xref: /openbmc/linux/arch/sparc/kernel/pci_sun4v.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
1 /* pci_sun4v.c: SUN4V specific PCI controller support.
2  *
3  * Copyright (C) 2006, 2007, 2008 David S. Miller (davem@davemloft.net)
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/types.h>
8 #include <linux/pci.h>
9 #include <linux/init.h>
10 #include <linux/slab.h>
11 #include <linux/interrupt.h>
12 #include <linux/percpu.h>
13 #include <linux/irq.h>
14 #include <linux/msi.h>
15 #include <linux/log2.h>
16 #include <linux/of_device.h>
17 
18 #include <asm/iommu.h>
19 #include <asm/irq.h>
20 #include <asm/hypervisor.h>
21 #include <asm/prom.h>
22 
23 #include "pci_impl.h"
24 #include "iommu_common.h"
25 
26 #include "pci_sun4v.h"
27 
28 #define DRIVER_NAME	"pci_sun4v"
29 #define PFX		DRIVER_NAME ": "
30 
31 static unsigned long vpci_major = 1;
32 static unsigned long vpci_minor = 1;
33 
34 #define PGLIST_NENTS	(PAGE_SIZE / sizeof(u64))
35 
36 struct iommu_batch {
37 	struct device	*dev;		/* Device mapping is for.	*/
38 	unsigned long	prot;		/* IOMMU page protections	*/
39 	unsigned long	entry;		/* Index into IOTSB.		*/
40 	u64		*pglist;	/* List of physical pages	*/
41 	unsigned long	npages;		/* Number of pages in list.	*/
42 };
43 
44 static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
45 static int iommu_batch_initialized;
46 
47 /* Interrupts must be disabled.  */
48 static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
49 {
50 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
51 
52 	p->dev		= dev;
53 	p->prot		= prot;
54 	p->entry	= entry;
55 	p->npages	= 0;
56 }
57 
58 /* Interrupts must be disabled.  */
59 static long iommu_batch_flush(struct iommu_batch *p)
60 {
61 	struct pci_pbm_info *pbm = p->dev->archdata.host_controller;
62 	unsigned long devhandle = pbm->devhandle;
63 	unsigned long prot = p->prot;
64 	unsigned long entry = p->entry;
65 	u64 *pglist = p->pglist;
66 	unsigned long npages = p->npages;
67 
68 	while (npages != 0) {
69 		long num;
70 
71 		num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
72 					  npages, prot, __pa(pglist));
73 		if (unlikely(num < 0)) {
74 			if (printk_ratelimit())
75 				printk("iommu_batch_flush: IOMMU map of "
76 				       "[%08lx:%08llx:%lx:%lx:%lx] failed with "
77 				       "status %ld\n",
78 				       devhandle, HV_PCI_TSBID(0, entry),
79 				       npages, prot, __pa(pglist), num);
80 			return -1;
81 		}
82 
83 		entry += num;
84 		npages -= num;
85 		pglist += num;
86 	}
87 
88 	p->entry = entry;
89 	p->npages = 0;
90 
91 	return 0;
92 }
93 
94 static inline void iommu_batch_new_entry(unsigned long entry)
95 {
96 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
97 
98 	if (p->entry + p->npages == entry)
99 		return;
100 	if (p->entry != ~0UL)
101 		iommu_batch_flush(p);
102 	p->entry = entry;
103 }
104 
105 /* Interrupts must be disabled.  */
106 static inline long iommu_batch_add(u64 phys_page)
107 {
108 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
109 
110 	BUG_ON(p->npages >= PGLIST_NENTS);
111 
112 	p->pglist[p->npages++] = phys_page;
113 	if (p->npages == PGLIST_NENTS)
114 		return iommu_batch_flush(p);
115 
116 	return 0;
117 }
118 
119 /* Interrupts must be disabled.  */
120 static inline long iommu_batch_end(void)
121 {
122 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
123 
124 	BUG_ON(p->npages >= PGLIST_NENTS);
125 
126 	return iommu_batch_flush(p);
127 }
128 
129 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
130 				   dma_addr_t *dma_addrp, gfp_t gfp)
