drivers/infiniband/hw/cxgb4/device.c

   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 #include <linux/math64.h>
  37
  38 #include <rdma/ib_verbs.h>
  39
  40 #include "iw_cxgb4.h"
  41
  42 #define DRV_VERSION "0.1"
  43
  44 MODULE_AUTHOR("Steve Wise");
  45 MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver");
  46 MODULE_LICENSE("Dual BSD/GPL");
  47 MODULE_VERSION(DRV_VERSION);
  48
  49 static int allow_db_fc_on_t5;
  50 module_param(allow_db_fc_on_t5, int, 0644);
  51 MODULE_PARM_DESC(allow_db_fc_on_t5,
  52                  "Allow DB Flow Control on T5 (default = 0)");
  53
  54 static int allow_db_coalescing_on_t5;
  55 module_param(allow_db_coalescing_on_t5, int, 0644);
  56 MODULE_PARM_DESC(allow_db_coalescing_on_t5,
  57                  "Allow DB Coalescing on T5 (default = 0)");
  58
  59 int c4iw_wr_log = 0;
  60 module_param(c4iw_wr_log, int, 0444);
  61 MODULE_PARM_DESC(c4iw_wr_log, "Enables logging of work request timing data.");
  62
  63 static int c4iw_wr_log_size_order = 12;
  64 module_param(c4iw_wr_log_size_order, int, 0444);
  65 MODULE_PARM_DESC(c4iw_wr_log_size_order,
  66                  "Number of entries (log2) in the work request timing log.");
  67
  68 struct uld_ctx {
  69         struct list_head entry;
  70         struct cxgb4_lld_info lldi;
  71         struct c4iw_dev *dev;
  72 };
  73
  74 static LIST_HEAD(uld_ctx_list);
  75 static DEFINE_MUTEX(dev_mutex);
  76
  77 #define DB_FC_RESUME_SIZE 64
  78 #define DB_FC_RESUME_DELAY 1
  79 #define DB_FC_DRAIN_THRESH 0
  80
  81 static struct dentry *c4iw_debugfs_root;
  82
  83 struct c4iw_debugfs_data {
  84         struct c4iw_dev *devp;
  85         char *buf;
  86         int bufsize;
  87         int pos;
  88 };
  89
  90 static int count_idrs(int id, void *p, void *data)
  91 {
  92         int *countp = data;
  93
  94         *countp = *countp + 1;
  95         return 0;
  96 }
  97
  98 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
  99                             loff_t *ppos)
 100 {
 101         struct c4iw_debugfs_data *d = file->private_data;
 102
 103         return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
 104 }
 105
 106 void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
 107 {
 108         struct wr_log_entry le;
 109         int idx;
 110
 111         if (!wq->rdev->wr_log)
 112                 return;
 113
 114         idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
 115                 (wq->rdev->wr_log_size - 1);
 116         le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
 117         getnstimeofday(&le.poll_host_ts);
 118         le.valid = 1;
 119         le.cqe_sge_ts = CQE_TS(cqe);
 120         if (SQ_TYPE(cqe)) {
 121                 le.qid = wq->sq.qid;
 122                 le.opcode = CQE_OPCODE(cqe);
 123                 le.post_host_ts = wq->sq.sw_sq[wq->sq.cidx].host_ts;
 124                 le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
 125                 le.wr_id = CQE_WRID_SQ_IDX(cqe);
 126         } else {
 127                 le.qid = wq->rq.qid;
 128                 le.opcode = FW_RI_RECEIVE;
 129                 le.post_host_ts = wq->rq.sw_rq[wq->rq.cidx].host_ts;
 130                 le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
 131                 le.wr_id = CQE_WRID_MSN(cqe);
 132         }
 133         wq->rdev->wr_log[idx] = le;
 134 }
 135
 136 static int wr_log_show(struct seq_file *seq, void *v)
 137 {
 138         struct c4iw_dev *dev = seq->private;
 139         struct timespec prev_ts = {0, 0};
 140         struct wr_log_entry *lep;
 141         int prev_ts_set = 0;
 142         int idx, end;
 143
 144 #define ts2ns(ts) div64_u64((ts) * dev->rdev.lldi.cclk_ps, 1000)
 145
 146         idx = atomic_read(&dev->rdev.wr_log_idx) &
 147                 (dev->rdev.wr_log_size - 1);
 148         end = idx - 1;
 149         if (end < 0)
 150                 end = dev->rdev.wr_log_size - 1;
 151         lep = &dev->rdev.wr_log[idx];
 152         while (idx != end) {
 153                 if (lep->valid) {
 154                         if (!prev_ts_set) {
 155                                 prev_ts_set = 1;
 156                                 prev_ts = lep->poll_host_ts;
 157                         }
 158                         seq_printf(seq, "%04u: sec %lu nsec %lu qid %u opcode "
 159                                    "%u %s 0x%x host_wr_delta sec %lu nsec %lu "
 160                                    "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
 161                                    "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
