1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * PowerNV Platform dependent EEH operations
4  *
5  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013.
6  */
7 
8 #include <linux/atomic.h>
9 #include <linux/debugfs.h>
10 #include <linux/delay.h>
11 #include <linux/export.h>
12 #include <linux/init.h>
13 #include <linux/interrupt.h>
14 #include <linux/list.h>
15 #include <linux/msi.h>
16 #include <linux/of.h>
17 #include <linux/pci.h>
18 #include <linux/proc_fs.h>
19 #include <linux/rbtree.h>
20 #include <linux/sched.h>
21 #include <linux/seq_file.h>
22 #include <linux/spinlock.h>
23 
24 #include <asm/eeh.h>
25 #include <asm/eeh_event.h>
26 #include <asm/firmware.h>
27 #include <asm/io.h>
28 #include <asm/iommu.h>
29 #include <asm/machdep.h>
30 #include <asm/msi_bitmap.h>
31 #include <asm/opal.h>
32 #include <asm/ppc-pci.h>
33 #include <asm/pnv-pci.h>
34 
35 #include "powernv.h"
36 #include "pci.h"
37 
38 static int eeh_event_irq = -EINVAL;
39 
40 void pnv_pcibios_bus_add_device(struct pci_dev *pdev)
41 {
42 	struct pci_dn *pdn = pci_get_pdn(pdev);
43 
44 	if (!pdev->is_virtfn)
45 		return;
46 
47 	/*
48 	 * The following operations will fail if VF's sysfs files
49 	 * aren't created or its resources aren't finalized.
50 	 */
51 	eeh_add_device_early(pdn);
52 	eeh_add_device_late(pdev);
53 	eeh_sysfs_add_device(pdev);
54 }
55 
56 static int pnv_eeh_init(void)
57 {
58 	struct pci_controller *hose;
59 	struct pnv_phb *phb;
60 	int max_diag_size = PNV_PCI_DIAG_BUF_SIZE;
61 
62 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
63 		pr_warn("%s: OPAL is required !\n",
64 			__func__);
65 		return -EINVAL;
66 	}
67 
68 	/* Set probe mode */
69 	eeh_add_flag(EEH_PROBE_MODE_DEV);
70 
71 	/*
72 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
73 	 * doesn't do that. So we have to selectively enable I/O
74 	 * prior to collecting error log.
75 	 */
76 	list_for_each_entry(hose, &hose_list, list_node) {
77 		phb = hose->private_data;
78 
79 		if (phb->model == PNV_PHB_MODEL_P7IOC)
80 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
81 
82 		if (phb->diag_data_size > max_diag_size)
83 			max_diag_size = phb->diag_data_size;
84 
85 		/*
86 		 * PE#0 should be regarded as valid by EEH core
87 		 * if it's not the reserved one. Currently, we
88 		 * have the reserved PE#255 and PE#127 for PHB3
89 		 * and P7IOC separately. So we should regard
90 		 * PE#0 as valid for PHB3 and P7IOC.
91 		 */
92 		if (phb->ioda.reserved_pe_idx != 0)
93 			eeh_add_flag(EEH_VALID_PE_ZERO);
94 
95 		break;
96 	}
97 
98 	eeh_set_pe_aux_size(max_diag_size);
99 	ppc_md.pcibios_bus_add_device = pnv_pcibios_bus_add_device;
100 
101 	return 0;
102 }
103 
104 static irqreturn_t pnv_eeh_event(int irq, void *data)
105 {
106 	/*
107 	 * We simply send a special EEH event if EEH has been
108 	 * enabled. We don't care about EEH events until we've
109 	 * finished processing the outstanding ones. Event processing
110 	 * gets unmasked in next_error() if EEH is enabled.
111 	 */
112 	disable_irq_nosync(irq);
113 
114 	if (eeh_enabled())
115 		eeh_send_failure_event(NULL);
116 
117 	return IRQ_HANDLED;
118 }
119 
120 #ifdef CONFIG_DEBUG_FS
121 static ssize_t pnv_eeh_ei_write(struct file *filp,
122 				const char __user *user_buf,
123 				size_t count, loff_t *ppos)
124 {
125 	struct pci_controller *hose = filp->private_data;
126 	struct eeh_pe *pe;
127 	int pe_no, type, func;
128 	unsigned long addr, mask;
129 	char buf[50];
130 	int ret;
131 
132 	if (!eeh_ops || !eeh_ops->err_inject)
133 		return -ENXIO;
134 
135 	/* Copy over argument buffer */
136 	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
137 	if (!ret)
138 		return -EFAULT;
139 
140 	/* Retrieve parameters */
141 	ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
142 		     &pe_no, &type, &func, &addr, &mask);
143 	if (ret != 5)
144 		return -EINVAL;
145 
146 	/* Retrieve PE */
147 	pe = eeh_pe_get(hose, pe_no, 0);
148 	if (!pe)
149 		return -ENODEV;
150 
151 	/* Do error injection */
152 	ret = eeh_ops->err_inject(pe, type, func, addr, mask);
153 	return ret < 0 ? ret : count;
154 }
155 
156 static const struct file_operations pnv_eeh_ei_fops = {
157 	.open	= simple_open,
158 	.llseek	= no_llseek,
159 	.write	= pnv_eeh_ei_write,
160 };
161 
162 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val)
163 {
164 	struct pci_controller *hose = data;
165 	struct pnv_phb *phb = hose->private_data;
166 
167 	out_be64(phb->regs + offset, val);
168 	return 0;
169 }
170 
171 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val)
172 {
173 	struct pci_controller *hose = data;
174 	struct pnv_phb *phb = hose->private_data;
175 
176 	*val = in_be64(phb->regs + offset);
177 	return 0;
178 }
179 
180 #define PNV_EEH_DBGFS_ENTRY(name, reg)				\
181 static int pnv_eeh_dbgfs_set_##name(void *data, u64 val)	\
182 {								\
183 	return pnv_eeh_dbgfs_set(data, reg, val);		\
184 }								\
185 								\
186 static int pnv_eeh_dbgfs_get_##name(void *data, u64 *val)	\
187 {								\
188 	return pnv_eeh_dbgfs_get(data, reg, val);		\
189 }								\
190 								\
191 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_dbgfs_ops_##name,		\
192 			pnv_eeh_dbgfs_get_##name,		\
193                         pnv_eeh_dbgfs_set_##name,		\
194 			"0x%llx\n")
195 
196 PNV_EEH_DBGFS_ENTRY(outb, 0xD10);
197 PNV_EEH_DBGFS_ENTRY(inbA, 0xD90);
198 PNV_EEH_DBGFS_ENTRY(inbB, 0xE10);
199 
200 #endif /* CONFIG_DEBUG_FS */
201 
202 /**
203  * pnv_eeh_post_init - EEH platform dependent post initialization
204  *
205  * EEH platform dependent post initialization on powernv. When
206  * the function is called, the EEH PEs and devices should have
207  * been built. If the I/O cache staff has been built, EEH is
208  * ready to supply service.
