1 /*
2  * The file intends to implement the platform dependent EEH operations on
3  * powernv platform. Actually, the powernv was created in order to fully
4  * hypervisor support.
5  *
6  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  */
13 
14 #include <linux/atomic.h>
15 #include <linux/debugfs.h>
16 #include <linux/delay.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/interrupt.h>
20 #include <linux/list.h>
21 #include <linux/msi.h>
22 #include <linux/of.h>
23 #include <linux/pci.h>
24 #include <linux/proc_fs.h>
25 #include <linux/rbtree.h>
26 #include <linux/sched.h>
27 #include <linux/seq_file.h>
28 #include <linux/spinlock.h>
29 
30 #include <asm/eeh.h>
31 #include <asm/eeh_event.h>
32 #include <asm/firmware.h>
33 #include <asm/io.h>
34 #include <asm/iommu.h>
35 #include <asm/machdep.h>
36 #include <asm/msi_bitmap.h>
37 #include <asm/opal.h>
38 #include <asm/ppc-pci.h>
39 #include <asm/pnv-pci.h>
40 
41 #include "powernv.h"
42 #include "pci.h"
43 
44 static int eeh_event_irq = -EINVAL;
45 
46 void pnv_pcibios_bus_add_device(struct pci_dev *pdev)
47 {
48 	struct pci_dn *pdn = pci_get_pdn(pdev);
49 
50 	if (!pdev->is_virtfn)
51 		return;
52 
53 	/*
54 	 * The following operations will fail if VF's sysfs files
55 	 * aren't created or its resources aren't finalized.
56 	 */
57 	eeh_add_device_early(pdn);
58 	eeh_add_device_late(pdev);
59 	eeh_sysfs_add_device(pdev);
60 }
61 
62 static int pnv_eeh_init(void)
63 {
64 	struct pci_controller *hose;
65 	struct pnv_phb *phb;
66 	int max_diag_size = PNV_PCI_DIAG_BUF_SIZE;
67 
68 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
69 		pr_warn("%s: OPAL is required !\n",
70 			__func__);
71 		return -EINVAL;
72 	}
73 
74 	/* Set probe mode */
75 	eeh_add_flag(EEH_PROBE_MODE_DEV);
76 
77 	/*
78 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
79 	 * doesn't do that. So we have to selectively enable I/O
80 	 * prior to collecting error log.
81 	 */
82 	list_for_each_entry(hose, &hose_list, list_node) {
83 		phb = hose->private_data;
84 
85 		if (phb->model == PNV_PHB_MODEL_P7IOC)
86 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
87 
88 		if (phb->diag_data_size > max_diag_size)
89 			max_diag_size = phb->diag_data_size;
90 
91 		/*
92 		 * PE#0 should be regarded as valid by EEH core
93 		 * if it's not the reserved one. Currently, we
94 		 * have the reserved PE#255 and PE#127 for PHB3
95 		 * and P7IOC separately. So we should regard
96 		 * PE#0 as valid for PHB3 and P7IOC.
97 		 */
98 		if (phb->ioda.reserved_pe_idx != 0)
99 			eeh_add_flag(EEH_VALID_PE_ZERO);
100 
101 		break;
102 	}
103 
104 	eeh_set_pe_aux_size(max_diag_size);
105 	ppc_md.pcibios_bus_add_device = pnv_pcibios_bus_add_device;
106 
107 	return 0;
108 }
109 
110 static irqreturn_t pnv_eeh_event(int irq, void *data)
111 {
112 	/*
113 	 * We simply send a special EEH event if EEH has been
114 	 * enabled. We don't care about EEH events until we've
115 	 * finished processing the outstanding ones. Event processing
116 	 * gets unmasked in next_error() if EEH is enabled.
117 	 */
118 	disable_irq_nosync(irq);
119 
120 	if (eeh_enabled())
121 		eeh_send_failure_event(NULL);
122 
123 	return IRQ_HANDLED;
124 }
125 
126 #ifdef CONFIG_DEBUG_FS
127 static ssize_t pnv_eeh_ei_write(struct file *filp,
128 				const char __user *user_buf,
129 				size_t count, loff_t *ppos)
130 {
131 	struct pci_controller *hose = filp->private_data;
132 	struct eeh_pe *pe;
133 	int pe_no, type, func;
134 	unsigned long addr, mask;
135 	char buf[50];
136 	int ret;
137 
138 	if (!eeh_ops || !eeh_ops->err_inject)
139 		return -ENXIO;
140 
141 	/* Copy over argument buffer */
142 	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
143 	if (!ret)
144 		return -EFAULT;
145 
146 	/* Retrieve parameters */
147 	ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
148 		     &pe_no, &type, &func, &addr, &mask);
149 	if (ret != 5)
150 		return -EINVAL;
151 
152 	/* Retrieve PE */
153 	pe = eeh_pe_get(hose, pe_no, 0);
154 	if (!pe)
155 		return -ENODEV;
156 
157 	/* Do error injection */
158 	ret = eeh_ops->err_inject(pe, type, func, addr, mask);
159 	return ret < 0 ? ret : count;
160 }
161 
162 static const struct file_operations pnv_eeh_ei_fops = {
163 	.open	= simple_open,
164 	.llseek	= no_llseek,
165 	.write	= pnv_eeh_ei_write,
166 };
167 
168 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val)
169 {
170 	struct pci_controller *hose = data;
171 	struct pnv_phb *phb = hose->private_data;
172 
173 	out_be64(phb->regs + offset, val);
174 	return 0;
175 }
176 
177 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val)
178 {
179 	struct pci_controller *hose = data;
180 	struct pnv_phb *phb = hose->private_data;
181 
182 	*val = in_be64(phb->regs + offset);
183 	return 0;
184 }
185 
186 #define PNV_EEH_DBGFS_ENTRY(name, reg)				\
187 static int pnv_eeh_dbgfs_set_##name(void *data, u64 val)	\
188 {								\
189 	return pnv_eeh_dbgfs_set(data, reg, val);		\
190 }								\
191 								\
192 static int pnv_eeh_dbgfs_get_##name(void *data, u64 *val)	\
193 {								\
194 	return pnv_eeh_dbgfs_get(data, reg, val);		\
195 }								\
196 								\
197 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_dbgfs_ops_##name,		\
198 			pnv_eeh_dbgfs_get_##name,		\
199                         pnv_eeh_dbgfs_set_##name,		\
200 			"0x%llx\n")
201 
202 PNV_EEH_DBGFS_ENTRY(outb, 0xD10);
203 PNV_EEH_DBGFS_ENTRY(inbA, 0xD90);
204 PNV_EEH_DBGFS_ENTRY(inbB, 0xE10);
205 
206 #endif /* CONFIG_DEBUG_FS */
207 
208 /**
209  * pnv_eeh_post_init - EEH platform dependent post initialization
210  *
211  * EEH platform dependent post initialization on powernv. When
212  * the function is called, the EEH PEs and devices should have
213  * been built. If the I/O cache staff has been built, EEH is
214  * ready to supply service.
