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 sg_table *sg, 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 rdev->mman.bdev.dev_mapping = rdev->ddev->dev_mapping;
120 type = ttm_bo_type_kernel;
122 type = ttm_bo_type_sg;
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));
139 bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL);
142 r = drm_gem_object_init(rdev->ddev, &bo->gem_base, size);
148 bo->gem_base.driver_private = NULL;
149 bo->surface_reg = -1;
150 INIT_LIST_HEAD(&bo->list);
151 INIT_LIST_HEAD(&bo->va);
154 radeon_ttm_placement_from_domain(bo, domain);
155 /* Kernel allocation are uninterruptible */
156 down_read(&rdev->pm.mclk_lock);
157 r = ttm_bo_init(&rdev->mman.bdev, &bo->tbo, size, type,
158 &bo->placement, page_align, 0, !kernel, NULL,
159 acc_size, sg, &radeon_ttm_bo_destroy);
160 up_read(&rdev->pm.mclk_lock);
161 if (unlikely(r != 0)) {
162 if (r != -ERESTARTSYS) {
163 if (domain == RADEON_GEM_DOMAIN_VRAM) {
164 domain |= RADEON_GEM_DOMAIN_GTT;
168 "object_init failed for (%lu, 0x%08X)\n",
175 trace_radeon_bo_create(bo);
180 int radeon_bo_kmap(struct radeon_bo *bo, void **ptr)
191 r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap);
195 bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
199 radeon_bo_check_tiling(bo, 0, 0);
203 void radeon_bo_kunmap(struct radeon_bo *bo)
205 if (bo->kptr == NULL)
208 radeon_bo_check_tiling(bo, 0, 0);
209 ttm_bo_kunmap(&bo->kmap);
212 void radeon_bo_unref(struct radeon_bo **bo)
214 struct ttm_buffer_object *tbo;
215 struct radeon_device *rdev;
221 down_read(&rdev->pm.mclk_lock);
223 up_read(&rdev->pm.mclk_lock);
228 int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset,
236 *gpu_addr = radeon_bo_gpu_offset(bo);
238 if (max_offset != 0) {
241 if (domain == RADEON_GEM_DOMAIN_VRAM)
242 domain_start = bo->rdev->mc.vram_start;
244 domain_start = bo->rdev->mc.gtt_start;
245 WARN_ON_ONCE(max_offset <
246 (radeon_bo_gpu_offset(bo) - domain_start));
251 radeon_ttm_placement_from_domain(bo, domain);
252 if (domain == RADEON_GEM_DOMAIN_VRAM) {
253 /* force to pin into visible video ram */
254 bo->placement.lpfn = bo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
257 u64 lpfn = max_offset >> PAGE_SHIFT;
259 if (!bo->placement.lpfn)
260 bo->placement.lpfn = bo->rdev->mc.gtt_size >> PAGE_SHIFT;
262 if (lpfn < bo->placement.lpfn)
263 bo->placement.lpfn = lpfn;
265 for (i = 0; i < bo->placement.num_placement; i++)
266 bo->placements[i] |= TTM_PL_FLAG_NO_EVICT;
267 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
268 if (likely(r == 0)) {
270 if (gpu_addr != NULL)
271 *gpu_addr = radeon_bo_gpu_offset(bo);
273 if (unlikely(r != 0))
274 dev_err(bo->rdev->dev, "%p pin failed\n", bo);
278 int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
280 return radeon_bo_pin_restricted(bo, domain, 0, gpu_addr);
283 int radeon_bo_unpin(struct radeon_bo *bo)
287 if (!bo->pin_count) {
288 dev_warn(bo->rdev->dev, "%p unpin not necessary\n", bo);
294 for (i = 0; i < bo->placement.num_placement; i++)
295 bo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT;
296 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
297 if (unlikely(r != 0))
298 dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo);
302 int radeon_bo_evict_vram(struct radeon_device *rdev)
304 /* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
305 if (0 && (rdev->flags & RADEON_IS_IGP)) {
306 if (rdev->mc.igp_sideport_enabled == false)
307 /* Useless to evict on IGP chips */
310 return ttm_bo_evict_mm(&rdev->mman.bdev, TTM_PL_VRAM);
313 void radeon_bo_force_delete(struct radeon_device *rdev)
