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 	if (eeh_has_flag(EEH_POSTPONED_PROBE)) {
227 		eeh_clear_flag(EEH_POSTPONED_PROBE);
228 		if (eeh_enabled())
229 			pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
230 		else
231 			pr_info("EEH: No capable adapters found\n");
232 	}
233 
234 	/* Register OPAL event notifier */
235 	eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR));
236 	if (eeh_event_irq < 0) {
237 		pr_err("%s: Can't register OPAL event interrupt (%d)\n",
238 		       __func__, eeh_event_irq);
239 		return eeh_event_irq;
240 	}
241 
242 	ret = request_irq(eeh_event_irq, pnv_eeh_event,
243 			  IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL);
244 	if (ret < 0) {
245 		irq_dispose_mapping(eeh_event_irq);
246 		pr_err("%s: Can't request OPAL event interrupt (%d)\n",
247 		       __func__, eeh_event_irq);
248 		return ret;
249 	}
250 
251 	if (!eeh_enabled())
252 		disable_irq(eeh_event_irq);
253 
254 	list_for_each_entry(hose, &hose_list, list_node) {
255 		phb = hose->private_data;
256 
257 		/*
258 		 * If EEH is enabled, we're going to rely on that.
259 		 * Otherwise, we restore to conventional mechanism
260 		 * to clear frozen PE during PCI config access.
261 		 */
262 		if (eeh_enabled())
263 			phb->flags |= PNV_PHB_FLAG_EEH;
264 		else
265 			phb->flags &= ~PNV_PHB_FLAG_EEH;
266 
267 		/* Create debugfs entries */
268 #ifdef CONFIG_DEBUG_FS
269 		if (phb->has_dbgfs || !phb->dbgfs)
270 			continue;
271 
272 		phb->has_dbgfs = 1;
273 		debugfs_create_file("err_injct", 0200,
274 				    phb->dbgfs, hose,
275 				    &pnv_eeh_ei_fops);
276 
277 		debugfs_create_file("err_injct_outbound", 0600,
278 				    phb->dbgfs, hose,
279 				    &pnv_eeh_dbgfs_ops_outb);
280 		debugfs_create_file("err_injct_inboundA", 0600,
281 				    phb->dbgfs, hose,
282 				    &pnv_eeh_dbgfs_ops_inbA);
283 		debugfs_create_file("err_injct_inboundB", 0600,
284 				    phb->dbgfs, hose,
285 				    &pnv_eeh_dbgfs_ops_inbB);
286 #endif /* CONFIG_DEBUG_FS */
287 	}
288 
289 	return ret;
290 }
291 
292 static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap)
293 {
294 	int pos = PCI_CAPABILITY_LIST;
295 	int cnt = 48;   /* Maximal number of capabilities */
296 	u32 status, id;
297 
298 	if (!pdn)
299 		return 0;
300 
301 	/* Check if the device supports capabilities */
302 	pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status);
303 	if (!(status & PCI_STATUS_CAP_LIST))
304 		return 0;
305 
306 	while (cnt--) {
307 		pnv_pci_cfg_read(pdn, pos, 1, &pos);
308 		if (pos < 0x40)
309 			break;
310 
311 		pos &= ~3;
312 		pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
313 		if (id == 0xff)
314 			break;
315 
316 		/* Found */
317 		if (id == cap)
318 			return pos;
319 
320 		/* Next one */
321 		pos += PCI_CAP_LIST_NEXT;
322 	}
323 
324 	return 0;
325 }
326 
327 static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap)
328 {
329 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
330 	u32 header;
331 	int pos = 256, ttl = (4096 - 256) / 8;
332 
333 	if (!edev || !edev->pcie_cap)
334 		return 0;
335 	if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
336 		return 0;
337 	else if (!header)
338 		return 0;
339 
340 	while (ttl-- > 0) {
341 		if (PCI_EXT_CAP_ID(header) == cap && pos)
342 			return pos;
343 
344 		pos = PCI_EXT_CAP_NEXT(header);
345 		if (pos < 256)
346 			break;
347 
348 		if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
349 			break;
350 	}
351 
352 	return 0;
353 }
354 
355 /**
356  * pnv_eeh_probe - Do probe on PCI device
357  * @pdn: PCI device node
358  * @data: unused
359  *
360  * When EEH module is installed during system boot, all PCI devices
361  * are checked one by one to see if it supports EEH. The function
362  * is introduced for the purpose. By default, EEH has been enabled
363  * on all PCI devices. That's to say, we only need do necessary
364  * initialization on the corresponding eeh device and create PE
365  * accordingly.
366  *
367  * It's notable that's unsafe to retrieve the EEH device through
368  * the corresponding PCI device. During the PCI device hotplug, which
369  * was possiblly triggered by EEH core, the binding between EEH device
370  * and the PCI device isn't built yet.
371  */
372 static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
373 {
374 	struct pci_controller *hose = pdn->phb;
375 	struct pnv_phb *phb = hose->private_data;
376 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
377 	uint32_t pcie_flags;
378 	int ret;
379 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
380 
381 	/*
382 	 * When probing the root bridge, which doesn't have any
383 	 * subordinate PCI devices. We don't have OF node for
384 	 * the root bridge. So it's not reasonable to continue
385 	 * the probing.
386 	 */
387 	if (!edev || edev->pe)
388 		return NULL;
389 
390 	/* Skip for PCI-ISA bridge */
391 	if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
392 		return NULL;
393 
394 	/* Skip if we haven't probed yet */
395 	if (phb->ioda.pe_rmap[config_addr] == IODA_INVALID_PE) {
396 		eeh_add_flag(EEH_POSTPONED_PROBE);
397 		return NULL;
398 	}
399 
400 	/* Initialize eeh device */
401 	edev->class_code = pdn->class_code;
402 	edev->mode	&= 0xFFFFFF00;
403 	edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
404 	edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
405 	edev->af_cap   = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF);
406 	edev->aer_cap  = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
407 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
408 		edev->mode |= EEH_DEV_BRIDGE;
409 		if (edev->pcie_cap) {
410 			pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
411 					 2, &pcie_flags);
412 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
413 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
414 				edev->mode |= EEH_DEV_ROOT_PORT;
415 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
416 				edev->mode |= EEH_DEV_DS_PORT;
417 		}
418 	}
419 
420 	edev->pe_config_addr = phb->ioda.pe_rmap[config_addr];
421 
422 	/* Create PE */
423 	ret = eeh_add_to_parent_pe(edev);
424 	if (ret) {
425 		pr_warn("%s: Can't add PCI dev %04x:%02x:%02x.%01x to parent PE (%x)\n",
426 			__func__, hose->global_number, pdn->busno,
427 			PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn), ret);
428 		return NULL;
429 	}
430 
431 	/*
432 	 * If the PE contains any one of following adapters, the
433 	 * PCI config space can't be accessed when dumping EEH log.
