xref: /openbmc/linux/arch/powerpc/platforms/powernv/pci.c (revision 8cc7581cdb84a232468c41bc417183a423dfbb07)
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/opal.h>
30 #include <asm/iommu.h>
31 #include <asm/tce.h>
32 #include <asm/firmware.h>
33 #include <asm/eeh_event.h>
34 #include <asm/eeh.h>
35 
36 #include "powernv.h"
37 #include "pci.h"
38 
39 #ifdef CONFIG_PCI_MSI
40 int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
41 {
42 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
43 	struct pnv_phb *phb = hose->private_data;
44 	struct msi_desc *entry;
45 	struct msi_msg msg;
46 	int hwirq;
47 	unsigned int virq;
48 	int rc;
49 
50 	if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
51 		return -ENODEV;
52 
53 	if (pdev->no_64bit_msi && !phb->msi32_support)
54 		return -ENODEV;
55 
56 	for_each_pci_msi_entry(entry, pdev) {
57 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
58 			pr_warn("%s: Supports only 64-bit MSIs\n",
59 				pci_name(pdev));
60 			return -ENXIO;
61 		}
62 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
63 		if (hwirq < 0) {
64 			pr_warn("%s: Failed to find a free MSI\n",
65 				pci_name(pdev));
66 			return -ENOSPC;
67 		}
68 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
69 		if (virq == NO_IRQ) {
70 			pr_warn("%s: Failed to map MSI to linux irq\n",
71 				pci_name(pdev));
72 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
73 			return -ENOMEM;
74 		}
75 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
76 				    virq, entry->msi_attrib.is_64, &msg);
77 		if (rc) {
78 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
79 			irq_dispose_mapping(virq);
80 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
81 			return rc;
82 		}
83 		irq_set_msi_desc(virq, entry);
84 		pci_write_msi_msg(virq, &msg);
85 	}
86 	return 0;
87 }
88 
89 void pnv_teardown_msi_irqs(struct pci_dev *pdev)
90 {
91 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
92 	struct pnv_phb *phb = hose->private_data;
93 	struct msi_desc *entry;
94 	irq_hw_number_t hwirq;
95 
96 	if (WARN_ON(!phb))
97 		return;
98 
99 	for_each_pci_msi_entry(entry, pdev) {
100 		if (entry->irq == NO_IRQ)
101 			continue;
102 		hwirq = virq_to_hw(entry->irq);
103 		irq_set_msi_desc(entry->irq, NULL);
104 		irq_dispose_mapping(entry->irq);
105 		msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
106 	}
107 }
108 #endif /* CONFIG_PCI_MSI */
109 
110 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
111 					 struct OpalIoPhbErrorCommon *common)
112 {
113 	struct OpalIoP7IOCPhbErrorData *data;
114 	int i;
115 
116 	data = (struct OpalIoP7IOCPhbErrorData *)common;
117 	pr_info("P7IOC PHB#%d Diag-data (Version: %d)\n",
118 		hose->global_number, be32_to_cpu(common->version));
119 
120 	if (data->brdgCtl)
121 		pr_info("brdgCtl:     %08x\n",
122 			be32_to_cpu(data->brdgCtl));
123 	if (data->portStatusReg || data->rootCmplxStatus ||
124 	    data->busAgentStatus)
125 		pr_info("UtlSts:      %08x %08x %08x\n",
126 			be32_to_cpu(data->portStatusReg),
127 			be32_to_cpu(data->rootCmplxStatus),
128 			be32_to_cpu(data->busAgentStatus));
129 	if (data->deviceStatus || data->slotStatus   ||
130 	    data->linkStatus   || data->devCmdStatus ||
131 	    data->devSecStatus)
132 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
