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>
32 #include <subdev/fb.h>
33 #include <subdev/vm.h>
34 #include <subdev/bar.h>
36 #include "nouveau_drm.h"
37 #include "nouveau_dma.h"
38 #include "nouveau_fence.h"
40 #include "nouveau_bo.h"
41 #include "nouveau_ttm.h"
42 #include "nouveau_gem.h"
45 * NV10-NV40 tiling helpers
49 nv10_bo_update_tile_region(struct drm_device
*dev
, struct nouveau_drm_tile
*reg
,
50 u32 addr
, u32 size
, u32 pitch
, u32 flags
)
52 struct nouveau_drm
*drm
= nouveau_drm(dev
);
53 int i
= reg
- drm
->tile
.reg
;
54 struct nouveau_fb
*pfb
= nouveau_fb(drm
->device
);
55 struct nouveau_fb_tile
*tile
= &pfb
->tile
.region
[i
];
56 struct nouveau_engine
*engine
;
58 nouveau_fence_unref(®
->fence
);
61 pfb
->tile
.fini(pfb
, i
, tile
);
64 pfb
->tile
.init(pfb
, i
, addr
, size
, pitch
, flags
, tile
);
66 pfb
->tile
.prog(pfb
, i
, tile
);
68 if ((engine
= nouveau_engine(pfb
, NVDEV_ENGINE_GR
)))
69 engine
->tile_prog(engine
, i
);
70 if ((engine
= nouveau_engine(pfb
, NVDEV_ENGINE_MPEG
)))
71 engine
->tile_prog(engine
, i
);
74 static struct nouveau_drm_tile
*
75 nv10_bo_get_tile_region(struct drm_device
*dev
, int i
)
77 struct nouveau_drm
*drm
= nouveau_drm(dev
);
78 struct nouveau_drm_tile
*tile
= &drm
->tile
.reg
[i
];
80 spin_lock(&drm
->tile
.lock
);
83 (!tile
->fence
|| nouveau_fence_done(tile
->fence
)))
88 spin_unlock(&drm
->tile
.lock
);
93 nv10_bo_put_tile_region(struct drm_device
*dev
, struct nouveau_drm_tile
*tile
,
94 struct nouveau_fence
*fence
)
96 struct nouveau_drm
*drm
= nouveau_drm(dev
);
99 spin_lock(&drm
->tile
.lock
);
101 /* Mark it as pending. */
103 nouveau_fence_ref(fence
);
107 spin_unlock(&drm
->tile
.lock
);
111 static struct nouveau_drm_tile
*
112 nv10_bo_set_tiling(struct drm_device
*dev
, u32 addr
,
113 u32 size
, u32 pitch
, u32 flags
)
115 struct nouveau_drm
*drm
= nouveau_drm(dev
);
116 struct nouveau_fb
*pfb
= nouveau_fb(drm
->device
);
117 struct nouveau_drm_tile
*tile
, *found
= NULL
;
120 for (i
= 0; i
< pfb
->tile
.regions
; i
++) {
121 tile
= nv10_bo_get_tile_region(dev
, i
);
123 if (pitch
&& !found
) {
127 } else if (tile
&& pfb
->tile
.region
[i
].pitch
) {
128 /* Kill an unused tile region. */
129 nv10_bo_update_tile_region(dev
, tile
, 0, 0, 0, 0);
132 nv10_bo_put_tile_region(dev
, tile
, NULL
);
136 nv10_bo_update_tile_region(dev
, found
, addr
, size
,
142 nouveau_bo_del_ttm(struct ttm_buffer_object
*bo
)
144 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
145 struct drm_device
*dev
= drm
->dev
;
146 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
148 if (unlikely(nvbo
->gem
))
149 DRM_ERROR("bo %p still attached to GEM object\n", bo
);
150 nv10_bo_put_tile_region(dev
, nvbo
->tile
, NULL
);
155 nouveau_bo_fixup_align(struct nouveau_bo
*nvbo
, u32 flags
,
156 int *align
, int *size
)
158 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
159 struct nouveau_device
*device
= nv_device(drm
->device
);
161 if (device
->card_type
< NV_50
) {
162 if (nvbo
->tile_mode
) {
163 if (device
->chipset
>= 0x40) {
165 *size
= roundup(*size
, 64 * nvbo
->tile_mode
);
167 } else if (device
->chipset
>= 0x30) {
169 *size
= roundup(*size
, 64 * nvbo
->tile_mode
);
171 } else if (device
->chipset
>= 0x20) {
173 *size
= roundup(*size
, 64 * nvbo
->tile_mode
);
175 } else if (device
->chipset
>= 0x10) {
177 *size
= roundup(*size
, 32 * nvbo
->tile_mode
);
181 *size
= roundup(*size
, (1 << nvbo
->page_shift
));
182 *align
= max((1 << nvbo
->page_shift
), *align
);
185 *size
= roundup(*size
, PAGE_SIZE
);
189 nouveau_bo_new(struct drm_device
*dev
, int size
, int align
,
190 uint32_t flags
, uint32_t tile_mode
, uint32_t tile_flags
,
192 struct nouveau_bo
**pnvbo
)
194 struct nouveau_drm
*drm
= nouveau_drm(dev
);
195 struct nouveau_bo
*nvbo
;
198 int type
= ttm_bo_type_device
;
201 type
= ttm_bo_type_sg
;
203 nvbo
= kzalloc(sizeof(struct nouveau_bo
), GFP_KERNEL
);
206 INIT_LIST_HEAD(&nvbo
->head
);
207 INIT_LIST_HEAD(&nvbo
->entry
);
208 INIT_LIST_HEAD(&nvbo
->vma_list
);
209 nvbo
->tile_mode
= tile_mode
;
210 nvbo
->tile_flags
= tile_flags
;
211 nvbo
->bo
.bdev
= &drm
->ttm
.bdev
;
213 nvbo
->page_shift
= 12;
214 if (drm
->client
.base
.vm
) {
215 if (!(flags
& TTM_PL_FLAG_TT
) && size
> 256 * 1024)
216 nvbo
->page_shift
= drm
->client
.base
.vm
->vmm
->lpg_shift
;
219 nouveau_bo_fixup_align(nvbo
, flags
