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 int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
164 {
165 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
166 	struct pnv_phb *phb = hose->private_data;
167 	struct msi_desc *entry;
168 	struct msi_msg msg;
169 	int hwirq;
170 	unsigned int virq;
171 	int rc;
172 
173 	if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
174 		return -ENODEV;
175 
176 	if (pdev->no_64bit_msi && !phb->msi32_support)
177 		return -ENODEV;
178 
179 	for_each_pci_msi_entry(entry, pdev) {
180 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
181 			pr_warn("%s: Supports only 64-bit MSIs\n",
182 				pci_name(pdev));
183 			return -ENXIO;
184 		}
185 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
186 		if (hwirq < 0) {
187 			pr_warn("%s: Failed to find a free MSI\n",
188 				pci_name(pdev));
189 			return -ENOSPC;
190 		}
191 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
192 		if (!virq) {
193 			pr_warn("%s: Failed to map MSI to linux irq\n",
194 				pci_name(pdev));
195 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
196 			return -ENOMEM;
197 		}
198 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
199 				    virq, entry->msi_attrib.is_64, &msg);
200 		if (rc) {
201 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
202 			irq_dispose_mapping(virq);
203 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
204 			return rc;
205 		}
206 		irq_set_msi_desc(virq, entry);
207 		pci_write_msi_msg(virq, &msg);
208 	}
209 	return 0;
210 }
211 
212 void pnv_teardown_msi_irqs(struct pci_dev *pdev)
213 {
214 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
215 	struct pnv_phb *phb = hose->private_data;
216 	struct msi_desc *entry;
217 	irq_hw_number_t hwirq;
218 
219 	if (WARN_ON(!phb))
220 		return;
221 
222 	for_each_pci_msi_entry(entry, pdev) {
223 		if (!entry->irq)
224 			continue;
225 		hwirq = virq_to_hw(entry->irq);
226 		irq_set_msi_desc(entry->irq, NULL);
227 		irq_dispose_mapping(entry->irq);
228 		msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
229 	}
230 }
231 
232 /* Nicely print the contents of the PE State Tables (PEST). */
233 static void pnv_pci_dump_pest(__be64 pestA[], __be64 pestB[], int pest_size)
234 {
235 	__be64 prevA = ULONG_MAX, prevB = ULONG_MAX;
236 	bool dup = false;
237 	int i;
238 
239 	for (i = 0; i < pest_size; i++) {
240 		__be64 peA = be64_to_cpu(pestA[i]);
241 		__be64 peB = be64_to_cpu(pestB[i]);
242 
243 		if (peA != prevA || peB != prevB) {
244 			if (dup) {
245 				pr_info("PE[..%03x] A/B: as above\n", i-1);
246 				dup = false;
247 			}
248 			prevA = peA;
249 			prevB = peB;
250 			if (peA & PNV_IODA_STOPPED_STATE ||
251 			    peB & PNV_IODA_STOPPED_STATE)
252 				pr_info("PE[%03x] A/B: %016llx %016llx\n",
253 					i, peA, peB);
254 		} else if (!dup && (peA & PNV_IODA_STOPPED_STATE ||
255 				    peB & PNV_IODA_STOPPED_STATE)) {
256 			dup = true;
257 		}
258 	}
259 }
260 
261 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
262 					 struct OpalIoPhbErrorCommon *common)
263 {
264 	struct OpalIoP7IOCPhbErrorData *data;
265 
266 	data = (struct OpalIoP7IOCPhbErrorData *)common;
267 	pr_info("P7IOC PHB#%x Diag-data (Version: %d)\n",
268 		hose->global_number, be32_to_cpu(common->version));
269 
270 	if (data->brdgCtl)
271 		pr_info("brdgCtl:     %08x\n",
272 			be32_to_cpu(data->brdgCtl));
