xref: /linux/arch/powerpc/platforms/powernv/eeh-powernv.c (revision a36e9f5cfe9eb3a1dce8769c7058251c42705357)
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/irqdomain.h>
15 #include <linux/list.h>
16 #include <linux/msi.h>
17 #include <linux/of.h>
18 #include <linux/pci.h>
19 #include <linux/proc_fs.h>
20 #include <linux/rbtree.h>
21 #include <linux/sched.h>
22 #include <linux/seq_file.h>
23 #include <linux/spinlock.h>
24 
25 #include <asm/eeh.h>
26 #include <asm/eeh_event.h>
27 #include <asm/firmware.h>
28 #include <asm/io.h>
29 #include <asm/iommu.h>
30 #include <asm/machdep.h>
31 #include <asm/msi_bitmap.h>
32 #include <asm/opal.h>
33 #include <asm/ppc-pci.h>
34 #include <asm/pnv-pci.h>
35 
36 #include "powernv.h"
37 #include "pci.h"
38 #include "../../../../drivers/pci/pci.h"
39 
40 static int eeh_event_irq = -EINVAL;
41 
42 static void pnv_pcibios_bus_add_device(struct pci_dev *pdev)
43 {
44 	dev_dbg(&pdev->dev, "EEH: Setting up device\n");
45 	eeh_probe_device(pdev);
46 }
47 
48 static irqreturn_t pnv_eeh_event(int irq, void *data)
49 {
50 	/*
51 	 * We simply send a special EEH event if EEH has been
52 	 * enabled. We don't care about EEH events until we've
53 	 * finished processing the outstanding ones. Event processing
54 	 * gets unmasked in next_error() if EEH is enabled.
55 	 */
56 	disable_irq_nosync(irq);
57 
58 	if (eeh_enabled())
59 		eeh_send_failure_event(NULL);
60 
61 	return IRQ_HANDLED;
62 }
63 
64 #ifdef CONFIG_DEBUG_FS
65 static ssize_t pnv_eeh_ei_write(struct file *filp,
66 				const char __user *user_buf,
67 				size_t count, loff_t *ppos)
68 {
69 	struct pci_controller *hose = filp->private_data;
70 	struct eeh_pe *pe;
71 	int pe_no, type, func;
72 	unsigned long addr, mask;
73 	char buf[50];
74 	int ret;
75 
76 	if (!eeh_ops || !eeh_ops->err_inject)
77 		return -ENXIO;
78 
79 	/* Copy over argument buffer */
80 	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
81 	if (!ret)
82 		return -EFAULT;
83 
84 	/* Retrieve parameters */
85 	ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
86 		     &pe_no, &type, &func, &addr, &mask);
87 	if (ret != 5)
88 		return -EINVAL;
89 
90 	/* Retrieve PE */
91 	pe = eeh_pe_get(hose, pe_no);
92 	if (!pe)
93 		return -ENODEV;
94 
95 	/* Do error injection */
96 	ret = eeh_ops->err_inject(pe, type, func, addr, mask);
97 	return ret < 0 ? ret : count;
98 }
99 
100 static const struct file_operations pnv_eeh_ei_fops = {
101 	.open	= simple_open,
102 	.llseek	= no_llseek,
103 	.write	= pnv_eeh_ei_write,
104 };
105 
106 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val)
107 {
108 	struct pci_controller *hose = data;
109 	struct pnv_phb *phb = hose->private_data;
110 
111 	out_be64(phb->regs + offset, val);
112 	return 0;
113 }
114 
115 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val)
116 {
117 	struct pci_controller *hose = data;
118 	struct pnv_phb *phb = hose->private_data;
119 
120 	*val = in_be64(phb->regs + offset);
121 	return 0;
122 }
123 
124 #define PNV_EEH_DBGFS_ENTRY(name, reg)				\
125 static int pnv_eeh_dbgfs_set_##name(void *data, u64 val)	\
126 {								\
127 	return pnv_eeh_dbgfs_set(data, reg, val);		\
128 }								\
129 								\
130 static int pnv_eeh_dbgfs_get_##name(void *data, u64 *val)	\
131 {								\
132 	return pnv_eeh_dbgfs_get(data, reg, val);		\
133 }								\
134 								\
135 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_dbgfs_ops_##name,		\
136 			pnv_eeh_dbgfs_get_##name,		\
137                         pnv_eeh_dbgfs_set_##name,		\
138 			"0x%llx\n")
139 
140 PNV_EEH_DBGFS_ENTRY(outb, 0xD10);
141 PNV_EEH_DBGFS_ENTRY(inbA, 0xD90);
142 PNV_EEH_DBGFS_ENTRY(inbB, 0xE10);
143 
144 #endif /* CONFIG_DEBUG_FS */
145 
146 static void pnv_eeh_enable_phbs(void)
147 {
148 	struct pci_controller *hose;
149 	struct pnv_phb *phb;
150 
151 	list_for_each_entry(hose, &hose_list, list_node) {
152 		phb = hose->private_data;
153 		/*
154 		 * If EEH is enabled, we're going to rely on that.
155 		 * Otherwise, we restore to conventional mechanism
156 		 * to clear frozen PE during PCI config access.
157 		 */
158 		if (eeh_enabled())
159 			phb->flags |= PNV_PHB_FLAG_EEH;
160 		else
161 			phb->flags &= ~PNV_PHB_FLAG_EEH;
162 	}
163 }
164 
165 /**
166  * pnv_eeh_post_init - EEH platform dependent post initialization
167  *
168  * EEH platform dependent post initialization on powernv. When
169  * the function is called, the EEH PEs and devices should have
170  * been built. If the I/O cache staff has been built, EEH is
171  * ready to supply service.
