2 * Copyright 2007 Dave Airlied
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:
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
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.
25 * Authors: Dave Airlied <airlied@linux.ie>
26 * Ben Skeggs <darktama@iinet.net.au>
27 * Jeremy Kolb <jkolb@brandeis.edu>
30 #include <core/engine.h>
31 #include <linux/swiotlb.h>
33 #include <subdev/fb.h>
34 #include <subdev/vm.h>
35 #include <subdev/bar.h>
37 #include "nouveau_drm.h"
38 #include "nouveau_dma.h"
39 #include "nouveau_fence.h"
41 #include "nouveau_bo.h"
42 #include "nouveau_ttm.h"
43 #include "nouveau_gem.h"
46 * NV10-NV40 tiling helpers
50 nv10_bo_update_tile_region(struct drm_device
*dev
, struct nouveau_drm_tile
*reg
,
51 u32 addr
, u32 size
, u32 pitch
, u32 flags
)
53 struct nouveau_drm
*drm
= nouveau_drm(dev
);
54 int i
= reg
- drm
->tile
.reg
;
55 struct nouveau_fb
*pfb
= nouveau_fb(drm
->device
);
56 struct nouveau_fb_tile
*tile
= &pfb
->tile
.region
[i
];
57 struct nouveau_engine
*engine
;
59 nouveau_fence_unref(®
->fence
);
62 pfb
->tile
.fini(pfb
, i
, tile
);
65 pfb
->tile
.init(pfb
, i
, addr
, size
, pitch
, flags
, tile
);
67 pfb
->tile
.prog(pfb
, i
, tile
);
69 if ((engine
= nouveau_engine(pfb
, NVDEV_ENGINE_GR
)))
70 engine
->tile_prog(engine
, i
);
71 if ((engine
= nouveau_engine(pfb
, NVDEV_ENGINE_MPEG
)))
72 engine
->tile_prog(engine
, i
);
75 static struct nouveau_drm_tile
*
76 nv10_bo_get_tile_region(struct drm_device
*dev
, int i
)
78 struct nouveau_drm
*drm
= nouveau_drm(dev
);
79 struct nouveau_drm_tile
*tile
= &drm
->tile
.reg
[i
];
81 spin_lock(&drm
->tile
.lock
);
84 (!tile
->fence
|| nouveau_fence_done(tile
->fence
)))
89 spin_unlock(&drm
->tile
.lock
);
94 nv10_bo_put_tile_region(struct drm_device
*dev
, struct nouveau_drm_tile
*tile
,
95 struct nouveau_fence
*fence
)
97 struct nouveau_drm
*drm
= nouveau_drm(dev
);
100 spin_lock(&drm
->tile
.lock
);
102 /* Mark it as pending. */
104 nouveau_fence_ref(fence
);
108 spin_unlock(&drm
->tile
.lock
);
112 static struct nouveau_drm_tile
*
113 nv10_bo_set_tiling(struct drm_device
*dev
, u32 addr
,
114 u32 size
, u32 pitch
, u32 flags
)
116 struct nouveau_drm
*drm
= nouveau_drm(dev
);
117 struct nouveau_fb
*pfb
= nouveau_fb(drm
->device
);
118 struct nouveau_drm_tile
*tile
, *found
= NULL
;
121 for (i
= 0; i
< pfb
->tile
.regions
; i
++) {
122 tile
= nv10_bo_get_tile_region(dev
, i
);
124 if (pitch
&& !found
) {
128 } else if (tile
&& pfb
->tile
.region
[i
].pitch
) {
129 /* Kill an unused tile region. */
130 nv10_bo_update_tile_region(dev
, tile
, 0, 0, 0, 0);
133 nv10_bo_put_tile_region(dev
, tile
, NULL
);
137 nv10_bo_update_tile_region(dev
, found
, addr
, size
,
143 nouveau_bo_del_ttm(struct ttm_buffer_object
*bo
)
145 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
146 struct drm_device
*dev
= drm
->dev
;
147 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
149 if (unlikely(nvbo
->gem
))
150 DRM_ERROR("bo %p still attached to GEM object\n", bo
);
151 WARN_ON(nvbo
->pin_refcnt
> 0);
152 nv10_bo_put_tile_region(dev
, nvbo
->tile
, NULL
);
157 nouveau_bo_fixup_align(struct nouveau_bo
*nvbo
, u32 flags
,
158 int *align
, int *size
)
160 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
161 struct nouveau_device
*device
= nv_device(drm
->device
);
163 if (device
->card_type
< NV_50
) {
164 if (nvbo
->tile_mode
) {
165 if (device
->chipset
>= 0x40) {
167 *size
= roundup(*size
, 64 * nvbo
->tile_mode
);
169 } else if (device
->chipset
>= 0x30) {
171 *size
= roundup(*size
, 64 * nvbo
->tile_mode
);
173 } else if (device
->chipset
>= 0x20) {
175 *size
= roundup(*size
, 64 * nvbo
->tile_mode
);
177 } else if (device
->chipset
>= 0x10) {
179 *size
= roundup(*size
, 32 * nvbo
->tile_mode
);
183 *size
= roundup(*size
, (1 << nvbo
->page_shift
));
184 *align
= max((1 << nvbo
->page_shift
), *align
);
187 *size
= roundup(*size
, PAGE_SIZE
);
191 nouveau_bo_new(struct drm_device
*dev
, int size
, int align
,
192 uint32_t flags
, uint32_t tile_mode
, uint32_t tile_flags
,
194 struct nouveau_bo
**pnvbo
)
196 struct nouveau_drm
*drm
= nouveau_drm(dev
);
197 struct nouveau_bo
*nvbo
;
200 int type
= ttm_bo_type_device
;
203 type
= ttm_bo_type_sg
;
205 nvbo
= kzalloc(sizeof(struct nouveau_bo
), GFP_KERNEL
);
208 INIT_LIST_HEAD(&nvbo
->head
);
209 INIT_LIST_HEAD(&nvbo
->entry
);
210 INIT_LIST_HEAD(&nvbo
->vma_list
);
211 nvbo
->tile_mode