209  */
210 int pnv_eeh_post_init(void)
211 {
212 	struct pci_controller *hose;
213 	struct pnv_phb *phb;
214 	int ret = 0;
215 
216 	/* Probe devices & build address cache */
217 	eeh_probe_devices();
218 	eeh_addr_cache_build();
219 
220 	/* Register OPAL event notifier */
221 	eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
222 	if (eeh_event_irq < 0) {
223 		pr_err("%s: Can't register OPAL event interrupt (%d)\n",
224 		       __func__, eeh_event_irq);
225 		return eeh_event_irq;
226 	}
227 
228 	ret = request_irq(eeh_event_irq, pnv_eeh_event,
229 			  IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
230 	if (ret < 0) {
231 		irq_dispose_mapping(eeh_event_irq);
232 		pr_err("%s: Can't request OPAL event interrupt (%d)\n",
233 		       __func__, eeh_event_irq);
234 		return ret;
235 	}
236 
237 	if (!eeh_enabled())
238 		disable_irq(eeh_event_irq);
239 
240 	list_for_each_entry(hose, &hose_list, list_node) {
241 		phb = hose->private_data;
242 
243 		/*
244 		 * If EEH is enabled, we're going to rely on that.
245 		 * Otherwise, we restore to conventional mechanism
246 		 * to clear frozen PE during PCI config access.
247 		 */
248 		if (eeh_enabled())
249 			phb->flags |= PNV_PHB_FLAG_EEH;
250 		else
251 			phb->flags &= ~PNV_PHB_FLAG_EEH;
252 
253 		/* Create debugfs entries */
254 #ifdef CONFIG_DEBUG_FS
255 		if (phb->has_dbgfs || !phb->dbgfs)
256 			continue;
257 
258 		phb->has_dbgfs = 1;
259 		debugfs_create_file("err_injct", 0200,
260 				    phb->dbgfs, hose,
261 				    &pnv_eeh_ei_fops);
262 
263 		debugfs_create_file("err_injct_outbound", 0600,
264 				    phb->dbgfs, hose,
265 				    &pnv_eeh_dbgfs_ops_outb);
266 		debugfs_create_file("err_injct_inboundA", 0600,
267 				    phb->dbgfs, hose,
268 				    &pnv_eeh_dbgfs_ops_inbA);
269 		debugfs_create_file("err_injct_inboundB", 0600,
270 				    phb->dbgfs, hose,
271 				    &pnv_eeh_dbgfs_ops_inbB);
272 #endif /* CONFIG_DEBUG_FS */
273 	}
274 
275 	return ret;
276 }
277 
278 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
279 {
280 	int pos = PCI_CAPABILITY_LIST;
281 	int cnt = 48;   /* Maximal number of capabilities */
282 	u32 status, id;
283 
284 	if (!pdn)
285 		return 0;
286 
287 	/* Check if the device supports capabilities */
288 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
289 	if (!(status & PCI_STATUS_CAP_LIST))
290 		return 0;
291 
292 	while (cnt--) {
293 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
294 		if (pos < 0x40)
295 			break;
296 
297 		pos &= ~3;
298 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
299 		if (id == 0xff)
300 			break;
301 
302 		/* Found */
303 		if (id == cap)
304 			return pos;
305 
306 		/* Next one */
307 		pos += PCI_CAP_LIST_NEXT;
308 	}
309 
310 	return 0;
311 }
312 
313 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
314 {
315 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
316 	u32 header;
317 	int pos = 256, ttl = (4096 - 256) / 8;
318 
319 	if (!edev || !edev->pcie_cap)
320 		return 0;
321 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
322 		return 0;
323 	else if (!header)
324 		return 0;
325 
326 	while (ttl-- > 0) {
327 		if (PCI_EXT_CAP_ID(header) == cap && pos)
328 			return pos;
329 
330 		pos = PCI_EXT_CAP_NEXT(header);
331 		if (pos < 256)
332 			break;
333 
334 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
335 			break;
336 	}
337 
338 	return 0;
339 }
340 
341 /**
342  * pnv_eeh_probe - Do probe on PCI device
343  * @pdn: PCI device node
344  * @data: unused
345  *
346  * When EEH module is installed during system boot, all PCI devices
347  * are checked one by one to see if it supports EEH. The function
348  * is introduced for the purpose. By default, EEH has been enabled
349  * on all PCI devices. That's to say, we only need do necessary
350  * initialization on the corresponding eeh device and create PE
351  * accordingly.
352  *
353  * It's notable that's unsafe to retrieve the EEH device through
354  * the corresponding PCI device. During the PCI device hotplug, which
355  * was possiblly triggered by EEH core, the binding between EEH device
356  * and the PCI device isn't built yet.
357  */
358 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
359 {
360 	struct pci_controller *hose = pdn->phb;
361 	struct pnv_phb *phb = hose->private_data;
362 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
363 	uint32_t pcie_flags;
364 	int ret;
365 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
366 
367 	/*
368 	 * When probing the root bridge, which doesn't have any
369 	 * subordinate PCI devices. We don't have OF node for
370 	 * the root bridge. So it's not reasonable to continue
371 	 * the probing.
372 	 */
373 	if (!edev || edev->pe)
374 		return NULL;
375 
376 	/* Skip for PCI-ISA bridge */
377 	if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
378 		return NULL;
379 
380 	/* Initialize eeh device */
381 	edev->class_code = pdn->class_code;
382 	edev->mode	&= 0xFFFFFF00;
383 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
384 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
385 	edev->af_cap   = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF);
386 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
387 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
388 		edev->mode |= EEH_DEV_BRIDGE;
389 		if (edev->pcie_cap) {
390 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
391 					 2, &pcie_flags);
392 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
393 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
394 				edev->mode |= EEH_DEV_ROOT_PORT;
395 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
396 				edev->mode |= EEH_DEV_DS_PORT;
397 		}
398 	}
399 
400 	edev->pe_config_addr = phb->ioda.pe_rmap[config_addr];
401 
402 	/* Create PE */
403 	ret = eeh_add_to_parent_pe(edev);
404 	if (ret) {
405 		pr_warn("%s: Can't add PCI dev %04x:%02x:%02x.%01x to parent PE (%x)\n",
406 			__func__, hose->global_number, pdn->busno,
407 			PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn), ret);
408 		return NULL;
409 	}
410 
411 	/*
412 	 * If the PE contains any one of following adapters, the
413 	 * PCI config space can't be accessed when dumping EEH log.
414 	 * Otherwise, we will run into fenced PHB caused by shortage
415 	 * of outbound credits in the adapter. The PCI config access
416 	 * should be blocked until PE reset. MMIO access is dropped
417 	 * by hardware certainly. In order to drop PCI config requests,
418 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
419 	 * will be checked in the backend for PE state retrival. If
420 	 * the PE becomes frozen for the first time and the flag has
421 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
422 	 * that PE to block its config space.
