xref: /openbmc/linux/arch/sparc/kernel/pci_sun4v.c (revision 63dc02bd)
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/export.h>
16 #include <linux/log2.h>
17 #include <linux/of_device.h>
18 
19 #include <asm/iommu.h>
20 #include <asm/irq.h>
21 #include <asm/hypervisor.h>
22 #include <asm/prom.h>
23 
24 #include "pci_impl.h"
25 #include "iommu_common.h"
26 
27 #include "pci_sun4v.h"
28 
29 #define DRIVER_NAME	"pci_sun4v"
30 #define PFX		DRIVER_NAME ": "
31 
32 static unsigned long vpci_major = 1;
33 static unsigned long vpci_minor = 1;
34 
35 #define PGLIST_NENTS	(PAGE_SIZE / sizeof(u64))
36 
37 struct iommu_batch {
38 	struct device	*dev;		/* Device mapping is for.	*/
39 	unsigned long	prot;		/* IOMMU page protections	*/
40 	unsigned long	entry;		/* Index into IOTSB.		*/
41 	u64		*pglist;	/* List of physical pages	*/
42 	unsigned long	npages;		/* Number of pages in list.	*/
43 };
44 
45 static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
46 static int iommu_batch_initialized;
47 
48 /* Interrupts must be disabled.  */
49 static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
50 {
51 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
52 
53 	p->dev		= dev;
54 	p->prot		= prot;
55 	p->entry	= entry;
56 	p->npages	= 0;
57 }
58 
59 /* Interrupts must be disabled.  */
60 static long iommu_batch_flush(struct iommu_batch *p)
61 {
62 	struct pci_pbm_info *pbm = p->dev->archdata.host_controller;
63 	unsigned long devhandle = pbm->devhandle;
64 	unsigned long prot = p->prot;
65 	unsigned long entry = p->entry;
66 	u64 *pglist = p->pglist;
67 	unsigned long npages = p->npages;
68 
69 	while (npages != 0) {
70 		long num;
71 
72 		num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
73 					  npages, prot, __pa(pglist));
74 		if (unlikely(num < 0)) {
75 			if (printk_ratelimit())
76 				printk("iommu_batch_flush: IOMMU map of "
77 				       "[%08lx:%08llx:%lx:%lx:%lx] failed with "
78 				       "status %ld\n",
79 				       devhandle, HV_PCI_TSBID(0, entry),
80 				       npages, prot, __pa(pglist), num);
81 			return -1;
82 		}
83 
84 		entry += num;
85 		npages -= num;
86 		pglist += num;
87 	}
88 
89 	p->entry = entry;
90 	p->npages = 0;
91 
92 	return 0;
93 }
94 
95 static inline void iommu_batch_new_entry(unsigned long entry)
96 {
97 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
98 
99 	if (p->entry + p->npages == entry)
100 		return;
101 	if (p->entry != ~0UL)
102 		iommu_batch_flush(p);
103 	p->entry = entry;
104 }
105 
106 /* Interrupts must be disabled.  */
107 static inline long iommu_batch_add(u64 phys_page)
108 {
109 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
110 
111 	BUG_ON(p->npages >= PGLIST_NENTS);
112 
113 	p->pglist[p->npages++] = phys_page;
114 	if (p->npages == PGLIST_NENTS)
115 		return iommu_batch_flush(p);
116 
117 	return 0;
118 }
119 
120 /* Interrupts must be disabled.  */
121 static inline long iommu_batch_end(void)
122 {
123 	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
124 
125 	BUG_ON(p->npages >= PGLIST_NENTS);
126 
127 	return iommu_batch_flush(p);
128 }
129 
130 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
131 				   dma_addr_t *dma_addrp, gfp_t gfp,
132 				   struct dma_attrs *attrs)
133 {
134 	unsigned long flags, order, first_page, npages, n;
135 	struct iommu *iommu;
136 	struct page *page;
137 	void *ret;
138 	long entry;
139 	int nid;
140 
141 	size = IO_PAGE_ALIGN(size);