= tile_mode
;
212 nvbo
->tile_flags
= tile_flags
;
213 nvbo
->bo
.bdev
= &drm
->ttm
.bdev
;
215 nvbo
->page_shift
= 12;
216 if (drm
->client
.base
.vm
) {
217 if (!(flags
& TTM_PL_FLAG_TT
) && size
> 256 * 1024)
218 nvbo
->page_shift
= drm
->client
.base
.vm
->vmm
->lpg_shift
;
221 nouveau_bo_fixup_align(nvbo
, flags
, &align
, &size
);
222 nvbo
->bo
.mem
.num_pages
= size
>> PAGE_SHIFT
;
223 nouveau_bo_placement_set(nvbo
, flags
, 0);
225 acc_size
= ttm_bo_dma_acc_size(&drm
->ttm
.bdev
, size
,
226 sizeof(struct nouveau_bo
));
228 ret
= ttm_bo_init(&drm
->ttm
.bdev
, &nvbo
->bo
, size
,
229 type
, &nvbo
->placement
,
230 align
>> PAGE_SHIFT
, false, NULL
, acc_size
, sg
,
233 /* ttm will call nouveau_bo_del_ttm if it fails.. */
242 set_placement_list(uint32_t *pl
, unsigned *n
, uint32_t type
, uint32_t flags
)
246 if (type
& TTM_PL_FLAG_VRAM
)
247 pl
[(*n
)++] = TTM_PL_FLAG_VRAM
| flags
;
248 if (type
& TTM_PL_FLAG_TT
)
249 pl
[(*n
)++] = TTM_PL_FLAG_TT
| flags
;
250 if (type
& TTM_PL_FLAG_SYSTEM
)
251 pl
[(*n
)++] = TTM_PL_FLAG_SYSTEM
| flags
;
255 set_placement_range(struct nouveau_bo
*nvbo
, uint32_t type
)
257 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
258 struct nouveau_fb
*pfb
= nouveau_fb(drm
->device
);
259 u32 vram_pages
= pfb
->ram
->size
>> PAGE_SHIFT
;
261 if (nv_device(drm
->device
)->card_type
== NV_10
&&
262 nvbo
->tile_mode
&& (type
& TTM_PL_FLAG_VRAM
) &&
263 nvbo
->bo
.mem
.num_pages
< vram_pages
/ 4) {
265 * Make sure that the color and depth buffers are handled
266 * by independent memory controller units. Up to a 9x
267 * speed up when alpha-blending and depth-test are enabled
270 if (nvbo
->tile_flags
& NOUVEAU_GEM_TILE_ZETA
) {
271 nvbo
->placement
.fpfn
= vram_pages
/ 2;
272 nvbo
->placement
.lpfn
= ~0;
274 nvbo
->placement
.fpfn
= 0;
275 nvbo
->placement
.lpfn
= vram_pages
/ 2;
281 nouveau_bo_placement_set(struct nouveau_bo
*nvbo
, uint32_t type
, uint32_t busy
)
283 struct ttm_placement
*pl
= &nvbo
->placement
;
284 uint32_t flags
= TTM_PL_MASK_CACHING
|
285 (nvbo
->pin_refcnt
? TTM_PL_FLAG_NO_EVICT
: 0);
287 pl
->placement
= nvbo
->placements
;
288 set_placement_list(nvbo
->placements
, &pl
->num_placement
,
291 pl
->busy_placement
= nvbo
->busy_placements
;
292 set_placement_list(nvbo
->busy_placements
, &pl
->num_busy_placement
,
295 set_placement_range(nvbo
, type
);
299 nouveau_bo_pin(struct nouveau_bo
*nvbo
, uint32_t memtype
)
301 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
302 struct ttm_buffer_object
*bo
= &nvbo
->bo
;
305 ret
= ttm_bo_reserve(bo
, false, false, false, 0);
309 if (nvbo
->pin_refcnt
&& !(memtype
& (1 << bo
->mem
.mem_type
))) {
310 NV_ERROR(drm
, "bo %p pinned elsewhere: 0x%08x vs 0x%08x\n", bo
,
311 1 << bo
->mem
.mem_type
, memtype
);
316 if (nvbo
->pin_refcnt
++)
319 nouveau_bo_placement_set(nvbo
, memtype
, 0);
321 ret
= nouveau_bo_validate(nvbo
, false, false);
323 switch (bo
->mem
.mem_type
) {
325 drm
->gem
.vram_available
-= bo
->mem
.size
;
328 drm
->gem
.gart_available
-= bo
->mem
.size
;
335 ttm_bo_unreserve(bo
);
340 nouveau_bo_unpin(struct nouveau_bo
*nvbo
)
342 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
343 struct ttm_buffer_object
*bo
= &nvbo
->bo
;
346 ret
= ttm_bo_reserve(bo
, false, false, false, 0);
350 ref
= --nvbo
->pin_refcnt
;
351 WARN_ON_ONCE(ref
< 0);
355 nouveau_bo_placement_set(nvbo
, bo
->mem
.placement
, 0);
357 ret
= nouveau_bo_validate(nvbo
, false, false);
359 switch (bo
->mem
.mem_type
) {
361 drm
->gem
.vram_available
+= bo
->mem
.size
;
364 drm
->gem
.gart_available
+= bo
->mem
.size
;
372 ttm_bo_unreserve(bo
);
377 nouveau_bo_map(struct nouveau_bo
*nvbo
)
381 ret
= ttm_bo_reserve(&nvbo
->bo
, false, false, false, 0);
385 ret
= ttm_bo_kmap(&nvbo
->bo
, 0, nvbo
->bo
.mem
.num_pages
, &nvbo
->kmap
);
386 ttm_bo_unreserve(&nvbo
->bo
);
391 nouveau_bo_unmap(struct nouveau_bo
*nvbo
)
394 ttm_bo_kunmap(&nvbo
->kmap
);
398 nouveau_bo_validate(struct nouveau_bo
*nvbo
, bool interruptible
,
403 ret
= ttm_bo_validate(&nvbo
->bo
, &nvbo
->placement
,
404 interruptible
, no_wait_gpu
);
412 nouveau_bo_rd16(struct nouveau_bo
*nvbo
, unsigned index
)
415 u16
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
418 return ioread16_native((void __force __iomem
*)mem
);
424 nouveau_bo_wr16(struct nouveau_bo
*nvbo
, unsigned index
, u16 val
)
427 u16
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
