Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6] / arch / sparc64 / kernel / pci_sun4v.c
1 /* pci_sun4v.c: SUN4V specific PCI controller support.
2  *
3  * Copyright (C) 2006, 2007 David S. Miller (davem@davemloft.net)
4  */
5
6 #include <linux/kernel.h>
7 #include <linux/types.h>
8 #include <linux/pci.h>
9 #include <linux/init.h>
10 #include <linux/slab.h>
11 #include <linux/interrupt.h>
12 #include <linux/percpu.h>
13 #include <linux/irq.h>
14 #include <linux/msi.h>
15 #include <linux/log2.h>
16
17 #include <asm/iommu.h>
18 #include <asm/irq.h>
19 #include <asm/upa.h>
20 #include <asm/pstate.h>
21 #include <asm/oplib.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 static unsigned long vpci_major = 1;
31 static unsigned long vpci_minor = 1;
32
33 #define PGLIST_NENTS    (PAGE_SIZE / sizeof(u64))
34
35 struct iommu_batch {
36         struct device   *dev;           /* Device mapping is for.       */
37         unsigned long   prot;           /* IOMMU page protections       */
38         unsigned long   entry;          /* Index into IOTSB.            */
39         u64             *pglist;        /* List of physical pages       */
40         unsigned long   npages;         /* Number of pages in list.     */
41 };
42
43 static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
44
45 /* Interrupts must be disabled.  */
46 static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
47 {
48         struct iommu_batch *p = &__get_cpu_var(iommu_batch);
49
50         p->dev          = dev;
51         p->prot         = prot;
52         p->entry        = entry;
53         p->npages       = 0;
54 }
55
56 /* Interrupts must be disabled.  */
57 static long iommu_batch_flush(struct iommu_batch *p)
58 {
59         struct pci_pbm_info *pbm = p->dev->archdata.host_controller;
60         unsigned long devhandle = pbm->devhandle;
61         unsigned long prot = p->prot;
62         unsigned long entry = p->entry;
63         u64 *pglist = p->pglist;
64         unsigned long npages = p->npages;
65
66         while (npages != 0) {
67                 long num;
68
69                 num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
70                                           npages, prot, __pa(pglist));
71                 if (unlikely(num < 0)) {
72                         if (printk_ratelimit())
73                                 printk("iommu_batch_flush: IOMMU map of "
74                                        "[%08lx:%08lx:%lx:%lx:%lx] failed with "
75                                        "status %ld\n",
76                                        devhandle, HV_PCI_TSBID(0, entry),
77                                        npages, prot, __pa(pglist), num);
78                         return -1;
79                 }
80
81                 entry += num;
82                 npages -= num;
83                 pglist += num;
84         }
85
86         p->entry = entry;
87         p->npages = 0;
88
89         return 0;
90 }
91
92 /* Interrupts must be disabled.  */
93 static inline long iommu_batch_add(u64 phys_page)
94 {
95         struct iommu_batch *p = &__get_cpu_var(iommu_batch);
96
97         BUG_ON(p->npages >= PGLIST_NENTS);
98
99         p->pglist[p->npages++] = phys_page;
100         if (p->npages == PGLIST_NENTS)
101                 return iommu_batch_flush(p);
102
103         return 0;
104 }
105
106 /* Interrupts must be disabled.  */
107 static inline long iommu_batch_end(void)
108 {
109         struct iommu_batch *p = &__get_cpu_var(iommu_batch);
110
111         BUG_ON(p->npages >= PGLIST_NENTS);
112
113         return iommu_batch_flush(p);
114 }
115
116 static long arena_alloc(struct iommu_arena *arena, unsigned long npages)
117 {
118         unsigned long n, i, start, end, limit;
119         int pass;
120
121         limit = arena->limit;
122         start = arena->hint;
123         pass = 0;
124
125 again:
126         n = find_next_zero_bit(arena->map, limit, start);
127         end = n + npages;
128         if (unlikely(end >= limit)) {
129                 if (likely(pass < 1)) {
130                         limit = start;
131                         start = 0;
132                         pass++;
133                         goto again;
134                 } else {
135                         /* Scanned the whole thing, give up. */
136                         return -1;
137                 }
138         }
139
140         for (i = n; i < end; i++) {
