1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * The file intends to implement the platform dependent EEH operations on
4  * powernv platform. Actually, the powernv was created in order to fully
5  * hypervisor support.
6  *
7  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013.
8  */
9 
10 #include <linux/atomic.h>
11 #include <linux/debugfs.h>
12 #include <linux/delay.h>
13 #include <linux/export.h>
14 #include <linux/init.h>
15 #include <linux/interrupt.h>
16 #include <linux/list.h>
17 #include <linux/msi.h>
18 #include <linux/of.h>
19 #include <linux/pci.h>
20 #include <linux/proc_fs.h>
21 #include <linux/rbtree.h>
22 #include <linux/sched.h>
23 #include <linux/seq_file.h>
24 #include <linux/spinlock.h>
25 
26 #include <asm/eeh.h>
27 #include <asm/eeh_event.h>
28 #include <asm/firmware.h>
29 #include <asm/io.h>
30 #include <asm/iommu.h>
31 #include <asm/machdep.h>
32 #include <asm/msi_bitmap.h>
33 #include <asm/opal.h>
34 #include <asm/ppc-pci.h>
35 #include <asm/pnv-pci.h>
36 
37 #include "powernv.h"
38 #include "pci.h"
39 
40 static int eeh_event_irq = -EINVAL;
41 
42 void pnv_pcibios_bus_add_device(struct pci_dev *pdev)
43 {
44 	struct pci_dn *pdn = pci_get_pdn(pdev);
45 
46 	if (!pdev->is_virtfn)
47 		return;
48 
49 	/*
50 	 * The following operations will fail if VF's sysfs files
51 	 * aren't created or its resources aren't finalized.
52 	 */
53 	eeh_add_device_early(pdn);
54 	eeh_add_device_late(pdev);
55 	eeh_sysfs_add_device(pdev);
56 }
57 
58 static int pnv_eeh_init(void)
59 {
60 	struct pci_controller *hose;
61 	struct pnv_phb *phb;
62 	int max_diag_size = PNV_PCI_DIAG_BUF_SIZE;
63 
64 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
65 		pr_warn("%s: OPAL is required !\n",
66 			__func__);
67 		return -EINVAL;
68 	}
69 
70 	/* Set probe mode */
71 	eeh_add_flag(EEH_PROBE_MODE_DEV);
72 
73 	/*
74 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
75 	 * doesn't do that. So we have to selectively enable I/O
76 	 * prior to collecting error log.
77 	 */
78 	list_for_each_entry(hose, &hose_list, list_node) {
79 		phb = hose->private_data;
80 
81 		if (phb->model == PNV_PHB_MODEL_P7IOC)
82 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
83 
84 		if (phb->diag_data_size > max_diag_size)
85 			max_diag_size = phb->diag_data_size;
86 
87 		/*
88 		 * PE#0 should be regarded as valid by EEH core
89 		 * if it's not the reserved one. Currently, we
90 		 * have the reserved PE#255 and PE#127 for PHB3
91 		 * and P7IOC separately. So we should regard
92 		 * PE#0 as valid for PHB3 and P7IOC.
93 		 */
94 		if (phb->ioda.reserved_pe_idx != 0)
95 			eeh_add_flag(EEH_VALID_PE_ZERO);
96 
97 		break;
98 	}
99 
100 	eeh_set_pe_aux_size(max_diag_size);
101 	ppc_md.pcibios_bus_add_device = pnv_pcibios_bus_add_device;
102 
103 	return 0;
104 }
105 
106 static irqreturn_t pnv_eeh_event(int irq, void *data)
107 {
108 	/*
109 	 * We simply send a special EEH event if EEH has been
110 	 * enabled. We don't care about EEH events until we've
111 	 * finished processing the outstanding ones. Event processing
112 	 * gets unmasked in next_error() if EEH is enabled.
113 	 */
114 	disable_irq_nosync(irq);
115 
116 	if (eeh_enabled())
117 		eeh_send_failure_event(NULL);
118 
119 	return IRQ_HANDLED;
120 }
121 
122 #ifdef CONFIG_DEBUG_FS
123 static ssize_t pnv_eeh_ei_write(struct file *filp,
124 				const char __user *user_buf,
125 				size_t count, loff_t *ppos)
126 {
127 	struct pci_controller *hose = filp->private_data;
128 	struct eeh_pe *pe;
129 	int pe_no, type, func;
130 	unsigned long addr, mask;
131 	char buf[50];
132 	int ret;
133 
134 	if (!eeh_ops || !eeh_ops->err_inject)
135 		return -ENXIO;
136 
137 	/* Copy over argument buffer */
138 	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
139 	if (!ret)
140 		return -EFAULT;
141 
142 	/* Retrieve parameters */
143 	ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
144 		     &pe_no, &type, &func, &addr, &mask);
145 	if (ret != 5)
146 		return -EINVAL;
147 
148 	/* Retrieve PE */
149 	pe = eeh_pe_get(hose, pe_no, 0);
150 	if (!pe)
151 		return -ENODEV;
152 
153 	/* Do error injection */
154 	ret = eeh_ops->err_inject(pe, type, func, addr, mask);
155 	return ret < 0 ? ret : count;
156 }
157 
158 static const struct file_operations pnv_eeh_ei_fops = {
159 	.open	= simple_open,
160 	.llseek	= no_llseek,
161 	.write	= pnv_eeh_ei_write,
162 };
163 
164 static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val)
165 {
166 	struct pci_controller *hose = data;
167 	struct pnv_phb *phb = hose->private_data;
168 
169 	out_be64(phb->regs + offset, val);
170 	return 0;
171 }
172 
173 static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val)
174 {
175 	struct pci_controller *hose = data;
176 	struct pnv_phb *phb = hose->private_data;
177 
178 	*val = in_be64(phb->regs + offset);
179 	return 0;
180 }
181 
182 #define PNV_EEH_DBGFS_ENTRY(name, reg)				\
183 static int pnv_eeh_dbgfs_set_##name(void *data, u64 val)	\
184 {								\
185 	return pnv_eeh_dbgfs_set(data, reg, val);		\
186 }								\
187 								\
188 static int pnv_eeh_dbgfs_get_##name(void *data, u64 *val)	\
189 {								\
190 	return pnv_eeh_dbgfs_get(data, reg, val);		\
191 }								\
192 								\
193 DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_dbgfs_ops_##name,		\
194 			pnv_eeh_dbgfs_get_##name,		\
195                         pnv_eeh_dbgfs_set_##name,		\
196 			"0x%llx\n")
197 
198 PNV_EEH_DBGFS_ENTRY(outb, 0xD10);
199 PNV_EEH_DBGFS_ENTRY(inbA, 0xD90);
200 PNV_EEH_DBGFS_ENTRY(inbB, 0xE10);
201 
202 #endif /* CONFIG_DEBUG_FS */
203 
204 /**
205  * pnv_eeh_post_init - EEH platform dependent post initialization
206  *
207  * EEH platform dependent post initialization on powernv. When
208  * the function is called, the EEH PEs and devices should have
209  * been built. If the I/O cache staff has been built, EEH is
210  * ready to supply service.
