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 <ttm/ttm_bo_api.h>
33 #include <ttm/ttm_bo_driver.h>
34 #include <ttm/ttm_placement.h>
35 #include <ttm/ttm_module.h>
37 #include <drm/radeon_drm.h>
38 #include "radeon_reg.h"
41 #define DRM_FILE_PAGE_OFFSET (0x100000000ULL >> PAGE_SHIFT)
43 static struct radeon_device *radeon_get_rdev(struct ttm_bo_device *bdev)
45 struct radeon_mman *mman;
46 struct radeon_device *rdev;
48 mman = container_of(bdev, struct radeon_mman, bdev);
49 rdev = container_of(mman, struct radeon_device, mman);
57 static int radeon_ttm_mem_global_init(struct ttm_global_reference *ref)
59 return ttm_mem_global_init(ref->object);
62 static void radeon_ttm_mem_global_release(struct ttm_global_reference *ref)
64 ttm_mem_global_release(ref->object);
67 static int radeon_ttm_global_init(struct radeon_device *rdev)
69 struct ttm_global_reference *global_ref;
72 rdev->mman.mem_global_referenced = false;
73 global_ref = &rdev->mman.mem_global_ref;
74 global_ref->global_type = TTM_GLOBAL_TTM_MEM;
75 global_ref->size = sizeof(struct ttm_mem_global);
76 global_ref->init = &radeon_ttm_mem_global_init;
77 global_ref->release = &radeon_ttm_mem_global_release;
78 r = ttm_global_item_ref(global_ref);
80 DRM_ERROR("Failed referencing a global TTM memory object.\n");
83 rdev->mman.mem_global_referenced = true;
87 static void radeon_ttm_global_fini(struct radeon_device *rdev)
89 if (rdev->mman.mem_global_referenced) {
90 ttm_global_item_unref(&rdev->mman.mem_global_ref);
91 rdev->mman.mem_global_referenced = false;
95 struct ttm_backend *radeon_ttm_backend_create(struct radeon_device *rdev);
97 static struct ttm_backend*
98 radeon_create_ttm_backend_entry(struct ttm_bo_device *bdev)
100 struct radeon_device *rdev;
102 rdev = radeon_get_rdev(bdev);
104 if (rdev->flags & RADEON_IS_AGP) {
105 return ttm_agp_backend_init(bdev, rdev->ddev->agp->bridge);
109 return radeon_ttm_backend_create(rdev);
113 static int radeon_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
118 static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
119 struct ttm_mem_type_manager *man)
121 struct radeon_device *rdev;
123 rdev = radeon_get_rdev(bdev);
128 man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
129 man->available_caching = TTM_PL_MASK_CACHING;
130 man->default_caching = TTM_PL_FLAG_CACHED;
134 man->available_caching = TTM_PL_MASK_CACHING;
135 man->default_caching = TTM_PL_FLAG_CACHED;
137 if (rdev->flags & RADEON_IS_AGP) {
138 if (!(drm_core_has_AGP(rdev->ddev) && rdev->ddev->agp)) {
139 DRM_ERROR("AGP is not enabled for memory type %u\n",
143 man->io_offset = rdev->mc.agp_base;
144 man->io_size = rdev->mc.gtt_size;
146 man->flags = TTM_MEMTYPE_FLAG_NEEDS_IOREMAP |
147 TTM_MEMTYPE_FLAG_MAPPABLE;
148 man->available_caching = TTM_PL_FLAG_UNCACHED |
150 man->default_caching = TTM_PL_FLAG_WC;
157 man->flags = TTM_MEMTYPE_FLAG_MAPPABLE |
158 TTM_MEMTYPE_FLAG_CMA;
162 /* "On-card" video ram */
164 man->flags = TTM_MEMTYPE_FLAG_FIXED |
165 TTM_MEMTYPE_FLAG_NEEDS_IOREMAP |
166 TTM_MEMTYPE_FLAG_MAPPABLE;
167 man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
168 man->default_caching = TTM_PL_FLAG_WC;
170 man->io_offset = rdev->mc.aper_base;
171 man->io_size = rdev->mc.aper_size;
174 DRM_ERROR("Unsupported memory type %u\n", (unsigned)type);
180 static uint32_t radeon_evict_flags(struct ttm_buffer_object *bo)
182 uint32_t cur_placement = bo->mem.placement & ~TTM_PL_MASK_MEMTYPE;
184 switch (bo->mem.mem_type) {
186 return (cur_placement & ~TTM_PL_MASK_CACHING) |