430 iowrite16_native(val
, (void __force __iomem
*)mem
);
436 nouveau_bo_rd32(struct nouveau_bo
*nvbo
, unsigned index
)
439 u32
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
442 return ioread32_native((void __force __iomem
*)mem
);
448 nouveau_bo_wr32(struct nouveau_bo
*nvbo
, unsigned index
, u32 val
)
451 u32
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
454 iowrite32_native(val
, (void __force __iomem
*)mem
);
459 static struct ttm_tt
*
460 nouveau_ttm_tt_create(struct ttm_bo_device
*bdev
, unsigned long size
,
461 uint32_t page_flags
, struct page
*dummy_read
)
464 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
465 struct drm_device
*dev
= drm
->dev
;
467 if (drm
->agp
.stat
== ENABLED
) {
468 return ttm_agp_tt_create(bdev
, dev
->agp
->bridge
, size
,
469 page_flags
, dummy_read
);
473 return nouveau_sgdma_create_ttm(bdev
, size
, page_flags
, dummy_read
);
477 nouveau_bo_invalidate_caches(struct ttm_bo_device
*bdev
, uint32_t flags
)
479 /* We'll do this from user space. */
484 nouveau_bo_init_mem_type(struct ttm_bo_device
*bdev
, uint32_t type
,
485 struct ttm_mem_type_manager
*man
)
487 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
491 man
->flags
= TTM_MEMTYPE_FLAG_MAPPABLE
;
492 man
->available_caching
= TTM_PL_MASK_CACHING
;
493 man
->default_caching
= TTM_PL_FLAG_CACHED
;
496 if (nv_device(drm
->device
)->card_type
>= NV_50
) {
497 man
->func
= &nouveau_vram_manager
;
498 man
->io_reserve_fastpath
= false;
499 man
->use_io_reserve_lru
= true;
501 man
->func
= &ttm_bo_manager_func
;
503 man
->flags
= TTM_MEMTYPE_FLAG_FIXED
|
504 TTM_MEMTYPE_FLAG_MAPPABLE
;
505 man
->available_caching
= TTM_PL_FLAG_UNCACHED
|
507 man
->default_caching
= TTM_PL_FLAG_WC
;
510 if (nv_device(drm
->device
)->card_type
>= NV_50
)
511 man
->func
= &nouveau_gart_manager
;
513 if (drm
->agp
.stat
!= ENABLED
)
514 man
->func
= &nv04_gart_manager
;
516 man
->func
= &ttm_bo_manager_func
;
518 if (drm
->agp
.stat
== ENABLED
) {
519 man
->flags
= TTM_MEMTYPE_FLAG_MAPPABLE
;
520 man
->available_caching
= TTM_PL_FLAG_UNCACHED
|
522 man
->default_caching
= TTM_PL_FLAG_WC
;
524 man
->flags
= TTM_MEMTYPE_FLAG_MAPPABLE
|
525 TTM_MEMTYPE_FLAG_CMA
;
526 man
->available_caching
= TTM_PL_MASK_CACHING
;
527 man
->default_caching
= TTM_PL_FLAG_CACHED
;
538 nouveau_bo_evict_flags(struct ttm_buffer_object
*bo
, struct ttm_placement
*pl
)
540 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
542 switch (bo
->mem
.mem_type
) {
544 nouveau_bo_placement_set(nvbo
, TTM_PL_FLAG_TT
,
548 nouveau_bo_placement_set(nvbo
, TTM_PL_FLAG_SYSTEM
, 0);
552 *pl
= nvbo
->placement
;
556 /* GPU-assisted copy using NV_MEMORY_TO_MEMORY_FORMAT, can access
557 * TTM_PL_{VRAM,TT} directly.
561 nouveau_bo_move_accel_cleanup(struct nouveau_channel
*chan
,
562 struct nouveau_bo
*nvbo
, bool evict
,
563 bool no_wait_gpu
, struct ttm_mem_reg
*new_mem
)
565 struct nouveau_fence
*fence
= NULL
;
568 ret
= nouveau_fence_new(chan
, false, &fence
);
572 ret
= ttm_bo_move_accel_cleanup(&nvbo
->bo
, fence
, evict
,
573 no_wait_gpu
, new_mem
);
574 nouveau_fence_unref(&fence
);
579 nve0_bo_move_init(struct nouveau_channel
*chan
, u32 handle
)
581 int ret
= RING_SPACE(chan
, 2);
583 BEGIN_NVC0(chan
, NvSubCopy
, 0x0000, 1);
584 OUT_RING (chan
, handle
& 0x0000ffff);
591 nve0_bo_move_copy(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
592 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
594 struct nouveau_mem
*node
= old_mem
->mm_node
;
595 int ret
= RING_SPACE(chan
, 10);
597 BEGIN_NVC0(chan
, NvSubCopy
, 0x0400, 8);
598 OUT_RING (chan
, upper_32_bits(node
->vma
[0].offset
));
599 OUT_RING (chan
, lower_32_bits(node
->vma
[0].offset
));
600 OUT_RING (chan
, upper_32_bits(node
->vma
[1].offset
));
601 OUT_RING (chan
, lower_32_bits(node
->vma
[1].offset
));
602 OUT_RING (chan
, PAGE_SIZE
);
603 OUT_RING (chan
, PAGE_SIZE
);
604 OUT_RING (chan
, PAGE_SIZE
);
605 OUT_RING (chan
, new_mem
->num_pages
);
606 BEGIN_IMC0(chan
, NvSubCopy
, 0x0300, 0x0386);
612 nvc0_bo_move_init(struct nouveau_channel
*chan
, u32 handle
)
614 int ret
= RING_SPACE(chan
, 2);
616 BEGIN_NVC0(chan
, NvSubCopy
, 0x0000, 1);
617 OUT_RING (chan
, handle
);
623 nvc0_bo_move_copy(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
624 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
626 struct nouveau_mem
*node
= old_mem
->mm_node
;
627 u64 src_offset
= node
->vma
[0].offset
;
