xref: /openbmc/linux/arch/sparc/kernel/pci_sun4v.c (revision 293d5b43)
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 #include <linux/iommu-common.h>
19 
20 #include <asm/iommu.h>
21 #include <asm/irq.h>
22 #include <asm/hypervisor.h>
23 #include <asm/prom.h>
24 
25 #include "pci_impl.h"
26 #include "iommu_common.h"
27 
28 #include "pci_sun4v.h"
29 
30 #define DRIVER_NAME	"pci_sun4v"
31 #define PFX		DRIVER_NAME ": "
32 
33 static unsigned long vpci_major = 1;
34 static unsigned long vpci_minor = 1;
35 
36 #define PGLIST_NENTS	(PAGE_SIZE / sizeof(u64))
37 
38 struct iommu_batch {
39 	struct device	*dev;		/* Device mapping is for.	*/
40 	unsigned long	prot;		/* IOMMU page protections	*/
41 	unsigned long	entry;		/* Index into IOTSB.		*/
42 	u64		*pglist;	/* List of physical pages	*/
43 	unsigned long	npages;		/* Number of pages in list.	*/
44 };
45 
46 static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
47 static int iommu_batch_initialized;
48 
49 /* Interrupts must be disabled.  */
50 static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
51 {
52 	struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
53 
54 	p->dev		= dev;
55 	p->prot		= prot;
56 	p->entry	= entry;
57 	p->npages	= 0;
58 }
59 
60 /* Interrupts must be disabled.  */
61 static long iommu_batch_flush(struct iommu_batch *p)
62 {
63 	struct pci_pbm_info *pbm = p->dev->archdata.host_controller;
64 	unsigned long devhandle = pbm->devhandle;
65 	unsigned long prot = p->prot;
66 	unsigned long entry = p->entry;
67 	u64 *pglist = p->pglist;
68 	unsigned long npages = p->npages;
69 
70 	while (npages != 0) {
71 		long num;
72 
73 		num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
74 					  npages, prot, __pa(pglist));
75 		if (unlikely(num < 0)) {
76 			if (printk_ratelimit())
77 				printk("iommu_batch_flush: IOMMU map of "
78 				       "[%08lx:%08llx:%lx:%lx:%lx] failed with "
79 				       "status %ld\n",
80 				       devhandle, HV_PCI_TSBID(0, entry),
81 				       npages, prot, __pa(pglist), num);
82 			return -1;
83 		}
84 
85 		entry += num;
86 		npages -= num;
87 		pglist += num;
88 	}
89 
90 	p->entry = entry;
91 	p->npages = 0;
92 
93 	return 0;
94 }
95 
96 static inline void iommu_batch_new_entry(unsigned long entry)
97 {
98 	struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
99 
100 	if (p->entry + p->npages == entry)
101 		return;
102 	if (p->entry != ~0UL)
103 		iommu_batch_flush(p);
104 	p->entry = entry;
105 }
106 
107 /* Interrupts must be disabled.  */
108 static inline long iommu_batch_add(u64 phys_page)
109 {
110 	struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
111 
112 	BUG_ON(p->npages >= PGLIST_NENTS);
113 
114 	p->pglist[p->npages++] = phys_page;
115 	if (p->npages == PGLIST_NENTS)
116 		return iommu_batch_flush(p);
117 
118 	return 0;
119 }
120 
121 /* Interrupts must be disabled.  */
122 static inline long iommu_batch_end(void)
123 {
124 	struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
125 
126 	BUG_ON(p->npages >= PGLIST_NENTS);
127 
128 	return iommu_batch_flush(p);
129 }
130 
131 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
132 				   dma_addr_t *dma_addrp, gfp_t gfp,
133 				   unsigned long attrs)
134 {
135 	unsigned long flags, order, first_page, npages, n;
136 	struct iommu *iommu;
