]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - arch/sparc64/kernel/pci_sun4v.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:


52bd4563492d82b8345ba6862214cbf54d2651fb
[mirror_ubuntu-zesty-kernel.git] / 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
16 #include <asm/pbm.h>
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 #define PGLIST_NENTS (PAGE_SIZE / sizeof(u64))
31
32 struct pci_iommu_batch {
33 struct pci_dev *pdev; /* Device mapping is for. */
34 unsigned long prot; /* IOMMU page protections */
35 unsigned long entry; /* Index into IOTSB. */
36 u64 *pglist; /* List of physical pages */
37 unsigned long npages; /* Number of pages in list. */
38 };
39
40 static DEFINE_PER_CPU(struct pci_iommu_batch, pci_iommu_batch);
41
42 /* Interrupts must be disabled. */
43 static inline void pci_iommu_batch_start(struct pci_dev *pdev, unsigned long prot, unsigned long entry)
44 {
45 struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
46
47 p->pdev = pdev;
48 p->prot = prot;
49 p->entry = entry;
50 p->npages = 0;
51 }
52
53 /* Interrupts must be disabled. */
54 static long pci_iommu_batch_flush(struct pci_iommu_batch *p)
55 {
56 struct pci_pbm_info *pbm = p->pdev->dev.archdata.host_controller;
57 unsigned long devhandle = pbm->devhandle;
58 unsigned long prot = p->prot;
59 unsigned long entry = p->entry;
60 u64 *pglist = p->pglist;
61 unsigned long npages = p->npages;
62
63 while (npages != 0) {
64 long num;
65
66 num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
67 npages, prot, __pa(pglist));
68 if (unlikely(num < 0)) {
69 if (printk_ratelimit())
70 printk("pci_iommu_batch_flush: IOMMU map of "
71 "[%08lx:%08lx:%lx:%lx:%lx] failed with "
72 "status %ld\n",
73 devhandle, HV_PCI_TSBID(0, entry),
74 npages, prot, __pa(pglist), num);
75 return -1;
76 }
77
78 entry += num;
79 npages -= num;
80 pglist += num;
81 }
82
83 p->entry = entry;
84 p->npages = 0;
85
86 return 0;
87 }
88
89 /* Interrupts must be disabled. */
90 static inline long pci_iommu_batch_add(u64 phys_page)
91 {
92 struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
93
94 BUG_ON(p->npages >= PGLIST_NENTS);
95
96 p->pglist[p->npages++] = phys_page;
97 if (p->npages == PGLIST_NENTS)
98 return pci_iommu_batch_flush(p);
99
100 return 0;
101 }
102
103 /* Interrupts must be disabled. */
104 static inline long pci_iommu_batch_end(void)
105 {
106 struct pci_iommu_batch *p = &__get_cpu_var(pci_iommu_batch);
107
108 BUG_ON(p->npages >= PGLIST_NENTS);
109
110 return pci_iommu_batch_flush(p);
111 }
112
113 static long pci_arena_alloc(struct pci_iommu_arena *arena, unsigned long npages)
114 {
115 unsigned long n, i, start, end, limit;
116 int pass;
117
118 limit = arena->limit;
119 start = arena->hint;
120 pass = 0;
121
122 again:
123 n = find_next_zero_bit(arena->map, limit, start);
124 end = n + npages;
125 if (unlikely(end >= limit)) {
126 if (likely(pass < 1)) {
127 limit = start;
128 start = 0;
129 pass++;
130 goto again;
131 } else {
132 /* Scanned the whole thing, give up. */
133 return -1;
134 }
135 }
136
137 for (i = n; i < end; i++) {
138 if (test_bit(i, arena->map)) {
139 start = i + 1;
140 goto again;
141 }
142 }
143
144 for (i = n; i < end; i++)
145 __set_bit(i, arena->map);
146
147 arena->hint = end;
148
149 return n;
150 }
151
152 static void pci_arena_free(struct pci_iommu_arena *arena, unsigned long base, unsigned long npages)
153 {
154 unsigned long i;
155
156 for (i = base; i < (base + npages); i++)
157 __clear_bit(i, arena->map);
158 }
159
160 static void *pci_4v_alloc_consistent(struct pci_dev *pdev, size_t size, dma_addr_t *dma_addrp, gfp_t gfp)
161 {
162 struct pci_iommu *iommu;
163 unsigned long flags, order, first_page, npages, n;
164 void *ret;
165 long entry;
166
167 size = IO_PAGE_ALIGN(size);
168 order = get_order(size);
169 if (unlikely(order >= MAX_ORDER))
170 return NULL;
171
172 npages = size >> IO_PAGE_SHIFT;
173
174 first_page = __get_free_pages(gfp, order);
175 if (unlikely(first_page == 0UL))
176 return NULL;
177
178 memset((char *)first_page, 0, PAGE_SIZE << order);
179
180 iommu = pdev->dev.archdata.iommu;
181
182 spin_lock_irqsave(&iommu->lock, flags);
183 entry = pci_arena_alloc(&iommu->arena, npages);
184 spin_unlock_irqrestore(&iommu->lock, flags);
185
186 if (unlikely(entry < 0L))
187 goto arena_alloc_fail;
188
189 *dma_addrp = (iommu->page_table_map_base +
190 (entry << IO_PAGE_SHIFT));
191 ret = (void *) first_page;
192 first_page = __pa(first_page);
193
194 local_irq_save(flags);
195
196 pci_iommu_batch_start(pdev,
197 (HV_PCI_MAP_ATTR_READ |
198 HV_PCI_MAP_ATTR_WRITE),
199 entry);
200
201 for (n = 0; n < npages; n++) {
202 long err = pci_iommu_batch_add(first_page + (n * PAGE_SIZE));
203 if (unlikely(err < 0L))
204 goto iommu_map_fail;
205 }
206
