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 #include "../../../../drivers/pci/pci.h"
38 
39 static int eeh_event_irq = -EINVAL;
40 
41 void pnv_pcibios_bus_add_device(struct pci_dev *pdev)
42 {
43 	struct pci_dn *pdn = pci_get_pdn(pdev);
44 
45 	if (eeh_has_flag(EEH_FORCE_DISABLED))
46 		return;
47 
48 	dev_dbg(&pdev->dev, "EEH: Setting up device\n");
49 	eeh_add_device_early(pdn);
50 	eeh_add_device_late(pdev);
51 	eeh_sysfs_add_device(pdev);
52 }
53 
54 static int pnv_eeh_init(void)
55 {
56 	struct pci_controller *hose;
57 	struct pnv_phb *phb;
58 	int max_diag_size = PNV_PCI_DIAG_BUF_SIZE;
59 
60 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
61 		pr_warn("%s: OPAL is required !\n",
62 			__func__);
63 		return -EINVAL;
64 	}
65 
66 	/* Set probe mode */
67 	eeh_add_flag(EEH_PROBE_MODE_DEV);
68 
69 	/*
70 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
71 	 * doesn't do that. So we have to selectively enable I/O
72 	 * prior to collecting error log.
73 	 */
74 	list_for_each_entry(hose, &hose_list, list_node) {
75 		phb = hose->private_data;
76 
77 		if (phb->model == PNV_PHB_MODEL_P7IOC)
78 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
79 
80 		if (phb->diag_data_size > max_diag_size)
81 			max_diag_size = phb->diag_data_size;
82 
83 		/*
84 		 * PE#0 should be regarded as valid by EEH core
85 		 * if it's not the reserved one. Currently, we
86 		 * have the reserved PE#255 and PE#127 for PHB3
87 		 * and P7IOC separately. So we should regard
88 		 * PE#0 as valid for PHB3 and P7IOC.
89 		 */
90 		if (phb->ioda.reserved_pe_idx != 0)
91 			eeh_add_flag(EEH_VALID_PE_ZERO);
92 
93 		break;
94 	}
95 
96 	eeh_set_pe_aux_size(max_diag_size);
97 	ppc_md.pcibios_bus_add_device = pnv_pcibios_bus_add_device;
98 
99 	return 0;
100 }
101 
102 static irqreturn_t pnv_eeh_event(int irq, void *data)
103 {
104 	/*
105 	 * We simply send a special EEH event if EEH has been
106 	 * enabled. We don't care about EEH events until we've
107 	 * finished processing the outstanding ones. Event processing
108 	 * gets unmasked in next_error() if EEH is enabled.
109 	 */
110 	disable_irq_nosync(irq);
111 
112 	if (eeh_enabled())
113 		eeh_send_failure_event(NULL);
114 
115 	return IRQ_HANDLED;
116 }
117 
118 #ifdef CONFIG_DEBUG_FS
119 static ssize_t pnv_eeh_ei_write(struct file *filp,
120 				const char __user *user_buf,
121 				size_t count, loff_t *ppos)
122 {
123 	struct pci_controller *hose = filp->private_data;
124 	struct eeh_pe *pe;
125 	int pe_no, type, func;
126 	unsigned long addr, mask;
127 	char buf[50];
128 	int ret;
129 
130 	if (!eeh_ops || !eeh_ops->err_inject)
131 		return -ENXIO;
132 
133 	/* Copy over argument buffer */
134 	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
135 	if (!ret)
136 		return -EFAULT;
137 
138 	/* Retrieve parameters */
139 	ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
140 		     &pe_no, &type, &func, &addr, &mask);
141 	if (ret != 5)
142 		return -EINVAL;
143 
144 	/* Retrieve PE */
145 	pe = eeh_pe_get(hose, pe_no, 0);
146 	if (!pe)
147 		return -ENODEV;
148 
149 	/* Do error injection */
150 	ret = eeh_ops->err_inject(pe, type, func, addr, mask);
151 	return ret < 0 ? ret : count;
152 }
153 
154 static const struct file_operations pnv_eeh_ei_fops = {
155 	.open	= simple_open,
156 	.llseek	= no_llseek,
157 	.write	= pnv_eeh_ei_write,
158 };
159 
160 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val)
161 {
162 	struct pci_controller *hose = data;
163 	struct pnv_phb *phb = hose->private_data;
164 
165 	out_be64(phb->regs + offset, val);
166 	return 0;
167 }
168 
169 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val)
170 {
171 	struct pci_controller *hose = data;
172 	struct pnv_phb *phb = hose->private_data;
173 
174 	*val = in_be64(phb->regs + offset);
175 	return 0;
176 }
177 
178 #define PNV_EEH_DBGFS_ENTRY(name, reg)				\
179 static int pnv_eeh_dbgfs_set_##name(void *data, u64 val)	\
180 {								\
181 	return pnv_eeh_dbgfs_set(data, reg, val);		\
182 }								\
183 								\
184 static int pnv_eeh_dbgfs_get_##name(void *data, u64 *val)	\
185 {								\
186 	return pnv_eeh_dbgfs_get(data, reg, val);		\
187 }								\
188 								\
189 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_dbgfs_ops_##name,		\
190 			pnv_eeh_dbgfs_get_##name,		\
191                         pnv_eeh_dbgfs_set_##name,		\
192 			"0x%llx\n")
193 
194 PNV_EEH_DBGFS_ENTRY(outb, 0xD10);
195 PNV_EEH_DBGFS_ENTRY(inbA, 0xD90);
196 PNV_EEH_DBGFS_ENTRY(inbB, 0xE10);
197 
198 #endif /* CONFIG_DEBUG_FS */
199 
200 void pnv_eeh_enable_phbs(void)
201 {
202 	struct pci_controller *hose;
203 	struct pnv_phb *phb;
204 
205 	list_for_each_entry(hose, &hose_list, list_node) {
206 		phb = hose->private_data;
207 		/*
208 		 * If EEH is enabled, we're going to rely on that.
209 		 * Otherwise, we restore to conventional mechanism
210 		 * to clear frozen PE during PCI config access.
211 		 */
212 		if (eeh_enabled())
213 			phb->flags |= PNV_PHB_FLAG_EEH;
214 		else
215 			phb->flags &= ~PNV_PHB_FLAG_EEH;
216 	}
217 }
218 
219 /**
220  * pnv_eeh_post_init - EEH platform dependent post initialization
221  *
222  * EEH platform dependent post initialization on powernv. When
223  * the function is called, the EEH PEs and devices should have
224  * been built. If the I/O cache staff has been built, EEH is
225  * ready to supply service.