172  */
173 int pnv_eeh_post_init(void)
174 {
175 	struct pci_controller *hose;
176 	struct pnv_phb *phb;
177 	int ret = 0;
178 
179 	eeh_show_enabled();
180 
181 	/* Register OPAL event notifier */
182 	eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
183 	if (eeh_event_irq < 0) {
184 		pr_err("%s: Can't register OPAL event interrupt (%d)\n",
185 		       __func__, eeh_event_irq);
186 		return eeh_event_irq;
187 	}
188 
189 	ret = request_irq(eeh_event_irq, pnv_eeh_event,
190 			  IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
191 	if (ret < 0) {
192 		irq_dispose_mapping(eeh_event_irq);
193 		pr_err("%s: Can't request OPAL event interrupt (%d)\n",
194 		       __func__, eeh_event_irq);
195 		return ret;
196 	}
197 
198 	if (!eeh_enabled())
199 		disable_irq(eeh_event_irq);
200 
201 	pnv_eeh_enable_phbs();
202 
203 	list_for_each_entry(hose, &hose_list, list_node) {
204 		phb = hose->private_data;
205 
206 		/* Create debugfs entries */
207 #ifdef CONFIG_DEBUG_FS
208 		if (phb->has_dbgfs || !phb->dbgfs)
209 			continue;
210 
211 		phb->has_dbgfs = 1;
212 		debugfs_create_file("err_injct", 0200,
213 				    phb->dbgfs, hose,
214 				    &pnv_eeh_ei_fops);
215 
216 		debugfs_create_file("err_injct_outbound", 0600,
217 				    phb->dbgfs, hose,
218 				    &pnv_eeh_dbgfs_ops_outb);
219 		debugfs_create_file("err_injct_inboundA", 0600,
220 				    phb->dbgfs, hose,
221 				    &pnv_eeh_dbgfs_ops_inbA);
222 		debugfs_create_file("err_injct_inboundB", 0600,
223 				    phb->dbgfs, hose,
224 				    &pnv_eeh_dbgfs_ops_inbB);
225 #endif /* CONFIG_DEBUG_FS */
226 	}
227 
228 	return ret;
229 }
230 
231 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
232 {
233 	int pos = PCI_CAPABILITY_LIST;
234 	int cnt = 48;   /* Maximal number of capabilities */
235 	u32 status, id;
236 
237 	if (!pdn)
238 		return 0;
239 
240 	/* Check if the device supports capabilities */
241 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
242 	if (!(status & PCI_STATUS_CAP_LIST))
243 		return 0;
244 
245 	while (cnt--) {
246 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
247 		if (pos < 0x40)
248 			break;
249 
250 		pos &= ~3;
251 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
252 		if (id == 0xff)
253 			break;
254 
255 		/* Found */
256 		if (id == cap)
257 			return pos;
258 
259 		/* Next one */
260 		pos += PCI_CAP_LIST_NEXT;
261 	}
262 
263 	return 0;
264 }
265 
266 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
267 {
268 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
269 	u32 header;
270 	int pos = 256, ttl = (4096 - 256) / 8;
271 
272 	if (!edev || !edev->pcie_cap)
273 		return 0;
274 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
275 		return 0;
276 	else if (!header)
277 		return 0;
278 
279 	while (ttl-- > 0) {
280 		if (PCI_EXT_CAP_ID(header) == cap && pos)
281 			return pos;
282 
283 		pos = PCI_EXT_CAP_NEXT(header);
284 		if (pos < 256)
285 			break;
286 
287 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
288 			break;
289 	}
290 
291 	return 0;
292 }
293 
294 static struct eeh_pe *pnv_eeh_get_upstream_pe(struct pci_dev *pdev)
295 {
296 	struct pci_controller *hose = pdev->bus->sysdata;
297 	struct pnv_phb *phb = hose->private_data;
298 	struct pci_dev *parent = pdev->bus->self;
299 
300 #ifdef CONFIG_PCI_IOV
301 	/* for VFs we use the PF's PE as the upstream PE */
302 	if (pdev->is_virtfn)
303 		parent = pdev->physfn;
304 #endif
305 
306 	/* otherwise use the PE of our parent bridge */
307 	if (parent) {
308 		struct pnv_ioda_pe *ioda_pe = pnv_ioda_get_pe(parent);
309 
310 		return eeh_pe_get(phb->hose, ioda_pe->pe_number);
311 	}
312 
313 	return NULL;
314 }
315 
316 /**
317  * pnv_eeh_probe - Do probe on PCI device
318  * @pdev: pci_dev to probe
319  *
320  * Create, or find the existing, eeh_dev for this pci_dev.
321  */
322 static struct eeh_dev *pnv_eeh_probe(struct pci_dev *pdev)
323 {
324 	struct pci_dn *pdn = pci_get_pdn(pdev);
325 	struct pci_controller *hose = pdn->phb;
326 	struct pnv_phb *phb = hose->private_data;
327 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
328 	struct eeh_pe *upstream_pe;
329 	uint32_t pcie_flags;
330 	int ret;
331 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
332 
333 	/*
334 	 * When probing the root bridge, which doesn't have any
335 	 * subordinate PCI devices. We don't have OF node for
336 	 * the root bridge. So it's not reasonable to continue
337 	 * the probing.
338 	 */
339 	if (!edev || edev->pe)
340 		return NULL;
341 
342 	/* already configured? */
343 	if (edev->pdev) {
344 		pr_debug("%s: found existing edev for %04x:%02x:%02x.%01x\n",
345 			__func__, hose->global_number, config_addr >> 8,
346 			PCI_SLOT(config_addr), PCI_FUNC(config_addr));
347 		return edev;
348 	}
349 
350 	/* Skip for PCI-ISA bridge */
351 	if ((pdev->class >> 8) == PCI_CLASS_BRIDGE_ISA)
352 		return NULL;
353 
354 	eeh_edev_dbg(edev, "Probing device\n");
355 
356 	/* Initialize eeh device */
357 	edev->mode	&= 0xFFFFFF00;
358 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
359 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
360 	edev->af_cap   = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF);
361 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
362 	if ((pdev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
363 		edev->mode |= EEH_DEV_BRIDGE;
364 		if (edev->pcie_cap) {
365 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
366 					 2, &pcie_flags);
367 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
368 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
369 				edev->mode |= EEH_DEV_ROOT_PORT;
370 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
371 				edev->mode |= EEH_DEV_DS_PORT;
372 		}
373 	}
374 
375 	edev->pe_config_addr = phb->ioda.pe_rmap[config_addr];
376 
377 	upstream_pe = pnv_eeh_get_upstream_pe(pdev);
378 
379 	/* Create PE */
380 	ret = eeh_pe_tree_insert(edev, upstream_pe);
381 	if (ret) {
382 		eeh_edev_warn(edev, "Failed to add device to PE (code %d)\n", ret);
383 		return NULL;
384 	}
385 
386 	/*
387 	 * If the PE contains any one of following adapters, the
388 	 * PCI config space can't be accessed when dumping EEH log.
389 	 * Otherwise, we will run into fenced PHB caused by shortage
390 	 * of outbound credits in the adapter. The PCI config access
391 	 * should be blocked until PE reset. MMIO access is dropped
392 	 * by hardware certainly. In order to drop PCI config requests,
393 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
394 	 * will be checked in the backend for PE state retrieval. If
395 	 * the PE becomes frozen for the first time and the flag has
396 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
397 	 * that PE to block its config space.
398 	 *
399 	 * Broadcom BCM5718 2-ports NICs (14e4:1656)
400 	 * Broadcom Austin 4-ports NICs (14e4:1657)
401 	 * Broadcom Shiner 4-ports 1G NICs (14e4:168a)
402 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
403 	 */
404 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
405 	     pdn->device_id == 0x1656) ||
406 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
407 	     pdn->device_id == 0x1657) ||
408 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
409 	     pdn->device_id == 0x168a) ||
410 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
411 	     pdn->device_id == 0x168e))
412 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
413 
414 	/*
415 	 * Cache the PE primary bus, which can't be fetched when
416 	 * full hotplug is in progress. In that case, all child
417 	 * PCI devices of the PE are expected to be removed prior
418 	 * to PE reset.
419 	 */
420 	if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
421 		edev->pe->bus = pci_find_bus(hose->global_number,
422 					     pdn->busno);
423 		if (edev->pe->bus)
424 			edev->pe->state |= EEH_PE_PRI_BUS;
425 	}
426 
427 	/*
428 	 * Enable EEH explicitly so that we will do EEH check
429 	 * while accessing I/O stuff
430 	 */
431 	if (!eeh_has_flag(EEH_ENABLED)) {
432 		enable_irq(eeh_event_irq);
433 		pnv_eeh_enable_phbs();
434 		eeh_add_flag(EEH_ENABLED);
435 	}
436 
437 	/* Save memory bars */
438 	eeh_save_bars(edev);
439 
440 	eeh_edev_dbg(edev, "EEH enabled on device\n");
441 
442 	return edev;
443 }
444 
445 /**
446  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
447  * @pe: EEH PE
448  * @option: operation to be issued
449  *
450  * The function is used to control the EEH functionality globally.