434 	 * Otherwise, we will run into fenced PHB caused by shortage
435 	 * of outbound credits in the adapter. The PCI config access
436 	 * should be blocked until PE reset. MMIO access is dropped
437 	 * by hardware certainly. In order to drop PCI config requests,
438 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
439 	 * will be checked in the backend for PE state retrival. If
440 	 * the PE becomes frozen for the first time and the flag has
441 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
442 	 * that PE to block its config space.
443 	 *
444 	 * Broadcom BCM5718 2-ports NICs (14e4:1656)
445 	 * Broadcom Austin 4-ports NICs (14e4:1657)
446 	 * Broadcom Shiner 4-ports 1G NICs (14e4:168a)
447 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
448 	 */
449 	if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
450 	     pdn->device_id == 0x1656) ||
451 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
452 	     pdn->device_id == 0x1657) ||
453 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
454 	     pdn->device_id == 0x168a) ||
455 	    (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM &&
456 	     pdn->device_id == 0x168e))
457 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
458 
459 	/*
460 	 * Cache the PE primary bus, which can't be fetched when
461 	 * full hotplug is in progress. In that case, all child
462 	 * PCI devices of the PE are expected to be removed prior
463 	 * to PE reset.
464 	 */
465 	if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
466 		edev->pe->bus = pci_find_bus(hose->global_number,
467 					     pdn->busno);
468 		if (edev->pe->bus)
469 			edev->pe->state |= EEH_PE_PRI_BUS;
470 	}
471 
472 	/*
473 	 * Enable EEH explicitly so that we will do EEH check
474 	 * while accessing I/O stuff
475 	 */
476 	eeh_add_flag(EEH_ENABLED);
477 
478 	/* Save memory bars */
479 	eeh_save_bars(edev);
480 
481 	return NULL;
482 }
483 
484 /**
485  * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
486  * @pe: EEH PE
487  * @option: operation to be issued
488  *
489  * The function is used to control the EEH functionality globally.
490  * Currently, following options are support according to PAPR:
491  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
492  */
493 static int pnv_eeh_set_option(struct eeh_pe *pe, int option)
494 {
495 	struct pci_controller *hose = pe->phb;
496 	struct pnv_phb *phb = hose->private_data;
497 	bool freeze_pe = false;
498 	int opt;
499 	s64 rc;
500 
501 	switch (option) {
502 	case EEH_OPT_DISABLE:
503 		return -EPERM;
504 	case EEH_OPT_ENABLE:
505 		return 0;
506 	case EEH_OPT_THAW_MMIO:
507 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
508 		break;
509 	case EEH_OPT_THAW_DMA:
510 		opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
511 		break;
512 	case EEH_OPT_FREEZE_PE:
513 		freeze_pe = true;
514 		opt = OPAL_EEH_ACTION_SET_FREEZE_ALL;
515 		break;
516 	default:
517 		pr_warn("%s: Invalid option %d\n", __func__, option);
518 		return -EINVAL;
519 	}
520 
521 	/* Freeze master and slave PEs if PHB supports compound PEs */
522 	if (freeze_pe) {
523 		if (phb->freeze_pe) {
524 			phb->freeze_pe(phb, pe->addr);
525 			return 0;
526 		}
527 
528 		rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt);
529 		if (rc != OPAL_SUCCESS) {
530 			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
531 				__func__, rc, phb->hose->global_number,
532 				pe->addr);
533 			return -EIO;
534 		}
535 
536 		return 0;
537 	}
538 
539 	/* Unfreeze master and slave PEs if PHB supports */
540 	if (phb->unfreeze_pe)
541 		return phb->unfreeze_pe(phb, pe->addr, opt);
542 
543 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt);
544 	if (rc != OPAL_SUCCESS) {
545 		pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n",
546 			__func__, rc, option, phb->hose->global_number,
547 			pe->addr);
548 		return -EIO;
549 	}
550 
551 	return 0;
552 }
553 
554 /**
555  * pnv_eeh_get_pe_addr - Retrieve PE address
556  * @pe: EEH PE
557  *
558  * Retrieve the PE address according to the given tranditional
559  * PCI BDF (Bus/Device/Function) address.
560  */
561 static int pnv_eeh_get_pe_addr(struct eeh_pe *pe)
562 {
563 	return pe->addr;
564 }
565 
566 static void pnv_eeh_get_phb_diag(struct eeh_pe *pe)
567 {
568 	struct pnv_phb *phb = pe->phb->private_data;
569 	s64 rc;
570 
571 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
572 					 phb->diag_data_size);
573 	if (rc != OPAL_SUCCESS)
574 		pr_warn("%s: Failure %lld getting PHB#%x diag-data\n",
575 			__func__, rc, pe->phb->global_number);
576 }
577 
578 static int pnv_eeh_get_phb_state(struct eeh_pe *pe)
579 {
580 	struct pnv_phb *phb = pe->phb->private_data;
581 	u8 fstate;
582 	__be16 pcierr;
583 	s64 rc;
584 	int result = 0;
585 
586 	rc = opal_pci_eeh_freeze_status(phb->opal_id,
587 					pe->addr,
588 					&fstate,
589 					&pcierr,
590 					NULL);
591 	if (rc != OPAL_SUCCESS) {
592 		pr_warn("%s: Failure %lld getting PHB#%x state\n",
593 			__func__, rc, phb->hose->global_number);
594 		return EEH_STATE_NOT_SUPPORT;
595 	}
596 
597 	/*
598 	 * Check PHB state. If the PHB is frozen for the
599 	 * first time, to dump the PHB diag-data.
600 	 */
601 	if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
602 		result = (EEH_STATE_MMIO_ACTIVE  |
603 			  EEH_STATE_DMA_ACTIVE   |
604 			  EEH_STATE_MMIO_ENABLED |
605 			  EEH_STATE_DMA_ENABLED);
606 	} else if (!(pe->state & EEH_PE_ISOLATED)) {
607 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
608 		pnv_eeh_get_phb_diag(pe);
609 
610 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
611 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
612 	}
613 
614 	return result;
615 }
616 
617 static int pnv_eeh_get_pe_state(struct eeh_pe *pe)
618 {
619 	struct pnv_phb *phb = pe->phb->private_data;
620 	u8 fstate;
621 	__be16 pcierr;
622 	s64 rc;
623 	int result;
624 
625 	/*
626 	 * We don't clobber hardware frozen state until PE
627 	 * reset is completed. In order to keep EEH core
628 	 * moving forward, we have to return operational
629 	 * state during PE reset.
630 	 */
631 	if (pe->state & EEH_PE_RESET) {
632 		result = (EEH_STATE_MMIO_ACTIVE  |
633 			  EEH_STATE_DMA_ACTIVE   |
634 			  EEH_STATE_MMIO_ENABLED |
635 			  EEH_STATE_DMA_ENABLED);
636 		return result;
637 	}
638 
639 	/*
640 	 * Fetch PE state from hardware. If the PHB
641 	 * supports compound PE, let it handle that.
