drivers/staging/rdma/hfi1/mr.c

   1 /*
   2  *
   3  * This file is provided under a dual BSD/GPLv2 license.  When using or
   4  * redistributing this file, you may do so under either license.
   5  *
   6  * GPL LICENSE SUMMARY
   7  *
   8  * Copyright(c) 2015 Intel Corporation.
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of version 2 of the GNU General Public License as
  12  * published by the Free Software Foundation.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  *
  19  * BSD LICENSE
  20  *
  21  * Copyright(c) 2015 Intel Corporation.
  22  *
  23  * Redistribution and use in source and binary forms, with or without
  24  * modification, are permitted provided that the following conditions
  25  * are met:
  26  *
  27  *  - Redistributions of source code must retain the above copyright
  28  *    notice, this list of conditions and the following disclaimer.
  29  *  - Redistributions in binary form must reproduce the above copyright
  30  *    notice, this list of conditions and the following disclaimer in
  31  *    the documentation and/or other materials provided with the
  32  *    distribution.
  33  *  - Neither the name of Intel Corporation nor the names of its
  34  *    contributors may be used to endorse or promote products derived
  35  *    from this software without specific prior written permission.
  36  *
  37  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  38  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  39  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  40  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  41  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  42  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  43  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  44  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  45  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  46  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  47  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  48  *
  49  */
  50
  51 #include <rdma/ib_umem.h>
  52 #include <rdma/ib_smi.h>
  53
  54 #include "hfi.h"
  55
  56 /* Fast memory region */
  57 struct hfi1_fmr {
  58         struct ib_fmr ibfmr;
  59         struct hfi1_mregion mr;        /* must be last */
  60 };
  61
  62 static inline struct hfi1_fmr *to_ifmr(struct ib_fmr *ibfmr)
  63 {
  64         return container_of(ibfmr, struct hfi1_fmr, ibfmr);
  65 }
  66
  67 static int init_mregion(struct hfi1_mregion *mr, struct ib_pd *pd,
  68                         int count)
  69 {
  70         int m, i = 0;
  71         int rval = 0;
  72
  73         m = (count + HFI1_SEGSZ - 1) / HFI1_SEGSZ;
  74         for (; i < m; i++) {
  75                 mr->map[i] = kzalloc(sizeof(*mr->map[0]), GFP_KERNEL);
  76                 if (!mr->map[i])
  77                         goto bail;
  78         }
  79         mr->mapsz = m;
  80         init_completion(&mr->comp);
  81         /* count returning the ptr to user */
  82         atomic_set(&mr->refcount, 1);
  83         mr->pd = pd;
  84         mr->max_segs = count;
  85 out:
  86         return rval;
  87 bail:
  88         while (i)
  89                 kfree(mr->map[--i]);
  90         rval = -ENOMEM;
  91         goto out;
  92 }
  93
  94 static void deinit_mregion(struct hfi1_mregion *mr)
  95 {
  96         int i = mr->mapsz;
  97
  98         mr->mapsz = 0;
  99         while (i)
 100                 kfree(mr->map[--i]);
 101 }
 102
 103
 104 /**
 105  * hfi1_get_dma_mr - get a DMA memory region
 106  * @pd: protection domain for this memory region
 107  * @acc: access flags
 108  *
 109  * Returns the memory region on success, otherwise returns an errno.
 110  * Note that all DMA addresses should be created via the
 111  * struct ib_dma_mapping_ops functions (see dma.c).
