1 /* 2 * Copyright (c) 2009-2010 Chelsio, 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 #include <linux/module.h> 33 #include <linux/moduleparam.h> 34 #include <linux/debugfs.h> 35 #include <linux/vmalloc.h> 36 37 #include <rdma/ib_verbs.h> 38 39 #include "iw_cxgb4.h" 40 41 #define DRV_VERSION "0.1" 42 43 MODULE_AUTHOR("Steve Wise"); 44 MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver"); 45 MODULE_LICENSE("Dual BSD/GPL"); 46 MODULE_VERSION(DRV_VERSION); 47 48 static int allow_db_fc_on_t5; 49 module_param(allow_db_fc_on_t5, int, 0644); 50 MODULE_PARM_DESC(allow_db_fc_on_t5, 51 "Allow DB Flow Control on T5 (default = 0)"); 52 53 static int allow_db_coalescing_on_t5; 54 module_param(allow_db_coalescing_on_t5, int, 0644); 55 MODULE_PARM_DESC(allow_db_coalescing_on_t5, 56 "Allow DB Coalescing on T5 (default = 0)"); 57 58 struct uld_ctx { 59 struct list_head entry; 60 struct cxgb4_lld_info lldi; 61 struct c4iw_dev *dev; 62 }; 63 64 static LIST_HEAD(uld_ctx_list); 65 static DEFINE_MUTEX(dev_mutex); 66 67 static struct dentry *c4iw_debugfs_root; 68 69 struct c4iw_debugfs_data { 70 struct c4iw_dev *devp; 71 char *buf; 72 int bufsize; 73 int pos; 74 }; 75 76 static int count_idrs(int id, void *p, void *data) 77 { 78 int *countp = data; 79 80 *countp = *countp + 1; 81 return 0; 82 } 83 84 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count, 85 loff_t *ppos) 86 { 87 struct c4iw_debugfs_data *d = file->private_data; 88 89 return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos); 90 } 91 92 static int dump_qp(int id, void *p, void *data) 93 { 94 struct c4iw_qp *qp = p; 95 struct c4iw_debugfs_data *qpd = data; 96 int space; 97 int cc; 98 99 if (id != qp->wq.sq.qid) 100 return 0; 101 102 space = qpd->bufsize - qpd->pos - 1; 103 if (space == 0) 104 return 1; 105 106 if (qp->ep) { 107 if (qp->ep->com.local_addr.ss_family == AF_INET) { 108 struct sockaddr_in *lsin = (struct sockaddr_in *) 109 &qp->ep->com.local_addr; 110 struct sockaddr_in *rsin = (struct sockaddr_in *) 111 &qp->ep->com.remote_addr; 112 113 cc = snprintf(qpd->buf + qpd->pos, space, 114 "rc qp sq id %u rq id %u state %u " 115 "onchip %u ep tid %u state %u " 116 "%pI4:%u->%pI4:%u\n", 117 qp->wq.sq.qid, qp->wq.rq.qid, 118 (int)qp->attr.state, 119 qp->wq.sq.flags & T4_SQ_ONCHIP, 120 qp->ep->hwtid, (int)qp->ep->com.state, 121 &lsin->sin_addr, ntohs(lsin->sin_port), 122 &rsin->sin_addr, ntohs(rsin->sin_port)); 123 } else { 124 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *) 125 &qp->ep->com.local_addr; 126 struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *) 127 &qp->ep->com.remote_addr; 128 129 cc = snprintf(qpd->buf + qpd->pos, space, 130 "rc qp sq id %u rq id %u state %u " 131 "onchip %u ep tid %u state %u " 132 "%pI6:%u->%pI6:%u\n", 133 qp->wq.sq.qid, qp->wq.rq.qid, 134 (int)qp->attr.state, 135 qp->wq.sq.flags & T4_SQ_ONCHIP, 136 qp->ep->hwtid, (int)qp->ep->com.state, 137 &lsin6->sin6_addr, 138 ntohs(lsin6->sin6_port), 139 &rsin6->sin6_addr, 140 ntohs(rsin6->sin6_port)); 141 } 142 } else 143 cc = snprintf(qpd->buf + qpd->pos, space, 144 "qp sq id %u rq id %u state %u onchip %u\n", 145 qp->wq.sq.qid, qp->wq.rq.qid, 146 (int)qp->attr.state, 147 qp->wq.sq.flags & T4_SQ_ONCHIP); 148 if (cc < space) 149 qpd->pos += cc; 150 return 0; 151 } 152 153 static int qp_release(struct inode *inode, struct file *file) 154 { 155 struct c4iw_debugfs_data *qpd = file->private_data; 156 if (!qpd) { 157 printk(KERN_INFO "%s null qpd?\n", __func__); 158 return 0; 159 } 160 vfree(qpd->buf); 161 kfree(qpd); 162 return 0; 163 } 164 165 static int qp_open(struct inode *inode, struct file *file) 166 { 167 struct c4iw_debugfs_data *qpd; 168 int ret = 0; 169 int count = 1; 170 171 qpd = kmalloc(sizeof *qpd, GFP_KERNEL); 172 if (!qpd) { 173 ret = -ENOMEM; 174 goto out; 175 } 176 qpd->devp = inode->i_private; 177 qpd->pos = 0; 178 179 spin_lock_irq(&qpd->devp->lock); 180 idr_for_each(&qpd->devp->qpidr, count_idrs, &count); 181 spin_unlock_irq(&qpd->devp->lock); 182 183 qpd->bufsize = count * 128; 184 qpd->buf = vmalloc(qpd->bufsize); 185 if (!qpd->buf) { 186 ret = -ENOMEM; 187 goto err1; 188 } 189 190 spin_lock_irq(&qpd->devp->lock); 191 idr_for_each(&qpd->devp->qpidr, dump_qp, qpd); 192 spin_unlock_irq(&qpd->devp->lock); 193 194 