xref: /openbmc/linux/arch/powerpc/platforms/powernv/pci.c (revision 93707cbabcc8baf2b2b5f4a99c1f08ee83eb7abd)
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 static DEFINE_MUTEX(p2p_mutex);
41 
42 int pnv_pci_get_slot_id(struct device_node *np, uint64_t *id)
43 {
44 	struct device_node *parent = np;
45 	u32 bdfn;
46 	u64 phbid;
47 	int ret;
48 
49 	ret = of_property_read_u32(np, "reg", &bdfn);
50 	if (ret)
51 		return -ENXIO;
52 
53 	bdfn = ((bdfn & 0x00ffff00) >> 8);
54 	while ((parent = of_get_parent(parent))) {
55 		if (!PCI_DN(parent)) {
56 			of_node_put(parent);
57 			break;
58 		}
59 
60 		if (!of_device_is_compatible(parent, "ibm,ioda2-phb")) {
61 			of_node_put(parent);
62 			continue;
63 		}
64 
65 		ret = of_property_read_u64(parent, "ibm,opal-phbid", &phbid);
66 		if (ret) {
67 			of_node_put(parent);
68 			return -ENXIO;
69 		}
70 
71 		*id = PCI_SLOT_ID(phbid, bdfn);
72 		return 0;
73 	}
74 
75 	return -ENODEV;
76 }
77 EXPORT_SYMBOL_GPL(pnv_pci_get_slot_id);
78 
79 int pnv_pci_get_device_tree(uint32_t phandle, void *buf, uint64_t len)
80 {
81 	int64_t rc;
82 
83 	if (!opal_check_token(OPAL_GET_DEVICE_TREE))
84 		return -ENXIO;
85 
86 	rc = opal_get_device_tree(phandle, (uint64_t)buf, len);
87 	if (rc < OPAL_SUCCESS)
88 		return -EIO;
89 
90 	return rc;
91 }
92 EXPORT_SYMBOL_GPL(pnv_pci_get_device_tree);
93 
94 int pnv_pci_get_presence_state(uint64_t id, uint8_t *state)
95 {
96 	int64_t rc;
97 
98 	if (!opal_check_token(OPAL_PCI_GET_PRESENCE_STATE))
99 		return -ENXIO;
100 
101 	rc = opal_pci_get_presence_state(id, (uint64_t)state);
102 	if (rc != OPAL_SUCCESS)
103 		return -EIO;
104 
105 	return 0;
106 }
107 EXPORT_SYMBOL_GPL(pnv_pci_get_presence_state);
108 
109 int pnv_pci_get_power_state(uint64_t id, uint8_t *state)
110 {
111 	int64_t rc;
112 
113 	if (!opal_check_token(OPAL_PCI_GET_POWER_STATE))
114 		return -ENXIO;
115 
116 	rc = opal_pci_get_power_state(id, (uint64_t)state);
117 	if (rc != OPAL_SUCCESS)
118 		return -EIO;
119 
120 	return 0;
121 }
122 EXPORT_SYMBOL_GPL(pnv_pci_get_power_state);
123 
124 int pnv_pci_set_power_state(uint64_t id, uint8_t state, struct opal_msg *msg)
125 {
126 	struct opal_msg m;
127 	int token, ret;
128 	int64_t rc;
129 
130 	if (!opal_check_token(OPAL_PCI_SET_POWER_STATE))
131 		return -ENXIO;
132 
133 	token = opal_async_get_token_interruptible();
134 	if (unlikely(token < 0))
135 		return token;
136 
137 	rc = opal_pci_set_power_state(token, id, (uint64_t)&state);
138 	if (rc == OPAL_SUCCESS) {
139 		ret = 0;
140 		goto exit;
141 	} else if (rc != OPAL_ASYNC_COMPLETION) {
142 		ret = -EIO;
143 		goto exit;
144 	}
145 
146 	ret = opal_async_wait_response(token, &m);
147 	if (ret < 0)
148 		goto exit;
149 
150 	if (msg) {
151 		ret = 1;
152 		memcpy(msg, &m, sizeof(m));
153 	}
154 
155 exit:
156 	opal_async_release_token(token);
157 	return ret;
158 }
159 EXPORT_SYMBOL_GPL(pnv_pci_set_power_state);
160 
161 #ifdef CONFIG_PCI_MSI
162 int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
163 {
164 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
165 	struct pnv_phb *phb = hose->private_data;
166 	struct msi_desc *entry;
167 	struct msi_msg msg;
168 	int hwirq;
169 	unsigned int virq;
170 	int rc;
171 
172 	if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
173 		return -ENODEV;
174 
175 	if (pdev->no_64bit_msi && !phb->msi32_support)
176 		return -ENODEV;
177 
178 	for_each_pci_msi_entry(entry, pdev) {
179 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
180 			pr_warn("%s: Supports only 64-bit MSIs\n",
181 				pci_name(pdev));
182 			return -ENXIO;
183 		}
184 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
185 		if (hwirq < 0) {
186 			pr_warn("%s: Failed to find a free MSI\n",
187 				pci_name(pdev));
188 			return -ENOSPC;
189 		}
190 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
191 		if (!virq) {
192 			pr_warn("%s: Failed to map MSI to linux irq\n",
193 				pci_name(pdev));
194 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
195 			return -ENOMEM;
196 		}