207 if (unlikely(pci_iommu_batch_end() < 0L))
208 goto iommu_map_fail;
209
210 local_irq_restore(flags);
211
212 return ret;
213
214 iommu_map_fail:
215 /* Interrupts are disabled. */
216 spin_lock(&iommu->lock);
217 pci_arena_free(&iommu->arena, entry, npages);
218 spin_unlock_irqrestore(&iommu->lock, flags);
219
220 arena_alloc_fail:
221 free_pages(first_page, order);
222 return NULL;
223 }
224
225 static void pci_4v_free_consistent(struct pci_dev *pdev, size_t size, void *cpu, dma_addr_t dvma)
226 {
227 struct pci_pbm_info *pbm;
228 struct pci_iommu *iommu;
229 unsigned long flags, order, npages, entry;
230 u32 devhandle;
231
232 npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
233 iommu = pdev->dev.archdata.iommu;
234 pbm = pdev->dev.archdata.host_controller;
235 devhandle = pbm->devhandle;
236 entry = ((dvma - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
237
238 spin_lock_irqsave(&iommu->lock, flags);
239
240 pci_arena_free(&iommu->arena, entry, npages);
241
242 do {
243 unsigned long num;
244
245 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
246 npages);
247 entry += num;
248 npages -= num;
249 } while (npages != 0);
250
251 spin_unlock_irqrestore(&iommu->lock, flags);
252
253 order = get_order(size);
254 if (order < 10)
255 free_pages((unsigned long)cpu, order);
256 }
257
258 static dma_addr_t pci_4v_map_single(struct pci_dev *pdev, void *ptr, size_t sz, int direction)
259 {
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;
266
267 iommu = pdev->dev.archdata.iommu;
268
269 if (unlikely(direction == PCI_DMA_NONE))
270 goto bad;
271
272 oaddr = (unsigned long)ptr;
273 npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
274 npages >>= IO_PAGE_SHIFT;
275
276 spin_lock_irqsave(&iommu->lock, flags);
277 entry = pci_arena_alloc(&iommu->arena, npages);
278 spin_unlock_irqrestore(&iommu->lock, flags);
279
280 if (unlikely(entry < 0L))
281 goto bad;
282
283 bus_addr = (iommu->page_table_map_base +
284 (entry << IO_PAGE_SHIFT));
285 ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
286 base_paddr = __pa(oaddr & IO_PAGE_MASK);
287 prot = HV_PCI_MAP_ATTR_READ;
288 if (direction != PCI_DMA_TODEVICE)
289 prot |= HV_PCI_MAP_ATTR_WRITE;
290
291 local_irq_save(flags);
292
293 pci_iommu_batch_start(pdev, prot, entry);
294
295 for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
296 long err = pci_iommu_batch_add(base_paddr);
297 if (unlikely(err < 0L))
298 goto iommu_map_fail;
299 }
300 if (unlikely(pci_iommu_batch_end() < 0L))
301 goto iommu_map_fail;
302
303 local_irq_restore(flags);
304
305 return ret;
306
307 bad:
308 if (printk_ratelimit())
309 WARN_ON(1);
310 return PCI_DMA_ERROR_CODE;
311
312 iommu_map_fail:
313 /* Interrupts are disabled. */
314 spin_lock(&iommu->lock);
315 pci_arena_free(&iommu->arena, entry, npages);
316 spin_unlock_irqrestore(&iommu->lock, flags);
317
318 return PCI_DMA_ERROR_CODE;
319 }
320
321 static void pci_4v_unmap_single(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
322 {
323 struct pci_pbm_info *pbm;
324 struct pci_iommu *iommu;
325 unsigned long flags, npages;
326 long entry;
327 u32 devhandle;
328
329 if (unlikely(direction == PCI_DMA_NONE)) {
330 if (printk_ratelimit())
331 WARN_ON(1);
332 return;
333 }
334
335 iommu = pdev->dev.archdata.iommu;
336 pbm = pdev->dev.archdata.host_controller;
337 devhandle = pbm->devhandle;
338
339 npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
340 npages >>= IO_PAGE_SHIFT;
341 bus_addr &= IO_PAGE_MASK;
342
343 spin_lock_irqsave(&iommu->lock, flags);
344
345 entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
346 pci_arena_free(&iommu->arena, entry, npages);
347
348 do {
349 unsigned long num;
350
351 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
352 npages);
353 entry += num;
354 npages -= num;
355 } while (npages != 0);
356
357 spin_unlock_irqrestore(&iommu->lock, flags);
358 }
359
360 #define SG_ENT_PHYS_ADDRESS(SG) \
361 (__pa(page_address((SG)->page)) + (SG)->offset)
362
363 static inline long fill_sg(long entry, struct pci_dev *pdev,
364 struct scatterlist *sg,
365 int nused, int nelems, unsigned long prot)
366 {
367 struct scatterlist *dma_sg = sg;
368 struct scatterlist *sg_end = sg + nelems;
369 unsigned long flags;
370 int i;
371
372 local_irq_save(flags);
373
374 pci_iommu_batch_start(pdev, prot, entry);
375
376 for (i = 0; i < nused; i++) {
377 unsigned long pteval = ~0UL;
378 u32 dma_npages;
379
380 dma_npages = ((dma_sg->dma_address & (IO_PAGE_SIZE - 1UL)) +
381 dma_sg->dma_length +
382 ((IO_PAGE_SIZE - 1UL))) >> IO_PAGE_SHIFT;
383 do {
384 unsigned long offset;
385 signed int len;
386
387 /* If we are here, we know we have at least one
388 * more page to map. So walk forward until we
389 * hit a page crossing, and begin creating new
390 * mappings from that spot.
