1/* 2 * Copyright (c) 2004 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc. All rights reserved. 4 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved. 5 * Copyright (c) 2004 Voltaire, Inc. All rights reserved. 6 * 7 * This software is available to you under a choice of one of two 8 * licenses. You may choose to be licensed under the terms of the GNU 9 * General Public License (GPL) Version 2, available from the file 10 * COPYING in the main directory of this source tree, or the 11 * OpenIB.org BSD license below: 12 * 13 * Redistribution and use in source and binary forms, with or 14 * without modification, are permitted provided that the following 15 * conditions are met: 16 * 17 * - Redistributions of source code must retain the above 18 * copyright notice, this list of conditions and the following 19 * disclaimer. 20 * 21 * - Redistributions in binary form must reproduce the above 22 * copyright notice, this list of conditions and the following 23 * disclaimer in the documentation and/or other materials 24 * provided with the distribution. 25 * 26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 33 * SOFTWARE. 34 */ 35 36#include <linux/gfp.h> 37#include <linux/export.h> 38 39#include <linux/mlx4/cmd.h> 40#include <linux/mlx4/qp.h> 41 42#include "mlx4.h" 43#include "icm.h" 44 45/* QP to support BF should have bits 6,7 cleared */ 46#define MLX4_BF_QP_SKIP_MASK 0xc0 47#define MLX4_MAX_BF_QP_RANGE 0x40 48 49void mlx4_put_qp(struct mlx4_qp *qp) 50{ 51 if (refcount_dec_and_test(&qp->refcount)) 52 complete(&qp->free); 53} 54EXPORT_SYMBOL_GPL(mlx4_put_qp); 55 56void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type) 57{ 58 struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table; 59 struct mlx4_qp *qp; 60 61 spin_lock(&qp_table->lock); 62 63 qp = __mlx4_qp_lookup(dev, qpn); 64 if (qp) 65 refcount_inc(&qp->refcount); 66 67 spin_unlock(&qp_table->lock); 68 69 if (!qp) { 70 mlx4_dbg(dev, "Async event for none existent QP %08x\n", qpn); 71 return; 72 } 73 74 /* Need to call mlx4_put_qp() in event handler */ 75 qp->event(qp, event_type); 76} 77 78/* used for INIT/CLOSE port logic */ 79static int is_master_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0) 80{ 81 /* this procedure is called after we already know we are on the master */ 82 /* qp0 is either the proxy qp0, or the real qp0 */ 83 u32 pf_proxy_offset = dev->phys_caps.base_proxy_sqpn + 8 * mlx4_master_func_num(dev); 84 *proxy_qp0 = qp->qpn >= pf_proxy_offset && qp->qpn <= pf_proxy_offset + 1; 85 86 *real_qp0 = qp->qpn >= dev->phys_caps.base_sqpn && 87 qp->qpn <= dev->phys_caps.base_sqpn + 1; 88 89 return *real_qp0 || *proxy_qp0; 90} 91 92static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, 93 enum mlx4_qp_state cur_state, enum mlx4_qp_state new_state, 94 struct mlx4_qp_context *context, 95 enum mlx4_qp_optpar optpar, 96 int sqd_event, struct mlx4_qp *qp, int native) 97{ 98 static const u16 op[MLX4_QP_NUM_STATE][MLX4_QP_NUM_STATE] = { 99 [MLX4_QP_STATE_RST] = { 100 [MLX4_QP_STATE_RST] = MLX4_CMD_2RST_QP, 101 [MLX4_QP_STATE_ERR] = MLX4_CMD_2ERR_QP, 102 [MLX4_QP_STATE_INIT] = MLX4_CMD_RST2INIT_QP, 103 }, 104 [MLX4_QP_STATE_INIT] = { 105 [MLX4_QP_STATE_RST] = MLX4_CMD_2RST_QP, 106 [MLX4_QP_STATE_ERR] = MLX4_CMD_2ERR_QP, 107 [MLX4_QP_STATE_INIT] = MLX4_CMD_INIT2INIT_QP, 108 [MLX4_QP_STATE_RTR] = MLX4_CMD_INIT2RTR_QP, 109 }, 110 [MLX4_QP_STATE_RTR] = { 111 [MLX4_QP_STATE_RST] = MLX4_CMD_2RST_QP, 112 [MLX4_QP_STATE_ERR] = MLX4_CMD_2ERR_QP, 113 [MLX4_QP_STATE_RTS] = MLX4_CMD_RTR2RTS_QP, 114 }, 115 [MLX4_QP_STATE_RTS] = { 116 [MLX4_QP_STATE_RST] = MLX4_CMD_2RST_QP, 117 [MLX4_QP_STATE_ERR] = MLX4_CMD_2ERR_QP, 118 [MLX4_QP_STATE_RTS] = MLX4_CMD_RTS2RTS_QP, 119 [MLX4_QP_STATE_SQD] = MLX4_CMD_RTS2SQD_QP, 120 }, 121 [MLX4_QP_STATE_SQD] = { 122 [MLX4_QP_STATE_RST] = MLX4_CMD_2RST_QP, 123 [MLX4_QP_STATE_ERR] = MLX4_CMD_2ERR_QP, 124 [MLX4_QP_STATE_RTS] = MLX4_CMD_SQD2RTS_QP, 125 [MLX4_QP_STATE_SQD] = MLX4_CMD_SQD2SQD_QP, 126 }, 127 [MLX4_QP_STATE_SQER] = { 128 [MLX4_QP_STATE_RST] = MLX4_CMD_2RST_QP, 129 [MLX4_QP_STATE_ERR] = MLX4_CMD_2ERR_QP, 130 [MLX4_QP_STATE_RTS] = MLX4_CMD_SQERR2RTS_QP, 131 }, 132 [MLX4_QP_STATE_ERR] = { 133 [MLX4_QP_STATE_RST] = MLX4_CMD_2RST_QP, 134 [MLX4_QP_STATE_ERR] = MLX4_CMD_2ERR_QP, 135 } 136 }; 137 138 struct mlx4_priv *priv = mlx4_priv(dev); 139 struct mlx4_cmd_mailbox *mailbox; 140 int ret = 0; 141 int real_qp0 = 0; 142 int proxy_qp0 = 0; 143 u8 port; 144 145 if (cur_state >= MLX4_QP_NUM_STATE || new_state >= MLX4_QP_NUM_STATE || 146 !op[cur_state][new_state]) 147 return -EINVAL; 148 149 if (op[cur_state][new_state] == MLX4_CMD_2RST_QP) { 150 ret = mlx4_cmd(dev, 0, qp->qpn, 2, 151 MLX4_CMD_2RST_QP, MLX4_CMD_TIME_CLASS_A, native); 152 if (mlx4_is_master(dev) && cur_state != MLX4_QP_STATE_ERR && 153 cur_state != MLX4_QP_STATE_RST && 154 is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) { 155 port = (qp->qpn & 1) + 1; 156 if (proxy_qp0) 157 priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0; 158 else 159 priv->mfunc.master.qp0_state[port].qp0_active = 0; 160 } 161 return ret; 162 } 163 164 mailbox = mlx4_alloc_cmd_mailbox(dev); 165 if (IS_ERR(mailbox)) 166 return PTR_ERR(mailbox); 167 168 if (cur_state == MLX4_QP_STATE_RST && new_state == MLX4_QP_STATE_INIT) { 169 u64 mtt_addr = mlx4_mtt_addr(dev, mtt); 170 context->mtt_base_addr_h = mtt_addr >> 32; 171 context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff); 172 context->log_page_size = mtt->page_shift - MLX4_ICM_PAGE_SHIFT; 173 } 174 175 if ((cur_state == MLX4_QP_STATE_RTR) && 176 (new_state == MLX4_QP_STATE_RTS) && 177 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) 178 context->roce_entropy = 179 cpu_to_be16(mlx4_qp_roce_entropy(dev, qp->qpn)); 180 181 *(__be32 *) mailbox->buf = cpu_to_be32(optpar); 182 memcpy(mailbox->buf + 8, context, sizeof(*context)); 183 184 ((struct mlx4_qp_context *) (mailbox->buf + 8))->local_qpn = 185 cpu_to_be32(qp->qpn); 186 187 ret = mlx4_cmd(dev, mailbox->dma, 188 qp->qpn | (!!sqd_event << 31), 189 new_state == MLX4_QP_STATE_RST ? 2 : 0, 190 op[cur_state][new_state], MLX4_CMD_TIME_CLASS_C, native); 191 192 if (mlx4_is_master(dev) && is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) { 193 port = (qp->qpn & 1) + 1; 194 if (cur_state != MLX4_QP_STATE_ERR && 195 cur_state != MLX4_QP_STATE_RST && 196 new_state == MLX4_QP_STATE_ERR) { 197 if (proxy_qp0) 198 priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0; 199 else 200 priv->mfunc.master.qp0_state[port].qp0_active = 0; 201 } else if (new_state == MLX4_QP_STATE_RTR) { 202 if (proxy_qp0) 203 priv->mfunc.master.qp0_state[port].proxy_qp0_active = 1; 204 else 205 priv->mfunc.master.qp0_state[port].qp0_active = 1; 206 } 207 } 208 209 mlx4_free_cmd_mailbox(dev, mailbox); 210 return ret; 211} 212 213int mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, 214 enum mlx4_qp_state cur_state, enum mlx4_qp_state new_state, 215 struct mlx4_qp_context *context, 216 enum mlx4_qp_optpar optpar, 217 int sqd_event, struct mlx4_qp *qp) 218{ 219 return __mlx4_qp_modify(dev, mtt, cur_state, new_state, context, 220 optpar, sqd_event, qp, 0); 221} 222EXPORT_SYMBOL_GPL(mlx4_qp_modify); 223 224int __mlx4_qp_reserve_range(struct mlx4_dev *dev, int cnt, int align, 225 int *base, u8 flags) 226{ 227 u32 uid; 228 int bf_qp = !!