1 /* 2 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved. 3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved. 4 * 5 * This software is available to you under a choice of one of two 6 * licenses. You may choose to be licensed under the terms of the GNU 7 * General Public License (GPL) Version 2, available from the file 8 * COPYING in the main directory of this source tree, or the 9 * OpenIB.org BSD license below: 10 * 11 * Redistribution and use in source and binary forms, with or 12 * without modification, are permitted provided that the following 13 * conditions are met: 14 * 15 * - Redistributions of source code must retain the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer. 18 * 19 * - Redistributions in binary form must reproduce the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer in the documentation and/or other materials 22 * provided with the distribution. 23 * 24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 31 * SOFTWARE. 32 */ 33 34 #include <linux/slab.h> 35 #include <rdma/ib_user_verbs.h> 36 #include <rdma/iter.h> 37 38 #include "mlx4_ib.h" 39 40 static u32 convert_access(int acc) 41 { 42 return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX4_PERM_ATOMIC : 0) | 43 (acc & IB_ACCESS_REMOTE_WRITE ? MLX4_PERM_REMOTE_WRITE : 0) | 44 (acc & IB_ACCESS_REMOTE_READ ? MLX4_PERM_REMOTE_READ : 0) | 45 (acc & IB_ACCESS_LOCAL_WRITE ? MLX4_PERM_LOCAL_WRITE : 0) | 46 (acc & IB_ACCESS_MW_BIND ? MLX4_PERM_BIND_MW : 0) | 47 MLX4_PERM_LOCAL_READ; 48 } 49 50 static enum mlx4_mw_type to_mlx4_type(enum ib_mw_type type) 51 { 52 switch (type) { 53 case IB_MW_TYPE_1: return MLX4_MW_TYPE_1; 54 case IB_MW_TYPE_2: return MLX4_MW_TYPE_2; 55 default: return -1; 56 } 57 } 58 59 struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc) 60 { 61 struct mlx4_ib_mr *mr; 62 int err; 63 64 mr = kzalloc_obj(*mr); 65 if (!mr) 66 return ERR_PTR(-ENOMEM); 67 68 err = mlx4_mr_alloc(to_mdev(pd->device)->dev, to_mpd(pd)->pdn, 0, 69 ~0ull, convert_access(acc), 0, 0, &mr->mmr); 70 if (err) 71 goto err_free; 72 73 err = mlx4_mr_enable(to_mdev(pd->device)->dev, &mr->mmr); 74 if (err) 75 goto err_mr; 76 77 mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key; 78 mr->umem = NULL; 79 80 return &mr->ibmr; 81 82 err_mr: 83 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr); 84 85 err_free: 86 kfree(mr); 87 88 return ERR_PTR(err); 89 } 90 91 int mlx4_ib_umem_write_mtt(struct mlx4_ib_dev *dev, struct mlx4_mtt *mtt, 92 struct ib_umem *umem) 93 { 94 struct ib_block_iter biter; 95 int err, i = 0; 96 u64 addr; 97 98 rdma_umem_for_each_dma_block(umem, &biter, BIT(mtt->page_shift)) { 99 addr = rdma_block_iter_dma_address(&biter); 100 err = mlx4_write_mtt(dev->dev, mtt, i++, 1, &addr); 101 if (err) 102 return err; 103 } 104 return 0; 105 } 106 107 static struct ib_umem *mlx4_get_umem_mr(struct ib_device *device, u64 start, 108 u64 length, int access_flags) 109 { 110 /* 111 * Force registering the memory as writable if the underlying pages 112 * are writable. This is so rereg