142 	order = get_order(size);
143 	if (unlikely(order >= MAX_ORDER))
144 		return NULL;
145 
146 	npages = size >> IO_PAGE_SHIFT;
147 
148 	nid = dev->archdata.numa_node;
149 	page = alloc_pages_node(nid, gfp, order);
150 	if (unlikely(!page))
151 		return NULL;
152 
153 	first_page = (unsigned long) page_address(page);
154 	memset((char *)first_page, 0, PAGE_SIZE << order);
155 
156 	iommu = dev->archdata.iommu;
157 
158 	spin_lock_irqsave(&iommu->lock, flags);
159 	entry = iommu_range_alloc(dev, iommu, npages, NULL);
160 	spin_unlock_irqrestore(&iommu->lock, flags);
161 
162 	if (unlikely(entry == DMA_ERROR_CODE))
163 		goto range_alloc_fail;
164 
165 	*dma_addrp = (iommu->page_table_map_base +
166 		      (entry << IO_PAGE_SHIFT));
167 	ret = (void *) first_page;
168 	first_page = __pa(first_page);
169 
170 	local_irq_save(flags);
171 
172 	iommu_batch_start(dev,
173 			  (HV_PCI_MAP_ATTR_READ |
174 			   HV_PCI_MAP_ATTR_WRITE),
175 			  entry);
176 
177 	for (n = 0; n < npages; n++) {
178 		long err = iommu_batch_add(first_page + (n * PAGE_SIZE));
179 		if (unlikely(err < 0L))
180 			goto iommu_map_fail;
181 	}
182 
183 	if (unlikely(iommu_batch_end() < 0L))
184 		goto iommu_map_fail;
185 
186 	local_irq_restore(flags);
187 
188 	return ret;
189 
190 iommu_map_fail:
191 	/* Interrupts are disabled.  */
192 	spin_lock(&iommu->lock);
193 	iommu_range_free(iommu, *dma_addrp, npages);
194 	spin_unlock_irqrestore(&iommu->lock, flags);
195 
196 range_alloc_fail:
197 	free_pages(first_page, order);
198 	return NULL;
199 }
200 
201 static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
202 				 dma_addr_t dvma, struct dma_attrs *attrs)
203 {
204 	struct pci_pbm_info *pbm;
205 	struct iommu *iommu;
206 	unsigned long flags, order, npages, entry;
207 	u32 devhandle;
208 
209 	npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
210 	iommu = dev->archdata.iommu;
211 	pbm = dev->archdata.host_controller;
212 	devhandle = pbm->devhandle;
213 	entry = ((dvma - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
214 
215 	spin_lock_irqsave(&iommu->lock, flags);
216 
217 	iommu_range_free(iommu, dvma, npages);
218 
219 	do {
220 		unsigned long num;
221 
222 		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
223 					    npages);
224 		entry += num;
225 		npages -= num;
226 	} while (npages != 0);
227 
228 	spin_unlock_irqrestore(&iommu->lock, flags);
229 
230 	order = get_order(size);
231 	if (order < 10)
232 		free_pages((unsigned long)cpu, order);
233 }
234 
235 static dma_addr_t dma_4v_map_page(struct device *dev, struct page *page,
236 				  unsigned long offset, size_t sz,
237 				  enum dma_data_direction direction,
238 				  struct dma_attrs *attrs)
239 {
240 	struct iommu *iommu;
241 	unsigned long flags, npages, oaddr;
242 	unsigned long i, base_paddr;
243 	u32 bus_addr, ret;
244 	unsigned long prot;
245 	long entry;
246 
247 	iommu = dev->archdata.iommu;
248 
249 	if (unlikely(direction == DMA_NONE))
250 		goto bad;
251 
252 	oaddr = (unsigned long)(page_address(page) + offset);
253 	npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
254 	npages >>= IO_PAGE_SHIFT;
255 
256 	spin_lock_irqsave(&iommu->lock, flags);
257 	entry = iommu_range_alloc(dev, iommu, npages, NULL);
258 	spin_unlock_irqrestore(&iommu->lock, flags);
259 
260 	if (unlikely(entry == DMA_ERROR_CODE))
261 		goto bad;
262 
263 	bus_addr = (iommu->page_table_map_base +
