1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * The file intends to implement the platform dependent EEH operations on pseries.
4  * Actually, the pseries platform is built based on RTAS heavily. That means the
5  * pseries platform dependent EEH operations will be built on RTAS calls. The functions
6  * are derived from arch/powerpc/platforms/pseries/eeh.c and necessary cleanup has
7  * been done.
8  *
9  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2011.
10  * Copyright IBM Corporation 2001, 2005, 2006
11  * Copyright Dave Engebretsen & Todd Inglett 2001
12  * Copyright Linas Vepstas 2005, 2006
13  */
14 
15 #include <linux/atomic.h>
16 #include <linux/delay.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/list.h>
20 #include <linux/of.h>
21 #include <linux/pci.h>
22 #include <linux/proc_fs.h>
23 #include <linux/rbtree.h>
24 #include <linux/sched.h>
25 #include <linux/seq_file.h>
26 #include <linux/spinlock.h>
27 
28 #include <asm/eeh.h>
29 #include <asm/eeh_event.h>
30 #include <asm/io.h>
31 #include <asm/machdep.h>
32 #include <asm/ppc-pci.h>
33 #include <asm/rtas.h>
34 
35 /* RTAS tokens */
36 static int ibm_set_eeh_option;
37 static int ibm_set_slot_reset;
38 static int ibm_read_slot_reset_state;
39 static int ibm_read_slot_reset_state2;
40 static int ibm_slot_error_detail;
41 static int ibm_get_config_addr_info;
42 static int ibm_get_config_addr_info2;
43 static int ibm_configure_pe;
44 
45 #ifdef CONFIG_PCI_IOV
46 void pseries_pcibios_bus_add_device(struct pci_dev *pdev)
47 {
48 	struct pci_dn *pdn = pci_get_pdn(pdev);
49 	struct pci_dn *physfn_pdn;
50 	struct eeh_dev *edev;
51 
52 	if (!pdev->is_virtfn)
53 		return;
54 
55 	pdn->device_id  =  pdev->device;
56 	pdn->vendor_id  =  pdev->vendor;
57 	pdn->class_code =  pdev->class;
58 	/*
59 	 * Last allow unfreeze return code used for retrieval
60 	 * by user space in eeh-sysfs to show the last command
61 	 * completion from platform.
62 	 */
63 	pdn->last_allow_rc =  0;
64 	physfn_pdn      =  pci_get_pdn(pdev->physfn);
65 	pdn->pe_number  =  physfn_pdn->pe_num_map[pdn->vf_index];
66 	edev = pdn_to_eeh_dev(pdn);
67 
68 	/*
69 	 * The following operations will fail if VF's sysfs files
70 	 * aren't created or its resources aren't finalized.
71 	 */
72 	eeh_add_device_early(pdn);
73 	eeh_add_device_late(pdev);
74 	edev->pe_config_addr =  (pdn->busno << 16) | (pdn->devfn << 8);
75 	eeh_rmv_from_parent_pe(edev); /* Remove as it is adding to bus pe */
76 	eeh_add_to_parent_pe(edev);   /* Add as VF PE type */
77 	eeh_sysfs_add_device(pdev);
78 
79 }
80 #endif
81 
82 /*
83  * Buffer for reporting slot-error-detail rtas calls. Its here
84  * in BSS, and not dynamically alloced, so that it ends up in
85  * RMO where RTAS can access it.
86  */
87 static unsigned char slot_errbuf[RTAS_ERROR_LOG_MAX];
88 static DEFINE_SPINLOCK(slot_errbuf_lock);
89 static int eeh_error_buf_size;
90 
91 /**
92  * pseries_eeh_init - EEH platform dependent initialization
93  *
94  * EEH platform dependent initialization on pseries.
