xref: /openbmc/linux/arch/sparc/kernel/pci_sun4v.c (revision 4a44a19b)
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 = this_cpu_ptr(&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 = this_cpu_ptr(&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 = this_cpu_ptr(&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 = this_cpu_ptr(&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 pci_sun4v_scan_bus(struct pci_pbm_info *pbm, struct device *parent)
540 {
541 	struct property *prop;
542 	struct device_node *dp;
543 
544 	dp = pbm->op->dev.of_node;
545 	prop = of_find_property(dp, "66mhz-capable", NULL);
546 	pbm->is_66mhz_capable = (prop != NULL);
547 	pbm->pci_bus = pci_scan_one_pbm(pbm, parent);
548 
549 	/* XXX register error interrupt handlers XXX */
550 }
551 
552 static unsigned long probe_existing_entries(struct pci_pbm_info *pbm,
553 					    struct iommu *iommu)
554 {
555 	struct iommu_arena *arena = &iommu->arena;
556 	unsigned long i, cnt = 0;
557 	u32 devhandle;
558 
559 	devhandle = pbm->devhandle;
560 	for (i = 0; i < arena->limit; i++) {
561 		unsigned long ret, io_attrs, ra;
562 
563 		ret = pci_sun4v_iommu_getmap(devhandle,
564 					     HV_PCI_TSBID(0, i),
565 					     &io_attrs, &ra);
566 		if (ret == HV_EOK) {
567 			if (page_in_phys_avail(ra)) {
568 				pci_sun4v_iommu_demap(devhandle,
569 						      HV_PCI_TSBID(0, i), 1);
570 			} else {
571 				cnt++;
572 				__set_bit(i, arena->map);
573 			}
574 		}
575 	}
576 
577 	return cnt;
578 }
579 
580 static int pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
581 {
582 	static const u32 vdma_default[] = { 0x80000000, 0x80000000 };
583 	struct iommu *iommu = pbm->iommu;
584 	unsigned long num_tsb_entries, sz;
585 	u32 dma_mask, dma_offset;
586 	const u32 *vdma;
587 
588 	vdma = of_get_property(pbm->op->dev.of_node, "virtual-dma", NULL);
589 	if (!vdma)
590 		vdma = vdma_default;
591 
592 	if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) {
593 		printk(KERN_ERR PFX "Strange virtual-dma[%08x:%08x].\n",
594 		       vdma[0], vdma[1]);
595 		return -EINVAL;
596 	}
597 
598 	dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
599 	num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
600 
601 	dma_offset = vdma[0];
602 
603 	/* Setup initial software IOMMU state. */
604 	spin_lock_init(&iommu->lock);
605 	iommu->ctx_lowest_free = 1;
606 	iommu->page_table_map_base = dma_offset;
607 	iommu->dma_addr_mask = dma_mask;
608 
609 	/* Allocate and initialize the free area map.  */
610 	sz = (num_tsb_entries + 7) / 8;
611 	sz = (sz + 7UL) & ~7UL;
612 	iommu->arena.map = kzalloc(sz, GFP_KERNEL);
613 	if (!iommu->arena.map) {
614 		printk(KERN_ERR PFX "Error, kmalloc(arena.map) failed.\n");
615 		return -ENOMEM;
616 	}
617 	iommu->arena.limit = num_tsb_entries;
618 
619 	sz = probe_existing_entries(pbm, iommu);
620 	if (sz)
621 		printk("%s: Imported %lu TSB entries from OBP\n",
622 		       pbm->name, sz);
623 
624 	return 0;
625 }
626 
627 #ifdef CONFIG_PCI_MSI
628 struct pci_sun4v_msiq_entry {
629 	u64		version_type;
630 #define MSIQ_VERSION_MASK		0xffffffff00000000UL
631 #define MSIQ_VERSION_SHIFT		32
632 #define MSIQ_TYPE_MASK			0x00000000000000ffUL
633 #define MSIQ_TYPE_SHIFT			0
634 #define MSIQ_TYPE_NONE			0x00
635 #define MSIQ_TYPE_MSG			0x01
636 #define MSIQ_TYPE_MSI32			0x02
637 #define MSIQ_TYPE_MSI64			0x03
638 #define MSIQ_TYPE_INTX			0x08
639 #define MSIQ_TYPE_NONE2			0xff
640 
641 	u64		intx_sysino;
642 	u64		reserved1;
643 	u64		stick;
644 	u64		req_id;  /* bus/device/func */
645 #define MSIQ_REQID_BUS_MASK		0xff00UL
646 #define MSIQ_REQID_BUS_SHIFT		8
647 #define MSIQ_REQID_DEVICE_MASK		0x00f8UL
648 #define MSIQ_REQID_DEVICE_SHIFT		3
649 #define MSIQ_REQID_FUNC_MASK		0x0007UL
650 #define MSIQ_REQID_FUNC_SHIFT		0
651 
652 	u64		msi_address;
653 
654 	/* The format of this value is message type dependent.