133 			be32_to_cpu(data->deviceStatus),
134 			be32_to_cpu(data->slotStatus),
135 			be32_to_cpu(data->linkStatus),
136 			be32_to_cpu(data->devCmdStatus),
137 			be32_to_cpu(data->devSecStatus));
138 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
139 	    data->corrErrorStatus)
140 		pr_info("RootErrSts:  %08x %08x %08x\n",
141 			be32_to_cpu(data->rootErrorStatus),
142 			be32_to_cpu(data->uncorrErrorStatus),
143 			be32_to_cpu(data->corrErrorStatus));
144 	if (data->tlpHdr1 || data->tlpHdr2 ||
145 	    data->tlpHdr3 || data->tlpHdr4)
146 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
147 			be32_to_cpu(data->tlpHdr1),
148 			be32_to_cpu(data->tlpHdr2),
149 			be32_to_cpu(data->tlpHdr3),
150 			be32_to_cpu(data->tlpHdr4));
151 	if (data->sourceId || data->errorClass ||
152 	    data->correlator)
153 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
154 			be32_to_cpu(data->sourceId),
155 			be64_to_cpu(data->errorClass),
156 			be64_to_cpu(data->correlator));
157 	if (data->p7iocPlssr || data->p7iocCsr)
158 		pr_info("PhbSts:      %016llx %016llx\n",
159 			be64_to_cpu(data->p7iocPlssr),
160 			be64_to_cpu(data->p7iocCsr));
161 	if (data->lemFir)
162 		pr_info("Lem:         %016llx %016llx %016llx\n",
163 			be64_to_cpu(data->lemFir),
164 			be64_to_cpu(data->lemErrorMask),
165 			be64_to_cpu(data->lemWOF));
166 	if (data->phbErrorStatus)
167 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
168 			be64_to_cpu(data->phbErrorStatus),
169 			be64_to_cpu(data->phbFirstErrorStatus),
170 			be64_to_cpu(data->phbErrorLog0),
171 			be64_to_cpu(data->phbErrorLog1));
172 	if (data->mmioErrorStatus)
173 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
174 			be64_to_cpu(data->mmioErrorStatus),
175 			be64_to_cpu(data->mmioFirstErrorStatus),
176 			be64_to_cpu(data->mmioErrorLog0),
177 			be64_to_cpu(data->mmioErrorLog1));
178 	if (data->dma0ErrorStatus)
179 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
180 			be64_to_cpu(data->dma0ErrorStatus),
181 			be64_to_cpu(data->dma0FirstErrorStatus),
182 			be64_to_cpu(data->dma0ErrorLog0),
183 			be64_to_cpu(data->dma0ErrorLog1));
184 	if (data->dma1ErrorStatus)
185 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
186 			be64_to_cpu(data->dma1ErrorStatus),
187 			be64_to_cpu(data->dma1FirstErrorStatus),
188 			be64_to_cpu(data->dma1ErrorLog0),
189 			be64_to_cpu(data->dma1ErrorLog1));
190 
191 	for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
192 		if ((data->pestA[i] >> 63) == 0 &&
193 		    (data->pestB[i] >> 63) == 0)
194 			continue;
195 
196 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
197 			i, be64_to_cpu(data->pestA[i]),
198 			be64_to_cpu(data->pestB[i]));
199 	}
200 }
201 
202 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
203 					struct OpalIoPhbErrorCommon *common)
204 {
205 	struct OpalIoPhb3ErrorData *data;
206 	int i;
207 
208 	data = (struct OpalIoPhb3ErrorData*)common;
209 	pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n",
210 		hose->global_number, be32_to_cpu(common->version));
211 	if (data->brdgCtl)
212 		pr_info("brdgCtl:     %08x\n",
213 			be32_to_cpu(data->brdgCtl));
214 	if (data->portStatusReg || data->rootCmplxStatus ||
215 	    data->busAgentStatus)
216 		pr_info("UtlSts:      %08x %08x %08x\n",
217 			be32_to_cpu(data->portStatusReg),
