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/delay.h>
17 #include <linux/string.h>
18 #include <linux/init.h>
19 #include <linux/bootmem.h>
20 #include <linux/irq.h>
21 #include <linux/io.h>
22 #include <linux/msi.h>
23 
24 #include <asm/sections.h>
25 #include <asm/io.h>
26 #include <asm/prom.h>
27 #include <asm/pci-bridge.h>
28 #include <asm/machdep.h>
29 #include <asm/ppc-pci.h>
30 #include <asm/opal.h>
31 #include <asm/iommu.h>
32 #include <asm/tce.h>
33 
34 #include "powernv.h"
35 #include "pci.h"
36 
37 static int __pe_printk(const char *level, const struct pnv_ioda_pe *pe,
38 		       struct va_format *vaf)
39 {
40 	char pfix[32];
41 
42 	if (pe->pdev)
43 		strlcpy(pfix, dev_name(&pe->pdev->dev), sizeof(pfix));
44 	else
45 		sprintf(pfix, "%04x:%02x     ",
46 			pci_domain_nr(pe->pbus), pe->pbus->number);
47 	return printk("pci %s%s: [PE# %.3d] %pV", level, pfix, pe->pe_number, vaf);
48 }
49 
50 #define define_pe_printk_level(func, kern_level)		\
51 static int func(const struct pnv_ioda_pe *pe, const char *fmt, ...)	\
52 {								\
53 	struct va_format vaf;					\
54 	va_list args;						\
55 	int r;							\
56 								\
57 	va_start(args, fmt);					\
58 								\
59 	vaf.fmt = fmt;						\
60 	vaf.va = &args;						\
61 								\
62 	r = __pe_printk(kern_level, pe, &vaf);			\
63 	va_end(args);						\
64 								\
65 	return r;						\
66 }								\
67 
68 define_pe_printk_level(pe_err, KERN_ERR);
69 define_pe_printk_level(pe_warn, KERN_WARNING);
70 define_pe_printk_level(pe_info, KERN_INFO);
71 
72 static struct pci_dn *pnv_ioda_get_pdn(struct pci_dev *dev)
73 {
74 	struct device_node *np;
75 
76 	np = pci_device_to_OF_node(dev);
77 	if (!np)
78 		return NULL;
79 	return PCI_DN(np);
80 }
81 
82 static int __devinit pnv_ioda_alloc_pe(struct pnv_phb *phb)
83 {
84 	unsigned long pe;
85 
86 	do {
87 		pe = find_next_zero_bit(phb->ioda.pe_alloc,
88 					phb->ioda.total_pe, 0);
89 		if (pe >= phb->ioda.total_pe)
90 			return IODA_INVALID_PE;
91 	} while(test_and_set_bit(pe, phb->ioda.pe_alloc));
92 
93 	phb->ioda.pe_array[pe].pe_number = pe;
94 	return pe;
95 }
96 
97 static void __devinit pnv_ioda_free_pe(struct pnv_phb *phb, int pe)
98 {
99 	WARN_ON(phb->ioda.pe_array[pe].pdev);
100 
101 	memset(&phb->ioda.pe_array[pe], 0, sizeof(struct pnv_ioda_pe));
102 	clear_bit(pe, phb->ioda.pe_alloc);
103 }
104 
105 /* Currently those 2 are only used when MSIs are enabled, this will change
106  * but in the meantime, we need to protect them to avoid warnings
107  */
108 #ifdef CONFIG_PCI_MSI
109 static struct pnv_ioda_pe * __devinit pnv_ioda_get_pe(struct pci_dev *dev)
110 {
111 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
112 	struct pnv_phb *phb = hose->private_data;
113 	struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
114 
115 	if (!pdn)
116 		return NULL;
117 	if (pdn->pe_number == IODA_INVALID_PE)
118 		return NULL;
119 	return &phb->ioda.pe_array[pdn->pe_number];
120 }
121 #endif /* CONFIG_PCI_MSI */
122 
123 static int __devinit pnv_ioda_configure_pe(struct pnv_phb *phb,
124 					   struct pnv_ioda_pe *pe)
125 {
126 	struct pci_dev *parent;
127 	uint8_t bcomp, dcomp, fcomp;
128 	long rc, rid_end, rid;
129 
130 	/* Bus validation ? */
131 	if (pe->pbus) {
132 		int count;
133 
134 		dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER;
135 		fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER;
136 		parent = pe->pbus->self;
137 		if (pe->flags & PNV_IODA_PE_BUS_ALL)
138 			count = pe->pbus->busn_res.end - pe->pbus->busn_res.start + 1;
139 		else
140 			count = 1;
141 
142 		switch(count) {
143 		case  1: bcomp = OpalPciBusAll;		break;
144 		case  2: bcomp = OpalPciBus7Bits;	break;
145 		case  4: bcomp = OpalPciBus6Bits;	break;
146 		case  8: bcomp = OpalPciBus5Bits;	break;
147 		case 16: bcomp = OpalPciBus4Bits;	break;
148 		case 32: bcomp = OpalPciBus3Bits;	break;
149 		default:
150 			pr_err("%s: Number of subordinate busses %d"
151 			       " unsupported\n",
152 			       pci_name(pe->pbus->self), count);
153 			/* Do an exact match only */
154 			bcomp = OpalPciBusAll;
155 		}