628 u64 dst_offset
= node
->vma
[1].offset
;
629 u32 page_count
= new_mem
->num_pages
;
632 page_count
= new_mem
->num_pages
;
634 int line_count
= (page_count
> 8191) ? 8191 : page_count
;
636 ret
= RING_SPACE(chan
, 11);
640 BEGIN_NVC0(chan
, NvSubCopy
, 0x030c, 8);
641 OUT_RING (chan
, upper_32_bits(src_offset
));
642 OUT_RING (chan
, lower_32_bits(src_offset
));
643 OUT_RING (chan
, upper_32_bits(dst_offset
));
644 OUT_RING (chan
, lower_32_bits(dst_offset
));
645 OUT_RING (chan
, PAGE_SIZE
);
646 OUT_RING (chan
, PAGE_SIZE
);
647 OUT_RING (chan
, PAGE_SIZE
);
648 OUT_RING (chan
, line_count
);
649 BEGIN_NVC0(chan
, NvSubCopy
, 0x0300, 1);
650 OUT_RING (chan
, 0x00000110);
652 page_count
-= line_count
;
653 src_offset
+= (PAGE_SIZE
* line_count
);
654 dst_offset
+= (PAGE_SIZE
* line_count
);
661 nvc0_bo_move_m2mf(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
662 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
664 struct nouveau_mem
*node
= old_mem
->mm_node
;
665 u64 src_offset
= node
->vma
[0].offset
;
666 u64 dst_offset
= node
->vma
[1].offset
;
667 u32 page_count
= new_mem
->num_pages
;
670 page_count
= new_mem
->num_pages
;
672 int line_count
= (page_count
> 2047) ? 2047 : page_count
;
674 ret
= RING_SPACE(chan
, 12);
678 BEGIN_NVC0(chan
, NvSubCopy
, 0x0238, 2);
679 OUT_RING (chan
, upper_32_bits(dst_offset
));
680 OUT_RING (chan
, lower_32_bits(dst_offset
));
681 BEGIN_NVC0(chan
, NvSubCopy
, 0x030c, 6);
682 OUT_RING (chan
, upper_32_bits(src_offset
));
683 OUT_RING (chan
, lower_32_bits(src_offset
));
684 OUT_RING (chan
, PAGE_SIZE
); /* src_pitch */
685 OUT_RING (chan
, PAGE_SIZE
); /* dst_pitch */
686 OUT_RING (chan
, PAGE_SIZE
); /* line_length */
687 OUT_RING (chan
, line_count
);
688 BEGIN_NVC0(chan
, NvSubCopy
, 0x0300, 1);
689 OUT_RING (chan
, 0x00100110);
691 page_count
-= line_count
;
692 src_offset
+= (PAGE_SIZE
* line_count
);
693 dst_offset
+= (PAGE_SIZE
* line_count
);
700 nva3_bo_move_copy(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
701 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
703 struct nouveau_mem
*node
= old_mem
->mm_node
;
704 u64 src_offset
= node
->vma
[0].offset
;
705 u64 dst_offset
= node
->vma
[1].offset
;
706 u32 page_count
= new_mem
->num_pages
;
709 page_count
= new_mem
->num_pages
;
711 int line_count
= (page_count
> 8191) ? 8191 : page_count
;
713 ret
= RING_SPACE(chan
, 11);
717 BEGIN_NV04(chan
, NvSubCopy
, 0x030c, 8);
718 OUT_RING (chan
, upper_32_bits(src_offset
));
719 OUT_RING (chan
, lower_32_bits(src_offset
));
720 OUT_RING (chan
, upper_32_bits(dst_offset
));
721 OUT_RING (chan
, lower_32_bits(dst_offset
));
722 OUT_RING (chan
, PAGE_SIZE
);
723 OUT_RING (chan
, PAGE_SIZE
);
724 OUT_RING (chan
, PAGE_SIZE
);
725 OUT_RING (chan
, line_count
);
726 BEGIN_NV04(chan
, NvSubCopy
, 0x0300, 1);
727 OUT_RING (chan
, 0x00000110);
729 page_count
-= line_count
;
730 src_offset
+= (PAGE_SIZE
* line_count
);
731 dst_offset
+= (PAGE_SIZE
* line_count
);
738 nv98_bo_move_exec(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
739 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
741 struct nouveau_mem
*node
= old_mem
->mm_node
;
742 int ret
= RING_SPACE(chan
, 7);
744 BEGIN_NV04(chan
, NvSubCopy
, 0x0320, 6);
745 OUT_RING (chan
, upper_32_bits(node
->vma
[0].offset
));
746 OUT_RING (chan
, lower_32_bits(node
->vma
[0].offset
));
747 OUT_RING (chan
, upper_32_bits(node
->vma
[1].offset
));
748 OUT_RING (chan
, lower_32_bits(node
->vma
[1].offset
));
749 OUT_RING (chan
, 0x00000000 /* COPY */);
750 OUT_RING (chan
, new_mem
->num_pages
<< PAGE_SHIFT
);
756 nv84_bo_move_exec(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
757 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
759 struct nouveau_mem
*node
= old_mem
->mm_node
;
760 int ret
= RING_SPACE(chan
, 7);
762 BEGIN_NV04(chan
, NvSubCopy
, 0x0304, 6);
763 OUT_RING (chan
, new_mem
->num_pages
<< PAGE_SHIFT
);
764 OUT_RING (chan
, upper_32_bits(node
->vma
[0].offset
));
765 OUT_RING (chan
, lower_32_bits(node
->vma
[0].offset
));
766 OUT_RING (chan
, upper_32_bits(node
->vma
[1].offset
));
767 OUT_RING (chan
, lower_32_bits(node
->vma
[1].offset
));
768 OUT_RING (chan
, 0x00000000 /* MODE_COPY, QUERY_NONE */);
774 nv50_bo_move_init(struct nouveau_channel
*chan
, u32 handle
)
776 int ret
= RING_SPACE(chan
, 6);
778 BEGIN_NV04(chan
, NvSubCopy
, 0x0000, 1);
779 OUT_RING (chan
, handle
);