, &align
, &size
);
220 nvbo
->bo
.mem
.num_pages
= size
>> PAGE_SHIFT
;
221 nouveau_bo_placement_set(nvbo
, flags
, 0);
223 acc_size
= ttm_bo_dma_acc_size(&drm
->ttm
.bdev
, size
,
224 sizeof(struct nouveau_bo
));
226 ret
= ttm_bo_init(&drm
->ttm
.bdev
, &nvbo
->bo
, size
,
227 type
, &nvbo
->placement
,
228 align
>> PAGE_SHIFT
, 0, false, NULL
, acc_size
, sg
,
231 /* ttm will call nouveau_bo_del_ttm if it fails.. */
240 set_placement_list(uint32_t *pl
, unsigned *n
, uint32_t type
, uint32_t flags
)
244 if (type
& TTM_PL_FLAG_VRAM
)
245 pl
[(*n
)++] = TTM_PL_FLAG_VRAM
| flags
;
246 if (type
& TTM_PL_FLAG_TT
)
247 pl
[(*n
)++] = TTM_PL_FLAG_TT
| flags
;
248 if (type
& TTM_PL_FLAG_SYSTEM
)
249 pl
[(*n
)++] = TTM_PL_FLAG_SYSTEM
| flags
;
253 set_placement_range(struct nouveau_bo
*nvbo
, uint32_t type
)
255 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
256 struct nouveau_fb
*pfb
= nouveau_fb(drm
->device
);
257 u32 vram_pages
= pfb
->ram
.size
>> PAGE_SHIFT
;
259 if (nv_device(drm
->device
)->card_type
== NV_10
&&
260 nvbo
->tile_mode
&& (type
& TTM_PL_FLAG_VRAM
) &&
261 nvbo
->bo
.mem
.num_pages
< vram_pages
/ 4) {
263 * Make sure that the color and depth buffers are handled
264 * by independent memory controller units. Up to a 9x
265 * speed up when alpha-blending and depth-test are enabled
268 if (nvbo
->tile_flags
& NOUVEAU_GEM_TILE_ZETA
) {
269 nvbo
->placement
.fpfn
= vram_pages
/ 2;
270 nvbo
->placement
.lpfn
= ~0;
272 nvbo
->placement
.fpfn
= 0;
273 nvbo
->placement
.lpfn
= vram_pages
/ 2;
279 nouveau_bo_placement_set(struct nouveau_bo
*nvbo
, uint32_t type
, uint32_t busy
)
281 struct ttm_placement
*pl
= &nvbo
->placement
;
282 uint32_t flags
= TTM_PL_MASK_CACHING
|
283 (nvbo
->pin_refcnt
? TTM_PL_FLAG_NO_EVICT
: 0);
285 pl
->placement
= nvbo
->placements
;
286 set_placement_list(nvbo
->placements
, &pl
->num_placement
,
289 pl
->busy_placement
= nvbo
->busy_placements
;
290 set_placement_list(nvbo
->busy_placements
, &pl
->num_busy_placement
,
293 set_placement_range(nvbo
, type
);
297 nouveau_bo_pin(struct nouveau_bo
*nvbo
, uint32_t memtype
)
299 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
300 struct ttm_buffer_object
*bo
= &nvbo
->bo
;
303 if (nvbo
->pin_refcnt
&& !(memtype
& (1 << bo
->mem
.mem_type
))) {
304 NV_ERROR(drm
, "bo %p pinned elsewhere: 0x%08x vs 0x%08x\n", bo
,
305 1 << bo
->mem
.mem_type
, memtype
);
309 if (nvbo
->pin_refcnt
++)
312 ret
= ttm_bo_reserve(bo
, false, false, false, 0);
316 nouveau_bo_placement_set(nvbo
, memtype
, 0);
318 ret
= nouveau_bo_validate(nvbo
, false, false, false);
320 switch (bo
->mem
.mem_type
) {
322 drm
->gem
.vram_available
-= bo
->mem
.size
;
325 drm
->gem
.gart_available
-= bo
->mem
.size
;
331 ttm_bo_unreserve(bo
);
339 nouveau_bo_unpin(struct nouveau_bo
*nvbo
)
341 struct nouveau_drm
*drm
= nouveau_bdev(nvbo
->bo
.bdev
);
342 struct ttm_buffer_object
*bo
= &nvbo
->bo
;
345 if (--nvbo
->pin_refcnt
)
348 ret
= ttm_bo_reserve(bo
, false, false, false, 0);
352 nouveau_bo_placement_set(nvbo
, bo
->mem
.placement
, 0);
354 ret
= nouveau_bo_validate(nvbo
, false, false, false);
356 switch (bo
->mem
.mem_type
) {
358 drm
->gem
.vram_available
+= bo
->mem
.size
;
361 drm
->gem
.gart_available
+= bo
->mem
.size
;
368 ttm_bo_unreserve(bo
);
373 nouveau_bo_map(struct nouveau_bo
*nvbo
)
377 ret
= ttm_bo_reserve(&nvbo
->bo
, false, false, false, 0);
381 ret
= ttm_bo_kmap(&nvbo
->bo
, 0, nvbo
->bo
.mem
.num_pages
, &nvbo
->kmap
);
382 ttm_bo_unreserve(&nvbo
->bo
);
387 nouveau_bo_unmap(struct nouveau_bo
*nvbo
)
390 ttm_bo_kunmap(&nvbo
->kmap
);
394 nouveau_bo_validate(struct nouveau_bo
*nvbo
, bool interruptible
,
395 bool no_wait_reserve
, bool no_wait_gpu
)
399 ret
= ttm_bo_validate(&nvbo
->bo
, &nvbo
->placement
, interruptible
,
400 no_wait_reserve
, no_wait_gpu
);
408 nouveau_bo_rd16(struct nouveau_bo
*nvbo
, unsigned index
)
411 u16
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
414 return ioread16_native((void __force __iomem
*)mem
);
420 nouveau_bo_wr16(struct nouveau_bo
*nvbo
, unsigned index
, u16 val
)
423 u16
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
426 iowrite16_native(val
, (void __force __iomem
*)mem
);
432 nouveau_bo_rd32(struct nouveau_bo
*nvbo
, unsigned index
)
435 u32
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
438 return ioread32_native((void __force __iomem
