Mike Marciniszyn | 7724105 | 2015-07-30 15:17:43 -0400 | [diff] [blame] | 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 <linux/err.h> |
| 52 | #include <linux/slab.h> |
| 53 | #include <linux/vmalloc.h> |
| 54 | |
| 55 | #include "verbs.h" |
| 56 | |
| 57 | /** |
| 58 | * hfi1_post_srq_receive - post a receive on a shared receive queue |
| 59 | * @ibsrq: the SRQ to post the receive on |
| 60 | * @wr: the list of work requests to post |
| 61 | * @bad_wr: A pointer to the first WR to cause a problem is put here |
| 62 | * |
| 63 | * This may be called from interrupt context. |
| 64 | */ |
| 65 | int hfi1_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr, |
| 66 | struct ib_recv_wr **bad_wr) |
| 67 | { |
| 68 | struct hfi1_srq *srq = to_isrq(ibsrq); |
| 69 | struct hfi1_rwq *wq; |
| 70 | unsigned long flags; |
| 71 | int ret; |
| 72 | |
| 73 | for (; wr; wr = wr->next) { |
| 74 | struct hfi1_rwqe *wqe; |
| 75 | u32 next; |
| 76 | int i; |
| 77 | |
| 78 | if ((unsigned) wr->num_sge > srq->rq.max_sge) { |
| 79 | *bad_wr = wr; |
| 80 | ret = -EINVAL; |
| 81 | goto bail; |
| 82 | } |
| 83 | |
| 84 | spin_lock_irqsave(&srq->rq.lock, flags); |
| 85 | wq = srq->rq.wq; |
| 86 | next = wq->head + 1; |
| 87 | if (next >= srq->rq.size) |
| 88 | next = 0; |
| 89 | if (next == wq->tail) { |
| 90 | spin_unlock_irqrestore(&srq->rq.lock, flags); |
| 91 | *bad_wr = wr; |
| 92 | ret = -ENOMEM; |
| 93 | goto bail; |
| 94 | } |
| 95 | |
| 96 | wqe = get_rwqe_ptr(&srq->rq, wq->head); |
| 97 | wqe->wr_id = wr->wr_id; |
| 98 | wqe->num_sge = wr->num_sge; |
| 99 | for (i = 0; i < wr->num_sge; i++) |
| 100 | wqe->sg_list[i] = wr->sg_list[i]; |
| 101 | /* Make sure queue entry is written before the head index. */ |
| 102 | smp_wmb(); |
| 103 | wq->head = next; |
| 104 | spin_unlock_irqrestore(&srq->rq.lock, flags); |
| 105 | } |
| 106 | ret = 0; |
| 107 | |
| 108 | bail: |
| 109 | return ret; |
| 110 | } |
| 111 | |
| 112 | /** |
| 113 | * hfi1_create_srq - create a shared receive queue |
| 114 | * @ibpd: the protection domain of the SRQ to create |
| 115 | * @srq_init_attr: the attributes of the SRQ |
| 116 | * @udata: data from libibverbs when creating a user SRQ |
| 117 | */ |
| 118 | struct ib_srq *hfi1_create_srq(struct ib_pd *ibpd, |
| 119 | struct ib_srq_init_attr *srq_init_attr, |
| 120 | struct ib_udata *udata) |
| 121 | { |
| 122 | struct hfi1_ibdev *dev = to_idev(ibpd->device); |
| 123 | struct hfi1_srq *srq; |
| 124 | u32 sz; |
| 125 | struct ib_srq *ret; |
| 126 | |
| 127 | if (srq_init_attr->srq_type != IB_SRQT_BASIC) { |
| 128 | ret = ERR_PTR(-ENOSYS); |
| 129 | goto done; |
| 130 | } |
| 131 | |
| 132 | if (srq_init_attr->attr.max_sge == 0 || |