(flags & (u8)MLX4_RESERVE_ETH_BF_QP); 229 230 struct mlx4_priv *priv = mlx4_priv(dev); 231 struct mlx4_qp_table *qp_table = &priv->qp_table; 232 233 if (cnt > MLX4_MAX_BF_QP_RANGE && bf_qp) 234 return -ENOMEM; 235 236 uid = MLX4_QP_TABLE_ZONE_GENERAL; 237 if (flags & (u8)MLX4_RESERVE_A0_QP) { 238 if (bf_qp) 239 uid = MLX4_QP_TABLE_ZONE_RAW_ETH; 240 else 241 uid = MLX4_QP_TABLE_ZONE_RSS; 242 } 243 244 *base = mlx4_zone_alloc_entries(qp_table->zones, uid, cnt, align, 245 bf_qp ? MLX4_BF_QP_SKIP_MASK : 0, NULL); 246 if (*base == -1) 247 return -ENOMEM; 248 249 return 0; 250} 251 252int mlx4_qp_reserve_range(struct mlx4_dev *dev, int cnt, int align, 253 int *base, u8 flags, u8 usage) 254{ 255 u32 in_modifier = RES_QP | (((u32)usage & 3) << 30); 256 u64 in_param = 0; 257 u64 out_param; 258 int err; 259 260 /* Turn off all unsupported QP allocation flags */ 261 flags &= dev->caps.alloc_res_qp_mask; 262 263 if (mlx4_is_mfunc(dev)) { 264 set_param_l(&in_param, (((u32)flags) << 24) | (u32)cnt); 265 set_param_h(&in_param, align); 266 err = mlx4_cmd_imm(dev, in_param, &out_param, 267 in_modifier, RES_OP_RESERVE, 268 MLX4_CMD_ALLOC_RES, 269 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); 270 if (err) 271 return err; 272 273 *base = get_param_l(&out_param); 274 return 0; 275 } 276 return __mlx4_qp_reserve_range(dev, cnt, align, base, flags); 277} 278EXPORT_SYMBOL_GPL(mlx4_qp_reserve_range); 279 280void __mlx4_qp_release_range(struct mlx4_dev *dev, int base_qpn, int cnt) 281{ 282 struct mlx4_priv *priv = mlx4_priv(dev); 283 struct mlx4_qp_table *qp_table = &priv->qp_table; 284 285 if (mlx4_is_qp_reserved(dev, (u32) base_qpn)) 286 return; 287 mlx4_zone_free_entries_unique(qp_table->zones, base_qpn, cnt); 288} 289 290void mlx4_qp_release_range(struct mlx4_dev *dev, int base_qpn, int cnt) 291{ 292 u64 in_param = 0; 293 int err; 294 295 if (!cnt) 296 return; 297 298 if (mlx4_is_mfunc(dev)) { 299 set_param_l(&in_param, base_qpn); 300 set_param_h(&in_param, cnt); 301 err = mlx4_cmd(dev, in_param, RES_QP, RES_OP_RESERVE, 302 MLX4_CMD_FREE_RES, 303 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); 304 if (err) { 305 mlx4_warn(dev, "Failed to release qp range base:%d cnt:%d\n", 306 base_qpn, cnt); 307 } 308 } else 309 __mlx4_qp_release_range(dev, base_qpn, cnt); 310} 311EXPORT_SYMBOL_GPL(mlx4_qp_release_range); 312 313int __mlx4_qp_alloc_icm(struct mlx4_dev *dev, int qpn) 314{ 315 struct mlx4_priv *priv = mlx4_priv(dev); 316 struct mlx4_qp_table *qp_table = &priv->qp_table; 317 int err; 318 319 err = mlx4_table_get(dev, &qp_table->qp_table, qpn); 320 if (err) 321 goto err_out; 322 323 err = mlx4_table_get(dev, &qp_table->auxc_table, qpn); 324 if (err) 325 goto err_put_qp; 326 327 err = mlx4_table_get(dev, &qp_table->altc_table, qpn); 328 if (err) 329 goto err_put_auxc; 330 331 err = mlx4_table_get(dev, &qp_table->rdmarc_table, qpn); 332 if (err) 333 goto err_put_altc; 334 335 err = mlx4_table_get(dev, &qp_table->cmpt_table, qpn); 336 if (err) 337 goto err_put_rdmarc; 338 339 return 0; 340 341err_put_rdmarc: 342 mlx4_table_put(dev, &qp_table->rdmarc_table, qpn); 343 344err_put_altc: 345 mlx4_table_put(dev, &qp_table->altc_table, qpn); 346 347err_put_auxc: 348 mlx4_table_put(dev, &qp_table->auxc_table, qpn); 349 350err_put_qp: 351 mlx4_table_put(dev, &qp_table->qp_table, qpn); 352 353err_out: 354 return err; 355} 356 357static int mlx4_qp_alloc_icm(struct mlx4_dev *dev, int qpn) 358{ 359 u64 param = 0; 360 361 if (mlx4_is_mfunc(dev)) { 362 set_param_l(¶m, qpn); 363 return mlx4_cmd_imm(dev, param, ¶m, RES_QP, RES_OP_MAP_ICM, 364 MLX4_CMD_ALLOC_RES, MLX4_CMD_TIME_CLASS_A, 365 MLX4_CMD_WRAPPED); 366 } 367 return __mlx4_qp_alloc_icm(dev, qpn); 368} 369 370void __mlx4_qp_free_icm(struct mlx4_dev *dev, int qpn) 371{ 372 struct mlx4_priv *priv = mlx4_priv(dev); 373 struct mlx4_qp_table *qp_table = &priv->qp_table; 374 375 mlx4_table_put(dev, &qp_table->cmpt_table, qpn); 376 mlx4_table_put(dev, &qp_table->rdmarc_table, qpn); 377 mlx4_table_put(dev, &qp_table->altc_table, qpn); 378 mlx4_table_put(dev, &qp_table->auxc_table, qpn); 379 mlx4_table_put(dev, &qp_table->qp_table, qpn); 380} 381 382static void mlx4_qp_free_icm(struct mlx4_dev *dev, int qpn) 383{ 384 u64 in_param = 0; 385 386 if (mlx4_is_mfunc(dev)) { 387 set_param_l(&in_param, qpn); 388 if (mlx4_cmd(dev, in_param, RES_QP, RES_OP_MAP_ICM, 389 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A, 390 MLX4_CMD_WRAPPED)) 391 mlx4_warn(dev, "Failed to free icm of qp:%d\n", qpn); 392 } else 393 __mlx4_qp_free_icm(dev, qpn); 394} 395 396struct mlx4_qp *mlx4_qp_lookup(struct mlx4_dev *dev, u32 qpn) 397{ 398 struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table; 399 struct mlx4_qp *qp; 400 401 spin_lock_irq(&qp_table->lock); 402 403 qp = __mlx4_qp_lookup(dev, qpn); 404 405 spin_unlock_irq(&qp_table->lock); 406 return qp; 407} 408 409int mlx4_qp_alloc(struct mlx4_dev *dev, int qpn, struct mlx4_qp *qp) 410{ 411 struct mlx4_priv *priv = mlx4_priv(dev); 412 struct mlx4_qp_table *qp_table = &priv->qp_table; 413 int err; 414 415 if (!qpn) 416 return -EINVAL; 417 418 qp->qpn = qpn; 419 420 err = mlx4_qp_alloc_icm(dev, qpn); 421 if (err) 422 return err; 423 424 spin_lock_irq(&qp_table->lock); 425 err = radix_tree_insert(&dev->qp_table_tree, qp->qpn & 426 (dev->caps.num_qps - 1), qp); 427 spin_unlock_irq(&qp_table->lock); 428 if (err) 429 goto err_icm; 430 431 refcount_set(&qp->refcount, 1); 432 init_completion(&qp->free); 433 434 return 0; 435 436err_icm: 437 mlx4_qp_free_icm(dev, qpn); 438 return err; 439} 440 441EXPORT_SYMBOL_GPL(mlx4_qp_alloc); 442 443int mlx4_update_qp(struct mlx4_dev *dev, u32 qpn, 444 enum mlx4_update_qp_attr attr, 445 struct mlx4_update_qp_params *params) 446{ 447 struct mlx4_cmd_mailbox *mailbox; 448 struct mlx4_update_qp_context *cmd; 449 u64 pri_addr_path_mask = 0; 450 u64 qp_mask = 0; 451 int err = 0; 452 453 if (!attr || (attr & ~MLX4_UPDATE_QP_SUPPORTED_ATTRS)) 454 return -EINVAL; 455 456 mailbox = mlx4_alloc_cmd_mailbox(dev); 457 if (IS_ERR(mailbox)) 458 return PTR_ERR(mailbox); 459 460 cmd = (struct mlx4_update_qp_context *)mailbox->buf; 461 462 if (attr & MLX4_UPDATE_QP_SMAC) { 463 pri_addr_path_mask |= 1ULL << MLX4_UPD_QP_PATH_MASK_MAC_INDEX; 464 cmd->qp_context.pri_path.grh_mylmc = params->smac_index; 465 } 466 467 if (attr & MLX4_UPDATE_QP_ETH_SRC_CHECK_MC_LB) { 468 if (!(dev->caps.flags2 469 & MLX4_DEV_CAP_FLAG2_UPDATE_QP_SRC_CHECK_LB)) { 470 mlx4_warn(dev, 471 "Trying to set src check LB, but it isn't supported\n"); 472 err = -EOPNOTSUPP; 473 goto out; 474 } 475 pri_addr_path_mask |= 476 1ULL << MLX4_UPD_QP_PATH_MASK_ETH_SRC_CHECK_MC_LB; 477 if (params->flags & 478 MLX4_UPDATE_QP_PARAMS_FLAGS_ETH_CHECK_MC_LB) { 479 cmd->qp_context.pri_path.fl |= 480 MLX4_FL_ETH_SRC_CHECK_MC_LB; 481 } 482 } 483 484 if (attr & MLX4_UPDATE_QP_VSD) { 485 qp_mask |= 1ULL << MLX4_UPD_QP_MASK_VSD; 486 if (params->flags & MLX4_UPDATE_QP_PARAMS_FLAGS_VSD_ENABLE) 487 cmd->qp_context.param3 |= cpu_to_be32(MLX4_STRIP_VLAN); 