95  */
96 static int pseries_eeh_init(void)
97 {
98 	/* figure out EEH RTAS function call tokens */
99 	ibm_set_eeh_option		= rtas_token("ibm,set-eeh-option");
100 	ibm_set_slot_reset		= rtas_token("ibm,set-slot-reset");
101 	ibm_read_slot_reset_state2	= rtas_token("ibm,read-slot-reset-state2");
102 	ibm_read_slot_reset_state	= rtas_token("ibm,read-slot-reset-state");
103 	ibm_slot_error_detail		= rtas_token("ibm,slot-error-detail");
104 	ibm_get_config_addr_info2	= rtas_token("ibm,get-config-addr-info2");
105 	ibm_get_config_addr_info	= rtas_token("ibm,get-config-addr-info");
106 	ibm_configure_pe		= rtas_token("ibm,configure-pe");
107 
108 	/*
109 	 * ibm,configure-pe and ibm,configure-bridge have the same semantics,
110 	 * however ibm,configure-pe can be faster.  If we can't find
111 	 * ibm,configure-pe then fall back to using ibm,configure-bridge.
112 	 */
113 	if (ibm_configure_pe == RTAS_UNKNOWN_SERVICE)
114 		ibm_configure_pe 	= rtas_token("ibm,configure-bridge");
115 
116 	/*
117 	 * Necessary sanity check. We needn't check "get-config-addr-info"
118 	 * and its variant since the old firmware probably support address
119 	 * of domain/bus/slot/function for EEH RTAS operations.
120 	 */
121 	if (ibm_set_eeh_option == RTAS_UNKNOWN_SERVICE		||
122 	    ibm_set_slot_reset == RTAS_UNKNOWN_SERVICE		||
123 	    (ibm_read_slot_reset_state2 == RTAS_UNKNOWN_SERVICE &&
124 	     ibm_read_slot_reset_state == RTAS_UNKNOWN_SERVICE)	||
125 	    ibm_slot_error_detail == RTAS_UNKNOWN_SERVICE	||
126 	    ibm_configure_pe == RTAS_UNKNOWN_SERVICE) {
127 		pr_info("EEH functionality not supported\n");
128 		return -EINVAL;
129 	}
130 
131 	/* Initialize error log lock and size */
132 	spin_lock_init(&slot_errbuf_lock);
133 	eeh_error_buf_size = rtas_token("rtas-error-log-max");
134 	if (eeh_error_buf_size == RTAS_UNKNOWN_SERVICE) {
135 		pr_info("%s: unknown EEH error log size\n",
136 			__func__);
137 		eeh_error_buf_size = 1024;
138 	} else if (eeh_error_buf_size > RTAS_ERROR_LOG_MAX) {
139 		pr_info("%s: EEH error log size %d exceeds the maximal %d\n",
140 			__func__, eeh_error_buf_size, RTAS_ERROR_LOG_MAX);
141 		eeh_error_buf_size = RTAS_ERROR_LOG_MAX;
142 	}
143 
144 	/* Set EEH probe mode */
145 	eeh_add_flag(EEH_PROBE_MODE_DEVTREE | EEH_ENABLE_IO_FOR_LOG);
146 
147 #ifdef CONFIG_PCI_IOV
148 	/* Set EEH machine dependent code */
149 	ppc_md.pcibios_bus_add_device = pseries_pcibios_bus_add_device;
150 #endif
151 
152 	return 0;
153 }
154 
155 static int pseries_eeh_cap_start(struct pci_dn *pdn)
156 {
157 	u32 status;
158 
159 	if (!pdn)
160 		return 0;
161 
162 	rtas_read_config(pdn, PCI_STATUS, 2, &status);
163 	if (!(status & PCI_STATUS_CAP_LIST))
164 		return 0;
165 
166 	return PCI_CAPABILITY_LIST;
167 }
168 
169 
170 static int pseries_eeh_find_cap(struct pci_dn *pdn, int cap)
171 {
172 	int pos = pseries_eeh_cap_start(pdn);
173 	int cnt = 48;	/* Maximal number of capabilities */
174 	u32 id;
175 
176 	if (!pos)
177 		return 0;
178 
179         while (cnt--) {
180 		rtas_read_config(pdn, pos, 1, &pos);
181 		if (pos < 0x40)
182 			break;
183 		pos &= ~3;
184 		rtas_read_config(pdn, pos + PCI_CAP_LIST_ID, 1, &id);
185 		if (id == 0xff)
186 			break;