423 	 *
424 	 * Broadcom BCM5718 2-ports NICs (14e4:1656)
425 	 * Broadcom Austin 4-ports NICs (14e4:1657)
426 	 * Broadcom Shiner 4-ports 1G NICs (14e4:168a)
427 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
428 	 */
429 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
430 	     pdn->device_id == 0x1656) ||
431 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
432 	     pdn->device_id == 0x1657) ||
433 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
434 	     pdn->device_id == 0x168a) ||
435 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
436 	     pdn->device_id == 0x168e))
437 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
438 
439 	/*
440 	 * Cache the PE primary bus, which can't be fetched when
441 	 * full hotplug is in progress. In that case, all child
442 	 * PCI devices of the PE are expected to be removed prior
443 	 * to PE reset.
444 	 */
445 	if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
446 		edev->pe->bus = pci_find_bus(hose->global_number,
447 					     pdn->busno);
448 		if (edev->pe->bus)
449 			edev->pe->state |= EEH_PE_PRI_BUS;
450 	}
451 
452 	/*
453 	 * Enable EEH explicitly so that we will do EEH check
454 	 * while accessing I/O stuff
455 	 */
456 	eeh_add_flag(EEH_ENABLED);
457 
458 	/* Save memory bars */
459 	eeh_save_bars(edev);
460 
461 	return NULL;
462 }
463 
464 /**
465  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
466  * @pe: EEH PE
467  * @option: operation to be issued
468  *
469  * The function is used to control the EEH functionality globally.
470  * Currently, following options are support according to PAPR:
471  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
472  */
473 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
474 {
475 	struct pci_controller *hose = pe->phb;
476 	struct pnv_phb *phb = hose->private_data;
477 	bool freeze_pe = false;
478 	int opt;
479 	s64 rc;
480 
481 	switch (option) {
482 	case EEH_OPT_DISABLE:
483 		return -EPERM;
484 	case EEH_OPT_ENABLE:
485 		return 0;
486 	case EEH_OPT_THAW_MMIO:
487 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
488 		break;
489 	case EEH_OPT_THAW_DMA:
490 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
491 		break;
492 	case EEH_OPT_FREEZE_PE:
493 		freeze_pe = true;
494 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
495 		break;
496 	default:
497 		pr_warn("%s: Invalid option %d\n", __func__, option);
498 		return -EINVAL;
499 	}
500 
501 	/* Freeze master and slave PEs if PHB supports compound PEs */
502 	if (freeze_pe) {
503 		if (phb->freeze_pe) {
504 			phb->freeze_pe(phb, pe->addr);
505 			return 0;
506 		}
507 
508 		rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt);
509 		if (rc != OPAL_SUCCESS) {
510 			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
511 				__func__, rc, phb->hose->global_number,
512 				pe->addr);
513 			return -EIO;
514 		}
515 
516 		return 0;
517 	}
518 
519 	/* Unfreeze master and slave PEs if PHB supports */
520 	if (phb->unfreeze_pe)
521 		return phb->unfreeze_pe(phb, pe->addr, opt);
522 
523 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt);
524 	if (rc != OPAL_SUCCESS) {
525 		pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n",
526 			__func__, rc, option, phb->hose->global_number,
527 			pe->addr);
528 		return -EIO;
529 	}
530 
531 	return 0;
532 }
533 
534 /**
535  * pnv_eeh_get_pe_addr - Retrieve PE address
536  * @pe: EEH PE
537  *
538  * Retrieve the PE address according to the given tranditional
539  * PCI BDF (Bus/Device/Function) address.
540  */
541 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe)
542 {
543 	return pe->addr;
544 }
545 
546 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
547 {
548 	struct pnv_phb *phb = pe->phb->private_data;
549 	s64 rc;
550 
551 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
552 					 phb->diag_data_size);
553 	if (rc != OPAL_SUCCESS)
554 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
555 			__func__, rc, pe->phb->global_number);
556 }
557 
558 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
559 {
560 	struct pnv_phb *phb = pe->phb->private_data;
561 	u8 fstate = 0;
562 	__be16 pcierr = 0;
563 	s64 rc;
564 	int result = 0;
565 
566 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
567 					pe->addr,
568 					&fstate,
569 					&pcierr,
570 					NULL);
571 	if (rc != OPAL_SUCCESS) {
572 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
573 			__func__, rc, phb->hose->global_number);
574 		return EEH_STATE_NOT_SUPPORT;
575 	}
576 
577 	/*
578 	 * Check PHB state. If the PHB is frozen for the
579 	 * first time, to dump the PHB diag-data.
580 	 */
581 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
582 		result = (EEH_STATE_MMIO_ACTIVE  |
583 			  EEH_STATE_DMA_ACTIVE   |
584 			  EEH_STATE_MMIO_ENABLED |
585 			  EEH_STATE_DMA_ENABLED);
586 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
587 		eeh_pe_mark_isolated(pe);
588 		pnv_eeh_get_phb_diag(pe);
589 
590 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
591 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
592 	}
593 
594 	return result;
595 }
596 
597 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
598 {
599 	struct pnv_phb *phb = pe->phb->private_data;
600 	u8 fstate = 0;
601 	__be16 pcierr = 0;
602 	s64 rc;
603 	int result;
604 
605 	/*
606 	 * We don't clobber hardware frozen state until PE
607 	 * reset is completed. In order to keep EEH core
608 	 * moving forward, we have to return operational
609 	 * state during PE reset.
610 	 */
611 	if (pe->state & EEH_PE_RESET) {
612 		result = (EEH_STATE_MMIO_ACTIVE  |
613 			  EEH_STATE_DMA_ACTIVE   |
614 			  EEH_STATE_MMIO_ENABLED |
615 			  EEH_STATE_DMA_ENABLED);
616 		return result;
617 	}
618 
619 	/*
620 	 * Fetch PE state from hardware. If the PHB
621 	 * supports compound PE, let it handle that.
622 	 */
623 	if (phb->get_pe_state) {
624 		fstate = phb->get_pe_state(phb, pe->addr);
625 	} else {
626 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
627 						pe->addr,
628 						&fstate,
629 						&pcierr,
630 						NULL);
631 		if (rc != OPAL_SUCCESS) {
632 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
633 				__func__, rc, phb->hose->global_number,
634 				pe->addr);
635 			return EEH_STATE_NOT_SUPPORT;
636 		}
637 	}
638 
639 	/* Figure out state */
640 	switch (fstate) {
641 	case OPAL_EEH_STOPPED_NOT_FROZEN:
642 		result = (EEH_STATE_MMIO_ACTIVE  |
643 			  EEH_STATE_DMA_ACTIVE   |
644 			  EEH_STATE_MMIO_ENABLED |
645 			  EEH_STATE_DMA_ENABLED);
646 		break;
647 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
648 		result = (EEH_STATE_DMA_ACTIVE |
649 			  EEH_STATE_DMA_ENABLED);
650 		break;
651 	case OPAL_EEH_STOPPED_DMA_FREEZE:
652 		result = (EEH_STATE_MMIO_ACTIVE |
653 			  EEH_STATE_MMIO_ENABLED);
654 		break;
655 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
656 		result = 0;
657 		break;
658 	case OPAL_EEH_STOPPED_RESET:
659 		result = EEH_STATE_RESET_ACTIVE;
660 		break;
661 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
662 		result = EEH_STATE_UNAVAILABLE;
663 		break;
664 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
665 		result = EEH_STATE_NOT_SUPPORT;
666 		break;
667 	default:
668 		result = EEH_STATE_NOT_SUPPORT;
669 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
670 			__func__, phb->hose->global_number,
671 			pe->addr, fstate);
672 	}
673 
674 	/*
675 	 * If PHB supports compound PE, to freeze all
676 	 * slave PEs for consistency.