211  */
212 int pnv_eeh_post_init(void)
213 {
214 	struct pci_controller *hose;
215 	struct pnv_phb *phb;
216 	int ret = 0;
217 
218 	/* Probe devices & build address cache */
219 	eeh_probe_devices();
220 	eeh_addr_cache_build();
221 
222 	/* Register OPAL event notifier */
223 	eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
224 	if (eeh_event_irq < 0) {
225 		pr_err("%s: Can't register OPAL event interrupt (%d)\n",
226 		       __func__, eeh_event_irq);
227 		return eeh_event_irq;
228 	}
229 
230 	ret = request_irq(eeh_event_irq, pnv_eeh_event,
231 			  IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
232 	if (ret < 0) {
233 		irq_dispose_mapping(eeh_event_irq);
234 		pr_err("%s: Can't request OPAL event interrupt (%d)\n",
235 		       __func__, eeh_event_irq);
236 		return ret;
237 	}
238 
239 	if (!eeh_enabled())
240 		disable_irq(eeh_event_irq);
241 
242 	list_for_each_entry(hose, &hose_list, list_node) {
243 		phb = hose->private_data;
244 
245 		/*
246 		 * If EEH is enabled, we're going to rely on that.
247 		 * Otherwise, we restore to conventional mechanism
248 		 * to clear frozen PE during PCI config access.
249 		 */
250 		if (eeh_enabled())
251 			phb->flags |= PNV_PHB_FLAG_EEH;
252 		else
253 			phb->flags &= ~PNV_PHB_FLAG_EEH;
254 
255 		/* Create debugfs entries */
256 #ifdef CONFIG_DEBUG_FS
257 		if (phb->has_dbgfs || !phb->dbgfs)
258 			continue;
259 
260 		phb->has_dbgfs = 1;
261 		debugfs_create_file("err_injct", 0200,
262 				    phb->dbgfs, hose,
263 				    &pnv_eeh_ei_fops);
264 
265 		debugfs_create_file("err_injct_outbound", 0600,
266 				    phb->dbgfs, hose,
267 				    &pnv_eeh_dbgfs_ops_outb);
268 		debugfs_create_file("err_injct_inboundA", 0600,
269 				    phb->dbgfs, hose,
270 				    &pnv_eeh_dbgfs_ops_inbA);
271 		debugfs_create_file("err_injct_inboundB", 0600,
272 				    phb->dbgfs, hose,
273 				    &pnv_eeh_dbgfs_ops_inbB);
274 #endif /* CONFIG_DEBUG_FS */
275 	}
276 
277 	return ret;
278 }
279 
280 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
281 {
282 	int pos = PCI_CAPABILITY_LIST;
283 	int cnt = 48;   /* Maximal number of capabilities */
284 	u32 status, id;
285 
286 	if (!pdn)
287 		return 0;
288 
289 	/* Check if the device supports capabilities */
290 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
291 	if (!(status & PCI_STATUS_CAP_LIST))
292 		return 0;
293 
294 	while (cnt--) {
295 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
296 		if (pos < 0x40)
297 			break;
298 
299 		pos &= ~3;
300 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
301 		if (id == 0xff)
302 			break;
303 
304 		/* Found */
305 		if (id == cap)
306 			return pos;
307 
308 		/* Next one */
309 		pos += PCI_CAP_LIST_NEXT;
310 	}
311 
312 	return 0;
313 }
314 
315 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
316 {
317 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
318 	u32 header;
319 	int pos = 256, ttl = (4096 - 256) / 8;
320 
321 	if (!edev || !edev->pcie_cap)
322 		return 0;
323 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
324 		return 0;
325 	else if (!header)
326 		return 0;
327 
328 	while (ttl-- > 0) {
329 		if (PCI_EXT_CAP_ID(header) == cap && pos)
330 			return pos;
331 
332 		pos = PCI_EXT_CAP_NEXT(header);
333 		if (pos < 256)
334 			break;
335 
336 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
337 			break;
338 	}
339 
340 	return 0;
341 }
342 
343 /**
344  * pnv_eeh_probe - Do probe on PCI device
345  * @pdn: PCI device node
346  * @data: unused
347  *
348  * When EEH module is installed during system boot, all PCI devices
349  * are checked one by one to see if it supports EEH. The function
350  * is introduced for the purpose. By default, EEH has been enabled
351  * on all PCI devices. That's to say, we only need do necessary
352  * initialization on the corresponding eeh device and create PE
353  * accordingly.
354  *
355  * It's notable that's unsafe to retrieve the EEH device through
356  * the corresponding PCI device. During the PCI device hotplug, which
357  * was possiblly triggered by EEH core, the binding between EEH device
358  * and the PCI device isn't built yet.
359  */
360 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
361 {
362 	struct pci_controller *hose = pdn->phb;
363 	struct pnv_phb *phb = hose->private_data;
364 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
365 	uint32_t pcie_flags;
366 	int ret;
367 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
368 
369 	/*
370 	 * When probing the root bridge, which doesn't have any
371 	 * subordinate PCI devices. We don't have OF node for
372 	 * the root bridge. So it's not reasonable to continue
373 	 * the probing.
374 	 */
375 	if (!edev || edev->pe)
376 		return NULL;
377 
378 	/* Skip for PCI-ISA bridge */
379 	if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
380 		return NULL;
381 
382 	/* Initialize eeh device */
383 	edev->class_code = pdn->class_code;
384 	edev->mode	&= 0xFFFFFF00;
385 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
386 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
387 	edev->af_cap   = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF);
388 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
389 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
390 		edev->mode |= EEH_DEV_BRIDGE;
391 		if (edev->pcie_cap) {
392 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
393 					 2, &pcie_flags);
394 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
395 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
396 				edev->mode |= EEH_DEV_ROOT_PORT;
397 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
398 				edev->mode |= EEH_DEV_DS_PORT;
399 		}
400 	}
401 
402 	edev->pe_config_addr = phb->ioda.pe_rmap[config_addr];
403 
404 	/* Create PE */
405 	ret = eeh_add_to_parent_pe(edev);
406 	if (ret) {
407 		pr_warn("%s: Can't add PCI dev %04x:%02x:%02x.%01x to parent PE (%x)\n",
408 			__func__, hose->global_number, pdn->busno,
409 			PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn), ret);
410 		return NULL;
411 	}
412 
413 	/*
414 	 * If the PE contains any one of following adapters, the
415 	 * PCI config space can't be accessed when dumping EEH log.
416 	 * Otherwise, we will run into fenced PHB caused by shortage
417 	 * of outbound credits in the adapter. The PCI config access
418 	 * should be blocked until PE reset. MMIO access is dropped
419 	 * by hardware certainly. In order to drop PCI config requests,
420 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
421 	 * will be checked in the backend for PE state retrival. If
422 	 * the PE becomes frozen for the first time and the flag has
423 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
424 	 * that PE to block its config space.
