percpu, x86: don't use PMD_SIZE as embedded atom_size on 32bit
[zen-stable.git] / drivers / gpu / drm / radeon / radeon_object.c
blobfd2dc0e1e3f2587719d6f8c6bd62201ddd0e36ca
1 /*
2 * Copyright 2009 Jerome Glisse.
3 * All Rights Reserved.
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
23 * of the Software.
27 * Authors:
28 * Jerome Glisse <glisse@freedesktop.org>
29 * Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
30 * Dave Airlie
32 #include <linux/list.h>
33 #include <linux/slab.h>
34 #include <drm/drmP.h>
35 #include "radeon_drm.h"
36 #include "radeon.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);
59 kfree(bo_va);
63 static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo)
65 struct radeon_bo *bo;
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);
74 kfree(bo);
77 bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo)
79 if (bo->destroy == &radeon_ttm_bo_destroy)
80 return true;
81 return false;
84 void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain)
86 u32 c = 0;
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 |
94 TTM_PL_FLAG_VRAM;
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;
99 if (!c)
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;
113 size_t acc_size;
114 int r;
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;
121 if (kernel) {
122 type = ttm_bo_type_kernel;
123 } else {
124 type = ttm_bo_type_device;
126 *bo_ptr = NULL;
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);
133 return -ENOMEM;
136 acc_size = ttm_bo_dma_acc_size(&rdev->mman.bdev, size,
137 sizeof(struct radeon_bo));
139 retry:
140 bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL);
141 if (bo == NULL)
142 return -ENOMEM;
143 r = drm_gem_object_init(rdev->ddev, &bo->gem_base, size);
144 if (unlikely(r)) {
145 kfree(bo);
146 return r;
148 bo->rdev = rdev;
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;
164 goto retry;
166 dev_err(rdev->dev,
167 "object_init failed for (%lu, 0x%08X)\n",
168 size, domain);
170 return r;
172 *bo_ptr = bo;
174 trace_radeon_bo_create(bo);
176 return 0;
179 int radeon_bo_kmap(struct radeon_bo *bo, void **ptr)
181 bool is_iomem;
182 int r;
184 if (bo->kptr) {
185 if (ptr) {
186 *ptr = bo->kptr;
188 return 0;
190 r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap);
191 if (r) {
192 return r;
194 bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
195 if (ptr) {
196 *ptr = bo->kptr;
198 radeon_bo_check_tiling(bo, 0, 0);
199 return 0;
202 void radeon_bo_kunmap(struct radeon_bo *bo)
204 if (bo->kptr == NULL)
205 return;
206 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;
216 if ((*bo) == NULL)
217 return;
218 rdev = (*bo)->rdev;
219 tbo = &((*bo)->tbo);
220 mutex_lock(&rdev->vram_mutex);
221 ttm_bo_unref(&tbo);
222 mutex_unlock(&rdev->vram_mutex);
223 if (tbo == NULL)
224 *bo = NULL;
227 int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset,
228 u64 *gpu_addr)
230 int r, i;
232 if (bo->pin_count) {
233 bo->pin_count++;
234 if (gpu_addr)
235 *gpu_addr = radeon_bo_gpu_offset(bo);
236 WARN_ON_ONCE(max_offset != 0);
237 return 0;
239 radeon_ttm_placement_from_domain(bo, domain);
240 if (domain == RADEON_GEM_DOMAIN_VRAM) {
241 /* force to pin into visible video ram */
242 bo->placement.lpfn = bo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
244 if (max_offset) {
245 u64 lpfn = max_offset >> PAGE_SHIFT;
247 if (!bo->placement.lpfn)
248 bo->placement.lpfn = bo->rdev->mc.gtt_size >> PAGE_SHIFT;
250 if (lpfn < bo->placement.lpfn)
251 bo->placement.lpfn = lpfn;
253 for (i = 0; i < bo->placement.num_placement; i++)
254 bo->placements[i] |= TTM_PL_FLAG_NO_EVICT;
255 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
256 if (likely(r == 0)) {
257 bo->pin_count = 1;
258 if (gpu_addr != NULL)
259 *gpu_addr = radeon_bo_gpu_offset(bo);
261 if (unlikely(r != 0))
262 dev_err(bo->rdev->dev, "%p pin failed\n", bo);
263 return r;
266 int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
268 return radeon_bo_pin_restricted(bo, domain, 0, gpu_addr);
271 int radeon_bo_unpin(struct radeon_bo *bo)
273 int r, i;
275 if (!bo->pin_count) {
276 dev_warn(bo->rdev->dev, "%p unpin not necessary\n", bo);