677 	 *
678 	 * If the PE is switching to frozen state for the
679 	 * first time, to dump the PHB diag-data.
680 	 */
681 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
682 	    !(result & EEH_STATE_UNAVAILABLE) &&
683 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
684 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
685 	    !(pe->state & EEH_PE_ISOLATED)) {
686 		if (phb->freeze_pe)
687 			phb->freeze_pe(phb, pe->addr);
688 
689 		eeh_pe_mark_isolated(pe);
690 		pnv_eeh_get_phb_diag(pe);
691 
692 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
693 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
694 	}
695 
696 	return result;
697 }
698 
699 /**
700  * pnv_eeh_get_state - Retrieve PE state
701  * @pe: EEH PE
702  * @delay: delay while PE state is temporarily unavailable
703  *
704  * Retrieve the state of the specified PE. For IODA-compitable
705  * platform, it should be retrieved from IODA table. Therefore,
706  * we prefer passing down to hardware implementation to handle
707  * it.
708  */
709 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
710 {
711 	int ret;
712 
713 	if (pe->type & EEH_PE_PHB)
714 		ret = pnv_eeh_get_phb_state(pe);
715 	else
716 		ret = pnv_eeh_get_pe_state(pe);
717 
718 	if (!delay)
719 		return ret;
720 
721 	/*
722 	 * If the PE state is temporarily unavailable,
723 	 * to inform the EEH core delay for default
724 	 * period (1 second)
725 	 */
726 	*delay = 0;
727 	if (ret & EEH_STATE_UNAVAILABLE)
728 		*delay = 1000;
729 
730 	return ret;
731 }
732 
733 static s64 pnv_eeh_poll(unsigned long id)
734 {
735 	s64 rc = OPAL_HARDWARE;
736 
737 	while (1) {
738 		rc = opal_pci_poll(id);
739 		if (rc <= 0)
740 			break;
741 
742 		if (system_state < SYSTEM_RUNNING)
743 			udelay(1000 * rc);
744 		else
745 			msleep(rc);
746 	}
747 
748 	return rc;
749 }
750 
751 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
752 {
753 	struct pnv_phb *phb = hose->private_data;
754 	s64 rc = OPAL_HARDWARE;
755 
756 	pr_debug("%s: Reset PHB#%x, option=%d\n",
757 		 __func__, hose->global_number, option);
758 
759 	/* Issue PHB complete reset request */
760 	if (option == EEH_RESET_FUNDAMENTAL ||
761 	    option == EEH_RESET_HOT)
762 		rc = opal_pci_reset(phb->opal_id,
763 				    OPAL_RESET_PHB_COMPLETE,
764 				    OPAL_ASSERT_RESET);
765 	else if (option == EEH_RESET_DEACTIVATE)
766 		rc = opal_pci_reset(phb->opal_id,
767 				    OPAL_RESET_PHB_COMPLETE,
768 				    OPAL_DEASSERT_RESET);
769 	if (rc < 0)
770 		goto out;
771 
772 	/*
773 	 * Poll state of the PHB until the request is done
774 	 * successfully. The PHB reset is usually PHB complete
775 	 * reset followed by hot reset on root bus. So we also
776 	 * need the PCI bus settlement delay.
777 	 */
778 	if (rc > 0)
779 		rc = pnv_eeh_poll(phb->opal_id);
780 	if (option == EEH_RESET_DEACTIVATE) {
781 		if (system_state < SYSTEM_RUNNING)
782 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
783 		else
784 			msleep(EEH_PE_RST_SETTLE_TIME);
785 	}
786 out:
787 	if (rc != OPAL_SUCCESS)
788 		return -EIO;
789 
790 	return 0;
791 }
792 
793 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
794 {
795 	struct pnv_phb *phb = hose->private_data;
796 	s64 rc = OPAL_HARDWARE;
797 
798 	pr_debug("%s: Reset PHB#%x, option=%d\n",
799 		 __func__, hose->global_number, option);
800 
801 	/*
802 	 * During the reset deassert time, we needn't care
803 	 * the reset scope because the firmware does nothing
804 	 * for fundamental or hot reset during deassert phase.
