1 /*
2  * Support PCI/PCIe on PowerNV platforms
3  *
4  * Currently supports only P5IOC2
5  *
6  * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
7  *
8  * This program is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU General Public License
10  * as published by the Free Software Foundation; either version
11  * 2 of the License, or (at your option) any later version.
12  */
13 
14 #include <linux/kernel.h>
15 #include <linux/pci.h>
16 #include <linux/delay.h>
17 #include <linux/string.h>
18 #include <linux/init.h>
19 #include <linux/irq.h>
20 #include <linux/io.h>
21 #include <linux/msi.h>
22 #include <linux/iommu.h>
23 
24 #include <asm/sections.h>
25 #include <asm/io.h>
26 #include <asm/prom.h>
27 #include <asm/pci-bridge.h>
28 #include <asm/machdep.h>
29 #include <asm/msi_bitmap.h>
30 #include <asm/ppc-pci.h>
31 #include <asm/opal.h>
32 #include <asm/iommu.h>
33 #include <asm/tce.h>
34 #include <asm/firmware.h>
35 #include <asm/eeh_event.h>
36 #include <asm/eeh.h>
37 
38 #include "powernv.h"
39 #include "pci.h"
40 
41 /* Delay in usec */
42 #define PCI_RESET_DELAY_US	3000000
43 
44 #define cfg_dbg(fmt...)	do { } while(0)
45 //#define cfg_dbg(fmt...)	printk(fmt)
46 
47 #ifdef CONFIG_PCI_MSI
48 static int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
49 {
50 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
51 	struct pnv_phb *phb = hose->private_data;
52 	struct msi_desc *entry;
53 	struct msi_msg msg;
54 	int hwirq;
55 	unsigned int virq;
56 	int rc;
57 
58 	if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
59 		return -ENODEV;
60 
61 	if (pdev->no_64bit_msi && !phb->msi32_support)
62 		return -ENODEV;
63 
64 	list_for_each_entry(entry, &pdev->msi_list, list) {
65 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
66 			pr_warn("%s: Supports only 64-bit MSIs\n",
67 				pci_name(pdev));
68 			return -ENXIO;
69 		}
70 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
71 		if (hwirq < 0) {
72 			pr_warn("%s: Failed to find a free MSI\n",
73 				pci_name(pdev));
74 			return -ENOSPC;
75 		}
76 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
77 		if (virq == NO_IRQ) {
78 			pr_warn("%s: Failed to map MSI to linux irq\n",
79 				pci_name(pdev));
80 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
81 			return -ENOMEM;
82 		}
83 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
84 				    virq, entry->msi_attrib.is_64, &msg);
85 		if (rc) {
86 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
87 			irq_dispose_mapping(virq);
88 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
89 			return rc;
90 		}
91 		irq_set_msi_desc(virq, entry);
92 		pci_write_msi_msg(virq, &msg);
93 	}
94 	return 0;
95 }
96 
97 static void pnv_teardown_msi_irqs(struct pci_dev *pdev)
98 {
99 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
100 	struct pnv_phb *phb = hose->private_data;
101 	struct msi_desc *entry;
102 
103 	if (WARN_ON(!phb))
104 		return;
105 
106 	list_for_each_entry(entry, &pdev->msi_list, list) {
107 		if (entry->irq == NO_IRQ)
108 			continue;
109 		irq_set_msi_desc(entry->irq, NULL);
110 		msi_bitmap_free_hwirqs(&phb->msi_bmp,
111 			virq_to_hw(entry->irq) - phb->msi_base, 1);
112 		irq_dispose_mapping(entry->irq);
113 	}
114 }
115 #endif /* CONFIG_PCI_MSI */
116 
117 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