156 		rid_end = pe->rid + (count << 8);
157 	} else {
158 		parent = pe->pdev->bus->self;
159 		bcomp = OpalPciBusAll;
160 		dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
161 		fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER;
162 		rid_end = pe->rid + 1;
163 	}
164 
165 	/* Associate PE in PELT */
166 	rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid,
167 			     bcomp, dcomp, fcomp, OPAL_MAP_PE);
168 	if (rc) {
169 		pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc);
170 		return -ENXIO;
171 	}
172 	opal_pci_eeh_freeze_clear(phb->opal_id, pe->pe_number,
173 				  OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
174 
175 	/* Add to all parents PELT-V */
176 	while (parent) {
177 		struct pci_dn *pdn = pnv_ioda_get_pdn(parent);
178 		if (pdn && pdn->pe_number != IODA_INVALID_PE) {
179 			rc = opal_pci_set_peltv(phb->opal_id, pdn->pe_number,
180 						pe->pe_number, OPAL_ADD_PE_TO_DOMAIN);
181 			/* XXX What to do in case of error ? */
182 		}
183 		parent = parent->bus->self;
184 	}
185 	/* Setup reverse map */
186 	for (rid = pe->rid; rid < rid_end; rid++)
187 		phb->ioda.pe_rmap[rid] = pe->pe_number;
188 
189 	/* Setup one MVTs on IODA1 */
190 	if (phb->type == PNV_PHB_IODA1) {
191 		pe->mve_number = pe->pe_number;
192 		rc = opal_pci_set_mve(phb->opal_id, pe->mve_number,
193 				      pe->pe_number);
194 		if (rc) {
195 			pe_err(pe, "OPAL error %ld setting up MVE %d\n",
196 			       rc, pe->mve_number);
197 			pe->mve_number = -1;
198 		} else {
199 			rc = opal_pci_set_mve_enable(phb->opal_id,
200 						     pe->mve_number, OPAL_ENABLE_MVE);
201 			if (rc) {
202 				pe_err(pe, "OPAL error %ld enabling MVE %d\n",
203 				       rc, pe->mve_number);
204 				pe->mve_number = -1;
205 			}
206 		}
207 	} else if (phb->type == PNV_PHB_IODA2)
208 		pe->mve_number = 0;
209 
210 	return 0;
211 }
212 
213 static void __devinit pnv_ioda_link_pe_by_weight(struct pnv_phb *phb,
214 						 struct pnv_ioda_pe *pe)
215 {
216 	struct pnv_ioda_pe *lpe;
217 
218 	list_for_each_entry(lpe, &phb->ioda.pe_dma_list, dma_link) {
219 		if (lpe->dma_weight < pe->dma_weight) {
220 			list_add_tail(&pe->dma_link, &lpe->dma_link);
221 			return;
222 		}
223 	}
224 	list_add_tail(&pe->dma_link, &phb->ioda.pe_dma_list);
225 }
226 
227 static unsigned int pnv_ioda_dma_weight(struct pci_dev *dev)
228 {
229 	/* This is quite simplistic. The "base" weight of a device
230 	 * is 10. 0 means no DMA is to be accounted for it.
231 	 */
232 
233 	/* If it's a bridge, no DMA */
234 	if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
235 		return 0;
236 
237 	/* Reduce the weight of slow USB controllers */
238 	if (dev->class == PCI_CLASS_SERIAL_USB_UHCI ||
239 	    dev->class == PCI_CLASS_SERIAL_USB_OHCI ||
240 	    dev->class == PCI_CLASS_SERIAL_USB_EHCI)
241 		return 3;
242 
243 	/* Increase the weight of RAID (includes Obsidian) */
244 	if ((dev->class >> 8) == PCI_CLASS_STORAGE_RAID)
245 		return 15;
246 
247 	/* Default */
248 	return 10;
249 }
250 
251 #if 0
252 static struct pnv_ioda_pe * __devinit pnv_ioda_setup_dev_PE(struct pci_dev *dev)
253 {
254 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
255 	struct pnv_phb *phb = hose->private_data;
256 	struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
257 	struct pnv_ioda_pe *pe;
258 	int pe_num;
259 
260 	if (!pdn) {
261 		pr_err("%s: Device tree node not associated properly\n",
262 			   pci_name(dev));
263 		return NULL;
264 	}
265 	if (pdn->pe_number != IODA_INVALID_PE)
266 		return NULL;
267 
268 	/* PE#0 has been pre-set */
269 	if (dev->bus->number == 0)
270 		pe_num = 0;
271 	else
272 		pe_num = pnv_ioda_alloc_pe(phb);
273 	if (pe_num == IODA_INVALID_PE) {
274 		pr_warning("%s: Not enough PE# available, disabling device\n",
275 			   pci_name(dev));
276 		return NULL;
277 	}
278 
279 	/* NOTE: We get only one ref to the pci_dev for the pdn, not for the
280 	 * pointer in the PE data structure, both should be destroyed at the
281 	 * same time. However, this needs to be looked at more closely again
282 	 * once we actually start removing things (Hotplug, SR-IOV, ...)