642 	 */
643 	if (phb->get_pe_state) {
644 		fstate = phb->get_pe_state(phb, pe->addr);
645 	} else {
646 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
647 						pe->addr,
648 						&fstate,
649 						&pcierr,
650 						NULL);
651 		if (rc != OPAL_SUCCESS) {
652 			pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
653 				__func__, rc, phb->hose->global_number,
654 				pe->addr);
655 			return EEH_STATE_NOT_SUPPORT;
656 		}
657 	}
658 
659 	/* Figure out state */
660 	switch (fstate) {
661 	case OPAL_EEH_STOPPED_NOT_FROZEN:
662 		result = (EEH_STATE_MMIO_ACTIVE  |
663 			  EEH_STATE_DMA_ACTIVE   |
664 			  EEH_STATE_MMIO_ENABLED |
665 			  EEH_STATE_DMA_ENABLED);
666 		break;
667 	case OPAL_EEH_STOPPED_MMIO_FREEZE:
668 		result = (EEH_STATE_DMA_ACTIVE |
669 			  EEH_STATE_DMA_ENABLED);
670 		break;
671 	case OPAL_EEH_STOPPED_DMA_FREEZE:
672 		result = (EEH_STATE_MMIO_ACTIVE |
673 			  EEH_STATE_MMIO_ENABLED);
674 		break;
675 	case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
676 		result = 0;
677 		break;
678 	case OPAL_EEH_STOPPED_RESET:
679 		result = EEH_STATE_RESET_ACTIVE;
680 		break;
681 	case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
682 		result = EEH_STATE_UNAVAILABLE;
683 		break;
684 	case OPAL_EEH_STOPPED_PERM_UNAVAIL:
685 		result = EEH_STATE_NOT_SUPPORT;
686 		break;
687 	default:
688 		result = EEH_STATE_NOT_SUPPORT;
689 		pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
690 			__func__, phb->hose->global_number,
691 			pe->addr, fstate);
692 	}
693 
694 	/*
695 	 * If PHB supports compound PE, to freeze all
696 	 * slave PEs for consistency.
697 	 *
698 	 * If the PE is switching to frozen state for the
699 	 * first time, to dump the PHB diag-data.
700 	 */
701 	if (!(result & EEH_STATE_NOT_SUPPORT) &&
702 	    !(result & EEH_STATE_UNAVAILABLE) &&
703 	    !(result & EEH_STATE_MMIO_ACTIVE) &&
704 	    !(result & EEH_STATE_DMA_ACTIVE)  &&
705 	    !(pe->state & EEH_PE_ISOLATED)) {
706 		if (phb->freeze_pe)
707 			phb->freeze_pe(phb, pe->addr);
708 
709 		eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
710 		pnv_eeh_get_phb_diag(pe);
711 
712 		if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
713 			pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
714 	}
715 
716 	return result;
717 }
718 
719 /**
720  * pnv_eeh_get_state - Retrieve PE state
721  * @pe: EEH PE
722  * @delay: delay while PE state is temporarily unavailable
723  *
724  * Retrieve the state of the specified PE. For IODA-compitable
725  * platform, it should be retrieved from IODA table. Therefore,
726  * we prefer passing down to hardware implementation to handle
727  * it.
728  */
729 static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay)
730 {
731 	int ret;
732 
733 	if (pe->type & EEH_PE_PHB)
734 		ret = pnv_eeh_get_phb_state(pe);
735 	else
736 		ret = pnv_eeh_get_pe_state(pe);
737 
738 	if (!delay)
739 		return ret;
740 
741 	/*
742 	 * If the PE state is temporarily unavailable,
743 	 * to inform the EEH core delay for default
744 	 * period (1 second)
745 	 */
746 	*delay = 0;
747 	if (ret & EEH_STATE_UNAVAILABLE)
748 		*delay = 1000;
749 
750 	return ret;
751 }
752 
753 static s64 pnv_eeh_poll(unsigned long id)
754 {
755 	s64 rc = OPAL_HARDWARE;
756 
757 	while (1) {
758 		rc = opal_pci_poll(id);
759 		if (rc <= 0)
760 			break;
761 
762 		if (system_state < SYSTEM_RUNNING)
763 			udelay(1000 * rc);
764 		else
765 			msleep(rc);
766 	}
767 
768 	return rc;
769 }
770 
771 int pnv_eeh_phb_reset(struct pci_controller *hose, int option)
772 {
773 	struct pnv_phb *phb = hose->private_data;
774 	s64 rc = OPAL_HARDWARE;
775 
776 	pr_debug("%s: Reset PHB#%x, option=%d\n",
777 		 __func__, hose->global_number, option);
778 
779 	/* Issue PHB complete reset request */
780 	if (option == EEH_RESET_FUNDAMENTAL ||
781 	    option == EEH_RESET_HOT)
782 		rc = opal_pci_reset(phb->opal_id,
783 				    OPAL_RESET_PHB_COMPLETE,
784 				    OPAL_ASSERT_RESET);
785 	else if (option == EEH_RESET_DEACTIVATE)
786 		rc = opal_pci_reset(phb->opal_id,
787 				    OPAL_RESET_PHB_COMPLETE,
788 				    OPAL_DEASSERT_RESET);
789 	if (rc < 0)
790 		goto out;
791 
792 	/*
793 	 * Poll state of the PHB until the request is done
794 	 * successfully. The PHB reset is usually PHB complete
795 	 * reset followed by hot reset on root bus. So we also
796 	 * need the PCI bus settlement delay.
797 	 */
798 	if (rc > 0)
799 		rc = pnv_eeh_poll(phb->opal_id);
800 	if (option == EEH_RESET_DEACTIVATE) {
801 		if (system_state < SYSTEM_RUNNING)
802 			udelay(1000 * EEH_PE_RST_SETTLE_TIME);
803 		else
804 			msleep(EEH_PE_RST_SETTLE_TIME);
805 	}
806 out:
807 	if (rc != OPAL_SUCCESS)
808 		return -EIO;
809 
810 	return 0;
811 }
812 
813 static int pnv_eeh_root_reset(struct pci_controller *hose, int option)
814 {
815 	struct pnv_phb *phb = hose->private_data;
816 	s64 rc = OPAL_HARDWARE;
817 
818 	pr_debug("%s: Reset PHB#%x, option=%d\n",
819 		 __func__, hose->global_number, option);
820 
821 	/*
822 	 * During the reset deassert time, we needn't care
823 	 * the reset scope because the firmware does nothing
824 	 * for fundamental or hot reset during deassert phase.