| 133 | srq_init_attr->attr.max_sge > hfi1_max_srq_sges || |
| 134 | srq_init_attr->attr.max_wr == 0 || |
| 135 | srq_init_attr->attr.max_wr > hfi1_max_srq_wrs) { |
| 136 | ret = ERR_PTR(-EINVAL); |
| 137 | goto done; |
| 138 | } |
| 139 | |
| 140 | srq = kmalloc(sizeof(*srq), GFP_KERNEL); |
| 141 | if (!srq) { |
| 142 | ret = ERR_PTR(-ENOMEM); |
| 143 | goto done; |
| 144 | } |
| 145 | |
| 146 | /* |
| 147 | * Need to use vmalloc() if we want to support large #s of entries. |
| 148 | */ |
| 149 | srq->rq.size = srq_init_attr->attr.max_wr + 1; |
| 150 | srq->rq.max_sge = srq_init_attr->attr.max_sge; |
| 151 | sz = sizeof(struct ib_sge) * srq->rq.max_sge + |
| 152 | sizeof(struct hfi1_rwqe); |
| 153 | srq->rq.wq = vmalloc_user(sizeof(struct hfi1_rwq) + srq->rq.size * sz); |
| 154 | if (!srq->rq.wq) { |
| 155 | ret = ERR_PTR(-ENOMEM); |
| 156 | goto bail_srq; |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * Return the address of the RWQ as the offset to mmap. |
| 161 | * See hfi1_mmap() for details. |
| 162 | */ |
| 163 | if (udata && udata->outlen >= sizeof(__u64)) { |
| 164 | int err; |
| 165 | u32 s = sizeof(struct hfi1_rwq) + srq->rq.size * sz; |
| 166 | |
| 167 | srq->ip = |
| 168 | hfi1_create_mmap_info(dev, s, ibpd->uobject->context, |
| 169 | srq->rq.wq); |
| 170 | if (!srq->ip) { |
| 171 | ret = ERR_PTR(-ENOMEM); |
| 172 | goto bail_wq; |
| 173 | } |
| 174 | |
| 175 | err = ib_copy_to_udata(udata, &srq->ip->offset, |
| 176 | sizeof(srq->ip->offset)); |
| 177 | if (err) { |
| 178 | ret = ERR_PTR(err); |
| 179 | goto bail_ip; |
| 180 | } |
| 181 | } else |
| 182 | srq->ip = NULL; |
| 183 | |
| 184 | /* |
| 185 | * ib_create_srq() will initialize srq->ibsrq. |
| 186 | */ |
| 187 | spin_lock_init(&srq->rq.lock); |
| 188 | srq->rq.wq->head = 0; |
| 189 | srq->rq.wq->tail = 0; |
| 190 | srq->limit = srq_init_attr->attr.srq_limit; |
| 191 | |
| 192 | spin_lock(&dev->n_srqs_lock); |
| 193 | if (dev->n_srqs_allocated == hfi1_max_srqs) { |
| 194 | spin_unlock(&dev->n_srqs_lock); |
| 195 | ret = ERR_PTR(-ENOMEM); |
| 196 | goto bail_ip; |
| 197 | } |
| 198 | |
| 199 | dev->n_srqs_allocated++; |
| 200 | spin_unlock(&dev->n_srqs_lock); |
| 201 | |
| 202 | if (srq->ip) { |
| 203 | spin_lock_irq(&dev->pending_lock); |
| 204 | list_add(&srq->ip->pending_mmaps, &dev->pending_mmaps); |
| 205 | spin_unlock_irq(&dev->pending_lock); |
| 206 | } |
| 207 | |
| 208 | ret = &srq->ibsrq; |
| 209 | goto done; |
| 210 | |
| 211 | bail_ip: |
| 212 | kfree(srq->ip); |
| 213 | bail_wq: |
| 214 | vfree(srq->rq.wq); |
| 215 | bail_srq: |
| 216 | kfree(srq); |
| 217 | done: |
| 218 | return ret; |
| 219 | } |
| 220 | |
| 221 | /** |
| 222 | * hfi1_modify_srq - modify a shared receive queue |