315 struct radeon_bo *bo, *n;
317 if (list_empty(&rdev->gem.objects)) {
320 dev_err(rdev->dev, "Userspace still has active objects !\n");
321 list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) {
322 mutex_lock(&rdev->ddev->struct_mutex);
323 dev_err(rdev->dev, "%p %p %lu %lu force free\n",
324 &bo->gem_base, bo, (unsigned long)bo->gem_base.size,
325 *((unsigned long *)&bo->gem_base.refcount));
326 mutex_lock(&bo->rdev->gem.mutex);
327 list_del_init(&bo->list);
328 mutex_unlock(&bo->rdev->gem.mutex);
329 /* this should unref the ttm bo */
330 drm_gem_object_unreference(&bo->gem_base);
331 mutex_unlock(&rdev->ddev->struct_mutex);
335 int radeon_bo_init(struct radeon_device *rdev)
337 /* Add an MTRR for the VRAM */
338 rdev->mc.vram_mtrr = mtrr_add(rdev->mc.aper_base, rdev->mc.aper_size,
339 MTRR_TYPE_WRCOMB, 1);
340 DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
341 rdev->mc.mc_vram_size >> 20,
342 (unsigned long long)rdev->mc.aper_size >> 20);
343 DRM_INFO("RAM width %dbits %cDR\n",
344 rdev->mc.vram_width, rdev->mc.vram_is_ddr ? 'D' : 'S');
345 return radeon_ttm_init(rdev);
348 void radeon_bo_fini(struct radeon_device *rdev)
350 radeon_ttm_fini(rdev);
353 void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
354 struct list_head *head)
357 list_add(&lobj->tv.head, head);
359 list_add_tail(&lobj->tv.head, head);
363 int radeon_bo_list_validate(struct list_head *head)
365 struct radeon_bo_list *lobj;
366 struct radeon_bo *bo;
370 r = ttm_eu_reserve_buffers(head);
371 if (unlikely(r != 0)) {
374 list_for_each_entry(lobj, head, tv.head) {
376 if (!bo->pin_count) {
377 domain = lobj->wdomain ? lobj->wdomain : lobj->rdomain;
380 radeon_ttm_placement_from_domain(bo, domain);
381 r = ttm_bo_validate(&bo->tbo, &bo->placement,
384 if (r != -ERESTARTSYS && domain == RADEON_GEM_DOMAIN_VRAM) {
385 domain |= RADEON_GEM_DOMAIN_GTT;
391 lobj->gpu_offset = radeon_bo_gpu_offset(bo);
392 lobj->tiling_flags = bo->tiling_flags;
397 int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
398 struct vm_area_struct *vma)
400 return ttm_fbdev_mmap(vma, &bo->tbo);
403 int radeon_bo_get_surface_reg(struct radeon_bo *bo)
405 struct radeon_device *rdev = bo->rdev;
406 struct radeon_surface_reg *reg;
407 struct radeon_bo *old_object;
411 BUG_ON(!atomic_read(&bo->tbo.reserved));
413 if (!bo->tiling_flags)
416 if (bo->surface_reg >= 0) {
417 reg = &rdev->surface_regs[bo->surface_reg];
423 for (i = 0; i < RADEON_GEM_MAX_SURFACES; i++) {
425 reg = &rdev->surface_regs[i];
429 old_object = reg->bo;
430 if (old_object->pin_count == 0)
434 /* if we are all out */
435 if (i == RADEON_GEM_MAX_SURFACES) {
438 /* find someone with a surface reg and nuke their BO */
439 reg = &rdev->surface_regs[steal];
440 old_object = reg->bo;
441 /* blow away the mapping */
442 DRM_DEBUG("stealing surface reg %d from %p\n", steal, old_object);
443 ttm_bo_unmap_virtual(&old_object->tbo);
444 old_object->surface_reg = -1;
452 radeon_set_surface_reg(rdev, i, bo->tiling_flags, bo->pitch,
453 bo->tbo.mem.start << PAGE_SHIFT,
454 bo->tbo.num_pages << PAGE_SHIFT);
458 static void radeon_bo_clear_surface_reg(struct radeon_bo *bo)
460 struct radeon_device *rdev = bo->rdev;
461 struct radeon_surface_reg *reg;
463 if (bo->surface_reg == -1)
466 reg = &rdev->surface_regs[bo->surface_reg];
467 radeon_clear_surface_reg(rdev, bo->surface_reg);
470 bo->surface_reg = -1;
473 int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
474 uint32_t tiling_flags, uint32_t pitch)
476 struct radeon_device *rdev = bo->rdev;
479 if (rdev->family >= CHIP_CEDAR) {