137 	struct page *page;
138 	void *ret;
139 	long entry;
140 	int nid;
141 
142 	size = IO_PAGE_ALIGN(size);
143 	order = get_order(size);
144 	if (unlikely(order >= MAX_ORDER))
145 		return NULL;
146 
147 	npages = size >> IO_PAGE_SHIFT;
148 
149 	nid = dev->archdata.numa_node;
150 	page = alloc_pages_node(nid, gfp, order);
151 	if (unlikely(!page))
152 		return NULL;
153 
154 	first_page = (unsigned long) page_address(page);
155 	memset((char *)first_page, 0, PAGE_SIZE << order);
156 
157 	iommu = dev->archdata.iommu;
158 
159 	entry = iommu_tbl_range_alloc(dev, &iommu->tbl, npages, NULL,
160 				      (unsigned long)(-1), 0);
161 
162 	if (unlikely(entry == IOMMU_ERROR_CODE))
163 		goto range_alloc_fail;
164 
165 	*dma_addrp = (iommu->tbl.table_map_base + (entry << IO_PAGE_SHIFT));
166 	ret = (void *) first_page;
167 	first_page = __pa(first_page);
168 
169 	local_irq_save(flags);
170 
171 	iommu_batch_start(dev,
172 			  (HV_PCI_MAP_ATTR_READ |
173 			   HV_PCI_MAP_ATTR_WRITE),
174 			  entry);
175 
176 	for (n = 0; n < npages; n++) {
177 		long err = iommu_batch_add(first_page + (n * PAGE_SIZE));
178 		if (unlikely(err < 0L))
179 			goto iommu_map_fail;
180 	}
181 
182 	if (unlikely(iommu_batch_end() < 0L))
183 		goto iommu_map_fail;
184 
185 	local_irq_restore(flags);
186 
187 	return ret;
188 
189 iommu_map_fail:
190 	iommu_tbl_range_free(&iommu->tbl, *dma_addrp, npages, IOMMU_ERROR_CODE);
191 
192 range_alloc_fail:
193 	free_pages(first_page, order);
194 	return NULL;
195 }
196 
197 static void dma_4v_iommu_demap(void *demap_arg, unsigned long entry,
198 			       unsigned long npages)
199 {
200 	u32 devhandle = *(u32 *)demap_arg;
201 	unsigned long num, flags;
202 
203 	local_irq_save(flags);
204 	do {
205 		num = pci_sun4v_iommu_demap(devhandle,
206 					    HV_PCI_TSBID(0, entry),
207 					    npages);
208 
209 		entry += num;
210 		npages -= num;
211 	} while (npages != 0);
212 	local_irq_restore(flags);
213 }
214 
215 static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
216 				 dma_addr_t dvma, unsigned long attrs)
217 {
218 	struct pci_pbm_info *pbm;
219 	struct iommu *iommu;
220 	unsigned long order, npages, entry;
221 	u32 devhandle;
222 
223 	npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
224 	iommu = dev->archdata.iommu;
225 	pbm = dev->archdata.host_controller;
226 	devhandle = pbm->devhandle;
227 	entry = ((dvma - iommu->tbl.table_map_base) >> IO_PAGE_SHIFT);
228 	dma_4v_iommu_demap(&devhandle, entry, npages);
229 	iommu_tbl_range_free(&iommu->tbl, dvma, npages, IOMMU_ERROR_CODE);
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 				  unsigned long 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 	entry = iommu_tbl_range_alloc(dev, &iommu->tbl, npages, NULL,
257 				      (unsigned long)(-1), 0);
258 
259 	if (unlikely(entry == IOMMU_ERROR_CODE))
260 		goto bad;
261 
262 	bus_addr = (iommu->tbl.table_map_base + (entry << IO_PAGE_SHIFT));
263 	ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
264 	base_paddr = __pa(oaddr & IO_PAGE_MASK);
265 	prot = HV_PCI_MAP_ATTR_READ;
266 	if (direction != DMA_TO_DEVICE)
267 		prot |= HV_PCI_MAP_ATTR_WRITE;
268 
269 	local_irq_save(flags);
270 
271 	iommu_batch_start(dev, prot, entry);
272 
273 	for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
