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#undef DEBUG
13
14#include <linux/kernel.h>
15#include <linux/pci.h>
16#include <linux/crash_dump.h>
17#include <linux/debugfs.h>
18#include <linux/delay.h>
19#include <linux/string.h>
20#include <linux/init.h>
21#include <linux/bootmem.h>
22#include <linux/irq.h>
23#include <linux/io.h>
24#include <linux/msi.h>
25#include <linux/memblock.h>
26#include <linux/iommu.h>
27#include <linux/rculist.h>
28#include <linux/sizes.h>
29
30#include <asm/sections.h>
31#include <asm/io.h>
32#include <asm/prom.h>
33#include <asm/pci-bridge.h>
34#include <asm/machdep.h>
35#include <asm/msi_bitmap.h>
36#include <asm/ppc-pci.h>
37#include <asm/opal.h>
38#include <asm/iommu.h>
39#include <asm/tce.h>
40#include <asm/xics.h>
41#include <asm/debug.h>
42#include <asm/firmware.h>
43#include <asm/pnv-pci.h>
44#include <asm/mmzone.h>
45
46#include <misc/cxl-base.h>
47
48#include "powernv.h"
49#include "pci.h"
50
51/* 256M DMA window, 4K TCE pages, 8 bytes TCE */
52#define TCE32_TABLE_SIZE	((0x10000000 / 0x1000) * 8)
53
54#define POWERNV_IOMMU_DEFAULT_LEVELS	1
55#define POWERNV_IOMMU_MAX_LEVELS	5
56
57static void pnv_pci_ioda2_table_free_pages(struct iommu_table *tbl);
58
59static void pe_level_printk(const struct pnv_ioda_pe *pe, const char *level,
60			    const char *fmt, ...)
61{
62	struct va_format vaf;
63	va_list args;
64	char pfix[32];
65
66	va_start(args, fmt);
67
68	vaf.fmt = fmt;
69	vaf.va = &args;
70
71	if (pe->flags & PNV_IODA_PE_DEV)
72		strlcpy(pfix, dev_name(&pe->pdev->dev), sizeof(pfix));
73	else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL))
74		sprintf(pfix, "%04x:%02x     ",
75			pci_domain_nr(pe->pbus), pe->pbus->number);
76#ifdef CONFIG_PCI_IOV
77	else if (pe->flags & PNV_IODA_PE_VF)
78		sprintf(pfix, "%04x:%02x:%2x.%d",
79			pci_domain_nr(pe->parent_dev->bus),
80			(pe->rid & 0xff00) >> 8,
81			PCI_SLOT(pe->rid), PCI_FUNC(pe->rid));
82#endif /* CONFIG_PCI_IOV*/
83
84	printk("%spci %s: [PE# %.3d] %pV",
85	       level, pfix, pe->pe_number, &vaf);
86
87	va_end(args);
88}
89
90#define pe_err(pe, fmt, ...)					\
91	pe_level_printk(pe, KERN_ERR, fmt, ##__VA_ARGS__)
92#define pe_warn(pe, fmt, ...)					\
93	pe_level_printk(pe, KERN_WARNING, fmt, ##__VA_ARGS__)
94#define pe_info(pe, fmt, ...)					\
95	pe_level_printk(pe, KERN_INFO, fmt, ##__VA_ARGS__)
96
97static bool pnv_iommu_bypass_disabled __read_mostly;
98
99static int __init iommu_setup(char *str)
100{
101	if (!str)
102		return -EINVAL;
103
104	while (*str) {
105		if (!strncmp(str, "nobypass", 8)) {
106			pnv_iommu_bypass_disabled = true;
107			pr_info("PowerNV: IOMMU bypass window disabled.\n");
108			break;
109		}
110		str += strcspn(str, ",");
111		if (*str == ',')
112			str++;
113	}
114
115	return 0;
116}
117early_param("iommu", iommu_setup);
118
119/*
120 * stdcix is only supposed to be used in hypervisor real mode as per
121 * the architecture spec
122 */
123static inline void __raw_rm_writeq(u64 val, volatile void __iomem *paddr)
124{
125	__asm__ __volatile__("stdcix %0,0,%1"
126		: : "r" (val), "r" (paddr) : "memory");
127}
128
129static inline bool pnv_pci_is_mem_pref_64(unsigned long flags)
130{
131	return ((flags & (IORESOURCE_MEM_64 | IORESOURCE_PREFETCH)) ==
132		(IORESOURCE_MEM_64 | IORESOURCE_PREFETCH));
133}
134
135static void pnv_ioda_reserve_pe(struct pnv_phb *phb, int pe_no)
136{
137	if (!(pe_no >= 0 && pe_no < phb->ioda.total_pe)) {
138		pr_warn("%s: Invalid PE %d on PHB#%x\n",
139			__func__, pe_no, phb->hose->global_number);
140		return;
141	}
142
143	if (test_and_set_bit(pe_no, phb->ioda.pe_alloc))
144		pr_debug("%s: PE %d was reserved on PHB#%x\n",
145			 __func__, pe_no, phb->hose->global_number);
146
147	phb->ioda.pe_array[pe_no].phb = phb;
148	phb->ioda.pe_array[pe_no].pe_number = pe_no;
149}
150
151static int pnv_ioda_alloc_pe(struct pnv_phb *phb)
152{
153	unsigned long pe;
154
155	do {
156		pe = find_next_zero_bit(phb->ioda.pe_alloc,
157					phb->ioda.total_pe, 0);
158		if (pe >= phb->ioda.total_pe)
159			return IODA_INVALID_PE;
160	} while(test_and_set_bit(pe, phb->ioda.pe_alloc));
161
162	phb->ioda.pe_array[pe].phb = phb;
163	phb->ioda.pe_array[pe].pe_number = pe;
164	return pe;
165}
166
167static void pnv_ioda_free_pe(struct pnv_phb *phb, int pe)
168{
169	WARN_ON(phb->ioda.pe_array[pe].pdev);
170
171	memset(&phb->ioda.pe_array[pe], 0, sizeof(struct pnv_ioda_pe));
172	clear_bit(pe, phb->ioda.pe_alloc);
173}
174
175/* The default M64 BAR is shared by all PEs */
176static int pnv_ioda2_init_m64(struct pnv_phb *phb)
177{
178	const char *desc;
179	struct resource *r;
180	s64 rc;
181
182	/* Configure the default M64 BAR */
183	rc = opal_pci_set_phb_mem_window(phb->opal_id,
184					 OPAL_M64_WINDOW_TYPE,
185					 phb->ioda.m64_bar_idx,
186					 phb->ioda.m64_base,
187					 0, /* unused */
188					 phb->ioda.m64_size);
189	if (rc != OPAL_SUCCESS) {
190		desc = "configuring";
191		goto fail;
192	}
193
194	/* Enable the default M64 BAR */
195	rc = opal_pci_phb_mmio_enable(phb->opal_id,
196				      OPAL_M64_WINDOW_TYPE,
197				      phb->ioda.m64_bar_idx,
198				      OPAL_ENABLE_M64_SPLIT);
199	if (rc != OPAL_SUCCESS) {
200		desc = "enabling";
201		goto fail;
202	}
203
204	/* Mark the M64 BAR assigned */
205	set_bit(phb->ioda.m64_bar_idx, &phb->ioda.m64_bar_alloc);
206
207	/*
208	 * Strip off the segment used by the reserved PE, which is
209	 * expected to be 0 or last one of PE capabicity.
210	 */
211	r = &phb->hose->mem_resources[1];
212	if (phb->ioda.reserved_pe == 0)
213		r->start += phb->ioda.m64_segsize;
214	else if (phb->ioda.reserved_pe == (phb->ioda.total_pe - 1))
215		r->end -= phb->ioda.m64_segsize;
216	else
217		pr_warn("  Cannot strip M64 segment for reserved PE#%d\n",
218			phb->ioda.reserved_pe);
219
220	return 0;
221
222fail:
223	pr_warn("  Failure %lld %s M64 BAR#%d\n",
224		rc, desc, phb->ioda.m64_bar_idx);
225	opal_pci_phb_mmio_enable(phb->opal_id,
226				 OPAL_M64_WINDOW_TYPE,
227				 phb->ioda.m64_bar_idx,
228				 OPAL_DISABLE_M64);
229	return -EIO;
230}
231
232static void pnv_ioda2_reserve_dev_m64_pe(struct pci_dev *pdev,
233					 unsigned long *pe_bitmap)
234{
235	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
236	struct pnv_phb *phb = hose->private_data;
237	struct resource *r;
238	resource_size_t base, sgsz, start, end;
239	int segno, i;
240
241	base = phb->ioda.m64_base;
242	sgsz = phb->ioda.m64_segsize;
243	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
244		r = &pdev->resource[i];
245		if (!r->parent || !pnv_pci_is_mem_pref_64(r->flags))
246			continue;
247
248		start = _ALIGN_DOWN(r->start - base, sgsz);
249		end = _ALIGN_UP(r->end - base, sgsz);
250		for (segno = start / sgsz; segno < end / sgsz; segno++) {
251			if (pe_bitmap)
252				set_bit(segno, pe_bitmap);
253			else
254				pnv_ioda_reserve_pe(phb, segno);
255		}
256	}
257}
258
259static void pnv_ioda2_reserve_m64_pe(struct pci_bus *bus,
260				     unsigned long *pe_bitmap,
261				     bool all)
262{
263	struct pci_dev *pdev;
264
265	list_for_each_entry(pdev, &bus->devices, bus_list) {
266		pnv_ioda2_reserve_dev_m64_pe(pdev, pe_bitmap);
267
268		if (all && pdev->subordinate)
269			pnv_ioda2_reserve_m64_pe(pdev->subordinate,
270						 pe_bitmap, all);
271	}
272}
273
274static int pnv_ioda2_pick_m64_pe(struct pci_bus *bus, bool all)
275{
276	struct pci_controller *hose = pci_bus_to_host(bus);
277	struct pnv_phb *phb = hose->private_data;
278	struct pnv_ioda_pe *master_pe, *pe;
279	unsigned long size, *pe_alloc;
280	int i;
281
282	/* Root bus shouldn't use M64 */
283	if (pci_is_root_bus(bus))
284		return IODA_INVALID_PE;
285
286	/* Allocate bitmap */
287	size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long));
288	pe_alloc = kzalloc(size, GFP_KERNEL);
289	if (!pe_alloc) {
290		pr_warn("%s: Out of memory !\n",
291			__func__);
292		return IODA_INVALID_PE;
293	}
294
295	/* Figure out reserved PE numbers by the PE */
296	pnv_ioda2_reserve_m64_pe(bus, pe_alloc, all);
297
298	/*
299	 * the current bus might not own M64 window and that's all
300	 * contributed by its child buses. For the case, we needn't
301	 * pick M64 dependent PE#.
302	 */
303	if (bitmap_empty(pe_alloc, phb->ioda.total_pe)) {
304		kfree(pe_alloc);
305		return IODA_INVALID_PE;
306	}
307
308	/*
309	 * Figure out the master PE and put all slave PEs to master
310	 * PE's list to form compound PE.
311	 */
312	master_pe = NULL;
313	i = -1;
314	while ((i = find_next_bit(pe_alloc, phb->ioda.total_pe, i + 1)) <
315		phb->ioda.total_pe) {
316		pe = &phb->ioda.pe_array[i];
317
318		if (!master_pe) {
319			pe->flags |= PNV_IODA_PE_MASTER;
320			INIT_LIST_HEAD(&pe->slaves);
321			master_pe = pe;
322		} else {
323			pe->flags |= PNV_IODA_PE_SLAVE;
324			pe->master = master_pe;
325			list_add_tail(&pe->list, &master_pe->slaves);
326		}
327	}
328
329	kfree(pe_alloc);
330	return master_pe->pe_number;
331}
332
333static void __init pnv_ioda_parse_m64_window(struct pnv_phb *phb)
334{
335	struct pci_controller *hose = phb->hose;
336	struct device_node *dn = hose->dn;
337	struct resource *res;
338	const u32 *r;
339	u64 pci_addr;
340
341	/* FIXME: Support M64 for P7IOC */
342	if (phb->type != PNV_PHB_IODA2) {
343		pr_info("  Not support M64 window\n");
344		return;
345	}
346
347	if (!firmware_has_feature(FW_FEATURE_OPALv3)) {
348		pr_info("  Firmware too old to support M64 window\n");
349		return;
350	}
351
352	r = of_get_property(dn, "ibm,opal-m64-window", NULL);
353	if (!r) {
354		pr_info("  No <ibm,opal-m64-window> on %s\n",
355			dn->full_name);
356		return;
357	}
358
359	res = &hose->mem_resources[1];
360	res->start = of_translate_address(dn, r + 2);
361	res->end = res->start + of_read_number(r + 4, 2) - 1;
362	res->flags = (IORESOURCE_MEM | IORESOURCE_MEM_64 | IORESOURCE_PREFETCH);
363	pci_addr = of_read_number(r, 2);
364	hose->mem_offset[1] = res->start - pci_addr;
365
366	phb->ioda.m64_size = resource_size(res);
367	phb->ioda.m64_segsize = phb->ioda.m64_size / phb->ioda.total_pe;
368	phb->ioda.m64_base = pci_addr;
369
370	pr_info(" MEM64 0x%016llx..0x%016llx -> 0x%016llx\n",
371			res->start, res->end, pci_addr);
372
373	/* Use last M64 BAR to cover M64 window */
374	phb->ioda.m64_bar_idx = 15;
375	phb->init_m64 = pnv_ioda2_init_m64;
376	phb->reserve_m64_pe = pnv_ioda2_reserve_m64_pe;
377	phb->pick_m64_pe = pnv_ioda2_pick_m64_pe;
378}
379
380static void pnv_ioda_freeze_pe(struct pnv_phb *phb, int pe_no)
381{
382	struct pnv_ioda_pe *pe = &phb->ioda.pe_array[pe_no];
383	struct pnv_ioda_pe *slave;
384	s64 rc;
385
386	/* Fetch master PE */
387	if (pe->flags & PNV_IODA_PE_SLAVE) {
388		pe = pe->master;
389		if (WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER)))
390			return;
391
392		pe_no = pe->pe_number;
393	}
394
395	/* Freeze master PE */
396	rc = opal_pci_eeh_freeze_set(phb->opal_id,
397				     pe_no,
398				     OPAL_EEH_ACTION_SET_FREEZE_ALL);
399	if (rc != OPAL_SUCCESS) {
400		pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
401			__func__, rc, phb->hose->global_number, pe_no);
402		return;
403	}
404
405	/* Freeze slave PEs */
406	if (!(pe->flags & PNV_IODA_PE_MASTER))
407		return;
408
409	list_for_each_entry(slave, &pe->slaves, list) {
410		rc = opal_pci_eeh_freeze_set(phb->opal_id,
411					     slave->pe_number,
412					     OPAL_EEH_ACTION_SET_FREEZE_ALL);
413		if (rc != OPAL_SUCCESS)
414			pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
415				__func__, rc, phb->hose->global_number,
416				slave->pe_number);
417	}
418}
419
420static int pnv_ioda_unfreeze_pe(struct pnv_phb *phb, int pe_no, int opt)
421{
422	struct pnv_ioda_pe *pe, *slave;
423	s64 rc;
424
425	/* Find master PE */
426	pe = &phb->ioda.pe_array[pe_no];
427	if (pe->flags & PNV_IODA_PE_SLAVE) {
428		pe = pe->master;
429		WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER));
430		pe_no = pe->pe_number;
431	}
432
433	/* Clear frozen state for master PE */
434	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no, opt);
435	if (rc != OPAL_SUCCESS) {
436		pr_warn("%s: Failure %lld clear %d on PHB#%x-PE#%x\n",
437			__func__, rc, opt, phb->hose->global_number, pe_no);
438		return -EIO;
439	}
440
441	if (!(pe->flags & PNV_IODA_PE_MASTER))
442		return 0;
443
444	/* Clear frozen state for slave PEs */
445	list_for_each_entry(slave, &pe->slaves, list) {
446		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
447					     slave->pe_number,
448					     opt);
449		if (rc != OPAL_SUCCESS) {
450			pr_warn("%s: Failure %lld clear %d on PHB#%x-PE#%x\n",
451				__func__, rc, opt, phb->hose->global_number,
452				slave->pe_number);
453			return -EIO;
454		}
455	}
456
457	return 0;
458}
459
460static int pnv_ioda_get_pe_state(struct pnv_phb *phb, int pe_no)
461{
462	struct pnv_ioda_pe *slave, *pe;
463	u8 fstate, state;
464	__be16 pcierr;
465	s64 rc;
466
467	/* Sanity check on PE number */
468	if (pe_no < 0 || pe_no >= phb->ioda.total_pe)
469		return OPAL_EEH_STOPPED_PERM_UNAVAIL;
470
471	/*
472	 * Fetch the master PE and the PE instance might be
473	 * not initialized yet.