 162                                    "cqe_poll_delta_ns %llu\n",
 163                                    idx,
 164                                    timespec_sub(lep->poll_host_ts,
 165                                                 prev_ts).tv_sec,
 166                                    timespec_sub(lep->poll_host_ts,
 167                                                 prev_ts).tv_nsec,
 168                                    lep->qid, lep->opcode,
 169                                    lep->opcode == FW_RI_RECEIVE ?
 170                                                         "msn" : "wrid",
 171                                    lep->wr_id,
 172                                    timespec_sub(lep->poll_host_ts,
 173                                                 lep->post_host_ts).tv_sec,
 174                                    timespec_sub(lep->poll_host_ts,
 175                                                 lep->post_host_ts).tv_nsec,
 176                                    lep->post_sge_ts, lep->cqe_sge_ts,
 177                                    lep->poll_sge_ts,
 178                                    ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
 179                                    ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
 180                         prev_ts = lep->poll_host_ts;
 181                 }
 182                 idx++;
 183                 if (idx > (dev->rdev.wr_log_size - 1))
 184                         idx = 0;
 185                 lep = &dev->rdev.wr_log[idx];
 186         }
 187 #undef ts2ns
 188         return 0;
 189 }
 190
 191 static int wr_log_open(struct inode *inode, struct file *file)
 192 {
 193         return single_open(file, wr_log_show, inode->i_private);
 194 }
 195
 196 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
 197                             size_t count, loff_t *pos)
 198 {
 199         struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
 200         int i;
 201
 202         if (dev->rdev.wr_log)
 203                 for (i = 0; i < dev->rdev.wr_log_size; i++)
 204                         dev->rdev.wr_log[i].valid = 0;
 205         return count;
 206 }
 207
 208 static const struct file_operations wr_log_debugfs_fops = {
 209         .owner   = THIS_MODULE,
 210         .open    = wr_log_open,
 211         .release = single_release,
 212         .read    = seq_read,
 213         .llseek  = seq_lseek,
 214         .write   = wr_log_clear,
 215 };
 216
 217 static int dump_qp(int id, void *p, void *data)
 218 {
 219         struct c4iw_qp *qp = p;
 220         struct c4iw_debugfs_data *qpd = data;
 221         int space;
 222         int cc;
 223
 224         if (id != qp->wq.sq.qid)
 225                 return 0;
 226
 227         space = qpd->bufsize - qpd->pos - 1;
 228         if (space == 0)
 229                 return 1;
 230
 231         if (qp->ep) {
 232                 if (qp->ep->com.local_addr.ss_family == AF_INET) {
 233                         struct sockaddr_in *lsin = (struct sockaddr_in *)
 234                                 &qp->ep->com.cm_id->local_addr;
 235                         struct sockaddr_in *rsin = (struct sockaddr_in *)
 236                                 &qp->ep->com.cm_id->remote_addr;
 237                         struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
 238                                 &qp->ep->com.cm_id->m_local_addr;
 239                         struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
 240                                 &qp->ep->com.cm_id->m_remote_addr;
 241
 242                         cc = snprintf(qpd->buf + qpd->pos, space,
 243                                       "rc qp sq id %u rq id %u state %u "
 244                                       "onchip %u ep tid %u state %u "
 245                                       "%pI4:%u/%u->%pI4:%u/%u\n",
 246                                       qp->wq.sq.qid, qp->wq.rq.qid,
 247                                       (int)qp->attr.state,
 248                                       qp->wq.sq.flags & T4_SQ_ONCHIP,
 249                                       qp->ep->hwtid, (int)qp->ep->com.state,
 250                                       &lsin->sin_addr, ntohs(lsin->sin_port),
 251                                       ntohs(mapped_lsin->sin_port),
 252                                       &rsin->sin_addr, ntohs(rsin->sin_port),
 253                                       ntohs(mapped_rsin->sin_port));
 254                 } else {
 255                         struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
 256                                 &qp->ep->com.cm_id->local_addr;