131 {
132 	unsigned long flags, order, first_page, npages, n;
133 	struct iommu *iommu;
134 	struct page *page;
135 	void *ret;
136 	long entry;
137 	int nid;
138 
139 	size = IO_PAGE_ALIGN(size);
140 	order = get_order(size);
141 	if (unlikely(order >= MAX_ORDER))
142 		return NULL;
143 
144 	npages = size >> IO_PAGE_SHIFT;
145 
146 	nid = dev->archdata.numa_node;
147 	page = alloc_pages_node(nid, gfp, order);
148 	if (unlikely(!page))
149 		return NULL;
150 
151 	first_page = (unsigned long) page_address(page);
152 	memset((char *)first_page, 0, PAGE_SIZE << order);
153 
154 	iommu = dev->archdata.iommu;
155 
156 	spin_lock_irqsave(&iommu->lock, flags);
157 	entry = iommu_range_alloc(dev, iommu, npages, NULL);
158 	spin_unlock_irqrestore(&iommu->lock, flags);
159 
160 	if (unlikely(entry == DMA_ERROR_CODE))
161 		goto range_alloc_fail;
162 
163 	*dma_addrp = (iommu->page_table_map_base +
164 		      (entry << IO_PAGE_SHIFT));
165 	ret = (void *) first_page;
166 	first_page = __pa(first_page);
167 
168 	local_irq_save(flags);
169 
170 	iommu_batch_start(dev,
171 			  (HV_PCI_MAP_ATTR_READ |
172 			   HV_PCI_MAP_ATTR_WRITE),
173 			  entry);
174 
175 	for (n = 0; n < npages; n++) {
176 		long err = iommu_batch_add(first_page + (n * PAGE_SIZE));
177 		if (unlikely(err < 0L))
178 			goto iommu_map_fail;
179 	}
180 
181 	if (unlikely(iommu_batch_end() < 0L))
182 		goto iommu_map_fail;
183 
184 	local_irq_restore(flags);
185 
186 	return ret;
187 
188 iommu_map_fail:
189 	/* Interrupts are disabled.  */
190 	spin_lock(&iommu->lock);
191 	iommu_range_free(iommu, *dma_addrp, npages);
192 	spin_unlock_irqrestore(&iommu->lock, flags);
193 
194 range_alloc_fail:
195 	free_pages(first_page, order);
196 	return NULL;
197 }
198 
199 static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
200 				 dma_addr_t dvma)
201 {
202 	struct pci_pbm_info *pbm;
203 	struct iommu *iommu;
204 	unsigned long flags, order, npages, entry;
205 	u32 devhandle;
206 
207 	npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
208 	iommu = dev->archdata.iommu;
209 	pbm = dev->archdata.host_controller;
210 	devhandle = pbm->devhandle;
211 	entry = ((dvma - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
212 
213 	spin_lock_irqsave(&iommu->lock, flags);
214 
215 	iommu_range_free(iommu, dvma, npages);
216 
217 	do {
218 		unsigned long num;
219 
220 		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
221 					    npages);
222 		entry += num;
223 		npages -= num;
224 	} while (npages != 0);
225 
226 	spin_unlock_irqrestore(&iommu->lock, flags);
227 
228 	order = get_order(size);
229 	if (order < 10)
230 		free_pages((unsigned long)cpu, order);
231 }
232 
233 static dma_addr_t dma_4v_map_single(struct device *dev, void *ptr, size_t sz,
234 				    enum dma_data_direction direction)
235 {
236 	struct iommu *iommu;
237 	unsigned long flags, npages, oaddr;
238 	unsigned long i, base_paddr;
239 	u32 bus_addr, ret;
240 	unsigned long prot;
241 	long entry;
242 
243 	iommu = dev->archdata.iommu;
244 
245 	if (unlikely(direction == DMA_NONE))
246 		goto bad;
247 
248 	oaddr = (unsigned long)ptr;
249 	npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
250 	npages >>= IO_PAGE_SHIFT;
251 
252 	spin_lock_irqsave(&iommu->lock, flags);
253 	entry = iommu_range_alloc(dev, iommu, npages, NULL);
254 	spin_unlock_irqrestore(&iommu->lock, flags);
255 
256 	if (unlikely(entry == DMA_ERROR_CODE))
257 		goto bad;
258 
259 	bus_addr = (iommu->page_table_map_base +
260 		    (entry << IO_PAGE_SHIFT));
