2 * Copyright 2009 Jerome Glisse.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
28 * Jerome Glisse <glisse@freedesktop.org>
29 * Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
32 #include <linux/list.h>
33 #include <linux/slab.h>
35 #include "radeon_drm.h"
37 #include "radeon_trace.h"
40 int radeon_ttm_init(struct radeon_device
*rdev
);
41 void radeon_ttm_fini(struct radeon_device
*rdev
);
42 static void radeon_bo_clear_surface_reg(struct radeon_bo
*bo
);
45 * To exclude mutual BO access we rely on bo_reserve exclusion, as all
46 * function are calling it.
49 void radeon_bo_clear_va(struct radeon_bo
*bo
)
51 struct radeon_bo_va
*bo_va
, *tmp
;
53 list_for_each_entry_safe(bo_va
, tmp
, &bo
->va
, bo_list
) {
54 /* remove from all vm address space */
55 radeon_vm_bo_rmv(bo
->rdev
, bo_va
->vm
, bo
);
59 static void radeon_ttm_bo_destroy(struct ttm_buffer_object
*tbo
)
63 bo
= container_of(tbo
, struct radeon_bo
, tbo
);
64 mutex_lock(&bo
->rdev
->gem
.mutex
);
65 list_del_init(&bo
->list
);
66 mutex_unlock(&bo
->rdev
->gem
.mutex
);
67 radeon_bo_clear_surface_reg(bo
);
68 radeon_bo_clear_va(bo
);
69 drm_gem_object_release(&bo
->gem_base
);
73 bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object
*bo
)
75 if (bo
->destroy
== &radeon_ttm_bo_destroy
)
80 void radeon_ttm_placement_from_domain(struct radeon_bo
*rbo
, u32 domain
)
84 rbo
->placement
.fpfn
= 0;
85 rbo
->placement
.lpfn
= 0;
86 rbo
->placement
.placement
= rbo
->placements
;
87 rbo
->placement
.busy_placement
= rbo
->placements
;
88 if (domain
& RADEON_GEM_DOMAIN_VRAM
)
89 rbo
->placements
[c
++] = TTM_PL_FLAG_WC
| TTM_PL_FLAG_UNCACHED
|
91 if (domain
& RADEON_GEM_DOMAIN_GTT
)
92 rbo
->placements
[c
++] = TTM_PL_MASK_CACHING
| TTM_PL_FLAG_TT
;
93 if (domain
& RADEON_GEM_DOMAIN_CPU
)
94 rbo
->placements
[c
++] = TTM_PL_MASK_CACHING
| TTM_PL_FLAG_SYSTEM
;
96 rbo
->placements
[c
++] = TTM_PL_MASK_CACHING
| TTM_PL_FLAG_SYSTEM
;
97 rbo
->placement
.num_placement
= c
;
98 rbo
->placement
.num_busy_placement
= c
;
101 int radeon_bo_create(struct radeon_device
*rdev
,
102 unsigned long size
, int byte_align
, bool kernel
, u32 domain
,
103 struct sg_table
*sg
, struct radeon_bo
**bo_ptr
)
105 struct radeon_bo
*bo
;
106 enum ttm_bo_type type
;
107 unsigned long page_align
= roundup(byte_align
, PAGE_SIZE
) >> PAGE_SHIFT
;
108 unsigned long max_size
= 0;
112 size
= ALIGN(size
, PAGE_SIZE
);
114 rdev
->mman
.bdev
.dev_mapping
= rdev
->ddev
->dev_mapping
;
116 type
= ttm_bo_type_kernel
;
118 type
= ttm_bo_type_sg
;
120 type
= ttm_bo_type_device
;
124 /* maximun bo size is the minimun btw visible vram and gtt size */
125 max_size
= min(rdev
->mc
.visible_vram_size
, rdev
->mc
.gtt_size
);
126 if ((page_align
<< PAGE_SHIFT
) >= max_size
) {
127 printk(KERN_WARNING
"%s:%d alloc size %ldM bigger than %ldMb limit\n",
