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 /* Delay in usec */
40 #define PCI_RESET_DELAY_US	3000000
41 
42 #define cfg_dbg(fmt...)	do { } while(0)
43 //#define cfg_dbg(fmt...)	printk(fmt)
44 
45 #ifdef CONFIG_PCI_MSI
46 int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
47 {
48 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
49 	struct pnv_phb *phb = hose->private_data;
50 	struct msi_desc *entry;
51 	struct msi_msg msg;
52 	int hwirq;
53 	unsigned int virq;
54 	int rc;
55 
56 	if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
57 		return -ENODEV;
58 
59 	if (pdev->no_64bit_msi && !phb->msi32_support)
60 		return -ENODEV;
61 
62 	for_each_pci_msi_entry(entry, pdev) {
63 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
64 			pr_warn("%s: Supports only 64-bit MSIs\n",
65 				pci_name(pdev));
66 			return -ENXIO;
67 		}
68 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
69 		if (hwirq < 0) {
70 			pr_warn("%s: Failed to find a free MSI\n",
71 				pci_name(pdev));
72 			return -ENOSPC;
73 		}
74 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
75 		if (virq == NO_IRQ) {
76 			pr_warn("%s: Failed to map MSI to linux irq\n",
77 				pci_name(pdev));
78 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
79 			return -ENOMEM;
80 		}
81 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
82 				    virq, entry->msi_attrib.is_64, &msg);
83 		if (rc) {
84 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
85 			irq_dispose_mapping(virq);
86 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
87 			return rc;
88 		}
89 		irq_set_msi_desc(virq, entry);
90 		pci_write_msi_msg(virq, &msg);
91 	}
92 	return 0;
93 }
94 
95 void pnv_teardown_msi_irqs(struct pci_dev *pdev)
96 {
97 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
98 	struct pnv_phb *phb = hose->private_data;
99 	struct msi_desc *entry;
100 	irq_hw_number_t hwirq;
101 
102 	if (WARN_ON(!phb))
103 		return;
104 
105 	for_each_pci_msi_entry(entry, pdev) {
106 		if (entry->irq == NO_IRQ)
107 			continue;
108 		hwirq = virq_to_hw(entry->irq);
109 		irq_set_msi_desc(entry->irq, NULL);
110 		irq_dispose_mapping(entry->irq);
111 		msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
112 	}
113 }
114 #endif /* CONFIG_PCI_MSI */
115 
116 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
117 					 struct OpalIoPhbErrorCommon *common)
118 {
119 	struct OpalIoP7IOCPhbErrorData *data;
120 	int i;
121 
122 	data = (struct OpalIoP7IOCPhbErrorData *)common;
123 	pr_info("P7IOC PHB#%d Diag-data (Version: %d)\n",
124 		hose->global_number, be32_to_cpu(common->version));
125 
126 	if (data->brdgCtl)
127 		pr_info("brdgCtl:     %08x\n",
128 			be32_to_cpu(data->brdgCtl));
129 	if (data->portStatusReg || data->rootCmplxStatus ||
130 	    data->busAgentStatus)
131 		pr_info("UtlSts:      %08x %08x %08x\n",
132 			be32_to_cpu(data->portStatusReg),
133 			be32_to_cpu(data->rootCmplxStatus),
134 			be32_to_cpu(data->busAgentStatus));
135 	if (data->deviceStatus || data->slotStatus   ||
136 	    data->linkStatus   || data->devCmdStatus ||
137 	    data->devSecStatus)
138 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
139 			be32_to_cpu(data->deviceStatus),
140 			be32_to_cpu(data->slotStatus),
141 			be32_to_cpu(data->linkStatus),
142 			be32_to_cpu(data->devCmdStatus),
143 			be32_to_cpu(data->devSecStatus));
144 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
145 	    data->corrErrorStatus)
146 		pr_info("RootErrSts:  %08x %08x %08x\n",
147 			be32_to_cpu(data->rootErrorStatus),