192 static int radeon_verify_access(struct ttm_buffer_object *bo, struct file *filp)
197 static void radeon_move_null(struct ttm_buffer_object *bo,
198 struct ttm_mem_reg *new_mem)
200 struct ttm_mem_reg *old_mem = &bo->mem;
202 BUG_ON(old_mem->mm_node != NULL);
204 new_mem->mm_node = NULL;
207 static int radeon_move_blit(struct ttm_buffer_object *bo,
208 bool evict, int no_wait,
209 struct ttm_mem_reg *new_mem,
210 struct ttm_mem_reg *old_mem)
212 struct radeon_device *rdev;
213 uint64_t old_start, new_start;
214 struct radeon_fence *fence;
217 rdev = radeon_get_rdev(bo->bdev);
218 r = radeon_fence_create(rdev, &fence);
222 old_start = old_mem->mm_node->start << PAGE_SHIFT;
223 new_start = new_mem->mm_node->start << PAGE_SHIFT;
225 switch (old_mem->mem_type) {
227 old_start += rdev->mc.vram_location;
230 old_start += rdev->mc.gtt_location;
233 DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
236 switch (new_mem->mem_type) {
238 new_start += rdev->mc.vram_location;
241 new_start += rdev->mc.gtt_location;
244 DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
247 if (!rdev->cp.ready) {
248 DRM_ERROR("Trying to move memory with CP turned off.\n");
251 r = radeon_copy(rdev, old_start, new_start, new_mem->num_pages, fence);
252 /* FIXME: handle copy error */
253 r = ttm_bo_move_accel_cleanup(bo, (void *)fence, NULL,
254 evict, no_wait, new_mem);
255 radeon_fence_unref(&fence);
259 static int radeon_move_vram_ram(struct ttm_buffer_object *bo,
260 bool evict, bool interruptible, bool no_wait,
261 struct ttm_mem_reg *new_mem)
263 struct radeon_device *rdev;
264 struct ttm_mem_reg *old_mem = &bo->mem;
265 struct ttm_mem_reg tmp_mem;
266 uint32_t proposed_placement;
269 rdev = radeon_get_rdev(bo->bdev);
271 tmp_mem.mm_node = NULL;
272 proposed_placement = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING;
273 r = ttm_bo_mem_space(bo, proposed_placement, &tmp_mem,
274 interruptible, no_wait);
278 r = ttm_tt_bind(bo->ttm, &tmp_mem);
282 r = radeon_move_blit(bo, true, no_wait, &tmp_mem, old_mem);
286 r = ttm_bo_move_ttm(bo, true, no_wait, new_mem);
288 if (tmp_mem.mm_node) {
289 spin_lock(&rdev->mman.bdev.lru_lock);
290 drm_mm_put_block(tmp_mem.mm_node);
291 spin_unlock(&rdev->mman.bdev.lru_lock);
297 static int radeon_move_ram_vram(struct ttm_buffer_object *bo,
298 bool evict, bool interruptible, bool no_wait,
299 struct ttm_mem_reg *new_mem)
301 struct radeon_device *rdev;
302 struct ttm_mem_reg *old_mem = &bo->mem;
303 struct ttm_mem_reg tmp_mem;
304 uint32_t proposed_flags;
307 rdev = radeon_get_rdev(bo->bdev);
309 tmp_mem.mm_node = NULL;
310 proposed_flags = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING;
311 r = ttm_bo_mem_space(bo, proposed_flags, &tmp_mem,
312 interruptible, no_wait);
316 r = ttm_bo_move_ttm(bo, true, no_wait, &tmp_mem);
320 r = radeon_move_blit(bo, true, no_wait, new_mem, old_mem);
325 if (tmp_mem.mm_node) {
326 spin_lock(&rdev->mman.bdev.lru_lock);
327 drm_mm_put_block(tmp_mem.mm_node);
328 spin_unlock(&rdev->mman.bdev.lru_lock);
334 static int radeon_bo_move(struct ttm_buffer_object *bo,
335 bool evict, bool interruptible, bool no_wait,
336 struct ttm_mem_reg *new_mem)
338 struct radeon_device *rdev;
339 struct ttm_mem_reg *old_mem = &bo->mem;
342 rdev = radeon_get_rdev(bo->bdev);
343 if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
344 radeon_move_null(bo, new_mem);
347 if ((old_mem->mem_type == TTM_PL_TT &&
348 new_mem->mem_type == TTM_PL_SYSTEM) ||
349 (old_mem->mem_type == TTM_PL_SYSTEM &&
350 new_mem->mem_type == TTM_PL_TT)) {