215  */
216 int pnv_eeh_post_init(void)
217 {
218 	struct pci_controller *hose;
219 	struct pnv_phb *phb;
220 	int ret = 0;
221 
222 	/* Probe devices & build address cache */
223 	eeh_probe_devices();
224 	eeh_addr_cache_build();
225 
226 	/* Register OPAL event notifier */
227 	eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
228 	if (eeh_event_irq < 0) {
229 		pr_err("%s: Can't register OPAL event interrupt (%d)\n",
230 		       __func__, eeh_event_irq);
231 		return eeh_event_irq;
232 	}
233 
234 	ret = request_irq(eeh_event_irq, pnv_eeh_event,
235 			  IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
236 	if (ret < 0) {
237 		irq_dispose_mapping(eeh_event_irq);
238 		pr_err("%s: Can't request OPAL event interrupt (%d)\n",
239 		       __func__, eeh_event_irq);
240 		return ret;
241 	}
242 
243 	if (!eeh_enabled())
244 		disable_irq(eeh_event_irq);
245 
246 	list_for_each_entry(hose, &hose_list, list_node) {
247 		phb = hose->private_data;
248 
249 		/*
250 		 * If EEH is enabled, we're going to rely on that.
251 		 * Otherwise, we restore to conventional mechanism
252 		 * to clear frozen PE during PCI config access.
253 		 */
254 		if (eeh_enabled())
255 			phb->flags |= PNV_PHB_FLAG_EEH;
256 		else
257 			phb->flags &= ~PNV_PHB_FLAG_EEH;
258 
259 		/* Create debugfs entries */
260 #ifdef CONFIG_DEBUG_FS
261 		if (phb->has_dbgfs || !phb->dbgfs)
262 			continue;
263 
264 		phb->has_dbgfs = 1;
265 		debugfs_create_file("err_injct", 0200,
266 				    phb->dbgfs, hose,
267 				    &pnv_eeh_ei_fops);
268 
269 		debugfs_create_file("err_injct_outbound", 0600,
270 				    phb->dbgfs, hose,
271 				    &pnv_eeh_dbgfs_ops_outb);
272 		debugfs_create_file("err_injct_inboundA", 0600,
273 				    phb->dbgfs, hose,
274 				    &pnv_eeh_dbgfs_ops_inbA);
275 		debugfs_create_file("err_injct_inboundB", 0600,
276 				    phb->dbgfs, hose,
277 				    &pnv_eeh_dbgfs_ops_inbB);
278 #endif /* CONFIG_DEBUG_FS */
279 	}
280 
281 	return ret;
282 }
283 
284 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
285 {
286 	int pos = PCI_CAPABILITY_LIST;
287 	int cnt = 48;   /* Maximal number of capabilities */
288 	u32 status, id;
289 
290 	if (!pdn)
291 		return 0;
292 
293 	/* Check if the device supports capabilities */
294 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
295 	if (!(status & PCI_STATUS_CAP_LIST))
296 		return 0;
297 
298 	while (cnt--) {
299 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
300 		if (pos < 0x40)
301 			break;
302 
303 		pos &= ~3;
304 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
305 		if (id == 0xff)
306 			break;
307 
308 		/* Found */
309 		if (id == cap)
310 			return pos;
311 
312 		/* Next one */
313 		pos += PCI_CAP_LIST_NEXT;
314 	}
315 
316 	return 0;
317 }
318 
319 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
320 {
321 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
322 	u32 header;
323 	int pos = 256, ttl = (4096 - 256) / 8;
324 
325 	if (!edev || !edev->pcie_cap)
326 		return 0;
327 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
328 		return 0;
329 	else if (!header)
330 		return 0;
331 
332 	while (ttl-- > 0) {
333 		if (PCI_EXT_CAP_ID(header) == cap && pos)
334 			return pos;
335 
336 		pos = PCI_EXT_CAP_NEXT(header);
337 		if (pos < 256)
338 			break;
339 
340 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
341 			break;
342 	}
343 
344 	return 0;
345 }
346 
347 /**
348  * pnv_eeh_probe - Do probe on PCI device
349  * @pdn: PCI device node
350  * @data: unused
351  *
352  * When EEH module is installed during system boot, all PCI devices
353  * are checked one by one to see if it supports EEH. The function
354  * is introduced for the purpose. By default, EEH has been enabled
355  * on all PCI devices. That's to say, we only need do necessary
356  * initialization on the corresponding eeh device and create PE
357  * accordingly.
358  *
359  * It's notable that's unsafe to retrieve the EEH device through
360  * the corresponding PCI device. During the PCI device hotplug, which
361  * was possiblly triggered by EEH core, the binding between EEH device
362  * and the PCI device isn't built yet.
363  */
364 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
365 {
366 	struct pci_controller *hose = pdn->phb;
367 	struct pnv_phb *phb = hose->private_data;
368 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
369 	uint32_t pcie_flags;
370 	int ret;
371 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
372 
373 	/*
374 	 * When probing the root bridge, which doesn't have any
375 	 * subordinate PCI devices. We don't have OF node for
376 	 * the root bridge. So it's not reasonable to continue
377 	 * the probing.
378 	 */
379 	if (!edev || edev->pe)
380 		return NULL;
381 
382 	/* Skip for PCI-ISA bridge */
383 	if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
384 		return NULL;
385 
386 	/* Skip if we haven't probed yet */
387 	if (phb->ioda.pe_rmap[config_addr] == IODA_INVALID_PE)
388 		return NULL;
389 
390 	/* Initialize eeh device */
391 	edev->class_code = pdn->class_code;
392 	edev->mode	&= 0xFFFFFF00;
393 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
394 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
395 	edev->af_cap   = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF);
396 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
397 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
398 		edev->mode |= EEH_DEV_BRIDGE;
399 		if (edev->pcie_cap) {
400 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
401 					 2, &pcie_flags);
402 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
403 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
404 				edev->mode |= EEH_DEV_ROOT_PORT;
405 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
406 				edev->mode |= EEH_DEV_DS_PORT;
407 		}
408 	}
409 
410 	edev->pe_config_addr = phb->ioda.pe_rmap[config_addr];
411 
412 	/* Create PE */
413 	ret = eeh_add_to_parent_pe(edev);
414 	if (ret) {
415 		pr_warn("%s: Can't add PCI dev %04x:%02x:%02x.%01x to parent PE (%x)\n",
416 			__func__, hose->global_number, pdn->busno,
417 			PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn), ret);
418 		return NULL;
419 	}
420 
421 	/*
422 	 * If the PE contains any one of following adapters, the
423 	 * PCI config space can't be accessed when dumping EEH log.
424 	 * Otherwise, we will run into fenced PHB caused by shortage
425 	 * of outbound credits in the adapter. The PCI config access
426 	 * should be blocked until PE reset. MMIO access is dropped
427 	 * by hardware certainly. In order to drop PCI config requests,
428 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
429 	 * will be checked in the backend for PE state retrival. If
430 	 * the PE becomes frozen for the first time and the flag has
431 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
432 	 * that PE to block its config space.
433 	 *
434 	 * Broadcom BCM5718 2-ports NICs (14e4:1656)
435 	 * Broadcom Austin 4-ports NICs (14e4:1657)
436 	 * Broadcom Shiner 4-ports 1G NICs (14e4:168a)
437 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
438 	 */
439 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
440 	     pdn->device_id == 0x1656) ||
441 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
442 	     pdn->device_id == 0x1657) ||
443 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
444 	     pdn->device_id == 0x168a) ||
445 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
446 	     pdn->device_id == 0x168e))
447 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
448 
449 	/*
450 	 * Cache the PE primary bus, which can't be fetched when
451 	 * full hotplug is in progress. In that case, all child
452 	 * PCI devices of the PE are expected to be removed prior
453 	 * to PE reset.