197 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
198 				    virq, entry->msi_attrib.is_64, &msg);
199 		if (rc) {
200 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
201 			irq_dispose_mapping(virq);
202 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
203 			return rc;
204 		}
205 		irq_set_msi_desc(virq, entry);
206 		pci_write_msi_msg(virq, &msg);
207 	}
208 	return 0;
209 }
210 
211 void pnv_teardown_msi_irqs(struct pci_dev *pdev)
212 {
213 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
214 	struct pnv_phb *phb = hose->private_data;
215 	struct msi_desc *entry;
216 	irq_hw_number_t hwirq;
217 
218 	if (WARN_ON(!phb))
219 		return;
220 
221 	for_each_pci_msi_entry(entry, pdev) {
222 		if (!entry->irq)
223 			continue;
224 		hwirq = virq_to_hw(entry->irq);
225 		irq_set_msi_desc(entry->irq, NULL);
226 		irq_dispose_mapping(entry->irq);
227 		msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
228 	}
229 }
230 #endif /* CONFIG_PCI_MSI */
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;
604 	__be16	pcierr;
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 static __be64 *pnv_tce(struct iommu_table *tbl, long idx)
804 {
805 	__be64 *tmp = ((__be64 *)tbl->it_base);
806 	int  level = tbl->it_indirect_levels;
807 	const long shift = ilog2(tbl->it_level_size);
808 	unsigned long mask = (tbl->it_level_size - 1) << (level * shift);
809 
810 	while (level) {
811 		int n = (idx & mask) >> (level * shift);
812 		unsigned long tce = be64_to_cpu(tmp[n]);
813 
814 		tmp = __va(tce & ~(TCE_PCI_READ | TCE_PCI_WRITE));
815 		idx &= ~mask;
816 		mask >>= shift;
817 		--level;
818 	}
819 
820 	return tmp + idx;
821 }
822 
823 int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
824 		unsigned long uaddr, enum dma_data_direction direction,
825 		unsigned long attrs)
826 {
827 	u64 proto_tce = iommu_direction_to_tce_perm(direction);
828 	u64 rpn = __pa(uaddr) >> tbl->it_page_shift;
829 	long i;
830 
831 	if (proto_tce & TCE_PCI_WRITE)
832 		proto_tce |= TCE_PCI_READ;
833 
834 	for (i = 0; i < npages; i++) {
835 		unsigned long newtce = proto_tce |
836 			((rpn + i) << tbl->it_page_shift);
837 		unsigned long idx = index - tbl->it_offset + i;
838 
839 		*(pnv_tce(tbl, idx)) = cpu_to_be64(newtce);
840 	}
841 
842 	return 0;
843 }
844 
845 #ifdef CONFIG_IOMMU_API
846 int pnv_tce_xchg(struct iommu_table *tbl, long index,
847 		unsigned long *hpa, enum dma_data_direction *direction)
848 {
849 	u64 proto_tce = iommu_direction_to_tce_perm(*direction);
850 	unsigned long newtce = *hpa | proto_tce, oldtce;
851 	unsigned long idx = index - tbl->it_offset;
852 
853 	BUG_ON(*hpa & ~IOMMU_PAGE_MASK(tbl));
854 
855 	if (newtce & TCE_PCI_WRITE)
856 		newtce |= TCE_PCI_READ;
857 
858 	oldtce = be64_to_cpu(xchg(pnv_tce(tbl, idx), cpu_to_be64(newtce)));
859 	*hpa = oldtce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
860 	*direction = iommu_tce_direction(oldtce);
861 
862 	return 0;
863 }
864 #endif
865 
866 void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
867 {
868 	long i;
869 
870 	for (i = 0; i < npages; i++) {
871 		unsigned long idx = index - tbl->it_offset + i;
872 
873 		*(pnv_tce(tbl, idx)) = cpu_to_be64(0);
874 	}
875 }
876 
877 unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
878 {
879 	return be64_to_cpu(*(pnv_tce(tbl, index - tbl->it_offset)));
880 }
881 
882 struct iommu_table *pnv_pci_table_alloc(int nid)
883 {
884 	struct iommu_table *tbl;
885 
886 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
887 	if (!tbl)
888 		return NULL;
889 
890 	INIT_LIST_HEAD_RCU(&tbl->it_group_list);
891 	kref_init(&tbl->it_kref);
892 
893 	return tbl;
894 }
895 
896 long pnv_pci_link_table_and_group(int node, int num,
897 		struct iommu_table *tbl,
898 		struct iommu_table_group *table_group)
899 {
900 	struct iommu_table_group_link *tgl = NULL;
901 
902 	if (WARN_ON(!tbl || !table_group))
903 		return -EINVAL;
904 
905 	tgl = kzalloc_node(sizeof(struct iommu_table_group_link), GFP_KERNEL,
906 			node);
907 	if (!tgl)
908 		return -ENOMEM;
909 