351 /* bind is enought */
352 radeon_move_null(bo, new_mem);
355 if (!rdev->cp.ready) {
357 DRM_ERROR("CP is not ready use memcpy.\n");
358 return ttm_bo_move_memcpy(bo, evict, no_wait, new_mem);
361 if (old_mem->mem_type == TTM_PL_VRAM &&
362 new_mem->mem_type == TTM_PL_SYSTEM) {
363 return radeon_move_vram_ram(bo, evict, interruptible,
365 } else if (old_mem->mem_type == TTM_PL_SYSTEM &&
366 new_mem->mem_type == TTM_PL_VRAM) {
367 return radeon_move_ram_vram(bo, evict, interruptible,
370 r = radeon_move_blit(bo, evict, no_wait, new_mem, old_mem);
378 const uint32_t radeon_mem_prios[] = {
384 const uint32_t radeon_busy_prios[] = {
390 static int radeon_sync_obj_wait(void *sync_obj, void *sync_arg,
391 bool lazy, bool interruptible)
393 return radeon_fence_wait((struct radeon_fence *)sync_obj, interruptible);
396 static int radeon_sync_obj_flush(void *sync_obj, void *sync_arg)
401 static void radeon_sync_obj_unref(void **sync_obj)
403 radeon_fence_unref((struct radeon_fence **)sync_obj);
406 static void *radeon_sync_obj_ref(void *sync_obj)
408 return radeon_fence_ref((struct radeon_fence *)sync_obj);
411 static bool radeon_sync_obj_signaled(void *sync_obj, void *sync_arg)
413 return radeon_fence_signaled((struct radeon_fence *)sync_obj);
416 static struct ttm_bo_driver radeon_bo_driver = {
417 .mem_type_prio = radeon_mem_prios,
418 .mem_busy_prio = radeon_busy_prios,
419 .num_mem_type_prio = ARRAY_SIZE(radeon_mem_prios),
420 .num_mem_busy_prio = ARRAY_SIZE(radeon_busy_prios),
421 .create_ttm_backend_entry = &radeon_create_ttm_backend_entry,
422 .invalidate_caches = &radeon_invalidate_caches,
423 .init_mem_type = &radeon_init_mem_type,
424 .evict_flags = &radeon_evict_flags,
425 .move = &radeon_bo_move,
426 .verify_access = &radeon_verify_access,
427 .sync_obj_signaled = &radeon_sync_obj_signaled,
428 .sync_obj_wait = &radeon_sync_obj_wait,
429 .sync_obj_flush = &radeon_sync_obj_flush,
430 .sync_obj_unref = &radeon_sync_obj_unref,
431 .sync_obj_ref = &radeon_sync_obj_ref,
434 int radeon_ttm_init(struct radeon_device *rdev)
438 r = radeon_ttm_global_init(rdev);
442 /* No others user of address space so set it to 0 */
443 r = ttm_bo_device_init(&rdev->mman.bdev,
444 rdev->mman.mem_global_ref.object,
445 &radeon_bo_driver, DRM_FILE_PAGE_OFFSET);
447 DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
450 r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM, 0,
451 ((rdev->mc.aper_size) >> PAGE_SHIFT));
453 DRM_ERROR("Failed initializing VRAM heap.\n");
456 r = radeon_object_create(rdev, NULL, 256 * 1024, true,
457 RADEON_GEM_DOMAIN_VRAM, false,
458 &rdev->stollen_vga_memory);
462 r = radeon_object_pin(rdev->stollen_vga_memory, RADEON_GEM_DOMAIN_VRAM, NULL);
464 radeon_object_unref(&rdev->stollen_vga_memory);
467 DRM_INFO("radeon: %uM of VRAM memory ready\n",
468 rdev->mc.vram_size / (1024 * 1024));
469 r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_TT, 0,
470 ((rdev->mc.gtt_size) >> PAGE_SHIFT));
472 DRM_ERROR("Failed initializing GTT heap.\n");
475 DRM_INFO("radeon: %uM of GTT memory ready.\n",
476 rdev->mc.gtt_size / (1024 * 1024));
477 if (unlikely(rdev->mman.bdev.dev_mapping == NULL)) {
478 rdev->mman.bdev.dev_mapping = rdev->ddev->dev_mapping;
483 void radeon_ttm_fini(struct radeon_device *rdev)
485 if (rdev->stollen_vga_memory) {
486 radeon_object_unpin(rdev->stollen_vga_memory);
487 radeon_object_unref(&rdev->stollen_vga_memory);
489 ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_VRAM);
490 ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_TT);