454 	 */
455 	if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
456 		edev->pe->bus = pci_find_bus(hose->global_number,
457 					     pdn->busno);
458 		if (edev->pe->bus)
459 			edev->pe->state |= EEH_PE_PRI_BUS;
460 	}
461 
462 	/*
463 	 * Enable EEH explicitly so that we will do EEH check
464 	 * while accessing I/O stuff
465 	 */
466 	eeh_add_flag(EEH_ENABLED);
467 
468 	/* Save memory bars */
469 	eeh_save_bars(edev);
470 
471 	return NULL;
472 }
473 
474 /**
475  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
476  * @pe: EEH PE
477  * @option: operation to be issued
478  *
479  * The function is used to control the EEH functionality globally.
480  * Currently, following options are support according to PAPR:
481  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
482  */
483 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
484 {
485 	struct pci_controller *hose = pe->phb;
486 	struct pnv_phb *phb = hose->private_data;
487 	bool freeze_pe = false;
488 	int opt;
489 	s64 rc;
490 
491 	switch (option) {
492 	case EEH_OPT_DISABLE:
493 		return -EPERM;
494 	case EEH_OPT_ENABLE:
495 		return 0;
496 	case EEH_OPT_THAW_MMIO:
497 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
498 		break;
499 	case EEH_OPT_THAW_DMA:
500 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
501 		break;
502 	case EEH_OPT_FREEZE_PE:
503 		freeze_pe = true;
504 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
505 		break;
506 	default:
507 		pr_warn("%s: Invalid option %d\n", __func__, option);
508 		return -EINVAL;
509 	}
510 
511 	/* Freeze master and slave PEs if PHB supports compound PEs */
512 	if (freeze_pe) {
513 		if (phb->freeze_pe) {
514 			phb->freeze_pe(phb, pe->addr);
515 			return 0;
516 		}
517 
518 		rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt);
519 		if (rc != OPAL_SUCCESS) {
520 			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
521 				__func__, rc, phb->hose->global_number,
522 				pe->addr);
523 			return -EIO;
524 		}
525 
526 		return 0;
527 	}
528 
529 	/* Unfreeze master and slave PEs if PHB supports */
530 	if (phb->unfreeze_pe)
531 		return phb->unfreeze_pe(phb, pe->addr, opt);
532 
533 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt);
534 	if (rc != OPAL_SUCCESS) {
535 		pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n",
536 			__func__, rc, option, phb->hose->global_number,
537 			pe->addr);
538 		return -EIO;
539 	}
540 
541 	return 0;
542 }
543 
544 /**
545  * pnv_eeh_get_pe_addr - Retrieve PE address
546  * @pe: EEH PE
547  *
548  * Retrieve the PE address according to the given tranditional
549  * PCI BDF (Bus/Device/Function) address.
550  */
551 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe)
552 {
553 	return pe->addr;
554 }
555 
556 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
557 {
558 	struct pnv_phb *phb = pe->phb->private_data;
559 	s64 rc;
560 
561 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
562 					 phb->diag_data_size);
563 	if (rc != OPAL_SUCCESS)
564 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
565 			__func__, rc, pe->phb->global_number);
566 }
567 
568 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
569 {
570 	struct pnv_phb *phb = pe->phb->private_data;
571 	u8 fstate;
572 	__be16 pcierr;
573 	s64 rc;
574 	int result = 0;
575 
576 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
577 					pe->addr,
578 					&fstate,
579 					&pcierr,
580 					NULL);
581 	if (rc != OPAL_SUCCESS) {
582 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
583 			__func__, rc, phb->hose->global_number);
584 		return EEH_STATE_NOT_SUPPORT;
585 	}
586 
587 	/*
588 	 * Check PHB state. If the PHB is frozen for the
589 	 * first time, to dump the PHB diag-data.
590 	 */
591 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
592 		result = (EEH_STATE_MMIO_ACTIVE  |
593 			  EEH_STATE_DMA_ACTIVE   |
594 			  EEH_STATE_MMIO_ENABLED |
595 			  EEH_STATE_DMA_ENABLED);
596 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
597 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
598 		pnv_eeh_get_phb_diag(pe);
599 
600 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
601 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
602 	}
603 
604 	return result;
605 }
606 
607 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
608 {
609 	struct pnv_phb *phb = pe->phb->private_data;
610 	u8 fstate;
611 	__be16 pcierr;
612 	s64 rc;
613 	int result;
614 
615 	/*
616 	 * We don't clobber hardware frozen state until PE
617 	 * reset is completed. In order to keep EEH core
618 	 * moving forward, we have to return operational
619 	 * state during PE reset.
620 	 */
621 	if (pe->state & EEH_PE_RESET) {
622 		result = (EEH_STATE_MMIO_ACTIVE  |
623 			  EEH_STATE_DMA_ACTIVE   |
624 			  EEH_STATE_MMIO_ENABLED |
625 			  EEH_STATE_DMA_ENABLED);
626 		return result;
627 	}
628 
629 	/*
630 	 * Fetch PE state from hardware. If the PHB
631 	 * supports compound PE, let it handle that.
632 	 */
633 	if (phb->get_pe_state) {
634 		fstate = phb->get_pe_state(phb, pe->addr);
635 	} else {
636 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
637 						pe->addr,
638 						&fstate,
639 						&pcierr,
640 						NULL);
641 		if (rc != OPAL_SUCCESS) {
642 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
643 				__func__, rc, phb->hose->global_number,
644 				pe->addr);
645 			return EEH_STATE_NOT_SUPPORT;
646 		}
647 	}
648 
649 	/* Figure out state */
650 	switch (fstate) {
651 	case OPAL_EEH_STOPPED_NOT_FROZEN:
652 		result = (EEH_STATE_MMIO_ACTIVE  |
653 			  EEH_STATE_DMA_ACTIVE   |
654 			  EEH_STATE_MMIO_ENABLED |
655 			  EEH_STATE_DMA_ENABLED);
656 		break;
657 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
658 		result = (EEH_STATE_DMA_ACTIVE |
659 			  EEH_STATE_DMA_ENABLED);
660 		break;
661 	case OPAL_EEH_STOPPED_DMA_FREEZE:
662 		result = (EEH_STATE_MMIO_ACTIVE |
663 			  EEH_STATE_MMIO_ENABLED);
664 		break;
665 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
666 		result = 0;
667 		break;
668 	case OPAL_EEH_STOPPED_RESET:
669 		result = EEH_STATE_RESET_ACTIVE;
670 		break;
671 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
672 		result = EEH_STATE_UNAVAILABLE;
673 		break;
674 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
675 		result = EEH_STATE_NOT_SUPPORT;
676 		break;
677 	default:
678 		result = EEH_STATE_NOT_SUPPORT;
679 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
680 			__func__, phb->hose->global_number,
681 			pe->addr, fstate);
682 	}
683 
684 	/*
685 	 * If PHB supports compound PE, to freeze all
686 	 * slave PEs for consistency.
687 	 *
688 	 * If the PE is switching to frozen state for the
689 	 * first time, to dump the PHB diag-data.