118 					 struct OpalIoPhbErrorCommon *common)
119 {
120 	struct OpalIoP7IOCPhbErrorData *data;
121 	int i;
122 
123 	data = (struct OpalIoP7IOCPhbErrorData *)common;
124 	pr_info("P7IOC PHB#%d Diag-data (Version: %d)\n",
125 		hose->global_number, be32_to_cpu(common->version));
126 
127 	if (data->brdgCtl)
128 		pr_info("brdgCtl:     %08x\n",
129 			be32_to_cpu(data->brdgCtl));
130 	if (data->portStatusReg || data->rootCmplxStatus ||
131 	    data->busAgentStatus)
132 		pr_info("UtlSts:      %08x %08x %08x\n",
133 			be32_to_cpu(data->portStatusReg),
134 			be32_to_cpu(data->rootCmplxStatus),
135 			be32_to_cpu(data->busAgentStatus));
136 	if (data->deviceStatus || data->slotStatus   ||
137 	    data->linkStatus   || data->devCmdStatus ||
138 	    data->devSecStatus)
139 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
140 			be32_to_cpu(data->deviceStatus),
141 			be32_to_cpu(data->slotStatus),
142 			be32_to_cpu(data->linkStatus),
143 			be32_to_cpu(data->devCmdStatus),
144 			be32_to_cpu(data->devSecStatus));
145 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
146 	    data->corrErrorStatus)
147 		pr_info("RootErrSts:  %08x %08x %08x\n",
148 			be32_to_cpu(data->rootErrorStatus),
149 			be32_to_cpu(data->uncorrErrorStatus),
150 			be32_to_cpu(data->corrErrorStatus));
151 	if (data->tlpHdr1 || data->tlpHdr2 ||
152 	    data->tlpHdr3 || data->tlpHdr4)
153 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
154 			be32_to_cpu(data->tlpHdr1),
155 			be32_to_cpu(data->tlpHdr2),
156 			be32_to_cpu(data->tlpHdr3),
157 			be32_to_cpu(data->tlpHdr4));
158 	if (data->sourceId || data->errorClass ||
159 	    data->correlator)
160 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
161 			be32_to_cpu(data->sourceId),
162 			be64_to_cpu(data->errorClass),
163 			be64_to_cpu(data->correlator));
164 	if (data->p7iocPlssr || data->p7iocCsr)
165 		pr_info("PhbSts:      %016llx %016llx\n",
166 			be64_to_cpu(data->p7iocPlssr),
167 			be64_to_cpu(data->p7iocCsr));
168 	if (data->lemFir)
169 		pr_info("Lem:         %016llx %016llx %016llx\n",
170 			be64_to_cpu(data->lemFir),
171 			be64_to_cpu(data->lemErrorMask),
172 			be64_to_cpu(data->lemWOF));
173 	if (data->phbErrorStatus)
174 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
175 			be64_to_cpu(data->phbErrorStatus),
176 			be64_to_cpu(data->phbFirstErrorStatus),
177 			be64_to_cpu(data->phbErrorLog0),
178 			be64_to_cpu(data->phbErrorLog1));
179 	if (data->mmioErrorStatus)
180 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
181 			be64_to_cpu(data->mmioErrorStatus),
182 			be64_to_cpu(data->mmioFirstErrorStatus),
183 			be64_to_cpu(data->mmioErrorLog0),
184 			be64_to_cpu(data->mmioErrorLog1));
185 	if (data->dma0ErrorStatus)
186 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
187 			be64_to_cpu(data->dma0ErrorStatus),
188 			be64_to_cpu(data->dma0FirstErrorStatus),
189 			be64_to_cpu(data->dma0ErrorLog0),
190 			be64_to_cpu(data->dma0ErrorLog1));
191 	if (data->dma1ErrorStatus)
192 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
193 			be64_to_cpu(data->dma1ErrorStatus),
194 			be64_to_cpu(data->dma1FirstErrorStatus),
195 			be64_to_cpu(data->dma1ErrorLog0),
196 			be64_to_cpu(data->dma1ErrorLog1));
197 
198 	for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
199 		if ((data->pestA[i] >> 63) == 0 &&
