?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
OMAPDSS: VENC: fix NULL pointer dereference in DSS2 VENC sysfs debug attr on OMAP4
[zen-stable.git] / drivers / gpu / drm / nouveau / nouveau_bo.c
blobec54364ac828f1bd69c13e42cf50507db5e913f6
1 /*
2 * Copyright 2007 Dave Airlied
3 * All Rights Reserved.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 */
24 /*
25 * Authors: Dave Airlied <airlied@linux.ie>
26 * Ben Skeggs <darktama@iinet.net.au>
27 * Jeremy Kolb <jkolb@brandeis.edu>
28 */
29
30 #include "drmP.h"
31 #include "ttm/ttm_page_alloc.h"
32
33 #include "nouveau_drm.h"
34 #include "nouveau_drv.h"
35 #include "nouveau_dma.h"
36 #include "nouveau_mm.h"
37 #include "nouveau_vm.h"
38
39 #include <linux/log2.h>
40 #include <linux/slab.h>
41
42 static void
43 nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
44 {
45 struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
46 struct drm_device *dev = dev_priv->dev;
47 struct nouveau_bo *nvbo = nouveau_bo(bo);
48
49 if (unlikely(nvbo->gem))
50 DRM_ERROR("bo %p still attached to GEM object\n", bo);
51
52 nv10_mem_put_tile_region(dev, nvbo->tile, NULL);
53 kfree(nvbo);
54 }
55
56 static void
57 nouveau_bo_fixup_align(struct nouveau_bo *nvbo, u32 flags,
58 int *align, int *size)
59 {
60 struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev);
61
62 if (dev_priv->card_type < NV_50) {
63 if (nvbo->tile_mode) {
64 if (dev_priv->chipset >= 0x40) {
65 *align = 65536;
66 *size = roundup(*size, 64 * nvbo->tile_mode);
67
68 } else if (dev_priv->chipset >= 0x30) {
69 *align = 32768;
70 *size = roundup(*size, 64 * nvbo->tile_mode);
71
72 } else if (dev_priv->chipset >= 0x20) {
73 *align = 16384;
74 *size = roundup(*size, 64 * nvbo->tile_mode);
75
76 } else if (dev_priv->chipset >= 0x10) {
77 *align = 16384;
78 *size = roundup(*size, 32 * nvbo->tile_mode);
79 }
80 }
81 } else {
82 *size = roundup(*size, (1 << nvbo->page_shift));
83 *align = max((1 << nvbo->page_shift), *align);
84 }
85
86 *size = roundup(*size, PAGE_SIZE);
87 }
88
89 int
90 nouveau_bo_new(struct drm_device *dev, int size, int align,
91 uint32_t flags, uint32_t tile_mode, uint32_t tile_flags,
92 struct nouveau_bo **pnvbo)
93 {
94 struct drm_nouveau_private *dev_priv = dev->dev_private;
95 struct nouveau_bo *nvbo;
96 size_t acc_size;
97 int ret;
98
99 nvbo = kzalloc(sizeof(struct nouveau_bo), GFP_KERNEL);
100 if (!nvbo)
101 return -ENOMEM;
102 INIT_LIST_HEAD(&nvbo->head);
103 INIT_LIST_HEAD(&nvbo->entry);
104 INIT_LIST_HEAD(&nvbo->vma_list);
105 nvbo->tile_mode = tile_mode;
106 nvbo->tile_flags = tile_flags;
107 nvbo->bo.bdev = &dev_priv->ttm.bdev;
108
109 nvbo->page_shift = 12;
110 if (dev_priv->bar1_vm) {
111 if (!(flags & TTM_PL_FLAG_TT) && size > 256 * 1024)
112 nvbo->page_shift = dev_priv->bar1_vm->lpg_shift;
113 }
114
115 nouveau_bo_fixup_align(nvbo, flags, &align, &size);
116 nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
117 nouveau_bo_placement_set(nvbo, flags, 0);
118
119 acc_size = ttm_bo_dma_acc_size(&dev_priv->ttm.bdev, size,
120 sizeof(struct nouveau_bo));
121
122 ret = ttm_bo_init(&dev_priv->ttm.bdev, &nvbo->bo, size,
123 ttm_bo_type_device, &nvbo->placement,
124 align >> PAGE_SHIFT, 0, false, NULL, acc_size,
125 nouveau_bo_del_ttm);
126 if (ret) {
127 /* ttm will call nouveau_bo_del_ttm if it fails.. */
128 return ret;
129 }
130
131 *pnvbo = nvbo;
132 return 0;
133 }
134
135 static void
136 set_placement_list(uint32_t *pl, unsigned *n, uint32_t type, uint32_t flags)
137 {
138 *n = 0;
139
140 if (type & TTM_PL_FLAG_VRAM)
141 pl[(*n)++] = TTM_PL_FLAG_VRAM | flags;
142 if (type & TTM_PL_FLAG_TT)
143 pl[(*n)++] = TTM_PL_FLAG_TT | flags;
144 if (type & TTM_PL_FLAG_SYSTEM)
145 pl[(*n)++] = TTM_PL_FLAG_SYSTEM | flags;
146 }
147
148 static void
149 set_placement_range(struct nouveau_bo *nvbo, uint32_t type)
150 {
151 struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev);
152 int vram_pages = dev_priv->vram_size >> PAGE_SHIFT;
153
154 if (dev_priv->card_type == NV_10 &&
155 nvbo->tile_mode && (type & TTM_PL_FLAG_VRAM) &&
156 nvbo->bo.mem.num_pages < vram_pages / 4) {
157 /*
158 * Make sure that the color and depth buffers are handled
159 * by independent memory controller units. Up to a 9x
160 * speed up when alpha-blending and depth-test are enabled
161 * at the same time.
