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/bootmem.h>
20 #include <linux/irq.h>
21 #include <linux/io.h>
22 #include <linux/msi.h>
23 #include <linux/iommu.h>
24 
25 #include <asm/sections.h>
26 #include <asm/io.h>
27 #include <asm/prom.h>
28 #include <asm/pci-bridge.h>
29 #include <asm/machdep.h>
30 #include <asm/msi_bitmap.h>
31 #include <asm/ppc-pci.h>
32 #include <asm/opal.h>
33 #include <asm/iommu.h>
34 #include <asm/tce.h>
35 #include <asm/firmware.h>
36 #include <asm/eeh_event.h>
37 #include <asm/eeh.h>
38 
39 #include "powernv.h"
40 #include "pci.h"
41 
42 /* Delay in usec */
43 #define PCI_RESET_DELAY_US	3000000
44 
45 #define cfg_dbg(fmt...)	do { } while(0)
46 //#define cfg_dbg(fmt...)	printk(fmt)
47 
48 #ifdef CONFIG_PCI_MSI
49 static int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
50 {
51 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
52 	struct pnv_phb *phb = hose->private_data;
53 	struct msi_desc *entry;
54 	struct msi_msg msg;
55 	int hwirq;
56 	unsigned int virq;
57 	int rc;
58 
59 	if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
60 		return -ENODEV;
61 
62 	if (pdev->no_64bit_msi && !phb->msi32_support)
63 		return -ENODEV;
64 
65 	list_for_each_entry(entry, &pdev->msi_list, list) {
66 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
67 			pr_warn("%s: Supports only 64-bit MSIs\n",
68 				pci_name(pdev));
69 			return -ENXIO;
70 		}
71 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
72 		if (hwirq < 0) {
73 			pr_warn("%s: Failed to find a free MSI\n",
74 				pci_name(pdev));
75 			return -ENOSPC;
76 		}
77 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
78 		if (virq == NO_IRQ) {
79 			pr_warn("%s: Failed to map MSI to linux irq\n",
80 				pci_name(pdev));
81 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
82 			return -ENOMEM;
83 		}
84 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
85 				    virq, entry->msi_attrib.is_64, &msg);
86 		if (rc) {
87 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
88 			irq_dispose_mapping(virq);
89 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
90 			return rc;
91 		}
92 		irq_set_msi_desc(virq, entry);
93 		write_msi_msg(virq, &msg);
94 	}
95 	return 0;
96 }
97 
98 static void pnv_teardown_msi_irqs(struct pci_dev *pdev)
99 {
100 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
101 	struct pnv_phb *phb = hose->private_data;
102 	struct msi_desc *entry;
103 
104 	if (WARN_ON(!phb))
105 		return;
106 
107 	list_for_each_entry(entry, &pdev->msi_list, list) {
108 		if (entry->irq == NO_IRQ)
109 			continue;
110 		irq_set_msi_desc(entry->irq, NULL);
111 		msi_bitmap_free_hwirqs(&phb->msi_bmp,
112 			virq_to_hw(entry->irq) - phb->msi_base, 1);
113 		irq_dispose_mapping(entry->irq);
114 	}
115 }
116 #endif /* CONFIG_PCI_MSI */
117 
118 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
119 					 struct OpalIoPhbErrorCommon *common)
120 {
121 	struct OpalIoP7IOCPhbErrorData *data;
122 	int i;
123 
124 	data = (struct OpalIoP7IOCPhbErrorData *)common;
125 	pr_info("P7IOC PHB#%d Diag-data (Version: %d)\n",
126 		hose->global_number, be32_to_cpu(common->version));
127 
128 	if (data->brdgCtl)
129 		pr_info("brdgCtl:     %08x\n",
130 			be32_to_cpu(data->brdgCtl));
131 	if (data->portStatusReg || data->rootCmplxStatus ||
132 	    data->busAgentStatus)
133 		pr_info("UtlSts:      %08x %08x %08x\n",
134 			be32_to_cpu(data->portStatusReg),
135 			be32_to_cpu(data->rootCmplxStatus),
136 			be32_to_cpu(data->busAgentStatus));
137 	if (data->deviceStatus || data->slotStatus   ||
138 	    data->linkStatus   || data->devCmdStatus ||
139 	    data->devSecStatus)
140 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
141 			be32_to_cpu(data->deviceStatus),
142 			be32_to_cpu(data->slotStatus),
143 			be32_to_cpu(data->linkStatus),
144 			be32_to_cpu(data->devCmdStatus),
145 			be32_to_cpu(data->devSecStatus));
146 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