226  */
227 int pnv_eeh_post_init(void)
228 {
229 	struct pci_controller *hose;
230 	struct pnv_phb *phb;
231 	int ret = 0;
232 
233 	eeh_show_enabled();
234 
235 	/* Register OPAL event notifier */
236 	eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
237 	if (eeh_event_irq < 0) {
238 		pr_err("%s: Can't register OPAL event interrupt (%d)\n",
239 		       __func__, eeh_event_irq);
240 		return eeh_event_irq;
241 	}
242 
243 	ret = request_irq(eeh_event_irq, pnv_eeh_event,
244 			  IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
245 	if (ret < 0) {
246 		irq_dispose_mapping(eeh_event_irq);
247 		pr_err("%s: Can't request OPAL event interrupt (%d)\n",
248 		       __func__, eeh_event_irq);
249 		return ret;
250 	}
251 
252 	if (!eeh_enabled())
253 		disable_irq(eeh_event_irq);
254 
255 	pnv_eeh_enable_phbs();
256 
257 	list_for_each_entry(hose, &hose_list, list_node) {
258 		phb = hose->private_data;
259 
260 		/* Create debugfs entries */
261 #ifdef CONFIG_DEBUG_FS
262 		if (phb->has_dbgfs || !phb->dbgfs)
263 			continue;
264 
265 		phb->has_dbgfs = 1;
266 		debugfs_create_file("err_injct", 0200,
267 				    phb->dbgfs, hose,
268 				    &pnv_eeh_ei_fops);
269 
270 		debugfs_create_file("err_injct_outbound", 0600,
271 				    phb->dbgfs, hose,
272 				    &pnv_eeh_dbgfs_ops_outb);
273 		debugfs_create_file("err_injct_inboundA", 0600,
274 				    phb->dbgfs, hose,
275 				    &pnv_eeh_dbgfs_ops_inbA);
276 		debugfs_create_file("err_injct_inboundB", 0600,
277 				    phb->dbgfs, hose,
278 				    &pnv_eeh_dbgfs_ops_inbB);
279 #endif /* CONFIG_DEBUG_FS */
280 	}
281 
282 	return ret;
283 }
284 
285 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
286 {
287 	int pos = PCI_CAPABILITY_LIST;
288 	int cnt = 48;   /* Maximal number of capabilities */
289 	u32 status, id;
290 
291 	if (!pdn)
292 		return 0;
293 
294 	/* Check if the device supports capabilities */
295 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
296 	if (!(status & PCI_STATUS_CAP_LIST))
297 		return 0;
298 
299 	while (cnt--) {
300 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
301 		if (pos < 0x40)
302 			break;
303 
304 		pos &= ~3;
305 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
306 		if (id == 0xff)
307 			break;
308 
309 		/* Found */
310 		if (id == cap)
311 			return pos;
312 
313 		/* Next one */
314 		pos += PCI_CAP_LIST_NEXT;
315 	}
316 
317 	return 0;
318 }
319 
320 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
321 {
322 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
323 	u32 header;
324 	int pos = 256, ttl = (4096 - 256) / 8;
325 
326 	if (!edev || !edev->pcie_cap)
327 		return 0;
328 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
329 		return 0;
330 	else if (!header)
331 		return 0;
332 
333 	while (ttl-- > 0) {
334 		if (PCI_EXT_CAP_ID(header) == cap && pos)
335 			return pos;
336 
337 		pos = PCI_EXT_CAP_NEXT(header);
338 		if (pos < 256)
339 			break;
340 
341 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
342 			break;
343 	}
344 
345 	return 0;
346 }
347 
348 /**
349  * pnv_eeh_probe - Do probe on PCI device
350  * @pdn: PCI device node
351  * @data: unused
352  *
353  * When EEH module is installed during system boot, all PCI devices
354  * are checked one by one to see if it supports EEH. The function
355  * is introduced for the purpose. By default, EEH has been enabled
356  * on all PCI devices. That's to say, we only need do necessary
357  * initialization on the corresponding eeh device and create PE
358  * accordingly.
359  *
360  * It's notable that's unsafe to retrieve the EEH device through
361  * the corresponding PCI device. During the PCI device hotplug, which
362  * was possiblly triggered by EEH core, the binding between EEH device
363  * and the PCI device isn't built yet.
364  */
365 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
366 {
367 	struct pci_controller *hose = pdn->phb;
368 	struct pnv_phb *phb = hose->private_data;
369 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
370 	uint32_t pcie_flags;
371 	int ret;
372 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
373 
374 	/*
375 	 * When probing the root bridge, which doesn't have any
376 	 * subordinate PCI devices. We don't have OF node for
377 	 * the root bridge. So it's not reasonable to continue
378 	 * the probing.
379 	 */
380 	if (!edev || edev->pe)
381 		return NULL;
382 
383 	/* Skip for PCI-ISA bridge */
384 	if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
385 		return NULL;
386 
387 	eeh_edev_dbg(edev, "Probing device\n");
388 
389 	/* Initialize eeh device */
390 	edev->class_code = pdn->class_code;
391 	edev->mode	&= 0xFFFFFF00;
392 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
393 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
394 	edev->af_cap   = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF);
395 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
396 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
397 		edev->mode |= EEH_DEV_BRIDGE;
398 		if (edev->pcie_cap) {
399 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
400 					 2, &pcie_flags);
401 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
402 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
403 				edev->mode |= EEH_DEV_ROOT_PORT;
404 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
405 				edev->mode |= EEH_DEV_DS_PORT;
406 		}
407 	}
408 
409 	edev->pe_config_addr = phb->ioda.pe_rmap[config_addr];
410 
411 	/* Create PE */
412 	ret = eeh_add_to_parent_pe(edev);
413 	if (ret) {
414 		eeh_edev_warn(edev, "Failed to add device to PE (code %d)\n", ret);
415 		return NULL;
416 	}
417 
418 	/*
419 	 * If the PE contains any one of following adapters, the
420 	 * PCI config space can't be accessed when dumping EEH log.
421 	 * Otherwise, we will run into fenced PHB caused by shortage
422 	 * of outbound credits in the adapter. The PCI config access
423 	 * should be blocked until PE reset. MMIO access is dropped
424 	 * by hardware certainly. In order to drop PCI config requests,
425 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
426 	 * will be checked in the backend for PE state retrival. If
427 	 * the PE becomes frozen for the first time and the flag has
428 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
429 	 * that PE to block its config space.
430 	 *
431 	 * Broadcom BCM5718 2-ports NICs (14e4:1656)
432 	 * Broadcom Austin 4-ports NICs (14e4:1657)
433 	 * Broadcom Shiner 4-ports 1G NICs (14e4:168a)
434 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
435 	 */
436 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
437 	     pdn->device_id == 0x1656) ||
438 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
439 	     pdn->device_id == 0x1657) ||
440 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
441 	     pdn->device_id == 0x168a) ||
442 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
443 	     pdn->device_id == 0x168e))
444 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
445 
446 	/*
447 	 * Cache the PE primary bus, which can't be fetched when
448 	 * full hotplug is in progress. In that case, all child
449 	 * PCI devices of the PE are expected to be removed prior
450 	 * to PE reset.
451 	 */
452 	if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
453 		edev->pe->bus = pci_find_bus(hose->global_number,
454 					     pdn->busno);
455 		if (edev->pe->bus)
456 			edev->pe->state |= EEH_PE_PRI_BUS;
457 	}
458 
459 	/*
460 	 * Enable EEH explicitly so that we will do EEH check
461 	 * while accessing I/O stuff
462 	 */
463 	if (!eeh_has_flag(EEH_ENABLED)) {
464 		enable_irq(eeh_event_irq);
465 		pnv_eeh_enable_phbs();
466 		eeh_add_flag(EEH_ENABLED);
467 	}
468 
469 	/* Save memory bars */
470 	eeh_save_bars(edev);
471 
472 	eeh_edev_dbg(edev, "EEH enabled on device\n");
473 
474 	return NULL;
475 }
476 
477 /**
478  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
479  * @pe: EEH PE
480  * @option: operation to be issued
481  *
482  * The function is used to control the EEH functionality globally.