*)mem
);
444 nouveau_bo_wr32(struct nouveau_bo
*nvbo
, unsigned index
, u32 val
)
447 u32
*mem
= ttm_kmap_obj_virtual(&nvbo
->kmap
, &is_iomem
);
450 iowrite32_native(val
, (void __force __iomem
*)mem
);
455 static struct ttm_tt
*
456 nouveau_ttm_tt_create(struct ttm_bo_device
*bdev
, unsigned long size
,
457 uint32_t page_flags
, struct page
*dummy_read
)
459 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
460 struct drm_device
*dev
= drm
->dev
;
462 if (drm
->agp
.stat
== ENABLED
) {
463 return ttm_agp_tt_create(bdev
, dev
->agp
->bridge
, size
,
464 page_flags
, dummy_read
);
467 return nouveau_sgdma_create_ttm(bdev
, size
, page_flags
, dummy_read
);
471 nouveau_bo_invalidate_caches(struct ttm_bo_device
*bdev
, uint32_t flags
)
473 /* We'll do this from user space. */
478 nouveau_bo_init_mem_type(struct ttm_bo_device
*bdev
, uint32_t type
,
479 struct ttm_mem_type_manager
*man
)
481 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
485 man
->flags
= TTM_MEMTYPE_FLAG_MAPPABLE
;
486 man
->available_caching
= TTM_PL_MASK_CACHING
;
487 man
->default_caching
= TTM_PL_FLAG_CACHED
;
490 if (nv_device(drm
->device
)->card_type
>= NV_50
) {
491 man
->func
= &nouveau_vram_manager
;
492 man
->io_reserve_fastpath
= false;
493 man
->use_io_reserve_lru
= true;
495 man
->func
= &ttm_bo_manager_func
;
497 man
->flags
= TTM_MEMTYPE_FLAG_FIXED
|
498 TTM_MEMTYPE_FLAG_MAPPABLE
;
499 man
->available_caching
= TTM_PL_FLAG_UNCACHED
|
501 man
->default_caching
= TTM_PL_FLAG_WC
;
504 if (nv_device(drm
->device
)->card_type
>= NV_50
)
505 man
->func
= &nouveau_gart_manager
;
507 if (drm
->agp
.stat
!= ENABLED
)
508 man
->func
= &nv04_gart_manager
;
510 man
->func
= &ttm_bo_manager_func
;
512 if (drm
->agp
.stat
== ENABLED
) {
513 man
->flags
= TTM_MEMTYPE_FLAG_MAPPABLE
;
514 man
->available_caching
= TTM_PL_FLAG_UNCACHED
|
516 man
->default_caching
= TTM_PL_FLAG_WC
;
518 man
->flags
= TTM_MEMTYPE_FLAG_MAPPABLE
|
519 TTM_MEMTYPE_FLAG_CMA
;
520 man
->available_caching
= TTM_PL_MASK_CACHING
;
521 man
->default_caching
= TTM_PL_FLAG_CACHED
;
532 nouveau_bo_evict_flags(struct ttm_buffer_object
*bo
, struct ttm_placement
*pl
)
534 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
536 switch (bo
->mem
.mem_type
) {
538 nouveau_bo_placement_set(nvbo
, TTM_PL_FLAG_TT
,
542 nouveau_bo_placement_set(nvbo
, TTM_PL_FLAG_SYSTEM
, 0);
546 *pl
= nvbo
->placement
;
550 /* GPU-assisted copy using NV_MEMORY_TO_MEMORY_FORMAT, can access
551 * TTM_PL_{VRAM,TT} directly.
555 nouveau_bo_move_accel_cleanup(struct nouveau_channel
*chan
,
556 struct nouveau_bo
*nvbo
, bool evict
,
557 bool no_wait_reserve
, bool no_wait_gpu
,
558 struct ttm_mem_reg
*new_mem
)
560 struct nouveau_fence
*fence
= NULL
;
563 ret
= nouveau_fence_new(chan
, &fence
);
567 ret
= ttm_bo_move_accel_cleanup(&nvbo
->bo
, fence
, NULL
, evict
,
568 no_wait_reserve
, no_wait_gpu
, new_mem
);
569 nouveau_fence_unref(&fence
);
574 nve0_bo_move_copy(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
575 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
577 struct nouveau_mem
*node
= old_mem
->mm_node
;
578 int ret
= RING_SPACE(chan
, 10);
580 BEGIN_NVC0(chan
, NvSubCopy
, 0x0400, 8);
581 OUT_RING (chan
, upper_32_bits(node
->vma
[0].offset
));
582 OUT_RING (chan
, lower_32_bits(node
->vma
[0].offset
));
583 OUT_RING (chan
, upper_32_bits(node
->vma
[1].offset
));
584 OUT_RING (chan
, lower_32_bits(node
->vma
[1].offset
));
585 OUT_RING (chan
, PAGE_SIZE
);
586 OUT_RING (chan
, PAGE_SIZE
);
587 OUT_RING (chan
, PAGE_SIZE
);
588 OUT_RING (chan
, new_mem
->num_pages
);
589 BEGIN_IMC0(chan
, NvSubCopy
, 0x0300, 0x0386);
595 nvc0_bo_move_init(struct nouveau_channel
*chan
, u32 handle
)
597 int ret
= RING_SPACE(chan
, 2);
599 BEGIN_NVC0(chan
, NvSubCopy
, 0x0000, 1);
600 OUT_RING (chan
, handle
);
606 nvc0_bo_move_copy(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
607 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
609 struct nouveau_mem
*node
= old_mem
->mm_node
;
610 u64 src_offset
= node
->vma
[0].offset
;
611 u64 dst_offset
= node
->vma
[1].offset
;
612 u32 page_count
= new_mem
->num_pages
;
615 page_count
= new_mem
->num_pages
;
617 int line_count
= (page_count
> 8191) ? 8191 : page_count
;
619 ret
= RING_SPACE(chan
, 11);
623 BEGIN_NVC0(chan
, NvSubCopy
, 0x030c, 8);
624 OUT_RING (chan
, upper_32_bits(src_offset
));