474	 */
475	pe = &phb->ioda.pe_array[pe_no];
476	if (pe->flags & PNV_IODA_PE_SLAVE) {
477		pe = pe->master;
478		WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER));
479		pe_no = pe->pe_number;
480	}
481
482	/* Check the master PE */
483	rc = opal_pci_eeh_freeze_status(phb->opal_id, pe_no,
484					&state, &pcierr, NULL);
485	if (rc != OPAL_SUCCESS) {
486		pr_warn("%s: Failure %lld getting "
487			"PHB#%x-PE#%x state\n",
488			__func__, rc,
489			phb->hose->global_number, pe_no);
490		return OPAL_EEH_STOPPED_TEMP_UNAVAIL;
491	}
492
493	/* Check the slave PE */
494	if (!(pe->flags & PNV_IODA_PE_MASTER))
495		return state;
496
497	list_for_each_entry(slave, &pe->slaves, list) {
498		rc = opal_pci_eeh_freeze_status(phb->opal_id,
499						slave->pe_number,
500						&fstate,
501						&pcierr,
502						NULL);
503		if (rc != OPAL_SUCCESS) {
504			pr_warn("%s: Failure %lld getting "
505				"PHB#%x-PE#%x state\n",
506				__func__, rc,
507				phb->hose->global_number, slave->pe_number);
508			return OPAL_EEH_STOPPED_TEMP_UNAVAIL;
509		}
510
511		/*
512		 * Override the result based on the ascending
513		 * priority.
514		 */
515		if (fstate > state)
516			state = fstate;
517	}
518
519	return state;
520}
521
522/* Currently those 2 are only used when MSIs are enabled, this will change
523 * but in the meantime, we need to protect them to avoid warnings
524 */
525#ifdef CONFIG_PCI_MSI
526static struct pnv_ioda_pe *pnv_ioda_get_pe(struct pci_dev *dev)
527{
528	struct pci_controller *hose = pci_bus_to_host(dev->bus);
529	struct pnv_phb *phb = hose->private_data;
530	struct pci_dn *pdn = pci_get_pdn(dev);
531
532	if (!pdn)
533		return NULL;
534	if (pdn->pe_number == IODA_INVALID_PE)
535		return NULL;
536	return &phb->ioda.pe_array[pdn->pe_number];
537}
538#endif /* CONFIG_PCI_MSI */
539
540static int pnv_ioda_set_one_peltv(struct pnv_phb *phb,
541				  struct pnv_ioda_pe *parent,
542				  struct pnv_ioda_pe *child,
543				  bool is_add)
544{
545	const char *desc = is_add ? "adding" : "removing";
546	uint8_t op = is_add ? OPAL_ADD_PE_TO_DOMAIN :
547			      OPAL_REMOVE_PE_FROM_DOMAIN;
548	struct pnv_ioda_pe *slave;
549	long rc;
550
551	/* Parent PE affects child PE */
552	rc = opal_pci_set_peltv(phb->opal_id, parent->pe_number,
553				child->pe_number, op);
554	if (rc != OPAL_SUCCESS) {
555		pe_warn(child, "OPAL error %ld %s to parent PELTV\n",
556			rc, desc);
557		return -ENXIO;
558	}
559
560	if (!(child->flags & PNV_IODA_PE_MASTER))
561		return 0;
562
563	/* Compound case: parent PE affects slave PEs */
564	list_for_each_entry(slave, &child->slaves, list) {
565		rc = opal_pci_set_peltv(phb->opal_id, parent->pe_number,
566					slave->pe_number, op);
567		if (rc != OPAL_SUCCESS) {
568			pe_warn(slave, "OPAL error %ld %s to parent PELTV\n",
569				rc, desc);
570			return -ENXIO;
571		}
572	}
573
574	return 0;
575}
576
577static int pnv_ioda_set_peltv(struct pnv_phb *phb,
578			      struct pnv_ioda_pe *pe,
579			      bool is_add)
580{
581	struct pnv_ioda_pe *slave;
582	struct pci_dev *pdev = NULL;
583	int ret;
584
585	/*
586	 * Clear PE frozen state. If it's master PE, we need
587	 * clear slave PE frozen state as well.
588	 */
589	if (is_add) {
590		opal_pci_eeh_freeze_clear(phb->opal_id, pe->pe_number,
591					  OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
592		if (pe->flags & PNV_IODA_PE_MASTER) {
593			list_for_each_entry(slave, &pe->slaves, list)
594				opal_pci_eeh_freeze_clear(phb->opal_id,
595							  slave->pe_number,
596							  OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
597		}
598	}
599
600	/*
601	 * Associate PE in PELT. We need add the PE into the
602	 * corresponding PELT-V as well. Otherwise, the error
603	 * originated from the PE might contribute to other
604	 * PEs.
605	 */
606	ret = pnv_ioda_set_one_peltv(phb, pe, pe, is_add);
607	if (ret)
608		return ret;
609
610	/* For compound PEs, any one affects all of them */
611	if (pe->flags & PNV_IODA_PE_MASTER) {
612		list_for_each_entry(slave, &pe->slaves, list) {
613			ret = pnv_ioda_set_one_peltv(phb, slave, pe, is_add);
614			if (ret)
615				return ret;
616		}
617	}
618
619	if (pe->flags & (PNV_IODA_PE_BUS_ALL | PNV_IODA_PE_BUS))
620		pdev = pe->pbus->self;
621	else if (pe->flags & PNV_IODA_PE_DEV)
622		pdev = pe->pdev->bus->self;
623#ifdef CONFIG_PCI_IOV
624	else if (pe->flags & PNV_IODA_PE_VF)
625		pdev = pe->parent_dev;
626#endif /* CONFIG_PCI_IOV */
627	while (pdev) {
628		struct pci_dn *pdn = pci_get_pdn(pdev);
629		struct pnv_ioda_pe *parent;
630
631		if (pdn && pdn->pe_number != IODA_INVALID_PE) {
632			parent = &phb->ioda.pe_array[pdn->pe_number];
633			ret = pnv_ioda_set_one_peltv(phb, parent, pe, is_add);
634			if (ret)
635				return ret;
636		}
637
638		pdev = pdev->bus->self;
639	}
640
641	return 0;
642}
643
644#ifdef CONFIG_PCI_IOV
645static int pnv_ioda_deconfigure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
646{
647	struct pci_dev *parent;
648	uint8_t bcomp, dcomp, fcomp;
649	int64_t rc;
650	long rid_end, rid;
651
652	/* Currently, we just deconfigure VF PE. Bus PE will always there.*/
653	if (pe->pbus) {
654		int count;
655
656		dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER;
657		fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER;
658		parent = pe->pbus->self;
659		if (pe->flags & PNV_IODA_PE_BUS_ALL)
660			count = pe->pbus->busn_res.end - pe->pbus->busn_res.start + 1;
661		else
662			count = 1;
663
664		switch(count) {
665		case  1: bcomp = OpalPciBusAll;         break;
666		case  2: bcomp = OpalPciBus7Bits;       break;
667		case  4: bcomp = OpalPciBus6Bits;       break;
668		case  8: bcomp = OpalPciBus5Bits;       break;
669		case 16: bcomp = OpalPciBus4Bits;       break;
670		case 32: bcomp = OpalPciBus3Bits;       break;
671		default:
672			dev_err(&pe->pbus->dev, "Number of subordinate buses %d unsupported\n",
673			        count);
674			/* Do an exact match only */
675			bcomp = OpalPciBusAll;
676		}
677		rid_end = pe->rid + (count << 8);
678	} else {
679		if (pe->flags & PNV_IODA_PE_VF)
680			parent = pe->parent_dev;
681		else
682			parent = pe->pdev->bus->self;
683		bcomp = OpalPciBusAll;
684		dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
685		fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER;
686		rid_end = pe->rid + 1;
687	}
688
689	/* Clear the reverse map */
690	for (rid = pe->rid; rid < rid_end; rid++)
691		phb->ioda.pe_rmap[rid] = 0;
692
693	/* Release from all parents PELT-V */
694	while (parent) {
695		struct pci_dn *pdn = pci_get_pdn(parent);
696		if (pdn && pdn->pe_number != IODA_INVALID_PE) {
697			rc = opal_pci_set_peltv(phb->opal_id, pdn->pe_number,
698						pe->pe_number, OPAL_REMOVE_PE_FROM_DOMAIN);
699			/* XXX What to do in case of error ? */
700		}
701		parent = parent->bus->self;
702	}
703
704	opal_pci_eeh_freeze_clear(phb->opal_id, pe->pe_number,
705				  OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
706
707	/* Disassociate PE in PELT */
708	rc = opal_pci_set_peltv(phb->opal_id, pe->pe_number,
709				pe->pe_number, OPAL_REMOVE_PE_FROM_DOMAIN);
710	if (rc)
711		pe_warn(pe, "OPAL error %ld remove self from PELTV\n", rc);
712	rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid,
713			     bcomp, dcomp, fcomp, OPAL_UNMAP_PE);
714	if (rc)
715		pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc);
716
717	pe->pbus = NULL;
718	pe->pdev = NULL;
719	pe->parent_dev = NULL;
720
721	return 0;
722}
723#endif /* CONFIG_PCI_IOV */
724
725static int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
726{
727	struct pci_dev *parent;
728	uint8_t bcomp, dcomp, fcomp;
729	long rc, rid_end, rid;
730
731	/* Bus validation ? */
732	if (pe->pbus) {
733		int count;
734
735		dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER;
736		fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER;
737		parent = pe->pbus->self;
738		if (pe->flags & PNV_IODA_PE_BUS_ALL)
739			count = pe->pbus->busn_res.end - pe->pbus->busn_res.start + 1;
740		else
741			count = 1;
742
743		switch(count) {
744		case  1: bcomp = OpalPciBusAll;		break;
745		case  2: bcomp = OpalPciBus7Bits;	break;
746		case  4: bcomp = OpalPciBus6Bits;	break;
747		case  8: bcomp = OpalPciBus5Bits;	break;
748		case 16: bcomp = OpalPciBus4Bits;	break;
749		case 32: bcomp = OpalPciBus3Bits;	break;
750		default:
751			dev_err(&pe->pbus->dev, "Number of subordinate buses %d unsupported\n",
752			        count);
753			/* Do an exact match only */
754			bcomp = OpalPciBusAll;
755		}
756		rid_end = pe->rid + (count << 8);
757	} else {
758#ifdef CONFIG_PCI_IOV
759		if (pe->flags & PNV_IODA_PE_VF)
760			parent = pe->parent_dev;
761		else
762#endif /* CONFIG_PCI_IOV */
763			parent = pe->pdev->bus->self;
764		bcomp = OpalPciBusAll;
765		dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
766		fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER;
767		rid_end = pe->rid + 1;
768	}
769
770	/*
771	 * Associate PE in PELT. We need add the PE into the
772	 * corresponding PELT-V as well. Otherwise, the error
773	 * originated from the PE might contribute to other
774	 * PEs.
775	 */
776	rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid,
777			     bcomp, dcomp, fcomp, OPAL_MAP_PE);
778	if (rc) {
779		pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc);
780		return -ENXIO;
781	}
782
783	/* Configure PELTV */
784	pnv_ioda_set_peltv(phb, pe, true);
785
786	/* Setup reverse map */
787	for (rid = pe->rid; rid < rid_end; rid++)
788		phb->ioda.pe_rmap[rid] = pe->pe_number;
789
790	/* Setup one MVTs on IODA1 */
791	if (phb->type != PNV_PHB_IODA1) {
792		pe->mve_number = 0;
793		goto out;
794	}
795
796	pe->mve_number = pe->pe_number;
797	rc = opal_pci_set_mve(phb->opal_id, pe->mve_number, pe->pe_number);
798	if (rc != OPAL_SUCCESS) {
799		pe_err(pe, "OPAL error %ld setting up MVE %d\n",
800		       rc, pe->mve_number);
801		pe->mve_number = -1;
802	} else {
803		rc = opal_pci_set_mve_enable(phb->opal_id,
804					     pe->mve_number, OPAL_ENABLE_MVE);
805		if (rc) {
806			pe_err(pe, "OPAL error %ld enabling MVE %d\n",
807			       rc, pe->mve_number);
808			pe->mve_number = -1;
809		}
810	}
811
812out:
813	return 0;
814}
815
816static void pnv_ioda_link_pe_by_weight(struct pnv_phb *phb,
817				       struct pnv_ioda_pe *pe)
818{
819	struct pnv_ioda_pe *lpe;
820
821	list_for_each_entry(lpe, &phb->ioda.pe_dma_list, dma_link) {
822		if (lpe->dma_weight < pe->dma_weight) {
823			list_add_tail(&pe->dma_link, &lpe->dma_link);
824			return;
825		}
826	}
827	list_add_tail(&pe->dma_link, &phb->ioda.pe_dma_list);
828}
829
830static unsigned int pnv_ioda_dma_weight(struct pci_dev *dev)
831{
832	/* This is quite simplistic. The "base" weight of a device
833	 * is 10. 0 means no DMA is to be accounted for it.
834	 */
835
836	/* If it's a bridge, no DMA */
837	if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
838		return 0;
839
840	/* Reduce the weight of slow USB controllers */
841	if (dev->class == PCI_CLASS_SERIAL_USB_UHCI ||
842	    dev->class == PCI_CLASS_SERIAL_USB_OHCI ||
843	    dev->class == PCI_CLASS_SERIAL_USB_EHCI)
844		return 3;
845
846	/* Increase the weight of RAID (includes Obsidian) */
847	if ((dev->class >> 8) == PCI_CLASS_STORAGE_RAID)
848		return 15;
849
850	/* Default */
851	return 10;
852}
853
854#ifdef CONFIG_PCI_IOV
855static int pnv_pci_vf_resource_shift(struct pci_dev *dev, int offset)
856{
857	struct pci_dn *pdn = pci_get_pdn(dev);
858	int i;
859	struct resource *res, res2;
860	resource_size_t size;
861	u16 num_vfs;
862
863	if (!dev->is_physfn)
864		return -EINVAL;
865
866	/*
867	 * "offset" is in VFs.  The M64 windows are sized so that when they
868	 * are segmented, each segment is the same size as the IOV BAR.
869	 * Each segment is in a separate PE, and the high order bits of the
870	 * address are the PE number.  Therefore, each VF's BAR is in a
871	 * separate PE, and changing the IOV BAR start address changes the
872	 * range of PEs the VFs are in.
873	 */
874	num_vfs = pdn->num_vfs;
875	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
876		res = &dev->resource[i + PCI_IOV_RESOURCES];
877		if (!res->flags || !res->parent)
878			continue;
879
880		if (!pnv_pci_is_mem_pref_64(res->flags))
881			continue;
882
883		/*
884		 * The actual IOV BAR range is determined by the start address
885		 * and the actual size for num_vfs VFs BAR.  This check is to
886		 * make sure that after shifting, the range will not overlap
887		 * with another device.