391 */
392 for (;;) {
393 unsigned long tmp;
394
395 tmp = SG_ENT_PHYS_ADDRESS(sg);
396 len = sg->length;
397 if (((tmp ^ pteval) >> IO_PAGE_SHIFT) != 0UL) {
398 pteval = tmp & IO_PAGE_MASK;
399 offset = tmp & (IO_PAGE_SIZE - 1UL);
400 break;
401 }
402 if (((tmp ^ (tmp + len - 1UL)) >> IO_PAGE_SHIFT) != 0UL) {
403 pteval = (tmp + IO_PAGE_SIZE) & IO_PAGE_MASK;
404 offset = 0UL;
405 len -= (IO_PAGE_SIZE - (tmp & (IO_PAGE_SIZE - 1UL)));
406 break;
407 }
408 sg++;
409 }
410
411 pteval = (pteval & IOPTE_PAGE);
412 while (len > 0) {
413 long err;
414
415 err = pci_iommu_batch_add(pteval);
416 if (unlikely(err < 0L))
417 goto iommu_map_failed;
418
419 pteval += IO_PAGE_SIZE;
420 len -= (IO_PAGE_SIZE - offset);
421 offset = 0;
422 dma_npages--;
423 }
424
425 pteval = (pteval & IOPTE_PAGE) + len;
426 sg++;
427
428 /* Skip over any tail mappings we've fully mapped,
429 * adjusting pteval along the way. Stop when we
430 * detect a page crossing event.
431 */
432 while (sg < sg_end &&
433 (pteval << (64 - IO_PAGE_SHIFT)) != 0UL &&
434 (pteval == SG_ENT_PHYS_ADDRESS(sg)) &&
435 ((pteval ^
436 (SG_ENT_PHYS_ADDRESS(sg) + sg->length - 1UL)) >> IO_PAGE_SHIFT) == 0UL) {
437 pteval += sg->length;
438 sg++;
439 }
440 if ((pteval << (64 - IO_PAGE_SHIFT)) == 0UL)
441 pteval = ~0UL;
442 } while (dma_npages != 0);
443 dma_sg++;
444 }
445
446 if (unlikely(pci_iommu_batch_end() < 0L))
447 goto iommu_map_failed;
448
449 local_irq_restore(flags);
450 return 0;
451
452 iommu_map_failed:
453 local_irq_restore(flags);
454 return -1L;
455 }
456
457 static int pci_4v_map_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
458 {
459 struct pci_iommu *iommu;
460 unsigned long flags, npages, prot;
461 u32 dma_base;
462 struct scatterlist *sgtmp;
463 long entry, err;
464 int used;
465
466 /* Fast path single entry scatterlists. */
467 if (nelems == 1) {
468 sglist->dma_address =
469 pci_4v_map_single(pdev,
470 (page_address(sglist->page) + sglist->offset),
471 sglist->length, direction);
472 if (unlikely(sglist->dma_address == PCI_DMA_ERROR_CODE))
473 return 0;
474 sglist->dma_length = sglist->length;
475 return 1;
476 }
477
478 iommu = pdev->dev.archdata.iommu;
479
480 if (unlikely(direction == PCI_DMA_NONE))
481 goto bad;
482
483 /* Step 1: Prepare scatter list. */
484 npages = prepare_sg(sglist, nelems);
485
486 /* Step 2: Allocate a cluster and context, if necessary. */
487 spin_lock_irqsave(&iommu->lock, flags);
488 entry = pci_arena_alloc(&iommu->arena, npages);
489 spin_unlock_irqrestore(&iommu->lock, flags);
490
491 if (unlikely(entry < 0L))
492 goto bad;
493
494 dma_base = iommu->page_table_map_base +
495 (entry << IO_PAGE_SHIFT);
496
497 /* Step 3: Normalize DMA addresses. */
498 used = nelems;
499
500 sgtmp = sglist;
501 while (used && sgtmp->dma_length) {
502 sgtmp->dma_address += dma_base;
503 sgtmp++;
504 used--;
505 }
506 used = nelems - used;
507
508 /* Step 4: Create the mappings. */
509 prot = HV_PCI_MAP_ATTR_READ;
510 if (direction != PCI_DMA_TODEVICE)
511 prot |= HV_PCI_MAP_ATTR_WRITE;
512
513 err = fill_sg(entry, pdev, sglist, used, nelems, prot);
514 if (unlikely(err < 0L))
515 goto iommu_map_failed;
516
517 return used;
518
519 bad:
520 if (printk_ratelimit())
521 WARN_ON(1);
522 return 0;
523
524 iommu_map_failed:
525 spin_lock_irqsave(&iommu->lock, flags);
526 pci_arena_free(&iommu->arena, entry, npages);
527 spin_unlock_irqrestore(&iommu->lock, flags);
528
529 return 0;
530 }
531
532 static void pci_4v_unmap_sg(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
533 {
534 struct pci_pbm_info *pbm;
535 struct pci_iommu *iommu;
536 unsigned long flags, i, npages;
537 long entry;
538 u32 devhandle, bus_addr;
539
540 if (unlikely(direction == PCI_DMA_NONE)) {
541 if (printk_ratelimit())
542 WARN_ON(1);
543 }
544
545 iommu = pdev->dev.archdata.iommu;
546 pbm = pdev->dev.archdata.host_controller;
547 devhandle = pbm->devhandle;
548
549 bus_addr = sglist->dma_address & IO_PAGE_MASK;
550
551 for (i = 1; i < nelems; i++)
552 if (sglist[i].dma_length == 0)
553 break;
554 i--;
555 npages = (IO_PAGE_ALIGN(sglist[i].dma_address + sglist[i].dma_length) -
556 bus_addr) >> IO_PAGE_SHIFT;
557
558 entry = ((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