625 OUT_RING (chan
, lower_32_bits(src_offset
));
626 OUT_RING (chan
, upper_32_bits(dst_offset
));
627 OUT_RING (chan
, lower_32_bits(dst_offset
));
628 OUT_RING (chan
, PAGE_SIZE
);
629 OUT_RING (chan
, PAGE_SIZE
);
630 OUT_RING (chan
, PAGE_SIZE
);
631 OUT_RING (chan
, line_count
);
632 BEGIN_NVC0(chan
, NvSubCopy
, 0x0300, 1);
633 OUT_RING (chan
, 0x00000110);
635 page_count
-= line_count
;
636 src_offset
+= (PAGE_SIZE
* line_count
);
637 dst_offset
+= (PAGE_SIZE
* line_count
);
644 nvc0_bo_move_m2mf(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
645 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
647 struct nouveau_mem
*node
= old_mem
->mm_node
;
648 u64 src_offset
= node
->vma
[0].offset
;
649 u64 dst_offset
= node
->vma
[1].offset
;
650 u32 page_count
= new_mem
->num_pages
;
653 page_count
= new_mem
->num_pages
;
655 int line_count
= (page_count
> 2047) ? 2047 : page_count
;
657 ret
= RING_SPACE(chan
, 12);
661 BEGIN_NVC0(chan
, NvSubCopy
, 0x0238, 2);
662 OUT_RING (chan
, upper_32_bits(dst_offset
));
663 OUT_RING (chan
, lower_32_bits(dst_offset
));
664 BEGIN_NVC0(chan
, NvSubCopy
, 0x030c, 6);
665 OUT_RING (chan
, upper_32_bits(src_offset
));
666 OUT_RING (chan
, lower_32_bits(src_offset
));
667 OUT_RING (chan
, PAGE_SIZE
); /* src_pitch */
668 OUT_RING (chan
, PAGE_SIZE
); /* dst_pitch */
669 OUT_RING (chan
, PAGE_SIZE
); /* line_length */
670 OUT_RING (chan
, line_count
);
671 BEGIN_NVC0(chan
, NvSubCopy
, 0x0300, 1);
672 OUT_RING (chan
, 0x00100110);
674 page_count
-= line_count
;
675 src_offset
+= (PAGE_SIZE
* line_count
);
676 dst_offset
+= (PAGE_SIZE
* line_count
);
683 nva3_bo_move_copy(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
684 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
686 struct nouveau_mem
*node
= old_mem
->mm_node
;
687 u64 src_offset
= node
->vma
[0].offset
;
688 u64 dst_offset
= node
->vma
[1].offset
;
689 u32 page_count
= new_mem
->num_pages
;
692 page_count
= new_mem
->num_pages
;
694 int line_count
= (page_count
> 8191) ? 8191 : page_count
;
696 ret
= RING_SPACE(chan
, 11);
700 BEGIN_NV04(chan
, NvSubCopy
, 0x030c, 8);
701 OUT_RING (chan
, upper_32_bits(src_offset
));
702 OUT_RING (chan
, lower_32_bits(src_offset
));
703 OUT_RING (chan
, upper_32_bits(dst_offset
));
704 OUT_RING (chan
, lower_32_bits(dst_offset
));
705 OUT_RING (chan
, PAGE_SIZE
);
706 OUT_RING (chan
, PAGE_SIZE
);
707 OUT_RING (chan
, PAGE_SIZE
);
708 OUT_RING (chan
, line_count
);
709 BEGIN_NV04(chan
, NvSubCopy
, 0x0300, 1);
710 OUT_RING (chan
, 0x00000110);
712 page_count
-= line_count
;
713 src_offset
+= (PAGE_SIZE
* line_count
);
714 dst_offset
+= (PAGE_SIZE
* line_count
);
721 nv98_bo_move_exec(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
722 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
724 struct nouveau_mem
*node
= old_mem
->mm_node
;
725 int ret
= RING_SPACE(chan
, 7);
727 BEGIN_NV04(chan
, NvSubCopy
, 0x0320, 6);
728 OUT_RING (chan
, upper_32_bits(node
->vma
[0].offset
));
729 OUT_RING (chan
, lower_32_bits(node
->vma
[0].offset
));
730 OUT_RING (chan
, upper_32_bits(node
->vma
[1].offset
));
731 OUT_RING (chan
, lower_32_bits(node
->vma
[1].offset
));
732 OUT_RING (chan
, 0x00000000 /* COPY */);
733 OUT_RING (chan
, new_mem
->num_pages
<< PAGE_SHIFT
);
739 nv84_bo_move_exec(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
740 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
742 struct nouveau_mem
*node
= old_mem
->mm_node
;
743 int ret
= RING_SPACE(chan
, 7);
745 BEGIN_NV04(chan
, NvSubCopy
, 0x0304, 6);
746 OUT_RING (chan
, new_mem
->num_pages
<< PAGE_SHIFT
);
747 OUT_RING (chan
, upper_32_bits(node
->vma
[0].offset
));
748 OUT_RING (chan
, lower_32_bits(node
->vma
[0].offset
));
749 OUT_RING (chan
, upper_32_bits(node
->vma
[1].offset
));
750 OUT_RING (chan
, lower_32_bits(node
->vma
[1].offset
));
751 OUT_RING (chan
, 0x00000000 /* MODE_COPY, QUERY_NONE */);
757 nv50_bo_move_init(struct nouveau_channel
*chan
, u32 handle
)
759 int ret
= RING_SPACE(chan
, 6);
761 BEGIN_NV04(chan
, NvSubCopy
, 0x0000, 1);
762 OUT_RING (chan
, handle
);
763 BEGIN_NV04(chan
, NvSubCopy
, 0x0180, 3);
764 OUT_RING (chan
, NvNotify0
);
765 OUT_RING (chan
, NvDmaFB
);
766 OUT_RING (chan
, NvDmaFB
);
773 nv50_bo_move_m2mf(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
774 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