690 	 */
691 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
692 	    !(result & EEH_STATE_UNAVAILABLE) &&
693 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
694 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
695 	    !(pe->state & EEH_PE_ISOLATED)) {
696 		if (phb->freeze_pe)
697 			phb->freeze_pe(phb, pe->addr);
698 
699 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
700 		pnv_eeh_get_phb_diag(pe);
701 
702 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
703 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
704 	}
705 
706 	return result;
707 }
708 
709 /**
710  * pnv_eeh_get_state - Retrieve PE state
711  * @pe: EEH PE
712  * @delay: delay while PE state is temporarily unavailable
713  *
714  * Retrieve the state of the specified PE. For IODA-compitable
715  * platform, it should be retrieved from IODA table. Therefore,
716  * we prefer passing down to hardware implementation to handle
717  * it.
718  */
719 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
720 {
721 	int ret;
722 
723 	if (pe->type & EEH_PE_PHB)
724 		ret = pnv_eeh_get_phb_state(pe);
725 	else
726 		ret = pnv_eeh_get_pe_state(pe);
727 
728 	if (!delay)
729 		return ret;
730 
731 	/*
732 	 * If the PE state is temporarily unavailable,
733 	 * to inform the EEH core delay for default
734 	 * period (1 second)
735 	 */
736 	*delay = 0;
737 	if (ret & EEH_STATE_UNAVAILABLE)
738 		*delay = 1000;
739 
740 	return ret;
741 }
742 
743 static s64 pnv_eeh_poll(unsigned long id)
744 {
745 	s64 rc = OPAL_HARDWARE;
746 
747 	while (1) {
748 		rc = opal_pci_poll(id);
749 		if (rc <= 0)
750 			break;
751 
752 		if (system_state < SYSTEM_RUNNING)
753 			udelay(1000 * rc);
754 		else
755 			msleep(rc);
756 	}
757 
758 	return rc;
759 }
760 
761 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
762 {
763 	struct pnv_phb *phb = hose->private_data;
764 	s64 rc = OPAL_HARDWARE;
765 
766 	pr_debug("%s: Reset PHB#%x, option=%d\n",
767 		 __func__, hose->global_number, option);
768 
769 	/* Issue PHB complete reset request */
770 	if (option == EEH_RESET_FUNDAMENTAL ||
771 	    option == EEH_RESET_HOT)
772 		rc = opal_pci_reset(phb->opal_id,
773 				    OPAL_RESET_PHB_COMPLETE,
774 				    OPAL_ASSERT_RESET);
775 	else if (option == EEH_RESET_DEACTIVATE)
776 		rc = opal_pci_reset(phb->opal_id,
777 				    OPAL_RESET_PHB_COMPLETE,
778 				    OPAL_DEASSERT_RESET);
779 	if (rc < 0)
780 		goto out;
781 
782 	/*
783 	 * Poll state of the PHB until the request is done
784 	 * successfully. The PHB reset is usually PHB complete
785 	 * reset followed by hot reset on root bus. So we also
786 	 * need the PCI bus settlement delay.
787 	 */
788 	if (rc > 0)
789 		rc = pnv_eeh_poll(phb->opal_id);
790 	if (option == EEH_RESET_DEACTIVATE) {
791 		if (system_state < SYSTEM_RUNNING)
792 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
793 		else
794 			msleep(EEH_PE_RST_SETTLE_TIME);
795 	}
796 out:
797 	if (rc != OPAL_SUCCESS)
798 		return -EIO;
799 
800 	return 0;
801 }
802 
803 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
804 {
805 	struct pnv_phb *phb = hose->private_data;
806 	s64 rc = OPAL_HARDWARE;
807 
808 	pr_debug("%s: Reset PHB#%x, option=%d\n",
809 		 __func__, hose->global_number, option);
810 
811 	/*
812 	 * During the reset deassert time, we needn't care
813 	 * the reset scope because the firmware does nothing
814 	 * for fundamental or hot reset during deassert phase.
815 	 */
816 	if (option == EEH_RESET_FUNDAMENTAL)
817 		rc = opal_pci_reset(phb->opal_id,
818 				    OPAL_RESET_PCI_FUNDAMENTAL,
819 				    OPAL_ASSERT_RESET);
820 	else if (option == EEH_RESET_HOT)
821 		rc = opal_pci_reset(phb->opal_id,
822 				    OPAL_RESET_PCI_HOT,
823 				    OPAL_ASSERT_RESET);
824 	else if (option == EEH_RESET_DEACTIVATE)
825 		rc = opal_pci_reset(phb->opal_id,
826 				    OPAL_RESET_PCI_HOT,
827 				    OPAL_DEASSERT_RESET);
828 	if (rc < 0)
829 		goto out;
830 
831 	/* Poll state of the PHB until the request is done */
832 	if (rc > 0)
833 		rc = pnv_eeh_poll(phb->opal_id);
834 	if (option == EEH_RESET_DEACTIVATE)
835 		msleep(EEH_PE_RST_SETTLE_TIME);
836 out:
837 	if (rc != OPAL_SUCCESS)
838 		return -EIO;
839 
840 	return 0;
841 }
842 
843 static int __pnv_eeh_bridge_reset(struct pci_dev *dev, int option)
844 {
845 	struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
846 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
847 	int aer = edev ? edev->aer_cap : 0;
848 	u32 ctrl;
849 
850 	pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
851 		 __func__, pci_domain_nr(dev->bus),
852 		 dev->bus->number, option);
853 
854 	switch (option) {
855 	case EEH_RESET_FUNDAMENTAL:
856 	case EEH_RESET_HOT:
857 		/* Don't report linkDown event */
858 		if (aer) {
859 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
860 					     4, &ctrl);
861 			ctrl |= PCI_ERR_UNC_SURPDN;
862 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
863 					      4, ctrl);
864 		}
865 
866 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
867 		ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
868 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
869 
870 		msleep(EEH_PE_RST_HOLD_TIME);
871 		break;
872 	case EEH_RESET_DEACTIVATE:
873 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
874 		ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
875 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
876 
877 		msleep(EEH_PE_RST_SETTLE_TIME);
878 
879 		/* Continue reporting linkDown event */
880 		if (aer) {
881 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
882 					     4, &ctrl);
883 			ctrl &= ~PCI_ERR_UNC_SURPDN;
884 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
885 					      4, ctrl);
886 		}
887 
888 		break;
889 	}
890 
891 	return 0;
892 }
893 
894 static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option)
895 {
896 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
897 	struct pnv_phb *phb = hose->private_data;
898 	struct device_node *dn = pci_device_to_OF_node(pdev);
899 	uint64_t id = PCI_SLOT_ID(phb->opal_id,
900 				  (pdev->bus->number << 8) | pdev->devfn);
901 	uint8_t scope;
902 	int64_t rc;
903 
904 	/* Hot reset to the bus if firmware cannot handle */
905 	if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL))
906 		return __pnv_eeh_bridge_reset(pdev, option);
907 
908 	switch (option) {
909 	case EEH_RESET_FUNDAMENTAL:
910 		scope = OPAL_RESET_PCI_FUNDAMENTAL;
911 		break;
912 	case EEH_RESET_HOT:
913 		scope = OPAL_RESET_PCI_HOT;
914 		break;
915 	case EEH_RESET_DEACTIVATE:
916 		return 0;
917 	default:
918 		dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n",
919 			__func__, option);
920 		return -EINVAL;
921 	}
922 