274 		long err = iommu_batch_add(base_paddr);
275 		if (unlikely(err < 0L))
276 			goto iommu_map_fail;
277 	}
278 	if (unlikely(iommu_batch_end() < 0L))
279 		goto iommu_map_fail;
280 
281 	local_irq_restore(flags);
282 
283 	return ret;
284 
285 bad:
286 	if (printk_ratelimit())
287 		WARN_ON(1);
288 	return DMA_ERROR_CODE;
289 
290 iommu_map_fail:
291 	iommu_tbl_range_free(&iommu->tbl, bus_addr, npages, IOMMU_ERROR_CODE);
292 	return DMA_ERROR_CODE;
293 }
294 
295 static void dma_4v_unmap_page(struct device *dev, dma_addr_t bus_addr,
296 			      size_t sz, enum dma_data_direction direction,
297 			      unsigned long attrs)
298 {
299 	struct pci_pbm_info *pbm;
300 	struct iommu *iommu;
301 	unsigned long npages;
302 	long entry;
303 	u32 devhandle;
304 
305 	if (unlikely(direction == DMA_NONE)) {
306 		if (printk_ratelimit())
307 			WARN_ON(1);
308 		return;
309 	}
310 
311 	iommu = dev->archdata.iommu;
312 	pbm = dev->archdata.host_controller;
313 	devhandle = pbm->devhandle;
314 
315 	npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
316 	npages >>= IO_PAGE_SHIFT;
317 	bus_addr &= IO_PAGE_MASK;
318 	entry = (bus_addr - iommu->tbl.table_map_base) >> IO_PAGE_SHIFT;
319 	dma_4v_iommu_demap(&devhandle, entry, npages);
320 	iommu_tbl_range_free(&iommu->tbl, bus_addr, npages, IOMMU_ERROR_CODE);
321 }
322 
323 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
324 			 int nelems, enum dma_data_direction direction,
325 			 unsigned long attrs)
326 {
327 	struct scatterlist *s, *outs, *segstart;
328 	unsigned long flags, handle, prot;
329 	dma_addr_t dma_next = 0, dma_addr;
330 	unsigned int max_seg_size;
331 	unsigned long seg_boundary_size;
332 	int outcount, incount, i;
333 	struct iommu *iommu;
334 	unsigned long base_shift;
335 	long err;
336 
337 	BUG_ON(direction == DMA_NONE);
338 
339 	iommu = dev->archdata.iommu;
340 	if (nelems == 0 || !iommu)
341 		return 0;
342 
343 	prot = HV_PCI_MAP_ATTR_READ;
344 	if (direction != DMA_TO_DEVICE)
345 		prot |= HV_PCI_MAP_ATTR_WRITE;
346 
347 	outs = s = segstart = &sglist[0];
348 	outcount = 1;
349 	incount = nelems;
350 	handle = 0;
351 
352 	/* Init first segment length for backout at failure */
353 	outs->dma_length = 0;
354 
355 	local_irq_save(flags);
356 
357 	iommu_batch_start(dev, prot, ~0UL);
358 
359 	max_seg_size = dma_get_max_seg_size(dev);
360 	seg_boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
361 				  IO_PAGE_SIZE) >> IO_PAGE_SHIFT;
362 	base_shift = iommu->tbl.table_map_base >> IO_PAGE_SHIFT;
363 	for_each_sg(sglist, s, nelems, i) {
364 		unsigned long paddr, npages, entry, out_entry = 0, slen;
365 
366 		slen = s->length;
367 		/* Sanity check */
368 		if (slen == 0) {
369 			dma_next = 0;
370 			continue;
371 		}
372 		/* Allocate iommu entries for that segment */
373 		paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
374 		npages = iommu_num_pages(paddr, slen, IO_PAGE_SIZE);
375 		entry = iommu_tbl_range_alloc(dev, &iommu->tbl, npages,
376 					      &handle, (unsigned long)(-1), 0);
377 
378 		/* Handle failure */
379 		if (unlikely(entry == IOMMU_ERROR_CODE)) {
380 			if (printk_ratelimit())
381 				printk(KERN_INFO "iommu_alloc failed, iommu %p paddr %lx"
382 				       " npages %lx\n", iommu, paddr, npages);
383 			goto iommu_map_failed;
384 		}
385 
