1 /*
2  * Support PCI/PCIe on PowerNV platforms
3  *
4  * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/kernel.h>
13 #include <linux/pci.h>
14 #include <linux/delay.h>
15 #include <linux/string.h>
16 #include <linux/init.h>
17 #include <linux/irq.h>
18 #include <linux/io.h>
19 #include <linux/msi.h>
20 #include <linux/iommu.h>
21 
22 #include <asm/sections.h>
23 #include <asm/io.h>
24 #include <asm/prom.h>
25 #include <asm/pci-bridge.h>
26 #include <asm/machdep.h>
27 #include <asm/msi_bitmap.h>
28 #include <asm/ppc-pci.h>
29 #include <asm/pnv-pci.h>
30 #include <asm/opal.h>
31 #include <asm/iommu.h>
32 #include <asm/tce.h>
33 #include <asm/firmware.h>
34 #include <asm/eeh_event.h>
35 #include <asm/eeh.h>
36 
37 #include "powernv.h"
38 #include "pci.h"
39 
40 int pnv_pci_get_slot_id(struct device_node *np, uint64_t *id)
41 {
42 	struct device_node *parent = np;
43 	u32 bdfn;
44 	u64 phbid;
45 	int ret;
46 
47 	ret = of_property_read_u32(np, "reg", &bdfn);
48 	if (ret)
49 		return -ENXIO;
50 
51 	bdfn = ((bdfn & 0x00ffff00) >> 8);
52 	while ((parent = of_get_parent(parent))) {
53 		if (!PCI_DN(parent)) {
54 			of_node_put(parent);
55 			break;
56 		}
57 
58 		if (!of_device_is_compatible(parent, "ibm,ioda2-phb")) {
59 			of_node_put(parent);
60 			continue;
61 		}
62 
63 		ret = of_property_read_u64(parent, "ibm,opal-phbid", &phbid);
64 		if (ret) {
65 			of_node_put(parent);
66 			return -ENXIO;
67 		}
68 
69 		*id = PCI_SLOT_ID(phbid, bdfn);
70 		return 0;
71 	}
72 
73 	return -ENODEV;
74 }
75 EXPORT_SYMBOL_GPL(pnv_pci_get_slot_id);
76 
77 int pnv_pci_get_device_tree(uint32_t phandle, void *buf, uint64_t len)
78 {
79 	int64_t rc;
80 
81 	if (!opal_check_token(OPAL_GET_DEVICE_TREE))
82 		return -ENXIO;
83 
84 	rc = opal_get_device_tree(phandle, (uint64_t)buf, len);
85 	if (rc < OPAL_SUCCESS)
86 		return -EIO;
87 
88 	return rc;
89 }
90 EXPORT_SYMBOL_GPL(pnv_pci_get_device_tree);
91 
92 int pnv_pci_get_presence_state(uint64_t id, uint8_t *state)
93 {
94 	int64_t rc;
95 
96 	if (!opal_check_token(OPAL_PCI_GET_PRESENCE_STATE))
97 		return -ENXIO;
98 
99 	rc = opal_pci_get_presence_state(id, (uint64_t)state);
100 	if (rc != OPAL_SUCCESS)
101 		return -EIO;
102 
103 	return 0;
104 }
105 EXPORT_SYMBOL_GPL(pnv_pci_get_presence_state);
106 
107 int pnv_pci_get_power_state(uint64_t id, uint8_t *state)
108 {
109 	int64_t rc;
110 
111 	if (!opal_check_token(OPAL_PCI_GET_POWER_STATE))
112 		return -ENXIO;
113 
114 	rc = opal_pci_get_power_state(id, (uint64_t)state);
115 	if (rc != OPAL_SUCCESS)
116 		return -EIO;
117 
118 	return 0;
119 }
120 EXPORT_SYMBOL_GPL(pnv_pci_get_power_state);
121 
122 int pnv_pci_set_power_state(uint64_t id, uint8_t state, struct opal_msg *msg)
123 {
124 	struct opal_msg m;
125 	int token, ret;
126 	int64_t rc;
127 
128 	if (!opal_check_token(OPAL_PCI_SET_POWER_STATE))
129 		return -ENXIO;
130 
131 	token = opal_async_get_token_interruptible();
132 	if (unlikely(token < 0))
133 		return token;
134 
135 	rc = opal_pci_set_power_state(token, id, (uint64_t)&state);
136 	if (rc == OPAL_SUCCESS) {
137 		ret = 0;
138 		goto exit;
139 	} else if (rc != OPAL_ASYNC_COMPLETION) {
140 		ret = -EIO;
141 		goto exit;
142 	}
143 
144 	ret = opal_async_wait_response(token, &m);
145 	if (ret < 0)
146 		goto exit;
147 
148 	if (msg) {
149 		ret = 1;
150 		memcpy(msg, &m, sizeof(m));
151 	}
152 
153 exit:
154 	opal_async_release_token(token);
155 	return ret;
156 }
157 EXPORT_SYMBOL_GPL(pnv_pci_set_power_state);
158 
159 #ifdef CONFIG_PCI_MSI
160 int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
161 {
162 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
163 	struct pnv_phb *phb = hose->private_data;
164 	struct msi_desc *entry;
165 	struct msi_msg msg;
166 	int hwirq;
167 	unsigned int virq;
168 	int rc;
169 
170 	if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
171 		return -ENODEV;
172 
173 	if (pdev->no_64bit_msi && !phb->msi32_support)
174 		return -ENODEV;
175 
176 	for_each_pci_msi_entry(entry, pdev) {