273 	if (data->portStatusReg || data->rootCmplxStatus ||
274 	    data->busAgentStatus)
275 		pr_info("UtlSts:      %08x %08x %08x\n",
276 			be32_to_cpu(data->portStatusReg),
277 			be32_to_cpu(data->rootCmplxStatus),
278 			be32_to_cpu(data->busAgentStatus));
279 	if (data->deviceStatus || data->slotStatus   ||
280 	    data->linkStatus   || data->devCmdStatus ||
281 	    data->devSecStatus)
282 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
283 			be32_to_cpu(data->deviceStatus),
284 			be32_to_cpu(data->slotStatus),
285 			be32_to_cpu(data->linkStatus),
286 			be32_to_cpu(data->devCmdStatus),
287 			be32_to_cpu(data->devSecStatus));
288 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
289 	    data->corrErrorStatus)
290 		pr_info("RootErrSts:  %08x %08x %08x\n",
291 			be32_to_cpu(data->rootErrorStatus),
292 			be32_to_cpu(data->uncorrErrorStatus),
293 			be32_to_cpu(data->corrErrorStatus));
294 	if (data->tlpHdr1 || data->tlpHdr2 ||
295 	    data->tlpHdr3 || data->tlpHdr4)
296 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
297 			be32_to_cpu(data->tlpHdr1),
298 			be32_to_cpu(data->tlpHdr2),
299 			be32_to_cpu(data->tlpHdr3),
300 			be32_to_cpu(data->tlpHdr4));
301 	if (data->sourceId || data->errorClass ||
302 	    data->correlator)
303 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
304 			be32_to_cpu(data->sourceId),
305 			be64_to_cpu(data->errorClass),
306 			be64_to_cpu(data->correlator));
307 	if (data->p7iocPlssr || data->p7iocCsr)
308 		pr_info("PhbSts:      %016llx %016llx\n",
309 			be64_to_cpu(data->p7iocPlssr),
310 			be64_to_cpu(data->p7iocCsr));
311 	if (data->lemFir)
312 		pr_info("Lem:         %016llx %016llx %016llx\n",
313 			be64_to_cpu(data->lemFir),
314 			be64_to_cpu(data->lemErrorMask),
315 			be64_to_cpu(data->lemWOF));
316 	if (data->phbErrorStatus)
317 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
318 			be64_to_cpu(data->phbErrorStatus),
319 			be64_to_cpu(data->phbFirstErrorStatus),
320 			be64_to_cpu(data->phbErrorLog0),
321 			be64_to_cpu(data->phbErrorLog1));
322 	if (data->mmioErrorStatus)
323 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
324 			be64_to_cpu(data->mmioErrorStatus),
325 			be64_to_cpu(data->mmioFirstErrorStatus),
326 			be64_to_cpu(data->mmioErrorLog0),
327 			be64_to_cpu(data->mmioErrorLog1));
328 	if (data->dma0ErrorStatus)
329 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
330 			be64_to_cpu(data->dma0ErrorStatus),
331 			be64_to_cpu(data->dma0FirstErrorStatus),
332 			be64_to_cpu(data->dma0ErrorLog0),
333 			be64_to_cpu(data->dma0ErrorLog1));
334 	if (data->dma1ErrorStatus)
335 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
336 			be64_to_cpu(data->dma1ErrorStatus),
337 			be64_to_cpu(data->dma1FirstErrorStatus),
338 			be64_to_cpu(data->dma1ErrorLog0),
339 			be64_to_cpu(data->dma1ErrorLog1));
340 
341 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_P7IOC_NUM_PEST_REGS);
342 }
343 
344 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
345 					struct OpalIoPhbErrorCommon *common)
346 {
347 	struct OpalIoPhb3ErrorData *data;
348 
349 	data = (struct OpalIoPhb3ErrorData*)common;
350 	pr_info("PHB3 PHB#%x Diag-data (Version: %d)\n",
351 		hose->global_number, be32_to_cpu(common->version));
352 	if (data->brdgCtl)
353 		pr_info("brdgCtl:     %08x\n",
354 			be32_to_cpu(data->brdgCtl));