218 			be32_to_cpu(data->rootCmplxStatus),
219 			be32_to_cpu(data->busAgentStatus));
220 	if (data->deviceStatus || data->slotStatus   ||
221 	    data->linkStatus   || data->devCmdStatus ||
222 	    data->devSecStatus)
223 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
224 			be32_to_cpu(data->deviceStatus),
225 			be32_to_cpu(data->slotStatus),
226 			be32_to_cpu(data->linkStatus),
227 			be32_to_cpu(data->devCmdStatus),
228 			be32_to_cpu(data->devSecStatus));
229 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
230 	    data->corrErrorStatus)
231 		pr_info("RootErrSts:  %08x %08x %08x\n",
232 			be32_to_cpu(data->rootErrorStatus),
233 			be32_to_cpu(data->uncorrErrorStatus),
234 			be32_to_cpu(data->corrErrorStatus));
235 	if (data->tlpHdr1 || data->tlpHdr2 ||
236 	    data->tlpHdr3 || data->tlpHdr4)
237 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
238 			be32_to_cpu(data->tlpHdr1),
239 			be32_to_cpu(data->tlpHdr2),
240 			be32_to_cpu(data->tlpHdr3),
241 			be32_to_cpu(data->tlpHdr4));
242 	if (data->sourceId || data->errorClass ||
243 	    data->correlator)
244 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
245 			be32_to_cpu(data->sourceId),
246 			be64_to_cpu(data->errorClass),
247 			be64_to_cpu(data->correlator));
248 	if (data->nFir)
249 		pr_info("nFir:        %016llx %016llx %016llx\n",
250 			be64_to_cpu(data->nFir),
251 			be64_to_cpu(data->nFirMask),
252 			be64_to_cpu(data->nFirWOF));
253 	if (data->phbPlssr || data->phbCsr)
254 		pr_info("PhbSts:      %016llx %016llx\n",
255 			be64_to_cpu(data->phbPlssr),
256 			be64_to_cpu(data->phbCsr));
257 	if (data->lemFir)
258 		pr_info("Lem:         %016llx %016llx %016llx\n",
259 			be64_to_cpu(data->lemFir),
260 			be64_to_cpu(data->lemErrorMask),
261 			be64_to_cpu(data->lemWOF));
262 	if (data->phbErrorStatus)
263 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
264 			be64_to_cpu(data->phbErrorStatus),
265 			be64_to_cpu(data->phbFirstErrorStatus),
266 			be64_to_cpu(data->phbErrorLog0),
267 			be64_to_cpu(data->phbErrorLog1));
268 	if (data->mmioErrorStatus)
269 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
270 			be64_to_cpu(data->mmioErrorStatus),
271 			be64_to_cpu(data->mmioFirstErrorStatus),
272 			be64_to_cpu(data->mmioErrorLog0),
273 			be64_to_cpu(data->mmioErrorLog1));
274 	if (data->dma0ErrorStatus)
275 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
276 			be64_to_cpu(data->dma0ErrorStatus),
277 			be64_to_cpu(data->dma0FirstErrorStatus),
278 			be64_to_cpu(data->dma0ErrorLog0),
279 			be64_to_cpu(data->dma0ErrorLog1));
280 	if (data->dma1ErrorStatus)
281 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
282 			be64_to_cpu(data->dma1ErrorStatus),
283 			be64_to_cpu(data->dma1FirstErrorStatus),
284 			be64_to_cpu(data->dma1ErrorLog0),
285 			be64_to_cpu(data->dma1ErrorLog1));
286 
287 	for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
288 		if ((be64_to_cpu(data->pestA[i]) >> 63) == 0 &&
289 		    (be64_to_cpu(data->pestB[i]) >> 63) == 0)
290 			continue;
291 
292 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
293 				i, be64_to_cpu(data->pestA[i]),
294 				be64_to_cpu(data->pestB[i]));
295 	}
296 }
297 
298 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
299 				unsigned char *log_buff)
