Merge remote-tracking branch 'moduleh/module.h-split'
[linux-2.6/next.git] / arch / sparc / kernel / pci_sun4v.c
blobb272cda35a0125904bc6f17eb1c5d14054dec24e
1 /* pci_sun4v.c: SUN4V specific PCI controller support.
3 * Copyright (C) 2006, 2007, 2008 David S. Miller (davem@davemloft.net)
4 */
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>
19 #include <asm/iommu.h>
20 #include <asm/irq.h>
21 #include <asm/hypervisor.h>
22 #include <asm/prom.h>
24 #include "pci_impl.h"
25 #include "iommu_common.h"
27 #include "pci_sun4v.h"
29 #define DRIVER_NAME "pci_sun4v"
30 #define PFX DRIVER_NAME ": "
32 static unsigned long vpci_major = 1;
33 static unsigned long vpci_minor = 1;
35 #define PGLIST_NENTS (PAGE_SIZE / sizeof(u64))
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. */
45 static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
46 static int iommu_batch_initialized;
48 /* Interrupts must be disabled. */
49 static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
51 struct iommu_batch *p = &__get_cpu_var(iommu_batch);
53 p->dev = dev;
54 p->prot = prot;
55 p->entry = entry;
56 p->npages = 0;
59 /* Interrupts must be disabled. */
60 static long iommu_batch_flush(struct iommu_batch *p)
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;
69 while (npages != 0) {
70 long num;
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;
84 entry += num;
85 npages -= num;
86 pglist += num;
89 p->entry = entry;
90 p->npages = 0;
92 return 0;
95 static inline void iommu_batch_new_entry(unsigned long entry)
97 struct iommu_batch *p = &__get_cpu_var(iommu_batch);
99 if (p->entry + p->npages == entry)
100 return;
101 if (p->entry != ~0UL)
102 iommu_batch_flush(p);
103 p->entry = entry;
106 /* Interrupts must be disabled. */
107 static inline long iommu_batch_add(u64 phys_page)
109 struct iommu_batch *p = &__get_cpu_var(iommu_batch);
111 BUG_ON(p->npages >= PGLIST_NENTS);
113 p->pglist[p->npages++] = phys_page;
114 if (p->npages == PGLIST_NENTS)
115 return iommu_batch_flush(p);
117 return 0;
120 /* Interrupts must be disabled. */
121 static inline long iommu_batch_end(void)
123 struct iommu_batch *p = &__get_cpu_var(iommu_batch);
125 BUG_ON(p->npages >= PGLIST_NENTS);
127 return iommu_batch_flush(p);
130 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
131 dma_addr_t *dma_addrp, gfp_t gfp)
133 unsigned long flags, order, first_page, npages, n;
134 struct iommu *iommu;
135 struct page *page;
136 void *ret;
137 long entry;
138 int nid;
140 size = IO_PAGE_ALIGN(size);
141 order = get_order(size);
142 if (unlikely(order >= MAX_ORDER))
143 return NULL;
145 npages = size >> IO_PAGE_SHIFT;
147 nid = dev->archdata.numa_node;
148 page = alloc_pages_node(nid, gfp, order);
149 if (unlikely(!page))
150 return NULL;
152 first_page = (unsigned long) page_address(page);
153 memset((char *)first_page, 0, PAGE_SIZE << order);
155 iommu = dev->archdata.iommu;
157 spin_lock_irqsave(&iommu->lock, flags);
158 entry = iommu_range_alloc(dev, iommu, npages, NULL);
159 spin_unlock_irqrestore(&iommu->lock, flags);
161 if (unlikely(entry == DMA_ERROR_CODE))
162 goto range_alloc_fail;
164 *dma_addrp = (iommu->page_table_map_base +
165 (entry << IO_PAGE_SHIFT));
166 ret = (void *) first_page;
167 first_page = __pa(first_page);
169 local_irq_save(flags);
171 iommu_batch_start(dev,
172 (HV_PCI_MAP_ATTR_READ |
173 HV_PCI_MAP_ATTR_WRITE),
174 entry);
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;
182 if (unlikely(iommu_batch_end() < 0L))
183 goto iommu_map_fail;
185 local_irq_restore(flags);
187 return ret;
189 iommu_map_fail:
190 /* Interrupts are disabled. */
191 spin_lock(&iommu->lock);
192 iommu_range_free(iommu, *dma_addrp, npages);
193 spin_unlock_irqrestore(&iommu->lock, flags);
195 range_alloc_fail:
196 free_pages(first_page, order);
197 return NULL;
200 static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
201 dma_addr_t dvma)
203 struct pci_pbm_info *pbm;
