xref: /linux/arch/powerpc/platforms/powernv/pci.c (revision 0a94608f0f7de9b1135ffea3546afe68eafef57f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Support PCI/PCIe on PowerNV platforms
4  *
5  * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/pci.h>
10 #include <linux/delay.h>
11 #include <linux/string.h>
12 #include <linux/init.h>
13 #include <linux/irq.h>
14 #include <linux/io.h>
15 #include <linux/msi.h>
16 #include <linux/iommu.h>
17 #include <linux/sched/mm.h>
18 
19 #include <asm/sections.h>
20 #include <asm/io.h>
21 #include <asm/prom.h>
22 #include <asm/pci-bridge.h>
23 #include <asm/machdep.h>
24 #include <asm/msi_bitmap.h>
25 #include <asm/ppc-pci.h>
26 #include <asm/pnv-pci.h>
27 #include <asm/opal.h>
28 #include <asm/iommu.h>
29 #include <asm/tce.h>
30 #include <asm/firmware.h>
31 #include <asm/eeh_event.h>
32 #include <asm/eeh.h>
33 
34 #include "powernv.h"
35 #include "pci.h"
36 
37 static DEFINE_MUTEX(tunnel_mutex);
38 
39 int pnv_pci_get_slot_id(struct device_node *np, uint64_t *id)
40 {
41 	struct device_node *node = np;
42 	u32 bdfn;
43 	u64 phbid;
44 	int ret;
45 
46 	ret = of_property_read_u32(np, "reg", &bdfn);
47 	if (ret)
48 		return -ENXIO;
49 
50 	bdfn = ((bdfn & 0x00ffff00) >> 8);
51 	for (node = np; node; node = of_get_parent(node)) {
52 		if (!PCI_DN(node)) {
53 			of_node_put(node);
54 			break;
55 		}
56 
57 		if (!of_device_is_compatible(node, "ibm,ioda2-phb") &&
58 		    !of_device_is_compatible(node, "ibm,ioda3-phb") &&
59 		    !of_device_is_compatible(node, "ibm,ioda2-npu2-opencapi-phb")) {
60 			of_node_put(node);
61 			continue;
62 		}
63 
64 		ret = of_property_read_u64(node, "ibm,opal-phbid", &phbid);
65 		if (ret) {
66 			of_node_put(node);
67 			return -ENXIO;
68 		}
69 
70 		if (of_device_is_compatible(node, "ibm,ioda2-npu2-opencapi-phb"))
71 			*id = PCI_PHB_SLOT_ID(phbid);
72 		else
73 			*id = PCI_SLOT_ID(phbid, bdfn);
74 		return 0;
75 	}
76 
77 	return -ENODEV;
78 }
79 EXPORT_SYMBOL_GPL(pnv_pci_get_slot_id);
80 
81 int pnv_pci_get_device_tree(uint32_t phandle, void *buf, uint64_t len)
82 {
83 	int64_t rc;
84 
85 	if (!opal_check_token(OPAL_GET_DEVICE_TREE))
86 		return -ENXIO;
87 
88 	rc = opal_get_device_tree(phandle, (uint64_t)buf, len);
89 	if (rc < OPAL_SUCCESS)
90 		return -EIO;
91 
92 	return rc;
93 }
94 EXPORT_SYMBOL_GPL(pnv_pci_get_device_tree);
95 
96 int pnv_pci_get_presence_state(uint64_t id, uint8_t *state)
97 {
98 	int64_t rc;
99 
100 	if (!opal_check_token(OPAL_PCI_GET_PRESENCE_STATE))
101 		return -ENXIO;
102 
103 	rc = opal_pci_get_presence_state(id, (uint64_t)state);
104 	if (rc != OPAL_SUCCESS)
105 		return -EIO;
106 
107 	return 0;
108 }
109 EXPORT_SYMBOL_GPL(pnv_pci_get_presence_state);
110 
111 int pnv_pci_get_power_state(uint64_t id, uint8_t *state)
112 {
113 	int64_t rc;
114 
115 	if (!opal_check_token(OPAL_PCI_GET_POWER_STATE))
116 		return -ENXIO;
117 
118 	rc = opal_pci_get_power_state(id, (uint64_t)state);
119 	if (rc != OPAL_SUCCESS)
120 		return -EIO;
121 
122 	return 0;
123 }
124 EXPORT_SYMBOL_GPL(pnv_pci_get_power_state);
125 
126 int pnv_pci_set_power_state(uint64_t id, uint8_t state, struct opal_msg *msg)
127 {
128 	struct opal_msg m;
129 	int token, ret;
130 	int64_t rc;
131 
132 	if (!opal_check_token(OPAL_PCI_SET_POWER_STATE))
133 		return -ENXIO;
134 
135 	token = opal_async_get_token_interruptible();
136 	if (unlikely(token < 0))
137 		return token;
138 
139 	rc = opal_pci_set_power_state(token, id, (uint64_t)&state);
140 	if (rc == OPAL_SUCCESS) {
141 		ret = 0;
142 		goto exit;
