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
48int 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	for_each_pci_msi_entry(entry, pdev) {
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
97void 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	irq_hw_number_t hwirq;
103
104	if (WARN_ON(!phb))
105		return;
106
107	for_each_pci_msi_entry(entry, pdev) {
108		if (entry->irq == NO_IRQ)
109			continue;
110		hwirq = virq_to_hw(entry->irq);
111		irq_set_msi_desc(entry->irq, NULL);
112		irq_dispose_mapping(entry->irq);
113		msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
114	}
115}
116#endif /* CONFIG_PCI_MSI */
117
118static 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
210static 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
306void 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
328static 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
370static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
371{
372	struct pnv_phb *phb = pdn->phb->private_data;
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 = pdn->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		(pdn->busno << 8) | (pdn->devfn), 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
428int pnv_pci_cfg_read(struct pci_dn *pdn,
429		     int where, int size, u32 *val)
430{
431	struct pnv_phb *phb = pdn->phb->private_data;
432	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
433	s64 rc;
434
435	switch (size) {
436	case 1: {
437		u8 v8;
438		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
439		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
440		break;
441	}
442	case 2: {
443		__be16 v16;
444		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
445						   &v16);
446		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
447		break;
448	}
449	case 4: {
450		__be32 v32;
451		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
452		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
453		break;
454	}
455	default:
456		return PCIBIOS_FUNC_NOT_SUPPORTED;
457	}
458
459	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
460		__func__, pdn->busno, pdn->devfn, where, size, *val);
461	return PCIBIOS_SUCCESSFUL;
462}
463
464int pnv_pci_cfg_write(struct pci_dn *pdn,
465		      int where, int size, u32 val)
466{
467	struct pnv_phb *phb = pdn->phb->private_data;
468	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
469
470	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
471		pdn->busno, pdn->devfn, where, size, val);
472	switch (size) {
473	case 1:
474		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
475		break;
476	case 2:
477		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
478		break;
479	case 4:
480		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
481		break;
482	default:
483		return PCIBIOS_FUNC_NOT_SUPPORTED;
484	}
485
486	return PCIBIOS_SUCCESSFUL;
487}
488
489#if CONFIG_EEH
490static bool pnv_pci_cfg_check(struct pci_dn *pdn)
491{
492	struct eeh_dev *edev = NULL;
493	struct pnv_phb *phb = pdn->phb->private_data;
494
495	/* EEH not enabled ? */
496	if (!(phb->flags & PNV_PHB_FLAG_EEH))
497		return true;
498
499	/* PE reset or device removed ? */
500	edev = pdn->edev;
501	if (edev) {
502		if (edev->pe &&
503		    (edev->pe->state & EEH_PE_CFG_BLOCKED))
504			return false;
505
506		if (edev->mode & EEH_DEV_REMOVED)
507			return false;
508	}
509
510	return true;
511}
512#else
513static inline pnv_pci_cfg_check(struct pci_dn *pdn)
514{
515	return true;
516}
517#endif /* CONFIG_EEH */
518
519static int pnv_pci_read_config(struct pci_bus *bus,
520			       unsigned int devfn,
521			       int where, int size, u32 *val)
522{
523	struct pci_dn *pdn;
524	struct pnv_phb *phb;
525	int ret;
526
527	*val = 0xFFFFFFFF;
528	pdn = pci_get_pdn_by_devfn(bus, devfn);
529	if (!pdn)
530		return PCIBIOS_DEVICE_NOT_FOUND;
531
532	if (!pnv_pci_cfg_check(pdn))
533		return PCIBIOS_DEVICE_NOT_FOUND;
534
535	ret = pnv_pci_cfg_read(pdn, where, size, val);
536	phb = pdn->phb->private_data;
537	if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