 257                         struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
 258                                 &qp->ep->com.cm_id->remote_addr;
 259                         struct sockaddr_in6 *mapped_lsin6 =
 260                                 (struct sockaddr_in6 *)
 261                                 &qp->ep->com.cm_id->m_local_addr;
 262                         struct sockaddr_in6 *mapped_rsin6 =
 263                                 (struct sockaddr_in6 *)
 264                                 &qp->ep->com.cm_id->m_remote_addr;
 265
 266                         cc = snprintf(qpd->buf + qpd->pos, space,
 267                                       "rc qp sq id %u rq id %u state %u "
 268                                       "onchip %u ep tid %u state %u "
 269                                       "%pI6:%u/%u->%pI6:%u/%u\n",
 270                                       qp->wq.sq.qid, qp->wq.rq.qid,
 271                                       (int)qp->attr.state,
 272                                       qp->wq.sq.flags & T4_SQ_ONCHIP,
 273                                       qp->ep->hwtid, (int)qp->ep->com.state,
 274                                       &lsin6->sin6_addr,
 275                                       ntohs(lsin6->sin6_port),
 276                                       ntohs(mapped_lsin6->sin6_port),
 277                                       &rsin6->sin6_addr,
 278                                       ntohs(rsin6->sin6_port),
 279                                       ntohs(mapped_rsin6->sin6_port));
 280                 }
 281         } else
 282                 cc = snprintf(qpd->buf + qpd->pos, space,
 283                              "qp sq id %u rq id %u state %u onchip %u\n",
 284                               qp->wq.sq.qid, qp->wq.rq.qid,
 285                               (int)qp->attr.state,
 286                               qp->wq.sq.flags & T4_SQ_ONCHIP);
 287         if (cc < space)
 288                 qpd->pos += cc;
 289         return 0;
 290 }
 291
 292 static int qp_release(struct inode *inode, struct file *file)
 293 {
 294         struct c4iw_debugfs_data *qpd = file->private_data;
 295         if (!qpd) {
 296                 printk(KERN_INFO "%s null qpd?\n", __func__);
 297                 return 0;
 298         }
 299         vfree(qpd->buf);
 300         kfree(qpd);
 301         return 0;
 302 }
 303
 304 static int qp_open(struct inode *inode, struct file *file)
 305 {
 306         struct c4iw_debugfs_data *qpd;
 307         int count = 1;
 308
 309         qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
 310         if (!qpd)
 311                 return -ENOMEM;
 312
 313         qpd->devp = inode->i_private;
 314         qpd->pos = 0;
 315
 316         spin_lock_irq(&qpd->devp->lock);
 317         idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
 318         spin_unlock_irq(&qpd->devp->lock);
 319
 320         qpd->bufsize = count * 180;
 321         qpd->buf = vmalloc(qpd->bufsize);
 322         if (!qpd->buf) {
 323                 kfree(qpd);
 324                 return -ENOMEM;
 325         }
 326
 327         spin_lock_irq(&qpd->devp->lock);
 328         idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
 329         spin_unlock_irq(&qpd->devp->lock);
 330
 331         qpd->buf[qpd->pos++] = 0;
 332         file->private_data = qpd;
 333         return 0;
 334 }
 335
 336 static const struct file_operations qp_debugfs_fops = {
 337         .owner   = THIS_MODULE,
 338         .open    = qp_open,
 339         .release = qp_release,
 340         .read    = debugfs_read,
 341         .llseek  = default_llseek,
 342 };
 343
 344 static int dump_stag(int id, void *p, void *data)
 345 {
 346         struct c4iw_debugfs_data *stagd = data;
 347         int space;
 348         int cc;
 349         struct fw_ri_tpte tpte;
 350         int ret;
 351
 352         space = stagd->bufsize - stagd->pos - 1;
 353         if (space == 0)
 354                 return 1;
 355
 356         ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
 357                               (__be32 *)&tpte);
 358         if (ret) {
 359                 dev_err(&stagd->devp->rdev.lldi.pdev->dev,
 360                         "%s cxgb4_read_tpte err %d\n", __func__, ret);
 361                 return ret;
 362         }
 363         cc = snprintf(stagd->buf + stagd->pos, space,
 364                       "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
 365                       "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
 366                       (u32)id<<8,