488 } 489 490 if (attr & MLX4_UPDATE_QP_RATE_LIMIT) { 491 qp_mask |= 1ULL << MLX4_UPD_QP_MASK_RATE_LIMIT; 492 cmd->qp_context.rate_limit_params = cpu_to_be16((params->rate_unit << 14) | params->rate_val); 493 } 494 495 if (attr & MLX4_UPDATE_QP_QOS_VPORT) { 496 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QOS_VPP)) { 497 mlx4_warn(dev, "Granular QoS per VF is not enabled\n"); 498 err = -EOPNOTSUPP; 499 goto out; 500 } 501 502 qp_mask |= 1ULL << MLX4_UPD_QP_MASK_QOS_VPP; 503 cmd->qp_context.qos_vport = params->qos_vport; 504 } 505 506 cmd->primary_addr_path_mask = cpu_to_be64(pri_addr_path_mask); 507 cmd->qp_mask = cpu_to_be64(qp_mask); 508 509 err = mlx4_cmd(dev, mailbox->dma, qpn & 0xffffff, 0, 510 MLX4_CMD_UPDATE_QP, MLX4_CMD_TIME_CLASS_A, 511 MLX4_CMD_NATIVE); 512out: 513 mlx4_free_cmd_mailbox(dev, mailbox); 514 return err; 515} 516EXPORT_SYMBOL_GPL(mlx4_update_qp); 517 518void mlx4_qp_remove(struct mlx4_dev *dev, struct mlx4_qp *qp) 519{ 520 struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table; 521 unsigned long flags; 522 523 spin_lock_irqsave(&qp_table->lock, flags); 524 radix_tree_delete(&dev->qp_table_tree, qp->qpn & (dev->caps.num_qps - 1)); 525 spin_unlock_irqrestore(&qp_table->lock, flags); 526} 527EXPORT_SYMBOL_GPL(mlx4_qp_remove); 528 529void mlx4_qp_free(struct mlx4_dev *dev, struct mlx4_qp *qp) 530{ 531 mlx4_put_qp(qp); 532 wait_for_completion(&qp->free); 533 534 mlx4_qp_free_icm(dev, qp->qpn); 535} 536EXPORT_SYMBOL_GPL(mlx4_qp_free); 537 538static int mlx4_CONF_SPECIAL_QP(struct mlx4_dev *dev, u32 base_qpn) 539{ 540 return mlx4_cmd(dev, 0, base_qpn, 0, MLX4_CMD_CONF_SPECIAL_QP, 541 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE); 542} 543 544#define MLX4_QP_TABLE_RSS_ETH_PRIORITY 2 545#define MLX4_QP_TABLE_RAW_ETH_PRIORITY 1 546#define MLX4_QP_TABLE_RAW_ETH_SIZE 256 547 548static int mlx4_create_zones(struct mlx4_dev *dev, 549 u32 reserved_bottom_general, 550 u32 reserved_top_general, 551 u32 reserved_bottom_rss, 552 u32 start_offset_rss, 553 u32 max_table_offset) 554{ 555 struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table; 556 struct mlx4_bitmap (*bitmap)[MLX4_QP_TABLE_ZONE_NUM] = NULL; 557 int bitmap_initialized = 0; 558 u32 last_offset; 559 int k; 560 int err; 561 562 qp_table->zones = mlx4_zone_allocator_create(MLX4_ZONE_ALLOC_FLAGS_NO_OVERLAP); 563 564 if (NULL == qp_table->zones) 565 return -ENOMEM; 566 567 bitmap = kmalloc(sizeof(*bitmap), GFP_KERNEL); 568 569 if (NULL == bitmap) { 570 err = -ENOMEM; 571 goto free_zone; 572 } 573 574 err = mlx4_bitmap_init(*bitmap + MLX4_QP_TABLE_ZONE_GENERAL, dev->caps.num_qps, 575 (1 << 23) - 1, reserved_bottom_general, 576 reserved_top_general); 577 578 if (err) 579 goto free_bitmap; 580 581 ++bitmap_initialized; 582 583 err = mlx4_zone_add_one(qp_table->zones, *bitmap + MLX4_QP_TABLE_ZONE_GENERAL, 584 MLX4_ZONE_FALLBACK_TO_HIGHER_PRIO | 585 MLX4_ZONE_USE_RR, 0, 586 0, qp_table->zones_uids + MLX4_QP_TABLE_ZONE_GENERAL); 587 588 if (err) 589 goto free_bitmap; 590 591 err = mlx4_bitmap_init(*bitmap + MLX4_QP_TABLE_ZONE_RSS, 592 reserved_bottom_rss, 593 reserved_bottom_rss - 1, 594 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 595 reserved_bottom_rss - start_offset_rss); 596 597 if (err) 598 goto free_bitmap; 599 600 ++bitmap_initialized; 601 602 err = mlx4_zone_add_one(qp_table->zones, *bitmap + MLX4_QP_TABLE_ZONE_RSS, 603 MLX4_ZONE_ALLOW_ALLOC_FROM_LOWER_PRIO | 604 MLX4_ZONE_ALLOW_ALLOC_FROM_EQ_PRIO | 605 MLX4_ZONE_USE_RR, MLX4_QP_TABLE_RSS_ETH_PRIORITY, 606 0, qp_table->zones_uids + MLX4_QP_TABLE_ZONE_RSS); 