283 	 *
284 	 * At some point we want to remove the PDN completely anyways
285 	 */
286 	pe = &phb->ioda.pe_array[pe_num];
287 	pci_dev_get(dev);
288 	pdn->pcidev = dev;
289 	pdn->pe_number = pe_num;
290 	pe->pdev = dev;
291 	pe->pbus = NULL;
292 	pe->tce32_seg = -1;
293 	pe->mve_number = -1;
294 	pe->rid = dev->bus->number << 8 | pdn->devfn;
295 
296 	pe_info(pe, "Associated device to PE\n");
297 
298 	if (pnv_ioda_configure_pe(phb, pe)) {
299 		/* XXX What do we do here ? */
300 		if (pe_num)
301 			pnv_ioda_free_pe(phb, pe_num);
302 		pdn->pe_number = IODA_INVALID_PE;
303 		pe->pdev = NULL;
304 		pci_dev_put(dev);
305 		return NULL;
306 	}
307 
308 	/* Assign a DMA weight to the device */
309 	pe->dma_weight = pnv_ioda_dma_weight(dev);
310 	if (pe->dma_weight != 0) {
311 		phb->ioda.dma_weight += pe->dma_weight;
312 		phb->ioda.dma_pe_count++;
313 	}
314 
315 	/* Link the PE */
316 	pnv_ioda_link_pe_by_weight(phb, pe);
317 
318 	return pe;
319 }
320 #endif /* Useful for SRIOV case */
321 
322 static void pnv_ioda_setup_same_PE(struct pci_bus *bus, struct pnv_ioda_pe *pe)
323 {
324 	struct pci_dev *dev;
325 
326 	list_for_each_entry(dev, &bus->devices, bus_list) {
327 		struct pci_dn *pdn = pnv_ioda_get_pdn(dev);
328 
329 		if (pdn == NULL) {
330 			pr_warn("%s: No device node associated with device !\n",
331 				pci_name(dev));
332 			continue;
333 		}
334 		pci_dev_get(dev);
335 		pdn->pcidev = dev;
336 		pdn->pe_number = pe->pe_number;
337 		pe->dma_weight += pnv_ioda_dma_weight(dev);
338 		if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate)
339 			pnv_ioda_setup_same_PE(dev->subordinate, pe);
340 	}
341 }
342 
343 /*
344  * There're 2 types of PCI bus sensitive PEs: One that is compromised of
345  * single PCI bus. Another one that contains the primary PCI bus and its
346  * subordinate PCI devices and buses. The second type of PE is normally
347  * orgiriated by PCIe-to-PCI bridge or PLX switch downstream ports.
348  */
349 static void __devinit pnv_ioda_setup_bus_PE(struct pci_bus *bus, int all)
350 {
351 	struct pci_controller *hose = pci_bus_to_host(bus);
352 	struct pnv_phb *phb = hose->private_data;
353 	struct pnv_ioda_pe *pe;
354 	int pe_num;
355 
356 	pe_num = pnv_ioda_alloc_pe(phb);
357 	if (pe_num == IODA_INVALID_PE) {
358 		pr_warning("%s: Not enough PE# available for PCI bus %04x:%02x\n",
359 			__func__, pci_domain_nr(bus), bus->number);
360 		return;
361 	}
362 
363 	pe = &phb->ioda.pe_array[pe_num];
364 	pe->flags = (all ? PNV_IODA_PE_BUS_ALL : PNV_IODA_PE_BUS);
365 	pe->pbus = bus;
366 	pe->pdev = NULL;
367 	pe->tce32_seg = -1;
368 	pe->mve_number = -1;
369 	pe->rid = bus->busn_res.start << 8;
370 	pe->dma_weight = 0;
371 
372 	if (all)
373 		pe_info(pe, "Secondary bus %d..%d associated with PE#%d\n",
374 			bus->busn_res.start, bus->busn_res.end, pe_num);
375 	else
376 		pe_info(pe, "Secondary bus %d associated with PE#%d\n",
377 			bus->busn_res.start, pe_num);
378 
379 	if (pnv_ioda_configure_pe(phb, pe)) {
380 		/* XXX What do we do here ? */
381 		if (pe_num)
382 			pnv_ioda_free_pe(phb, pe_num);
383 		pe->pbus = NULL;
384 		return;
385 	}
386 
387 	/* Associate it with all child devices */
388 	pnv_ioda_setup_same_PE(bus, pe);
389 
390 	/* Put PE to the list */
391 	list_add_tail(&pe->list, &phb->ioda.pe_list);
392 
393 	/* Account for one DMA PE if at least one DMA capable device exist
394 	 * below the bridge
395 	 */
396 	if (pe->dma_weight != 0) {
397 		phb->ioda.dma_weight += pe->dma_weight;
398 		phb->ioda.dma_pe_count++;
399 	}
400 
401 	/* Link the PE */
402 	pnv_ioda_link_pe_by_weight(phb, pe);
403 }
404 
405 static void __devinit pnv_ioda_setup_PEs(struct pci_bus *bus)
406 {
407 	struct pci_dev *dev;
408 
409 	pnv_ioda_setup_bus_PE(bus, 0);
410 
411 	list_for_each_entry(dev, &bus->devices, bus_list) {
412 		if (dev->subordinate) {
413 			if (pci_pcie_type(dev) == PCI_EXP_TYPE_PCI_BRIDGE)
414 				pnv_ioda_setup_bus_PE(dev->subordinate, 1);
415 			else
416 				pnv_ioda_setup_PEs(dev->subordinate);
417 		}
418 	}
419 }
420 
421 /*
422  * Configure PEs so that the downstream PCI buses and devices
423  * could have their associated PE#. Unfortunately, we didn't
424  * figure out the way to identify the PLX bridge yet. So we
425  * simply put the PCI bus and the subordinate behind the root
426  * port to PE# here. The game rule here is expected to be changed
427  * as soon as we can detected PLX bridge correctly.