923 	rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET);
924 	if (rc <= OPAL_SUCCESS)
925 		goto out;
926 
927 	rc = pnv_eeh_poll(id);
928 out:
929 	return (rc == OPAL_SUCCESS) ? 0 : -EIO;
930 }
931 
932 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
933 {
934 	struct pci_controller *hose;
935 
936 	if (pci_is_root_bus(dev->bus)) {
937 		hose = pci_bus_to_host(dev->bus);
938 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
939 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
940 	} else {
941 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
942 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
943 	}
944 }
945 
946 static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type,
947 				     int pos, u16 mask)
948 {
949 	int i, status = 0;
950 
951 	/* Wait for Transaction Pending bit to be cleared */
952 	for (i = 0; i < 4; i++) {
953 		eeh_ops->read_config(pdn, pos, 2, &status);
954 		if (!(status & mask))
955 			return;
956 
957 		msleep((1 << i) * 100);
958 	}
959 
960 	pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n",
961 		__func__, type,
962 		pdn->phb->global_number, pdn->busno,
963 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
964 }
965 
966 static int pnv_eeh_do_flr(struct pci_dn *pdn, int option)
967 {
968 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
969 	u32 reg = 0;
970 
971 	if (WARN_ON(!edev->pcie_cap))
972 		return -ENOTTY;
973 
974 	eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP, 4, &reg);
975 	if (!(reg & PCI_EXP_DEVCAP_FLR))
976 		return -ENOTTY;
977 
978 	switch (option) {
979 	case EEH_RESET_HOT:
980 	case EEH_RESET_FUNDAMENTAL:
981 		pnv_eeh_wait_for_pending(pdn, "",
982 					 edev->pcie_cap + PCI_EXP_DEVSTA,
983 					 PCI_EXP_DEVSTA_TRPND);
984 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
985 				     4, &reg);
986 		reg |= PCI_EXP_DEVCTL_BCR_FLR;
987 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
988 				      4, reg);
989 		msleep(EEH_PE_RST_HOLD_TIME);
990 		break;
991 	case EEH_RESET_DEACTIVATE:
992 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
993 				     4, &reg);
994 		reg &= ~PCI_EXP_DEVCTL_BCR_FLR;
995 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
996 				      4, reg);
997 		msleep(EEH_PE_RST_SETTLE_TIME);
998 		break;
999 	}
1000 
1001 	return 0;
1002 }
1003 
1004 static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option)
1005 {
1006 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1007 	u32 cap = 0;
1008 
1009 	if (WARN_ON(!edev->af_cap))
1010 		return -ENOTTY;
1011 
1012 	eeh_ops->read_config(pdn, edev->af_cap + PCI_AF_CAP, 1, &cap);
1013 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
1014 		return -ENOTTY;
1015 
1016 	switch (option) {
1017 	case EEH_RESET_HOT:
1018 	case EEH_RESET_FUNDAMENTAL:
1019 		/*
1020 		 * Wait for Transaction Pending bit to clear. A word-aligned
1021 		 * test is used, so we use the conrol offset rather than status
1022 		 * and shift the test bit to match.
1023 		 */
1024 		pnv_eeh_wait_for_pending(pdn, "AF",
1025 					 edev->af_cap + PCI_AF_CTRL,
1026 					 PCI_AF_STATUS_TP << 8);
1027 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL,
1028 				      1, PCI_AF_CTRL_FLR);
1029 		msleep(EEH_PE_RST_HOLD_TIME);
1030 		break;
1031 	case EEH_RESET_DEACTIVATE:
1032 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL, 1, 0);
1033 		msleep(EEH_PE_RST_SETTLE_TIME);
1034 		break;
1035 	}
1036 
1037 	return 0;
1038 }
1039 
1040 static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option)
1041 {
1042 	struct eeh_dev *edev;
1043 	struct pci_dn *pdn;
1044 	int ret;
1045 
1046 	/* The VF PE should have only one child device */
1047 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, list);
1048 	pdn = eeh_dev_to_pdn(edev);
1049 	if (!pdn)
1050 		return -ENXIO;
1051 
1052 	ret = pnv_eeh_do_flr(pdn, option);
1053 	if (!ret)
1054 		return ret;
1055 
1056 	return pnv_eeh_do_af_flr(pdn, option);
1057 }
1058 
1059 /**
1060  * pnv_eeh_reset - Reset the specified PE
1061  * @pe: EEH PE
1062  * @option: reset option
1063  *
1064  * Do reset on the indicated PE. For PCI bus sensitive PE,
1065  * we need to reset the parent p2p bridge. The PHB has to
1066  * be reinitialized if the p2p bridge is root bridge. For
1067  * PCI device sensitive PE, we will try to reset the device
1068  * through FLR. For now, we don't have OPAL APIs to do HARD
1069  * reset yet, so all reset would be SOFT (HOT) reset.
1070  */
1071 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
1072 {
1073 	struct pci_controller *hose = pe->phb;
1074 	struct pnv_phb *phb;
1075 	struct pci_bus *bus;
1076 	int64_t rc;
1077 
1078 	/*
1079 	 * For PHB reset, we always have complete reset. For those PEs whose
1080 	 * primary bus derived from root complex (root bus) or root port
1081 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
1082 	 * For other PEs, we always have hot reset on the PE primary bus.
1083 	 *
1084 	 * Here, we have different design to pHyp, which always clear the
1085 	 * frozen state during PE reset. However, the good idea here from
1086 	 * benh is to keep frozen state before we get PE reset done completely
1087 	 * (until BAR restore). With the frozen state, HW drops illegal IO
1088 	 * or MMIO access, which can incur recrusive frozen PE during PE
1089 	 * reset. The side effect is that EEH core has to clear the frozen
1090 	 * state explicitly after BAR restore.
1091 	 */
1092 	if (pe->type & EEH_PE_PHB)
1093 		return pnv_eeh_phb_reset(hose, option);
1094 
1095 	/*
1096 	 * The frozen PE might be caused by PAPR error injection
1097 	 * registers, which are expected to be cleared after hitting
1098 	 * frozen PE as stated in the hardware spec. Unfortunately,
1099 	 * that's not true on P7IOC. So we have to clear it manually
1100 	 * to avoid recursive EEH errors during recovery.
1101 	 */
1102 	phb = hose->private_data;
1103 	if (phb->model == PNV_PHB_MODEL_P7IOC &&
1104 	    (option == EEH_RESET_HOT ||
1105 	     option == EEH_RESET_FUNDAMENTAL)) {
1106 		rc = opal_pci_reset(phb->opal_id,
1107 				    OPAL_RESET_PHB_ERROR,
1108 				    OPAL_ASSERT_RESET);
1109 		if (rc != OPAL_SUCCESS) {
1110 			pr_warn("%s: Failure %lld clearing error injection registers\n",
1111 				__func__, rc);
1112 			return -EIO;
1113 		}
1114 	}
1115 
1116 	if (pe->type & EEH_PE_VF)
1117 		return pnv_eeh_reset_vf_pe(pe, option);
1118 
1119 	bus = eeh_pe_bus_get(pe);
1120 	if (!bus) {
1121 		pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
1122 			__func__, pe->phb->global_number, pe->addr);
1123 		return -EIO;
1124 	}
1125 
1126 	/*
1127 	 * If dealing with the root bus (or the bus underneath the
1128 	 * root port), we reset the bus underneath the root port.
