[SPARC64]: Kill ino_bucket->pil
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / sparc64 / kernel / pci_sun4v.c
blobb97c81ba88359f008b88eb30740902393431eb55
1 /* pci_sun4v.c: SUN4V specific PCI controller support.
3 * Copyright (C) 2006 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>
14 #include <asm/pbm.h>
15 #include <asm/iommu.h>
16 #include <asm/irq.h>
17 #include <asm/upa.h>
18 #include <asm/pstate.h>
19 #include <asm/oplib.h>
20 #include <asm/hypervisor.h>
22 #include "pci_impl.h"
23 #include "iommu_common.h"
25 #include "pci_sun4v.h"
27 #define PGLIST_NENTS (PAGE_SIZE / sizeof(u64))
29 struct pci_iommu_batch {
30 struct pci_dev *pdev; /* Device mapping is for. */
31 unsigned long prot; /* IOMMU page protections */
32 unsigned long entry; /* Index into IOTSB. */
33 u64 *pglist; /* List of physical pages */
34 unsigned long npages; /* Number of pages in list. */
37 static DEFINE_PER_CPU(struct pci_iommu_batch, pci_iommu_batch);
39 /* Interrupts must be disabled. */
40 static inline void pci_iommu_batch_start(struct pci_dev *pdev, unsigned long prot, unsigned long entry)
42 struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
44 p->pdev = pdev;
45 p->prot = prot;
46 p->entry = entry;
47 p->npages = 0;
50 /* Interrupts must be disabled. */
51 static long pci_iommu_batch_flush(struct pci_iommu_batch *p)
53 struct pcidev_cookie *pcp = p->pdev->sysdata;
54 unsigned long devhandle = pcp->pbm->devhandle;
55 unsigned long prot = p->prot;
56 unsigned long entry = p->entry;
57 u64 *pglist = p->pglist;
58 unsigned long npages = p->npages;
60 while (npages != 0) {
61 long num;
63 num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
64 npages, prot, __pa(pglist));
65 if (unlikely(num < 0)) {
66 if (printk_ratelimit())
67 printk("pci_iommu_batch_flush: IOMMU map of "
68 "[%08lx:%08lx:%lx:%lx:%lx] failed with "
69 "status %ld\n",
70 devhandle, HV_PCI_TSBID(0, entry),
71 npages, prot, __pa(pglist), num);
72 return -1;
75 entry += num;
76 npages -= num;
77 pglist += num;
80 p->entry = entry;
81 p->npages = 0;
83 return 0;
86 /* Interrupts must be disabled. */
87 static inline long pci_iommu_batch_add(u64 phys_page)
89 struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
91 BUG_ON(p->npages >= PGLIST_NENTS);
93 p->pglist[p->npages++] = phys_page;
94 if (p->npages == PGLIST_NENTS)
95 return pci_iommu_batch_flush(p);
97 return 0;
100 /* Interrupts must be disabled. */
101 static inline long pci_iommu_batch_end(void)
103 struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
105 BUG_ON(p->npages >= PGLIST_NENTS);
107 return pci_iommu_batch_flush(p);
110 static long pci_arena_alloc(struct pci_iommu_arena *arena, unsigned long npages)
112 unsigned long n, i, start, end, limit;
113 int pass;
115 limit = arena->limit;
116 start = arena->hint;
117 pass = 0;
119 again:
120 n = find_next_zero_bit(arena->map, limit, start);
121 end = n + npages;
122 if (unlikely(end >= limit)) {
123 if (likely(pass < 1)) {
124 limit = start;
125 start = 0;
126 pass++;
127 goto again;
128 } else {
129 /* Scanned the whole thing, give up. */
130 return -1;
134 for (i = n; i < end; i++) {
135 if (test_bit(i, arena->map)) {
136 start = i + 1;
137 goto again;
141 for (i = n; i < end; i++)
142 __set_bit(i, arena->map);
144 arena->hint = end;
146 return n;
149 static void pci_arena_free(struct pci_iommu_arena *arena, unsigned long base, unsigned long npages)
151 unsigned long i;
153 for (i = base; i < (base + npages); i++)
154 __clear_bit(i, arena->map);
157 static void *pci_4v_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_addrp, gfp_t gfp)
159 struct pcidev_cookie *pcp;
160 struct pci_iommu *iommu;
161 unsigned long flags, order, first_page, npages, n;
162 void *ret;
163 long entry;
165 size = IO_PAGE_ALIGN(size);
166 order = get_order(size);
167 if (unlikely(order >= MAX_ORDER))
168 return NULL;
170 npages = size >> IO_PAGE_SHIFT;
172 first_page = __get_free_pages(gfp, order);
173 if (unlikely(first_page == 0UL))
174 return NULL;
176 memset((char *)first_page, 0, PAGE_SIZE << order);
178 pcp = pdev->sysdata;
179 iommu = pcp->pbm->iommu;
181 spin_lock_irqsave(&iommu->lock, flags);
182 entry = pci_arena_alloc(&iommu->arena, npages);
183 spin_unlock_irqrestore(&iommu->lock, flags);
185 if (unlikely(entry < 0L))
186 goto arena_alloc_fail;
188 *dma_addrp = (iommu->page_table_map_base +
189 (entry << IO_PAGE_SHIFT));
190 ret = (void *) first_page;
191 first_page = __pa(first_page);
193 local_irq_save(flags);
195 pci_iommu_batch_start(pdev,
196 (HV_PCI_MAP_ATTR_READ |
197 HV_PCI_MAP_ATTR_WRITE),
198 entry);
200 for (n = 0; n < npages; n++) {
201 long err = pci_iommu_batch_add(first_page + (n * PAGE_SIZE));