141                 if (test_bit(i, arena->map)) {
142                         start = i + 1;
143                         goto again;
144                 }
145         }
146
147         for (i = n; i < end; i++)
148                 __set_bit(i, arena->map);
149
150         arena->hint = end;
151
152         return n;
153 }
154
155 static void arena_free(struct iommu_arena *arena, unsigned long base,
156                        unsigned long npages)
157 {
158         unsigned long i;
159
160         for (i = base; i < (base + npages); i++)
161                 __clear_bit(i, arena->map);
162 }
163
164 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
165                                    dma_addr_t *dma_addrp, gfp_t gfp)
166 {
167         struct iommu *iommu;
168         unsigned long flags, order, first_page, npages, n;
169         void *ret;
170         long entry;
171
172         size = IO_PAGE_ALIGN(size);
173         order = get_order(size);
174         if (unlikely(order >= MAX_ORDER))
175                 return NULL;
176
177         npages = size >> IO_PAGE_SHIFT;
178
179         first_page = __get_free_pages(gfp, order);
180         if (unlikely(first_page == 0UL))
181                 return NULL;
182
183         memset((char *)first_page, 0, PAGE_SIZE << order);
184
185         iommu = dev->archdata.iommu;
186
187         spin_lock_irqsave(&iommu->lock, flags);
188         entry = arena_alloc(&iommu->arena, npages);
189         spin_unlock_irqrestore(&iommu->lock, flags);
190
191         if (unlikely(entry < 0L))
192                 goto arena_alloc_fail;
193
194         *dma_addrp = (iommu->page_table_map_base +
195                       (entry << IO_PAGE_SHIFT));
196         ret = (void *) first_page;
197         first_page = __pa(first_page);
198
199         local_irq_save(flags);
200
201         iommu_batch_start(dev,
202                           (HV_PCI_MAP_ATTR_READ |
203                            HV_PCI_MAP_ATTR_WRITE),
204                           entry);
205
206         for (n = 0; n < npages; n++) {
207                 long err = iommu_batch_add(first_page + (n * PAGE_SIZE));
208                 if (unlikely(err < 0L))
209                         goto iommu_map_fail;
210         }
211
212         if (unlikely(iommu_batch_end() < 0L))
213                 goto iommu_map_fail;
214
215         local_irq_restore(flags);
216
217         return ret;
218
219 iommu_map_fail:
220         /* Interrupts are disabled.  */
221         spin_lock(&iommu->lock);
222         arena_free(&iommu->arena, entry, npages);
223         spin_unlock_irqrestore(&iommu->lock, flags);
224
225 arena_alloc_fail:
226         free_pages(first_page, order);
227         return NULL;
228 }
229
230 static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
231                                  dma_addr_t dvma)
232 {
233         struct pci_pbm_info *pbm;
234         struct iommu *iommu;
235         unsigned long flags, order, npages, entry;
236         u32 devhandle;
237
238         npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
239         iommu = dev->archdata.iommu;
240         pbm = dev->archdata.host_controller;
241         devhandle = pbm->devhandle;
242         entry = ((dvma - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
243
244         spin_lock_irqsave(&iommu->lock, flags);
245
246         arena_free(&iommu->arena, entry, npages);
247
248         do {
249                 unsigned long num;
250
251                 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
252                                             npages);
253                 entry += num;
254                 npages -= num;
255         } while (npages != 0);
256
257         spin_unlock_irqrestore(&iommu->lock, flags);
258
259         order = get_order(size);
260         if (order < 10)
261                 free_pages((unsigned long)cpu, order);
262 }
263
264 static dma_addr_t dma_4v_map_single(struct device *dev, void *ptr, size_t sz,
265                                     enum dma_data_direction direction)
266 {
267         struct iommu *iommu;
268         unsigned long flags, npages, oaddr;
269         unsigned long i, base_paddr;
270         u32 bus_addr, ret;
271         unsigned long prot;
272         long entry;
273
274         iommu = dev->archdata.iommu;
275
276         if (unlikely(direction == DMA_NONE))
277                 goto bad;
278