177 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
178 			pr_warn("%s: Supports only 64-bit MSIs\n",
179 				pci_name(pdev));
180 			return -ENXIO;
181 		}
182 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
183 		if (hwirq < 0) {
184 			pr_warn("%s: Failed to find a free MSI\n",
185 				pci_name(pdev));
186 			return -ENOSPC;
187 		}
188 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
189 		if (!virq) {
190 			pr_warn("%s: Failed to map MSI to linux irq\n",
191 				pci_name(pdev));
192 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
193 			return -ENOMEM;
194 		}
195 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
196 				    virq, entry->msi_attrib.is_64, &msg);
197 		if (rc) {
198 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
199 			irq_dispose_mapping(virq);
200 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
201 			return rc;
202 		}
203 		irq_set_msi_desc(virq, entry);
204 		pci_write_msi_msg(virq, &msg);
205 	}
206 	return 0;
207 }
208 
209 void pnv_teardown_msi_irqs(struct pci_dev *pdev)
210 {
211 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
212 	struct pnv_phb *phb = hose->private_data;
213 	struct msi_desc *entry;
214 	irq_hw_number_t hwirq;
215 
216 	if (WARN_ON(!phb))
217 		return;
218 
219 	for_each_pci_msi_entry(entry, pdev) {
220 		if (!entry->irq)
221 			continue;
222 		hwirq = virq_to_hw(entry->irq);
223 		irq_set_msi_desc(entry->irq, NULL);
224 		irq_dispose_mapping(entry->irq);
225 		msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
226 	}
227 }
228 #endif /* CONFIG_PCI_MSI */
229 
230 /* Nicely print the contents of the PE State Tables (PEST). */
231 static void pnv_pci_dump_pest(__be64 pestA[], __be64 pestB[], int pest_size)
232 {
233 	__be64 prevA = ULONG_MAX, prevB = ULONG_MAX;
234 	bool dup = false;
235 	int i;
236 
237 	for (i = 0; i < pest_size; i++) {
238 		__be64 peA = be64_to_cpu(pestA[i]);
239 		__be64 peB = be64_to_cpu(pestB[i]);
240 
241 		if (peA != prevA || peB != prevB) {
242 			if (dup) {
243 				pr_info("PE[..%03x] A/B: as above\n", i-1);
244 				dup = false;
245 			}
246 			prevA = peA;
247 			prevB = peB;
248 			if (peA & PNV_IODA_STOPPED_STATE ||
249 			    peB & PNV_IODA_STOPPED_STATE)
250 				pr_info("PE[%03x] A/B: %016llx %016llx\n",
251 					i, peA, peB);
252 		} else if (!dup && (peA & PNV_IODA_STOPPED_STATE ||
253 				    peB & PNV_IODA_STOPPED_STATE)) {
254 			dup = true;
255 		}
256 	}
257 }
258 
259 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
260 					 struct OpalIoPhbErrorCommon *common)
261 {
262 	struct OpalIoP7IOCPhbErrorData *data;
263 
264 	data = (struct OpalIoP7IOCPhbErrorData *)common;
265 	pr_info("P7IOC PHB#%x Diag-data (Version: %d)\n",
266 		hose->global_number, be32_to_cpu(common->version));
267 
268 	if (data->brdgCtl)
269 		pr_info("brdgCtl:     %08x\n",
270 			be32_to_cpu(data->brdgCtl));
271 	if (data->portStatusReg || data->rootCmplxStatus ||
272 	    data->busAgentStatus)
273 		pr_info("UtlSts:      %08x %08x %08x\n",
274 			be32_to_cpu(data->portStatusReg),
275 			be32_to_cpu(data->rootCmplxStatus),
276 			be32_to_cpu(data->busAgentStatus));
277 	if (data->deviceStatus || data->slotStatus   ||
278 	    data->linkStatus   || data->devCmdStatus ||
279 	    data->devSecStatus)
280 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
281 			be32_to_cpu(data->deviceStatus),
282 			be32_to_cpu(data->slotStatus),
283 			be32_to_cpu(data->linkStatus),
284 			be32_to_cpu(data->devCmdStatus),
285 			be32_to_cpu(data->devSecStatus));
286 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
287 	    data->corrErrorStatus)
288 		pr_info("RootErrSts:  %08x %08x %08x\n",
289 			be32_to_cpu(data->rootErrorStatus),
290 			be32_to_cpu(data->uncorrErrorStatus),
291 			be32_to_cpu(data->corrErrorStatus));
292 	if (data->tlpHdr1 || data->tlpHdr2 ||
293 	    data->tlpHdr3 || data->tlpHdr4)
294 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
295 			be32_to_cpu(data->tlpHdr1),
296 			be32_to_cpu(data->tlpHdr2),
297 			be32_to_cpu(data->tlpHdr3),