202 if (unlikely(err < 0L))
203 goto iommu_map_fail;
206 if (unlikely(pci_iommu_batch_end() < 0L))
207 goto iommu_map_fail;
209 local_irq_restore(flags);
211 return ret;
213 iommu_map_fail:
214 /* Interrupts are disabled. */
215 spin_lock(&iommu->lock);
216 pci_arena_free(&iommu->arena, entry, npages);
217 spin_unlock_irqrestore(&iommu->lock, flags);
219 arena_alloc_fail:
220 free_pages(first_page, order);
221 return NULL;
224 static void pci_4v_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_t dvma)
226 struct pcidev_cookie *pcp;
227 struct pci_iommu *iommu;
228 unsigned long flags, order, npages, entry;
229 u32 devhandle;
231 npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
232 pcp = pdev->sysdata;
233 iommu = pcp->pbm->iommu;
234 devhandle = pcp->pbm->devhandle;
235 entry = ((dvma - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
237 spin_lock_irqsave(&iommu->lock, flags);
239 pci_arena_free(&iommu->arena, entry, npages);
241 do {
242 unsigned long num;
244 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
245 npages);
246 entry += num;
247 npages -= num;
248 } while (npages != 0);
250 spin_unlock_irqrestore(&iommu->lock, flags);
252 order = get_order(size);
253 if (order < 10)
254 free_pages((unsigned long)cpu, order);
257 static dma_addr_t pci_4v_map_single(struct pci_dev *pdev, void *ptr, size_t sz, int direction)
259 struct pcidev_cookie *pcp;
260 struct pci_iommu *iommu;
261 unsigned long flags, npages, oaddr;
262 unsigned long i, base_paddr;
263 u32 bus_addr, ret;
264 unsigned long prot;
265 long entry;
267 pcp = pdev->sysdata;
268 iommu = pcp->pbm->iommu;
270 if (unlikely(direction == PCI_DMA_NONE))
271 goto bad;
273 oaddr = (unsigned long)ptr;
274 npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
275 npages >>= IO_PAGE_SHIFT;
277 spin_lock_irqsave(&iommu->lock, flags);
278 entry = pci_arena_alloc(&iommu->arena, npages);
279 spin_unlock_irqrestore(&iommu->lock, flags);
281 if (unlikely(entry < 0L))
282 goto bad;
284 bus_addr = (iommu->page_table_map_base +
285 (entry << IO_PAGE_SHIFT));
286 ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
287 base_paddr = __pa(oaddr & IO_PAGE_MASK);
288 prot = HV_PCI_MAP_ATTR_READ;
289 if (direction != PCI_DMA_TODEVICE)
290 prot |= HV_PCI_MAP_ATTR_WRITE;
292 local_irq_save(flags);
294 pci_iommu_batch_start(pdev, prot, entry);
296 for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
297 long err = pci_iommu_batch_add(base_paddr);
298 if (unlikely(err < 0L))
299 goto iommu_map_fail;
301 if (unlikely(pci_iommu_batch_end() < 0L))
302 goto iommu_map_fail;
304 local_irq_restore(flags);
306 return ret;
308 bad:
309 if (printk_ratelimit())
310 WARN_ON(1);
311 return PCI_DMA_ERROR_CODE;
313 iommu_map_fail:
314 /* Interrupts are disabled. */
315 spin_lock(&iommu->lock);
316 pci_arena_free(&iommu->arena, entry, npages);
317 spin_unlock_irqrestore(&iommu->lock, flags);
319 return PCI_DMA_ERROR_CODE;
322 static void pci_4v_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
324 struct pcidev_cookie *pcp;
325 struct pci_iommu *iommu;
326 unsigned long flags, npages;
327 long entry;
328 u32 devhandle;
330 if (unlikely(direction == PCI_DMA_NONE)) {
331 if (printk_ratelimit())
332 WARN_ON(1);
333 return;
336 pcp = pdev->sysdata;
337 iommu = pcp->pbm->iommu;
338 devhandle = pcp->pbm->devhandle;
340 npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
341 npages >>= IO_PAGE_SHIFT;
342 bus_addr &= IO_PAGE_MASK;
344 spin_lock_irqsave(&iommu->lock, flags);
346 entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
347 pci_arena_free(&iommu->arena, entry, npages);
349 do {
350 unsigned long num;
352 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
353 npages);
354 entry += num;
355 npages -= num;
356 } while (npages != 0);
358 spin_unlock_irqrestore(&iommu->lock, flags);
361 #define SG_ENT_PHYS_ADDRESS(SG) \
362 (__pa(page_address((SG)->page)) + (SG)->offset)
364 static inline long fill_sg(long entry, struct pci_dev *pdev,
365 struct scatterlist *sg,
366 int nused, int nelems, unsigned long prot)
368 struct scatterlist *dma_sg = sg;
369 struct scatterlist *sg_end = sg + nelems;
370 unsigned long flags;
371 int i;
373 local_irq_save(flags);
375 pci_iommu_batch_start(pdev, prot, entry);
377 for (i = 0; i < nused; i++) {
378 unsigned long pteval = ~0UL;
379 u32 dma_npages;
381 dma_npages = ((dma_sg->dma_address & (IO_PAGE_SIZE - 1UL)) +
382 dma_sg->dma_length +
383 ((IO_PAGE_SIZE - 1UL))) >> IO_PAGE_SHIFT;
384 do {
385 unsigned long offset;
386 signed int len;
388 /* If we are here, we know we have at least one
389 * more page to map. So walk forward until we
390 * hit a page crossing, and begin creating new
391 * mappings from that spot.