200 		    (data->pestB[i] >> 63) == 0)
201 			continue;
202 
203 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
204 			i, be64_to_cpu(data->pestA[i]),
205 			be64_to_cpu(data->pestB[i]));
206 	}
207 }
208 
209 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
210 					struct OpalIoPhbErrorCommon *common)
211 {
212 	struct OpalIoPhb3ErrorData *data;
213 	int i;
214 
215 	data = (struct OpalIoPhb3ErrorData*)common;
216 	pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n",
217 		hose->global_number, be32_to_cpu(common->version));
218 	if (data->brdgCtl)
219 		pr_info("brdgCtl:     %08x\n",
220 			be32_to_cpu(data->brdgCtl));
221 	if (data->portStatusReg || data->rootCmplxStatus ||
222 	    data->busAgentStatus)
223 		pr_info("UtlSts:      %08x %08x %08x\n",
224 			be32_to_cpu(data->portStatusReg),
225 			be32_to_cpu(data->rootCmplxStatus),
226 			be32_to_cpu(data->busAgentStatus));
227 	if (data->deviceStatus || data->slotStatus   ||
228 	    data->linkStatus   || data->devCmdStatus ||
229 	    data->devSecStatus)
230 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
231 			be32_to_cpu(data->deviceStatus),
232 			be32_to_cpu(data->slotStatus),
233 			be32_to_cpu(data->linkStatus),
234 			be32_to_cpu(data->devCmdStatus),
235 			be32_to_cpu(data->devSecStatus));
236 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
237 	    data->corrErrorStatus)
238 		pr_info("RootErrSts:  %08x %08x %08x\n",
239 			be32_to_cpu(data->rootErrorStatus),
240 			be32_to_cpu(data->uncorrErrorStatus),
241 			be32_to_cpu(data->corrErrorStatus));
242 	if (data->tlpHdr1 || data->tlpHdr2 ||
243 	    data->tlpHdr3 || data->tlpHdr4)
244 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
245 			be32_to_cpu(data->tlpHdr1),
246 			be32_to_cpu(data->tlpHdr2),
247 			be32_to_cpu(data->tlpHdr3),
248 			be32_to_cpu(data->tlpHdr4));
249 	if (data->sourceId || data->errorClass ||
250 	    data->correlator)
251 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
252 			be32_to_cpu(data->sourceId),
253 			be64_to_cpu(data->errorClass),
254 			be64_to_cpu(data->correlator));
255 	if (data->nFir)
256 		pr_info("nFir:        %016llx %016llx %016llx\n",
257 			be64_to_cpu(data->nFir),
258 			be64_to_cpu(data->nFirMask),
259 			be64_to_cpu(data->nFirWOF));
260 	if (data->phbPlssr || data->phbCsr)
261 		pr_info("PhbSts:      %016llx %016llx\n",
262 			be64_to_cpu(data->phbPlssr),
263 			be64_to_cpu(data->phbCsr));
264 	if (data->lemFir)
265 		pr_info("Lem:         %016llx %016llx %016llx\n",
266 			be64_to_cpu(data->lemFir),
267 			be64_to_cpu(data->lemErrorMask),
268 			be64_to_cpu(data->lemWOF));
269 	if (data->phbErrorStatus)
270 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
271 			be64_to_cpu(data->phbErrorStatus),
272 			be64_to_cpu(data->phbFirstErrorStatus),
273 			be64_to_cpu(data->phbErrorLog0),
274 			be64_to_cpu(data->phbErrorLog1));
275 	if (data->mmioErrorStatus)
276 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
277 			be64_to_cpu(data->mmioErrorStatus),
278 			be64_to_cpu(data->mmioFirstErrorStatus),
279 			be64_to_cpu(data->mmioErrorLog0),
280 			be64_to_cpu(data->mmioErrorLog1));
281 	if (data->dma0ErrorStatus)
282 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
283 			be64_to_cpu(data->dma0ErrorStatus),
284 			be64_to_cpu(data->dma0FirstErrorStatus),