559
560 spin_lock_irqsave(&iommu->lock, flags);
561
562 pci_arena_free(&iommu->arena, entry, npages);
563
564 do {
565 unsigned long num;
566
567 num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
568 npages);
569 entry += num;
570 npages -= num;
571 } while (npages != 0);
572
573 spin_unlock_irqrestore(&iommu->lock, flags);
574 }
575
576 static void pci_4v_dma_sync_single_for_cpu(struct pci_dev *pdev, dma_addr_t bus_addr, size_t sz, int direction)
577 {
578 /* Nothing to do... */
579 }
580
581 static void pci_4v_dma_sync_sg_for_cpu(struct pci_dev *pdev, struct scatterlist *sglist, int nelems, int direction)
582 {
583 /* Nothing to do... */
584 }
585
586 const struct pci_iommu_ops pci_sun4v_iommu_ops = {
587 .alloc_consistent = pci_4v_alloc_consistent,
588 .free_consistent = pci_4v_free_consistent,
589 .map_single = pci_4v_map_single,
590 .unmap_single = pci_4v_unmap_single,
591 .map_sg = pci_4v_map_sg,
592 .unmap_sg = pci_4v_unmap_sg,
593 .dma_sync_single_for_cpu = pci_4v_dma_sync_single_for_cpu,
594 .dma_sync_sg_for_cpu = pci_4v_dma_sync_sg_for_cpu,
595 };
596
597 static inline int pci_sun4v_out_of_range(struct pci_pbm_info *pbm, unsigned int bus, unsigned int device, unsigned int func)
598 {
599 if (bus < pbm->pci_first_busno ||
600 bus > pbm->pci_last_busno)
601 return 1;
602 return 0;
603 }
604
605 static int pci_sun4v_read_pci_cfg(struct pci_bus *bus_dev, unsigned int devfn,
606 int where, int size, u32 *value)
607 {
608 struct pci_pbm_info *pbm = bus_dev->sysdata;
609 u32 devhandle = pbm->devhandle;
610 unsigned int bus = bus_dev->number;
611 unsigned int device = PCI_SLOT(devfn);
612 unsigned int func = PCI_FUNC(devfn);
613 unsigned long ret;
614
615 if (bus_dev == pbm->pci_bus && devfn == 0x00)
616 return pci_host_bridge_read_pci_cfg(bus_dev, devfn, where,
617 size, value);
618 if (pci_sun4v_out_of_range(pbm, bus, device, func)) {
619 ret = ~0UL;
620 } else {
621 ret = pci_sun4v_config_get(devhandle,
622 HV_PCI_DEVICE_BUILD(bus, device, func),
623 where, size);
624 #if 0
625 printk("rcfg: [%x:%x:%x:%d]=[%lx]\n",
626 devhandle, HV_PCI_DEVICE_BUILD(bus, device, func),
627 where, size, ret);
628 #endif
629 }
630 switch (size) {
631 case 1:
632 *value = ret & 0xff;
633 break;
634 case 2:
635 *value = ret & 0xffff;
636 break;
637 case 4:
638 *value = ret & 0xffffffff;
639 break;
640 };
641
642
643 return PCIBIOS_SUCCESSFUL;
644 }
645
646 static int pci_sun4v_write_pci_cfg(struct pci_bus *bus_dev, unsigned int devfn,
647 int where, int size, u32 value)
648 {
649 struct pci_pbm_info *pbm = bus_dev->sysdata;
650 u32 devhandle = pbm->devhandle;
651 unsigned int bus = bus_dev->number;
652 unsigned int device = PCI_SLOT(devfn);
653 unsigned int func = PCI_FUNC(devfn);
654 unsigned long ret;
655
656 if (bus_dev == pbm->pci_bus && devfn == 0x00)
657 return pci_host_bridge_write_pci_cfg(bus_dev, devfn, where,
658 size, value);
659 if (pci_sun4v_out_of_range(pbm, bus, device, func)) {
660 /* Do nothing. */
661 } else {
662 ret = pci_sun4v_config_put(devhandle,
663 HV_PCI_DEVICE_BUILD(bus, device, func),
664 where, size, value);
665 #if 0
666 printk("wcfg: [%x:%x:%x:%d] v[%x] == [%lx]\n",
667 devhandle, HV_PCI_DEVICE_BUILD(bus, device, func),
668 where, size, value, ret);
669 #endif
670 }
671 return PCIBIOS_SUCCESSFUL;
672 }
673
674 static struct pci_ops pci_sun4v_ops = {
675 .read = pci_sun4v_read_pci_cfg,
676 .write = pci_sun4v_write_pci_cfg,
677 };
678
679
680 static void pbm_scan_bus(struct pci_controller_info *p,
681 struct pci_pbm_info *pbm)
682 {
683 pbm->pci_bus = pci_scan_one_pbm(pbm);
684 }
685
686 static void pci_sun4v_scan_bus(struct pci_controller_info *p)
687 {
688 struct property *prop;
689 struct device_node *dp;
690
691 if ((dp = p->pbm_A.prom_node) != NULL) {
692 prop = of_find_property(dp, "66mhz-capable", NULL);
693 p->pbm_A.is_66mhz_capable = (prop != NULL);
694
695 pbm_scan_bus(p, &p->pbm_A);
696 }
697 if ((dp = p->pbm_B.prom_node) != NULL) {
698 prop = of_find_property(dp, "66mhz-capable", NULL);
699 p->pbm_B.is_66mhz_capable = (prop != NULL);
700
701 pbm_scan_bus(p, &p->pbm_B);
702 }
703
704 /* XXX register error interrupt handlers XXX */
705 }
706
