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 static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo)
53 bo = container_of(tbo, struct radeon_bo, tbo);
54 mutex_lock(&bo->rdev->gem.mutex);
55 list_del_init(&bo->list);
56 mutex_unlock(&bo->rdev->gem.mutex);
57 radeon_bo_clear_surface_reg(bo);
58 drm_gem_object_release(&bo->gem_base);
62 bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo)
64 if (bo->destroy == &radeon_ttm_bo_destroy)
69 void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain)
73 rbo->placement.fpfn = 0;
74 rbo->placement.lpfn = 0;
75 rbo->placement.placement = rbo->placements;
76 rbo->placement.busy_placement = rbo->placements;
77 if (domain & RADEON_GEM_DOMAIN_VRAM)
78 rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
80 if (domain & RADEON_GEM_DOMAIN_GTT)
81 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
82 if (domain & RADEON_GEM_DOMAIN_CPU)
83 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
85 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
86 rbo->placement.num_placement = c;
87 rbo->placement.num_busy_placement = c;
90 int radeon_bo_create(struct radeon_device *rdev,
91 unsigned long size, int byte_align, bool kernel, u32 domain,
92 struct radeon_bo **bo_ptr)
95 enum ttm_bo_type type;
96 unsigned long page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
97 unsigned long max_size = 0;
100 size = ALIGN(size, PAGE_SIZE);
102 if (unlikely(rdev->mman.bdev.dev_mapping == NULL)) {
103 rdev->mman.bdev.dev_mapping = rdev->ddev->dev_mapping;
106 type = ttm_bo_type_kernel;
108 type = ttm_bo_type_device;
112 /* maximun bo size is the minimun btw visible vram and gtt size */
113 max_size = min(rdev->mc.visible_vram_size, rdev->mc.gtt_size);
114 if ((page_align << PAGE_SHIFT) >= max_size) {
115 printk(KERN_WARNING "%s:%d alloc size %ldM bigger than %ldMb limit\n",
116 __func__, __LINE__, page_align >> (20 - PAGE_SHIFT), max_size >> 20);
121 bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL);
124 r = drm_gem_object_init(rdev->ddev, &bo->gem_base, size);
130 bo->gem_base.driver_private = NULL;
131 bo->surface_reg = -1;
132 INIT_LIST_HEAD(&bo->list);
133 radeon_ttm_placement_from_domain(bo, domain);
134 /* Kernel allocation are uninterruptible */
135 mutex_lock(&rdev->vram_mutex);
136 r = ttm_bo_init(&rdev->mman.bdev, &bo->tbo, size, type,
137 &bo->placement, page_align, 0, !kernel, NULL, size,
138 &radeon_ttm_bo_destroy);
139 mutex_unlock(&rdev->vram_mutex);
140 if (unlikely(r != 0)) {
141 if (r != -ERESTARTSYS) {
142 if (domain == RADEON_GEM_DOMAIN_VRAM) {
143 domain |= RADEON_GEM_DOMAIN_GTT;
147 "object_init failed for (%lu, 0x%08X)\n",
154 trace_radeon_bo_create(bo);
159 int radeon_bo_kmap(struct radeon_bo *bo, void **ptr)
170 r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap);
174 bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
178 radeon_bo_check_tiling(bo, 0, 0);
182 void radeon_bo_kunmap(struct radeon_bo *bo)
184 if (bo->kptr == NULL)
187 radeon_bo_check_tiling(bo, 0, 0);
188 ttm_bo_kunmap(&bo->kmap);
191 void radeon_bo_unref(struct radeon_bo **bo)
193 struct ttm_buffer_object *tbo;
194 struct radeon_device *rdev;
200 mutex_lock(&rdev->vram_mutex);
202 mutex_unlock(&rdev->vram_mutex);
207 int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset,
215 *gpu_addr = radeon_bo_gpu_offset(bo);
216 WARN_ON_ONCE(max_offset != 0);
219 radeon_ttm_placement_from_domain(bo, domain);
220 if (domain == RADEON_GEM_DOMAIN_VRAM) {
221 /* force to pin into visible video ram */
222 bo->placement.lpfn = bo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
225 u64 lpfn = max_offset >> PAGE_SHIFT;
227 if (!bo->placement.lpfn)
228 bo->placement.lpfn = bo->rdev->mc.gtt_size >> PAGE_SHIFT;