143 	} else if (rc != OPAL_ASYNC_COMPLETION) {
144 		ret = -EIO;
145 		goto exit;
146 	}
147 
148 	ret = opal_async_wait_response(token, &m);
149 	if (ret < 0)
150 		goto exit;
151 
152 	if (msg) {
153 		ret = 1;
154 		memcpy(msg, &m, sizeof(m));
155 	}
156 
157 exit:
158 	opal_async_release_token(token);
159 	return ret;
160 }
161 EXPORT_SYMBOL_GPL(pnv_pci_set_power_state);
162 
163 /* Nicely print the contents of the PE State Tables (PEST). */
164 static void pnv_pci_dump_pest(__be64 pestA[], __be64 pestB[], int pest_size)
165 {
166 	__be64 prevA = ULONG_MAX, prevB = ULONG_MAX;
167 	bool dup = false;
168 	int i;
169 
170 	for (i = 0; i < pest_size; i++) {
171 		__be64 peA = be64_to_cpu(pestA[i]);
172 		__be64 peB = be64_to_cpu(pestB[i]);
173 
174 		if (peA != prevA || peB != prevB) {
175 			if (dup) {
176 				pr_info("PE[..%03x] A/B: as above\n", i-1);
177 				dup = false;
178 			}
179 			prevA = peA;
180 			prevB = peB;
181 			if (peA & PNV_IODA_STOPPED_STATE ||
182 			    peB & PNV_IODA_STOPPED_STATE)
183 				pr_info("PE[%03x] A/B: %016llx %016llx\n",
184 					i, peA, peB);
185 		} else if (!dup && (peA & PNV_IODA_STOPPED_STATE ||
186 				    peB & PNV_IODA_STOPPED_STATE)) {
187 			dup = true;
188 		}
189 	}
190 }
191 
192 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
193 					 struct OpalIoPhbErrorCommon *common)
194 {
195 	struct OpalIoP7IOCPhbErrorData *data;
196 
197 	data = (struct OpalIoP7IOCPhbErrorData *)common;
198 	pr_info("P7IOC PHB#%x Diag-data (Version: %d)\n",
199 		hose->global_number, be32_to_cpu(common->version));
200 
201 	if (data->brdgCtl)
202 		pr_info("brdgCtl:     %08x\n",
203 			be32_to_cpu(data->brdgCtl));
204 	if (data->portStatusReg || data->rootCmplxStatus ||
205 	    data->busAgentStatus)
206 		pr_info("UtlSts:      %08x %08x %08x\n",
207 			be32_to_cpu(data->portStatusReg),
208 			be32_to_cpu(data->rootCmplxStatus),
209 			be32_to_cpu(data->busAgentStatus));
210 	if (data->deviceStatus || data->slotStatus   ||
211 	    data->linkStatus   || data->devCmdStatus ||
212 	    data->devSecStatus)
213 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
214 			be32_to_cpu(data->deviceStatus),
215 			be32_to_cpu(data->slotStatus),
216 			be32_to_cpu(data->linkStatus),
217 			be32_to_cpu(data->devCmdStatus),
218 			be32_to_cpu(data->devSecStatus));
219 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
220 	    data->corrErrorStatus)
221 		pr_info("RootErrSts:  %08x %08x %08x\n",
222 			be32_to_cpu(data->rootErrorStatus),
223 			be32_to_cpu(data->uncorrErrorStatus),
224 			be32_to_cpu(data->corrErrorStatus));
225 	if (data->tlpHdr1 || data->tlpHdr2 ||
226 	    data->tlpHdr3 || data->tlpHdr4)
227 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
228 			be32_to_cpu(data->tlpHdr1),
229 			be32_to_cpu(data->tlpHdr2),
230 			be32_to_cpu(data->tlpHdr3),
231 			be32_to_cpu(data->tlpHdr4));
232 	if (data->sourceId || data->errorClass ||
233 	    data->correlator)
234 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
235 			be32_to_cpu(data->sourceId),
236 			be64_to_cpu(data->errorClass),
237 			be64_to_cpu(data->correlator));
238 	if (data->p7iocPlssr || data->p7iocCsr)
239 		pr_info("PhbSts:      %016llx %016llx\n",
240 			be64_to_cpu(data->p7iocPlssr),
241 			be64_to_cpu(data->p7iocCsr));
242 	if (data->lemFir)
243 		pr_info("Lem:         %016llx %016llx %016llx\n",
244 			be64_to_cpu(data->lemFir),
245 			be64_to_cpu(data->lemErrorMask),
246 			be64_to_cpu(data->lemWOF));
247 	if (data->phbErrorStatus)