607 608 if (err) 609 goto free_bitmap; 610 611 last_offset = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW]; 612 /* We have a single zone for the A0 steering QPs area of the FW. This area 613 * needs to be split into subareas. One set of subareas is for RSS QPs 614 * (in which qp number bits 6 and/or 7 are set); the other set of subareas 615 * is for RAW_ETH QPs, which require that both bits 6 and 7 are zero. 616 * Currently, the values returned by the FW (A0 steering area starting qp number 617 * and A0 steering area size) are such that there are only two subareas -- one 618 * for RSS and one for RAW_ETH. 619 */ 620 for (k = MLX4_QP_TABLE_ZONE_RSS + 1; k < sizeof(*bitmap)/sizeof((*bitmap)[0]); 621 k++) { 622 int size; 623 u32 offset = start_offset_rss; 624 u32 bf_mask; 625 u32 requested_size; 626 627 /* Assuming MLX4_BF_QP_SKIP_MASK is consecutive ones, this calculates 628 * a mask of all LSB bits set until (and not including) the first 629 * set bit of MLX4_BF_QP_SKIP_MASK. For example, if MLX4_BF_QP_SKIP_MASK 630 * is 0xc0, bf_mask will be 0x3f. 631 */ 632 bf_mask = (MLX4_BF_QP_SKIP_MASK & ~(MLX4_BF_QP_SKIP_MASK - 1)) - 1; 633 requested_size = min((u32)MLX4_QP_TABLE_RAW_ETH_SIZE, bf_mask + 1); 634 635 if (((last_offset & MLX4_BF_QP_SKIP_MASK) && 636 ((int)(max_table_offset - last_offset)) >= 637 roundup_pow_of_two(MLX4_BF_QP_SKIP_MASK)) || 638 (!(last_offset & MLX4_BF_QP_SKIP_MASK) && 639 !((last_offset + requested_size - 1) & 640 MLX4_BF_QP_SKIP_MASK))) 641 size = requested_size; 642 else { 643 u32 candidate_offset = 644 (last_offset | MLX4_BF_QP_SKIP_MASK | bf_mask) + 1; 645 646 if (last_offset & MLX4_BF_QP_SKIP_MASK) 647 last_offset = candidate_offset; 648 649 /* From this point, the BF bits are 0 */ 650 651 if (last_offset > max_table_offset) { 652 /* need to skip */ 653 size = -1; 654 } else { 655 size = min3(max_table_offset - last_offset, 656 bf_mask - (last_offset & bf_mask), 657 requested_size); 658 if (size < requested_size) { 659 int candidate_size; 660 661 candidate_size = min3( 662 max_table_offset - candidate_offset, 663 bf_mask - (last_offset & bf_mask), 664 requested_size); 665 666 /* We will not take this path if last_offset was 667 * already set above to candidate_offset 668 */ 669 if (candidate_size > size) { 670 last_offset = candidate_offset; 671 size = candidate_size; 672 } 673 } 674 } 675 } 676 677 if (size > 0) { 678 /* mlx4_bitmap_alloc_range will find a contiguous range of "size" 679 * QPs in which both bits 6 and 7 are zero, because we pass it the 680 * MLX4_BF_SKIP_MASK). 681 */ 682 offset = mlx4_bitmap_alloc_range( 683 *bitmap + MLX4_QP_TABLE_ZONE_RSS, 684 size, 1, 685 MLX4_BF_QP_SKIP_MASK); 686 687 if (offset == (u32)-1) { 688 err = -ENOMEM; 689 break; 690 } 691 692 last_offset = offset + size; 693 694 err = mlx4_bitmap_init(*bitmap + k, roundup_pow_of_two(size), 695 roundup_pow_of_two(size) - 1, 0, 696 roundup_pow_of_two(size) - size); 697 } else { 698 /* Add an empty bitmap, we'll allocate from different zones (since 699 * at least one is reserved) 700 */ 701 err = mlx4_bitmap_init(*bitmap + k, 1, 702 MLX4_QP_TABLE_RAW_ETH_SIZE - 1, 0, 703 0); 704 if (!err) 705 mlx4_bitmap_alloc_range(*bitmap + k, 1, 1, 0); 706 } 707 708 if (err) 709 break; 710 711 ++bitmap_initialized; 712 713 err = mlx4_zone_add_one(qp_table->zones, *bitmap + k, 714 MLX4_ZONE_ALLOW_ALLOC_FROM_LOWER_PRIO | 715 MLX4_ZONE_ALLOW_ALLOC_FROM_EQ_PRIO | 716 MLX4_ZONE_USE_RR, MLX4_QP_TABLE_RAW_ETH_PRIORITY, 717 offset, qp_table->zones_uids + k); 718 719 if (err) 720 break; 721 } 722 723 if (err) 724 