300 {
301 	struct OpalIoPhbErrorCommon *common;
302 
303 	if (!hose || !log_buff)
304 		return;
305 
306 	common = (struct OpalIoPhbErrorCommon *)log_buff;
307 	switch (be32_to_cpu(common->ioType)) {
308 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
309 		pnv_pci_dump_p7ioc_diag_data(hose, common);
310 		break;
311 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
312 		pnv_pci_dump_phb3_diag_data(hose, common);
313 		break;
314 	default:
315 		pr_warn("%s: Unrecognized ioType %d\n",
316 			__func__, be32_to_cpu(common->ioType));
317 	}
318 }
319 
320 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
321 {
322 	unsigned long flags, rc;
323 	int has_diag, ret = 0;
324 
325 	spin_lock_irqsave(&phb->lock, flags);
326 
327 	/* Fetch PHB diag-data */
328 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
329 					 PNV_PCI_DIAG_BUF_SIZE);
330 	has_diag = (rc == OPAL_SUCCESS);
331 
332 	/* If PHB supports compound PE, to handle it */
333 	if (phb->unfreeze_pe) {
334 		ret = phb->unfreeze_pe(phb,
335 				       pe_no,
336 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
337 	} else {
338 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
339 					     pe_no,
340 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
341 		if (rc) {
342 			pr_warn("%s: Failure %ld clearing frozen "
343 				"PHB#%x-PE#%x\n",
344 				__func__, rc, phb->hose->global_number,
345 				pe_no);
346 			ret = -EIO;
347 		}
348 	}
349 
350 	/*
351 	 * For now, let's only display the diag buffer when we fail to clear
352 	 * the EEH status. We'll do more sensible things later when we have
353 	 * proper EEH support. We need to make sure we don't pollute ourselves
354 	 * with the normal errors generated when probing empty slots
355 	 */
356 	if (has_diag && ret)
357 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
358 
359 	spin_unlock_irqrestore(&phb->lock, flags);
360 }
361 
362 static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
363 {
364 	struct pnv_phb *phb = pdn->phb->private_data;
365 	u8	fstate;
366 	__be16	pcierr;
367 	unsigned int pe_no;
368 	s64	rc;
369 
370 	/*
371 	 * Get the PE#. During the PCI probe stage, we might not
372 	 * setup that yet. So all ER errors should be mapped to
373 	 * reserved PE.
374 	 */
375 	pe_no = pdn->pe_number;
376 	if (pe_no == IODA_INVALID_PE) {
377 		pe_no = phb->ioda.reserved_pe_idx;
378 	}
379 
380 	/*
381 	 * Fetch frozen state. If the PHB support compound PE,
382 	 * we need handle that case.
383 	 */
384 	if (phb->get_pe_state) {
385 		fstate = phb->get_pe_state(phb, pe_no);
386 	} else {
387 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
388 						pe_no,
389 						&fstate,
390 						&pcierr,
391 						NULL);
392 		if (rc) {
393 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
394 				__func__, rc, phb->hose->global_number, pe_no);
395 			return;
396 		}
397 	}
398 
399 	pr_devel(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
400 		 (pdn->busno << 8) | (pdn->devfn), pe_no, fstate);
401 
402 	/* Clear the frozen state if applicable */
403 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
404 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
405 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
406 		/*
407 		 * If PHB supports compound PE, freeze it for
408 		 * consistency.