825 	 */
826 	if (option == EEH_RESET_FUNDAMENTAL)
827 		rc = opal_pci_reset(phb->opal_id,
828 				    OPAL_RESET_PCI_FUNDAMENTAL,
829 				    OPAL_ASSERT_RESET);
830 	else if (option == EEH_RESET_HOT)
831 		rc = opal_pci_reset(phb->opal_id,
832 				    OPAL_RESET_PCI_HOT,
833 				    OPAL_ASSERT_RESET);
834 	else if (option == EEH_RESET_DEACTIVATE)
835 		rc = opal_pci_reset(phb->opal_id,
836 				    OPAL_RESET_PCI_HOT,
837 				    OPAL_DEASSERT_RESET);
838 	if (rc < 0)
839 		goto out;
840 
841 	/* Poll state of the PHB until the request is done */
842 	if (rc > 0)
843 		rc = pnv_eeh_poll(phb->opal_id);
844 	if (option == EEH_RESET_DEACTIVATE)
845 		msleep(EEH_PE_RST_SETTLE_TIME);
846 out:
847 	if (rc != OPAL_SUCCESS)
848 		return -EIO;
849 
850 	return 0;
851 }
852 
853 static int __pnv_eeh_bridge_reset(struct pci_dev *dev, int option)
854 {
855 	struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn);
856 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
857 	int aer = edev ? edev->aer_cap : 0;
858 	u32 ctrl;
859 
860 	pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
861 		 __func__, pci_domain_nr(dev->bus),
862 		 dev->bus->number, option);
863 
864 	switch (option) {
865 	case EEH_RESET_FUNDAMENTAL:
866 	case EEH_RESET_HOT:
867 		/* Don't report linkDown event */
868 		if (aer) {
869 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
870 					     4, &ctrl);
871 			ctrl |= PCI_ERR_UNC_SURPDN;
872 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
873 					      4, ctrl);
874 		}
875 
876 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
877 		ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
878 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
879 
880 		msleep(EEH_PE_RST_HOLD_TIME);
881 		break;
882 	case EEH_RESET_DEACTIVATE:
883 		eeh_ops->read_config(pdn, PCI_BRIDGE_CONTROL, 2, &ctrl);
884 		ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
885 		eeh_ops->write_config(pdn, PCI_BRIDGE_CONTROL, 2, ctrl);
886 
887 		msleep(EEH_PE_RST_SETTLE_TIME);
888 
889 		/* Continue reporting linkDown event */
890 		if (aer) {
891 			eeh_ops->read_config(pdn, aer + PCI_ERR_UNCOR_MASK,
892 					     4, &ctrl);
893 			ctrl &= ~PCI_ERR_UNC_SURPDN;
894 			eeh_ops->write_config(pdn, aer + PCI_ERR_UNCOR_MASK,
895 					      4, ctrl);
896 		}
897 
898 		break;
899 	}
900 
901 	return 0;
902 }
903 
904 static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option)
905 {
906 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
907 	struct pnv_phb *phb = hose->private_data;
908 	struct device_node *dn = pci_device_to_OF_node(pdev);
909 	uint64_t id = PCI_SLOT_ID(phb->opal_id,
910 				  (pdev->bus->number << 8) | pdev->devfn);
911 	uint8_t scope;
912 	int64_t rc;
913 
914 	/* Hot reset to the bus if firmware cannot handle */
915 	if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL))
916 		return __pnv_eeh_bridge_reset(pdev, option);
917 
918 	switch (option) {
919 	case EEH_RESET_FUNDAMENTAL:
920 		scope = OPAL_RESET_PCI_FUNDAMENTAL;
921 		break;
922 	case EEH_RESET_HOT:
923 		scope = OPAL_RESET_PCI_HOT;
924 		break;
925 	case EEH_RESET_DEACTIVATE:
926 		return 0;
927 	default:
928 		dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n",
929 			__func__, option);
930 		return -EINVAL;
931 	}
932 
933 	rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET);
934 	if (rc <= OPAL_SUCCESS)
935 		goto out;
936 
937 	rc = pnv_eeh_poll(id);
938 out:
939 	return (rc == OPAL_SUCCESS) ? 0 : -EIO;
940 }
941 
942 void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
943 {
944 	struct pci_controller *hose;
945 
946 	if (pci_is_root_bus(dev->bus)) {
947 		hose = pci_bus_to_host(dev->bus);
948 		pnv_eeh_root_reset(hose, EEH_RESET_HOT);
949 		pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
950 	} else {
951 		pnv_eeh_bridge_reset(dev, EEH_RESET_HOT);
952 		pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
953 	}
954 }
955 
956 static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type,
957 				     int pos, u16 mask)
958 {
959 	int i, status = 0;
960 
961 	/* Wait for Transaction Pending bit to be cleared */
962 	for (i = 0; i < 4; i++) {
963 		eeh_ops->read_config(pdn, pos, 2, &status);
964 		if (!(status & mask))
965 			return;
966 
967 		msleep((1 << i) * 100);
968 	}
969 
970 	pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n",
971 		__func__, type,
972 		pdn->phb->global_number, pdn->busno,
973 		PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn));
974 }
975 
976 static int pnv_eeh_do_flr(struct pci_dn *pdn, int option)
977 {
978 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
979 	u32 reg = 0;
980 
981 	if (WARN_ON(!edev->pcie_cap))
982 		return -ENOTTY;
983 
984 	eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCAP, 4, &reg);
985 	if (!(reg & PCI_EXP_DEVCAP_FLR))
986 		return -ENOTTY;
987 
988 	switch (option) {
989 	case EEH_RESET_HOT:
990 	case EEH_RESET_FUNDAMENTAL:
991 		pnv_eeh_wait_for_pending(pdn, "",
992 					 edev->pcie_cap + PCI_EXP_DEVSTA,
993 					 PCI_EXP_DEVSTA_TRPND);
994 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
995 				     4, &reg);
996 		reg |= PCI_EXP_DEVCTL_BCR_FLR;
997 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
998 				      4, reg);
999 		msleep(EEH_PE_RST_HOLD_TIME);
1000 		break;
1001 	case EEH_RESET_DEACTIVATE:
1002 		eeh_ops->read_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
1003 				     4, &reg);
1004 		reg &= ~PCI_EXP_DEVCTL_BCR_FLR;
1005 		eeh_ops->write_config(pdn, edev->pcie_cap + PCI_EXP_DEVCTL,
1006 				      4, reg);
1007 		msleep(EEH_PE_RST_SETTLE_TIME);
1008 		break;
1009 	}
1010 
1011 	return 0;
1012 }
1013 
1014 static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option)
1015 {
1016 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1017 	u32 cap = 0;
1018 
1019 	if (WARN_ON(!edev->af_cap))
1020 		return -ENOTTY;
1021 
1022 	eeh_ops->read_config(pdn, edev->af_cap + PCI_AF_CAP, 1, &cap);
1023 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
1024 		return -ENOTTY;
1025 
1026 	switch (option) {
1027 	case EEH_RESET_HOT:
1028 	case EEH_RESET_FUNDAMENTAL:
1029 		/*
1030 		 * Wait for Transaction Pending bit to clear. A word-aligned
1031 		 * test is used, so we use the conrol offset rather than status
1032 		 * and shift the test bit to match.