428  */
429 static void __devinit pnv_pci_ioda_setup_PEs(void)
430 {
431 	struct pci_controller *hose, *tmp;
432 
433 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
434 		pnv_ioda_setup_PEs(hose->bus);
435 	}
436 }
437 
438 static void __devinit pnv_pci_ioda_dma_dev_setup(struct pnv_phb *phb,
439 						 struct pci_dev *dev)
440 {
441 	/* We delay DMA setup after we have assigned all PE# */
442 }
443 
444 static void __devinit pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe,
445 					     struct pci_bus *bus)
446 {
447 	struct pci_dev *dev;
448 
449 	list_for_each_entry(dev, &bus->devices, bus_list) {
450 		set_iommu_table_base(&dev->dev, &pe->tce32_table);
451 		if (dev->subordinate)
452 			pnv_ioda_setup_bus_dma(pe, dev->subordinate);
453 	}
454 }
455 
456 static void __devinit pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb,
457 						struct pnv_ioda_pe *pe,
458 						unsigned int base,
459 						unsigned int segs)
460 {
461 
462 	struct page *tce_mem = NULL;
463 	const __be64 *swinvp;
464 	struct iommu_table *tbl;
465 	unsigned int i;
466 	int64_t rc;
467 	void *addr;
468 
469 	/* 256M DMA window, 4K TCE pages, 8 bytes TCE */
470 #define TCE32_TABLE_SIZE	((0x10000000 / 0x1000) * 8)
471 
472 	/* XXX FIXME: Handle 64-bit only DMA devices */
473 	/* XXX FIXME: Provide 64-bit DMA facilities & non-4K TCE tables etc.. */
474 	/* XXX FIXME: Allocate multi-level tables on PHB3 */
475 
476 	/* We shouldn't already have a 32-bit DMA associated */
477 	if (WARN_ON(pe->tce32_seg >= 0))
478 		return;
479 
480 	/* Grab a 32-bit TCE table */
481 	pe->tce32_seg = base;
482 	pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n",
483 		(base << 28), ((base + segs) << 28) - 1);
484 
485 	/* XXX Currently, we allocate one big contiguous table for the
486 	 * TCEs. We only really need one chunk per 256M of TCE space
487 	 * (ie per segment) but that's an optimization for later, it
488 	 * requires some added smarts with our get/put_tce implementation
489 	 */
490 	tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL,
491 				   get_order(TCE32_TABLE_SIZE * segs));
492 	if (!tce_mem) {
493 		pe_err(pe, " Failed to allocate a 32-bit TCE memory\n");
494 		goto fail;
495 	}
496 	addr = page_address(tce_mem);
497 	memset(addr, 0, TCE32_TABLE_SIZE * segs);
498 
499 	/* Configure HW */
500 	for (i = 0; i < segs; i++) {
501 		rc = opal_pci_map_pe_dma_window(phb->opal_id,
502 					      pe->pe_number,
503 					      base + i, 1,
504 					      __pa(addr) + TCE32_TABLE_SIZE * i,
505 					      TCE32_TABLE_SIZE, 0x1000);
506 		if (rc) {
507 			pe_err(pe, " Failed to configure 32-bit TCE table,"
508 			       " err %ld\n", rc);
509 			goto fail;
510 		}
511 	}
512 
513 	/* Setup linux iommu table */
514 	tbl = &pe->tce32_table;
515 	pnv_pci_setup_iommu_table(tbl, addr, TCE32_TABLE_SIZE * segs,
516 				  base << 28);
517 
518 	/* OPAL variant of P7IOC SW invalidated TCEs */
519 	swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL);
520 	if (swinvp) {
521 		/* We need a couple more fields -- an address and a data
522 		 * to or.  Since the bus is only printed out on table free
523 		 * errors, and on the first pass the data will be a relative
524 		 * bus number, print that out instead.