298 			be32_to_cpu(data->tlpHdr4));
299 	if (data->sourceId || data->errorClass ||
300 	    data->correlator)
301 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
302 			be32_to_cpu(data->sourceId),
303 			be64_to_cpu(data->errorClass),
304 			be64_to_cpu(data->correlator));
305 	if (data->p7iocPlssr || data->p7iocCsr)
306 		pr_info("PhbSts:      %016llx %016llx\n",
307 			be64_to_cpu(data->p7iocPlssr),
308 			be64_to_cpu(data->p7iocCsr));
309 	if (data->lemFir)
310 		pr_info("Lem:         %016llx %016llx %016llx\n",
311 			be64_to_cpu(data->lemFir),
312 			be64_to_cpu(data->lemErrorMask),
313 			be64_to_cpu(data->lemWOF));
314 	if (data->phbErrorStatus)
315 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
316 			be64_to_cpu(data->phbErrorStatus),
317 			be64_to_cpu(data->phbFirstErrorStatus),
318 			be64_to_cpu(data->phbErrorLog0),
319 			be64_to_cpu(data->phbErrorLog1));
320 	if (data->mmioErrorStatus)
321 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
322 			be64_to_cpu(data->mmioErrorStatus),
323 			be64_to_cpu(data->mmioFirstErrorStatus),
324 			be64_to_cpu(data->mmioErrorLog0),
325 			be64_to_cpu(data->mmioErrorLog1));
326 	if (data->dma0ErrorStatus)
327 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
328 			be64_to_cpu(data->dma0ErrorStatus),
329 			be64_to_cpu(data->dma0FirstErrorStatus),
330 			be64_to_cpu(data->dma0ErrorLog0),
331 			be64_to_cpu(data->dma0ErrorLog1));
332 	if (data->dma1ErrorStatus)
333 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
334 			be64_to_cpu(data->dma1ErrorStatus),
335 			be64_to_cpu(data->dma1FirstErrorStatus),
336 			be64_to_cpu(data->dma1ErrorLog0),
337 			be64_to_cpu(data->dma1ErrorLog1));
338 
339 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_P7IOC_NUM_PEST_REGS);
340 }
341 
342 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
343 					struct OpalIoPhbErrorCommon *common)
344 {
345 	struct OpalIoPhb3ErrorData *data;
346 
347 	data = (struct OpalIoPhb3ErrorData*)common;
348 	pr_info("PHB3 PHB#%x Diag-data (Version: %d)\n",
349 		hose->global_number, be32_to_cpu(common->version));
350 	if (data->brdgCtl)
351 		pr_info("brdgCtl:     %08x\n",
352 			be32_to_cpu(data->brdgCtl));
353 	if (data->portStatusReg || data->rootCmplxStatus ||
354 	    data->busAgentStatus)
355 		pr_info("UtlSts:      %08x %08x %08x\n",
356 			be32_to_cpu(data->portStatusReg),
357 			be32_to_cpu(data->rootCmplxStatus),
358 			be32_to_cpu(data->busAgentStatus));
359 	if (data->deviceStatus || data->slotStatus   ||
360 	    data->linkStatus   || data->devCmdStatus ||
361 	    data->devSecStatus)
362 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
363 			be32_to_cpu(data->deviceStatus),
364 			be32_to_cpu(data->slotStatus),
365 			be32_to_cpu(data->linkStatus),
366 			be32_to_cpu(data->devCmdStatus),
367 			be32_to_cpu(data->devSecStatus));
368 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
369 	    data->corrErrorStatus)
370 		pr_info("RootErrSts:  %08x %08x %08x\n",
371 			be32_to_cpu(data->rootErrorStatus),
372 			be32_to_cpu(data->uncorrErrorStatus),
373 			be32_to_cpu(data->corrErrorStatus));
374 	if (data->tlpHdr1 || data->tlpHdr2 ||
375 	    data->tlpHdr3 || data->tlpHdr4)
376 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
377 			be32_to_cpu(data->tlpHdr1),
378 			be32_to_cpu(data->tlpHdr2),
379 			be32_to_cpu(data->tlpHdr3),
380 			be32_to_cpu(data->tlpHdr4));
381 	if (data->sourceId || data->errorClass ||
382 	    data->correlator)
383 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
384 			be32_to_cpu(data->sourceId),
385 			be64_to_cpu(data->errorClass),
386 			be64_to_cpu(data->correlator));
387 	if (data->nFir)
388 		pr_info("nFir:        %016llx %016llx %016llx\n",
389 			be64_to_cpu(data->nFir),
390 			be64_to_cpu(data->nFirMask),
391 			be64_to_cpu(data->nFirWOF));
392 	if (data->phbPlssr || data->phbCsr)
393 		pr_info("PhbSts:      %016llx %016llx\n",
394 			be64_to_cpu(data->phbPlssr),
395 			be64_to_cpu(data->phbCsr));
396 	if (data->lemFir)
