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 mutex_lock(&bo_va->vm->mutex);
56 list_del(&bo_va->vm_list);
57 mutex_unlock(&bo_va->vm->mutex);
58 list_del(&bo_va->bo_list);
63 static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo)
67 bo = container_of(tbo, struct radeon_bo, tbo);
68 mutex_lock(&bo->rdev->gem.mutex);
69 list_del_init(&bo->list);
70 mutex_unlock(&bo->rdev->gem.mutex);
71 radeon_bo_clear_surface_reg(bo);
72 radeon_bo_clear_va(bo);
73 drm_gem_object_release(&bo->gem_base);
77 bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo)
79 if (bo->destroy == &radeon_ttm_bo_destroy)
84 void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain)
88 rbo->placement.fpfn = 0;
89 rbo->placement.lpfn = 0;
90 rbo->placement.placement = rbo->placements;
91 rbo->placement.busy_placement = rbo->placements;
92 if (domain & RADEON_GEM_DOMAIN_VRAM)
93 rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
95 if (domain & RADEON_GEM_DOMAIN_GTT)
96 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
97 if (domain & RADEON_GEM_DOMAIN_CPU)
98 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
100 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
101 rbo->placement.num_placement = c;
102 rbo->placement.num_busy_placement = c;
105 int radeon_bo_create(struct radeon_device *rdev,
106 unsigned long size, int byte_align, bool kernel, u32 domain,
107 struct radeon_bo **bo_ptr)
109 struct radeon_bo *bo;
110 enum ttm_bo_type type;
111 unsigned long page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
112 unsigned long max_size = 0;
116 size = ALIGN(size, PAGE_SIZE);
118 if (unlikely(rdev->mman.bdev.dev_mapping == NULL)) {
119 rdev->mman.bdev.dev_mapping = rdev->ddev->dev_mapping;
122 type = ttm_bo_type_kernel;
124 type = ttm_bo_type_device;
128 /* maximun bo size is the minimun btw visible vram and gtt size */
129 max_size = min(rdev->mc.visible_vram_size, rdev->mc.gtt_size);
130 if ((page_align << PAGE_SHIFT) >= max_size) {
131 printk(KERN_WARNING "%s:%d alloc size %ldM bigger than %ldMb limit\n",
132 __func__, __LINE__, page_align >> (20 - PAGE_SHIFT), max_size >> 20);
136 acc_size = ttm_bo_dma_acc_size(&rdev->mman.bdev, size,
137 sizeof(struct radeon_bo));
140 bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL);
143 r = drm_gem_object_init(rdev->ddev, &bo->gem_base, size);
149 bo->gem_base.driver_private = NULL;
150 bo->surface_reg = -1;
151 INIT_LIST_HEAD(&bo->list);
152 INIT_LIST_HEAD(&bo->va);
153 radeon_ttm_placement_from_domain(bo, domain);
154 /* Kernel allocation are uninterruptible */
155 mutex_lock(&rdev->vram_mutex);
156 r = ttm_bo_init(&rdev->mman.bdev, &bo->tbo, size, type,
157 &bo->placement, page_align, 0, !kernel, NULL,
158 acc_size, &radeon_ttm_bo_destroy);
159 mutex_unlock(&rdev->vram_mutex);
160 if (unlikely(r != 0)) {
161 if (r != -ERESTARTSYS) {
162 if (domain == RADEON_GEM_DOMAIN_VRAM) {
163 domain |= RADEON_GEM_DOMAIN_GTT;
167 "object_init failed for (%lu, 0x%08X)\n",
174 trace_radeon_bo_create(bo);
179 int radeon_bo_kmap(struct radeon_bo *bo, void **ptr)
190 r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap);
194 bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
198 radeon_bo_check_tiling(bo, 0, 0);
202 void radeon_bo_kunmap(struct radeon_bo *bo)
204 if (bo->kptr == NULL)
207 radeon_bo_check_tiling(bo, 0, 0);
208 ttm_bo_kunmap(&bo->kmap);
211 void radeon_bo_unref(struct radeon_bo **bo)
213 struct ttm_buffer_object *tbo;
214 struct radeon_device *rdev;
220 mutex_lock(&rdev->vram_mutex);
222 mutex_unlock(&rdev->vram_mutex);
227 int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset,
235 *gpu_addr = radeon_bo_gpu_offset(bo);
237 if (max_offset != 0) {
240 if (domain == RADEON_GEM_DOMAIN_VRAM)
241 domain_start = bo->rdev->mc.vram_start;
243 domain_start = bo->rdev->mc.gtt_start;
244 WARN_ON_ONCE((*gpu_addr - domain_start) > max_offset);
249 radeon_ttm_placement_from_domain(bo, domain);
250 if (domain == RADEON_GEM_DOMAIN_VRAM) {
251 /* force to pin into visible video ram */
252 bo->placement.lpfn = bo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
255 u64 lpfn = max_offset >> PAGE_SHIFT;
257 if (!bo->placement.lpfn)
258 bo->placement.lpfn = bo->rdev->mc.gtt_size >> PAGE_SHIFT;