285 			be64_to_cpu(data->dma0ErrorLog0),
286 			be64_to_cpu(data->dma0ErrorLog1));
287 	if (data->dma1ErrorStatus)
288 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
289 			be64_to_cpu(data->dma1ErrorStatus),
290 			be64_to_cpu(data->dma1FirstErrorStatus),
291 			be64_to_cpu(data->dma1ErrorLog0),
292 			be64_to_cpu(data->dma1ErrorLog1));
293 
294 	for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
295 		if ((be64_to_cpu(data->pestA[i]) >> 63) == 0 &&
296 		    (be64_to_cpu(data->pestB[i]) >> 63) == 0)
297 			continue;
298 
299 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
300 				i, be64_to_cpu(data->pestA[i]),
301 				be64_to_cpu(data->pestB[i]));
302 	}
303 }
304 
305 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
306 				unsigned char *log_buff)
307 {
308 	struct OpalIoPhbErrorCommon *common;
309 
310 	if (!hose || !log_buff)
311 		return;
312 
313 	common = (struct OpalIoPhbErrorCommon *)log_buff;
314 	switch (be32_to_cpu(common->ioType)) {
315 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
316 		pnv_pci_dump_p7ioc_diag_data(hose, common);
317 		break;
318 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
319 		pnv_pci_dump_phb3_diag_data(hose, common);
320 		break;
321 	default:
322 		pr_warn("%s: Unrecognized ioType %d\n",
323 			__func__, be32_to_cpu(common->ioType));
324 	}
325 }
326 
327 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
328 {
329 	unsigned long flags, rc;
330 	int has_diag, ret = 0;
331 
332 	spin_lock_irqsave(&phb->lock, flags);
333 
334 	/* Fetch PHB diag-data */
335 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
336 					 PNV_PCI_DIAG_BUF_SIZE);
337 	has_diag = (rc == OPAL_SUCCESS);
338 
339 	/* If PHB supports compound PE, to handle it */
340 	if (phb->unfreeze_pe) {
341 		ret = phb->unfreeze_pe(phb,
342 				       pe_no,
343 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
344 	} else {
345 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
346 					     pe_no,
347 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
348 		if (rc) {
349 			pr_warn("%s: Failure %ld clearing frozen "
350 				"PHB#%x-PE#%x\n",
351 				__func__, rc, phb->hose->global_number,
352 				pe_no);
353 			ret = -EIO;
354 		}
355 	}
356 
357 	/*
358 	 * For now, let's only display the diag buffer when we fail to clear
359 	 * the EEH status. We'll do more sensible things later when we have
360 	 * proper EEH support. We need to make sure we don't pollute ourselves
361 	 * with the normal errors generated when probing empty slots
362 	 */
363 	if (has_diag && ret)
364 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
365 
366 	spin_unlock_irqrestore(&phb->lock, flags);
367 }
368 
369 static void pnv_pci_config_check_eeh(struct pnv_phb *phb,
370 				     struct device_node *dn)
371 {
372 	u8	fstate;
373 	__be16	pcierr;
374 	int	pe_no;
375 	s64	rc;
376 
377 	/*
378 	 * Get the PE#. During the PCI probe stage, we might not
379 	 * setup that yet. So all ER errors should be mapped to
380 	 * reserved PE.
381 	 */
382 	pe_no = PCI_DN(dn)->pe_number;
383 	if (pe_no == IODA_INVALID_PE) {
384 		if (phb->type == PNV_PHB_P5IOC2)
385 			pe_no = 0;
386 		else
387 			pe_no = phb->ioda.reserved_pe;
388 	}
389 
390 	/*
391 	 * Fetch frozen state. If the PHB support compound PE,
392 	 * we need handle that case.