goto free_bitmap; 725 726 qp_table->bitmap_gen = *bitmap; 727 728 return err; 729 730free_bitmap: 731 for (k = 0; k < bitmap_initialized; k++) 732 mlx4_bitmap_cleanup(*bitmap + k); 733 kfree(bitmap); 734free_zone: 735 mlx4_zone_allocator_destroy(qp_table->zones); 736 return err; 737} 738 739static void mlx4_cleanup_qp_zones(struct mlx4_dev *dev) 740{ 741 struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table; 742 743 if (qp_table->zones) { 744 int i; 745 746 for (i = 0; 747 i < ARRAY_SIZE(qp_table->zones_uids); 748 i++) { 749 struct mlx4_bitmap *bitmap = 750 mlx4_zone_get_bitmap(qp_table->zones, 751 qp_table->zones_uids[i]); 752 753 mlx4_zone_remove_one(qp_table->zones, qp_table->zones_uids[i]); 754 if (NULL == bitmap) 755 continue; 756 757 mlx4_bitmap_cleanup(bitmap); 758 } 759 mlx4_zone_allocator_destroy(qp_table->zones); 760 kfree(qp_table->bitmap_gen); 761 qp_table->bitmap_gen = NULL; 762 qp_table->zones = NULL; 763 } 764} 765 766int mlx4_init_qp_table(struct mlx4_dev *dev) 767{ 768 struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table; 769 int err; 770 int reserved_from_top = 0; 771 int reserved_from_bot; 772 int k; 773 int fixed_reserved_from_bot_rv = 0; 774 int bottom_reserved_for_rss_bitmap; 775 u32 max_table_offset = dev->caps.dmfs_high_rate_qpn_base + 776 dev->caps.dmfs_high_rate_qpn_range; 777 778 spin_lock_init(&qp_table->lock); 779 INIT_RADIX_TREE(&dev->qp_table_tree, GFP_ATOMIC); 780 if (mlx4_is_slave(dev)) 781 return 0; 782 783 /* We reserve 2 extra QPs per port for the special QPs. The 784 * block of special QPs must be aligned to a multiple of 8, so 785 * round up. 786 * 787 * We also reserve the MSB of the 24-bit QP number to indicate 788 * that a QP is an XRC QP. 789 */ 790 for (k = 0; k <= MLX4_QP_REGION_BOTTOM; k++) 791 fixed_reserved_from_bot_rv += dev->caps.reserved_qps_cnt[k]; 792 793 if (fixed_reserved_from_bot_rv < max_table_offset) 794 fixed_reserved_from_bot_rv = max_table_offset; 795 796 /* We reserve at least 1 extra for bitmaps that we don't have enough space for*/ 797 bottom_reserved_for_rss_bitmap = 798 roundup_pow_of_two(fixed_reserved_from_bot_rv + 1); 799 dev->phys_caps.base_sqpn = ALIGN(bottom_reserved_for_rss_bitmap, 8); 800 801 { 802 int sort[MLX4_NUM_QP_REGION]; 803 int i, j; 804 int last_base = dev->caps.num_qps; 805 806 for (i = 1; i < MLX4_NUM_QP_REGION; ++i) 807 sort[i] = i; 808 809 for (i = MLX4_NUM_QP_REGION; i > MLX4_QP_REGION_BOTTOM; --i) { 810 for (j = MLX4_QP_REGION_BOTTOM + 2; j < i; ++j) { 811 if (dev->caps.reserved_qps_cnt[sort[j]] > 812 dev->caps.reserved_qps_cnt[sort[j - 1]]) 813 swap(sort[j], sort[j - 1]); 814 } 815 } 816 817 for (i = MLX4_QP_REGION_BOTTOM + 1; i < MLX4_NUM_QP_REGION; ++i) { 818 last_base -= dev->caps.reserved_qps_cnt[sort[i]]; 819 dev->caps.reserved_qps_base[sort[i]] = last_base; 820 reserved_from_top += 821 dev->caps.reserved_qps_cnt[sort[i]]; 822 } 823 } 824 825 /* Reserve 8 real SQPs in both native and SRIOV modes. 826 * In addition, in SRIOV mode, reserve 8 proxy SQPs per function 827 * (for all PFs and VFs), and 8 corresponding tunnel QPs. 828 * Each proxy SQP works opposite its own tunnel QP. 829 * 830 * The QPs are arranged as follows: 831 * a. 8 real SQPs 832 * b. All the proxy SQPs (8 per function) 833 * c. All the tunnel QPs (8 per function) 834 */ 835 reserved_from_bot = mlx4_num_reserved_sqps(dev); 836 if (reserved_from_bot + reserved_from_top > dev->caps.num_qps) { 837 mlx4_err(dev, "Number of reserved QPs is higher than number of QPs\n"); 838 return -EINVAL; 839 } 840 841 err = mlx4_create_zones(dev, reserved_from_bot, reserved_from_bot, 