can change the access permissions 113 * from readable to writable without having to run through ib_umem_get_va 114 * again 115 */ 116 if (!ib_access_writable(access_flags)) { 117 unsigned long untagged_start = untagged_addr(start); 118 struct vm_area_struct *vma; 119 120 mmap_read_lock(current->mm); 121 /* 122 * FIXME: Ideally this would iterate over all the vmas that 123 * cover the memory, but for now it requires a single vma to 124 * entirely cover the MR to support RO mappings. 125 */ 126 vma = find_vma(current->mm, untagged_start); 127 if (vma && vma->vm_end >= untagged_start + length && 128 vma->vm_start <= untagged_start) { 129 if (vma->vm_flags & VM_WRITE) 130 access_flags |= IB_ACCESS_LOCAL_WRITE; 131 } else { 132 access_flags |= IB_ACCESS_LOCAL_WRITE; 133 } 134 135 mmap_read_unlock(current->mm); 136 } 137 138 return ib_umem_get_va(device, start, length, access_flags); 139 } 140 141 struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 142 u64 virt_addr, int access_flags, 143 struct ib_dmah *dmah, 144 struct ib_udata *udata) 145 { 146 struct mlx4_ib_dev *dev = to_mdev(pd->device); 147 struct mlx4_ib_mr *mr; 148 int shift; 149 int err; 150 int n; 151 152 if (dmah) 153 return ERR_PTR(-EOPNOTSUPP); 154 155 mr = kzalloc_obj(*mr); 156 if (!mr) 157 return ERR_PTR(-ENOMEM); 158 159 mr->umem = mlx4_get_umem_mr(pd->device, start, length, access_flags); 160 if (IS_ERR(mr->umem)) { 161 err = PTR_ERR(mr->umem); 162 goto err_free; 163 } 164 165 shift = mlx4_ib_umem_calc_optimal_mtt_size(mr->umem, start, &n); 166 if (shift < 0) { 167 err = shift; 168 goto err_umem; 169 } 170 171 err = mlx4_mr_alloc(dev->dev, to_mpd(pd)->pdn, virt_addr, length, 172 convert_access(access_flags), n, shift, &mr->mmr); 173 if (err) 174 goto err_umem; 175 176 err = mlx4_ib_umem_write_mtt(dev, &mr->mmr.mtt, mr->umem); 177 if (err) 178 goto err_mr; 179 180 err = mlx4_mr_enable(dev->dev, &mr->mmr); 181 if (err) 182 goto err_mr; 183 184 mr->access_flags = access_flags; 185 mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key; 186 mr->ibmr.page_size = 1U << shift; 187 188 return &mr->ibmr; 189 190 err_mr: 191 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr); 192 193 err_umem: 194 ib_umem_release(mr->umem); 195 196 err_free: 197 kfree(mr); 198 199 return ERR_PTR(err); 200 } 201 202 struct ib_mr *mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags, u64 start, 203 u64 length, u64 virt_addr, 204 int mr_access_flags, struct ib_pd *pd, 205 struct ib_udata *udata) 206 { 207 struct mlx4_ib_dev *dev = to_mdev(mr->device); 208 struct mlx4_ib_mr *mmr = to_mmr(mr); 209 struct mlx4_mpt_entry *mpt_entry; 210 struct mlx4_mpt_entry **pmpt_entry = &mpt_entry; 211 int err; 212 213 err = ib_umem_check_rereg(mmr->umem, flags, mr_access_flags); 214 if (err) 215 return ERR_PTR(err); 216 217 /* Since we synchronize this call and mlx4_ib_dereg_mr via uverbs, 218 * we assume that the calls can't run concurrently. Otherwise, a 219 * race exists. 