888		 */
889		size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
890		res2.flags = res->flags;
891		res2.start = res->start + (size * offset);
892		res2.end = res2.start + (size * num_vfs) - 1;
893
894		if (res2.end > res->end) {
895			dev_err(&dev->dev, "VF BAR%d: %pR would extend past %pR (trying to enable %d VFs shifted by %d)\n",
896				i, &res2, res, num_vfs, offset);
897			return -EBUSY;
898		}
899	}
900
901	/*
902	 * After doing so, there would be a "hole" in the /proc/iomem when
903	 * offset is a positive value. It looks like the device return some
904	 * mmio back to the system, which actually no one could use it.
905	 */
906	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
907		res = &dev->resource[i + PCI_IOV_RESOURCES];
908		if (!res->flags || !res->parent)
909			continue;
910
911		if (!pnv_pci_is_mem_pref_64(res->flags))
912			continue;
913
914		size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
915		res2 = *res;
916		res->start += size * offset;
917
918		dev_info(&dev->dev, "VF BAR%d: %pR shifted to %pR (%sabling %d VFs shifted by %d)\n",
919			 i, &res2, res, (offset > 0) ? "En" : "Dis",
920			 num_vfs, offset);
921		pci_update_resource(dev, i + PCI_IOV_RESOURCES);
922	}
923	return 0;
924}
925#endif /* CONFIG_PCI_IOV */
926
927#if 0
928static struct pnv_ioda_pe *pnv_ioda_setup_dev_PE(struct pci_dev *dev)
929{
930	struct pci_controller *hose = pci_bus_to_host(dev->bus);
931	struct pnv_phb *phb = hose->private_data;
932	struct pci_dn *pdn = pci_get_pdn(dev);
933	struct pnv_ioda_pe *pe;
934	int pe_num;
935
936	if (!pdn) {
937		pr_err("%s: Device tree node not associated properly\n",
938			   pci_name(dev));
939		return NULL;
940	}
941	if (pdn->pe_number != IODA_INVALID_PE)
942		return NULL;
943
944	/* PE#0 has been pre-set */
945	if (dev->bus->number == 0)
946		pe_num = 0;
947	else
948		pe_num = pnv_ioda_alloc_pe(phb);
949	if (pe_num == IODA_INVALID_PE) {
950		pr_warning("%s: Not enough PE# available, disabling device\n",
951			   pci_name(dev));
952		return NULL;
953	}
954
955	/* NOTE: We get only one ref to the pci_dev for the pdn, not for the
956	 * pointer in the PE data structure, both should be destroyed at the
957	 * same time. However, this needs to be looked at more closely again
958	 * once we actually start removing things (Hotplug, SR-IOV, ...)
959	 *
960	 * At some point we want to remove the PDN completely anyways
961	 */
962	pe = &phb->ioda.pe_array[pe_num];
963	pci_dev_get(dev);
964	pdn->pcidev = dev;
965	pdn->pe_number = pe_num;
966	pe->pdev = dev;
967	pe->pbus = NULL;
968	pe->tce32_seg = -1;
969	pe->mve_number = -1;
970	pe->rid = dev->bus->number << 8 | pdn->devfn;
971
972	pe_info(pe, "Associated device to PE\n");
973
974	if (pnv_ioda_configure_pe(phb, pe)) {
975		/* XXX What do we do here ? */
976		if (pe_num)
977			pnv_ioda_free_pe(phb, pe_num);
978		pdn->pe_number = IODA_INVALID_PE;
979		pe->pdev = NULL;
980		pci_dev_put(dev);
981		return NULL;
982	}
983
984	/* Assign a DMA weight to the device */
985	pe->dma_weight = pnv_ioda_dma_weight(dev);
986	if (pe->dma_weight != 0) {
987		phb->ioda.dma_weight += pe->dma_weight;
988		phb->ioda.dma_pe_count++;
989	}
990
991	/* Link the PE */
992	pnv_ioda_link_pe_by_weight(phb, pe);
993
994	return pe;
995}
996#endif /* Useful for SRIOV case */
997
998static void pnv_ioda_setup_same_PE(struct pci_bus *bus, struct pnv_ioda_pe *pe)
999{
1000	struct pci_dev *dev;
1001
1002	list_for_each_entry(dev, &bus->devices, bus_list) {
1003		struct pci_dn *pdn = pci_get_pdn(dev);
1004
1005		if (pdn == NULL) {
1006			pr_warn("%s: No device node associated with device !\n",
1007				pci_name(dev));
1008			continue;
1009		}
1010		pdn->pe_number = pe->pe_number;
1011		pe->dma_weight += pnv_ioda_dma_weight(dev);
1012		if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate)
1013			pnv_ioda_setup_same_PE(dev->subordinate, pe);
1014	}
1015}
1016
1017/*
1018 * There're 2 types of PCI bus sensitive PEs: One that is compromised of
1019 * single PCI bus. Another one that contains the primary PCI bus and its
1020 * subordinate PCI devices and buses. The second type of PE is normally
1021 * orgiriated by PCIe-to-PCI bridge or PLX switch downstream ports.
1022 */
1023static void pnv_ioda_setup_bus_PE(struct pci_bus *bus, bool all)
1024{
1025	struct pci_controller *hose = pci_bus_to_host(bus);
1026	struct pnv_phb *phb = hose->private_data;
1027	struct pnv_ioda_pe *pe;
1028	int pe_num = IODA_INVALID_PE;
1029
1030	/* Check if PE is determined by M64 */
1031	if (phb->pick_m64_pe)
1032		pe_num = phb->pick_m64_pe(bus, all);
1033
1034	/* The PE number isn't pinned by M64 */
1035	if (pe_num == IODA_INVALID_PE)
1036		pe_num = pnv_ioda_alloc_pe(phb);
1037
1038	if (pe_num == IODA_INVALID_PE) {
1039		pr_warning("%s: Not enough PE# available for PCI bus %04x:%02x\n",
1040			__func__, pci_domain_nr(bus), bus->number);
1041		return;
1042	}
1043
1044	pe = &phb->ioda.pe_array[pe_num];
1045	pe->flags |= (all ? PNV_IODA_PE_BUS_ALL : PNV_IODA_PE_BUS);
1046	pe->pbus = bus;
1047	pe->pdev = NULL;
1048	pe->tce32_seg = -1;
1049	pe->mve_number = -1;
1050	pe->rid = bus->busn_res.start << 8;
1051	pe->dma_weight = 0;
1052
1053	if (all)
1054		pe_info(pe, "Secondary bus %d..%d associated with PE#%d\n",
1055			bus->busn_res.start, bus->busn_res.end, pe_num);
1056	else
1057		pe_info(pe, "Secondary bus %d associated with PE#%d\n",
1058			bus->busn_res.start, pe_num);
1059
1060	if (pnv_ioda_configure_pe(phb, pe)) {
1061		/* XXX What do we do here ? */
1062		if (pe_num)
1063			pnv_ioda_free_pe(phb, pe_num);
1064		pe->pbus = NULL;
1065		return;
1066	}
1067
1068	/* Associate it with all child devices */
1069	pnv_ioda_setup_same_PE(bus, pe);
1070
1071	/* Put PE to the list */
1072	list_add_tail(&pe->list, &phb->ioda.pe_list);
1073
1074	/* Account for one DMA PE if at least one DMA capable device exist
1075	 * below the bridge
1076	 */
1077	if (pe->dma_weight != 0) {
1078		phb->ioda.dma_weight += pe->dma_weight;
1079		phb->ioda.dma_pe_count++;
1080	}
1081
1082	/* Link the PE */
1083	pnv_ioda_link_pe_by_weight(phb, pe);
1084}
1085
1086static void pnv_ioda_setup_PEs(struct pci_bus *bus)
1087{
1088	struct pci_dev *dev;
1089
1090	pnv_ioda_setup_bus_PE(bus, false);
1091
1092	list_for_each_entry(dev, &bus->devices, bus_list) {
1093		if (dev->subordinate) {
1094			if (pci_pcie_type(dev) == PCI_EXP_TYPE_PCI_BRIDGE)
1095				pnv_ioda_setup_bus_PE(dev->subordinate, true);
1096			else
1097				pnv_ioda_setup_PEs(dev->subordinate);
1098		}
1099	}
1100}
1101
1102/*
1103 * Configure PEs so that the downstream PCI buses and devices
1104 * could have their associated PE#. Unfortunately, we didn't
1105 * figure out the way to identify the PLX bridge yet. So we
1106 * simply put the PCI bus and the subordinate behind the root
1107 * port to PE# here. The game rule here is expected to be changed
1108 * as soon as we can detected PLX bridge correctly.
1109 */
1110static void pnv_pci_ioda_setup_PEs(void)
1111{
1112	struct pci_controller *hose, *tmp;
1113	struct pnv_phb *phb;
1114
1115	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1116		phb = hose->private_data;
1117
1118		/* M64 layout might affect PE allocation */
1119		if (phb->reserve_m64_pe)
1120			phb->reserve_m64_pe(hose->bus, NULL, true);
1121
1122		pnv_ioda_setup_PEs(hose->bus);
1123	}
1124}
1125
1126#ifdef CONFIG_PCI_IOV
1127static int pnv_pci_vf_release_m64(struct pci_dev *pdev)
1128{
1129	struct pci_bus        *bus;
1130	struct pci_controller *hose;
1131	struct pnv_phb        *phb;
1132	struct pci_dn         *pdn;
1133	int                    i, j;
1134
1135	bus = pdev->bus;
1136	hose = pci_bus_to_host(bus);
1137	phb = hose->private_data;
1138	pdn = pci_get_pdn(pdev);
1139
1140	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
1141		for (j = 0; j < M64_PER_IOV; j++) {
1142			if (pdn->m64_wins[i][j] == IODA_INVALID_M64)
1143				continue;
1144			opal_pci_phb_mmio_enable(phb->opal_id,
1145				OPAL_M64_WINDOW_TYPE, pdn->m64_wins[i][j], 0);
1146			clear_bit(pdn->m64_wins[i][j], &phb->ioda.m64_bar_alloc);
1147			pdn->m64_wins[i][j] = IODA_INVALID_M64;
1148		}
1149
1150	return 0;
1151}
1152
1153static int pnv_pci_vf_assign_m64(struct pci_dev *pdev, u16 num_vfs)
1154{
1155	struct pci_bus        *bus;
1156	struct pci_controller *hose;
1157	struct pnv_phb        *phb;
1158	struct pci_dn         *pdn;
1159	unsigned int           win;
1160	struct resource       *res;
1161	int                    i, j;
1162	int64_t                rc;
1163	int                    total_vfs;
1164	resource_size_t        size, start;
1165	int                    pe_num;
1166	int                    vf_groups;
1167	int                    vf_per_group;
1168
1169	bus = pdev->bus;
1170	hose = pci_bus_to_host(bus);
1171	phb = hose->private_data;
1172	pdn = pci_get_pdn(pdev);
1173	total_vfs = pci_sriov_get_totalvfs(pdev);
1174
1175	/* Initialize the m64_wins to IODA_INVALID_M64 */
1176	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
1177		for (j = 0; j < M64_PER_IOV; j++)
1178			pdn->m64_wins[i][j] = IODA_INVALID_M64;
1179
1180	if (pdn->m64_per_iov == M64_PER_IOV) {
1181		vf_groups = (num_vfs <= M64_PER_IOV) ? num_vfs: M64_PER_IOV;
1182		vf_per_group = (num_vfs <= M64_PER_IOV)? 1:
1183			roundup_pow_of_two(num_vfs) / pdn->m64_per_iov;
1184	} else {
1185		vf_groups = 1;
1186		vf_per_group = 1;
1187	}
1188
1189	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
1190		res = &pdev->resource[i + PCI_IOV_RESOURCES];
1191		if (!res->flags || !res->parent)
1192			continue;
1193
1194		if (!pnv_pci_is_mem_pref_64(res->flags))
1195			continue;
1196
1197		for (j = 0; j < vf_groups; j++) {
1198			do {
1199				win = find_next_zero_bit(&phb->ioda.m64_bar_alloc,
1200						phb->ioda.m64_bar_idx + 1, 0);
1201
1202				if (win >= phb->ioda.m64_bar_idx + 1)
1203					goto m64_failed;
1204			} while (test_and_set_bit(win, &phb->ioda.m64_bar_alloc));
1205
1206			pdn->m64_wins[i][j] = win;
1207
1208			if (pdn->m64_per_iov == M64_PER_IOV) {
1209				size = pci_iov_resource_size(pdev,
1210							PCI_IOV_RESOURCES + i);
1211				size = size * vf_per_group;
1212				start = res->start + size * j;
1213			} else {
1214				size = resource_size(res);
1215				start = res->start;
1216			}
1217
1218			/* Map the M64 here */
1219			if (pdn->m64_per_iov == M64_PER_IOV) {
1220				pe_num = pdn->offset + j;
1221				rc = opal_pci_map_pe_mmio_window(phb->opal_id,
1222						pe_num, OPAL_M64_WINDOW_TYPE,
1223						pdn->m64_wins[i][j], 0);
1224			}
1225
1226			rc = opal_pci_set_phb_mem_window(phb->opal_id,
1227						 OPAL_M64_WINDOW_TYPE,
1228						 pdn->m64_wins[i][j],
1229						 start,
1230						 0, /* unused */
1231						 size);
1232
1233
1234			if (rc != OPAL_SUCCESS) {
1235				dev_err(&pdev->dev, "Failed to map M64 window #%d: %lld\n",
1236					win, rc);
1237				goto m64_failed;
1238			}
1239
1240			if (pdn->m64_per_iov == M64_PER_IOV)
1241				rc = opal_pci_phb_mmio_enable(phb->opal_id,
1242				     OPAL_M64_WINDOW_TYPE, pdn->m64_wins[i][j], 2);
1243			else
1244				rc = opal_pci_phb_mmio_enable(phb->opal_id,
1245				     OPAL_M64_WINDOW_TYPE, pdn->m64_wins[i][j], 1);
1246
1247			if (rc != OPAL_SUCCESS) {
1248				dev_err(&pdev->dev, "Failed to enable M64 window #%d: %llx\n",
1249					win, rc);
1250				goto m64_failed;
1251			}
1252		}
1253	}
1254	return 0;
1255
1256m64_failed:
1257	pnv_pci_vf_release_m64(pdev);
1258	return -EBUSY;
1259}
1260
1261static long pnv_pci_ioda2_unset_window(struct iommu_table_group *table_group,
1262		int num);
1263static void pnv_pci_ioda2_set_bypass(struct pnv_ioda_pe *pe, bool enable);
1264