425 	 *
426 	 * Broadcom BCM5718 2-ports NICs (14e4:1656)
427 	 * Broadcom Austin 4-ports NICs (14e4:1657)
428 	 * Broadcom Shiner 4-ports 1G NICs (14e4:168a)
429 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
430 	 */
431 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
432 	     pdn->device_id == 0x1656) ||
433 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
434 	     pdn->device_id == 0x1657) ||
435 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
436 	     pdn->device_id == 0x168a) ||
437 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
438 	     pdn->device_id == 0x168e))
439 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
440 
441 	/*
442 	 * Cache the PE primary bus, which can't be fetched when
443 	 * full hotplug is in progress. In that case, all child
444 	 * PCI devices of the PE are expected to be removed prior
445 	 * to PE reset.
446 	 */
447 	if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
448 		edev->pe->bus = pci_find_bus(hose->global_number,
449 					     pdn->busno);
450 		if (edev->pe->bus)
451 			edev->pe->state |= EEH_PE_PRI_BUS;
452 	}
453 
454 	/*
455 	 * Enable EEH explicitly so that we will do EEH check
456 	 * while accessing I/O stuff
457 	 */
458 	eeh_add_flag(EEH_ENABLED);
459 
460 	/* Save memory bars */
461 	eeh_save_bars(edev);
462 
463 	return NULL;
464 }
465 
466 /**
467  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
468  * @pe: EEH PE
469  * @option: operation to be issued
470  *
471  * The function is used to control the EEH functionality globally.
472  * Currently, following options are support according to PAPR:
473  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
474  */
475 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
476 {
477 	struct pci_controller *hose = pe->phb;
478 	struct pnv_phb *phb = hose->private_data;
479 	bool freeze_pe = false;
480 	int opt;
481 	s64 rc;
482 
483 	switch (option) {
484 	case EEH_OPT_DISABLE:
485 		return -EPERM;
486 	case EEH_OPT_ENABLE:
487 		return 0;
488 	case EEH_OPT_THAW_MMIO:
489 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
490 		break;
491 	case EEH_OPT_THAW_DMA:
492 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
493 		break;
494 	case EEH_OPT_FREEZE_PE:
495 		freeze_pe = true;
496 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
497 		break;
498 	default:
499 		pr_warn("%s: Invalid option %d\n", __func__, option);
500 		return -EINVAL;
501 	}
502 
503 	/* Freeze master and slave PEs if PHB supports compound PEs */
504 	if (freeze_pe) {
505 		if (phb->freeze_pe) {
506 			phb->freeze_pe(phb, pe->addr);
507 			return 0;
508 		}
509 
510 		rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt);
511 		if (rc != OPAL_SUCCESS) {
512 			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
513 				__func__, rc, phb->hose->global_number,
514 				pe->addr);
515 			return -EIO;
516 		}
517 
518 		return 0;
519 	}
520 
521 	/* Unfreeze master and slave PEs if PHB supports */
522 	if (phb->unfreeze_pe)
523 		return phb->unfreeze_pe(phb, pe->addr, opt);
524 
525 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt);
526 	if (rc != OPAL_SUCCESS) {
527 		pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n",
528 			__func__, rc, option, phb->hose->global_number,
529 			pe->addr);
530 		return -EIO;
531 	}
532 
533 	return 0;
534 }
535 
536 /**
537  * pnv_eeh_get_pe_addr - Retrieve PE address
538  * @pe: EEH PE
539  *
540  * Retrieve the PE address according to the given tranditional
541  * PCI BDF (Bus/Device/Function) address.
542  */
543 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe)
544 {
545 	return pe->addr;
546 }
547 
548 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
549 {
550 	struct pnv_phb *phb = pe->phb->private_data;
551 	s64 rc;
552 
553 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
554 					 phb->diag_data_size);
555 	if (rc != OPAL_SUCCESS)
556 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
557 			__func__, rc, pe->phb->global_number);
558 }
559 
560 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
561 {
562 	struct pnv_phb *phb = pe->phb->private_data;
563 	u8 fstate = 0;
564 	__be16 pcierr = 0;
565 	s64 rc;
566 	int result = 0;
567 
568 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
569 					pe->addr,
570 					&fstate,
571 					&pcierr,
572 					NULL);
573 	if (rc != OPAL_SUCCESS) {
574 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
575 			__func__, rc, phb->hose->global_number);
576 		return EEH_STATE_NOT_SUPPORT;
577 	}
578 
579 	/*
580 	 * Check PHB state. If the PHB is frozen for the
581 	 * first time, to dump the PHB diag-data.
582 	 */
583 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
584 		result = (EEH_STATE_MMIO_ACTIVE  |
585 			  EEH_STATE_DMA_ACTIVE   |
586 			  EEH_STATE_MMIO_ENABLED |
587 			  EEH_STATE_DMA_ENABLED);
588 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
589 		eeh_pe_mark_isolated(pe);
590 		pnv_eeh_get_phb_diag(pe);
591 
592 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
593 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
594 	}
595 
596 	return result;
597 }
598 
599 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
600 {
601 	struct pnv_phb *phb = pe->phb->private_data;
602 	u8 fstate = 0;
603 	__be16 pcierr = 0;
604 	s64 rc;
605 	int result;
606 
607 	/*
608 	 * We don't clobber hardware frozen state until PE
609 	 * reset is completed. In order to keep EEH core
610 	 * moving forward, we have to return operational
611 	 * state during PE reset.
612 	 */
613 	if (pe->state & EEH_PE_RESET) {
614 		result = (EEH_STATE_MMIO_ACTIVE  |
615 			  EEH_STATE_DMA_ACTIVE   |
616 			  EEH_STATE_MMIO_ENABLED |
617 			  EEH_STATE_DMA_ENABLED);
618 		return result;
619 	}
620 
621 	/*
622 	 * Fetch PE state from hardware. If the PHB
623 	 * supports compound PE, let it handle that.
624 	 */
625 	if (phb->get_pe_state) {
626 		fstate = phb->get_pe_state(phb, pe->addr);
627 	} else {
628 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
629 						pe->addr,
630 						&fstate,
631 						&pcierr,
632 						NULL);
633 		if (rc != OPAL_SUCCESS) {
634 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
635 				__func__, rc, phb->hose->global_number,
636 				pe->addr);
637 			return EEH_STATE_NOT_SUPPORT;
638 		}
639 	}
640 
641 	/* Figure out state */
642 	switch (fstate) {
643 	case OPAL_EEH_STOPPED_NOT_FROZEN:
644 		result = (EEH_STATE_MMIO_ACTIVE  |
645 			  EEH_STATE_DMA_ACTIVE   |
646 			  EEH_STATE_MMIO_ENABLED |
647 			  EEH_STATE_DMA_ENABLED);
648 		break;
649 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
650 		result = (EEH_STATE_DMA_ACTIVE |
651 			  EEH_STATE_DMA_ENABLED);
652 		break;
653 	case OPAL_EEH_STOPPED_DMA_FREEZE:
654 		result = (EEH_STATE_MMIO_ACTIVE |
655 			  EEH_STATE_MMIO_ENABLED);
656 		break;
657 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
658 		result = 0;
659 		break;
660 	case OPAL_EEH_STOPPED_RESET:
661 		result = EEH_STATE_RESET_ACTIVE;
662 		break;
663 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
664 		result = EEH_STATE_UNAVAILABLE;
665 		break;
666 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
667 		result = EEH_STATE_NOT_SUPPORT;
668 		break;
669 	default:
670 		result = EEH_STATE_NOT_SUPPORT;
671 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
672 			__func__, phb->hose->global_number,
673 			pe->addr, fstate);
674 	}
675 
676 	/*
677 	 * If PHB supports compound PE, to freeze all
678 	 * slave PEs for consistency.