393 for (;;) {
394 unsigned long tmp;
396 tmp = SG_ENT_PHYS_ADDRESS(sg);
397 len = sg->length;
398 if (((tmp ^ pteval) >> IO_PAGE_SHIFT) != 0UL) {
399 pteval = tmp & IO_PAGE_MASK;
400 offset = tmp & (IO_PAGE_SIZE - 1UL);
401 break;
403 if (((tmp ^ (tmp + len - 1UL)) >> IO_PAGE_SHIFT) != 0UL) {
404 pteval = (tmp + IO_PAGE_SIZE) & IO_PAGE_MASK;
405 offset = 0UL;
406 len -= (IO_PAGE_SIZE - (tmp & (IO_PAGE_SIZE - 1UL)));
407 break;
409 sg++;
412 pteval = (pteval & IOPTE_PAGE);
413 while (len > 0) {
414 long err;
416 err = pci_iommu_batch_add(pteval);
417 if (unlikely(err < 0L))
418 goto iommu_map_failed;
420 pteval += IO_PAGE_SIZE;
421 len -= (IO_PAGE_SIZE - offset);
422 offset = 0;
423 dma_npages--;
426 pteval = (pteval & IOPTE_PAGE) + len;
427 sg++;
429 /* Skip over any tail mappings we've fully mapped,
430 * adjusting pteval along the way. Stop when we
431 * detect a page crossing event.
433 while (sg < sg_end &&
434 (pteval << (64 - IO_PAGE_SHIFT)) != 0UL &&
435 (pteval == SG_ENT_PHYS_ADDRESS(sg)) &&
436 ((pteval ^
437 (SG_ENT_PHYS_ADDRESS(sg) + sg->length - 1UL)) >> IO_PAGE_SHIFT) == 0UL) {
438 pteval += sg->length;
439 sg++;
441 if ((pteval << (64 - IO_PAGE_SHIFT)) == 0UL)
442 pteval = ~0UL;
443 } while (dma_npages != 0);
444 dma_sg++;
447 if (unlikely(pci_iommu_batch_end() < 0L))
448 goto iommu_map_failed;
450 local_irq_restore(flags);
451 return 0;
453 iommu_map_failed:
454 local_irq_restore(flags);
455 return -1L;
458 static int pci_4v_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
460 struct pcidev_cookie *pcp;
461 struct pci_iommu *iommu;
462 unsigned long flags, npages, prot;
463 u32 dma_base;
464 struct scatterlist *sgtmp;
465 long entry, err;
466 int used;
468 /* Fast path single entry scatterlists. */
469 if (nelems == 1) {
470 sglist->dma_address =
471 pci_4v_map_single(pdev,
472 (page_address(sglist->page) + sglist->offset),
473 sglist->length, direction);
474 if (unlikely(sglist->dma_address == PCI_DMA_ERROR_CODE))
475 return 0;
476 sglist->dma_length = sglist->length;
477 return 1;
480 pcp = pdev->sysdata;
481 iommu = pcp->pbm->iommu;
483 if (unlikely(direction == PCI_DMA_NONE))
484 goto bad;
486 /* Step 1: Prepare scatter list. */
487 npages = prepare_sg(sglist, nelems);
489 /* Step 2: Allocate a cluster and context, if necessary. */
490 spin_lock_irqsave(&iommu->lock, flags);
491 entry = pci_arena_alloc(&iommu->arena, npages);
492 spin_unlock_irqrestore(&iommu->lock, flags);
494 if (unlikely(entry < 0L))
495 goto bad;
497 dma_base = iommu->page_table_map_base +
498 (entry << IO_PAGE_SHIFT);
500 /* Step 3: Normalize DMA addresses. */
501 used = nelems;
503 sgtmp = sglist;
504 while (used && sgtmp->dma_length) {
505 sgtmp->dma_address += dma_base;
506 sgtmp++;
507 used--;
509 used = nelems - used;
511 /* Step 4: Create the mappings. */
512 prot = HV_PCI_MAP_ATTR_READ;
513 if (direction != PCI_DMA_TODEVICE)
514 prot |= HV_PCI_MAP_ATTR_WRITE;
516 err = fill_sg(entry, pdev, sglist, used, nelems, prot);
517 if (unlikely(err < 0L))
518 goto iommu_map_failed;
520 return used;
522 bad:
523 if (printk_ratelimit())
524 WARN_ON(1);
525 return 0;
527 iommu_map_failed:
528 spin_lock_irqsave(&iommu->lock, flags);
529 pci_arena_free(&iommu->arena, entry, npages);
530 spin_unlock_irqrestore(&iommu->lock, flags);
532 return 0;
535 static void pci_4v_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
537 struct pcidev_cookie *pcp;
538 struct pci_iommu *iommu;
539 unsigned long flags, i, npages;
540 long entry;
541 u32 devhandle, bus_addr;
543 if (unlikely(direction == PCI_DMA_NONE)) {
544 if (printk_ratelimit())
545 WARN_ON(1);
548 pcp = pdev->sysdata;
549 iommu = pcp->pbm->iommu;
550 devhandle = pcp->pbm->devhandle;
552 bus_addr = sglist->dma_address & IO_PAGE_MASK;
554 for (i = 1; i < nelems; i++)
555 if (sglist[i].dma_length == 0)
556 break;
557 i--;
558 npages = (IO_PAGE_ALIGN(sglist[i].dma_address + sglist[i].dma_length) -
559 bus_addr) >> IO_PAGE_SHIFT;
561 entry = ((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
563 spin_lock_irqsave(&iommu->lock, flags);
565 pci_arena_free(&iommu->arena, entry, npages);
567 do {
568 unsigned long num;
570 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
571 npages);
572 entry += num;
573 npages -= num;
574 } while (npages != 0);
576 spin_unlock_irqrestore(&iommu->lock, flags);
579 static void pci_4v_dma_sync_single_for_cpu(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
581 /* Nothing to do... */
584 static void pci_4v_dma_sync_sg_for_cpu(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
586 /* Nothing to do... */
589 struct pci_iommu_ops pci_sun4v_iommu_ops = {
590 .alloc_consistent = pci_4v_alloc_consistent,
591 .free_consistent = pci_4v_free_consistent,
592 .map_single = pci_4v_map_single,
593 .unmap_single = pci_4v_unmap_single,
594 .map_sg = pci_4v_map_sg,
595 .unmap_sg = pci_4v_unmap_sg,
596 .dma_sync_single_for_cpu = pci_4v_dma_sync_single_for_cpu,
597 .dma_sync_sg_for_cpu = pci_4v_dma_sync_sg_for_cpu,
600 /* SUN4V PCI configuration space accessors. */
602 struct pdev_entry {
603 struct pdev_entry *next;
604 u32 devhandle;
605 unsigned int bus;
606 unsigned int device;
607 unsigned int func;
610 #define PDEV_HTAB_SIZE 16
611 #define PDEV_HTAB_MASK (PDEV_HTAB_SIZE - 1)
612 static struct pdev_entry *pdev_htab[PDEV_HTAB_SIZE];
614 static inline unsigned int pdev_hashfn(u32 devhandle, unsigned int bus, unsigned int device, unsigned int func)
616 unsigned int val;
618 val = (devhandle ^ (devhandle >> 4));
619 val ^= bus;
620 val ^= device;
621 val ^= func;
623 return val & PDEV_HTAB_MASK;
626 static int pdev_htab_add(u32 devhandle, unsigned int bus, unsigned int device, unsigned int func)
628 struct pdev_entry *p = kmalloc(sizeof(*p), GFP_KERNEL);
629 struct pdev_entry **slot;
631 if (!p)
632 return -ENOMEM;
634 slot = &pdev_htab[pdev_hashfn(devhandle, bus, device, func)];
635 p->next = *slot;
636 *slot = p;
638 p->devhandle = devhandle;
639 p->bus = bus;
640 p->device = device;
641 p->func = func;
643 return 0;
646 /* Recursively descend into the OBP device tree, rooted at toplevel_node,
647 * looking for a PCI device matching bus and devfn.