397 		pr_info("Lem:         %016llx %016llx %016llx\n",
398 			be64_to_cpu(data->lemFir),
399 			be64_to_cpu(data->lemErrorMask),
400 			be64_to_cpu(data->lemWOF));
401 	if (data->phbErrorStatus)
402 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
403 			be64_to_cpu(data->phbErrorStatus),
404 			be64_to_cpu(data->phbFirstErrorStatus),
405 			be64_to_cpu(data->phbErrorLog0),
406 			be64_to_cpu(data->phbErrorLog1));
407 	if (data->mmioErrorStatus)
408 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
409 			be64_to_cpu(data->mmioErrorStatus),
410 			be64_to_cpu(data->mmioFirstErrorStatus),
411 			be64_to_cpu(data->mmioErrorLog0),
412 			be64_to_cpu(data->mmioErrorLog1));
413 	if (data->dma0ErrorStatus)
414 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
415 			be64_to_cpu(data->dma0ErrorStatus),
416 			be64_to_cpu(data->dma0FirstErrorStatus),
417 			be64_to_cpu(data->dma0ErrorLog0),
418 			be64_to_cpu(data->dma0ErrorLog1));
419 	if (data->dma1ErrorStatus)
420 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
421 			be64_to_cpu(data->dma1ErrorStatus),
422 			be64_to_cpu(data->dma1FirstErrorStatus),
423 			be64_to_cpu(data->dma1ErrorLog0),
424 			be64_to_cpu(data->dma1ErrorLog1));
425 
426 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB3_NUM_PEST_REGS);
427 }
428 
429 static void pnv_pci_dump_phb4_diag_data(struct pci_controller *hose,
430 					struct OpalIoPhbErrorCommon *common)
431 {
432 	struct OpalIoPhb4ErrorData *data;
433 
434 	data = (struct OpalIoPhb4ErrorData*)common;
435 	pr_info("PHB4 PHB#%d Diag-data (Version: %d)\n",
436 		hose->global_number, be32_to_cpu(common->version));
437 	if (data->brdgCtl)
438 		pr_info("brdgCtl:    %08x\n",
439 			be32_to_cpu(data->brdgCtl));
440 	if (data->deviceStatus || data->slotStatus   ||
441 	    data->linkStatus   || data->devCmdStatus ||
442 	    data->devSecStatus)
443 		pr_info("RootSts:    %08x %08x %08x %08x %08x\n",
444 			be32_to_cpu(data->deviceStatus),
445 			be32_to_cpu(data->slotStatus),
446 			be32_to_cpu(data->linkStatus),
447 			be32_to_cpu(data->devCmdStatus),
448 			be32_to_cpu(data->devSecStatus));
449 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
450 	    data->corrErrorStatus)
451 		pr_info("RootErrSts: %08x %08x %08x\n",
452 			be32_to_cpu(data->rootErrorStatus),
453 			be32_to_cpu(data->uncorrErrorStatus),
454 			be32_to_cpu(data->corrErrorStatus));
455 	if (data->tlpHdr1 || data->tlpHdr2 ||
456 	    data->tlpHdr3 || data->tlpHdr4)
457 		pr_info("RootErrLog: %08x %08x %08x %08x\n",
458 			be32_to_cpu(data->tlpHdr1),
459 			be32_to_cpu(data->tlpHdr2),
460 			be32_to_cpu(data->tlpHdr3),
461 			be32_to_cpu(data->tlpHdr4));
462 	if (data->sourceId)
463 		pr_info("sourceId:   %08x\n", be32_to_cpu(data->sourceId));
464 	if (data->nFir)
465 		pr_info("nFir:       %016llx %016llx %016llx\n",
466 			be64_to_cpu(data->nFir),
467 			be64_to_cpu(data->nFirMask),
468 			be64_to_cpu(data->nFirWOF));
469 	if (data->phbPlssr || data->phbCsr)
470 		pr_info("PhbSts:     %016llx %016llx\n",
471 			be64_to_cpu(data->phbPlssr),
472 			be64_to_cpu(data->phbCsr));
473 	if (data->lemFir)
474 		pr_info("Lem:        %016llx %016llx %016llx\n",
475 			be64_to_cpu(data->lemFir),
476 			be64_to_cpu(data->lemErrorMask),
477 			be64_to_cpu(data->lemWOF));
478 	if (data->phbErrorStatus)
479 		pr_info("PhbErr:     %016llx %016llx %016llx %016llx\n",
480 			be64_to_cpu(data->phbErrorStatus),
481 			be64_to_cpu(data->phbFirstErrorStatus),
482 			be64_to_cpu(data->phbErrorLog0),
483 			be64_to_cpu(data->phbErrorLog1));
484 	if (data->phbTxeErrorStatus)
485 		pr_info("PhbTxeErr:  %016llx %016llx %016llx %016llx\n",
486 			be64_to_cpu(data->phbTxeErrorStatus),
487 			be64_to_cpu(data->phbTxeFirstErrorStatus),
488 			be64_to_cpu(data->phbTxeErrorLog0),
489 			be64_to_cpu(data->phbTxeErrorLog1));
490 	if (data->phbRxeArbErrorStatus)
491 		pr_info("RxeArbErr:  %016llx %016llx %016llx %016llx\n",
492 			be64_to_cpu(data->phbRxeArbErrorStatus),
493 			be64_to_cpu(data->phbRxeArbFirstErrorStatus),