1265static void pnv_pci_ioda2_release_dma_pe(struct pci_dev *dev, struct pnv_ioda_pe *pe)
1266{
1267	struct iommu_table    *tbl;
1268	int64_t               rc;
1269
1270	tbl = pe->table_group.tables[0];
1271	rc = pnv_pci_ioda2_unset_window(&pe->table_group, 0);
1272	if (rc)
1273		pe_warn(pe, "OPAL error %ld release DMA window\n", rc);
1274
1275	pnv_pci_ioda2_set_bypass(pe, false);
1276	if (pe->table_group.group) {
1277		iommu_group_put(pe->table_group.group);
1278		BUG_ON(pe->table_group.group);
1279	}
1280	pnv_pci_ioda2_table_free_pages(tbl);
1281	iommu_free_table(tbl, of_node_full_name(dev->dev.of_node));
1282}
1283
1284static void pnv_ioda_release_vf_PE(struct pci_dev *pdev, u16 num_vfs)
1285{
1286	struct pci_bus        *bus;
1287	struct pci_controller *hose;
1288	struct pnv_phb        *phb;
1289	struct pnv_ioda_pe    *pe, *pe_n;
1290	struct pci_dn         *pdn;
1291	u16                    vf_index;
1292	int64_t                rc;
1293
1294	bus = pdev->bus;
1295	hose = pci_bus_to_host(bus);
1296	phb = hose->private_data;
1297	pdn = pci_get_pdn(pdev);
1298
1299	if (!pdev->is_physfn)
1300		return;
1301
1302	if (pdn->m64_per_iov == M64_PER_IOV && num_vfs > M64_PER_IOV) {
1303		int   vf_group;
1304		int   vf_per_group;
1305		int   vf_index1;
1306
1307		vf_per_group = roundup_pow_of_two(num_vfs) / pdn->m64_per_iov;
1308
1309		for (vf_group = 0; vf_group < M64_PER_IOV; vf_group++)
1310			for (vf_index = vf_group * vf_per_group;
1311				vf_index < (vf_group + 1) * vf_per_group &&
1312				vf_index < num_vfs;
1313				vf_index++)
1314				for (vf_index1 = vf_group * vf_per_group;
1315					vf_index1 < (vf_group + 1) * vf_per_group &&
1316					vf_index1 < num_vfs;
1317					vf_index1++){
1318
1319					rc = opal_pci_set_peltv(phb->opal_id,
1320						pdn->offset + vf_index,
1321						pdn->offset + vf_index1,
1322						OPAL_REMOVE_PE_FROM_DOMAIN);
1323
1324					if (rc)
1325					    dev_warn(&pdev->dev, "%s: Failed to unlink same group PE#%d(%lld)\n",
1326						__func__,
1327						pdn->offset + vf_index1, rc);
1328				}
1329	}
1330
1331	list_for_each_entry_safe(pe, pe_n, &phb->ioda.pe_list, list) {
1332		if (pe->parent_dev != pdev)
1333			continue;
1334
1335		pnv_pci_ioda2_release_dma_pe(pdev, pe);
1336
1337		/* Remove from list */
1338		mutex_lock(&phb->ioda.pe_list_mutex);
1339		list_del(&pe->list);
1340		mutex_unlock(&phb->ioda.pe_list_mutex);
1341
1342		pnv_ioda_deconfigure_pe(phb, pe);
1343
1344		pnv_ioda_free_pe(phb, pe->pe_number);
1345	}
1346}
1347
1348void pnv_pci_sriov_disable(struct pci_dev *pdev)
1349{
1350	struct pci_bus        *bus;
1351	struct pci_controller *hose;
1352	struct pnv_phb        *phb;
1353	struct pci_dn         *pdn;
1354	struct pci_sriov      *iov;
1355	u16 num_vfs;
1356
1357	bus = pdev->bus;
1358	hose = pci_bus_to_host(bus);
1359	phb = hose->private_data;
1360	pdn = pci_get_pdn(pdev);
1361	iov = pdev->sriov;
1362	num_vfs = pdn->num_vfs;
1363
1364	/* Release VF PEs */
1365	pnv_ioda_release_vf_PE(pdev, num_vfs);
1366
1367	if (phb->type == PNV_PHB_IODA2) {
1368		if (pdn->m64_per_iov == 1)
1369			pnv_pci_vf_resource_shift(pdev, -pdn->offset);
1370
1371		/* Release M64 windows */
1372		pnv_pci_vf_release_m64(pdev);
1373
1374		/* Release PE numbers */
1375		bitmap_clear(phb->ioda.pe_alloc, pdn->offset, num_vfs);
1376		pdn->offset = 0;
1377	}
1378}
1379
1380static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb,
1381				       struct pnv_ioda_pe *pe);
1382static void pnv_ioda_setup_vf_PE(struct pci_dev *pdev, u16 num_vfs)
1383{
1384	struct pci_bus        *bus;
1385	struct pci_controller *hose;
1386	struct pnv_phb        *phb;
1387	struct pnv_ioda_pe    *pe;
1388	int                    pe_num;
1389	u16                    vf_index;
1390	struct pci_dn         *pdn;
1391	int64_t                rc;
1392
1393	bus = pdev->bus;
1394	hose = pci_bus_to_host(bus);
1395	phb = hose->private_data;
1396	pdn = pci_get_pdn(pdev);
1397
1398	if (!pdev->is_physfn)
1399		return;
1400
1401	/* Reserve PE for each VF */
1402	for (vf_index = 0; vf_index < num_vfs; vf_index++) {
1403		pe_num = pdn->offset + vf_index;
1404
1405		pe = &phb->ioda.pe_array[pe_num];
1406		pe->pe_number = pe_num;
1407		pe->phb = phb;
1408		pe->flags = PNV_IODA_PE_VF;
1409		pe->pbus = NULL;
1410		pe->parent_dev = pdev;
1411		pe->tce32_seg = -1;
1412		pe->mve_number = -1;
1413		pe->rid = (pci_iov_virtfn_bus(pdev, vf_index) << 8) |
1414			   pci_iov_virtfn_devfn(pdev, vf_index);
1415
1416		pe_info(pe, "VF %04d:%02d:%02d.%d associated with PE#%d\n",
1417			hose->global_number, pdev->bus->number,
1418			PCI_SLOT(pci_iov_virtfn_devfn(pdev, vf_index)),
1419			PCI_FUNC(pci_iov_virtfn_devfn(pdev, vf_index)), pe_num);
1420
1421		if (pnv_ioda_configure_pe(phb, pe)) {
1422			/* XXX What do we do here ? */
1423			if (pe_num)
1424				pnv_ioda_free_pe(phb, pe_num);
1425			pe->pdev = NULL;
1426			continue;
1427		}
1428
1429		/* Put PE to the list */
1430		mutex_lock(&phb->ioda.pe_list_mutex);
1431		list_add_tail(&pe->list, &phb->ioda.pe_list);
1432		mutex_unlock(&phb->ioda.pe_list_mutex);
1433
1434		pnv_pci_ioda2_setup_dma_pe(phb, pe);
1435	}
1436
1437	if (pdn->m64_per_iov == M64_PER_IOV && num_vfs > M64_PER_IOV) {
1438		int   vf_group;
1439		int   vf_per_group;
1440		int   vf_index1;
1441
1442		vf_per_group = roundup_pow_of_two(num_vfs) / pdn->m64_per_iov;
1443
1444		for (vf_group = 0; vf_group < M64_PER_IOV; vf_group++) {
1445			for (vf_index = vf_group * vf_per_group;
1446			     vf_index < (vf_group + 1) * vf_per_group &&
1447			     vf_index < num_vfs;
1448			     vf_index++) {
1449				for (vf_index1 = vf_group * vf_per_group;
1450				     vf_index1 < (vf_group + 1) * vf_per_group &&
1451				     vf_index1 < num_vfs;
1452				     vf_index1++) {
1453
1454					rc = opal_pci_set_peltv(phb->opal_id,
1455						pdn->offset + vf_index,
1456						pdn->offset + vf_index1,
1457						OPAL_ADD_PE_TO_DOMAIN);
1458
1459					if (rc)
1460					    dev_warn(&pdev->dev, "%s: Failed to link same group PE#%d(%lld)\n",
1461						__func__,
1462						pdn->offset + vf_index1, rc);
1463				}
1464			}
1465		}
1466	}
1467}
1468
1469int pnv_pci_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
1470{
1471	struct pci_bus        *bus;
1472	struct pci_controller *hose;
1473	struct pnv_phb        *phb;
1474	struct pci_dn         *pdn;
1475	int                    ret;
1476
1477	bus = pdev->bus;
1478	hose = pci_bus_to_host(bus);
1479	phb = hose->private_data;
1480	pdn = pci_get_pdn(pdev);
1481
1482	if (phb->type == PNV_PHB_IODA2) {
1483		/* Calculate available PE for required VFs */
1484		mutex_lock(&phb->ioda.pe_alloc_mutex);
1485		pdn->offset = bitmap_find_next_zero_area(
1486			phb->ioda.pe_alloc, phb->ioda.total_pe,
1487			0, num_vfs, 0);
1488		if (pdn->offset >= phb->ioda.total_pe) {
1489			mutex_unlock(&phb->ioda.pe_alloc_mutex);
1490			dev_info(&pdev->dev, "Failed to enable VF%d\n", num_vfs);
1491			pdn->offset = 0;
1492			return -EBUSY;
1493		}
1494		bitmap_set(phb->ioda.pe_alloc, pdn->offset, num_vfs);
1495		pdn->num_vfs = num_vfs;
1496		mutex_unlock(&phb->ioda.pe_alloc_mutex);
1497
1498		/* Assign M64 window accordingly */
1499		ret = pnv_pci_vf_assign_m64(pdev, num_vfs);
1500		if (ret) {
1501			dev_info(&pdev->dev, "Not enough M64 window resources\n");
1502			goto m64_failed;
1503		}
1504
1505		/*
1506		 * When using one M64 BAR to map one IOV BAR, we need to shift
1507		 * the IOV BAR according to the PE# allocated to the VFs.
1508		 * Otherwise, the PE# for the VF will conflict with others.
1509		 */
1510		if (pdn->m64_per_iov == 1) {
1511			ret = pnv_pci_vf_resource_shift(pdev, pdn->offset);
1512			if (ret)
1513				goto m64_failed;
1514		}
1515	}
1516
1517	/* Setup VF PEs */
1518	pnv_ioda_setup_vf_PE(pdev, num_vfs);
1519
1520	return 0;
1521
1522m64_failed:
1523	bitmap_clear(phb->ioda.pe_alloc, pdn->offset, num_vfs);
1524	pdn->offset = 0;
1525
1526	return ret;
1527}
1528
1529int pcibios_sriov_disable(struct pci_dev *pdev)
1530{
1531	pnv_pci_sriov_disable(pdev);
1532
1533	/* Release PCI data */
1534	remove_dev_pci_data(pdev);
1535	return 0;
1536}
1537
1538int pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
1539{
1540	/* Allocate PCI data */
1541	add_dev_pci_data(pdev);
1542
1543	pnv_pci_sriov_enable(pdev, num_vfs);
1544	return 0;
1545}
1546#endif /* CONFIG_PCI_IOV */
1547
1548static void pnv_pci_ioda_dma_dev_setup(struct pnv_phb *phb, struct pci_dev *pdev)
1549{
1550	struct pci_dn *pdn = pci_get_pdn(pdev);
1551	struct pnv_ioda_pe *pe;
1552
1553	/*
1554	 * The function can be called while the PE#
1555	 * hasn't been assigned. Do nothing for the
1556	 * case.
1557	 */
1558	if (!pdn || pdn->pe_number == IODA_INVALID_PE)
1559		return;
1560
1561	pe = &phb->ioda.pe_array[pdn->pe_number];
1562	WARN_ON(get_dma_ops(&pdev->dev) != &dma_iommu_ops);
1563	set_dma_offset(&pdev->dev, pe->tce_bypass_base);
1564	set_iommu_table_base(&pdev->dev, pe->table_group.tables[0]);
1565	/*
1566	 * Note: iommu_add_device() will fail here as
1567	 * for physical PE: the device is already added by now;
1568	 * for virtual PE: sysfs entries are not ready yet and
1569	 * tce_iommu_bus_notifier will add the device to a group later.
1570	 */
1571}
1572
1573static int pnv_pci_ioda_dma_set_mask(struct pci_dev *pdev, u64 dma_mask)
1574{
1575	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
1576	struct pnv_phb *phb = hose->private_data;
1577	struct pci_dn *pdn = pci_get_pdn(pdev);
1578	struct pnv_ioda_pe *pe;
1579	uint64_t top;
1580	bool bypass = false;
1581
1582	if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE))
1583		return -ENODEV;;
1584
1585	pe = &phb->ioda.pe_array[pdn->pe_number];
1586	if (pe->tce_bypass_enabled) {
1587		top = pe->tce_bypass_base + memblock_end_of_DRAM() - 1;
1588		bypass = (dma_mask >= top);
1589	}
1590
1591	if (bypass) {
1592		dev_info(&pdev->dev, "Using 64-bit DMA iommu bypass\n");
1593		set_dma_ops(&pdev->dev, &dma_direct_ops);
1594	} else {
1595		dev_info(&pdev->dev, "Using 32-bit DMA via iommu\n");
1596		set_dma_ops(&pdev->dev, &dma_iommu_ops);
1597	}
1598	*pdev->dev.dma_mask = dma_mask;
1599	return 0;
1600}
1601
1602static u64 pnv_pci_ioda_dma_get_required_mask(struct pci_dev *pdev)
1603{
1604	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
1605	struct pnv_phb *phb = hose->private_data;
1606	struct pci_dn *pdn = pci_get_pdn(pdev);
1607	struct pnv_ioda_pe *pe;
1608	u64 end, mask;
1609
1610	if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE))
1611		return 0;
1612
1613	pe = &phb->ioda.pe_array[pdn->pe_number];
1614	if (!pe->tce_bypass_enabled)
1615		return __dma_get_required_mask(&pdev->dev);
1616
1617
1618	end = pe->tce_bypass_base + memblock_end_of_DRAM();
1619	mask = 1ULL << (fls64(end) - 1);
1620	mask += mask - 1;
1621
1622	return mask;
1623}
1624
1625static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe,
1626				   struct pci_bus *bus)
1627{
1628	struct pci_dev *dev;
1629
1630	list_for_each_entry(dev, &bus->devices, bus_list) {
1631		set_iommu_table_base(&dev->dev, pe->table_group.tables[0]);
1632		set_dma_offset(&dev->dev, pe->tce_bypass_base);
1633		iommu_add_device(&dev->dev);
1634
1635		if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate)
1636			pnv_ioda_setup_bus_dma(pe, dev->subordinate);
1637	}
1638}
1639
1640static void pnv_pci_ioda1_tce_invalidate(struct iommu_table *tbl,
1641		unsigned long index, unsigned long npages, bool rm)
1642{
1643	struct iommu_table_group_link *tgl = list_first_entry_or_null(
1644			&tbl->it_group_list, struct iommu_table_group_link,
1645			next);
1646	struct pnv_ioda_pe *pe = container_of(tgl->table_group,
1647			struct pnv_ioda_pe, table_group);
1648	__be64 __iomem *invalidate = rm ?