1033 		 */
1034 		pnv_eeh_wait_for_pending(pdn, "AF",
1035 					 edev->af_cap + PCI_AF_CTRL,
1036 					 PCI_AF_STATUS_TP << 8);
1037 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL,
1038 				      1, PCI_AF_CTRL_FLR);
1039 		msleep(EEH_PE_RST_HOLD_TIME);
1040 		break;
1041 	case EEH_RESET_DEACTIVATE:
1042 		eeh_ops->write_config(pdn, edev->af_cap + PCI_AF_CTRL, 1, 0);
1043 		msleep(EEH_PE_RST_SETTLE_TIME);
1044 		break;
1045 	}
1046 
1047 	return 0;
1048 }
1049 
1050 static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option)
1051 {
1052 	struct eeh_dev *edev;
1053 	struct pci_dn *pdn;
1054 	int ret;
1055 
1056 	/* The VF PE should have only one child device */
1057 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, list);
1058 	pdn = eeh_dev_to_pdn(edev);
1059 	if (!pdn)
1060 		return -ENXIO;
1061 
1062 	ret = pnv_eeh_do_flr(pdn, option);
1063 	if (!ret)
1064 		return ret;
1065 
1066 	return pnv_eeh_do_af_flr(pdn, option);
1067 }
1068 
1069 /**
1070  * pnv_eeh_reset - Reset the specified PE
1071  * @pe: EEH PE
1072  * @option: reset option
1073  *
1074  * Do reset on the indicated PE. For PCI bus sensitive PE,
1075  * we need to reset the parent p2p bridge. The PHB has to
1076  * be reinitialized if the p2p bridge is root bridge. For
1077  * PCI device sensitive PE, we will try to reset the device
1078  * through FLR. For now, we don't have OPAL APIs to do HARD
1079  * reset yet, so all reset would be SOFT (HOT) reset.
1080  */
1081 static int pnv_eeh_reset(struct eeh_pe *pe, int option)
1082 {
1083 	struct pci_controller *hose = pe->phb;
1084 	struct pnv_phb *phb;
1085 	struct pci_bus *bus;
1086 	int64_t rc;
1087 
1088 	/*
1089 	 * For PHB reset, we always have complete reset. For those PEs whose
1090 	 * primary bus derived from root complex (root bus) or root port
1091 	 * (usually bus#1), we apply hot or fundamental reset on the root port.
1092 	 * For other PEs, we always have hot reset on the PE primary bus.
1093 	 *
1094 	 * Here, we have different design to pHyp, which always clear the
1095 	 * frozen state during PE reset. However, the good idea here from
1096 	 * benh is to keep frozen state before we get PE reset done completely
1097 	 * (until BAR restore). With the frozen state, HW drops illegal IO
1098 	 * or MMIO access, which can incur recrusive frozen PE during PE
1099 	 * reset. The side effect is that EEH core has to clear the frozen
1100 	 * state explicitly after BAR restore.
1101 	 */
1102 	if (pe->type & EEH_PE_PHB)
1103 		return pnv_eeh_phb_reset(hose, option);
1104 
1105 	/*
1106 	 * The frozen PE might be caused by PAPR error injection
1107 	 * registers, which are expected to be cleared after hitting
1108 	 * frozen PE as stated in the hardware spec. Unfortunately,
1109 	 * that's not true on P7IOC. So we have to clear it manually
1110 	 * to avoid recursive EEH errors during recovery.
1111 	 */
1112 	phb = hose->private_data;
1113 	if (phb->model == PNV_PHB_MODEL_P7IOC &&
1114 	    (option == EEH_RESET_HOT ||
1115 	     option == EEH_RESET_FUNDAMENTAL)) {
1116 		rc = opal_pci_reset(phb->opal_id,
1117 				    OPAL_RESET_PHB_ERROR,
1118 				    OPAL_ASSERT_RESET);
1119 		if (rc != OPAL_SUCCESS) {
1120 			pr_warn("%s: Failure %lld clearing error injection registers\n",
1121 				__func__, rc);
1122 			return -EIO;
1123 		}
1124 	}
1125 
1126 	if (pe->type & EEH_PE_VF)
1127 		return pnv_eeh_reset_vf_pe(pe, option);
1128 
1129 	bus = eeh_pe_bus_get(pe);
1130 	if (!bus) {
1131 		pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n",
1132 			__func__, pe->phb->global_number, pe->addr);
1133 		return -EIO;
1134 	}
1135 
1136 	/*
1137 	 * If dealing with the root bus (or the bus underneath the
1138 	 * root port), we reset the bus underneath the root port.
1139 	 *
1140 	 * The cxl driver depends on this behaviour for bi-modal card
1141 	 * switching.
1142 	 */
1143 	if (pci_is_root_bus(bus) ||
1144 	    pci_is_root_bus(bus->parent))
1145 		return pnv_eeh_root_reset(hose, option);
1146 
1147 	return pnv_eeh_bridge_reset(bus->self, option);
1148 }
1149 
1150 /**
1151  * pnv_eeh_wait_state - Wait for PE state
1152  * @pe: EEH PE
1153  * @max_wait: maximal period in millisecond
1154  *
1155  * Wait for the state of associated PE. It might take some time
1156  * to retrieve the PE's state.
1157  */
1158 static int pnv_eeh_wait_state(struct eeh_pe *pe, int max_wait)
1159 {
1160 	int ret;
1161 	int mwait;
1162 
1163 	while (1) {
1164 		ret = pnv_eeh_get_state(pe, &mwait);
1165 
1166 		/*
1167 		 * If the PE's state is temporarily unavailable,
1168 		 * we have to wait for the specified time. Otherwise,
1169 		 * the PE's state will be returned immediately.
1170 		 */
1171 		if (ret != EEH_STATE_UNAVAILABLE)
1172 			return ret;
1173 
1174 		if (max_wait <= 0) {
1175 			pr_warn("%s: Timeout getting PE#%x's state (%d)\n",
1176 				__func__, pe->addr, max_wait);
1177 			return EEH_STATE_NOT_SUPPORT;
1178 		}
1179 
1180 		max_wait -= mwait;
1181 		msleep(mwait);
1182 	}
1183 
1184 	return EEH_STATE_NOT_SUPPORT;
1185 }
1186 
1187 /**
1188  * pnv_eeh_get_log - Retrieve error log
1189  * @pe: EEH PE
1190  * @severity: temporary or permanent error log
1191  * @drv_log: driver log to be combined with retrieved error log
1192  * @len: length of driver log
1193  *
1194  * Retrieve the temporary or permanent error from the PE.
1195  */
1196 static int pnv_eeh_get_log(struct eeh_pe *pe, int severity,
1197 			   char *drv_log, unsigned long len)
1198 {
1199 	if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
1200 		pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
1201 
1202 	return 0;
1203 }
1204 
1205 /**
1206  * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
1207  * @pe: EEH PE
1208  *
1209  * The function will be called to reconfigure the bridges included
1210  * in the specified PE so that the mulfunctional PE would be recovered
1211  * again.
1212  */
1213 static int pnv_eeh_configure_bridge(struct eeh_pe *pe)
1214 {
1215 	return 0;
1216 }
1217 
1218 /**
1219  * pnv_pe_err_inject - Inject specified error to the indicated PE
1220  * @pe: the indicated PE
1221  * @type: error type
1222  * @func: specific error type
1223  * @addr: address
1224  * @mask: address mask
1225  *
1226  * The routine is called to inject specified error, which is
1227  * determined by @type and @func, to the indicated PE for
1228  * testing purpose.