220 */ 221 err = mlx4_mr_hw_get_mpt(dev->dev, &mmr->mmr, &pmpt_entry); 222 if (err) 223 return ERR_PTR(err); 224 225 if (flags & IB_MR_REREG_PD) { 226 err = mlx4_mr_hw_change_pd(dev->dev, *pmpt_entry, 227 to_mpd(pd)->pdn); 228 229 if (err) 230 goto release_mpt_entry; 231 } 232 233 if (flags & IB_MR_REREG_ACCESS) { 234 if (ib_access_writable(mr_access_flags) && 235 !mmr->umem->writable) { 236 err = -EPERM; 237 goto release_mpt_entry; 238 } 239 240 err = mlx4_mr_hw_change_access(dev->dev, *pmpt_entry, 241 convert_access(mr_access_flags)); 242 243 if (err) 244 goto release_mpt_entry; 245 } else { 246 mr_access_flags = mmr->access_flags; 247 } 248 249 if (flags & IB_MR_REREG_TRANS) { 250 int shift; 251 int n; 252 253 mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr); 254 ib_umem_release(mmr->umem); 255 mmr->umem = mlx4_get_umem_mr(mr->device, start, length, 256 mr_access_flags); 257 if (IS_ERR(mmr->umem)) { 258 err = PTR_ERR(mmr->umem); 259 /* Prevent mlx4_ib_dereg_mr from free'ing invalid pointer */ 260 mmr->umem = NULL; 261 goto release_mpt_entry; 262 } 263 n = ib_umem_num_dma_blocks(mmr->umem, PAGE_SIZE); 264 shift = PAGE_SHIFT; 265 266 err = mlx4_mr_rereg_mem_write(dev->dev, &mmr->mmr, 267 virt_addr, length, n, shift, 268 *pmpt_entry); 269 if (err) { 270 ib_umem_release(mmr->umem); 271 goto release_mpt_entry; 272 } 273 mmr->mmr.iova = virt_addr; 274 mmr->mmr.size = length; 275 276 err = mlx4_ib_umem_write_mtt(dev, &mmr->mmr.mtt, mmr->umem); 277 if (err) { 278 mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr); 279 ib_umem_release(mmr->umem); 280 goto release_mpt_entry; 281 } 282 } 283 284 /* If we couldn't transfer the MR to the HCA, just remember to 285 * return a failure. But dereg_mr will free the resources. 286 */ 287 err = mlx4_mr_hw_write_mpt(dev->dev, &mmr->mmr, pmpt_entry); 288 if (!err && flags & IB_MR_REREG_ACCESS) { 289 mmr->access_flags = mr_access_flags; 290 mmr->mmr.access = convert_access(mr_access_flags); 291 } 292 293 release_mpt_entry: 294 mlx4_mr_hw_put_mpt(dev->dev, pmpt_entry); 295 if (err) 296 return ERR_PTR(err); 297 return NULL; 298 } 299 300 static int 301 mlx4_alloc_priv_pages(struct ib_device *device, 302 struct mlx4_ib_mr *mr, 303 int max_pages) 304 { 305 int ret; 306 307 /* Ensure that size is aligned to DMA cacheline 308 * requirements. 309 * max_pages is limited to MLX4_MAX_FAST_REG_PAGES 310 * so page_map_size will never cross PAGE_SIZE. 311 */ 312 mr->page_map_size = roundup(max_pages * sizeof(u64), 313 MLX4_MR_PAGES_ALIGN); 314 315 /* Prevent cross page boundary allocation. */ 316 mr->pages = kzalloc(PAGE_SIZE, GFP_KERNEL); 317 if (!mr->pages) 318 return -ENOMEM; 319 320 mr->page_map = dma_map_single(device->dev.parent, mr->pages, 321 mr->page_map_size, DMA_TO_DEVICE); 322 323 if (dma_mapping_error(device->dev.parent, mr->page_map)) { 324 ret = -ENOMEM; 325 goto err; 326 } 327 328 return 0; 329 330 err: 331 kfree(mr->pages); 332 return ret; 333 } 334 335 static void 336 mlx4_free_priv_pages(struct mlx4_ib_mr *mr) 337 { 338 if (mr->pages) { 339 struct ib_device *device = mr->ibmr.device; 340 341 