1 /* 2 * Copyright 2010 Red Hat Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 * Authors: Ben Skeggs 23 */ 24 #include "priv.h" 25 26 #include <core/gpuobj.h> 27 #include <subdev/fb.h> 28 29 void 30 nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node) 31 { 32 struct nvkm_vm *vm = vma->vm; 33 struct nvkm_mmu *mmu = vm->mmu; 34 struct nvkm_mm_node *r = node->mem; 35 int big = vma->node->type != mmu->func->spg_shift; 36 u32 offset = vma->node->offset + (delta >> 12); 37 u32 bits = vma->node->type - 12; 38 u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; 39 u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; 40 u32 max = 1 << (mmu->func->pgt_bits - bits); 41 u32 end, len; 42 43 delta = 0; 44 while (r) { 45 u64 phys = (u64)r->offset << 12; 46 u32 num = r->length >> bits; 47 48 while (num) { 49 struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; 50 51 end = (pte + num); 52 if (unlikely(end >= max)) 53 end = max; 54 len = end - pte; 55 56 mmu->func->map(vma, pgt, node, pte, len, phys, delta); 57 58 num -= len; 59 pte += len; 60 if (unlikely(end >= max)) { 61 phys += len << (bits + 12); 62 pde++; 63 pte = 0; 64 } 65 66 delta += (u64)len << vma->node->type; 67 } 68 r = r->next; 69 }; 70 71 mmu->func->flush(vm); 72 } 73 74 static void 75 nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length, 76 struct nvkm_mem *mem) 77 { 78 struct nvkm_vm *vm = vma->vm; 79 struct nvkm_mmu *mmu = vm->mmu; 80 int big = vma->node->type != mmu->func->spg_shift; 81 u32 offset = vma->node->offset + (delta >> 12); 82 u32 bits = vma->node->type - 12; 83 u32 num = length >> vma->node->type; 84 u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; 85 u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; 86 u32 max = 1 << (mmu->func->pgt_bits - bits); 87 unsigned m, sglen; 88 u32 end, len; 89 int i; 90 struct scatterlist *sg; 91 92 for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) { 93 struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; 94 sglen = sg_dma_len(sg) >> PAGE_SHIFT; 95 96 end = pte + sglen; 97 if (unlikely(end >= max)) 98 end = max; 99 len = end - pte; 100 101 for (m = 0; m < len; m++) { 102 dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT); 103 104 mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr); 105 num--; 106 pte++; 107 108 if (num == 0) 109 goto finish; 110 } 111 if (unlikely(end >= max)) { 112 pde++; 113 pte = 0; 114 } 115 if (m < sglen) { 116 for (; m < sglen; m++) { 117 dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT); 118 119 mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr); 120 num--; 121 pte++; 122 if (num == 0) 123 goto finish; 124 } 125 } 126 127 } 128 finish: 129 mmu->func->flush(vm); 130 } 131 132 static void 133 nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length, 134 struct nvkm_mem *mem) 135 { 136 struct nvkm_vm *vm = vma->vm; 137 struct nvkm_mmu *mmu = vm->mmu; 138 dma_addr_t *list = mem->pages; 139 int big = vma->node->type != mmu->func->spg_shift; 140 u32 offset = vma->node->offset + (delta >> 12); 141 u32 bits = vma->node->type - 12; 142 u32 num = length >> vma->node->type; 143 u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; 144 u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; 