187 		if (id == cap)
188 			return pos;
189 		pos += PCI_CAP_LIST_NEXT;
190 	}
191 
192 	return 0;
193 }
194 
195 static int pseries_eeh_find_ecap(struct pci_dn *pdn, int cap)
196 {
197 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
198 	u32 header;
199 	int pos = 256;
200 	int ttl = (4096 - 256) / 8;
201 
202 	if (!edev || !edev->pcie_cap)
203 		return 0;
204 	if (rtas_read_config(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
205 		return 0;
206 	else if (!header)
207 		return 0;
208 
209 	while (ttl-- > 0) {
210 		if (PCI_EXT_CAP_ID(header) == cap && pos)
211 			return pos;
212 
213 		pos = PCI_EXT_CAP_NEXT(header);
214 		if (pos < 256)
215 			break;
216 
217 		if (rtas_read_config(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL)
218 			break;
219 	}
220 
221 	return 0;
222 }
223 
224 /**
225  * pseries_eeh_probe - EEH probe on the given device
226  * @pdn: PCI device node
227  * @data: Unused
228  *
229  * When EEH module is installed during system boot, all PCI devices
230  * are checked one by one to see if it supports EEH. The function
231  * is introduced for the purpose.
232  */
233 static void *pseries_eeh_probe(struct pci_dn *pdn, void *data)
234 {
235 	struct eeh_dev *edev;
236 	struct eeh_pe pe;
237 	u32 pcie_flags;
238 	int enable = 0;
239 	int ret;
240 
241 	/* Retrieve OF node and eeh device */
242 	edev = pdn_to_eeh_dev(pdn);
243 	if (!edev || edev->pe)
244 		return NULL;
245 
246 	/* Check class/vendor/device IDs */
247 	if (!pdn->vendor_id || !pdn->device_id || !pdn->class_code)
248 		return NULL;
249 
250 	/* Skip for PCI-ISA bridge */
251         if ((pdn->class_code >> 8) == PCI_CLASS_BRIDGE_ISA)
252 		return NULL;
253 
254 	/*
255 	 * Update class code and mode of eeh device. We need
256 	 * correctly reflects that current device is root port
257 	 * or PCIe switch downstream port.
258 	 */
259 	edev->class_code = pdn->class_code;
260 	edev->pcix_cap = pseries_eeh_find_cap(pdn, PCI_CAP_ID_PCIX);
261 	edev->pcie_cap = pseries_eeh_find_cap(pdn, PCI_CAP_ID_EXP);
262 	edev->aer_cap = pseries_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR);
263 	edev->mode &= 0xFFFFFF00;
264 	if ((edev->class_code >> 8) == PCI_CLASS_BRIDGE_PCI) {
265 		edev->mode |= EEH_DEV_BRIDGE;
266 		if (edev->pcie_cap) {
267 			rtas_read_config(pdn, edev->pcie_cap + PCI_EXP_FLAGS,
268 					 2, &pcie_flags);
269 			pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4;
270 			if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT)
271 				edev->mode |= EEH_DEV_ROOT_PORT;
272 			else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM)
273 				edev->mode |= EEH_DEV_DS_PORT;
274 		}
275 	}
276 
277 	/* Initialize the fake PE */
278 	memset(&pe, 0, sizeof(struct eeh_pe));
279 	pe.phb = pdn->phb;
280 	pe.config_addr = (pdn->busno << 16) | (pdn->devfn << 8);
281 
282 	/* Enable EEH on the device */
283 	ret = eeh_ops->set_option(&pe, EEH_OPT_ENABLE);
284 	if (!ret) {
285 		/* Retrieve PE address */
286 		edev->pe_config_addr = eeh_ops->get_pe_addr(&pe);
287 		pe.addr = edev->pe_config_addr;
288 
289 		/* Some older systems (Power4) allow the ibm,set-eeh-option
290 		 * call to succeed even on nodes where EEH is not supported.