162 */
163 if (nvbo->tile_flags & NOUVEAU_GEM_TILE_ZETA) {
164 nvbo->placement.fpfn = vram_pages / 2;
165 nvbo->placement.lpfn = ~0;
166 } else {
167 nvbo->placement.fpfn = 0;
168 nvbo->placement.lpfn = vram_pages / 2;
169 }
170 }
171 }
172
173 void
174 nouveau_bo_placement_set(struct nouveau_bo *nvbo, uint32_t type, uint32_t busy)
175 {
176 struct ttm_placement *pl = &nvbo->placement;
177 uint32_t flags = TTM_PL_MASK_CACHING |
178 (nvbo->pin_refcnt ? TTM_PL_FLAG_NO_EVICT : 0);
179
180 pl->placement = nvbo->placements;
181 set_placement_list(nvbo->placements, &pl->num_placement,
182 type, flags);
183
184 pl->busy_placement = nvbo->busy_placements;
185 set_placement_list(nvbo->busy_placements, &pl->num_busy_placement,
186 type | busy, flags);
187
188 set_placement_range(nvbo, type);
189 }
190
191 int
192 nouveau_bo_pin(struct nouveau_bo *nvbo, uint32_t memtype)
193 {
194 struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev);
195 struct ttm_buffer_object *bo = &nvbo->bo;
196 int ret;
197
198 if (nvbo->pin_refcnt && !(memtype & (1 << bo->mem.mem_type))) {
199 NV_ERROR(nouveau_bdev(bo->bdev)->dev,
200 "bo %p pinned elsewhere: 0x%08x vs 0x%08x\n", bo,
201 1 << bo->mem.mem_type, memtype);
202 return -EINVAL;
203 }
204
205 if (nvbo->pin_refcnt++)
206 return 0;
207
208 ret = ttm_bo_reserve(bo, false, false, false, 0);
209 if (ret)
210 goto out;
211
212 nouveau_bo_placement_set(nvbo, memtype, 0);
213
214 ret = nouveau_bo_validate(nvbo, false, false, false);
215 if (ret == 0) {
216 switch (bo->mem.mem_type) {
217 case TTM_PL_VRAM:
218 dev_priv->fb_aper_free -= bo->mem.size;
219 break;
220 case TTM_PL_TT:
221 dev_priv->gart_info.aper_free -= bo->mem.size;
222 break;
223 default:
224 break;
225 }
226 }
227 ttm_bo_unreserve(bo);
228 out:
229 if (unlikely(ret))
230 nvbo->pin_refcnt--;
231 return ret;
232 }
233
234 int
235 nouveau_bo_unpin(struct nouveau_bo *nvbo)
236 {
237 struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev);
238 struct ttm_buffer_object *bo = &nvbo->bo;
239 int ret;
240
241 if (--nvbo->pin_refcnt)
242 return 0;
243
244 ret = ttm_bo_reserve(bo, false, false, false, 0);
245 if (ret)
246 return ret;
247
248 nouveau_bo_placement_set(nvbo, bo->mem.placement, 0);
249
250 ret = nouveau_bo_validate(nvbo, false, false, false);
251 if (ret == 0) {
252 switch (bo->mem.mem_type) {
253 case TTM_PL_VRAM:
254 dev_priv->fb_aper_free += bo->mem.size;
255 break;
256 case TTM_PL_TT:
257 dev_priv->gart_info.aper_free += bo->mem.size;
258 break;
259 default:
260 break;
261 }
262 }
263
264 ttm_bo_unreserve(bo);
265 return ret;
266 }
267
268 int
269 nouveau_bo_map(struct nouveau_bo *nvbo)
270 {
271 int ret;
272
273 ret = ttm_bo_reserve(&nvbo->bo, false, false, false, 0);
274 if (ret)
275 return ret;
276
277 ret = ttm_bo_kmap(&nvbo->bo, 0, nvbo->bo.mem.num_pages, &nvbo->kmap);
278 ttm_bo_unreserve(&nvbo->bo);
279 return ret;
280 }
281
282 void
283 nouveau_bo_unmap(struct nouveau_bo *nvbo)
284 {
285 if (nvbo)
286 ttm_bo_kunmap(&nvbo->kmap);
287 }
288
289 int
290 nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
291 bool no_wait_reserve, bool no_wait_gpu)
292 {
293 int ret;
294
295 ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement, interruptible,
296 no_wait_reserve, no_wait_gpu);
297 if (ret)
298 return ret;
299
300 return 0;
301 }
302
303 u16
304 nouveau_bo_rd16(struct nouveau_bo *nvbo, unsigned index)
305 {
306 bool is_iomem;
307 u16 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem);
308 mem = &mem[index];
309 if (is_iomem)
310 return ioread16_native((void __force __iomem *)mem);
311 else
312 return *mem;
313 }
314
315 void
316 nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val)
317 {
318 bool is_iomem;
319 u16 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem);
320 mem = &mem[index];
321 if (is_iomem)
322 iowrite16_native(val, (void __force __iomem *)mem);