355 	if (data->portStatusReg || data->rootCmplxStatus ||
356 	    data->busAgentStatus)
357 		pr_info("UtlSts:      %08x %08x %08x\n",
358 			be32_to_cpu(data->portStatusReg),
359 			be32_to_cpu(data->rootCmplxStatus),
360 			be32_to_cpu(data->busAgentStatus));
361 	if (data->deviceStatus || data->slotStatus   ||
362 	    data->linkStatus   || data->devCmdStatus ||
363 	    data->devSecStatus)
364 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
365 			be32_to_cpu(data->deviceStatus),
366 			be32_to_cpu(data->slotStatus),
367 			be32_to_cpu(data->linkStatus),
368 			be32_to_cpu(data->devCmdStatus),
369 			be32_to_cpu(data->devSecStatus));
370 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
371 	    data->corrErrorStatus)
372 		pr_info("RootErrSts:  %08x %08x %08x\n",
373 			be32_to_cpu(data->rootErrorStatus),
374 			be32_to_cpu(data->uncorrErrorStatus),
375 			be32_to_cpu(data->corrErrorStatus));
376 	if (data->tlpHdr1 || data->tlpHdr2 ||
377 	    data->tlpHdr3 || data->tlpHdr4)
378 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
379 			be32_to_cpu(data->tlpHdr1),
380 			be32_to_cpu(data->tlpHdr2),
381 			be32_to_cpu(data->tlpHdr3),
382 			be32_to_cpu(data->tlpHdr4));
383 	if (data->sourceId || data->errorClass ||
384 	    data->correlator)
385 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
386 			be32_to_cpu(data->sourceId),
387 			be64_to_cpu(data->errorClass),
388 			be64_to_cpu(data->correlator));
389 	if (data->nFir)
390 		pr_info("nFir:        %016llx %016llx %016llx\n",
391 			be64_to_cpu(data->nFir),
392 			be64_to_cpu(data->nFirMask),
393 			be64_to_cpu(data->nFirWOF));
394 	if (data->phbPlssr || data->phbCsr)
395 		pr_info("PhbSts:      %016llx %016llx\n",
396 			be64_to_cpu(data->phbPlssr),
397 			be64_to_cpu(data->phbCsr));
398 	if (data->lemFir)
399 		pr_info("Lem:         %016llx %016llx %016llx\n",
400 			be64_to_cpu(data->lemFir),
401 			be64_to_cpu(data->lemErrorMask),
402 			be64_to_cpu(data->lemWOF));
403 	if (data->phbErrorStatus)
404 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
405 			be64_to_cpu(data->phbErrorStatus),
406 			be64_to_cpu(data->phbFirstErrorStatus),
407 			be64_to_cpu(data->phbErrorLog0),
408 			be64_to_cpu(data->phbErrorLog1));
409 	if (data->mmioErrorStatus)
410 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
411 			be64_to_cpu(data->mmioErrorStatus),
412 			be64_to_cpu(data->mmioFirstErrorStatus),
413 			be64_to_cpu(data->mmioErrorLog0),
414 			be64_to_cpu(data->mmioErrorLog1));
415 	if (data->dma0ErrorStatus)
416 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
417 			be64_to_cpu(data->dma0ErrorStatus),
418 			be64_to_cpu(data->dma0FirstErrorStatus),
419 			be64_to_cpu(data->dma0ErrorLog0),
420 			be64_to_cpu(data->dma0ErrorLog1));
421 	if (data->dma1ErrorStatus)
422 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
423 			be64_to_cpu(data->dma1ErrorStatus),
424 			be64_to_cpu(data->dma1FirstErrorStatus),
425 			be64_to_cpu(data->dma1ErrorLog0),
426 			be64_to_cpu(data->dma1ErrorLog1));
427 
428 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB3_NUM_PEST_REGS);
429 }
430 
431 static void pnv_pci_dump_phb4_diag_data(struct pci_controller *hose,
432 					struct OpalIoPhbErrorCommon *common)
433 {
434 	struct OpalIoPhb4ErrorData *data;
435 
436 	data = (struct OpalIoPhb4ErrorData*)common;
437 	pr_info("PHB4 PHB#%d Diag-data (Version: %d)\n",