261 	ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
262 	base_paddr = __pa(oaddr & IO_PAGE_MASK);
263 	prot = HV_PCI_MAP_ATTR_READ;
264 	if (direction != DMA_TO_DEVICE)
265 		prot |= HV_PCI_MAP_ATTR_WRITE;
266 
267 	local_irq_save(flags);
268 
269 	iommu_batch_start(dev, prot, entry);
270 
271 	for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
272 		long err = iommu_batch_add(base_paddr);
273 		if (unlikely(err < 0L))
274 			goto iommu_map_fail;
275 	}
276 	if (unlikely(iommu_batch_end() < 0L))
277 		goto iommu_map_fail;
278 
279 	local_irq_restore(flags);
280 
281 	return ret;
282 
283 bad:
284 	if (printk_ratelimit())
285 		WARN_ON(1);
286 	return DMA_ERROR_CODE;
287 
288 iommu_map_fail:
289 	/* Interrupts are disabled.  */
290 	spin_lock(&iommu->lock);
291 	iommu_range_free(iommu, bus_addr, npages);
292 	spin_unlock_irqrestore(&iommu->lock, flags);
293 
294 	return DMA_ERROR_CODE;
295 }
296 
297 static void dma_4v_unmap_single(struct device *dev, dma_addr_t bus_addr,
298 				size_t sz, enum dma_data_direction direction)
299 {
300 	struct pci_pbm_info *pbm;
301 	struct iommu *iommu;
302 	unsigned long flags, npages;
303 	long entry;
304 	u32 devhandle;
305 
306 	if (unlikely(direction == DMA_NONE)) {
307 		if (printk_ratelimit())
308 			WARN_ON(1);
309 		return;
310 	}
311 
312 	iommu = dev->archdata.iommu;
313 	pbm = dev->archdata.host_controller;
314 	devhandle = pbm->devhandle;
315 
316 	npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
317 	npages >>= IO_PAGE_SHIFT;
318 	bus_addr &= IO_PAGE_MASK;
319 
320 	spin_lock_irqsave(&iommu->lock, flags);
321 
322 	iommu_range_free(iommu, bus_addr, npages);
323 
324 	entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
325 	do {
326 		unsigned long num;
327 
328 		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
329 					    npages);
330 		entry += num;
331 		npages -= num;
332 	} while (npages != 0);
333 
334 	spin_unlock_irqrestore(&iommu->lock, flags);
335 }
336 
337 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
338 			 int nelems, enum dma_data_direction direction)
339 {
340 	struct scatterlist *s, *outs, *segstart;
341 	unsigned long flags, handle, prot;
342 	dma_addr_t dma_next = 0, dma_addr;
343 	unsigned int max_seg_size;
344 	unsigned long seg_boundary_size;
345 	int outcount, incount, i;
346 	struct iommu *iommu;
347 	unsigned long base_shift;
348 	long err;
349 
350 	BUG_ON(direction == DMA_NONE);
351 
352 	iommu = dev->archdata.iommu;
353 	if (nelems == 0 || !iommu)
354 		return 0;
355 
356 	prot = HV_PCI_MAP_ATTR_READ;
357 	if (direction != DMA_TO_DEVICE)
358 		prot |= HV_PCI_MAP_ATTR_WRITE;
359 
360 	outs = s = segstart = &sglist[0];
361 	outcount = 1;
362 	incount = nelems;
363 	handle = 0;
364 
365 	/* Init first segment length for backout at failure */
366 	outs->dma_length = 0;
367 
368 	spin_lock_irqsave(&iommu->lock, flags);
369 
370 	iommu_batch_start(dev, prot, ~0UL);
371 
372 	max_seg_size = dma_get_max_seg_size(dev);
373 	seg_boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
374 				  IO_PAGE_SIZE) >> IO_PAGE_SHIFT;
375 	base_shift = iommu->page_table_map_base >> IO_PAGE_SHIFT;
376 	for_each_sg(sglist, s, nelems, i) {
377 		unsigned long paddr, npages, entry, out_entry = 0, slen;
378 
379 		slen = s->length;
380 		/* Sanity check */
381 		if (slen == 0) {
382 			dma_next = 0;
383 			continue;
384 		}
385 		/* Allocate iommu entries for that segment */
386 		paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
387 		npages = iommu_num_pages(paddr, slen, IO_PAGE_SIZE);