| 223 | * @ibsrq: the SRQ to modify |
| 224 | * @attr: the new attributes of the SRQ |
| 225 | * @attr_mask: indicates which attributes to modify |
| 226 | * @udata: user data for libibverbs.so |
| 227 | */ |
| 228 | int hfi1_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, |
| 229 | enum ib_srq_attr_mask attr_mask, |
| 230 | struct ib_udata *udata) |
| 231 | { |
| 232 | struct hfi1_srq *srq = to_isrq(ibsrq); |
| 233 | struct hfi1_rwq *wq; |
| 234 | int ret = 0; |
| 235 | |
| 236 | if (attr_mask & IB_SRQ_MAX_WR) { |
| 237 | struct hfi1_rwq *owq; |
| 238 | struct hfi1_rwqe *p; |
| 239 | u32 sz, size, n, head, tail; |
| 240 | |
| 241 | /* Check that the requested sizes are below the limits. */ |
| 242 | if ((attr->max_wr > hfi1_max_srq_wrs) || |
| 243 | ((attr_mask & IB_SRQ_LIMIT) ? |
| 244 | attr->srq_limit : srq->limit) > attr->max_wr) { |
| 245 | ret = -EINVAL; |
| 246 | goto bail; |
| 247 | } |
| 248 | |
| 249 | sz = sizeof(struct hfi1_rwqe) + |
| 250 | srq->rq.max_sge * sizeof(struct ib_sge); |
| 251 | size = attr->max_wr + 1; |
| 252 | wq = vmalloc_user(sizeof(struct hfi1_rwq) + size * sz); |
| 253 | if (!wq) { |
| 254 | ret = -ENOMEM; |
| 255 | goto bail; |
| 256 | } |
| 257 | |
| 258 | /* Check that we can write the offset to mmap. */ |
| 259 | if (udata && udata->inlen >= sizeof(__u64)) { |
| 260 | __u64 offset_addr; |
| 261 | __u64 offset = 0; |
| 262 | |
| 263 | ret = ib_copy_from_udata(&offset_addr, udata, |
| 264 | sizeof(offset_addr)); |
| 265 | if (ret) |
| 266 | goto bail_free; |
| 267 | udata->outbuf = |
| 268 | (void __user *) (unsigned long) offset_addr; |
| 269 | ret = ib_copy_to_udata(udata, &offset, |
| 270 | sizeof(offset)); |
| 271 | if (ret) |
| 272 | goto bail_free; |
| 273 | } |
| 274 | |
| 275 | spin_lock_irq(&srq->rq.lock); |
| 276 | /* |
| 277 | * validate head and tail pointer values and compute |
| 278 | * the number of remaining WQEs. |
| 279 | */ |
| 280 | owq = srq->rq.wq; |
| 281 | head = owq->head; |
| 282 | tail = owq->tail; |
| 283 | if (head >= srq->rq.size || tail >= srq->rq.size) { |
| 284 | ret = -EINVAL; |
| 285 | goto bail_unlock; |
| 286 | } |
| 287 | n = head; |
| 288 | if (n < tail) |
| 289 | n += srq->rq.size - tail; |
| 290 | else |
| 291 | n -= tail; |
| 292 | if (size <= n) { |
| 293 | ret = -EINVAL; |
| 294 | goto bail_unlock; |
| 295 | } |
| 296 | n = 0; |
| 297 | p = wq->wq; |
| 298 | while (tail != head) { |
| 299 | struct hfi1_rwqe *wqe; |
| 300 | int i; |
| 301 | |
| 302 | wqe = get_rwqe_ptr(&srq->rq, tail); |
| 303 | p->wr_id = wqe->wr_id; |
| 304 | p->num_sge = wqe->num_sge; |
| 305 | for (i = 0; i < wqe->num_sge; i++) |
| 306 | p->sg_list[i] = wqe->sg_list[i]; |
| 307 | n++; |