264 		    (entry << IO_PAGE_SHIFT));
265 	ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
266 	base_paddr = __pa(oaddr & IO_PAGE_MASK);
267 	prot = HV_PCI_MAP_ATTR_READ;
268 	if (direction != DMA_TO_DEVICE)
269 		prot |= HV_PCI_MAP_ATTR_WRITE;
270 
271 	local_irq_save(flags);
272 
273 	iommu_batch_start(dev, prot, entry);
274 
275 	for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
276 		long err = iommu_batch_add(base_paddr);
277 		if (unlikely(err < 0L))
278 			goto iommu_map_fail;
279 	}
280 	if (unlikely(iommu_batch_end() < 0L))
281 		goto iommu_map_fail;
282 
283 	local_irq_restore(flags);
284 
285 	return ret;
286 
287 bad:
288 	if (printk_ratelimit())
289 		WARN_ON(1);
290 	return DMA_ERROR_CODE;
291 
292 iommu_map_fail:
293 	/* Interrupts are disabled.  */
294 	spin_lock(&iommu->lock);
295 	iommu_range_free(iommu, bus_addr, npages);
296 	spin_unlock_irqrestore(&iommu->lock, flags);
297 
298 	return DMA_ERROR_CODE;
299 }
300 
301 static void dma_4v_unmap_page(struct device *dev, dma_addr_t bus_addr,
302 			      size_t sz, enum dma_data_direction direction,
303 			      struct dma_attrs *attrs)
304 {
305 	struct pci_pbm_info *pbm;
306 	struct iommu *iommu;
307 	unsigned long flags, npages;
308 	long entry;
309 	u32 devhandle;
310 
311 	if (unlikely(direction == DMA_NONE)) {
312 		if (printk_ratelimit())
313 			WARN_ON(1);
314 		return;
315 	}
316 
317 	iommu = dev->archdata.iommu;
318 	pbm = dev->archdata.host_controller;
319 	devhandle = pbm->devhandle;
320 
321 	npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
322 	npages >>= IO_PAGE_SHIFT;
323 	bus_addr &= IO_PAGE_MASK;
324 
325 	spin_lock_irqsave(&iommu->lock, flags);
326 
327 	iommu_range_free(iommu, bus_addr, npages);
328 
329 	entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
330 	do {
331 		unsigned long num;
332 
333 		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
334 					    npages);
335 		entry += num;
336 		npages -= num;
337 	} while (npages != 0);
338 
339 	spin_unlock_irqrestore(&iommu->lock, flags);
340 }
341 
342 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
343 			 int nelems, enum dma_data_direction direction,
344 			 struct dma_attrs *attrs)
345 {
346 	struct scatterlist *s, *outs, *segstart;
347 	unsigned long flags, handle, prot;
348 	dma_addr_t dma_next = 0, dma_addr;
349 	unsigned int max_seg_size;
350 	unsigned long seg_boundary_size;
351 	int outcount, incount, i;
352 	struct iommu *iommu;
353 	unsigned long base_shift;
354 	long err;
355 
356 	BUG_ON(direction == DMA_NONE);
357 
358 	iommu = dev->archdata.iommu;
359 	if (nelems == 0 || !iommu)
360 		return 0;
361 
362 	prot = HV_PCI_MAP_ATTR_READ;
363 	if (direction != DMA_TO_DEVICE)
364 		prot |= HV_PCI_MAP_ATTR_WRITE;
365 
366 	outs = s = segstart = &sglist[0];
367 	outcount = 1;
368 	incount = nelems;
369 	handle = 0;
370 
371 	/* Init first segment length for backout at failure */
372 	outs->dma_length = 0;
373 
374 	spin_lock_irqsave(&iommu->lock, flags);
375 
376 	iommu_batch_start(dev, prot, ~0UL);
377 
378 	max_seg_size = dma_get_max_seg_size(dev);
379 	seg_boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
380 				  IO_PAGE_SIZE) >> IO_PAGE_SHIFT;
381 	base_shift = iommu->page_table_map_base >> IO_PAGE_SHIFT;
382 	for_each_sg(sglist, s, nelems, i) {
383 		unsigned long paddr, npages, entry, out_entry = 0, slen;
384 
385 		slen = s->length;
386 		/* Sanity check */
387 		if (slen == 0) {