230 if (lpfn < bo->placement.lpfn)
231 bo->placement.lpfn = lpfn;
233 for (i = 0; i < bo->placement.num_placement; i++)
234 bo->placements[i] |= TTM_PL_FLAG_NO_EVICT;
235 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
236 if (likely(r == 0)) {
238 if (gpu_addr != NULL)
239 *gpu_addr = radeon_bo_gpu_offset(bo);
241 if (unlikely(r != 0))
242 dev_err(bo->rdev->dev, "%p pin failed\n", bo);
246 int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
248 return radeon_bo_pin_restricted(bo, domain, 0, gpu_addr);
251 int radeon_bo_unpin(struct radeon_bo *bo)
255 if (!bo->pin_count) {
256 dev_warn(bo->rdev->dev, "%p unpin not necessary\n", bo);
262 for (i = 0; i < bo->placement.num_placement; i++)
263 bo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT;
264 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
265 if (unlikely(r != 0))
266 dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo);
270 int radeon_bo_evict_vram(struct radeon_device *rdev)
272 /* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
273 if (0 && (rdev->flags & RADEON_IS_IGP)) {
274 if (rdev->mc.igp_sideport_enabled == false)
275 /* Useless to evict on IGP chips */
278 return ttm_bo_evict_mm(&rdev->mman.bdev, TTM_PL_VRAM);
281 void radeon_bo_force_delete(struct radeon_device *rdev)
283 struct radeon_bo *bo, *n;
285 if (list_empty(&rdev->gem.objects)) {
288 dev_err(rdev->dev, "Userspace still has active objects !\n");
289 list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) {
290 mutex_lock(&rdev->ddev->struct_mutex);
291 dev_err(rdev->dev, "%p %p %lu %lu force free\n",
292 &bo->gem_base, bo, (unsigned long)bo->gem_base.size,
293 *((unsigned long *)&bo->gem_base.refcount));
294 mutex_lock(&bo->rdev->gem.mutex);
295 list_del_init(&bo->list);
296 mutex_unlock(&bo->rdev->gem.mutex);
297 /* this should unref the ttm bo */
298 drm_gem_object_unreference(&bo->gem_base);
299 mutex_unlock(&rdev->ddev->struct_mutex);
303 int radeon_bo_init(struct radeon_device *rdev)
305 /* Add an MTRR for the VRAM */
306 rdev->mc.vram_mtrr = mtrr_add(rdev->mc.aper_base, rdev->mc.aper_size,
307 MTRR_TYPE_WRCOMB, 1);
308 DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
309 rdev->mc.mc_vram_size >> 20,
310 (unsigned long long)rdev->mc.aper_size >> 20);
311 DRM_INFO("RAM width %dbits %cDR\n",
312 rdev->mc.vram_width, rdev->mc.vram_is_ddr ? 'D' : 'S');
313 return radeon_ttm_init(rdev);
316 void radeon_bo_fini(struct radeon_device *rdev)
318 radeon_ttm_fini(rdev);
321 void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
322 struct list_head *head)
325 list_add(&lobj->tv.head, head);
327 list_add_tail(&lobj->tv.head, head);
331 int radeon_bo_list_validate(struct list_head *head)
333 struct radeon_bo_list *lobj;
334 struct radeon_bo *bo;
338 r = ttm_eu_reserve_buffers(head);
339 if (unlikely(r != 0)) {
342 list_for_each_entry(lobj, head, tv.head) {
344 if (!bo->pin_count) {
345 domain = lobj->wdomain ? lobj->wdomain : lobj->rdomain;
348 radeon_ttm_placement_from_domain(bo, domain);
349 r = ttm_bo_validate(&bo->tbo, &bo->placement,
352 if (r != -ERESTARTSYS && domain == RADEON_GEM_DOMAIN_VRAM) {
353 domain |= RADEON_GEM_DOMAIN_GTT;
359 lobj->gpu_offset = radeon_bo_gpu_offset(bo);
360 lobj->tiling_flags = bo->tiling_flags;
365 int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
366 struct vm_area_struct *vma)
368 return ttm_fbdev_mmap(vma, &bo->tbo);
371 int radeon_bo_get_surface_reg(struct radeon_bo *bo)
373 struct radeon_device *rdev = bo->rdev;
374 struct radeon_surface_reg *reg;
375 struct radeon_bo *old_object;
379 BUG_ON(!atomic_read(&bo->tbo.reserved));
381 if (!bo->tiling_flags)
384 if (bo->surface_reg >= 0) {
385 reg = &rdev->surface_regs[bo->surface_reg];
391 for (i = 0; i < RADEON_GEM_MAX_SURFACES; i++) {