145 u32 max = 1 << (mmu->func->pgt_bits - bits); 146 u32 end, len; 147 148 while (num) { 149 struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; 150 151 end = (pte + num); 152 if (unlikely(end >= max)) 153 end = max; 154 len = end - pte; 155 156 mmu->func->map_sg(vma, pgt, mem, pte, len, list); 157 158 num -= len; 159 pte += len; 160 list += len; 161 if (unlikely(end >= max)) { 162 pde++; 163 pte = 0; 164 } 165 } 166 167 mmu->func->flush(vm); 168 } 169 170 void 171 nvkm_vm_map(struct nvkm_vma *vma, struct nvkm_mem *node) 172 { 173 if (node->sg) 174 nvkm_vm_map_sg_table(vma, 0, node->size << 12, node); 175 else 176 if (node->pages) 177 nvkm_vm_map_sg(vma, 0, node->size << 12, node); 178 else 179 nvkm_vm_map_at(vma, 0, node); 180 } 181 182 void 183 nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length) 184 { 185 struct nvkm_vm *vm = vma->vm; 186 struct nvkm_mmu *mmu = vm->mmu; 187 int big = vma->node->type != mmu->func->spg_shift; 188 u32 offset = vma->node->offset + (delta >> 12); 189 u32 bits = vma->node->type - 12; 190 u32 num = length >> vma->node->type; 191 u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; 192 u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; 193 u32 max = 1 << (mmu->func->pgt_bits - bits); 194 u32 end, len; 195 196 while (num) { 197 struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; 198 199 end = (pte + num); 200 if (unlikely(end >= max)) 201 end = max; 202 len = end - pte; 203 204 mmu->func->unmap(vma, pgt, pte, len); 205 206 num -= len; 207 pte += len; 208 if (unlikely(end >= max)) { 209 pde++; 210 pte = 0; 211 } 212 } 213 214 mmu->func->flush(vm); 215 } 216 217 void 218 nvkm_vm_unmap(struct nvkm_vma *vma) 219 { 220 nvkm_vm_unmap_at(vma, 0, (u64)vma->node->length << 12); 221 } 222 223 static void 224 nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde) 225 { 226 struct nvkm_mmu *mmu = vm->mmu; 227 struct nvkm_vm_pgd *vpgd; 228 struct nvkm_vm_pgt *vpgt; 229 struct nvkm_memory *pgt; 230 u32 pde; 231 232 for (pde = fpde; pde <= lpde; pde++) { 233 vpgt = &vm->pgt[pde - vm->fpde]; 234 if (--vpgt->refcount[big]) 235 continue; 236 237 pgt = vpgt->mem[big]; 238 vpgt->mem[big] = NULL; 239 240 list_for_each_entry(vpgd, &vm->pgd_list, head) { 241 mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem); 242 } 243 244 mmu->func->flush(vm); 245 246 nvkm_memory_del(&pgt); 247 } 248 } 249 250 static int 251 nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type) 252 { 253 struct nvkm_mmu *mmu = vm->mmu; 254 struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde]; 255 struct nvkm_vm_pgd *vpgd; 256 int big = (type != mmu->func->spg_shift); 257 u32 pgt_size; 258 int ret; 259 260 pgt_size = (1 << (mmu->func->pgt_bits + 12)) >> type; 261 pgt_size *= 8; 262 263 ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST, 264 pgt_size, 0x1000, true, &vpgt->mem[big]); 265 if (unlikely(ret)) 266 return ret; 267 268 list_for_each_entry(vpgd, &vm->pgd_list, head) { 269 mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem); 270 } 271 272 vpgt->refcount[big]++; 273 return 0; 274 } 275 276 int 277 nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access, 278 struct nvkm_vma *vma) 279 { 280 struct nvkm_mmu *mmu = vm->mmu; 281 u32 align = (1 << page_shift) >> 12; 282 u32 msize = size >> 12; 283 u32 fpde, lpde, pde; 284 int ret; 285 286 mutex_lock(&vm->mutex); 287 ret = nvkm_mm_head(&vm->mm, 0, page_shift, msize, msize, align, 288 &vma->node); 289 if (unlikely(ret != 0)) { 290 mutex_unlock(&vm->mutex); 291 return ret; 292 } 293 294 fpde = (vma->node->offset >> mmu->func->pgt_bits); 295 lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits; 296 297 for (pde = fpde; pde <= lpde; pde++) { 298 struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde]; 299 int big = (vma->node->type != mmu->func->spg_shift); 300 301 if (likely(vpgt->refcount[big])) { 302 vpgt->refcount[big]++; 303 continue; 304 } 305 306 ret = nvkm_vm_map_pgt(vm, pde, vma->node->type); 307 if (ret) { 308 if (pde != fpde) 309 nvkm_vm_unmap_pgt(vm, big, fpde, pde - 1); 310 nvkm_mm_free(&vm->mm, &vma->node); 311 mutex_unlock(&vm->mutex); 312 return ret; 313 } 314 } 315 mutex_unlock(&vm->mutex); 316 317 vma->vm = NULL; 318 nvkm_vm_ref(vm, &vma->vm, NULL); 319 vma->offset = (u64)vma->node->offset << 12; 320 vma->access = access; 321 return 0; 322 } 323 324 void 325 nvkm_vm_put(struct nvkm_vma *vma) 326 { 327 struct nvkm_mmu *mmu; 328 struct nvkm_vm *vm; 329 u32 fpde, lpde; 330 331 if (unlikely(vma->node == NULL)) 332 return; 333 vm = vma->vm; 334 mmu = vm->mmu; 335 336 fpde = (vma->node->offset >> mmu->func->pgt_bits); 337 lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits; 338 339 mutex_lock(&vm->mutex); 340 nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->func->spg_shift, fpde, lpde); 341 nvkm_mm_free(&vm->mm, &vma->node); 342 mutex_unlock(&vm->mutex); 343 344 nvkm_vm_ref(NULL, &vma->vm, NULL); 345 } 346 347 int 348 nvkm_vm_boot(struct nvkm_vm *vm, u64 size) 349 { 350 struct nvkm_mmu *mmu = vm->mmu; 351 struct nvkm_memory *pgt; 352 int ret; 353 354 ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST, 355 (size >> mmu->func->spg_shift) * 8, 0x1000, true, &pgt); 356 if (ret == 0) { 357 vm->pgt[0].refcount[0] = 1; 358 vm->pgt[0].mem[0] = pgt; 359 nvkm_memory_boot(pgt, vm); 360 } 361 362 return ret; 363 } 364 365 int 366 nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset, 367 u32 block, struct lock_class_key *key, struct nvkm_vm **pvm) 368 { 369 static struct lock_class_key _key; 370 struct nvkm_vm *vm; 371 u64 mm_length = (offset + length) - mm_offset; 372 int ret; 373 374 vm = kzalloc(sizeof(*vm), GFP_KERNEL); 375 if (!vm) 376 return -ENOMEM; 377 378 __mutex_init(&vm->mutex, "&vm->mutex", key ? key : &_key); 379 INIT_LIST_HEAD(&vm->pgd_list); 380 vm->mmu = mmu; 381 kref_init(&vm->refcount); 382 vm->fpde = offset >> (mmu->func->pgt_bits + 12); 383 vm->lpde = (offset + length - 1) >> (mmu->func->pgt_bits + 12); 384 385 vm->pgt = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt)); 386 if (!vm->pgt) { 387 kfree(vm); 388 return -ENOMEM; 389 } 390 391 ret = nvkm_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12, 392 block >> 12); 393 if (ret) { 394 vfree(vm->pgt); 395 kfree(vm); 396 return ret; 397 } 398 399 *pvm = vm; 400 401 return 0; 402 } 403 404 int 405 nvkm_vm_new(struct nvkm_device *device, u64 offset, u64 length, u64 mm_offset, 406 struct