277 return 0;
279 bo->pin_count--;
280 if (bo->pin_count)
281 return 0;
282 for (i = 0; i < bo->placement.num_placement; i++)
283 bo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT;
284 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
285 if (unlikely(r != 0))
286 dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo);
287 return r;
290 int radeon_bo_evict_vram(struct radeon_device *rdev)
292 /* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
293 if (0 && (rdev->flags & RADEON_IS_IGP)) {
294 if (rdev->mc.igp_sideport_enabled == false)
295 /* Useless to evict on IGP chips */
296 return 0;
298 return ttm_bo_evict_mm(&rdev->mman.bdev, TTM_PL_VRAM);
301 void radeon_bo_force_delete(struct radeon_device *rdev)
303 struct radeon_bo *bo, *n;
305 if (list_empty(&rdev->gem.objects)) {
306 return;
308 dev_err(rdev->dev, "Userspace still has active objects !\n");
309 list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) {
310 mutex_lock(&rdev->ddev->struct_mutex);
311 dev_err(rdev->dev, "%p %p %lu %lu force free\n",
312 &bo->gem_base, bo, (unsigned long)bo->gem_base.size,
313 *((unsigned long *)&bo->gem_base.refcount));
314 mutex_lock(&bo->rdev->gem.mutex);
315 list_del_init(&bo->list);
316 mutex_unlock(&bo->rdev->gem.mutex);
317 /* this should unref the ttm bo */
318 drm_gem_object_unreference(&bo->gem_base);
319 mutex_unlock(&rdev->ddev->struct_mutex);
323 int radeon_bo_init(struct radeon_device *rdev)
325 /* Add an MTRR for the VRAM */
326 rdev->mc.vram_mtrr = mtrr_add(rdev->mc.aper_base, rdev->mc.aper_size,
327 MTRR_TYPE_WRCOMB, 1);
328 DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
329 rdev->mc.mc_vram_size >> 20,
330 (unsigned long long)rdev->mc.aper_size >> 20);
331 DRM_INFO("RAM width %dbits %cDR\n",
332 rdev->mc.vram_width, rdev->mc.vram_is_ddr ? 'D' : 'S');
333 return radeon_ttm_init(rdev);
336 void radeon_bo_fini(struct radeon_device *rdev)
338 radeon_ttm_fini(rdev);
341 void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
342 struct list_head *head)
344 if (lobj->wdomain) {
345 list_add(&lobj->tv.head, head);
346 } else {
347 list_add_tail(&lobj->tv.head, head);
351 int radeon_bo_list_validate(struct list_head *head)
353 struct radeon_bo_list *lobj;
354 struct radeon_bo *bo;
355 u32 domain;
356 int r;
358 r = ttm_eu_reserve_buffers(head);
359 if (unlikely(r != 0)) {
360 return r;
362 list_for_each_entry(lobj, head, tv.head) {
363 bo = lobj->bo;
364 if (!bo->pin_count) {
365 domain = lobj->wdomain ? lobj->wdomain : lobj->rdomain;
367 retry:
368 radeon_ttm_placement_from_domain(bo, domain);
369 r = ttm_bo_validate(&bo->tbo, &bo->placement,
370 true, false, false);
371 if (unlikely(r)) {
372 if (r != -ERESTARTSYS && domain == RADEON_GEM_DOMAIN_VRAM) {
373 domain |= RADEON_GEM_DOMAIN_GTT;
374 goto retry;
376 return r;
379 lobj->gpu_offset = radeon_bo_gpu_offset(bo);
380 lobj->tiling_flags = bo->tiling_flags;
382 return 0;
385 int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
386 struct vm_area_struct *vma)
388 return ttm_fbdev_mmap(vma, &bo->tbo);
391 int radeon_bo_get_surface_reg(struct radeon_bo *bo)
393 struct radeon_device *rdev = bo->rdev;
394 struct radeon_surface_reg *reg;
395 struct radeon_bo *old_object;
396 int steal;
397 int i;
399 BUG_ON(!atomic_read(&bo->tbo.reserved));
401 if (!bo->tiling_flags)
402 return 0;
404 if (bo->surface_reg >= 0) {
405 reg = &rdev->surface_regs[bo->surface_reg];
406 i = bo->surface_reg;
407 goto out;
410 steal = -1;
411 for (i = 0; i < RADEON_GEM_MAX_SURFACES; i++) {
413 reg = &rdev->surface_regs[i];
414 if (!reg->bo)
415 break;
417 old_object = reg->bo;
418 if (old_object->pin_count == 0)
419 steal = i;
422 /* if we are all out */
423 if (i == RADEON_GEM_MAX_SURFACES) {
424 if (steal == -1)
425 return -ENOMEM;
426 /* find someone with a surface reg and nuke their BO */
427 reg = &rdev->surface_regs[steal];
428 old_object = reg->bo;
429 /* blow away the mapping */
430 DRM_DEBUG("stealing surface reg %d from %p\n", steal, old_object);
431 ttm_bo_unmap_virtual(&old_object->tbo);
432 old_object->surface_reg = -1;
433 i = steal;
436 bo->surface_reg = i;
437 reg->bo = bo;
439 out:
440 radeon_set_surface_reg(rdev, i, bo->tiling_flags, bo->pitch,
441 bo->tbo.mem.start << PAGE_SHIFT,
442 bo->tbo.num_pages << PAGE_SHIFT);
443 return 0;
446 static void radeon_bo_clear_surface_reg(struct radeon_bo *bo)
448 struct radeon_device *rdev = bo->rdev;
449 struct radeon_surface_reg *reg;
451 if (bo->surface_reg == -1)
452 return;
454 reg = &rdev->surface_regs[bo->surface_reg];
455 radeon_clear_surface_reg(rdev, bo->surface_reg);
457 reg->bo = NULL;
458 bo->surface_reg = -1;
461 int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
462 uint32_t tiling_flags, uint32_t pitch)
464 int r;
466 r = radeon_bo_reserve(bo, false);
467 if (unlikely(r != 0))
468 return r;
469 bo->tiling_flags = tiling_flags;
470 bo->pitch = pitch;
471 radeon_bo_unreserve(bo);
472 return 0;
475 void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
476 uint32_t *tiling_flags,
477 uint32_t *pitch)
479 BUG_ON(!atomic_read(&bo->tbo.reserved));
480 if (tiling_flags)
481 *tiling_flags = bo->tiling_flags;
482 if (pitch)
483 *pitch = bo->pitch;
486 int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
487 bool force_drop)
489 BUG_ON(!atomic_read(&bo->tbo.reserved));
491 if (!(bo->tiling_flags & RADEON_TILING_SURFACE))
492 return 0;
494 if (force_drop) {
495 radeon_bo_clear_surface_reg(bo);
496 return 0;
499 if (bo->tbo.mem.mem_type != TTM_PL_VRAM) {
500 if (!has_moved)
501 return 0;
503 if (bo->surface_reg >= 0)
504 radeon_bo_clear_surface_reg(bo);
505 return 0;
508 if ((bo->surface_reg >= 0) && !has_moved)
509 return 0;
511 return radeon_bo_get_surface_reg(bo);
514 void radeon_bo_move_notify(struct ttm_buffer_object *bo,
515 struct ttm_mem_reg *mem)
517 struct radeon_bo *rbo;
518 if (!radeon_ttm_bo_is_radeon_bo(bo))
519 return;
520 rbo = container_of(bo, struct radeon_bo, tbo);
521 radeon_bo_check_tiling(rbo, 0, 1);
522 radeon_vm_bo_invalidate(rbo->rdev, rbo);
525 int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
527 struct radeon_device *rdev;
528 struct radeon_bo *rbo;
529 unsigned long offset, size;
530 int r;
532 if (!radeon_ttm_bo_is_radeon_bo(bo))
533 return 0;
534 rbo = container_of(bo, struct radeon_bo, tbo);
535 radeon_bo_check_tiling(rbo, 0, 0);
536 rdev = rbo->rdev;
537 if (bo->mem.mem_type == TTM_PL_VRAM) {
538 size = bo->mem.num_pages << PAGE_SHIFT;
539 offset = bo->mem.start << PAGE_SHIFT;
540 if ((offset + size) > rdev->mc.visible_vram_size) {
541 /* hurrah the memory is not visible ! */
542 radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
543 rbo->placement.lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
544 r = ttm_bo_validate(bo, &rbo->placement, false, true, false);
545 if (unlikely(r != 0))
546 return r;
547 offset = bo->mem.start << PAGE_SHIFT;
548 /* this should not happen */
549 if ((offset + size) > rdev->mc.visible_vram_size)
550 return -EINVAL;
553 return 0;
556 int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
558 int r;
560 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
561 if (unlikely(r != 0))
562 return r;
563 spin_lock(&bo->tbo.bdev->fence_lock);
564 if (mem_type)
565 *mem_type = bo->tbo.mem.mem_type;
566 if (bo->tbo.sync_obj)
567 r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
568 spin_unlock(&bo->tbo.bdev->fence_lock);
569 ttm_bo_unreserve(&bo->tbo);
570 return r;
575 * radeon_bo_reserve - reserve bo
576 * @bo: bo structure
577 * @no_wait: don't sleep while trying to reserve (return -EBUSY)
579 * Returns:
580 * -EBUSY: buffer is busy and @no_wait is true
581 * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by
582 * a signal. Release all buffer reservations and return to user-space.
584 int radeon_bo_reserve(struct radeon_bo *bo, bool no_wait)
586 int r;
588 r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
589 if (unlikely(r != 0)) {
590 if (r != -ERESTARTSYS)
591 dev_err(bo->rdev->dev, "%p reserve failed\n", bo);
592 return r;
594 return 0;
597 /* object have to be reserved */
598 struct radeon_bo_va *radeon_bo_va(struct radeon_bo *rbo, struct radeon_vm *vm)
600 struct radeon_bo_va *bo_va;
602 list_for_each_entry(bo_va, &rbo->va, bo_list) {
603 if (bo_va->vm == vm) {
604 return bo_va;
607 return NULL;