393 reg = &rdev->surface_regs[i];
397 old_object = reg->bo;
398 if (old_object->pin_count == 0)
402 /* if we are all out */
403 if (i == RADEON_GEM_MAX_SURFACES) {
406 /* find someone with a surface reg and nuke their BO */
407 reg = &rdev->surface_regs[steal];
408 old_object = reg->bo;
409 /* blow away the mapping */
410 DRM_DEBUG("stealing surface reg %d from %p\n", steal, old_object);
411 ttm_bo_unmap_virtual(&old_object->tbo);
412 old_object->surface_reg = -1;
420 radeon_set_surface_reg(rdev, i, bo->tiling_flags, bo->pitch,
421 bo->tbo.mem.start << PAGE_SHIFT,
422 bo->tbo.num_pages << PAGE_SHIFT);
426 static void radeon_bo_clear_surface_reg(struct radeon_bo *bo)
428 struct radeon_device *rdev = bo->rdev;
429 struct radeon_surface_reg *reg;
431 if (bo->surface_reg == -1)
434 reg = &rdev->surface_regs[bo->surface_reg];
435 radeon_clear_surface_reg(rdev, bo->surface_reg);
438 bo->surface_reg = -1;
441 int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
442 uint32_t tiling_flags, uint32_t pitch)
446 r = radeon_bo_reserve(bo, false);
447 if (unlikely(r != 0))
449 bo->tiling_flags = tiling_flags;
451 radeon_bo_unreserve(bo);
455 void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
456 uint32_t *tiling_flags,
459 BUG_ON(!atomic_read(&bo->tbo.reserved));
461 *tiling_flags = bo->tiling_flags;
466 int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
469 BUG_ON(!atomic_read(&bo->tbo.reserved));
471 if (!(bo->tiling_flags & RADEON_TILING_SURFACE))
475 radeon_bo_clear_surface_reg(bo);
479 if (bo->tbo.mem.mem_type != TTM_PL_VRAM) {
483 if (bo->surface_reg >= 0)
484 radeon_bo_clear_surface_reg(bo);
488 if ((bo->surface_reg >= 0) && !has_moved)
491 return radeon_bo_get_surface_reg(bo);
494 void radeon_bo_move_notify(struct ttm_buffer_object *bo,
495 struct ttm_mem_reg *mem)
497 struct radeon_bo *rbo;
498 if (!radeon_ttm_bo_is_radeon_bo(bo))
500 rbo = container_of(bo, struct radeon_bo, tbo);
501 radeon_bo_check_tiling(rbo, 0, 1);
504 int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
506 struct radeon_device *rdev;
507 struct radeon_bo *rbo;
508 unsigned long offset, size;
511 if (!radeon_ttm_bo_is_radeon_bo(bo))
513 rbo = container_of(bo, struct radeon_bo, tbo);
514 radeon_bo_check_tiling(rbo, 0, 0);
516 if (bo->mem.mem_type == TTM_PL_VRAM) {
517 size = bo->mem.num_pages << PAGE_SHIFT;
518 offset = bo->mem.start << PAGE_SHIFT;
519 if ((offset + size) > rdev->mc.visible_vram_size) {
520 /* hurrah the memory is not visible ! */
521 radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
522 rbo->placement.lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
523 r = ttm_bo_validate(bo, &rbo->placement, false, true, false);
524 if (unlikely(r != 0))
526 offset = bo->mem.start << PAGE_SHIFT;
527 /* this should not happen */
528 if ((offset + size) > rdev->mc.visible_vram_size)
535 int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
539 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
540 if (unlikely(r != 0))
542 spin_lock(&bo->tbo.bdev->fence_lock);
544 *mem_type = bo->tbo.mem.mem_type;
545 if (bo->tbo.sync_obj)
546 r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
547 spin_unlock(&bo->tbo.bdev->fence_lock);
548 ttm_bo_unreserve(&bo->tbo);
554 * radeon_bo_reserve - reserve bo
556 * @no_wait: don't sleep while trying to reserve (return -EBUSY)
559 * -EBUSY: buffer is busy and @no_wait is true
560 * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by
561 * a signal. Release all buffer reservations and return to user-space.
563 int radeon_bo_reserve(struct radeon_bo *bo, bool no_wait)
567 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
568 if (unlikely(r != 0)) {
569 if (r != -ERESTARTSYS)
570 dev_err(bo->rdev->dev, "%p reserve failed\n", bo);