1 /* 2 * Copyright 2009 Jerome Glisse. 3 * All Rights Reserved. 4 * 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: 12 * 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. 20 * 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. 24 * 25 */ 26 /* 27 * Authors: 28 * Jerome Glisse <glisse@freedesktop.org> 29 * Thomas Hellstrom <thomas-at-tungstengraphics-dot-com> 30 * Dave Airlie 31 */ 32 #include <linux/list.h> 33 #include <drm/drmP.h> 34 #include "radeon_drm.h" 35 #include "radeon.h" 36 37 38 int radeon_ttm_init(struct radeon_device *rdev); 39 void radeon_ttm_fini(struct radeon_device *rdev); 40 static void radeon_bo_clear_surface_reg(struct radeon_bo *bo); 41 42 /* 43 * To exclude mutual BO access we rely on bo_reserve exclusion, as all 44 * function are calling it. 45 */ 46 47 static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo) 48 { 49 struct radeon_bo *bo; 50 51 bo = container_of(tbo, struct radeon_bo, tbo); 52 mutex_lock(&bo->rdev->gem.mutex); 53 list_del_init(&bo->list); 54 mutex_unlock(&bo->rdev->gem.mutex); 55 radeon_bo_clear_surface_reg(bo); 56 kfree(bo); 57 } 58 59 bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo) 60 { 61 if (bo->destroy == &radeon_ttm_bo_destroy) 62 return true; 63 return false; 64 } 65 66 void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain) 67 { 68 u32 c = 0; 69 70 rbo->placement.fpfn = 0; 71 rbo->placement.lpfn = 0; 72 rbo->placement.placement = rbo->placements; 73 rbo->placement.busy_placement = rbo->placements; 74 if (domain & RADEON_GEM_DOMAIN_VRAM) 75 rbo->placements[c++] = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | 76 TTM_PL_FLAG_VRAM; 77 if (domain & RADEON_GEM_DOMAIN_GTT) 78 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT; 79 if (domain & RADEON_GEM_DOMAIN_CPU) 80 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM; 81 if (!c) 82 rbo->placements[c++] = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM; 83 rbo->placement.num_placement = c; 84 rbo->placement.num_busy_placement = c; 85 } 86 87 int radeon_bo_create(struct radeon_device *rdev, struct drm_gem_object *gobj, 88 unsigned long size, bool kernel, u32 domain, 89 struct radeon_bo **bo_ptr) 90 { 91 struct radeon_bo *bo; 92 enum ttm_bo_type type; 93 int r; 94 95 if (unlikely(rdev->mman.bdev.dev_mapping == NULL)) { 96 rdev->mman.bdev.dev_mapping = rdev->ddev->dev_mapping; 97 } 98 if (kernel) { 99 type = ttm_bo_type_kernel; 100 } else { 101 type = ttm_bo_type_device; 102 } 103 *bo_ptr = NULL; 104 bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL); 105 if (bo == NULL) 106 return -ENOMEM; 107 bo->rdev = rdev; 108 bo->gobj = gobj; 109 bo->surface_reg = -1; 110 INIT_LIST_HEAD(&bo->list); 111 112 radeon_ttm_placement_from_domain(bo, domain); 113 /* Kernel allocation are uninterruptible */ 114 r = ttm_bo_init(&rdev->mman.bdev, &bo->tbo, size, type, 115 &bo->placement, 0, 0, !kernel, NULL, size, 116 &radeon_ttm_bo_destroy); 117 if (unlikely(r != 0)) { 118 if (r != -ERESTARTSYS) 119 dev_err(rdev->dev, 120 "object_init failed for (%lu, 0x%08X)\n", 121 size, domain); 122 return r; 123 } 124 *bo_ptr = bo; 125 if (gobj) { 126 mutex_lock(&bo->rdev->gem.mutex); 