279         oaddr = (unsigned long)ptr;
280         npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
281         npages >>= IO_PAGE_SHIFT;
282
283         spin_lock_irqsave(&iommu->lock, flags);
284         entry = arena_alloc(&iommu->arena, npages);
285         spin_unlock_irqrestore(&iommu->lock, flags);
286
287         if (unlikely(entry < 0L))
288                 goto bad;
289
290         bus_addr = (iommu->page_table_map_base +
291                     (entry << IO_PAGE_SHIFT));
292         ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
293         base_paddr = __pa(oaddr & IO_PAGE_MASK);
294         prot = HV_PCI_MAP_ATTR_READ;
295         if (direction != DMA_TO_DEVICE)
296                 prot |= HV_PCI_MAP_ATTR_WRITE;
297
298         local_irq_save(flags);
299
300         iommu_batch_start(dev, prot, entry);
301
302         for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
303                 long err = iommu_batch_add(base_paddr);
304                 if (unlikely(err < 0L))
305                         goto iommu_map_fail;
306         }
307         if (unlikely(iommu_batch_end() < 0L))
308                 goto iommu_map_fail;
309
310         local_irq_restore(flags);
311
312         return ret;
313
314 bad:
315         if (printk_ratelimit())
316                 WARN_ON(1);
317         return DMA_ERROR_CODE;
318
319 iommu_map_fail:
320         /* Interrupts are disabled.  */
321         spin_lock(&iommu->lock);
322         arena_free(&iommu->arena, entry, npages);
323         spin_unlock_irqrestore(&iommu->lock, flags);
324
325         return DMA_ERROR_CODE;
326 }
327
328 static void dma_4v_unmap_single(struct device *dev, dma_addr_t bus_addr,
329                                 size_t sz, enum dma_data_direction direction)
330 {
331         struct pci_pbm_info *pbm;
332         struct iommu *iommu;
333         unsigned long flags, npages;
334         long entry;
335         u32 devhandle;
336
337         if (unlikely(direction == DMA_NONE)) {
338                 if (printk_ratelimit())
339                         WARN_ON(1);
340                 return;
341         }
342
343         iommu = dev->archdata.iommu;
344         pbm = dev->archdata.host_controller;
345         devhandle = pbm->devhandle;
346
347         npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
348         npages >>= IO_PAGE_SHIFT;
349         bus_addr &= IO_PAGE_MASK;
350
351         spin_lock_irqsave(&iommu->lock, flags);
352
353         entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
354         arena_free(&iommu->arena, entry, npages);
355
356         do {
357                 unsigned long num;
358
359                 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
360                                             npages);
361                 entry += num;
362                 npages -= num;
363         } while (npages != 0);
364
365         spin_unlock_irqrestore(&iommu->lock, flags);
366 }
367
368 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
369                          int nelems, enum dma_data_direction direction)
370 {
371         unsigned long flags, npages, i, prot;
372         struct scatterlist *sg;
373         struct iommu *iommu;
374         long entry, err;
375         u32 dma_base;
376
377         /* Fast path single entry scatterlists. */
378         if (nelems == 1) {
379                 sglist->dma_address =
380                         dma_4v_map_single(dev, sg_virt(sglist),
381                                           sglist->length, direction);
382                 if (unlikely(sglist->dma_address == DMA_ERROR_CODE))
383                         return 0;
384                 sglist->dma_length = sglist->length;
385                 return 1;
386         }
387
388         iommu = dev->archdata.iommu;
389         
390         if (unlikely(direction == DMA_NONE))
391                 goto bad;
392
393         npages = calc_npages(sglist, nelems);
394
395         spin_lock_irqsave(&iommu->lock, flags);
396         entry = arena_alloc(&iommu->arena, npages);
397         spin_unlock_irqrestore(&iommu->lock, flags);
398
399         if (unlikely(entry < 0L))
400                 goto bad;
401
402         dma_base = iommu->page_table_map_base +
403                 (entry << IO_PAGE_SHIFT);
404
405         prot = HV_PCI_MAP_ATTR_READ;
406         if (direction != DMA_TO_DEVICE)
407                 prot |= HV_PCI_MAP_ATTR_WRITE;