148 			be32_to_cpu(data->uncorrErrorStatus),
149 			be32_to_cpu(data->corrErrorStatus));
150 	if (data->tlpHdr1 || data->tlpHdr2 ||
151 	    data->tlpHdr3 || data->tlpHdr4)
152 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
153 			be32_to_cpu(data->tlpHdr1),
154 			be32_to_cpu(data->tlpHdr2),
155 			be32_to_cpu(data->tlpHdr3),
156 			be32_to_cpu(data->tlpHdr4));
157 	if (data->sourceId || data->errorClass ||
158 	    data->correlator)
159 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
160 			be32_to_cpu(data->sourceId),
161 			be64_to_cpu(data->errorClass),
162 			be64_to_cpu(data->correlator));
163 	if (data->p7iocPlssr || data->p7iocCsr)
164 		pr_info("PhbSts:      %016llx %016llx\n",
165 			be64_to_cpu(data->p7iocPlssr),
166 			be64_to_cpu(data->p7iocCsr));
167 	if (data->lemFir)
168 		pr_info("Lem:         %016llx %016llx %016llx\n",
169 			be64_to_cpu(data->lemFir),
170 			be64_to_cpu(data->lemErrorMask),
171 			be64_to_cpu(data->lemWOF));
172 	if (data->phbErrorStatus)
173 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
174 			be64_to_cpu(data->phbErrorStatus),
175 			be64_to_cpu(data->phbFirstErrorStatus),
176 			be64_to_cpu(data->phbErrorLog0),
177 			be64_to_cpu(data->phbErrorLog1));
178 	if (data->mmioErrorStatus)
179 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
180 			be64_to_cpu(data->mmioErrorStatus),
181 			be64_to_cpu(data->mmioFirstErrorStatus),
182 			be64_to_cpu(data->mmioErrorLog0),
183 			be64_to_cpu(data->mmioErrorLog1));
184 	if (data->dma0ErrorStatus)
185 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
186 			be64_to_cpu(data->dma0ErrorStatus),
187 			be64_to_cpu(data->dma0FirstErrorStatus),
188 			be64_to_cpu(data->dma0ErrorLog0),
189 			be64_to_cpu(data->dma0ErrorLog1));
190 	if (data->dma1ErrorStatus)
191 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
192 			be64_to_cpu(data->dma1ErrorStatus),
193 			be64_to_cpu(data->dma1FirstErrorStatus),
194 			be64_to_cpu(data->dma1ErrorLog0),
195 			be64_to_cpu(data->dma1ErrorLog1));
196 
197 	for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
198 		if ((data->pestA[i] >> 63) == 0 &&
199 		    (data->pestB[i] >> 63) == 0)
200 			continue;
201 
202 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
203 			i, be64_to_cpu(data->pestA[i]),
204 			be64_to_cpu(data->pestB[i]));
205 	}
206 }
207 
208 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
209 					struct OpalIoPhbErrorCommon *common)
210 {
211 	struct OpalIoPhb3ErrorData *data;
212 	int i;
213 
214 	data = (struct OpalIoPhb3ErrorData*)common;
215 	pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n",
216 		hose->global_number, be32_to_cpu(common->version));
217 	if (data->brdgCtl)
218 		pr_info("brdgCtl:     %08x\n",
219 			be32_to_cpu(data->brdgCtl));
220 	if (data->portStatusReg || data->rootCmplxStatus ||
221 	    data->busAgentStatus)
222 		pr_info("UtlSts:      %08x %08x %08x\n",
223 			be32_to_cpu(data->portStatusReg),
224 			be32_to_cpu(data->rootCmplxStatus),
225 			be32_to_cpu(data->busAgentStatus));
226 	if (data->deviceStatus || data->slotStatus   ||
227 	    data->linkStatus   || data->devCmdStatus ||
228 	    data->devSecStatus)
229 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
230 			be32_to_cpu(data->deviceStatus),
231 			be32_to_cpu(data->slotStatus),
232 			be32_to_cpu(data->linkStatus),
233 			be32_to_cpu(data->devCmdStatus),
234 			be32_to_cpu(data->devSecStatus));
235 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
236 	    data->corrErrorStatus)