204 struct iommu *iommu;
205 unsigned long flags, order, npages, entry;
206 u32 devhandle;
208 npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
209 iommu = dev->archdata.iommu;
210 pbm = dev->archdata.host_controller;
211 devhandle = pbm->devhandle;
212 entry = ((dvma - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
214 spin_lock_irqsave(&iommu->lock, flags);
216 iommu_range_free(iommu, dvma, npages);
218 do {
219 unsigned long num;
221 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
222 npages);
223 entry += num;
224 npages -= num;
225 } while (npages != 0);
227 spin_unlock_irqrestore(&iommu->lock, flags);
229 order = get_order(size);
230 if (order < 10)
231 free_pages((unsigned long)cpu, order);
234 static dma_addr_t dma_4v_map_page(struct device *dev, struct page *page,
235 unsigned long offset, size_t sz,
236 enum dma_data_direction direction,
237 struct dma_attrs *attrs)
239 struct iommu *iommu;
240 unsigned long flags, npages, oaddr;
241 unsigned long i, base_paddr;
242 u32 bus_addr, ret;
243 unsigned long prot;
244 long entry;
246 iommu = dev->archdata.iommu;
248 if (unlikely(direction == DMA_NONE))
249 goto bad;
251 oaddr = (unsigned long)(page_address(page) + offset);
252 npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
253 npages >>= IO_PAGE_SHIFT;
255 spin_lock_irqsave(&iommu->lock, flags);
256 entry = iommu_range_alloc(dev, iommu, npages, NULL);
257 spin_unlock_irqrestore(&iommu->lock, flags);
259 if (unlikely(entry == DMA_ERROR_CODE))
260 goto bad;
262 bus_addr = (iommu->page_table_map_base +
263 (entry << IO_PAGE_SHIFT));
264 ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
265 base_paddr = __pa(oaddr & IO_PAGE_MASK);
266 prot = HV_PCI_MAP_ATTR_READ;
267 if (direction != DMA_TO_DEVICE)
268 prot |= HV_PCI_MAP_ATTR_WRITE;
270 local_irq_save(flags);
272 iommu_batch_start(dev, prot, entry);
274 for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
275 long err = iommu_batch_add(base_paddr);
276 if (unlikely(err < 0L))
277 goto iommu_map_fail;
279 if (unlikely(iommu_batch_end() < 0L))
280 goto iommu_map_fail;
282 local_irq_restore(flags);
284 return ret;
286 bad:
287 if (printk_ratelimit())
288 WARN_ON(1);
289 return DMA_ERROR_CODE;
291 iommu_map_fail:
292 /* Interrupts are disabled. */
293 spin_lock(&iommu->lock);
294 iommu_range_free(iommu, bus_addr, npages);
295 spin_unlock_irqrestore(&iommu->lock, flags);
297 return DMA_ERROR_CODE;
300 static void dma_4v_unmap_page(struct device *dev, dma_addr_t bus_addr,
301 size_t sz, enum dma_data_direction direction,
302 struct dma_attrs *attrs)
304 struct pci_pbm_info *pbm;
305 struct iommu *iommu;
306 unsigned long flags, npages;
307 long entry;
308 u32 devhandle;
310 if (unlikely(direction == DMA_NONE)) {
311 if (printk_ratelimit())
312 WARN_ON(1);
313 return;
316 iommu = dev->archdata.iommu;
317 pbm = dev->archdata.host_controller;
318 devhandle = pbm->devhandle;
320 npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
321 npages >>= IO_PAGE_SHIFT;
322 bus_addr &= IO_PAGE_MASK;
324 spin_lock_irqsave(&iommu->lock, flags);
326 iommu_range_free(iommu, bus_addr, npages);
328 entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
329 do {
330 unsigned long num;
332 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
333 npages);
334 entry += num;
335 npages -= num;
336 } while (npages != 0);
338 spin_unlock_irqrestore(&iommu->lock, flags);
341 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
342 int nelems, enum dma_data_direction direction,
343 struct dma_attrs *attrs)
345 struct scatterlist *s, *outs, *segstart;
346 unsigned long flags, handle, prot;
347 dma_addr_t dma_next = 0, dma_addr;
348 unsigned int max_seg_size;
349 unsigned long seg_boundary_size;
350 int outcount, incount, i;
351 struct iommu *iommu;
352 unsigned long base_shift;
353 long err;
355 BUG_ON(direction == DMA_NONE);
357 iommu = dev->archdata.iommu;
358 if (nelems == 0 || !iommu)