1229  */
1230 static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
1231 			      unsigned long addr, unsigned long mask)
1232 {
1233 	struct pci_controller *hose = pe->phb;
1234 	struct pnv_phb *phb = hose->private_data;
1235 	s64 rc;
1236 
1237 	if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
1238 	    type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
1239 		pr_warn("%s: Invalid error type %d\n",
1240 			__func__, type);
1241 		return -ERANGE;
1242 	}
1243 
1244 	if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
1245 	    func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
1246 		pr_warn("%s: Invalid error function %d\n",
1247 			__func__, func);
1248 		return -ERANGE;
1249 	}
1250 
1251 	/* Firmware supports error injection ? */
1252 	if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
1253 		pr_warn("%s: Firmware doesn't support error injection\n",
1254 			__func__);
1255 		return -ENXIO;
1256 	}
1257 
1258 	/* Do error injection */
1259 	rc = opal_pci_err_inject(phb->opal_id, pe->addr,
1260 				 type, func, addr, mask);
1261 	if (rc != OPAL_SUCCESS) {
1262 		pr_warn("%s: Failure %lld injecting error "
1263 			"%d-%d to PHB#%x-PE#%x\n",
1264 			__func__, rc, type, func,
1265 			hose->global_number, pe->addr);
1266 		return -EIO;
1267 	}
1268 
1269 	return 0;
1270 }
1271 
1272 static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn)
1273 {
1274 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1275 
1276 	if (!edev || !edev->pe)
1277 		return false;
1278 
1279 	/*
1280 	 * We will issue FLR or AF FLR to all VFs, which are contained
1281 	 * in VF PE. It relies on the EEH PCI config accessors. So we
1282 	 * can't block them during the window.
1283 	 */
1284 	if (edev->physfn && (edev->pe->state & EEH_PE_RESET))
1285 		return false;
1286 
1287 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
1288 		return true;
1289 
1290 	return false;
1291 }
1292 
1293 static int pnv_eeh_read_config(struct pci_dn *pdn,
1294 			       int where, int size, u32 *val)
1295 {
1296 	if (!pdn)
1297 		return PCIBIOS_DEVICE_NOT_FOUND;
1298 
1299 	if (pnv_eeh_cfg_blocked(pdn)) {
1300 		*val = 0xFFFFFFFF;
1301 		return PCIBIOS_SET_FAILED;
1302 	}
1303 
1304 	return pnv_pci_cfg_read(pdn, where, size, val);
1305 }
1306 
1307 static int pnv_eeh_write_config(struct pci_dn *pdn,
1308 				int where, int size, u32 val)
1309 {
1310 	if (!pdn)
1311 		return PCIBIOS_DEVICE_NOT_FOUND;
1312 
1313 	if (pnv_eeh_cfg_blocked(pdn))
1314 		return PCIBIOS_SET_FAILED;
1315 
1316 	return pnv_pci_cfg_write(pdn, where, size, val);
1317 }
1318 
1319 static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data)
1320 {
1321 	/* GEM */
1322 	if (data->gemXfir || data->gemRfir ||
1323 	    data->gemRirqfir || data->gemMask || data->gemRwof)
1324 		pr_info("  GEM: %016llx %016llx %016llx %016llx %016llx\n",
1325 			be64_to_cpu(data->gemXfir),
1326 			be64_to_cpu(data->gemRfir),
1327 			be64_to_cpu(data->gemRirqfir),
1328 			be64_to_cpu(data->gemMask),
1329 			be64_to_cpu(data->gemRwof));
1330 
1331 	/* LEM */
1332 	if (data->lemFir || data->lemErrMask ||
1333 	    data->lemAction0 || data->lemAction1 || data->lemWof)
1334 		pr_info("  LEM: %016llx %016llx %016llx %016llx %016llx\n",
1335 			be64_to_cpu(data->lemFir),
1336 			be64_to_cpu(data->lemErrMask),
1337 			be64_to_cpu(data->lemAction0),
1338 			be64_to_cpu(data->lemAction1),
1339 			be64_to_cpu(data->lemWof));
1340 }
1341 
1342 static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose)
1343 {
1344 	struct pnv_phb *phb = hose->private_data;
1345 	struct OpalIoP7IOCErrorData *data =
1346 		(struct OpalIoP7IOCErrorData*)phb->diag_data;
1347 	long rc;
1348 
1349 	rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
1350 	if (rc != OPAL_SUCCESS) {
1351 		pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
1352 			__func__, phb->hub_id, rc);
1353 		return;
1354 	}
1355 
1356 	switch (be16_to_cpu(data->type)) {
1357 	case OPAL_P7IOC_DIAG_TYPE_RGC:
1358 		pr_info("P7IOC diag-data for RGC\n\n");
1359 		pnv_eeh_dump_hub_diag_common(data);
1360 		if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
1361 			pr_info("  RGC: %016llx %016llx\n",
1362 				be64_to_cpu(data->rgc.rgcStatus),
1363 				be64_to_cpu(data->rgc.rgcLdcp));
1364 		break;
1365 	case OPAL_P7IOC_DIAG_TYPE_BI:
1366 		pr_info("P7IOC diag-data for BI %s\n\n",
1367 			data->bi.biDownbound ? "Downbound" : "Upbound");
1368 		pnv_eeh_dump_hub_diag_common(data);
1369 		if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
1370 		    data->bi.biLdcp2 || data->bi.biFenceStatus)
1371 			pr_info("  BI:  %016llx %016llx %016llx %016llx\n",
1372 				be64_to_cpu(data->bi.biLdcp0),
1373 				be64_to_cpu(data->bi.biLdcp1),
1374 				be64_to_cpu(data->bi.biLdcp2),
1375 				be64_to_cpu(data->bi.biFenceStatus));
1376 		break;
1377 	case OPAL_P7IOC_DIAG_TYPE_CI:
1378 		pr_info("P7IOC diag-data for CI Port %d\n\n",
1379 			data->ci.ciPort);
1380 		pnv_eeh_dump_hub_diag_common(data);
1381 		if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
1382 			pr_info("  CI:  %016llx %016llx\n",
1383 				be64_to_cpu(data->ci.ciPortStatus),
1384 				be64_to_cpu(data->ci.ciPortLdcp));
1385 		break;
1386 	case OPAL_P7IOC_DIAG_TYPE_MISC:
1387 		pr_info("P7IOC diag-data for MISC\n\n");
1388 		pnv_eeh_dump_hub_diag_common(data);
1389 		break;
1390 	case OPAL_P7IOC_DIAG_TYPE_I2C:
1391 		pr_info("P7IOC diag-data for I2C\n\n");
1392 		pnv_eeh_dump_hub_diag_common(data);
1393 		break;
1394 	default:
1395 		pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
1396 			__func__, phb->hub_id, data->type);
1397 	}
1398 }
1399 
1400 static int pnv_eeh_get_pe(struct pci_controller *hose,
1401 			  u16 pe_no, struct eeh_pe **pe)
1402 {
1403 	struct pnv_phb *phb = hose->private_data;
1404 	struct pnv_ioda_pe *pnv_pe;
1405 	struct eeh_pe *dev_pe;
1406 
1407 	/*
1408 	 * If PHB supports compound PE, to fetch
1409 	 * the master PE because slave PE is invisible
1410 	 * to EEH core.