438 		hose->global_number, be32_to_cpu(common->version));
439 	if (data->brdgCtl)
440 		pr_info("brdgCtl:    %08x\n",
441 			be32_to_cpu(data->brdgCtl));
442 	if (data->deviceStatus || data->slotStatus   ||
443 	    data->linkStatus   || data->devCmdStatus ||
444 	    data->devSecStatus)
445 		pr_info("RootSts:    %08x %08x %08x %08x %08x\n",
446 			be32_to_cpu(data->deviceStatus),
447 			be32_to_cpu(data->slotStatus),
448 			be32_to_cpu(data->linkStatus),
449 			be32_to_cpu(data->devCmdStatus),
450 			be32_to_cpu(data->devSecStatus));
451 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
452 	    data->corrErrorStatus)
453 		pr_info("RootErrSts: %08x %08x %08x\n",
454 			be32_to_cpu(data->rootErrorStatus),
455 			be32_to_cpu(data->uncorrErrorStatus),
456 			be32_to_cpu(data->corrErrorStatus));
457 	if (data->tlpHdr1 || data->tlpHdr2 ||
458 	    data->tlpHdr3 || data->tlpHdr4)
459 		pr_info("RootErrLog: %08x %08x %08x %08x\n",
460 			be32_to_cpu(data->tlpHdr1),
461 			be32_to_cpu(data->tlpHdr2),
462 			be32_to_cpu(data->tlpHdr3),
463 			be32_to_cpu(data->tlpHdr4));
464 	if (data->sourceId)
465 		pr_info("sourceId:   %08x\n", be32_to_cpu(data->sourceId));
466 	if (data->nFir)
467 		pr_info("nFir:       %016llx %016llx %016llx\n",
468 			be64_to_cpu(data->nFir),
469 			be64_to_cpu(data->nFirMask),
470 			be64_to_cpu(data->nFirWOF));
471 	if (data->phbPlssr || data->phbCsr)
472 		pr_info("PhbSts:     %016llx %016llx\n",
473 			be64_to_cpu(data->phbPlssr),
474 			be64_to_cpu(data->phbCsr));
475 	if (data->lemFir)
476 		pr_info("Lem:        %016llx %016llx %016llx\n",
477 			be64_to_cpu(data->lemFir),
478 			be64_to_cpu(data->lemErrorMask),
479 			be64_to_cpu(data->lemWOF));
480 	if (data->phbErrorStatus)
481 		pr_info("PhbErr:     %016llx %016llx %016llx %016llx\n",
482 			be64_to_cpu(data->phbErrorStatus),
483 			be64_to_cpu(data->phbFirstErrorStatus),
484 			be64_to_cpu(data->phbErrorLog0),
485 			be64_to_cpu(data->phbErrorLog1));
486 	if (data->phbTxeErrorStatus)
487 		pr_info("PhbTxeErr:  %016llx %016llx %016llx %016llx\n",
488 			be64_to_cpu(data->phbTxeErrorStatus),
489 			be64_to_cpu(data->phbTxeFirstErrorStatus),
490 			be64_to_cpu(data->phbTxeErrorLog0),
491 			be64_to_cpu(data->phbTxeErrorLog1));
492 	if (data->phbRxeArbErrorStatus)
493 		pr_info("RxeArbErr:  %016llx %016llx %016llx %016llx\n",
494 			be64_to_cpu(data->phbRxeArbErrorStatus),
495 			be64_to_cpu(data->phbRxeArbFirstErrorStatus),
496 			be64_to_cpu(data->phbRxeArbErrorLog0),
497 			be64_to_cpu(data->phbRxeArbErrorLog1));
498 	if (data->phbRxeMrgErrorStatus)
499 		pr_info("RxeMrgErr:  %016llx %016llx %016llx %016llx\n",
500 			be64_to_cpu(data->phbRxeMrgErrorStatus),
501 			be64_to_cpu(data->phbRxeMrgFirstErrorStatus),
502 			be64_to_cpu(data->phbRxeMrgErrorLog0),
503 			be64_to_cpu(data->phbRxeMrgErrorLog1));
504 	if (data->phbRxeTceErrorStatus)
505 		pr_info("RxeTceErr:  %016llx %016llx %016llx %016llx\n",
506 			be64_to_cpu(data->phbRxeTceErrorStatus),
507 			be64_to_cpu(data->phbRxeTceFirstErrorStatus),
508 			be64_to_cpu(data->phbRxeTceErrorLog0),
509 			be64_to_cpu(data->phbRxeTceErrorLog1));
510 
511 	if (data->phbPblErrorStatus)
512 		pr_info("PblErr:     %016llx %016llx %016llx %016llx\n",
513 			be64_to_cpu(data->phbPblErrorStatus),
514 			be64_to_cpu(data->phbPblFirstErrorStatus),
515 			be64_to_cpu(data->phbPblErrorLog0),