483  * Currently, following options are support according to PAPR:
484  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
485  */
486 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
487 {
488 	struct pci_controller *hose = pe->phb;
489 	struct pnv_phb *phb = hose->private_data;
490 	bool freeze_pe = false;
491 	int opt;
492 	s64 rc;
493 
494 	switch (option) {
495 	case EEH_OPT_DISABLE:
496 		return -EPERM;
497 	case EEH_OPT_ENABLE:
498 		return 0;
499 	case EEH_OPT_THAW_MMIO:
500 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
501 		break;
502 	case EEH_OPT_THAW_DMA:
503 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
504 		break;
505 	case EEH_OPT_FREEZE_PE:
506 		freeze_pe = true;
507 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
508 		break;
509 	default:
510 		pr_warn("%s: Invalid option %d\n", __func__, option);
511 		return -EINVAL;
512 	}
513 
514 	/* Freeze master and slave PEs if PHB supports compound PEs */
515 	if (freeze_pe) {
516 		if (phb->freeze_pe) {
517 			phb->freeze_pe(phb, pe->addr);
518 			return 0;
519 		}
520 
521 		rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt);
522 		if (rc != OPAL_SUCCESS) {
523 			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
524 				__func__, rc, phb->hose->global_number,
525 				pe->addr);
526 			return -EIO;
527 		}
528 
529 		return 0;
530 	}
531 
532 	/* Unfreeze master and slave PEs if PHB supports */
533 	if (phb->unfreeze_pe)
534 		return phb->unfreeze_pe(phb, pe->addr, opt);
535 
536 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt);
537 	if (rc != OPAL_SUCCESS) {
538 		pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n",
539 			__func__, rc, option, phb->hose->global_number,
540 			pe->addr);
541 		return -EIO;
542 	}
543 
544 	return 0;
545 }
546 
547 /**
548  * pnv_eeh_get_pe_addr - Retrieve PE address
549  * @pe: EEH PE
550  *
551  * Retrieve the PE address according to the given tranditional
552  * PCI BDF (Bus/Device/Function) address.
553  */
554 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe)
555 {
556 	return pe->addr;
557 }
558 
559 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
560 {
561 	struct pnv_phb *phb = pe->phb->private_data;
562 	s64 rc;
563 
564 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
565 					 phb->diag_data_size);
566 	if (rc != OPAL_SUCCESS)
567 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
568 			__func__, rc, pe->phb->global_number);
569 }
570 
571 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
572 {
573 	struct pnv_phb *phb = pe->phb->private_data;
574 	u8 fstate = 0;
575 	__be16 pcierr = 0;
576 	s64 rc;
577 	int result = 0;
578 
579 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
580 					pe->addr,
581 					&fstate,
582 					&pcierr,
583 					NULL);
584 	if (rc != OPAL_SUCCESS) {
585 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
586 			__func__, rc, phb->hose->global_number);
587 		return EEH_STATE_NOT_SUPPORT;
588 	}
589 
590 	/*
591 	 * Check PHB state. If the PHB is frozen for the
592 	 * first time, to dump the PHB diag-data.
593 	 */
594 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
595 		result = (EEH_STATE_MMIO_ACTIVE  |
596 			  EEH_STATE_DMA_ACTIVE   |
597 			  EEH_STATE_MMIO_ENABLED |
598 			  EEH_STATE_DMA_ENABLED);
599 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
600 		eeh_pe_mark_isolated(pe);
601 		pnv_eeh_get_phb_diag(pe);
602 
603 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
604 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
605 	}
606 
607 	return result;
608 }
609 
610 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
611 {
612 	struct pnv_phb *phb = pe->phb->private_data;
613 	u8 fstate = 0;
614 	__be16 pcierr = 0;
615 	s64 rc;
616 	int result;
617 
618 	/*
619 	 * We don't clobber hardware frozen state until PE
620 	 * reset is completed. In order to keep EEH core
621 	 * moving forward, we have to return operational
622 	 * state during PE reset.
623 	 */
624 	if (pe->state & EEH_PE_RESET) {
625 		result = (EEH_STATE_MMIO_ACTIVE  |
626 			  EEH_STATE_DMA_ACTIVE   |
627 			  EEH_STATE_MMIO_ENABLED |
628 			  EEH_STATE_DMA_ENABLED);
629 		return result;
630 	}
631 
632 	/*
633 	 * Fetch PE state from hardware. If the PHB
634 	 * supports compound PE, let it handle that.
635 	 */
636 	if (phb->get_pe_state) {
637 		fstate = phb->get_pe_state(phb, pe->addr);
638 	} else {
639 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
640 						pe->addr,
641 						&fstate,
642 						&pcierr,
643 						NULL);
644 		if (rc != OPAL_SUCCESS) {
645 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
646 				__func__, rc, phb->hose->global_number,
647 				pe->addr);
648 			return EEH_STATE_NOT_SUPPORT;
649 		}
650 	}
651 
652 	/* Figure out state */
653 	switch (fstate) {
654 	case OPAL_EEH_STOPPED_NOT_FROZEN:
655 		result = (EEH_STATE_MMIO_ACTIVE  |
656 			  EEH_STATE_DMA_ACTIVE   |
657 			  EEH_STATE_MMIO_ENABLED |
658 			  EEH_STATE_DMA_ENABLED);
659 		break;
660 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
661 		result = (EEH_STATE_DMA_ACTIVE |
662 			  EEH_STATE_DMA_ENABLED);
663 		break;
664 	case OPAL_EEH_STOPPED_DMA_FREEZE:
665 		result = (EEH_STATE_MMIO_ACTIVE |
666 			  EEH_STATE_MMIO_ENABLED);
667 		break;
668 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
669 		result = 0;
670 		break;
671 	case OPAL_EEH_STOPPED_RESET:
672 		result = EEH_STATE_RESET_ACTIVE;
673 		break;
674 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
675 		result = EEH_STATE_UNAVAILABLE;
676 		break;
677 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
678 		result = EEH_STATE_NOT_SUPPORT;
679 		break;
680 	default:
681 		result = EEH_STATE_NOT_SUPPORT;
682 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
683 			__func__, phb->hose->global_number,
684 			pe->addr, fstate);
685 	}
686 
687 	/*
688 	 * If PHB supports compound PE, to freeze all
689 	 * slave PEs for consistency.
690 	 *
691 	 * If the PE is switching to frozen state for the
692 	 * first time, to dump the PHB diag-data.
693 	 */
694 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
695 	    !(result & EEH_STATE_UNAVAILABLE) &&
696 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
697 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
698 	    !(pe->state & EEH_PE_ISOLATED)) {
699 		if (phb->freeze_pe)
700 			phb->freeze_pe(phb, pe->addr);
701 
702 		eeh_pe_mark_isolated(pe);
703 		pnv_eeh_get_phb_diag(pe);
704 
705 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
706 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
707 	}
708 
709 	return result;
710 }
711 
712 /**
713  * pnv_eeh_get_state - Retrieve PE state
714  * @pe: EEH PE
715  * @delay: delay while PE state is temporarily unavailable
716  *
717  * Retrieve the state of the specified PE. For IODA-compitable
718  * platform, it should be retrieved from IODA table. Therefore,
719  * we prefer passing down to hardware implementation to handle
720  * it.