679 	 *
680 	 * If the PE is switching to frozen state for the
681 	 * first time, to dump the PHB diag-data.
682 	 */
683 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
684 	    !(result & EEH_STATE_UNAVAILABLE) &&
685 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
686 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
687 	    !(pe->state & EEH_PE_ISOLATED)) {
688 		if (phb->freeze_pe)
689 			phb->freeze_pe(phb, pe->addr);
690 
691 		eeh_pe_mark_isolated(pe);
692 		pnv_eeh_get_phb_diag(pe);
693 
694 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
695 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
696 	}
697 
698 	return result;
699 }
700 
701 /**
702  * pnv_eeh_get_state - Retrieve PE state
703  * @pe: EEH PE
704  * @delay: delay while PE state is temporarily unavailable
705  *
706  * Retrieve the state of the specified PE. For IODA-compitable
707  * platform, it should be retrieved from IODA table. Therefore,
708  * we prefer passing down to hardware implementation to handle
709  * it.
710  */
711 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
712 {
713 	int ret;
714 
715 	if (pe->type & EEH_PE_PHB)
716 		ret = pnv_eeh_get_phb_state(pe);
717 	else
718 		ret = pnv_eeh_get_pe_state(pe);
719 
720 	if (!delay)
721 		return ret;
722 
723 	/*
724 	 * If the PE state is temporarily unavailable,
725 	 * to inform the EEH core delay for default
726 	 * period (1 second)
727 	 */
728 	*delay = 0;
729 	if (ret & EEH_STATE_UNAVAILABLE)
730 		*delay = 1000;
731 
732 	return ret;
733 }
734 
735 static s64 pnv_eeh_poll(unsigned long id)
736 {
737 	s64 rc = OPAL_HARDWARE;
738 
739 	while (1) {
740 		rc = opal_pci_poll(id);
741 		if (rc <= 0)
742 			break;
743 
744 		if (system_state < SYSTEM_RUNNING)
745 			udelay(1000 * rc);
746 		else
747 			msleep(rc);
748 	}
749 
750 	return rc;
751 }
752 
753 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
754 {
755 	struct pnv_phb *phb = hose->private_data;
756 	s64 rc = OPAL_HARDWARE;
757 
758 	pr_debug("%s: Reset PHB#%x, option=%d\n",
759 		 __func__, hose->global_number, option);
760 
761 	/* Issue PHB complete reset request */
762 	if (option == EEH_RESET_FUNDAMENTAL ||
763 	    option == EEH_RESET_HOT)
764 		rc = opal_pci_reset(phb->opal_id,
765 				    OPAL_RESET_PHB_COMPLETE,
766 				    OPAL_ASSERT_RESET);
767 	else if (option == EEH_RESET_DEACTIVATE)
768 		rc = opal_pci_reset(phb->opal_id,
769 				    OPAL_RESET_PHB_COMPLETE,
770 				    OPAL_DEASSERT_RESET);
771 	if (rc < 0)
772 		goto out;
773 
774 	/*
775 	 * Poll state of the PHB until the request is done
776 	 * successfully. The PHB reset is usually PHB complete
777 	 * reset followed by hot reset on root bus. So we also
778 	 * need the PCI bus settlement delay.
779 	 */
780 	if (rc > 0)
781 		rc = pnv_eeh_poll(phb->opal_id);
782 	if (option == EEH_RESET_DEACTIVATE) {
783 		if (system_state < SYSTEM_RUNNING)
784 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
785 		else
786 			msleep(EEH_PE_RST_SETTLE_TIME);
787 	}
788 out:
789 	if (rc != OPAL_SUCCESS)
790 		return -EIO;
791 
792 	return 0;
793 }
794 
795 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
796 {
797 	struct pnv_phb *phb = hose->private_data;
798 	s64 rc = OPAL_HARDWARE;
799 
800 	pr_debug("%s: Reset PHB#%x, option=%d\n",
801 		 __func__, hose->global_number, option);
802 
803 	/*
804 	 * During the reset deassert time, we needn't care
805 	 * the reset scope because the firmware does nothing
806 	 * for fundamental or hot reset during deassert phase.
807 	 */
808 	if (option == EEH_RESET_FUNDAMENTAL)
809 		rc = opal_pci_reset(phb->opal_id,
810 				    OPAL_RESET_PCI_FUNDAMENTAL,
811 				    OPAL_ASSERT_RESET);
812 	else if (option == EEH_RESET_HOT)
813 		rc = opal_pci_reset(phb->opal_id,
814 				    OPAL_RESET_PCI_HOT,
815 				    OPAL_ASSERT_RESET);
816 	else if (option == EEH_RESET_DEACTIVATE)
817 		rc = opal_pci_reset(phb->opal_id,
818 				    OPAL_RESET_PCI_HOT,
819 				    OPAL_DEASSERT_RESET);
820 	if (rc < 0)
821 		goto out;
822 
823 	/* Poll state of the PHB until the request is done */
824 	if (rc > 0)
825 		rc = pnv_eeh_poll(phb->opal_id);
826 	if (option == EEH_RESET_DEACTIVATE)
827 		msleep(EEH_PE_RST_SETTLE_TIME);
828 out:
829 	if (rc != OPAL_SUCCESS)
830 		return -EIO;
831 
832 	return 0;
833 }
834 
835 static int __pnv_eeh_bridge_reset(struct pci_dev *dev, int option)
836 {
837 	struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
838 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
839 	int aer = edev ? edev->aer_cap : 0;
840 	u32 ctrl;
841 
842 	pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
843 		 __func__, pci_domain_nr(dev->bus),
844 		 dev->bus->number, option);
845 
846 	switch (option) {
847 	case EEH_RESET_FUNDAMENTAL:
848 	case EEH_RESET_HOT:
849 		/* Don't report linkDown event */
850 		if (aer) {
851 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
852 					     4, &ctrl);
853 			ctrl |= PCI_ERR_UNC_SURPDN;
854 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
855 					      4, ctrl);
856 		}
857 
858 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
859 		ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
860 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
861 
862 		msleep(EEH_PE_RST_HOLD_TIME);
863 		break;
864 	case EEH_RESET_DEACTIVATE:
865 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
866 		ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
867 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
868 
869 		msleep(EEH_PE_RST_SETTLE_TIME);
870 
871 		/* Continue reporting linkDown event */
872 		if (aer) {
873 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
874 					     4, &ctrl);
875 			ctrl &= ~PCI_ERR_UNC_SURPDN;
876 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
877 					      4, ctrl);
878 		}
879 
880 		break;
881 	}
882 
883 	return 0;
884 }
885 
886 static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option)
887 {
888 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
889 	struct pnv_phb *phb = hose->private_data;
890 	struct device_node *dn = pci_device_to_OF_node(pdev);
891 	uint64_t id = PCI_SLOT_ID(phb->opal_id,
892 				  (pdev->bus->number << 8) | pdev->devfn);
893 	uint8_t scope;
894 	int64_t rc;