494 			be64_to_cpu(data->phbRxeArbErrorLog0),
495 			be64_to_cpu(data->phbRxeArbErrorLog1));
496 	if (data->phbRxeMrgErrorStatus)
497 		pr_info("RxeMrgErr:  %016llx %016llx %016llx %016llx\n",
498 			be64_to_cpu(data->phbRxeMrgErrorStatus),
499 			be64_to_cpu(data->phbRxeMrgFirstErrorStatus),
500 			be64_to_cpu(data->phbRxeMrgErrorLog0),
501 			be64_to_cpu(data->phbRxeMrgErrorLog1));
502 	if (data->phbRxeTceErrorStatus)
503 		pr_info("RxeTceErr:  %016llx %016llx %016llx %016llx\n",
504 			be64_to_cpu(data->phbRxeTceErrorStatus),
505 			be64_to_cpu(data->phbRxeTceFirstErrorStatus),
506 			be64_to_cpu(data->phbRxeTceErrorLog0),
507 			be64_to_cpu(data->phbRxeTceErrorLog1));
508 
509 	if (data->phbPblErrorStatus)
510 		pr_info("PblErr:     %016llx %016llx %016llx %016llx\n",
511 			be64_to_cpu(data->phbPblErrorStatus),
512 			be64_to_cpu(data->phbPblFirstErrorStatus),
513 			be64_to_cpu(data->phbPblErrorLog0),
514 			be64_to_cpu(data->phbPblErrorLog1));
515 	if (data->phbPcieDlpErrorStatus)
516 		pr_info("PcieDlp:    %016llx %016llx %016llx\n",
517 			be64_to_cpu(data->phbPcieDlpErrorLog1),
518 			be64_to_cpu(data->phbPcieDlpErrorLog2),
519 			be64_to_cpu(data->phbPcieDlpErrorStatus));
520 	if (data->phbRegbErrorStatus)
521 		pr_info("RegbErr:    %016llx %016llx %016llx %016llx\n",
522 			be64_to_cpu(data->phbRegbErrorStatus),
523 			be64_to_cpu(data->phbRegbFirstErrorStatus),
524 			be64_to_cpu(data->phbRegbErrorLog0),
525 			be64_to_cpu(data->phbRegbErrorLog1));
526 
527 
528 	pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB4_NUM_PEST_REGS);
529 }
530 
531 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
532 				unsigned char *log_buff)
533 {
534 	struct OpalIoPhbErrorCommon *common;
535 
536 	if (!hose || !log_buff)
537 		return;
538 
539 	common = (struct OpalIoPhbErrorCommon *)log_buff;
540 	switch (be32_to_cpu(common->ioType)) {
541 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
542 		pnv_pci_dump_p7ioc_diag_data(hose, common);
543 		break;
544 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
545 		pnv_pci_dump_phb3_diag_data(hose, common);
546 		break;
547 	case OPAL_PHB_ERROR_DATA_TYPE_PHB4:
548 		pnv_pci_dump_phb4_diag_data(hose, common);
549 		break;
550 	default:
551 		pr_warn("%s: Unrecognized ioType %d\n",
552 			__func__, be32_to_cpu(common->ioType));
553 	}
554 }
555 
556 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
557 {
558 	unsigned long flags, rc;
559 	int has_diag, ret = 0;
560 
561 	spin_lock_irqsave(&phb->lock, flags);
562 
563 	/* Fetch PHB diag-data */
564 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag_data,
565 					 phb->diag_data_size);
566 	has_diag = (rc == OPAL_SUCCESS);
567 
568 	/* If PHB supports compound PE, to handle it */
569 	if (phb->unfreeze_pe) {
570 		ret = phb->unfreeze_pe(phb,
571 				       pe_no,
572 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
573 	} else {
574 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
575 					     pe_no,
576 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
577 		if (rc) {
578 			pr_warn("%s: Failure %ld clearing frozen "
579 				"PHB#%x-PE#%x\n",
580 				__func__, rc, phb->hose->global_number,
581 				pe_no);
582 			ret = -EIO;
583 		}
584 	}
585 
586 	/*
587 	 * For now, let's only display the diag buffer when we fail to clear
588 	 * the EEH status. We'll do more sensible things later when we have
589 	 * proper EEH support. We need to make sure we don't pollute ourselves
590 	 * with the normal errors generated when probing empty slots
591 	 */
592 	if (has_diag && ret)
593 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag_data);
594 
595 	spin_unlock_irqrestore(&phb->lock, flags);
596 }
597 
598 static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
599 {
600 	struct pnv_phb *phb = pdn->phb->private_data;
601 	u8	fstate;
602 	__be16	pcierr;
603 	unsigned int pe_no;
604 	s64	rc;
605 
606 	/*
607 	 * Get the PE#. During the PCI probe stage, we might not
608 	 * setup that yet. So all ER errors should be mapped to
609 	 * reserved PE.