qpd->buf[qpd->pos++] = 0; 195 file->private_data = qpd; 196 goto out; 197 err1: 198 kfree(qpd); 199 out: 200 return ret; 201 } 202 203 static const struct file_operations qp_debugfs_fops = { 204 .owner = THIS_MODULE, 205 .open = qp_open, 206 .release = qp_release, 207 .read = debugfs_read, 208 .llseek = default_llseek, 209 }; 210 211 static int dump_stag(int id, void *p, void *data) 212 { 213 struct c4iw_debugfs_data *stagd = data; 214 int space; 215 int cc; 216 217 space = stagd->bufsize - stagd->pos - 1; 218 if (space == 0) 219 return 1; 220 221 cc = snprintf(stagd->buf + stagd->pos, space, "0x%x\n", id<<8); 222 if (cc < space) 223 stagd->pos += cc; 224 return 0; 225 } 226 227 static int stag_release(struct inode *inode, struct file *file) 228 { 229 struct c4iw_debugfs_data *stagd = file->private_data; 230 if (!stagd) { 231 printk(KERN_INFO "%s null stagd?\n", __func__); 232 return 0; 233 } 234 kfree(stagd->buf); 235 kfree(stagd); 236 return 0; 237 } 238 239 static int stag_open(struct inode *inode, struct file *file) 240 { 241 struct c4iw_debugfs_data *stagd; 242 int ret = 0; 243 int count = 1; 244 245 stagd = kmalloc(sizeof *stagd, GFP_KERNEL); 246 if (!stagd) { 247 ret = -ENOMEM; 248 goto out; 249 } 250 stagd->devp = inode->i_private; 251 stagd->pos = 0; 252 253 spin_lock_irq(&stagd->devp->lock); 254 idr_for_each(&stagd->devp->mmidr, count_idrs, &count); 255 spin_unlock_irq(&stagd->devp->lock); 256 257 stagd->bufsize = count * sizeof("0x12345678\n"); 258 stagd->buf = kmalloc(stagd->bufsize, GFP_KERNEL); 259 if (!stagd->buf) { 260 ret = -ENOMEM; 261 goto err1; 262 } 263 264 spin_lock_irq(&stagd->devp->lock); 265 idr_for_each(&stagd->devp->mmidr, dump_stag, stagd); 266 spin_unlock_irq(&stagd->devp->lock); 267 268 stagd->buf[stagd->pos++] = 0; 269 file->private_data = stagd; 270 goto out; 271 err1: 272 kfree(stagd); 273 out: 274 return ret; 275 } 276 277 static const struct file_operations stag_debugfs_fops = { 278 .owner = THIS_MODULE, 279 .open = stag_open, 280 .release = stag_release, 281 .read = debugfs_read, 282 .llseek = default_llseek, 283 }; 284 285 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY"}; 286 287 static int stats_show(struct seq_file *seq, void *v) 288 { 289 struct c4iw_dev *dev = seq->private; 290 291 seq_printf(seq, " Object: %10s %10s %10s %10s\n", "Total", "Current", 292 "Max", "Fail"); 293 seq_printf(seq, " PDID: %10llu %10llu %10llu %10llu\n", 294 dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur, 295 dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail); 296 seq_printf(seq, " QID: %10llu %10llu %10llu %10llu\n", 297 dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur, 298 dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail); 299 seq_printf(seq, " TPTMEM: %10llu %10llu %10llu %10llu\n", 300 dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur, 301 dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail); 302 seq_printf(seq, " PBLMEM: %10llu %10llu %10llu %10llu\n", 303 dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur, 304 dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail); 305 seq_printf(seq, " RQTMEM: %10llu %10llu %10llu %10llu\n", 306 dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur, 307 dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail); 308 seq_printf(seq, " OCQPMEM: %10llu %10llu %10llu %10llu\n", 309 dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur, 310 dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail); 311 seq_printf(seq, " DB FULL: %10llu\n", dev->rdev.stats.db_full); 312 seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty); 313 seq_printf(seq, " DB DROP: %10llu\n", dev->rdev.stats.db_drop); 314 seq_printf(seq, " DB State: %s Transitions %llu\n", 315 db_state_str[dev->db_state], 316 dev->rdev.stats.db_state_transitions); 317 seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full); 318 seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n", 319 dev->rdev.stats.act_ofld_conn_fails); 320 seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n", 321 dev->rdev.stats.pas_ofld_conn_fails); 322 return 0; 323 } 324 325 static int stats_open(struct inode *inode, struct file *file) 326 { 327 return single_open(file, stats_show, inode->i_private); 328 } 329 330 static ssize_t