323 else
324 *mem = val;
325 }
326
327 u32
328 nouveau_bo_rd32(struct nouveau_bo *nvbo, unsigned index)
329 {
330 bool is_iomem;
331 u32 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem);
332 mem = &mem[index];
333 if (is_iomem)
334 return ioread32_native((void __force __iomem *)mem);
335 else
336 return *mem;
337 }
338
339 void
340 nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val)
341 {
342 bool is_iomem;
343 u32 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem);
344 mem = &mem[index];
345 if (is_iomem)
346 iowrite32_native(val, (void __force __iomem *)mem);
347 else
348 *mem = val;
349 }
350
351 static struct ttm_tt *
352 nouveau_ttm_tt_create(struct ttm_bo_device *bdev,
353 unsigned long size, uint32_t page_flags,
354 struct page *dummy_read_page)
355 {
356 struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
357 struct drm_device *dev = dev_priv->dev;
358
359 switch (dev_priv->gart_info.type) {
360 #if __OS_HAS_AGP
361 case NOUVEAU_GART_AGP:
362 return ttm_agp_tt_create(bdev, dev->agp->bridge,
363 size, page_flags, dummy_read_page);
364 #endif
365 case NOUVEAU_GART_PDMA:
366 case NOUVEAU_GART_HW:
367 return nouveau_sgdma_create_ttm(bdev, size, page_flags,
368 dummy_read_page);
369 default:
370 NV_ERROR(dev, "Unknown GART type %d\n",
371 dev_priv->gart_info.type);
372 break;
373 }
374
375 return NULL;
376 }
377
378 static int
379 nouveau_bo_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
380 {
381 /* We'll do this from user space. */
382 return 0;
383 }
384
385 static int
386 nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
387 struct ttm_mem_type_manager *man)
388 {
389 struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
390 struct drm_device *dev = dev_priv->dev;
391
392 switch (type) {
393 case TTM_PL_SYSTEM:
394 man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
395 man->available_caching = TTM_PL_MASK_CACHING;
396 man->default_caching = TTM_PL_FLAG_CACHED;
397 break;
398 case TTM_PL_VRAM:
399 if (dev_priv->card_type >= NV_50) {
400 man->func = &nouveau_vram_manager;
401 man->io_reserve_fastpath = false;
402 man->use_io_reserve_lru = true;
403 } else {
404 man->func = &ttm_bo_manager_func;
405 }
406 man->flags = TTM_MEMTYPE_FLAG_FIXED |
407 TTM_MEMTYPE_FLAG_MAPPABLE;
408 man->available_caching = TTM_PL_FLAG_UNCACHED |
409 TTM_PL_FLAG_WC;
410 man->default_caching = TTM_PL_FLAG_WC;
411 break;
412 case TTM_PL_TT:
413 if (dev_priv->card_type >= NV_50)
414 man->func = &nouveau_gart_manager;
415 else
416 man->func = &ttm_bo_manager_func;
417 switch (dev_priv->gart_info.type) {
418 case NOUVEAU_GART_AGP:
419 man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
420 man->available_caching = TTM_PL_FLAG_UNCACHED |
421 TTM_PL_FLAG_WC;
422 man->default_caching = TTM_PL_FLAG_WC;
423 break;
424 case NOUVEAU_GART_PDMA:
425 case NOUVEAU_GART_HW:
426 man->flags = TTM_MEMTYPE_FLAG_MAPPABLE |
427 TTM_MEMTYPE_FLAG_CMA;
428 man->available_caching = TTM_PL_MASK_CACHING;
429 man->default_caching = TTM_PL_FLAG_CACHED;
430 break;
431 default:
432 NV_ERROR(dev, "Unknown GART type: %d\n",
433 dev_priv->gart_info.type);
434 return -EINVAL;
435 }
436 break;
437 default:
438 NV_ERROR(dev, "Unsupported memory type %u\n", (unsigned)type);
439 return -EINVAL;
440 }
441 return 0;
442 }
443
444 static void
445 nouveau_bo_evict_flags(struct ttm_buffer_object *bo, struct ttm_placement *pl)
446 {
447 struct nouveau_bo *nvbo = nouveau_bo(bo);
448
449 switch (bo->mem.mem_type) {
450 case TTM_PL_VRAM:
451 nouveau_bo_placement_set(nvbo, TTM_PL_FLAG_TT,
452 TTM_PL_FLAG_SYSTEM);
453 break;
454 default:
455 nouveau_bo_placement_set(nvbo, TTM_PL_FLAG_SYSTEM, 0);
456 break;
457 }
458
459 *pl = nvbo->placement;
460 }
461
462
463 /* GPU-assisted copy using NV_MEMORY_TO_MEMORY_FORMAT, can access
464 * TTM_PL_{VRAM,TT} directly.