707 static unsigned long probe_existing_entries(struct pci_pbm_info *pbm,
708 struct pci_iommu *iommu)
709 {
710 struct pci_iommu_arena *arena = &iommu->arena;
711 unsigned long i, cnt = 0;
712 u32 devhandle;
713
714 devhandle = pbm->devhandle;
715 for (i = 0; i < arena->limit; i++) {
716 unsigned long ret, io_attrs, ra;
717
718 ret = pci_sun4v_iommu_getmap(devhandle,
719 HV_PCI_TSBID(0, i),
720 &io_attrs, &ra);
721 if (ret == HV_EOK) {
722 if (page_in_phys_avail(ra)) {
723 pci_sun4v_iommu_demap(devhandle,
724 HV_PCI_TSBID(0, i), 1);
725 } else {
726 cnt++;
727 __set_bit(i, arena->map);
728 }
729 }
730 }
731
732 return cnt;
733 }
734
735 static void pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
736 {
737 struct pci_iommu *iommu = pbm->iommu;
738 struct property *prop;
739 unsigned long num_tsb_entries, sz;
740 u32 vdma[2], dma_mask, dma_offset;
741 int tsbsize;
742
743 prop = of_find_property(pbm->prom_node, "virtual-dma", NULL);
744 if (prop) {
745 u32 *val = prop->value;
746
747 vdma[0] = val[0];
748 vdma[1] = val[1];
749 } else {
750 /* No property, use default values. */
751 vdma[0] = 0x80000000;
752 vdma[1] = 0x80000000;
753 }
754
755 dma_mask = vdma[0];
756 switch (vdma[1]) {
757 case 0x20000000:
758 dma_mask |= 0x1fffffff;
759 tsbsize = 64;
760 break;
761
762 case 0x40000000:
763 dma_mask |= 0x3fffffff;
764 tsbsize = 128;
765 break;
766
767 case 0x80000000:
768 dma_mask |= 0x7fffffff;
769 tsbsize = 256;
770 break;
771
772 default:
773 prom_printf("PCI-SUN4V: strange virtual-dma size.\n");
774 prom_halt();
775 };
776
777 tsbsize *= (8 * 1024);
778
779 num_tsb_entries = tsbsize / sizeof(iopte_t);
780
781 dma_offset = vdma[0];
782
783 /* Setup initial software IOMMU state. */
784 spin_lock_init(&iommu->lock);
785 iommu->ctx_lowest_free = 1;
786 iommu->page_table_map_base = dma_offset;
787 iommu->dma_addr_mask = dma_mask;
788
789 /* Allocate and initialize the free area map. */
790 sz = num_tsb_entries / 8;
791 sz = (sz + 7UL) & ~7UL;
792 iommu->arena.map = kzalloc(sz, GFP_KERNEL);
793 if (!iommu->arena.map) {
794 prom_printf("PCI_IOMMU: Error, kmalloc(arena.map) failed.\n");
795 prom_halt();
796 }
797 iommu->arena.limit = num_tsb_entries;
798
799 sz = probe_existing_entries(pbm, iommu);
800 if (sz)
801 printk("%s: Imported %lu TSB entries from OBP\n",
802 pbm->name, sz);
803 }
804
805 static void pci_sun4v_get_bus_range(struct pci_pbm_info *pbm)
806 {
807 struct property *prop;
808 unsigned int *busrange;
809
810 prop = of_find_property(pbm->prom_node, "bus-range", NULL);
811
812 busrange = prop->value;
813
814 pbm->pci_first_busno = busrange[0];
815 pbm->pci_last_busno = busrange[1];
816
817 }
818
819 #ifdef CONFIG_PCI_MSI
820 struct pci_sun4v_msiq_entry {
821 u64 version_type;
822 #define MSIQ_VERSION_MASK 0xffffffff00000000UL
823 #define MSIQ_VERSION_SHIFT 32
824 #define MSIQ_TYPE_MASK 0x00000000000000ffUL
825 #define MSIQ_TYPE_SHIFT 0
826 #define MSIQ_TYPE_NONE 0x00
827 #define MSIQ_TYPE_MSG 0x01
828 #define MSIQ_TYPE_MSI32 0x02
829 #define MSIQ_TYPE_MSI64 0x03
830 #define MSIQ_TYPE_INTX 0x08
831 #define MSIQ_TYPE_NONE2 0xff
832
833 u64 intx_sysino;
834 u64 reserved1;
835 u64 stick;
836 u64 req_id; /* bus/device/func */
837 #define MSIQ_REQID_BUS_MASK 0xff00UL
838 #define MSIQ_REQID_BUS_SHIFT 8
839 #define MSIQ_REQID_DEVICE_MASK 0x00f8UL
840 #define MSIQ_REQID_DEVICE_SHIFT 3
841 #define MSIQ_REQID_FUNC_MASK 0x0007UL
842 #define MSIQ_REQID_FUNC_SHIFT 0
843
844 u64 msi_address;
845
846 /* The format of this value is message type dependant.
847 * For MSI bits 15:0 are the data from the MSI packet.
848 * For MSI-X bits 31:0 are the data from the MSI packet.
849 * For MSG, the message code and message routing code where:
850 * bits 39:32 is the bus/device/fn of the msg target-id
851 * bits 18:16 is the message routing code
852 * bits 7:0 is the message code
853 * For INTx the low order 2-bits are:
854 * 00 - INTA
855 * 01 - INTB
856 * 10 - INTC
857 * 11 - INTD
858 */
859 u64 msi_data;
860
861 u64 reserved2;
862 };
863
864 /* For now this just runs as a pre-handler for the real interrupt handler.
865 * So we just walk through the queue and ACK all the entries, update the
866 * head pointer, and return.