610 	 */
611 	pe_no = pdn->pe_number;
612 	if (pe_no == IODA_INVALID_PE) {
613 		pe_no = phb->ioda.reserved_pe_idx;
614 	}
615 
616 	/*
617 	 * Fetch frozen state. If the PHB support compound PE,
618 	 * we need handle that case.
619 	 */
620 	if (phb->get_pe_state) {
621 		fstate = phb->get_pe_state(phb, pe_no);
622 	} else {
623 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
624 						pe_no,
625 						&fstate,
626 						&pcierr,
627 						NULL);
628 		if (rc) {
629 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
630 				__func__, rc, phb->hose->global_number, pe_no);
631 			return;
632 		}
633 	}
634 
635 	pr_devel(" -> EEH check, bdfn=%04x PE#%x fstate=%x\n",
636 		 (pdn->busno << 8) | (pdn->devfn), pe_no, fstate);
637 
638 	/* Clear the frozen state if applicable */
639 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
640 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
641 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
642 		/*
643 		 * If PHB supports compound PE, freeze it for
644 		 * consistency.
645 		 */
646 		if (phb->freeze_pe)
647 			phb->freeze_pe(phb, pe_no);
648 
649 		pnv_pci_handle_eeh_config(phb, pe_no);
650 	}
651 }
652 
653 int pnv_pci_cfg_read(struct pci_dn *pdn,
654 		     int where, int size, u32 *val)
655 {
656 	struct pnv_phb *phb = pdn->phb->private_data;
657 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
658 	s64 rc;
659 
660 	switch (size) {
661 	case 1: {
662 		u8 v8;
663 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
664 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
665 		break;
666 	}
667 	case 2: {
668 		__be16 v16;
669 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
670 						   &v16);
671 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
672 		break;
673 	}
674 	case 4: {
675 		__be32 v32;
676 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
677 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
678 		break;
679 	}
680 	default:
681 		return PCIBIOS_FUNC_NOT_SUPPORTED;
682 	}
683 
684 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
685 		 __func__, pdn->busno, pdn->devfn, where, size, *val);
686 	return PCIBIOS_SUCCESSFUL;
687 }
688 
689 int pnv_pci_cfg_write(struct pci_dn *pdn,
690 		      int where, int size, u32 val)
691 {
692 	struct pnv_phb *phb = pdn->phb->private_data;
693 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
694 
695 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
696 		 __func__, pdn->busno, pdn->devfn, where, size, val);
697 	switch (size) {
698 	case 1:
699 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
700 		break;
701 	case 2:
702 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
703 		break;
704 	case 4:
705 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
706 		break;
707 	default:
708 		return PCIBIOS_FUNC_NOT_SUPPORTED;
709 	}
710 
711 	return PCIBIOS_SUCCESSFUL;
712 }
713 
714 #if CONFIG_EEH
715 static bool pnv_pci_cfg_check(struct pci_dn *pdn)
716 {
717 	struct eeh_dev *edev = NULL;
718 	struct pnv_phb *phb = pdn->phb->private_data;
719 
720 	/* EEH not enabled ? */
721 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
722 		return true;
723 
724 	/* PE reset or device removed ? */
725 	edev = pdn->edev;
726 	if (edev) {
727 		if (edev->pe &&
728 		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
729 			return false;
730 
731 		if (edev->mode & EEH_DEV_REMOVED)
732 			return false;
733 	}
734 
735 	return true;
736 }
737 #else
738 static inline pnv_pci_cfg_check(struct pci_dn *pdn)
739 {
740 	return true;
741 }
742 #endif /* CONFIG_EEH */
743 
744 static int pnv_pci_read_config(struct pci_bus *bus,
745 			       unsigned int devfn,
746 			       int where, int size, u32 *val)
747 {
748 	struct pci_dn *pdn;
749 	struct pnv_phb *phb;
750 	int ret;
751 
752 	*val = 0xFFFFFFFF;
753 	pdn = pci_get_pdn_by_devfn(bus, devfn);
754 	if (!pdn)
755 		return PCIBIOS_DEVICE_NOT_FOUND;
756 
757 	if (!pnv_pci_cfg_check(pdn))
758 		return PCIBIOS_DEVICE_NOT_FOUND;
759 
760 	ret = pnv_pci_cfg_read(pdn, where, size, val);
761 	phb = pdn->phb->private_data;
762 	if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
763 		if (*val == EEH_IO_ERROR_VALUE(size) &&
764 		    eeh_dev_check_failure(pdn->edev))
765                         return PCIBIOS_DEVICE_NOT_FOUND;