465 */
466
467 static int
468 nouveau_bo_move_accel_cleanup(struct nouveau_channel *chan,
469 struct nouveau_bo *nvbo, bool evict,
470 bool no_wait_reserve, bool no_wait_gpu,
471 struct ttm_mem_reg *new_mem)
472 {
473 struct nouveau_fence *fence = NULL;
474 int ret;
475
476 ret = nouveau_fence_new(chan, &fence, true);
477 if (ret)
478 return ret;
479
480 ret = ttm_bo_move_accel_cleanup(&nvbo->bo, fence, NULL, evict,
481 no_wait_reserve, no_wait_gpu, new_mem);
482 nouveau_fence_unref(&fence);
483 return ret;
484 }
485
486 static int
487 nvc0_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
488 struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
489 {
490 struct nouveau_mem *node = old_mem->mm_node;
491 u64 src_offset = node->vma[0].offset;
492 u64 dst_offset = node->vma[1].offset;
493 u32 page_count = new_mem->num_pages;
494 int ret;
495
496 page_count = new_mem->num_pages;
497 while (page_count) {
498 int line_count = (page_count > 2047) ? 2047 : page_count;
499
500 ret = RING_SPACE(chan, 12);
501 if (ret)
502 return ret;
503
504 BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0238, 2);
505 OUT_RING (chan, upper_32_bits(dst_offset));
506 OUT_RING (chan, lower_32_bits(dst_offset));
507 BEGIN_NVC0(chan, 2, NvSubM2MF, 0x030c, 6);
508 OUT_RING (chan, upper_32_bits(src_offset));
509 OUT_RING (chan, lower_32_bits(src_offset));
510 OUT_RING (chan, PAGE_SIZE); /* src_pitch */
511 OUT_RING (chan, PAGE_SIZE); /* dst_pitch */
512 OUT_RING (chan, PAGE_SIZE); /* line_length */
513 OUT_RING (chan, line_count);
514 BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0300, 1);
515 OUT_RING (chan, 0x00100110);
516
517 page_count -= line_count;
518 src_offset += (PAGE_SIZE * line_count);
519 dst_offset += (PAGE_SIZE * line_count);
520 }
521
522 return 0;
523 }
524
525 static int
526 nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
527 struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
528 {
529 struct nouveau_mem *node = old_mem->mm_node;
530 struct nouveau_bo *nvbo = nouveau_bo(bo);
531 u64 length = (new_mem->num_pages << PAGE_SHIFT);
532 u64 src_offset = node->vma[0].offset;
533 u64 dst_offset = node->vma[1].offset;
534 int ret;
535
536 while (length) {
537 u32 amount, stride, height;
538
539 amount = min(length, (u64)(4 * 1024 * 1024));
540 stride = 16 * 4;
541 height = amount / stride;
542
543 if (new_mem->mem_type == TTM_PL_VRAM &&
544 nouveau_bo_tile_layout(nvbo)) {
545 ret = RING_SPACE(chan, 8);
546 if (ret)
547 return ret;
548
549 BEGIN_RING(chan, NvSubM2MF, 0x0200, 7);
550 OUT_RING (chan, 0);
551 OUT_RING (chan, 0);
552 OUT_RING (chan, stride);
553 OUT_RING (chan, height);
554 OUT_RING (chan, 1);
555 OUT_RING (chan, 0);
556 OUT_RING (chan, 0);
557 } else {
558 ret = RING_SPACE(chan, 2);
559 if (ret)
560 return ret;
561
562 BEGIN_RING(chan, NvSubM2MF, 0x0200, 1);
563 OUT_RING (chan, 1);
564 }
565 if (old_mem->mem_type == TTM_PL_VRAM &&
566 nouveau_bo_tile_layout(nvbo)) {
567 ret = RING_SPACE(chan, 8);
568 if (ret)
569 return ret;
570
571 BEGIN_RING(chan, NvSubM2MF, 0x021c, 7);
572 OUT_RING (chan, 0);
573 OUT_RING (chan, 0);
574 OUT_RING (chan, stride);
575 OUT_RING (chan, height);
576 OUT_RING (chan, 1);
577 OUT_RING (chan, 0);
578 OUT_RING (chan, 0);
579 } else {
580 ret = RING_SPACE(chan, 2);
581 if (ret)
582 return ret;
583
584 BEGIN_RING(chan, NvSubM2MF, 0x021c, 1);
585 OUT_RING (chan, 1);
586 }
587
588 ret = RING_SPACE(chan, 14);
589 if (ret)
590 return ret;
591
592 BEGIN_RING(chan, NvSubM2MF, 0x0238, 2);
593 OUT_RING (chan, upper_32_bits(src_offset));
594 OUT_RING (chan, upper_32_bits(dst_offset));
595 BEGIN_RING(chan, NvSubM2MF, 0x030c, 8);
596 OUT_RING (chan, lower_32_bits(src_offset));
597 OUT_RING (chan, lower_32_bits(dst_offset));
598 OUT_RING (chan, stride);
599 OUT_RING (chan, stride);
600 OUT_RING (chan, stride);
601 OUT_RING (chan, height);
602 OUT_RING (chan, 0x00000101);
603 OUT_RING (chan, 0x00000000);
604 BEGIN_RING(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_NOP, 1);
605 OUT_RING (chan, 0);
606
607 length -= amount;
608 src_offset += amount;
609 dst_offset += amount;
610 }
611
612 return 0;
613 }
614
615 static inline uint32_t
616 nouveau_bo_mem_ctxdma(struct ttm_buffer_object *bo,