237 		pr_info("RootErrSts:  %08x %08x %08x\n",
238 			be32_to_cpu(data->rootErrorStatus),
239 			be32_to_cpu(data->uncorrErrorStatus),
240 			be32_to_cpu(data->corrErrorStatus));
241 	if (data->tlpHdr1 || data->tlpHdr2 ||
242 	    data->tlpHdr3 || data->tlpHdr4)
243 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
244 			be32_to_cpu(data->tlpHdr1),
245 			be32_to_cpu(data->tlpHdr2),
246 			be32_to_cpu(data->tlpHdr3),
247 			be32_to_cpu(data->tlpHdr4));
248 	if (data->sourceId || data->errorClass ||
249 	    data->correlator)
250 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
251 			be32_to_cpu(data->sourceId),
252 			be64_to_cpu(data->errorClass),
253 			be64_to_cpu(data->correlator));
254 	if (data->nFir)
255 		pr_info("nFir:        %016llx %016llx %016llx\n",
256 			be64_to_cpu(data->nFir),
257 			be64_to_cpu(data->nFirMask),
258 			be64_to_cpu(data->nFirWOF));
259 	if (data->phbPlssr || data->phbCsr)
260 		pr_info("PhbSts:      %016llx %016llx\n",
261 			be64_to_cpu(data->phbPlssr),
262 			be64_to_cpu(data->phbCsr));
263 	if (data->lemFir)
264 		pr_info("Lem:         %016llx %016llx %016llx\n",
265 			be64_to_cpu(data->lemFir),
266 			be64_to_cpu(data->lemErrorMask),
267 			be64_to_cpu(data->lemWOF));
268 	if (data->phbErrorStatus)
269 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
270 			be64_to_cpu(data->phbErrorStatus),
271 			be64_to_cpu(data->phbFirstErrorStatus),
272 			be64_to_cpu(data->phbErrorLog0),
273 			be64_to_cpu(data->phbErrorLog1));
274 	if (data->mmioErrorStatus)
275 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
276 			be64_to_cpu(data->mmioErrorStatus),
277 			be64_to_cpu(data->mmioFirstErrorStatus),
278 			be64_to_cpu(data->mmioErrorLog0),
279 			be64_to_cpu(data->mmioErrorLog1));
280 	if (data->dma0ErrorStatus)
281 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
282 			be64_to_cpu(data->dma0ErrorStatus),
283 			be64_to_cpu(data->dma0FirstErrorStatus),
284 			be64_to_cpu(data->dma0ErrorLog0),
285 			be64_to_cpu(data->dma0ErrorLog1));
286 	if (data->dma1ErrorStatus)
287 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
288 			be64_to_cpu(data->dma1ErrorStatus),
289 			be64_to_cpu(data->dma1FirstErrorStatus),
290 			be64_to_cpu(data->dma1ErrorLog0),
291 			be64_to_cpu(data->dma1ErrorLog1));
292 
293 	for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
294 		if ((be64_to_cpu(data->pestA[i]) >> 63) == 0 &&
295 		    (be64_to_cpu(data->pestB[i]) >> 63) == 0)
296 			continue;
297 
298 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
299 				i, be64_to_cpu(data->pestA[i]),
300 				be64_to_cpu(data->pestB[i]));
301 	}
302 }
303 
304 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
305 				unsigned char *log_buff)
306 {
307 	struct OpalIoPhbErrorCommon *common;
308 
309 	if (!hose || !log_buff)
310 		return;
311 
312 	common = (struct OpalIoPhbErrorCommon *)log_buff;
313 	switch (be32_to_cpu(common->ioType)) {
314 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
315 		pnv_pci_dump_p7ioc_diag_data(hose, common);
316 		break;
317 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
318 		pnv_pci_dump_phb3_diag_data(hose, common);
319 		break;
320 	default:
321 		pr_warn("%s: Unrecognized ioType %d\n",
322 			__func__, be32_to_cpu(common->ioType));
323 	}
324 }
325 
326 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
327 {
328 	unsigned long flags, rc;
329 	int has_diag, ret = 0;
330 
331 	spin_lock_irqsave(&phb->lock, flags);
332 
333 	/* Fetch PHB diag-data */
334 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