408
409         local_irq_save(flags);
410
411         iommu_batch_start(dev, prot, entry);
412
413         for_each_sg(sglist, sg, nelems, i) {
414                 unsigned long paddr = SG_ENT_PHYS_ADDRESS(sg);
415                 unsigned long slen = sg->length;
416                 unsigned long this_npages;
417
418                 this_npages = iommu_num_pages(paddr, slen);
419
420                 sg->dma_address = dma_base | (paddr & ~IO_PAGE_MASK);
421                 sg->dma_length = slen;
422
423                 paddr &= IO_PAGE_MASK;
424                 while (this_npages--) {
425                         err = iommu_batch_add(paddr);
426                         if (unlikely(err < 0L)) {
427                                 local_irq_restore(flags);
428                                 goto iommu_map_failed;
429                         }
430
431                         paddr += IO_PAGE_SIZE;
432                         dma_base += IO_PAGE_SIZE;
433                 }
434         }
435
436         err = iommu_batch_end();
437
438         local_irq_restore(flags);
439
440         if (unlikely(err < 0L))
441                 goto iommu_map_failed;
442
443         return nelems;
444
445 bad:
446         if (printk_ratelimit())
447                 WARN_ON(1);
448         return 0;
449
450 iommu_map_failed:
451         spin_lock_irqsave(&iommu->lock, flags);
452         arena_free(&iommu->arena, entry, npages);
453         spin_unlock_irqrestore(&iommu->lock, flags);
454
455         return 0;
456 }
457
458 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
459                             int nelems, enum dma_data_direction direction)
460 {
461         unsigned long flags, npages;
462         struct pci_pbm_info *pbm;
463         u32 devhandle, bus_addr;
464         struct iommu *iommu;
465         long entry;
466
467         if (unlikely(direction == DMA_NONE)) {
468                 if (printk_ratelimit())
469                         WARN_ON(1);
470         }
471
472         iommu = dev->archdata.iommu;
473         pbm = dev->archdata.host_controller;
474         devhandle = pbm->devhandle;
475         
476         bus_addr = sglist->dma_address & IO_PAGE_MASK;
477
478         npages = calc_npages(sglist, nelems);
479
480         entry = ((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
481
482         spin_lock_irqsave(&iommu->lock, flags);
483
484         arena_free(&iommu->arena, entry, npages);
485
486         do {
487                 unsigned long num;
488
489                 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
490                                             npages);
491                 entry += num;
492                 npages -= num;
493         } while (npages != 0);
494
495         spin_unlock_irqrestore(&iommu->lock, flags);
496 }
497
498 static void dma_4v_sync_single_for_cpu(struct device *dev,
499                                        dma_addr_t bus_addr, size_t sz,
500                                        enum dma_data_direction direction)
501 {
502         /* Nothing to do... */
503 }
504
505 static void dma_4v_sync_sg_for_cpu(struct device *dev,
506                                    struct scatterlist *sglist, int nelems,
507                                    enum dma_data_direction direction)
508 {
509         /* Nothing to do... */
510 }
511
512 const struct dma_ops sun4v_dma_ops = {
513         .alloc_coherent                 = dma_4v_alloc_coherent,
514         .free_coherent                  = dma_4v_free_coherent,
515         .map_single                     = dma_4v_map_single,
516         .unmap_single                   = dma_4v_unmap_single,
517         .map_sg                         = dma_4v_map_sg,
518         .unmap_sg                       = dma_4v_unmap_sg,
519         .sync_single_for_cpu            = dma_4v_sync_single_for_cpu,
520         .sync_sg_for_cpu                = dma_4v_sync_sg_for_cpu,
521 };
522
523 static void __init pci_sun4v_scan_bus(struct pci_pbm_info *pbm)
524 {
525         struct property *prop;
526         struct device_node *dp;
527
528         dp = pbm->prom_node;
529         prop = of_find_property(dp, "66mhz-capable", NULL);
530         pbm->is_66mhz_capable = (prop != NULL);
531         pbm->pci_bus = pci_scan_one_pbm(pbm);
532
533         /* XXX register error interrupt handlers XXX */
534 }
535
536 static unsigned long __init probe_existing_entries(struct pci_pbm_info *pbm,