248 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
249 			be64_to_cpu(data->phbErrorStatus),
250 			be64_to_cpu(data->phbFirstErrorStatus),
251 			be64_to_cpu(data->phbErrorLog0),
252 			be64_to_cpu(data->phbErrorLog1));
253 	if (data->mmioErrorStatus)
254 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
255 			be64_to_cpu(data->mmioErrorStatus),
256 			be64_to_cpu(data->mmioFirstErrorStatus),
257 			be64_to_cpu(data->mmioErrorLog0),
258 			be64_to_cpu(data->mmioErrorLog1));
259 	if (data->dma0ErrorStatus)
260 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
261 			be64_to_cpu(data->dma0ErrorStatus),
262 			be64_to_cpu(data->dma0FirstErrorStatus),
263 			be64_to_cpu(data->dma0ErrorLog0),
264 			be64_to_cpu(data->dma0ErrorLog1));
265 	if (data->dma1ErrorStatus)
266 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
267 			be64_to_cpu(data->dma1ErrorStatus),
268 			be64_to_cpu(data->dma1FirstErrorStatus),
269 			be64_to_cpu(data->dma1ErrorLog0),
270 			be64_to_cpu(data->dma1ErrorLog1));
271 
272 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_P7IOC_NUM_PEST_REGS);
273 }
274 
275 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
276 					struct OpalIoPhbErrorCommon *common)
277 {
278 	struct OpalIoPhb3ErrorData *data;
279 
280 	data = (struct OpalIoPhb3ErrorData*)common;
281 	pr_info("PHB3 PHB#%x Diag-data (Version: %d)\n",
282 		hose->global_number, be32_to_cpu(common->version));
283 	if (data->brdgCtl)
284 		pr_info("brdgCtl:     %08x\n",
285 			be32_to_cpu(data->brdgCtl));
286 	if (data->portStatusReg || data->rootCmplxStatus ||
287 	    data->busAgentStatus)
288 		pr_info("UtlSts:      %08x %08x %08x\n",
289 			be32_to_cpu(data->portStatusReg),
290 			be32_to_cpu(data->rootCmplxStatus),
291 			be32_to_cpu(data->busAgentStatus));
292 	if (data->deviceStatus || data->slotStatus   ||
293 	    data->linkStatus   || data->devCmdStatus ||
294 	    data->devSecStatus)
295 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
296 			be32_to_cpu(data->deviceStatus),
297 			be32_to_cpu(data->slotStatus),
298 			be32_to_cpu(data->linkStatus),
299 			be32_to_cpu(data->devCmdStatus),
300 			be32_to_cpu(data->devSecStatus));
301 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
302 	    data->corrErrorStatus)
303 		pr_info("RootErrSts:  %08x %08x %08x\n",
304 			be32_to_cpu(data->rootErrorStatus),
305 			be32_to_cpu(data->uncorrErrorStatus),
306 			be32_to_cpu(data->corrErrorStatus));
307 	if (data->tlpHdr1 || data->tlpHdr2 ||
308 	    data->tlpHdr3 || data->tlpHdr4)
309 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
310 			be32_to_cpu(data->tlpHdr1),
311 			be32_to_cpu(data->tlpHdr2),
312 			be32_to_cpu(data->tlpHdr3),
313 			be32_to_cpu(data->tlpHdr4));
314 	if (data->sourceId || data->errorClass ||
315 	    data->correlator)
316 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
317 			be32_to_cpu(data->sourceId),
318 			be64_to_cpu(data->errorClass),
319 			be64_to_cpu(data->correlator));
320 	if (data->nFir)
321 		pr_info("nFir:        %016llx %016llx %016llx\n",
322 			be64_to_cpu(data->nFir),
323 			be64_to_cpu(data->nFirMask),
324 			be64_to_cpu(data->nFirWOF));
325 	if (data->phbPlssr || data->phbCsr)
326 		pr_info("PhbSts:      %016llx %016llx\n",
327 			be64_to_cpu(data->phbPlssr),
328 			be64_to_cpu(data->phbCsr));
329 	if (data->lemFir)
330 		pr_info("Lem:         %016llx %016llx %016llx\n",
331 			be64_to_cpu(data->lemFir),
332 			be64_to_cpu(data->lemErrorMask),
333 			be64_to_cpu(data->lemWOF));
334 	if (data->phbErrorStatus)