538		if (*val == EEH_IO_ERROR_VALUE(size) &&
539		    eeh_dev_check_failure(pdn->edev))
540                        return PCIBIOS_DEVICE_NOT_FOUND;
541	} else {
542		pnv_pci_config_check_eeh(pdn);
543	}
544
545	return ret;
546}
547
548static int pnv_pci_write_config(struct pci_bus *bus,
549				unsigned int devfn,
550				int where, int size, u32 val)
551{
552	struct pci_dn *pdn;
553	struct pnv_phb *phb;
554	int ret;
555
556	pdn = pci_get_pdn_by_devfn(bus, devfn);
557	if (!pdn)
558		return PCIBIOS_DEVICE_NOT_FOUND;
559
560	if (!pnv_pci_cfg_check(pdn))
561		return PCIBIOS_DEVICE_NOT_FOUND;
562
563	ret = pnv_pci_cfg_write(pdn, where, size, val);
564	phb = pdn->phb->private_data;
565	if (!(phb->flags & PNV_PHB_FLAG_EEH))
566		pnv_pci_config_check_eeh(pdn);
567
568	return ret;
569}
570
571struct pci_ops pnv_pci_ops = {
572	.read  = pnv_pci_read_config,
573	.write = pnv_pci_write_config,
574};
575
576static __be64 *pnv_tce(struct iommu_table *tbl, long idx)
577{
578	__be64 *tmp = ((__be64 *)tbl->it_base);
579	int  level = tbl->it_indirect_levels;
580	const long shift = ilog2(tbl->it_level_size);
581	unsigned long mask = (tbl->it_level_size - 1) << (level * shift);
582
583	while (level) {
584		int n = (idx & mask) >> (level * shift);
585		unsigned long tce = be64_to_cpu(tmp[n]);
586
587		tmp = __va(tce & ~(TCE_PCI_READ | TCE_PCI_WRITE));
588		idx &= ~mask;
589		mask >>= shift;
590		--level;
591	}
592
593	return tmp + idx;
594}
595
596int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
597		unsigned long uaddr, enum dma_data_direction direction,
598		struct dma_attrs *attrs)
599{
600	u64 proto_tce = iommu_direction_to_tce_perm(direction);
601	u64 rpn = __pa(uaddr) >> tbl->it_page_shift;
602	long i;
603
604	if (proto_tce & TCE_PCI_WRITE)
605		proto_tce |= TCE_PCI_READ;
606
607	for (i = 0; i < npages; i++) {
608		unsigned long newtce = proto_tce |
609			((rpn + i) << tbl->it_page_shift);
610		unsigned long idx = index - tbl->it_offset + i;
611
612		*(pnv_tce(tbl, idx)) = cpu_to_be64(newtce);
613	}
614
615	return 0;
616}
617
618#ifdef CONFIG_IOMMU_API
619int pnv_tce_xchg(struct iommu_table *tbl, long index,
620		unsigned long *hpa, enum dma_data_direction *direction)
621{
622	u64 proto_tce = iommu_direction_to_tce_perm(*direction);
623	unsigned long newtce = *hpa | proto_tce, oldtce;
624	unsigned long idx = index - tbl->it_offset;
625
626	BUG_ON(*hpa & ~IOMMU_PAGE_MASK(tbl));
627
628	if (newtce & TCE_PCI_WRITE)
629		newtce |= TCE_PCI_READ;
630
631	oldtce = xchg(pnv_tce(tbl, idx), cpu_to_be64(newtce));
632	*hpa = be64_to_cpu(oldtce) & ~(TCE_PCI_READ | TCE_PCI_WRITE);
633	*direction = iommu_tce_direction(oldtce);
634
635	return 0;
636}
637#endif
638
639void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
640{
641	long i;
642
643	for (i = 0; i < npages; i++) {
644		unsigned long idx = index - tbl->it_offset + i;
645
646		*(pnv_tce(tbl, idx)) = cpu_to_be64(0);
647	}
648}
649
650unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
651{
652	return *(pnv_tce(tbl, index - tbl->it_offset));
653}
654
655struct iommu_table *pnv_pci_table_alloc(int nid)
656{
657	struct iommu_table *tbl;
658
659	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
660	INIT_LIST_HEAD_RCU(&tbl->it_group_list);
661
662	return tbl;
663}
664
665long pnv_pci_link_table_and_group(int node, int num,
666		struct iommu_table *tbl,
667		struct iommu_table_group *table_group)
668{
669	struct iommu_table_group_link *tgl = NULL;
670
671	if (WARN_ON(!tbl || !table_group))
672		return -EINVAL;
673
674	tgl = kzalloc_node(sizeof(struct iommu_table_group_link), GFP_KERNEL,
675			node);
676	if (!tgl)
677		return -ENOMEM;
678
679	tgl->table_group = table_group;
680	list_add_rcu(&tgl->next, &tbl->it_group_list);
681
682	table_group->tables[num] = tbl;
683
684	return 0;
685}
686
687static void pnv_iommu_table_group_link_free(struct rcu_head *head)
688{
689	struct iommu_table_group_link *tgl = container_of(head,
690			struct iommu_table_group_link, rcu);
691
692	kfree(tgl);
693}
694
695void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