1649		(__be64 __iomem *)pe->phb->ioda.tce_inval_reg_phys :
1650		pe->phb->ioda.tce_inval_reg;
1651	unsigned long start, end, inc;
1652	const unsigned shift = tbl->it_page_shift;
1653
1654	start = __pa(((__be64 *)tbl->it_base) + index - tbl->it_offset);
1655	end = __pa(((__be64 *)tbl->it_base) + index - tbl->it_offset +
1656			npages - 1);
1657
1658	/* BML uses this case for p6/p7/galaxy2: Shift addr and put in node */
1659	if (tbl->it_busno) {
1660		start <<= shift;
1661		end <<= shift;
1662		inc = 128ull << shift;
1663		start |= tbl->it_busno;
1664		end |= tbl->it_busno;
1665	} else if (tbl->it_type & TCE_PCI_SWINV_PAIR) {
1666		/* p7ioc-style invalidation, 2 TCEs per write */
1667		start |= (1ull << 63);
1668		end |= (1ull << 63);
1669		inc = 16;
1670        } else {
1671		/* Default (older HW) */
1672                inc = 128;
1673	}
1674
1675        end |= inc - 1;	/* round up end to be different than start */
1676
1677        mb(); /* Ensure above stores are visible */
1678        while (start <= end) {
1679		if (rm)
1680			__raw_rm_writeq(cpu_to_be64(start), invalidate);
1681		else
1682			__raw_writeq(cpu_to_be64(start), invalidate);
1683                start += inc;
1684        }
1685
1686	/*
1687	 * The iommu layer will do another mb() for us on build()
1688	 * and we don't care on free()
1689	 */
1690}
1691
1692static int pnv_ioda1_tce_build(struct iommu_table *tbl, long index,
1693		long npages, unsigned long uaddr,
1694		enum dma_data_direction direction,
1695		struct dma_attrs *attrs)
1696{
1697	int ret = pnv_tce_build(tbl, index, npages, uaddr, direction,
1698			attrs);
1699
1700	if (!ret && (tbl->it_type & TCE_PCI_SWINV_CREATE))
1701		pnv_pci_ioda1_tce_invalidate(tbl, index, npages, false);
1702
1703	return ret;
1704}
1705
1706#ifdef CONFIG_IOMMU_API
1707static int pnv_ioda1_tce_xchg(struct iommu_table *tbl, long index,
1708		unsigned long *hpa, enum dma_data_direction *direction)
1709{
1710	long ret = pnv_tce_xchg(tbl, index, hpa, direction);
1711
1712	if (!ret && (tbl->it_type &
1713			(TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE)))
1714		pnv_pci_ioda1_tce_invalidate(tbl, index, 1, false);
1715
1716	return ret;
1717}
1718#endif
1719
1720static void pnv_ioda1_tce_free(struct iommu_table *tbl, long index,
1721		long npages)
1722{
1723	pnv_tce_free(tbl, index, npages);
1724
1725	if (tbl->it_type & TCE_PCI_SWINV_FREE)
1726		pnv_pci_ioda1_tce_invalidate(tbl, index, npages, false);
1727}
1728
1729static struct iommu_table_ops pnv_ioda1_iommu_ops = {
1730	.set = pnv_ioda1_tce_build,
1731#ifdef CONFIG_IOMMU_API
1732	.exchange = pnv_ioda1_tce_xchg,
1733#endif
1734	.clear = pnv_ioda1_tce_free,
1735	.get = pnv_tce_get,
1736};
1737
1738static inline void pnv_pci_ioda2_tce_invalidate_entire(struct pnv_ioda_pe *pe)
1739{
1740	/* 01xb - invalidate TCEs that match the specified PE# */
1741	unsigned long val = (0x4ull << 60) | (pe->pe_number & 0xFF);
1742	struct pnv_phb *phb = pe->phb;
1743
1744	if (!phb->ioda.tce_inval_reg)
1745		return;
1746
1747	mb(); /* Ensure above stores are visible */
1748	__raw_writeq(cpu_to_be64(val), phb->ioda.tce_inval_reg);
1749}
1750
1751static void pnv_pci_ioda2_do_tce_invalidate(unsigned pe_number, bool rm,
1752		__be64 __iomem *invalidate, unsigned shift,
1753		unsigned long index, unsigned long npages)
1754{
1755	unsigned long start, end, inc;
1756
1757	/* We'll invalidate DMA address in PE scope */
1758	start = 0x2ull << 60;
1759	start |= (pe_number & 0xFF);
1760	end = start;
1761
1762	/* Figure out the start, end and step */
1763	start |= (index << shift);
1764	end |= ((index + npages - 1) << shift);
1765	inc = (0x1ull << shift);
1766	mb();
1767
1768	while (start <= end) {
1769		if (rm)
1770			__raw_rm_writeq(cpu_to_be64(start), invalidate);
1771		else
1772			__raw_writeq(cpu_to_be64(start), invalidate);
1773		start += inc;
1774	}
1775}
1776
1777static void pnv_pci_ioda2_tce_invalidate(struct iommu_table *tbl,
1778		unsigned long index, unsigned long npages, bool rm)
1779{
1780	struct iommu_table_group_link *tgl;
1781
1782	list_for_each_entry_rcu(tgl, &tbl->it_group_list, next) {
1783		struct pnv_ioda_pe *pe = container_of(tgl->table_group,
1784				struct pnv_ioda_pe, table_group);
1785		__be64 __iomem *invalidate = rm ?
1786			(__be64 __iomem *)pe->phb->ioda.tce_inval_reg_phys :
1787			pe->phb->ioda.tce_inval_reg;
1788
1789		pnv_pci_ioda2_do_tce_invalidate(pe->pe_number, rm,
1790			invalidate, tbl->it_page_shift,
1791			index, npages);
1792	}
1793}
1794
1795static int pnv_ioda2_tce_build(struct iommu_table *tbl, long index,
1796		long npages, unsigned long uaddr,
1797		enum dma_data_direction direction,
1798		struct dma_attrs *attrs)
1799{
1800	int ret = pnv_tce_build(tbl, index, npages, uaddr, direction,
1801			attrs);
1802
1803	if (!ret && (tbl->it_type & TCE_PCI_SWINV_CREATE))
1804		pnv_pci_ioda2_tce_invalidate(tbl, index, npages, false);
1805
1806	return ret;
1807}
1808
1809#ifdef CONFIG_IOMMU_API
1810static int pnv_ioda2_tce_xchg(struct iommu_table *tbl, long index,
1811		unsigned long *hpa, enum dma_data_direction *direction)
1812{
1813	long ret = pnv_tce_xchg(tbl, index, hpa, direction);
1814
1815	if (!ret && (tbl->it_type &
1816			(TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE)))
1817		pnv_pci_ioda2_tce_invalidate(tbl, index, 1, false);
1818
1819	return ret;
1820}
1821#endif
1822
1823static void pnv_ioda2_tce_free(struct iommu_table *tbl, long index,
1824		long npages)
1825{
1826	pnv_tce_free(tbl, index, npages);
1827
1828	if (tbl->it_type & TCE_PCI_SWINV_FREE)
1829		pnv_pci_ioda2_tce_invalidate(tbl, index, npages, false);
1830}
1831
1832static void pnv_ioda2_table_free(struct iommu_table *tbl)
1833{
1834	pnv_pci_ioda2_table_free_pages(tbl);
1835	iommu_free_table(tbl, "pnv");
1836}
1837
1838static struct iommu_table_ops pnv_ioda2_iommu_ops = {
1839	.set = pnv_ioda2_tce_build,
1840#ifdef CONFIG_IOMMU_API
1841	.exchange = pnv_ioda2_tce_xchg,
1842#endif
1843	.clear = pnv_ioda2_tce_free,
1844	.get = pnv_tce_get,
1845	.free = pnv_ioda2_table_free,
1846};
1847
1848static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb,
1849				      struct pnv_ioda_pe *pe, unsigned int base,
1850				      unsigned int segs)
1851{
1852
1853	struct page *tce_mem = NULL;
1854	struct iommu_table *tbl;
1855	unsigned int i;
1856	int64_t rc;
1857	void *addr;
1858
1859	/* XXX FIXME: Handle 64-bit only DMA devices */
1860	/* XXX FIXME: Provide 64-bit DMA facilities & non-4K TCE tables etc.. */
1861	/* XXX FIXME: Allocate multi-level tables on PHB3 */
1862
1863	/* We shouldn't already have a 32-bit DMA associated */
1864	if (WARN_ON(pe->tce32_seg >= 0))
1865		return;
1866
1867	tbl = pnv_pci_table_alloc(phb->hose->node);
1868	iommu_register_group(&pe->table_group, phb->hose->global_number,
1869			pe->pe_number);
1870	pnv_pci_link_table_and_group(phb->hose->node, 0, tbl, &pe->table_group);
1871
1872	/* Grab a 32-bit TCE table */
1873	pe->tce32_seg = base;
1874	pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n",
1875		(base << 28), ((base + segs) << 28) - 1);
1876
1877	/* XXX Currently, we allocate one big contiguous table for the
1878	 * TCEs. We only really need one chunk per 256M of TCE space
1879	 * (ie per segment) but that's an optimization for later, it
1880	 * requires some added smarts with our get/put_tce implementation
1881	 */
1882	tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL,
1883				   get_order(TCE32_TABLE_SIZE * segs));
1884	if (!tce_mem) {
1885		pe_err(pe, " Failed to allocate a 32-bit TCE memory\n");
1886		goto fail;
1887	}
1888	addr = page_address(tce_mem);
1889	memset(addr, 0, TCE32_TABLE_SIZE * segs);
1890
1891	/* Configure HW */
1892	for (i = 0; i < segs; i++) {
1893		rc = opal_pci_map_pe_dma_window(phb->opal_id,
1894					      pe->pe_number,
1895					      base + i, 1,
1896					      __pa(addr) + TCE32_TABLE_SIZE * i,
1897					      TCE32_TABLE_SIZE, 0x1000);
1898		if (rc) {
1899			pe_err(pe, " Failed to configure 32-bit TCE table,"
1900			       " err %ld\n", rc);
1901			goto fail;
1902		}
1903	}
1904
1905	/* Setup linux iommu table */
1906	pnv_pci_setup_iommu_table(tbl, addr, TCE32_TABLE_SIZE * segs,
1907				  base << 28, IOMMU_PAGE_SHIFT_4K);
1908
1909	/* OPAL variant of P7IOC SW invalidated TCEs */
1910	if (phb->ioda.tce_inval_reg)
1911		tbl->it_type |= (TCE_PCI_SWINV_CREATE |
1912				 TCE_PCI_SWINV_FREE   |
1913				 TCE_PCI_SWINV_PAIR);
1914
1915	tbl->it_ops = &pnv_ioda1_iommu_ops;
1916	pe->table_group.tce32_start = tbl->it_offset << tbl->it_page_shift;
1917	pe->table_group.tce32_size = tbl->it_size << tbl->it_page_shift;
1918	iommu_init_table(tbl, phb->hose->node);
1919
1920	if (pe->flags & PNV_IODA_PE_DEV) {
1921		/*
1922		 * Setting table base here only for carrying iommu_group
1923		 * further down to let iommu_add_device() do the job.
1924		 * pnv_pci_ioda_dma_dev_setup will override it later anyway.
1925		 */
1926		set_iommu_table_base(&pe->pdev->dev, tbl);
1927		iommu_add_device(&pe->pdev->dev);
1928	} else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL))
1929		pnv_ioda_setup_bus_dma(pe, pe->pbus);
1930
1931	return;
1932 fail:
1933	/* XXX Failure: Try to fallback to 64-bit only ? */
1934	if (pe->tce32_seg >= 0)
1935		pe->tce32_seg = -1;
1936	if (tce_mem)
1937		__free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs));
1938	if (tbl) {
1939		pnv_pci_unlink_table_and_group(tbl, &pe->table_group);
1940		iommu_free_table(tbl, "pnv");
1941	}
1942}
1943
1944static long pnv_pci_ioda2_set_window(struct iommu_table_group *table_group,
1945		int num, struct iommu_table *tbl)
1946{
1947	struct pnv_ioda_pe *pe = container_of(table_group, struct pnv_ioda_pe,
1948			table_group);
1949	struct pnv_phb *phb = pe->phb;
1950	int64_t rc;
1951	const unsigned long size = tbl->it_indirect_levels ?
1952			tbl->it_level_size : tbl->it_size;
1953	const __u64 start_addr = tbl->it_offset << tbl->it_page_shift;
1954	const __u64 win_size = tbl->it_size << tbl->it_page_shift;
1955
1956	pe_info(pe, "Setting up window#%d %llx..%llx pg=%x\n", num,
1957			start_addr, start_addr + win_size - 1,
1958			IOMMU_PAGE_SIZE(tbl));
1959
1960	/*
1961	 * Map TCE table through TVT. The TVE index is the PE number
1962	 * shifted by 1 bit for 32-bits DMA space.
1963	 */
1964	rc = opal_pci_map_pe_dma_window(phb->opal_id,
1965			pe->pe_number,
1966			(pe->pe_number << 1) + num,
1967			tbl->it_indirect_levels + 1,
1968			__pa(tbl->it_base),
1969			size << 3,
1970			IOMMU_PAGE_SIZE(tbl));
1971	if (rc) {
1972		pe_err(pe, "Failed to configure TCE table, err %ld\n", rc);
1973		return rc;
1974	}
1975
1976	pnv_pci_link_table_and_group(phb->hose->node, num,
1977			tbl, &pe->table_group);
1978	pnv_pci_ioda2_tce_invalidate_entire(pe);
1979
1980	return 0;
1981}
1982
1983static void pnv_pci_ioda2_set_bypass(struct pnv_ioda_pe *pe, bool enable)
1984{
1985	uint16_t window_id = (pe->pe_number << 1 ) + 1;
1986	int64_t rc;
1987
1988	pe_info(pe, "%sabling 64-bit DMA bypass\n", enable ? "En" : "Dis");
1989	if (enable) {
1990		phys_addr_t top = memblock_end_of_DRAM();
1991
1992		top = roundup_pow_of_two(top);
1993		rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id,
1994						     pe->pe_number,
1995						     window_id,
1996						     pe->tce_bypass_base,
1997						     top);
1998	} else {
1999		rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id,
2000						     pe->pe_number,
2001						     window_id,
2002						     pe->tce_bypass_base,
2003						     0);
2004	}
2005	if (rc)
2006		pe_err(pe, "OPAL error %lld configuring bypass window\n", rc);
2007	else
2008		pe->tce_bypass_enabled = enable;
2009}
2010
2011static long pnv_pci_ioda2_table_alloc_pages(int nid, __u64 bus_offset,
2012		__u32 page_shift, __u64 window_size, __u32 levels,
2013		struct iommu_table *tbl);
2014
2015static long pnv_pci_ioda2_create_table(struct iommu_table_group *table_group,
2016		int num, __u32 page_shift, __u64 window_size, __u32 levels,
2017		struct iommu_table **ptbl)
2018{
2019	struct pnv_ioda_pe *pe = container_of(table_group, struct pnv_ioda_pe,
2020			table_group);
2021	int nid = pe->phb->hose->node;
2022	__u64 bus_offset = num ? pe->tce_bypass_base : table_group->tce32_start;
2023	long ret;
2024	struct iommu_table *tbl;
2025
2026	tbl = pnv_pci_table_alloc(nid);
2027	if (!tbl)
2028		return -ENOMEM;
2029
2030	ret = pnv_pci_ioda2_table_alloc_pages(nid,
2031			bus_offset, page_shift, window_size,
2032			levels, tbl);
2033	if (ret) {
2034		iommu_free_table(tbl, "pnv");
2035		return ret;
2036	}
2037
2038	tbl->it_ops = &pnv_ioda2_iommu_ops;
2039	if (pe->phb->ioda.tce_inval_reg)
2040		tbl->it_type |= (TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE);
2041
2042	*ptbl = tbl;
2043
2044	return 0;
2045}
2046
2047static long pnv_pci_ioda2_setup_default_config(struct pnv_ioda_pe *pe)
2048{
2049	struct iommu_table *tbl = NULL;
2050	long rc;
2051
2052	/*
2053	 * crashkernel= specifies the kdump kernel's maximum memory at
2054	 * some offset and there is no guaranteed the result is a power
2055	 * of 2, which will cause errors later.
2056	 */
2057	const u64 max_memory = __rounddown_pow_of_two(memory_hotplug_max());
2058
2059	/*
2060	 * In memory constrained environments, e.g. kdump kernel, the
2061	 * DMA window can be larger than available memory, which will
2062	 * cause errors later.
2063	 */
2064	const u64 window_size = min((u64)pe->table_group.tce32_size, max_memory);
2065
2066	rc = pnv_pci_ioda2_create_table(&pe->table_group, 0,
2067			IOMMU_PAGE_SHIFT_4K,
2068			window_size,
2069			POWERNV_IOMMU_DEFAULT_LEVELS, &tbl);
2070	if (rc) {
2071		pe_err(pe, "Failed to create 32-bit TCE table, err %ld",
2072				rc);
2073		return rc;
2074	}
2075
2076	iommu_init_table(tbl, pe->phb->hose->node);
2077
2078	rc = pnv_pci_ioda2_set_window(&pe->table_group, 0, tbl);
2079	if (rc) {
2080		pe_err(pe, "Failed to configure 32-bit TCE table, err %ld\n",
2081				rc);
2082		pnv_ioda2_table_free(tbl);
2083		return rc;
2084	}
2085
2086	if (!pnv_iommu_bypass_disabled)
2087		pnv_pci_ioda2_set_bypass(pe, true);
2088
2089	/* OPAL variant of PHB3 invalidated TCEs */
2090	if (pe->phb->ioda.tce_inval_reg)
2091		tbl->it_type |= (TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE);
2092
2093	/*
2094	 * Setting table base here only for carrying iommu_group
2095	 * further down to let iommu_add_device() do the job.