147 	    data->corrErrorStatus)
148 		pr_info("RootErrSts:  %08x %08x %08x\n",
149 			be32_to_cpu(data->rootErrorStatus),
150 			be32_to_cpu(data->uncorrErrorStatus),
151 			be32_to_cpu(data->corrErrorStatus));
152 	if (data->tlpHdr1 || data->tlpHdr2 ||
153 	    data->tlpHdr3 || data->tlpHdr4)
154 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
155 			be32_to_cpu(data->tlpHdr1),
156 			be32_to_cpu(data->tlpHdr2),
157 			be32_to_cpu(data->tlpHdr3),
158 			be32_to_cpu(data->tlpHdr4));
159 	if (data->sourceId || data->errorClass ||
160 	    data->correlator)
161 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
162 			be32_to_cpu(data->sourceId),
163 			be64_to_cpu(data->errorClass),
164 			be64_to_cpu(data->correlator));
165 	if (data->p7iocPlssr || data->p7iocCsr)
166 		pr_info("PhbSts:      %016llx %016llx\n",
167 			be64_to_cpu(data->p7iocPlssr),
168 			be64_to_cpu(data->p7iocCsr));
169 	if (data->lemFir)
170 		pr_info("Lem:         %016llx %016llx %016llx\n",
171 			be64_to_cpu(data->lemFir),
172 			be64_to_cpu(data->lemErrorMask),
173 			be64_to_cpu(data->lemWOF));
174 	if (data->phbErrorStatus)
175 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
176 			be64_to_cpu(data->phbErrorStatus),
177 			be64_to_cpu(data->phbFirstErrorStatus),
178 			be64_to_cpu(data->phbErrorLog0),
179 			be64_to_cpu(data->phbErrorLog1));
180 	if (data->mmioErrorStatus)
181 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
182 			be64_to_cpu(data->mmioErrorStatus),
183 			be64_to_cpu(data->mmioFirstErrorStatus),
184 			be64_to_cpu(data->mmioErrorLog0),
185 			be64_to_cpu(data->mmioErrorLog1));
186 	if (data->dma0ErrorStatus)
187 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
188 			be64_to_cpu(data->dma0ErrorStatus),
189 			be64_to_cpu(data->dma0FirstErrorStatus),
190 			be64_to_cpu(data->dma0ErrorLog0),
191 			be64_to_cpu(data->dma0ErrorLog1));
192 	if (data->dma1ErrorStatus)
193 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
194 			be64_to_cpu(data->dma1ErrorStatus),
195 			be64_to_cpu(data->dma1FirstErrorStatus),
196 			be64_to_cpu(data->dma1ErrorLog0),
197 			be64_to_cpu(data->dma1ErrorLog1));
198 
199 	for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
200 		if ((data->pestA[i] >> 63) == 0 &&
201 		    (data->pestB[i] >> 63) == 0)
202 			continue;
203 
204 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
205 			i, be64_to_cpu(data->pestA[i]),
206 			be64_to_cpu(data->pestB[i]));
207 	}
208 }
209 
210 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
211 					struct OpalIoPhbErrorCommon *common)
212 {
213 	struct OpalIoPhb3ErrorData *data;
214 	int i;
215 
216 	data = (struct OpalIoPhb3ErrorData*)common;
217 	pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n",
218 		hose->global_number, be32_to_cpu(common->version));
219 	if (data->brdgCtl)
220 		pr_info("brdgCtl:     %08x\n",
221 			be32_to_cpu(data->brdgCtl));
222 	if (data->portStatusReg || data->rootCmplxStatus ||
223 	    data->busAgentStatus)
224 		pr_info("UtlSts:      %08x %08x %08x\n",
225 			be32_to_cpu(data->portStatusReg),
226 			be32_to_cpu(data->rootCmplxStatus),
227 			be32_to_cpu(data->busAgentStatus));
228 	if (data->deviceStatus || data->slotStatus   ||
229 	    data->linkStatus   || data->devCmdStatus ||
230 	    data->devSecStatus)
231 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
232 			be32_to_cpu(data->deviceStatus),
233 			be32_to_cpu(data->slotStatus),
234 			be32_to_cpu(data->linkStatus),
235 			be32_to_cpu(data->devCmdStatus),
236 			be32_to_cpu(data->devSecStatus));
237 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
238 	    data->corrErrorStatus)
239 		pr_info("RootErrSts:  %08x %08x %08x\n",
240 			be32_to_cpu(data->rootErrorStatus),
241 			be32_to_cpu(data->uncorrErrorStatus),
242 			be32_to_cpu(data->corrErrorStatus));
243 	if (data->tlpHdr1 || data->tlpHdr2 ||
244 	    data->tlpHdr3 || data->tlpHdr4)