388 			dma_next = 0;
389 			continue;
390 		}
391 		/* Allocate iommu entries for that segment */
392 		paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
393 		npages = iommu_num_pages(paddr, slen, IO_PAGE_SIZE);
394 		entry = iommu_range_alloc(dev, iommu, npages, &handle);
395 
396 		/* Handle failure */
397 		if (unlikely(entry == DMA_ERROR_CODE)) {
398 			if (printk_ratelimit())
399 				printk(KERN_INFO "iommu_alloc failed, iommu %p paddr %lx"
400 				       " npages %lx\n", iommu, paddr, npages);
401 			goto iommu_map_failed;
402 		}
403 
404 		iommu_batch_new_entry(entry);
405 
406 		/* Convert entry to a dma_addr_t */
407 		dma_addr = iommu->page_table_map_base +
408 			(entry << IO_PAGE_SHIFT);
409 		dma_addr |= (s->offset & ~IO_PAGE_MASK);
410 
411 		/* Insert into HW table */
412 		paddr &= IO_PAGE_MASK;
413 		while (npages--) {
414 			err = iommu_batch_add(paddr);
415 			if (unlikely(err < 0L))
416 				goto iommu_map_failed;
417 			paddr += IO_PAGE_SIZE;
418 		}
419 
420 		/* If we are in an open segment, try merging */
421 		if (segstart != s) {
422 			/* We cannot merge if:
423 			 * - allocated dma_addr isn't contiguous to previous allocation
424 			 */
425 			if ((dma_addr != dma_next) ||
426 			    (outs->dma_length + s->length > max_seg_size) ||
427 			    (is_span_boundary(out_entry, base_shift,
428 					      seg_boundary_size, outs, s))) {
429 				/* Can't merge: create a new segment */
430 				segstart = s;
431 				outcount++;
432 				outs = sg_next(outs);
433 			} else {
434 				outs->dma_length += s->length;
435 			}
436 		}
437 
438 		if (segstart == s) {
439 			/* This is a new segment, fill entries */
440 			outs->dma_address = dma_addr;
441 			outs->dma_length = slen;
442 			out_entry = entry;
443 		}
444 
445 		/* Calculate next page pointer for contiguous check */
446 		dma_next = dma_addr + slen;
447 	}
448 
449 	err = iommu_batch_end();
450 
451 	if (unlikely(err < 0L))
452 		goto iommu_map_failed;
453 
454 	spin_unlock_irqrestore(&iommu->lock, flags);
455 
456 	if (outcount < incount) {
457 		outs = sg_next(outs);
458 		outs->dma_address = DMA_ERROR_CODE;
459 		outs->dma_length = 0;
460 	}
461 
462 	return outcount;
463 
464 iommu_map_failed:
465 	for_each_sg(sglist, s, nelems, i) {
466 		if (s->dma_length != 0) {
467 			unsigned long vaddr, npages;
468 
469 			vaddr = s->dma_address & IO_PAGE_MASK;
470 			npages = iommu_num_pages(s->dma_address, s->dma_length,
471 						 IO_PAGE_SIZE);
472 			iommu_range_free(iommu, vaddr, npages);
473 			/* XXX demap? XXX */
474 			s->dma_address = DMA_ERROR_CODE;
475 			s->dma_length = 0;
476 		}
477 		if (s == outs)
478 			break;
479 	}
480 	spin_unlock_irqrestore(&iommu->lock, flags);
481 
482 	return 0;
483 }
484 
485 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
486 			    int nelems, enum dma_data_direction direction,
487 			    struct dma_attrs *attrs)
488 {
489 	struct pci_pbm_info *pbm;
490 	struct scatterlist *sg;
491 	struct iommu *iommu;
492 	unsigned long flags;
493 	u32 devhandle;
494 
495 	BUG_ON(direction == DMA_NONE);
496 
497 	iommu = dev->archdata.iommu;
498 	pbm = dev->archdata.host_controller;
499 	devhandle = pbm->devhandle;
500 
501 	spin_lock_irqsave(&iommu->lock, flags);
502 
503 	sg = sglist;
504 	while (nelems--) {
505 		dma_addr_t dma_handle = sg->dma_address;
506 		unsigned int len = sg->dma_length;
507 		unsigned long npages, entry;
508 
509 		if (!len)
510 			break;