 112  */
 113 struct ib_mr *hfi1_get_dma_mr(struct ib_pd *pd, int acc)
 114 {
 115         struct hfi1_mr *mr = NULL;
 116         struct ib_mr *ret;
 117         int rval;
 118
 119         if (to_ipd(pd)->user) {
 120                 ret = ERR_PTR(-EPERM);
 121                 goto bail;
 122         }
 123
 124         mr = kzalloc(sizeof(*mr), GFP_KERNEL);
 125         if (!mr) {
 126                 ret = ERR_PTR(-ENOMEM);
 127                 goto bail;
 128         }
 129
 130         rval = init_mregion(&mr->mr, pd, 0);
 131         if (rval) {
 132                 ret = ERR_PTR(rval);
 133                 goto bail;
 134         }
 135
 136
 137         rval = hfi1_alloc_lkey(&mr->mr, 1);
 138         if (rval) {
 139                 ret = ERR_PTR(rval);
 140                 goto bail_mregion;
 141         }
 142
 143         mr->mr.access_flags = acc;
 144         ret = &mr->ibmr;
 145 done:
 146         return ret;
 147
 148 bail_mregion:
 149         deinit_mregion(&mr->mr);
 150 bail:
 151         kfree(mr);
 152         goto done;
 153 }
 154
 155 static struct hfi1_mr *alloc_mr(int count, struct ib_pd *pd)
 156 {
 157         struct hfi1_mr *mr;
 158         int rval = -ENOMEM;
 159         int m;
 160
 161         /* Allocate struct plus pointers to first level page tables. */
 162         m = (count + HFI1_SEGSZ - 1) / HFI1_SEGSZ;
 163         mr = kzalloc(sizeof(*mr) + m * sizeof(mr->mr.map[0]), GFP_KERNEL);
 164         if (!mr)
 165                 goto bail;
 166
 167         rval = init_mregion(&mr->mr, pd, count);
 168         if (rval)
 169                 goto bail;
 170         /*
 171          * ib_reg_phys_mr() will initialize mr->ibmr except for
 172          * lkey and rkey.
 173          */
 174         rval = hfi1_alloc_lkey(&mr->mr, 0);
 175         if (rval)
 176                 goto bail_mregion;
 177         mr->ibmr.lkey = mr->mr.lkey;
 178         mr->ibmr.rkey = mr->mr.lkey;
 179 done:
 180         return mr;
 181
 182 bail_mregion:
 183         deinit_mregion(&mr->mr);
 184 bail:
 185         kfree(mr);
 186         mr = ERR_PTR(rval);
 187         goto done;
 188 }
 189
 190 /**
 191  * hfi1_reg_phys_mr - register a physical memory region
 192  * @pd: protection domain for this memory region
 193  * @buffer_list: pointer to the list of physical buffers to register
 194  * @num_phys_buf: the number of physical buffers to register
 195  * @iova_start: the starting address passed over IB which maps to this MR
 196  *
 197  * Returns the memory region on success, otherwise returns an errno.
 198  */
 199 struct ib_mr *hfi1_reg_phys_mr(struct ib_pd *pd,
 200                                struct ib_phys_buf *buffer_list,
 201                                int num_phys_buf, int acc, u64 *iova_start)
 202 {
 203         struct hfi1_mr *mr;
 204         int n, m, i;
 205         struct ib_mr *ret;
 206
 207         mr = alloc_mr(num_phys_buf, pd);
 208         if (IS_ERR(mr)) {
 209                 ret = (struct ib_mr *)mr;
 210                 goto bail;
 211         }
 212
 213         mr->mr.user_base = *iova_start;
 214         mr->mr.iova = *iova_start;
 215         mr->mr.access_flags = acc;
 216
 217         m = 0;
 218         n = 0;
 219         for (i = 0; i < num_phys_buf; i++) {
 220                 mr->mr.map[m]->segs[n].vaddr = (void *) buffer_list[i].addr;
 221                 mr->mr.map[m]->segs[n].length = buffer_list[i].size;
 222                 mr->mr.length += buffer_list[i].size;
 223                 n++;
 224                 if (n == HFI1_SEGSZ) {
 225                         m++;
 226                         n = 0;
 227                 }
 228         }
 229
 230         ret = &mr->ibmr;
 231
 232 bail:
 233         return ret;
 234 }
 235
 236 /**
 237  * hfi1_reg_user_mr - register a userspace memory region
 238  * @pd: protection domain for this memory region
 239  * @start: starting userspace address
 240  * @length: length of region to register
 241  * @mr_access_flags: access flags for this memory region
 242  * @udata: unused by the driver
 243  *
 244  * Returns the memory region on success, otherwise returns an errno.