260 if (lpfn < bo->placement.lpfn)
261 bo->placement.lpfn = lpfn;
263 for (i = 0; i < bo->placement.num_placement; i++)
264 bo->placements[i] |= TTM_PL_FLAG_NO_EVICT;
265 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
266 if (likely(r == 0)) {
268 if (gpu_addr != NULL)
269 *gpu_addr = radeon_bo_gpu_offset(bo);
271 if (unlikely(r != 0))
272 dev_err(bo->rdev->dev, "%p pin failed\n", bo);
276 int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
278 return radeon_bo_pin_restricted(bo, domain, 0, gpu_addr);
281 int radeon_bo_unpin(struct radeon_bo *bo)
285 if (!bo->pin_count) {
286 dev_warn(bo->rdev->dev, "%p unpin not necessary\n", bo);
292 for (i = 0; i < bo->placement.num_placement; i++)
293 bo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT;
294 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
295 if (unlikely(r != 0))
296 dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo);
300 int radeon_bo_evict_vram(struct radeon_device *rdev)
302 /* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
303 if (0 && (rdev->flags & RADEON_IS_IGP)) {
304 if (rdev->mc.igp_sideport_enabled == false)
305 /* Useless to evict on IGP chips */
308 return ttm_bo_evict_mm(&rdev->mman.bdev, TTM_PL_VRAM);
311 void radeon_bo_force_delete(struct radeon_device *rdev)
313 struct radeon_bo *bo, *n;
315 if (list_empty(&rdev->gem.objects)) {
318 dev_err(rdev->dev, "Userspace still has active objects !\n");
319 list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) {
320 mutex_lock(&rdev->ddev->struct_mutex);
321 dev_err(rdev->dev, "%p %p %lu %lu force free\n",
322 &bo->gem_base, bo, (unsigned long)bo->gem_base.size,
323 *((unsigned long *)&bo->gem_base.refcount));
324 mutex_lock(&bo->rdev->gem.mutex);
325 list_del_init(&bo->list);
326 mutex_unlock(&bo->rdev->gem.mutex);
327 /* this should unref the ttm bo */
328 drm_gem_object_unreference(&bo->gem_base);
329 mutex_unlock(&rdev->ddev->struct_mutex);
333 int radeon_bo_init(struct radeon_device *rdev)
335 /* Add an MTRR for the VRAM */
336 rdev->mc.vram_mtrr = mtrr_add(rdev->mc.aper_base, rdev->mc.aper_size,
337 MTRR_TYPE_WRCOMB, 1);
338 DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
339 rdev->mc.mc_vram_size >> 20,
340 (unsigned long long)rdev->mc.aper_size >> 20);
341 DRM_INFO("RAM width %dbits %cDR\n",
342 rdev->mc.vram_width, rdev->mc.vram_is_ddr ? 'D' : 'S');
343 return radeon_ttm_init(rdev);
346 void radeon_bo_fini(struct radeon_device *rdev)
348 radeon_ttm_fini(rdev);
351 void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
352 struct list_head *head)
355 list_add(&lobj->tv.head, head);
357 list_add_tail(&lobj->tv.head, head);
361 int radeon_bo_list_validate(struct list_head *head)
363 struct radeon_bo_list *lobj;
364 struct radeon_bo *bo;
368 r = ttm_eu_reserve_buffers(head);
369 if (unlikely(r != 0)) {
372 list_for_each_entry(lobj, head, tv.head) {
374 if (!bo->pin_count) {
375 domain = lobj->wdomain ? lobj->wdomain : lobj->rdomain;
378 radeon_ttm_placement_from_domain(bo, domain);
379 r = ttm_bo_validate(&bo->tbo, &bo->placement,
382 if (r != -ERESTARTSYS && domain == RADEON_GEM_DOMAIN_VRAM) {
383 domain |= RADEON_GEM_DOMAIN_GTT;
389 lobj->gpu_offset = radeon_bo_gpu_offset(bo);
390 lobj->tiling_flags = bo->tiling_flags;
395 int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
396 struct vm_area_struct *vma)
398 return ttm_fbdev_mmap(vma, &bo->tbo);
401 int radeon_bo_get_surface_reg(struct radeon_bo *bo)
403 struct radeon_device *rdev = bo->rdev;
404 struct radeon_surface_reg *reg;
405 struct radeon_bo *old_object;
409 BUG_ON(!atomic_read(&bo->tbo.reserved));
411 if (!bo->tiling_flags)
414 if (bo->surface_reg >= 0) {
415 reg = &rdev->surface_regs[bo->surface_reg];
421 for (i = 0; i < RADEON_GEM_MAX_SURFACES; i++) {
423 reg = &rdev->surface_regs[i];
427 old_object = reg->bo;
428 if (old_object->pin_count == 0)
432 /* if we are all out */
433 if (i == RADEON_GEM_MAX_SURFACES) {
436 /* find someone with a surface reg and nuke their BO */
437 reg = &rdev->surface_regs[steal];
438 old_object = reg->bo;
439 /* blow away the mapping */
440 DRM_DEBUG("stealing surface reg %d from %p\n", steal, old_object);
441 ttm_bo_unmap_virtual(&old_object->tbo);
442 old_object->surface_reg = -1;