696		struct iommu_table_group *table_group)
697{
698	long i;
699	bool found;
700	struct iommu_table_group_link *tgl;
701
702	if (!tbl || !table_group)
703		return;
704
705	/* Remove link to a group from table's list of attached groups */
706	found = false;
707	list_for_each_entry_rcu(tgl, &tbl->it_group_list, next) {
708		if (tgl->table_group == table_group) {
709			list_del_rcu(&tgl->next);
710			call_rcu(&tgl->rcu, pnv_iommu_table_group_link_free);
711			found = true;
712			break;
713		}
714	}
715	if (WARN_ON(!found))
716		return;
717
718	/* Clean a pointer to iommu_table in iommu_table_group::tables[] */
719	found = false;
720	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
721		if (table_group->tables[i] == tbl) {
722			table_group->tables[i] = NULL;
723			found = true;
724			break;
725		}
726	}
727	WARN_ON(!found);
728}
729
730void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
731			       void *tce_mem, u64 tce_size,
732			       u64 dma_offset, unsigned page_shift)
733{
734	tbl->it_blocksize = 16;
735	tbl->it_base = (unsigned long)tce_mem;
736	tbl->it_page_shift = page_shift;
737	tbl->it_offset = dma_offset >> tbl->it_page_shift;
738	tbl->it_index = 0;
739	tbl->it_size = tce_size >> 3;
740	tbl->it_busno = 0;
741	tbl->it_type = TCE_PCI;
742}
743
744void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
745{
746	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
747	struct pnv_phb *phb = hose->private_data;
748#ifdef CONFIG_PCI_IOV
749	struct pnv_ioda_pe *pe;
750	struct pci_dn *pdn;
751
752	/* Fix the VF pdn PE number */
753	if (pdev->is_virtfn) {
754		pdn = pci_get_pdn(pdev);
755		WARN_ON(pdn->pe_number != IODA_INVALID_PE);
756		list_for_each_entry(pe, &phb->ioda.pe_list, list) {
757			if (pe->rid == ((pdev->bus->number << 8) |
758			    (pdev->devfn & 0xff))) {
759				pdn->pe_number = pe->pe_number;
760				pe->pdev = pdev;
761				break;
762			}
763		}
764	}
765#endif /* CONFIG_PCI_IOV */
766
767	if (phb && phb->dma_dev_setup)
768		phb->dma_dev_setup(phb, pdev);
769}
770
771void pnv_pci_dma_bus_setup(struct pci_bus *bus)
772{
773	struct pci_controller *hose = bus->sysdata;
774	struct pnv_phb *phb = hose->private_data;
775	struct pnv_ioda_pe *pe;
776
777	list_for_each_entry(pe, &phb->ioda.pe_list, list) {
778		if (!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)))
779			continue;
780
781		if (!pe->pbus)
782			continue;
783
784		if (bus->number == ((pe->rid >> 8) & 0xFF)) {
785			pe->pbus = bus;
786			break;
787		}
788	}
789}
790
791void pnv_pci_shutdown(void)
792{
793	struct pci_controller *hose;
794
795	list_for_each_entry(hose, &hose_list, list_node)
796		if (hose->controller_ops.shutdown)
797			hose->controller_ops.shutdown(hose);
798}
799
800/* Fixup wrong class code in p7ioc and p8 root complex */
801static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
802{
803	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
804}
805DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
806
807void __init pnv_pci_init(void)
808{
809	struct device_node *np;
810	bool found_ioda = false;
811
812	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
813
814	/* If we don't have OPAL, eg. in sim, just skip PCI probe */
815	if (!firmware_has_feature(FW_FEATURE_OPAL))
816		return;
817
818	/* Look for IODA IO-Hubs. We don't support mixing IODA
819	 * and p5ioc2 due to the need to change some global
820	 * probing flags
821	 */
822	for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
823		pnv_pci_init_ioda_hub(np);
824		found_ioda = true;
825	}
826
827	/* Look for p5ioc2 IO-Hubs */
828	if (!found_ioda)
829		for_each_compatible_node(np, NULL, "ibm,p5ioc2")
830			pnv_pci_init_p5ioc2_hub(np);
831
832	/* Look for ioda2 built-in PHB3's */
833	for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
834		pnv_pci_init_ioda2_phb(np);
835
836	/* Setup the linkage between OF nodes and PHBs */
837	pci_devs_phb_init();
838
839	/* Configure IOMMU DMA hooks */
840	set_pci_dma_ops(&dma_iommu_ops);
841}
842
843machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
844