721  */
722 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
723 {
724 	int ret;
725 
726 	if (pe->type & EEH_PE_PHB)
727 		ret = pnv_eeh_get_phb_state(pe);
728 	else
729 		ret = pnv_eeh_get_pe_state(pe);
730 
731 	if (!delay)
732 		return ret;
733 
734 	/*
735 	 * If the PE state is temporarily unavailable,
736 	 * to inform the EEH core delay for default
737 	 * period (1 second)
738 	 */
739 	*delay = 0;
740 	if (ret & EEH_STATE_UNAVAILABLE)
741 		*delay = 1000;
742 
743 	return ret;
744 }
745 
746 static s64 pnv_eeh_poll(unsigned long id)
747 {
748 	s64 rc = OPAL_HARDWARE;
749 
750 	while (1) {
751 		rc = opal_pci_poll(id);
752 		if (rc <= 0)
753 			break;
754 
755 		if (system_state < SYSTEM_RUNNING)
756 			udelay(1000 * rc);
757 		else
758 			msleep(rc);
759 	}
760 
761 	return rc;
762 }
763 
764 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
765 {
766 	struct pnv_phb *phb = hose->private_data;
767 	s64 rc = OPAL_HARDWARE;
768 
769 	pr_debug("%s: Reset PHB#%x, option=%d\n",
770 		 __func__, hose->global_number, option);
771 
772 	/* Issue PHB complete reset request */
773 	if (option == EEH_RESET_FUNDAMENTAL ||
774 	    option == EEH_RESET_HOT)
775 		rc = opal_pci_reset(phb->opal_id,
776 				    OPAL_RESET_PHB_COMPLETE,
777 				    OPAL_ASSERT_RESET);
778 	else if (option == EEH_RESET_DEACTIVATE)
779 		rc = opal_pci_reset(phb->opal_id,
780 				    OPAL_RESET_PHB_COMPLETE,
781 				    OPAL_DEASSERT_RESET);
782 	if (rc < 0)
783 		goto out;
784 
785 	/*
786 	 * Poll state of the PHB until the request is done
787 	 * successfully. The PHB reset is usually PHB complete
788 	 * reset followed by hot reset on root bus. So we also
789 	 * need the PCI bus settlement delay.
790 	 */
791 	if (rc > 0)
792 		rc = pnv_eeh_poll(phb->opal_id);
793 	if (option == EEH_RESET_DEACTIVATE) {
794 		if (system_state < SYSTEM_RUNNING)
795 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
796 		else
797 			msleep(EEH_PE_RST_SETTLE_TIME);
798 	}
799 out:
800 	if (rc != OPAL_SUCCESS)
801 		return -EIO;
802 
803 	return 0;
804 }
805 
806 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
807 {
808 	struct pnv_phb *phb = hose->private_data;
809 	s64 rc = OPAL_HARDWARE;
810 
811 	pr_debug("%s: Reset PHB#%x, option=%d\n",
812 		 __func__, hose->global_number, option);
813 
814 	/*
815 	 * During the reset deassert time, we needn't care
816 	 * the reset scope because the firmware does nothing
817 	 * for fundamental or hot reset during deassert phase.
818 	 */
819 	if (option == EEH_RESET_FUNDAMENTAL)
820 		rc = opal_pci_reset(phb->opal_id,
821 				    OPAL_RESET_PCI_FUNDAMENTAL,
822 				    OPAL_ASSERT_RESET);
823 	else if (option == EEH_RESET_HOT)
824 		rc = opal_pci_reset(phb->opal_id,
825 				    OPAL_RESET_PCI_HOT,
826 				    OPAL_ASSERT_RESET);
827 	else if (option == EEH_RESET_DEACTIVATE)
828 		rc = opal_pci_reset(phb->opal_id,
829 				    OPAL_RESET_PCI_HOT,
830 				    OPAL_DEASSERT_RESET);
831 	if (rc < 0)
832 		goto out;
833 
834 	/* Poll state of the PHB until the request is done */
835 	if (rc > 0)
836 		rc = pnv_eeh_poll(phb->opal_id);
837 	if (option == EEH_RESET_DEACTIVATE)
838 		msleep(EEH_PE_RST_SETTLE_TIME);
839 out:
840 	if (rc != OPAL_SUCCESS)
841 		return -EIO;
842 
843 	return 0;
844 }
845 
846 static int __pnv_eeh_bridge_reset(struct pci_dev *dev, int option)
847 {
848 	struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
849 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
850 	int aer = edev ? edev->aer_cap : 0;
851 	u32 ctrl;
852 
853 	pr_debug("%s: Secondary Reset PCI bus %04x:%02x with option %d\n",
854 		 __func__, pci_domain_nr(dev->bus),
855 		 dev->bus->number, option);
856 
857 	switch (option) {
858 	case EEH_RESET_FUNDAMENTAL:
859 	case EEH_RESET_HOT:
860 		/* Don't report linkDown event */
861 		if (aer) {
862 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
863 					     4, &ctrl);
864 			ctrl |= PCI_ERR_UNC_SURPDN;
865 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
866 					      4, ctrl);
867 		}
868 
869 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
870 		ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
871 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
872 
873 		msleep(EEH_PE_RST_HOLD_TIME);
874 		break;
875 	case EEH_RESET_DEACTIVATE:
876 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
877 		ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
878 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
879 
880 		msleep(EEH_PE_RST_SETTLE_TIME);
881 
882 		/* Continue reporting linkDown event */
883 		if (aer) {
884 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
885 					     4, &ctrl);
886 			ctrl &= ~PCI_ERR_UNC_SURPDN;
887 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
888 					      4, ctrl);
889 		}
890 
891 		break;
892 	}
893 
894 	return 0;
895 }
896 
897 static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option)
898 {
899 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
900 	struct pnv_phb *phb = hose->private_data;
901 	struct device_node *dn = pci_device_to_OF_node(pdev);
902 	uint64_t id = PCI_SLOT_ID(phb->opal_id,
903 				  (pdev->bus->number << 8) | pdev->devfn);
904 	uint8_t scope;
905 	int64_t rc;
906 
907 	/* Hot reset to the bus if firmware cannot handle */
908 	if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL))
909 		return __pnv_eeh_bridge_reset(pdev, option);
910 
911 	pr_debug("%s: FW reset PCI bus %04x:%02x with option %d\n",
912 		 __func__, pci_domain_nr(pdev->bus),
913 		 pdev->bus->number, option);
914 
915 	switch (option) {
916 	case EEH_RESET_FUNDAMENTAL:
917 		scope = OPAL_RESET_PCI_FUNDAMENTAL;
918 		break;
919 	case EEH_RESET_HOT:
920 		scope = OPAL_RESET_PCI_HOT;
921 		break;
922 	case EEH_RESET_DEACTIVATE:
923 		return 0;
924 	default:
925 		dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n",
926 			__func__, option);
927 		return -EINVAL;
928 	}
929 
930 	rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET);
931 	if (rc <= OPAL_SUCCESS)
932 		goto out;
933 
934 	rc = pnv_eeh_poll(id);
935 out:
936 	return (rc == OPAL_SUCCESS) ? 0 : -EIO;
937 }
938 
939 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
940 {
941 	struct pci_controller *hose;
942 
943 	if (pci_is_root_bus(dev->bus)) {
944 		hose = pci_bus_to_host(dev->bus);
945 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
946 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
947 	} else {