649 static int obp_find(struct linux_prom_pci_registers *pregs, int toplevel_node, unsigned int bus, unsigned int devfn)
651 toplevel_node = prom_getchild(toplevel_node);
653 while (toplevel_node != 0) {
654 int ret = obp_find(pregs, toplevel_node, bus, devfn);
656 if (ret != 0)
657 return ret;
659 ret = prom_getproperty(toplevel_node, "reg", (char *) pregs,
660 sizeof(*pregs) * PROMREG_MAX);
661 if (ret == 0 || ret == -1)
662 goto next_sibling;
664 if (((pregs[0].phys_hi >> 16) & 0xff) == bus &&
665 ((pregs[0].phys_hi >> 8) & 0xff) == devfn)
666 break;
668 next_sibling:
669 toplevel_node = prom_getsibling(toplevel_node);
672 return toplevel_node;
675 static int pdev_htab_populate(struct pci_pbm_info *pbm)
677 struct linux_prom_pci_registers pr[PROMREG_MAX];
678 u32 devhandle = pbm->devhandle;
679 unsigned int bus;
681 for (bus = pbm->pci_first_busno; bus <= pbm->pci_last_busno; bus++) {
682 unsigned int devfn;
684 for (devfn = 0; devfn < 256; devfn++) {
685 unsigned int device = PCI_SLOT(devfn);
686 unsigned int func = PCI_FUNC(devfn);
688 if (obp_find(pr, pbm->prom_node, bus, devfn)) {
689 int err = pdev_htab_add(devhandle, bus,
690 device, func);
691 if (err)
692 return err;
697 return 0;
700 static struct pdev_entry *pdev_find(u32 devhandle, unsigned int bus, unsigned int device, unsigned int func)
702 struct pdev_entry *p;
704 p = pdev_htab[pdev_hashfn(devhandle, bus, device, func)];
705 while (p) {
706 if (p->devhandle == devhandle &&
707 p->bus == bus &&
708 p->device == device &&
709 p->func == func)
710 break;
712 p = p->next;
715 return p;
718 static inline int pci_sun4v_out_of_range(struct pci_pbm_info *pbm, unsigned int bus, unsigned int device, unsigned int func)
720 if (bus < pbm->pci_first_busno ||
721 bus > pbm->pci_last_busno)
722 return 1;
723 return pdev_find(pbm->devhandle, bus, device, func) == NULL;
726 static int pci_sun4v_read_pci_cfg(struct pci_bus *bus_dev, unsigned int devfn,
727 int where, int size, u32 *value)
729 struct pci_pbm_info *pbm = bus_dev->sysdata;
730 u32 devhandle = pbm->devhandle;
731 unsigned int bus = bus_dev->number;
732 unsigned int device = PCI_SLOT(devfn);
733 unsigned int func = PCI_FUNC(devfn);
734 unsigned long ret;
736 if (pci_sun4v_out_of_range(pbm, bus, device, func)) {
737 ret = ~0UL;
738 } else {
739 ret = pci_sun4v_config_get(devhandle,
740 HV_PCI_DEVICE_BUILD(bus, device, func),
741 where, size);
742 #if 0
743 printk("rcfg: [%x:%x:%x:%d]=[%lx]\n",
744 devhandle, HV_PCI_DEVICE_BUILD(bus, device, func),
745 where, size, ret);
746 #endif
748 switch (size) {
749 case 1:
750 *value = ret & 0xff;
751 break;
752 case 2:
753 *value = ret & 0xffff;
754 break;
755 case 4:
756 *value = ret & 0xffffffff;
757 break;
761 return PCIBIOS_SUCCESSFUL;
764 static int pci_sun4v_write_pci_cfg(struct pci_bus *bus_dev, unsigned int devfn,
765 int where, int size, u32 value)
767 struct pci_pbm_info *pbm = bus_dev->sysdata;
768 u32 devhandle = pbm->devhandle;
769 unsigned int bus = bus_dev->number;
770 unsigned int device = PCI_SLOT(devfn);
771 unsigned int func = PCI_FUNC(devfn);
772 unsigned long ret;
774 if (pci_sun4v_out_of_range(pbm, bus, device, func)) {
775 /* Do nothing. */
776 } else {
777 ret = pci_sun4v_config_put(devhandle,
778 HV_PCI_DEVICE_BUILD(bus, device, func),
779 where, size, value);
780 #if 0
781 printk("wcfg: [%x:%x:%x:%d] v[%x] == [%lx]\n",