359 return 0;
361 prot = HV_PCI_MAP_ATTR_READ;
362 if (direction != DMA_TO_DEVICE)
363 prot |= HV_PCI_MAP_ATTR_WRITE;
365 outs = s = segstart = &sglist[0];
366 outcount = 1;
367 incount = nelems;
368 handle = 0;
370 /* Init first segment length for backout at failure */
371 outs->dma_length = 0;
373 spin_lock_irqsave(&iommu->lock, flags);
375 iommu_batch_start(dev, prot, ~0UL);
377 max_seg_size = dma_get_max_seg_size(dev);
378 seg_boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
379 IO_PAGE_SIZE) >> IO_PAGE_SHIFT;
380 base_shift = iommu->page_table_map_base >> IO_PAGE_SHIFT;
381 for_each_sg(sglist, s, nelems, i) {
382 unsigned long paddr, npages, entry, out_entry = 0, slen;
384 slen = s->length;
385 /* Sanity check */
386 if (slen == 0) {
387 dma_next = 0;
388 continue;
390 /* Allocate iommu entries for that segment */
391 paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
392 npages = iommu_num_pages(paddr, slen, IO_PAGE_SIZE);
393 entry = iommu_range_alloc(dev, iommu, npages, &handle);
395 /* Handle failure */
396 if (unlikely(entry == DMA_ERROR_CODE)) {
397 if (printk_ratelimit())
398 printk(KERN_INFO "iommu_alloc failed, iommu %p paddr %lx"
399 " npages %lx\n", iommu, paddr, npages);
400 goto iommu_map_failed;
403 iommu_batch_new_entry(entry);
405 /* Convert entry to a dma_addr_t */
406 dma_addr = iommu->page_table_map_base +
407 (entry << IO_PAGE_SHIFT);
408 dma_addr |= (s->offset & ~IO_PAGE_MASK);
410 /* Insert into HW table */
411 paddr &= IO_PAGE_MASK;
412 while (npages--) {
413 err = iommu_batch_add(paddr);
414 if (unlikely(err < 0L))
415 goto iommu_map_failed;
416 paddr += IO_PAGE_SIZE;
419 /* If we are in an open segment, try merging */
420 if (segstart != s) {
421 /* We cannot merge if:
422 * - allocated dma_addr isn't contiguous to previous allocation
424 if ((dma_addr != dma_next) ||
425 (outs->dma_length + s->length > max_seg_size) ||
426 (is_span_boundary(out_entry, base_shift,
427 seg_boundary_size, outs, s))) {
428 /* Can't merge: create a new segment */
429 segstart = s;
430 outcount++;
431 outs = sg_next(outs);
432 } else {
433 outs->dma_length += s->length;
437 if (segstart == s) {
438 /* This is a new segment, fill entries */
439 outs->dma_address = dma_addr;
440 outs->dma_length = slen;
441 out_entry = entry;
444 /* Calculate next page pointer for contiguous check */
445 dma_next = dma_addr + slen;
448 err = iommu_batch_end();
450 if (unlikely(err < 0L))
451 goto iommu_map_failed;
453 spin_unlock_irqrestore(&iommu->lock, flags);
455 if (outcount < incount) {
456 outs = sg_next(outs);
457 outs->dma_address = DMA_ERROR_CODE;
458 outs->dma_length = 0;
461 return outcount;
463 iommu_map_failed:
464 for_each_sg(sglist, s, nelems, i) {
465 if (s->dma_length != 0) {
466 unsigned long vaddr, npages;
468 vaddr = s->dma_address & IO_PAGE_MASK;
469 npages = iommu_num_pages(s->dma_address, s->dma_length,
470 IO_PAGE_SIZE);
471 iommu_range_free(iommu, vaddr, npages);
472 /* XXX demap? XXX */
473 s->dma_address = DMA_ERROR_CODE;
474 s->dma_length = 0;
476 if (s == outs)
477 break;
479 spin_unlock_irqrestore(&iommu->lock, flags);
481 return 0;
484 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
485 int nelems, enum dma_data_direction direction,
486 struct dma_attrs *attrs)
488 struct pci_pbm_info *pbm;
489 struct scatterlist *sg;
490 struct iommu *iommu;
491 unsigned long flags;
492 u32 devhandle;
494 BUG_ON(direction == DMA_NONE);
496 iommu = dev->archdata.iommu;
497 pbm = dev->archdata.host_controller;
498 devhandle = pbm->devhandle;
500 spin_lock_irqsave(&iommu->lock, flags);
502 sg = sglist;
503 while (nelems--) {
504 dma_addr_t dma_handle = sg->dma_address;
505 unsigned int len = sg->dma_length;
506 unsigned long npages, entry;
508 if (!len)
509 break;
510 npages = iommu_num_pages(dma_handle, len, IO_PAGE_SIZE);
511 iommu_range_free(iommu, dma_handle, npages);
513 entry = ((dma_handle - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