776 struct nouveau_mem
*node
= old_mem
->mm_node
;
777 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
778 u64 length
= (new_mem
->num_pages
<< PAGE_SHIFT
);
779 u64 src_offset
= node
->vma
[0].offset
;
780 u64 dst_offset
= node
->vma
[1].offset
;
784 u32 amount
, stride
, height
;
786 amount
= min(length
, (u64
)(4 * 1024 * 1024));
788 height
= amount
/ stride
;
790 if (new_mem
->mem_type
== TTM_PL_VRAM
&&
791 nouveau_bo_tile_layout(nvbo
)) {
792 ret
= RING_SPACE(chan
, 8);
796 BEGIN_NV04(chan
, NvSubCopy
, 0x0200, 7);
799 OUT_RING (chan
, stride
);
800 OUT_RING (chan
, height
);
805 ret
= RING_SPACE(chan
, 2);
809 BEGIN_NV04(chan
, NvSubCopy
, 0x0200, 1);
812 if (old_mem
->mem_type
== TTM_PL_VRAM
&&
813 nouveau_bo_tile_layout(nvbo
)) {
814 ret
= RING_SPACE(chan
, 8);
818 BEGIN_NV04(chan
, NvSubCopy
, 0x021c, 7);
821 OUT_RING (chan
, stride
);
822 OUT_RING (chan
, height
);
827 ret
= RING_SPACE(chan
, 2);
831 BEGIN_NV04(chan
, NvSubCopy
, 0x021c, 1);
835 ret
= RING_SPACE(chan
, 14);
839 BEGIN_NV04(chan
, NvSubCopy
, 0x0238, 2);
840 OUT_RING (chan
, upper_32_bits(src_offset
));
841 OUT_RING (chan
, upper_32_bits(dst_offset
));
842 BEGIN_NV04(chan
, NvSubCopy
, 0x030c, 8);
843 OUT_RING (chan
, lower_32_bits(src_offset
));
844 OUT_RING (chan
, lower_32_bits(dst_offset
));
845 OUT_RING (chan
, stride
);
846 OUT_RING (chan
, stride
);
847 OUT_RING (chan
, stride
);
848 OUT_RING (chan
, height
);
849 OUT_RING (chan
, 0x00000101);
850 OUT_RING (chan
, 0x00000000);
851 BEGIN_NV04(chan
, NvSubCopy
, NV_MEMORY_TO_MEMORY_FORMAT_NOP
, 1);
855 src_offset
+= amount
;
856 dst_offset
+= amount
;
863 nv04_bo_move_init(struct nouveau_channel
*chan
, u32 handle
)
865 int ret
= RING_SPACE(chan
, 4);
867 BEGIN_NV04(chan
, NvSubCopy
, 0x0000, 1);
868 OUT_RING (chan
, handle
);
869 BEGIN_NV04(chan
, NvSubCopy
, 0x0180, 1);
870 OUT_RING (chan
, NvNotify0
);
876 static inline uint32_t
877 nouveau_bo_mem_ctxdma(struct ttm_buffer_object
*bo
,
878 struct nouveau_channel
*chan
, struct ttm_mem_reg
*mem
)
880 if (mem
->mem_type
== TTM_PL_TT
)
886 nv04_bo_move_m2mf(struct nouveau_channel
*chan
, struct ttm_buffer_object
*bo
,
887 struct ttm_mem_reg
*old_mem
, struct ttm_mem_reg
*new_mem
)
889 u32 src_offset
= old_mem
->start
<< PAGE_SHIFT
;
890 u32 dst_offset
= new_mem
->start
<< PAGE_SHIFT
;
891 u32 page_count
= new_mem
->num_pages
;
894 ret
= RING_SPACE(chan
, 3);
898 BEGIN_NV04(chan
, NvSubCopy
, NV_MEMORY_TO_MEMORY_FORMAT_DMA_SOURCE
, 2);
899 OUT_RING (chan
, nouveau_bo_mem_ctxdma(bo
, chan
, old_mem
));
900 OUT_RING (chan
, nouveau_bo_mem_ctxdma(bo
, chan
, new_mem
));
902 page_count
= new_mem
->num_pages
;
904 int line_count
= (page_count
> 2047) ? 2047 : page_count
;
906 ret
= RING_SPACE(chan
, 11);
910 BEGIN_NV04(chan
, NvSubCopy
,
911 NV_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN
, 8);
912 OUT_RING (chan
, src_offset
);
913 OUT_RING (chan
, dst_offset
);
914 OUT_RING (chan
, PAGE_SIZE
); /* src_pitch */
915 OUT_RING (chan
, PAGE_SIZE
); /* dst_pitch */
916 OUT_RING (chan
, PAGE_SIZE
); /* line_length */
917 OUT_RING (chan
, line_count
);
918 OUT_RING (chan
, 0x00000101);
919 OUT_RING (chan
, 0x00000000);
920 BEGIN_NV04(chan
, NvSubCopy
, NV_MEMORY_TO_MEMORY_FORMAT_NOP
, 1);
923 page_count
-= line_count
;
924 src_offset
+= (PAGE_SIZE
* line_count
);
925 dst_offset
+= (PAGE_SIZE
* line_count
);
932 nouveau_vma_getmap(struct nouveau_channel
*chan
, struct nouveau_bo
*nvbo
,
933 struct ttm_mem_reg
*mem
, struct nouveau_vma
*vma
)
935 struct nouveau_mem
*node
= mem
->mm_node
;
938 ret
= nouveau_vm_get(nv_client(chan
->cli
)->vm
, mem
->num_pages
<<
939 PAGE_SHIFT
, node
->page_shift
,
940 NV_MEM_ACCESS_RW
, vma
);
944 if (mem
->mem_type
== TTM_PL_VRAM
)
945 nouveau_vm_map(vma
, node
);
947 nouveau_vm_map_sg(vma
, 0, mem
->num_pages
<< PAGE_SHIFT
, node
);
953 nouveau_bo_move_m2mf(struct ttm_buffer_object
*bo
, int evict
, bool intr
,
954 bool no_wait_reserve
, bool no_wait_gpu
,
955 struct ttm_mem_reg
*new_mem
)
957 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
958 struct nouveau_channel
*chan
= chan
= drm
->channel
;
959 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
960 struct ttm_mem_reg
*old_mem
= &bo
->mem
;
963 mutex_lock(&chan
->cli
->mutex
);
965 /* create temporary vmas for the transfer and attach them to the
966 * old nouveau_mem node, these will get cleaned up after ttm has
