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