127 list_add_tail(&bo->list, &rdev->gem.objects); 128 mutex_unlock(&bo->rdev->gem.mutex); 129 } 130 return 0; 131 } 132 133 int radeon_bo_kmap(struct radeon_bo *bo, void **ptr) 134 { 135 bool is_iomem; 136 int r; 137 138 if (bo->kptr) { 139 if (ptr) { 140 *ptr = bo->kptr; 141 } 142 return 0; 143 } 144 r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap); 145 if (r) { 146 return r; 147 } 148 bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem); 149 if (ptr) { 150 *ptr = bo->kptr; 151 } 152 radeon_bo_check_tiling(bo, 0, 0); 153 return 0; 154 } 155 156 void radeon_bo_kunmap(struct radeon_bo *bo) 157 { 158 if (bo->kptr == NULL) 159 return; 160 bo->kptr = NULL; 161 radeon_bo_check_tiling(bo, 0, 0); 162 ttm_bo_kunmap(&bo->kmap); 163 } 164 165 void radeon_bo_unref(struct radeon_bo **bo) 166 { 167 struct ttm_buffer_object *tbo; 168 169 if ((*bo) == NULL) 170 return; 171 tbo = &((*bo)->tbo); 172 ttm_bo_unref(&tbo); 173 if (tbo == NULL) 174 *bo = NULL; 175 } 176 177 int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr) 178 { 179 int r, i; 180 181 radeon_ttm_placement_from_domain(bo, domain); 182 if (bo->pin_count) { 183 bo->pin_count++; 184 if (gpu_addr) 185 *gpu_addr = radeon_bo_gpu_offset(bo); 186 return 0; 187 } 188 radeon_ttm_placement_from_domain(bo, domain); 189 for (i = 0; i < bo->placement.num_placement; i++) 190 bo->placements[i] |= TTM_PL_FLAG_NO_EVICT; 191 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false); 192 if (likely(r == 0)) { 193 bo->pin_count = 1; 194 if (gpu_addr != NULL) 195 *gpu_addr = radeon_bo_gpu_offset(bo); 196 } 197 if (unlikely(r != 0)) 198 dev_err(bo->rdev->dev, "%p pin failed\n", bo); 199 return r; 200 } 201 202 int radeon_bo_unpin(struct radeon_bo *bo) 203 { 204 int r, i; 205 206 if (!bo->pin_count) { 207 dev_warn(bo->rdev->dev, "%p unpin not necessary\n", bo); 208 return 0; 209 } 210 bo->pin_count--; 211 if (bo->pin_count) 212 return 0; 213 for (i = 0; i < bo->placement.num_placement; i++) 214 bo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT; 215 r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false); 216 if (unlikely(r != 0)) 217 dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo); 218 return r; 219 } 220 221 int radeon_bo_evict_vram(struct radeon_device *rdev) 222 { 223 /* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */ 224 if (0 && (rdev->flags & RADEON_IS_IGP)) { 225 if (rdev->mc.igp_sideport_enabled == false) 226 /* Useless to evict on IGP chips */ 227 return 0; 228 } 229 return ttm_bo_evict_mm(&rdev->mman.bdev, TTM_PL_VRAM); 230 } 231 232 void radeon_bo_force_delete(struct radeon_device *rdev) 233 { 234 struct radeon_bo *bo, *n; 235 struct drm_gem_object *gobj; 236 237 if (list_empty(&rdev->gem.objects)) { 238 return; 239 } 240 dev_err(rdev->dev, "Userspace still has active objects !