1411 	 */
1412 	pnv_pe = &phb->ioda.pe_array[pe_no];
1413 	if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
1414 		pnv_pe = pnv_pe->master;
1415 		WARN_ON(!pnv_pe ||
1416 			!(pnv_pe->flags & PNV_IODA_PE_MASTER));
1417 		pe_no = pnv_pe->pe_number;
1418 	}
1419 
1420 	/* Find the PE according to PE# */
1421 	dev_pe = eeh_pe_get(hose, pe_no, 0);
1422 	if (!dev_pe)
1423 		return -EEXIST;
1424 
1425 	/* Freeze the (compound) PE */
1426 	*pe = dev_pe;
1427 	if (!(dev_pe->state & EEH_PE_ISOLATED))
1428 		phb->freeze_pe(phb, pe_no);
1429 
1430 	/*
1431 	 * At this point, we're sure the (compound) PE should
1432 	 * have been frozen. However, we still need poke until
1433 	 * hitting the frozen PE on top level.
1434 	 */
1435 	dev_pe = dev_pe->parent;
1436 	while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
1437 		int ret;
1438 		ret = eeh_ops->get_state(dev_pe, NULL);
1439 		if (ret <= 0 || eeh_state_active(ret)) {
1440 			dev_pe = dev_pe->parent;
1441 			continue;
1442 		}
1443 
1444 		/* Frozen parent PE */
1445 		*pe = dev_pe;
1446 		if (!(dev_pe->state & EEH_PE_ISOLATED))
1447 			phb->freeze_pe(phb, dev_pe->addr);
1448 
1449 		/* Next one */
1450 		dev_pe = dev_pe->parent;
1451 	}
1452 
1453 	return 0;
1454 }
1455 
1456 /**
1457  * pnv_eeh_next_error - Retrieve next EEH error to handle
1458  * @pe: Affected PE
1459  *
1460  * The function is expected to be called by EEH core while it gets
1461  * special EEH event (without binding PE). The function calls to
1462  * OPAL APIs for next error to handle. The informational error is
1463  * handled internally by platform. However, the dead IOC, dead PHB,
1464  * fenced PHB and frozen PE should be handled by EEH core eventually.
1465  */
1466 static int pnv_eeh_next_error(struct eeh_pe **pe)
1467 {
1468 	struct pci_controller *hose;
1469 	struct pnv_phb *phb;
1470 	struct eeh_pe *phb_pe, *parent_pe;
1471 	__be64 frozen_pe_no;
1472 	__be16 err_type, severity;
1473 	long rc;
1474 	int state, ret = EEH_NEXT_ERR_NONE;
1475 
1476 	/*
1477 	 * While running here, it's safe to purge the event queue. The
1478 	 * event should still be masked.
1479 	 */
1480 	eeh_remove_event(NULL, false);
1481 
1482 	list_for_each_entry(hose, &hose_list, list_node) {
1483 		/*
1484 		 * If the subordinate PCI buses of the PHB has been
1485 		 * removed or is exactly under error recovery, we
1486 		 * needn't take care of it any more.
1487 		 */
1488 		phb = hose->private_data;
1489 		phb_pe = eeh_phb_pe_get(hose);
1490 		if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
1491 			continue;
1492 
1493 		rc = opal_pci_next_error(phb->opal_id,
1494 					 &frozen_pe_no, &err_type, &severity);
1495 		if (rc != OPAL_SUCCESS) {
1496 			pr_devel("%s: Invalid return value on "
1497 				 "PHB#%x (0x%lx) from opal_pci_next_error",
1498 				 __func__, hose->global_number, rc);
1499 			continue;
1500 		}
1501 
1502 		/* If the PHB doesn't have error, stop processing */
1503 		if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
1504 		    be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
1505 			pr_devel("%s: No error found on PHB#%x\n",
1506 				 __func__, hose->global_number);
1507 			continue;
1508 		}
1509 
1510 		/*
1511 		 * Processing the error. We're expecting the error with
1512 		 * highest priority reported upon multiple errors on the
1513 		 * specific PHB.
1514 		 */
1515 		pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
1516 			__func__, be16_to_cpu(err_type),
1517 			be16_to_cpu(severity), be64_to_cpu(frozen_pe_no),
1518 			hose->global_number);
1519 		switch (be16_to_cpu(err_type)) {
1520 		case OPAL_EEH_IOC_ERROR:
1521 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
1522 				pr_err("EEH: dead IOC detected\n");
1523 				ret = EEH_NEXT_ERR_DEAD_IOC;
1524 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1525 				pr_info("EEH: IOC informative error "
1526 					"detected\n");
1527 				pnv_eeh_get_and_dump_hub_diag(hose);
1528 				ret = EEH_NEXT_ERR_NONE;
1529 			}
1530 
1531 			break;
1532 		case OPAL_EEH_PHB_ERROR:
1533 			if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
1534 				*pe = phb_pe;
1535 				pr_err("EEH: dead PHB#%x detected, "
1536 				       "location: %s\n",
1537 					hose->global_number,
1538 					eeh_pe_loc_get(phb_pe));
1539 				ret = EEH_NEXT_ERR_DEAD_PHB;
1540 			} else if (be16_to_cpu(severity) ==
1541 				   OPAL_EEH_SEV_PHB_FENCED) {
1542 				*pe = phb_pe;
1543 				pr_err("EEH: Fenced PHB#%x detected, "
1544 				       "location: %s\n",
1545 					hose->global_number,
1546 					eeh_pe_loc_get(phb_pe));
1547 				ret = EEH_NEXT_ERR_FENCED_PHB;
1548 			} else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
1549 				pr_info("EEH: PHB#%x informative error "
1550 					"detected, location: %s\n",
1551 					hose->global_number,
1552 					eeh_pe_loc_get(phb_pe));
1553 				pnv_eeh_get_phb_diag(phb_pe);
1554 				pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
1555 				ret = EEH_NEXT_ERR_NONE;
1556 			}
1557 
1558 			break;
1559 		case OPAL_EEH_PE_ERROR:
1560 			/*
1561 			 * If we can't find the corresponding PE, we
1562 			 * just try to unfreeze.