895 
896 	/* Hot reset to the bus if firmware cannot handle */
897 	if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL))
898 		return __pnv_eeh_bridge_reset(pdev, option);
899 
900 	switch (option) {
901 	case EEH_RESET_FUNDAMENTAL:
902 		scope = OPAL_RESET_PCI_FUNDAMENTAL;
903 		break;
904 	case EEH_RESET_HOT:
905 		scope = OPAL_RESET_PCI_HOT;
906 		break;
907 	case EEH_RESET_DEACTIVATE:
908 		return 0;
909 	default:
910 		dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n",
911 			__func__, option);
912 		return -EINVAL;
913 	}
914 
915 	rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET);
916 	if (rc <= OPAL_SUCCESS)
917 		goto out;
918 
919 	rc = pnv_eeh_poll(id);
920 out:
921 	return (rc == OPAL_SUCCESS) ? 0 : -EIO;
922 }
923 
924 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
925 {
926 	struct pci_controller *hose;
927 
928 	if (pci_is_root_bus(dev->bus)) {
929 		hose = pci_bus_to_host(dev->bus);
930 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
931 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
932 	} else {
933 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
934 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
935 	}
936 }
937 
938 static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type,
939 				     int pos, u16 mask)
940 {
941 	int i, status = 0;
942 
943 	/* Wait for Transaction Pending bit to be cleared */
944 	for (i = 0; i < 4; i++) {
945 		eeh_ops->read_config(pdn, pos, 2, &status);
946 		if (!(status & mask))
947 			return;
948 
949 		msleep((1 << i) * 100);
950 	}
951 
952 	pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n",
953 		__func__, type,
954 		pdn->phb->global_number, pdn->busno,
955 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
956 }
957 
958 static int pnv_eeh_do_flr(struct pci_dn *pdn, int option)
959 {
960 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
961 	u32 reg = 0;
962 
963 	if (WARN_ON(!edev->pcie_cap))
964 		return -ENOTTY;
965 
966 	eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP, 4, &reg);
967 	if (!(reg & PCI_EXP_DEVCAP_FLR))
968 		return -ENOTTY;
969 
970 	switch (option) {
971 	case EEH_RESET_HOT:
972 	case EEH_RESET_FUNDAMENTAL:
973 		pnv_eeh_wait_for_pending(pdn, "",
974 					 edev->pcie_cap + PCI_EXP_DEVSTA,
975 					 PCI_EXP_DEVSTA_TRPND);
976 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
977 				     4, &reg);
978 		reg |= PCI_EXP_DEVCTL_BCR_FLR;
979 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
980 				      4, reg);
981 		msleep(EEH_PE_RST_HOLD_TIME);
982 		break;
983 	case EEH_RESET_DEACTIVATE:
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_SETTLE_TIME);
990 		break;
991 	}
992 
993 	return 0;
994 }
995 
996 static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option)
997 {
998 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
999 	u32 cap = 0;
1000 
1001 	if (WARN_ON(!edev->af_cap))
1002 		return -ENOTTY;
1003 
1004 	eeh_ops->read_config(pdn, edev->af_cap + PCI_AF_CAP, 1, &cap);
1005 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
1006 		return -ENOTTY;
1007 
1008 	switch (option) {
1009 	case EEH_RESET_HOT:
1010 	case EEH_RESET_FUNDAMENTAL:
1011 		/*
1012 		 * Wait for Transaction Pending bit to clear. A word-aligned
1013 		 * test is used, so we use the conrol offset rather than status
1014 		 * and shift the test bit to match.
1015 		 */
1016 		pnv_eeh_wait_for_pending(pdn, "AF",
1017 					 edev->af_cap + PCI_AF_CTRL,
1018 					 PCI_AF_STATUS_TP << 8);
1019 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL,
1020 				      1, PCI_AF_CTRL_FLR);
1021 		msleep(EEH_PE_RST_HOLD_TIME);
1022 		break;
1023 	case EEH_RESET_DEACTIVATE:
1024 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL, 1, 0);
1025 		msleep(EEH_PE_RST_SETTLE_TIME);
1026 		break;
1027 	}
1028 
1029 	return 0;
1030 }
1031 
1032 static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option)
1033 {
1034 	struct eeh_dev *edev;
1035 	struct pci_dn *pdn;
1036 	int ret;
1037 
1038 	/* The VF PE should have only one child device */
1039 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, entry);
1040 	pdn = eeh_dev_to_pdn(edev);
1041 	if (!pdn)
1042 		return -ENXIO;
1043 
1044 	ret = pnv_eeh_do_flr(pdn, option);
1045 	if (!ret)
1046 		return ret;
1047 
1048 	return pnv_eeh_do_af_flr(pdn, option);
1049 }
1050 
1051 /**
1052  * pnv_eeh_reset - Reset the specified PE
1053  * @pe: EEH PE
1054  * @option: reset option
1055  *
1056  * Do reset on the indicated PE. For PCI bus sensitive PE,
1057  * we need to reset the parent p2p bridge. The PHB has to
1058  * be reinitialized if the p2p bridge is root bridge. For
1059  * PCI device sensitive PE, we will try to reset the device
1060  * through FLR. For now, we don't have OPAL APIs to do HARD
1061  * reset yet, so all reset would be SOFT (HOT) reset.
1062  */
1063 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
1064 {
1065 	struct pci_controller *hose = pe->phb;
1066 	struct pnv_phb *phb;
1067 	struct pci_bus *bus;
1068 	int64_t rc;
1069 
1070 	/*
1071 	 * For PHB reset, we always have complete reset. For those PEs whose
1072 	 * primary bus derived from root complex (root bus) or root port
1073 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
1074 	 * For other PEs, we always have hot reset on the PE primary bus.
1075 	 *
1076 	 * Here, we have different design to pHyp, which always clear the
1077 	 * frozen state during PE reset. However, the good idea here from
1078 	 * benh is to keep frozen state before we get PE reset done completely
1079 	 * (until BAR restore). With the frozen state, HW drops illegal IO
1080 	 * or MMIO access, which can incur recrusive frozen PE during PE
1081 	 * reset. The side effect is that EEH core has to clear the frozen
1082 	 * state explicitly after BAR restore.
1083 	 */
1084 	if (pe->type & EEH_PE_PHB)
1085 		return pnv_eeh_phb_reset(hose, option);
1086 
1087 	/*
1088 	 * The frozen PE might be caused by PAPR error injection
1089 	 * registers, which are expected to be cleared after hitting
1090 	 * frozen PE as stated in the hardware spec. Unfortunately,
1091 	 * that's not true on P7IOC. So we have to clear it manually
1092 	 * to avoid recursive EEH errors during recovery.