stats_clear(struct file *file, const char __user *buf, 331 size_t count, loff_t *pos) 332 { 333 struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private; 334 335 mutex_lock(&dev->rdev.stats.lock); 336 dev->rdev.stats.pd.max = 0; 337 dev->rdev.stats.pd.fail = 0; 338 dev->rdev.stats.qid.max = 0; 339 dev->rdev.stats.qid.fail = 0; 340 dev->rdev.stats.stag.max = 0; 341 dev->rdev.stats.stag.fail = 0; 342 dev->rdev.stats.pbl.max = 0; 343 dev->rdev.stats.pbl.fail = 0; 344 dev->rdev.stats.rqt.max = 0; 345 dev->rdev.stats.rqt.fail = 0; 346 dev->rdev.stats.ocqp.max = 0; 347 dev->rdev.stats.ocqp.fail = 0; 348 dev->rdev.stats.db_full = 0; 349 dev->rdev.stats.db_empty = 0; 350 dev->rdev.stats.db_drop = 0; 351 dev->rdev.stats.db_state_transitions = 0; 352 dev->rdev.stats.tcam_full = 0; 353 dev->rdev.stats.act_ofld_conn_fails = 0; 354 dev->rdev.stats.pas_ofld_conn_fails = 0; 355 mutex_unlock(&dev->rdev.stats.lock); 356 return count; 357 } 358 359 static const struct file_operations stats_debugfs_fops = { 360 .owner = THIS_MODULE, 361 .open = stats_open, 362 .release = single_release, 363 .read = seq_read, 364 .llseek = seq_lseek, 365 .write = stats_clear, 366 }; 367 368 static int dump_ep(int id, void *p, void *data) 369 { 370 struct c4iw_ep *ep = p; 371 struct c4iw_debugfs_data *epd = data; 372 int space; 373 int cc; 374 375 space = epd->bufsize - epd->pos - 1; 376 if (space == 0) 377 return 1; 378 379 if (ep->com.local_addr.ss_family == AF_INET) { 380 struct sockaddr_in *lsin = (struct sockaddr_in *) 381 &ep->com.local_addr; 382 struct sockaddr_in *rsin = (struct sockaddr_in *) 383 &ep->com.remote_addr; 384 385 cc = snprintf(epd->buf + epd->pos, space, 386 "ep %p cm_id %p qp %p state %d flags 0x%lx " 387 "history 0x%lx hwtid %d atid %d " 388 "%pI4:%d <-> %pI4:%d\n", 389 ep, ep->com.cm_id, ep->com.qp, 390 (int)ep->com.state, ep->com.flags, 391 ep->com.history, ep->hwtid, ep->atid, 392 &lsin->sin_addr, ntohs(lsin->sin_port), 393 &rsin->sin_addr, ntohs(rsin->sin_port)); 394 } else { 395 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *) 396 &ep->com.local_addr; 397 struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *) 398 &ep->com.remote_addr; 399 400 cc = snprintf(epd->buf + epd->pos, space, 401 "ep %p cm_id %p qp %p state %d flags 0x%lx " 402 "history 0x%lx hwtid %d atid %d " 403 "%pI6:%d <-> %pI6:%d\n", 404 ep, ep->com.cm_id, ep->com.qp, 405 (int)ep->com.state, ep->com.flags, 406 ep->com.history, ep->hwtid, ep->atid, 407 &lsin6->sin6_addr, ntohs(lsin6->sin6_port), 408 &rsin6->sin6_addr, ntohs(rsin6->sin6_port)); 409 } 410 if (cc < space) 411 epd->pos += cc; 412 return 0; 413 } 414 415 static int dump_listen_ep(int id, void *p, void *data) 416 { 417 struct c4iw_listen_ep *ep = p; 418 struct c4iw_debugfs_data *epd = data; 419 int space; 420 int cc; 421 422 space = epd->bufsize - epd->pos - 1; 423 if (space == 0) 424 return 1; 425 426 if (ep->com.local_addr.ss_family == AF_INET) { 427 struct sockaddr_in *lsin = (struct sockaddr_in *) 428 &ep->com.local_addr; 429 430 cc = snprintf(epd->buf + epd->pos, space, 431 "ep %p cm_id %p state %d flags 0x%lx stid %d " 432 "backlog %d %pI4:%d\n", 433 ep, ep->com.cm_id, (int)ep->com.state, 434 ep->com.flags, ep->stid, ep->backlog, 435 &lsin->sin_addr, ntohs(lsin->sin_port)); 436 } else { 437 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *) 438 &ep->com.local_addr; 439 440 cc = snprintf(epd->buf + epd->pos, space, 441 "ep %p cm_id %p state %d flags 0x%lx stid %d " 442 "backlog %d %pI6:%d\n", 443 ep, ep->com.cm_id, (int)ep->com.state, 444 ep->com.flags, ep->stid, ep->backlog, 445 &lsin6->sin6_addr, ntohs(lsin6->sin6_port)); 446 } 447 if (cc < space) 448 epd->pos += cc; 449 return 0; 450 } 451 452 static int ep_release(struct inode *inode, struct file *file) 453 { 454 struct c4iw_debugfs_data *epd = file->private_data; 455 if (!epd) { 456 pr_info("%s null qpd?