805 	 */
806 	if (option == EEH_RESET_FUNDAMENTAL)
807 		rc = opal_pci_reset(phb->opal_id,
808 				    OPAL_RESET_PCI_FUNDAMENTAL,
809 				    OPAL_ASSERT_RESET);
810 	else if (option == EEH_RESET_HOT)
811 		rc = opal_pci_reset(phb->opal_id,
812 				    OPAL_RESET_PCI_HOT,
813 				    OPAL_ASSERT_RESET);
814 	else if (option == EEH_RESET_DEACTIVATE)
815 		rc = opal_pci_reset(phb->opal_id,
816 				    OPAL_RESET_PCI_HOT,
817 				    OPAL_DEASSERT_RESET);
818 	if (rc < 0)
819 		goto out;
820 
821 	/* Poll state of the PHB until the request is done */
822 	if (rc > 0)
823 		rc = pnv_eeh_poll(phb->opal_id);
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 static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option)
885 {
886 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
887 	struct pnv_phb *phb = hose->private_data;
888 	struct device_node *dn = pci_device_to_OF_node(pdev);
889 	uint64_t id = PCI_SLOT_ID(phb->opal_id,
890 				  (pdev->bus->number << 8) | pdev->devfn);
891 	uint8_t scope;
892 	int64_t rc;
893 
894 	/* Hot reset to the bus if firmware cannot handle */
895 	if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL))
896 		return __pnv_eeh_bridge_reset(pdev, option);
897 
898 	switch (option) {
899 	case EEH_RESET_FUNDAMENTAL:
900 		scope = OPAL_RESET_PCI_FUNDAMENTAL;
901 		break;
902 	case EEH_RESET_HOT:
903 		scope = OPAL_RESET_PCI_HOT;
904 		break;
905 	case EEH_RESET_DEACTIVATE:
906 		return 0;
907 	default:
908 		dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n",
909 			__func__, option);
910 		return -EINVAL;
911 	}
912 
913 	rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET);
914 	if (rc <= OPAL_SUCCESS)
915 		goto out;
916 
917 	rc = pnv_eeh_poll(id);
918 out:
919 	return (rc == OPAL_SUCCESS) ? 0 : -EIO;
920 }
921 
922 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
923 {
924 	struct pci_controller *hose;
925 
926 	if (pci_is_root_bus(dev->bus)) {
927 		hose = pci_bus_to_host(dev->bus);
928 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
929 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
930 	} else {
931 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
932 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
933 	}
934 }
935 
936 static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type,
937 				     int pos, u16 mask)
938 {
939 	int i, status = 0;
940 
941 	/* Wait for Transaction Pending bit to be cleared */
942 	for (i = 0; i < 4; i++) {
943 		eeh_ops->read_config(pdn, pos, 2, &status);
944 		if (!(status & mask))
945 			return;
946 
947 		msleep((1 << i) * 100);
948 	}
949 
950 	pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n",
951 		__func__, type,
952 		pdn->phb->global_number, pdn->busno,
953 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
954 }
955 
956 static int pnv_eeh_do_flr(struct pci_dn *pdn, int option)
957 {
958 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
959 	u32 reg = 0;
960 
961 	if (WARN_ON(!edev->pcie_cap))
962 		return -ENOTTY;
963 
964 	eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP, 4, &reg);
965 	if (!(reg & PCI_EXP_DEVCAP_FLR))
966 		return -ENOTTY;
967 
968 	switch (option) {
969 	case EEH_RESET_HOT:
970 	case EEH_RESET_FUNDAMENTAL:
971 		pnv_eeh_wait_for_pending(pdn, "",
972 					 edev->pcie_cap + PCI_EXP_DEVSTA,
973 					 PCI_EXP_DEVSTA_TRPND);
974 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
975 				     4, &reg);
976 		reg |= PCI_EXP_DEVCTL_BCR_FLR;
977 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
978 				      4, reg);
979 		msleep(EEH_PE_RST_HOLD_TIME);
980 		break;
981 	case EEH_RESET_DEACTIVATE:
982 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
983 				     4, &reg);
984 		reg &= ~PCI_EXP_DEVCTL_BCR_FLR;
985 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
986 				      4, reg);
987 		msleep(EEH_PE_RST_SETTLE_TIME);
988 		break;
989 	}
990 
991 	return 0;
992 }
993 
994 static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option)
995 {
996 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
997 	u32 cap = 0;
998 
999 	if (WARN_ON(!edev->af_cap))
1000 		return -ENOTTY;
1001 
1002 	eeh_ops->read_config(pdn, edev->af_cap + PCI_AF_CAP, 1, &cap);
1003 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
1004 		return -ENOTTY;
1005 
1006 	switch (option) {
1007 	case EEH_RESET_HOT:
1008 	case EEH_RESET_FUNDAMENTAL:
1009 		/*
1010 		 * Wait for Transaction Pending bit to clear. A word-aligned
1011 		 * test is used, so we use the conrol offset rather than status
1012 		 * and shift the test bit to match.
1013 		 */
1014 		pnv_eeh_wait_for_pending(pdn, "AF",
1015 					 edev->af_cap + PCI_AF_CTRL,
1016 					 PCI_AF_STATUS_TP << 8);
1017 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL,
1018 				      1, PCI_AF_CTRL_FLR);
1019 		msleep(EEH_PE_RST_HOLD_TIME);
1020 		break;
1021 	case EEH_RESET_DEACTIVATE:
1022 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL, 1, 0);
1023 		msleep(EEH_PE_RST_SETTLE_TIME);
1024 		break;
1025 	}
1026 
1027 	return 0;
1028 }
1029 
1030 static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option)
1031 {
1032 	struct eeh_dev *edev;
1033 	struct pci_dn *pdn;
1034 	int ret;
1035 
1036 	/* The VF PE should have only one child device */
1037 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, entry);
1038 	pdn = eeh_dev_to_pdn(edev);
1039 	if (!pdn)
1040 		return -ENXIO;
1041 
1042 	ret = pnv_eeh_do_flr(pdn, option);
1043 	if (!ret)
1044 		return ret;
1045 
1046 	return pnv_eeh_do_af_flr(pdn, option);
1047 }
1048 
1049 /**
1050  * pnv_eeh_reset - Reset the specified PE
1051  * @pe: EEH PE
1052  * @option: reset option
1053  *
1054  * Do reset on the indicated PE. For PCI bus sensitive PE,
1055  * we need to reset the parent p2p bridge. The PHB has to
1056  * be reinitialized if the p2p bridge is root bridge. For
1057  * PCI device sensitive PE, we will try to reset the device
1058  * through FLR. For now, we don't have OPAL APIs to do HARD
1059  * reset yet, so all reset would be SOFT (HOT) reset.
1060  */
1061 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
1062 {
1063 	struct pci_controller *hose = pe->phb;
1064 	struct pnv_phb *phb;
1065 	struct pci_bus *bus;
1066 	int64_t rc;
1067 
1068 	/*
1069 	 * For PHB reset, we always have complete reset. For those PEs whose
1070 	 * primary bus derived from root complex (root bus) or root port
1071 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
1072 	 * For other PEs, we always have hot reset on the PE primary bus.
1073 	 *
1074 	 * Here, we have different design to pHyp, which always clear the
1075 	 * frozen state during PE reset. However, the good idea here from
1076 	 * benh is to keep frozen state before we get PE reset done completely
1077 	 * (until BAR restore). With the frozen state, HW drops illegal IO
1078 	 * or MMIO access, which can incur recrusive frozen PE during PE
1079 	 * reset. The side effect is that EEH core has to clear the frozen
1080 	 * state explicitly after BAR restore.
1081 	 */
1082 	if (pe->type & EEH_PE_PHB)
1083 		return pnv_eeh_phb_reset(hose, option);
1084 
1085 	/*
1086 	 * The frozen PE might be caused by PAPR error injection
1087 	 * registers, which are expected to be cleared after hitting
1088 	 * frozen PE as stated in the hardware spec. Unfortunately,
1089 	 * that's not true on P7IOC. So we have to clear it manually
1090 	 * to avoid recursive EEH errors during recovery.
1091 	 */
1092 	phb = hose->private_data;
1093 	if (phb->model == PNV_PHB_MODEL_P7IOC &&
1094 	    (option == EEH_RESET_HOT ||
1095 	     option == EEH_RESET_FUNDAMENTAL)) {
1096 		rc = opal_pci_reset(phb->opal_id,
1097 				    OPAL_RESET_PHB_ERROR,
1098 				    OPAL_ASSERT_RESET);
1099 		if (rc != OPAL_SUCCESS) {
1100 			pr_warn("%s: Failure %lld clearing error injection registers\n",
1101 				__func__, rc);
1102 			return -EIO;
1103 		}
1104 	}
1105 
1106 	if (pe->type & EEH_PE_VF)
1107 		return pnv_eeh_reset_vf_pe(pe, option);
1108 
1109 	bus = eeh_pe_bus_get(pe);
1110 	if (!bus) {
1111 		pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
1112 			__func__, pe->phb->global_number, pe->addr);
1113 		return -EIO;
1114 	}
1115 
1116 	/*
1117 	 * If dealing with the root bus (or the bus underneath the
1118 	 * root port), we reset the bus underneath the root port.