291 		 * Verify support explicitly.
292 		 */
293 		ret = eeh_ops->get_state(&pe, NULL);
294 		if (ret > 0 && ret != EEH_STATE_NOT_SUPPORT)
295 			enable = 1;
296 
297 		if (enable) {
298 			eeh_add_flag(EEH_ENABLED);
299 			eeh_add_to_parent_pe(edev);
300 
301 			pr_debug("%s: EEH enabled on %02x:%02x.%01x PHB#%x-PE#%x\n",
302 				__func__, pdn->busno, PCI_SLOT(pdn->devfn),
303 				PCI_FUNC(pdn->devfn), pe.phb->global_number,
304 				pe.addr);
305 		} else if (pdn->parent && pdn_to_eeh_dev(pdn->parent) &&
306 			   (pdn_to_eeh_dev(pdn->parent))->pe) {
307 			/* This device doesn't support EEH, but it may have an
308 			 * EEH parent, in which case we mark it as supported.
309 			 */
310 			edev->pe_config_addr = pdn_to_eeh_dev(pdn->parent)->pe_config_addr;
311 			eeh_add_to_parent_pe(edev);
312 		}
313 	}
314 
315 	/* Save memory bars */
316 	eeh_save_bars(edev);
317 
318 	return NULL;
319 }
320 
321 /**
322  * pseries_eeh_set_option - Initialize EEH or MMIO/DMA reenable
323  * @pe: EEH PE
324  * @option: operation to be issued
325  *
326  * The function is used to control the EEH functionality globally.
327  * Currently, following options are support according to PAPR:
328  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
329  */
330 static int pseries_eeh_set_option(struct eeh_pe *pe, int option)
331 {
332 	int ret = 0;
333 	int config_addr;
334 
335 	/*
336 	 * When we're enabling or disabling EEH functioality on
337 	 * the particular PE, the PE config address is possibly
338 	 * unavailable. Therefore, we have to figure it out from
339 	 * the FDT node.
340 	 */
341 	switch (option) {
342 	case EEH_OPT_DISABLE:
343 	case EEH_OPT_ENABLE:
344 	case EEH_OPT_THAW_MMIO:
345 	case EEH_OPT_THAW_DMA:
346 		config_addr = pe->config_addr;
347 		if (pe->addr)
348 			config_addr = pe->addr;
349 		break;
350 	case EEH_OPT_FREEZE_PE:
351 		/* Not support */
352 		return 0;
353 	default:
354 		pr_err("%s: Invalid option %d\n",
355 			__func__, option);
356 		return -EINVAL;
357 	}
358 
359 	ret = rtas_call(ibm_set_eeh_option, 4, 1, NULL,
360 			config_addr, BUID_HI(pe->phb->buid),
361 			BUID_LO(pe->phb->buid), option);
362 
363 	return ret;
364 }
365 
366 /**
367  * pseries_eeh_get_pe_addr - Retrieve PE address
368  * @pe: EEH PE
369  *
370  * Retrieve the assocated PE address. Actually, there're 2 RTAS
371  * function calls dedicated for the purpose. We need implement
372  * it through the new function and then the old one. Besides,
373  * you should make sure the config address is figured out from
374  * FDT node before calling the function.
375  *
376  * It's notable that zero'ed return value means invalid PE config
377  * address.
378  */
379 static int pseries_eeh_get_pe_addr(struct eeh_pe *pe)
380 {
381 	int ret = 0;
382 	int rets[3];
383 
384 	if (ibm_get_config_addr_info2 != RTAS_UNKNOWN_SERVICE) {
385 		/*
386 		 * First of all, we need to make sure there has one PE
387 		 * associated with the device. Otherwise, PE address is
388 		 * meaningless.