655 	 * For MSI bits 15:0 are the data from the MSI packet.
656 	 * For MSI-X bits 31:0 are the data from the MSI packet.
657 	 * For MSG, the message code and message routing code where:
658 	 * 	bits 39:32 is the bus/device/fn of the msg target-id
659 	 *	bits 18:16 is the message routing code
660 	 *	bits 7:0 is the message code
661 	 * For INTx the low order 2-bits are:
662 	 *	00 - INTA
663 	 *	01 - INTB
664 	 *	10 - INTC
665 	 *	11 - INTD
666 	 */
667 	u64		msi_data;
668 
669 	u64		reserved2;
670 };
671 
672 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
673 			      unsigned long *head)
674 {
675 	unsigned long err, limit;
676 
677 	err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
678 	if (unlikely(err))
679 		return -ENXIO;
680 
681 	limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
682 	if (unlikely(*head >= limit))
683 		return -EFBIG;
684 
685 	return 0;
686 }
687 
688 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
689 				 unsigned long msiqid, unsigned long *head,
690 				 unsigned long *msi)
691 {
692 	struct pci_sun4v_msiq_entry *ep;
693 	unsigned long err, type;
694 
695 	/* Note: void pointer arithmetic, 'head' is a byte offset  */
696 	ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
697 				 (pbm->msiq_ent_count *
698 				  sizeof(struct pci_sun4v_msiq_entry))) +
699 	      *head);
700 
701 	if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
702 		return 0;
703 
704 	type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
705 	if (unlikely(type != MSIQ_TYPE_MSI32 &&
706 		     type != MSIQ_TYPE_MSI64))
707 		return -EINVAL;
708 
709 	*msi = ep->msi_data;
710 
711 	err = pci_sun4v_msi_setstate(pbm->devhandle,
712 				     ep->msi_data /* msi_num */,
713 				     HV_MSISTATE_IDLE);
714 	if (unlikely(err))
715 		return -ENXIO;
716 
717 	/* Clear the entry.  */
718 	ep->version_type &= ~MSIQ_TYPE_MASK;
719 
720 	(*head) += sizeof(struct pci_sun4v_msiq_entry);
721 	if (*head >=
722 	    (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
723 		*head = 0;
724 
725 	return 1;
726 }
727 
728 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
729 			      unsigned long head)
730 {
731 	unsigned long err;
732 
733 	err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
734 	if (unlikely(err))
735 		return -EINVAL;
736 
737 	return 0;
738 }
739 
740 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
741 			       unsigned long msi, int is_msi64)
742 {
743 	if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
744 				  (is_msi64 ?