1119 	 *
1120 	 * The cxl driver depends on this behaviour for bi-modal card
1121 	 * switching.
1122 	 */
1123 	if (pci_is_root_bus(bus) ||
1124 	    pci_is_root_bus(bus->parent))
1125 		return pnv_eeh_root_reset(hose, option);
1126 
1127 	return pnv_eeh_bridge_reset(bus->self, option);
1128 }
1129 
1130 /**
1131  * pnv_eeh_get_log - Retrieve error log
1132  * @pe: EEH PE
1133  * @severity: temporary or permanent error log
1134  * @drv_log: driver log to be combined with retrieved error log
1135  * @len: length of driver log
1136  *
1137  * Retrieve the temporary or permanent error from the PE.
1138  */
1139 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1140 			   char *drv_log, unsigned long len)
1141 {
1142 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1143 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1144 
1145 	return 0;
1146 }
1147 
1148 /**
1149  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1150  * @pe: EEH PE
1151  *
1152  * The function will be called to reconfigure the bridges included
1153  * in the specified PE so that the mulfunctional PE would be recovered
1154  * again.
1155  */
1156 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1157 {
1158 	return 0;
1159 }
1160 
1161 /**
1162  * pnv_pe_err_inject - Inject specified error to the indicated PE
1163  * @pe: the indicated PE
1164  * @type: error type
1165  * @func: specific error type
1166  * @addr: address
1167  * @mask: address mask
1168  *
1169  * The routine is called to inject specified error, which is
1170  * determined by @type and @func, to the indicated PE for
1171  * testing purpose.
1172  */
1173 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1174 			      unsigned long addr, unsigned long mask)
1175 {
1176 	struct pci_controller *hose = pe->phb;
1177 	struct pnv_phb *phb = hose->private_data;
1178 	s64 rc;
1179 
1180 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1181 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1182 		pr_warn("%s: Invalid error type %d\n",
1183 			__func__, type);
1184 		return -ERANGE;
1185 	}
1186 
1187 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1188 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1189 		pr_warn("%s: Invalid error function %d\n",
1190 			__func__, func);
1191 		return -ERANGE;
1192 	}
1193 
1194 	/* Firmware supports error injection ? */
1195 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1196 		pr_warn("%s: Firmware doesn't support error injection\n",
1197 			__func__);
1198 		return -ENXIO;
1199 	}
1200 
1201 	/* Do error injection */
1202 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1203 				 type, func, addr, mask);
1204 	if (rc != OPAL_SUCCESS) {
1205 		pr_warn("%s: Failure %lld injecting error "
1206 			"%d-%d to PHB#%x-PE#%x\n",
1207 			__func__, rc, type, func,
1208 			hose->global_number, pe->addr);
1209 		return -EIO;
1210 	}
1211 
1212 	return 0;
1213 }
1214 
1215 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1216 {
1217 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1218 
1219 	if (!edev || !edev->pe)
1220 		return false;
1221 
1222 	/*
1223 	 * We will issue FLR or AF FLR to all VFs, which are contained
1224 	 * in VF PE. It relies on the EEH PCI config accessors. So we
1225 	 * can't block them during the window.
1226 	 */
1227 	if (edev->physfn && (edev->pe->state & EEH_PE_RESET))
1228 		return false;
1229 
1230 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1231 		return true;
1232 
1233 	return false;
1234 }
1235 
1236 static int pnv_eeh_read_config(struct pci_dn *pdn,
1237 			       int where, int size, u32 *val)
1238 {
1239 	if (!pdn)
1240 		return PCIBIOS_DEVICE_NOT_FOUND;
1241 
1242 	if (pnv_eeh_cfg_blocked(pdn)) {
1243 		*val = 0xFFFFFFFF;
1244 		return PCIBIOS_SET_FAILED;
1245 	}
1246 
1247 	return pnv_pci_cfg_read(pdn, where, size, val);
1248 }
1249 
1250 static int pnv_eeh_write_config(struct pci_dn *pdn,
1251 				int where, int size, u32 val)
1252 {
1253 	if (!pdn)
1254 		return PCIBIOS_DEVICE_NOT_FOUND;
1255 
1256 	if (pnv_eeh_cfg_blocked(pdn))
1257 		return PCIBIOS_SET_FAILED;
1258 
1259 	return pnv_pci_cfg_write(pdn, where, size, val);
1260 }
1261 
1262 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1263 {
1264 	/* GEM */
1265 	if (data->gemXfir || data->gemRfir ||
1266 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1267 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1268 			be64_to_cpu(data->gemXfir),
1269 			be64_to_cpu(data->gemRfir),
1270 			be64_to_cpu(data->gemRirqfir),
1271 			be64_to_cpu(data->gemMask),
1272 			be64_to_cpu(data->gemRwof));
1273 
1274 	/* LEM */
1275 	if (data->lemFir || data->lemErrMask ||
1276 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1277 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1278 			be64_to_cpu(data->lemFir),
1279 			be64_to_cpu(data->lemErrMask),
1280 			be64_to_cpu(data->lemAction0),
1281 			be64_to_cpu(data->lemAction1),
1282 			be64_to_cpu(data->lemWof));
1283 }
1284 
1285 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1286 {
1287 	struct pnv_phb *phb = hose->private_data;
1288 	struct OpalIoP7IOCErrorData *data =
1289 		(struct OpalIoP7IOCErrorData*)phb->diag_data;
1290 	long rc;
1291 
1292 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1293 	if (rc != OPAL_SUCCESS) {
1294 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1295 			__func__, phb->hub_id, rc);
1296 		return;
1297 	}
1298 
1299 	switch (be16_to_cpu(data->type)) {
1300 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1301 		pr_info("P7IOC diag-data for RGC\n\n");
1302 		pnv_eeh_dump_hub_diag_common(data);
1303 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1304 			pr_info("  RGC: %016llx %016llx\n",
1305 				be64_to_cpu(data->rgc.rgcStatus),
1306 				be64_to_cpu(data->rgc.rgcLdcp));
1307 		break;
1308 	case OPAL_P7IOC_DIAG_TYPE_BI:
1309 		pr_info("P7IOC diag-data for BI %s\n\n",
1310 			data->bi.biDownbound ? "Downbound" : "Upbound");
1311 		pnv_eeh_dump_hub_diag_common(data);
1312 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1313 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1314 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1315 				be64_to_cpu(data->bi.biLdcp0),
1316 				be64_to_cpu(data->bi.biLdcp1),
1317 				be64_to_cpu(data->bi.biLdcp2),
1318 				be64_to_cpu(data->bi.biFenceStatus));
1319 		break;
1320 	case OPAL_P7IOC_DIAG_TYPE_CI:
1321 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1322 			data->ci.ciPort);
1323 		pnv_eeh_dump_hub_diag_common(data);
1324 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1325 			pr_info("  CI:  %016llx %016llx\n",
1326 				be64_to_cpu(data->ci.ciPortStatus),
1327 				be64_to_cpu(data->ci.ciPortLdcp));
1328 		break;
1329 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1330 		pr_info("P7IOC diag-data for MISC\n\n");
1331 		pnv_eeh_dump_hub_diag_common(data);
1332 		break;
1333 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1334 		pr_info("P7IOC diag-data for I2C\n\n");
1335 		pnv_eeh_dump_hub_diag_common(data);
1336 		break;
1337 	default:
1338 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1339 			__func__, phb->hub_id, data->type);
1340 	}
1341 }
1342 
1343 static int pnv_eeh_get_pe(struct pci_controller *hose,
1344 			  u16 pe_no, struct eeh_pe **pe)
1345 {
1346 	struct pnv_phb *phb = hose->private_data;
1347 	struct pnv_ioda_pe *pnv_pe;
1348 	struct eeh_pe *dev_pe;
1349 
1350 	/*
1351 	 * If PHB supports compound PE, to fetch
1352 	 * the master PE because slave PE is invisible
1353 	 * to EEH core.