525 		 */
526 		tbl->it_busno = 0;
527 		tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
528 		tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE
529 			| TCE_PCI_SWINV_PAIR;
530 	}
531 	iommu_init_table(tbl, phb->hose->node);
532 
533 	if (pe->pdev)
534 		set_iommu_table_base(&pe->pdev->dev, tbl);
535 	else
536 		pnv_ioda_setup_bus_dma(pe, pe->pbus);
537 
538 	return;
539  fail:
540 	/* XXX Failure: Try to fallback to 64-bit only ? */
541 	if (pe->tce32_seg >= 0)
542 		pe->tce32_seg = -1;
543 	if (tce_mem)
544 		__free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs));
545 }
546 
547 static void __devinit pnv_ioda_setup_dma(struct pnv_phb *phb)
548 {
549 	struct pci_controller *hose = phb->hose;
550 	unsigned int residual, remaining, segs, tw, base;
551 	struct pnv_ioda_pe *pe;
552 
553 	/* If we have more PE# than segments available, hand out one
554 	 * per PE until we run out and let the rest fail. If not,
555 	 * then we assign at least one segment per PE, plus more based
556 	 * on the amount of devices under that PE
557 	 */
558 	if (phb->ioda.dma_pe_count > phb->ioda.tce32_count)
559 		residual = 0;
560 	else
561 		residual = phb->ioda.tce32_count -
562 			phb->ioda.dma_pe_count;
563 
564 	pr_info("PCI: Domain %04x has %ld available 32-bit DMA segments\n",
565 		hose->global_number, phb->ioda.tce32_count);
566 	pr_info("PCI: %d PE# for a total weight of %d\n",
567 		phb->ioda.dma_pe_count, phb->ioda.dma_weight);
568 
569 	/* Walk our PE list and configure their DMA segments, hand them
570 	 * out one base segment plus any residual segments based on
571 	 * weight
572 	 */
573 	remaining = phb->ioda.tce32_count;
574 	tw = phb->ioda.dma_weight;
575 	base = 0;
576 	list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) {
577 		if (!pe->dma_weight)
578 			continue;
579 		if (!remaining) {
580 			pe_warn(pe, "No DMA32 resources available\n");
581 			continue;
582 		}
583 		segs = 1;
584 		if (residual) {
585 			segs += ((pe->dma_weight * residual)  + (tw / 2)) / tw;
586 			if (segs > remaining)
587 				segs = remaining;
588 		}
589 		pe_info(pe, "DMA weight %d, assigned %d DMA32 segments\n",
590 			pe->dma_weight, segs);
591 		pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs);
592 		remaining -= segs;
593 		base += segs;
594 	}
595 }
596 
597 #ifdef CONFIG_PCI_MSI
598 static int pnv_pci_ioda_msi_setup(struct pnv_phb *phb, struct pci_dev *dev,
599 				  unsigned int hwirq, unsigned int is_64,
600 				  struct msi_msg *msg)
601 {
602 	struct pnv_ioda_pe *pe = pnv_ioda_get_pe(dev);
603 	unsigned int xive_num = hwirq - phb->msi_base;
604 	uint64_t addr64;
605 	uint32_t addr32, data;
606 	int rc;
607 
608 	/* No PE assigned ? bail out ... no MSI for you ! */
609 	if (pe == NULL)
610 		return -ENXIO;
611 
612 	/* Check if we have an MVE */
613 	if (pe->mve_number < 0)
614 		return -ENXIO;
615 
616 	/* Assign XIVE to PE */
617 	rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num);
618 	if (rc) {
619 		pr_warn("%s: OPAL error %d setting XIVE %d PE\n",
620 			pci_name(dev), rc, xive_num);
621 		return -EIO;
622 	}
623 
624 	if (is_64) {
625 		rc = opal_get_msi_64(phb->opal_id, pe->mve_number, xive_num, 1,
626 				     &addr64, &data);
627 		if (rc) {
628 			pr_warn("%s: OPAL error %d getting 64-bit MSI data\n",
629 				pci_name(dev), rc);
630 			return -EIO;
631 		}
632 		msg->address_hi = addr64 >> 32;
633 		msg->address_lo = addr64 & 0xfffffffful;
634 	} else {
635 		rc = opal_get_msi_32(phb->opal_id, pe->mve_number, xive_num, 1,
636 				     &addr32, &data);
637 		if (rc) {
638 			pr_warn("%s: OPAL error %d getting 32-bit MSI data\n",
639 				pci_name(dev), rc);
640 			return -EIO;
641 		}
642 		msg->address_hi = 0;
643 		msg->address_lo = addr32;
644 	}
645 	msg->data = data;
646 
647 	pr_devel("%s: %s-bit MSI on hwirq %x (xive #%d),"
648 		 " address=%x_%08x data=%x PE# %d\n",
649 		 pci_name(dev), is_64 ? "64" : "32", hwirq, xive_num,
650 		 msg->address_hi, msg->address_lo, data, pe->pe_number);
651 
652 	return 0;
653 }
654 
655 static void pnv_pci_init_ioda_msis(struct pnv_phb *phb)
656 {
657 	unsigned int bmap_size;
658 	const __be32 *prop = of_get_property(phb->hose->dn,
659 					     "ibm,opal-msi-ranges", NULL);
660 	if (!prop) {
661 		/* BML Fallback */
662 		prop = of_get_property(phb->hose->dn, "msi-ranges", NULL);
663 	}
664 	if (!prop)
665 		return;
666 
667 	phb->msi_base = be32_to_cpup(prop);
668 	phb->msi_count = be32_to_cpup(prop + 1);
669 	bmap_size = BITS_TO_LONGS(phb->msi_count) * sizeof(unsigned long);
670 	phb->msi_map = zalloc_maybe_bootmem(bmap_size, GFP_KERNEL);
671 	if (!phb->msi_map) {
672 		pr_err("PCI %d: Failed to allocate MSI bitmap !\n",
673 		       phb->hose->global_number);
674 		return;
675 	}
676 	phb->msi_setup = pnv_pci_ioda_msi_setup;
677 	phb->msi32_support = 1;
678 	pr_info("  Allocated bitmap for %d MSIs (base IRQ 0x%x)\n",
679 		phb->msi_count, phb->msi_base);
680 }
681 #else
682 static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) { }
683 #endif /* CONFIG_PCI_MSI */
684 
685 /*
686  * This function is supposed to be called on basis of PE from top
687  * to bottom style. So the the I/O or MMIO segment assigned to
688  * parent PE could be overrided by its child PEs if necessary.