128 __func__
, __LINE__
, page_align
>> (20 - PAGE_SHIFT
), max_size
>> 20);
132 acc_size
= ttm_bo_dma_acc_size(&rdev
->mman
.bdev
, size
,
133 sizeof(struct radeon_bo
));
136 bo
= kzalloc(sizeof(struct radeon_bo
), GFP_KERNEL
);
139 r
= drm_gem_object_init(rdev
->ddev
, &bo
->gem_base
, size
);
145 bo
->gem_base
.driver_private
= NULL
;
146 bo
->surface_reg
= -1;
147 INIT_LIST_HEAD(&bo
->list
);
148 INIT_LIST_HEAD(&bo
->va
);
149 radeon_ttm_placement_from_domain(bo
, domain
);
150 /* Kernel allocation are uninterruptible */
151 down_read(&rdev
->pm
.mclk_lock
);
152 r
= ttm_bo_init(&rdev
->mman
.bdev
, &bo
->tbo
, size
, type
,
153 &bo
->placement
, page_align
, 0, !kernel
, NULL
,
154 acc_size
, sg
, &radeon_ttm_bo_destroy
);
155 up_read(&rdev
->pm
.mclk_lock
);
156 if (unlikely(r
!= 0)) {
157 if (r
!= -ERESTARTSYS
) {
158 if (domain
== RADEON_GEM_DOMAIN_VRAM
) {
159 domain
|= RADEON_GEM_DOMAIN_GTT
;
163 "object_init failed for (%lu, 0x%08X)\n",
170 trace_radeon_bo_create(bo
);
175 int radeon_bo_kmap(struct radeon_bo
*bo
, void **ptr
)
186 r
= ttm_bo_kmap(&bo
->tbo
, 0, bo
->tbo
.num_pages
, &bo
->kmap
);
190 bo
->kptr
= ttm_kmap_obj_virtual(&bo
->kmap
, &is_iomem
);
194 radeon_bo_check_tiling(bo
, 0, 0);
198 void radeon_bo_kunmap(struct radeon_bo
*bo
)
200 if (bo
->kptr
== NULL
)
203 radeon_bo_check_tiling(bo
, 0, 0);
204 ttm_bo_kunmap(&bo
->kmap
);
207 void radeon_bo_unref(struct radeon_bo
**bo
)
209 struct ttm_buffer_object
*tbo
;
210 struct radeon_device
*rdev
;
216 down_read(&rdev
->pm
.mclk_lock
);
218 up_read(&rdev
->pm
.mclk_lock
);
223 int radeon_bo_pin_restricted(struct radeon_bo
*bo
, u32 domain
, u64 max_offset
,
231 *gpu_addr
= radeon_bo_gpu_offset(bo
);
233 if (max_offset
!= 0) {
236 if (domain
== RADEON_GEM_DOMAIN_VRAM
)
237 domain_start
= bo
->rdev
->mc
.vram_start
;
239 domain_start
= bo
->rdev
->mc
.gtt_start
;
240 WARN_ON_ONCE(max_offset
<
241 (radeon_bo_gpu_offset(bo
) - domain_start
));
246 radeon_ttm_placement_from_domain(bo
, domain
);
247 if (domain
== RADEON_GEM_DOMAIN_VRAM
) {
248 /* force to pin into visible video ram */
249 bo
->placement
.lpfn
= bo
->rdev
->mc
.visible_vram_size
>> PAGE_SHIFT
;
252 u64 lpfn
= max_offset
>> PAGE_SHIFT
;
254 if (!bo
->placement
.lpfn
)
255 bo
->placement
.lpfn
= bo
->rdev
->mc
.gtt_size
>> PAGE_SHIFT
;
257 if (lpfn
< bo
->placement
.lpfn
)
258 bo
->placement
.lpfn
= lpfn
;
260 for (i
= 0; i
< bo
->placement
.num_placement
; i
++)
261 bo
->placements
[i
] |= TTM_PL_FLAG_NO_EVICT
;
262 r
= ttm_bo_validate(&bo
->tbo
, &bo
->placement
, false, false, false);
263 if (likely(r
== 0)) {
265 if (gpu_addr
!= NULL
)
266 *gpu_addr
= radeon_bo_gpu_offset(bo
);
268 if (unlikely(r
!= 0))
269 dev_err(bo
->rdev
->dev
, "%p pin failed\n", bo
);
273 int radeon_bo_pin(struct radeon_bo
*bo