388 		entry = iommu_range_alloc(dev, iommu, npages, &handle);
389 
390 		/* Handle failure */
391 		if (unlikely(entry == DMA_ERROR_CODE)) {
392 			if (printk_ratelimit())
393 				printk(KERN_INFO "iommu_alloc failed, iommu %p paddr %lx"
394 				       " npages %lx\n", iommu, paddr, npages);
395 			goto iommu_map_failed;
396 		}
397 
398 		iommu_batch_new_entry(entry);
399 
400 		/* Convert entry to a dma_addr_t */
401 		dma_addr = iommu->page_table_map_base +
402 			(entry << IO_PAGE_SHIFT);
403 		dma_addr |= (s->offset & ~IO_PAGE_MASK);
404 
405 		/* Insert into HW table */
406 		paddr &= IO_PAGE_MASK;
407 		while (npages--) {
408 			err = iommu_batch_add(paddr);
409 			if (unlikely(err < 0L))
410 				goto iommu_map_failed;
411 			paddr += IO_PAGE_SIZE;
412 		}
413 
414 		/* If we are in an open segment, try merging */
415 		if (segstart != s) {
416 			/* We cannot merge if:
417 			 * - allocated dma_addr isn't contiguous to previous allocation
418 			 */
419 			if ((dma_addr != dma_next) ||
420 			    (outs->dma_length + s->length > max_seg_size) ||
421 			    (is_span_boundary(out_entry, base_shift,
422 					      seg_boundary_size, outs, s))) {
423 				/* Can't merge: create a new segment */
424 				segstart = s;
425 				outcount++;
426 				outs = sg_next(outs);
427 			} else {
428 				outs->dma_length += s->length;
429 			}
430 		}
431 
432 		if (segstart == s) {
433 			/* This is a new segment, fill entries */
434 			outs->dma_address = dma_addr;
435 			outs->dma_length = slen;
436 			out_entry = entry;
437 		}
438 
439 		/* Calculate next page pointer for contiguous check */
440 		dma_next = dma_addr + slen;
441 	}
442 
443 	err = iommu_batch_end();
444 
445 	if (unlikely(err < 0L))
446 		goto iommu_map_failed;
447 
448 	spin_unlock_irqrestore(&iommu->lock, flags);
449 
450 	if (outcount < incount) {
451 		outs = sg_next(outs);
452 		outs->dma_address = DMA_ERROR_CODE;
453 		outs->dma_length = 0;
454 	}
455 
456 	return outcount;
457 
458 iommu_map_failed:
459 	for_each_sg(sglist, s, nelems, i) {
460 		if (s->dma_length != 0) {
461 			unsigned long vaddr, npages;
462 
463 			vaddr = s->dma_address & IO_PAGE_MASK;
464 			npages = iommu_num_pages(s->dma_address, s->dma_length,
465 						 IO_PAGE_SIZE);
466 			iommu_range_free(iommu, vaddr, npages);
467 			/* XXX demap? XXX */
468 			s->dma_address = DMA_ERROR_CODE;
469 			s->dma_length = 0;
470 		}
471 		if (s == outs)
472 			break;
473 	}
474 	spin_unlock_irqrestore(&iommu->lock, flags);
475 
476 	return 0;
477 }
478 
479 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
480 			    int nelems, enum dma_data_direction direction)
481 {
482 	struct pci_pbm_info *pbm;
483 	struct scatterlist *sg;
484 	struct iommu *iommu;
485 	unsigned long flags;
486 	u32 devhandle;
487 
488 	BUG_ON(direction == DMA_NONE);
489 
490 	iommu = dev->archdata.iommu;
491 	pbm = dev->archdata.host_controller;
492 	devhandle = pbm->devhandle;
493 
494 	spin_lock_irqsave(&iommu->lock, flags);
495 
496 	sg = sglist;
497 	while (nelems--) {
498 		dma_addr_t dma_handle = sg->dma_address;
499 		unsigned int len = sg->dma_length;
500 		unsigned long npages, entry;
501 
502 		if (!len)
503 			break;
504 		npages = iommu_num_pages(dma_handle, len, IO_PAGE_SIZE);
505 		iommu_range_free(iommu, dma_handle, npages);
506 
507 		entry = ((dma_handle - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
508 		while (npages) {
509 			unsigned long num;
510 
511 			num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
512 						    npages);