dma_unmap_single(device->dev.parent, mr->page_map, 342 mr->page_map_size, DMA_TO_DEVICE); 343 kfree(mr->pages); 344 mr->pages = NULL; 345 } 346 } 347 348 int mlx4_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) 349 { 350 struct mlx4_ib_mr *mr = to_mmr(ibmr); 351 int ret; 352 353 mlx4_free_priv_pages(mr); 354 355 ret = mlx4_mr_free(to_mdev(ibmr->device)->dev, &mr->mmr); 356 if (ret) 357 return ret; 358 if (mr->umem) 359 ib_umem_release(mr->umem); 360 kfree(mr); 361 362 return 0; 363 } 364 365 int mlx4_ib_alloc_mw(struct ib_mw *ibmw, struct ib_udata *udata) 366 { 367 struct mlx4_ib_dev *dev = to_mdev(ibmw->device); 368 struct mlx4_ib_mw *mw = to_mmw(ibmw); 369 int err; 370 371 err = mlx4_mw_alloc(dev->dev, to_mpd(ibmw->pd)->pdn, 372 to_mlx4_type(ibmw->type), &mw->mmw); 373 if (err) 374 return err; 375 376 err = mlx4_mw_enable(dev->dev, &mw->mmw); 377 if (err) 378 goto err_mw; 379 380 ibmw->rkey = mw->mmw.key; 381 return 0; 382 383 err_mw: 384 mlx4_mw_free(dev->dev, &mw->mmw); 385 return err; 386 } 387 388 int mlx4_ib_dealloc_mw(struct ib_mw *ibmw) 389 { 390 struct mlx4_ib_mw *mw = to_mmw(ibmw); 391 392 mlx4_mw_free(to_mdev(ibmw->device)->dev, &mw->mmw); 393 return 0; 394 } 395 396 struct ib_mr *mlx4_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type, 397 u32 max_num_sg) 398 { 399 struct mlx4_ib_dev *dev = to_mdev(pd->device); 400 struct mlx4_ib_mr *mr; 401 int err; 402 403 if (mr_type != IB_MR_TYPE_MEM_REG || 404 max_num_sg > MLX4_MAX_FAST_REG_PAGES) 405 return ERR_PTR(-EINVAL); 406 407 mr = kzalloc_obj(*mr); 408 if (!mr) 409 return ERR_PTR(-ENOMEM); 410 411 err = mlx4_mr_alloc(dev->dev, to_mpd(pd)->pdn, 0, 0, 0, 412 max_num_sg, 0, &mr->mmr); 413 if (err) 414 goto err_free; 415 416 err = mlx4_alloc_priv_pages(pd->device, mr, max_num_sg); 417 if (err) 418 goto err_free_mr; 419 420 mr->max_pages = max_num_sg; 421 err = mlx4_mr_enable(dev->dev, &mr->mmr); 422 if (err) 423 goto err_free_pl; 424 425 mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key; 426 mr->umem = NULL; 427 428 return &mr->ibmr; 429 430 err_free_pl: 431 mr->ibmr.device = pd->device; 432 mlx4_free_priv_pages(mr); 433 err_free_mr: 434 (void) mlx4_mr_free(dev->dev, &mr->mmr); 435 err_free: 436 kfree(mr); 437 return ERR_PTR(err); 438 } 439 440 static int mlx4_set_page(struct ib_mr *ibmr, u64 addr) 441 { 442 struct mlx4_ib_mr *mr = to_mmr(ibmr); 443 444 if (unlikely(mr->npages == mr->max_pages)) 445 return -ENOMEM; 446 447 mr->pages[mr->npages++] = cpu_to_be64(addr | MLX4_MTT_FLAG_PRESENT); 448 449 return 0; 450 } 451 452 int mlx4_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents, 453 unsigned int *sg_offset) 454 { 455 struct mlx4_ib_mr *mr = to_mmr(ibmr); 456 int rc; 457 458 mr->npages = 0; 459 460 ib_dma_sync_single_for_cpu(ibmr->device, mr->page_map, 461 mr->page_map_size, DMA_TO_DEVICE); 462 463 rc = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, mlx4_set_page); 464 465 ib_dma_sync_single_for_device(ibmr->device, mr->page_map, 466 mr->page_map_size, DMA_TO_DEVICE); 467 468 return rc; 469 } 470