617 struct nouveau_channel *chan, struct ttm_mem_reg *mem)
618 {
619 if (mem->mem_type == TTM_PL_TT)
620 return chan->gart_handle;
621 return chan->vram_handle;
622 }
623
624 static int
625 nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
626 struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
627 {
628 u32 src_offset = old_mem->start << PAGE_SHIFT;
629 u32 dst_offset = new_mem->start << PAGE_SHIFT;
630 u32 page_count = new_mem->num_pages;
631 int ret;
632
633 ret = RING_SPACE(chan, 3);
634 if (ret)
635 return ret;
636
637 BEGIN_RING(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_DMA_SOURCE, 2);
638 OUT_RING (chan, nouveau_bo_mem_ctxdma(bo, chan, old_mem));
639 OUT_RING (chan, nouveau_bo_mem_ctxdma(bo, chan, new_mem));
640
641 page_count = new_mem->num_pages;
642 while (page_count) {
643 int line_count = (page_count > 2047) ? 2047 : page_count;
644
645 ret = RING_SPACE(chan, 11);
646 if (ret)
647 return ret;
648
649 BEGIN_RING(chan, NvSubM2MF,
650 NV_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 8);
651 OUT_RING (chan, src_offset);
652 OUT_RING (chan, dst_offset);
653 OUT_RING (chan, PAGE_SIZE); /* src_pitch */
654 OUT_RING (chan, PAGE_SIZE); /* dst_pitch */
655 OUT_RING (chan, PAGE_SIZE); /* line_length */
656 OUT_RING (chan, line_count);
657 OUT_RING (chan, 0x00000101);
658 OUT_RING (chan, 0x00000000);
659 BEGIN_RING(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_NOP, 1);
660 OUT_RING (chan, 0);
661
662 page_count -= line_count;
663 src_offset += (PAGE_SIZE * line_count);
664 dst_offset += (PAGE_SIZE * line_count);
665 }
666
667 return 0;
668 }
669
670 static int
671 nouveau_vma_getmap(struct nouveau_channel *chan, struct nouveau_bo *nvbo,
672 struct ttm_mem_reg *mem, struct nouveau_vma *vma)
673 {
674 struct nouveau_mem *node = mem->mm_node;
675 int ret;
676
677 ret = nouveau_vm_get(chan->vm, mem->num_pages << PAGE_SHIFT,
678 node->page_shift, NV_MEM_ACCESS_RO, vma);
679 if (ret)
680 return ret;
681
682 if (mem->mem_type == TTM_PL_VRAM)
683 nouveau_vm_map(vma, node);
684 else
685 nouveau_vm_map_sg(vma, 0, mem->num_pages << PAGE_SHIFT, node);
686
687 return 0;
688 }
689
690 static int
691 nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
692 bool no_wait_reserve, bool no_wait_gpu,
693 struct ttm_mem_reg *new_mem)
694 {
695 struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
696 struct nouveau_bo *nvbo = nouveau_bo(bo);
697 struct ttm_mem_reg *old_mem = &bo->mem;
698 struct nouveau_channel *chan;
699 int ret;
700
701 chan = nvbo->channel;
702 if (!chan) {
703 chan = dev_priv->channel;
704 mutex_lock_nested(&chan->mutex, NOUVEAU_KCHANNEL_MUTEX);
705 }
706
707 /* create temporary vmas for the transfer and attach them to the
708 * old nouveau_mem node, these will get cleaned up after ttm has
709 * destroyed the ttm_mem_reg
710 */
711 if (dev_priv->card_type >= NV_50) {
712 struct nouveau_mem *node = old_mem->mm_node;
713
714 ret = nouveau_vma_getmap(chan, nvbo, old_mem, &node->vma[0]);
715 if (ret)
716 goto out;
717
718 ret = nouveau_vma_getmap(chan, nvbo, new_mem, &node->vma[1]);
719 if (ret)
720 goto out;
721 }
722
723 if (dev_priv->card_type < NV_50)
724 ret = nv04_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
725 else
726 if (dev_priv->card_type < NV_C0)
727 ret = nv50_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
728 else
729 ret = nvc0_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
730 if (ret == 0) {
731 ret = nouveau_bo_move_accel_cleanup(chan, nvbo, evict,
732 no_wait_reserve,
733 no_wait_gpu, new_mem);
734 }
735
736 out:
737 if (chan == dev_priv->channel)
738 mutex_unlock(&chan->mutex);
739 return ret;
740 }
741
742 static int
743 nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
744 bool no_wait_reserve, bool no_wait_gpu,
745 struct ttm_mem_reg *new_mem)
746 {
747 u32 placement_memtype = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING;
748 struct ttm_placement placement;
749 struct ttm_mem_reg tmp_mem;
750 int ret;
751
752 placement.fpfn = placement.lpfn = 0;
753 placement.num_placement = placement.num_busy_placement = 1;
754 placement.placement = placement.busy_placement = &placement_memtype;
755
756 tmp_mem = *new_mem;