1354 	 */
1355 	pnv_pe = &phb->ioda.pe_array[pe_no];
1356 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1357 		pnv_pe = pnv_pe->master;
1358 		WARN_ON(!pnv_pe ||
1359 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1360 		pe_no = pnv_pe->pe_number;
1361 	}
1362 
1363 	/* Find the PE according to PE# */
1364 	dev_pe = eeh_pe_get(hose, pe_no, 0);
1365 	if (!dev_pe)
1366 		return -EEXIST;
1367 
1368 	/* Freeze the (compound) PE */
1369 	*pe = dev_pe;
1370 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1371 		phb->freeze_pe(phb, pe_no);
1372 
1373 	/*
1374 	 * At this point, we're sure the (compound) PE should
1375 	 * have been frozen. However, we still need poke until
1376 	 * hitting the frozen PE on top level.
1377 	 */
1378 	dev_pe = dev_pe->parent;
1379 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1380 		int ret;
1381 		ret = eeh_ops->get_state(dev_pe, NULL);
1382 		if (ret <= 0 || eeh_state_active(ret)) {
1383 			dev_pe = dev_pe->parent;
1384 			continue;
1385 		}
1386 
1387 		/* Frozen parent PE */
1388 		*pe = dev_pe;
1389 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1390 			phb->freeze_pe(phb, dev_pe->addr);
1391 
1392 		/* Next one */
1393 		dev_pe = dev_pe->parent;
1394 	}
1395 
1396 	return 0;
1397 }
1398 
1399 /**
1400  * pnv_eeh_next_error - Retrieve next EEH error to handle
1401  * @pe: Affected PE
1402  *
1403  * The function is expected to be called by EEH core while it gets
1404  * special EEH event (without binding PE). The function calls to
1405  * OPAL APIs for next error to handle. The informational error is
1406  * handled internally by platform. However, the dead IOC, dead PHB,
1407  * fenced PHB and frozen PE should be handled by EEH core eventually.
1408  */
1409 static int pnv_eeh_next_error(struct eeh_pe **pe)
1410 {
1411 	struct pci_controller *hose;
1412 	struct pnv_phb *phb;
1413 	struct eeh_pe *phb_pe, *parent_pe;
1414 	__be64 frozen_pe_no;
1415 	__be16 err_type, severity;
1416 	long rc;
1417 	int state, ret = EEH_NEXT_ERR_NONE;
1418 
1419 	/*
1420 	 * While running here, it's safe to purge the event queue. The
1421 	 * event should still be masked.
1422 	 */
1423 	eeh_remove_event(NULL, false);
1424 
1425 	list_for_each_entry(hose, &hose_list, list_node) {
1426 		/*
1427 		 * If the subordinate PCI buses of the PHB has been
1428 		 * removed or is exactly under error recovery, we
1429 		 * needn't take care of it any more.
1430 		 */
1431 		phb = hose->private_data;
1432 		phb_pe = eeh_phb_pe_get(hose);
1433 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1434 			continue;
1435 
1436 		rc = opal_pci_next_error(phb->opal_id,
1437 					 &frozen_pe_no, &err_type, &severity);
1438 		if (rc != OPAL_SUCCESS) {
1439 			pr_devel("%s: Invalid return value on "
1440 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1441 				 __func__, hose->global_number, rc);
1442 			continue;
1443 		}
1444 
1445 		/* If the PHB doesn't have error, stop processing */
1446 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1447 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1448 			pr_devel("%s: No error found on PHB#%x\n",
1449 				 __func__, hose->global_number);
1450 			continue;
1451 		}
1452 
1453 		/*
1454 		 * Processing the error. We're expecting the error with
1455 		 * highest priority reported upon multiple errors on the
1456 		 * specific PHB.
1457 		 */
1458 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1459 			__func__, be16_to_cpu(err_type),
1460 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1461 			hose->global_number);
1462 		switch (be16_to_cpu(err_type)) {
1463 		case OPAL_EEH_IOC_ERROR:
1464 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1465 				pr_err("EEH: dead IOC detected\n");
1466 				ret = EEH_NEXT_ERR_DEAD_IOC;
1467 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1468 				pr_info("EEH: IOC informative error "
1469 					"detected\n");
1470 				pnv_eeh_get_and_dump_hub_diag(hose);
1471 				ret = EEH_NEXT_ERR_NONE;
1472 			}
1473 
1474 			break;
1475 		case OPAL_EEH_PHB_ERROR:
1476 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1477 				*pe = phb_pe;
1478 				pr_err("EEH: dead PHB#%x detected, "
1479 				       "location: %s\n",
1480 					hose->global_number,
1481 					eeh_pe_loc_get(phb_pe));
1482 				ret = EEH_NEXT_ERR_DEAD_PHB;
1483 			} else if (be16_to_cpu(severity) ==
1484 				   OPAL_EEH_SEV_PHB_FENCED) {
1485 				*pe = phb_pe;
1486 				pr_err("EEH: Fenced PHB#%x detected, "
1487 				       "location: %s\n",
1488 					hose->global_number,
1489 					eeh_pe_loc_get(phb_pe));
1490 				ret = EEH_NEXT_ERR_FENCED_PHB;
1491 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1492 				pr_info("EEH: PHB#%x informative error "
1493 					"detected, location: %s\n",
1494 					hose->global_number,
1495 					eeh_pe_loc_get(phb_pe));
1496 				pnv_eeh_get_phb_diag(phb_pe);
1497 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1498 				ret = EEH_NEXT_ERR_NONE;
1499 			}
1500 
1501 			break;
1502 		case OPAL_EEH_PE_ERROR:
1503 			/*
1504 			 * If we can't find the corresponding PE, we
1505 			 * just try to unfreeze.