745 				   HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
746 		return -ENXIO;
747 	if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
748 		return -ENXIO;
749 	if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
750 		return -ENXIO;
751 	return 0;
752 }
753 
754 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
755 {
756 	unsigned long err, msiqid;
757 
758 	err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
759 	if (err)
760 		return -ENXIO;
761 
762 	pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
763 
764 	return 0;
765 }
766 
767 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
768 {
769 	unsigned long q_size, alloc_size, pages, order;
770 	int i;
771 
772 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
773 	alloc_size = (pbm->msiq_num * q_size);
774 	order = get_order(alloc_size);
775 	pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
776 	if (pages == 0UL) {
777 		printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
778 		       order);
779 		return -ENOMEM;
780 	}
781 	memset((char *)pages, 0, PAGE_SIZE << order);
782 	pbm->msi_queues = (void *) pages;
783 
784 	for (i = 0; i < pbm->msiq_num; i++) {
785 		unsigned long err, base = __pa(pages + (i * q_size));
786 		unsigned long ret1, ret2;
787 
788 		err = pci_sun4v_msiq_conf(pbm->devhandle,
789 					  pbm->msiq_first + i,
790 					  base, pbm->msiq_ent_count);
791 		if (err) {
792 			printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
793 			       err);
794 			goto h_error;
795 		}
796 
797 		err = pci_sun4v_msiq_info(pbm->devhandle,
798 					  pbm->msiq_first + i,
799 					  &ret1, &ret2);
800 		if (err) {
801 			printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
802 			       err);
803 			goto h_error;
804 		}
805 		if (ret1 != base || ret2 != pbm->msiq_ent_count) {
806 			printk(KERN_ERR "MSI: Bogus qconf "
807 			       "expected[%lx:%x] got[%lx:%lx]\n",
808 			       base, pbm->msiq_ent_count,
809 			       ret1, ret2);
810 			goto h_error;
811 		}
812 	}
813 
814 	return 0;
815 
816 h_error:
817 	free_pages(pages, order);
818 	return -EINVAL;
819 }
820 
821 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
822 {
823 	unsigned long q_size, alloc_size, pages, order;
824 	int i;
825 
826 	for (i = 0; i < pbm->msiq_num; i++) {
827 		unsigned long msiqid = pbm->msiq_first + i;
828 
829 		(void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
830 	}
831 
832 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
833 	alloc_size = (pbm->msiq_num * q_size);
834 	order = get_order(alloc_size);
835 
836 	pages = (unsigned long) pbm->msi_queues;
837 
838 	free_pages(pages, order);
839 
840 	pbm->msi_queues = NULL;
841 }
842 
843 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
844 				    unsigned long msiqid,
845 				    unsigned long devino)
846 {
847 	unsigned int irq = sun4v_build_irq(pbm->devhandle, devino);
848 
849 	if (!irq)
850 		return -ENOMEM;
851 
852 	if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
853 		return -EINVAL;
854 	if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
855 		return -EINVAL;
856 
857 	return irq;
858 }
859 
860 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
861 	.get_head	=	pci_sun4v_get_head,
862 	.dequeue_msi	=	pci_sun4v_dequeue_msi,
863 	.set_head	=	pci_sun4v_set_head,
864 	.msi_setup	=	pci_sun4v_msi_setup,
865 	.msi_teardown	=	pci_sun4v_msi_teardown,
866 	.msiq_alloc	=	pci_sun4v_msiq_alloc,
867 	.msiq_free	=	pci_sun4v_msiq_free,
868 	.msiq_build_irq	=	pci_sun4v_msiq_build_irq,
869 };
870 
871 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
872 {