1129 	 *
1130 	 * The cxl driver depends on this behaviour for bi-modal card
1131 	 * switching.
1132 	 */
1133 	if (pci_is_root_bus(bus) ||
1134 	    pci_is_root_bus(bus->parent))
1135 		return pnv_eeh_root_reset(hose, option);
1136 
1137 	return pnv_eeh_bridge_reset(bus->self, option);
1138 }
1139 
1140 /**
1141  * pnv_eeh_wait_state - Wait for PE state
1142  * @pe: EEH PE
1143  * @max_wait: maximal period in millisecond
1144  *
1145  * Wait for the state of associated PE. It might take some time
1146  * to retrieve the PE's state.
1147  */
1148 static int pnv_eeh_wait_state(struct eeh_pe *pe, int max_wait)
1149 {
1150 	int ret;
1151 	int mwait;
1152 
1153 	while (1) {
1154 		ret = pnv_eeh_get_state(pe, &mwait);
1155 
1156 		/*
1157 		 * If the PE's state is temporarily unavailable,
1158 		 * we have to wait for the specified time. Otherwise,
1159 		 * the PE's state will be returned immediately.
1160 		 */
1161 		if (ret != EEH_STATE_UNAVAILABLE)
1162 			return ret;
1163 
1164 		if (max_wait <= 0) {
1165 			pr_warn("%s: Timeout getting PE#%x's state (%d)\n",
1166 				__func__, pe->addr, max_wait);
1167 			return EEH_STATE_NOT_SUPPORT;
1168 		}
1169 
1170 		max_wait -= mwait;
1171 		msleep(mwait);
1172 	}
1173 
1174 	return EEH_STATE_NOT_SUPPORT;
1175 }
1176 
1177 /**
1178  * pnv_eeh_get_log - Retrieve error log
1179  * @pe: EEH PE
1180  * @severity: temporary or permanent error log
1181  * @drv_log: driver log to be combined with retrieved error log
1182  * @len: length of driver log
1183  *
1184  * Retrieve the temporary or permanent error from the PE.
1185  */
1186 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1187 			   char *drv_log, unsigned long len)
1188 {
1189 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1190 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1191 
1192 	return 0;
1193 }
1194 
1195 /**
1196  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1197  * @pe: EEH PE
1198  *
1199  * The function will be called to reconfigure the bridges included
1200  * in the specified PE so that the mulfunctional PE would be recovered
1201  * again.
1202  */
1203 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1204 {
1205 	return 0;
1206 }
1207 
1208 /**
1209  * pnv_pe_err_inject - Inject specified error to the indicated PE
1210  * @pe: the indicated PE
1211  * @type: error type
1212  * @func: specific error type
1213  * @addr: address
1214  * @mask: address mask
1215  *
1216  * The routine is called to inject specified error, which is
1217  * determined by @type and @func, to the indicated PE for
1218  * testing purpose.
1219  */
1220 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1221 			      unsigned long addr, unsigned long mask)
1222 {
1223 	struct pci_controller *hose = pe->phb;
1224 	struct pnv_phb *phb = hose->private_data;
1225 	s64 rc;
1226 
1227 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1228 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1229 		pr_warn("%s: Invalid error type %d\n",
1230 			__func__, type);
1231 		return -ERANGE;
1232 	}
1233 
1234 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1235 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1236 		pr_warn("%s: Invalid error function %d\n",
1237 			__func__, func);
1238 		return -ERANGE;
1239 	}
1240 
1241 	/* Firmware supports error injection ? */
1242 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1243 		pr_warn("%s: Firmware doesn't support error injection\n",
1244 			__func__);
1245 		return -ENXIO;
1246 	}
1247 
1248 	/* Do error injection */
1249 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1250 				 type, func, addr, mask);
1251 	if (rc != OPAL_SUCCESS) {
1252 		pr_warn("%s: Failure %lld injecting error "
1253 			"%d-%d to PHB#%x-PE#%x\n",
1254 			__func__, rc, type, func,
1255 			hose->global_number, pe->addr);
1256 		return -EIO;
1257 	}
1258 
1259 	return 0;
1260 }
1261 
1262 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1263 {
1264 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1265 
1266 	if (!edev || !edev->pe)
1267 		return false;
1268 
1269 	/*
1270 	 * We will issue FLR or AF FLR to all VFs, which are contained
1271 	 * in VF PE. It relies on the EEH PCI config accessors. So we
1272 	 * can't block them during the window.
1273 	 */
1274 	if (edev->physfn && (edev->pe->state & EEH_PE_RESET))
1275 		return false;
1276 
1277 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1278 		return true;
1279 
1280 	return false;
1281 }
1282 
1283 static int pnv_eeh_read_config(struct pci_dn *pdn,
1284 			       int where, int size, u32 *val)
1285 {
1286 	if (!pdn)
1287 		return PCIBIOS_DEVICE_NOT_FOUND;
1288 
1289 	if (pnv_eeh_cfg_blocked(pdn)) {
1290 		*val = 0xFFFFFFFF;
1291 		return PCIBIOS_SET_FAILED;
1292 	}
1293 
1294 	return pnv_pci_cfg_read(pdn, where, size, val);
1295 }
1296 
1297 static int pnv_eeh_write_config(struct pci_dn *pdn,
1298 				int where, int size, u32 val)
1299 {
1300 	if (!pdn)
1301 		return PCIBIOS_DEVICE_NOT_FOUND;
1302 
1303 	if (pnv_eeh_cfg_blocked(pdn))
1304 		return PCIBIOS_SET_FAILED;
1305 
1306 	return pnv_pci_cfg_write(pdn, where, size, val);
1307 }
1308 
1309 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1310 {
1311 	/* GEM */
1312 	if (data->gemXfir || data->gemRfir ||
1313 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1314 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1315 			be64_to_cpu(data->gemXfir),
1316 			be64_to_cpu(data->gemRfir),
1317 			be64_to_cpu(data->gemRirqfir),
1318 			be64_to_cpu(data->gemMask),
1319 			be64_to_cpu(data->gemRwof));
1320 
1321 	/* LEM */
1322 	if (data->lemFir || data->lemErrMask ||
1323 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1324 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1325 			be64_to_cpu(data->lemFir),
1326 			be64_to_cpu(data->lemErrMask),
1327 			be64_to_cpu(data->lemAction0),
1328 			be64_to_cpu(data->lemAction1),
1329 			be64_to_cpu(data->lemWof));
1330 }
1331 
1332 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1333 {
1334 	struct pnv_phb *phb = hose->private_data;
1335 	struct OpalIoP7IOCErrorData *data =
1336 		(struct OpalIoP7IOCErrorData*)phb->diag_data;
1337 	long rc;
1338 
1339 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1340 	if (rc != OPAL_SUCCESS) {
1341 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1342 			__func__, phb->hub_id, rc);
1343 		return;
1344 	}
1345 
1346 	switch (be16_to_cpu(data->type)) {
1347 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1348 		pr_info("P7IOC diag-data for RGC\n\n");
1349 		pnv_eeh_dump_hub_diag_common(data);
1350 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1351 			pr_info("  RGC: %016llx %016llx\n",
1352 				be64_to_cpu(data->rgc.rgcStatus),
1353 				be64_to_cpu(data->rgc.rgcLdcp));
1354 		break;
1355 	case OPAL_P7IOC_DIAG_TYPE_BI:
1356 		pr_info("P7IOC diag-data for BI %s\n\n",
1357 			data->bi.biDownbound ? "Downbound" : "Upbound");
1358 		pnv_eeh_dump_hub_diag_common(data);
1359 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1360 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1361 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1362 				be64_to_cpu(data->bi.biLdcp0),
1363 				be64_to_cpu(data->bi.biLdcp1),
1364 				be64_to_cpu(data->bi.biLdcp2),
1365 				be64_to_cpu(data->bi.biFenceStatus));
1366 		break;
1367 	case OPAL_P7IOC_DIAG_TYPE_CI:
1368 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1369 			data->ci.ciPort);
1370 		pnv_eeh_dump_hub_diag_common(data);
1371 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1372 			pr_info("  CI:  %016llx %016llx\n",
1373 				be64_to_cpu(data->ci.ciPortStatus),
1374 				be64_to_cpu(data->ci.ciPortLdcp));
1375 		break;
1376 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1377 		pr_info("P7IOC diag-data for MISC\n\n");
1378 		pnv_eeh_dump_hub_diag_common(data);
1379 		break;
1380 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1381 		pr_info("P7IOC diag-data for I2C\n\n");
1382 		pnv_eeh_dump_hub_diag_common(data);
1383 		break;
1384 	default:
1385 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1386 			__func__, phb->hub_id, data->type);
1387 	}
1388 }
1389 
1390 static int pnv_eeh_get_pe(struct pci_controller *hose,
1391 			  u16 pe_no, struct eeh_pe **pe)
1392 {
1393 	struct pnv_phb *phb = hose->private_data;
1394 	struct pnv_ioda_pe *pnv_pe;
1395 	struct eeh_pe *dev_pe;
1396 
1397 	/*
1398 	 * If PHB supports compound PE, to fetch
1399 	 * the master PE because slave PE is invisible
1400 	 * to EEH core.