 245  */
 246 struct ib_mr *hfi1_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 247                                u64 virt_addr, int mr_access_flags,
 248                                struct ib_udata *udata)
 249 {
 250         struct hfi1_mr *mr;
 251         struct ib_umem *umem;
 252         struct scatterlist *sg;
 253         int n, m, entry;
 254         struct ib_mr *ret;
 255
 256         if (length == 0) {
 257                 ret = ERR_PTR(-EINVAL);
 258                 goto bail;
 259         }
 260
 261         umem = ib_umem_get(pd->uobject->context, start, length,
 262                            mr_access_flags, 0);
 263         if (IS_ERR(umem))
 264                 return (void *) umem;
 265
 266         n = umem->nmap;
 267
 268         mr = alloc_mr(n, pd);
 269         if (IS_ERR(mr)) {
 270                 ret = (struct ib_mr *)mr;
 271                 ib_umem_release(umem);
 272                 goto bail;
 273         }
 274
 275         mr->mr.user_base = start;
 276         mr->mr.iova = virt_addr;
 277         mr->mr.length = length;
 278         mr->mr.offset = ib_umem_offset(umem);
 279         mr->mr.access_flags = mr_access_flags;
 280         mr->umem = umem;
 281
 282         if (is_power_of_2(umem->page_size))
 283                 mr->mr.page_shift = ilog2(umem->page_size);
 284         m = 0;
 285         n = 0;
 286         for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
 287                         void *vaddr;
 288
 289                         vaddr = page_address(sg_page(sg));
 290                         if (!vaddr) {
 291                                 ret = ERR_PTR(-EINVAL);
 292                                 goto bail;
 293                         }
 294                         mr->mr.map[m]->segs[n].vaddr = vaddr;
 295                         mr->mr.map[m]->segs[n].length = umem->page_size;
 296                         n++;
 297                         if (n == HFI1_SEGSZ) {
 298                                 m++;
 299                                 n = 0;
 300                         }
 301         }
 302         ret = &mr->ibmr;
 303
 304 bail:
 305         return ret;
 306 }
 307
 308 /**
 309  * hfi1_dereg_mr - unregister and free a memory region
 310  * @ibmr: the memory region to free
 311  *
 312  * Returns 0 on success.
 313  *
 314  * Note that this is called to free MRs created by hfi1_get_dma_mr()
 315  * or hfi1_reg_user_mr().
 316  */
 317 int hfi1_dereg_mr(struct ib_mr *ibmr)
 318 {
 319         struct hfi1_mr *mr = to_imr(ibmr);
 320         int ret = 0;
 321         unsigned long timeout;
 322
 323         hfi1_free_lkey(&mr->mr);
 324
 325         hfi1_put_mr(&mr->mr); /* will set completion if last */
 326         timeout = wait_for_completion_timeout(&mr->mr.comp,
 327                 5 * HZ);
 328         if (!timeout) {
 329                 dd_dev_err(
 330                         dd_from_ibdev(mr->mr.pd->device),
 331                         "hfi1_dereg_mr timeout mr %p pd %p refcount %u\n",
 332                         mr, mr->mr.pd, atomic_read(&mr->mr.refcount));
 333                 hfi1_get_mr(&mr->mr);
 334                 ret = -EBUSY;
 335                 goto out;
 336         }
 337         deinit_mregion(&mr->mr);
 338         if (mr->umem)
 339                 ib_umem_release(mr->umem);
 340         kfree(mr);
 341 out:
 342         return ret;
 343 }
 344
 345 /*
 346  * Allocate a memory region usable with the
 347  * IB_WR_FAST_REG_MR send work request.