948 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
949 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
950 	}
951 }
952 
953 static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type,
954 				     int pos, u16 mask)
955 {
956 	int i, status = 0;
957 
958 	/* Wait for Transaction Pending bit to be cleared */
959 	for (i = 0; i < 4; i++) {
960 		eeh_ops->read_config(pdn, pos, 2, &status);
961 		if (!(status & mask))
962 			return;
963 
964 		msleep((1 << i) * 100);
965 	}
966 
967 	pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n",
968 		__func__, type,
969 		pdn->phb->global_number, pdn->busno,
970 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
971 }
972 
973 static int pnv_eeh_do_flr(struct pci_dn *pdn, int option)
974 {
975 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
976 	u32 reg = 0;
977 
978 	if (WARN_ON(!edev->pcie_cap))
979 		return -ENOTTY;
980 
981 	eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP, 4, &reg);
982 	if (!(reg & PCI_EXP_DEVCAP_FLR))
983 		return -ENOTTY;
984 
985 	switch (option) {
986 	case EEH_RESET_HOT:
987 	case EEH_RESET_FUNDAMENTAL:
988 		pnv_eeh_wait_for_pending(pdn, "",
989 					 edev->pcie_cap + PCI_EXP_DEVSTA,
990 					 PCI_EXP_DEVSTA_TRPND);
991 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
992 				     4, &reg);
993 		reg |= PCI_EXP_DEVCTL_BCR_FLR;
994 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
995 				      4, reg);
996 		msleep(EEH_PE_RST_HOLD_TIME);
997 		break;
998 	case EEH_RESET_DEACTIVATE:
999 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
1000 				     4, &reg);
1001 		reg &= ~PCI_EXP_DEVCTL_BCR_FLR;
1002 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
1003 				      4, reg);
1004 		msleep(EEH_PE_RST_SETTLE_TIME);
1005 		break;
1006 	}
1007 
1008 	return 0;
1009 }
1010 
1011 static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option)
1012 {
1013 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1014 	u32 cap = 0;
1015 
1016 	if (WARN_ON(!edev->af_cap))
1017 		return -ENOTTY;
1018 
1019 	eeh_ops->read_config(pdn, edev->af_cap + PCI_AF_CAP, 1, &cap);
1020 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
1021 		return -ENOTTY;
1022 
1023 	switch (option) {
1024 	case EEH_RESET_HOT:
1025 	case EEH_RESET_FUNDAMENTAL:
1026 		/*
1027 		 * Wait for Transaction Pending bit to clear. A word-aligned
1028 		 * test is used, so we use the conrol offset rather than status
1029 		 * and shift the test bit to match.
1030 		 */
1031 		pnv_eeh_wait_for_pending(pdn, "AF",
1032 					 edev->af_cap + PCI_AF_CTRL,
1033 					 PCI_AF_STATUS_TP << 8);
1034 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL,
1035 				      1, PCI_AF_CTRL_FLR);
1036 		msleep(EEH_PE_RST_HOLD_TIME);
1037 		break;
1038 	case EEH_RESET_DEACTIVATE:
1039 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL, 1, 0);
1040 		msleep(EEH_PE_RST_SETTLE_TIME);
1041 		break;
1042 	}
1043 
1044 	return 0;
1045 }
1046 
1047 static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option)
1048 {
1049 	struct eeh_dev *edev;
1050 	struct pci_dn *pdn;
1051 	int ret;
1052 
1053 	/* The VF PE should have only one child device */
1054 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, entry);
1055 	pdn = eeh_dev_to_pdn(edev);
1056 	if (!pdn)
1057 		return -ENXIO;
1058 
1059 	ret = pnv_eeh_do_flr(pdn, option);
1060 	if (!ret)
1061 		return ret;
1062 
1063 	return pnv_eeh_do_af_flr(pdn, option);
1064 }
1065 
1066 /**
1067  * pnv_eeh_reset - Reset the specified PE
1068  * @pe: EEH PE
1069  * @option: reset option
1070  *
1071  * Do reset on the indicated PE. For PCI bus sensitive PE,
1072  * we need to reset the parent p2p bridge. The PHB has to
1073  * be reinitialized if the p2p bridge is root bridge. For
1074  * PCI device sensitive PE, we will try to reset the device
1075  * through FLR. For now, we don't have OPAL APIs to do HARD
1076  * reset yet, so all reset would be SOFT (HOT) reset.
1077  */
1078 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
1079 {
1080 	struct pci_controller *hose = pe->phb;
1081 	struct pnv_phb *phb;
1082 	struct pci_bus *bus;
1083 	int64_t rc;
1084 
1085 	/*
1086 	 * For PHB reset, we always have complete reset. For those PEs whose
1087 	 * primary bus derived from root complex (root bus) or root port
1088 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
1089 	 * For other PEs, we always have hot reset on the PE primary bus.
1090 	 *
1091 	 * Here, we have different design to pHyp, which always clear the
1092 	 * frozen state during PE reset. However, the good idea here from
1093 	 * benh is to keep frozen state before we get PE reset done completely
1094 	 * (until BAR restore). With the frozen state, HW drops illegal IO
1095 	 * or MMIO access, which can incur recrusive frozen PE during PE
1096 	 * reset. The side effect is that EEH core has to clear the frozen
1097 	 * state explicitly after BAR restore.
1098 	 */
1099 	if (pe->type & EEH_PE_PHB)
1100 		return pnv_eeh_phb_reset(hose, option);
1101 
1102 	/*
1103 	 * The frozen PE might be caused by PAPR error injection
1104 	 * registers, which are expected to be cleared after hitting
1105 	 * frozen PE as stated in the hardware spec. Unfortunately,
1106 	 * that's not true on P7IOC. So we have to clear it manually
1107 	 * to avoid recursive EEH errors during recovery.
1108 	 */
1109 	phb = hose->private_data;
1110 	if (phb->model == PNV_PHB_MODEL_P7IOC &&
1111 	    (option == EEH_RESET_HOT ||
1112 	     option == EEH_RESET_FUNDAMENTAL)) {
1113 		rc = opal_pci_reset(phb->opal_id,
1114 				    OPAL_RESET_PHB_ERROR,
1115 				    OPAL_ASSERT_RESET);
1116 		if (rc != OPAL_SUCCESS) {
1117 			pr_warn("%s: Failure %lld clearing error injection registers\n",
1118 				__func__, rc);
1119 			return -EIO;
1120 		}
1121 	}
1122 
1123 	if (pe->type & EEH_PE_VF)
1124 		return pnv_eeh_reset_vf_pe(pe, option);
1125 
1126 	bus = eeh_pe_bus_get(pe);
1127 	if (!bus) {
1128 		pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
1129 			__func__, pe->phb->global_number, pe->addr);
1130 		return -EIO;
1131 	}
1132 
1133 	if (pci_is_root_bus(bus))
1134 		return pnv_eeh_root_reset(hose, option);
1135 
1136 	/*
1137 	 * For hot resets try use the generic PCI error recovery reset
1138 	 * functions. These correctly handles the case where the secondary
1139 	 * bus is behind a hotplug slot and it will use the slot provided
1140 	 * reset methods to prevent spurious hotplug events during the reset.