451  * Currently, following options are support according to PAPR:
452  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
453  */
454 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
455 {
456 	struct pci_controller *hose = pe->phb;
457 	struct pnv_phb *phb = hose->private_data;
458 	bool freeze_pe = false;
459 	int opt;
460 	s64 rc;
461 
462 	switch (option) {
463 	case EEH_OPT_DISABLE:
464 		return -EPERM;
465 	case EEH_OPT_ENABLE:
466 		return 0;
467 	case EEH_OPT_THAW_MMIO:
468 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
469 		break;
470 	case EEH_OPT_THAW_DMA:
471 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
472 		break;
473 	case EEH_OPT_FREEZE_PE:
474 		freeze_pe = true;
475 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
476 		break;
477 	default:
478 		pr_warn("%s: Invalid option %d\n", __func__, option);
479 		return -EINVAL;
480 	}
481 
482 	/* Freeze master and slave PEs if PHB supports compound PEs */
483 	if (freeze_pe) {
484 		if (phb->freeze_pe) {
485 			phb->freeze_pe(phb, pe->addr);
486 			return 0;
487 		}
488 
489 		rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt);
490 		if (rc != OPAL_SUCCESS) {
491 			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
492 				__func__, rc, phb->hose->global_number,
493 				pe->addr);
494 			return -EIO;
495 		}
496 
497 		return 0;
498 	}
499 
500 	/* Unfreeze master and slave PEs if PHB supports */
501 	if (phb->unfreeze_pe)
502 		return phb->unfreeze_pe(phb, pe->addr, opt);
503 
504 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt);
505 	if (rc != OPAL_SUCCESS) {
506 		pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n",
507 			__func__, rc, option, phb->hose->global_number,
508 			pe->addr);
509 		return -EIO;
510 	}
511 
512 	return 0;
513 }
514 
515 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
516 {
517 	struct pnv_phb *phb = pe->phb->private_data;
518 	s64 rc;
519 
520 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
521 					 phb->diag_data_size);
522 	if (rc != OPAL_SUCCESS)
523 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
524 			__func__, rc, pe->phb->global_number);
525 }
526 
527 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
528 {
529 	struct pnv_phb *phb = pe->phb->private_data;
530 	u8 fstate = 0;
531 	__be16 pcierr = 0;
532 	s64 rc;
533 	int result = 0;
534 
535 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
536 					pe->addr,
537 					&fstate,
538 					&pcierr,
539 					NULL);
540 	if (rc != OPAL_SUCCESS) {
541 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
542 			__func__, rc, phb->hose->global_number);
543 		return EEH_STATE_NOT_SUPPORT;
544 	}
545 
546 	/*
547 	 * Check PHB state. If the PHB is frozen for the
548 	 * first time, to dump the PHB diag-data.
549 	 */
550 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
551 		result = (EEH_STATE_MMIO_ACTIVE  |
552 			  EEH_STATE_DMA_ACTIVE   |
553 			  EEH_STATE_MMIO_ENABLED |
554 			  EEH_STATE_DMA_ENABLED);
555 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
556 		eeh_pe_mark_isolated(pe);
557 		pnv_eeh_get_phb_diag(pe);
558 
559 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
560 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
561 	}
562 
563 	return result;
564 }
565 
566 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
567 {
568 	struct pnv_phb *phb = pe->phb->private_data;
569 	u8 fstate = 0;
570 	__be16 pcierr = 0;
571 	s64 rc;
572 	int result;
573 
574 	/*
575 	 * We don't clobber hardware frozen state until PE
576 	 * reset is completed. In order to keep EEH core
577 	 * moving forward, we have to return operational
578 	 * state during PE reset.
579 	 */
580 	if (pe->state & EEH_PE_RESET) {
581 		result = (EEH_STATE_MMIO_ACTIVE  |
582 			  EEH_STATE_DMA_ACTIVE   |
583 			  EEH_STATE_MMIO_ENABLED |
584 			  EEH_STATE_DMA_ENABLED);
585 		return result;
586 	}
587 
588 	/*
589 	 * Fetch PE state from hardware. If the PHB
590 	 * supports compound PE, let it handle that.
591 	 */
592 	if (phb->get_pe_state) {
593 		fstate = phb->get_pe_state(phb, pe->addr);
594 	} else {
595 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
596 						pe->addr,
597 						&fstate,
598 						&pcierr,
599 						NULL);
600 		if (rc != OPAL_SUCCESS) {
601 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
602 				__func__, rc, phb->hose->global_number,
603 				pe->addr);
604 			return EEH_STATE_NOT_SUPPORT;
605 		}
606 	}
607 
608 	/* Figure out state */
609 	switch (fstate) {
610 	case OPAL_EEH_STOPPED_NOT_FROZEN:
611 		result = (EEH_STATE_MMIO_ACTIVE  |
612 			  EEH_STATE_DMA_ACTIVE   |
613 			  EEH_STATE_MMIO_ENABLED |
614 			  EEH_STATE_DMA_ENABLED);
615 		break;
616 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
617 		result = (EEH_STATE_DMA_ACTIVE |
618 			  EEH_STATE_DMA_ENABLED);
619 		break;
620 	case OPAL_EEH_STOPPED_DMA_FREEZE:
621 		result = (EEH_STATE_MMIO_ACTIVE |
622 			  EEH_STATE_MMIO_ENABLED);
623 		break;
624 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
625 		result = 0;
626 		break;
627 	case OPAL_EEH_STOPPED_RESET:
628 		result = EEH_STATE_RESET_ACTIVE;
629 		break;
630 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
631 		result = EEH_STATE_UNAVAILABLE;
632 		break;
633 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
634 		result = EEH_STATE_NOT_SUPPORT;
635 		break;
636 	default:
637 		result = EEH_STATE_NOT_SUPPORT;
638 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
639 			__func__, phb->hose->global_number,
640 			pe->addr, fstate);
641 	}
642 
643 	/*
644 	 * If PHB supports compound PE, to freeze all
645 	 * slave PEs for consistency.
646 	 *
647 	 * If the PE is switching to frozen state for the
648 	 * first time, to dump the PHB diag-data.
649 	 */
650 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
651 	    !(result & EEH_STATE_UNAVAILABLE) &&
652 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
653 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
654 	    !(pe->state & EEH_PE_ISOLATED)) {
655 		if (phb->freeze_pe)
656 			phb->freeze_pe(phb, pe->addr);
657 
658 		eeh_pe_mark_isolated(pe);
659 		pnv_eeh_get_phb_diag(pe);
660 
661 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
662 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
663 	}
664 
665 	return result;
666 }
667 
668 /**
669  * pnv_eeh_get_state - Retrieve PE state
670  * @pe: EEH PE
671  * @delay: delay while PE state is temporarily unavailable
672  *
673  * Retrieve the state of the specified PE. For IODA-compitable
674  * platform, it should be retrieved from IODA table. Therefore,
675  * we prefer passing down to hardware implementation to handle
676  * it.