780 BEGIN_NV04(chan
, NvSubCopy
, 0x0180, 3);
781 OUT_RING (chan
, NvNotify0
);
782 OUT_RING (chan
, NvDmaFB
);
783 OUT_RING (chan
, NvDmaFB
);
790 nv50_bo_move_m2mf(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
791 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
793 struct nouveau_mem
*node
= old_mem
->mm_node
;
794 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
795 u64 length
= (new_mem
->num_pages
<< PAGE_SHIFT
);
796 u64 src_offset
= node
->vma
[0].offset
;
797 u64 dst_offset
= node
->vma
[1].offset
;
801 u32 amount
, stride
, height
;
803 amount
= min(length
, (u64
)(4 * 1024 * 1024));
805 height
= amount
/ stride
;
807 if (old_mem
->mem_type
== TTM_PL_VRAM
&&
808 nouveau_bo_tile_layout(nvbo
)) {
809 ret
= RING_SPACE(chan
, 8);
813 BEGIN_NV04(chan
, NvSubCopy
, 0x0200, 7);
816 OUT_RING (chan
, stride
);
817 OUT_RING (chan
, height
);
822 ret
= RING_SPACE(chan
, 2);
826 BEGIN_NV04(chan
, NvSubCopy
, 0x0200, 1);
829 if (new_mem
->mem_type
== TTM_PL_VRAM
&&
830 nouveau_bo_tile_layout(nvbo
)) {
831 ret
= RING_SPACE(chan
, 8);
835 BEGIN_NV04(chan
, NvSubCopy
, 0x021c, 7);
838 OUT_RING (chan
, stride
);
839 OUT_RING (chan
, height
);
844 ret
= RING_SPACE(chan
, 2);
848 BEGIN_NV04(chan
, NvSubCopy
, 0x021c, 1);
852 ret
= RING_SPACE(chan
, 14);
856 BEGIN_NV04(chan
, NvSubCopy
, 0x0238, 2);
857 OUT_RING (chan
, upper_32_bits(src_offset
));
858 OUT_RING (chan
, upper_32_bits(dst_offset
));
859 BEGIN_NV04(chan
, NvSubCopy
, 0x030c, 8);
860 OUT_RING (chan
, lower_32_bits(src_offset
));
861 OUT_RING (chan
, lower_32_bits(dst_offset
));
862 OUT_RING (chan
, stride
);
863 OUT_RING (chan
, stride
);
864 OUT_RING (chan
, stride
);
865 OUT_RING (chan
, height
);
866 OUT_RING (chan
, 0x00000101);
867 OUT_RING (chan
, 0x00000000);
868 BEGIN_NV04(chan
, NvSubCopy
, NV_MEMORY_TO_MEMORY_FORMAT_NOP
, 1);
872 src_offset
+= amount
;
873 dst_offset
+= amount
;
880 nv04_bo_move_init(struct nouveau_channel
*chan
, u32 handle
)
882 int ret
= RING_SPACE(chan
, 4);
884 BEGIN_NV04(chan
, NvSubCopy
, 0x0000, 1);
885 OUT_RING (chan
, handle
);
886 BEGIN_NV04(chan
, NvSubCopy
, 0x0180, 1);
887 OUT_RING (chan
, NvNotify0
);
893 static inline uint32_t
894 nouveau_bo_mem_ctxdma(struct ttm_buffer_object
*bo
,
895 struct nouveau_channel
*chan
, struct ttm_mem_reg
*mem
)
897 if (mem
->mem_type
== TTM_PL_TT
)
903 nv04_bo_move_m2mf(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
904 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
906 u32 src_offset
= old_mem
->start
<< PAGE_SHIFT
;
907 u32 dst_offset
= new_mem
->start
<< PAGE_SHIFT
;
908 u32 page_count
= new_mem
->num_pages
;
911 ret
= RING_SPACE(chan
, 3);
915 BEGIN_NV04(chan
, NvSubCopy
, NV_MEMORY_TO_MEMORY_FORMAT_DMA_SOURCE
, 2);
916 OUT_RING (chan
, nouveau_bo_mem_ctxdma(bo
, chan
, old_mem
));
917 OUT_RING (chan
, nouveau_bo_mem_ctxdma(bo
, chan
, new_mem
));
919 page_count
= new_mem
->num_pages
;
921 int line_count
= (page_count
> 2047) ? 2047 : page_count
;
923 ret
= RING_SPACE(chan
, 11);
927 BEGIN_NV04(chan
, NvSubCopy
,
928 NV_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN
, 8);
929 OUT_RING (chan
, src_offset
);
930 OUT_RING (chan
, dst_offset
);
931 OUT_RING (chan
, PAGE_SIZE
); /* src_pitch */
932 OUT_RING (chan
, PAGE_SIZE
); /* dst_pitch */
933 OUT_RING (chan
, PAGE_SIZE
); /* line_length */
934 OUT_RING (chan
, line_count
);
935 OUT_RING (chan
, 0x00000101);
936 OUT_RING (chan
, 0x00000000);
937 BEGIN_NV04(chan
, NvSubCopy
, NV_MEMORY_TO_MEMORY_FORMAT_NOP
, 1);
940 page_count
-= line_count
;
941 src_offset
+= (PAGE_SIZE
* line_count
);
942 dst_offset
+= (PAGE_SIZE
* line_count
);
949 nouveau_vma_getmap(struct nouveau_channel
*chan
, struct nouveau_bo
*nvbo
,
950 struct ttm_mem_reg
*mem
, struct nouveau_vma
*vma
)
952 struct nouveau_mem
*node
= mem
->mm_node
;
955 ret
= nouveau_vm_get(nv_client(chan
->cli
)->vm
, mem
->num_pages
<<
956 PAGE_SHIFT
, node
->page_shift
,
957 NV_MEM_ACCESS_RW
, vma
);
961 if (mem
->mem_type
== TTM_PL_VRAM
)
962 nouveau_vm_map(vma
, node
);
964 nouveau_vm_map_sg(vma
, 0, mem
->num_pages
<< PAGE_SHIFT
, node
);
970 nouveau_bo_move_m2mf(struct ttm_buffer_object
*bo
, int evict
, bool intr
,
971 bool no_wait_gpu
, struct ttm_mem_reg
*new_mem
)
973 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
974 struct nouveau_channel
*chan
= chan
= drm
->ttm
.chan
;
975 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