766 	} else {
767 		pnv_pci_config_check_eeh(pdn);
768 	}
769 
770 	return ret;
771 }
772 
773 static int pnv_pci_write_config(struct pci_bus *bus,
774 				unsigned int devfn,
775 				int where, int size, u32 val)
776 {
777 	struct pci_dn *pdn;
778 	struct pnv_phb *phb;
779 	int ret;
780 
781 	pdn = pci_get_pdn_by_devfn(bus, devfn);
782 	if (!pdn)
783 		return PCIBIOS_DEVICE_NOT_FOUND;
784 
785 	if (!pnv_pci_cfg_check(pdn))
786 		return PCIBIOS_DEVICE_NOT_FOUND;
787 
788 	ret = pnv_pci_cfg_write(pdn, where, size, val);
789 	phb = pdn->phb->private_data;
790 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
791 		pnv_pci_config_check_eeh(pdn);
792 
793 	return ret;
794 }
795 
796 struct pci_ops pnv_pci_ops = {
797 	.read  = pnv_pci_read_config,
798 	.write = pnv_pci_write_config,
799 };
800 
801 static __be64 *pnv_tce(struct iommu_table *tbl, long idx)
802 {
803 	__be64 *tmp = ((__be64 *)tbl->it_base);
804 	int  level = tbl->it_indirect_levels;
805 	const long shift = ilog2(tbl->it_level_size);
806 	unsigned long mask = (tbl->it_level_size - 1) << (level * shift);
807 
808 	while (level) {
809 		int n = (idx & mask) >> (level * shift);
810 		unsigned long tce = be64_to_cpu(tmp[n]);
811 
812 		tmp = __va(tce & ~(TCE_PCI_READ | TCE_PCI_WRITE));
813 		idx &= ~mask;
814 		mask >>= shift;
815 		--level;
816 	}
817 
818 	return tmp + idx;
819 }
820 
821 int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
822 		unsigned long uaddr, enum dma_data_direction direction,
823 		unsigned long attrs)
824 {
825 	u64 proto_tce = iommu_direction_to_tce_perm(direction);
826 	u64 rpn = __pa(uaddr) >> tbl->it_page_shift;
827 	long i;
828 
829 	if (proto_tce & TCE_PCI_WRITE)
830 		proto_tce |= TCE_PCI_READ;
831 
832 	for (i = 0; i < npages; i++) {
833 		unsigned long newtce = proto_tce |
834 			((rpn + i) << tbl->it_page_shift);
835 		unsigned long idx = index - tbl->it_offset + i;
836 
837 		*(pnv_tce(tbl, idx)) = cpu_to_be64(newtce);
838 	}
839 
840 	return 0;
841 }
842 
843 #ifdef CONFIG_IOMMU_API
844 int pnv_tce_xchg(struct iommu_table *tbl, long index,
845 		unsigned long *hpa, enum dma_data_direction *direction)
846 {
847 	u64 proto_tce = iommu_direction_to_tce_perm(*direction);
848 	unsigned long newtce = *hpa | proto_tce, oldtce;
849 	unsigned long idx = index - tbl->it_offset;
850 
851 	BUG_ON(*hpa & ~IOMMU_PAGE_MASK(tbl));
852 
853 	if (newtce & TCE_PCI_WRITE)
854 		newtce |= TCE_PCI_READ;
855 
856 	oldtce = be64_to_cpu(xchg(pnv_tce(tbl, idx), cpu_to_be64(newtce)));
857 	*hpa = oldtce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
858 	*direction = iommu_tce_direction(oldtce);
859 
860 	return 0;
861 }
862 #endif
863 
864 void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
865 {
866 	long i;
867 
868 	for (i = 0; i < npages; i++) {
869 		unsigned long idx = index - tbl->it_offset + i;
870 
871 		*(pnv_tce(tbl, idx)) = cpu_to_be64(0);
872 	}
873 }
874 
875 unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
876 {
877 	return be64_to_cpu(*(pnv_tce(tbl, index - tbl->it_offset)));
878 }
879 
880 struct iommu_table *pnv_pci_table_alloc(int nid)
881 {
882 	struct iommu_table *tbl;
883 
884 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
885 	if (!tbl)
886 		return NULL;
887 
888 	INIT_LIST_HEAD_RCU(&tbl->it_group_list);
889 	kref_init(&tbl->it_kref);
890 
891 	return tbl;
892 }
893 
894 long pnv_pci_link_table_and_group(int node, int num,
895 		struct iommu_table *tbl,
896 		struct iommu_table_group *table_group)
897 {
898 	struct iommu_table_group_link *tgl = NULL;
899 
900 	if (WARN_ON(!tbl || !table_group))
901 		return -EINVAL;
902 
903 	tgl = kzalloc_node(sizeof(struct iommu_table_group_link), GFP_KERNEL,
904 			node);
905 	if (!tgl)
906 		return -ENOMEM;
907 
908 	tgl->table_group = table_group;
909 	list_add_rcu(&tgl->next, &tbl->it_group_list);
910 
911 	table_group->tables[num] = tbl;
912 
913 	return 0;
914 }
915 
916 static void pnv_iommu_table_group_link_free(struct rcu_head *head)
917 {
918 	struct iommu_table_group_link *tgl = container_of(head,
919 			struct iommu_table_group_link, rcu);
920 
921 	kfree(tgl);
922 }
923 
924 void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
925 		struct iommu_table_group *table_group)
926 {
927 	long i;
928 	bool found;