513 			entry += num;
514 			npages -= num;
515 		}
516 
517 		sg = sg_next(sg);
518 	}
519 
520 	spin_unlock_irqrestore(&iommu->lock, flags);
521 }
522 
523 static void dma_4v_sync_single_for_cpu(struct device *dev,
524 				       dma_addr_t bus_addr, size_t sz,
525 				       enum dma_data_direction direction)
526 {
527 	/* Nothing to do... */
528 }
529 
530 static void dma_4v_sync_sg_for_cpu(struct device *dev,
531 				   struct scatterlist *sglist, int nelems,
532 				   enum dma_data_direction direction)
533 {
534 	/* Nothing to do... */
535 }
536 
537 static const struct dma_ops sun4v_dma_ops = {
538 	.alloc_coherent			= dma_4v_alloc_coherent,
539 	.free_coherent			= dma_4v_free_coherent,
540 	.map_single			= dma_4v_map_single,
541 	.unmap_single			= dma_4v_unmap_single,
542 	.map_sg				= dma_4v_map_sg,
543 	.unmap_sg			= dma_4v_unmap_sg,
544 	.sync_single_for_cpu		= dma_4v_sync_single_for_cpu,
545 	.sync_sg_for_cpu		= dma_4v_sync_sg_for_cpu,
546 };
547 
548 static void __init pci_sun4v_scan_bus(struct pci_pbm_info *pbm,
549 				      struct device *parent)
550 {
551 	struct property *prop;
552 	struct device_node *dp;
553 
554 	dp = pbm->op->node;
555 	prop = of_find_property(dp, "66mhz-capable", NULL);
556 	pbm->is_66mhz_capable = (prop != NULL);
557 	pbm->pci_bus = pci_scan_one_pbm(pbm, parent);
558 
559 	/* XXX register error interrupt handlers XXX */
560 }
561 
562 static unsigned long __init probe_existing_entries(struct pci_pbm_info *pbm,
563 						   struct iommu *iommu)
564 {
565 	struct iommu_arena *arena = &iommu->arena;
566 	unsigned long i, cnt = 0;
567 	u32 devhandle;
568 
569 	devhandle = pbm->devhandle;
570 	for (i = 0; i < arena->limit; i++) {
571 		unsigned long ret, io_attrs, ra;
572 
573 		ret = pci_sun4v_iommu_getmap(devhandle,
574 					     HV_PCI_TSBID(0, i),
575 					     &io_attrs, &ra);
576 		if (ret == HV_EOK) {
577 			if (page_in_phys_avail(ra)) {
578 				pci_sun4v_iommu_demap(devhandle,
579 						      HV_PCI_TSBID(0, i), 1);
580 			} else {
581 				cnt++;
582 				__set_bit(i, arena->map);
583 			}
584 		}
585 	}
586 
587 	return cnt;
588 }
589 
590 static int __init pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
591 {
592 	static const u32 vdma_default[] = { 0x80000000, 0x80000000 };
593 	struct iommu *iommu = pbm->iommu;
594 	unsigned long num_tsb_entries, sz, tsbsize;
595 	u32 dma_mask, dma_offset;
596 	const u32 *vdma;
597 
598 	vdma = of_get_property(pbm->op->node, "virtual-dma", NULL);
599 	if (!vdma)
600 		vdma = vdma_default;
601 
602 	if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) {
603 		printk(KERN_ERR PFX "Strange virtual-dma[%08x:%08x].\n",
604 		       vdma[0], vdma[1]);
605 		return -EINVAL;
606 	};
607 
608 	dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
609 	num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
610 	tsbsize = num_tsb_entries * sizeof(iopte_t);
611 
612 	dma_offset = vdma[0];
613 
614 	/* Setup initial software IOMMU state. */
615 	spin_lock_init(&iommu->lock);
616 	iommu->ctx_lowest_free = 1;
617 	iommu->page_table_map_base = dma_offset;
618 	iommu->dma_addr_mask = dma_mask;
619 
620 	/* Allocate and initialize the free area map.  */
621 	sz = (num_tsb_entries + 7) / 8;
622 	sz = (sz + 7UL) & ~7UL;
623 	iommu->arena.map = kzalloc(sz, GFP_KERNEL);
624 	if (!iommu->arena.map) {
625 		printk(KERN_ERR PFX "Error, kmalloc(arena.map) failed.\n");
626 		return -ENOMEM;
627 	}
628 	iommu->arena.limit = num_tsb_entries;
629 
630 	sz = probe_existing_entries(pbm, iommu);
631 	if (sz)