867 *
868 * In the longer term it would be nice to do something more integrated
869 * wherein we can pass in some of this MSI info to the drivers. This
870 * would be most useful for PCIe fabric error messages, although we could
871 * invoke those directly from the loop here in order to pass the info around.
872 */
873 static void pci_sun4v_msi_prehandler(unsigned int ino, void *data1, void *data2)
874 {
875 struct pci_pbm_info *pbm = data1;
876 struct pci_sun4v_msiq_entry *base, *ep;
877 unsigned long msiqid, orig_head, head, type, err;
878
879 msiqid = (unsigned long) data2;
880
881 head = 0xdeadbeef;
882 err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, &head);
883 if (unlikely(err))
884 goto hv_error_get;
885
886 if (unlikely(head >= (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry))))
887 goto bad_offset;
888
889 head /= sizeof(struct pci_sun4v_msiq_entry);
890 orig_head = head;
891 base = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
892 (pbm->msiq_ent_count *
893 sizeof(struct pci_sun4v_msiq_entry))));
894 ep = &base[head];
895 while ((ep->version_type & MSIQ_TYPE_MASK) != 0) {
896 type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
897 if (unlikely(type != MSIQ_TYPE_MSI32 &&
898 type != MSIQ_TYPE_MSI64))
899 goto bad_type;
900
901 pci_sun4v_msi_setstate(pbm->devhandle,
902 ep->msi_data /* msi_num */,
903 HV_MSISTATE_IDLE);
904
905 /* Clear the entry. */
906 ep->version_type &= ~MSIQ_TYPE_MASK;
907
908 /* Go to next entry in ring. */
909 head++;
910 if (head >= pbm->msiq_ent_count)
911 head = 0;
912 ep = &base[head];
913 }
914
915 if (likely(head != orig_head)) {
916 /* ACK entries by updating head pointer. */
917 head *= sizeof(struct pci_sun4v_msiq_entry);
918 err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
919 if (unlikely(err))
920 goto hv_error_set;
921 }
922 return;
923
924 hv_error_set:
925 printk(KERN_EMERG "MSI: Hypervisor set head gives error %lu\n", err);
926 goto hv_error_cont;
927
928 hv_error_get:
929 printk(KERN_EMERG "MSI: Hypervisor get head gives error %lu\n", err);
930
931 hv_error_cont:
932 printk(KERN_EMERG "MSI: devhandle[%x] msiqid[%lx] head[%lu]\n",
933 pbm->devhandle, msiqid, head);
934 return;
935
936 bad_offset:
937 printk(KERN_EMERG "MSI: Hypervisor gives bad offset %lx max(%lx)\n",
938 head, pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry));
939 return;
940
941 bad_type:
942 printk(KERN_EMERG "MSI: Entry has bad type %lx\n", type);
943 return;
944 }
945
946 static int msi_bitmap_alloc(struct pci_pbm_info *pbm)
947 {
948 unsigned long size, bits_per_ulong;
949
950 bits_per_ulong = sizeof(unsigned long) * 8;
951 size = (pbm->msi_num + (bits_per_ulong - 1)) & ~(bits_per_ulong - 1);
952 size /= 8;
953 BUG_ON(size % sizeof(unsigned long));
954
955 pbm->msi_bitmap = kzalloc(size, GFP_KERNEL);
956 if (!pbm->msi_bitmap)
957 return -ENOMEM;
958
959 return 0;
960 }
961
962 static void msi_bitmap_free(struct pci_pbm_info *pbm)
963 {
964 kfree(pbm->msi_bitmap);
965 pbm->msi_bitmap = NULL;
966 }
967
968 static int msi_queue_alloc(struct pci_pbm_info *pbm)
969 {
970 unsigned long q_size, alloc_size, pages, order;
971 int i;
972
973 q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
974 alloc_size = (pbm->msiq_num * q_size);
975 order = get_order(alloc_size);
976 pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
977 if (pages == 0UL) {
978 printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
979 order);
980 return -ENOMEM;
981 }
982 memset((char *)pages, 0, PAGE_SIZE << order);
983 pbm->msi_queues = (void *) pages;
984
985 for (i = 0; i < pbm->msiq_num; i++) {
986 unsigned long err, base = __pa(pages + (i * q_size));
987 unsigned long ret1, ret2;
988
989 err = pci_sun4v_msiq_conf(pbm->devhandle,
990 pbm->msiq_first + i,
991 base, pbm->msiq_ent_count);
992 if (err) {
993 printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
994 err);
995 goto h_error;
996 }
997
998 err = pci_sun4v_msiq_info(pbm->devhandle,
999 pbm->msiq_first + i,
1000 &ret1, &ret2);
1001 if (err) {
1002 printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
1003 err);
1004 goto h_error;
1005 }
1006 if (ret1 != base || ret2 != pbm->msiq_ent_count) {
1007 printk(KERN_ERR "MSI: Bogus qconf "
1008 "expected[%lx:%x] got[%lx:%lx]\n",
1009 base, pbm->msiq_ent_count,
1010 ret1, ret2);
1011 goto h_error;
1012 }
1013 }
1014
1015 return 0;
1016
1017 h_error:
1018 free_pages(pages, order);