\n", __func__); 457 return 0; 458 } 459 vfree(epd->buf); 460 kfree(epd); 461 return 0; 462 } 463 464 static int ep_open(struct inode *inode, struct file *file) 465 { 466 struct c4iw_debugfs_data *epd; 467 int ret = 0; 468 int count = 1; 469 470 epd = kmalloc(sizeof(*epd), GFP_KERNEL); 471 if (!epd) { 472 ret = -ENOMEM; 473 goto out; 474 } 475 epd->devp = inode->i_private; 476 epd->pos = 0; 477 478 spin_lock_irq(&epd->devp->lock); 479 idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count); 480 idr_for_each(&epd->devp->atid_idr, count_idrs, &count); 481 idr_for_each(&epd->devp->stid_idr, count_idrs, &count); 482 spin_unlock_irq(&epd->devp->lock); 483 484 epd->bufsize = count * 160; 485 epd->buf = vmalloc(epd->bufsize); 486 if (!epd->buf) { 487 ret = -ENOMEM; 488 goto err1; 489 } 490 491 spin_lock_irq(&epd->devp->lock); 492 idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd); 493 idr_for_each(&epd->devp->atid_idr, dump_ep, epd); 494 idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd); 495 spin_unlock_irq(&epd->devp->lock); 496 497 file->private_data = epd; 498 goto out; 499 err1: 500 kfree(epd); 501 out: 502 return ret; 503 } 504 505 static const struct file_operations ep_debugfs_fops = { 506 .owner = THIS_MODULE, 507 .open = ep_open, 508 .release = ep_release, 509 .read = debugfs_read, 510 }; 511 512 static int setup_debugfs(struct c4iw_dev *devp) 513 { 514 struct dentry *de; 515 516 if (!devp->debugfs_root) 517 return -1; 518 519 de = debugfs_create_file("qps", S_IWUSR, devp->debugfs_root, 520 (void *)devp, &qp_debugfs_fops); 521 if (de && de->d_inode) 522 de->d_inode->i_size = 4096; 523 524 de = debugfs_create_file("stags", S_IWUSR, devp->debugfs_root, 525 (void *)devp, &stag_debugfs_fops); 526 if (de && de->d_inode) 527 de->d_inode->i_size = 4096; 528 529 de = debugfs_create_file("stats", S_IWUSR, devp->debugfs_root, 530 (void *)devp, &stats_debugfs_fops); 531 if (de && de->d_inode) 532 de->d_inode->i_size = 4096; 533 534 de = debugfs_create_file("eps", S_IWUSR, devp->debugfs_root, 535 (void *)devp, &ep_debugfs_fops); 536 if (de && de->d_inode) 537 de->d_inode->i_size = 4096; 538 539 return 0; 540 } 541 542 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev, 543 struct c4iw_dev_ucontext *uctx) 544 { 545 struct list_head *pos, *nxt; 546 struct c4iw_qid_list *entry; 547 548 mutex_lock(&uctx->lock); 549 list_for_each_safe(pos, nxt, &uctx->qpids) { 550 entry = list_entry(pos, struct c4iw_qid_list, entry); 551 list_del_init(&entry->entry); 552 if (!(entry->qid & rdev->qpmask)) { 553 c4iw_put_resource(&rdev->resource.qid_table, 554 entry->qid); 555 mutex_lock(&rdev->stats.lock); 556 rdev->stats.qid.cur -= rdev->qpmask + 1; 557 mutex_unlock(&rdev->stats.lock); 558 } 559 kfree(entry); 560 } 561 562 list_for_each_safe(pos, nxt, &uctx->qpids) { 563 entry = list_entry(pos, struct c4iw_qid_list, entry); 564 list_del_init(&entry->entry); 565 kfree(entry); 566 } 567 mutex_unlock(&uctx->lock); 568 } 569 570 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev, 571 struct c4iw_dev_ucontext *uctx) 572 { 573 INIT_LIST_HEAD(&uctx->qpids); 574 INIT_LIST_HEAD(&uctx->cqids); 575 mutex_init(&uctx->lock); 576 } 577 578 /* Caller takes care of locking if needed */ 579 static int c4iw_rdev_open(struct c4iw_rdev *rdev) 580 { 581 int err; 582 583 c4iw_init_dev_ucontext(rdev, &rdev->uctx); 584 585 /* 586 * qpshift is the number of bits to shift the qpid left in order 587 * to get the correct address of the doorbell for that qp. 588 */ 589 rdev->qpshift = PAGE_SHIFT - ilog2(rdev->lldi.udb_density); 590 rdev->qpmask = rdev->lldi.udb_density - 1; 591 rdev->cqshift = PAGE_SHIFT - ilog2(rdev->lldi.ucq_density); 592 rdev->cqmask = rdev->lldi.ucq_density - 1; 593 PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d " 594 "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x " 595 "qp qid start %u size %u cq qid start %u size %u\n", 596 __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start, 597 rdev->lldi.vr->stag.size, c4iw_num_stags(rdev), 598 rdev->lldi.vr->pbl.start, 599 rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start, 600 rdev->lldi.vr->rq.size, 601 rdev->lldi.vr->qp.start, 602 rdev->lldi.vr->qp.size, 603 rdev->lldi.vr->cq.start, 604 rdev->lldi.vr->cq.size); 605 PDBG("udb len 0x%x udb base %llx db_reg %p gts_reg %p qpshift %lu " 606 "qpmask 0x%x cqshift %lu cqmask 0x%x\n", 607 (unsigned)pci_resource_len(rdev->lldi.pdev, 2), 608 (u64)pci_resource_start(rdev->lldi.pdev, 2), 609 rdev->lldi.db_reg, 610 rdev->lldi.gts_reg, 611 rdev->qpshift, rdev->qpmask, 612 rdev->cqshift, rdev->cqmask); 613 614 if (c4iw_num_stags(rdev) == 0) { 615 err = -EINVAL; 616 goto err1; 617 } 618 619 rdev->stats.pd.total = T4_MAX_NUM_PD; 620 rdev->stats.stag.total = rdev->lldi.vr->stag.size; 621 rdev->stats.pbl.total = rdev->lldi.vr->pbl.size; 622 rdev->stats.rqt.total = rdev->lldi.vr->rq.size; 623 rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size; 624 rdev->stats.qid.total = rdev->lldi.vr->qp.size; 625 626 err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD); 627 if (err) { 628 printk(KERN_ERR MOD "error %d initializing resources\n", err); 629 goto err1; 630 } 631 err = c4iw_pblpool_create(rdev); 632 if (err) { 633 printk(KERN_ERR MOD "error %d initializing pbl pool\n", err); 634 goto err2; 635 } 636 err = c4iw_rqtpool_create(rdev); 637 if (err) { 638 printk(KERN_ERR MOD "error %d initializing rqt pool\n", err); 639 goto err3; 640 } 641 err = c4iw_ocqp_pool_create(rdev); 642 if (err) { 643 printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err); 644 goto err4; 645 } 646 return 0; 647 err4: 648 c4iw_rqtpool_destroy(rdev); 649 err3: 650 c4iw_pblpool_destroy(rdev); 651 err2: 652 c4iw_destroy_resource(&rdev->resource); 653 err1: 654 return err; 655 } 656 657 static void c4iw_rdev_close(struct c4iw_rdev *rdev) 658 { 659 c4iw_pblpool_destroy(rdev); 660 c4iw_rqtpool_destroy(rdev); 661 c4iw_destroy_resource(&rdev->resource); 662 } 663 664 static void c4iw_dealloc(struct uld_ctx *ctx) 665 { 666 c4iw_rdev_close(&ctx->dev->rdev); 667 idr_destroy(&ctx->dev->cqidr); 668 idr_destroy(&ctx->dev->qpidr); 669 idr_destroy(&ctx->dev->mmidr); 670 idr_destroy(&ctx->dev->hwtid_idr); 671 idr_destroy(&ctx->dev->stid_idr); 672 idr_destroy(&ctx->dev->atid_idr); 673 iounmap(ctx->dev->rdev.oc_mw_kva); 674 ib_dealloc_device(&ctx->dev->ibdev); 675 ctx->dev = NULL; 676 } 677 678 static void c4iw_remove(struct uld_ctx *ctx) 679 { 680 PDBG("%s c4iw_dev %p\n", __func__, ctx->dev); 681 c4iw_unregister_device(ctx->dev); 682 c4iw_dealloc(ctx); 683 } 684 685 static int rdma_supported(const struct cxgb4_lld_info *infop) 686 { 687 return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 && 688 infop->vr->rq.size > 0 && infop->vr->qp.size > 0 && 689 infop->vr->cq.size > 0; 690 } 691 692 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop) 693 { 694 struct c4iw_dev *devp; 695 int ret; 696 697 if (!rdma_supported(infop)) { 698 printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n", 699 pci_name(infop->pdev)); 700 return ERR_PTR(-ENOSYS); 701 } 702 if (!ocqp_supported(infop)) 703 pr_info("%s: On-Chip Queues not supported on this device.\n", 704 pci_name(infop->pdev)); 705 706 if (!is_t4(infop->adapter_type)) { 707 if (!allow_db_fc_on_t5) { 708 db_fc_threshold = 100000; 709 pr_info("DB Flow Control Disabled.\n"); 710 } 711 712 if (!allow_db_coalescing_on_t5) { 713 db_coalescing_threshold = -1; 714 pr_info("DB Coalescing Disabled.\n"); 715 } 716 } 717 718 devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp)); 719 if (!devp) { 720 printk(KERN_ERR MOD "Cannot allocate ib device\n"); 721 return ERR_PTR(-ENOMEM); 722 } 723 devp->rdev.lldi = *infop; 724 725 devp->rdev.oc_mw_pa = pci_resource_start(devp->rdev.lldi.pdev, 2) + 726 (pci_resource_len(devp->rdev.lldi.pdev, 2) - 727 roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size)); 728 devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa, 729 devp->rdev.lldi.vr->ocq.size); 730 731 PDBG(KERN_INFO MOD "ocq memory: " 732 "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n", 733 devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size, 734 devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva); 735 736 ret = c4iw_rdev_open(&devp->rdev); 737 if (ret) { 738 printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret); 739 ib_dealloc_device(&devp->ibdev); 740 return ERR_PTR(ret); 741 } 742 743 idr_init(&devp->cqidr); 744 idr_init(&devp->qpidr); 745 idr_init(&devp->mmidr); 746 idr_init(&devp->hwtid_idr); 747 idr_init(&devp->stid_idr); 748 idr_init(&devp->atid_idr); 749 spin_lock_init(&devp->lock); 750 mutex_init(&devp->rdev.stats.lock); 751 mutex_init(&devp->db_mutex); 752 753 if (c4iw_debugfs_root) { 754 devp->debugfs_root = debugfs_create_dir( 755 pci_name(devp->rdev.lldi.pdev), 756 c4iw_debugfs_root); 757 setup_debugfs(devp); 758 } 759 return devp; 760 } 761 762 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop) 763 { 764 struct uld_ctx *ctx; 765 static int vers_printed; 766 int i; 767 768 if (!vers_printed++) 769 pr_info("Chelsio T4/T5 RDMA Driver - version %s\n", 770 DRV_VERSION); 771 772 ctx = kzalloc(sizeof *ctx, GFP_KERNEL); 773 if (!ctx) { 774 ctx = ERR_PTR(-ENOMEM); 775 goto out; 776 } 777 ctx->lldi = *infop; 778 779 PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n", 780 __func__, pci_name(ctx->lldi.pdev), 781 ctx->lldi.nchan, ctx->lldi.nrxq, 782 ctx->lldi.ntxq, ctx->lldi.nports); 783 784 mutex_lock(&dev_mutex); 785 list_add_tail(&ctx->entry, &uld_ctx_list); 786 mutex_unlock(&dev_mutex); 787 788 for (i = 0; i < ctx->lldi.nrxq; i++) 789 PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]); 790 out: 791 return ctx; 792 } 793 794 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl, 795 const __be64 *rsp, 796 u32 pktshift) 797 { 798 struct sk_buff *skb; 799 800 /* 801 * Allocate space for cpl_pass_accept_req which will be synthesized by 802 * driver. Once the driver synthesizes the request the skb will go 803 * through the regular cpl_pass_accept_req processing. 804 * The math here assumes sizeof cpl_pass_accept_req >= sizeof 805 * cpl_rx_pkt. 806 */ 807 skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) + 808 sizeof(struct rss_header) - pktshift, GFP_ATOMIC); 809 if (unlikely(!skb)) 810 return NULL; 811 812 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) + 813 sizeof(struct rss_header) - pktshift); 814 815 /* 816 * This skb will contain: 817 * rss_header from the rspq descriptor (1 flit) 818 * cpl_rx_pkt struct from the rspq descriptor (2 flits) 819 * space for the difference between the size of an 820 * rx_pkt and pass_accept_req cpl (1 flit) 821 * the packet data from the gl 822 */ 823 skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) + 824 sizeof(struct rss_header)); 825 skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) + 826 sizeof(struct cpl_pass_accept_req), 827 gl->va + pktshift, 828 gl->tot_len - pktshift); 829 return skb; 830 } 831 832 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl, 833 const __be64 *rsp) 834 { 835 unsigned int opcode = *(u8 *)rsp; 836 struct sk_buff *skb; 837 838 if (opcode != CPL_RX_PKT) 839 goto out; 840 841 skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift); 842 if (skb == NULL) 843 goto out; 844 845 if (c4iw_handlers[opcode] == NULL) { 846 pr_info("%s no handler opcode 0x%x...\n", __func__, 847 opcode); 848 kfree_skb(skb); 849 goto out; 850 } 851 c4iw_handlers[opcode](dev, skb); 852 return 1; 853 out: 854 return 0; 855 } 856 857 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp, 858 const struct pkt_gl *gl) 859 { 860 struct uld_ctx *ctx = handle; 861 struct c4iw_dev *dev = ctx->dev; 862 struct sk_buff *skb; 863 u8 opcode; 864 865 if (gl == NULL) { 866 /* omit RSS and rsp_ctrl at end of descriptor */ 867 unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8; 868 869 skb = alloc_skb(256, GFP_ATOMIC); 870 if (!skb) 871 goto nomem; 872 __skb_put(skb, len); 873 skb_copy_to_linear_data(skb, &rsp[1], len); 874 } else if (gl == CXGB4_MSG_AN) { 875 const struct rsp_ctrl *rc = (void *)rsp; 876 877 u32 qid = be32_to_cpu(rc->pldbuflen_qid); 878 c4iw_ev_handler(dev, qid); 879 return 0; 880 } else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) { 881 if (recv_rx_pkt(dev, gl, rsp)) 882 return 0; 883 884 pr_info("%s: unexpected FL contents at %p, " \ 885 "RSS %#llx, FL %#llx, len %u\n", 886 pci_name(ctx->lldi.pdev), gl->va, 887 (unsigned long long)be64_to_cpu(*rsp), 888 (unsigned long long)be64_to_cpu( 889 *(__force __be64 *)gl->va), 890 gl->tot_len); 891 892 return 0; 893 } else { 894 skb = cxgb4_pktgl_to_skb(gl, 128, 128); 895 if (unlikely(!skb)) 896 goto nomem; 897 } 898 899 opcode = *(u8 *)rsp; 900 if (c4iw_handlers[opcode]) 901 c4iw_handlers[opcode](dev, skb); 902 else 903 pr_info("%s no handler opcode 0x%x...