409 		 */
410 		if (phb->freeze_pe)
411 			phb->freeze_pe(phb, pe_no);
412 
413 		pnv_pci_handle_eeh_config(phb, pe_no);
414 	}
415 }
416 
417 int pnv_pci_cfg_read(struct pci_dn *pdn,
418 		     int where, int size, u32 *val)
419 {
420 	struct pnv_phb *phb = pdn->phb->private_data;
421 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
422 	s64 rc;
423 
424 	switch (size) {
425 	case 1: {
426 		u8 v8;
427 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
428 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
429 		break;
430 	}
431 	case 2: {
432 		__be16 v16;
433 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
434 						   &v16);
435 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
436 		break;
437 	}
438 	case 4: {
439 		__be32 v32;
440 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
441 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
442 		break;
443 	}
444 	default:
445 		return PCIBIOS_FUNC_NOT_SUPPORTED;
446 	}
447 
448 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
449 		 __func__, pdn->busno, pdn->devfn, where, size, *val);
450 	return PCIBIOS_SUCCESSFUL;
451 }
452 
453 int pnv_pci_cfg_write(struct pci_dn *pdn,
454 		      int where, int size, u32 val)
455 {
456 	struct pnv_phb *phb = pdn->phb->private_data;
457 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
458 
459 	pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
460 		 __func__, pdn->busno, pdn->devfn, where, size, val);
461 	switch (size) {
462 	case 1:
463 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
464 		break;
465 	case 2:
466 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
467 		break;
468 	case 4:
469 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
470 		break;
471 	default:
472 		return PCIBIOS_FUNC_NOT_SUPPORTED;
473 	}
474 
475 	return PCIBIOS_SUCCESSFUL;
476 }
477 
478 #if CONFIG_EEH
479 static bool pnv_pci_cfg_check(struct pci_dn *pdn)
480 {
481 	struct eeh_dev *edev = NULL;
482 	struct pnv_phb *phb = pdn->phb->private_data;
483 
484 	/* EEH not enabled ? */
485 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
486 		return true;
487 
488 	/* PE reset or device removed ? */
489 	edev = pdn->edev;
490 	if (edev) {
491 		if (edev->pe &&
492 		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
493 			return false;
494 
495 		if (edev->mode & EEH_DEV_REMOVED)
496 			return false;
497 	}
498 
499 	return true;
500 }
501 #else
502 static inline pnv_pci_cfg_check(struct pci_dn *pdn)
503 {
504 	return true;
505 }
506 #endif /* CONFIG_EEH */
507 
508 static int pnv_pci_read_config(struct pci_bus *bus,
509 			       unsigned int devfn,
510 			       int where, int size, u32 *val)
511 {
512 	struct pci_dn *pdn;
513 	struct pnv_phb *phb;
514 	int ret;
515 
516 	*val = 0xFFFFFFFF;
517 	pdn = pci_get_pdn_by_devfn(bus, devfn);
518 	if (!pdn)
519 		return PCIBIOS_DEVICE_NOT_FOUND;
520 
521 	if (!pnv_pci_cfg_check(pdn))
522 		return PCIBIOS_DEVICE_NOT_FOUND;
523 
524 	ret = pnv_pci_cfg_read(pdn, where, size, val);
525 	phb = pdn->phb->private_data;
526 	if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
527 		if (*val == EEH_IO_ERROR_VALUE(size) &&
528 		    eeh_dev_check_failure(pdn->edev))
529                         return PCIBIOS_DEVICE_NOT_FOUND;
530 	} else {
531 		pnv_pci_config_check_eeh(pdn);
532 	}
533 
534 	return ret;
535 }
536 
537 static int pnv_pci_write_config(struct pci_bus *bus,
538 				unsigned int devfn,
539 				int where, int size, u32 val)
540 {
541 	struct pci_dn *pdn;
542 	struct pnv_phb *phb;
543 	int ret;
544 
545 	pdn = pci_get_pdn_by_devfn(bus, devfn);
546 	if (!pdn)
547 		return PCIBIOS_DEVICE_NOT_FOUND;
548 
549 	if (!pnv_pci_cfg_check(pdn))
550 		return PCIBIOS_DEVICE_NOT_FOUND;
551 