1141 	 *
1142 	 * Fundemental resets need to be handled internally to EEH since the
1143 	 * PCI core doesn't really have a concept of a fundemental reset,
1144 	 * mainly because there's no standard way to generate one. Only a
1145 	 * few devices require an FRESET so it should be fine.
1146 	 */
1147 	if (option != EEH_RESET_FUNDAMENTAL) {
1148 		/*
1149 		 * NB: Skiboot and pnv_eeh_bridge_reset() also no-op the
1150 		 *     de-assert step. It's like the OPAL reset API was
1151 		 *     poorly designed or something...
1152 		 */
1153 		if (option == EEH_RESET_DEACTIVATE)
1154 			return 0;
1155 
1156 		rc = pci_bus_error_reset(bus->self);
1157 		if (!rc)
1158 			return 0;
1159 	}
1160 
1161 	/* otherwise, use the generic bridge reset. this might call into FW */
1162 	if (pci_is_root_bus(bus->parent))
1163 		return pnv_eeh_root_reset(hose, option);
1164 	return pnv_eeh_bridge_reset(bus->self, option);
1165 }
1166 
1167 /**
1168  * pnv_eeh_get_log - Retrieve error log
1169  * @pe: EEH PE
1170  * @severity: temporary or permanent error log
1171  * @drv_log: driver log to be combined with retrieved error log
1172  * @len: length of driver log
1173  *
1174  * Retrieve the temporary or permanent error from the PE.
1175  */
1176 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1177 			   char *drv_log, unsigned long len)
1178 {
1179 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1180 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1181 
1182 	return 0;
1183 }
1184 
1185 /**
1186  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1187  * @pe: EEH PE
1188  *
1189  * The function will be called to reconfigure the bridges included
1190  * in the specified PE so that the mulfunctional PE would be recovered
1191  * again.
1192  */
1193 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1194 {
1195 	return 0;
1196 }
1197 
1198 /**
1199  * pnv_pe_err_inject - Inject specified error to the indicated PE
1200  * @pe: the indicated PE
1201  * @type: error type
1202  * @func: specific error type
1203  * @addr: address
1204  * @mask: address mask
1205  *
1206  * The routine is called to inject specified error, which is
1207  * determined by @type and @func, to the indicated PE for
1208  * testing purpose.
1209  */
1210 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1211 			      unsigned long addr, unsigned long mask)
1212 {
1213 	struct pci_controller *hose = pe->phb;
1214 	struct pnv_phb *phb = hose->private_data;
1215 	s64 rc;
1216 
1217 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1218 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1219 		pr_warn("%s: Invalid error type %d\n",
1220 			__func__, type);
1221 		return -ERANGE;
1222 	}
1223 
1224 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1225 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1226 		pr_warn("%s: Invalid error function %d\n",
1227 			__func__, func);
1228 		return -ERANGE;
1229 	}
1230 
1231 	/* Firmware supports error injection ? */
1232 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1233 		pr_warn("%s: Firmware doesn't support error injection\n",
1234 			__func__);
1235 		return -ENXIO;
1236 	}
1237 
1238 	/* Do error injection */
1239 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1240 				 type, func, addr, mask);
1241 	if (rc != OPAL_SUCCESS) {
1242 		pr_warn("%s: Failure %lld injecting error "
1243 			"%d-%d to PHB#%x-PE#%x\n",
1244 			__func__, rc, type, func,
1245 			hose->global_number, pe->addr);
1246 		return -EIO;
1247 	}
1248 
1249 	return 0;
1250 }
1251 
1252 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1253 {
1254 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1255 
1256 	if (!edev || !edev->pe)
1257 		return false;
1258 
1259 	/*
1260 	 * We will issue FLR or AF FLR to all VFs, which are contained
1261 	 * in VF PE. It relies on the EEH PCI config accessors. So we
1262 	 * can't block them during the window.
1263 	 */
1264 	if (edev->physfn && (edev->pe->state & EEH_PE_RESET))
1265 		return false;
1266 
1267 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1268 		return true;
1269 
1270 	return false;
1271 }
1272 
1273 static int pnv_eeh_read_config(struct pci_dn *pdn,
1274 			       int where, int size, u32 *val)
1275 {
1276 	if (!pdn)
1277 		return PCIBIOS_DEVICE_NOT_FOUND;
1278 
1279 	if (pnv_eeh_cfg_blocked(pdn)) {
1280 		*val = 0xFFFFFFFF;
1281 		return PCIBIOS_SET_FAILED;
1282 	}
1283 
1284 	return pnv_pci_cfg_read(pdn, where, size, val);
1285 }
1286 
1287 static int pnv_eeh_write_config(struct pci_dn *pdn,
1288 				int where, int size, u32 val)
1289 {
1290 	if (!pdn)
1291 		return PCIBIOS_DEVICE_NOT_FOUND;
1292 
1293 	if (pnv_eeh_cfg_blocked(pdn))
1294 		return PCIBIOS_SET_FAILED;
1295 
1296 	return pnv_pci_cfg_write(pdn, where, size, val);
1297 }
1298 
1299 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1300 {
1301 	/* GEM */
1302 	if (data->gemXfir || data->gemRfir ||
1303 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1304 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1305 			be64_to_cpu(data->gemXfir),
1306 			be64_to_cpu(data->gemRfir),
1307 			be64_to_cpu(data->gemRirqfir),
1308 			be64_to_cpu(data->gemMask),
1309 			be64_to_cpu(data->gemRwof));
1310 
1311 	/* LEM */
1312 	if (data->lemFir || data->lemErrMask ||
1313 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1314 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1315 			be64_to_cpu(data->lemFir),
1316 			be64_to_cpu(data->lemErrMask),
1317 			be64_to_cpu(data->lemAction0),
1318 			be64_to_cpu(data->lemAction1),
1319 			be64_to_cpu(data->lemWof));
1320 }
1321 
1322 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1323 {
1324 	struct pnv_phb *phb = hose->private_data;
1325 	struct OpalIoP7IOCErrorData *data =
1326 		(struct OpalIoP7IOCErrorData*)phb->diag_data;
1327 	long rc;
1328 
1329 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1330 	if (rc != OPAL_SUCCESS) {
1331 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1332 			__func__, phb->hub_id, rc);
1333 		return;
1334 	}
1335 
1336 	switch (be16_to_cpu(data->type)) {
1337 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1338 		pr_info("P7IOC diag-data for RGC\n\n");
1339 		pnv_eeh_dump_hub_diag_common(data);
1340 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1341 			pr_info("  RGC: %016llx %016llx\n",
1342 				be64_to_cpu(data->rgc.rgcStatus),
1343 				be64_to_cpu(data->rgc.rgcLdcp));
1344 		break;
1345 	case OPAL_P7IOC_DIAG_TYPE_BI:
1346 		pr_info("P7IOC diag-data for BI %s\n\n",
1347 			data->bi.biDownbound ? "Downbound" : "Upbound");
1348 		pnv_eeh_dump_hub_diag_common(data);
1349 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1350 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1351 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1352 				be64_to_cpu(data->bi.biLdcp0),
1353 				be64_to_cpu(data->bi.biLdcp1),
1354 				be64_to_cpu(data->bi.biLdcp2),
1355 				be64_to_cpu(data->bi.biFenceStatus));
1356 		break;
1357 	case OPAL_P7IOC_DIAG_TYPE_CI:
1358 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1359 			data->ci.ciPort);
1360 		pnv_eeh_dump_hub_diag_common(data);
1361 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1362 			pr_info("  CI:  %016llx %016llx\n",
1363 				be64_to_cpu(data->ci.ciPortStatus),
1364 				be64_to_cpu(data->ci.ciPortLdcp));
1365 		break;
1366 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1367 		pr_info("P7IOC diag-data for MISC\n\n");
1368 		pnv_eeh_dump_hub_diag_common(data);
1369 		break;
1370 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1371 		pr_info("P7IOC diag-data for I2C\n\n");
1372 		pnv_eeh_dump_hub_diag_common(data);
1373 		break;
1374 	default:
1375 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1376 			__func__, phb->hub_id, data->type);
1377 	}
1378 }
1379 
1380 static int pnv_eeh_get_pe(struct pci_controller *hose,
1381 			  u16 pe_no, struct eeh_pe **pe)
1382 {
1383 	struct pnv_phb *phb = hose->private_data;
1384 	struct pnv_ioda_pe *pnv_pe;
1385 	struct eeh_pe *dev_pe;
1386 
1387 	/*
1388 	 * If PHB supports compound PE, to fetch
1389 	 * the master PE because slave PE is invisible
1390 	 * to EEH core.