335 					 PNV_PCI_DIAG_BUF_SIZE);
336 	has_diag = (rc == OPAL_SUCCESS);
337 
338 	/* If PHB supports compound PE, to handle it */
339 	if (phb->unfreeze_pe) {
340 		ret = phb->unfreeze_pe(phb,
341 				       pe_no,
342 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
343 	} else {
344 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
345 					     pe_no,
346 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
347 		if (rc) {
348 			pr_warn("%s: Failure %ld clearing frozen "
349 				"PHB#%x-PE#%x\n",
350 				__func__, rc, phb->hose->global_number,
351 				pe_no);
352 			ret = -EIO;
353 		}
354 	}
355 
356 	/*
357 	 * For now, let's only display the diag buffer when we fail to clear
358 	 * the EEH status. We'll do more sensible things later when we have
359 	 * proper EEH support. We need to make sure we don't pollute ourselves
360 	 * with the normal errors generated when probing empty slots
361 	 */
362 	if (has_diag && ret)
363 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
364 
365 	spin_unlock_irqrestore(&phb->lock, flags);
366 }
367 
368 static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
369 {
370 	struct pnv_phb *phb = pdn->phb->private_data;
371 	u8	fstate;
372 	__be16	pcierr;
373 	int	pe_no;
374 	s64	rc;
375 
376 	/*
377 	 * Get the PE#. During the PCI probe stage, we might not
378 	 * setup that yet. So all ER errors should be mapped to
379 	 * reserved PE.
380 	 */
381 	pe_no = pdn->pe_number;
382 	if (pe_no == IODA_INVALID_PE) {
383 		if (phb->type == PNV_PHB_P5IOC2)
384 			pe_no = 0;
385 		else
386 			pe_no = phb->ioda.reserved_pe;
387 	}
388 
389 	/*
390 	 * Fetch frozen state. If the PHB support compound PE,
391 	 * we need handle that case.
392 	 */
393 	if (phb->get_pe_state) {
394 		fstate = phb->get_pe_state(phb, pe_no);
395 	} else {
396 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
397 						pe_no,
398 						&fstate,
399 						&pcierr,
400 						NULL);
401 		if (rc) {
402 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
403 				__func__, rc, phb->hose->global_number, pe_no);
404 			return;
405 		}
406 	}
407 
408 	cfg_dbg(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
409 		(pdn->busno << 8) | (pdn->devfn), pe_no, fstate);
410 
411 	/* Clear the frozen state if applicable */
412 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
413 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
414 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
415 		/*
416 		 * If PHB supports compound PE, freeze it for
417 		 * consistency.
418 		 */
419 		if (phb->freeze_pe)
420 			phb->freeze_pe(phb, pe_no);
421 
422 		pnv_pci_handle_eeh_config(phb, pe_no);
423 	}
424 }
425 
426 int pnv_pci_cfg_read(struct pci_dn *pdn,
427 		     int where, int size, u32 *val)
428 {
429 	struct pnv_phb *phb = pdn->phb->private_data;
430 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
431 	s64 rc;
432 
433 	switch (size) {
434 	case 1: {
435 		u8 v8;
436 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
437 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
438 		break;
439 	}
440 	case 2: {
441 		__be16 v16;
442 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
443 						   &v16);
444 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
445 		break;
446 	}
447 	case 4: {
448 		__be32 v32;
449 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
450 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
451 		break;
452 	}
453 	default:
454 		return PCIBIOS_FUNC_NOT_SUPPORTED;
455 	}
456 
457 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