491 ttm_bo_device_release(&rdev->mman.bdev);
492 radeon_gart_fini(rdev);
493 radeon_ttm_global_fini(rdev);
494 DRM_INFO("radeon: ttm finalized\n");
497 static struct vm_operations_struct radeon_ttm_vm_ops;
498 static struct vm_operations_struct *ttm_vm_ops = NULL;
500 static int radeon_ttm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
502 struct ttm_buffer_object *bo;
505 bo = (struct ttm_buffer_object *)vma->vm_private_data;
507 return VM_FAULT_NOPAGE;
509 r = ttm_vm_ops->fault(vma, vmf);
513 int radeon_mmap(struct file *filp, struct vm_area_struct *vma)
515 struct drm_file *file_priv;
516 struct radeon_device *rdev;
519 if (unlikely(vma->vm_pgoff < DRM_FILE_PAGE_OFFSET)) {
520 return drm_mmap(filp, vma);
523 file_priv = (struct drm_file *)filp->private_data;
524 rdev = file_priv->minor->dev->dev_private;
528 r = ttm_bo_mmap(filp, vma, &rdev->mman.bdev);
529 if (unlikely(r != 0)) {
532 if (unlikely(ttm_vm_ops == NULL)) {
533 ttm_vm_ops = vma->vm_ops;
534 radeon_ttm_vm_ops = *ttm_vm_ops;
535 radeon_ttm_vm_ops.fault = &radeon_ttm_fault;
537 vma->vm_ops = &radeon_ttm_vm_ops;
543 * TTM backend functions.
545 struct radeon_ttm_backend {
546 struct ttm_backend backend;
547 struct radeon_device *rdev;
548 unsigned long num_pages;
550 struct page *dummy_read_page;
556 static int radeon_ttm_backend_populate(struct ttm_backend *backend,
557 unsigned long num_pages,
559 struct page *dummy_read_page)
561 struct radeon_ttm_backend *gtt;
563 gtt = container_of(backend, struct radeon_ttm_backend, backend);
565 gtt->num_pages = num_pages;
566 gtt->dummy_read_page = dummy_read_page;
567 gtt->populated = true;
571 static void radeon_ttm_backend_clear(struct ttm_backend *backend)
573 struct radeon_ttm_backend *gtt;
575 gtt = container_of(backend, struct radeon_ttm_backend, backend);
578 gtt->dummy_read_page = NULL;
579 gtt->populated = false;
584 static int radeon_ttm_backend_bind(struct ttm_backend *backend,
585 struct ttm_mem_reg *bo_mem)
587 struct radeon_ttm_backend *gtt;
590 gtt = container_of(backend, struct radeon_ttm_backend, backend);
591 gtt->offset = bo_mem->mm_node->start << PAGE_SHIFT;
592 if (!gtt->num_pages) {
593 WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n", gtt->num_pages, bo_mem, backend);
595 r = radeon_gart_bind(gtt->rdev, gtt->offset,
596 gtt->num_pages, gtt->pages);
598 DRM_ERROR("failed to bind %lu pages at 0x%08X\n",
599 gtt->num_pages, gtt->offset);
606 static int radeon_ttm_backend_unbind(struct ttm_backend *backend)
608 struct radeon_ttm_backend *gtt;
610 gtt = container_of(backend, struct radeon_ttm_backend, backend);
611 radeon_gart_unbind(gtt->rdev, gtt->offset, gtt->num_pages);
616 static void radeon_ttm_backend_destroy(struct ttm_backend *backend)
618 struct radeon_ttm_backend *gtt;
620 gtt = container_of(backend, struct radeon_ttm_backend, backend);
622 radeon_ttm_backend_unbind(backend);
627 static struct ttm_backend_func radeon_backend_func = {
628 .populate = &radeon_ttm_backend_populate,
629 .clear = &radeon_ttm_backend_clear,
630 .bind = &radeon_ttm_backend_bind,
631 .unbind = &radeon_ttm_backend_unbind,
632 .destroy = &radeon_ttm_backend_destroy,
635 struct ttm_backend *radeon_ttm_backend_create(struct radeon_device *rdev)
637 struct radeon_ttm_backend *gtt;
639 gtt = kzalloc(sizeof(struct radeon_ttm_backend), GFP_KERNEL);
643 gtt->backend.bdev = &rdev->mman.bdev;
644 gtt->backend.flags = 0;
645 gtt->backend.func = &radeon_backend_func;
649 gtt->dummy_read_page = NULL;
650 gtt->populated = false;
652 return >t->backend;