386 		iommu_batch_new_entry(entry);
387 
388 		/* Convert entry to a dma_addr_t */
389 		dma_addr = iommu->tbl.table_map_base + (entry << IO_PAGE_SHIFT);
390 		dma_addr |= (s->offset & ~IO_PAGE_MASK);
391 
392 		/* Insert into HW table */
393 		paddr &= IO_PAGE_MASK;
394 		while (npages--) {
395 			err = iommu_batch_add(paddr);
396 			if (unlikely(err < 0L))
397 				goto iommu_map_failed;
398 			paddr += IO_PAGE_SIZE;
399 		}
400 
401 		/* If we are in an open segment, try merging */
402 		if (segstart != s) {
403 			/* We cannot merge if:
404 			 * - allocated dma_addr isn't contiguous to previous allocation
405 			 */
406 			if ((dma_addr != dma_next) ||
407 			    (outs->dma_length + s->length > max_seg_size) ||
408 			    (is_span_boundary(out_entry, base_shift,
409 					      seg_boundary_size, outs, s))) {
410 				/* Can't merge: create a new segment */
411 				segstart = s;
412 				outcount++;
413 				outs = sg_next(outs);
414 			} else {
415 				outs->dma_length += s->length;
416 			}
417 		}
418 
419 		if (segstart == s) {
420 			/* This is a new segment, fill entries */
421 			outs->dma_address = dma_addr;
422 			outs->dma_length = slen;
423 			out_entry = entry;
424 		}
425 
426 		/* Calculate next page pointer for contiguous check */
427 		dma_next = dma_addr + slen;
428 	}
429 
430 	err = iommu_batch_end();
431 
432 	if (unlikely(err < 0L))
433 		goto iommu_map_failed;
434 
435 	local_irq_restore(flags);
436 
437 	if (outcount < incount) {
438 		outs = sg_next(outs);
439 		outs->dma_address = DMA_ERROR_CODE;
440 		outs->dma_length = 0;
441 	}
442 
443 	return outcount;
444 
445 iommu_map_failed:
446 	for_each_sg(sglist, s, nelems, i) {
447 		if (s->dma_length != 0) {
448 			unsigned long vaddr, npages;
449 
450 			vaddr = s->dma_address & IO_PAGE_MASK;
451 			npages = iommu_num_pages(s->dma_address, s->dma_length,
452 						 IO_PAGE_SIZE);
453 			iommu_tbl_range_free(&iommu->tbl, vaddr, npages,
454 					     IOMMU_ERROR_CODE);
455 			/* XXX demap? XXX */
456 			s->dma_address = DMA_ERROR_CODE;
457 			s->dma_length = 0;
458 		}
459 		if (s == outs)
460 			break;
461 	}
462 	local_irq_restore(flags);
463 
464 	return 0;
465 }
466 
467 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
468 			    int nelems, enum dma_data_direction direction,
469 			    unsigned long attrs)
470 {
471 	struct pci_pbm_info *pbm;
472 	struct scatterlist *sg;
473 	struct iommu *iommu;
474 	unsigned long flags, entry;
475 	u32 devhandle;
476 
477 	BUG_ON(direction == DMA_NONE);
478 
479 	iommu = dev->archdata.iommu;
480 	pbm = dev->archdata.host_controller;
481 	devhandle = pbm->devhandle;
482 
483 	local_irq_save(flags);
484 
485 	sg = sglist;
486 	while (nelems--) {
487 		dma_addr_t dma_handle = sg->dma_address;
488 		unsigned int len = sg->dma_length;
489 		unsigned long npages;
490 		struct iommu_map_table *tbl = &iommu->tbl;
491 		unsigned long shift = IO_PAGE_SHIFT;
492 
493 		if (!len)
494 			break;
495 		npages = iommu_num_pages(dma_handle, len, IO_PAGE_SIZE);
496 		entry = ((dma_handle - tbl->table_map_base) >> shift);
497 		dma_4v_iommu_demap(&devhandle, entry, npages);
498 		iommu_tbl_range_free(&iommu->tbl, dma_handle, npages,
499 				     IOMMU_ERROR_CODE);
500 		sg = sg_next(sg);
501 	}
502 
503 	local_irq_restore(flags);
504 }
505 