458 		__func__, pdn->busno, pdn->devfn, where, size, *val);
459 	return PCIBIOS_SUCCESSFUL;
460 }
461 
462 int pnv_pci_cfg_write(struct pci_dn *pdn,
463 		      int where, int size, u32 val)
464 {
465 	struct pnv_phb *phb = pdn->phb->private_data;
466 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
467 
468 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
469 		pdn->busno, pdn->devfn, where, size, val);
470 	switch (size) {
471 	case 1:
472 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
473 		break;
474 	case 2:
475 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
476 		break;
477 	case 4:
478 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
479 		break;
480 	default:
481 		return PCIBIOS_FUNC_NOT_SUPPORTED;
482 	}
483 
484 	return PCIBIOS_SUCCESSFUL;
485 }
486 
487 #if CONFIG_EEH
488 static bool pnv_pci_cfg_check(struct pci_dn *pdn)
489 {
490 	struct eeh_dev *edev = NULL;
491 	struct pnv_phb *phb = pdn->phb->private_data;
492 
493 	/* EEH not enabled ? */
494 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
495 		return true;
496 
497 	/* PE reset or device removed ? */
498 	edev = pdn->edev;
499 	if (edev) {
500 		if (edev->pe &&
501 		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
502 			return false;
503 
504 		if (edev->mode & EEH_DEV_REMOVED)
505 			return false;
506 	}
507 
508 	return true;
509 }
510 #else
511 static inline pnv_pci_cfg_check(struct pci_dn *pdn)
512 {
513 	return true;
514 }
515 #endif /* CONFIG_EEH */
516 
517 static int pnv_pci_read_config(struct pci_bus *bus,
518 			       unsigned int devfn,
519 			       int where, int size, u32 *val)
520 {
521 	struct pci_dn *pdn;
522 	struct pnv_phb *phb;
523 	int ret;
524 
525 	*val = 0xFFFFFFFF;
526 	pdn = pci_get_pdn_by_devfn(bus, devfn);
527 	if (!pdn)
528 		return PCIBIOS_DEVICE_NOT_FOUND;
529 
530 	if (!pnv_pci_cfg_check(pdn))
531 		return PCIBIOS_DEVICE_NOT_FOUND;
532 
533 	ret = pnv_pci_cfg_read(pdn, where, size, val);
534 	phb = pdn->phb->private_data;
535 	if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
536 		if (*val == EEH_IO_ERROR_VALUE(size) &&
537 		    eeh_dev_check_failure(pdn->edev))
538                         return PCIBIOS_DEVICE_NOT_FOUND;
539 	} else {
540 		pnv_pci_config_check_eeh(pdn);
541 	}
542 
543 	return ret;
544 }
545 
546 static int pnv_pci_write_config(struct pci_bus *bus,
547 				unsigned int devfn,
548 				int where, int size, u32 val)
549 {
550 	struct pci_dn *pdn;
551 	struct pnv_phb *phb;
552 	int ret;
553 
554 	pdn = pci_get_pdn_by_devfn(bus, devfn);
555 	if (!pdn)
556 		return PCIBIOS_DEVICE_NOT_FOUND;
557 
558 	if (!pnv_pci_cfg_check(pdn))
559 		return PCIBIOS_DEVICE_NOT_FOUND;
560 
561 	ret = pnv_pci_cfg_write(pdn, where, size, val);
562 	phb = pdn->phb->private_data;
563 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
564 		pnv_pci_config_check_eeh(pdn);
565 
566 	return ret;
567 }
568 
569 struct pci_ops pnv_pci_ops = {
570 	.read  = pnv_pci_read_config,
571 	.write = pnv_pci_write_config,
572 };
573 
574 static __be64 *pnv_tce(struct iommu_table *tbl, long idx)
575 {
576 	__be64 *tmp = ((__be64 *)tbl->it_base);
577 	int  level = tbl->it_indirect_levels;
578 	const long shift = ilog2(tbl->it_level_size);
579 	unsigned long mask = (tbl->it_level_size - 1) << (level * shift);
580 
581 	while (level) {
582 		int n = (idx & mask) >> (level * shift);
583 		unsigned long tce = be64_to_cpu(tmp[n]);
584 
585 		tmp = __va(tce & ~(TCE_PCI_READ | TCE_PCI_WRITE));
586 		idx &= ~mask;
587 		mask >>= shift;
588 		--level;