335 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
336 			be64_to_cpu(data->phbErrorStatus),
337 			be64_to_cpu(data->phbFirstErrorStatus),
338 			be64_to_cpu(data->phbErrorLog0),
339 			be64_to_cpu(data->phbErrorLog1));
340 	if (data->mmioErrorStatus)
341 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
342 			be64_to_cpu(data->mmioErrorStatus),
343 			be64_to_cpu(data->mmioFirstErrorStatus),
344 			be64_to_cpu(data->mmioErrorLog0),
345 			be64_to_cpu(data->mmioErrorLog1));
346 	if (data->dma0ErrorStatus)
347 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
348 			be64_to_cpu(data->dma0ErrorStatus),
349 			be64_to_cpu(data->dma0FirstErrorStatus),
350 			be64_to_cpu(data->dma0ErrorLog0),
351 			be64_to_cpu(data->dma0ErrorLog1));
352 	if (data->dma1ErrorStatus)
353 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
354 			be64_to_cpu(data->dma1ErrorStatus),
355 			be64_to_cpu(data->dma1FirstErrorStatus),
356 			be64_to_cpu(data->dma1ErrorLog0),
357 			be64_to_cpu(data->dma1ErrorLog1));
358 
359 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB3_NUM_PEST_REGS);
360 }
361 
362 static void pnv_pci_dump_phb4_diag_data(struct pci_controller *hose,
363 					struct OpalIoPhbErrorCommon *common)
364 {
365 	struct OpalIoPhb4ErrorData *data;
366 
367 	data = (struct OpalIoPhb4ErrorData*)common;
368 	pr_info("PHB4 PHB#%d Diag-data (Version: %d)\n",
369 		hose->global_number, be32_to_cpu(common->version));
370 	if (data->brdgCtl)
371 		pr_info("brdgCtl:    %08x\n",
372 			be32_to_cpu(data->brdgCtl));
373 	if (data->deviceStatus || data->slotStatus   ||
374 	    data->linkStatus   || data->devCmdStatus ||
375 	    data->devSecStatus)
376 		pr_info("RootSts:    %08x %08x %08x %08x %08x\n",
377 			be32_to_cpu(data->deviceStatus),
378 			be32_to_cpu(data->slotStatus),
379 			be32_to_cpu(data->linkStatus),
380 			be32_to_cpu(data->devCmdStatus),
381 			be32_to_cpu(data->devSecStatus));
382 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
383 	    data->corrErrorStatus)
384 		pr_info("RootErrSts: %08x %08x %08x\n",
385 			be32_to_cpu(data->rootErrorStatus),
386 			be32_to_cpu(data->uncorrErrorStatus),
387 			be32_to_cpu(data->corrErrorStatus));
388 	if (data->tlpHdr1 || data->tlpHdr2 ||
389 	    data->tlpHdr3 || data->tlpHdr4)
390 		pr_info("RootErrLog: %08x %08x %08x %08x\n",
391 			be32_to_cpu(data->tlpHdr1),
392 			be32_to_cpu(data->tlpHdr2),
393 			be32_to_cpu(data->tlpHdr3),
394 			be32_to_cpu(data->tlpHdr4));
395 	if (data->sourceId)
396 		pr_info("sourceId:   %08x\n", be32_to_cpu(data->sourceId));
397 	if (data->nFir)
398 		pr_info("nFir:       %016llx %016llx %016llx\n",
399 			be64_to_cpu(data->nFir),
400 			be64_to_cpu(data->nFirMask),
401 			be64_to_cpu(data->nFirWOF));
402 	if (data->phbPlssr || data->phbCsr)
403 		pr_info("PhbSts:     %016llx %016llx\n",
404 			be64_to_cpu(data->phbPlssr),
405 			be64_to_cpu(data->phbCsr));
406 	if (data->lemFir)
407 		pr_info("Lem:        %016llx %016llx %016llx\n",
408 			be64_to_cpu(data->lemFir),
409 			be64_to_cpu(data->lemErrorMask),
410 			be64_to_cpu(data->lemWOF));
411 	if (data->phbErrorStatus)
412 		pr_info("PhbErr:     %016llx %016llx %016llx %016llx\n",
413 			be64_to_cpu(data->phbErrorStatus),
414 			be64_to_cpu(data->phbFirstErrorStatus),
415 			be64_to_cpu(data->phbErrorLog0),
416 			be64_to_cpu(data->phbErrorLog1));
417 	if (data->phbTxeErrorStatus)
418 		pr_info("PhbTxeErr:  %016llx %016llx %016llx %016llx\n",
419 			be64_to_cpu(data->phbTxeErrorStatus),
420 			be64_to_cpu(data->phbTxeFirstErrorStatus),