514 while (npages) {
515 unsigned long num;
517 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
518 npages);
519 entry += num;
520 npages -= num;
523 sg = sg_next(sg);
526 spin_unlock_irqrestore(&iommu->lock, flags);
529 static struct dma_map_ops sun4v_dma_ops = {
530 .alloc_coherent = dma_4v_alloc_coherent,
531 .free_coherent = dma_4v_free_coherent,
532 .map_page = dma_4v_map_page,
533 .unmap_page = dma_4v_unmap_page,
534 .map_sg = dma_4v_map_sg,
535 .unmap_sg = dma_4v_unmap_sg,
538 static void __devinit pci_sun4v_scan_bus(struct pci_pbm_info *pbm,
539 struct device *parent)
541 struct property *prop;
542 struct device_node *dp;
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);
549 /* XXX register error interrupt handlers XXX */
552 static unsigned long __devinit probe_existing_entries(struct pci_pbm_info *pbm,
553 struct iommu *iommu)
555 struct iommu_arena *arena = &iommu->arena;
556 unsigned long i, cnt = 0;
557 u32 devhandle;
559 devhandle = pbm->devhandle;
560 for (i = 0; i < arena->limit; i++) {
561 unsigned long ret, io_attrs, ra;
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);
577 return cnt;
580 static int __devinit pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
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;
588 vdma = of_get_property(pbm->op->dev.of_node, "virtual-dma", NULL);
589 if (!vdma)
590 vdma = vdma_default;
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;
598 dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
599 num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
601 dma_offset = vdma[0];
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;
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;
617 iommu->arena.limit = num_tsb_entries;
619 sz = probe_existing_entries(pbm, iommu);
620 if (sz)
621 printk("%s: Imported %lu TSB entries from OBP\n",
622 pbm->name, sz);
624 return 0;
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
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
652 u64 msi_address;
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
667 u64 msi_data;
669 u64 reserved2;
672 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
673 unsigned long *head)
675 unsigned long err, limit;
677 err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
678 if (unlikely(err))
679 return -ENXIO;
681 limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
682 if (unlikely(*head >= limit))
683 return -EFBIG;
685 return 0;
688 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
689 unsigned long msiqid, unsigned long *head,
690 unsigned long *msi)
692 struct pci_sun4v_msiq_entry *ep;
693 unsigned long err, type;
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);
701 if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
702 return 0;
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;
709 *msi = ep->msi_data;
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;
717 /* Clear the entry. */
718 ep->version_type &= ~MSIQ_TYPE_MASK;
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;
725 return 1;
728 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
729 unsigned long head)
731 unsigned long err;
733 err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
734 if (unlikely(err))
735 return -EINVAL;
737 return 0;
740 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
741 unsigned long msi, int is_msi64)
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;
754 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
756 unsigned long err, msiqid;
758 err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
759 if (err)
760 return -ENXIO;
762 pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
764 return 0;
767 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
769 unsigned long q_size, alloc_size, pages, order;
770 int i;
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;
781 memset((char *)pages, 0, PAGE_SIZE << order);