245 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
246 			be32_to_cpu(data->tlpHdr1),
247 			be32_to_cpu(data->tlpHdr2),
248 			be32_to_cpu(data->tlpHdr3),
249 			be32_to_cpu(data->tlpHdr4));
250 	if (data->sourceId || data->errorClass ||
251 	    data->correlator)
252 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
253 			be32_to_cpu(data->sourceId),
254 			be64_to_cpu(data->errorClass),
255 			be64_to_cpu(data->correlator));
256 	if (data->nFir)
257 		pr_info("nFir:        %016llx %016llx %016llx\n",
258 			be64_to_cpu(data->nFir),
259 			be64_to_cpu(data->nFirMask),
260 			be64_to_cpu(data->nFirWOF));
261 	if (data->phbPlssr || data->phbCsr)
262 		pr_info("PhbSts:      %016llx %016llx\n",
263 			be64_to_cpu(data->phbPlssr),
264 			be64_to_cpu(data->phbCsr));
265 	if (data->lemFir)
266 		pr_info("Lem:         %016llx %016llx %016llx\n",
267 			be64_to_cpu(data->lemFir),
268 			be64_to_cpu(data->lemErrorMask),
269 			be64_to_cpu(data->lemWOF));
270 	if (data->phbErrorStatus)
271 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
272 			be64_to_cpu(data->phbErrorStatus),
273 			be64_to_cpu(data->phbFirstErrorStatus),
274 			be64_to_cpu(data->phbErrorLog0),
275 			be64_to_cpu(data->phbErrorLog1));
276 	if (data->mmioErrorStatus)
277 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
278 			be64_to_cpu(data->mmioErrorStatus),
279 			be64_to_cpu(data->mmioFirstErrorStatus),
280 			be64_to_cpu(data->mmioErrorLog0),
281 			be64_to_cpu(data->mmioErrorLog1));
282 	if (data->dma0ErrorStatus)
283 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
284 			be64_to_cpu(data->dma0ErrorStatus),
285 			be64_to_cpu(data->dma0FirstErrorStatus),
286 			be64_to_cpu(data->dma0ErrorLog0),
287 			be64_to_cpu(data->dma0ErrorLog1));
288 	if (data->dma1ErrorStatus)
289 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
290 			be64_to_cpu(data->dma1ErrorStatus),
291 			be64_to_cpu(data->dma1FirstErrorStatus),
292 			be64_to_cpu(data->dma1ErrorLog0),
293 			be64_to_cpu(data->dma1ErrorLog1));
294 
295 	for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
296 		if ((be64_to_cpu(data->pestA[i]) >> 63) == 0 &&
297 		    (be64_to_cpu(data->pestB[i]) >> 63) == 0)
298 			continue;
299 
300 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
301 				i, be64_to_cpu(data->pestA[i]),
302 				be64_to_cpu(data->pestB[i]));
303 	}
304 }
305 
306 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
307 				unsigned char *log_buff)
308 {
309 	struct OpalIoPhbErrorCommon *common;
310 
311 	if (!hose || !log_buff)
312 		return;
313 
314 	common = (struct OpalIoPhbErrorCommon *)log_buff;
315 	switch (be32_to_cpu(common->ioType)) {
316 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
317 		pnv_pci_dump_p7ioc_diag_data(hose, common);
318 		break;
319 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
320 		pnv_pci_dump_phb3_diag_data(hose, common);
321 		break;
322 	default:
323 		pr_warn("%s: Unrecognized ioType %d\n",
324 			__func__, be32_to_cpu(common->ioType));
325 	}
326 }
327 
328 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
329 {
330 	unsigned long flags, rc;
331 	int has_diag, ret = 0;
332 
333 	spin_lock_irqsave(&phb->lock, flags);
334 
335 	/* Fetch PHB diag-data */
336 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
337 					 PNV_PCI_DIAG_BUF_SIZE);
338 	has_diag = (rc == OPAL_SUCCESS);
339 
340 	/* If PHB supports compound PE, to handle it */
341 	if (phb->unfreeze_pe) {
342 		ret = phb->unfreeze_pe(phb,
343 				       pe_no,
344 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
345 	} else {
346 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
347 					     pe_no,
348 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
349 		if (rc) {
350 			pr_warn("%s: Failure %ld clearing frozen "
351 				"PHB#%x-PE#%x\n",
352 				__func__, rc, phb->hose->global_number,
353 				pe_no);
354 			ret = -EIO;
355 		}
356 	}
357 
358 	/*