757 tmp_mem.mm_node = NULL;
758 ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait_reserve, no_wait_gpu);
759 if (ret)
760 return ret;
761
762 ret = ttm_tt_bind(bo->ttm, &tmp_mem);
763 if (ret)
764 goto out;
765
766 ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_reserve, no_wait_gpu, &tmp_mem);
767 if (ret)
768 goto out;
769
770 ret = ttm_bo_move_ttm(bo, true, no_wait_reserve, no_wait_gpu, new_mem);
771 out:
772 ttm_bo_mem_put(bo, &tmp_mem);
773 return ret;
774 }
775
776 static int
777 nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
778 bool no_wait_reserve, bool no_wait_gpu,
779 struct ttm_mem_reg *new_mem)
780 {
781 u32 placement_memtype = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING;
782 struct ttm_placement placement;
783 struct ttm_mem_reg tmp_mem;
784 int ret;
785
786 placement.fpfn = placement.lpfn = 0;
787 placement.num_placement = placement.num_busy_placement = 1;
788 placement.placement = placement.busy_placement = &placement_memtype;
789
790 tmp_mem = *new_mem;
791 tmp_mem.mm_node = NULL;
792 ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait_reserve, no_wait_gpu);
793 if (ret)
794 return ret;
795
796 ret = ttm_bo_move_ttm(bo, true, no_wait_reserve, no_wait_gpu, &tmp_mem);
797 if (ret)
798 goto out;
799
800 ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_reserve, no_wait_gpu, new_mem);
801 if (ret)
802 goto out;
803
804 out:
805 ttm_bo_mem_put(bo, &tmp_mem);
806 return ret;
807 }
808
809 static void
810 nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
811 {
812 struct nouveau_bo *nvbo = nouveau_bo(bo);
813 struct nouveau_vma *vma;
814
815 /* ttm can now (stupidly) pass the driver bos it didn't create... */
816 if (bo->destroy != nouveau_bo_del_ttm)
817 return;
818
819 list_for_each_entry(vma, &nvbo->vma_list, head) {
820 if (new_mem && new_mem->mem_type == TTM_PL_VRAM) {
821 nouveau_vm_map(vma, new_mem->mm_node);
822 } else
823 if (new_mem && new_mem->mem_type == TTM_PL_TT &&
824 nvbo->page_shift == vma->vm->spg_shift) {
825 nouveau_vm_map_sg(vma, 0, new_mem->
826 num_pages << PAGE_SHIFT,
827 new_mem->mm_node);
828 } else {
829 nouveau_vm_unmap(vma);
830 }
831 }
832 }
833
834 static int
835 nouveau_bo_vm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem,
836 struct nouveau_tile_reg **new_tile)
837 {
838 struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
839 struct drm_device *dev = dev_priv->dev;
840 struct nouveau_bo *nvbo = nouveau_bo(bo);
841 u64 offset = new_mem->start << PAGE_SHIFT;
842
843 *new_tile = NULL;
844 if (new_mem->mem_type != TTM_PL_VRAM)
845 return 0;
846
847 if (dev_priv->card_type >= NV_10) {
848 *new_tile = nv10_mem_set_tiling(dev, offset, new_mem->size,
849 nvbo->tile_mode,
850 nvbo->tile_flags);
851 }
852
853 return 0;
854 }
855
856 static void
857 nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo,
858 struct nouveau_tile_reg *new_tile,
859 struct nouveau_tile_reg **old_tile)
860 {
861 struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
862 struct drm_device *dev = dev_priv->dev;
863
864 nv10_mem_put_tile_region(dev, *old_tile, bo->sync_obj);
865 *old_tile = new_tile;
866 }
867
868 static int
869 nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
870 bool no_wait_reserve, bool no_wait_gpu,
871 struct ttm_mem_reg *new_mem)
872 {
873 struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
874 struct nouveau_bo *nvbo = nouveau_bo(bo);
875 struct ttm_mem_reg *old_mem = &bo->mem;
876 struct nouveau_tile_reg *new_tile = NULL;
877 int ret = 0;
878
879 if (dev_priv->card_type < NV_50) {
880 ret = nouveau_bo_vm_bind(bo, new_mem, &new_tile);
881 if (ret)
882 return ret;
883 }
884
885 /* Fake bo copy. */
886 if (old_mem->mem_type == TTM_PL_SYSTEM && !bo->ttm) {
887 BUG_ON(bo->mem.mm_node != NULL);
888 bo->mem = *new_mem;
889 new_mem->mm_node = NULL;
890 goto out;
891 }
892
893 /* Software copy if the card isn't up and running yet. */
894 if (!dev_priv->channel) {
895 ret = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
896 goto out;
897 }
898
899 /* Hardware assisted copy. */
900 if (new_mem->mem_type == TTM_PL_SYSTEM)
901 ret = nouveau_bo_move_flipd(bo, evict, intr, no_wait_reserve, no_wait_gpu, new_mem);
