1 /* 2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 4 * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc. All rights reserved. 5 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved. 6 * Copyright (c) 2004 Voltaire, Inc. All rights reserved. 7 * 8 * This software is available to you under a choice of one of two 9 * licenses. You may choose to be licensed under the terms of the GNU 10 * General Public License (GPL) Version 2, available from the file 11 * COPYING in the main directory of this source tree, or the 12 * OpenIB.org BSD license below: 13 * 14 * Redistribution and use in source and binary forms, with or 15 * without modification, are permitted provided that the following 16 * conditions are met: 17 * 18 * - Redistributions of source code must retain the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer. 21 * 22 * - Redistributions in binary form must reproduce the above 23 * copyright notice, this list of conditions and the following 24 * disclaimer in the documentation and/or other materials 25 * provided with the distribution. 26 * 27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 34 * SOFTWARE. 35 */ 36 37 #include <linux/hardirq.h> 38 #include <linux/export.h> 39 40 #include <linux/mlx4/cmd.h> 41 #include <linux/mlx4/cq.h> 42 43 #include "mlx4.h" 44 #include "icm.h" 45 46 #define MLX4_CQ_STATUS_OK ( 0 << 28) 47 #define MLX4_CQ_STATUS_OVERFLOW ( 9 << 28) 48 #define MLX4_CQ_STATUS_WRITE_FAIL (10 << 28) 49 #define MLX4_CQ_FLAG_CC ( 1 << 18) 50 #define MLX4_CQ_FLAG_OI ( 1 << 17) 51 #define MLX4_CQ_STATE_ARMED ( 9 << 8) 52 #define MLX4_CQ_STATE_ARMED_SOL ( 6 << 8) 53 #define MLX4_EQ_STATE_FIRED (10 << 8) 54 55 #define TASKLET_MAX_TIME 2 56 #define TASKLET_MAX_TIME_JIFFIES msecs_to_jiffies(TASKLET_MAX_TIME) 57 58 void mlx4_cq_tasklet_cb(unsigned long data) 59 { 60 unsigned long flags; 61 unsigned long end = jiffies + TASKLET_MAX_TIME_JIFFIES; 62 struct mlx4_eq_tasklet *ctx = (struct mlx4_eq_tasklet *)data; 63 struct mlx4_cq *mcq, *temp; 64 65 spin_lock_irqsave(&ctx->lock, flags); 66 list_splice_tail_init(&ctx->list, &ctx->process_list); 67 spin_unlock_irqrestore(&ctx->lock, flags); 68 69 list_for_each_entry_safe(mcq, temp, &ctx->process_list, tasklet_ctx.list) { 70 list_del_init(&mcq->tasklet_ctx.list); 71 mcq->tasklet_ctx.comp(mcq); 72 if (refcount_dec_and_test(&mcq->refcount)) 73 complete(&mcq->free); 74 if (time_after(jiffies, end)) 75 break; 76 } 77 78 if (!list_empty(&ctx->process_list)) 79 tasklet_schedule(&ctx->task); 80 } 81 82 static void mlx4_add_cq_to_tasklet(struct mlx4_cq *cq) 83 { 84 struct mlx4_eq_tasklet *tasklet_ctx = cq->tasklet_ctx.priv; 85 unsigned long flags; 86 bool kick; 87 88 spin_lock_irqsave(&tasklet_ctx->lock, flags); 89 /* When migrating CQs between EQs will be implemented, please note 90 * that you need to sync this point. It is possible that 91 * while migrating a CQ, completions on the old EQs could 92 * still arrive. 93 */ 94 if (list_empty_careful(&cq->tasklet_ctx.list)) { 95 refcount_inc(&cq->refcount); 96 kick = list_empty(&tasklet_ctx->list); 97 list_add_tail(&cq->tasklet_ctx.list, &tasklet_ctx->list); 98 if (kick) 99 tasklet_schedule(&tasklet_ctx->task); 100 } 101 spin_unlock_irqrestore(&tasklet_ctx->lock, flags); 102 } 103 104 void mlx4_cq_completion(struct mlx4_dev *dev, u32 cqn) 105 { 106 struct mlx4_cq *cq; 107 108 rcu_read_lock(); 109 cq = radix_tree_lookup(&mlx4_priv(dev)->cq_table.tree, 110 cqn & (dev->caps.num_cqs - 1)); 111 rcu_read_unlock(); 112 113 if (!cq) { 114 mlx4_dbg(dev, "Completion event for bogus CQ %08x\n", cqn); 115 return; 116 } 117 118 /* Acessing the CQ outside of rcu_read_lock is safe, because 119 * the CQ is freed only after interrupt handling is completed. 