782 pbm->msi_queues = (void *) pages;
784 for (i = 0; i < pbm->msiq_num; i++) {
785 unsigned long err, base = __pa(pages + (i * q_size));
786 unsigned long ret1, ret2;
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;
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;
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;
814 return 0;
816 h_error:
817 free_pages(pages, order);
818 return -EINVAL;
821 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
823 unsigned long q_size, alloc_size, pages, order;
824 int i;
826 for (i = 0; i < pbm->msiq_num; i++) {
827 unsigned long msiqid = pbm->msiq_first + i;
829 (void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
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);
836 pages = (unsigned long) pbm->msi_queues;
838 free_pages(pages, order);
840 pbm->msi_queues = NULL;
843 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
844 unsigned long msiqid,
845 unsigned long devino)
847 unsigned int irq = sun4v_build_irq(pbm->devhandle, devino);
849 if (!irq)
850 return -ENOMEM;
852 if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
853 return -EINVAL;
854 if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
855 return -EINVAL;
857 return irq;
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,
871 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
873 sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
875 #else /* CONFIG_PCI_MSI */
876 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
879 #endif /* !(CONFIG_PCI_MSI) */
881 static int __devinit pci_sun4v_pbm_init(struct pci_pbm_info *pbm,
882 struct platform_device *op, u32 devhandle)
884 struct device_node *dp = op->dev.of_node;
885 int err;
887 pbm->numa_node = of_node_to_nid(dp);
889 pbm->pci_ops = &sun4v_pci_ops;
890 pbm->config_space_reg_bits = 12;
892 pbm->index = pci_num_pbms++;
894 pbm->op = op;
896 pbm->devhandle = devhandle;
898 pbm->name = dp->full_name;
900 printk("%s: SUN4V PCI Bus Module\n", pbm->name);
901 printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node);
903 pci_determine_mem_io_space(pbm);
905 pci_get_pbm_props(pbm);
907 err = pci_sun4v_iommu_init(pbm);
908 if (err)
909 return err;
911 pci_sun4v_msi_init(pbm);
913 pci_sun4v_scan_bus(pbm, &op->dev);
915 pbm->next = pci_pbm_root;
916 pci_pbm_root = pbm;
918 return 0;
921 static int __devinit pci_sun4v_probe(struct platform_device *op)
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;
931 dp = op->dev.of_node;
933 if (!hvapi_negotiated++) {
934 err = sun4v_hvapi_register(HV_GRP_PCI,
935 vpci_major,
936 &vpci_minor);
938 if (err) {
939 printk(KERN_ERR PFX "Could not register hvapi, "
940 "err=%d\n", err);
941 return err;
943 printk(KERN_INFO PFX "Registered hvapi major[%lu] minor[%lu]\n",
944 vpci_major, vpci_minor);
946 dma_ops = &sun4v_dma_ops;
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;
955 devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
957 err = -ENOMEM;
958 if (!iommu_batch_initialized) {
959 for_each_possible_cpu(i) {
960 unsigned long page = get_zeroed_page(GFP_KERNEL);
962 if (!page)
963 goto out_err;
965 per_cpu(iommu_batch, i).pglist = (u64 *) page;
967 iommu_batch_initialized = 1;
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;
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;
982 pbm->iommu = iommu;
984 err = pci_sun4v_pbm_init(pbm, op, devhandle);
985 if (err)
986 goto out_free_iommu;
988 dev_set_drvdata(&op->dev, pbm);
990 return 0;
992 out_free_iommu:
993 kfree(pbm->iommu);
995 out_free_controller:
996 kfree(pbm);
998 out_err:
999 return err;
1002 static const struct of_device_id pci_sun4v_match[] = {
1004 .name = "pci",
1005 .compatible = "SUNW,sun4v-pci",
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,
1016 .probe = pci_sun4v_probe,
1019 static int __init pci_sun4v_init(void)
1021 return platform_driver_register(&pci_sun4v_driver);
1024 subsys_initcall(pci_sun4v_init);