511 		npages = iommu_num_pages(dma_handle, len, IO_PAGE_SIZE);
512 		iommu_range_free(iommu, dma_handle, npages);
513 
514 		entry = ((dma_handle - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
515 		while (npages) {
516 			unsigned long num;
517 
518 			num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
519 						    npages);
520 			entry += num;
521 			npages -= num;
522 		}
523 
524 		sg = sg_next(sg);
525 	}
526 
527 	spin_unlock_irqrestore(&iommu->lock, flags);
528 }
529 
530 static struct dma_map_ops sun4v_dma_ops = {
531 	.alloc				= dma_4v_alloc_coherent,
532 	.free				= dma_4v_free_coherent,
533 	.map_page			= dma_4v_map_page,
534 	.unmap_page			= dma_4v_unmap_page,
535 	.map_sg				= dma_4v_map_sg,
536 	.unmap_sg			= dma_4v_unmap_sg,
537 };
538 
539 static void __devinit pci_sun4v_scan_bus(struct pci_pbm_info *pbm,
540 					 struct device *parent)
541 {
542 	struct property *prop;
543 	struct device_node *dp;
544 
545 	dp = pbm->op->dev.of_node;
546 	prop = of_find_property(dp, "66mhz-capable", NULL);
547 	pbm->is_66mhz_capable = (prop != NULL);
548 	pbm->pci_bus = pci_scan_one_pbm(pbm, parent);
549 
550 	/* XXX register error interrupt handlers XXX */
551 }
552 
553 static unsigned long __devinit probe_existing_entries(struct pci_pbm_info *pbm,
554 						      struct iommu *iommu)
555 {
556 	struct iommu_arena *arena = &iommu->arena;
557 	unsigned long i, cnt = 0;
558 	u32 devhandle;
559 
560 	devhandle = pbm->devhandle;
561 	for (i = 0; i < arena->limit; i++) {
562 		unsigned long ret, io_attrs, ra;
563 
564 		ret = pci_sun4v_iommu_getmap(devhandle,
565 					     HV_PCI_TSBID(0, i),
566 					     &io_attrs, &ra);
567 		if (ret == HV_EOK) {
568 			if (page_in_phys_avail(ra)) {
569 				pci_sun4v_iommu_demap(devhandle,
570 						      HV_PCI_TSBID(0, i), 1);
571 			} else {
572 				cnt++;
573 				__set_bit(i, arena->map);
574 			}
575 		}
576 	}
577 
578 	return cnt;
579 }
580 
581 static int __devinit pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
582 {
583 	static const u32 vdma_default[] = { 0x80000000, 0x80000000 };
584 	struct iommu *iommu = pbm->iommu;
585 	unsigned long num_tsb_entries, sz;
586 	u32 dma_mask, dma_offset;
587 	const u32 *vdma;
588 
589 	vdma = of_get_property(pbm->op->dev.of_node, "virtual-dma", NULL);
590 	if (!vdma)
591 		vdma = vdma_default;
592 
593 	if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) {
594 		printk(KERN_ERR PFX "Strange virtual-dma[%08x:%08x].\n",
595 		       vdma[0], vdma[1]);
596 		return -EINVAL;
597 	};
598 
599 	dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
600 	num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
601 
602 	dma_offset = vdma[0];
603 
604 	/* Setup initial software IOMMU state. */
605 	spin_lock_init(&iommu->lock);
606 	iommu->ctx_lowest_free = 1;
607 	iommu->page_table_map_base = dma_offset;
608 	iommu->dma_addr_mask = dma_mask;
609 
610 	/* Allocate and initialize the free area map.  */
611 	sz = (num_tsb_entries + 7) / 8;
612 	sz = (sz + 7UL) & ~7UL;
613 	iommu->arena.map = kzalloc(sz, GFP_KERNEL);
614 	if (!iommu->arena.map) {
615 		printk(KERN_ERR PFX "Error, kmalloc(arena.map) failed.\n");
616 		return -ENOMEM;
617 	}
618 	iommu->arena.limit = num_tsb_entries;
619 
620 	sz = probe_existing_entries(pbm, iommu);
621 	if (sz)
622 		printk("%s: Imported %lu TSB entries from OBP\n",
623 		       pbm->name, sz);
624 
625 	return 0;
626 }
627 