1093 	 */
1094 	phb = hose->private_data;
1095 	if (phb->model == PNV_PHB_MODEL_P7IOC &&
1096 	    (option == EEH_RESET_HOT ||
1097 	     option == EEH_RESET_FUNDAMENTAL)) {
1098 		rc = opal_pci_reset(phb->opal_id,
1099 				    OPAL_RESET_PHB_ERROR,
1100 				    OPAL_ASSERT_RESET);
1101 		if (rc != OPAL_SUCCESS) {
1102 			pr_warn("%s: Failure %lld clearing error injection registers\n",
1103 				__func__, rc);
1104 			return -EIO;
1105 		}
1106 	}
1107 
1108 	if (pe->type & EEH_PE_VF)
1109 		return pnv_eeh_reset_vf_pe(pe, option);
1110 
1111 	bus = eeh_pe_bus_get(pe);
1112 	if (!bus) {
1113 		pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
1114 			__func__, pe->phb->global_number, pe->addr);
1115 		return -EIO;
1116 	}
1117 
1118 	/*
1119 	 * If dealing with the root bus (or the bus underneath the
1120 	 * root port), we reset the bus underneath the root port.
1121 	 *
1122 	 * The cxl driver depends on this behaviour for bi-modal card
1123 	 * switching.
1124 	 */
1125 	if (pci_is_root_bus(bus) ||
1126 	    pci_is_root_bus(bus->parent))
1127 		return pnv_eeh_root_reset(hose, option);
1128 
1129 	return pnv_eeh_bridge_reset(bus->self, option);
1130 }
1131 
1132 /**
1133  * pnv_eeh_get_log - Retrieve error log
1134  * @pe: EEH PE
1135  * @severity: temporary or permanent error log
1136  * @drv_log: driver log to be combined with retrieved error log
1137  * @len: length of driver log
1138  *
1139  * Retrieve the temporary or permanent error from the PE.
1140  */
1141 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1142 			   char *drv_log, unsigned long len)
1143 {
1144 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1145 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1146 
1147 	return 0;
1148 }
1149 
1150 /**
1151  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1152  * @pe: EEH PE
1153  *
1154  * The function will be called to reconfigure the bridges included
1155  * in the specified PE so that the mulfunctional PE would be recovered
1156  * again.
1157  */
1158 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1159 {
1160 	return 0;
1161 }
1162 
1163 /**
1164  * pnv_pe_err_inject - Inject specified error to the indicated PE
1165  * @pe: the indicated PE
1166  * @type: error type
1167  * @func: specific error type
1168  * @addr: address
1169  * @mask: address mask
1170  *
1171  * The routine is called to inject specified error, which is
1172  * determined by @type and @func, to the indicated PE for
1173  * testing purpose.
1174  */
1175 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1176 			      unsigned long addr, unsigned long mask)
1177 {
1178 	struct pci_controller *hose = pe->phb;
1179 	struct pnv_phb *phb = hose->private_data;
1180 	s64 rc;
1181 
1182 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1183 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1184 		pr_warn("%s: Invalid error type %d\n",
1185 			__func__, type);
1186 		return -ERANGE;
1187 	}
1188 
1189 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1190 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1191 		pr_warn("%s: Invalid error function %d\n",
1192 			__func__, func);
1193 		return -ERANGE;
1194 	}
1195 
1196 	/* Firmware supports error injection ? */
1197 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1198 		pr_warn("%s: Firmware doesn't support error injection\n",
1199 			__func__);
1200 		return -ENXIO;
1201 	}
1202 
1203 	/* Do error injection */
1204 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1205 				 type, func, addr, mask);
1206 	if (rc != OPAL_SUCCESS) {
1207 		pr_warn("%s: Failure %lld injecting error "
1208 			"%d-%d to PHB#%x-PE#%x\n",
1209 			__func__, rc, type, func,
1210 			hose->global_number, pe->addr);
1211 		return -EIO;
1212 	}
1213 
1214 	return 0;
1215 }
1216 
1217 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1218 {
1219 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1220 
1221 	if (!edev || !edev->pe)
1222 		return false;
1223 
1224 	/*
1225 	 * We will issue FLR or AF FLR to all VFs, which are contained
1226 	 * in VF PE. It relies on the EEH PCI config accessors. So we
1227 	 * can't block them during the window.
1228 	 */
1229 	if (edev->physfn && (edev->pe->state & EEH_PE_RESET))
1230 		return false;
1231 
1232 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1233 		return true;
1234 
1235 	return false;
1236 }
1237 
1238 static int pnv_eeh_read_config(struct pci_dn *pdn,
1239 			       int where, int size, u32 *val)
1240 {
1241 	if (!pdn)
1242 		return PCIBIOS_DEVICE_NOT_FOUND;
1243 
1244 	if (pnv_eeh_cfg_blocked(pdn)) {
1245 		*val = 0xFFFFFFFF;
1246 		return PCIBIOS_SET_FAILED;
1247 	}
1248 
1249 	return pnv_pci_cfg_read(pdn, where, size, val);
1250 }
1251 
1252 static int pnv_eeh_write_config(struct pci_dn *pdn,
1253 				int where, int size, u32 val)
1254 {
1255 	if (!pdn)
1256 		return PCIBIOS_DEVICE_NOT_FOUND;
1257 
1258 	if (pnv_eeh_cfg_blocked(pdn))
1259 		return PCIBIOS_SET_FAILED;
1260 
1261 	return pnv_pci_cfg_write(pdn, where, size, val);
1262 }
1263 
1264 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1265 {
1266 	/* GEM */
1267 	if (data->gemXfir || data->gemRfir ||
1268 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1269 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1270 			be64_to_cpu(data->gemXfir),
1271 			be64_to_cpu(data->gemRfir),
1272 			be64_to_cpu(data->gemRirqfir),
1273 			be64_to_cpu(data->gemMask),
1274 			be64_to_cpu(data->gemRwof));
1275 
1276 	/* LEM */
1277 	if (data->lemFir || data->lemErrMask ||
1278 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1279 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1280 			be64_to_cpu(data->lemFir),
1281 			be64_to_cpu(data->lemErrMask),
1282 			be64_to_cpu(data->lemAction0),
1283 			be64_to_cpu(data->lemAction1),
1284 			be64_to_cpu(data->lemWof));
1285 }
1286 
1287 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1288 {
1289 	struct pnv_phb *phb = hose->private_data;
1290 	struct OpalIoP7IOCErrorData *data =
1291 		(struct OpalIoP7IOCErrorData*)phb->diag_data;
1292 	long rc;
1293 
1294 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1295 	if (rc != OPAL_SUCCESS) {
1296 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1297 			__func__, phb->hub_id, rc);
1298 		return;
1299 	}
1300 
1301 	switch (be16_to_cpu(data->type)) {
1302 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1303 		pr_info("P7IOC diag-data for RGC\n\n");
1304 		pnv_eeh_dump_hub_diag_common(data);
1305 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1306 			pr_info("  RGC: %016llx %016llx\n",
1307 				be64_to_cpu(data->rgc.rgcStatus),
1308 				be64_to_cpu(data->rgc.rgcLdcp));
1309 		break;
1310 	case OPAL_P7IOC_DIAG_TYPE_BI:
1311 		pr_info("P7IOC diag-data for BI %s\n\n",
1312 			data->bi.biDownbound ? "Downbound" : "Upbound");
1313 		pnv_eeh_dump_hub_diag_common(data);
1314 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1315 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1316 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1317 				be64_to_cpu(data->bi.biLdcp0),
1318 				be64_to_cpu(data->bi.biLdcp1),
1319 				be64_to_cpu(data->bi.biLdcp2),
1320 				be64_to_cpu(data->bi.biFenceStatus));
1321 		break;
1322 	case OPAL_P7IOC_DIAG_TYPE_CI:
1323 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1324 			data->ci.ciPort);
1325 		pnv_eeh_dump_hub_diag_common(data);
1326 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1327 			pr_info("  CI:  %016llx %016llx\n",
1328 				be64_to_cpu(data->ci.ciPortStatus),
1329 				be64_to_cpu(data->ci.ciPortLdcp));
1330 		break;
1331 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1332 		pr_info("P7IOC diag-data for MISC\n\n");
1333 		pnv_eeh_dump_hub_diag_common(data);
1334 		break;
1335 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1336 		pr_info("P7IOC diag-data for I2C\n\n");
1337 		pnv_eeh_dump_hub_diag_common(data);
1338 		break;
1339 	default:
1340 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1341 			__func__, phb->hub_id, data->type);
1342 	}
1343 }
1344 
1345 static int pnv_eeh_get_pe(struct pci_controller *hose,
1346 			  u16 pe_no, struct eeh_pe **pe)
1347 {
1348 	struct pnv_phb *phb = hose->private_data;
1349 	struct pnv_ioda_pe *pnv_pe;
1350 	struct eeh_pe *dev_pe;
1351 
1352 	/*
1353 	 * If PHB supports compound PE, to fetch
1354 	 * the master PE because slave PE is invisible
1355 	 * to EEH core.