516 			be64_to_cpu(data->phbPblErrorLog1));
517 	if (data->phbPcieDlpErrorStatus)
518 		pr_info("PcieDlp:    %016llx %016llx %016llx\n",
519 			be64_to_cpu(data->phbPcieDlpErrorLog1),
520 			be64_to_cpu(data->phbPcieDlpErrorLog2),
521 			be64_to_cpu(data->phbPcieDlpErrorStatus));
522 	if (data->phbRegbErrorStatus)
523 		pr_info("RegbErr:    %016llx %016llx %016llx %016llx\n",
524 			be64_to_cpu(data->phbRegbErrorStatus),
525 			be64_to_cpu(data->phbRegbFirstErrorStatus),
526 			be64_to_cpu(data->phbRegbErrorLog0),
527 			be64_to_cpu(data->phbRegbErrorLog1));
528 
529 
530 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB4_NUM_PEST_REGS);
531 }
532 
533 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
534 				unsigned char *log_buff)
535 {
536 	struct OpalIoPhbErrorCommon *common;
537 
538 	if (!hose || !log_buff)
539 		return;
540 
541 	common = (struct OpalIoPhbErrorCommon *)log_buff;
542 	switch (be32_to_cpu(common->ioType)) {
543 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
544 		pnv_pci_dump_p7ioc_diag_data(hose, common);
545 		break;
546 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
547 		pnv_pci_dump_phb3_diag_data(hose, common);
548 		break;
549 	case OPAL_PHB_ERROR_DATA_TYPE_PHB4:
550 		pnv_pci_dump_phb4_diag_data(hose, common);
551 		break;
552 	default:
553 		pr_warn("%s: Unrecognized ioType %d\n",
554 			__func__, be32_to_cpu(common->ioType));
555 	}
556 }
557 
558 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
559 {
560 	unsigned long flags, rc;
561 	int has_diag, ret = 0;
562 
563 	spin_lock_irqsave(&phb->lock, flags);
564 
565 	/* Fetch PHB diag-data */
566 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag_data,
567 					 phb->diag_data_size);
568 	has_diag = (rc == OPAL_SUCCESS);
569 
570 	/* If PHB supports compound PE, to handle it */
571 	if (phb->unfreeze_pe) {
572 		ret = phb->unfreeze_pe(phb,
573 				       pe_no,
574 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
575 	} else {
576 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
577 					     pe_no,
578 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
579 		if (rc) {
580 			pr_warn("%s: Failure %ld clearing frozen "
581 				"PHB#%x-PE#%x\n",
582 				__func__, rc, phb->hose->global_number,
583 				pe_no);
584 			ret = -EIO;
585 		}
586 	}
587 
588 	/*
589 	 * For now, let's only display the diag buffer when we fail to clear
590 	 * the EEH status. We'll do more sensible things later when we have
591 	 * proper EEH support. We need to make sure we don't pollute ourselves
592 	 * with the normal errors generated when probing empty slots
593 	 */
594 	if (has_diag && ret)
595 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag_data);
596 
597 	spin_unlock_irqrestore(&phb->lock, flags);
598 }
599 
600 static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
601 {
602 	struct pnv_phb *phb = pdn->phb->private_data;
603 	u8	fstate = 0;
604 	__be16	pcierr = 0;
605 	unsigned int pe_no;
606 	s64	rc;
607 
608 	/*
609 	 * Get the PE#. During the PCI probe stage, we might not
610 	 * setup that yet. So all ER errors should be mapped to
611 	 * reserved PE.
612 	 */
613 	pe_no = pdn->pe_number;
614 	if (pe_no == IODA_INVALID_PE) {
615 		pe_no = phb->ioda.reserved_pe_idx;
616 	}
617 
618 	/*
619 	 * Fetch frozen state. If the PHB support compound PE,
620 	 * we need handle that case.