2096	 * pnv_pci_ioda_dma_dev_setup will override it later anyway.
2097	 */
2098	if (pe->flags & PNV_IODA_PE_DEV)
2099		set_iommu_table_base(&pe->pdev->dev, tbl);
2100
2101	return 0;
2102}
2103
2104#if defined(CONFIG_IOMMU_API) || defined(CONFIG_PCI_IOV)
2105static long pnv_pci_ioda2_unset_window(struct iommu_table_group *table_group,
2106		int num)
2107{
2108	struct pnv_ioda_pe *pe = container_of(table_group, struct pnv_ioda_pe,
2109			table_group);
2110	struct pnv_phb *phb = pe->phb;
2111	long ret;
2112
2113	pe_info(pe, "Removing DMA window #%d\n", num);
2114
2115	ret = opal_pci_map_pe_dma_window(phb->opal_id, pe->pe_number,
2116			(pe->pe_number << 1) + num,
2117			0/* levels */, 0/* table address */,
2118			0/* table size */, 0/* page size */);
2119	if (ret)
2120		pe_warn(pe, "Unmapping failed, ret = %ld\n", ret);
2121	else
2122		pnv_pci_ioda2_tce_invalidate_entire(pe);
2123
2124	pnv_pci_unlink_table_and_group(table_group->tables[num], table_group);
2125
2126	return ret;
2127}
2128#endif
2129
2130#ifdef CONFIG_IOMMU_API
2131static unsigned long pnv_pci_ioda2_get_table_size(__u32 page_shift,
2132		__u64 window_size, __u32 levels)
2133{
2134	unsigned long bytes = 0;
2135	const unsigned window_shift = ilog2(window_size);
2136	unsigned entries_shift = window_shift - page_shift;
2137	unsigned table_shift = entries_shift + 3;
2138	unsigned long tce_table_size = max(0x1000UL, 1UL << table_shift);
2139	unsigned long direct_table_size;
2140
2141	if (!levels || (levels > POWERNV_IOMMU_MAX_LEVELS) ||
2142			(window_size > memory_hotplug_max()) ||
2143			!is_power_of_2(window_size))
2144		return 0;
2145
2146	/* Calculate a direct table size from window_size and levels */
2147	entries_shift = (entries_shift + levels - 1) / levels;
2148	table_shift = entries_shift + 3;
2149	table_shift = max_t(unsigned, table_shift, PAGE_SHIFT);
2150	direct_table_size =  1UL << table_shift;
2151
2152	for ( ; levels; --levels) {
2153		bytes += _ALIGN_UP(tce_table_size, direct_table_size);
2154
2155		tce_table_size /= direct_table_size;
2156		tce_table_size <<= 3;
2157		tce_table_size = _ALIGN_UP(tce_table_size, direct_table_size);
2158	}
2159
2160	return bytes;
2161}
2162
2163static void pnv_ioda2_take_ownership(struct iommu_table_group *table_group)
2164{
2165	struct pnv_ioda_pe *pe = container_of(table_group, struct pnv_ioda_pe,
2166						table_group);
2167	/* Store @tbl as pnv_pci_ioda2_unset_window() resets it */
2168	struct iommu_table *tbl = pe->table_group.tables[0];
2169
2170	pnv_pci_ioda2_set_bypass(pe, false);
2171	pnv_pci_ioda2_unset_window(&pe->table_group, 0);
2172	pnv_ioda2_table_free(tbl);
2173}
2174
2175static void pnv_ioda2_release_ownership(struct iommu_table_group *table_group)
2176{
2177	struct pnv_ioda_pe *pe = container_of(table_group, struct pnv_ioda_pe,
2178						table_group);
2179
2180	pnv_pci_ioda2_setup_default_config(pe);
2181}
2182
2183static struct iommu_table_group_ops pnv_pci_ioda2_ops = {
2184	.get_table_size = pnv_pci_ioda2_get_table_size,
2185	.create_table = pnv_pci_ioda2_create_table,
2186	.set_window = pnv_pci_ioda2_set_window,
2187	.unset_window = pnv_pci_ioda2_unset_window,
2188	.take_ownership = pnv_ioda2_take_ownership,
2189	.release_ownership = pnv_ioda2_release_ownership,
2190};
2191#endif
2192
2193static void pnv_pci_ioda_setup_opal_tce_kill(struct pnv_phb *phb)
2194{
2195	const __be64 *swinvp;
2196
2197	/* OPAL variant of PHB3 invalidated TCEs */
2198	swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL);
2199	if (!swinvp)
2200		return;
2201
2202	phb->ioda.tce_inval_reg_phys = be64_to_cpup(swinvp);
2203	phb->ioda.tce_inval_reg = ioremap(phb->ioda.tce_inval_reg_phys, 8);
2204}
2205
2206static __be64 *pnv_pci_ioda2_table_do_alloc_pages(int nid, unsigned shift,
2207		unsigned levels, unsigned long limit,
2208		unsigned long *current_offset, unsigned long *total_allocated)
2209{
2210	struct page *tce_mem = NULL;
2211	__be64 *addr, *tmp;
2212	unsigned order = max_t(unsigned, shift, PAGE_SHIFT) - PAGE_SHIFT;
2213	unsigned long allocated = 1UL << (order + PAGE_SHIFT);
2214	unsigned entries = 1UL << (shift - 3);
2215	long i;
2216
2217	tce_mem = alloc_pages_node(nid, GFP_KERNEL, order);
2218	if (!tce_mem) {
2219		pr_err("Failed to allocate a TCE memory, order=%d\n", order);
2220		return NULL;
2221	}
2222	addr = page_address(tce_mem);
2223	memset(addr, 0, allocated);
2224	*total_allocated += allocated;
2225
2226	--levels;
2227	if (!levels) {
2228		*current_offset += allocated;
2229		return addr;
2230	}
2231
2232	for (i = 0; i < entries; ++i) {
2233		tmp = pnv_pci_ioda2_table_do_alloc_pages(nid, shift,
2234				levels, limit, current_offset, total_allocated);
2235		if (!tmp)
2236			break;
2237
2238		addr[i] = cpu_to_be64(__pa(tmp) |
2239				TCE_PCI_READ | TCE_PCI_WRITE);
2240
2241		if (*current_offset >= limit)
2242			break;
2243	}
2244
2245	return addr;
2246}
2247
2248static void pnv_pci_ioda2_table_do_free_pages(__be64 *addr,
2249		unsigned long size, unsigned level);
2250
2251static long pnv_pci_ioda2_table_alloc_pages(int nid, __u64 bus_offset,
2252		__u32 page_shift, __u64 window_size, __u32 levels,
2253		struct iommu_table *tbl)
2254{
2255	void *addr;
2256	unsigned long offset = 0, level_shift, total_allocated = 0;
2257	const unsigned window_shift = ilog2(window_size);
2258	unsigned entries_shift = window_shift - page_shift;
2259	unsigned table_shift = max_t(unsigned, entries_shift + 3, PAGE_SHIFT);
2260	const unsigned long tce_table_size = 1UL << table_shift;
2261
2262	if (!levels || (levels > POWERNV_IOMMU_MAX_LEVELS))
2263		return -EINVAL;
2264
2265	if ((window_size > memory_hotplug_max()) || !is_power_of_2(window_size))
2266		return -EINVAL;
2267
2268	/* Adjust direct table size from window_size and levels */
2269	entries_shift = (entries_shift + levels - 1) / levels;
2270	level_shift = entries_shift + 3;
2271	level_shift = max_t(unsigned, level_shift, PAGE_SHIFT);
2272
2273	/* Allocate TCE table */
2274	addr = pnv_pci_ioda2_table_do_alloc_pages(nid, level_shift,
2275			levels, tce_table_size, &offset, &total_allocated);
2276
2277	/* addr==NULL means that the first level allocation failed */
2278	if (!addr)
2279		return -ENOMEM;
2280
2281	/*
2282	 * First level was allocated but some lower level failed as
2283	 * we did not allocate as much as we wanted,
2284	 * release partially allocated table.
2285	 */
2286	if (offset < tce_table_size) {
2287		pnv_pci_ioda2_table_do_free_pages(addr,
2288				1ULL << (level_shift - 3), levels - 1);
2289		return -ENOMEM;
2290	}
2291
2292	/* Setup linux iommu table */
2293	pnv_pci_setup_iommu_table(tbl, addr, tce_table_size, bus_offset,
2294			page_shift);
2295	tbl->it_level_size = 1ULL << (level_shift - 3);
2296	tbl->it_indirect_levels = levels - 1;
2297	tbl->it_allocated_size = total_allocated;
2298
2299	pr_devel("Created TCE table: ws=%08llx ts=%lx @%08llx\n",
2300			window_size, tce_table_size, bus_offset);
2301
2302	return 0;
2303}
2304
2305static void pnv_pci_ioda2_table_do_free_pages(__be64 *addr,
2306		unsigned long size, unsigned level)
2307{
2308	const unsigned long addr_ul = (unsigned long) addr &
2309			~(TCE_PCI_READ | TCE_PCI_WRITE);
2310
2311	if (level) {
2312		long i;
2313		u64 *tmp = (u64 *) addr_ul;
2314
2315		for (i = 0; i < size; ++i) {
2316			unsigned long hpa = be64_to_cpu(tmp[i]);
2317
2318			if (!(hpa & (TCE_PCI_READ | TCE_PCI_WRITE)))
2319				continue;
2320
2321			pnv_pci_ioda2_table_do_free_pages(__va(hpa), size,
2322					level - 1);
2323		}
2324	}
2325
2326	free_pages(addr_ul, get_order(size << 3));
2327}
2328
2329static void pnv_pci_ioda2_table_free_pages(struct iommu_table *tbl)
2330{
2331	const unsigned long size = tbl->it_indirect_levels ?
2332			tbl->it_level_size : tbl->it_size;
2333
2334	if (!tbl->it_size)
2335		return;
2336
2337	pnv_pci_ioda2_table_do_free_pages((__be64 *)tbl->it_base, size,
2338			tbl->it_indirect_levels);
2339}
2340
2341static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb,
2342				       struct pnv_ioda_pe *pe)
2343{
2344	int64_t rc;
2345
2346	/* We shouldn't already have a 32-bit DMA associated */
2347	if (WARN_ON(pe->tce32_seg >= 0))
2348		return;
2349
2350	/* TVE #1 is selected by PCI address bit 59 */
2351	pe->tce_bypass_base = 1ull << 59;
2352
2353	iommu_register_group(&pe->table_group, phb->hose->global_number,
2354			pe->pe_number);
2355
2356	/* The PE will reserve all possible 32-bits space */
2357	pe->tce32_seg = 0;
2358	pe_info(pe, "Setting up 32-bit TCE table at 0..%08x\n",
2359		phb->ioda.m32_pci_base);
2360
2361	/* Setup linux iommu table */
2362	pe->table_group.tce32_start = 0;
2363	pe->table_group.tce32_size = phb->ioda.m32_pci_base;
2364	pe->table_group.max_dynamic_windows_supported =
2365			IOMMU_TABLE_GROUP_MAX_TABLES;
2366	pe->table_group.max_levels = POWERNV_IOMMU_MAX_LEVELS;
2367	pe->table_group.pgsizes = SZ_4K | SZ_64K | SZ_16M;
2368#ifdef CONFIG_IOMMU_API
2369	pe->table_group.ops = &pnv_pci_ioda2_ops;
2370#endif
2371
2372	rc = pnv_pci_ioda2_setup_default_config(pe);
2373	if (rc) {
2374		if (pe->tce32_seg >= 0)
2375			pe->tce32_seg = -1;
2376		return;
2377	}
2378
2379	if (pe->flags & PNV_IODA_PE_DEV)
2380		iommu_add_device(&pe->pdev->dev);
2381	else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL))
2382		pnv_ioda_setup_bus_dma(pe, pe->pbus);
2383}
2384
2385static void pnv_ioda_setup_dma(struct pnv_phb *phb)
2386{
2387	struct pci_controller *hose = phb->hose;
2388	unsigned int residual, remaining, segs, tw, base;
2389	struct pnv_ioda_pe *pe;
2390
2391	/* If we have more PE# than segments available, hand out one
2392	 * per PE until we run out and let the rest fail. If not,
2393	 * then we assign at least one segment per PE, plus more based
2394	 * on the amount of devices under that PE
2395	 */
2396	if (phb->ioda.dma_pe_count > phb->ioda.tce32_count)
2397		residual = 0;
2398	else
2399		residual = phb->ioda.tce32_count -
2400			phb->ioda.dma_pe_count;
2401
2402	pr_info("PCI: Domain %04x has %ld available 32-bit DMA segments\n",
2403		hose->global_number, phb->ioda.tce32_count);
2404	pr_info("PCI: %d PE# for a total weight of %d\n",
2405		phb->ioda.dma_pe_count, phb->ioda.dma_weight);
2406
2407	pnv_pci_ioda_setup_opal_tce_kill(phb);
2408
2409	/* Walk our PE list and configure their DMA segments, hand them
2410	 * out one base segment plus any residual segments based on
2411	 * weight
2412	 */
2413	remaining = phb->ioda.tce32_count;
2414	tw = phb->ioda.dma_weight;
2415	base = 0;
2416	list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) {
2417		if (!pe->dma_weight)
2418			continue;
2419		if (!remaining) {
2420			pe_warn(pe, "No DMA32 resources available\n");
2421			continue;
2422		}
2423		segs = 1;
2424		if (residual) {
2425			segs += ((pe->dma_weight * residual)  + (tw / 2)) / tw;
2426			if (segs > remaining)
2427				segs = remaining;
2428		}
2429
2430		/*
2431		 * For IODA2 compliant PHB3, we needn't care about the weight.
2432		 * The all available 32-bits DMA space will be assigned to
2433		 * the specific PE.
2434		 */
2435		if (phb->type == PNV_PHB_IODA1) {
2436			pe_info(pe, "DMA weight %d, assigned %d DMA32 segments\n",
2437				pe->dma_weight, segs);
2438			pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs);
2439		} else {
2440			pe_info(pe, "Assign DMA32 space\n");
2441			segs = 0;
2442			pnv_pci_ioda2_setup_dma_pe(phb, pe);
2443		}
2444
2445		remaining -= segs;
2446		base += segs;
2447	}
2448}
2449
2450#ifdef CONFIG_PCI_MSI
2451static void pnv_ioda2_msi_eoi(struct irq_data *d)
2452{
2453	unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
2454	struct irq_chip *chip = irq_data_get_irq_chip(d);
2455	struct pnv_phb *phb = container_of(chip, struct pnv_phb,
2456					   ioda.irq_chip);
2457	int64_t rc;
2458
2459	rc = opal_pci_msi_eoi(phb->opal_id, hw_irq);
2460	WARN_ON_ONCE(rc);
2461
2462	icp_native_eoi(d);
2463}
2464
2465
2466static void set_msi_irq_chip(struct pnv_phb *phb, unsigned int virq)
2467{
2468	struct irq_data *idata;
2469	struct irq_chip *ichip;
2470
2471	if (phb->type != PNV_PHB_IODA2)
2472		return;
2473
2474	if (!phb->ioda.irq_chip_init) {
2475		/*
2476		 * First time we setup an MSI IRQ, we need to setup the
2477		 * corresponding IRQ chip to route correctly.