120 */ 121 ++cq->arm_sn; 122 123 cq->comp(cq); 124 } 125 126 void mlx4_cq_event(struct mlx4_dev *dev, u32 cqn, int event_type) 127 { 128 struct mlx4_cq_table *cq_table = &mlx4_priv(dev)->cq_table; 129 struct mlx4_cq *cq; 130 131 rcu_read_lock(); 132 cq = radix_tree_lookup(&cq_table->tree, cqn & (dev->caps.num_cqs - 1)); 133 rcu_read_unlock(); 134 135 if (!cq) { 136 mlx4_dbg(dev, "Async event for bogus CQ %08x\n", cqn); 137 return; 138 } 139 140 /* Acessing the CQ outside of rcu_read_lock is safe, because 141 * the CQ is freed only after interrupt handling is completed. 142 */ 143 cq->event(cq, event_type); 144 } 145 146 static int mlx4_SW2HW_CQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox, 147 int cq_num, u8 opmod) 148 { 149 return mlx4_cmd(dev, mailbox->dma, cq_num, opmod, 150 MLX4_CMD_SW2HW_CQ, MLX4_CMD_TIME_CLASS_A, 151 MLX4_CMD_WRAPPED); 152 } 153 154 static int mlx4_MODIFY_CQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox, 155 int cq_num, u32 opmod) 156 { 157 return mlx4_cmd(dev, mailbox->dma, cq_num, opmod, MLX4_CMD_MODIFY_CQ, 158 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); 159 } 160 161 static int mlx4_HW2SW_CQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox, 162 int cq_num) 163 { 164 return mlx4_cmd_box(dev, 0, mailbox ? mailbox->dma : 0, 165 cq_num, mailbox ? 0 : 1, MLX4_CMD_HW2SW_CQ, 166 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); 167 } 168 169 int mlx4_cq_modify(struct mlx4_dev *dev, struct mlx4_cq *cq, 170 u16 count, u16 period) 171 { 172 struct mlx4_cmd_mailbox *mailbox; 173 struct mlx4_cq_context *cq_context; 174 int err; 175 176 mailbox = mlx4_alloc_cmd_mailbox(dev); 177 if (IS_ERR(mailbox)) 178 return PTR_ERR(mailbox); 179 180 cq_context = mailbox->buf; 181 cq_context->cq_max_count = cpu_to_be16(count); 182 cq_context->cq_period = cpu_to_be16(period); 183 184 err = mlx4_MODIFY_CQ(dev, mailbox, cq->cqn, 1); 185 186 mlx4_free_cmd_mailbox(dev, mailbox); 187 return err; 188 } 189 EXPORT_SYMBOL_GPL(mlx4_cq_modify); 190 191 int mlx4_cq_resize(struct mlx4_dev *dev, struct mlx4_cq *cq, 192 int entries, struct mlx4_mtt *mtt) 193 { 194 struct mlx4_cmd_mailbox *mailbox; 195 struct mlx4_cq_context *cq_context; 196 u64 mtt_addr; 197 int err; 198 199 mailbox = mlx4_alloc_cmd_mailbox(dev); 200 if (IS_ERR(mailbox)) 201 return PTR_ERR(mailbox); 202 203 cq_context = mailbox->buf; 204 cq_context->logsize_usrpage = cpu_to_be32(ilog2(entries) << 24); 205 cq_context->log_page_size = mtt->page_shift - 12; 206 mtt_addr = mlx4_mtt_addr(dev, mtt); 207 cq_context->mtt_base_addr_h = mtt_addr >> 32; 208 cq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff); 209 210 err = mlx4_MODIFY_CQ(dev, mailbox, cq->cqn, 0); 211 212 mlx4_free_cmd_mailbox(dev, mailbox); 213 return err; 214 } 215 EXPORT_SYMBOL_GPL(mlx4_cq_resize); 216 217 int __mlx4_cq_alloc_icm(struct mlx4_dev *dev, int *cqn) 218 { 219 struct mlx4_priv *priv = mlx4_priv(dev); 220 struct mlx4_cq_table *cq_table = &priv->cq_table; 221 int err; 222 223 *cqn = mlx4_bitmap_alloc(&cq_table->bitmap); 224 if (*cqn == -1) 225 return -ENOMEM; 226 227 err = mlx4_table_get(dev, &cq_table->table, *cqn); 228 if (err) 229 goto err_out; 230 231 err = mlx4_table_get(dev, &cq_table->cmpt_table, *cqn); 232 if (err) 233 goto err_put; 234 return 0; 235 236 err_put: 237 mlx4_table_put(dev, &cq_table->table, *cqn); 238 239 err_out: 240 mlx4_bitmap_free(&cq_table->bitmap, *cqn, MLX4_NO_RR); 241 return err; 242 } 243 244 static int mlx4_cq_alloc_icm(struct mlx4_dev *dev, int *cqn, u8 usage) 245 { 246 u32 in_modifier = RES_CQ | (((u32)usage & 3) << 30); 247 u64 out_param; 248 int err; 249 250 if (mlx4_is_mfunc(dev)) { 251 err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier, 252 RES_OP_RESERVE_AND_MAP, MLX4_CMD_ALLOC_RES, 253 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); 254 if (err) 255 return err; 256 else { 257 *cqn = get_param_l(&out_param); 258 return 0; 259 } 260 } 261 return __mlx4_cq_alloc_icm(dev, cqn); 262 } 263 264 void __mlx4_cq_free_icm(struct mlx4_dev *dev, int cqn) 265 { 266 struct mlx4_priv *priv = mlx4_priv(dev); 267 struct mlx4_cq_table *cq_table = &priv->cq_table; 268 269 mlx4_table_put(dev, &cq_table->cmpt_table, cqn); 270 mlx4_table_put(dev, &cq_table->table, cqn); 271 mlx4_bitmap_free(&cq_table->bitmap, cqn, MLX4_NO_RR); 272 } 273 274 static void mlx4_cq_free_icm(struct mlx4_dev *dev, int cqn) 275 { 276 u64 in_param = 0; 277 int err; 278 279 if (mlx4_is_mfunc(dev)) { 280 set_param_l(&in_param, cqn); 281 err = mlx4_cmd(dev, in_param, RES_CQ, RES_OP_RESERVE_AND_MAP, 282 MLX4_CMD_FREE_RES, 283 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); 284 if (err) 285 mlx4_warn(dev, "Failed freeing cq:%d\n", cqn); 286 } else 287 __mlx4_cq_free_icm(dev, cqn); 288 } 289 290 static int mlx4_init_user_cqes(void *buf, int entries, int cqe_size) 291 { 292 int entries_per_copy = PAGE_SIZE / cqe_size; 293 void *init_ents; 294 int err = 0; 295 int i; 296 297 init_ents = kmalloc(PAGE_SIZE, GFP_KERNEL); 298 if (!init_ents) 299 return -ENOMEM; 300 301 /* Populate a list of CQ entries to reduce the number of 302 * copy_to_user calls. 0xcc is the initialization value 303 * required by the FW. 304 */ 305 memset(init_ents, 0xcc, PAGE_SIZE); 306 307 if (entries_per_copy < entries) { 308 for (i = 0; i < entries / entries_per_copy; i++) { 309 err = copy_to_user(buf, init_ents, PAGE_SIZE); 310 if (err) 311 goto out; 312 313 buf += PAGE_SIZE; 314 } 315 } else { 316 err = copy_to_user(buf, init_ents, entries * cqe_size); 317 } 318 319 out: 320 kfree(init_ents); 321 322 return err; 323 } 324 325 static void mlx4_init_kernel_cqes(struct mlx4_buf *buf, 326 int entries, 327 int cqe_size) 328 { 329 int i; 330 331 if (buf->nbufs == 1) 332 memset(buf->direct.buf, 0xcc, entries * cqe_size); 333 else 334 for (i = 0; i < buf->npages; i++) 335 memset(buf->page_list[i].buf, 0xcc, 336 1UL << buf->page_shift); 337 } 338 339 int mlx4_cq_alloc(struct mlx4_dev *dev, int nent, 340 struct mlx4_mtt *mtt, struct mlx4_uar *uar, u64 db_rec, 341 struct mlx4_cq *cq, unsigned vector, int collapsed, 342 int timestamp_en, void *buf_addr, bool user_cq) 343 { 344 bool sw_cq_init = dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SW_CQ_INIT; 345 struct mlx4_priv *priv = mlx4_priv(dev); 346 struct mlx4_cq_table *cq_table = &priv->cq_table; 347 struct mlx4_cmd_mailbox *mailbox; 348 struct mlx4_cq_context *cq_context; 349 u64 mtt_addr; 350 int err; 351 352 if (vector >= dev->caps.num_comp_vectors) 353 return -EINVAL; 354 355 cq->vector = vector; 356 357 err = mlx4_cq_alloc_icm(dev, &cq->cqn, cq->usage); 358 if (err) 359 return err; 360 361 spin_lock(&cq_table->lock); 362 err = radix_tree_insert(&cq_table->tree, cq->cqn, cq); 363 spin_unlock(&cq_table->lock); 