Do error injection */
1078 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1079 				 type, func, addr, mask);
1080 	if (rc != OPAL_SUCCESS) {
1081 		pr_warn("%s: Failure %lld injecting error "
1082 			"%d-%d to PHB#%x-PE#%x\n",
1083 			__func__, rc, type, func,
1084 			hose->global_number, pe->addr);
1085 		return -EIO;
1086 	}
1087 
1088 	return 0;
1089 }
1090 
1091 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1092 {
1093 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1094 
1095 	if (!edev || !edev->pe)
1096 		return false;
1097 
1098 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1099 		return true;
1100 
1101 	return false;
1102 }
1103 
1104 static int pnv_eeh_read_config(struct pci_dn *pdn,
1105 			       int where, int size, u32 *val)
1106 {
1107 	if (!pdn)
1108 		return PCIBIOS_DEVICE_NOT_FOUND;
1109 
1110 	if (pnv_eeh_cfg_blocked(pdn)) {
1111 		*val = 0xFFFFFFFF;
1112 		return PCIBIOS_SET_FAILED;
1113 	}
1114 
1115 	return pnv_pci_cfg_read(pdn, where, size, val);
1116 }
1117 
1118 static int pnv_eeh_write_config(struct pci_dn *pdn,
1119 				int where, int size, u32 val)
1120 {
1121 	if (!pdn)
1122 		return PCIBIOS_DEVICE_NOT_FOUND;
1123 
1124 	if (pnv_eeh_cfg_blocked(pdn))
1125 		return PCIBIOS_SET_FAILED;
1126 
1127 	return pnv_pci_cfg_write(pdn, where, size, val);
1128 }
1129 
1130 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1131 {
1132 	/* GEM */
1133 	if (data->gemXfir || data->gemRfir ||
1134 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1135 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1136 			be64_to_cpu(data->gemXfir),
1137 			be64_to_cpu(data->gemRfir),
1138 			be64_to_cpu(data->gemRirqfir),
1139 			be64_to_cpu(data->gemMask),
1140 			be64_to_cpu(data->gemRwof));
1141 
1142 	/* LEM */
1143 	if (data->lemFir || data->lemErrMask ||
1144 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1145 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1146 			be64_to_cpu(data->lemFir),
1147 			be64_to_cpu(data->lemErrMask),
1148 			be64_to_cpu(data->lemAction0),
1149 			be64_to_cpu(data->lemAction1),
1150 			be64_to_cpu(data->lemWof));
1151 }
1152 
1153 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1154 {
1155 	struct pnv_phb *phb = hose->private_data;
1156 	struct OpalIoP7IOCErrorData *data = &phb->diag.hub_diag;
1157 	long rc;
1158 
1159 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1160 	if (rc != OPAL_SUCCESS) {
1161 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1162 			__func__, phb->hub_id, rc);
1163 		return;
1164 	}
1165 
1166 	switch (data->type) {
1167 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1168 		pr_info("P7IOC diag-data for RGC\n\n");
1169 		pnv_eeh_dump_hub_diag_common(data);
1170 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1171 			pr_info("  RGC: %016llx %016llx\n",
1172 				be64_to_cpu(data->rgc.rgcStatus),
1173 				be64_to_cpu(data->rgc.rgcLdcp));
1174 		break;
1175 	case OPAL_P7IOC_DIAG_TYPE_BI:
1176 		pr_info("P7IOC diag-data for BI %s\n\n",
1177 			data->bi.biDownbound ? "Downbound" : "Upbound");
1178 		pnv_eeh_dump_hub_diag_common(data);
1179 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1180 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1181 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1182 				be64_to_cpu(data->bi.biLdcp0),
1183 				be64_to_cpu(data->bi.biLdcp1),
1184 				be64_to_cpu(data->bi.biLdcp2),
1185 				be64_to_cpu(data->bi.biFenceStatus));
1186 		break;
1187 	case OPAL_P7IOC_DIAG_TYPE_CI:
1188 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1189 			data->ci.ciPort);
1190 		pnv_eeh_dump_hub_diag_common(data);
1191 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1192 			pr_info("  CI:  %016llx %016llx\n",
1193 				be64_to_cpu(data->ci.ciPortStatus),
1194 				be64_to_cpu(data->ci.ciPortLdcp));
1195 		break;
1196 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1197 		pr_info("P7IOC diag-data for MISC\n\n");
1198 		pnv_eeh_dump_hub_diag_common(data);
1199 		break;
1200 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1201 		pr_info("P7IOC diag-data for I2C\n\n");
1202 		pnv_eeh_dump_hub_diag_common(data);
1203 		break;
1204 	default:
1205 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1206 			__func__, phb->hub_id, data->type);
1207 	}
1208 }
1209 
1210 static int pnv_eeh_get_pe(struct pci_controller *hose,
1211 			  u16 pe_no, struct eeh_pe **pe)
1212 {
1213 	struct pnv_phb *phb = hose->private_data;
1214 	struct pnv_ioda_pe *pnv_pe;
1215 	struct eeh_pe *dev_pe;
1216 	struct eeh_dev edev;
1217 
1218 	/*
1219 	 * If PHB supports compound PE, to fetch
1220 	 * the master PE because slave PE is invisible
1221 	 * to EEH core.