782 devhandle, HV_PCI_DEVICE_BUILD(bus, device, func),
783 where, size, value, ret);
784 #endif
786 return PCIBIOS_SUCCESSFUL;
789 static struct pci_ops pci_sun4v_ops = {
790 .read = pci_sun4v_read_pci_cfg,
791 .write = pci_sun4v_write_pci_cfg,
795 static void pbm_scan_bus(struct pci_controller_info *p,
796 struct pci_pbm_info *pbm)
798 struct pcidev_cookie *cookie = kmalloc(sizeof(*cookie), GFP_KERNEL);
800 if (!cookie) {
801 prom_printf("%s: Critical allocation failure.\n", pbm->name);
802 prom_halt();
805 /* All we care about is the PBM. */
806 memset(cookie, 0, sizeof(*cookie));
807 cookie->pbm = pbm;
809 pbm->pci_bus = pci_scan_bus(pbm->pci_first_busno, p->pci_ops, pbm);
810 #if 0
811 pci_fixup_host_bridge_self(pbm->pci_bus);
812 pbm->pci_bus->self->sysdata = cookie;
813 #endif
814 pci_fill_in_pbm_cookies(pbm->pci_bus, pbm,
815 pbm->prom_node);
816 pci_record_assignments(pbm, pbm->pci_bus);
817 pci_assign_unassigned(pbm, pbm->pci_bus);
818 pci_fixup_irq(pbm, pbm->pci_bus);
819 pci_determine_66mhz_disposition(pbm, pbm->pci_bus);
820 pci_setup_busmastering(pbm, pbm->pci_bus);
823 static void pci_sun4v_scan_bus(struct pci_controller_info *p)
825 if (p->pbm_A.prom_node) {
826 p->pbm_A.is_66mhz_capable =
827 prom_getbool(p->pbm_A.prom_node, "66mhz-capable");
829 pbm_scan_bus(p, &p->pbm_A);
831 if (p->pbm_B.prom_node) {
832 p->pbm_B.is_66mhz_capable =
833 prom_getbool(p->pbm_B.prom_node, "66mhz-capable");
835 pbm_scan_bus(p, &p->pbm_B);
838 /* XXX register error interrupt handlers XXX */
841 static unsigned int pci_sun4v_irq_build(struct pci_pbm_info *pbm,
842 struct pci_dev *pdev,
843 unsigned int devino)
845 u32 devhandle = pbm->devhandle;
847 return sun4v_build_irq(devhandle, devino, IBF_PCI);
850 static void pci_sun4v_base_address_update(struct pci_dev *pdev, int resource)
852 struct pcidev_cookie *pcp = pdev->sysdata;
853 struct pci_pbm_info *pbm = pcp->pbm;
854 struct resource *res, *root;
855 u32 reg;
856 int where, size, is_64bit;
858 res = &pdev->resource[resource];
859 if (resource < 6) {
860 where = PCI_BASE_ADDRESS_0 + (resource * 4);
861 } else if (resource == PCI_ROM_RESOURCE) {
862 where = pdev->rom_base_reg;
863 } else {
864 /* Somebody might have asked allocation of a non-standard resource */
865 return;
868 /* XXX 64-bit MEM handling is not %100 correct... XXX */
869 is_64bit = 0;
870 if (res->flags & IORESOURCE_IO)
871 root = &pbm->io_space;
872 else {
873 root = &pbm->mem_space;
874 if ((res->flags & PCI_BASE_ADDRESS_MEM_TYPE_MASK)
875 == PCI_BASE_ADDRESS_MEM_TYPE_64)
876 is_64bit = 1;
879 size = res->end - res->start;
880 pci_read_config_dword(pdev, where, &reg);
881 reg = ((reg & size) |
882 (((u32)(res->start - root->start)) & ~size));
883 if (resource == PCI_ROM_RESOURCE) {
884 reg |= PCI_ROM_ADDRESS_ENABLE;
885 res->flags |= IORESOURCE_ROM_ENABLE;
887 pci_write_config_dword(pdev, where, reg);
889 /* This knows that the upper 32-bits of the address
890 * must be zero. Our PCI common layer enforces this.
892 if (is_64bit)
893 pci_write_config_dword(pdev, where + 4, 0);
896 static void pci_sun4v_resource_adjust(struct pci_dev *pdev,
897 struct resource *res,
898 struct resource *root)
900 res->start += root->start;
901 res->end += root->start;
904 /* Use ranges property to determine where PCI MEM, I/O, and Config
905 * space are for this PCI bus module.