689  */
690 static void __devinit pnv_ioda_setup_pe_seg(struct pci_controller *hose,
691 				struct pnv_ioda_pe *pe)
692 {
693 	struct pnv_phb *phb = hose->private_data;
694 	struct pci_bus_region region;
695 	struct resource *res;
696 	int i, index;
697 	int rc;
698 
699 	/*
700 	 * NOTE: We only care PCI bus based PE for now. For PCI
701 	 * device based PE, for example SRIOV sensitive VF should
702 	 * be figured out later.
703 	 */
704 	BUG_ON(!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)));
705 
706 	pci_bus_for_each_resource(pe->pbus, res, i) {
707 		if (!res || !res->flags ||
708 		    res->start > res->end)
709 			continue;
710 
711 		if (res->flags & IORESOURCE_IO) {
712 			region.start = res->start - phb->ioda.io_pci_base;
713 			region.end   = res->end - phb->ioda.io_pci_base;
714 			index = region.start / phb->ioda.io_segsize;
715 
716 			while (index < phb->ioda.total_pe &&
717 			       region.start <= region.end) {
718 				phb->ioda.io_segmap[index] = pe->pe_number;
719 				rc = opal_pci_map_pe_mmio_window(phb->opal_id,
720 					pe->pe_number, OPAL_IO_WINDOW_TYPE, 0, index);
721 				if (rc != OPAL_SUCCESS) {
722 					pr_err("%s: OPAL error %d when mapping IO "
723 					       "segment #%d to PE#%d\n",
724 					       __func__, rc, index, pe->pe_number);
725 					break;
726 				}
727 
728 				region.start += phb->ioda.io_segsize;
729 				index++;
730 			}
731 		} else if (res->flags & IORESOURCE_MEM) {
732 			region.start = res->start -
733 				       hose->pci_mem_offset -
734 				       phb->ioda.m32_pci_base;
735 			region.end   = res->end -
736 				       hose->pci_mem_offset -
737 				       phb->ioda.m32_pci_base;
738 			index = region.start / phb->ioda.m32_segsize;
739 
740 			while (index < phb->ioda.total_pe &&
741 			       region.start <= region.end) {
742 				phb->ioda.m32_segmap[index] = pe->pe_number;
743 				rc = opal_pci_map_pe_mmio_window(phb->opal_id,
744 					pe->pe_number, OPAL_M32_WINDOW_TYPE, 0, index);
745 				if (rc != OPAL_SUCCESS) {
746 					pr_err("%s: OPAL error %d when mapping M32 "
747 					       "segment#%d to PE#%d",
748 					       __func__, rc, index, pe->pe_number);
749 					break;
750 				}
751 
752 				region.start += phb->ioda.m32_segsize;
753 				index++;
754 			}
755 		}
756 	}
757 }
758 
759 static void __devinit pnv_pci_ioda_setup_seg(void)
760 {
761 	struct pci_controller *tmp, *hose;
762 	struct pnv_phb *phb;
763 	struct pnv_ioda_pe *pe;
764 
765 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
766 		phb = hose->private_data;
767 		list_for_each_entry(pe, &phb->ioda.pe_list, list) {
768 			pnv_ioda_setup_pe_seg(hose, pe);
769 		}
770 	}
771 }
772 
773 static void __devinit pnv_pci_ioda_setup_DMA(void)
774 {
775 	struct pci_controller *hose, *tmp;
776 	struct pnv_phb *phb;
777 
778 	list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
779 		pnv_ioda_setup_dma(hose->private_data);
780 
781 		/* Mark the PHB initialization done */
782 		phb = hose->private_data;
783 		phb->initialized = 1;
784 	}
785 }
786 
787 static void __devinit pnv_pci_ioda_fixup(void)
788 {
789 	pnv_pci_ioda_setup_PEs();
790 	pnv_pci_ioda_setup_seg();
791 	pnv_pci_ioda_setup_DMA();
792 }
793 
794 /*
795  * Returns the alignment for I/O or memory windows for P2P
796  * bridges. That actually depends on how PEs are segmented.
797  * For now, we return I/O or M32 segment size for PE sensitive
798  * P2P bridges. Otherwise, the default values (4KiB for I/O,
799  * 1MiB for memory) will be returned.
800  *
801  * The current PCI bus might be put into one PE, which was
802  * create against the parent PCI bridge. For that case, we
803  * needn't enlarge the alignment so that we can save some
804  * resources.