632 		printk("%s: Imported %lu TSB entries from OBP\n",
633 		       pbm->name, sz);
634 
635 	return 0;
636 }
637 
638 #ifdef CONFIG_PCI_MSI
639 struct pci_sun4v_msiq_entry {
640 	u64		version_type;
641 #define MSIQ_VERSION_MASK		0xffffffff00000000UL
642 #define MSIQ_VERSION_SHIFT		32
643 #define MSIQ_TYPE_MASK			0x00000000000000ffUL
644 #define MSIQ_TYPE_SHIFT			0
645 #define MSIQ_TYPE_NONE			0x00
646 #define MSIQ_TYPE_MSG			0x01
647 #define MSIQ_TYPE_MSI32			0x02
648 #define MSIQ_TYPE_MSI64			0x03
649 #define MSIQ_TYPE_INTX			0x08
650 #define MSIQ_TYPE_NONE2			0xff
651 
652 	u64		intx_sysino;
653 	u64		reserved1;
654 	u64		stick;
655 	u64		req_id;  /* bus/device/func */
656 #define MSIQ_REQID_BUS_MASK		0xff00UL
657 #define MSIQ_REQID_BUS_SHIFT		8
658 #define MSIQ_REQID_DEVICE_MASK		0x00f8UL
659 #define MSIQ_REQID_DEVICE_SHIFT		3
660 #define MSIQ_REQID_FUNC_MASK		0x0007UL
661 #define MSIQ_REQID_FUNC_SHIFT		0
662 
663 	u64		msi_address;
664 
665 	/* The format of this value is message type dependent.
666 	 * For MSI bits 15:0 are the data from the MSI packet.
667 	 * For MSI-X bits 31:0 are the data from the MSI packet.
668 	 * For MSG, the message code and message routing code where:
669 	 * 	bits 39:32 is the bus/device/fn of the msg target-id
670 	 *	bits 18:16 is the message routing code
671 	 *	bits 7:0 is the message code
672 	 * For INTx the low order 2-bits are:
673 	 *	00 - INTA
674 	 *	01 - INTB
675 	 *	10 - INTC
676 	 *	11 - INTD
677 	 */
678 	u64		msi_data;
679 
680 	u64		reserved2;
681 };
682 
683 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
684 			      unsigned long *head)
685 {
686 	unsigned long err, limit;
687 
688 	err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
689 	if (unlikely(err))
690 		return -ENXIO;
691 
692 	limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
693 	if (unlikely(*head >= limit))
694 		return -EFBIG;
695 
696 	return 0;
697 }
698 
699 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
700 				 unsigned long msiqid, unsigned long *head,
701 				 unsigned long *msi)
702 {
703 	struct pci_sun4v_msiq_entry *ep;
704 	unsigned long err, type;
705 
706 	/* Note: void pointer arithmetic, 'head' is a byte offset  */
707 	ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
708 				 (pbm->msiq_ent_count *
709 				  sizeof(struct pci_sun4v_msiq_entry))) +
710 	      *head);
711 
712 	if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
713 		return 0;
714 
715 	type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
716 	if (unlikely(type != MSIQ_TYPE_MSI32 &&
717 		     type != MSIQ_TYPE_MSI64))
718 		return -EINVAL;
719 
720 	*msi = ep->msi_data;
721 
722 	err = pci_sun4v_msi_setstate(pbm->devhandle,
723 				     ep->msi_data /* msi_num */,
724 				     HV_MSISTATE_IDLE);
725 	if (unlikely(err))
726 		return -ENXIO;
727 
728 	/* Clear the entry.  */
729 	ep->version_type &= ~MSIQ_TYPE_MASK;
730 
731 	(*head) += sizeof(struct pci_sun4v_msiq_entry);
732 	if (*head >=
733 	    (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
734 		*head = 0;
735 
736 	return 1;
737 }
738 
739 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
740 			      unsigned long head)
741 {
742 	unsigned long err;
743 
744 	err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
745 	if (unlikely(err))
746 		return -EINVAL;
747 
748 	return 0;
749 }
750 
751 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
752 			       unsigned long msi, int is_msi64)
753 {
754 	if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
755 				  (is_msi64 ?