842 bottom_reserved_for_rss_bitmap, 843 fixed_reserved_from_bot_rv, 844 max_table_offset); 845 846 if (err) 847 return err; 848 849 if (mlx4_is_mfunc(dev)) { 850 /* for PPF use */ 851 dev->phys_caps.base_proxy_sqpn = dev->phys_caps.base_sqpn + 8; 852 dev->phys_caps.base_tunnel_sqpn = dev->phys_caps.base_sqpn + 8 + 8 * MLX4_MFUNC_MAX; 853 854 /* In mfunc, calculate proxy and tunnel qp offsets for the PF here, 855 * since the PF does not call mlx4_slave_caps */ 856 dev->caps.spec_qps = kcalloc(dev->caps.num_ports, 857 sizeof(*dev->caps.spec_qps), 858 GFP_KERNEL); 859 if (!dev->caps.spec_qps) { 860 err = -ENOMEM; 861 goto err_mem; 862 } 863 864 for (k = 0; k < dev->caps.num_ports; k++) { 865 dev->caps.spec_qps[k].qp0_proxy = dev->phys_caps.base_proxy_sqpn + 866 8 * mlx4_master_func_num(dev) + k; 867 dev->caps.spec_qps[k].qp0_tunnel = dev->caps.spec_qps[k].qp0_proxy + 8 * MLX4_MFUNC_MAX; 868 dev->caps.spec_qps[k].qp1_proxy = dev->phys_caps.base_proxy_sqpn + 869 8 * mlx4_master_func_num(dev) + MLX4_MAX_PORTS + k; 870 dev->caps.spec_qps[k].qp1_tunnel = dev->caps.spec_qps[k].qp1_proxy + 8 * MLX4_MFUNC_MAX; 871 } 872 } 873 874 875 err = mlx4_CONF_SPECIAL_QP(dev, dev->phys_caps.base_sqpn); 876 if (err) 877 goto err_mem; 878 879 return err; 880 881err_mem: 882 kfree(dev->caps.spec_qps); 883 dev->caps.spec_qps = NULL; 884 mlx4_cleanup_qp_zones(dev); 885 return err; 886} 887 888void mlx4_cleanup_qp_table(struct mlx4_dev *dev) 889{ 890 if (mlx4_is_slave(dev)) 891 return; 892 893 mlx4_CONF_SPECIAL_QP(dev, 0); 894 895 mlx4_cleanup_qp_zones(dev); 896} 897 898int mlx4_qp_query(struct mlx4_dev *dev, struct mlx4_qp *qp, 899 struct mlx4_qp_context *context) 900{ 901 struct mlx4_cmd_mailbox *mailbox; 902 int err; 903 904 mailbox = mlx4_alloc_cmd_mailbox(dev); 905 if (IS_ERR(mailbox)) 906 return PTR_ERR(mailbox); 907 908 err = mlx4_cmd_box(dev, 0, mailbox->dma, qp->qpn, 0, 909 MLX4_CMD_QUERY_QP, MLX4_CMD_TIME_CLASS_A, 910 MLX4_CMD_WRAPPED); 911 if (!err) 912 memcpy(context, mailbox->buf + 8, sizeof(*context)); 913 914 mlx4_free_cmd_mailbox(dev, mailbox); 915 return err; 916} 917EXPORT_SYMBOL_GPL(mlx4_qp_query); 918 919int mlx4_qp_to_ready(struct mlx4_dev *dev, struct mlx4_mtt *mtt, 920 struct mlx4_qp_context *context, 921 struct mlx4_qp *qp, enum mlx4_qp_state *qp_state) 922{ 923 int err; 924 int i; 925 static const enum mlx4_qp_state states[] = { 926 MLX4_QP_STATE_RST, 927 MLX4_QP_STATE_INIT, 928 MLX4_QP_STATE_RTR, 929 MLX4_QP_STATE_RTS 930 }; 931 932 for (i = 0; i < ARRAY_SIZE(states) - 1; i++) { 933 context->flags &= cpu_to_be32(~(0xf << 28)); 934 context->flags |= cpu_to_be32(states[i + 1] << 28); 935 if (states[i + 1] != MLX4_QP_STATE_RTR) 936 context->params2 &= ~cpu_to_be32(MLX4_QP_BIT_FPP); 937 err = mlx4_qp_modify(dev, mtt, states[i], states[i + 1], 938 context, 0, 0, qp); 939 if (err) { 940 mlx4_err(dev, "Failed to bring QP to state: %d with error: %d\n", 941 states[i + 1], err); 942 return err; 943 } 944 945 *qp_state = states[i + 1]; 946 } 947 948 return 0; 949} 950EXPORT_SYMBOL_GPL(mlx4_qp_to_ready); 951 952u16 mlx4_qp_roce_entropy(struct mlx4_dev *dev, u32 qpn) 953{ 954 struct mlx4_qp_context context; 955 struct mlx4_qp qp; 956 int err; 957 958 qp.qpn = qpn; 959 err = mlx4_qp_query(dev, &qp, &context); 960 if (!err) { 961 u32 dest_qpn = be32_to_cpu(context.remote_qpn) & 0xffffff; 962 u16 folded_dst = folded_qp(dest_qpn); 963 u16 folded_src = folded_qp(qpn); 964 965 return (dest_qpn != qpn) ? 966 ((folded_dst ^ folded_src) | 0xC000) : 967 folded_src | 0xC000; 968 } 969 return 0xdead; 970} 971