1356 	 */
1357 	pnv_pe = &phb->ioda.pe_array[pe_no];
1358 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1359 		pnv_pe = pnv_pe->master;
1360 		WARN_ON(!pnv_pe ||
1361 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1362 		pe_no = pnv_pe->pe_number;
1363 	}
1364 
1365 	/* Find the PE according to PE# */
1366 	dev_pe = eeh_pe_get(hose, pe_no, 0);
1367 	if (!dev_pe)
1368 		return -EEXIST;
1369 
1370 	/* Freeze the (compound) PE */
1371 	*pe = dev_pe;
1372 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1373 		phb->freeze_pe(phb, pe_no);
1374 
1375 	/*
1376 	 * At this point, we're sure the (compound) PE should
1377 	 * have been frozen. However, we still need poke until
1378 	 * hitting the frozen PE on top level.
1379 	 */
1380 	dev_pe = dev_pe->parent;
1381 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1382 		int ret;
1383 		ret = eeh_ops->get_state(dev_pe, NULL);
1384 		if (ret <= 0 || eeh_state_active(ret)) {
1385 			dev_pe = dev_pe->parent;
1386 			continue;
1387 		}
1388 
1389 		/* Frozen parent PE */
1390 		*pe = dev_pe;
1391 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1392 			phb->freeze_pe(phb, dev_pe->addr);
1393 
1394 		/* Next one */
1395 		dev_pe = dev_pe->parent;
1396 	}
1397 
1398 	return 0;
1399 }
1400 
1401 /**
1402  * pnv_eeh_next_error - Retrieve next EEH error to handle
1403  * @pe: Affected PE
1404  *
1405  * The function is expected to be called by EEH core while it gets
1406  * special EEH event (without binding PE). The function calls to
1407  * OPAL APIs for next error to handle. The informational error is
1408  * handled internally by platform. However, the dead IOC, dead PHB,
1409  * fenced PHB and frozen PE should be handled by EEH core eventually.
1410  */
1411 static int pnv_eeh_next_error(struct eeh_pe **pe)
1412 {
1413 	struct pci_controller *hose;
1414 	struct pnv_phb *phb;
1415 	struct eeh_pe *phb_pe, *parent_pe;
1416 	__be64 frozen_pe_no;
1417 	__be16 err_type, severity;
1418 	long rc;
1419 	int state, ret = EEH_NEXT_ERR_NONE;
1420 
1421 	/*
1422 	 * While running here, it's safe to purge the event queue. The
1423 	 * event should still be masked.
1424 	 */
1425 	eeh_remove_event(NULL, false);
1426 
1427 	list_for_each_entry(hose, &hose_list, list_node) {
1428 		/*
1429 		 * If the subordinate PCI buses of the PHB has been
1430 		 * removed or is exactly under error recovery, we
1431 		 * needn't take care of it any more.
1432 		 */
1433 		phb = hose->private_data;
1434 		phb_pe = eeh_phb_pe_get(hose);
1435 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1436 			continue;
1437 
1438 		rc = opal_pci_next_error(phb->opal_id,
1439 					 &frozen_pe_no, &err_type, &severity);
1440 		if (rc != OPAL_SUCCESS) {
1441 			pr_devel("%s: Invalid return value on "
1442 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1443 				 __func__, hose->global_number, rc);
1444 			continue;
1445 		}
1446 
1447 		/* If the PHB doesn't have error, stop processing */
1448 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1449 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1450 			pr_devel("%s: No error found on PHB#%x\n",
1451 				 __func__, hose->global_number);
1452 			continue;
1453 		}
1454 
1455 		/*
1456 		 * Processing the error. We're expecting the error with
1457 		 * highest priority reported upon multiple errors on the
1458 		 * specific PHB.
1459 		 */
1460 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1461 			__func__, be16_to_cpu(err_type),
1462 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1463 			hose->global_number);
1464 		switch (be16_to_cpu(err_type)) {
1465 		case OPAL_EEH_IOC_ERROR:
1466 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1467 				pr_err("EEH: dead IOC detected\n");
1468 				ret = EEH_NEXT_ERR_DEAD_IOC;
1469 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1470 				pr_info("EEH: IOC informative error "
1471 					"detected\n");
1472 				pnv_eeh_get_and_dump_hub_diag(hose);
1473 				ret = EEH_NEXT_ERR_NONE;
1474 			}
1475 
1476 			break;
1477 		case OPAL_EEH_PHB_ERROR:
1478 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1479 				*pe = phb_pe;
1480 				pr_err("EEH: dead PHB#%x detected, "
1481 				       "location: %s\n",
1482 					hose->global_number,
1483 					eeh_pe_loc_get(phb_pe));
1484 				ret = EEH_NEXT_ERR_DEAD_PHB;
1485 			} else if (be16_to_cpu(severity) ==
1486 				   OPAL_EEH_SEV_PHB_FENCED) {
1487 				*pe = phb_pe;
1488 				pr_err("EEH: Fenced PHB#%x detected, "
1489 				       "location: %s\n",
1490 					hose->global_number,
1491 					eeh_pe_loc_get(phb_pe));
1492 				ret = EEH_NEXT_ERR_FENCED_PHB;
1493 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1494 				pr_info("EEH: PHB#%x informative error "
1495 					"detected, location: %s\n",
1496 					hose->global_number,
1497 					eeh_pe_loc_get(phb_pe));
1498 				pnv_eeh_get_phb_diag(phb_pe);
1499 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1500 				ret = EEH_NEXT_ERR_NONE;
1501 			}
1502 
1503 			break;
1504 		case OPAL_EEH_PE_ERROR:
1505 			/*
1506 			 * If we can't find the corresponding PE, we
1507 			 * just try to unfreeze.