2478		 */
2479		idata = irq_get_irq_data(virq);
2480		ichip = irq_data_get_irq_chip(idata);
2481		phb->ioda.irq_chip_init = 1;
2482		phb->ioda.irq_chip = *ichip;
2483		phb->ioda.irq_chip.irq_eoi = pnv_ioda2_msi_eoi;
2484	}
2485	irq_set_chip(virq, &phb->ioda.irq_chip);
2486}
2487
2488#ifdef CONFIG_CXL_BASE
2489
2490struct device_node *pnv_pci_get_phb_node(struct pci_dev *dev)
2491{
2492	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2493
2494	return of_node_get(hose->dn);
2495}
2496EXPORT_SYMBOL(pnv_pci_get_phb_node);
2497
2498int pnv_phb_to_cxl_mode(struct pci_dev *dev, uint64_t mode)
2499{
2500	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2501	struct pnv_phb *phb = hose->private_data;
2502	struct pnv_ioda_pe *pe;
2503	int rc;
2504
2505	pe = pnv_ioda_get_pe(dev);
2506	if (!pe)
2507		return -ENODEV;
2508
2509	pe_info(pe, "Switching PHB to CXL\n");
2510
2511	rc = opal_pci_set_phb_cxl_mode(phb->opal_id, mode, pe->pe_number);
2512	if (rc)
2513		dev_err(&dev->dev, "opal_pci_set_phb_cxl_mode failed: %i\n", rc);
2514
2515	return rc;
2516}
2517EXPORT_SYMBOL(pnv_phb_to_cxl_mode);
2518
2519/* Find PHB for cxl dev and allocate MSI hwirqs?
2520 * Returns the absolute hardware IRQ number
2521 */
2522int pnv_cxl_alloc_hwirqs(struct pci_dev *dev, int num)
2523{
2524	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2525	struct pnv_phb *phb = hose->private_data;
2526	int hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, num);
2527
2528	if (hwirq < 0) {
2529		dev_warn(&dev->dev, "Failed to find a free MSI\n");
2530		return -ENOSPC;
2531	}
2532
2533	return phb->msi_base + hwirq;
2534}
2535EXPORT_SYMBOL(pnv_cxl_alloc_hwirqs);
2536
2537void pnv_cxl_release_hwirqs(struct pci_dev *dev, int hwirq, int num)
2538{
2539	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2540	struct pnv_phb *phb = hose->private_data;
2541
2542	msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, num);
2543}
2544EXPORT_SYMBOL(pnv_cxl_release_hwirqs);
2545
2546void pnv_cxl_release_hwirq_ranges(struct cxl_irq_ranges *irqs,
2547				  struct pci_dev *dev)
2548{
2549	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2550	struct pnv_phb *phb = hose->private_data;
2551	int i, hwirq;
2552
2553	for (i = 1; i < CXL_IRQ_RANGES; i++) {
2554		if (!irqs->range[i])
2555			continue;
2556		pr_devel("cxl release irq range 0x%x: offset: 0x%lx  limit: %ld\n",
2557			 i, irqs->offset[i],
2558			 irqs->range[i]);
2559		hwirq = irqs->offset[i] - phb->msi_base;
2560		msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq,
2561				       irqs->range[i]);
2562	}
2563}
2564EXPORT_SYMBOL(pnv_cxl_release_hwirq_ranges);
2565
2566int pnv_cxl_alloc_hwirq_ranges(struct cxl_irq_ranges *irqs,
2567			       struct pci_dev *dev, int num)
2568{
2569	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2570	struct pnv_phb *phb = hose->private_data;
2571	int i, hwirq, try;
2572
2573	memset(irqs, 0, sizeof(struct cxl_irq_ranges));
2574
2575	/* 0 is reserved for the multiplexed PSL DSI interrupt */
2576	for (i = 1; i < CXL_IRQ_RANGES && num; i++) {
2577		try = num;
2578		while (try) {
2579			hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, try);
2580			if (hwirq >= 0)
2581				break;
2582			try /= 2;
2583		}
2584		if (!try)
2585			goto fail;
2586
2587		irqs->offset[i] = phb->msi_base + hwirq;
2588		irqs->range[i] = try;
2589		pr_devel("cxl alloc irq range 0x%x: offset: 0x%lx  limit: %li\n",
2590			 i, irqs->offset[i], irqs->range[i]);
2591		num -= try;
2592	}
2593	if (num)
2594		goto fail;
2595
2596	return 0;
2597fail:
2598	pnv_cxl_release_hwirq_ranges(irqs, dev);
2599	return -ENOSPC;
2600}
2601EXPORT_SYMBOL(pnv_cxl_alloc_hwirq_ranges);
2602
2603int pnv_cxl_get_irq_count(struct pci_dev *dev)
2604{
2605	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2606	struct pnv_phb *phb = hose->private_data;
2607
2608	return phb->msi_bmp.irq_count;
2609}
2610EXPORT_SYMBOL(pnv_cxl_get_irq_count);
2611
2612int pnv_cxl_ioda_msi_setup(struct pci_dev *dev, unsigned int hwirq,
2613			   unsigned int virq)
2614{
2615	struct pci_controller *hose = pci_bus_to_host(dev->bus);
2616	struct pnv_phb *phb = hose->private_data;
2617	unsigned int xive_num = hwirq - phb->msi_base;
2618	struct pnv_ioda_pe *pe;
2619	int rc;
2620
2621	if (!(pe = pnv_ioda_get_pe(dev)))
2622		return -ENODEV;
2623
2624	/* Assign XIVE to PE */
2625	rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num);
2626	if (rc) {
2627		pe_warn(pe, "%s: OPAL error %d setting msi_base 0x%x "
2628			"hwirq 0x%x XIVE 0x%x PE\n",
2629			pci_name(dev), rc, phb->msi_base, hwirq, xive_num);
2630		return -EIO;
2631	}
2632	set_msi_irq_chip(phb, virq);
2633
2634	return 0;
2635}
2636EXPORT_SYMBOL(pnv_cxl_ioda_msi_setup);
2637#endif
2638
2639static int pnv_pci_ioda_msi_setup(struct pnv_phb *phb, struct pci_dev *dev,
2640				  unsigned int hwirq, unsigned int virq,
2641				  unsigned int is_64, struct msi_msg *msg)
2642{
2643	struct pnv_ioda_pe *pe = pnv_ioda_get_pe(dev);
2644	unsigned int xive_num = hwirq - phb->msi_base;
2645	__be32 data;
2646	int rc;
2647
2648	/* No PE assigned ? bail out ... no MSI for you ! */
2649	if (pe == NULL)
2650		return -ENXIO;
2651
2652	/* Check if we have an MVE */
2653	if (pe->mve_number < 0)
2654		return -ENXIO;
2655
2656	/* Force 32-bit MSI on some broken devices */
2657	if (dev->no_64bit_msi)
2658		is_64 = 0;
2659
2660	/* Assign XIVE to PE */
2661	rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num);
2662	if (rc) {
2663		pr_warn("%s: OPAL error %d setting XIVE %d PE\n",
2664			pci_name(dev), rc, xive_num);
2665		return -EIO;
2666	}
2667
2668	if (is_64) {
2669		__be64 addr64;
2670
2671		rc = opal_get_msi_64(phb->opal_id, pe->mve_number, xive_num, 1,
2672				     &addr64, &data);
2673		if (rc) {
2674			pr_warn("%s: OPAL error %d getting 64-bit MSI data\n",
2675				pci_name(dev), rc);
2676			return -EIO;
2677		}
2678		msg->address_hi = be64_to_cpu(addr64) >> 32;
2679		msg->address_lo = be64_to_cpu(addr64) & 0xfffffffful;
2680	} else {
2681		__be32 addr32;
2682
2683		rc = opal_get_msi_32(phb->opal_id, pe->mve_number, xive_num, 1,
2684				     &addr32, &data);
2685		if (rc) {
2686			pr_warn("%s: OPAL error %d getting 32-bit MSI data\n",
2687				pci_name(dev), rc);
2688			return -EIO;
2689		}
2690		msg->address_hi = 0;
2691		msg->address_lo = be32_to_cpu(addr32);
2692	}
2693	msg->data = be32_to_cpu(data);
2694
2695	set_msi_irq_chip(phb, virq);
2696
2697	pr_devel("%s: %s-bit MSI on hwirq %x (xive #%d),"
2698		 " address=%x_%08x data=%x PE# %d\n",
2699		 pci_name(dev), is_64 ? "64" : "32", hwirq, xive_num,
2700		 msg->address_hi, msg->address_lo, data, pe->pe_number);
2701
2702	return 0;
2703}
2704
2705static void pnv_pci_init_ioda_msis(struct pnv_phb *phb)
2706{
2707	unsigned int count;
2708	const __be32 *prop = of_get_property(phb->hose->dn,
2709					     "ibm,opal-msi-ranges", NULL);
2710	if (!prop) {
2711		/* BML Fallback */
2712		prop = of_get_property(phb->hose->dn, "msi-ranges", NULL);
2713	}
2714	if (!prop)
2715		return;
2716
2717	phb->msi_base = be32_to_cpup(prop);
2718	count = be32_to_cpup(prop + 1);
2719	if (msi_bitmap_alloc(&phb->msi_bmp, count, phb->hose->dn)) {
2720		pr_err("PCI %d: Failed to allocate MSI bitmap !\n",
2721		       phb->hose->global_number);
2722		return;
2723	}
2724
2725	phb->msi_setup = pnv_pci_ioda_msi_setup;
2726	phb->msi32_support = 1;
2727	pr_info("  Allocated bitmap for %d MSIs (base IRQ 0x%x)\n",
2728		count, phb->msi_base);
2729}
2730#else
2731static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) { }
2732#endif /* CONFIG_PCI_MSI */
2733
2734#ifdef CONFIG_PCI_IOV
2735static void pnv_pci_ioda_fixup_iov_resources(struct pci_dev *pdev)
2736{
2737	struct pci_controller *hose;
2738	struct pnv_phb *phb;
2739	struct resource *res;
2740	int i;
2741	resource_size_t size;
2742	struct pci_dn *pdn;
2743	int mul, total_vfs;
2744
2745	if (!pdev->is_physfn || pdev->is_added)
2746		return;
2747
2748	hose = pci_bus_to_host(pdev->bus);
2749	phb = hose->private_data;
2750
2751	pdn = pci_get_pdn(pdev);
2752	pdn->vfs_expanded = 0;
2753
2754	total_vfs = pci_sriov_get_totalvfs(pdev);
2755	pdn->m64_per_iov = 1;
2756	mul = phb->ioda.total_pe;
2757
2758	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
2759		res = &pdev->resource[i + PCI_IOV_RESOURCES];
2760		if (!res->flags || res->parent)
2761			continue;
2762		if (!pnv_pci_is_mem_pref_64(res->flags)) {
2763			dev_warn(&pdev->dev, " non M64 VF BAR%d: %pR\n",
2764				 i, res);
2765			continue;
2766		}
2767
2768		size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES);
2769
2770		/* bigger than 64M */
2771		if (size > (1 << 26)) {
2772			dev_info(&pdev->dev, "PowerNV: VF BAR%d: %pR IOV size is bigger than 64M, roundup power2\n",
2773				 i, res);
2774			pdn->m64_per_iov = M64_PER_IOV;
2775			mul = roundup_pow_of_two(total_vfs);
2776			break;
2777		}
2778	}
2779
2780	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
2781		res = &pdev->resource[i + PCI_IOV_RESOURCES];
2782		if (!res->flags || res->parent)
2783			continue;
2784		if (!pnv_pci_is_mem_pref_64(res->flags)) {
2785			dev_warn(&pdev->dev, "Skipping expanding VF BAR%d: %pR\n",
2786				 i, res);
2787			continue;
2788		}
2789
2790		dev_dbg(&pdev->dev, " Fixing VF BAR%d: %pR to\n", i, res);
2791		size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES);
2792		res->end = res->start + size * mul - 1;
2793		dev_dbg(&pdev->dev, "                       %pR\n", res);
2794		dev_info(&pdev->dev, "VF BAR%d: %pR (expanded to %d VFs for PE alignment)",
2795			 i, res, mul);
2796	}
2797	pdn->vfs_expanded = mul;
2798}
2799#endif /* CONFIG_PCI_IOV */
2800
2801/*
2802 * This function is supposed to be called on basis of PE from top
2803 * to bottom style. So the the I/O or MMIO segment assigned to
2804 * parent PE could be overrided by its child PEs if necessary.
2805 */
2806static void pnv_ioda_setup_pe_seg(struct pci_controller *hose,
2807				  struct pnv_ioda_pe *pe)
2808{
2809	struct pnv_phb *phb = hose->private_data;
2810	struct pci_bus_region region;
2811	struct resource *res;
2812	int i, index;
2813	int rc;
2814
2815	/*
2816	 * NOTE: We only care PCI bus based PE for now. For PCI
2817	 * device based PE, for example SRIOV sensitive VF should
2818	 * be figured out later.
2819	 */
2820	BUG_ON(!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)));
2821
2822	pci_bus_for_each_resource(pe->pbus, res, i) {
2823		if (!res || !res->flags ||
2824		    res->start > res->end)
2825			continue;
2826
2827		if (res->flags & IORESOURCE_IO) {
2828			region.start = res->start - phb->ioda.io_pci_base;
2829			region.end   = res->end - phb->ioda.io_pci_base;
2830			index = region.start / phb->ioda.io_segsize;
2831
2832			while (index < phb->ioda.total_pe &&
2833			       region.start <= region.end) {
2834				phb->ioda.io_segmap[index] = pe->pe_number;
2835				rc = opal_pci_map_pe_mmio_window(phb->opal_id,
2836					pe->pe_number, OPAL_IO_WINDOW_TYPE, 0, index);
2837				if (rc != OPAL_SUCCESS) {
2838					pr_err("%s: OPAL error %d when mapping IO "
2839					       "segment #%d to PE#%d\n",
2840					       __func__, rc, index, pe->pe_number);
2841					break;
2842				}
2843
2844				region.start += phb->ioda.io_segsize;
2845				index++;
2846			}
2847		} else if ((res->flags & IORESOURCE_MEM) &&
2848			   !pnv_pci_is_mem_pref_64(res->flags)) {
2849			region.start = res->start -
2850				       hose->mem_offset[0] -
2851				       phb->ioda.m32_pci_base;
2852			region.end   = res->end -
2853				       hose->mem_offset[0] -
2854				       phb->ioda.m32_pci_base;
2855			index = region.start / phb->ioda.m32_segsize;
2856
2857			while (index < phb->ioda.total_pe &&
2858			       region.start <= region.end) {
2859				phb->ioda.m32_segmap[index] = pe->pe_number;
2860				rc = opal_pci_map_pe_mmio_window(phb->opal_id,
2861					pe->pe_number, OPAL_M32_WINDOW_TYPE, 0, index);
2862				if (rc != OPAL_SUCCESS) {
2863					pr_err("%s: OPAL error %d when mapping M32 "
2864					       "segment#%d to PE#%d",
2865					       __func__, rc, index, pe->pe_number);
2866					break;
2867				}
2868
2869				region.start += phb->ioda.m32_segsize;
2870				index++;
2871			}
2872		}
2873	}
2874}
2875
2876static void pnv_pci_ioda_setup_seg(void)
2877{
2878	struct pci_controller *tmp, *hose;
2879	struct pnv_phb *phb;
2880	struct pnv_ioda_pe *pe;
2881
2882	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
2883		phb = hose->private_data;
2884		list_for_each_entry(pe, &phb->ioda.pe_list, list) {
2885			pnv_ioda_setup_pe_seg(hose, pe);
2886		}
2887	}
2888}
2889
2890static void pnv_pci_ioda_setup_DMA(void)
2891{
2892	struct pci_controller *hose, *tmp;
2893	struct pnv_phb *phb;
2894
2895	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
2896		pnv_ioda_setup_dma(hose->private_data);
2897
2898		/* Mark the PHB initialization done */
2899		phb = hose->private_data;
2900		phb->initialized = 1;
2901	}
2902}
2903
2904static void pnv_pci_ioda_create_dbgfs(void)
2905{
2906#ifdef CONFIG_DEBUG_FS
2907	struct pci_controller *hose, *tmp;
2908	struct pnv_phb *phb;
2909	char name[16];
2910
2911	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
2912		phb = hose->private_data;
2913
2914		sprintf(name, "PCI%04x", hose->global_number);
2915		phb->dbgfs = debugfs_create_dir(name, powerpc_debugfs_root);
2916		if (!phb->dbgfs)
2917			pr_warning("%s: Error on creating debugfs on PHB#%x\n",
2918				__func__, hose->global_number);
2919	}
2920#endif /* CONFIG_DEBUG_FS */
2921}
2922
2923static void pnv_pci_ioda_fixup(void)
2924{
2925	pnv_pci_ioda_setup_PEs();
2926	pnv_pci_ioda_setup_seg();
2927	pnv_pci_ioda_setup_DMA();
2928
2929	pnv_pci_ioda_create_dbgfs();
2930
2931#ifdef CONFIG_EEH
2932	eeh_init();
2933	eeh_addr_cache_build();
2934#endif
2935}
2936
2937/*
2938 * Returns the alignment for I/O or memory windows for P2P
2939 * bridges. That actually depends on how PEs are segmented.