| 308 | p = (struct hfi1_rwqe *)((char *)p + sz); |
| 309 | if (++tail >= srq->rq.size) |
| 310 | tail = 0; |
| 311 | } |
| 312 | srq->rq.wq = wq; |
| 313 | srq->rq.size = size; |
| 314 | wq->head = n; |
| 315 | wq->tail = 0; |
| 316 | if (attr_mask & IB_SRQ_LIMIT) |
| 317 | srq->limit = attr->srq_limit; |
| 318 | spin_unlock_irq(&srq->rq.lock); |
| 319 | |
| 320 | vfree(owq); |
| 321 | |
| 322 | if (srq->ip) { |
| 323 | struct hfi1_mmap_info *ip = srq->ip; |
| 324 | struct hfi1_ibdev *dev = to_idev(srq->ibsrq.device); |
| 325 | u32 s = sizeof(struct hfi1_rwq) + size * sz; |
| 326 | |
| 327 | hfi1_update_mmap_info(dev, ip, s, wq); |
| 328 | |
| 329 | /* |
| 330 | * Return the offset to mmap. |
| 331 | * See hfi1_mmap() for details. |
| 332 | */ |
| 333 | if (udata && udata->inlen >= sizeof(__u64)) { |
| 334 | ret = ib_copy_to_udata(udata, &ip->offset, |
| 335 | sizeof(ip->offset)); |
| 336 | if (ret) |
| 337 | goto bail; |
| 338 | } |
| 339 | |
| 340 | /* |
| 341 | * Put user mapping info onto the pending list |
| 342 | * unless it already is on the list. |
| 343 | */ |
| 344 | spin_lock_irq(&dev->pending_lock); |
| 345 | if (list_empty(&ip->pending_mmaps)) |
| 346 | list_add(&ip->pending_mmaps, |
| 347 | &dev->pending_mmaps); |
| 348 | spin_unlock_irq(&dev->pending_lock); |
| 349 | } |
| 350 | } else if (attr_mask & IB_SRQ_LIMIT) { |
| 351 | spin_lock_irq(&srq->rq.lock); |
| 352 | if (attr->srq_limit >= srq->rq.size) |
| 353 | ret = -EINVAL; |
| 354 | else |
| 355 | srq->limit = attr->srq_limit; |
| 356 | spin_unlock_irq(&srq->rq.lock); |
| 357 | } |
| 358 | goto bail; |
| 359 | |
| 360 | bail_unlock: |
| 361 | spin_unlock_irq(&srq->rq.lock); |
| 362 | bail_free: |
| 363 | vfree(wq); |
| 364 | bail: |
| 365 | return ret; |
| 366 | } |
| 367 | |
| 368 | int hfi1_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr) |
| 369 | { |
| 370 | struct hfi1_srq *srq = to_isrq(ibsrq); |
| 371 | |
| 372 | attr->max_wr = srq->rq.size - 1; |
| 373 | attr->max_sge = srq->rq.max_sge; |
| 374 | attr->srq_limit = srq->limit; |
| 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | /** |
| 379 | * hfi1_destroy_srq - destroy a shared receive queue |
| 380 | * @ibsrq: the SRQ to destroy |
| 381 | */ |
| 382 | int hfi1_destroy_srq(struct ib_srq *ibsrq) |
| 383 | { |
| 384 | struct hfi1_srq *srq = to_isrq(ibsrq); |
| 385 | struct hfi1_ibdev *dev = to_idev(ibsrq->device); |
| 386 | |
| 387 | spin_lock(&dev->n_srqs_lock); |
| 388 | dev->n_srqs_allocated--; |
| 389 | spin_unlock(&dev->n_srqs_lock); |
| 390 | if (srq->ip) |
| 391 | kref_put(&srq->ip->ref, hfi1_release_mmap_info); |
| 392 | else |
| 393 | vfree(srq->rq.wq); |
| 394 | kfree(srq); |
| 395 | |
| 396 | return 0; |
| 397 | } |