389 		 */
390 		ret = rtas_call(ibm_get_config_addr_info2, 4, 2, rets,
391 				pe->config_addr, BUID_HI(pe->phb->buid),
392 				BUID_LO(pe->phb->buid), 1);
393 		if (ret || (rets[0] == 0))
394 			return 0;
395 
396 		/* Retrieve the associated PE config address */
397 		ret = rtas_call(ibm_get_config_addr_info2, 4, 2, rets,
398 				pe->config_addr, BUID_HI(pe->phb->buid),
399 				BUID_LO(pe->phb->buid), 0);
400 		if (ret) {
401 			pr_warn("%s: Failed to get address for PHB#%x-PE#%x\n",
402 				__func__, pe->phb->global_number, pe->config_addr);
403 			return 0;
404 		}
405 
406 		return rets[0];
407 	}
408 
409 	if (ibm_get_config_addr_info != RTAS_UNKNOWN_SERVICE) {
410 		ret = rtas_call(ibm_get_config_addr_info, 4, 2, rets,
411 				pe->config_addr, BUID_HI(pe->phb->buid),
412 				BUID_LO(pe->phb->buid), 0);
413 		if (ret) {
414 			pr_warn("%s: Failed to get address for PHB#%x-PE#%x\n",
415 				__func__, pe->phb->global_number, pe->config_addr);
416 			return 0;
417 		}
418 
419 		return rets[0];
420 	}
421 
422 	return ret;
423 }
424 
425 /**
426  * pseries_eeh_get_state - Retrieve PE state
427  * @pe: EEH PE
428  * @delay: suggested time to wait if state is unavailable
429  *
430  * Retrieve the state of the specified PE. On RTAS compliant
431  * pseries platform, there already has one dedicated RTAS function
432  * for the purpose. It's notable that the associated PE config address
433  * might be ready when calling the function. Therefore, endeavour to
434  * use the PE config address if possible. Further more, there're 2
435  * RTAS calls for the purpose, we need to try the new one and back
436  * to the old one if the new one couldn't work properly.
437  */
438 static int pseries_eeh_get_state(struct eeh_pe *pe, int *delay)
439 {
440 	int config_addr;
441 	int ret;
442 	int rets[4];
443 	int result;
444 
445 	/* Figure out PE config address if possible */
446 	config_addr = pe->config_addr;
447 	if (pe->addr)
448 		config_addr = pe->addr;
449 
450 	if (ibm_read_slot_reset_state2 != RTAS_UNKNOWN_SERVICE) {
451 		ret = rtas_call(ibm_read_slot_reset_state2, 3, 4, rets,
452 				config_addr, BUID_HI(pe->phb->buid),
453 				BUID_LO(pe->phb->buid));
454 	} else if (ibm_read_slot_reset_state != RTAS_UNKNOWN_SERVICE) {
455 		/* Fake PE unavailable info */
456 		rets[2] = 0;
457 		ret = rtas_call(ibm_read_slot_reset_state, 3, 3, rets,
458 				config_addr, BUID_HI(pe->phb->buid),
459 				BUID_LO(pe->phb->buid));
460 	} else {
461 		return EEH_STATE_NOT_SUPPORT;
462 	}
463 
464 	if (ret)
465 		return ret;
466 
467 	/* Parse the result out */
468 	if (!rets[1])
469 		return EEH_STATE_NOT_SUPPORT;
470 
471 	switch(rets[0]) {
472 	case 0:
473 		result = EEH_STATE_MMIO_ACTIVE |
474 			 EEH_STATE_DMA_ACTIVE;
475 		break;
476 	case 1:
477 		result = EEH_STATE_RESET_ACTIVE |
478 			 EEH_STATE_MMIO_ACTIVE  |
479 			 EEH_STATE_DMA_ACTIVE;
480 		break;
481 	case 2:
482 		result = 0;
483 		break;