\n", __func__, 904 opcode); 905 906 return 0; 907 nomem: 908 return -1; 909 } 910 911 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state) 912 { 913 struct uld_ctx *ctx = handle; 914 915 PDBG("%s new_state %u\n", __func__, new_state); 916 switch (new_state) { 917 case CXGB4_STATE_UP: 918 printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev)); 919 if (!ctx->dev) { 920 int ret; 921 922 ctx->dev = c4iw_alloc(&ctx->lldi); 923 if (IS_ERR(ctx->dev)) { 924 printk(KERN_ERR MOD 925 "%s: initialization failed: %ld\n", 926 pci_name(ctx->lldi.pdev), 927 PTR_ERR(ctx->dev)); 928 ctx->dev = NULL; 929 break; 930 } 931 ret = c4iw_register_device(ctx->dev); 932 if (ret) { 933 printk(KERN_ERR MOD 934 "%s: RDMA registration failed: %d\n", 935 pci_name(ctx->lldi.pdev), ret); 936 c4iw_dealloc(ctx); 937 } 938 } 939 break; 940 case CXGB4_STATE_DOWN: 941 printk(KERN_INFO MOD "%s: Down\n", 942 pci_name(ctx->lldi.pdev)); 943 if (ctx->dev) 944 c4iw_remove(ctx); 945 break; 946 case CXGB4_STATE_START_RECOVERY: 947 printk(KERN_INFO MOD "%s: Fatal Error\n", 948 pci_name(ctx->lldi.pdev)); 949 if (ctx->dev) { 950 struct ib_event event; 951 952 ctx->dev->rdev.flags |= T4_FATAL_ERROR; 953 memset(&event, 0, sizeof event); 954 event.event = IB_EVENT_DEVICE_FATAL; 955 event.device = &ctx->dev->ibdev; 956 ib_dispatch_event(&event); 957 c4iw_remove(ctx); 958 } 959 break; 960 case CXGB4_STATE_DETACH: 961 printk(KERN_INFO MOD "%s: Detach\n", 962 pci_name(ctx->lldi.pdev)); 963 if (ctx->dev) 964 c4iw_remove(ctx); 965 break; 966 } 967 return 0; 968 } 969 970 static int disable_qp_db(int id, void *p, void *data) 971 { 972 struct c4iw_qp *qp = p; 973 974 t4_disable_wq_db(&qp->wq); 975 return 0; 976 } 977 978 static void stop_queues(struct uld_ctx *ctx) 979 { 980 spin_lock_irq(&ctx->dev->lock); 981 if (ctx->dev->db_state == NORMAL) { 982 ctx->dev->rdev.stats.db_state_transitions++; 983 ctx->dev->db_state = FLOW_CONTROL; 984 idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL); 985 } 986 spin_unlock_irq(&ctx->dev->lock); 987 } 988 989 static int enable_qp_db(int id, void *p, void *data) 990 { 991 struct c4iw_qp *qp = p; 992 993 t4_enable_wq_db(&qp->wq); 994 return 0; 995 } 996 997 static void resume_queues(struct uld_ctx *ctx) 998 { 999 spin_lock_irq(&ctx->dev->lock); 1000 if (ctx->dev->qpcnt <= db_fc_threshold && 1001 ctx->dev->db_state == FLOW_CONTROL) { 1002 ctx->dev->db_state = NORMAL; 1003 ctx->dev->rdev.stats.db_state_transitions++; 1004 idr_for_each(&ctx->dev->qpidr, enable_qp_db, NULL); 1005 } 1006 spin_unlock_irq(&ctx->dev->lock); 1007 } 1008 1009 struct qp_list { 1010 unsigned idx; 1011 struct c4iw_qp **qps; 1012 }; 1013 1014 static int add_and_ref_qp(int id, void *p, void *data) 1015 { 1016 struct qp_list *qp_listp = data; 1017 struct c4iw_qp *qp = p; 1018 1019 c4iw_qp_add_ref(&qp->ibqp); 1020 qp_listp->qps[qp_listp->idx++] = qp; 1021 return 0; 1022 } 1023 1024 static int count_qps(int id, void *p, void *data) 1025 { 1026 unsigned *countp = data; 1027 (*countp)++; 1028 return 0; 1029 } 1030 1031 static void deref_qps(struct qp_list qp_list) 1032 { 1033 int idx; 1034 1035 for (idx = 0; idx < qp_list.idx; idx++) 1036 c4iw_qp_rem_ref(&qp_list.qps[idx]->ibqp); 1037 } 1038 1039 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list) 1040 { 1041 int idx; 1042 int ret; 1043 1044 for (idx = 0; idx < qp_list->idx; idx++) { 1045 struct c4iw_qp *qp = qp_list->qps[idx]; 1046 1047 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0], 1048 qp->wq.sq.qid, 1049 t4_sq_host_wq_pidx(&qp->wq), 1050 t4_sq_wq_size(&qp->wq)); 1051 if (ret) { 1052 printk(KERN_ERR MOD "%s: Fatal error - " 1053 "DB overflow recovery failed - " 1054 "error syncing SQ qid %u\n", 1055 pci_name(ctx->lldi.pdev), qp->wq.sq.qid); 1056 return; 1057 } 1058 1059 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0], 1060 qp->wq.rq.qid, 1061 t4_rq_host_wq_pidx(&qp->wq), 1062 t4_rq_wq_size(&qp->wq)); 1063 1064 if (ret) { 1065 printk(KERN_ERR MOD "%s: Fatal error - " 1066 "DB overflow recovery failed - " 1067 "error syncing RQ qid %u\n", 1068 pci_name(ctx->lldi.pdev), qp->wq.rq.qid); 1069 return; 1070 } 1071 1072 /* Wait for the dbfifo to drain */ 1073 while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) { 1074 set_current_state(TASK_UNINTERRUPTIBLE); 1075 schedule_timeout(usecs_to_jiffies(10)); 1076 } 1077 } 1078 } 1079 1080 static void recover_queues(struct uld_ctx *ctx) 1081 { 1082 int count = 0; 1083 struct qp_list qp_list; 1084 int ret; 1085 1086 /* lock out kernel db ringers */ 1087 mutex_lock(&ctx->dev->db_mutex); 1088 1089 /* put all queues in to recovery mode */ 1090 spin_lock_irq(&ctx->dev->lock); 1091 ctx->dev->db_state = RECOVERY; 1092 ctx->dev->rdev.stats.db_state_transitions++; 