, u32 domain
, u64
*gpu_addr
)
275 return radeon_bo_pin_restricted(bo
, domain
, 0, gpu_addr
);
278 int radeon_bo_unpin(struct radeon_bo
*bo
)
282 if (!bo
->pin_count
) {
283 dev_warn(bo
->rdev
->dev
, "%p unpin not necessary\n", bo
);
289 for (i
= 0; i
< bo
->placement
.num_placement
; i
++)
290 bo
->placements
[i
] &= ~TTM_PL_FLAG_NO_EVICT
;
291 r
= ttm_bo_validate(&bo
->tbo
, &bo
->placement
, false, false, false);
292 if (unlikely(r
!= 0))
293 dev_err(bo
->rdev
->dev
, "%p validate failed for unpin\n", bo
);
297 int radeon_bo_evict_vram(struct radeon_device
*rdev
)
299 /* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
300 if (0 && (rdev
->flags
& RADEON_IS_IGP
)) {
301 if (rdev
->mc
.igp_sideport_enabled
== false)
302 /* Useless to evict on IGP chips */
305 return ttm_bo_evict_mm(&rdev
->mman
.bdev
, TTM_PL_VRAM
);
308 void radeon_bo_force_delete(struct radeon_device
*rdev
)
310 struct radeon_bo
*bo
, *n
;
312 if (list_empty(&rdev
->gem
.objects
)) {
315 dev_err(rdev
->dev
, "Userspace still has active objects !\n");
316 list_for_each_entry_safe(bo
, n
, &rdev
->gem
.objects
, list
) {
317 mutex_lock(&rdev
->ddev
->struct_mutex
);
318 dev_err(rdev
->dev
, "%p %p %lu %lu force free\n",
319 &bo
->gem_base
, bo
, (unsigned long)bo
->gem_base
.size
,
320 *((unsigned long *)&bo
->gem_base
.refcount
));
321 mutex_lock(&bo
->rdev
->gem
.mutex
);
322 list_del_init(&bo
->list
);
323 mutex_unlock(&bo
->rdev
->gem
.mutex
);
324 /* this should unref the ttm bo */
325 drm_gem_object_unreference(&bo
->gem_base
);
326 mutex_unlock(&rdev
->ddev
->struct_mutex
);
330 int radeon_bo_init(struct radeon_device
*rdev
)
332 /* Add an MTRR for the VRAM */
333 rdev
->mc
.vram_mtrr
= mtrr_add(rdev
->mc
.aper_base
, rdev
->mc
.aper_size
,
334 MTRR_TYPE_WRCOMB
, 1);
335 DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
336 rdev
->mc
.mc_vram_size
>> 20,
337 (unsigned long long)rdev
->mc
.aper_size
>> 20);
338 DRM_INFO("RAM width %dbits %cDR\n",
339 rdev
->mc
.vram_width
, rdev
->mc
.vram_is_ddr
? 'D' : 'S');
340 return radeon_ttm_init(rdev
);
343 void radeon_bo_fini(struct radeon_device
*rdev
)
345 radeon_ttm_fini(rdev
);
348 void radeon_bo_list_add_object(struct radeon_bo_list
*lobj
,
349 struct list_head
*head
)
352 list_add(&lobj
->tv
.head
, head
);
354 list_add_tail(&lobj
->tv
.head
, head
);
358 int radeon_bo_list_validate(struct list_head
*head
)
360 struct radeon_bo_list
*lobj
;
361 struct radeon_bo
*bo
;
365 r
= ttm_eu_reserve_buffers(head
);
366 if (unlikely(r
!= 0)) {
369 list_for_each_entry(lobj
, head
, tv
.head
) {
371 if (!bo
->pin_count
) {
372 domain
= lobj
->wdomain
? lobj
->wdomain
: lobj
->rdomain
;
375 radeon_ttm_placement_from_domain(bo
, domain
);
376 r
= ttm_bo_validate(&bo
->tbo
, &bo
->placement
,
379 if (r
!= -ERESTARTSYS
&& domain