393 	 */
394 	if (phb->get_pe_state) {
395 		fstate = phb->get_pe_state(phb, pe_no);
396 	} else {
397 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
398 						pe_no,
399 						&fstate,
400 						&pcierr,
401 						NULL);
402 		if (rc) {
403 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
404 				__func__, rc, phb->hose->global_number, pe_no);
405 			return;
406 		}
407 	}
408 
409 	cfg_dbg(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
410 		(PCI_DN(dn)->busno << 8) | (PCI_DN(dn)->devfn),
411 		pe_no, fstate);
412 
413 	/* Clear the frozen state if applicable */
414 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
415 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
416 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
417 		/*
418 		 * If PHB supports compound PE, freeze it for
419 		 * consistency.
420 		 */
421 		if (phb->freeze_pe)
422 			phb->freeze_pe(phb, pe_no);
423 
424 		pnv_pci_handle_eeh_config(phb, pe_no);
425 	}
426 }
427 
428 int pnv_pci_cfg_read(struct device_node *dn,
429 		     int where, int size, u32 *val)
430 {
431 	struct pci_dn *pdn = PCI_DN(dn);
432 	struct pnv_phb *phb = pdn->phb->private_data;
433 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
434 	s64 rc;
435 
436 	switch (size) {
437 	case 1: {
438 		u8 v8;
439 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
440 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
441 		break;
442 	}
443 	case 2: {
444 		__be16 v16;
445 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
446 						   &v16);
447 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
448 		break;
449 	}
450 	case 4: {
451 		__be32 v32;
452 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
453 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
454 		break;
455 	}
456 	default:
457 		return PCIBIOS_FUNC_NOT_SUPPORTED;
458 	}
459 
460 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
461 		__func__, pdn->busno, pdn->devfn, where, size, *val);
462 	return PCIBIOS_SUCCESSFUL;
463 }
464 
465 int pnv_pci_cfg_write(struct device_node *dn,
466 		      int where, int size, u32 val)
467 {
468 	struct pci_dn *pdn = PCI_DN(dn);
469 	struct pnv_phb *phb = pdn->phb->private_data;
470 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
471 
472 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
473 		pdn->busno, pdn->devfn, where, size, val);
474 	switch (size) {
475 	case 1:
476 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
477 		break;
478 	case 2:
479 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
480 		break;
481 	case 4:
482 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
483 		break;
484 	default:
485 		return PCIBIOS_FUNC_NOT_SUPPORTED;
486 	}
487 
488 	return PCIBIOS_SUCCESSFUL;
489 }
490 
491 #if CONFIG_EEH
492 static bool pnv_pci_cfg_check(struct pci_controller *hose,
493 			      struct device_node *dn)
494 {
495 	struct eeh_dev *edev = NULL;
496 	struct pnv_phb *phb = hose->private_data;
497 
498 	/* EEH not enabled ? */
499 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
500 		return true;
501 
502 	/* PE reset or device removed ? */
503 	edev = of_node_to_eeh_dev(dn);
504 	if (edev) {
505 		if (edev->pe &&
506 		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
507 			return false;
508 
509 		if (edev->mode & EEH_DEV_REMOVED)
510 			return false;
511 	}
512 
513 	return true;
514 }
515 #else
516 static inline pnv_pci_cfg_check(struct pci_controller *hose,
517 				struct device_node *dn)
518 {
519 	return true;
520 }
521 #endif /* CONFIG_EEH */
522 
523 static int pnv_pci_read_config(struct pci_bus *bus,
524 			       unsigned int devfn,
525 			       int where, int size, u32 *val)