506 static struct dma_map_ops sun4v_dma_ops = {
507 	.alloc				= dma_4v_alloc_coherent,
508 	.free				= dma_4v_free_coherent,
509 	.map_page			= dma_4v_map_page,
510 	.unmap_page			= dma_4v_unmap_page,
511 	.map_sg				= dma_4v_map_sg,
512 	.unmap_sg			= dma_4v_unmap_sg,
513 };
514 
515 static void pci_sun4v_scan_bus(struct pci_pbm_info *pbm, struct device *parent)
516 {
517 	struct property *prop;
518 	struct device_node *dp;
519 
520 	dp = pbm->op->dev.of_node;
521 	prop = of_find_property(dp, "66mhz-capable", NULL);
522 	pbm->is_66mhz_capable = (prop != NULL);
523 	pbm->pci_bus = pci_scan_one_pbm(pbm, parent);
524 
525 	/* XXX register error interrupt handlers XXX */
526 }
527 
528 static unsigned long probe_existing_entries(struct pci_pbm_info *pbm,
529 					    struct iommu_map_table *iommu)
530 {
531 	struct iommu_pool *pool;
532 	unsigned long i, pool_nr, cnt = 0;
533 	u32 devhandle;
534 
535 	devhandle = pbm->devhandle;
536 	for (pool_nr = 0; pool_nr < iommu->nr_pools; pool_nr++) {
537 		pool = &(iommu->pools[pool_nr]);
538 		for (i = pool->start; i <= pool->end; i++) {
539 			unsigned long ret, io_attrs, ra;
540 
541 			ret = pci_sun4v_iommu_getmap(devhandle,
542 						     HV_PCI_TSBID(0, i),
543 						     &io_attrs, &ra);
544 			if (ret == HV_EOK) {
545 				if (page_in_phys_avail(ra)) {
546 					pci_sun4v_iommu_demap(devhandle,
547 							      HV_PCI_TSBID(0,
548 							      i), 1);
549 				} else {
550 					cnt++;
551 					__set_bit(i, iommu->map);
552 				}
553 			}
554 		}
555 	}
556 	return cnt;
557 }
558 
559 static int pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
560 {
561 	static const u32 vdma_default[] = { 0x80000000, 0x80000000 };
562 	struct iommu *iommu = pbm->iommu;
563 	unsigned long num_tsb_entries, sz;
564 	u32 dma_mask, dma_offset;
565 	const u32 *vdma;
566 
567 	vdma = of_get_property(pbm->op->dev.of_node, "virtual-dma", NULL);
568 	if (!vdma)
569 		vdma = vdma_default;
570 
571 	if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) {
572 		printk(KERN_ERR PFX "Strange virtual-dma[%08x:%08x].\n",
573 		       vdma[0], vdma[1]);
574 		return -EINVAL;
575 	}
576 
577 	dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
578 	num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
579 
580 	dma_offset = vdma[0];
581 
582 	/* Setup initial software IOMMU state. */
583 	spin_lock_init(&iommu->lock);
584 	iommu->ctx_lowest_free = 1;
585 	iommu->tbl.table_map_base = dma_offset;
586 	iommu->dma_addr_mask = dma_mask;
587 
588 	/* Allocate and initialize the free area map.  */
589 	sz = (num_tsb_entries + 7) / 8;
590 	sz = (sz + 7UL) & ~7UL;
591 	iommu->tbl.map = kzalloc(sz, GFP_KERNEL);
592 	if (!iommu->tbl.map) {
593 		printk(KERN_ERR PFX "Error, kmalloc(arena.map) failed.\n");
594 		return -ENOMEM;
595 	}
596 	iommu_tbl_pool_init(&iommu->tbl, num_tsb_entries, IO_PAGE_SHIFT,
597 			    NULL, false /* no large_pool */,
598 			    0 /* default npools */,
599 			    false /* want span boundary checking */);
600 	sz = probe_existing_entries(pbm, &iommu->tbl);
601 	if (sz)
602 		printk("%s: Imported %lu TSB entries from OBP\n",
603 		       pbm->name, sz);
604 
605 	return 0;
606 }
607 
608 #ifdef CONFIG_PCI_MSI
609 struct pci_sun4v_msiq_entry {
610 	u64		version_type;
611 #define MSIQ_VERSION_MASK		0xffffffff00000000UL