1391 	 */
1392 	pnv_pe = &phb->ioda.pe_array[pe_no];
1393 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1394 		pnv_pe = pnv_pe->master;
1395 		WARN_ON(!pnv_pe ||
1396 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1397 		pe_no = pnv_pe->pe_number;
1398 	}
1399 
1400 	/* Find the PE according to PE# */
1401 	dev_pe = eeh_pe_get(hose, pe_no, 0);
1402 	if (!dev_pe)
1403 		return -EEXIST;
1404 
1405 	/* Freeze the (compound) PE */
1406 	*pe = dev_pe;
1407 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1408 		phb->freeze_pe(phb, pe_no);
1409 
1410 	/*
1411 	 * At this point, we're sure the (compound) PE should
1412 	 * have been frozen. However, we still need poke until
1413 	 * hitting the frozen PE on top level.
1414 	 */
1415 	dev_pe = dev_pe->parent;
1416 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1417 		int ret;
1418 		ret = eeh_ops->get_state(dev_pe, NULL);
1419 		if (ret <= 0 || eeh_state_active(ret)) {
1420 			dev_pe = dev_pe->parent;
1421 			continue;
1422 		}
1423 
1424 		/* Frozen parent PE */
1425 		*pe = dev_pe;
1426 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1427 			phb->freeze_pe(phb, dev_pe->addr);
1428 
1429 		/* Next one */
1430 		dev_pe = dev_pe->parent;
1431 	}
1432 
1433 	return 0;
1434 }
1435 
1436 /**
1437  * pnv_eeh_next_error - Retrieve next EEH error to handle
1438  * @pe: Affected PE
1439  *
1440  * The function is expected to be called by EEH core while it gets
1441  * special EEH event (without binding PE). The function calls to
1442  * OPAL APIs for next error to handle. The informational error is
1443  * handled internally by platform. However, the dead IOC, dead PHB,
1444  * fenced PHB and frozen PE should be handled by EEH core eventually.
1445  */
1446 static int pnv_eeh_next_error(struct eeh_pe **pe)
1447 {
1448 	struct pci_controller *hose;
1449 	struct pnv_phb *phb;
1450 	struct eeh_pe *phb_pe, *parent_pe;
1451 	__be64 frozen_pe_no;
1452 	__be16 err_type, severity;
1453 	long rc;
1454 	int state, ret = EEH_NEXT_ERR_NONE;
1455 
1456 	/*
1457 	 * While running here, it's safe to purge the event queue. The
1458 	 * event should still be masked.
1459 	 */
1460 	eeh_remove_event(NULL, false);
1461 
1462 	list_for_each_entry(hose, &hose_list, list_node) {
1463 		/*
1464 		 * If the subordinate PCI buses of the PHB has been
1465 		 * removed or is exactly under error recovery, we
1466 		 * needn't take care of it any more.
1467 		 */
1468 		phb = hose->private_data;
1469 		phb_pe = eeh_phb_pe_get(hose);
1470 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1471 			continue;
1472 
1473 		rc = opal_pci_next_error(phb->opal_id,
1474 					 &frozen_pe_no, &err_type, &severity);
1475 		if (rc != OPAL_SUCCESS) {
1476 			pr_devel("%s: Invalid return value on "
1477 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1478 				 __func__, hose->global_number, rc);
1479 			continue;
1480 		}
1481 
1482 		/* If the PHB doesn't have error, stop processing */
1483 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1484 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1485 			pr_devel("%s: No error found on PHB#%x\n",
1486 				 __func__, hose->global_number);
1487 			continue;
1488 		}
1489 
1490 		/*
1491 		 * Processing the error. We're expecting the error with
1492 		 * highest priority reported upon multiple errors on the
1493 		 * specific PHB.
1494 		 */
1495 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1496 			__func__, be16_to_cpu(err_type),
1497 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1498 			hose->global_number);
1499 		switch (be16_to_cpu(err_type)) {
1500 		case OPAL_EEH_IOC_ERROR:
1501 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1502 				pr_err("EEH: dead IOC detected\n");
1503 				ret = EEH_NEXT_ERR_DEAD_IOC;
1504 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1505 				pr_info("EEH: IOC informative error "
1506 					"detected\n");
1507 				pnv_eeh_get_and_dump_hub_diag(hose);
1508 				ret = EEH_NEXT_ERR_NONE;
1509 			}
1510 
1511 			break;
1512 		case OPAL_EEH_PHB_ERROR:
1513 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1514 				*pe = phb_pe;
1515 				pr_err("EEH: dead PHB#%x detected, "
1516 				       "location: %s\n",
1517 					hose->global_number,
1518 					eeh_pe_loc_get(phb_pe));
1519 				ret = EEH_NEXT_ERR_DEAD_PHB;
1520 			} else if (be16_to_cpu(severity) ==
1521 				   OPAL_EEH_SEV_PHB_FENCED) {
1522 				*pe = phb_pe;
1523 				pr_err("EEH: Fenced PHB#%x detected, "
1524 				       "location: %s\n",
1525 					hose->global_number,
1526 					eeh_pe_loc_get(phb_pe));
1527 				ret = EEH_NEXT_ERR_FENCED_PHB;
1528 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1529 				pr_info("EEH: PHB#%x informative error "
1530 					"detected, location: %s\n",
1531 					hose->global_number,
1532 					eeh_pe_loc_get(phb_pe));
1533 				pnv_eeh_get_phb_diag(phb_pe);
1534 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1535 				ret = EEH_NEXT_ERR_NONE;
1536 			}
1537 
1538 			break;
1539 		case OPAL_EEH_PE_ERROR:
1540 			/*
1541 			 * If we can't find the corresponding PE, we
1542 			 * just try to unfreeze.