756 				   HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
757 		return -ENXIO;
758 	if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
759 		return -ENXIO;
760 	if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
761 		return -ENXIO;
762 	return 0;
763 }
764 
765 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
766 {
767 	unsigned long err, msiqid;
768 
769 	err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
770 	if (err)
771 		return -ENXIO;
772 
773 	pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
774 
775 	return 0;
776 }
777 
778 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
779 {
780 	unsigned long q_size, alloc_size, pages, order;
781 	int i;
782 
783 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
784 	alloc_size = (pbm->msiq_num * q_size);
785 	order = get_order(alloc_size);
786 	pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
787 	if (pages == 0UL) {
788 		printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
789 		       order);
790 		return -ENOMEM;
791 	}
792 	memset((char *)pages, 0, PAGE_SIZE << order);
793 	pbm->msi_queues = (void *) pages;
794 
795 	for (i = 0; i < pbm->msiq_num; i++) {
796 		unsigned long err, base = __pa(pages + (i * q_size));
797 		unsigned long ret1, ret2;
798 
799 		err = pci_sun4v_msiq_conf(pbm->devhandle,
800 					  pbm->msiq_first + i,
801 					  base, pbm->msiq_ent_count);
802 		if (err) {
803 			printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
804 			       err);
805 			goto h_error;
806 		}
807 
808 		err = pci_sun4v_msiq_info(pbm->devhandle,
809 					  pbm->msiq_first + i,
810 					  &ret1, &ret2);
811 		if (err) {
812 			printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
813 			       err);
814 			goto h_error;
815 		}
816 		if (ret1 != base || ret2 != pbm->msiq_ent_count) {
817 			printk(KERN_ERR "MSI: Bogus qconf "
818 			       "expected[%lx:%x] got[%lx:%lx]\n",
819 			       base, pbm->msiq_ent_count,
820 			       ret1, ret2);
821 			goto h_error;
822 		}
823 	}
824 
825 	return 0;
826 
827 h_error:
828 	free_pages(pages, order);
829 	return -EINVAL;
830 }
831 
832 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
833 {
834 	unsigned long q_size, alloc_size, pages, order;
835 	int i;
836 
837 	for (i = 0; i < pbm->msiq_num; i++) {
838 		unsigned long msiqid = pbm->msiq_first + i;
839 
840 		(void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
841 	}
842 
843 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
844 	alloc_size = (pbm->msiq_num * q_size);
845 	order = get_order(alloc_size);
846 
847 	pages = (unsigned long) pbm->msi_queues;
848 
849 	free_pages(pages, order);
850 
851 	pbm->msi_queues = NULL;
852 }
853 
854 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
855 				    unsigned long msiqid,
856 				    unsigned long devino)
857 {
858 	unsigned int virt_irq = sun4v_build_irq(pbm->devhandle, devino);
859 
860 	if (!virt_irq)
861 		return -ENOMEM;
862 
863 	if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
864 		return -EINVAL;
865 	if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
866 		return -EINVAL;
867 
868 	return virt_irq;
869 }
870 
871 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
872 	.get_head	=	pci_sun4v_get_head,
873 	.dequeue_msi	=	pci_sun4v_dequeue_msi,
874 	.set_head	=	pci_sun4v_set_head,
875 	.msi_setup	=	pci_sun4v_msi_setup,
876 	.msi_teardown	=	pci_sun4v_msi_teardown,
877 	.msiq_alloc	=	pci_sun4v_msiq_alloc,
878 	.msiq_free	=	pci_sun4v_msiq_free,
879 	.msiq_build_irq	=	pci_sun4v_msiq_build_irq,
880 };
881 
882 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