805  */
806 static resource_size_t pnv_pci_window_alignment(struct pci_bus *bus,
807 						unsigned long type)
808 {
809 	struct pci_dev *bridge;
810 	struct pci_controller *hose = pci_bus_to_host(bus);
811 	struct pnv_phb *phb = hose->private_data;
812 	int num_pci_bridges = 0;
813 
814 	bridge = bus->self;
815 	while (bridge) {
816 		if (pci_pcie_type(bridge) == PCI_EXP_TYPE_PCI_BRIDGE) {
817 			num_pci_bridges++;
818 			if (num_pci_bridges >= 2)
819 				return 1;
820 		}
821 
822 		bridge = bridge->bus->self;
823 	}
824 
825 	/* We need support prefetchable memory window later */
826 	if (type & IORESOURCE_MEM)
827 		return phb->ioda.m32_segsize;
828 
829 	return phb->ioda.io_segsize;
830 }
831 
832 /* Prevent enabling devices for which we couldn't properly
833  * assign a PE
834  */
835 static int __devinit pnv_pci_enable_device_hook(struct pci_dev *dev)
836 {
837 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
838 	struct pnv_phb *phb = hose->private_data;
839 	struct pci_dn *pdn;
840 
841 	/* The function is probably called while the PEs have
842 	 * not be created yet. For example, resource reassignment
843 	 * during PCI probe period. We just skip the check if
844 	 * PEs isn't ready.
845 	 */
846 	if (!phb->initialized)
847 		return 0;
848 
849 	pdn = pnv_ioda_get_pdn(dev);
850 	if (!pdn || pdn->pe_number == IODA_INVALID_PE)
851 		return -EINVAL;
852 
853 	return 0;
854 }
855 
856 static u32 pnv_ioda_bdfn_to_pe(struct pnv_phb *phb, struct pci_bus *bus,
857 			       u32 devfn)
858 {
859 	return phb->ioda.pe_rmap[(bus->number << 8) | devfn];
860 }
861 
862 void __init pnv_pci_init_ioda1_phb(struct device_node *np)
863 {
864 	struct pci_controller *hose;
865 	static int primary = 1;
866 	struct pnv_phb *phb;
867 	unsigned long size, m32map_off, iomap_off, pemap_off;
868 	const u64 *prop64;
869 	u64 phb_id;
870 	void *aux;
871 	long rc;
872 
873 	pr_info(" Initializing IODA OPAL PHB %s\n", np->full_name);
874 
875 	prop64 = of_get_property(np, "ibm,opal-phbid", NULL);
876 	if (!prop64) {
877 		pr_err("  Missing \"ibm,opal-phbid\" property !\n");
878 		return;
879 	}
880 	phb_id = be64_to_cpup(prop64);
881 	pr_debug("  PHB-ID  : 0x%016llx\n", phb_id);
882 
883 	phb = alloc_bootmem(sizeof(struct pnv_phb));
884 	if (phb) {
885 		memset(phb, 0, sizeof(struct pnv_phb));
886 		phb->hose = hose = pcibios_alloc_controller(np);
887 	}
888 	if (!phb || !phb->hose) {
889 		pr_err("PCI: Failed to allocate PCI controller for %s\n",
890 		       np->full_name);
891 		return;
892 	}
893 
894 	spin_lock_init(&phb->lock);
895 	/* XXX Use device-tree */
896 	hose->first_busno = 0;
897 	hose->last_busno = 0xff;
898 	hose->private_data = phb;
899 	phb->opal_id = phb_id;
900 	phb->type = PNV_PHB_IODA1;
901 
902 	/* Detect specific models for error handling */
903 	if (of_device_is_compatible(np, "ibm,p7ioc-pciex"))
904 		phb->model = PNV_PHB_MODEL_P7IOC;
905 	else
906 		phb->model = PNV_PHB_MODEL_UNKNOWN;
907 
908 	/* We parse "ranges" now since we need to deduce the register base
909 	 * from the IO base
910 	 */
911 	pci_process_bridge_OF_ranges(phb->hose, np, primary);
912 	primary = 0;
913 
914 	/* Magic formula from Milton */
915 	phb->regs = of_iomap(np, 0);
916 	if (phb->regs == NULL)
917 		pr_err("  Failed to map registers !\n");
918 
919 
920 	/* XXX This is hack-a-thon. This needs to be changed so that:
921 	 *  - we obtain stuff like PE# etc... from device-tree
922 	 *  - we properly re-allocate M32 ourselves
923 	 *    (the OFW one isn't very good)
924 	 */
925 
926 	/* Initialize more IODA stuff */
927 	phb->ioda.total_pe = 128;
928 
929 	phb->ioda.m32_size = resource_size(&hose->mem_resources[0]);
930 	/* OFW Has already off top 64k of M32 space (MSI space) */
931 	phb->ioda.m32_size += 0x10000;
932 
933 	phb->ioda.m32_segsize = phb->ioda.m32_size / phb->ioda.total_pe;
934 	phb->ioda.m32_pci_base = hose->mem_resources[0].start -
935 		hose->pci_mem_offset;
936 	phb->ioda.io_size = hose->pci_io_size;
937 	phb->ioda.io_segsize = phb->ioda.io_size / phb->ioda.total_pe;