552 	ret = pnv_pci_cfg_write(pdn, where, size, val);
553 	phb = pdn->phb->private_data;
554 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
555 		pnv_pci_config_check_eeh(pdn);
556 
557 	return ret;
558 }
559 
560 struct pci_ops pnv_pci_ops = {
561 	.read  = pnv_pci_read_config,
562 	.write = pnv_pci_write_config,
563 };
564 
565 static __be64 *pnv_tce(struct iommu_table *tbl, long idx)
566 {
567 	__be64 *tmp = ((__be64 *)tbl->it_base);
568 	int  level = tbl->it_indirect_levels;
569 	const long shift = ilog2(tbl->it_level_size);
570 	unsigned long mask = (tbl->it_level_size - 1) << (level * shift);
571 
572 	while (level) {
573 		int n = (idx & mask) >> (level * shift);
574 		unsigned long tce = be64_to_cpu(tmp[n]);
575 
576 		tmp = __va(tce & ~(TCE_PCI_READ | TCE_PCI_WRITE));
577 		idx &= ~mask;
578 		mask >>= shift;
579 		--level;
580 	}
581 
582 	return tmp + idx;
583 }
584 
585 int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
586 		unsigned long uaddr, enum dma_data_direction direction,
587 		struct dma_attrs *attrs)
588 {
589 	u64 proto_tce = iommu_direction_to_tce_perm(direction);
590 	u64 rpn = __pa(uaddr) >> tbl->it_page_shift;
591 	long i;
592 
593 	if (proto_tce & TCE_PCI_WRITE)
594 		proto_tce |= TCE_PCI_READ;
595 
596 	for (i = 0; i < npages; i++) {
597 		unsigned long newtce = proto_tce |
598 			((rpn + i) << tbl->it_page_shift);
599 		unsigned long idx = index - tbl->it_offset + i;
600 
601 		*(pnv_tce(tbl, idx)) = cpu_to_be64(newtce);
602 	}
603 
604 	return 0;
605 }
606 
607 #ifdef CONFIG_IOMMU_API
608 int pnv_tce_xchg(struct iommu_table *tbl, long index,
609 		unsigned long *hpa, enum dma_data_direction *direction)
610 {
611 	u64 proto_tce = iommu_direction_to_tce_perm(*direction);
612 	unsigned long newtce = *hpa | proto_tce, oldtce;
613 	unsigned long idx = index - tbl->it_offset;
614 
615 	BUG_ON(*hpa & ~IOMMU_PAGE_MASK(tbl));
616 
617 	if (newtce & TCE_PCI_WRITE)
618 		newtce |= TCE_PCI_READ;
619 
620 	oldtce = xchg(pnv_tce(tbl, idx), cpu_to_be64(newtce));
621 	*hpa = be64_to_cpu(oldtce) & ~(TCE_PCI_READ | TCE_PCI_WRITE);
622 	*direction = iommu_tce_direction(oldtce);
623 
624 	return 0;
625 }
626 #endif
627 
628 void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
629 {
630 	long i;
631 
632 	for (i = 0; i < npages; i++) {
633 		unsigned long idx = index - tbl->it_offset + i;
634 
635 		*(pnv_tce(tbl, idx)) = cpu_to_be64(0);
636 	}
637 }
638 
639 unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
640 {
641 	return *(pnv_tce(tbl, index - tbl->it_offset));
642 }
643 
644 struct iommu_table *pnv_pci_table_alloc(int nid)
645 {
646 	struct iommu_table *tbl;
647 
648 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
649 	INIT_LIST_HEAD_RCU(&tbl->it_group_list);
650 
651 	return tbl;
652 }
653 
654 long pnv_pci_link_table_and_group(int node, int num,
655 		struct iommu_table *tbl,
656 		struct iommu_table_group *table_group)
657 {
658 	struct iommu_table_group_link *tgl = NULL;
659 
660 	if (WARN_ON(!tbl || !table_group))
661 		return -EINVAL;
662 
663 	tgl = kzalloc_node(sizeof(struct iommu_table_group_link), GFP_KERNEL,
664 			node);
665 	if (!tgl)
666 		return -ENOMEM;
667 
668 	tgl->table_group = table_group;
669 	list_add_rcu(&tgl->next, &tbl->it_group_list);
670 
671 	table_group->tables[num] = tbl;
672 
673 	return 0;
674 }
675 
676 static void pnv_iommu_table_group_link_free(struct rcu_head *head)
677 {
678 	struct iommu_table_group_link *tgl = container_of(head,
679 			struct iommu_table_group_link, rcu);
680 
681 	kfree(tgl);
682 }
683 
684 void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
685 		struct iommu_table_group *table_group)
686 {
687 	long i;
688 	bool found;
689 	struct iommu_table_group_link *tgl;
690 
691 	if (!tbl || !table_group)