967 * destroyed the ttm_mem_reg
969 if (nv_device(drm
->device
)->card_type
>= NV_50
) {
970 struct nouveau_mem
*node
= old_mem
->mm_node
;
972 ret
= nouveau_vma_getmap(chan
, nvbo
, old_mem
, &node
->vma
[0]);
976 ret
= nouveau_vma_getmap(chan
, nvbo
, new_mem
, &node
->vma
[1]);
981 ret
= drm
->ttm
.move(chan
, bo
, &bo
->mem
, new_mem
);
983 ret
= nouveau_bo_move_accel_cleanup(chan
, nvbo
, evict
,
985 no_wait_gpu
, new_mem
);
989 mutex_unlock(&chan
->cli
->mutex
);
994 nouveau_bo_move_init(struct nouveau_channel
*chan
)
996 struct nouveau_cli
*cli
= chan
->cli
;
997 struct nouveau_drm
*drm
= chan
->drm
;
998 static const struct {
1002 int (*exec
)(struct nouveau_channel
*,
1003 struct ttm_buffer_object
*,
1004 struct ttm_mem_reg
*, struct ttm_mem_reg
*);
1005 int (*init
)(struct nouveau_channel
*, u32 handle
);
1007 { "COPY", 0, 0xa0b5, nve0_bo_move_copy
, nvc0_bo_move_init
},
1008 { "COPY1", 5, 0x90b8, nvc0_bo_move_copy
, nvc0_bo_move_init
},
1009 { "COPY0", 4, 0x90b5, nvc0_bo_move_copy
, nvc0_bo_move_init
},
1010 { "COPY", 0, 0x85b5, nva3_bo_move_copy
, nv50_bo_move_init
},
1011 { "CRYPT", 0, 0x74c1, nv84_bo_move_exec
, nv50_bo_move_init
},
1012 { "M2MF", 0, 0x9039, nvc0_bo_move_m2mf
, nvc0_bo_move_init
},
1013 { "M2MF", 0, 0x5039, nv50_bo_move_m2mf
, nv50_bo_move_init
},
1014 { "M2MF", 0, 0x0039, nv04_bo_move_m2mf
, nv04_bo_move_init
},
1016 { "CRYPT", 0, 0x88b4, nv98_bo_move_exec
, nv50_bo_move_init
},
1017 }, *mthd
= _methods
;
1018 const char *name
= "CPU";
1022 struct nouveau_object
*object
;
1023 u32 handle
= (mthd
->engine
<< 16) | mthd
->oclass
;
1025 ret
= nouveau_object_new(nv_object(cli
), chan
->handle
, handle
,
1026 mthd
->oclass
, NULL
, 0, &object
);
1028 ret
= mthd
->init(chan
, handle
);
1030 nouveau_object_del(nv_object(cli
),
1031 chan
->handle
, handle
);
1035 drm
->ttm
.move
= mthd
->exec
;
1039 } while ((++mthd
)->exec
);
1041 NV_INFO(drm
, "MM: using %s for buffer copies\n", name
);
1045 nouveau_bo_move_flipd(struct ttm_buffer_object
*bo
, bool evict
, bool intr
,
1046 bool no_wait_reserve
, bool no_wait_gpu
,
1047 struct ttm_mem_reg
*new_mem
)
1049 u32 placement_memtype
= TTM_PL_FLAG_TT
| TTM_PL_MASK_CACHING
;
1050 struct ttm_placement placement
;
1051 struct ttm_mem_reg tmp_mem
;
1054 placement
.fpfn
= placement
.lpfn
= 0;
1055 placement
.num_placement
= placement
.num_busy_placement
= 1;
1056 placement
.placement
= placement
.busy_placement
= &placement_memtype
;
1059 tmp_mem
.mm_node
= NULL
;
1060 ret
= ttm_bo_mem_space(bo
, &placement
, &tmp_mem
, intr
, no_wait_reserve
, no_wait_gpu
);
1064 ret
= ttm_tt_bind(bo
->ttm
, &tmp_mem
);
1068 ret
= nouveau_bo_move_m2mf(bo
, true, intr
, no_wait_reserve
, no_wait_gpu
, &tmp_mem
);
1072 ret
= ttm_bo_move_ttm(bo
, true, no_wait_reserve
, no_wait_gpu
, new_mem
);
1074 ttm_bo_mem_put(bo
, &tmp_mem
);
1079 nouveau_bo_move_flips(struct ttm_buffer_object
*bo
, bool evict
, bool intr
,
1080 bool no_wait_reserve
, bool no_wait_gpu
,
1081 struct ttm_mem_reg
*new_mem
)
1083 u32 placement_memtype
= TTM_PL_FLAG_TT
| TTM_PL_MASK_CACHING
;
1084 struct ttm_placement placement
;
1085 struct ttm_mem_reg tmp_mem
;
1088 placement
.fpfn
= placement
.lpfn
= 0;
1089 placement
.num_placement
= placement
.num_busy_placement
= 1;
1090 placement
.placement
= placement
.busy_placement
= &placement_memtype
;
1093 tmp_mem
.mm_node
= NULL
;
1094 ret
= ttm_bo_mem_space(bo
, &placement
, &tmp_mem
, intr
, no_wait_reserve
, no_wait_gpu
);
1098 ret
= ttm_bo_move_ttm(bo
, true, no_wait_reserve
, no_wait_gpu
, &tmp_mem
);
1102 ret
= nouveau_bo_move_m2mf(bo
, true, intr
, no_wait_reserve
, no_wait_gpu
, new_mem
);
1107 ttm_bo_mem_put(bo
, &tmp_mem
);
1112 nouveau_bo_move_ntfy(struct ttm_buffer_object
*bo
, struct ttm_mem_reg
*new_mem
)
1114 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1115 struct nouveau_vma
*vma
;
1117 /* ttm can now (stupidly) pass the driver bos it didn't create... */
1118 if (bo
->destroy
!= nouveau_bo_del_ttm
)
1121 list_for_each_entry(vma
, &nvbo
->vma_list
, head
) {
1122 if (new_mem
&& new_mem
->mem_type
== TTM_PL_VRAM
) {
1123 nouveau_vm_map(vma
, new_mem
->mm_node
);
1125 if (new_mem
&& new_mem
->mem_type
== TTM_PL_TT
&&
1126 nvbo
->page_shift
== vma
->vm
->vmm
->spg_shift
) {
1127 if (((struct nouveau_mem
*)new_mem
->mm_node
)->sg
)
1128 nouveau_vm_map_sg_table(vma
, 0, new_mem
->
1129 num_pages