873 	sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
874 }
875 #else /* CONFIG_PCI_MSI */
876 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
877 {
878 }
879 #endif /* !(CONFIG_PCI_MSI) */
880 
881 static int pci_sun4v_pbm_init(struct pci_pbm_info *pbm,
882 			      struct platform_device *op, u32 devhandle)
883 {
884 	struct device_node *dp = op->dev.of_node;
885 	int err;
886 
887 	pbm->numa_node = of_node_to_nid(dp);
888 
889 	pbm->pci_ops = &sun4v_pci_ops;
890 	pbm->config_space_reg_bits = 12;
891 
892 	pbm->index = pci_num_pbms++;
893 
894 	pbm->op = op;
895 
896 	pbm->devhandle = devhandle;
897 
898 	pbm->name = dp->full_name;
899 
900 	printk("%s: SUN4V PCI Bus Module\n", pbm->name);
901 	printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node);
902 
903 	pci_determine_mem_io_space(pbm);
904 
905 	pci_get_pbm_props(pbm);
906 
907 	err = pci_sun4v_iommu_init(pbm);
908 	if (err)
909 		return err;
910 
911 	pci_sun4v_msi_init(pbm);
912 
913 	pci_sun4v_scan_bus(pbm, &op->dev);
914 
915 	pbm->next = pci_pbm_root;
916 	pci_pbm_root = pbm;
917 
918 	return 0;
919 }
920 
921 static int pci_sun4v_probe(struct platform_device *op)
922 {
923 	const struct linux_prom64_registers *regs;
924 	static int hvapi_negotiated = 0;
925 	struct pci_pbm_info *pbm;
926 	struct device_node *dp;
927 	struct iommu *iommu;
928 	u32 devhandle;
929 	int i, err;
930 
931 	dp = op->dev.of_node;
932 
933 	if (!hvapi_negotiated++) {
934 		err = sun4v_hvapi_register(HV_GRP_PCI,
935 					   vpci_major,
936 					   &vpci_minor);
937 
938 		if (err) {
939 			printk(KERN_ERR PFX "Could not register hvapi, "
940 			       "err=%d\n", err);
941 			return err;
942 		}
943 		printk(KERN_INFO PFX "Registered hvapi major[%lu] minor[%lu]\n",
944 		       vpci_major, vpci_minor);
945 
946 		dma_ops = &sun4v_dma_ops;
947 	}
948 
949 	regs = of_get_property(dp, "reg", NULL);
950 	err = -ENODEV;
951 	if (!regs) {
952 		printk(KERN_ERR PFX "Could not find config registers\n");
953 		goto out_err;
954 	}
955 	devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
956 
957 	err = -ENOMEM;
958 	if (!iommu_batch_initialized) {
959 		for_each_possible_cpu(i) {
960 			unsigned long page = get_zeroed_page(GFP_KERNEL);
961 
962 			if (!page)
963 				goto out_err;
964 
965 			per_cpu(iommu_batch, i).pglist = (u64 *) page;
966 		}
967 		iommu_batch_initialized = 1;
968 	}
969 
970 	pbm = kzalloc(sizeof(*pbm), GFP_KERNEL);
971 	if (!pbm) {
972 		printk(KERN_ERR PFX "Could not allocate pci_pbm_info\n");
973 		goto out_err;
974 	}
975 
976 	iommu = kzalloc(sizeof(struct iommu), GFP_KERNEL);
977 	if (!iommu) {
978 		printk(KERN_ERR PFX "Could not allocate pbm iommu\n");
979 		goto out_free_controller;
980 	}
981 
982 	pbm->iommu = iommu;
983 
984 	err = pci_sun4v_pbm_init(pbm, op, devhandle);
985 	if (err)
986 		goto out_free_iommu;
987 
988 	dev_set_drvdata(&op->dev, pbm);
989 
990 	return 0;
991 
992 out_free_iommu:
993 	kfree(pbm->iommu);
994 
995 out_free_controller:
996 	kfree(pbm);
997 
998 out_err:
999 	return err;
1000 }
1001 
1002 static const struct of_device_id pci_sun4v_match[] = {
1003 	{
1004 		.name = "pci",
1005 		.compatible = "SUNW,sun4v-pci",
1006 	},
1007 	{},
1008 };
1009 
1010 static struct platform_driver pci_sun4v_driver = {
1011 	.driver = {
1012 		.name = DRIVER_NAME,
1013 		.owner = THIS_MODULE,
1014 		.of_match_table = pci_sun4v_match,
1015 	},
1016 	.probe		= pci_sun4v_probe,
1017 };
1018 
1019 static int __init pci_sun4v_init(void)
1020 {
1021 	return platform_driver_register(&pci_sun4v_driver);
1022 }
1023 
1024 subsys_initcall(pci_sun4v_init);
1025