628 #ifdef CONFIG_PCI_MSI
629 struct pci_sun4v_msiq_entry {
630 	u64		version_type;
631 #define MSIQ_VERSION_MASK		0xffffffff00000000UL
632 #define MSIQ_VERSION_SHIFT		32
633 #define MSIQ_TYPE_MASK			0x00000000000000ffUL
634 #define MSIQ_TYPE_SHIFT			0
635 #define MSIQ_TYPE_NONE			0x00
636 #define MSIQ_TYPE_MSG			0x01
637 #define MSIQ_TYPE_MSI32			0x02
638 #define MSIQ_TYPE_MSI64			0x03
639 #define MSIQ_TYPE_INTX			0x08
640 #define MSIQ_TYPE_NONE2			0xff
641 
642 	u64		intx_sysino;
643 	u64		reserved1;
644 	u64		stick;
645 	u64		req_id;  /* bus/device/func */
646 #define MSIQ_REQID_BUS_MASK		0xff00UL
647 #define MSIQ_REQID_BUS_SHIFT		8
648 #define MSIQ_REQID_DEVICE_MASK		0x00f8UL
649 #define MSIQ_REQID_DEVICE_SHIFT		3
650 #define MSIQ_REQID_FUNC_MASK		0x0007UL
651 #define MSIQ_REQID_FUNC_SHIFT		0
652 
653 	u64		msi_address;
654 
655 	/* The format of this value is message type dependent.
656 	 * For MSI bits 15:0 are the data from the MSI packet.
657 	 * For MSI-X bits 31:0 are the data from the MSI packet.
658 	 * For MSG, the message code and message routing code where:
659 	 * 	bits 39:32 is the bus/device/fn of the msg target-id
660 	 *	bits 18:16 is the message routing code
661 	 *	bits 7:0 is the message code
662 	 * For INTx the low order 2-bits are:
663 	 *	00 - INTA
664 	 *	01 - INTB
665 	 *	10 - INTC
666 	 *	11 - INTD
667 	 */
668 	u64		msi_data;
669 
670 	u64		reserved2;
671 };
672 
673 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
674 			      unsigned long *head)
675 {
676 	unsigned long err, limit;
677 
678 	err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
679 	if (unlikely(err))
680 		return -ENXIO;
681 
682 	limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
683 	if (unlikely(*head >= limit))
684 		return -EFBIG;
685 
686 	return 0;
687 }
688 
689 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
690 				 unsigned long msiqid, unsigned long *head,
691 				 unsigned long *msi)
692 {
693 	struct pci_sun4v_msiq_entry *ep;
694 	unsigned long err, type;
695 
696 	/* Note: void pointer arithmetic, 'head' is a byte offset  */
697 	ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
698 				 (pbm->msiq_ent_count *
699 				  sizeof(struct pci_sun4v_msiq_entry))) +
700 	      *head);
701 
702 	if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
703 		return 0;
704 
705 	type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
706 	if (unlikely(type != MSIQ_TYPE_MSI32 &&
707 		     type != MSIQ_TYPE_MSI64))
708 		return -EINVAL;
709 
710 	*msi = ep->msi_data;
711 
712 	err = pci_sun4v_msi_setstate(pbm->devhandle,
713 				     ep->msi_data /* msi_num */,
714 				     HV_MSISTATE_IDLE);
715 	if (unlikely(err))
716 		return -ENXIO;
717 
718 	/* Clear the entry.  */
719 	ep->version_type &= ~MSIQ_TYPE_MASK;
720 
721 	(*head) += sizeof(struct pci_sun4v_msiq_entry);
722 	if (*head >=
723 	    (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
724 		*head = 0;
725 
726 	return 1;
727 }
728 
729 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
730 			      unsigned long head)
731 {
732 	unsigned long err;
733 
734 	err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
735 	if (unlikely(err))
736 		return -EINVAL;
737 
738 	return 0;
739 }
740 
741 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
742 			       unsigned long msi, int is_msi64)
743 {
744 	if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
745 				  (is_msi64 ?