1401 	 */
1402 	pnv_pe = &phb->ioda.pe_array[pe_no];
1403 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1404 		pnv_pe = pnv_pe->master;
1405 		WARN_ON(!pnv_pe ||
1406 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1407 		pe_no = pnv_pe->pe_number;
1408 	}
1409 
1410 	/* Find the PE according to PE# */
1411 	dev_pe = eeh_pe_get(hose, pe_no, 0);
1412 	if (!dev_pe)
1413 		return -EEXIST;
1414 
1415 	/* Freeze the (compound) PE */
1416 	*pe = dev_pe;
1417 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1418 		phb->freeze_pe(phb, pe_no);
1419 
1420 	/*
1421 	 * At this point, we're sure the (compound) PE should
1422 	 * have been frozen. However, we still need poke until
1423 	 * hitting the frozen PE on top level.
1424 	 */
1425 	dev_pe = dev_pe->parent;
1426 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1427 		int ret;
1428 		int active_flags = (EEH_STATE_MMIO_ACTIVE |
1429 				    EEH_STATE_DMA_ACTIVE);
1430 
1431 		ret = eeh_ops->get_state(dev_pe, NULL);
1432 		if (ret <= 0 || (ret & active_flags) == active_flags) {
1433 			dev_pe = dev_pe->parent;
1434 			continue;
1435 		}
1436 
1437 		/* Frozen parent PE */
1438 		*pe = dev_pe;
1439 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1440 			phb->freeze_pe(phb, dev_pe->addr);
1441 
1442 		/* Next one */
1443 		dev_pe = dev_pe->parent;
1444 	}
1445 
1446 	return 0;
1447 }
1448 
1449 /**
1450  * pnv_eeh_next_error - Retrieve next EEH error to handle
1451  * @pe: Affected PE
1452  *
1453  * The function is expected to be called by EEH core while it gets
1454  * special EEH event (without binding PE). The function calls to
1455  * OPAL APIs for next error to handle. The informational error is
1456  * handled internally by platform. However, the dead IOC, dead PHB,
1457  * fenced PHB and frozen PE should be handled by EEH core eventually.
1458  */
1459 static int pnv_eeh_next_error(struct eeh_pe **pe)
1460 {
1461 	struct pci_controller *hose;
1462 	struct pnv_phb *phb;
1463 	struct eeh_pe *phb_pe, *parent_pe;
1464 	__be64 frozen_pe_no;
1465 	__be16 err_type, severity;
1466 	int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
1467 	long rc;
1468 	int state, ret = EEH_NEXT_ERR_NONE;
1469 
1470 	/*
1471 	 * While running here, it's safe to purge the event queue. The
1472 	 * event should still be masked.
1473 	 */
1474 	eeh_remove_event(NULL, false);
1475 
1476 	list_for_each_entry(hose, &hose_list, list_node) {
1477 		/*
1478 		 * If the subordinate PCI buses of the PHB has been
1479 		 * removed or is exactly under error recovery, we
1480 		 * needn't take care of it any more.
1481 		 */
1482 		phb = hose->private_data;
1483 		phb_pe = eeh_phb_pe_get(hose);
1484 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1485 			continue;
1486 
1487 		rc = opal_pci_next_error(phb->opal_id,
1488 					 &frozen_pe_no, &err_type, &severity);
1489 		if (rc != OPAL_SUCCESS) {
1490 			pr_devel("%s: Invalid return value on "
1491 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1492 				 __func__, hose->global_number, rc);
1493 			continue;
1494 		}
1495 
1496 		/* If the PHB doesn't have error, stop processing */
1497 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1498 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1499 			pr_devel("%s: No error found on PHB#%x\n",
1500 				 __func__, hose->global_number);
1501 			continue;
1502 		}
1503 
1504 		/*
1505 		 * Processing the error. We're expecting the error with
1506 		 * highest priority reported upon multiple errors on the
1507 		 * specific PHB.
1508 		 */
1509 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1510 			__func__, be16_to_cpu(err_type),
1511 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1512 			hose->global_number);
1513 		switch (be16_to_cpu(err_type)) {
1514 		case OPAL_EEH_IOC_ERROR:
1515 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1516 				pr_err("EEH: dead IOC detected\n");
1517 				ret = EEH_NEXT_ERR_DEAD_IOC;
1518 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1519 				pr_info("EEH: IOC informative error "
1520 					"detected\n");
1521 				pnv_eeh_get_and_dump_hub_diag(hose);
1522 				ret = EEH_NEXT_ERR_NONE;
1523 			}
1524 
1525 			break;
1526 		case OPAL_EEH_PHB_ERROR:
1527 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1528 				*pe = phb_pe;
1529 				pr_err("EEH: dead PHB#%x detected, "
1530 				       "location: %s\n",
1531 					hose->global_number,
1532 					eeh_pe_loc_get(phb_pe));
1533 				ret = EEH_NEXT_ERR_DEAD_PHB;
1534 			} else if (be16_to_cpu(severity) ==
1535 				   OPAL_EEH_SEV_PHB_FENCED) {
1536 				*pe = phb_pe;
1537 				pr_err("EEH: Fenced PHB#%x detected, "
1538 				       "location: %s\n",
1539 					hose->global_number,
1540 					eeh_pe_loc_get(phb_pe));
1541 				ret = EEH_NEXT_ERR_FENCED_PHB;
1542 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1543 				pr_info("EEH: PHB#%x informative error "
1544 					"detected, location: %s\n",
1545 					hose->global_number,
1546 					eeh_pe_loc_get(phb_pe));
1547 				pnv_eeh_get_phb_diag(phb_pe);
1548 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1549 				ret = EEH_NEXT_ERR_NONE;
1550 			}
1551 
1552 			break;
1553 		case OPAL_EEH_PE_ERROR:
1554 			/*
1555 			 * If we can't find the corresponding PE, we
1556 			 * just try to unfreeze.