421 			be64_to_cpu(data->phbTxeErrorLog0),
422 			be64_to_cpu(data->phbTxeErrorLog1));
423 	if (data->phbRxeArbErrorStatus)
424 		pr_info("RxeArbErr:  %016llx %016llx %016llx %016llx\n",
425 			be64_to_cpu(data->phbRxeArbErrorStatus),
426 			be64_to_cpu(data->phbRxeArbFirstErrorStatus),
427 			be64_to_cpu(data->phbRxeArbErrorLog0),
428 			be64_to_cpu(data->phbRxeArbErrorLog1));
429 	if (data->phbRxeMrgErrorStatus)
430 		pr_info("RxeMrgErr:  %016llx %016llx %016llx %016llx\n",
431 			be64_to_cpu(data->phbRxeMrgErrorStatus),
432 			be64_to_cpu(data->phbRxeMrgFirstErrorStatus),
433 			be64_to_cpu(data->phbRxeMrgErrorLog0),
434 			be64_to_cpu(data->phbRxeMrgErrorLog1));
435 	if (data->phbRxeTceErrorStatus)
436 		pr_info("RxeTceErr:  %016llx %016llx %016llx %016llx\n",
437 			be64_to_cpu(data->phbRxeTceErrorStatus),
438 			be64_to_cpu(data->phbRxeTceFirstErrorStatus),
439 			be64_to_cpu(data->phbRxeTceErrorLog0),
440 			be64_to_cpu(data->phbRxeTceErrorLog1));
441 
442 	if (data->phbPblErrorStatus)
443 		pr_info("PblErr:     %016llx %016llx %016llx %016llx\n",
444 			be64_to_cpu(data->phbPblErrorStatus),
445 			be64_to_cpu(data->phbPblFirstErrorStatus),
446 			be64_to_cpu(data->phbPblErrorLog0),
447 			be64_to_cpu(data->phbPblErrorLog1));
448 	if (data->phbPcieDlpErrorStatus)
449 		pr_info("PcieDlp:    %016llx %016llx %016llx\n",
450 			be64_to_cpu(data->phbPcieDlpErrorLog1),
451 			be64_to_cpu(data->phbPcieDlpErrorLog2),
452 			be64_to_cpu(data->phbPcieDlpErrorStatus));
453 	if (data->phbRegbErrorStatus)
454 		pr_info("RegbErr:    %016llx %016llx %016llx %016llx\n",
455 			be64_to_cpu(data->phbRegbErrorStatus),
456 			be64_to_cpu(data->phbRegbFirstErrorStatus),
457 			be64_to_cpu(data->phbRegbErrorLog0),
458 			be64_to_cpu(data->phbRegbErrorLog1));
459 
460 
461 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB4_NUM_PEST_REGS);
462 }
463 
464 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
465 				unsigned char *log_buff)
466 {
467 	struct OpalIoPhbErrorCommon *common;
468 
469 	if (!hose || !log_buff)
470 		return;
471 
472 	common = (struct OpalIoPhbErrorCommon *)log_buff;
473 	switch (be32_to_cpu(common->ioType)) {
474 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
475 		pnv_pci_dump_p7ioc_diag_data(hose, common);
476 		break;
477 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
478 		pnv_pci_dump_phb3_diag_data(hose, common);
479 		break;
480 	case OPAL_PHB_ERROR_DATA_TYPE_PHB4:
481 		pnv_pci_dump_phb4_diag_data(hose, common);
482 		break;
483 	default:
484 		pr_warn("%s: Unrecognized ioType %d\n",
485 			__func__, be32_to_cpu(common->ioType));
486 	}
487 }
488 
489 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
490 {
491 	unsigned long flags, rc;
492 	int has_diag, ret = 0;
493 
494 	spin_lock_irqsave(&phb->lock, flags);
495 
496 	/* Fetch PHB diag-data */
497 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag_data,
498 					 phb->diag_data_size);
499 	has_diag = (rc == OPAL_SUCCESS);
500 
501 	/* If PHB supports compound PE, to handle it */
502 	if (phb->unfreeze_pe) {
503 		ret = phb->unfreeze_pe(phb,
504 				       pe_no,
505 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
506 	} else {
507 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
508 					     pe_no,
509 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
510 		if (rc) {
511 			pr_warn("%s: Failure %ld clearing frozen "
512 				"PHB#%x-PE#%x\n",
513 				__func__, rc, phb->hose->global_number,
514 				pe_no);
515 			ret = -EIO;
516 		}
517 	}
518 
519 	/*