 348  *
 349  * Return the memory region on success, otherwise return an errno.
 350  */
 351 struct ib_mr *hfi1_alloc_mr(struct ib_pd *pd,
 352                             enum ib_mr_type mr_type,
 353                             u32 max_num_sg)
 354 {
 355         struct hfi1_mr *mr;
 356
 357         if (mr_type != IB_MR_TYPE_MEM_REG)
 358                 return ERR_PTR(-EINVAL);
 359
 360         mr = alloc_mr(max_num_sg, pd);
 361         if (IS_ERR(mr))
 362                 return (struct ib_mr *)mr;
 363
 364         return &mr->ibmr;
 365 }
 366
 367 struct ib_fast_reg_page_list *
 368 hfi1_alloc_fast_reg_page_list(struct ib_device *ibdev, int page_list_len)
 369 {
 370         unsigned size = page_list_len * sizeof(u64);
 371         struct ib_fast_reg_page_list *pl;
 372
 373         if (size > PAGE_SIZE)
 374                 return ERR_PTR(-EINVAL);
 375
 376         pl = kzalloc(sizeof(*pl), GFP_KERNEL);
 377         if (!pl)
 378                 return ERR_PTR(-ENOMEM);
 379
 380         pl->page_list = kzalloc(size, GFP_KERNEL);
 381         if (!pl->page_list)
 382                 goto err_free;
 383
 384         return pl;
 385
 386 err_free:
 387         kfree(pl);
 388         return ERR_PTR(-ENOMEM);
 389 }
 390
 391 void hfi1_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl)
 392 {
 393         kfree(pl->page_list);
 394         kfree(pl);
 395 }
 396
 397 /**
 398  * hfi1_alloc_fmr - allocate a fast memory region
 399  * @pd: the protection domain for this memory region
 400  * @mr_access_flags: access flags for this memory region
 401  * @fmr_attr: fast memory region attributes
 402  *
 403  * Returns the memory region on success, otherwise returns an errno.
 404  */
 405 struct ib_fmr *hfi1_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
 406                               struct ib_fmr_attr *fmr_attr)
 407 {
 408         struct hfi1_fmr *fmr;
 409         int m;
 410         struct ib_fmr *ret;
 411         int rval = -ENOMEM;
 412
 413         /* Allocate struct plus pointers to first level page tables. */
 414         m = (fmr_attr->max_pages + HFI1_SEGSZ - 1) / HFI1_SEGSZ;
 415         fmr = kzalloc(sizeof(*fmr) + m * sizeof(fmr->mr.map[0]), GFP_KERNEL);
 416         if (!fmr)
 417                 goto bail;
 418
 419         rval = init_mregion(&fmr->mr, pd, fmr_attr->max_pages);
 420         if (rval)
 421                 goto bail;
 422
 423         /*
 424          * ib_alloc_fmr() will initialize fmr->ibfmr except for lkey &
 425          * rkey.
 426          */
 427         rval = hfi1_alloc_lkey(&fmr->mr, 0);
 428         if (rval)
 429                 goto bail_mregion;
 430         fmr->ibfmr.rkey = fmr->mr.lkey;
 431         fmr->ibfmr.lkey = fmr->mr.lkey;
 432         /*
 433          * Resources are allocated but no valid mapping (RKEY can't be
 434          * used).