== RADEON_GEM_DOMAIN_VRAM
) {
380 domain
|= RADEON_GEM_DOMAIN_GTT
;
386 lobj
->gpu_offset
= radeon_bo_gpu_offset(bo
);
387 lobj
->tiling_flags
= bo
->tiling_flags
;
392 int radeon_bo_fbdev_mmap(struct radeon_bo
*bo
,
393 struct vm_area_struct
*vma
)
395 return ttm_fbdev_mmap(vma
, &bo
->tbo
);
398 int radeon_bo_get_surface_reg(struct radeon_bo
*bo
)
400 struct radeon_device
*rdev
= bo
->rdev
;
401 struct radeon_surface_reg
*reg
;
402 struct radeon_bo
*old_object
;
406 BUG_ON(!atomic_read(&bo
->tbo
.reserved
));
408 if (!bo
->tiling_flags
)
411 if (bo
->surface_reg
>= 0) {
412 reg
= &rdev
->surface_regs
[bo
->surface_reg
];
418 for (i
= 0; i
< RADEON_GEM_MAX_SURFACES
; i
++) {
420 reg
= &rdev
->surface_regs
[i
];
424 old_object
= reg
->bo
;
425 if (old_object
->pin_count
== 0)
429 /* if we are all out */
430 if (i
== RADEON_GEM_MAX_SURFACES
) {
433 /* find someone with a surface reg and nuke their BO */
434 reg
= &rdev
->surface_regs
[steal
];
435 old_object
= reg
->bo
;
436 /* blow away the mapping */
437 DRM_DEBUG("stealing surface reg %d from %p\n", steal
, old_object
);
438 ttm_bo_unmap_virtual(&old_object
->tbo
);
439 old_object
->surface_reg
= -1;
447 radeon_set_surface_reg(rdev
, i
, bo
->tiling_flags
, bo
->pitch
,
448 bo
->tbo
.mem
.start
<< PAGE_SHIFT
,
449 bo
->tbo
.num_pages
<< PAGE_SHIFT
);
453 static void radeon_bo_clear_surface_reg(struct radeon_bo
*bo
)
455 struct radeon_device
*rdev
= bo
->rdev
;
456 struct radeon_surface_reg
*reg
;
458 if (bo
->surface_reg
== -1)
461 reg
= &rdev
->surface_regs
[bo
->surface_reg
];
462 radeon_clear_surface_reg(rdev
, bo
->surface_reg
);
465 bo
->surface_reg
= -1;
468 int radeon_bo_set_tiling_flags(struct radeon_bo
*bo
,
469 uint32_t tiling_flags
, uint32_t pitch
)
471 struct radeon_device
*rdev
= bo
->rdev
;
474 if (rdev
->family
>= CHIP_CEDAR
) {
475 unsigned bankw
, bankh
, mtaspect
, tilesplit
, stilesplit
;
477 bankw
= (tiling_flags
>> RADEON_TILING_EG_BANKW_SHIFT
) & RADEON_TILING_EG_BANKW_MASK
;
478 bankh
= (tiling_flags
>> RADEON_TILING_EG_BANKH_SHIFT
) & RADEON_TILING_EG_BANKH_MASK
;
479 mtaspect
= (tiling_flags
>> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT
) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK
;
480 tilesplit
= (tiling_flags
>> RADEON_TILING_EG_TILE_SPLIT_SHIFT
) & RADEON_TILING_EG_TILE_SPLIT_MASK
;
481 stilesplit
= (tiling_flags
>> RADEON_TILING_EG_STENCIL_TILE_SPLIT_SHIFT
) & RADEON_TILING_EG_STENCIL_TILE_SPLIT_MASK
;
515 if (stilesplit
> 6) {
519 r
= radeon_bo_reserve(bo
, false);
520 if (unlikely(r
!= 0))
522 bo
->tiling_flags
= tiling_flags
;
524 radeon_bo_unreserve(bo
);
528 void radeon_bo_get_tiling_flags(struct radeon_bo
*bo
,
529 uint32_t *tiling_flags
,
532 BUG_ON(!atomic_read(&bo
->tbo
.reserved
));
534 *tiling_flags
= bo
->tiling_flags
;
539 int radeon_bo_check_tiling(struct radeon_bo