677  */
678 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
679 {
680 	int ret;
681 
682 	if (pe->type & EEH_PE_PHB)
683 		ret = pnv_eeh_get_phb_state(pe);
684 	else
685 		ret = pnv_eeh_get_pe_state(pe);
686 
687 	if (!delay)
688 		return ret;
689 
690 	/*
691 	 * If the PE state is temporarily unavailable,
692 	 * to inform the EEH core delay for default
693 	 * period (1 second)
694 	 */
695 	*delay = 0;
696 	if (ret & EEH_STATE_UNAVAILABLE)
697 		*delay = 1000;
698 
699 	return ret;
700 }
701 
702 static s64 pnv_eeh_poll(unsigned long id)
703 {
704 	s64 rc = OPAL_HARDWARE;
705 
706 	while (1) {
707 		rc = opal_pci_poll(id);
708 		if (rc <= 0)
709 			break;
710 
711 		if (system_state < SYSTEM_RUNNING)
712 			udelay(1000 * rc);
713 		else
714 			msleep(rc);
715 	}
716 
717 	return rc;
718 }
719 
720 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
721 {
722 	struct pnv_phb *phb = hose->private_data;
723 	s64 rc = OPAL_HARDWARE;
724 
725 	pr_debug("%s: Reset PHB#%x, option=%d\n",
726 		 __func__, hose->global_number, option);
727 
728 	/* Issue PHB complete reset request */
729 	if (option == EEH_RESET_FUNDAMENTAL ||
730 	    option == EEH_RESET_HOT)
731 		rc = opal_pci_reset(phb->opal_id,
732 				    OPAL_RESET_PHB_COMPLETE,
733 				    OPAL_ASSERT_RESET);
734 	else if (option == EEH_RESET_DEACTIVATE)
735 		rc = opal_pci_reset(phb->opal_id,
736 				    OPAL_RESET_PHB_COMPLETE,
737 				    OPAL_DEASSERT_RESET);
738 	if (rc < 0)
739 		goto out;
740 
741 	/*
742 	 * Poll state of the PHB until the request is done
743 	 * successfully. The PHB reset is usually PHB complete
744 	 * reset followed by hot reset on root bus. So we also
745 	 * need the PCI bus settlement delay.
746 	 */
747 	if (rc > 0)
748 		rc = pnv_eeh_poll(phb->opal_id);
749 	if (option == EEH_RESET_DEACTIVATE) {
750 		if (system_state < SYSTEM_RUNNING)
751 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
752 		else
753 			msleep(EEH_PE_RST_SETTLE_TIME);
754 	}
755 out:
756 	if (rc != OPAL_SUCCESS)
757 		return -EIO;
758 
759 	return 0;
760 }
761 
762 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
763 {
764 	struct pnv_phb *phb = hose->private_data;
765 	s64 rc = OPAL_HARDWARE;
766 
767 	pr_debug("%s: Reset PHB#%x, option=%d\n",
768 		 __func__, hose->global_number, option);
769 
770 	/*
771 	 * During the reset deassert time, we needn't care
772 	 * the reset scope because the firmware does nothing
773 	 * for fundamental or hot reset during deassert phase.
774 	 */
775 	if (option == EEH_RESET_FUNDAMENTAL)
776 		rc = opal_pci_reset(phb->opal_id,
777 				    OPAL_RESET_PCI_FUNDAMENTAL,
778 				    OPAL_ASSERT_RESET);
779 	else if (option == EEH_RESET_HOT)
780 		rc = opal_pci_reset(phb->opal_id,
781 				    OPAL_RESET_PCI_HOT,
782 				    OPAL_ASSERT_RESET);
783 	else if (option == EEH_RESET_DEACTIVATE)
784 		rc = opal_pci_reset(phb->opal_id,
785 				    OPAL_RESET_PCI_HOT,
786 				    OPAL_DEASSERT_RESET);
787 	if (rc < 0)
788 		goto out;
789 
790 	/* Poll state of the PHB until the request is done */
791 	if (rc > 0)
792 		rc = pnv_eeh_poll(phb->opal_id);
793 	if (option == EEH_RESET_DEACTIVATE)
794 		msleep(EEH_PE_RST_SETTLE_TIME);
795 out:
796 	if (rc != OPAL_SUCCESS)
797 		return -EIO;
798 
799 	return 0;
800 }
801 
802 static int __pnv_eeh_bridge_reset(struct pci_dev *dev, int option)
803 {
804 	struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
805 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
806 	int aer = edev ? edev->aer_cap : 0;
807 	u32 ctrl;
808 
809 	pr_debug("%s: Secondary Reset PCI bus %04x:%02x with option %d\n",
810 		 __func__, pci_domain_nr(dev->bus),
811 		 dev->bus->number, option);
812 
813 	switch (option) {
814 	case EEH_RESET_FUNDAMENTAL:
815 	case EEH_RESET_HOT:
816 		/* Don't report linkDown event */
817 		if (aer) {
818 			eeh_ops->read_config(edev, aer + PCI_ERR_UNCOR_MASK,
819 					     4, &ctrl);
820 			ctrl |= PCI_ERR_UNC_SURPDN;
821 			eeh_ops->write_config(edev, aer + PCI_ERR_UNCOR_MASK,
822 					      4, ctrl);
823 		}
824 
825 		eeh_ops->read_config(edev, PCI_BRIDGE_CONTROL, 2, &ctrl);
826 		ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
827 		eeh_ops->write_config(edev, PCI_BRIDGE_CONTROL, 2, ctrl);
828 
829 		msleep(EEH_PE_RST_HOLD_TIME);
830 		break;
831 	case EEH_RESET_DEACTIVATE:
832 		eeh_ops->read_config(edev, PCI_BRIDGE_CONTROL, 2, &ctrl);
833 		ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
834 		eeh_ops->write_config(edev, PCI_BRIDGE_CONTROL, 2, ctrl);
835 
836 		msleep(EEH_PE_RST_SETTLE_TIME);
837 
838 		/* Continue reporting linkDown event */
839 		if (aer) {
840 			eeh_ops->read_config(edev, aer + PCI_ERR_UNCOR_MASK,
841 					     4, &ctrl);
842 			ctrl &= ~PCI_ERR_UNC_SURPDN;
843 			eeh_ops->write_config(edev, aer + PCI_ERR_UNCOR_MASK,
844 					      4, ctrl);
845 		}
846 
847 		break;
848 	}
849 
850 	return 0;
851 }
852 
853 static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option)
854 {
855 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
856 	struct pnv_phb *phb = hose->private_data;
857 	struct device_node *dn = pci_device_to_OF_node(pdev);
858 	uint64_t id = PCI_SLOT_ID(phb->opal_id, pci_dev_id(pdev));
859 	uint8_t scope;
860 	int64_t rc;
861 
862 	/* Hot reset to the bus if firmware cannot handle */
863 	if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL))
864 		return __pnv_eeh_bridge_reset(pdev, option);
865 
866 	pr_debug("%s: FW reset PCI bus %04x:%02x with option %d\n",
867 		 __func__, pci_domain_nr(pdev->bus),
868 		 pdev->bus->number, option);
869 
870 	switch (option) {
871 	case EEH_RESET_FUNDAMENTAL:
872 		scope = OPAL_RESET_PCI_FUNDAMENTAL;
873 		break;
874 	case EEH_RESET_HOT:
875 		scope = OPAL_RESET_PCI_HOT;
876 		break;
877 	case EEH_RESET_DEACTIVATE:
878 		return 0;
879 	default:
880 		dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n",
881 			__func__, option);
882 		return -EINVAL;
883 	}
884 
885 	rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET);
886 	if (rc <= OPAL_SUCCESS)
887 		goto out;
888 
889 	rc = pnv_eeh_poll(id);
890 out:
891 	return (rc == OPAL_SUCCESS) ? 0 : -EIO;
892 }
893 
894 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
895 {
896 	struct pci_controller *hose;
897 
898 	if (pci_is_root_bus(dev->bus)) {
899 		hose = pci_bus_to_host(dev->bus);
900 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
901 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
902 	} else {
903 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