907 static void pci_sun4v_determine_mem_io_space(struct pci_pbm_info *pbm)
909 int i, saw_mem, saw_io;
911 saw_mem = saw_io = 0;
912 for (i = 0; i < pbm->num_pbm_ranges; i++) {
913 struct linux_prom_pci_ranges *pr = &pbm->pbm_ranges[i];
914 unsigned long a;
915 int type;
917 type = (pr->child_phys_hi >> 24) & 0x3;
918 a = (((unsigned long)pr->parent_phys_hi << 32UL) |
919 ((unsigned long)pr->parent_phys_lo << 0UL));
921 switch (type) {
922 case 1:
923 /* 16-bit IO space, 16MB */
924 pbm->io_space.start = a;
925 pbm->io_space.end = a + ((16UL*1024UL*1024UL) - 1UL);
926 pbm->io_space.flags = IORESOURCE_IO;
927 saw_io = 1;
928 break;
930 case 2:
931 /* 32-bit MEM space, 2GB */
932 pbm->mem_space.start = a;
933 pbm->mem_space.end = a + (0x80000000UL - 1UL);
934 pbm->mem_space.flags = IORESOURCE_MEM;
935 saw_mem = 1;
936 break;
938 case 3:
939 /* XXX 64-bit MEM handling XXX */
941 default:
942 break;
946 if (!saw_io || !saw_mem) {
947 prom_printf("%s: Fatal error, missing %s PBM range.\n",
948 pbm->name,
949 (!saw_io ? "IO" : "MEM"));
950 prom_halt();
953 printk("%s: PCI IO[%lx] MEM[%lx]\n",
954 pbm->name,
955 pbm->io_space.start,
956 pbm->mem_space.start);
959 static void pbm_register_toplevel_resources(struct pci_controller_info *p,
960 struct pci_pbm_info *pbm)
962 pbm->io_space.name = pbm->mem_space.name = pbm->name;
964 request_resource(&ioport_resource, &pbm->io_space);
965 request_resource(&iomem_resource, &pbm->mem_space);
966 pci_register_legacy_regions(&pbm->io_space,
967 &pbm->mem_space);
970 static unsigned long probe_existing_entries(struct pci_pbm_info *pbm,
971 struct pci_iommu *iommu)
973 struct pci_iommu_arena *arena = &iommu->arena;
974 unsigned long i, cnt = 0;
975 u32 devhandle;
977 devhandle = pbm->devhandle;
978 for (i = 0; i < arena->limit; i++) {
979 unsigned long ret, io_attrs, ra;
981 ret = pci_sun4v_iommu_getmap(devhandle,
982 HV_PCI_TSBID(0, i),
983 &io_attrs, &ra);
984 if (ret == HV_EOK) {
985 cnt++;
986 __set_bit(i, arena->map);
990 return cnt;
993 static void pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
995 struct pci_iommu *iommu = pbm->iommu;
996 unsigned long num_tsb_entries, sz;
997 u32 vdma[2], dma_mask, dma_offset;
998 int err, tsbsize;
1000 err = prom_getproperty(pbm->prom_node, "virtual-dma",
1001 (char *)&vdma[0], sizeof(vdma));
1002 if (err == 0 || err == -1) {
1003 /* No property, use default values. */
1004 vdma[0] = 0x80000000;
1005 vdma[1] = 0x80000000;
1008 dma_mask = vdma[0];
1009 switch (vdma[1]) {
1010 case 0x20000000:
1011 dma_mask |= 0x1fffffff;
1012 tsbsize = 64;
1013 break;
1015 case 0x40000000:
1016 dma_mask |= 0x3fffffff;
1017 tsbsize = 128;
1018 break;
1020 case 0x80000000:
1021 dma_mask |= 0x7fffffff;
1022 tsbsize = 256;
1023 break;
1025 default:
1026 prom_printf("PCI-SUN4V: strange virtual-dma size.\n");
1027 prom_halt();
1030 tsbsize *= (8 * 1024);
1032 num_tsb_entries = tsbsize / sizeof(iopte_t);
1034 dma_offset = vdma[0];
1036 /* Setup initial software IOMMU state. */
1037 spin_lock_init(&iommu->lock);
1038 iommu->ctx_lowest_free = 1;
1039 iommu->page_table_map_base = dma_offset;
1040 iommu->dma_addr_mask = dma_mask;
1042 /* Allocate and initialize the free area map. */
1043 sz = num_tsb_entries / 8;
1044 sz = (sz + 7UL) & ~7UL;
1045 iommu->arena.map = kmalloc(sz, GFP_KERNEL);
1046 if (!iommu->arena.map) {
1047 prom_printf("PCI_IOMMU: Error, kmalloc(arena.map) failed.\n");
1048 prom_halt();
1050 memset(iommu->arena.map, 0, sz);
1051 iommu->arena.limit = num_tsb_entries;
1053 sz = probe_existing_entries(pbm, iommu);
1055 printk("%s: TSB entries [%lu], existing mapings [%lu]\n",
1056 pbm->name, num_tsb_entries, sz);
1059 static void pci_sun4v_get_bus_range(struct pci_pbm_info *pbm)
1061 unsigned int busrange[2];
1062 int prom_node = pbm->prom_node;
1063 int err;
1065 err = prom_getproperty(prom_node, "bus-range",
1066 (char *)&busrange[0],
1067 sizeof(busrange));
1068 if (err == 0 || err == -1) {
1069 prom_printf("%s: Fatal error, no bus-range.\n", pbm->name);
1070 prom_halt();
1073 pbm->pci_first_busno = busrange[0];
1074 pbm->pci_last_busno = busrange[1];
1078 static void pci_sun4v_pbm_init(struct pci_controller_info *p, int prom_node, u32 devhandle)
1080 struct pci_pbm_info *pbm;
1081 int err, i;
1083 if (devhandle & 0x40)
1084 pbm = &p->pbm_B;
1085 else
1086 pbm = &p->pbm_A;
1088 pbm->parent = p;
1089 pbm->prom_node = prom_node;
1090 pbm->pci_first_slot = 1;
1092 pbm->devhandle = devhandle;
1094 sprintf(pbm->name, "SUN4V-PCI%d PBM%c",
1095 p->index, (pbm == &p->pbm_A ? 'A' : 'B'));
1097 printk("%s: devhandle[%x] prom_node[%x:%x]\n",
1098 pbm->name, pbm->devhandle,
1099 pbm->prom_node, prom_getchild(pbm->prom_node));
1101 prom_getstring(prom_node, "name",
1102 pbm->prom_name, sizeof(pbm->prom_name));
1104 err = prom_getproperty(prom_node, "ranges",
1105 (char *) pbm->pbm_ranges,
1106 sizeof(pbm->pbm_ranges));
1107 if (err == 0 || err == -1) {
1108 prom_printf("%s: Fatal error, no ranges property.\n",
1109 pbm->name);
1110 prom_halt();
1113 pbm->num_pbm_ranges =
1114 (err / sizeof(struct linux_prom_pci_ranges));
1116 /* Mask out the top 8 bits of the ranges, leaving the real
1117 * physical address.