1543 			 */
1544 			if (pnv_eeh_get_pe(hose,
1545 				be64_to_cpu(frozen_pe_no), pe)) {
1546 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1547 					hose->global_number, be64_to_cpu(frozen_pe_no));
1548 				pr_info("EEH: PHB location: %s\n",
1549 					eeh_pe_loc_get(phb_pe));
1550 
1551 				/* Dump PHB diag-data */
1552 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1553 					phb->diag_data, phb->diag_data_size);
1554 				if (rc == OPAL_SUCCESS)
1555 					pnv_pci_dump_phb_diag_data(hose,
1556 							phb->diag_data);
1557 
1558 				/* Try best to clear it */
1559 				opal_pci_eeh_freeze_clear(phb->opal_id,
1560 					be64_to_cpu(frozen_pe_no),
1561 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1562 				ret = EEH_NEXT_ERR_NONE;
1563 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1564 				   eeh_pe_passed(*pe)) {
1565 				ret = EEH_NEXT_ERR_NONE;
1566 			} else {
1567 				pr_err("EEH: Frozen PE#%x "
1568 				       "on PHB#%x detected\n",
1569 				       (*pe)->addr,
1570 					(*pe)->phb->global_number);
1571 				pr_err("EEH: PE location: %s, "
1572 				       "PHB location: %s\n",
1573 				       eeh_pe_loc_get(*pe),
1574 				       eeh_pe_loc_get(phb_pe));
1575 				ret = EEH_NEXT_ERR_FROZEN_PE;
1576 			}
1577 
1578 			break;
1579 		default:
1580 			pr_warn("%s: Unexpected error type %d\n",
1581 				__func__, be16_to_cpu(err_type));
1582 		}
1583 
1584 		/*
1585 		 * EEH core will try recover from fenced PHB or
1586 		 * frozen PE. In the time for frozen PE, EEH core
1587 		 * enable IO path for that before collecting logs,
1588 		 * but it ruins the site. So we have to dump the
1589 		 * log in advance here.
1590 		 */
1591 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1592 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1593 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1594 			eeh_pe_mark_isolated(*pe);
1595 			pnv_eeh_get_phb_diag(*pe);
1596 
1597 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1598 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1599 							   (*pe)->data);
1600 		}
1601 
1602 		/*
1603 		 * We probably have the frozen parent PE out there and
1604 		 * we need have to handle frozen parent PE firstly.
1605 		 */
1606 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1607 			parent_pe = (*pe)->parent;
1608 			while (parent_pe) {
1609 				/* Hit the ceiling ? */
1610 				if (parent_pe->type & EEH_PE_PHB)
1611 					break;
1612 
1613 				/* Frozen parent PE ? */
1614 				state = eeh_ops->get_state(parent_pe, NULL);
1615 				if (state > 0 && !eeh_state_active(state))
1616 					*pe = parent_pe;
1617 
1618 				/* Next parent level */
1619 				parent_pe = parent_pe->parent;
1620 			}
1621 
1622 			/* We possibly migrate to another PE */
1623 			eeh_pe_mark_isolated(*pe);
1624 		}
1625 
1626 		/*
1627 		 * If we have no errors on the specific PHB or only
1628 		 * informative error there, we continue poking it.
1629 		 * Otherwise, we need actions to be taken by upper
1630 		 * layer.
1631 		 */
1632 		if (ret > EEH_NEXT_ERR_INF)
1633 			break;
1634 	}
1635 
1636 	/* Unmask the event */
1637 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1638 		enable_irq(eeh_event_irq);
1639 
1640 	return ret;
1641 }
1642 
1643 static int pnv_eeh_restore_config(struct pci_dn *pdn)
1644 {
1645 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1646 	struct pnv_phb *phb;
1647 	s64 ret = 0;
1648 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
1649 
1650 	if (!edev)
1651 		return -EEXIST;
1652 
1653 	/*
1654 	 * We have to restore the PCI config space after reset since the
1655 	 * firmware can't see SRIOV VFs.
1656 	 *
1657 	 * FIXME: The MPS, error routing rules, timeout setting are worthy
1658 	 * to be exported by firmware in extendible way.
1659 	 */
1660 	if (edev->physfn) {
1661 		ret = eeh_restore_vf_config(pdn);
1662 	} else {
1663 		phb = pdn->phb->private_data;
1664 		ret = opal_pci_reinit(phb->opal_id,
1665 				      OPAL_REINIT_PCI_DEV, config_addr);
1666 	}
1667 
1668 	if (ret) {
1669 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1670 			__func__, config_addr, ret);
1671 		return -EIO;
1672 	}
1673 
1674 	return ret;
1675 }
1676 
1677 static struct eeh_ops pnv_eeh_ops = {
1678 	.name                   = "powernv",
1679 	.init                   = pnv_eeh_init,
1680 	.probe			= pnv_eeh_probe,
1681 	.set_option             = pnv_eeh_set_option,
1682 	.get_pe_addr            = pnv_eeh_get_pe_addr,
1683 	.get_state              = pnv_eeh_get_state,
1684 	.reset                  = pnv_eeh_reset,
1685 	.get_log                = pnv_eeh_get_log,
1686 	.configure_bridge       = pnv_eeh_configure_bridge,
1687 	.err_inject		= pnv_eeh_err_inject,
1688 	.read_config            = pnv_eeh_read_config,
1689 	.write_config           = pnv_eeh_write_config,
1690 	.next_error		= pnv_eeh_next_error,
1691 	.restore_config		= pnv_eeh_restore_config,
1692 	.notify_resume		= NULL
1693 };
1694 
1695 #ifdef CONFIG_PCI_IOV
1696 static void pnv_pci_fixup_vf_mps(struct pci_dev *pdev)
1697 {
1698 	struct pci_dn *pdn = pci_get_pdn(pdev);
1699 	int parent_mps;
1700 
1701 	if (!pdev->is_virtfn)
1702 		return;
1703 
1704 	/* Synchronize MPS for VF and PF */
1705 	parent_mps = pcie_get_mps(pdev->physfn);
1706 	if ((128 << pdev->pcie_mpss) >= parent_mps)
1707 		pcie_set_mps(pdev, parent_mps);
1708 	pdn->mps = pcie_get_mps(pdev);
1709 }
1710 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pnv_pci_fixup_vf_mps);
1711 #endif /* CONFIG_PCI_IOV */
1712 
1713 /**
1714  * eeh_powernv_init - Register platform dependent EEH operations
1715  *
1716  * EEH initialization on powernv platform. This function should be
1717  * called before any EEH related functions.
1718  */
1719 static int __init eeh_powernv_init(void)
1720 {
1721 	int ret = -EINVAL;
1722 
1723 	ret = eeh_ops_register(&pnv_eeh_ops);
1724 	if (!ret)
1725 		pr_info("EEH: PowerNV platform initialized\n");
1726 	else
1727 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
1728 
1729 	return ret;
1730 }
1731 machine_early_initcall(powernv, eeh_powernv_init);
1732