526 {
527 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
528 	struct pci_dn *pdn;
529 	struct pnv_phb *phb;
530 	bool found = false;
531 	int ret;
532 
533 	*val = 0xFFFFFFFF;
534 	for (dn = busdn->child; dn; dn = dn->sibling) {
535 		pdn = PCI_DN(dn);
536 		if (pdn && pdn->devfn == devfn) {
537 			phb = pdn->phb->private_data;
538 			found = true;
539 			break;
540 		}
541 	}
542 
543 	if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
544 		return PCIBIOS_DEVICE_NOT_FOUND;
545 
546 	ret = pnv_pci_cfg_read(dn, where, size, val);
547 	if (phb->flags & PNV_PHB_FLAG_EEH) {
548 		if (*val == EEH_IO_ERROR_VALUE(size) &&
549 		    eeh_dev_check_failure(of_node_to_eeh_dev(dn)))
550                         return PCIBIOS_DEVICE_NOT_FOUND;
551 	} else {
552 		pnv_pci_config_check_eeh(phb, dn);
553 	}
554 
555 	return ret;
556 }
557 
558 static int pnv_pci_write_config(struct pci_bus *bus,
559 				unsigned int devfn,
560 				int where, int size, u32 val)
561 {
562 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
563 	struct pci_dn *pdn;
564 	struct pnv_phb *phb;
565 	bool found = false;
566 	int ret;
567 
568 	for (dn = busdn->child; dn; dn = dn->sibling) {
569 		pdn = PCI_DN(dn);
570 		if (pdn && pdn->devfn == devfn) {
571 			phb = pdn->phb->private_data;
572 			found = true;
573 			break;
574 		}
575 	}
576 
577 	if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
578 		return PCIBIOS_DEVICE_NOT_FOUND;
579 
580 	ret = pnv_pci_cfg_write(dn, where, size, val);
581 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
582 		pnv_pci_config_check_eeh(phb, dn);
583 
584 	return ret;
585 }
586 
587 struct pci_ops pnv_pci_ops = {
588 	.read  = pnv_pci_read_config,
589 	.write = pnv_pci_write_config,
590 };
591 
592 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
593 			 unsigned long uaddr, enum dma_data_direction direction,
594 			 struct dma_attrs *attrs, bool rm)
595 {
596 	u64 proto_tce;
597 	__be64 *tcep, *tces;
598 	u64 rpn;
599 
600 	proto_tce = TCE_PCI_READ; // Read allowed
601 
602 	if (direction != DMA_TO_DEVICE)
603 		proto_tce |= TCE_PCI_WRITE;
604 
605 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
606 	rpn = __pa(uaddr) >> tbl->it_page_shift;
607 
608 	while (npages--)
609 		*(tcep++) = cpu_to_be64(proto_tce |
610 				(rpn++ << tbl->it_page_shift));
611 
612 	/* Some implementations won't cache invalid TCEs and thus may not
613 	 * need that flush. We'll probably turn it_type into a bit mask
614 	 * of flags if that becomes the case
615 	 */
616 	if (tbl->it_type & TCE_PCI_SWINV_CREATE)
617 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
618 
619 	return 0;
620 }
621 
622 static int pnv_tce_build_vm(struct iommu_table *tbl, long index, long npages,
623 			    unsigned long uaddr,
624 			    enum dma_data_direction direction,
625 			    struct dma_attrs *attrs)
626 {
627 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs,
628 			false);
629 }
630 
631 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages,
632 		bool rm)
633 {
634 	__be64 *tcep, *tces;
635 
636 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
637 
638 	while (npages--)
639 		*(tcep++) = cpu_to_be64(0);
640 
641 	if (tbl->it_type & TCE_PCI_SWINV_FREE)
642 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
643 }
644 
645 static void pnv_tce_free_vm(struct iommu_table *tbl, long index, long npages)
646 {
647 	pnv_tce_free(tbl, index, npages, false);
648 }
649 
650 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
651 {
652 	return ((u64 *)tbl->it_base)[index - tbl->it_offset];
653 }
654 
655 static int pnv_tce_build_rm(struct iommu_table *tbl, long index, long npages,