1093 idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL); 1094 spin_unlock_irq(&ctx->dev->lock); 1095 1096 /* slow everybody down */ 1097 set_current_state(TASK_UNINTERRUPTIBLE); 1098 schedule_timeout(usecs_to_jiffies(1000)); 1099 1100 /* Wait for the dbfifo to completely drain. */ 1101 while (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1) > 0) { 1102 set_current_state(TASK_UNINTERRUPTIBLE); 1103 schedule_timeout(usecs_to_jiffies(10)); 1104 } 1105 1106 /* flush the SGE contexts */ 1107 ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]); 1108 if (ret) { 1109 printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n", 1110 pci_name(ctx->lldi.pdev)); 1111 goto out; 1112 } 1113 1114 /* Count active queues so we can build a list of queues to recover */ 1115 spin_lock_irq(&ctx->dev->lock); 1116 idr_for_each(&ctx->dev->qpidr, count_qps, &count); 1117 1118 qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC); 1119 if (!qp_list.qps) { 1120 printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n", 1121 pci_name(ctx->lldi.pdev)); 1122 spin_unlock_irq(&ctx->dev->lock); 1123 goto out; 1124 } 1125 qp_list.idx = 0; 1126 1127 /* add and ref each qp so it doesn't get freed */ 1128 idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list); 1129 1130 spin_unlock_irq(&ctx->dev->lock); 1131 1132 /* now traverse the list in a safe context to recover the db state*/ 1133 recover_lost_dbs(ctx, &qp_list); 1134 1135 /* we're almost done! deref the qps and clean up */ 1136 deref_qps(qp_list); 1137 kfree(qp_list.qps); 1138 1139 /* Wait for the dbfifo to completely drain again */ 1140 while (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1) > 0) { 1141 set_current_state(TASK_UNINTERRUPTIBLE); 1142 schedule_timeout(usecs_to_jiffies(10)); 1143 } 1144 1145 /* resume the queues */ 1146 spin_lock_irq(&ctx->dev->lock); 1147 if (ctx->dev->qpcnt > db_fc_threshold) 1148 ctx->dev->db_state = FLOW_CONTROL; 1149 else { 1150 ctx->dev->db_state = NORMAL; 1151 idr_for_each(&ctx->dev->qpidr, enable_qp_db, NULL); 1152 } 1153 ctx->dev->rdev.stats.db_state_transitions++; 1154 spin_unlock_irq(&ctx->dev->lock); 1155 1156 out: 1157 /* start up kernel db ringers again */ 1158 mutex_unlock(&ctx->dev->db_mutex); 1159 } 1160 1161 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...) 1162 { 1163 struct uld_ctx *ctx = handle; 1164 1165 switch (control) { 1166 case CXGB4_CONTROL_DB_FULL: 1167 stop_queues(ctx); 1168 mutex_lock(&ctx->dev->rdev.stats.lock); 1169 ctx->dev->rdev.stats.db_full++; 1170 mutex_unlock(&ctx->dev->rdev.stats.lock); 1171 break; 1172 case CXGB4_CONTROL_DB_EMPTY: 1173 resume_queues(ctx); 1174 mutex_lock(&ctx->dev->rdev.stats.lock); 1175 ctx->dev->rdev.stats.db_empty++; 1176 mutex_unlock(&ctx->dev->rdev.stats.lock); 1177 break; 1178 case CXGB4_CONTROL_DB_DROP: 1179 recover_queues(ctx); 1180 mutex_lock(&ctx->dev->rdev.stats.lock); 1181 ctx->dev->rdev.stats.db_drop++; 1182 mutex_unlock(&ctx->dev->rdev.stats.lock); 1183 break; 1184 default: 1185 printk(KERN_WARNING MOD "%s: unknown control cmd %u\n", 1186 pci_name(ctx->lldi.pdev), control); 1187 break; 1188 } 1189 return 0; 1190 } 1191 1192 static struct cxgb4_uld_info c4iw_uld_info = { 1193 .name = DRV_NAME, 1194 .add = c4iw_uld_add, 1195 .rx_handler = c4iw_uld_rx_handler, 1196 .state_change = c4iw_uld_state_change, 1197 .control = c4iw_uld_control, 1198 }; 1199 1200 static int __init c4iw_init_module(void) 1201 { 1202 int err; 1203 1204 err = c4iw_cm_init(); 1205 if (err) 1206 return err; 1207 1208 c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL); 1209 if (!c4iw_debugfs_root) 1210 printk(KERN_WARNING MOD 1211 "could not create debugfs entry, continuing\n"); 1212 1213 cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info); 1214 1215 return 0; 1216 } 1217 1218 static void __exit c4iw_exit_module(void) 1219 { 1220 struct uld_ctx *ctx, *tmp; 1221 1222 mutex_lock(&dev_mutex); 1223 list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) { 1224 if (ctx->dev) 1225 c4iw_remove(ctx); 1226 kfree(ctx); 1227 } 1228 mutex_unlock(&dev_mutex); 1229 cxgb4_unregister_uld(CXGB4_ULD_RDMA); 1230 c4iw_cm_term(); 1231 debugfs_remove_recursive(c4iw_debugfs_root); 1232 } 1233 1234 module_init(c4iw_init_module); 1235 module_exit(c4iw_exit_module); 1236