746 				   HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
747 		return -ENXIO;
748 	if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
749 		return -ENXIO;
750 	if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
751 		return -ENXIO;
752 	return 0;
753 }
754 
755 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
756 {
757 	unsigned long err, msiqid;
758 
759 	err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
760 	if (err)
761 		return -ENXIO;
762 
763 	pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
764 
765 	return 0;
766 }
767 
768 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
769 {
770 	unsigned long q_size, alloc_size, pages, order;
771 	int i;
772 
773 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
774 	alloc_size = (pbm->msiq_num * q_size);
775 	order = get_order(alloc_size);
776 	pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
777 	if (pages == 0UL) {
778 		printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
779 		       order);
780 		return -ENOMEM;
781 	}
782 	memset((char *)pages, 0, PAGE_SIZE << order);
783 	pbm->msi_queues = (void *) pages;
784 
785 	for (i = 0; i < pbm->msiq_num; i++) {
786 		unsigned long err, base = __pa(pages + (i * q_size));
787 		unsigned long ret1, ret2;
788 
789 		err = pci_sun4v_msiq_conf(pbm->devhandle,
790 					  pbm->msiq_first + i,
791 					  base, pbm->msiq_ent_count);
792 		if (err) {
793 			printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
794 			       err);
795 			goto h_error;
796 		}
797 
798 		err = pci_sun4v_msiq_info(pbm->devhandle,
799 					  pbm->msiq_first + i,
800 					  &ret1, &ret2);
801 		if (err) {
802 			printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
803 			       err);
804 			goto h_error;
805 		}
806 		if (ret1 != base || ret2 != pbm->msiq_ent_count) {
807 			printk(KERN_ERR "MSI: Bogus qconf "
808 			       "expected[%lx:%x] got[%lx:%lx]\n",
809 			       base, pbm->msiq_ent_count,
810 			       ret1, ret2);
811 			goto h_error;
812 		}
813 	}
814 
815 	return 0;
816 
817 h_error:
818 	free_pages(pages, order);
819 	return -EINVAL;
820 }
821 
822 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
823 {
824 	unsigned long q_size, alloc_size, pages, order;
825 	int i;
826 
827 	for (i = 0; i < pbm->msiq_num; i++) {
828 		unsigned long msiqid = pbm->msiq_first + i;
829 
830 		(void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
831 	}
832 
833 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
834 	alloc_size = (pbm->msiq_num * q_size);
835 	order = get_order(alloc_size);
836 
837 	pages = (unsigned long) pbm->msi_queues;
838 
839 	free_pages(pages, order);
840 
841 	pbm->msi_queues = NULL;
842 }
843 
844 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
845 				    unsigned long msiqid,
846 				    unsigned long devino)
847 {
848 	unsigned int irq = sun4v_build_irq(pbm->devhandle, devino);
849 
850 	if (!irq)
851 		return -ENOMEM;
852 
853 	if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
854 		return -EINVAL;
855 	if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
856 		return -EINVAL;
857 
858 	return irq;
859 }
860 
861 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
862 	.get_head	=	pci_sun4v_get_head,
863 	.dequeue_msi	=	pci_sun4v_dequeue_msi,
864 	.set_head	=	pci_sun4v_set_head,
865 	.msi_setup	=	pci_sun4v_msi_setup,
866 	.msi_teardown	=	pci_sun4v_msi_teardown,
867 	.msiq_alloc	=	pci_sun4v_msiq_alloc,
868 	.msiq_free	=	pci_sun4v_msiq_free,
869 	.msiq_build_irq	=	pci_sun4v_msiq_build_irq,
870 };
871 
872 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
873 {