589 	}
590 
591 	return tmp + idx;
592 }
593 
594 int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
595 		unsigned long uaddr, enum dma_data_direction direction,
596 		struct dma_attrs *attrs)
597 {
598 	u64 proto_tce = iommu_direction_to_tce_perm(direction);
599 	u64 rpn = __pa(uaddr) >> tbl->it_page_shift;
600 	long i;
601 
602 	if (proto_tce & TCE_PCI_WRITE)
603 		proto_tce |= TCE_PCI_READ;
604 
605 	for (i = 0; i < npages; i++) {
606 		unsigned long newtce = proto_tce |
607 			((rpn + i) << tbl->it_page_shift);
608 		unsigned long idx = index - tbl->it_offset + i;
609 
610 		*(pnv_tce(tbl, idx)) = cpu_to_be64(newtce);
611 	}
612 
613 	return 0;
614 }
615 
616 #ifdef CONFIG_IOMMU_API
617 int pnv_tce_xchg(struct iommu_table *tbl, long index,
618 		unsigned long *hpa, enum dma_data_direction *direction)
619 {
620 	u64 proto_tce = iommu_direction_to_tce_perm(*direction);
621 	unsigned long newtce = *hpa | proto_tce, oldtce;
622 	unsigned long idx = index - tbl->it_offset;
623 
624 	BUG_ON(*hpa & ~IOMMU_PAGE_MASK(tbl));
625 
626 	if (newtce & TCE_PCI_WRITE)
627 		newtce |= TCE_PCI_READ;
628 
629 	oldtce = xchg(pnv_tce(tbl, idx), cpu_to_be64(newtce));
630 	*hpa = be64_to_cpu(oldtce) & ~(TCE_PCI_READ | TCE_PCI_WRITE);
631 	*direction = iommu_tce_direction(oldtce);
632 
633 	return 0;
634 }
635 #endif
636 
637 void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
638 {
639 	long i;
640 
641 	for (i = 0; i < npages; i++) {
642 		unsigned long idx = index - tbl->it_offset + i;
643 
644 		*(pnv_tce(tbl, idx)) = cpu_to_be64(0);
645 	}
646 }
647 
648 unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
649 {
650 	return *(pnv_tce(tbl, index - tbl->it_offset));
651 }
652 
653 struct iommu_table *pnv_pci_table_alloc(int nid)
654 {
655 	struct iommu_table *tbl;
656 
657 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
658 	INIT_LIST_HEAD_RCU(&tbl->it_group_list);
659 
660 	return tbl;
661 }
662 
663 long pnv_pci_link_table_and_group(int node, int num,
664 		struct iommu_table *tbl,
665 		struct iommu_table_group *table_group)
666 {
667 	struct iommu_table_group_link *tgl = NULL;
668 
669 	if (WARN_ON(!tbl || !table_group))
670 		return -EINVAL;
671 
672 	tgl = kzalloc_node(sizeof(struct iommu_table_group_link), GFP_KERNEL,
673 			node);
674 	if (!tgl)
675 		return -ENOMEM;
676 
677 	tgl->table_group = table_group;
678 	list_add_rcu(&tgl->next, &tbl->it_group_list);
679 
680 	table_group->tables[num] = tbl;
681 
682 	return 0;
683 }
684 
685 static void pnv_iommu_table_group_link_free(struct rcu_head *head)
686 {
687 	struct iommu_table_group_link *tgl = container_of(head,
688 			struct iommu_table_group_link, rcu);
689 
690 	kfree(tgl);
691 }
692 
693 void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
694 		struct iommu_table_group *table_group)
695 {
696 	long i;
697 	bool found;
698 	struct iommu_table_group_link *tgl;
699 
700 	if (!tbl || !table_group)
701 		return;
702 
703 	/* Remove link to a group from table's list of attached groups */
704 	found = false;
705 	list_for_each_entry_rcu(tgl, &tbl->it_group_list, next) {
706 		if (tgl->table_group == table_group) {
707 			list_del_rcu(&tgl->next);
708 			call_rcu(&tgl->rcu, pnv_iommu_table_group_link_free);
709 			found = true;
710 			break;
711 		}
712 	}
713 	if (WARN_ON(!found))
714 		return;
715 
716 	/* Clean a pointer to iommu_table in iommu_table_group::tables[] */
717 	found = false;
718 	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
719 		if (table_group->tables[i] == tbl) {