902 else if (old_mem->mem_type == TTM_PL_SYSTEM)
903 ret = nouveau_bo_move_flips(bo, evict, intr, no_wait_reserve, no_wait_gpu, new_mem);
904 else
905 ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait_reserve, no_wait_gpu, new_mem);
906
907 if (!ret)
908 goto out;
909
910 /* Fallback to software copy. */
911 ret = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
912
913 out:
914 if (dev_priv->card_type < NV_50) {
915 if (ret)
916 nouveau_bo_vm_cleanup(bo, NULL, &new_tile);
917 else
918 nouveau_bo_vm_cleanup(bo, new_tile, &nvbo->tile);
919 }
920
921 return ret;
922 }
923
924 static int
925 nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp)
926 {
927 return 0;
928 }
929
930 static int
931 nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
932 {
933 struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
934 struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
935 struct drm_device *dev = dev_priv->dev;
936 int ret;
937
938 mem->bus.addr = NULL;
939 mem->bus.offset = 0;
940 mem->bus.size = mem->num_pages << PAGE_SHIFT;
941 mem->bus.base = 0;
942 mem->bus.is_iomem = false;
943 if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
944 return -EINVAL;
945 switch (mem->mem_type) {
946 case TTM_PL_SYSTEM:
947 /* System memory */
948 return 0;
949 case TTM_PL_TT:
950 #if __OS_HAS_AGP
951 if (dev_priv->gart_info.type == NOUVEAU_GART_AGP) {
952 mem->bus.offset = mem->start << PAGE_SHIFT;
953 mem->bus.base = dev_priv->gart_info.aper_base;
954 mem->bus.is_iomem = true;
955 }
956 #endif
957 break;
958 case TTM_PL_VRAM:
959 {
960 struct nouveau_mem *node = mem->mm_node;
961 u8 page_shift;
962
963 if (!dev_priv->bar1_vm) {
964 mem->bus.offset = mem->start << PAGE_SHIFT;
965 mem->bus.base = pci_resource_start(dev->pdev, 1);
966 mem->bus.is_iomem = true;
967 break;
968 }
969
970 if (dev_priv->card_type >= NV_C0)
971 page_shift = node->page_shift;
972 else
973 page_shift = 12;
974
975 ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size,
976 page_shift, NV_MEM_ACCESS_RW,
977 &node->bar_vma);
978 if (ret)
979 return ret;
980
981 nouveau_vm_map(&node->bar_vma, node);
982 if (ret) {
983 nouveau_vm_put(&node->bar_vma);
984 return ret;
985 }
986
987 mem->bus.offset = node->bar_vma.offset;
988 if (dev_priv->card_type == NV_50) /*XXX*/
989 mem->bus.offset -= 0x0020000000ULL;
990 mem->bus.base = pci_resource_start(dev->pdev, 1);
991 mem->bus.is_iomem = true;
992 }
993 break;
994 default:
995 return -EINVAL;
996 }
997 return 0;
998 }
999
1000 static void
1001 nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
1002 {
1003 struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
1004 struct nouveau_mem *node = mem->mm_node;
1005
1006 if (!dev_priv->bar1_vm || mem->mem_type != TTM_PL_VRAM)
1007 return;
1008
1009 if (!node->bar_vma.node)
1010 return;
1011
1012 nouveau_vm_unmap(&node->bar_vma);
1013 nouveau_vm_put(&node->bar_vma);
1014 }
1015
1016 static int
1017 nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
1018 {
1019 struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
1020 struct nouveau_bo *nvbo = nouveau_bo(bo);
1021
1022 /* as long as the bo isn't in vram, and isn't tiled, we've got
1023 * nothing to do here.
1024 */
1025 if (bo->mem.mem_type != TTM_PL_VRAM) {
1026 if (dev_priv->card_type < NV_50 ||
1027 !nouveau_bo_tile_layout(nvbo))
1028 return 0;
1029 }
1030
1031 /* make sure bo is in mappable vram */
1032 if (bo->mem.start + bo->mem.num_pages < dev_priv->fb_mappable_pages)
1033 return 0;
1034
1035
1036 nvbo->placement.fpfn = 0;
1037 nvbo->placement.lpfn = dev_priv->fb_mappable_pages;
1038 nouveau_bo_placement_set(nvbo, TTM_PL_VRAM, 0);
1039 return nouveau_bo_validate(nvbo, false, true, false);
1040 }
1041
1042 void
1043 nouveau_bo_fence(struct nouveau_bo *nvbo, struct nouveau_fence *fence)
1044 {
1045 struct nouveau_fence *old_fence;
1046
1047 if (likely(fence))
1048 nouveau_fence_ref(fence);
1049
1050 spin_lock(&nvbo->bo.bdev->fence_lock);
1051 old_fence = nvbo->bo.sync_obj;
1052 nvbo->bo.sync_obj = fence;
1053 spin_unlock(&nvbo->bo.bdev->fence_lock);
1054
1055 nouveau_fence_unref(&old_fence);
1056 }
1057
1058 static int
1059 nouveau_ttm_tt_populate(struct ttm_tt *ttm)
1060 {