359 	 * For now, let's only display the diag buffer when we fail to clear
360 	 * the EEH status. We'll do more sensible things later when we have
361 	 * proper EEH support. We need to make sure we don't pollute ourselves
362 	 * with the normal errors generated when probing empty slots
363 	 */
364 	if (has_diag && ret)
365 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
366 
367 	spin_unlock_irqrestore(&phb->lock, flags);
368 }
369 
370 static void pnv_pci_config_check_eeh(struct pnv_phb *phb,
371 				     struct device_node *dn)
372 {
373 	u8	fstate;
374 	__be16	pcierr;
375 	int	pe_no;
376 	s64	rc;
377 
378 	/*
379 	 * Get the PE#. During the PCI probe stage, we might not
380 	 * setup that yet. So all ER errors should be mapped to
381 	 * reserved PE.
382 	 */
383 	pe_no = PCI_DN(dn)->pe_number;
384 	if (pe_no == IODA_INVALID_PE) {
385 		if (phb->type == PNV_PHB_P5IOC2)
386 			pe_no = 0;
387 		else
388 			pe_no = phb->ioda.reserved_pe;
389 	}
390 
391 	/*
392 	 * Fetch frozen state. If the PHB support compound PE,
393 	 * we need handle that case.
394 	 */
395 	if (phb->get_pe_state) {
396 		fstate = phb->get_pe_state(phb, pe_no);
397 	} else {
398 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
399 						pe_no,
400 						&fstate,
401 						&pcierr,
402 						NULL);
403 		if (rc) {
404 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
405 				__func__, rc, phb->hose->global_number, pe_no);
406 			return;
407 		}
408 	}
409 
410 	cfg_dbg(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
411 		(PCI_DN(dn)->busno << 8) | (PCI_DN(dn)->devfn),
412 		pe_no, fstate);
413 
414 	/* Clear the frozen state if applicable */
415 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
416 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
417 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
418 		/*
419 		 * If PHB supports compound PE, freeze it for
420 		 * consistency.
421 		 */
422 		if (phb->freeze_pe)
423 			phb->freeze_pe(phb, pe_no);
424 
425 		pnv_pci_handle_eeh_config(phb, pe_no);
426 	}
427 }
428 
429 int pnv_pci_cfg_read(struct device_node *dn,
430 		     int where, int size, u32 *val)
431 {
432 	struct pci_dn *pdn = PCI_DN(dn);
433 	struct pnv_phb *phb = pdn->phb->private_data;
434 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
435 	s64 rc;
436 
437 	switch (size) {
438 	case 1: {
439 		u8 v8;
440 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
441 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
442 		break;
443 	}
444 	case 2: {
445 		__be16 v16;
446 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
447 						   &v16);
448 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
449 		break;
450 	}
451 	case 4: {
452 		__be32 v32;
453 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
454 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
455 		break;
456 	}
457 	default:
458 		return PCIBIOS_FUNC_NOT_SUPPORTED;
459 	}
460 
461 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
462 		__func__, pdn->busno, pdn->devfn, where, size, *val);
463 	return PCIBIOS_SUCCESSFUL;
464 }
465 
466 int pnv_pci_cfg_write(struct device_node *dn,
467 		      int where, int size, u32 val)
468 {
469 	struct pci_dn *pdn = PCI_DN(dn);
470 	struct pnv_phb *phb = pdn->phb->private_data;
471 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
472 
473 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
474 		pdn->busno, pdn->devfn, where, size, val);
475 	switch (size) {
476 	case 1:
477 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
478 		break;
479 	case 2:
480 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
481 		break;
482 	case 4:
483 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
484 		break;
485 	default:
486 		return PCIBIOS_FUNC_NOT_SUPPORTED;
487 	}
488 
489 	return PCIBIOS_SUCCESSFUL;
490 }
491 
492 #if CONFIG_EEH
493 static bool pnv_pci_cfg_check(struct pci_controller *hose,
494 			      struct device_node *dn)