904 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
905 	}
906 }
907 
908 static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type,
909 				     int pos, u16 mask)
910 {
911 	struct eeh_dev *edev = pdn->edev;
912 	int i, status = 0;
913 
914 	/* Wait for Transaction Pending bit to be cleared */
915 	for (i = 0; i < 4; i++) {
916 		eeh_ops->read_config(edev, pos, 2, &status);
917 		if (!(status & mask))
918 			return;
919 
920 		msleep((1 << i) * 100);
921 	}
922 
923 	pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n",
924 		__func__, type,
925 		pdn->phb->global_number, pdn->busno,
926 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
927 }
928 
929 static int pnv_eeh_do_flr(struct pci_dn *pdn, int option)
930 {
931 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
932 	u32 reg = 0;
933 
934 	if (WARN_ON(!edev->pcie_cap))
935 		return -ENOTTY;
936 
937 	eeh_ops->read_config(edev, edev->pcie_cap + PCI_EXP_DEVCAP, 4, &reg);
938 	if (!(reg & PCI_EXP_DEVCAP_FLR))
939 		return -ENOTTY;
940 
941 	switch (option) {
942 	case EEH_RESET_HOT:
943 	case EEH_RESET_FUNDAMENTAL:
944 		pnv_eeh_wait_for_pending(pdn, "",
945 					 edev->pcie_cap + PCI_EXP_DEVSTA,
946 					 PCI_EXP_DEVSTA_TRPND);
947 		eeh_ops->read_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL,
948 				     4, &reg);
949 		reg |= PCI_EXP_DEVCTL_BCR_FLR;
950 		eeh_ops->write_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL,
951 				      4, reg);
952 		msleep(EEH_PE_RST_HOLD_TIME);
953 		break;
954 	case EEH_RESET_DEACTIVATE:
955 		eeh_ops->read_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL,
956 				     4, &reg);
957 		reg &= ~PCI_EXP_DEVCTL_BCR_FLR;
958 		eeh_ops->write_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL,
959 				      4, reg);
960 		msleep(EEH_PE_RST_SETTLE_TIME);
961 		break;
962 	}
963 
964 	return 0;
965 }
966 
967 static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option)
968 {
969 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
970 	u32 cap = 0;
971 
972 	if (WARN_ON(!edev->af_cap))
973 		return -ENOTTY;
974 
975 	eeh_ops->read_config(edev, edev->af_cap + PCI_AF_CAP, 1, &cap);
976 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
977 		return -ENOTTY;
978 
979 	switch (option) {
980 	case EEH_RESET_HOT:
981 	case EEH_RESET_FUNDAMENTAL:
982 		/*
983 		 * Wait for Transaction Pending bit to clear. A word-aligned
984 		 * test is used, so we use the control offset rather than status
985 		 * and shift the test bit to match.
986 		 */
987 		pnv_eeh_wait_for_pending(pdn, "AF",
988 					 edev->af_cap + PCI_AF_CTRL,
989 					 PCI_AF_STATUS_TP << 8);
990 		eeh_ops->write_config(edev, edev->af_cap + PCI_AF_CTRL,
991 				      1, PCI_AF_CTRL_FLR);
992 		msleep(EEH_PE_RST_HOLD_TIME);
993 		break;
994 	case EEH_RESET_DEACTIVATE:
995 		eeh_ops->write_config(edev, edev->af_cap + PCI_AF_CTRL, 1, 0);
996 		msleep(EEH_PE_RST_SETTLE_TIME);
997 		break;
998 	}
999 
1000 	return 0;
1001 }
1002 
1003 static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option)
1004 {
1005 	struct eeh_dev *edev;
1006 	struct pci_dn *pdn;
1007 	int ret;
1008 
1009 	/* The VF PE should have only one child device */
1010 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, entry);
1011 	pdn = eeh_dev_to_pdn(edev);
1012 	if (!pdn)
1013 		return -ENXIO;
1014 
1015 	ret = pnv_eeh_do_flr(pdn, option);
1016 	if (!ret)
1017 		return ret;
1018 
1019 	return pnv_eeh_do_af_flr(pdn, option);
1020 }
1021 
1022 /**
1023  * pnv_eeh_reset - Reset the specified PE
1024  * @pe: EEH PE
1025  * @option: reset option
1026  *
1027  * Do reset on the indicated PE. For PCI bus sensitive PE,
1028  * we need to reset the parent p2p bridge. The PHB has to
1029  * be reinitialized if the p2p bridge is root bridge. For
1030  * PCI device sensitive PE, we will try to reset the device
1031  * through FLR. For now, we don't have OPAL APIs to do HARD
1032  * reset yet, so all reset would be SOFT (HOT) reset.
1033  */
1034 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
1035 {
1036 	struct pci_controller *hose = pe->phb;
1037 	struct pnv_phb *phb;
1038 	struct pci_bus *bus;
1039 	int64_t rc;
1040 
1041 	/*
1042 	 * For PHB reset, we always have complete reset. For those PEs whose
1043 	 * primary bus derived from root complex (root bus) or root port
1044 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
1045 	 * For other PEs, we always have hot reset on the PE primary bus.
1046 	 *
1047 	 * Here, we have different design to pHyp, which always clear the
1048 	 * frozen state during PE reset. However, the good idea here from
1049 	 * benh is to keep frozen state before we get PE reset done completely
1050 	 * (until BAR restore). With the frozen state, HW drops illegal IO
1051 	 * or MMIO access, which can incur recursive frozen PE during PE
1052 	 * reset. The side effect is that EEH core has to clear the frozen
1053 	 * state explicitly after BAR restore.
1054 	 */
1055 	if (pe->type & EEH_PE_PHB)
1056 		return pnv_eeh_phb_reset(hose, option);
1057 
1058 	/*
1059 	 * The frozen PE might be caused by PAPR error injection
1060 	 * registers, which are expected to be cleared after hitting
1061 	 * frozen PE as stated in the hardware spec. Unfortunately,
1062 	 * that's not true on P7IOC. So we have to clear it manually
1063 	 * to avoid recursive EEH errors during recovery.
1064 	 */
1065 	phb = hose->private_data;
1066 	if (phb->model == PNV_PHB_MODEL_P7IOC &&
1067 	    (option == EEH_RESET_HOT ||
1068 	     option == EEH_RESET_FUNDAMENTAL)) {
1069 		rc = opal_pci_reset(phb->opal_id,
1070 				    OPAL_RESET_PHB_ERROR,
1071 				    OPAL_ASSERT_RESET);
1072 		if (rc != OPAL_SUCCESS) {
1073 			pr_warn("%s: Failure %lld clearing error injection registers\n",
1074 				__func__, rc);
1075 			return -EIO;
1076 		}
1077 	}
1078 
1079 	if (pe->type & EEH_PE_VF)
1080 		return pnv_eeh_reset_vf_pe(pe, option);
1081 
1082 	bus = eeh_pe_bus_get(pe);
1083 	if (!bus) {
1084 		pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
1085 			__func__, pe->phb->global_number, pe->addr);
1086 		return -EIO;
1087 	}
1088 
1089 	if (pci_is_root_bus(bus))
1090 		return pnv_eeh_root_reset(hose, option);
1091 
1092 	/*
1093 	 * For hot resets try use the generic PCI error recovery reset
1094 	 * functions. These correctly handles the case where the secondary
1095 	 * bus is behind a hotplug slot and it will use the slot provided
1096 	 * reset methods to prevent spurious hotplug events during the reset.