976 struct ttm_mem_reg
*old_mem
= &bo
->mem
;
979 mutex_lock_nested(&chan
->cli
->mutex
, SINGLE_DEPTH_NESTING
);
981 /* create temporary vmas for the transfer and attach them to the
982 * old nouveau_mem node, these will get cleaned up after ttm has
983 * destroyed the ttm_mem_reg
985 if (nv_device(drm
->device
)->card_type
>= NV_50
) {
986 struct nouveau_mem
*node
= old_mem
->mm_node
;
988 ret
= nouveau_vma_getmap(chan
, nvbo
, old_mem
, &node
->vma
[0]);
992 ret
= nouveau_vma_getmap(chan
, nvbo
, new_mem
, &node
->vma
[1]);
997 ret
= drm
->ttm
.move(chan
, bo
, &bo
->mem
, new_mem
);
999 ret
= nouveau_bo_move_accel_cleanup(chan
, nvbo
, evict
,
1000 no_wait_gpu
, new_mem
);
1004 mutex_unlock(&chan
->cli
->mutex
);
1009 nouveau_bo_move_init(struct nouveau_drm
*drm
)
1011 static const struct {
1015 int (*exec
)(struct nouveau_channel
*,
1016 struct ttm_buffer_object
*,
1017 struct ttm_mem_reg
*, struct ttm_mem_reg
*);
1018 int (*init
)(struct nouveau_channel
*, u32 handle
);
1020 { "COPY", 4, 0xa0b5, nve0_bo_move_copy
, nve0_bo_move_init
},
1021 { "GRCE", 0, 0xa0b5, nve0_bo_move_copy
, nvc0_bo_move_init
},
1022 { "COPY1", 5, 0x90b8, nvc0_bo_move_copy
, nvc0_bo_move_init
},
1023 { "COPY0", 4, 0x90b5, nvc0_bo_move_copy
, nvc0_bo_move_init
},
1024 { "COPY", 0, 0x85b5, nva3_bo_move_copy
, nv50_bo_move_init
},
1025 { "CRYPT", 0, 0x74c1, nv84_bo_move_exec
, nv50_bo_move_init
},
1026 { "M2MF", 0, 0x9039, nvc0_bo_move_m2mf
, nvc0_bo_move_init
},
1027 { "M2MF", 0, 0x5039, nv50_bo_move_m2mf
, nv50_bo_move_init
},
1028 { "M2MF", 0, 0x0039, nv04_bo_move_m2mf
, nv04_bo_move_init
},
1030 { "CRYPT", 0, 0x88b4, nv98_bo_move_exec
, nv50_bo_move_init
},
1031 }, *mthd
= _methods
;
1032 const char *name
= "CPU";
1036 struct nouveau_object
*object
;
1037 struct nouveau_channel
*chan
;
1038 u32 handle
= (mthd
->engine
<< 16) | mthd
->oclass
;
1043 chan
= drm
->channel
;
1047 ret
= nouveau_object_new(nv_object(drm
), chan
->handle
, handle
,
1048 mthd
->oclass
, NULL
, 0, &object
);
1050 ret
= mthd
->init(chan
, handle
);
1052 nouveau_object_del(nv_object(drm
),
1053 chan
->handle
, handle
);
1057 drm
->ttm
.move
= mthd
->exec
;
1058 drm
->ttm
.chan
= chan
;
1062 } while ((++mthd
)->exec
);
1064 NV_INFO(drm
, "MM: using %s for buffer copies\n", name
);
1068 nouveau_bo_move_flipd(struct ttm_buffer_object
*bo
, bool evict
, bool intr
,
1069 bool no_wait_gpu
, struct ttm_mem_reg
*new_mem
)
1071 u32 placement_memtype
= TTM_PL_FLAG_TT
| TTM_PL_MASK_CACHING
;
1072 struct ttm_placement placement
;
1073 struct ttm_mem_reg tmp_mem
;
1076 placement
.fpfn
= placement
.lpfn
= 0;
1077 placement
.num_placement
= placement
.num_busy_placement
= 1;
1078 placement
.placement
= placement
.busy_placement
= &placement_memtype
;
1081 tmp_mem
.mm_node
= NULL
;
1082 ret
= ttm_bo_mem_space(bo
, &placement
, &tmp_mem
, intr
, no_wait_gpu
);
1086 ret
= ttm_tt_bind(bo
->ttm
, &tmp_mem
);
1090 ret
= nouveau_bo_move_m2mf(bo
, true, intr
, no_wait_gpu
, &tmp_mem
);
1094 ret
= ttm_bo_move_ttm(bo
, true, no_wait_gpu
, new_mem
);
1096 ttm_bo_mem_put(bo
, &tmp_mem
);
1101 nouveau_bo_move_flips(struct ttm_buffer_object
*bo
, bool evict
, bool intr
,
1102 bool no_wait_gpu
, struct ttm_mem_reg
*new_mem
)
1104 u32 placement_memtype
= TTM_PL_FLAG_TT
| TTM_PL_MASK_CACHING
;
1105 struct ttm_placement placement
;
1106 struct ttm_mem_reg tmp_mem
;
1109 placement
.fpfn
= placement
.lpfn
= 0;
1110 placement
.num_placement
= placement
.num_busy_placement
= 1;
1111 placement
.placement
= placement
.busy_placement
= &placement_memtype
;
1114 tmp_mem
.mm_node
= NULL
;
1115 ret
= ttm_bo_mem_space(bo
, &placement
, &tmp_mem
, intr
, no_wait_gpu
);
1119 ret
= ttm_bo_move_ttm(bo
, true, no_wait_gpu
, &tmp_mem
);
1123 ret
= nouveau_bo_move_m2mf(bo
, true, intr
, no_wait_gpu
, new_mem
);
1128 ttm_bo_mem_put(bo
, &tmp_mem
);
1133 nouveau_bo_move_ntfy(struct ttm_buffer_object
*bo
, struct ttm_mem_reg
*new_mem
)
1135 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1136 struct nouveau_vma
*vma
;
1138 /* ttm can now (stupidly) pass the driver bos it didn't create... */
1139 if (bo
->destroy
!= nouveau_bo_del_ttm
)
1142 list_for_each_entry(vma
, &nvbo
->vma_list
, head
) {
1143 if (new_mem
&& new_mem
->mem_type
== TTM_PL_VRAM
) {
1144 nouveau_vm_map(vma
, new_mem
->mm_node
);
1146 if (new_mem
&& new_mem
->mem_type
== TTM_PL_TT
&&
1147 nvbo
->page_shift
== vma
->vm
->vmm
->spg_shift
) {