537                                                    struct iommu *iommu)
538 {
539         struct iommu_arena *arena = &iommu->arena;
540         unsigned long i, cnt = 0;
541         u32 devhandle;
542
543         devhandle = pbm->devhandle;
544         for (i = 0; i < arena->limit; i++) {
545                 unsigned long ret, io_attrs, ra;
546
547                 ret = pci_sun4v_iommu_getmap(devhandle,
548                                              HV_PCI_TSBID(0, i),
549                                              &io_attrs, &ra);
550                 if (ret == HV_EOK) {
551                         if (page_in_phys_avail(ra)) {
552                                 pci_sun4v_iommu_demap(devhandle,
553                                                       HV_PCI_TSBID(0, i), 1);
554                         } else {
555                                 cnt++;
556                                 __set_bit(i, arena->map);
557                         }
558                 }
559         }
560
561         return cnt;
562 }
563
564 static void __init pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
565 {
566         struct iommu *iommu = pbm->iommu;
567         struct property *prop;
568         unsigned long num_tsb_entries, sz, tsbsize;
569         u32 vdma[2], dma_mask, dma_offset;
570
571         prop = of_find_property(pbm->prom_node, "virtual-dma", NULL);
572         if (prop) {
573                 u32 *val = prop->value;
574
575                 vdma[0] = val[0];
576                 vdma[1] = val[1];
577         } else {
578                 /* No property, use default values. */
579                 vdma[0] = 0x80000000;
580                 vdma[1] = 0x80000000;
581         }
582
583         if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) {
584                 prom_printf("PCI-SUN4V: strange virtual-dma[%08x:%08x].\n",
585                             vdma[0], vdma[1]);
586                 prom_halt();
587         };
588
589         dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
590         num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
591         tsbsize = num_tsb_entries * sizeof(iopte_t);
592
593         dma_offset = vdma[0];
594
595         /* Setup initial software IOMMU state. */
596         spin_lock_init(&iommu->lock);
597         iommu->ctx_lowest_free = 1;
598         iommu->page_table_map_base = dma_offset;
599         iommu->dma_addr_mask = dma_mask;
600
601         /* Allocate and initialize the free area map.  */
602         sz = (num_tsb_entries + 7) / 8;
603         sz = (sz + 7UL) & ~7UL;
604         iommu->arena.map = kzalloc(sz, GFP_KERNEL);
605         if (!iommu->arena.map) {
606                 prom_printf("PCI_IOMMU: Error, kmalloc(arena.map) failed.\n");
607                 prom_halt();
608         }
609         iommu->arena.limit = num_tsb_entries;
610
611         sz = probe_existing_entries(pbm, iommu);
612         if (sz)
613                 printk("%s: Imported %lu TSB entries from OBP\n",
614                        pbm->name, sz);
615 }
616
617 #ifdef CONFIG_PCI_MSI
618 struct pci_sun4v_msiq_entry {
619         u64             version_type;
620 #define MSIQ_VERSION_MASK               0xffffffff00000000UL
621 #define MSIQ_VERSION_SHIFT              32
622 #define MSIQ_TYPE_MASK                  0x00000000000000ffUL
623 #define MSIQ_TYPE_SHIFT                 0
624 #define MSIQ_TYPE_NONE                  0x00
625 #define MSIQ_TYPE_MSG                   0x01
626 #define MSIQ_TYPE_MSI32                 0x02
627 #define MSIQ_TYPE_MSI64                 0x03
628 #define MSIQ_TYPE_INTX                  0x08
629 #define MSIQ_TYPE_NONE2                 0xff
630
631         u64             intx_sysino;
632         u64             reserved1;
633         u64             stick;
634         u64             req_id;  /* bus/device/func */
635 #define MSIQ_REQID_BUS_MASK             0xff00UL
636 #define MSIQ_REQID_BUS_SHIFT            8
637 #define MSIQ_REQID_DEVICE_MASK          0x00f8UL
638 #define MSIQ_REQID_DEVICE_SHIFT         3
639 #define MSIQ_REQID_FUNC_MASK            0x0007UL
640 #define MSIQ_REQID_FUNC_SHIFT           0
641
642         u64             msi_address;
643
644         /* The format of this value is message type dependent.
645          * For MSI bits 15:0 are the data from the MSI packet.
646          * For MSI-X bits 31:0 are the data from the MSI packet.