612 #define MSIQ_VERSION_SHIFT		32
613 #define MSIQ_TYPE_MASK			0x00000000000000ffUL
614 #define MSIQ_TYPE_SHIFT			0
615 #define MSIQ_TYPE_NONE			0x00
616 #define MSIQ_TYPE_MSG			0x01
617 #define MSIQ_TYPE_MSI32			0x02
618 #define MSIQ_TYPE_MSI64			0x03
619 #define MSIQ_TYPE_INTX			0x08
620 #define MSIQ_TYPE_NONE2			0xff
621 
622 	u64		intx_sysino;
623 	u64		reserved1;
624 	u64		stick;
625 	u64		req_id;  /* bus/device/func */
626 #define MSIQ_REQID_BUS_MASK		0xff00UL
627 #define MSIQ_REQID_BUS_SHIFT		8
628 #define MSIQ_REQID_DEVICE_MASK		0x00f8UL
629 #define MSIQ_REQID_DEVICE_SHIFT		3
630 #define MSIQ_REQID_FUNC_MASK		0x0007UL
631 #define MSIQ_REQID_FUNC_SHIFT		0
632 
633 	u64		msi_address;
634 
635 	/* The format of this value is message type dependent.
636 	 * For MSI bits 15:0 are the data from the MSI packet.
637 	 * For MSI-X bits 31:0 are the data from the MSI packet.
638 	 * For MSG, the message code and message routing code where:
639 	 * 	bits 39:32 is the bus/device/fn of the msg target-id
640 	 *	bits 18:16 is the message routing code
641 	 *	bits 7:0 is the message code
642 	 * For INTx the low order 2-bits are:
643 	 *	00 - INTA
644 	 *	01 - INTB
645 	 *	10 - INTC
646 	 *	11 - INTD
647 	 */
648 	u64		msi_data;
649 
650 	u64		reserved2;
651 };
652 
653 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
654 			      unsigned long *head)
655 {
656 	unsigned long err, limit;
657 
658 	err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
659 	if (unlikely(err))
660 		return -ENXIO;
661 
662 	limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
663 	if (unlikely(*head >= limit))
664 		return -EFBIG;
665 
666 	return 0;
667 }
668 
669 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
670 				 unsigned long msiqid, unsigned long *head,
671 				 unsigned long *msi)
672 {
673 	struct pci_sun4v_msiq_entry *ep;
674 	unsigned long err, type;
675 
676 	/* Note: void pointer arithmetic, 'head' is a byte offset  */
677 	ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
678 				 (pbm->msiq_ent_count *
679 				  sizeof(struct pci_sun4v_msiq_entry))) +
680 	      *head);
681 
682 	if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
683 		return 0;
684 
685 	type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
686 	if (unlikely(type != MSIQ_TYPE_MSI32 &&
687 		     type != MSIQ_TYPE_MSI64))
688 		return -EINVAL;
689 
690 	*msi = ep->msi_data;
691 
692 	err = pci_sun4v_msi_setstate(pbm->devhandle,
693 				     ep->msi_data /* msi_num */,
694 				     HV_MSISTATE_IDLE);
695 	if (unlikely(err))
696 		return -ENXIO;
697 
698 	/* Clear the entry.  */
699 	ep->version_type &= ~MSIQ_TYPE_MASK;
700 
701 	(*head) += sizeof(struct pci_sun4v_msiq_entry);
702 	if (*head >=
703 	    (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
704 		*head = 0;
705 
706 	return 1;
707 }
708 
709 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
710 			      unsigned long head)
711 {
712 	unsigned long err;
713 
714 	err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
715 	if (unlikely(err))
716 		return -EINVAL;
717 
718 	return 0;
719 }
720 
721 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
722 			       unsigned long msi, int is_msi64)
723 {
724 	if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
725 				  (is_msi64 ?