1508 			 */
1509 			if (pnv_eeh_get_pe(hose,
1510 				be64_to_cpu(frozen_pe_no), pe)) {
1511 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1512 					hose->global_number, be64_to_cpu(frozen_pe_no));
1513 				pr_info("EEH: PHB location: %s\n",
1514 					eeh_pe_loc_get(phb_pe));
1515 
1516 				/* Dump PHB diag-data */
1517 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1518 					phb->diag_data, phb->diag_data_size);
1519 				if (rc == OPAL_SUCCESS)
1520 					pnv_pci_dump_phb_diag_data(hose,
1521 							phb->diag_data);
1522 
1523 				/* Try best to clear it */
1524 				opal_pci_eeh_freeze_clear(phb->opal_id,
1525 					be64_to_cpu(frozen_pe_no),
1526 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1527 				ret = EEH_NEXT_ERR_NONE;
1528 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1529 				   eeh_pe_passed(*pe)) {
1530 				ret = EEH_NEXT_ERR_NONE;
1531 			} else {
1532 				pr_err("EEH: Frozen PE#%x "
1533 				       "on PHB#%x detected\n",
1534 				       (*pe)->addr,
1535 					(*pe)->phb->global_number);
1536 				pr_err("EEH: PE location: %s, "
1537 				       "PHB location: %s\n",
1538 				       eeh_pe_loc_get(*pe),
1539 				       eeh_pe_loc_get(phb_pe));
1540 				ret = EEH_NEXT_ERR_FROZEN_PE;
1541 			}
1542 
1543 			break;
1544 		default:
1545 			pr_warn("%s: Unexpected error type %d\n",
1546 				__func__, be16_to_cpu(err_type));
1547 		}
1548 
1549 		/*
1550 		 * EEH core will try recover from fenced PHB or
1551 		 * frozen PE. In the time for frozen PE, EEH core
1552 		 * enable IO path for that before collecting logs,
1553 		 * but it ruins the site. So we have to dump the
1554 		 * log in advance here.
1555 		 */
1556 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1557 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1558 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1559 			eeh_pe_mark_isolated(*pe);
1560 			pnv_eeh_get_phb_diag(*pe);
1561 
1562 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1563 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1564 							   (*pe)->data);
1565 		}
1566 
1567 		/*
1568 		 * We probably have the frozen parent PE out there and
1569 		 * we need have to handle frozen parent PE firstly.
1570 		 */
1571 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1572 			parent_pe = (*pe)->parent;
1573 			while (parent_pe) {
1574 				/* Hit the ceiling ? */
1575 				if (parent_pe->type & EEH_PE_PHB)
1576 					break;
1577 
1578 				/* Frozen parent PE ? */
1579 				state = eeh_ops->get_state(parent_pe, NULL);
1580 				if (state > 0 && !eeh_state_active(state))
1581 					*pe = parent_pe;
1582 
1583 				/* Next parent level */
1584 				parent_pe = parent_pe->parent;
1585 			}
1586 
1587 			/* We possibly migrate to another PE */
1588 			eeh_pe_mark_isolated(*pe);
1589 		}
1590 
1591 		/*
1592 		 * If we have no errors on the specific PHB or only
1593 		 * informative error there, we continue poking it.
1594 		 * Otherwise, we need actions to be taken by upper
1595 		 * layer.
1596 		 */
1597 		if (ret > EEH_NEXT_ERR_INF)
1598 			break;
1599 	}
1600 
1601 	/* Unmask the event */
1602 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1603 		enable_irq(eeh_event_irq);
1604 
1605 	return ret;
1606 }
1607 
1608 static int pnv_eeh_restore_config(struct pci_dn *pdn)
1609 {
1610 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1611 	struct pnv_phb *phb;
1612 	s64 ret = 0;
1613 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
1614 
1615 	if (!edev)
1616 		return -EEXIST;
1617 
1618 	/*
1619 	 * We have to restore the PCI config space after reset since the
1620 	 * firmware can't see SRIOV VFs.
1621 	 *
1622 	 * FIXME: The MPS, error routing rules, timeout setting are worthy
1623 	 * to be exported by firmware in extendible way.
1624 	 */
1625 	if (edev->physfn) {
1626 		ret = eeh_restore_vf_config(pdn);
1627 	} else {
1628 		phb = pdn->phb->private_data;
1629 		ret = opal_pci_reinit(phb->opal_id,
1630 				      OPAL_REINIT_PCI_DEV, config_addr);
1631 	}
1632 
1633 	if (ret) {
1634 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1635 			__func__, config_addr, ret);
1636 		return -EIO;
1637 	}
1638 
1639 	return ret;
1640 }
1641 
1642 static struct eeh_ops pnv_eeh_ops = {
1643 	.name                   = "powernv",
1644 	.init                   = pnv_eeh_init,
1645 	.probe			= pnv_eeh_probe,
1646 	.set_option             = pnv_eeh_set_option,
1647 	.get_pe_addr            = pnv_eeh_get_pe_addr,
1648 	.get_state              = pnv_eeh_get_state,
1649 	.reset                  = pnv_eeh_reset,
1650 	.get_log                = pnv_eeh_get_log,
1651 	.configure_bridge       = pnv_eeh_configure_bridge,
1652 	.err_inject		= pnv_eeh_err_inject,
1653 	.read_config            = pnv_eeh_read_config,
1654 	.write_config           = pnv_eeh_write_config,
1655 	.next_error		= pnv_eeh_next_error,
1656 	.restore_config		= pnv_eeh_restore_config,
1657 	.notify_resume		= NULL
1658 };
1659 
1660 #ifdef CONFIG_PCI_IOV
1661 static void pnv_pci_fixup_vf_mps(struct pci_dev *pdev)
1662 {
1663 	struct pci_dn *pdn = pci_get_pdn(pdev);
1664 	int parent_mps;
1665 
1666 	if (!pdev->is_virtfn)
1667 		return;
1668 
1669 	/* Synchronize MPS for VF and PF */
1670 	parent_mps = pcie_get_mps(pdev->physfn);
1671 	if ((128 << pdev->pcie_mpss) >= parent_mps)
1672 		pcie_set_mps(pdev, parent_mps);
1673 	pdn->mps = pcie_get_mps(pdev);
1674 }
1675 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pnv_pci_fixup_vf_mps);
1676 #endif /* CONFIG_PCI_IOV */
1677 
1678 /**
1679  * eeh_powernv_init - Register platform dependent EEH operations
1680  *
1681  * EEH initialization on powernv platform. This function should be
1682  * called before any EEH related functions.
1683  */
1684 static int __init eeh_powernv_init(void)
1685 {
1686 	int ret = -EINVAL;
1687 
1688 	ret = eeh_ops_register(&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_early_initcall(powernv, eeh_powernv_init);
1697