1563 			 */
1564 			if (pnv_eeh_get_pe(hose,
1565 				be64_to_cpu(frozen_pe_no), pe)) {
1566 				pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1567 					hose->global_number, be64_to_cpu(frozen_pe_no));
1568 				pr_info("EEH: PHB location: %s\n",
1569 					eeh_pe_loc_get(phb_pe));
1570 
1571 				/* Dump PHB diag-data */
1572 				rc = opal_pci_get_phb_diag_data2(phb->opal_id,
1573 					phb->diag_data, phb->diag_data_size);
1574 				if (rc == OPAL_SUCCESS)
1575 					pnv_pci_dump_phb_diag_data(hose,
1576 							phb->diag_data);
1577 
1578 				/* Try best to clear it */
1579 				opal_pci_eeh_freeze_clear(phb->opal_id,
1580 					be64_to_cpu(frozen_pe_no),
1581 					OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1582 				ret = EEH_NEXT_ERR_NONE;
1583 			} else if ((*pe)->state & EEH_PE_ISOLATED ||
1584 				   eeh_pe_passed(*pe)) {
1585 				ret = EEH_NEXT_ERR_NONE;
1586 			} else {
1587 				pr_err("EEH: Frozen PE#%x "
1588 				       "on PHB#%x detected\n",
1589 				       (*pe)->addr,
1590 					(*pe)->phb->global_number);
1591 				pr_err("EEH: PE location: %s, "
1592 				       "PHB location: %s\n",
1593 				       eeh_pe_loc_get(*pe),
1594 				       eeh_pe_loc_get(phb_pe));
1595 				ret = EEH_NEXT_ERR_FROZEN_PE;
1596 			}
1597 
1598 			break;
1599 		default:
1600 			pr_warn("%s: Unexpected error type %d\n",
1601 				__func__, be16_to_cpu(err_type));
1602 		}
1603 
1604 		/*
1605 		 * EEH core will try recover from fenced PHB or
1606 		 * frozen PE. In the time for frozen PE, EEH core
1607 		 * enable IO path for that before collecting logs,
1608 		 * but it ruins the site. So we have to dump the
1609 		 * log in advance here.
1610 		 */
1611 		if ((ret == EEH_NEXT_ERR_FROZEN_PE  ||
1612 		    ret == EEH_NEXT_ERR_FENCED_PHB) &&
1613 		    !((*pe)->state & EEH_PE_ISOLATED)) {
1614 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1615 			pnv_eeh_get_phb_diag(*pe);
1616 
1617 			if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1618 				pnv_pci_dump_phb_diag_data((*pe)->phb,
1619 							   (*pe)->data);
1620 		}
1621 
1622 		/*
1623 		 * We probably have the frozen parent PE out there and
1624 		 * we need have to handle frozen parent PE firstly.
1625 		 */
1626 		if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1627 			parent_pe = (*pe)->parent;
1628 			while (parent_pe) {
1629 				/* Hit the ceiling ? */
1630 				if (parent_pe->type & EEH_PE_PHB)
1631 					break;
1632 
1633 				/* Frozen parent PE ? */
1634 				state = eeh_ops->get_state(parent_pe, NULL);
1635 				if (state > 0 && !eeh_state_active(state))
1636 					*pe = parent_pe;
1637 
1638 				/* Next parent level */
1639 				parent_pe = parent_pe->parent;
1640 			}
1641 
1642 			/* We possibly migrate to another PE */
1643 			eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1644 		}
1645 
1646 		/*
1647 		 * If we have no errors on the specific PHB or only
1648 		 * informative error there, we continue poking it.
1649 		 * Otherwise, we need actions to be taken by upper
1650 		 * layer.
1651 		 */
1652 		if (ret > EEH_NEXT_ERR_INF)
1653 			break;
1654 	}
1655 
1656 	/* Unmask the event */
1657 	if (ret == EEH_NEXT_ERR_NONE && eeh_enabled())
1658 		enable_irq(eeh_event_irq);
1659 
1660 	return ret;
1661 }
1662 
1663 static int pnv_eeh_restore_config(struct pci_dn *pdn)
1664 {
1665 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
1666 	struct pnv_phb *phb;
1667 	s64 ret = 0;
1668 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
1669 
1670 	if (!edev)
1671 		return -EEXIST;
1672 
1673 	/*
1674 	 * We have to restore the PCI config space after reset since the
1675 	 * firmware can't see SRIOV VFs.
1676 	 *
1677 	 * FIXME: The MPS, error routing rules, timeout setting are worthy
1678 	 * to be exported by firmware in extendible way.
1679 	 */
1680 	if (edev->physfn) {
1681 		ret = eeh_restore_vf_config(pdn);
1682 	} else {
1683 		phb = pdn->phb->private_data;
1684 		ret = opal_pci_reinit(phb->opal_id,
1685 				      OPAL_REINIT_PCI_DEV, config_addr);
1686 	}
1687 
1688 	if (ret) {
1689 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
1690 			__func__, config_addr, ret);
1691 		return -EIO;
1692 	}
1693 
1694 	return ret;
1695 }
1696 
1697 static struct eeh_ops pnv_eeh_ops = {
1698 	.name                   = "powernv",
1699 	.init                   = pnv_eeh_init,
1700 	.probe			= pnv_eeh_probe,
1701 	.set_option             = pnv_eeh_set_option,
1702 	.get_pe_addr            = pnv_eeh_get_pe_addr,
1703 	.get_state              = pnv_eeh_get_state,
1704 	.reset                  = pnv_eeh_reset,
1705 	.wait_state             = pnv_eeh_wait_state,
1706 	.get_log                = pnv_eeh_get_log,
1707 	.configure_bridge       = pnv_eeh_configure_bridge,
1708 	.err_inject		= pnv_eeh_err_inject,
1709 	.read_config            = pnv_eeh_read_config,
1710 	.write_config           = pnv_eeh_write_config,
1711 	.next_error		= pnv_eeh_next_error,
1712 	.restore_config		= pnv_eeh_restore_config,
1713 	.notify_resume		= NULL
1714 };
1715 
1716 #ifdef CONFIG_PCI_IOV
1717 static void pnv_pci_fixup_vf_mps(struct pci_dev *pdev)
1718 {
1719 	struct pci_dn *pdn = pci_get_pdn(pdev);
1720 	int parent_mps;
1721 
1722 	if (!pdev->is_virtfn)
1723 		return;
1724 
1725 	/* Synchronize MPS for VF and PF */
1726 	parent_mps = pcie_get_mps(pdev->physfn);
1727 	if ((128 << pdev->pcie_mpss) >= parent_mps)
1728 		pcie_set_mps(pdev, parent_mps);
1729 	pdn->mps = pcie_get_mps(pdev);
1730 }
1731 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pnv_pci_fixup_vf_mps);
1732 #endif /* CONFIG_PCI_IOV */
1733 
1734 /**
1735  * eeh_powernv_init - Register platform dependent EEH operations
1736  *
1737  * EEH initialization on powernv platform. This function should be
1738  * called before any EEH related functions.
1739  */
1740 static int __init eeh_powernv_init(void)
1741 {
1742 	int ret = -EINVAL;
1743 
1744 	ret = eeh_ops_register(&pnv_eeh_ops);
1745 	if (!ret)
1746 		pr_info("EEH: PowerNV platform initialized\n");
1747 	else
1748 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
1749 
1750 	return ret;
1751 }
1752 machine_early_initcall(powernv, eeh_powernv_init);
1753