480 unsigned bankw, bankh, mtaspect, tilesplit, stilesplit;
482 bankw = (tiling_flags >> RADEON_TILING_EG_BANKW_SHIFT) & RADEON_TILING_EG_BANKW_MASK;
483 bankh = (tiling_flags >> RADEON_TILING_EG_BANKH_SHIFT) & RADEON_TILING_EG_BANKH_MASK;
484 mtaspect = (tiling_flags >> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK;
485 tilesplit = (tiling_flags >> RADEON_TILING_EG_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_TILE_SPLIT_MASK;
486 stilesplit = (tiling_flags >> RADEON_TILING_EG_STENCIL_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_STENCIL_TILE_SPLIT_MASK;
520 if (stilesplit > 6) {
524 r = radeon_bo_reserve(bo, false);
525 if (unlikely(r != 0))
527 bo->tiling_flags = tiling_flags;
529 radeon_bo_unreserve(bo);
533 void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
534 uint32_t *tiling_flags,
537 BUG_ON(!atomic_read(&bo->tbo.reserved));
539 *tiling_flags = bo->tiling_flags;
544 int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
547 BUG_ON(!atomic_read(&bo->tbo.reserved));
549 if (!(bo->tiling_flags & RADEON_TILING_SURFACE))
553 radeon_bo_clear_surface_reg(bo);
557 if (bo->tbo.mem.mem_type != TTM_PL_VRAM) {
561 if (bo->surface_reg >= 0)
562 radeon_bo_clear_surface_reg(bo);
566 if ((bo->surface_reg >= 0) && !has_moved)
569 return radeon_bo_get_surface_reg(bo);
572 void radeon_bo_move_notify(struct ttm_buffer_object *bo,
573 struct ttm_mem_reg *mem)
575 struct radeon_bo *rbo;
576 if (!radeon_ttm_bo_is_radeon_bo(bo))
578 rbo = container_of(bo, struct radeon_bo, tbo);
579 radeon_bo_check_tiling(rbo, 0, 1);
580 radeon_vm_bo_invalidate(rbo->rdev, rbo);
583 int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
585 struct radeon_device *rdev;
586 struct radeon_bo *rbo;
587 unsigned long offset, size;
590 if (!radeon_ttm_bo_is_radeon_bo(bo))
592 rbo = container_of(bo, struct radeon_bo, tbo);
593 radeon_bo_check_tiling(rbo, 0, 0);
595 if (bo->mem.mem_type == TTM_PL_VRAM) {
596 size = bo->mem.num_pages << PAGE_SHIFT;
597 offset = bo->mem.start << PAGE_SHIFT;
598 if ((offset + size) > rdev->mc.visible_vram_size) {
599 /* hurrah the memory is not visible ! */
600 radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
601 rbo->placement.lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
602 r = ttm_bo_validate(bo, &rbo->placement, false, true, false);
603 if (unlikely(r != 0))
605 offset = bo->mem.start << PAGE_SHIFT;
606 /* this should not happen */
607 if ((offset + size) > rdev->mc.visible_vram_size)
614 int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
618 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
619 if (unlikely(r != 0))
621 spin_lock(&bo->tbo.bdev->fence_lock);
623 *mem_type = bo->tbo.mem.mem_type;
624 if (bo->tbo.sync_obj)
625 r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
626 spin_unlock(&bo->tbo.bdev->fence_lock);
627 ttm_bo_unreserve(&bo->tbo);
633 * radeon_bo_reserve - reserve bo
635 * @no_wait: don't sleep while trying to reserve (return -EBUSY)
638 * -EBUSY: buffer is busy and @no_wait is true
639 * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by
640 * a signal. Release all buffer reservations and return to user-space.
642 int radeon_bo_reserve(struct radeon_bo *bo, bool no_wait)
646 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
647 if (unlikely(r != 0)) {
648 if (r != -ERESTARTSYS)
649 dev_err(bo->rdev->dev, "%p reserve failed\n", bo);
655 /* object have to be reserved */
656 struct radeon_bo_va *radeon_bo_va(struct radeon_bo *rbo, struct radeon_vm *vm)
658 struct radeon_bo_va *bo_va;
660 list_for_each_entry(bo_va, &rbo->va, bo_list) {
661 if (bo_va->vm == vm) {