1222 	 */
1223 	pnv_pe = &phb->ioda.pe_array[pe_no];
1224 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1225 		pnv_pe = pnv_pe->master;
1226 		WARN_ON(!pnv_pe ||
1227 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1228 		pe_no = pnv_pe->pe_number;
1229 	}
1230 
1231 	/* Find the PE according to PE# */
1232 	memset(&edev, 0, sizeof(struct eeh_dev));
1233 	edev.phb = hose;
1234 	edev.pe_config_addr = pe_no;
1235 	dev_pe = eeh_pe_get(&edev);
1236 	if (!dev_pe)
1237 		return -EEXIST;
1238 
1239 	/* Freeze the (compound) PE */
1240 	*pe = dev_pe;
1241 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1242 		phb->freeze_pe(phb, pe_no);
1243 
1244 	/*
1245 	 * At this point, we're sure the (compound) PE should
1246 	 * have been frozen. However, we still need poke until
1247 	 * hitting the frozen PE on top level.
1248 	 */
1249 	dev_pe = dev_pe->parent;
1250 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1251 		int ret;
1252 		int active_flags = (EEH_STATE_MMIO_ACTIVE |
1253 				    EEH_STATE_DMA_ACTIVE);
1254 
1255 		ret = eeh_ops->get_state(dev_pe, NULL);
1256 		if (ret <= 0 || (ret & active_flags) == active_flags) {
1257 			dev_pe = dev_pe->parent;
1258 			continue;
1259 		}
1260 
1261 		/* Frozen parent PE */
1262 		*pe = dev_pe;
1263 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1264 			phb->freeze_pe(phb, dev_pe->addr);
1265 
1266 		/* Next one */
1267 		dev_pe = dev_pe->parent;
1268 	}
1269 
1270 	return 0;
1271 }
1272 
1273 /**
1274  * pnv_eeh_next_error - Retrieve next EEH error to handle
1275  * @pe: Affected PE
1276  *
1277  * The function is expected to be called by EEH core while it gets
1278  * special EEH event (without binding PE). The function calls to
1279  * OPAL APIs for next error to handle. The informational error is
1280  * handled internally by platform. However, the dead IOC, dead PHB,
1281  * fenced PHB and frozen PE should be handled by EEH core eventually.
1282  */
1283 static int pnv_eeh_next_error(struct eeh_pe **pe)
1284 {
1285 	struct pci_controller *hose;
1286 	struct pnv_phb *phb;
1287 	struct eeh_pe *phb_pe, *parent_pe;
1288 	__be64 frozen_pe_no;
1289 	__be16 err_type, severity;
1290 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1291 	long rc;
1292 	int state, ret = EEH_NEXT_ERR_NONE;
1293 
1294 	/*
1295 	 * While running here, it's safe to purge the event queue. The
1296 	 * event should still be masked.
1297 	 */
1298 	eeh_remove_event(NULL, false);
1299 
1300 	list_for_each_entry(hose, &hose_list, list_node) {
1301 		/*
1302 		 * If the subordinate PCI buses of the PHB has been
1303 		 * removed or is exactly under error recovery, we
1304 		 * needn't take care of it any more.