1097 	 *
1098 	 * Fundamental resets need to be handled internally to EEH since the
1099 	 * PCI core doesn't really have a concept of a fundamental reset,
1100 	 * mainly because there's no standard way to generate one. Only a
1101 	 * few devices require an FRESET so it should be fine.
1102 	 */
1103 	if (option != EEH_RESET_FUNDAMENTAL) {
1104 		/*
1105 		 * NB: Skiboot and pnv_eeh_bridge_reset() also no-op the
1106 		 *     de-assert step. It's like the OPAL reset API was
1107 		 *     poorly designed or something...
1108 		 */
1109 		if (option == EEH_RESET_DEACTIVATE)
1110 			return 0;
1111 
1112 		rc = pci_bus_error_reset(bus->self);
1113 		if (!rc)
1114 			return 0;
1115 	}
1116 
1117 	/* otherwise, use the generic bridge reset. this might call into FW */
1118 	if (pci_is_root_bus(bus->parent))
1119 		return pnv_eeh_root_reset(hose, option);
1120 	return pnv_eeh_bridge_reset(bus->self, option);
1121 }
1122 
1123 /**
1124  * pnv_eeh_get_log - Retrieve error log
1125  * @pe: EEH PE
1126  * @severity: temporary or permanent error log
1127  * @drv_log: driver log to be combined with retrieved error log
1128  * @len: length of driver log
1129  *
1130  * Retrieve the temporary or permanent error from the PE.
1131  */
1132 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1133 			   char *drv_log, unsigned long len)
1134 {
1135 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1136 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1137 
1138 	return 0;
1139 }
1140 
1141 /**
1142  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1143  * @pe: EEH PE
1144  *
1145  * The function will be called to reconfigure the bridges included
1146  * in the specified PE so that the mulfunctional PE would be recovered
1147  * again.
1148  */
1149 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1150 {
1151 	return 0;
1152 }
1153 
1154 /**
1155  * pnv_pe_err_inject - Inject specified error to the indicated PE
1156  * @pe: the indicated PE
1157  * @type: error type
1158  * @func: specific error type
1159  * @addr: address
1160  * @mask: address mask
1161  *
1162  * The routine is called to inject specified error, which is
1163  * determined by @type and @func, to the indicated PE for
1164  * testing purpose.
1165  */
1166 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1167 			      unsigned long addr, unsigned long mask)
1168 {
1169 	struct pci_controller *hose = pe->phb;
1170 	struct pnv_phb *phb = hose->private_data;
1171 	s64 rc;
1172 
1173 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1174 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1175 		pr_warn("%s: Invalid error type %d\n",
1176 			__func__, type);
1177 		return -ERANGE;
1178 	}
1179 
1180 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1181 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1182 		pr_warn("%s: Invalid error function %d\n",
1183 			__func__, func);
1184 		return -ERANGE;
1185 	}
1186 
1187 	/* Firmware supports error injection ? */
1188 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1189 		pr_warn("%s: Firmware doesn't support error injection\n",
1190 			__func__);
1191 		return -ENXIO;
1192 	}
1193 
1194 	/* Do error injection */
1195 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1196 				 type, func, addr, mask);
1197 	if (rc != OPAL_SUCCESS) {
1198 		pr_warn("%s: Failure %lld injecting error "
1199 			"%d-%d to PHB#%x-PE#%x\n",
1200 			__func__, rc, type, func,
1201 			hose->global_number, pe->addr);
1202 		return -EIO;
1203 	}
1204 
1205 	return 0;
1206 }
1207 
1208 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1209 {
1210 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1211 
1212 	if (!edev || !edev->pe)
1213 		return false;
1214 
1215 	/*
1216 	 * We will issue FLR or AF FLR to all VFs, which are contained
1217 	 * in VF PE. It relies on the EEH PCI config accessors. So we
1218 	 * can't block them during the window.
1219 	 */
1220 	if (edev->physfn && (edev->pe->state & EEH_PE_RESET))
1221 		return false;
1222 
1223 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1224 		return true;
1225 
1226 	return false;
1227 }
1228 
1229 static int pnv_eeh_read_config(struct eeh_dev *edev,
1230 			       int where, int size, u32 *val)
1231 {
1232 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
1233 
1234 	if (!pdn)
1235 		return PCIBIOS_DEVICE_NOT_FOUND;
1236 
1237 	if (pnv_eeh_cfg_blocked(pdn)) {
1238 		*val = 0xFFFFFFFF;
1239 		return PCIBIOS_SET_FAILED;
1240 	}
1241 
1242 	return pnv_pci_cfg_read(pdn, where, size, val);
1243 }
1244 
1245 static int pnv_eeh_write_config(struct eeh_dev *edev,
1246 				int where, int size, u32 val)
1247 {
1248 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
1249 
1250 	if (!pdn)
1251 		return PCIBIOS_DEVICE_NOT_FOUND;
1252 
1253 	if (pnv_eeh_cfg_blocked(pdn))
1254 		return PCIBIOS_SET_FAILED;
1255 
1256 	return pnv_pci_cfg_write(pdn, where, size, val);
1257 }
1258 
1259 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1260 {
1261 	/* GEM */
1262 	if (data->gemXfir || data->gemRfir ||
1263 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1264 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1265 			be64_to_cpu(data->gemXfir),
1266 			be64_to_cpu(data->gemRfir),
1267 			be64_to_cpu(data->gemRirqfir),
1268 			be64_to_cpu(data->gemMask),
1269 			be64_to_cpu(data->gemRwof));
1270 
1271 	/* LEM */
1272 	if (data->lemFir || data->lemErrMask ||
1273 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1274 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1275 			be64_to_cpu(data->lemFir),
1276 			be64_to_cpu(data->lemErrMask),
1277 			be64_to_cpu(data->lemAction0),
1278 			be64_to_cpu(data->lemAction1),
1279 			be64_to_cpu(data->lemWof));
1280 }
1281 
1282 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1283 {
1284 	struct pnv_phb *phb = hose->private_data;
1285 	struct OpalIoP7IOCErrorData *data =
1286 		(struct OpalIoP7IOCErrorData*)phb->diag_data;
1287 	long rc;
1288 
1289 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1290 	if (rc != OPAL_SUCCESS) {
1291 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1292 			__func__, phb->hub_id, rc);
1293 		return;
1294 	}
1295 
1296 	switch (be16_to_cpu(data->type)) {
1297 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1298 		pr_info("P7IOC diag-data for RGC\n\n");
1299 		pnv_eeh_dump_hub_diag_common(data);
1300 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1301 			pr_info("  RGC: %016llx %016llx\n",
1302 				be64_to_cpu(data->rgc.rgcStatus),
1303 				be64_to_cpu(data->rgc.rgcLdcp));
1304 		break;
1305 	case OPAL_P7IOC_DIAG_TYPE_BI:
1306 		pr_info("P7IOC diag-data for BI %s\n\n",
1307 			data->bi.biDownbound ? "Downbound" : "Upbound");
1308 		pnv_eeh_dump_hub_diag_common(data);
1309 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1310 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1311 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1312 				be64_to_cpu(data->bi.biLdcp0),
1313 				be64_to_cpu(data->bi.biLdcp1),
1314 				be64_to_cpu(data->bi.biLdcp2),
1315 				be64_to_cpu(data->bi.biFenceStatus));
1316 		break;
1317 	case OPAL_P7IOC_DIAG_TYPE_CI:
1318 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1319 			data->ci.ciPort);
1320 		pnv_eeh_dump_hub_diag_common(data);
1321 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1322 			pr_info("  CI:  %016llx %016llx\n",
1323 				be64_to_cpu(data->ci.ciPortStatus),
1324 				be64_to_cpu(data->ci.ciPortLdcp));
1325 		break;
1326 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1327 		pr_info("P7IOC diag-data for MISC\n\n");
1328 		pnv_eeh_dump_hub_diag_common(data);
1329 		break;
1330 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1331 		pr_info("P7IOC diag-data for I2C\n\n");
1332 		pnv_eeh_dump_hub_diag_common(data);
1333 		break;
1334 	default:
1335 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1336 			__func__, phb->hub_id, data->type);
1337 	}
1338 }
1339 
1340 static int pnv_eeh_get_pe(struct pci_controller *hose,
1341 			  u16 pe_no, struct eeh_pe **pe)
1342 {
1343 	struct pnv_phb *phb = hose->private_data;
1344 	struct pnv_ioda_pe *pnv_pe;
1345 	struct eeh_pe *dev_pe;
1346 
1347 	/*
1348 	 * If PHB supports compound PE, to fetch
1349 	 * the master PE because slave PE is invisible
1350 	 * to EEH core.