1148 if (((struct nouveau_mem
*)new_mem
->mm_node
)->sg
)
1149 nouveau_vm_map_sg_table(vma
, 0, new_mem
->
1150 num_pages
<< PAGE_SHIFT
,
1153 nouveau_vm_map_sg(vma
, 0, new_mem
->
1154 num_pages
<< PAGE_SHIFT
,
1157 nouveau_vm_unmap(vma
);
1163 nouveau_bo_vm_bind(struct ttm_buffer_object
*bo
, struct ttm_mem_reg
*new_mem
,
1164 struct nouveau_drm_tile
**new_tile
)
1166 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1167 struct drm_device
*dev
= drm
->dev
;
1168 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1169 u64 offset
= new_mem
->start
<< PAGE_SHIFT
;
1172 if (new_mem
->mem_type
!= TTM_PL_VRAM
)
1175 if (nv_device(drm
->device
)->card_type
>= NV_10
) {
1176 *new_tile
= nv10_bo_set_tiling(dev
, offset
, new_mem
->size
,
1185 nouveau_bo_vm_cleanup(struct ttm_buffer_object
*bo
,
1186 struct nouveau_drm_tile
*new_tile
,
1187 struct nouveau_drm_tile
**old_tile
)
1189 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1190 struct drm_device
*dev
= drm
->dev
;
1192 nv10_bo_put_tile_region(dev
, *old_tile
, bo
->sync_obj
);
1193 *old_tile
= new_tile
;
1197 nouveau_bo_move(struct ttm_buffer_object
*bo
, bool evict
, bool intr
,
1198 bool no_wait_gpu
, struct ttm_mem_reg
*new_mem
)
1200 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1201 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1202 struct ttm_mem_reg
*old_mem
= &bo
->mem
;
1203 struct nouveau_drm_tile
*new_tile
= NULL
;
1206 if (nv_device(drm
->device
)->card_type
< NV_50
) {
1207 ret
= nouveau_bo_vm_bind(bo
, new_mem
, &new_tile
);
1213 if (old_mem
->mem_type
== TTM_PL_SYSTEM
&& !bo
->ttm
) {
1214 BUG_ON(bo
->mem
.mm_node
!= NULL
);
1216 new_mem
->mm_node
= NULL
;
1220 /* CPU copy if we have no accelerated method available */
1221 if (!drm
->ttm
.move
) {
1222 ret
= ttm_bo_move_memcpy(bo
, evict
, no_wait_gpu
, new_mem
);
1226 /* Hardware assisted copy. */
1227 if (new_mem
->mem_type
== TTM_PL_SYSTEM
)
1228 ret
= nouveau_bo_move_flipd(bo
, evict
, intr
,
1229 no_wait_gpu
, new_mem
);
1230 else if (old_mem
->mem_type
== TTM_PL_SYSTEM
)
1231 ret
= nouveau_bo_move_flips(bo
, evict
, intr
,
1232 no_wait_gpu
, new_mem
);
1234 ret
= nouveau_bo_move_m2mf(bo
, evict
, intr
,
1235 no_wait_gpu
, new_mem
);
1240 /* Fallback to software copy. */
1241 ret
= ttm_bo_move_memcpy(bo
, evict
, no_wait_gpu
, new_mem
);
1244 if (nv_device(drm
->device
)->card_type
< NV_50
) {
1246 nouveau_bo_vm_cleanup(bo
, NULL
, &new_tile
);
1248 nouveau_bo_vm_cleanup(bo
, new_tile
, &nvbo
->tile
);
1255 nouveau_bo_verify_access(struct ttm_buffer_object
*bo
, struct file
*filp
)
1261 nouveau_ttm_io_mem_reserve(struct ttm_bo_device
*bdev
, struct ttm_mem_reg
*mem
)
1263 struct ttm_mem_type_manager
*man
= &bdev
->man
[mem
->mem_type
];
1264 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
1265 struct drm_device
*dev
= drm
->dev
;
1268 mem
->bus
.addr
= NULL
;
1269 mem
->bus
.offset
= 0;
1270 mem
->bus
.size
= mem
->num_pages
<< PAGE_SHIFT
;
1272 mem
->bus
.is_iomem
= false;
1273 if (!(man
->flags
& TTM_MEMTYPE_FLAG_MAPPABLE
))
1275 switch (mem
->mem_type
) {
1281 if (drm
->agp
.stat
== ENABLED
) {
1282 mem
->bus
.offset
= mem
->start
<< PAGE_SHIFT
;
1283 mem
->bus
.base
= drm
->agp
.base
;
1284 mem
->bus
.is_iomem
= !dev
->agp
->cant_use_aperture
;
1289 mem
->bus
.offset
= mem
->start
<< PAGE_SHIFT
;
1290 mem
->bus
.base
= pci_resource_start(dev
->pdev
, 1);
1291 mem
->bus
.is_iomem
= true;
1292 if (nv_device(drm
->device
)->card_type
>= NV_50
) {
1293 struct nouveau_bar
*bar
= nouveau_bar(drm
->device
);
1294 struct nouveau_mem
*node
= mem
->mm_node
;
1296 ret
= bar
->umap(bar
, node
, NV_MEM_ACCESS_RW
,
1301 mem
->bus
.offset
= node
->bar_vma
.offset
;
1311 nouveau_ttm_io_mem_free(struct ttm_bo_device
*bdev
, struct ttm_mem_reg
*mem
)
1313 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
1314 struct nouveau_bar
*bar
= nouveau_bar(drm
->device
);
1315 struct nouveau_mem
*node
= mem
->mm_node
;
1317 if (!node
->bar_vma
.node
)
1320 bar
->unmap(bar
, &node
->bar_vma
);
1324 nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object
*bo
)
1326 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1327 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1328 struct nouveau_device
*device
= nv_device(drm
->device
);
1329 u32 mappable
= pci_resource_len(device
->pdev
, 1) >> PAGE_SHIFT
;
1331 /* as long as the bo isn't in vram, and isn't tiled, we've got
1332 * nothing to do here.