2940 * For now, we return I/O or M32 segment size for PE sensitive
2941 * P2P bridges. Otherwise, the default values (4KiB for I/O,
2942 * 1MiB for memory) will be returned.
2943 *
2944 * The current PCI bus might be put into one PE, which was
2945 * create against the parent PCI bridge. For that case, we
2946 * needn't enlarge the alignment so that we can save some
2947 * resources.
2948 */
2949static resource_size_t pnv_pci_window_alignment(struct pci_bus *bus,
2950						unsigned long type)
2951{
2952	struct pci_dev *bridge;
2953	struct pci_controller *hose = pci_bus_to_host(bus);
2954	struct pnv_phb *phb = hose->private_data;
2955	int num_pci_bridges = 0;
2956
2957	bridge = bus->self;
2958	while (bridge) {
2959		if (pci_pcie_type(bridge) == PCI_EXP_TYPE_PCI_BRIDGE) {
2960			num_pci_bridges++;
2961			if (num_pci_bridges >= 2)
2962				return 1;
2963		}
2964
2965		bridge = bridge->bus->self;
2966	}
2967
2968	/* We fail back to M32 if M64 isn't supported */
2969	if (phb->ioda.m64_segsize &&
2970	    pnv_pci_is_mem_pref_64(type))
2971		return phb->ioda.m64_segsize;
2972	if (type & IORESOURCE_MEM)
2973		return phb->ioda.m32_segsize;
2974
2975	return phb->ioda.io_segsize;
2976}
2977
2978#ifdef CONFIG_PCI_IOV
2979static resource_size_t pnv_pci_iov_resource_alignment(struct pci_dev *pdev,
2980						      int resno)
2981{
2982	struct pci_dn *pdn = pci_get_pdn(pdev);
2983	resource_size_t align, iov_align;
2984
2985	iov_align = resource_size(&pdev->resource[resno]);
2986	if (iov_align)
2987		return iov_align;
2988
2989	align = pci_iov_resource_size(pdev, resno);
2990	if (pdn->vfs_expanded)
2991		return pdn->vfs_expanded * align;
2992
2993	return align;
2994}
2995#endif /* CONFIG_PCI_IOV */
2996
2997/* Prevent enabling devices for which we couldn't properly
2998 * assign a PE
2999 */
3000static bool pnv_pci_enable_device_hook(struct pci_dev *dev)
3001{
3002	struct pci_controller *hose = pci_bus_to_host(dev->bus);
3003	struct pnv_phb *phb = hose->private_data;
3004	struct pci_dn *pdn;
3005
3006	/* The function is probably called while the PEs have
3007	 * not be created yet. For example, resource reassignment
3008	 * during PCI probe period. We just skip the check if
3009	 * PEs isn't ready.
3010	 */
3011	if (!phb->initialized)
3012		return true;
3013
3014	pdn = pci_get_pdn(dev);
3015	if (!pdn || pdn->pe_number == IODA_INVALID_PE)
3016		return false;
3017
3018	return true;
3019}
3020
3021static u32 pnv_ioda_bdfn_to_pe(struct pnv_phb *phb, struct pci_bus *bus,
3022			       u32 devfn)
3023{
3024	return phb->ioda.pe_rmap[(bus->number << 8) | devfn];
3025}
3026
3027static void pnv_pci_ioda_shutdown(struct pci_controller *hose)
3028{
3029	struct pnv_phb *phb = hose->private_data;
3030
3031	opal_pci_reset(phb->opal_id, OPAL_RESET_PCI_IODA_TABLE,
3032		       OPAL_ASSERT_RESET);
3033}
3034
3035static const struct pci_controller_ops pnv_pci_ioda_controller_ops = {
3036       .dma_dev_setup = pnv_pci_dma_dev_setup,
3037       .dma_bus_setup = pnv_pci_dma_bus_setup,
3038#ifdef CONFIG_PCI_MSI
3039       .setup_msi_irqs = pnv_setup_msi_irqs,
3040       .teardown_msi_irqs = pnv_teardown_msi_irqs,
3041#endif
3042       .enable_device_hook = pnv_pci_enable_device_hook,
3043       .window_alignment = pnv_pci_window_alignment,
3044       .reset_secondary_bus = pnv_pci_reset_secondary_bus,
3045       .dma_set_mask = pnv_pci_ioda_dma_set_mask,
3046       .dma_get_required_mask = pnv_pci_ioda_dma_get_required_mask,
3047       .shutdown = pnv_pci_ioda_shutdown,
3048};
3049
3050static void __init pnv_pci_init_ioda_phb(struct device_node *np,
3051					 u64 hub_id, int ioda_type)
3052{
3053	struct pci_controller *hose;
3054	struct pnv_phb *phb;
3055	unsigned long size, m32map_off, pemap_off, iomap_off = 0;
3056	const __be64 *prop64;
3057	const __be32 *prop32;
3058	int len;
3059	u64 phb_id;
3060	void *aux;
3061	long rc;
3062
3063	pr_info("Initializing IODA%d OPAL PHB %s\n", ioda_type, np->full_name);
3064
3065	prop64 = of_get_property(np, "ibm,opal-phbid", NULL);
3066	if (!prop64) {
3067		pr_err("  Missing \"ibm,opal-phbid\" property !\n");
3068		return;
3069	}
3070	phb_id = be64_to_cpup(prop64);
3071	pr_debug("  PHB-ID  : 0x%016llx\n", phb_id);
3072
3073	phb = memblock_virt_alloc(sizeof(struct pnv_phb), 0);
3074
3075	/* Allocate PCI controller */
3076	phb->hose = hose = pcibios_alloc_controller(np);
3077	if (!phb->hose) {
3078		pr_err("  Can't allocate PCI controller for %s\n",
3079		       np->full_name);
3080		memblock_free(__pa(phb), sizeof(struct pnv_phb));
3081		return;
3082	}
3083
3084	spin_lock_init(&phb->lock);
3085	prop32 = of_get_property(np, "bus-range", &len);
3086	if (prop32 && len == 8) {
3087		hose->first_busno = be32_to_cpu(prop32[0]);
3088		hose->last_busno = be32_to_cpu(prop32[1]);
3089	} else {
3090		pr_warn("  Broken <bus-range> on %s\n", np->full_name);
3091		hose->first_busno = 0;
3092		hose->last_busno = 0xff;
3093	}
3094	hose->private_data = phb;
3095	phb->hub_id = hub_id;
3096	phb->opal_id = phb_id;
3097	phb->type = ioda_type;
3098	mutex_init(&phb->ioda.pe_alloc_mutex);
3099
3100	/* Detect specific models for error handling */
3101	if (of_device_is_compatible(np, "ibm,p7ioc-pciex"))
3102		phb->model = PNV_PHB_MODEL_P7IOC;
3103	else if (of_device_is_compatible(np, "ibm,power8-pciex"))
3104		phb->model = PNV_PHB_MODEL_PHB3;
3105	else
3106		phb->model = PNV_PHB_MODEL_UNKNOWN;
3107
3108	/* Parse 32-bit and IO ranges (if any) */
3109	pci_process_bridge_OF_ranges(hose, np, !hose->global_number);
3110
3111	/* Get registers */
3112	phb->regs = of_iomap(np, 0);
3113	if (phb->regs == NULL)
3114		pr_err("  Failed to map registers !\n");
3115
3116	/* Initialize more IODA stuff */
3117	phb->ioda.total_pe = 1;
3118	prop32 = of_get_property(np, "ibm,opal-num-pes", NULL);
3119	if (prop32)
3120		phb->ioda.total_pe = be32_to_cpup(prop32);
3121	prop32 = of_get_property(np, "ibm,opal-reserved-pe", NULL);
3122	if (prop32)
3123		phb->ioda.reserved_pe = be32_to_cpup(prop32);
3124
3125	/* Parse 64-bit MMIO range */
3126	pnv_ioda_parse_m64_window(phb);
3127
3128	phb->ioda.m32_size = resource_size(&hose->mem_resources[0]);
3129	/* FW Has already off top 64k of M32 space (MSI space) */
3130	phb->ioda.m32_size += 0x10000;
3131
3132	phb->ioda.m32_segsize = phb->ioda.m32_size / phb->ioda.total_pe;
3133	phb->ioda.m32_pci_base = hose->mem_resources[0].start - hose->mem_offset[0];
3134	phb->ioda.io_size = hose->pci_io_size;
3135	phb->ioda.io_segsize = phb->ioda.io_size / phb->ioda.total_pe;
3136	phb->ioda.io_pci_base = 0; /* XXX calculate this ? */
3137
3138	/* Allocate aux data & arrays. We don't have IO ports on PHB3 */
3139	size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long));
3140	m32map_off = size;
3141	size += phb->ioda.total_pe * sizeof(phb->ioda.m32_segmap[0]);
3142	if (phb->type == PNV_PHB_IODA1) {
3143		iomap_off = size;
3144		size += phb->ioda.total_pe * sizeof(phb->ioda.io_segmap[0]);
3145	}
3146	pemap_off = size;
3147	size += phb->ioda.total_pe * sizeof(struct pnv_ioda_pe);
3148	aux = memblock_virt_alloc(size, 0);
3149	phb->ioda.pe_alloc = aux;
3150	phb->ioda.m32_segmap = aux + m32map_off;
3151	if (phb->type == PNV_PHB_IODA1)
3152		phb->ioda.io_segmap = aux + iomap_off;
3153	phb->ioda.pe_array = aux + pemap_off;
3154	set_bit(phb->ioda.reserved_pe, phb->ioda.pe_alloc);
3155
3156	INIT_LIST_HEAD(&phb->ioda.pe_dma_list);
3157	INIT_LIST_HEAD(&phb->ioda.pe_list);
3158	mutex_init(&phb->ioda.pe_list_mutex);
3159
3160	/* Calculate how many 32-bit TCE segments we have */
3161	phb->ioda.tce32_count = phb->ioda.m32_pci_base >> 28;
3162
3163#if 0 /* We should really do that ... */
3164	rc = opal_pci_set_phb_mem_window(opal->phb_id,
3165					 window_type,
3166					 window_num,
3167					 starting_real_address,
3168					 starting_pci_address,
3169					 segment_size);
3170#endif
3171
3172	pr_info("  %03d (%03d) PE's M32: 0x%x [segment=0x%x]\n",
3173		phb->ioda.total_pe, phb->ioda.reserved_pe,
3174		phb->ioda.m32_size, phb->ioda.m32_segsize);
3175	if (phb->ioda.m64_size)
3176		pr_info("                 M64: 0x%lx [segment=0x%lx]\n",
3177			phb->ioda.m64_size, phb->ioda.m64_segsize);
3178	if (phb->ioda.io_size)
3179		pr_info("                  IO: 0x%x [segment=0x%x]\n",
3180			phb->ioda.io_size, phb->ioda.io_segsize);
3181
3182
3183	phb->hose->ops = &pnv_pci_ops;
3184	phb->get_pe_state = pnv_ioda_get_pe_state;
3185	phb->freeze_pe = pnv_ioda_freeze_pe;
3186	phb->unfreeze_pe = pnv_ioda_unfreeze_pe;
3187
3188	/* Setup RID -> PE mapping function */
3189	phb->bdfn_to_pe = pnv_ioda_bdfn_to_pe;
3190
3191	/* Setup TCEs */
3192	phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
3193
3194	/* Setup MSI support */
3195	pnv_pci_init_ioda_msis(phb);
3196
3197	/*
3198	 * We pass the PCI probe flag PCI_REASSIGN_ALL_RSRC here
3199	 * to let the PCI core do resource assignment. It's supposed
3200	 * that the PCI core will do correct I/O and MMIO alignment
3201	 * for the P2P bridge bars so that each PCI bus (excluding
3202	 * the child P2P bridges) can form individual PE.
3203	 */
3204	ppc_md.pcibios_fixup = pnv_pci_ioda_fixup;
3205	hose->controller_ops = pnv_pci_ioda_controller_ops;
3206
3207#ifdef CONFIG_PCI_IOV
3208	ppc_md.pcibios_fixup_sriov = pnv_pci_ioda_fixup_iov_resources;
3209	ppc_md.pcibios_iov_resource_alignment = pnv_pci_iov_resource_alignment;
3210#endif
3211
3212	pci_add_flags(PCI_REASSIGN_ALL_RSRC);
3213
3214	/* Reset IODA tables to a clean state */
3215	rc = opal_pci_reset(phb_id, OPAL_RESET_PCI_IODA_TABLE, OPAL_ASSERT_RESET);
3216	if (rc)
3217		pr_warning("  OPAL Error %ld performing IODA table reset !\n", rc);
3218
3219	/* If we're running in kdump kerenl, the previous kerenl never
3220	 * shutdown PCI devices correctly. We already got IODA table
3221	 * cleaned out. So we have to issue PHB reset to stop all PCI
3222	 * transactions from previous kerenl.
3223	 */
3224	if (is_kdump_kernel()) {
3225		pr_info("  Issue PHB reset ...\n");
3226		pnv_eeh_phb_reset(hose, EEH_RESET_FUNDAMENTAL);
3227		pnv_eeh_phb_reset(hose, EEH_RESET_DEACTIVATE);
3228	}
3229
3230	/* Remove M64 resource if we can't configure it successfully */
3231	if (!phb->init_m64 || phb->init_m64(phb))
3232		hose->mem_resources[1].flags = 0;
3233}
3234
3235void __init pnv_pci_init_ioda2_phb(struct device_node *np)
3236{
3237	pnv_pci_init_ioda_phb(np, 0, PNV_PHB_IODA2);
3238}
3239
3240void __init pnv_pci_init_ioda_hub(struct device_node *np)
3241{
3242	struct device_node *phbn;
3243	const __be64 *prop64;
3244	u64 hub_id;
3245
3246	pr_info("Probing IODA IO-Hub %s\n", np->full_name);
3247
3248	prop64 = of_get_property(np, "ibm,opal-hubid", NULL);
3249	if (!prop64) {
3250		pr_err(" Missing \"ibm,opal-hubid\" property !\n");
3251		return;
3252	}
3253	hub_id = be64_to_cpup(prop64);
3254	pr_devel(" HUB-ID : 0x%016llx\n", hub_id);
3255
3256	/* Count child PHBs */
3257	for_each_child_of_node(np, phbn) {
3258		/* Look for IODA1 PHBs */
3259		if (of_device_is_compatible(phbn, "ibm,ioda-phb"))
3260			pnv_pci_init_ioda_phb(phbn, hub_id, PNV_PHB_IODA1);
3261	}
3262}
3263