720 			table_group->tables[i] = NULL;
721 			found = true;
722 			break;
723 		}
724 	}
725 	WARN_ON(!found);
726 }
727 
728 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
729 			       void *tce_mem, u64 tce_size,
730 			       u64 dma_offset, unsigned page_shift)
731 {
732 	tbl->it_blocksize = 16;
733 	tbl->it_base = (unsigned long)tce_mem;
734 	tbl->it_page_shift = page_shift;
735 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
736 	tbl->it_index = 0;
737 	tbl->it_size = tce_size >> 3;
738 	tbl->it_busno = 0;
739 	tbl->it_type = TCE_PCI;
740 }
741 
742 void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
743 {
744 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
745 	struct pnv_phb *phb = hose->private_data;
746 #ifdef CONFIG_PCI_IOV
747 	struct pnv_ioda_pe *pe;
748 	struct pci_dn *pdn;
749 
750 	/* Fix the VF pdn PE number */
751 	if (pdev->is_virtfn) {
752 		pdn = pci_get_pdn(pdev);
753 		WARN_ON(pdn->pe_number != IODA_INVALID_PE);
754 		list_for_each_entry(pe, &phb->ioda.pe_list, list) {
755 			if (pe->rid == ((pdev->bus->number << 8) |
756 			    (pdev->devfn & 0xff))) {
757 				pdn->pe_number = pe->pe_number;
758 				pe->pdev = pdev;
759 				break;
760 			}
761 		}
762 	}
763 #endif /* CONFIG_PCI_IOV */
764 
765 	if (phb && phb->dma_dev_setup)
766 		phb->dma_dev_setup(phb, pdev);
767 }
768 
769 void pnv_pci_dma_bus_setup(struct pci_bus *bus)
770 {
771 	struct pci_controller *hose = bus->sysdata;
772 	struct pnv_phb *phb = hose->private_data;
773 	struct pnv_ioda_pe *pe;
774 
775 	list_for_each_entry(pe, &phb->ioda.pe_list, list) {
776 		if (!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)))
777 			continue;
778 
779 		if (!pe->pbus)
780 			continue;
781 
782 		if (bus->number == ((pe->rid >> 8) & 0xFF)) {
783 			pe->pbus = bus;
784 			break;
785 		}
786 	}
787 }
788 
789 void pnv_pci_shutdown(void)
790 {
791 	struct pci_controller *hose;
792 
793 	list_for_each_entry(hose, &hose_list, list_node)
794 		if (hose->controller_ops.shutdown)
795 			hose->controller_ops.shutdown(hose);
796 }
797 
798 /* Fixup wrong class code in p7ioc and p8 root complex */
799 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
800 {
801 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
802 }
803 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
804 
805 void __init pnv_pci_init(void)
806 {
807 	struct device_node *np;
808 	bool found_ioda = false;
809 
810 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
811 
812 	/* If we don't have OPAL, eg. in sim, just skip PCI probe */
813 	if (!firmware_has_feature(FW_FEATURE_OPAL))
814 		return;
815 
816 	/* Look for IODA IO-Hubs. We don't support mixing IODA
817 	 * and p5ioc2 due to the need to change some global
818 	 * probing flags
819 	 */
820 	for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
821 		pnv_pci_init_ioda_hub(np);
822 		found_ioda = true;
823 	}
824 
825 	/* Look for p5ioc2 IO-Hubs */
826 	if (!found_ioda)
827 		for_each_compatible_node(np, NULL, "ibm,p5ioc2")
828 			pnv_pci_init_p5ioc2_hub(np);
829 
830 	/* Look for ioda2 built-in PHB3's */
831 	for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
832 		pnv_pci_init_ioda2_phb(np);
833 
834 	/* Look for NPU PHBs */
835 	for_each_compatible_node(np, NULL, "ibm,ioda2-npu-phb")
836 		pnv_pci_init_npu_phb(np);
837 
838 	/* Setup the linkage between OF nodes and PHBs */
839 	pci_devs_phb_init();
840 
841 	/* Configure IOMMU DMA hooks */
842 	set_pci_dma_ops(&dma_iommu_ops);
843 }
844 
845 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
846