520 	 * For now, let's only display the diag buffer when we fail to clear
521 	 * the EEH status. We'll do more sensible things later when we have
522 	 * proper EEH support. We need to make sure we don't pollute ourselves
523 	 * with the normal errors generated when probing empty slots
524 	 */
525 	if (has_diag && ret)
526 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag_data);
527 
528 	spin_unlock_irqrestore(&phb->lock, flags);
529 }
530 
531 static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
532 {
533 	struct pnv_phb *phb = pdn->phb->private_data;
534 	u8	fstate = 0;
535 	__be16	pcierr = 0;
536 	unsigned int pe_no;
537 	s64	rc;
538 
539 	/*
540 	 * Get the PE#. During the PCI probe stage, we might not
541 	 * setup that yet. So all ER errors should be mapped to
542 	 * reserved PE.
543 	 */
544 	pe_no = pdn->pe_number;
545 	if (pe_no == IODA_INVALID_PE) {
546 		pe_no = phb->ioda.reserved_pe_idx;
547 	}
548 
549 	/*
550 	 * Fetch frozen state. If the PHB support compound PE,
551 	 * we need handle that case.
552 	 */
553 	if (phb->get_pe_state) {
554 		fstate = phb->get_pe_state(phb, pe_no);
555 	} else {
556 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
557 						pe_no,
558 						&fstate,
559 						&pcierr,
560 						NULL);
561 		if (rc) {
562 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
563 				__func__, rc, phb->hose->global_number, pe_no);
564 			return;
565 		}
566 	}
567 
568 	pr_devel(" -> EEH check, bdfn=%04x PE#%x fstate=%x\n",
569 		 (pdn->busno << 8) | (pdn->devfn), pe_no, fstate);
570 
571 	/* Clear the frozen state if applicable */
572 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
573 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
574 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
575 		/*
576 		 * If PHB supports compound PE, freeze it for
577 		 * consistency.
578 		 */
579 		if (phb->freeze_pe)
580 			phb->freeze_pe(phb, pe_no);
581 
582 		pnv_pci_handle_eeh_config(phb, pe_no);
583 	}
584 }
585 
586 int pnv_pci_cfg_read(struct pci_dn *pdn,
587 		     int where, int size, u32 *val)
588 {
589 	struct pnv_phb *phb = pdn->phb->private_data;
590 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
591 	s64 rc;
592 
593 	switch (size) {
594 	case 1: {
595 		u8 v8;
596 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
597 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
598 		break;
599 	}
600 	case 2: {
601 		__be16 v16;
602 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
603 						   &v16);
604 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
605 		break;
606 	}
607 	case 4: {
608 		__be32 v32;
609 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
610 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
611 		break;
612 	}
613 	default:
614 		return PCIBIOS_FUNC_NOT_SUPPORTED;
615 	}
616 
617 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
618 		 __func__, pdn->busno, pdn->devfn, where, size, *val);
619 	return PCIBIOS_SUCCESSFUL;
620 }
621 
622 int pnv_pci_cfg_write(struct pci_dn *pdn,
623 		      int where, int size, u32 val)
624 {
625 	struct pnv_phb *phb = pdn->phb->private_data;
626 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
627 
628 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
629 		 __func__, pdn->busno, pdn->devfn, where, size, val);
630 	switch (size) {
631 	case 1:
632 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
633 		break;
634 	case 2:
635 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
636 		break;
637 	case 4:
638 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
639 		break;
640 	default:
641 		return PCIBIOS_FUNC_NOT_SUPPORTED;
642 	}
643 
644 	return PCIBIOS_SUCCESSFUL;
645 }
646 
647 #ifdef CONFIG_EEH
648 static bool pnv_pci_cfg_check(struct pci_dn *pdn)
649 {
650 	struct eeh_dev *edev = NULL;
651 	struct pnv_phb *phb = pdn->phb->private_data;
652 
653 	/* EEH not enabled ? */
654 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
655 		return true;
656 
657 	/* PE reset or device removed ? */
658 	edev = pdn->edev;
659 	if (edev) {
660 		if (edev->pe &&
661 		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
662 			return false;
663 
664 		if (edev->mode & EEH_DEV_REMOVED)
665 			return false;
666 	}
667 
668 	return true;
669 }
670 #else
671 static inline pnv_pci_cfg_check(struct pci_dn *pdn)
672 {
673 	return true;
674 }
675 #endif /* CONFIG_EEH */
676 
677 static int pnv_pci_read_config(struct pci_bus *bus,
678 			       unsigned int devfn,
679 			       int where, int size, u32 *val)
680 {
681 	struct pci_dn *pdn;
682 	struct pnv_phb *phb;
683 	int ret;
684 
685 	*val = 0xFFFFFFFF;
686 	pdn = pci_get_pdn_by_devfn(bus, devfn);
687 	if (!pdn)
688 		return PCIBIOS_DEVICE_NOT_FOUND;
689 
690 	if (!pnv_pci_cfg_check(pdn))
691 		return PCIBIOS_DEVICE_NOT_FOUND;
692 
693 	ret = pnv_pci_cfg_read(pdn, where, size, val);
694 	phb = pdn->phb->private_data;
695 	if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
696 		if (*val == EEH_IO_ERROR_VALUE(size) &&
697 		    eeh_dev_check_failure(pdn->edev))
698                         return PCIBIOS_DEVICE_NOT_FOUND;
699 	} else {
700 		pnv_pci_config_check_eeh(pdn);
701 	}
702 
703 	return ret;
704 }
705 
706 static int pnv_pci_write_config(struct pci_bus *bus,
707 				unsigned int devfn,
708 				int where, int size, u32 val)
709 {
710 	struct pci_dn *pdn;
711 	struct pnv_phb *phb;
712 	int ret;
713 
714 	pdn = pci_get_pdn_by_devfn(bus, devfn);
715 	if (!pdn)
716 		return PCIBIOS_DEVICE_NOT_FOUND;
717 
718 	if (!pnv_pci_cfg_check(pdn))
719 		return PCIBIOS_DEVICE_NOT_FOUND;
720 
721 	ret = pnv_pci_cfg_write(pdn, where, size, val);
722 	phb = pdn->phb->private_data;
723 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
724 		pnv_pci_config_check_eeh(pdn);
725 
726 	return ret;
727 }
728 
729 struct pci_ops pnv_pci_ops = {
730 	.read  = pnv_pci_read_config,
731 	.write = pnv_pci_write_config,
732 };
733 
734 struct iommu_table *pnv_pci_table_alloc(int nid)
735 {
736 	struct iommu_table *tbl;
737 
738 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
739 	if (!tbl)
740 		return NULL;
741 
742 	INIT_LIST_HEAD_RCU(&tbl->it_group_list);
743 	kref_init(&tbl->it_kref);
744 
745 	return tbl;
746 }
747 
748 struct device_node *pnv_pci_get_phb_node(struct pci_dev *dev)
749 {
750 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
751 
752 	return of_node_get(hose->dn);
753 }
754 EXPORT_SYMBOL(pnv_pci_get_phb_node);
755 
756 int pnv_pci_set_tunnel_bar(struct pci_dev *dev, u64 addr, int enable)
757 {
758 	struct pnv_phb *phb = pci_bus_to_pnvhb(dev->bus);
759 	u64 tunnel_bar;
760 	__be64 val;
761 	int rc;
762 
763 	if (!opal_check_token(OPAL_PCI_GET_PBCQ_TUNNEL_BAR))
764 		return -ENXIO;
765 	if (!opal_check_token(OPAL_PCI_SET_PBCQ_TUNNEL_BAR))
766 		return -ENXIO;
767 
768 	mutex_lock(&tunnel_mutex);
769 	rc = opal_pci_get_pbcq_tunnel_bar(phb->opal_id, &val);
770 	if (rc != OPAL_SUCCESS) {
771 		rc = -EIO;
772 		goto out;
773 	}
774 	tunnel_bar = be64_to_cpu(val);
775 	if (enable) {
776 		/*
777 		* Only one device per PHB can use atomics.