495 {
496 	struct eeh_dev *edev = NULL;
497 	struct pnv_phb *phb = hose->private_data;
498 
499 	/* EEH not enabled ? */
500 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
501 		return true;
502 
503 	/* PE reset or device removed ? */
504 	edev = of_node_to_eeh_dev(dn);
505 	if (edev) {
506 		if (edev->pe &&
507 		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
508 			return false;
509 
510 		if (edev->mode & EEH_DEV_REMOVED)
511 			return false;
512 	}
513 
514 	return true;
515 }
516 #else
517 static inline pnv_pci_cfg_check(struct pci_controller *hose,
518 				struct device_node *dn)
519 {
520 	return true;
521 }
522 #endif /* CONFIG_EEH */
523 
524 static int pnv_pci_read_config(struct pci_bus *bus,
525 			       unsigned int devfn,
526 			       int where, int size, u32 *val)
527 {
528 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
529 	struct pci_dn *pdn;
530 	struct pnv_phb *phb;
531 	bool found = false;
532 	int ret;
533 
534 	*val = 0xFFFFFFFF;
535 	for (dn = busdn->child; dn; dn = dn->sibling) {
536 		pdn = PCI_DN(dn);
537 		if (pdn && pdn->devfn == devfn) {
538 			phb = pdn->phb->private_data;
539 			found = true;
540 			break;
541 		}
542 	}
543 
544 	if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
545 		return PCIBIOS_DEVICE_NOT_FOUND;
546 
547 	ret = pnv_pci_cfg_read(dn, where, size, val);
548 	if (phb->flags & PNV_PHB_FLAG_EEH) {
549 		if (*val == EEH_IO_ERROR_VALUE(size) &&
550 		    eeh_dev_check_failure(of_node_to_eeh_dev(dn)))
551                         return PCIBIOS_DEVICE_NOT_FOUND;
552 	} else {
553 		pnv_pci_config_check_eeh(phb, dn);
554 	}
555 
556 	return ret;
557 }
558 
559 static int pnv_pci_write_config(struct pci_bus *bus,
560 				unsigned int devfn,
561 				int where, int size, u32 val)
562 {
563 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
564 	struct pci_dn *pdn;
565 	struct pnv_phb *phb;
566 	bool found = false;
567 	int ret;
568 
569 	for (dn = busdn->child; dn; dn = dn->sibling) {
570 		pdn = PCI_DN(dn);
571 		if (pdn && pdn->devfn == devfn) {
572 			phb = pdn->phb->private_data;
573 			found = true;
574 			break;
575 		}
576 	}
577 
578 	if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
579 		return PCIBIOS_DEVICE_NOT_FOUND;
580 
581 	ret = pnv_pci_cfg_write(dn, where, size, val);
582 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
583 		pnv_pci_config_check_eeh(phb, dn);
584 
585 	return ret;
586 }
587 
588 struct pci_ops pnv_pci_ops = {
589 	.read  = pnv_pci_read_config,
590 	.write = pnv_pci_write_config,
591 };
592 
593 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
594 			 unsigned long uaddr, enum dma_data_direction direction,
595 			 struct dma_attrs *attrs, bool rm)
596 {
597 	u64 proto_tce;
598 	__be64 *tcep, *tces;
599 	u64 rpn;
600 
601 	proto_tce = TCE_PCI_READ; // Read allowed
602 
603 	if (direction != DMA_TO_DEVICE)
604 		proto_tce |= TCE_PCI_WRITE;
605 
606 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
607 	rpn = __pa(uaddr) >> tbl->it_page_shift;
608 
609 	while (npages--)
610 		*(tcep++) = cpu_to_be64(proto_tce |
611 				(rpn++ << tbl->it_page_shift));
612 
613 	/* Some implementations won't cache invalid TCEs and thus may not
614 	 * need that flush. We'll probably turn it_type into a bit mask
615 	 * of flags if that becomes the case
616 	 */
617 	if (tbl->it_type & TCE_PCI_SWINV_CREATE)
618 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
619 
620 	return 0;
621 }
622 
623 static int pnv_tce_build_vm(struct iommu_table *tbl, long index, long npages,
624 			    unsigned long uaddr,
625 			    enum dma_data_direction direction,
626 			    struct dma_attrs *attrs)
627 {
628 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs,
629 			false);
630 }
631 
632 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages,
633 		bool rm)
634 {
635 	__be64 *tcep, *tces;
636 
637 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
638 