1334 if (bo
->mem
.mem_type
!= TTM_PL_VRAM
) {
1335 if (nv_device(drm
->device
)->card_type
< NV_50
||
1336 !nouveau_bo_tile_layout(nvbo
))
1340 /* make sure bo is in mappable vram */
1341 if (bo
->mem
.start
+ bo
->mem
.num_pages
< mappable
)
1345 nvbo
->placement
.fpfn
= 0;
1346 nvbo
->placement
.lpfn
= mappable
;
1347 nouveau_bo_placement_set(nvbo
, TTM_PL_FLAG_VRAM
, 0);
1348 return nouveau_bo_validate(nvbo
, false, false);
1352 nouveau_ttm_tt_populate(struct ttm_tt
*ttm
)
1354 struct ttm_dma_tt
*ttm_dma
= (void *)ttm
;
1355 struct nouveau_drm
*drm
;
1356 struct drm_device
*dev
;
1359 bool slave
= !!(ttm
->page_flags
& TTM_PAGE_FLAG_SG
);
1361 if (ttm
->state
!= tt_unpopulated
)
1364 if (slave
&& ttm
->sg
) {
1365 /* make userspace faulting work */
1366 drm_prime_sg_to_page_addr_arrays(ttm
->sg
, ttm
->pages
,
1367 ttm_dma
->dma_address
, ttm
->num_pages
);
1368 ttm
->state
= tt_unbound
;
1372 drm
= nouveau_bdev(ttm
->bdev
);
1376 if (drm
->agp
.stat
== ENABLED
) {
1377 return ttm_agp_tt_populate(ttm
);
1381 #ifdef CONFIG_SWIOTLB
1382 if (swiotlb_nr_tbl()) {
1383 return ttm_dma_populate((void *)ttm
, dev
->dev
);
1387 r
= ttm_pool_populate(ttm
);
1392 for (i
= 0; i
< ttm
->num_pages
; i
++) {
1393 ttm_dma
->dma_address
[i
] = pci_map_page(dev
->pdev
, ttm
->pages
[i
],
1395 PCI_DMA_BIDIRECTIONAL
);
1396 if (pci_dma_mapping_error(dev
->pdev
, ttm_dma
->dma_address
[i
])) {
1398 pci_unmap_page(dev
->pdev
, ttm_dma
->dma_address
[i
],
1399 PAGE_SIZE
, PCI_DMA_BIDIRECTIONAL
);
1400 ttm_dma
->dma_address
[i
] = 0;
1402 ttm_pool_unpopulate(ttm
);
1410 nouveau_ttm_tt_unpopulate(struct ttm_tt
*ttm
)
1412 struct ttm_dma_tt
*ttm_dma
= (void *)ttm
;
1413 struct nouveau_drm
*drm
;
1414 struct drm_device
*dev
;
1416 bool slave
= !!(ttm
->page_flags
& TTM_PAGE_FLAG_SG
);
1421 drm
= nouveau_bdev(ttm
->bdev
);
1425 if (drm
->agp
.stat
== ENABLED
) {
1426 ttm_agp_tt_unpopulate(ttm
);
1431 #ifdef CONFIG_SWIOTLB
1432 if (swiotlb_nr_tbl()) {
1433 ttm_dma_unpopulate((void *)ttm
, dev
->dev
);
1438 for (i
= 0; i
< ttm
->num_pages
; i
++) {
1439 if (ttm_dma
->dma_address
[i
]) {
1440 pci_unmap_page(dev
->pdev
, ttm_dma
->dma_address
[i
],
1441 PAGE_SIZE
, PCI_DMA_BIDIRECTIONAL
);
1445 ttm_pool_unpopulate(ttm
);
1449 nouveau_bo_fence(struct nouveau_bo
*nvbo
, struct nouveau_fence
*fence
)
1451 struct nouveau_fence
*old_fence
= NULL
;
1454 nouveau_fence_ref(fence
);
1456 spin_lock(&nvbo
->bo
.bdev
->fence_lock
);
1457 old_fence
= nvbo
->bo
.sync_obj
;
1458 nvbo
->bo
.sync_obj
= fence
;
1459 spin_unlock(&nvbo
->bo
.bdev
->fence_lock
);
1461 nouveau_fence_unref(&old_fence
);
1465 nouveau_bo_fence_unref(void **sync_obj
)
1467 nouveau_fence_unref((struct nouveau_fence
**)sync_obj
);
1471 nouveau_bo_fence_ref(void *sync_obj
)
1473 return nouveau_fence_ref(sync_obj
);
1477 nouveau_bo_fence_signalled(void *sync_obj
)
1479 return nouveau_fence_done(sync_obj
);
1483 nouveau_bo_fence_wait(void *sync_obj
, bool lazy
, bool intr
)
1485 return nouveau_fence_wait(sync_obj
, lazy
, intr
);
1489 nouveau_bo_fence_flush(void *sync_obj
)
1494 struct ttm_bo_driver nouveau_bo_driver
= {
1495 .ttm_tt_create
= &nouveau_ttm_tt_create
,
1496 .ttm_tt_populate
= &nouveau_ttm_tt_populate
,
1497 .ttm_tt_unpopulate
= &nouveau_ttm_tt_unpopulate
,
1498 .invalidate_caches
= nouveau_bo_invalidate_caches
,
1499 .init_mem_type
= nouveau_bo_init_mem_type
,
1500 .evict_flags
= nouveau_bo_evict_flags
,
1501 .move_notify
= nouveau_bo_move_ntfy
,
1502 .move
= nouveau_bo_move
,
1503 .verify_access
= nouveau_bo_verify_access
,
1504 .sync_obj_signaled
= nouveau_bo_fence_signalled
,
1505 .sync_obj_wait
= nouveau_bo_fence_wait
,
1506 .sync_obj_flush
= nouveau_bo_fence_flush
,
1507 .sync_obj_unref
= nouveau_bo_fence_unref
,
1508 .sync_obj_ref
= nouveau_bo_fence_ref
,
1509 .fault_reserve_notify
= &nouveau_ttm_fault_reserve_notify
,
1510 .io_mem_reserve
= &nouveau_ttm_io_mem_reserve
,
1511 .io_mem_free
= &nouveau_ttm_io_mem_free
,
1514 struct nouveau_vma
*
1515 nouveau_bo_vma_find(struct nouveau_bo
*nvbo
, struct nouveau_vm
*vm
)
1517 struct nouveau_vma
*vma
;
1518 list_for_each_entry(vma
, &nvbo
->vma_list
, head
) {
1527 nouveau_bo_vma_add(struct nouveau_bo
*nvbo
, struct nouveau_vm
*vm
,
1528 struct nouveau_vma
*vma
)
1530 const u32 size
= nvbo
->bo
.mem
.num_pages
<< PAGE_SHIFT
;
1531 struct nouveau_mem
*node
= nvbo
->bo
.mem
.mm_node
;
1534 ret
= nouveau_vm_get(vm
, size
, nvbo
->page_shift
,
1535 NV_MEM_ACCESS_RW
, vma
);
1539 if (nvbo
->bo
.mem
.mem_type
== TTM_PL_VRAM
)
1540 nouveau_vm_map(vma
, nvbo
->bo
.mem
.mm_node
);
1541 else if (nvbo
->bo
.mem
.mem_type
== TTM_PL_TT
) {
1543 nouveau_vm_map_sg_table(vma
, 0, size
, node
);
1545 nouveau_vm_map_sg(vma
, 0, size
, node
);
1548 list_add_tail(&vma
->head
, &nvbo
->vma_list
);
1554 nouveau_bo_vma_del(struct nouveau_bo
*nvbo
, struct nouveau_vma
*vma
)
1557 if (nvbo
->bo
.mem
.mem_type
!= TTM_PL_SYSTEM
)
1558 nouveau_vm_unmap(vma
);
1559 nouveau_vm_put(vma
);
1560 list_del(&vma
->head
);