1 /* 2 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32 33 #include <linux/mlx4/qp.h> 34 #include <linux/mlx4/srq.h> 35 36 #include "mlx4_ib.h" 37 #include "user.h" 38 39 static void *get_wqe(struct mlx4_ib_srq *srq, int n) 40 { 41 return mlx4_buf_offset(&srq->buf, n << srq->msrq.wqe_shift); 42 } 43 44 static void mlx4_ib_srq_event(struct mlx4_srq *srq, enum mlx4_event type) 45 { 46 struct ib_event event; 47 struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq; 48 49 if (ibsrq->event_handler) { 50 event.device = ibsrq->device; 51 event.element.srq = ibsrq; 52 switch (type) { 53 case MLX4_EVENT_TYPE_SRQ_LIMIT: 54 event.event = IB_EVENT_SRQ_LIMIT_REACHED; 55 break; 56 case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR: 57 event.event = IB_EVENT_SRQ_ERR; 58 break; 59 default: 60 printk(KERN_WARNING "mlx4_ib: Unexpected event type %d " 61 "on SRQ %06x\n", type, srq->srqn); 62 return; 63 } 64 65 ibsrq->event_handler(&event, ibsrq->srq_context); 66 } 67 } 68 69 struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd, 70 struct ib_srq_init_attr *init_attr, 71 struct ib_udata *udata) 72 { 73 struct mlx4_ib_dev *dev = to_mdev(pd->device); 74 struct mlx4_ib_srq *srq; 75 struct mlx4_wqe_srq_next_seg *next; 76 int desc_size; 77 int buf_size; 78 int err; 79 int i; 80 81 /* Sanity check SRQ size before proceeding */ 82 if (init_attr->attr.max_wr >= dev->dev->caps.max_srq_wqes || 83 init_attr->attr.max_sge > dev->dev->caps.max_srq_sge) 84 return ERR_PTR(-EINVAL); 85 86 srq = kmalloc(sizeof *srq, GFP_KERNEL); 87 if (!srq) 88 return ERR_PTR(-ENOMEM); 89 90 mutex_init(&srq->mutex); 91 spin_lock_init(&srq->lock); 92 srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1); 93 srq->msrq.max_gs = init_attr->attr.max_sge; 94 95 desc_size = max(32UL, 96 roundup_pow_of_two(sizeof (struct mlx4_wqe_srq_next_seg) + 97 srq->msrq.max_gs * 98 sizeof (struct mlx4_wqe_data_seg))); 99 srq->msrq.wqe_shift = ilog2(desc_size); 100 101 buf_size = srq->msrq.max * desc_size; 102 103 if (pd->uobject) { 104 struct mlx4_ib_create_srq ucmd; 105 106 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) { 107 err = -EFAULT; 108 goto err_srq; 109 } 110 111 srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, 112 buf_size, 0); 113 if (IS_ERR(srq->umem)) { 114 err = PTR_ERR(srq->umem); 115 goto err_srq; 116 } 117 118 err = mlx4_mtt_init(dev->dev, ib_umem_page_count(srq->umem), 119 ilog2(srq->umem->page_size), &srq->mtt); 120 if (err) 121 goto err_buf; 122 123 err = mlx4_ib_umem_write_mtt(dev, &srq->mtt, srq->umem); 124 if (err) 125 goto err_mtt; 126 127 err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context), 128 ucmd.db_addr, &srq->db); 129 if (err) 130 goto err_mtt; 131 } else { 132 err = mlx4_ib_db_alloc(dev, &srq->db, 0); 133 if (err) 134 goto err_srq; 135 136 *srq->db.db = 0; 137 138 if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf)) { 139 err = -ENOMEM; 140 goto err_db; 141 } 142 143 srq->head = 0; 144 srq->tail = srq->msrq.max - 1; 145 srq->wqe_ctr = 0; 146 147 for (i = 0; i < srq->msrq.max; ++i) { 148 next = get_wqe(srq, i); 149 next->next_wqe_index = 150 cpu_to_be16((i + 1) & (srq->msrq.max - 1)); 151 } 152 153 err = mlx4_mtt_init(dev->dev, srq->buf.npages, srq->buf.page_shift, 154 &srq->mtt); 155 if (err) 156 goto err_buf; 157 158 err = mlx4_buf_write_mtt(dev->dev, &srq->mtt, &srq->buf); 159 if (err) 160 goto err_mtt; 161 162 srq->wrid = kmalloc(srq->msrq.max * sizeof (u64), GFP_KERNEL); 163 if (!srq->wrid) { 164 err = -ENOMEM; 165 goto err_mtt; 166 } 167 } 168 169 err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, &srq->mtt, 170 srq->db.dma, &srq->msrq); 171 if (err) 172 goto err_wrid; 173 174 srq->msrq.event = mlx4_ib_srq_event; 175 176 if (pd->uobject) 177 if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) { 178 err = -EFAULT; 179 goto err_wrid; 180 } 181 182 init_attr->attr.max_wr = srq->msrq.max - 1; 183 184 return &srq->ibsrq; 185 186 err_wrid: 187 if (pd->uobject) 188 mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db); 189 else 190 kfree(srq->wrid); 191 192 err_mtt: 193 mlx4_mtt_cleanup(dev->dev, &srq->mtt); 194 195 err_buf: 196 if (pd->uobject) 197 ib_umem_release(srq->umem); 198 else 199 mlx4_buf_free(dev->dev, buf_size, &srq->buf); 200 201 err_db: 202 if (!pd->uobject) 203 mlx4_ib_db_free(dev, &srq->db); 204 205 err_srq: 206 kfree(srq); 207 208 return ERR_PTR(err); 209 } 210 211 int mlx4_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, 212 enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) 213 { 214 struct mlx4_ib_dev *dev = to_mdev(ibsrq->device); 215 struct mlx4_ib_srq *srq = to_msrq(ibsrq); 216 int ret; 217 218 /* We don't support resizing SRQs (yet?) */ 219 if (attr_mask & IB_SRQ_MAX_WR) 220 return -EINVAL; 221 222 if (attr_mask & IB_SRQ_LIMIT) { 223 if (attr->srq_limit >= srq->msrq.max) 224 return -EINVAL; 225 226 mutex_lock(&srq->mutex); 227 ret = mlx4_srq_arm(dev->dev, &srq->msrq, attr->srq_limit); 228 mutex_unlock(&srq->mutex); 229 230 if (ret) 231 return ret; 232 } 233 234 return 0; 235 } 236 237 int mlx4_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr) 238 { 239 struct mlx4_ib_dev *dev = to_mdev(ibsrq->device); 240 struct mlx4_ib_srq *srq = to_msrq(ibsrq); 241 int ret; 242 int limit_watermark; 243 244 ret = mlx4_srq_query(dev->dev, &srq->msrq, &limit_watermark); 245 if (ret) 246 return ret; 247 248 srq_attr->srq_limit = limit_watermark; 249 srq_attr->max_wr = srq->msrq.max - 1; 250 srq_attr->max_sge = srq->msrq.max_gs; 251 252 return 0; 253 } 254 255 int mlx4_ib_destroy_srq(struct ib_srq *srq) 256 { 257 struct mlx4_ib_dev *dev = to_mdev(srq->device); 258 struct mlx4_ib_srq *msrq = to_msrq(srq); 259 260 mlx4_srq_free(dev->dev, &msrq->msrq); 261 mlx4_mtt_cleanup(dev->dev, &msrq->mtt); 262 263 if (srq->uobject) { 264 mlx4_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db); 265 ib_umem_release(msrq->umem); 266 } else { 267 kfree(msrq->wrid); 268 mlx4_buf_free(dev->dev, msrq->msrq.max << msrq->msrq.wqe_shift, 269 &msrq->buf); 270 mlx4_ib_db_free(dev, &msrq->db); 271 } 272 273 kfree(msrq); 274 275 return 0; 276 } 277 278 void mlx4_ib_free_srq_wqe(struct mlx4_ib_srq *srq, int wqe_index) 279 { 280 struct mlx4_wqe_srq_next_seg *next; 281 282 /* always called with interrupts disabled. */ 283 spin_lock(&srq->lock); 284 285 next = get_wqe(srq, srq->tail); 286 next->next_wqe_index = cpu_to_be16(wqe_index); 287 srq->tail = wqe_index; 288 289 spin_unlock(&srq->lock); 290 } 291 292 int mlx4_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, 293 struct ib_recv_wr **bad_wr) 294 { 295 struct mlx4_ib_srq *srq = to_msrq(ibsrq); 296 struct mlx4_wqe_srq_next_seg *next; 297 struct mlx4_wqe_data_seg *scat; 298 unsigned long flags; 299 int err = 0; 300 int nreq; 301 int i; 302 303 spin_lock_irqsave(&srq->lock, flags); 304 305 for (nreq = 0; wr; ++nreq, wr = wr->next) { 306 if (unlikely(wr->num_sge > srq->msrq.max_gs)) { 307 err = -EINVAL; 308 *bad_wr = wr; 309 break; 310 } 311 312 if (unlikely(srq->head == srq->tail)) { 313 err = -ENOMEM; 314 *bad_wr = wr; 315 break; 316 } 317 318 srq->wrid[srq->head] = wr->wr_id; 319 320 next = get_wqe(srq, srq->head); 321 srq->head = be16_to_cpu(next->next_wqe_index); 322 scat = (struct mlx4_wqe_data_seg *) (next + 1); 323 324 for (i = 0; i < wr->num_sge; ++i) { 325 scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length); 326 scat[i].lkey = cpu_to_be32(wr->sg_list[i].lkey); 327 scat[i].addr = cpu_to_be64(wr->sg_list[i].addr); 328 } 329 330 if (i < srq->msrq.max_gs) { 331 scat[i].byte_count = 0; 332 scat[i].lkey = cpu_to_be32(MLX4_INVALID_LKEY); 333 scat[i].addr = 0; 334 } 335 } 336 337 if (likely(nreq)) { 338 srq->wqe_ctr += nreq; 339 340 /* 341 * Make sure that descriptors are written before 342 * doorbell record. 343 */ 344 wmb(); 345 346 *srq->db.db = cpu_to_be32(srq->wqe_ctr); 347 } 348 349 spin_unlock_irqrestore(&srq->lock, flags); 350 351 return err; 352 } 353