692 		return;
693 
694 	/* Remove link to a group from table's list of attached groups */
695 	found = false;
696 	list_for_each_entry_rcu(tgl, &tbl->it_group_list, next) {
697 		if (tgl->table_group == table_group) {
698 			list_del_rcu(&tgl->next);
699 			call_rcu(&tgl->rcu, pnv_iommu_table_group_link_free);
700 			found = true;
701 			break;
702 		}
703 	}
704 	if (WARN_ON(!found))
705 		return;
706 
707 	/* Clean a pointer to iommu_table in iommu_table_group::tables[] */
708 	found = false;
709 	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
710 		if (table_group->tables[i] == tbl) {
711 			table_group->tables[i] = NULL;
712 			found = true;
713 			break;
714 		}
715 	}
716 	WARN_ON(!found);
717 }
718 
719 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
720 			       void *tce_mem, u64 tce_size,
721 			       u64 dma_offset, unsigned page_shift)
722 {
723 	tbl->it_blocksize = 16;
724 	tbl->it_base = (unsigned long)tce_mem;
725 	tbl->it_page_shift = page_shift;
726 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
727 	tbl->it_index = 0;
728 	tbl->it_size = tce_size >> 3;
729 	tbl->it_busno = 0;
730 	tbl->it_type = TCE_PCI;
731 }
732 
733 void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
734 {
735 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
736 	struct pnv_phb *phb = hose->private_data;
737 #ifdef CONFIG_PCI_IOV
738 	struct pnv_ioda_pe *pe;
739 	struct pci_dn *pdn;
740 
741 	/* Fix the VF pdn PE number */
742 	if (pdev->is_virtfn) {
743 		pdn = pci_get_pdn(pdev);
744 		WARN_ON(pdn->pe_number != IODA_INVALID_PE);
745 		list_for_each_entry(pe, &phb->ioda.pe_list, list) {
746 			if (pe->rid == ((pdev->bus->number << 8) |
747 			    (pdev->devfn & 0xff))) {
748 				pdn->pe_number = pe->pe_number;
749 				pe->pdev = pdev;
750 				break;
751 			}
752 		}
753 	}
754 #endif /* CONFIG_PCI_IOV */
755 
756 	if (phb && phb->dma_dev_setup)
757 		phb->dma_dev_setup(phb, pdev);
758 }
759 
760 void pnv_pci_dma_bus_setup(struct pci_bus *bus)
761 {
762 	struct pci_controller *hose = bus->sysdata;
763 	struct pnv_phb *phb = hose->private_data;
764 	struct pnv_ioda_pe *pe;
765 
766 	list_for_each_entry(pe, &phb->ioda.pe_list, list) {
767 		if (!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)))
768 			continue;
769 
770 		if (!pe->pbus)
771 			continue;
772 
773 		if (bus->number == ((pe->rid >> 8) & 0xFF)) {
774 			pe->pbus = bus;
775 			break;
776 		}
777 	}
778 }
779 
780 void pnv_pci_shutdown(void)
781 {
782 	struct pci_controller *hose;
783 
784 	list_for_each_entry(hose, &hose_list, list_node)
785 		if (hose->controller_ops.shutdown)
786 			hose->controller_ops.shutdown(hose);
787 }
788 
789 /* Fixup wrong class code in p7ioc and p8 root complex */
790 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
791 {
792 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
793 }
794 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
795 
796 void __init pnv_pci_init(void)
797 {
798 	struct device_node *np;
799 
800 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
801 
802 	/* If we don't have OPAL, eg. in sim, just skip PCI probe */
803 	if (!firmware_has_feature(FW_FEATURE_OPAL))
804 		return;
805 
806 	/* Look for IODA IO-Hubs. */
807 	for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
808 		pnv_pci_init_ioda_hub(np);
809 	}
810 
811 	/* Look for ioda2 built-in PHB3's */
812 	for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
813 		pnv_pci_init_ioda2_phb(np);
814 
815 	/* Look for NPU PHBs */
816 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu-phb")
817 		pnv_pci_init_npu_phb(np);
818 
819 	/* Configure IOMMU DMA hooks */
820 	set_pci_dma_ops(&dma_iommu_ops);
821 }
822 
823 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
824