1119 for (i = 0; i < pbm->num_pbm_ranges; i++)
1120 pbm->pbm_ranges[i].parent_phys_hi &= 0x0fffffff;
1122 pci_sun4v_determine_mem_io_space(pbm);
1123 pbm_register_toplevel_resources(p, pbm);
1125 err = prom_getproperty(prom_node, "interrupt-map",
1126 (char *)pbm->pbm_intmap,
1127 sizeof(pbm->pbm_intmap));
1128 if (err == 0 || err == -1) {
1129 prom_printf("%s: Fatal error, no interrupt-map property.\n",
1130 pbm->name);
1131 prom_halt();
1134 pbm->num_pbm_intmap = (err / sizeof(struct linux_prom_pci_intmap));
1135 err = prom_getproperty(prom_node, "interrupt-map-mask",
1136 (char *)&pbm->pbm_intmask,
1137 sizeof(pbm->pbm_intmask));
1138 if (err == 0 || err == -1) {
1139 prom_printf("%s: Fatal error, no interrupt-map-mask.\n",
1140 pbm->name);
1141 prom_halt();
1144 pci_sun4v_get_bus_range(pbm);
1145 pci_sun4v_iommu_init(pbm);
1147 pdev_htab_populate(pbm);
1150 void sun4v_pci_init(int node, char *model_name)
1152 struct pci_controller_info *p;
1153 struct pci_iommu *iommu;
1154 struct linux_prom64_registers regs;
1155 u32 devhandle;
1156 int i;
1158 prom_getproperty(node, "reg", (char *)&regs, sizeof(regs));
1159 devhandle = (regs.phys_addr >> 32UL) & 0x0fffffff;
1161 for (p = pci_controller_root; p; p = p->next) {
1162 struct pci_pbm_info *pbm;
1164 if (p->pbm_A.prom_node && p->pbm_B.prom_node)
1165 continue;
1167 pbm = (p->pbm_A.prom_node ?
1168 &p->pbm_A :
1169 &p->pbm_B);
1171 if (pbm->devhandle == (devhandle ^ 0x40)) {
1172 pci_sun4v_pbm_init(p, node, devhandle);
1173 return;
1177 for_each_possible_cpu(i) {
1178 unsigned long page = get_zeroed_page(GFP_ATOMIC);
1180 if (!page)
1181 goto fatal_memory_error;
1183 per_cpu(pci_iommu_batch, i).pglist = (u64 *) page;
1186 p = kmalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
1187 if (!p)
1188 goto fatal_memory_error;
1190 memset(p, 0, sizeof(*p));
1192 iommu = kmalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
1193 if (!iommu)
1194 goto fatal_memory_error;
1196 memset(iommu, 0, sizeof(*iommu));
1197 p->pbm_A.iommu = iommu;
1199 iommu = kmalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
1200 if (!iommu)
1201 goto fatal_memory_error;
1203 memset(iommu, 0, sizeof(*iommu));
1204 p->pbm_B.iommu = iommu;
1206 p->next = pci_controller_root;
1207 pci_controller_root = p;
1209 p->index = pci_num_controllers++;
1210 p->pbms_same_domain = 0;
1212 p->scan_bus = pci_sun4v_scan_bus;
1213 p->irq_build = pci_sun4v_irq_build;
1214 p->base_address_update = pci_sun4v_base_address_update;
1215 p->resource_adjust = pci_sun4v_resource_adjust;
1216 p->pci_ops = &pci_sun4v_ops;
1218 /* Like PSYCHO and SCHIZO we have a 2GB aligned area
1219 * for memory space.
1221 pci_memspace_mask = 0x7fffffffUL;
1223 pci_sun4v_pbm_init(p, node, devhandle);
1224 return;
1226 fatal_memory_error:
1227 prom_printf("SUN4V_PCI: Fatal memory allocation error.\n");
1228 prom_halt();