\n"); 241 list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) { 242 mutex_lock(&rdev->ddev->struct_mutex); 243 gobj = bo->gobj; 244 dev_err(rdev->dev, "%p %p %lu %lu force free\n", 245 gobj, bo, (unsigned long)gobj->size, 246 *((unsigned long *)&gobj->refcount)); 247 mutex_lock(&bo->rdev->gem.mutex); 248 list_del_init(&bo->list); 249 mutex_unlock(&bo->rdev->gem.mutex); 250 radeon_bo_unref(&bo); 251 gobj->driver_private = NULL; 252 drm_gem_object_unreference(gobj); 253 mutex_unlock(&rdev->ddev->struct_mutex); 254 } 255 } 256 257 int radeon_bo_init(struct radeon_device *rdev) 258 { 259 /* Add an MTRR for the VRAM */ 260 rdev->mc.vram_mtrr = mtrr_add(rdev->mc.aper_base, rdev->mc.aper_size, 261 MTRR_TYPE_WRCOMB, 1); 262 DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n", 263 rdev->mc.mc_vram_size >> 20, 264 (unsigned long long)rdev->mc.aper_size >> 20); 265 DRM_INFO("RAM width %dbits %cDR\n", 266 rdev->mc.vram_width, rdev->mc.vram_is_ddr ? 'D' : 'S'); 267 return radeon_ttm_init(rdev); 268 } 269 270 void radeon_bo_fini(struct radeon_device *rdev) 271 { 272 radeon_ttm_fini(rdev); 273 } 274 275 void radeon_bo_list_add_object(struct radeon_bo_list *lobj, 276 struct list_head *head) 277 { 278 if (lobj->wdomain) { 279 list_add(&lobj->list, head); 280 } else { 281 list_add_tail(&lobj->list, head); 282 } 283 } 284 285 int radeon_bo_list_reserve(struct list_head *head) 286 { 287 struct radeon_bo_list *lobj; 288 int r; 289 290 list_for_each_entry(lobj, head, list){ 291 r = radeon_bo_reserve(lobj->bo, false); 292 if (unlikely(r != 0)) 293 return r; 294 } 295 return 0; 296 } 297 298 void radeon_bo_list_unreserve(struct list_head *head) 299 { 300 struct radeon_bo_list *lobj; 301 302 list_for_each_entry(lobj, head, list) { 303 /* only unreserve object we successfully reserved */ 304 if (radeon_bo_is_reserved(lobj->bo)) 305 radeon_bo_unreserve(lobj->bo); 306 } 307 } 308 309 int radeon_bo_list_validate(struct list_head *head) 310 { 311 struct radeon_bo_list *lobj; 312 struct radeon_bo *bo; 313 int r; 314 315 r = radeon_bo_list_reserve(head); 316 if (unlikely(r != 0)) { 317 return r; 318 } 319 list_for_each_entry(lobj, head, list) { 320 bo = lobj->bo; 321 if (!bo->pin_count) { 322 if (lobj->wdomain) { 323 radeon_ttm_placement_from_domain(bo, 324 lobj->wdomain); 325 } else { 326 radeon_ttm_placement_from_domain(bo, 327 lobj->rdomain); 328 } 329 r = ttm_bo_validate(&bo->tbo, &bo->placement, 330 true, false); 331 if (unlikely(r)) 332 return r; 333 } 334 lobj->gpu_offset = radeon_bo_gpu_offset(bo); 335 lobj->tiling_flags = bo->tiling_flags; 336 } 337 return 0; 338 } 339 340 void radeon_bo_list_fence(struct list_head *head, void *fence) 341 { 342 struct radeon_bo_list *lobj; 343 struct radeon_bo *bo; 344 struct radeon_fence *old_fence = NULL; 345 346 list_for_each_entry(lobj, head, list) { 347 bo = lobj->bo; 348 spin_lock(&bo->tbo.lock); 349 old_fence = (struct radeon_fence *)bo->tbo.sync_obj; 350 bo->tbo.sync_obj = radeon_fence_ref(fence); 351 bo->tbo.sync_obj_arg = NULL; 352 spin_unlock(&bo->tbo.lock); 353 if (old_fence) { 354 radeon_fence_unref(&old_fence); 355 } 356 } 357 } 358 359 int radeon_bo_fbdev_mmap(struct radeon_bo *bo, 360 struct vm_area_struct *vma) 361 { 362 return ttm_fbdev_mmap(vma, &bo->tbo); 363 } 364 365 int radeon_bo_get_surface_reg(struct radeon_bo *bo) 366 { 367 struct radeon_device *rdev = bo->rdev; 368 struct radeon_surface_reg *reg; 369 struct radeon_bo *old_object; 370 int steal; 371 int i; 372 373 