910 	tgl->table_group = table_group;
911 	list_add_rcu(&tgl->next, &tbl->it_group_list);
912 
913 	table_group->tables[num] = tbl;
914 
915 	return 0;
916 }
917 
918 static void pnv_iommu_table_group_link_free(struct rcu_head *head)
919 {
920 	struct iommu_table_group_link *tgl = container_of(head,
921 			struct iommu_table_group_link, rcu);
922 
923 	kfree(tgl);
924 }
925 
926 void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
927 		struct iommu_table_group *table_group)
928 {
929 	long i;
930 	bool found;
931 	struct iommu_table_group_link *tgl;
932 
933 	if (!tbl || !table_group)
934 		return;
935 
936 	/* Remove link to a group from table's list of attached groups */
937 	found = false;
938 	list_for_each_entry_rcu(tgl, &tbl->it_group_list, next) {
939 		if (tgl->table_group == table_group) {
940 			list_del_rcu(&tgl->next);
941 			call_rcu(&tgl->rcu, pnv_iommu_table_group_link_free);
942 			found = true;
943 			break;
944 		}
945 	}
946 	if (WARN_ON(!found))
947 		return;
948 
949 	/* Clean a pointer to iommu_table in iommu_table_group::tables[] */
950 	found = false;
951 	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
952 		if (table_group->tables[i] == tbl) {
953 			table_group->tables[i] = NULL;
954 			found = true;
955 			break;
956 		}
957 	}
958 	WARN_ON(!found);
959 }
960 
961 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
962 			       void *tce_mem, u64 tce_size,
963 			       u64 dma_offset, unsigned page_shift)
964 {
965 	tbl->it_blocksize = 16;
966 	tbl->it_base = (unsigned long)tce_mem;
967 	tbl->it_page_shift = page_shift;
968 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
969 	tbl->it_index = 0;
970 	tbl->it_size = tce_size >> 3;
971 	tbl->it_busno = 0;
972 	tbl->it_type = TCE_PCI;
973 }
974 
975 void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
976 {
977 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
978 	struct pnv_phb *phb = hose->private_data;
979 #ifdef CONFIG_PCI_IOV
980 	struct pnv_ioda_pe *pe;
981 	struct pci_dn *pdn;
982 
983 	/* Fix the VF pdn PE number */
984 	if (pdev->is_virtfn) {
985 		pdn = pci_get_pdn(pdev);
986 		WARN_ON(pdn->pe_number != IODA_INVALID_PE);
987 		list_for_each_entry(pe, &phb->ioda.pe_list, list) {
988 			if (pe->rid == ((pdev->bus->number << 8) |
989 			    (pdev->devfn & 0xff))) {
990 				pdn->pe_number = pe->pe_number;
991 				pe->pdev = pdev;
992 				break;
993 			}
994 		}
995 	}
996 #endif /* CONFIG_PCI_IOV */
997 
998 	if (phb && phb->dma_dev_setup)
999 		phb->dma_dev_setup(phb, pdev);
1000 }
1001 
1002 void pnv_pci_dma_bus_setup(struct pci_bus *bus)
1003 {
1004 	struct pci_controller *hose = bus->sysdata;
1005 	struct pnv_phb *phb = hose->private_data;
1006 	struct pnv_ioda_pe *pe;
1007 
1008 	list_for_each_entry(pe, &phb->ioda.pe_list, list) {
1009 		if (!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)))
1010 			continue;
1011 
1012 		if (!pe->pbus)
1013 			continue;
1014 
1015 		if (bus->number == ((pe->rid >> 8) & 0xFF)) {
1016 			pe->pbus = bus;
1017 			break;
1018 		}
1019 	}
1020 }
1021 
1022 int pnv_pci_set_p2p(struct pci_dev *initiator, struct pci_dev *target, u64 desc)
1023 {
1024 	struct pci_controller *hose;
1025 	struct pnv_phb *phb_init, *phb_target;
1026 	struct pnv_ioda_pe *pe_init;
1027 	int rc;
1028 
1029 	if (!opal_check_token(OPAL_PCI_SET_P2P))
1030 		return -ENXIO;
1031 
1032 	hose = pci_bus_to_host(initiator->bus);
1033 	phb_init = hose->private_data;
1034 
1035 	hose = pci_bus_to_host(target->bus);
1036 	phb_target = hose->private_data;
1037 
1038 	pe_init = pnv_ioda_get_pe(initiator);
1039 	if (!pe_init)
1040 		return -ENODEV;
1041 
1042 	/*
1043 	 * Configuring the initiator's PHB requires to adjust its
1044 	 * TVE#1 setting. Since the same device can be an initiator
1045 	 * several times for different target devices, we need to keep
1046 	 * a reference count to know when we can restore the default
1047 	 * bypass setting on its TVE#1 when disabling. Opal is not
1048 	 * tracking PE states, so we add a reference count on the PE
1049 	 * in linux.