656 			    unsigned long uaddr,
657 			    enum dma_data_direction direction,
658 			    struct dma_attrs *attrs)
659 {
660 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs, true);
661 }
662 
663 static void pnv_tce_free_rm(struct iommu_table *tbl, long index, long npages)
664 {
665 	pnv_tce_free(tbl, index, npages, true);
666 }
667 
668 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
669 			       void *tce_mem, u64 tce_size,
670 			       u64 dma_offset, unsigned page_shift)
671 {
672 	tbl->it_blocksize = 16;
673 	tbl->it_base = (unsigned long)tce_mem;
674 	tbl->it_page_shift = page_shift;
675 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
676 	tbl->it_index = 0;
677 	tbl->it_size = tce_size >> 3;
678 	tbl->it_busno = 0;
679 	tbl->it_type = TCE_PCI;
680 }
681 
682 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
683 {
684 	struct iommu_table *tbl;
685 	const __be64 *basep, *swinvp;
686 	const __be32 *sizep;
687 
688 	basep = of_get_property(hose->dn, "linux,tce-base", NULL);
689 	sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
690 	if (basep == NULL || sizep == NULL) {
691 		pr_err("PCI: %s has missing tce entries !\n",
692 		       hose->dn->full_name);
693 		return NULL;
694 	}
695 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
696 	if (WARN_ON(!tbl))
697 		return NULL;
698 	pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
699 				  be32_to_cpup(sizep), 0, IOMMU_PAGE_SHIFT_4K);
700 	iommu_init_table(tbl, hose->node);
701 	iommu_register_group(tbl, pci_domain_nr(hose->bus), 0);
702 
703 	/* Deal with SW invalidated TCEs when needed (BML way) */
704 	swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
705 				 NULL);
706 	if (swinvp) {
707 		tbl->it_busno = be64_to_cpu(swinvp[1]);
708 		tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
709 		tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
710 	}
711 	return tbl;
712 }
713 
714 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
715 				       struct pci_dev *pdev)
716 {
717 	struct device_node *np = pci_bus_to_OF_node(hose->bus);
718 	struct pci_dn *pdn;
719 
720 	if (np == NULL)
721 		return;
722 	pdn = PCI_DN(np);
723 	if (!pdn->iommu_table)
724 		pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
725 	if (!pdn->iommu_table)
726 		return;
727 	set_iommu_table_base_and_group(&pdev->dev, pdn->iommu_table);
728 }
729 
730 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
731 {
732 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
733 	struct pnv_phb *phb = hose->private_data;
734 
735 	/* If we have no phb structure, try to setup a fallback based on
736 	 * the device-tree (RTAS PCI for example)
737 	 */
738 	if (phb && phb->dma_dev_setup)
739 		phb->dma_dev_setup(phb, pdev);
740 	else
741 		pnv_pci_dma_fallback_setup(hose, pdev);
742 }
743 
744 int pnv_pci_dma_set_mask(struct pci_dev *pdev, u64 dma_mask)
745 {
746 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
747 	struct pnv_phb *phb = hose->private_data;
748 
749 	if (phb && phb->dma_set_mask)
750 		return phb->dma_set_mask(phb, pdev, dma_mask);
751 	return __dma_set_mask(&pdev->dev, dma_mask);
752 }
753 
754 u64 pnv_pci_dma_get_required_mask(struct pci_dev *pdev)
755 {
756 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
757 	struct pnv_phb *phb = hose->private_data;
758 
759 	if (phb && phb->dma_get_required_mask)
760 		return phb->dma_get_required_mask(phb, pdev);
761 
762 	return __dma_get_required_mask(&pdev->dev);
763 }
764 
765 void pnv_pci_shutdown(void)
766 {
767 	struct pci_controller *hose;
768 
769 	list_for_each_entry(hose, &hose_list, list_node) {
770 		struct pnv_phb *phb = hose->private_data;