*bo
, bool has_moved
,
542 BUG_ON(!atomic_read(&bo
->tbo
.reserved
));
544 if (!(bo
->tiling_flags
& RADEON_TILING_SURFACE
))
548 radeon_bo_clear_surface_reg(bo
);
552 if (bo
->tbo
.mem
.mem_type
!= TTM_PL_VRAM
) {
556 if (bo
->surface_reg
>= 0)
557 radeon_bo_clear_surface_reg(bo
);
561 if ((bo
->surface_reg
>= 0) && !has_moved
)
564 return radeon_bo_get_surface_reg(bo
);
567 void radeon_bo_move_notify(struct ttm_buffer_object
*bo
,
568 struct ttm_mem_reg
*mem
)
570 struct radeon_bo
*rbo
;
571 if (!radeon_ttm_bo_is_radeon_bo(bo
))
573 rbo
= container_of(bo
, struct radeon_bo
, tbo
);
574 radeon_bo_check_tiling(rbo
, 0, 1);
575 radeon_vm_bo_invalidate(rbo
->rdev
, rbo
);
578 int radeon_bo_fault_reserve_notify(struct ttm_buffer_object
*bo
)
580 struct radeon_device
*rdev
;
581 struct radeon_bo
*rbo
;
582 unsigned long offset
, size
;
585 if (!radeon_ttm_bo_is_radeon_bo(bo
))
587 rbo
= container_of(bo
, struct radeon_bo
, tbo
);
588 radeon_bo_check_tiling(rbo
, 0, 0);
590 if (bo
->mem
.mem_type
== TTM_PL_VRAM
) {
591 size
= bo
->mem
.num_pages
<< PAGE_SHIFT
;
592 offset
= bo
->mem
.start
<< PAGE_SHIFT
;
593 if ((offset
+ size
) > rdev
->mc
.visible_vram_size
) {
594 /* hurrah the memory is not visible ! */
595 radeon_ttm_placement_from_domain(rbo
, RADEON_GEM_DOMAIN_VRAM
);
596 rbo
->placement
.lpfn
= rdev
->mc
.visible_vram_size
>> PAGE_SHIFT
;
597 r
= ttm_bo_validate(bo
, &rbo
->placement
, false, true, false);
598 if (unlikely(r
!= 0))
600 offset
= bo
->mem
.start
<< PAGE_SHIFT
;
601 /* this should not happen */
602 if ((offset
+ size
) > rdev
->mc
.visible_vram_size
)
609 int radeon_bo_wait(struct radeon_bo
*bo
, u32
*mem_type
, bool no_wait
)
613 r
= ttm_bo_reserve(&bo
->tbo
, true, no_wait
, false, 0);
614 if (unlikely(r
!= 0))
616 spin_lock(&bo
->tbo
.bdev
->fence_lock
);
618 *mem_type
= bo
->tbo
.mem
.mem_type
;
619 if (bo
->tbo
.sync_obj
)
620 r
= ttm_bo_wait(&bo
->tbo
, true, true, no_wait
);
621 spin_unlock(&bo
->tbo
.bdev
->fence_lock
);
622 ttm_bo_unreserve(&bo
->tbo
);
628 * radeon_bo_reserve - reserve bo
630 * @no_wait: don't sleep while trying to reserve (return -EBUSY)
633 * -EBUSY: buffer is busy and @no_wait is true
634 * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by
635 * a signal. Release all buffer reservations and return to user-space.
637 int radeon_bo_reserve(struct radeon_bo
*bo
, bool no_wait
)
641 r
= ttm_bo_reserve(&bo
->tbo
, true, no_wait
, false, 0);
642 if (unlikely(r
!= 0)) {
643 if (r
!= -ERESTARTSYS
)
644 dev_err(bo
->rdev
->dev
, "%p reserve failed\n", bo
);
650 /* object have to be reserved */
651 struct radeon_bo_va
*radeon_bo_va(struct radeon_bo
*rbo
, struct radeon_vm
*vm
)
653 struct radeon_bo_va
*bo_va
;
655 list_for_each_entry(bo_va
, &rbo
->va
, bo_list
) {
656 if (bo_va
->vm
== vm
) {