1305 		 */
1306 		phb = hose->private_data;
1307 		phb_pe = eeh_phb_pe_get(hose);
1308 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1309 			continue;
1310 
1311 		rc = opal_pci_next_error(phb->opal_id,
1312 					 &frozen_pe_no, &err_type, &severity);
1313 		if (rc != OPAL_SUCCESS) {
1314 			pr_devel("%s: Invalid return value on "
1315 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1316 				 __func__, hose->global_number, rc);
1317 			continue;
1318 		}
1319 
1320 		/* If the PHB doesn't have error, stop processing */
1321 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1322 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1323 			pr_devel("%s: No error found on PHB#%x\n",
1324 				 __func__, hose->global_number);
1325 			continue;
1326 		}
1327 
1328 		/*
1329 		 * Processing the error. We're expecting the error with
1330 		 * highest priority reported upon multiple errors on the
1331 		 * specific PHB.
1332 		 */
1333 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1334 			__func__, be16_to_cpu(err_type),
1335 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1336 			hose->global_number);
1337 		switch (be16_to_cpu(err_type)) {
1338 		case OPAL_EEH_IOC_ERROR:
1339 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1340 				pr_err("EEH: dead IOC detected\n");
1341 				ret = EEH_NEXT_ERR_DEAD_IOC;
1342 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1343 				pr_info("EEH: IOC informative error "
1344 					"detected\n");
1345 				pnv_eeh_get_and_dump_hub_diag(hose);
1346 				ret = EEH_NEXT_ERR_NONE;
1347 			}
1348 
1349 			break;
1350 		case OPAL_EEH_PHB_ERROR:
1351 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1352 				*pe = phb_pe;
1353 				pr_err("EEH: dead PHB#%x detected, "
1354 				       "location: %s\n",
1355 					hose->global_number,
1356 					eeh_pe_loc_get(phb_pe));
1357 				ret = EEH_NEXT_ERR_DEAD_PHB;
1358 			} else if (be16_to_cpu(severity) ==
1359 				   OPAL_EEH_SEV_PHB_FENCED) {
1360 				*pe = phb_pe;
1361 				pr_err("EEH: Fenced PHB#%x detected, "
1362 				       "location: %s\n",
1363 					hose->global_number,
1364 					eeh_pe_loc_get(phb_pe));
1365 				ret = EEH_NEXT_ERR_FENCED_PHB;
1366 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1367 				pr_info("EEH: PHB#%x informative error "
1368 					"detected, location: %s\n",
1369 					hose->global_number,
1370 					eeh_pe_loc_get(phb_pe));
1371 				pnv_eeh_get_phb_diag(phb_pe);
1372 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1373 				ret = EEH_NEXT_ERR_NONE;
1374 			}
1375 
1376 			break;
1377 		case OPAL_EEH_PE_ERROR:
1378 			/*
1379 			 * If we can't find the corresponding PE, we
1380 			 * just try to unfreeze.
1381 			 */
1382 			if (pnv_eeh_get_pe(hose,
1383 				be64_to_cpu(frozen_pe_no), pe)) {
1384 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1385 					hose->global_number, be64_to_cpu(frozen_pe_no));
1386 				pr_info("EEH: PHB location: %s\n",
1387 					eeh_pe_loc_get(phb_pe));
1388 
1389 				/* Dump PHB diag-data */
1390 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1391 					phb->diag.blob, PNV_PCI_DIAG_BUF_SIZE);
1392 				if (rc == OPAL_SUCCESS)
1393 					pnv_pci_dump_phb_diag_data(hose,
1394 							phb->diag.blob);
1395 
1396 				/* Try best to clear it */
1397 				opal_pci_eeh_freeze_clear(phb->opal_id,
1398 					frozen_pe_no,
1399 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1400 				ret = EEH_NEXT_ERR_NONE;
1401 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1402 				   eeh_pe_passed(*pe)) {
1403 				ret = EEH_NEXT_ERR_NONE;
1404 			} else {
1405 				pr_err("EEH: Frozen PE#%x "
1406 				       "on PHB#%x detected\n",
1407 				       (*pe)->addr,
1408 					(*pe)->phb->global_number);
1409 				pr_err("EEH: PE location: %s, "
1410 				       "PHB location: %s\n",
1411 				       eeh_pe_loc_get(*pe),
1412 				       eeh_pe_loc_get(phb_pe));
1413 				ret = EEH_NEXT_ERR_FROZEN_PE;
1414 			}
1415 
1416 			break;
1417 		default:
1418 			pr_warn("%s: Unexpected error type %d\n",
1419 				__func__, be16_to_cpu(err_type));
1420 		}
1421 
1422 		/*
1423 		 * EEH core will try recover from fenced PHB or
1424 		 * frozen PE. In the time for frozen PE, EEH core
1425 		 * enable IO path for that before collecting logs,
1426 		 * but it ruins the site. So we have to dump the
1427 		 * log in advance here.