778 		* Our policy is first-come, first-served.
779 		*/
780 		if (tunnel_bar) {
781 			if (tunnel_bar != addr)
782 				rc = -EBUSY;
783 			else
784 				rc = 0;	/* Setting same address twice is ok */
785 			goto out;
786 		}
787 	} else {
788 		/*
789 		* The device that owns atomics and wants to release
790 		* them must pass the same address with enable == 0.
791 		*/
792 		if (tunnel_bar != addr) {
793 			rc = -EPERM;
794 			goto out;
795 		}
796 		addr = 0x0ULL;
797 	}
798 	rc = opal_pci_set_pbcq_tunnel_bar(phb->opal_id, addr);
799 	rc = opal_error_code(rc);
800 out:
801 	mutex_unlock(&tunnel_mutex);
802 	return rc;
803 }
804 EXPORT_SYMBOL_GPL(pnv_pci_set_tunnel_bar);
805 
806 void pnv_pci_shutdown(void)
807 {
808 	struct pci_controller *hose;
809 
810 	list_for_each_entry(hose, &hose_list, list_node)
811 		if (hose->controller_ops.shutdown)
812 			hose->controller_ops.shutdown(hose);
813 }
814 
815 /* Fixup wrong class code in p7ioc and p8 root complex */
816 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
817 {
818 	dev->class = PCI_CLASS_BRIDGE_PCI_NORMAL;
819 }
820 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
821 
822 void __init pnv_pci_init(void)
823 {
824 	struct device_node *np;
825 
826 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
827 
828 	/* If we don't have OPAL, eg. in sim, just skip PCI probe */
829 	if (!firmware_has_feature(FW_FEATURE_OPAL))
830 		return;
831 
832 #ifdef CONFIG_PCIEPORTBUS
833 	/*
834 	 * On PowerNV PCIe devices are (currently) managed in cooperation
835 	 * with firmware. This isn't *strictly* required, but there's enough
836 	 * assumptions baked into both firmware and the platform code that
837 	 * it's unwise to allow the portbus services to be used.
838 	 *
839 	 * We need to fix this eventually, but for now set this flag to disable
840 	 * the portbus driver. The AER service isn't required since that AER
841 	 * events are handled via EEH. The pciehp hotplug driver can't work
842 	 * without kernel changes (and portbus binding breaks pnv_php). The
843 	 * other services also require some thinking about how we're going
844 	 * to integrate them.
845 	 */
846 	pcie_ports_disabled = true;
847 #endif
848 
849 	/* Look for IODA IO-Hubs. */
850 	for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
851 		pnv_pci_init_ioda_hub(np);
852 	}
853 
854 	/* Look for ioda2 built-in PHB3's */
855 	for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
856 		pnv_pci_init_ioda2_phb(np);
857 
858 	/* Look for ioda3 built-in PHB4's, we treat them as IODA2 */
859 	for_each_compatible_node(np, NULL, "ibm,ioda3-phb")
860 		pnv_pci_init_ioda2_phb(np);
861 
862 	/* Look for NPU2 OpenCAPI PHBs */
863 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-opencapi-phb")
864 		pnv_pci_init_npu2_opencapi_phb(np);
865 
866 	/* Configure IOMMU DMA hooks */
867 	set_pci_dma_ops(&dma_iommu_ops);
868 }
869 
870 static int pnv_tce_iommu_bus_notifier(struct notifier_block *nb,
871 		unsigned long action, void *data)
872 {
873 	struct device *dev = data;
874 
875 	switch (action) {
876 	case BUS_NOTIFY_DEL_DEVICE:
877 		iommu_del_device(dev);
878 		return 0;
879 	default:
880 		return 0;
881 	}
882 }
883 
884 static struct notifier_block pnv_tce_iommu_bus_nb = {
885 	.notifier_call = pnv_tce_iommu_bus_notifier,
886 };
887 
888 static int __init pnv_tce_iommu_bus_notifier_init(void)
889 {
890 	bus_register_notifier(&pci_bus_type, &pnv_tce_iommu_bus_nb);
891 	return 0;
892 }
893 machine_subsys_initcall_sync(powernv, pnv_tce_iommu_bus_notifier_init);
894