621 	 */
622 	if (phb->get_pe_state) {
623 		fstate = phb->get_pe_state(phb, pe_no);
624 	} else {
625 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
626 						pe_no,
627 						&fstate,
628 						&pcierr,
629 						NULL);
630 		if (rc) {
631 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
632 				__func__, rc, phb->hose->global_number, pe_no);
633 			return;
634 		}
635 	}
636 
637 	pr_devel(" -> EEH check, bdfn=%04x PE#%x fstate=%x\n",
638 		 (pdn->busno << 8) | (pdn->devfn), pe_no, fstate);
639 
640 	/* Clear the frozen state if applicable */
641 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
642 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
643 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
644 		/*
645 		 * If PHB supports compound PE, freeze it for
646 		 * consistency.
647 		 */
648 		if (phb->freeze_pe)
649 			phb->freeze_pe(phb, pe_no);
650 
651 		pnv_pci_handle_eeh_config(phb, pe_no);
652 	}
653 }
654 
655 int pnv_pci_cfg_read(struct pci_dn *pdn,
656 		     int where, int size, u32 *val)
657 {
658 	struct pnv_phb *phb = pdn->phb->private_data;
659 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
660 	s64 rc;
661 
662 	switch (size) {
663 	case 1: {
664 		u8 v8;
665 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
666 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
667 		break;
668 	}
669 	case 2: {
670 		__be16 v16;
671 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
672 						   &v16);
673 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
674 		break;
675 	}
676 	case 4: {
677 		__be32 v32;
678 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
679 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
680 		break;
681 	}
682 	default:
683 		return PCIBIOS_FUNC_NOT_SUPPORTED;
684 	}
685 
686 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
687 		 __func__, pdn->busno, pdn->devfn, where, size, *val);
688 	return PCIBIOS_SUCCESSFUL;
689 }
690 
691 int pnv_pci_cfg_write(struct pci_dn *pdn,
692 		      int where, int size, u32 val)
693 {
694 	struct pnv_phb *phb = pdn->phb->private_data;
695 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
696 
697 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
698 		 __func__, pdn->busno, pdn->devfn, where, size, val);
699 	switch (size) {
700 	case 1:
701 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
702 		break;
703 	case 2:
704 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
705 		break;
706 	case 4:
707 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
708 		break;
709 	default:
710 		return PCIBIOS_FUNC_NOT_SUPPORTED;
711 	}
712 
713 	return PCIBIOS_SUCCESSFUL;
714 }
715 
716 #if CONFIG_EEH
717 static bool pnv_pci_cfg_check(struct pci_dn *pdn)
718 {
719 	struct eeh_dev *edev = NULL;
720 	struct pnv_phb *phb = pdn->phb->private_data;
721 
722 	/* EEH not enabled ? */
723 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
724 		return true;
725 
726 	/* PE reset or device removed ? */
727 	edev = pdn->edev;
728 	if (edev) {
729 		if (edev->pe &&
730 		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
731 			return false;
732 
733 		if (edev->mode & EEH_DEV_REMOVED)
734 			return false;
735 	}
736 
737 	return true;
738 }
739 #else
740 static inline pnv_pci_cfg_check(struct pci_dn *pdn)
741 {
742 	return true;
743 }
744 #endif /* CONFIG_EEH */
745 
746 static int pnv_pci_read_config(struct pci_bus *bus,
747 			       unsigned int devfn,
748 			       int where, int size, u32 *val)
749 {
750 	struct pci_dn *pdn;
751 	struct pnv_phb *phb;
752 	int ret;
753 
754 	*val = 0xFFFFFFFF;
755 	pdn = pci_get_pdn_by_devfn(bus, devfn);
756 	if (!pdn)
757 		return PCIBIOS_DEVICE_NOT_FOUND;
758 
759 	if (!pnv_pci_cfg_check(pdn))
760 		return PCIBIOS_DEVICE_NOT_FOUND;
761 
762 	ret = pnv_pci_cfg_read(pdn, where, size, val);
763 	phb = pdn->phb->private_data;
764 	if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