<< PAGE_SHIFT
,
1132 nouveau_vm_map_sg(vma
, 0, new_mem
->
1133 num_pages
<< PAGE_SHIFT
,
1136 nouveau_vm_unmap(vma
);
1142 nouveau_bo_vm_bind(struct ttm_buffer_object
*bo
, struct ttm_mem_reg
*new_mem
,
1143 struct nouveau_drm_tile
**new_tile
)
1145 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1146 struct drm_device
*dev
= drm
->dev
;
1147 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1148 u64 offset
= new_mem
->start
<< PAGE_SHIFT
;
1151 if (new_mem
->mem_type
!= TTM_PL_VRAM
)
1154 if (nv_device(drm
->device
)->card_type
>= NV_10
) {
1155 *new_tile
= nv10_bo_set_tiling(dev
, offset
, new_mem
->size
,
1164 nouveau_bo_vm_cleanup(struct ttm_buffer_object
*bo
,
1165 struct nouveau_drm_tile
*new_tile
,
1166 struct nouveau_drm_tile
**old_tile
)
1168 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1169 struct drm_device
*dev
= drm
->dev
;
1171 nv10_bo_put_tile_region(dev
, *old_tile
, bo
->sync_obj
);
1172 *old_tile
= new_tile
;
1176 nouveau_bo_move(struct ttm_buffer_object
*bo
, bool evict
, bool intr
,
1177 bool no_wait_reserve
, bool no_wait_gpu
,
1178 struct ttm_mem_reg
*new_mem
)
1180 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1181 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1182 struct ttm_mem_reg
*old_mem
= &bo
->mem
;
1183 struct nouveau_drm_tile
*new_tile
= NULL
;
1186 if (nv_device(drm
->device
)->card_type
< NV_50
) {
1187 ret
= nouveau_bo_vm_bind(bo
, new_mem
, &new_tile
);
1193 if (old_mem
->mem_type
== TTM_PL_SYSTEM
&& !bo
->ttm
) {
1194 BUG_ON(bo
->mem
.mm_node
!= NULL
);
1196 new_mem
->mm_node
= NULL
;
1200 /* CPU copy if we have no accelerated method available */
1201 if (!drm
->ttm
.move
) {
1202 ret
= ttm_bo_move_memcpy(bo
, evict
, no_wait_reserve
, no_wait_gpu
, new_mem
);
1206 /* Hardware assisted copy. */
1207 if (new_mem
->mem_type
== TTM_PL_SYSTEM
)
1208 ret
= nouveau_bo_move_flipd(bo
, evict
, intr
, no_wait_reserve
, no_wait_gpu
, new_mem
);
1209 else if (old_mem
->mem_type
== TTM_PL_SYSTEM
)
1210 ret
= nouveau_bo_move_flips(bo
, evict
, intr
, no_wait_reserve
, no_wait_gpu
, new_mem
);
1212 ret
= nouveau_bo_move_m2mf(bo
, evict
, intr
, no_wait_reserve
, no_wait_gpu
, new_mem
);
1217 /* Fallback to software copy. */
1218 ret
= ttm_bo_move_memcpy(bo
, evict
, no_wait_reserve
, no_wait_gpu
, new_mem
);
1221 if (nv_device(drm
->device
)->card_type
< NV_50
) {
1223 nouveau_bo_vm_cleanup(bo
, NULL
, &new_tile
);
1225 nouveau_bo_vm_cleanup(bo
, new_tile
, &nvbo
->tile
);
1232 nouveau_bo_verify_access(struct ttm_buffer_object
*bo
, struct file
*filp
)
1238 nouveau_ttm_io_mem_reserve(struct ttm_bo_device
*bdev
, struct ttm_mem_reg
*mem
)
1240 struct ttm_mem_type_manager
*man
= &bdev
->man
[mem
->mem_type
];
1241 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
1242 struct drm_device
*dev
= drm
->dev
;
1245 mem
->bus
.addr
= NULL
;
1246 mem
->bus
.offset
= 0;
1247 mem
->bus
.size
= mem
->num_pages
<< PAGE_SHIFT
;
1249 mem
->bus
.is_iomem
= false;
1250 if (!(man
->flags
& TTM_MEMTYPE_FLAG_MAPPABLE
))
1252 switch (mem
->mem_type
) {
1258 if (drm
->agp
.stat
== ENABLED
) {
1259 mem
->bus
.offset
= mem
->start
<< PAGE_SHIFT
;
1260 mem
->bus
.base
= drm
->agp
.base
;
1261 mem
->bus
.is_iomem
= true;
1266 mem
->bus
.offset
= mem
->start
<< PAGE_SHIFT
;
1267 mem
->bus
.base
= pci_resource_start(dev
->pdev
, 1);
1268 mem
->bus
.is_iomem
= true;
1269 if (nv_device(drm
->device
)->card_type
>= NV_50
) {
1270 struct nouveau_bar
*bar
= nouveau_bar(drm
->device
);
1271 struct nouveau_mem
*node
= mem
->mm_node
;
1273 ret
= bar
->umap(bar
, node
, NV_MEM_ACCESS_RW
,
1278 mem
->bus
.offset
= node
->bar_vma
.offset
;
1288 nouveau_ttm_io_mem_free(struct ttm_bo_device
*bdev
, struct ttm_mem_reg
*mem
)
1290 struct nouveau_drm
*drm
= nouveau_bdev(bdev
);
1291 struct nouveau_bar
*bar
= nouveau_bar(drm
->device
);
1292 struct nouveau_mem
*node
= mem
->mm_node
;
1294 if (!node
->bar_vma
.node
)
1297 bar
->unmap(bar
, &node
->bar_vma
);
1301 nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object
*bo
)
1303 struct nouveau_drm
*drm
= nouveau_bdev(bo
->bdev
);
1304 struct nouveau_bo
*nvbo
= nouveau_bo(bo
);
1305 struct nouveau_device
*device
= nv_device(drm
->device
);
1306 u32 mappable
= pci_resource_len(device
->pdev
, 1) >> PAGE_SHIFT
;
1308 /* as long as the bo isn't in vram, and isn't tiled, we've got
1309 * nothing to do here.