1061 struct ttm_dma_tt *ttm_dma = (void *)ttm;
1062 struct drm_nouveau_private *dev_priv;
1063 struct drm_device *dev;
1064 unsigned i;
1065 int r;
1066
1067 if (ttm->state != tt_unpopulated)
1068 return 0;
1069
1070 dev_priv = nouveau_bdev(ttm->bdev);
1071 dev = dev_priv->dev;
1072
1073 #if __OS_HAS_AGP
1074 if (dev_priv->gart_info.type == NOUVEAU_GART_AGP) {
1075 return ttm_agp_tt_populate(ttm);
1076 }
1077 #endif
1078
1079 #ifdef CONFIG_SWIOTLB
1080 if (swiotlb_nr_tbl()) {
1081 return ttm_dma_populate((void *)ttm, dev->dev);
1082 }
1083 #endif
1084
1085 r = ttm_pool_populate(ttm);
1086 if (r) {
1087 return r;
1088 }
1089
1090 for (i = 0; i < ttm->num_pages; i++) {
1091 ttm_dma->dma_address[i] = pci_map_page(dev->pdev, ttm->pages[i],
1092 0, PAGE_SIZE,
1093 PCI_DMA_BIDIRECTIONAL);
1094 if (pci_dma_mapping_error(dev->pdev, ttm_dma->dma_address[i])) {
1095 while (--i) {
1096 pci_unmap_page(dev->pdev, ttm_dma->dma_address[i],
1097 PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
1098 ttm_dma->dma_address[i] = 0;
1099 }
1100 ttm_pool_unpopulate(ttm);
1101 return -EFAULT;
1102 }
1103 }
1104 return 0;
1105 }
1106
1107 static void
1108 nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
1109 {
1110 struct ttm_dma_tt *ttm_dma = (void *)ttm;
1111 struct drm_nouveau_private *dev_priv;
1112 struct drm_device *dev;
1113 unsigned i;
1114
1115 dev_priv = nouveau_bdev(ttm->bdev);
1116 dev = dev_priv->dev;
1117
1118 #if __OS_HAS_AGP
1119 if (dev_priv->gart_info.type == NOUVEAU_GART_AGP) {
1120 ttm_agp_tt_unpopulate(ttm);
1121 return;
1122 }
1123 #endif
1124
1125 #ifdef CONFIG_SWIOTLB
1126 if (swiotlb_nr_tbl()) {
1127 ttm_dma_unpopulate((void *)ttm, dev->dev);
1128 return;
1129 }
1130 #endif
1131
1132 for (i = 0; i < ttm->num_pages; i++) {
1133 if (ttm_dma->dma_address[i]) {
1134 pci_unmap_page(dev->pdev, ttm_dma->dma_address[i],
1135 PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
1136 }
1137 }
1138
1139 ttm_pool_unpopulate(ttm);
1140 }
1141
1142 struct ttm_bo_driver nouveau_bo_driver = {
1143 .ttm_tt_create = &nouveau_ttm_tt_create,
1144 .ttm_tt_populate = &nouveau_ttm_tt_populate,
1145 .ttm_tt_unpopulate = &nouveau_ttm_tt_unpopulate,
1146 .invalidate_caches = nouveau_bo_invalidate_caches,
1147 .init_mem_type = nouveau_bo_init_mem_type,
1148 .evict_flags = nouveau_bo_evict_flags,
1149 .move_notify = nouveau_bo_move_ntfy,
1150 .move = nouveau_bo_move,
1151 .verify_access = nouveau_bo_verify_access,
1152 .sync_obj_signaled = __nouveau_fence_signalled,
1153 .sync_obj_wait = __nouveau_fence_wait,
1154 .sync_obj_flush = __nouveau_fence_flush,
1155 .sync_obj_unref = __nouveau_fence_unref,
1156 .sync_obj_ref = __nouveau_fence_ref,
1157 .fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
1158 .io_mem_reserve = &nouveau_ttm_io_mem_reserve,
1159 .io_mem_free = &nouveau_ttm_io_mem_free,
1160 };
1161
1162 struct nouveau_vma *
1163 nouveau_bo_vma_find(struct nouveau_bo *nvbo, struct nouveau_vm *vm)
1164 {
1165 struct nouveau_vma *vma;
1166 list_for_each_entry(vma, &nvbo->vma_list, head) {
1167 if (vma->vm == vm)
1168 return vma;
1169 }
1170
1171 return NULL;
1172 }
1173
1174 int
1175 nouveau_bo_vma_add(struct nouveau_bo *nvbo, struct nouveau_vm *vm,
1176 struct nouveau_vma *vma)
1177 {
1178 const u32 size = nvbo->bo.mem.num_pages << PAGE_SHIFT;
1179 struct nouveau_mem *node = nvbo->bo.mem.mm_node;
1180 int ret;
1181
1182 ret = nouveau_vm_get(vm, size, nvbo->page_shift,
1183 NV_MEM_ACCESS_RW, vma);
1184 if (ret)
1185 return ret;
1186
1187 if (nvbo->bo.mem.mem_type == TTM_PL_VRAM)
1188 nouveau_vm_map(vma, nvbo->bo.mem.mm_node);
1189 else
1190 if (nvbo->bo.mem.mem_type == TTM_PL_TT)
1191 nouveau_vm_map_sg(vma, 0, size, node);
1192
1193 list_add_tail(&vma->head, &nvbo->vma_list);
1194 vma->refcount = 1;
1195 return 0;
1196 }
1197
1198 void
1199 nouveau_bo_vma_del(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
1200 {
1201 if (vma->node) {
1202 if (nvbo->bo.mem.mem_type != TTM_PL_SYSTEM) {
1203 spin_lock(&nvbo->bo.bdev->fence_lock);
1204 ttm_bo_wait(&nvbo->bo, false, false, false);
1205 spin_unlock(&nvbo->bo.bdev->fence_lock);
1206 nouveau_vm_unmap(vma);
1207 }
1208
1209 nouveau_vm_put(vma);
1210 list_del(&vma->head);
1211 }
1212 }