938 	phb->ioda.io_pci_base = 0; /* XXX calculate this ? */
939 
940 	/* Allocate aux data & arrays */
941 	size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long));
942 	m32map_off = size;
943 	size += phb->ioda.total_pe * sizeof(phb->ioda.m32_segmap[0]);
944 	iomap_off = size;
945 	size += phb->ioda.total_pe * sizeof(phb->ioda.io_segmap[0]);
946 	pemap_off = size;
947 	size += phb->ioda.total_pe * sizeof(struct pnv_ioda_pe);
948 	aux = alloc_bootmem(size);
949 	memset(aux, 0, size);
950 	phb->ioda.pe_alloc = aux;
951 	phb->ioda.m32_segmap = aux + m32map_off;
952 	phb->ioda.io_segmap = aux + iomap_off;
953 	phb->ioda.pe_array = aux + pemap_off;
954 	set_bit(0, phb->ioda.pe_alloc);
955 
956 	INIT_LIST_HEAD(&phb->ioda.pe_dma_list);
957 	INIT_LIST_HEAD(&phb->ioda.pe_list);
958 
959 	/* Calculate how many 32-bit TCE segments we have */
960 	phb->ioda.tce32_count = phb->ioda.m32_pci_base >> 28;
961 
962 	/* Clear unusable m64 */
963 	hose->mem_resources[1].flags = 0;
964 	hose->mem_resources[1].start = 0;
965 	hose->mem_resources[1].end = 0;
966 	hose->mem_resources[2].flags = 0;
967 	hose->mem_resources[2].start = 0;
968 	hose->mem_resources[2].end = 0;
969 
970 #if 0
971 	rc = opal_pci_set_phb_mem_window(opal->phb_id,
972 					 window_type,
973 					 window_num,
974 					 starting_real_address,
975 					 starting_pci_address,
976 					 segment_size);
977 #endif
978 
979 	pr_info("  %d PE's M32: 0x%x [segment=0x%x] IO: 0x%x [segment=0x%x]\n",
980 		phb->ioda.total_pe,
981 		phb->ioda.m32_size, phb->ioda.m32_segsize,
982 		phb->ioda.io_size, phb->ioda.io_segsize);
983 
984 	if (phb->regs)  {
985 		pr_devel(" BUID     = 0x%016llx\n", in_be64(phb->regs + 0x100));
986 		pr_devel(" PHB2_CR  = 0x%016llx\n", in_be64(phb->regs + 0x160));
987 		pr_devel(" IO_BAR   = 0x%016llx\n", in_be64(phb->regs + 0x170));
988 		pr_devel(" IO_BAMR  = 0x%016llx\n", in_be64(phb->regs + 0x178));
989 		pr_devel(" IO_SAR   = 0x%016llx\n", in_be64(phb->regs + 0x180));
990 		pr_devel(" M32_BAR  = 0x%016llx\n", in_be64(phb->regs + 0x190));
991 		pr_devel(" M32_BAMR = 0x%016llx\n", in_be64(phb->regs + 0x198));
992 		pr_devel(" M32_SAR  = 0x%016llx\n", in_be64(phb->regs + 0x1a0));
993 	}
994 	phb->hose->ops = &pnv_pci_ops;
995 
996 	/* Setup RID -> PE mapping function */
997 	phb->bdfn_to_pe = pnv_ioda_bdfn_to_pe;
998 
999 	/* Setup TCEs */
1000 	phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
1001 
1002 	/* Setup MSI support */
1003 	pnv_pci_init_ioda_msis(phb);
1004 
1005 	/*
1006 	 * We pass the PCI probe flag PCI_REASSIGN_ALL_RSRC here
1007 	 * to let the PCI core do resource assignment. It's supposed
1008 	 * that the PCI core will do correct I/O and MMIO alignment
1009 	 * for the P2P bridge bars so that each PCI bus (excluding
1010 	 * the child P2P bridges) can form individual PE.
1011 	 */
1012 	ppc_md.pcibios_fixup = pnv_pci_ioda_fixup;
1013 	ppc_md.pcibios_enable_device_hook = pnv_pci_enable_device_hook;
1014 	ppc_md.pcibios_window_alignment = pnv_pci_window_alignment;
1015 	pci_add_flags(PCI_REASSIGN_ALL_RSRC);
1016 
1017 	/* Reset IODA tables to a clean state */
1018 	rc = opal_pci_reset(phb_id, OPAL_PCI_IODA_TABLE_RESET, OPAL_ASSERT_RESET);
1019 	if (rc)
1020 		pr_warning("  OPAL Error %ld performing IODA table reset !\n", rc);
1021 	opal_pci_set_pe(phb_id, 0, 0, 7, 1, 1 , OPAL_MAP_PE);
1022 }
1023 
1024 void __init pnv_pci_init_ioda_hub(struct device_node *np)
1025 {
1026 	struct device_node *phbn;
1027 	const u64 *prop64;
1028 	u64 hub_id;
1029 
1030 	pr_info("Probing IODA IO-Hub %s\n", np->full_name);
1031 
1032 	prop64 = of_get_property(np, "ibm,opal-hubid", NULL);
1033 	if (!prop64) {
1034 		pr_err(" Missing \"ibm,opal-hubid\" property !\n");
1035 		return;
1036 	}
1037 	hub_id = be64_to_cpup(prop64);
1038 	pr_devel(" HUB-ID : 0x%016llx\n", hub_id);
1039 
1040 	/* Count child PHBs */
1041 	for_each_child_of_node(np, phbn) {
1042 		/* Look for IODA1 PHBs */
1043 		if (of_device_is_compatible(phbn, "ibm,ioda-phb"))
1044 			pnv_pci_init_ioda1_phb(phbn);
1045 	}
1046 }
1047