1019 return -EINVAL;
1020 }
1021
1022 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
1023 {
1024 const u32 *val;
1025 int len;
1026
1027 val = of_get_property(pbm->prom_node, "#msi-eqs", &len);
1028 if (!val || len != 4)
1029 goto no_msi;
1030 pbm->msiq_num = *val;
1031 if (pbm->msiq_num) {
1032 const struct msiq_prop {
1033 u32 first_msiq;
1034 u32 num_msiq;
1035 u32 first_devino;
1036 } *mqp;
1037 const struct msi_range_prop {
1038 u32 first_msi;
1039 u32 num_msi;
1040 } *mrng;
1041 const struct addr_range_prop {
1042 u32 msi32_high;
1043 u32 msi32_low;
1044 u32 msi32_len;
1045 u32 msi64_high;
1046 u32 msi64_low;
1047 u32 msi64_len;
1048 } *arng;
1049
1050 val = of_get_property(pbm->prom_node, "msi-eq-size", &len);
1051 if (!val || len != 4)
1052 goto no_msi;
1053
1054 pbm->msiq_ent_count = *val;
1055
1056 mqp = of_get_property(pbm->prom_node,
1057 "msi-eq-to-devino", &len);
1058 if (!mqp || len != sizeof(struct msiq_prop))
1059 goto no_msi;
1060
1061 pbm->msiq_first = mqp->first_msiq;
1062 pbm->msiq_first_devino = mqp->first_devino;
1063
1064 val = of_get_property(pbm->prom_node, "#msi", &len);
1065 if (!val || len != 4)
1066 goto no_msi;
1067 pbm->msi_num = *val;
1068
1069 mrng = of_get_property(pbm->prom_node, "msi-ranges", &len);
1070 if (!mrng || len != sizeof(struct msi_range_prop))
1071 goto no_msi;
1072 pbm->msi_first = mrng->first_msi;
1073
1074 val = of_get_property(pbm->prom_node, "msi-data-mask", &len);
1075 if (!val || len != 4)
1076 goto no_msi;
1077 pbm->msi_data_mask = *val;
1078
1079 val = of_get_property(pbm->prom_node, "msix-data-width", &len);
1080 if (!val || len != 4)
1081 goto no_msi;
1082 pbm->msix_data_width = *val;
1083
1084 arng = of_get_property(pbm->prom_node, "msi-address-ranges",
1085 &len);
1086 if (!arng || len != sizeof(struct addr_range_prop))
1087 goto no_msi;
1088 pbm->msi32_start = ((u64)arng->msi32_high << 32) |
1089 (u64) arng->msi32_low;
1090 pbm->msi64_start = ((u64)arng->msi64_high << 32) |
1091 (u64) arng->msi64_low;
1092 pbm->msi32_len = arng->msi32_len;
1093 pbm->msi64_len = arng->msi64_len;
1094
1095 if (msi_bitmap_alloc(pbm))
1096 goto no_msi;
1097
1098 if (msi_queue_alloc(pbm)) {
1099 msi_bitmap_free(pbm);
1100 goto no_msi;
1101 }
1102
1103 printk(KERN_INFO "%s: MSI Queue first[%u] num[%u] count[%u] "
1104 "devino[0x%x]\n",
1105 pbm->name,
1106 pbm->msiq_first, pbm->msiq_num,
1107 pbm->msiq_ent_count,
1108 pbm->msiq_first_devino);
1109 printk(KERN_INFO "%s: MSI first[%u] num[%u] mask[0x%x] "
1110 "width[%u]\n",
1111 pbm->name,
1112 pbm->msi_first, pbm->msi_num, pbm->msi_data_mask,
1113 pbm->msix_data_width);
1114 printk(KERN_INFO "%s: MSI addr32[0x%lx:0x%x] "
1115 "addr64[0x%lx:0x%x]\n",
1116 pbm->name,
1117 pbm->msi32_start, pbm->msi32_len,
1118 pbm->msi64_start, pbm->msi64_len);
1119 printk(KERN_INFO "%s: MSI queues at RA [%p]\n",
1120 pbm->name,
1121 pbm->msi_queues);
1122 }
1123
1124 return;
1125
1126 no_msi:
1127 pbm->msiq_num = 0;
1128 printk(KERN_INFO "%s: No MSI support.\n", pbm->name);
1129 }
1130
1131 static int alloc_msi(struct pci_pbm_info *pbm)
1132 {
1133 int i;
1134
1135 for (i = 0; i < pbm->msi_num; i++) {
1136 if (!test_and_set_bit(i, pbm->msi_bitmap))
1137 return i + pbm->msi_first;
1138 }
1139
1140 return -ENOENT;
1141 }
1142
1143 static void free_msi(struct pci_pbm_info *pbm, int msi_num)
1144 {
1145 msi_num -= pbm->msi_first;
1146 clear_bit(msi_num, pbm->msi_bitmap);
1147 }
1148
1149 static int pci_sun4v_setup_msi_irq(unsigned int *virt_irq_p,
1150 struct pci_dev *pdev,
1151 struct msi_desc *entry)
1152 {
1153 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
1154 unsigned long devino, msiqid;
1155 struct msi_msg msg;
1156 int msi_num, err;
1157
1158 *virt_irq_p = 0;
1159
1160 msi_num = alloc_msi(pbm);
1161 if (msi_num < 0)
1162 return msi_num;
1163
1164 devino = sun4v_build_msi(pbm->devhandle, virt_irq_p,
1165 pbm->msiq_first_devino,
1166 (pbm->msiq_first_devino +
1167 pbm->msiq_num));
1168 err = -ENOMEM;
1169 if (!devino)
1170 goto out_err;
1171
1172 set_irq_msi(*virt_irq_p, entry);
1173
1174 msiqid = ((devino - pbm->msiq_first_devino) +
1175 pbm->msiq_first);
1176
1177 err = -EINVAL;
1178 if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
1179 if (err)
1180 goto out_err;
1181
1182 if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
1183 goto out_err;
1184
1185 if (pci_sun4v_msi_setmsiq(pbm->devhandle,
1186 msi_num, msiqid,
1187 (entry->msi_attrib.is_64 ?