639 	while (npages--)
640 		*(tcep++) = cpu_to_be64(0);
641 
642 	if (tbl->it_type & TCE_PCI_SWINV_FREE)
643 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
644 }
645 
646 static void pnv_tce_free_vm(struct iommu_table *tbl, long index, long npages)
647 {
648 	pnv_tce_free(tbl, index, npages, false);
649 }
650 
651 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
652 {
653 	return ((u64 *)tbl->it_base)[index - tbl->it_offset];
654 }
655 
656 static int pnv_tce_build_rm(struct iommu_table *tbl, long index, long npages,
657 			    unsigned long uaddr,
658 			    enum dma_data_direction direction,
659 			    struct dma_attrs *attrs)
660 {
661 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs, true);
662 }
663 
664 static void pnv_tce_free_rm(struct iommu_table *tbl, long index, long npages)
665 {
666 	pnv_tce_free(tbl, index, npages, true);
667 }
668 
669 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
670 			       void *tce_mem, u64 tce_size,
671 			       u64 dma_offset, unsigned page_shift)
672 {
673 	tbl->it_blocksize = 16;
674 	tbl->it_base = (unsigned long)tce_mem;
675 	tbl->it_page_shift = page_shift;
676 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
677 	tbl->it_index = 0;
678 	tbl->it_size = tce_size >> 3;
679 	tbl->it_busno = 0;
680 	tbl->it_type = TCE_PCI;
681 }
682 
683 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
684 {
685 	struct iommu_table *tbl;
686 	const __be64 *basep, *swinvp;
687 	const __be32 *sizep;
688 
689 	basep = of_get_property(hose->dn, "linux,tce-base", NULL);
690 	sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
691 	if (basep == NULL || sizep == NULL) {
692 		pr_err("PCI: %s has missing tce entries !\n",
693 		       hose->dn->full_name);
694 		return NULL;
695 	}
696 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
697 	if (WARN_ON(!tbl))
698 		return NULL;
699 	pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
700 				  be32_to_cpup(sizep), 0, IOMMU_PAGE_SHIFT_4K);
701 	iommu_init_table(tbl, hose->node);
702 	iommu_register_group(tbl, pci_domain_nr(hose->bus), 0);
703 
704 	/* Deal with SW invalidated TCEs when needed (BML way) */
705 	swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
706 				 NULL);
707 	if (swinvp) {
708 		tbl->it_busno = be64_to_cpu(swinvp[1]);
709 		tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
710 		tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
711 	}
712 	return tbl;
713 }
714 
715 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
716 				       struct pci_dev *pdev)
717 {
718 	struct device_node *np = pci_bus_to_OF_node(hose->bus);
719 	struct pci_dn *pdn;
720 
721 	if (np == NULL)
722 		return;
723 	pdn = PCI_DN(np);
724 	if (!pdn->iommu_table)
725 		pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
726 	if (!pdn->iommu_table)
727 		return;
728 	set_iommu_table_base_and_group(&pdev->dev, pdn->iommu_table);
729 }
730 
731 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
732 {
733 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
734 	struct pnv_phb *phb = hose->private_data;
735 
736 	/* If we have no phb structure, try to setup a fallback based on
737 	 * the device-tree (RTAS PCI for example)
738 	 */
739 	if (phb && phb->dma_dev_setup)
740 		phb->dma_dev_setup(phb, pdev);
741 	else
742 		pnv_pci_dma_fallback_setup(hose, pdev);
743 }
744 
745 int pnv_pci_dma_set_mask(struct pci_dev *pdev, u64 dma_mask)
746 {
747 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
748 	struct pnv_phb *phb = hose->private_data;
749 
750 	if (phb && phb->dma_set_mask)
751 		return phb->dma_set_mask(phb, pdev, dma_mask);
752 	return __dma_set_mask(&pdev->dev, dma_mask);
753 }
754 
755 u64 pnv_pci_dma_get_required_mask(struct pci_dev *pdev)
756 {
757 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
758 	struct pnv_phb *phb = hose->private_data;
759 
760 	if (phb && phb->dma_get_required_mask)
761 		return phb->dma_get_required_mask(phb, pdev);