647          * For MSG, the message code and message routing code where:
648          *      bits 39:32 is the bus/device/fn of the msg target-id
649          *      bits 18:16 is the message routing code
650          *      bits 7:0 is the message code
651          * For INTx the low order 2-bits are:
652          *      00 - INTA
653          *      01 - INTB
654          *      10 - INTC
655          *      11 - INTD
656          */
657         u64             msi_data;
658
659         u64             reserved2;
660 };
661
662 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
663                               unsigned long *head)
664 {
665         unsigned long err, limit;
666
667         err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
668         if (unlikely(err))
669                 return -ENXIO;
670
671         limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
672         if (unlikely(*head >= limit))
673                 return -EFBIG;
674
675         return 0;
676 }
677
678 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
679                                  unsigned long msiqid, unsigned long *head,
680                                  unsigned long *msi)
681 {
682         struct pci_sun4v_msiq_entry *ep;
683         unsigned long err, type;
684
685         /* Note: void pointer arithmetic, 'head' is a byte offset  */
686         ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
687                                  (pbm->msiq_ent_count *
688                                   sizeof(struct pci_sun4v_msiq_entry))) +
689               *head);
690
691         if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
692                 return 0;
693
694         type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
695         if (unlikely(type != MSIQ_TYPE_MSI32 &&
696                      type != MSIQ_TYPE_MSI64))
697                 return -EINVAL;
698
699         *msi = ep->msi_data;
700
701         err = pci_sun4v_msi_setstate(pbm->devhandle,
702                                      ep->msi_data /* msi_num */,
703                                      HV_MSISTATE_IDLE);
704         if (unlikely(err))
705                 return -ENXIO;
706
707         /* Clear the entry.  */
708         ep->version_type &= ~MSIQ_TYPE_MASK;
709
710         (*head) += sizeof(struct pci_sun4v_msiq_entry);
711         if (*head >=
712             (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
713                 *head = 0;
714
715         return 1;
716 }
717
718 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
719                               unsigned long head)
720 {
721         unsigned long err;
722
723         err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
724         if (unlikely(err))
725                 return -EINVAL;
726
727         return 0;
728 }
729
730 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
731                                unsigned long msi, int is_msi64)
732 {
733         if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
734                                   (is_msi64 ?
735                                    HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
736                 return -ENXIO;
737         if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
738                 return -ENXIO;
739         if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
740                 return -ENXIO;
741         return 0;
742 }
743
744 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
745 {
746         unsigned long err, msiqid;
747
748         err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
749         if (err)
750                 return -ENXIO;
751
752         pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
753
754         return 0;
755 }
756
757 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
758 {
759         unsigned long q_size, alloc_size, pages, order;
760         int i;
761
762         q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
763         alloc_size = (pbm->msiq_num * q_size);
764         order = get_order(alloc_size);
765         pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
766         if (pages == 0UL) {
767                 printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
768                        order);
769                 return -ENOMEM;
770         }
771         memset((char *)pages, 0, PAGE_SIZE << order);
772         pbm->msi_queues = (void *) pages;
773
774         for (i = 0; i < pbm->msiq_num; i++) {
775                 unsigned long err, base = __pa(pages + (i * q_size));
776                 unsigned long ret1, ret2;
777
778                 err = pci_sun4v_msiq_conf(pbm->devhandle,
779                                           pbm->msiq_first + i,
780                                           base, pbm->msiq_ent_count);
781                 if (err) {
782                         printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
783                                err);
784                         goto h_error;
785                 }
786
787                 err = pci_sun4v_msiq_info(pbm->devhandle,
788                                           pbm->msiq_first + i,
789                                           &ret1, &ret2);
790                 if (err) {
791                         printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
792                                err);
793                         goto h_error;
794                 }
795                 if (ret1 != base || ret2 != pbm->msiq_ent_count) {
796                         printk(KERN_ERR "MSI: Bogus qconf "
797                                "expected[%lx:%x] got[%lx:%lx]\n",
798                                base, pbm->msiq_ent_count,
799                                ret1, ret2);