364 if (err) 365 goto err_icm; 366 367 mailbox = mlx4_alloc_cmd_mailbox(dev); 368 if (IS_ERR(mailbox)) { 369 err = PTR_ERR(mailbox); 370 goto err_radix; 371 } 372 373 cq_context = mailbox->buf; 374 cq_context->flags = cpu_to_be32(!!collapsed << 18); 375 if (timestamp_en) 376 cq_context->flags |= cpu_to_be32(1 << 19); 377 378 cq_context->logsize_usrpage = 379 cpu_to_be32((ilog2(nent) << 24) | 380 mlx4_to_hw_uar_index(dev, uar->index)); 381 cq_context->comp_eqn = priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(vector)].eqn; 382 cq_context->log_page_size = mtt->page_shift - MLX4_ICM_PAGE_SHIFT; 383 384 mtt_addr = mlx4_mtt_addr(dev, mtt); 385 cq_context->mtt_base_addr_h = mtt_addr >> 32; 386 cq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff); 387 cq_context->db_rec_addr = cpu_to_be64(db_rec); 388 389 if (sw_cq_init) { 390 if (user_cq) { 391 err = mlx4_init_user_cqes(buf_addr, nent, 392 dev->caps.cqe_size); 393 if (err) 394 sw_cq_init = false; 395 } else { 396 mlx4_init_kernel_cqes(buf_addr, nent, 397 dev->caps.cqe_size); 398 } 399 } 400 401 err = mlx4_SW2HW_CQ(dev, mailbox, cq->cqn, sw_cq_init); 402 403 mlx4_free_cmd_mailbox(dev, mailbox); 404 if (err) 405 goto err_radix; 406 407 cq->cons_index = 0; 408 cq->arm_sn = 1; 409 cq->uar = uar; 410 refcount_set(&cq->refcount, 1); 411 init_completion(&cq->free); 412 cq->comp = mlx4_add_cq_to_tasklet; 413 cq->tasklet_ctx.priv = 414 &priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(vector)].tasklet_ctx; 415 INIT_LIST_HEAD(&cq->tasklet_ctx.list); 416 417 418 cq->irq = priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(vector)].irq; 419 return 0; 420 421 err_radix: 422 spin_lock(&cq_table->lock); 423 radix_tree_delete(&cq_table->tree, cq->cqn); 424 spin_unlock(&cq_table->lock); 425 426 err_icm: 427 mlx4_cq_free_icm(dev, cq->cqn); 428 429 return err; 430 } 431 EXPORT_SYMBOL_GPL(mlx4_cq_alloc); 432 433 void mlx4_cq_free(struct mlx4_dev *dev, struct mlx4_cq *cq) 434 { 435 struct mlx4_priv *priv = mlx4_priv(dev); 436 struct mlx4_cq_table *cq_table = &priv->cq_table; 437 int err; 438 439 err = mlx4_HW2SW_CQ(dev, NULL, cq->cqn); 440 if (err) 441 mlx4_warn(dev, "HW2SW_CQ failed (%d) for CQN %06x\n", err, cq->cqn); 442 443 spin_lock(&cq_table->lock); 444 radix_tree_delete(&cq_table->tree, cq->cqn); 445 spin_unlock(&cq_table->lock); 446 447 synchronize_irq(priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(cq->vector)].irq); 448 if (priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(cq->vector)].irq != 449 priv->eq_table.eq[MLX4_EQ_ASYNC].irq) 450 synchronize_irq(priv->eq_table.eq[MLX4_EQ_ASYNC].irq); 451 452 if (refcount_dec_and_test(&cq->refcount)) 453 complete(&cq->free); 454 wait_for_completion(&cq->free); 455 456 mlx4_cq_free_icm(dev, cq->cqn); 457 } 458 EXPORT_SYMBOL_GPL(mlx4_cq_free); 459 460 int mlx4_init_cq_table(struct mlx4_dev *dev) 461 { 462 struct mlx4_cq_table *cq_table = &mlx4_priv(dev)->cq_table; 463 int err; 464 465 spin_lock_init(&cq_table->lock); 466 INIT_RADIX_TREE(&cq_table->tree, GFP_ATOMIC); 467 if (mlx4_is_slave(dev)) 468 return 0; 469 470 err = mlx4_bitmap_init(&cq_table->bitmap, dev->caps.num_cqs, 471 dev->caps.num_cqs - 1, dev->caps.reserved_cqs, 0); 472 if (err) 473 return err; 474 475 return 0; 476 } 477 478 void mlx4_cleanup_cq_table(struct mlx4_dev *dev) 479 { 480 if (mlx4_is_slave(dev)) 481 return; 482 /* Nothing to do to clean up radix_tree */ 483 mlx4_bitmap_cleanup(&mlx4_priv(dev)->cq_table.bitmap); 484 } 485