765 		if (*val == EEH_IO_ERROR_VALUE(size) &&
766 		    eeh_dev_check_failure(pdn->edev))
767                         return PCIBIOS_DEVICE_NOT_FOUND;
768 	} else {
769 		pnv_pci_config_check_eeh(pdn);
770 	}
771 
772 	return ret;
773 }
774 
775 static int pnv_pci_write_config(struct pci_bus *bus,
776 				unsigned int devfn,
777 				int where, int size, u32 val)
778 {
779 	struct pci_dn *pdn;
780 	struct pnv_phb *phb;
781 	int ret;
782 
783 	pdn = pci_get_pdn_by_devfn(bus, devfn);
784 	if (!pdn)
785 		return PCIBIOS_DEVICE_NOT_FOUND;
786 
787 	if (!pnv_pci_cfg_check(pdn))
788 		return PCIBIOS_DEVICE_NOT_FOUND;
789 
790 	ret = pnv_pci_cfg_write(pdn, where, size, val);
791 	phb = pdn->phb->private_data;
792 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
793 		pnv_pci_config_check_eeh(pdn);
794 
795 	return ret;
796 }
797 
798 struct pci_ops pnv_pci_ops = {
799 	.read  = pnv_pci_read_config,
800 	.write = pnv_pci_write_config,
801 };
802 
803 struct iommu_table *pnv_pci_table_alloc(int nid)
804 {
805 	struct iommu_table *tbl;
806 
807 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
808 	if (!tbl)
809 		return NULL;
810 
811 	INIT_LIST_HEAD_RCU(&tbl->it_group_list);
812 	kref_init(&tbl->it_kref);
813 
814 	return tbl;
815 }
816 
817 struct device_node *pnv_pci_get_phb_node(struct pci_dev *dev)
818 {
819 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
820 
821 	return of_node_get(hose->dn);
822 }
823 EXPORT_SYMBOL(pnv_pci_get_phb_node);
824 
825 int pnv_pci_set_tunnel_bar(struct pci_dev *dev, u64 addr, int enable)
826 {
827 	__be64 val;
828 	struct pci_controller *hose;
829 	struct pnv_phb *phb;
830 	u64 tunnel_bar;
831 	int rc;
832 
833 	if (!opal_check_token(OPAL_PCI_GET_PBCQ_TUNNEL_BAR))
834 		return -ENXIO;
835 	if (!opal_check_token(OPAL_PCI_SET_PBCQ_TUNNEL_BAR))
836 		return -ENXIO;
837 
838 	hose = pci_bus_to_host(dev->bus);
839 	phb = hose->private_data;
840 
841 	mutex_lock(&tunnel_mutex);
842 	rc = opal_pci_get_pbcq_tunnel_bar(phb->opal_id, &val);
843 	if (rc != OPAL_SUCCESS) {
844 		rc = -EIO;
845 		goto out;
846 	}
847 	tunnel_bar = be64_to_cpu(val);
848 	if (enable) {
849 		/*
850 		* Only one device per PHB can use atomics.
851 		* Our policy is first-come, first-served.
852 		*/
853 		if (tunnel_bar) {
854 			if (tunnel_bar != addr)
855 				rc = -EBUSY;
856 			else
857 				rc = 0;	/* Setting same address twice is ok */
858 			goto out;
859 		}
860 	} else {
861 		/*
862 		* The device that owns atomics and wants to release
863 		* them must pass the same address with enable == 0.
864 		*/
865 		if (tunnel_bar != addr) {
866 			rc = -EPERM;
867 			goto out;
868 		}
869 		addr = 0x0ULL;
870 	}
871 	rc = opal_pci_set_pbcq_tunnel_bar(phb->opal_id, addr);
872 	rc = opal_error_code(rc);
873 out:
874 	mutex_unlock(&tunnel_mutex);
875 	return rc;
876 }
877 EXPORT_SYMBOL_GPL(pnv_pci_set_tunnel_bar);
878 
879 void pnv_pci_shutdown(void)
880 {
881 	struct pci_controller *hose;
882 
883 	list_for_each_entry(hose, &hose_list, list_node)
884 		if (hose->controller_ops.shutdown)
885 			hose->controller_ops.shutdown(hose);
886 }
887 
888 /* Fixup wrong class code in p7ioc and p8 root complex */
889 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
890 {
891 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
892 }
893 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
894 
895 void __init pnv_pci_init(void)
896 {
897 	struct device_node *np;
898 
899 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
900 
901 	/* If we don't have OPAL, eg. in sim, just skip PCI probe */
902 	if (!firmware_has_feature(FW_FEATURE_OPAL))
903 		return;
904 
905 #ifdef CONFIG_PCIEPORTBUS
906 	/*
907 	 * On PowerNV PCIe devices are (currently) managed in cooperation
908 	 * with firmware. This isn't *strictly* required, but there's enough
909 	 * assumptions baked into both firmware and the platform code that
910 	 * it's unwise to allow the portbus services to be used.