 435          */
 436         fmr->mr.access_flags = mr_access_flags;
 437         fmr->mr.max_segs = fmr_attr->max_pages;
 438         fmr->mr.page_shift = fmr_attr->page_shift;
 439
 440         ret = &fmr->ibfmr;
 441 done:
 442         return ret;
 443
 444 bail_mregion:
 445         deinit_mregion(&fmr->mr);
 446 bail:
 447         kfree(fmr);
 448         ret = ERR_PTR(rval);
 449         goto done;
 450 }
 451
 452 /**
 453  * hfi1_map_phys_fmr - set up a fast memory region
 454  * @ibmfr: the fast memory region to set up
 455  * @page_list: the list of pages to associate with the fast memory region
 456  * @list_len: the number of pages to associate with the fast memory region
 457  * @iova: the virtual address of the start of the fast memory region
 458  *
 459  * This may be called from interrupt context.
 460  */
 461
 462 int hfi1_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
 463                       int list_len, u64 iova)
 464 {
 465         struct hfi1_fmr *fmr = to_ifmr(ibfmr);
 466         struct hfi1_lkey_table *rkt;
 467         unsigned long flags;
 468         int m, n, i;
 469         u32 ps;
 470         int ret;
 471
 472         i = atomic_read(&fmr->mr.refcount);
 473         if (i > 2)
 474                 return -EBUSY;
 475
 476         if (list_len > fmr->mr.max_segs) {
 477                 ret = -EINVAL;
 478                 goto bail;
 479         }
 480         rkt = &to_idev(ibfmr->device)->lk_table;
 481         spin_lock_irqsave(&rkt->lock, flags);
 482         fmr->mr.user_base = iova;
 483         fmr->mr.iova = iova;
 484         ps = 1 << fmr->mr.page_shift;
 485         fmr->mr.length = list_len * ps;
 486         m = 0;
 487         n = 0;
 488         for (i = 0; i < list_len; i++) {
 489                 fmr->mr.map[m]->segs[n].vaddr = (void *) page_list[i];
 490                 fmr->mr.map[m]->segs[n].length = ps;
 491                 if (++n == HFI1_SEGSZ) {
 492                         m++;
 493                         n = 0;
 494                 }
 495         }
 496         spin_unlock_irqrestore(&rkt->lock, flags);
 497         ret = 0;
 498
 499 bail:
 500         return ret;
 501 }
 502
 503 /**
 504  * hfi1_unmap_fmr - unmap fast memory regions
 505  * @fmr_list: the list of fast memory regions to unmap
 506  *
 507  * Returns 0 on success.
 508  */
 509 int hfi1_unmap_fmr(struct list_head *fmr_list)
 510 {
 511         struct hfi1_fmr *fmr;
 512         struct hfi1_lkey_table *rkt;
 513         unsigned long flags;
 514
 515         list_for_each_entry(fmr, fmr_list, ibfmr.list) {
 516                 rkt = &to_idev(fmr->ibfmr.device)->lk_table;
 517                 spin_lock_irqsave(&rkt->lock, flags);
 518                 fmr->mr.user_base = 0;
 519                 fmr->mr.iova = 0;
 520                 fmr->mr.length = 0;
 521                 spin_unlock_irqrestore(&rkt->lock, flags);
 522         }
 523         return 0;
 524 }
 525
 526 /**
 527  * hfi1_dealloc_fmr - deallocate a fast memory region
 528  * @ibfmr: the fast memory region to deallocate
 529  *
 530  * Returns 0 on success.
 531  */
 532 int hfi1_dealloc_fmr(struct ib_fmr *ibfmr)
 533 {
 534         struct hfi1_fmr *fmr = to_ifmr(ibfmr);
 535         int ret = 0;
 536         unsigned long timeout;
 537
 538         hfi1_free_lkey(&fmr->mr);
 539         hfi1_put_mr(&fmr->mr); /* will set completion if last */
 540         timeout = wait_for_completion_timeout(&fmr->mr.comp,
 541                 5 * HZ);
 542         if (!timeout) {
 543                 hfi1_get_mr(&fmr->mr);
 544                 ret = -EBUSY;
 545                 goto out;
 546         }
 547         deinit_mregion(&fmr->mr);
 548         kfree(fmr);
 549 out:
 550         return ret;
 551 }