883 {
884 	sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
885 }
886 #else /* CONFIG_PCI_MSI */
887 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
888 {
889 }
890 #endif /* !(CONFIG_PCI_MSI) */
891 
892 static int __init pci_sun4v_pbm_init(struct pci_pbm_info *pbm,
893 				     struct of_device *op, u32 devhandle)
894 {
895 	struct device_node *dp = op->node;
896 	int err;
897 
898 	pbm->numa_node = of_node_to_nid(dp);
899 
900 	pbm->pci_ops = &sun4v_pci_ops;
901 	pbm->config_space_reg_bits = 12;
902 
903 	pbm->index = pci_num_pbms++;
904 
905 	pbm->op = op;
906 
907 	pbm->devhandle = devhandle;
908 
909 	pbm->name = dp->full_name;
910 
911 	printk("%s: SUN4V PCI Bus Module\n", pbm->name);
912 	printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node);
913 
914 	pci_determine_mem_io_space(pbm);
915 
916 	pci_get_pbm_props(pbm);
917 
918 	err = pci_sun4v_iommu_init(pbm);
919 	if (err)
920 		return err;
921 
922 	pci_sun4v_msi_init(pbm);
923 
924 	pci_sun4v_scan_bus(pbm, &op->dev);
925 
926 	pbm->next = pci_pbm_root;
927 	pci_pbm_root = pbm;
928 
929 	return 0;
930 }
931 
932 static int __devinit pci_sun4v_probe(struct of_device *op,
933 				     const struct of_device_id *match)
934 {
935 	const struct linux_prom64_registers *regs;
936 	static int hvapi_negotiated = 0;
937 	struct pci_pbm_info *pbm;
938 	struct device_node *dp;
939 	struct iommu *iommu;
940 	u32 devhandle;
941 	int i, err;
942 
943 	dp = op->node;
944 
945 	if (!hvapi_negotiated++) {
946 		err = sun4v_hvapi_register(HV_GRP_PCI,
947 					   vpci_major,
948 					   &vpci_minor);
949 
950 		if (err) {
951 			printk(KERN_ERR PFX "Could not register hvapi, "
952 			       "err=%d\n", err);
953 			return err;
954 		}
955 		printk(KERN_INFO PFX "Registered hvapi major[%lu] minor[%lu]\n",
956 		       vpci_major, vpci_minor);
957 
958 		dma_ops = &sun4v_dma_ops;
959 	}
960 
961 	regs = of_get_property(dp, "reg", NULL);
962 	err = -ENODEV;
963 	if (!regs) {
964 		printk(KERN_ERR PFX "Could not find config registers\n");
965 		goto out_err;
966 	}
967 	devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
968 
969 	err = -ENOMEM;
970 	if (!iommu_batch_initialized) {
971 		for_each_possible_cpu(i) {
972 			unsigned long page = get_zeroed_page(GFP_KERNEL);
973 
974 			if (!page)
975 				goto out_err;
976 
977 			per_cpu(iommu_batch, i).pglist = (u64 *) page;
978 		}
979 		iommu_batch_initialized = 1;
980 	}
981 
982 	pbm = kzalloc(sizeof(*pbm), GFP_KERNEL);
983 	if (!pbm) {
984 		printk(KERN_ERR PFX "Could not allocate pci_pbm_info\n");
985 		goto out_err;
986 	}
987 
988 	iommu = kzalloc(sizeof(struct iommu), GFP_KERNEL);
989 	if (!iommu) {
990 		printk(KERN_ERR PFX "Could not allocate pbm iommu\n");
991 		goto out_free_controller;
992 	}
993 
994 	pbm->iommu = iommu;
995 
996 	err = pci_sun4v_pbm_init(pbm, op, devhandle);
997 	if (err)
998 		goto out_free_iommu;
999 
1000 	dev_set_drvdata(&op->dev, pbm);
1001 
1002 	return 0;
1003 
1004 out_free_iommu:
1005 	kfree(pbm->iommu);
1006 
1007 out_free_controller:
1008 	kfree(pbm);
1009 
1010 out_err:
1011 	return err;
1012 }
1013 
1014 static struct of_device_id __initdata pci_sun4v_match[] = {
1015 	{
1016 		.name = "pci",
1017 		.compatible = "SUNW,sun4v-pci",
1018 	},
1019 	{},
1020 };
1021 
1022 static struct of_platform_driver pci_sun4v_driver = {
1023 	.name		= DRIVER_NAME,
1024 	.match_table	= pci_sun4v_match,
1025 	.probe		= pci_sun4v_probe,
1026 };
1027 
1028 static int __init pci_sun4v_init(void)
1029 {
1030 	return of_register_driver(&pci_sun4v_driver, &of_bus_type);
1031 }
1032 
1033 subsys_initcall(pci_sun4v_init);
1034