726 				   HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
727 		return -ENXIO;
728 	if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
729 		return -ENXIO;
730 	if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
731 		return -ENXIO;
732 	return 0;
733 }
734 
735 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
736 {
737 	unsigned long err, msiqid;
738 
739 	err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
740 	if (err)
741 		return -ENXIO;
742 
743 	pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
744 
745 	return 0;
746 }
747 
748 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
749 {
750 	unsigned long q_size, alloc_size, pages, order;
751 	int i;
752 
753 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
754 	alloc_size = (pbm->msiq_num * q_size);
755 	order = get_order(alloc_size);
756 	pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
757 	if (pages == 0UL) {
758 		printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
759 		       order);
760 		return -ENOMEM;
761 	}
762 	memset((char *)pages, 0, PAGE_SIZE << order);
763 	pbm->msi_queues = (void *) pages;
764 
765 	for (i = 0; i < pbm->msiq_num; i++) {
766 		unsigned long err, base = __pa(pages + (i * q_size));
767 		unsigned long ret1, ret2;
768 
769 		err = pci_sun4v_msiq_conf(pbm->devhandle,
770 					  pbm->msiq_first + i,
771 					  base, pbm->msiq_ent_count);
772 		if (err) {
773 			printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
774 			       err);
775 			goto h_error;
776 		}
777 
778 		err = pci_sun4v_msiq_info(pbm->devhandle,
779 					  pbm->msiq_first + i,
780 					  &ret1, &ret2);
781 		if (err) {
782 			printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
783 			       err);
784 			goto h_error;
785 		}
786 		if (ret1 != base || ret2 != pbm->msiq_ent_count) {
787 			printk(KERN_ERR "MSI: Bogus qconf "
788 			       "expected[%lx:%x] got[%lx:%lx]\n",
789 			       base, pbm->msiq_ent_count,
790 			       ret1, ret2);
791 			goto h_error;
792 		}
793 	}
794 
795 	return 0;
796 
797 h_error:
798 	free_pages(pages, order);
799 	return -EINVAL;
800 }
801 
802 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
803 {
804 	unsigned long q_size, alloc_size, pages, order;
805 	int i;
806 
807 	for (i = 0; i < pbm->msiq_num; i++) {
808 		unsigned long msiqid = pbm->msiq_first + i;
809 
810 		(void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
811 	}
812 
813 	q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
814 	alloc_size = (pbm->msiq_num * q_size);
815 	order = get_order(alloc_size);
816 
817 	pages = (unsigned long) pbm->msi_queues;
818 
819 	free_pages(pages, order);
820 
821 	pbm->msi_queues = NULL;
822 }
823 
824 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
825 				    unsigned long msiqid,
826 				    unsigned long devino)
827 {
828 	unsigned int irq = sun4v_build_irq(pbm->devhandle, devino);
829 
830 	if (!irq)
831 		return -ENOMEM;
832 
833 	if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
834 		return -EINVAL;
835 	if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
836 		return -EINVAL;
837 
838 	return irq;
839 }
840 
841 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
842 	.get_head	=	pci_sun4v_get_head,
843 	.dequeue_msi	=	pci_sun4v_dequeue_msi,
844 	.set_head	=	pci_sun4v_set_head,
845 	.msi_setup	=	pci_sun4v_msi_setup,
846 	.msi_teardown	=	pci_sun4v_msi_teardown,
847 	.msiq_alloc	=	pci_sun4v_msiq_alloc,
848 	.msiq_free	=	pci_sun4v_msiq_free,
849 	.msiq_build_irq	=	pci_sun4v_msiq_build_irq,
850 };
851 
852 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