1188 HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
1189 goto out_err;
1190
1191 if (pci_sun4v_msi_setstate(pbm->devhandle, msi_num, HV_MSISTATE_IDLE))
1192 goto out_err;
1193
1194 if (pci_sun4v_msi_setvalid(pbm->devhandle, msi_num, HV_MSIVALID_VALID))
1195 goto out_err;
1196
1197 pdev->dev.archdata.msi_num = msi_num;
1198
1199 if (entry->msi_attrib.is_64) {
1200 msg.address_hi = pbm->msi64_start >> 32;
1201 msg.address_lo = pbm->msi64_start & 0xffffffff;
1202 } else {
1203 msg.address_hi = 0;
1204 msg.address_lo = pbm->msi32_start;
1205 }
1206 msg.data = msi_num;
1207 write_msi_msg(*virt_irq_p, &msg);
1208
1209 irq_install_pre_handler(*virt_irq_p,
1210 pci_sun4v_msi_prehandler,
1211 pbm, (void *) msiqid);
1212
1213 return 0;
1214
1215 out_err:
1216 free_msi(pbm, msi_num);
1217 sun4v_destroy_msi(*virt_irq_p);
1218 *virt_irq_p = 0;
1219 return err;
1220
1221 }
1222
1223 static void pci_sun4v_teardown_msi_irq(unsigned int virt_irq,
1224 struct pci_dev *pdev)
1225 {
1226 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
1227 unsigned long msiqid, err;
1228 unsigned int msi_num;
1229
1230 msi_num = pdev->dev.archdata.msi_num;
1231 err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi_num, &msiqid);
1232 if (err) {
1233 printk(KERN_ERR "%s: getmsiq gives error %lu\n",
1234 pbm->name, err);
1235 return;
1236 }
1237
1238 pci_sun4v_msi_setvalid(pbm->devhandle, msi_num, HV_MSIVALID_INVALID);
1239 pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_INVALID);
1240
1241 free_msi(pbm, msi_num);
1242
1243 /* The sun4v_destroy_msi() will liberate the devino and thus the MSIQ
1244 * allocation.
1245 */
1246 sun4v_destroy_msi(virt_irq);
1247 }
1248 #else /* CONFIG_PCI_MSI */
1249 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
1250 {
1251 }
1252 #endif /* !(CONFIG_PCI_MSI) */
1253
1254 static void pci_sun4v_pbm_init(struct pci_controller_info *p, struct device_node *dp, u32 devhandle)
1255 {
1256 struct pci_pbm_info *pbm;
1257
1258 if (devhandle & 0x40)
1259 pbm = &p->pbm_B;
1260 else
1261 pbm = &p->pbm_A;
1262
1263 pbm->parent = p;
1264 pbm->prom_node = dp;
1265
1266 pbm->devhandle = devhandle;
1267
1268 pbm->name = dp->full_name;
1269
1270 printk("%s: SUN4V PCI Bus Module\n", pbm->name);
1271
1272 pci_determine_mem_io_space(pbm);
1273
1274 pci_sun4v_get_bus_range(pbm);
1275 pci_sun4v_iommu_init(pbm);
1276 pci_sun4v_msi_init(pbm);
1277 }
1278
1279 void sun4v_pci_init(struct device_node *dp, char *model_name)
1280 {
1281 struct pci_controller_info *p;
1282 struct pci_iommu *iommu;
1283 struct property *prop;
1284 struct linux_prom64_registers *regs;
1285 u32 devhandle;
1286 int i;
1287
1288 prop = of_find_property(dp, "reg", NULL);
1289 regs = prop->value;
1290
1291 devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
1292
1293 for (p = pci_controller_root; p; p = p->next) {
1294 struct pci_pbm_info *pbm;
1295
1296 if (p->pbm_A.prom_node && p->pbm_B.prom_node)
1297 continue;
1298
1299 pbm = (p->pbm_A.prom_node ?
1300 &p->pbm_A :
1301 &p->pbm_B);
1302
1303 if (pbm->devhandle == (devhandle ^ 0x40)) {
1304 pci_sun4v_pbm_init(p, dp, devhandle);
1305 return;
1306 }
1307 }
1308
1309 for_each_possible_cpu(i) {
1310 unsigned long page = get_zeroed_page(GFP_ATOMIC);
1311
1312 if (!page)
1313 goto fatal_memory_error;
1314
1315 per_cpu(pci_iommu_batch, i).pglist = (u64 *) page;
1316 }
1317
1318 p = kzalloc(sizeof(struct pci_controller_info), GFP_ATOMIC);
1319 if (!p)
1320 goto fatal_memory_error;
1321
1322 iommu = kzalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
1323 if (!iommu)
1324 goto fatal_memory_error;
1325
1326 p->pbm_A.iommu = iommu;
1327
1328 iommu = kzalloc(sizeof(struct pci_iommu), GFP_ATOMIC);
1329 if (!iommu)
1330 goto fatal_memory_error;
1331
1332 p->pbm_B.iommu = iommu;
1333
1334 p->next = pci_controller_root;
1335 pci_controller_root = p;
1336
1337 p->index = pci_num_controllers++;
1338
1339 p->scan_bus = pci_sun4v_scan_bus;
1340 #ifdef CONFIG_PCI_MSI
1341 p->setup_msi_irq = pci_sun4v_setup_msi_irq;
1342 p->teardown_msi_irq = pci_sun4v_teardown_msi_irq;
1343 #endif
1344 p->pci_ops = &pci_sun4v_ops;
1345
1346 /* Like PSYCHO and SCHIZO we have a 2GB aligned area
1347 * for memory space.
1348 */
1349 pci_memspace_mask = 0x7fffffffUL;
1350
1351 pci_sun4v_pbm_init(p, dp, devhandle);
1352 return;
1353
1354 fatal_memory_error:
1355 prom_printf("SUN4V_PCI: Fatal memory allocation error.\n");
1356 prom_halt();
1357 }
ubuntu-zesty-kernel mirror