1506 			 */
1507 			if (pnv_eeh_get_pe(hose,
1508 				be64_to_cpu(frozen_pe_no), pe)) {
1509 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1510 					hose->global_number, be64_to_cpu(frozen_pe_no));
1511 				pr_info("EEH: PHB location: %s\n",
1512 					eeh_pe_loc_get(phb_pe));
1513 
1514 				/* Dump PHB diag-data */
1515 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1516 					phb->diag_data, phb->diag_data_size);
1517 				if (rc == OPAL_SUCCESS)
1518 					pnv_pci_dump_phb_diag_data(hose,
1519 							phb->diag_data);
1520 
1521 				/* Try best to clear it */
1522 				opal_pci_eeh_freeze_clear(phb->opal_id,
1523 					be64_to_cpu(frozen_pe_no),
1524 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1525 				ret = EEH_NEXT_ERR_NONE;
1526 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1527 				   eeh_pe_passed(*pe)) {
1528 				ret = EEH_NEXT_ERR_NONE;
1529 			} else {
1530 				pr_err("EEH: Frozen PE#%x "
1531 				       "on PHB#%x detected\n",
1532 				       (*pe)->addr,
1533 					(*pe)->phb->global_number);
1534 				pr_err("EEH: PE location: %s, "
1535 				       "PHB location: %s\n",
1536 				       eeh_pe_loc_get(*pe),
1537 				       eeh_pe_loc_get(phb_pe));
1538 				ret = EEH_NEXT_ERR_FROZEN_PE;
1539 			}
1540 
1541 			break;
1542 		default:
1543 			pr_warn("%s: Unexpected error type %d\n",
1544 				__func__, be16_to_cpu(err_type));
1545 		}
1546 
1547 		/*
1548 		 * EEH core will try recover from fenced PHB or
1549 		 * frozen PE. In the time for frozen PE, EEH core
1550 		 * enable IO path for that before collecting logs,
1551 		 * but it ruins the site. So we have to dump the
1552 		 * log in advance here.
1553 		 */
1554 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1555 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1556 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1557 			eeh_pe_mark_isolated(*pe);
1558 			pnv_eeh_get_phb_diag(*pe);
1559 
1560 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1561 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1562 							   (*pe)->data);
1563 		}
1564 
1565 		/*
1566 		 * We probably have the frozen parent PE out there and
1567 		 * we need have to handle frozen parent PE firstly.
1568 		 */
1569 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1570 			parent_pe = (*pe)->parent;
1571 			while (parent_pe) {
1572 				/* Hit the ceiling ? */
1573 				if (parent_pe->type & EEH_PE_PHB)
1574 					break;
1575 
1576 				/* Frozen parent PE ? */
1577 				state = eeh_ops->get_state(parent_pe, NULL);
1578 				if (state > 0 && !eeh_state_active(state))
1579 					*pe = parent_pe;
1580 
1581 				/* Next parent level */
1582 				parent_pe = parent_pe->parent;
1583 			}
1584 
1585 			/* We possibly migrate to another PE */
1586 			eeh_pe_mark_isolated(*pe);
1587 		}
1588 
1589 		/*
1590 		 * If we have no errors on the specific PHB or only
1591 		 * informative error there, we continue poking it.
1592 		 * Otherwise, we need actions to be taken by upper
1593 		 * layer.
1594 		 */
1595 		if (ret > EEH_NEXT_ERR_INF)
1596 			break;
1597 	}
1598 
1599 	/* Unmask the event */
1600 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1601 		enable_irq(eeh_event_irq);
1602 
1603 	return ret;
1604 }
1605 
1606 static int pnv_eeh_restore_config(struct pci_dn *pdn)
1607 {
1608 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1609 	struct pnv_phb *phb;
1610 	s64 ret = 0;
1611 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
1612 
1613 	if (!edev)
1614 		return -EEXIST;
1615 
1616 	/*
1617 	 * We have to restore the PCI config space after reset since the
1618 	 * firmware can't see SRIOV VFs.
1619 	 *
1620 	 * FIXME: The MPS, error routing rules, timeout setting are worthy
1621 	 * to be exported by firmware in extendible way.
1622 	 */
1623 	if (edev->physfn) {
1624 		ret = eeh_restore_vf_config(pdn);
1625 	} else {
1626 		phb = pdn->phb->private_data;
1627 		ret = opal_pci_reinit(phb->opal_id,
1628 				      OPAL_REINIT_PCI_DEV, config_addr);
1629 	}
1630 
1631 	if (ret) {
1632 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1633 			__func__, config_addr, ret);
1634 		return -EIO;
1635 	}
1636 
1637 	return ret;
1638 }
1639 
1640 static struct eeh_ops pnv_eeh_ops = {
1641 	.name                   = "powernv",
1642 	.init                   = pnv_eeh_init,
1643 	.probe			= pnv_eeh_probe,
1644 	.set_option             = pnv_eeh_set_option,
1645 	.get_pe_addr            = pnv_eeh_get_pe_addr,
1646 	.get_state              = pnv_eeh_get_state,
1647 	.reset                  = pnv_eeh_reset,
1648 	.get_log                = pnv_eeh_get_log,
1649 	.configure_bridge       = pnv_eeh_configure_bridge,
1650 	.err_inject		= pnv_eeh_err_inject,
1651 	.read_config            = pnv_eeh_read_config,
1652 	.write_config           = pnv_eeh_write_config,
1653 	.next_error		= pnv_eeh_next_error,
1654 	.restore_config		= pnv_eeh_restore_config,
1655 	.notify_resume		= NULL
1656 };
1657 
1658 #ifdef CONFIG_PCI_IOV
1659 static void pnv_pci_fixup_vf_mps(struct pci_dev *pdev)
1660 {
1661 	struct pci_dn *pdn = pci_get_pdn(pdev);
1662 	int parent_mps;
1663 
1664 	if (!pdev->is_virtfn)
1665 		return;
1666 
1667 	/* Synchronize MPS for VF and PF */
1668 	parent_mps = pcie_get_mps(pdev->physfn);
1669 	if ((128 << pdev->pcie_mpss) >= parent_mps)
1670 		pcie_set_mps(pdev, parent_mps);
1671 	pdn->mps = pcie_get_mps(pdev);
1672 }
1673 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pnv_pci_fixup_vf_mps);
1674 #endif /* CONFIG_PCI_IOV */
1675 
1676 /**
1677  * eeh_powernv_init - Register platform dependent EEH operations
1678  *
1679  * EEH initialization on powernv platform. This function should be
1680  * called before any EEH related functions.
1681  */
1682 static int __init eeh_powernv_init(void)
1683 {
1684 	int ret = -EINVAL;
1685 
1686 	ret = eeh_ops_register(&pnv_eeh_ops);
1687 	if (!ret)
1688 		pr_info("EEH: PowerNV platform initialized\n");
1689 	else
1690 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
1691 
1692 	return ret;
1693 }
1694 machine_early_initcall(powernv, eeh_powernv_init);
1695