1050 	 *
1051 	 * For the target, the configuration is per PHB, so we keep a
1052 	 * target reference count on the PHB.
1053 	 */
1054 	mutex_lock(&p2p_mutex);
1055 
1056 	if (desc & OPAL_PCI_P2P_ENABLE) {
1057 		/* always go to opal to validate the configuration */
1058 		rc = opal_pci_set_p2p(phb_init->opal_id, phb_target->opal_id,
1059 				      desc, pe_init->pe_number);
1060 
1061 		if (rc != OPAL_SUCCESS) {
1062 			rc = -EIO;
1063 			goto out;
1064 		}
1065 
1066 		pe_init->p2p_initiator_count++;
1067 		phb_target->p2p_target_count++;
1068 	} else {
1069 		if (!pe_init->p2p_initiator_count ||
1070 			!phb_target->p2p_target_count) {
1071 			rc = -EINVAL;
1072 			goto out;
1073 		}
1074 
1075 		if (--pe_init->p2p_initiator_count == 0)
1076 			pnv_pci_ioda2_set_bypass(pe_init, true);
1077 
1078 		if (--phb_target->p2p_target_count == 0) {
1079 			rc = opal_pci_set_p2p(phb_init->opal_id,
1080 					      phb_target->opal_id, desc,
1081 					      pe_init->pe_number);
1082 			if (rc != OPAL_SUCCESS) {
1083 				rc = -EIO;
1084 				goto out;
1085 			}
1086 		}
1087 	}
1088 	rc = 0;
1089 out:
1090 	mutex_unlock(&p2p_mutex);
1091 	return rc;
1092 }
1093 EXPORT_SYMBOL_GPL(pnv_pci_set_p2p);
1094 
1095 void pnv_pci_shutdown(void)
1096 {
1097 	struct pci_controller *hose;
1098 
1099 	list_for_each_entry(hose, &hose_list, list_node)
1100 		if (hose->controller_ops.shutdown)
1101 			hose->controller_ops.shutdown(hose);
1102 }
1103 
1104 /* Fixup wrong class code in p7ioc and p8 root complex */
1105 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
1106 {
1107 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
1108 }
1109 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
1110 
1111 void __init pnv_pci_init(void)
1112 {
1113 	struct device_node *np;
1114 
1115 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
1116 
1117 	/* If we don't have OPAL, eg. in sim, just skip PCI probe */
1118 	if (!firmware_has_feature(FW_FEATURE_OPAL))
1119 		return;
1120 
1121 	/* Look for IODA IO-Hubs. */
1122 	for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
1123 		pnv_pci_init_ioda_hub(np);
1124 	}
1125 
1126 	/* Look for ioda2 built-in PHB3's */
1127 	for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
1128 		pnv_pci_init_ioda2_phb(np);
1129 
1130 	/* Look for ioda3 built-in PHB4's, we treat them as IODA2 */
1131 	for_each_compatible_node(np, NULL, "ibm,ioda3-phb")
1132 		pnv_pci_init_ioda2_phb(np);
1133 
1134 	/* Look for NPU PHBs */
1135 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu-phb")
1136 		pnv_pci_init_npu_phb(np);
1137 
1138 	/*
1139 	 * Look for NPU2 PHBs which we treat mostly as NPU PHBs with
1140 	 * the exception of TCE kill which requires an OPAL call.
1141 	 */
1142 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-phb")
1143 		pnv_pci_init_npu_phb(np);
1144 
1145 	/* Look for NPU2 OpenCAPI PHBs */
1146 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-opencapi-phb")
1147 		pnv_pci_init_npu2_opencapi_phb(np);
1148 
1149 	/* Configure IOMMU DMA hooks */
1150 	set_pci_dma_ops(&dma_iommu_ops);
1151 }
1152 
1153 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
1154