771 
772 		if (phb && phb->shutdown)
773 			phb->shutdown(phb);
774 	}
775 }
776 
777 /* Fixup wrong class code in p7ioc and p8 root complex */
778 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
779 {
780 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
781 }
782 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
783 
784 static int pnv_pci_probe_mode(struct pci_bus *bus)
785 {
786 	struct pci_controller *hose = pci_bus_to_host(bus);
787 	const __be64 *tstamp;
788 	u64 now, target;
789 
790 
791 	/* We hijack this as a way to ensure we have waited long
792 	 * enough since the reset was lifted on the PCI bus
793 	 */
794 	if (bus != hose->bus)
795 		return PCI_PROBE_NORMAL;
796 	tstamp = of_get_property(hose->dn, "reset-clear-timestamp", NULL);
797 	if (!tstamp || !*tstamp)
798 		return PCI_PROBE_NORMAL;
799 
800 	now = mftb() / tb_ticks_per_usec;
801 	target = (be64_to_cpup(tstamp) / tb_ticks_per_usec)
802 		+ PCI_RESET_DELAY_US;
803 
804 	pr_devel("pci %04d: Reset target: 0x%llx now: 0x%llx\n",
805 		 hose->global_number, target, now);
806 
807 	if (now < target)
808 		msleep((target - now + 999) / 1000);
809 
810 	return PCI_PROBE_NORMAL;
811 }
812 
813 void __init pnv_pci_init(void)
814 {
815 	struct device_node *np;
816 
817 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
818 
819 	/* OPAL absent, try POPAL first then RTAS detection of PHBs */
820 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
821 #ifdef CONFIG_PPC_POWERNV_RTAS
822 		init_pci_config_tokens();
823 		find_and_init_phbs();
824 #endif /* CONFIG_PPC_POWERNV_RTAS */
825 	}
826 	/* OPAL is here, do our normal stuff */
827 	else {
828 		int found_ioda = 0;
829 
830 		/* Look for IODA IO-Hubs. We don't support mixing IODA
831 		 * and p5ioc2 due to the need to change some global
832 		 * probing flags
833 		 */
834 		for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
835 			pnv_pci_init_ioda_hub(np);
836 			found_ioda = 1;
837 		}
838 
839 		/* Look for p5ioc2 IO-Hubs */
840 		if (!found_ioda)
841 			for_each_compatible_node(np, NULL, "ibm,p5ioc2")
842 				pnv_pci_init_p5ioc2_hub(np);
843 
844 		/* Look for ioda2 built-in PHB3's */
845 		for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
846 			pnv_pci_init_ioda2_phb(np);
847 	}
848 
849 	/* Setup the linkage between OF nodes and PHBs */
850 	pci_devs_phb_init();
851 
852 	/* Configure IOMMU DMA hooks */
853 	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
854 	ppc_md.tce_build = pnv_tce_build_vm;
855 	ppc_md.tce_free = pnv_tce_free_vm;
856 	ppc_md.tce_build_rm = pnv_tce_build_rm;
857 	ppc_md.tce_free_rm = pnv_tce_free_rm;
858 	ppc_md.tce_get = pnv_tce_get;
859 	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
860 	set_pci_dma_ops(&dma_iommu_ops);
861 
862 	/* Configure MSIs */
863 #ifdef CONFIG_PCI_MSI
864 	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
865 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
866 #endif
867 }
868 
869 static int tce_iommu_bus_notifier(struct notifier_block *nb,
870 		unsigned long action, void *data)
871 {
872 	struct device *dev = data;
873 
874 	switch (action) {
875 	case BUS_NOTIFY_ADD_DEVICE:
876 		return iommu_add_device(dev);
877 	case BUS_NOTIFY_DEL_DEVICE:
878 		if (dev->iommu_group)
879 			iommu_del_device(dev);
880 		return 0;
881 	default:
882 		return 0;
883 	}
884 }
885 
886 static struct notifier_block tce_iommu_bus_nb = {
887 	.notifier_call = tce_iommu_bus_notifier,
888 };
889 
890 static int __init tce_iommu_bus_notifier_init(void)
891 {
892 	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
893 	return 0;
894 }
895 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
896