1428 		 */
1429 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1430 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1431 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1432 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1433 			pnv_eeh_get_phb_diag(*pe);
1434 
1435 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1436 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1437 							   (*pe)->data);
1438 		}
1439 
1440 		/*
1441 		 * We probably have the frozen parent PE out there and
1442 		 * we need have to handle frozen parent PE firstly.
1443 		 */
1444 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1445 			parent_pe = (*pe)->parent;
1446 			while (parent_pe) {
1447 				/* Hit the ceiling ? */
1448 				if (parent_pe->type & EEH_PE_PHB)
1449 					break;
1450 
1451 				/* Frozen parent PE ? */
1452 				state = eeh_ops->get_state(parent_pe, NULL);
1453 				if (state > 0 &&
1454 				    (state & active_flags) != active_flags)
1455 					*pe = parent_pe;
1456 
1457 				/* Next parent level */
1458 				parent_pe = parent_pe->parent;
1459 			}
1460 
1461 			/* We possibly migrate to another PE */
1462 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1463 		}
1464 
1465 		/*
1466 		 * If we have no errors on the specific PHB or only
1467 		 * informative error there, we continue poking it.
1468 		 * Otherwise, we need actions to be taken by upper
1469 		 * layer.
1470 		 */
1471 		if (ret > EEH_NEXT_ERR_INF)
1472 			break;
1473 	}
1474 
1475 	/* Unmask the event */
1476 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1477 		enable_irq(eeh_event_irq);
1478 
1479 	return ret;
1480 }
1481 
1482 static int pnv_eeh_restore_config(struct pci_dn *pdn)
1483 {
1484 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1485 	struct pnv_phb *phb;
1486 	s64 ret;
1487 
1488 	if (!edev)
1489 		return -EEXIST;
1490 
1491 	phb = edev->phb->private_data;
1492 	ret = opal_pci_reinit(phb->opal_id,
1493 			      OPAL_REINIT_PCI_DEV, edev->config_addr);
1494 	if (ret) {
1495 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1496 			__func__, edev->config_addr, ret);
1497 		return -EIO;
1498 	}
1499 
1500 	return 0;
1501 }
1502 
1503 static struct eeh_ops pnv_eeh_ops = {
1504 	.name                   = "powernv",
1505 	.init                   = pnv_eeh_init,
1506 	.post_init              = pnv_eeh_post_init,
1507 	.probe			= pnv_eeh_probe,
1508 	.set_option             = pnv_eeh_set_option,
1509 	.get_pe_addr            = pnv_eeh_get_pe_addr,
1510 	.get_state              = pnv_eeh_get_state,
1511 	.reset                  = pnv_eeh_reset,
1512 	.wait_state             = pnv_eeh_wait_state,
1513 	.get_log                = pnv_eeh_get_log,
1514 	.configure_bridge       = pnv_eeh_configure_bridge,
1515 	.err_inject		= pnv_eeh_err_inject,
1516 	.read_config            = pnv_eeh_read_config,
1517 	.write_config           = pnv_eeh_write_config,
1518 	.next_error		= pnv_eeh_next_error,
1519 	.restore_config		= pnv_eeh_restore_config
1520 };
1521 
1522 /**
1523  * eeh_powernv_init - Register platform dependent EEH operations
1524  *
1525  * EEH initialization on powernv platform. This function should be
1526  * called before any EEH related functions.
1527  */
1528 static int __init eeh_powernv_init(void)
1529 {
1530 	int ret = -EINVAL;
1531 
1532 	eeh_set_pe_aux_size(PNV_PCI_DIAG_BUF_SIZE);
1533 	ret = eeh_ops_register(&pnv_eeh_ops);
1534 	if (!ret)
1535 		pr_info("EEH: PowerNV platform initialized\n");
1536 	else
1537 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
1538 
1539 	return ret;
1540 }
1541 machine_early_initcall(powernv, eeh_powernv_init);
1542