1557 			 */
1558 			if (pnv_eeh_get_pe(hose,
1559 				be64_to_cpu(frozen_pe_no), pe)) {
1560 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1561 					hose->global_number, be64_to_cpu(frozen_pe_no));
1562 				pr_info("EEH: PHB location: %s\n",
1563 					eeh_pe_loc_get(phb_pe));
1564 
1565 				/* Dump PHB diag-data */
1566 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1567 					phb->diag_data, phb->diag_data_size);
1568 				if (rc == OPAL_SUCCESS)
1569 					pnv_pci_dump_phb_diag_data(hose,
1570 							phb->diag_data);
1571 
1572 				/* Try best to clear it */
1573 				opal_pci_eeh_freeze_clear(phb->opal_id,
1574 					be64_to_cpu(frozen_pe_no),
1575 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1576 				ret = EEH_NEXT_ERR_NONE;
1577 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1578 				   eeh_pe_passed(*pe)) {
1579 				ret = EEH_NEXT_ERR_NONE;
1580 			} else {
1581 				pr_err("EEH: Frozen PE#%x "
1582 				       "on PHB#%x detected\n",
1583 				       (*pe)->addr,
1584 					(*pe)->phb->global_number);
1585 				pr_err("EEH: PE location: %s, "
1586 				       "PHB location: %s\n",
1587 				       eeh_pe_loc_get(*pe),
1588 				       eeh_pe_loc_get(phb_pe));
1589 				ret = EEH_NEXT_ERR_FROZEN_PE;
1590 			}
1591 
1592 			break;
1593 		default:
1594 			pr_warn("%s: Unexpected error type %d\n",
1595 				__func__, be16_to_cpu(err_type));
1596 		}
1597 
1598 		/*
1599 		 * EEH core will try recover from fenced PHB or
1600 		 * frozen PE. In the time for frozen PE, EEH core
1601 		 * enable IO path for that before collecting logs,
1602 		 * but it ruins the site. So we have to dump the
1603 		 * log in advance here.
1604 		 */
1605 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1606 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1607 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1608 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1609 			pnv_eeh_get_phb_diag(*pe);
1610 
1611 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1612 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1613 							   (*pe)->data);
1614 		}
1615 
1616 		/*
1617 		 * We probably have the frozen parent PE out there and
1618 		 * we need have to handle frozen parent PE firstly.
1619 		 */
1620 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1621 			parent_pe = (*pe)->parent;
1622 			while (parent_pe) {
1623 				/* Hit the ceiling ? */
1624 				if (parent_pe->type & EEH_PE_PHB)
1625 					break;
1626 
1627 				/* Frozen parent PE ? */
1628 				state = eeh_ops->get_state(parent_pe, NULL);
1629 				if (state > 0 &&
1630 				    (state & active_flags) != active_flags)
1631 					*pe = parent_pe;
1632 
1633 				/* Next parent level */
1634 				parent_pe = parent_pe->parent;
1635 			}
1636 
1637 			/* We possibly migrate to another PE */
1638 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1639 		}
1640 
1641 		/*
1642 		 * If we have no errors on the specific PHB or only
1643 		 * informative error there, we continue poking it.
1644 		 * Otherwise, we need actions to be taken by upper
1645 		 * layer.
1646 		 */
1647 		if (ret > EEH_NEXT_ERR_INF)
1648 			break;
1649 	}
1650 
1651 	/* Unmask the event */
1652 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1653 		enable_irq(eeh_event_irq);
1654 
1655 	return ret;
1656 }
1657 
1658 static int pnv_eeh_restore_config(struct pci_dn *pdn)
1659 {
1660 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1661 	struct pnv_phb *phb;
1662 	s64 ret = 0;
1663 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
1664 
1665 	if (!edev)
1666 		return -EEXIST;
1667 
1668 	/*
1669 	 * We have to restore the PCI config space after reset since the
1670 	 * firmware can't see SRIOV VFs.
1671 	 *
1672 	 * FIXME: The MPS, error routing rules, timeout setting are worthy
1673 	 * to be exported by firmware in extendible way.
1674 	 */
1675 	if (edev->physfn) {
1676 		ret = eeh_restore_vf_config(pdn);
1677 	} else {
1678 		phb = pdn->phb->private_data;
1679 		ret = opal_pci_reinit(phb->opal_id,
1680 				      OPAL_REINIT_PCI_DEV, config_addr);
1681 	}
1682 
1683 	if (ret) {
1684 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1685 			__func__, config_addr, ret);
1686 		return -EIO;
1687 	}
1688 
1689 	return ret;
1690 }
1691 
1692 static struct eeh_ops pnv_eeh_ops = {
1693 	.name                   = "powernv",
1694 	.init                   = pnv_eeh_init,
1695 	.probe			= pnv_eeh_probe,
1696 	.set_option             = pnv_eeh_set_option,
1697 	.get_pe_addr            = pnv_eeh_get_pe_addr,
1698 	.get_state              = pnv_eeh_get_state,
1699 	.reset                  = pnv_eeh_reset,
1700 	.wait_state             = pnv_eeh_wait_state,
1701 	.get_log                = pnv_eeh_get_log,
1702 	.configure_bridge       = pnv_eeh_configure_bridge,
1703 	.err_inject		= pnv_eeh_err_inject,
1704 	.read_config            = pnv_eeh_read_config,
1705 	.write_config           = pnv_eeh_write_config,
1706 	.next_error		= pnv_eeh_next_error,
1707 	.restore_config		= pnv_eeh_restore_config,
1708 	.notify_resume		= NULL
1709 };
1710 
1711 #ifdef CONFIG_PCI_IOV
1712 static void pnv_pci_fixup_vf_mps(struct pci_dev *pdev)
1713 {
1714 	struct pci_dn *pdn = pci_get_pdn(pdev);
1715 	int parent_mps;
1716 
1717 	if (!pdev->is_virtfn)
1718 		return;
1719 
1720 	/* Synchronize MPS for VF and PF */
1721 	parent_mps = pcie_get_mps(pdev->physfn);
1722 	if ((128 << pdev->pcie_mpss) >= parent_mps)
1723 		pcie_set_mps(pdev, parent_mps);
1724 	pdn->mps = pcie_get_mps(pdev);
1725 }
1726 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pnv_pci_fixup_vf_mps);
1727 #endif /* CONFIG_PCI_IOV */
1728 
1729 /**
1730  * eeh_powernv_init - Register platform dependent EEH operations
1731  *
1732  * EEH initialization on powernv platform. This function should be
1733  * called before any EEH related functions.
1734  */
1735 static int __init eeh_powernv_init(void)
1736 {
1737 	int ret = -EINVAL;
1738 
1739 	ret = eeh_ops_register(&pnv_eeh_ops);
1740 	if (!ret)
1741 		pr_info("EEH: PowerNV platform initialized\n");
1742 	else
1743 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
1744 
1745 	return ret;
1746 }
1747 machine_early_initcall(powernv, eeh_powernv_init);
1748