484 	case 4:
485 		result = EEH_STATE_MMIO_ENABLED;
486 		break;
487 	case 5:
488 		if (rets[2]) {
489 			if (delay)
490 				*delay = rets[2];
491 			result = EEH_STATE_UNAVAILABLE;
492 		} else {
493 			result = EEH_STATE_NOT_SUPPORT;
494 		}
495 		break;
496 	default:
497 		result = EEH_STATE_NOT_SUPPORT;
498 	}
499 
500 	return result;
501 }
502 
503 /**
504  * pseries_eeh_reset - Reset the specified PE
505  * @pe: EEH PE
506  * @option: reset option
507  *
508  * Reset the specified PE
509  */
510 static int pseries_eeh_reset(struct eeh_pe *pe, int option)
511 {
512 	int config_addr;
513 	int ret;
514 
515 	/* Figure out PE address */
516 	config_addr = pe->config_addr;
517 	if (pe->addr)
518 		config_addr = pe->addr;
519 
520 	/* Reset PE through RTAS call */
521 	ret = rtas_call(ibm_set_slot_reset, 4, 1, NULL,
522 			config_addr, BUID_HI(pe->phb->buid),
523 			BUID_LO(pe->phb->buid), option);
524 
525 	/* If fundamental-reset not supported, try hot-reset */
526 	if (option == EEH_RESET_FUNDAMENTAL &&
527 	    ret == -8) {
528 		option = EEH_RESET_HOT;
529 		ret = rtas_call(ibm_set_slot_reset, 4, 1, NULL,
530 				config_addr, BUID_HI(pe->phb->buid),
531 				BUID_LO(pe->phb->buid), option);
532 	}
533 
534 	/* We need reset hold or settlement delay */
535 	if (option == EEH_RESET_FUNDAMENTAL ||
536 	    option == EEH_RESET_HOT)
537 		msleep(EEH_PE_RST_HOLD_TIME);
538 	else
539 		msleep(EEH_PE_RST_SETTLE_TIME);
540 
541 	return ret;
542 }
543 
544 /**
545  * pseries_eeh_get_log - Retrieve error log
546  * @pe: EEH PE
547  * @severity: temporary or permanent error log
548  * @drv_log: driver log to be combined with retrieved error log
549  * @len: length of driver log
550  *
551  * Retrieve the temporary or permanent error from the PE.
552  * Actually, the error will be retrieved through the dedicated
553  * RTAS call.
554  */
555 static int pseries_eeh_get_log(struct eeh_pe *pe, int severity, char *drv_log, unsigned long len)
556 {
557 	int config_addr;
558 	unsigned long flags;
559 	int ret;
560 
561 	spin_lock_irqsave(&slot_errbuf_lock, flags);
562 	memset(slot_errbuf, 0, eeh_error_buf_size);
563 
564 	/* Figure out the PE address */
565 	config_addr = pe->config_addr;
566 	if (pe->addr)
567 		config_addr = pe->addr;
568 
569 	ret = rtas_call(ibm_slot_error_detail, 8, 1, NULL, config_addr,
570 			BUID_HI(pe->phb->buid), BUID_LO(pe->phb->buid),
571 			virt_to_phys(drv_log), len,
572 			virt_to_phys(slot_errbuf), eeh_error_buf_size,
573 			severity);
574 	if (!ret)
575 		log_error(slot_errbuf, ERR_TYPE_RTAS_LOG, 0);
576 	spin_unlock_irqrestore(&slot_errbuf_lock, flags);
577 
578 	return ret;
579 }
580 
581 /**
582  * pseries_eeh_configure_bridge - Configure PCI bridges in the indicated PE
583  * @pe: EEH PE
584  *
585  * The function will be called to reconfigure the bridges included
586  * in the specified PE so that the mulfunctional PE would be recovered
587  * again.