853 {
854 	sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
855 }
856 #else /* CONFIG_PCI_MSI */
857 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
858 {
859 }
860 #endif /* !(CONFIG_PCI_MSI) */
861 
862 static int pci_sun4v_pbm_init(struct pci_pbm_info *pbm,
863 			      struct platform_device *op, u32 devhandle)
864 {
865 	struct device_node *dp = op->dev.of_node;
866 	int err;
867 
868 	pbm->numa_node = of_node_to_nid(dp);
869 
870 	pbm->pci_ops = &sun4v_pci_ops;
871 	pbm->config_space_reg_bits = 12;
872 
873 	pbm->index = pci_num_pbms++;
874 
875 	pbm->op = op;
876 
877 	pbm->devhandle = devhandle;
878 
879 	pbm->name = dp->full_name;
880 
881 	printk("%s: SUN4V PCI Bus Module\n", pbm->name);
882 	printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node);
883 
884 	pci_determine_mem_io_space(pbm);
885 
886 	pci_get_pbm_props(pbm);
887 
888 	err = pci_sun4v_iommu_init(pbm);
889 	if (err)
890 		return err;
891 
892 	pci_sun4v_msi_init(pbm);
893 
894 	pci_sun4v_scan_bus(pbm, &op->dev);
895 
896 	pbm->next = pci_pbm_root;
897 	pci_pbm_root = pbm;
898 
899 	return 0;
900 }
901 
902 static int pci_sun4v_probe(struct platform_device *op)
903 {
904 	const struct linux_prom64_registers *regs;
905 	static int hvapi_negotiated = 0;
906 	struct pci_pbm_info *pbm;
907 	struct device_node *dp;
908 	struct iommu *iommu;
909 	u32 devhandle;
910 	int i, err;
911 
912 	dp = op->dev.of_node;
913 
914 	if (!hvapi_negotiated++) {
915 		err = sun4v_hvapi_register(HV_GRP_PCI,
916 					   vpci_major,
917 					   &vpci_minor);
918 
919 		if (err) {
920 			printk(KERN_ERR PFX "Could not register hvapi, "
921 			       "err=%d\n", err);
922 			return err;
923 		}
924 		printk(KERN_INFO PFX "Registered hvapi major[%lu] minor[%lu]\n",
925 		       vpci_major, vpci_minor);
926 
927 		dma_ops = &sun4v_dma_ops;
928 	}
929 
930 	regs = of_get_property(dp, "reg", NULL);
931 	err = -ENODEV;
932 	if (!regs) {
933 		printk(KERN_ERR PFX "Could not find config registers\n");
934 		goto out_err;
935 	}
936 	devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
937 
938 	err = -ENOMEM;
939 	if (!iommu_batch_initialized) {
940 		for_each_possible_cpu(i) {
941 			unsigned long page = get_zeroed_page(GFP_KERNEL);
942 
943 			if (!page)
944 				goto out_err;
945 
946 			per_cpu(iommu_batch, i).pglist = (u64 *) page;
947 		}
948 		iommu_batch_initialized = 1;
949 	}
950 
951 	pbm = kzalloc(sizeof(*pbm), GFP_KERNEL);
952 	if (!pbm) {
953 		printk(KERN_ERR PFX "Could not allocate pci_pbm_info\n");
954 		goto out_err;
955 	}
956 
957 	iommu = kzalloc(sizeof(struct iommu), GFP_KERNEL);
958 	if (!iommu) {
959 		printk(KERN_ERR PFX "Could not allocate pbm iommu\n");
960 		goto out_free_controller;
961 	}
962 
963 	pbm->iommu = iommu;
964 
965 	err = pci_sun4v_pbm_init(pbm, op, devhandle);
966 	if (err)
967 		goto out_free_iommu;
968 
969 	dev_set_drvdata(&op->dev, pbm);
970 
971 	return 0;
972 
973 out_free_iommu:
974 	kfree(pbm->iommu);
975 
976 out_free_controller:
977 	kfree(pbm);
978 
979 out_err:
980 	return err;
981 }
982 
983 static const struct of_device_id pci_sun4v_match[] = {
984 	{
985 		.name = "pci",
986 		.compatible = "SUNW,sun4v-pci",
987 	},
988 	{},
989 };
990 
991 static struct platform_driver pci_sun4v_driver = {
992 	.driver = {
993 		.name = DRIVER_NAME,
994 		.of_match_table = pci_sun4v_match,
995 	},
996 	.probe		= pci_sun4v_probe,
997 };
998 
999 static int __init pci_sun4v_init(void)
1000 {
1001 	return platform_driver_register(&pci_sun4v_driver);
1002 }
1003 
1004 subsys_initcall(pci_sun4v_init);
1005