1311 if (bo
->mem
.mem_type
!= TTM_PL_VRAM
) {
1312 if (nv_device(drm
->device
)->card_type
< NV_50
||
1313 !nouveau_bo_tile_layout(nvbo
))
1317 /* make sure bo is in mappable vram */
1318 if (bo
->mem
.start
+ bo
->mem
.num_pages
< mappable
)
1322 nvbo
->placement
.fpfn
= 0;
1323 nvbo
->placement
.lpfn
= mappable
;
1324 nouveau_bo_placement_set(nvbo
, TTM_PL_FLAG_VRAM
, 0);
1325 return nouveau_bo_validate(nvbo
, false, true, false);
1329 nouveau_ttm_tt_populate(struct ttm_tt
*ttm
)
1331 struct ttm_dma_tt
*ttm_dma
= (void *)ttm
;
1332 struct nouveau_drm
*drm
;
1333 struct drm_device
*dev
;
1336 bool slave
= !!(ttm
->page_flags
& TTM_PAGE_FLAG_SG
);
1338 if (ttm
->state
!= tt_unpopulated
)
1341 if (slave
&& ttm
->sg
) {
1342 /* make userspace faulting work */
1343 drm_prime_sg_to_page_addr_arrays(ttm
->sg
, ttm
->pages
,
1344 ttm_dma
->dma_address
, ttm
->num_pages
);
1345 ttm
->state
= tt_unbound
;
1349 drm
= nouveau_bdev(ttm
->bdev
);
1353 if (drm
->agp
.stat
== ENABLED
) {
1354 return ttm_agp_tt_populate(ttm
);
1358 #ifdef CONFIG_SWIOTLB
1359 if (swiotlb_nr_tbl()) {
1360 return ttm_dma_populate((void *)ttm
, dev
->dev
);
1364 r
= ttm_pool_populate(ttm
);
1369 for (i
= 0; i
< ttm
->num_pages
; i
++) {
1370 ttm_dma
->dma_address
[i
] = pci_map_page(dev
->pdev
, ttm
->pages
[i
],
1372 PCI_DMA_BIDIRECTIONAL
);
1373 if (pci_dma_mapping_error(dev
->pdev
, ttm_dma
->dma_address
[i
])) {
1375 pci_unmap_page(dev
->pdev
, ttm_dma
->dma_address
[i
],
1376 PAGE_SIZE
, PCI_DMA_BIDIRECTIONAL
);
1377 ttm_dma
->dma_address
[i
] = 0;
1379 ttm_pool_unpopulate(ttm
);
1387 nouveau_ttm_tt_unpopulate(struct ttm_tt
*ttm
)
1389 struct ttm_dma_tt
*ttm_dma
= (void *)ttm
;
1390 struct nouveau_drm
*drm
;
1391 struct drm_device
*dev
;
1393 bool slave
= !!(ttm
->page_flags
& TTM_PAGE_FLAG_SG
);
1398 drm
= nouveau_bdev(ttm
->bdev
);
1402 if (drm
->agp
.stat
== ENABLED
) {
1403 ttm_agp_tt_unpopulate(ttm
);
1408 #ifdef CONFIG_SWIOTLB
1409 if (swiotlb_nr_tbl()) {
1410 ttm_dma_unpopulate((void *)ttm
, dev
->dev
);
1415 for (i
= 0; i
< ttm
->num_pages
; i
++) {
1416 if (ttm_dma
->dma_address
[i
]) {
1417 pci_unmap_page(dev
->pdev
, ttm_dma
->dma_address
[i
],
1418 PAGE_SIZE
, PCI_DMA_BIDIRECTIONAL
);
1422 ttm_pool_unpopulate(ttm
);
1426 nouveau_bo_fence(struct nouveau_bo
*nvbo
, struct nouveau_fence
*fence
)
1428 struct nouveau_fence
*old_fence
= NULL
;
1431 nouveau_fence_ref(fence
);
1433 spin_lock(&nvbo
->bo
.bdev
->fence_lock
);
1434 old_fence
= nvbo
->bo
.sync_obj
;
1435 nvbo
->bo
.sync_obj
= fence
;
1436 spin_unlock(&nvbo
->bo
.bdev
->fence_lock
);
1438 nouveau_fence_unref(&old_fence
);
1442 nouveau_bo_fence_unref(void **sync_obj
)
1444 nouveau_fence_unref((struct nouveau_fence
**)sync_obj
);
1448 nouveau_bo_fence_ref(void *sync_obj
)
1450 return nouveau_fence_ref(sync_obj
);
1454 nouveau_bo_fence_signalled(void *sync_obj
, void *sync_arg
)
1456 return nouveau_fence_done(sync_obj
);
1460 nouveau_bo_fence_wait(void *sync_obj
, void *sync_arg
, bool lazy
, bool intr
)
1462 return nouveau_fence_wait(sync_obj
, lazy
, intr
);
1466 nouveau_bo_fence_flush(void *sync_obj
, void *sync_arg
)
1471 struct ttm_bo_driver nouveau_bo_driver
= {
1472 .ttm_tt_create
= &nouveau_ttm_tt_create
,
1473 .ttm_tt_populate
= &nouveau_ttm_tt_populate
,
1474 .ttm_tt_unpopulate
= &nouveau_ttm_tt_unpopulate
,
1475 .invalidate_caches
= nouveau_bo_invalidate_caches
,
1476 .init_mem_type
= nouveau_bo_init_mem_type
,
1477 .evict_flags
= nouveau_bo_evict_flags
,
1478 .move_notify
= nouveau_bo_move_ntfy
,
1479 .move
= nouveau_bo_move
,
1480 .verify_access
= nouveau_bo_verify_access
,
1481 .sync_obj_signaled
= nouveau_bo_fence_signalled
,
1482 .sync_obj_wait
= nouveau_bo_fence_wait
,
1483 .sync_obj_flush
= nouveau_bo_fence_flush
,
1484 .sync_obj_unref
= nouveau_bo_fence_unref
,
1485 .sync_obj_ref
= nouveau_bo_fence_ref
,
1486 .fault_reserve_notify
= &nouveau_ttm_fault_reserve_notify
,
1487 .io_mem_reserve
= &nouveau_ttm_io_mem_reserve
,
1488 .io_mem_free
= &nouveau_ttm_io_mem_free
,
1491 struct nouveau_vma
*
1492 nouveau_bo_vma_find(struct nouveau_bo
*nvbo
, struct nouveau_vm
*vm
)
1494 struct nouveau_vma
*vma
;
1495 list_for_each_entry(vma
, &nvbo
->vma_list
, head
) {
1504 nouveau_bo_vma_add(struct nouveau_bo
*nvbo
, struct nouveau_vm
*vm
,
1505 struct nouveau_vma
*vma
)
1507 const u32 size
= nvbo
->bo
.mem
.num_pages
<< PAGE_SHIFT
;
1508 struct nouveau_mem
*node
= nvbo
->bo
.mem
.mm_node
;
1511 ret
= nouveau_vm_get(vm
, size
, nvbo
->page_shift
,
1512 NV_MEM_ACCESS_RW
, vma
);
1516 if (nvbo
->bo
.mem
.mem_type
== TTM_PL_VRAM
)
1517 nouveau_vm_map(vma
, nvbo
->bo
.mem
.mm_node
);
1518 else if (nvbo
->bo
.mem
.mem_type
== TTM_PL_TT
) {
1520 nouveau_vm_map_sg_table(vma
, 0, size
, node
);
1522 nouveau_vm_map_sg(vma
, 0, size
, node
);
1525 list_add_tail(&vma
->head
, &nvbo
->vma_list
);
1531 nouveau_bo_vma_del(struct nouveau_bo
*nvbo
, struct nouveau_vma
*vma
)
1534 if (nvbo
->bo
.mem
.mem_type
!= TTM_PL_SYSTEM
) {
1535 spin_lock(&nvbo
->bo
.bdev
->fence_lock
);
1536 ttm_bo_wait(&nvbo
->bo
, false, false, false);
1537 spin_unlock(&nvbo
->bo
.bdev
->fence_lock
);
1538 nouveau_vm_unmap(vma
);
1541 nouveau_vm_put(vma
);
1542 list_del(&vma
->head
);