929 	struct iommu_table_group_link *tgl;
930 
931 	if (!tbl || !table_group)
932 		return;
933 
934 	/* Remove link to a group from table's list of attached groups */
935 	found = false;
936 	list_for_each_entry_rcu(tgl, &tbl->it_group_list, next) {
937 		if (tgl->table_group == table_group) {
938 			list_del_rcu(&tgl->next);
939 			call_rcu(&tgl->rcu, pnv_iommu_table_group_link_free);
940 			found = true;
941 			break;
942 		}
943 	}
944 	if (WARN_ON(!found))
945 		return;
946 
947 	/* Clean a pointer to iommu_table in iommu_table_group::tables[] */
948 	found = false;
949 	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
950 		if (table_group->tables[i] == tbl) {
951 			table_group->tables[i] = NULL;
952 			found = true;
953 			break;
954 		}
955 	}
956 	WARN_ON(!found);
957 }
958 
959 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
960 			       void *tce_mem, u64 tce_size,
961 			       u64 dma_offset, unsigned page_shift)
962 {
963 	tbl->it_blocksize = 16;
964 	tbl->it_base = (unsigned long)tce_mem;
965 	tbl->it_page_shift = page_shift;
966 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
967 	tbl->it_index = 0;
968 	tbl->it_size = tce_size >> 3;
969 	tbl->it_busno = 0;
970 	tbl->it_type = TCE_PCI;
971 }
972 
973 void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
974 {
975 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
976 	struct pnv_phb *phb = hose->private_data;
977 #ifdef CONFIG_PCI_IOV
978 	struct pnv_ioda_pe *pe;
979 	struct pci_dn *pdn;
980 
981 	/* Fix the VF pdn PE number */
982 	if (pdev->is_virtfn) {
983 		pdn = pci_get_pdn(pdev);
984 		WARN_ON(pdn->pe_number != IODA_INVALID_PE);
985 		list_for_each_entry(pe, &phb->ioda.pe_list, list) {
986 			if (pe->rid == ((pdev->bus->number << 8) |
987 			    (pdev->devfn & 0xff))) {
988 				pdn->pe_number = pe->pe_number;
989 				pe->pdev = pdev;
990 				break;
991 			}
992 		}
993 	}
994 #endif /* CONFIG_PCI_IOV */
995 
996 	if (phb && phb->dma_dev_setup)
997 		phb->dma_dev_setup(phb, pdev);
998 }
999 
1000 void pnv_pci_dma_bus_setup(struct pci_bus *bus)
1001 {
1002 	struct pci_controller *hose = bus->sysdata;
1003 	struct pnv_phb *phb = hose->private_data;
1004 	struct pnv_ioda_pe *pe;
1005 
1006 	list_for_each_entry(pe, &phb->ioda.pe_list, list) {
1007 		if (!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)))
1008 			continue;
1009 
1010 		if (!pe->pbus)
1011 			continue;
1012 
1013 		if (bus->number == ((pe->rid >> 8) & 0xFF)) {
1014 			pe->pbus = bus;
1015 			break;
1016 		}
1017 	}
1018 }
1019 
1020 void pnv_pci_shutdown(void)
1021 {
1022 	struct pci_controller *hose;
1023 
1024 	list_for_each_entry(hose, &hose_list, list_node)
1025 		if (hose->controller_ops.shutdown)
1026 			hose->controller_ops.shutdown(hose);
1027 }
1028 
1029 /* Fixup wrong class code in p7ioc and p8 root complex */
1030 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
1031 {
1032 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
1033 }
1034 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
1035 
1036 void __init pnv_pci_init(void)
1037 {
1038 	struct device_node *np;
1039 
1040 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
1041 
1042 	/* If we don't have OPAL, eg. in sim, just skip PCI probe */
1043 	if (!firmware_has_feature(FW_FEATURE_OPAL))
1044 		return;
1045 
1046 	/* Look for IODA IO-Hubs. */
1047 	for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
1048 		pnv_pci_init_ioda_hub(np);
1049 	}
1050 
1051 	/* Look for ioda2 built-in PHB3's */
1052 	for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
1053 		pnv_pci_init_ioda2_phb(np);
1054 
1055 	/* Look for ioda3 built-in PHB4's, we treat them as IODA2 */
1056 	for_each_compatible_node(np, NULL, "ibm,ioda3-phb")
1057 		pnv_pci_init_ioda2_phb(np);
1058 
1059 	/* Look for NPU PHBs */
1060 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu-phb")
1061 		pnv_pci_init_npu_phb(np);
1062 
1063 	/*
1064 	 * Look for NPU2 PHBs which we treat mostly as NPU PHBs with
1065 	 * the exception of TCE kill which requires an OPAL call.
1066 	 */
1067 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-phb")
1068 		pnv_pci_init_npu_phb(np);
1069 
1070 	/* Configure IOMMU DMA hooks */
1071 	set_pci_dma_ops(&dma_iommu_ops);
1072 }
1073 
1074 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
1075