450 radeon_set_surface_reg(rdev, i, bo->tiling_flags, bo->pitch,
451 bo->tbo.mem.start << PAGE_SHIFT,
452 bo->tbo.num_pages << PAGE_SHIFT);
456 static void radeon_bo_clear_surface_reg(struct radeon_bo *bo)
458 struct radeon_device *rdev = bo->rdev;
459 struct radeon_surface_reg *reg;
461 if (bo->surface_reg == -1)
464 reg = &rdev->surface_regs[bo->surface_reg];
465 radeon_clear_surface_reg(rdev, bo->surface_reg);
468 bo->surface_reg = -1;
471 int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
472 uint32_t tiling_flags, uint32_t pitch)
474 struct radeon_device *rdev = bo->rdev;
477 if (rdev->family >= CHIP_CEDAR) {
478 unsigned bankw, bankh, mtaspect, tilesplit, stilesplit;
480 bankw = (tiling_flags >> RADEON_TILING_EG_BANKW_SHIFT) & RADEON_TILING_EG_BANKW_MASK;
481 bankh = (tiling_flags >> RADEON_TILING_EG_BANKH_SHIFT) & RADEON_TILING_EG_BANKH_MASK;
482 mtaspect = (tiling_flags >> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK;
483 tilesplit = (tiling_flags >> RADEON_TILING_EG_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_TILE_SPLIT_MASK;
484 stilesplit = (tiling_flags >> RADEON_TILING_EG_STENCIL_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_STENCIL_TILE_SPLIT_MASK;
518 if (stilesplit > 6) {
522 r = radeon_bo_reserve(bo, false);
523 if (unlikely(r != 0))
525 bo->tiling_flags = tiling_flags;
527 radeon_bo_unreserve(bo);
531 void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
532 uint32_t *tiling_flags,
535 BUG_ON(!atomic_read(&bo->tbo.reserved));
537 *tiling_flags = bo->tiling_flags;
542 int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
545 BUG_ON(!atomic_read(&bo->tbo.reserved));
547 if (!(bo->tiling_flags & RADEON_TILING_SURFACE))
551 radeon_bo_clear_surface_reg(bo);
555 if (bo->tbo.mem.mem_type != TTM_PL_VRAM) {
559 if (bo->surface_reg >= 0)
560 radeon_bo_clear_surface_reg(bo);
564 if ((bo->surface_reg >= 0) && !has_moved)
567 return radeon_bo_get_surface_reg(bo);
570 void radeon_bo_move_notify(struct ttm_buffer_object *bo,
571 struct ttm_mem_reg *mem)
573 struct radeon_bo *rbo;
574 if (!radeon_ttm_bo_is_radeon_bo(bo))
576 rbo = container_of(bo, struct radeon_bo, tbo);
577 radeon_bo_check_tiling(rbo, 0, 1);
578 radeon_vm_bo_invalidate(rbo->rdev, rbo);
581 int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
583 struct radeon_device *rdev;
584 struct radeon_bo *rbo;
585 unsigned long offset, size;
588 if (!radeon_ttm_bo_is_radeon_bo(bo))
590 rbo = container_of(bo, struct radeon_bo, tbo);
591 radeon_bo_check_tiling(rbo, 0, 0);
593 if (bo->mem.mem_type == TTM_PL_VRAM) {
594 size = bo->mem.num_pages << PAGE_SHIFT;
595 offset = bo->mem.start << PAGE_SHIFT;
596 if ((offset + size) > rdev->mc.visible_vram_size) {
597 /* hurrah the memory is not visible ! */
598 radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
599 rbo->placement.lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
600 r = ttm_bo_validate(bo, &rbo->placement, false, true, false);
601 if (unlikely(r != 0))
603 offset = bo->mem.start << PAGE_SHIFT;
604 /* this should not happen */
605 if ((offset + size) > rdev->mc.visible_vram_size)
612 int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
616 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
617 if (unlikely(r != 0))
619 spin_lock(&bo->tbo.bdev->fence_lock);
621 *mem_type = bo->tbo.mem.mem_type;
622 if (bo->tbo.sync_obj)
623 r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
624 spin_unlock(&bo->tbo.bdev->fence_lock);
625 ttm_bo_unreserve(&bo->tbo);
631 * radeon_bo_reserve - reserve bo
633 * @no_wait: don't sleep while trying to reserve (return -EBUSY)
636 * -EBUSY: buffer is busy and @no_wait is true
637 * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by
638 * a signal. Release all buffer reservations and return to user-space.
640 int radeon_bo_reserve(struct radeon_bo *bo, bool no_wait)
644 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
645 if (unlikely(r != 0)) {
646 if (r != -ERESTARTSYS)
647 dev_err(bo->rdev->dev, "%p reserve failed\n", bo);
653 /* object have to be reserved */
654 struct radeon_bo_va *radeon_bo_va(struct radeon_bo *rbo, struct radeon_vm *vm)
656 struct radeon_bo_va *bo_va;
658 list_for_each_entry(bo_va, &rbo->va, bo_list) {
659 if (bo_va->vm == vm) {