911 	 *
912 	 * We need to fix this eventually, but for now set this flag to disable
913 	 * the portbus driver. The AER service isn't required since that AER
914 	 * events are handled via EEH. The pciehp hotplug driver can't work
915 	 * without kernel changes (and portbus binding breaks pnv_php). The
916 	 * other services also require some thinking about how we're going
917 	 * to integrate them.
918 	 */
919 	pcie_ports_disabled = true;
920 #endif
921 
922 	/* Look for IODA IO-Hubs. */
923 	for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
924 		pnv_pci_init_ioda_hub(np);
925 	}
926 
927 	/* Look for ioda2 built-in PHB3's */
928 	for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
929 		pnv_pci_init_ioda2_phb(np);
930 
931 	/* Look for ioda3 built-in PHB4's, we treat them as IODA2 */
932 	for_each_compatible_node(np, NULL, "ibm,ioda3-phb")
933 		pnv_pci_init_ioda2_phb(np);
934 
935 	/* Look for NPU PHBs */
936 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu-phb")
937 		pnv_pci_init_npu_phb(np);
938 
939 	/*
940 	 * Look for NPU2 PHBs which we treat mostly as NPU PHBs with
941 	 * the exception of TCE kill which requires an OPAL call.
942 	 */
943 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-phb")
944 		pnv_pci_init_npu_phb(np);
945 
946 	/* Look for NPU2 OpenCAPI PHBs */
947 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-opencapi-phb")
948 		pnv_pci_init_npu2_opencapi_phb(np);
949 
950 	/* Configure IOMMU DMA hooks */
951 	set_pci_dma_ops(&dma_iommu_ops);
952 }
953 
954 static int pnv_tce_iommu_bus_notifier(struct notifier_block *nb,
955 		unsigned long action, void *data)
956 {
957 	struct device *dev = data;
958 	struct pci_dev *pdev;
959 	struct pci_dn *pdn;
960 	struct pnv_ioda_pe *pe;
961 	struct pci_controller *hose;
962 	struct pnv_phb *phb;
963 
964 	switch (action) {
965 	case BUS_NOTIFY_ADD_DEVICE:
966 		pdev = to_pci_dev(dev);
967 		pdn = pci_get_pdn(pdev);
968 		hose = pci_bus_to_host(pdev->bus);
969 		phb = hose->private_data;
970 
971 		WARN_ON_ONCE(!phb);
972 		if (!pdn || pdn->pe_number == IODA_INVALID_PE || !phb)
973 			return 0;
974 
975 		pe = &phb->ioda.pe_array[pdn->pe_number];
976 		if (!pe->table_group.group)
977 			return 0;
978 		iommu_add_device(&pe->table_group, dev);
979 		return 0;
980 	case BUS_NOTIFY_DEL_DEVICE:
981 		iommu_del_device(dev);
982 		return 0;
983 	default:
984 		return 0;
985 	}
986 }
987 
988 static struct notifier_block pnv_tce_iommu_bus_nb = {
989 	.notifier_call = pnv_tce_iommu_bus_notifier,
990 };
991 
992 static int __init pnv_tce_iommu_bus_notifier_init(void)
993 {
994 	bus_register_notifier(&pci_bus_type, &pnv_tce_iommu_bus_nb);
995 	return 0;
996 }
997 machine_subsys_initcall_sync(powernv, pnv_tce_iommu_bus_notifier_init);
998