874 	sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
875 }
876 #else /* CONFIG_PCI_MSI */
877 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
878 {
879 }
880 #endif /* !(CONFIG_PCI_MSI) */
881 
882 static int __devinit pci_sun4v_pbm_init(struct pci_pbm_info *pbm,
883 					struct platform_device *op, u32 devhandle)
884 {
885 	struct device_node *dp = op->dev.of_node;
886 	int err;
887 
888 	pbm->numa_node = of_node_to_nid(dp);
889 
890 	pbm->pci_ops = &sun4v_pci_ops;
891 	pbm->config_space_reg_bits = 12;
892 
893 	pbm->index = pci_num_pbms++;
894 
895 	pbm->op = op;
896 
897 	pbm->devhandle = devhandle;
898 
899 	pbm->name = dp->full_name;
900 
901 	printk("%s: SUN4V PCI Bus Module\n", pbm->name);
902 	printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node);
903 
904 	pci_determine_mem_io_space(pbm);
905 
906 	pci_get_pbm_props(pbm);
907 
908 	err = pci_sun4v_iommu_init(pbm);
909 	if (err)
910 		return err;
911 
912 	pci_sun4v_msi_init(pbm);
913 
914 	pci_sun4v_scan_bus(pbm, &op->dev);
915 
916 	pbm->next = pci_pbm_root;
917 	pci_pbm_root = pbm;
918 
919 	return 0;
920 }
921 
922 static int __devinit pci_sun4v_probe(struct platform_device *op)
923 {
924 	const struct linux_prom64_registers *regs;
925 	static int hvapi_negotiated = 0;
926 	struct pci_pbm_info *pbm;
927 	struct device_node *dp;
928 	struct iommu *iommu;
929 	u32 devhandle;
930 	int i, err;
931 
932 	dp = op->dev.of_node;
933 
934 	if (!hvapi_negotiated++) {
935 		err = sun4v_hvapi_register(HV_GRP_PCI,
936 					   vpci_major,
937 					   &vpci_minor);
938 
939 		if (err) {
940 			printk(KERN_ERR PFX "Could not register hvapi, "
941 			       "err=%d\n", err);
942 			return err;
943 		}
944 		printk(KERN_INFO PFX "Registered hvapi major[%lu] minor[%lu]\n",
945 		       vpci_major, vpci_minor);
946 
947 		dma_ops = &sun4v_dma_ops;
948 	}
949 
950 	regs = of_get_property(dp, "reg", NULL);
951 	err = -ENODEV;
952 	if (!regs) {
953 		printk(KERN_ERR PFX "Could not find config registers\n");
954 		goto out_err;
955 	}
956 	devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
957 
958 	err = -ENOMEM;
959 	if (!iommu_batch_initialized) {
960 		for_each_possible_cpu(i) {
961 			unsigned long page = get_zeroed_page(GFP_KERNEL);
962 
963 			if (!page)
964 				goto out_err;
965 
966 			per_cpu(iommu_batch, i).pglist = (u64 *) page;
967 		}
968 		iommu_batch_initialized = 1;
969 	}
970 
971 	pbm = kzalloc(sizeof(*pbm), GFP_KERNEL);
972 	if (!pbm) {
973 		printk(KERN_ERR PFX "Could not allocate pci_pbm_info\n");
974 		goto out_err;
975 	}
976 
977 	iommu = kzalloc(sizeof(struct iommu), GFP_KERNEL);
978 	if (!iommu) {
979 		printk(KERN_ERR PFX "Could not allocate pbm iommu\n");
980 		goto out_free_controller;
981 	}
982 
983 	pbm->iommu = iommu;
984 
985 	err = pci_sun4v_pbm_init(pbm, op, devhandle);
986 	if (err)
987 		goto out_free_iommu;
988 
989 	dev_set_drvdata(&op->dev, pbm);
990 
991 	return 0;
992 
993 out_free_iommu:
994 	kfree(pbm->iommu);
995 
996 out_free_controller:
997 	kfree(pbm);
998 
999 out_err:
1000 	return err;
1001 }
1002 
1003 static const struct of_device_id pci_sun4v_match[] = {
1004 	{
1005 		.name = "pci",
1006 		.compatible = "SUNW,sun4v-pci",
1007 	},
1008 	{},
1009 };
1010 
1011 static struct platform_driver pci_sun4v_driver = {
1012 	.driver = {
1013 		.name = DRIVER_NAME,
1014 		.owner = THIS_MODULE,
1015 		.of_match_table = pci_sun4v_match,
1016 	},
1017 	.probe		= pci_sun4v_probe,
1018 };
1019 
1020 static int __init pci_sun4v_init(void)
1021 {
1022 	return platform_driver_register(&pci_sun4v_driver);
1023 }
1024 
1025 subsys_initcall(pci_sun4v_init);
1026