588  */
589 static int pseries_eeh_configure_bridge(struct eeh_pe *pe)
590 {
591 	int config_addr;
592 	int ret;
593 	/* Waiting 0.2s maximum before skipping configuration */
594 	int max_wait = 200;
595 
596 	/* Figure out the PE address */
597 	config_addr = pe->config_addr;
598 	if (pe->addr)
599 		config_addr = pe->addr;
600 
601 	while (max_wait > 0) {
602 		ret = rtas_call(ibm_configure_pe, 3, 1, NULL,
603 				config_addr, BUID_HI(pe->phb->buid),
604 				BUID_LO(pe->phb->buid));
605 
606 		if (!ret)
607 			return ret;
608 
609 		/*
610 		 * If RTAS returns a delay value that's above 100ms, cut it
611 		 * down to 100ms in case firmware made a mistake.  For more
612 		 * on how these delay values work see rtas_busy_delay_time
613 		 */
614 		if (ret > RTAS_EXTENDED_DELAY_MIN+2 &&
615 		    ret <= RTAS_EXTENDED_DELAY_MAX)
616 			ret = RTAS_EXTENDED_DELAY_MIN+2;
617 
618 		max_wait -= rtas_busy_delay_time(ret);
619 
620 		if (max_wait < 0)
621 			break;
622 
623 		rtas_busy_delay(ret);
624 	}
625 
626 	pr_warn("%s: Unable to configure bridge PHB#%x-PE#%x (%d)\n",
627 		__func__, pe->phb->global_number, pe->addr, ret);
628 	return ret;
629 }
630 
631 /**
632  * pseries_eeh_read_config - Read PCI config space
633  * @pdn: PCI device node
634  * @where: PCI address
635  * @size: size to read
636  * @val: return value
637  *
638  * Read config space from the speicifed device
639  */
640 static int pseries_eeh_read_config(struct pci_dn *pdn, int where, int size, u32 *val)
641 {
642 	return rtas_read_config(pdn, where, size, val);
643 }
644 
645 /**
646  * pseries_eeh_write_config - Write PCI config space
647  * @pdn: PCI device node
648  * @where: PCI address
649  * @size: size to write
650  * @val: value to be written
651  *
652  * Write config space to the specified device
653  */
654 static int pseries_eeh_write_config(struct pci_dn *pdn, int where, int size, u32 val)
655 {
656 	return rtas_write_config(pdn, where, size, val);
657 }
658 
659 static int pseries_eeh_restore_config(struct pci_dn *pdn)
660 {
661 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
662 	s64 ret = 0;
663 
664 	if (!edev)
665 		return -EEXIST;
666 
667 	/*
668 	 * FIXME: The MPS, error routing rules, timeout setting are worthy
669 	 * to be exported by firmware in extendible way.
670 	 */
671 	if (edev->physfn)
672 		ret = eeh_restore_vf_config(pdn);
673 
674 	if (ret) {
675 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
676 			__func__, edev->pe_config_addr, ret);
677 		return -EIO;
678 	}
679 
680 	return ret;
681 }
682 
683 #ifdef CONFIG_PCI_IOV
684 int pseries_send_allow_unfreeze(struct pci_dn *pdn,
685 				u16 *vf_pe_array, int cur_vfs)
686 {
687 	int rc;
688 	int ibm_allow_unfreeze = rtas_token("ibm,open-sriov-allow-unfreeze");
689 	unsigned long buid, addr;
690 
691 	addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
692 	buid = pdn->phb->buid;
693 	spin_lock(&rtas_data_buf_lock);
694 	memcpy(rtas_data_buf, vf_pe_array, RTAS_DATA_BUF_SIZE);
695 	rc = rtas_call(ibm_allow_unfreeze, 5, 1, NULL,
696 		       addr,
697 		       BUID_HI(buid),
698 		       BUID_LO(buid),
699 		       rtas_data_buf, cur_vfs * sizeof(u16));
700 	spin_unlock(&rtas_data_buf_lock);
701 	if (rc)
702 		pr_warn("%s: Failed to allow unfreeze for PHB#%x-PE#%lx, rc=%x\n",
703 			__func__,