BUG_ON(!atomic_read(&bo->tbo.reserved)); 374 375 if (!bo->tiling_flags) 376 return 0; 377 378 if (bo->surface_reg >= 0) { 379 reg = &rdev->surface_regs[bo->surface_reg]; 380 i = bo->surface_reg; 381 goto out; 382 } 383 384 steal = -1; 385 for (i = 0; i < RADEON_GEM_MAX_SURFACES; i++) { 386 387 reg = &rdev->surface_regs[i]; 388 if (!reg->bo) 389 break; 390 391 old_object = reg->bo; 392 if (old_object->pin_count == 0) 393 steal = i; 394 } 395 396 /* if we are all out */ 397 if (i == RADEON_GEM_MAX_SURFACES) { 398 if (steal == -1) 399 return -ENOMEM; 400 /* find someone with a surface reg and nuke their BO */ 401 reg = &rdev->surface_regs[steal]; 402 old_object = reg->bo; 403 /* blow away the mapping */ 404 DRM_DEBUG("stealing surface reg %d from %p\n", steal, old_object); 405 ttm_bo_unmap_virtual(&old_object->tbo); 406 old_object->surface_reg = -1; 407 i = steal; 408 } 409 410 bo->surface_reg = i; 411 reg->bo = bo; 412 413 out: 414 radeon_set_surface_reg(rdev, i, bo->tiling_flags, bo->pitch, 415 bo->tbo.mem.mm_node->start << PAGE_SHIFT, 416 bo->tbo.num_pages << PAGE_SHIFT); 417 return 0; 418 } 419 420 static void radeon_bo_clear_surface_reg(struct radeon_bo *bo) 421 { 422 struct radeon_device *rdev = bo->rdev; 423 struct radeon_surface_reg *reg; 424 425 if (bo->surface_reg == -1) 426 return; 427 428 reg = &rdev->surface_regs[bo->surface_reg]; 429 radeon_clear_surface_reg(rdev, bo->surface_reg); 430 431 reg->bo = NULL; 432 bo->surface_reg = -1; 433 } 434 435 int radeon_bo_set_tiling_flags(struct radeon_bo *bo, 436 uint32_t tiling_flags, uint32_t pitch) 437 { 438 int r; 439 440 r = radeon_bo_reserve(bo, false); 441 if (unlikely(r != 0)) 442 return r; 443 bo->tiling_flags = tiling_flags; 444 bo->pitch = pitch; 445 radeon_bo_unreserve(bo); 446 return 0; 447 } 448 449 void radeon_bo_get_tiling_flags(struct radeon_bo *bo, 450 uint32_t *tiling_flags, 451 uint32_t *pitch) 452 { 453 BUG_ON(!atomic_read(&bo->tbo.reserved)); 454 if (tiling_flags) 455 *tiling_flags = bo->tiling_flags; 456 if (pitch) 457 *pitch = bo->pitch; 458 } 459 460 int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved, 461 bool force_drop) 462 { 463 BUG_ON(!atomic_read(&bo->tbo.reserved)); 464 465 if (!(bo->tiling_flags & RADEON_TILING_SURFACE)) 466 return 0; 467 468 if (force_drop) { 469 radeon_bo_clear_surface_reg(bo); 470 return 0; 471 } 472 473 if (bo->tbo.mem.mem_type != TTM_PL_VRAM) { 474 if (!has_moved) 475 return 0; 476 477 if (bo->surface_reg >= 0) 478 radeon_bo_clear_surface_reg(bo); 479 return 0; 480 } 481 482 if ((bo->surface_reg >= 0) && !has_moved) 483 return 0; 484 485 return radeon_bo_get_surface_reg(bo); 486 } 487 488 void radeon_bo_move_notify(struct ttm_buffer_object *bo, 489 struct ttm_mem_reg *mem) 490 { 491 struct radeon_bo *rbo; 492 if (!radeon_ttm_bo_is_radeon_bo(bo)) 493 return; 494 rbo = container_of(bo, struct radeon_bo, tbo); 495 radeon_bo_check_tiling(rbo, 0, 1); 496 } 497 498 void radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo) 499 { 500 struct radeon_bo *rbo; 501 if (!radeon_ttm_bo_is_radeon_bo(bo)) 502 return; 503 rbo = container_of(bo, struct radeon_bo, tbo); 504 radeon_bo_check_tiling(rbo, 0, 0); 505 } 506