762 
763 	return __dma_get_required_mask(&pdev->dev);
764 }
765 
766 void pnv_pci_shutdown(void)
767 {
768 	struct pci_controller *hose;
769 
770 	list_for_each_entry(hose, &hose_list, list_node) {
771 		struct pnv_phb *phb = hose->private_data;
772 
773 		if (phb && phb->shutdown)
774 			phb->shutdown(phb);
775 	}
776 }
777 
778 /* Fixup wrong class code in p7ioc and p8 root complex */
779 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
780 {
781 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
782 }
783 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
784 
785 static int pnv_pci_probe_mode(struct pci_bus *bus)
786 {
787 	struct pci_controller *hose = pci_bus_to_host(bus);
788 	const __be64 *tstamp;
789 	u64 now, target;
790 
791 
792 	/* We hijack this as a way to ensure we have waited long
793 	 * enough since the reset was lifted on the PCI bus
794 	 */
795 	if (bus != hose->bus)
796 		return PCI_PROBE_NORMAL;
797 	tstamp = of_get_property(hose->dn, "reset-clear-timestamp", NULL);
798 	if (!tstamp || !*tstamp)
799 		return PCI_PROBE_NORMAL;
800 
801 	now = mftb() / tb_ticks_per_usec;
802 	target = (be64_to_cpup(tstamp) / tb_ticks_per_usec)
803 		+ PCI_RESET_DELAY_US;
804 
805 	pr_devel("pci %04d: Reset target: 0x%llx now: 0x%llx\n",
806 		 hose->global_number, target, now);
807 
808 	if (now < target)
809 		msleep((target - now + 999) / 1000);
810 
811 	return PCI_PROBE_NORMAL;
812 }
813 
814 void __init pnv_pci_init(void)
815 {
816 	struct device_node *np;
817 
818 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
819 
820 	/* OPAL absent, try POPAL first then RTAS detection of PHBs */
821 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
822 #ifdef CONFIG_PPC_POWERNV_RTAS
823 		init_pci_config_tokens();
824 		find_and_init_phbs();
825 #endif /* CONFIG_PPC_POWERNV_RTAS */
826 	}
827 	/* OPAL is here, do our normal stuff */
828 	else {
829 		int found_ioda = 0;
830 
831 		/* Look for IODA IO-Hubs. We don't support mixing IODA
832 		 * and p5ioc2 due to the need to change some global
833 		 * probing flags
834 		 */
835 		for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
836 			pnv_pci_init_ioda_hub(np);
837 			found_ioda = 1;
838 		}
839 
840 		/* Look for p5ioc2 IO-Hubs */
841 		if (!found_ioda)
842 			for_each_compatible_node(np, NULL, "ibm,p5ioc2")
843 				pnv_pci_init_p5ioc2_hub(np);
844 
845 		/* Look for ioda2 built-in PHB3's */
846 		for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
847 			pnv_pci_init_ioda2_phb(np);
848 	}
849 
850 	/* Setup the linkage between OF nodes and PHBs */
851 	pci_devs_phb_init();
852 
853 	/* Configure IOMMU DMA hooks */
854 	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
855 	ppc_md.tce_build = pnv_tce_build_vm;
856 	ppc_md.tce_free = pnv_tce_free_vm;
857 	ppc_md.tce_build_rm = pnv_tce_build_rm;
858 	ppc_md.tce_free_rm = pnv_tce_free_rm;
859 	ppc_md.tce_get = pnv_tce_get;
860 	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
861 	set_pci_dma_ops(&dma_iommu_ops);
862 
863 	/* Configure MSIs */
864 #ifdef CONFIG_PCI_MSI
865 	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
866 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
867 #endif
868 }
869 
870 static int tce_iommu_bus_notifier(struct notifier_block *nb,
871 		unsigned long action, void *data)
872 {
873 	struct device *dev = data;
874 
875 	switch (action) {
876 	case BUS_NOTIFY_ADD_DEVICE:
877 		return iommu_add_device(dev);
878 	case BUS_NOTIFY_DEL_DEVICE:
879 		if (dev->iommu_group)
880 			iommu_del_device(dev);
881 		return 0;
882 	default:
883 		return 0;
884 	}
885 }
886 
887 static struct notifier_block tce_iommu_bus_nb = {
888 	.notifier_call = tce_iommu_bus_notifier,
889 };
890 
891 static int __init tce_iommu_bus_notifier_init(void)
892 {
893 	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
894 	return 0;
895 }
896 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
897