704 			pdn->phb->global_number, addr, rc);
705 	return rc;
706 }
707 
708 static int pseries_call_allow_unfreeze(struct eeh_dev *edev)
709 {
710 	struct pci_dn *pdn, *tmp, *parent, *physfn_pdn;
711 	int cur_vfs = 0, rc = 0, vf_index, bus, devfn;
712 	u16 *vf_pe_array;
713 
714 	vf_pe_array = kzalloc(RTAS_DATA_BUF_SIZE, GFP_KERNEL);
715 	if (!vf_pe_array)
716 		return -ENOMEM;
717 	if (pci_num_vf(edev->physfn ? edev->physfn : edev->pdev)) {
718 		if (edev->pdev->is_physfn) {
719 			cur_vfs = pci_num_vf(edev->pdev);
720 			pdn = eeh_dev_to_pdn(edev);
721 			parent = pdn->parent;
722 			for (vf_index = 0; vf_index < cur_vfs; vf_index++)
723 				vf_pe_array[vf_index] =
724 					cpu_to_be16(pdn->pe_num_map[vf_index]);
725 			rc = pseries_send_allow_unfreeze(pdn, vf_pe_array,
726 							 cur_vfs);
727 			pdn->last_allow_rc = rc;
728 			for (vf_index = 0; vf_index < cur_vfs; vf_index++) {
729 				list_for_each_entry_safe(pdn, tmp,
730 							 &parent->child_list,
731 							 list) {
732 					bus = pci_iov_virtfn_bus(edev->pdev,
733 								 vf_index);
734 					devfn = pci_iov_virtfn_devfn(edev->pdev,
735 								     vf_index);
736 					if (pdn->busno != bus ||
737 					    pdn->devfn != devfn)
738 						continue;
739 					pdn->last_allow_rc = rc;
740 				}
741 			}
742 		} else {
743 			pdn = pci_get_pdn(edev->pdev);
744 			vf_pe_array[0] = cpu_to_be16(pdn->pe_number);
745 			physfn_pdn = pci_get_pdn(edev->physfn);
746 			rc = pseries_send_allow_unfreeze(physfn_pdn,
747 							 vf_pe_array, 1);
748 			pdn->last_allow_rc = rc;
749 		}
750 	}
751 
752 	kfree(vf_pe_array);
753 	return rc;
754 }
755 
756 static int pseries_notify_resume(struct pci_dn *pdn)
757 {
758 	struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
759 
760 	if (!edev)
761 		return -EEXIST;
762 
763 	if (rtas_token("ibm,open-sriov-allow-unfreeze")
764 	    == RTAS_UNKNOWN_SERVICE)
765 		return -EINVAL;
766 
767 	if (edev->pdev->is_physfn || edev->pdev->is_virtfn)
768 		return pseries_call_allow_unfreeze(edev);
769 
770 	return 0;
771 }
772 #endif
773 
774 static struct eeh_ops pseries_eeh_ops = {
775 	.name			= "pseries",
776 	.init			= pseries_eeh_init,
777 	.probe			= pseries_eeh_probe,
778 	.set_option		= pseries_eeh_set_option,
779 	.get_pe_addr		= pseries_eeh_get_pe_addr,
780 	.get_state		= pseries_eeh_get_state,
781 	.reset			= pseries_eeh_reset,
782 	.get_log		= pseries_eeh_get_log,
783 	.configure_bridge       = pseries_eeh_configure_bridge,
784 	.err_inject		= NULL,
785 	.read_config		= pseries_eeh_read_config,
786 	.write_config		= pseries_eeh_write_config,
787 	.next_error		= NULL,
788 	.restore_config		= pseries_eeh_restore_config,
789 #ifdef CONFIG_PCI_IOV
790 	.notify_resume		= pseries_notify_resume
791 #endif
792 };
793 
794 /**
795  * eeh_pseries_init - Register platform dependent EEH operations
796  *
797  * EEH initialization on pseries platform. This function should be
798  * called before any EEH related functions.
799  */
800 static int __init eeh_pseries_init(void)
801 {
802 	int ret;
803 
804 	ret = eeh_ops_register(&pseries_eeh_ops);
805 	if (!ret)
806 		pr_info("EEH: pSeries platform initialized\n");
807 	else
808 		pr_info("EEH: pSeries platform initialization failure (%d)\n",
809 			ret);
810 
811 	return ret;
812 }
813 machine_early_initcall(pseries, eeh_pseries_init);
814