1351 	 */
1352 	pnv_pe = &phb->ioda.pe_array[pe_no];
1353 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1354 		pnv_pe = pnv_pe->master;
1355 		WARN_ON(!pnv_pe ||
1356 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1357 		pe_no = pnv_pe->pe_number;
1358 	}
1359 
1360 	/* Find the PE according to PE# */
1361 	dev_pe = eeh_pe_get(hose, pe_no);
1362 	if (!dev_pe)
1363 		return -EEXIST;
1364 
1365 	/* Freeze the (compound) PE */
1366 	*pe = dev_pe;
1367 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1368 		phb->freeze_pe(phb, pe_no);
1369 
1370 	/*
1371 	 * At this point, we're sure the (compound) PE should
1372 	 * have been frozen. However, we still need poke until
1373 	 * hitting the frozen PE on top level.
1374 	 */
1375 	dev_pe = dev_pe->parent;
1376 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1377 		int ret;
1378 		ret = eeh_ops->get_state(dev_pe, NULL);
1379 		if (ret <= 0 || eeh_state_active(ret)) {
1380 			dev_pe = dev_pe->parent;
1381 			continue;
1382 		}
1383 
1384 		/* Frozen parent PE */
1385 		*pe = dev_pe;
1386 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1387 			phb->freeze_pe(phb, dev_pe->addr);
1388 
1389 		/* Next one */
1390 		dev_pe = dev_pe->parent;
1391 	}
1392 
1393 	return 0;
1394 }
1395 
1396 /**
1397  * pnv_eeh_next_error - Retrieve next EEH error to handle
1398  * @pe: Affected PE
1399  *
1400  * The function is expected to be called by EEH core while it gets
1401  * special EEH event (without binding PE). The function calls to
1402  * OPAL APIs for next error to handle. The informational error is
1403  * handled internally by platform. However, the dead IOC, dead PHB,
1404  * fenced PHB and frozen PE should be handled by EEH core eventually.
1405  */
1406 static int pnv_eeh_next_error(struct eeh_pe **pe)
1407 {
1408 	struct pci_controller *hose;
1409 	struct pnv_phb *phb;
1410 	struct eeh_pe *phb_pe, *parent_pe;
1411 	__be64 frozen_pe_no;
1412 	__be16 err_type, severity;
1413 	long rc;
1414 	int state, ret = EEH_NEXT_ERR_NONE;
1415 
1416 	/*
1417 	 * While running here, it's safe to purge the event queue. The
1418 	 * event should still be masked.
1419 	 */
1420 	eeh_remove_event(NULL, false);
1421 
1422 	list_for_each_entry(hose, &hose_list, list_node) {
1423 		/*
1424 		 * If the subordinate PCI buses of the PHB has been
1425 		 * removed or is exactly under error recovery, we
1426 		 * needn't take care of it any more.
1427 		 */
1428 		phb = hose->private_data;
1429 		phb_pe = eeh_phb_pe_get(hose);
1430 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1431 			continue;
1432 
1433 		rc = opal_pci_next_error(phb->opal_id,
1434 					 &frozen_pe_no, &err_type, &severity);
1435 		if (rc != OPAL_SUCCESS) {
1436 			pr_devel("%s: Invalid return value on "
1437 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1438 				 __func__, hose->global_number, rc);
1439 			continue;
1440 		}
1441 
1442 		/* If the PHB doesn't have error, stop processing */
1443 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1444 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1445 			pr_devel("%s: No error found on PHB#%x\n",
1446 				 __func__, hose->global_number);
1447 			continue;
1448 		}
1449 
1450 		/*
1451 		 * Processing the error. We're expecting the error with
1452 		 * highest priority reported upon multiple errors on the
1453 		 * specific PHB.
1454 		 */
1455 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1456 			__func__, be16_to_cpu(err_type),
1457 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1458 			hose->global_number);
1459 		switch (be16_to_cpu(err_type)) {
1460 		case OPAL_EEH_IOC_ERROR:
1461 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1462 				pr_err("EEH: dead IOC detected\n");
1463 				ret = EEH_NEXT_ERR_DEAD_IOC;
1464 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1465 				pr_info("EEH: IOC informative error "
1466 					"detected\n");
1467 				pnv_eeh_get_and_dump_hub_diag(hose);
1468 				ret = EEH_NEXT_ERR_NONE;
1469 			}
1470 
1471 			break;
1472 		case OPAL_EEH_PHB_ERROR:
1473 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1474 				*pe = phb_pe;
1475 				pr_err("EEH: dead PHB#%x detected, "
1476 				       "location: %s\n",
1477 					hose->global_number,
1478 					eeh_pe_loc_get(phb_pe));
1479 				ret = EEH_NEXT_ERR_DEAD_PHB;
1480 			} else if (be16_to_cpu(severity) ==
1481 				   OPAL_EEH_SEV_PHB_FENCED) {
1482 				*pe = phb_pe;
1483 				pr_err("EEH: Fenced PHB#%x detected, "
1484 				       "location: %s\n",
1485 					hose->global_number,
1486 					eeh_pe_loc_get(phb_pe));
1487 				ret = EEH_NEXT_ERR_FENCED_PHB;
1488 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1489 				pr_info("EEH: PHB#%x informative error "
1490 					"detected, location: %s\n",
1491 					hose->global_number,
1492 					eeh_pe_loc_get(phb_pe));
1493 				pnv_eeh_get_phb_diag(phb_pe);
1494 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1495 				ret = EEH_NEXT_ERR_NONE;
1496 			}
1497 
1498 			break;
1499 		case OPAL_EEH_PE_ERROR:
1500 			/*
1501 			 * If we can't find the corresponding PE, we
1502 			 * just try to unfreeze.