800                         goto h_error;
801                 }
802         }
803
804         return 0;
805
806 h_error:
807         free_pages(pages, order);
808         return -EINVAL;
809 }
810
811 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
812 {
813         unsigned long q_size, alloc_size, pages, order;
814         int i;
815
816         for (i = 0; i < pbm->msiq_num; i++) {
817                 unsigned long msiqid = pbm->msiq_first + i;
818
819                 (void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
820         }
821
822         q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
823         alloc_size = (pbm->msiq_num * q_size);
824         order = get_order(alloc_size);
825
826         pages = (unsigned long) pbm->msi_queues;
827
828         free_pages(pages, order);
829
830         pbm->msi_queues = NULL;
831 }
832
833 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
834                                     unsigned long msiqid,
835                                     unsigned long devino)
836 {
837         unsigned int virt_irq = sun4v_build_irq(pbm->devhandle, devino);
838
839         if (!virt_irq)
840                 return -ENOMEM;
841
842         if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
843                 return -EINVAL;
844         if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
845                 return -EINVAL;
846
847         return virt_irq;
848 }
849
850 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
851         .get_head       =       pci_sun4v_get_head,
852         .dequeue_msi    =       pci_sun4v_dequeue_msi,
853         .set_head       =       pci_sun4v_set_head,
854         .msi_setup      =       pci_sun4v_msi_setup,
855         .msi_teardown   =       pci_sun4v_msi_teardown,
856         .msiq_alloc     =       pci_sun4v_msiq_alloc,
857         .msiq_free      =       pci_sun4v_msiq_free,
858         .msiq_build_irq =       pci_sun4v_msiq_build_irq,
859 };
860
861 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
862 {
863         sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
864 }
865 #else /* CONFIG_PCI_MSI */
866 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
867 {
868 }
869 #endif /* !(CONFIG_PCI_MSI) */
870
871 static void __init pci_sun4v_pbm_init(struct pci_controller_info *p,
872                                       struct device_node *dp, u32 devhandle)
873 {
874         struct pci_pbm_info *pbm;
875
876         if (devhandle & 0x40)
877                 pbm = &p->pbm_B;
878         else
879                 pbm = &p->pbm_A;
880
881         pbm->next = pci_pbm_root;
882         pci_pbm_root = pbm;
883
884         pbm->scan_bus = pci_sun4v_scan_bus;
885         pbm->pci_ops = &sun4v_pci_ops;
886         pbm->config_space_reg_bits = 12;
887
888         pbm->index = pci_num_pbms++;
889
890         pbm->parent = p;
891         pbm->prom_node = dp;
892
893         pbm->devhandle = devhandle;
894
895         pbm->name = dp->full_name;
896
897         printk("%s: SUN4V PCI Bus Module\n", pbm->name);
898
899         pci_determine_mem_io_space(pbm);
900
901         pci_get_pbm_props(pbm);
902         pci_sun4v_iommu_init(pbm);
903         pci_sun4v_msi_init(pbm);
904 }
905
906 void __init sun4v_pci_init(struct device_node *dp, char *model_name)
907 {
908         static int hvapi_negotiated = 0;
909         struct pci_controller_info *p;
910         struct pci_pbm_info *pbm;
911         struct iommu *iommu;
912         struct property *prop;
913         struct linux_prom64_registers *regs;
914         u32 devhandle;
915         int i;
916
917         if (!hvapi_negotiated++) {
918                 int err = sun4v_hvapi_register(HV_GRP_PCI,
919                                                vpci_major,
920                                                &vpci_minor);
921
922                 if (err) {
923                         prom_printf("SUN4V_PCI: Could not register hvapi, "
924                                     "err=%d\n", err);
925                         prom_halt();
926                 }
927                 printk("SUN4V_PCI: Registered hvapi major[%lu] minor[%lu]\n",
928                        vpci_major, vpci_minor);
929
930                 dma_ops = &sun4v_dma_ops;
931         }
932
933         prop = of_find_property(dp, "reg", NULL);
934         if (!prop) {
935                 prom_printf("SUN4V_PCI: Could not find config registers\n");
936                 prom_halt();
937         }
938         regs = prop->value;
939
940         devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
941
942         for (pbm = pci_pbm_root; pbm; pbm = pbm->next) {
943                 if (pbm->devhandle == (devhandle ^ 0x40)) {
944                         pci_sun4v_pbm_init(pbm->parent, dp, devhandle);
945                         return;
946                 }
947         }
948
949         for_each_possible_cpu(i) {
950                 unsigned long page = get_zeroed_page(GFP_ATOMIC);
951
952                 if (!page)
953                         goto fatal_memory_error;
954
955                 per_cpu(iommu_batch, i).pglist = (u64 *) page;
956         }
957
958         p = kzalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
959         if (!p)
960                 goto fatal_memory_error;
961
962         iommu = kzalloc(sizeof(struct iommu), GFP_ATOMIC);
963         if (!iommu)
964                 goto fatal_memory_error;
965
966         p->pbm_A.iommu = iommu;
967
968         iommu = kzalloc(sizeof(struct iommu), GFP_ATOMIC);
969         if (!iommu)
970                 goto fatal_memory_error;
971
972         p->pbm_B.iommu = iommu;
973
974         pci_sun4v_pbm_init(p, dp, devhandle);
975         return;
976
977 fatal_memory_error:
978         prom_printf("SUN4V_PCI: Fatal memory allocation error.\n");
979         prom_halt();
980 }