lock_class_key *key, struct nvkm_vm **pvm) 407 { 408 struct nvkm_mmu *mmu = device->mmu; 409 if (!mmu->func->create) 410 return -EINVAL; 411 return mmu->func->create(mmu, offset, length, mm_offset, key, pvm); 412 } 413 414 static int 415 nvkm_vm_link(struct nvkm_vm *vm, struct nvkm_gpuobj *pgd) 416 { 417 struct nvkm_mmu *mmu = vm->mmu; 418 struct nvkm_vm_pgd *vpgd; 419 int i; 420 421 if (!pgd) 422 return 0; 423 424 vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL); 425 if (!vpgd) 426 return -ENOMEM; 427 428 vpgd->obj = pgd; 429 430 mutex_lock(&vm->mutex); 431 for (i = vm->fpde; i <= vm->lpde; i++) 432 mmu->func->map_pgt(pgd, i, vm->pgt[i - vm->fpde].mem); 433 list_add(&vpgd->head, &vm->pgd_list); 434 mutex_unlock(&vm->mutex); 435 return 0; 436 } 437 438 static void 439 nvkm_vm_unlink(struct nvkm_vm *vm, struct nvkm_gpuobj *mpgd) 440 { 441 struct nvkm_vm_pgd *vpgd, *tmp; 442 443 if (!mpgd) 444 return; 445 446 mutex_lock(&vm->mutex); 447 list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) { 448 if (vpgd->obj == mpgd) { 449 list_del(&vpgd->head); 450 kfree(vpgd); 451 break; 452 } 453 } 454 mutex_unlock(&vm->mutex); 455 } 456 457 static void 458 nvkm_vm_del(struct kref *kref) 459 { 460 struct nvkm_vm *vm = container_of(kref, typeof(*vm), refcount); 461 struct nvkm_vm_pgd *vpgd, *tmp; 462 463 list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) { 464 nvkm_vm_unlink(vm, vpgd->obj); 465 } 466 467 nvkm_mm_fini(&vm->mm); 468 vfree(vm->pgt); 469 kfree(vm); 470 } 471 472 int 473 nvkm_vm_ref(struct nvkm_vm *ref, struct nvkm_vm **ptr, struct nvkm_gpuobj *pgd) 474 { 475 if (ref) { 476 int ret = nvkm_vm_link(ref, pgd); 477 if (ret) 478 return ret; 479 480 kref_get(&ref->refcount); 481 } 482 483 if (*ptr) { 484 nvkm_vm_unlink(*ptr, pgd); 485 kref_put(&(*ptr)->refcount, nvkm_vm_del); 486 } 487 488 *ptr = ref; 489 return 0; 490 } 491 492 static int 493 nvkm_mmu_oneinit(struct nvkm_subdev *subdev) 494 { 495 struct nvkm_mmu *mmu = nvkm_mmu(subdev); 496 if (mmu->func->oneinit) 497 return mmu->func->oneinit(mmu); 498 return 0; 499 } 500 501 static int 502 nvkm_mmu_init(struct nvkm_subdev *subdev) 503 { 504 struct nvkm_mmu *mmu = nvkm_mmu(subdev); 505 if (mmu->func->init) 506 mmu->func->init(mmu); 507 return 0; 508 } 509 510 static void * 511 nvkm_mmu_dtor(struct nvkm_subdev *subdev) 512 { 513 struct nvkm_mmu *mmu = nvkm_mmu(subdev); 514 if (mmu->func->dtor) 515 return mmu->func->dtor(mmu); 516 return mmu; 517 } 518 519 static const struct nvkm_subdev_func 520 nvkm_mmu = { 521 .dtor = nvkm_mmu_dtor, 522 .oneinit = nvkm_mmu_oneinit, 523 .init = nvkm_mmu_init, 524 }; 525 526 void 527 nvkm_mmu_ctor(const struct nvkm_mmu_func *func, struct nvkm_device *device, 528 int index, struct nvkm_mmu *mmu) 529 { 530 nvkm_subdev_ctor(&nvkm_mmu, device, index, &mmu->subdev); 531 mmu->func = func; 532 mmu->limit = func->limit; 533 mmu->dma_bits = func->dma_bits; 534 mmu->lpg_shift = func->lpg_shift; 535 } 536 537 int 538 nvkm_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device, 539 int index, struct nvkm_mmu **pmmu) 540 { 541 if (!(*pmmu = kzalloc(sizeof(**pmmu), GFP_KERNEL))) 542 return -ENOMEM; 543 nvkm_mmu_ctor(func, device, index, *pmmu); 544 return 0; 545 } 546