1503 			 */
1504 			if (pnv_eeh_get_pe(hose,
1505 				be64_to_cpu(frozen_pe_no), pe)) {
1506 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1507 					hose->global_number, be64_to_cpu(frozen_pe_no));
1508 				pr_info("EEH: PHB location: %s\n",
1509 					eeh_pe_loc_get(phb_pe));
1510 
1511 				/* Dump PHB diag-data */
1512 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1513 					phb->diag_data, phb->diag_data_size);
1514 				if (rc == OPAL_SUCCESS)
1515 					pnv_pci_dump_phb_diag_data(hose,
1516 							phb->diag_data);
1517 
1518 				/* Try best to clear it */
1519 				opal_pci_eeh_freeze_clear(phb->opal_id,
1520 					be64_to_cpu(frozen_pe_no),
1521 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1522 				ret = EEH_NEXT_ERR_NONE;
1523 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1524 				   eeh_pe_passed(*pe)) {
1525 				ret = EEH_NEXT_ERR_NONE;
1526 			} else {
1527 				pr_err("EEH: Frozen PE#%x "
1528 				       "on PHB#%x detected\n",
1529 				       (*pe)->addr,
1530 					(*pe)->phb->global_number);
1531 				pr_err("EEH: PE location: %s, "
1532 				       "PHB location: %s\n",
1533 				       eeh_pe_loc_get(*pe),
1534 				       eeh_pe_loc_get(phb_pe));
1535 				ret = EEH_NEXT_ERR_FROZEN_PE;
1536 			}
1537 
1538 			break;
1539 		default:
1540 			pr_warn("%s: Unexpected error type %d\n",
1541 				__func__, be16_to_cpu(err_type));
1542 		}
1543 
1544 		/*
1545 		 * EEH core will try recover from fenced PHB or
1546 		 * frozen PE. In the time for frozen PE, EEH core
1547 		 * enable IO path for that before collecting logs,
1548 		 * but it ruins the site. So we have to dump the
1549 		 * log in advance here.
1550 		 */
1551 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1552 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1553 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1554 			eeh_pe_mark_isolated(*pe);
1555 			pnv_eeh_get_phb_diag(*pe);
1556 
1557 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1558 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1559 							   (*pe)->data);
1560 		}
1561 
1562 		/*
1563 		 * We probably have the frozen parent PE out there and
1564 		 * we need have to handle frozen parent PE firstly.
1565 		 */
1566 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1567 			parent_pe = (*pe)->parent;
1568 			while (parent_pe) {
1569 				/* Hit the ceiling ? */
1570 				if (parent_pe->type & EEH_PE_PHB)
1571 					break;
1572 
1573 				/* Frozen parent PE ? */
1574 				state = eeh_ops->get_state(parent_pe, NULL);
1575 				if (state > 0 && !eeh_state_active(state))
1576 					*pe = parent_pe;
1577 
1578 				/* Next parent level */
1579 				parent_pe = parent_pe->parent;
1580 			}
1581 
1582 			/* We possibly migrate to another PE */
1583 			eeh_pe_mark_isolated(*pe);
1584 		}
1585 
1586 		/*
1587 		 * If we have no errors on the specific PHB or only
1588 		 * informative error there, we continue poking it.
1589 		 * Otherwise, we need actions to be taken by upper
1590 		 * layer.
1591 		 */
1592 		if (ret > EEH_NEXT_ERR_INF)
1593 			break;
1594 	}
1595 
1596 	/* Unmask the event */
1597 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1598 		enable_irq(eeh_event_irq);
1599 
1600 	return ret;
1601 }
1602 
1603 static int pnv_eeh_restore_config(struct eeh_dev *edev)
1604 {
1605 	struct pnv_phb *phb;
1606 	s64 ret = 0;
1607 
1608 	if (!edev)
1609 		return -EEXIST;
1610 
1611 	if (edev->physfn)
1612 		return 0;
1613 
1614 	phb = edev->controller->private_data;
1615 	ret = opal_pci_reinit(phb->opal_id,
1616 			      OPAL_REINIT_PCI_DEV, edev->bdfn);
1617 
1618 	if (ret) {
1619 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1620 			__func__, edev->bdfn, ret);
1621 		return -EIO;
1622 	}
1623 
1624 	return ret;
1625 }
1626 
1627 static struct eeh_ops pnv_eeh_ops = {
1628 	.name                   = "powernv",
1629 	.probe			= pnv_eeh_probe,
1630 	.set_option             = pnv_eeh_set_option,
1631 	.get_state              = pnv_eeh_get_state,
1632 	.reset                  = pnv_eeh_reset,
1633 	.get_log                = pnv_eeh_get_log,
1634 	.configure_bridge       = pnv_eeh_configure_bridge,
1635 	.err_inject		= pnv_eeh_err_inject,
1636 	.read_config            = pnv_eeh_read_config,
1637 	.write_config           = pnv_eeh_write_config,
1638 	.next_error		= pnv_eeh_next_error,
1639 	.restore_config		= pnv_eeh_restore_config,
1640 	.notify_resume		= NULL
1641 };
1642 
1643 /**
1644  * eeh_powernv_init - Register platform dependent EEH operations
1645  *
1646  * EEH initialization on powernv platform. This function should be
1647  * called before any EEH related functions.
1648  */
1649 static int __init eeh_powernv_init(void)
1650 {
1651 	int max_diag_size = PNV_PCI_DIAG_BUF_SIZE;
1652 	struct pci_controller *hose;
1653 	struct pnv_phb *phb;
1654 	int ret = -EINVAL;
1655 
1656 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
1657 		pr_warn("%s: OPAL is required !\n", __func__);
1658 		return -EINVAL;
1659 	}
1660 
1661 	/* Set probe mode */
1662 	eeh_add_flag(EEH_PROBE_MODE_DEV);
1663 
1664 	/*
1665 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
1666 	 * doesn't do that. So we have to selectively enable I/O
1667 	 * prior to collecting error log.
1668 	 */
1669 	list_for_each_entry(hose, &hose_list, list_node) {
1670 		phb = hose->private_data;
1671 
1672 		if (phb->model == PNV_PHB_MODEL_P7IOC)
1673 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
1674 
1675 		if (phb->diag_data_size > max_diag_size)
1676 			max_diag_size = phb->diag_data_size;
1677 
1678 		break;
1679 	}
1680 
1681 	/*
1682 	 * eeh_init() allocates the eeh_pe and its aux data buf so the
1683 	 * size needs to be set before calling eeh_init().
1684 	 */
1685 	eeh_set_pe_aux_size(max_diag_size);
1686 	ppc_md.pcibios_bus_add_device = pnv_pcibios_bus_add_device;
1687 
1688 	ret = eeh_init(&pnv_eeh_ops);
1689 	if (!ret)
1690 		pr_info("EEH: PowerNV platform initialized\n");
1691 	else
1692 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
1693 
1694 	return ret;
1695 }
1696 machine_arch_initcall(powernv, eeh_powernv_init);
1697