 367                       FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
 368                       FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
 369                       FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
 370                       FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
 371                       FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
 372                       FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
 373                       ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
 374                       ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
 375         if (cc < space)
 376                 stagd->pos += cc;
 377         return 0;
 378 }
 379
 380 static int stag_release(struct inode *inode, struct file *file)
 381 {
 382         struct c4iw_debugfs_data *stagd = file->private_data;
 383         if (!stagd) {
 384                 printk(KERN_INFO "%s null stagd?\n", __func__);
 385                 return 0;
 386         }
 387         vfree(stagd->buf);
 388         kfree(stagd);
 389         return 0;
 390 }
 391
 392 static int stag_open(struct inode *inode, struct file *file)
 393 {
 394         struct c4iw_debugfs_data *stagd;
 395         int ret = 0;
 396         int count = 1;
 397
 398         stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
 399         if (!stagd) {
 400                 ret = -ENOMEM;
 401                 goto out;
 402         }
 403         stagd->devp = inode->i_private;
 404         stagd->pos = 0;
 405
 406         spin_lock_irq(&stagd->devp->lock);
 407         idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
 408         spin_unlock_irq(&stagd->devp->lock);
 409
 410         stagd->bufsize = count * 256;
 411         stagd->buf = vmalloc(stagd->bufsize);
 412         if (!stagd->buf) {
 413                 ret = -ENOMEM;
 414                 goto err1;
 415         }
 416
 417         spin_lock_irq(&stagd->devp->lock);
 418         idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
 419         spin_unlock_irq(&stagd->devp->lock);
 420
 421         stagd->buf[stagd->pos++] = 0;
 422         file->private_data = stagd;
 423         goto out;
 424 err1:
 425         kfree(stagd);
 426 out:
 427         return ret;
 428 }
 429
 430 static const struct file_operations stag_debugfs_fops = {
 431         .owner   = THIS_MODULE,
 432         .open    = stag_open,
 433         .release = stag_release,
 434         .read    = debugfs_read,
 435         .llseek  = default_llseek,
 436 };
 437
 438 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
 439
 440 static int stats_show(struct seq_file *seq, void *v)
 441 {
 442         struct c4iw_dev *dev = seq->private;
 443
 444         seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
 445                    "Max", "Fail");
 446         seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
 447                         dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
 448                         dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
 449         seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
 450                         dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
 451                         dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
 452         seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
 453                         dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
 454                         dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
 455         seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
 456                         dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
 457                         dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
 458         seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
 459                         dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
 460                         dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
 461         seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
 462                         dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
 463                         dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
 464         seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
 465         seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
 466         seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
 467         seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
 468                    db_state_str[dev->db_state],
 469                    dev->rdev.stats.db_state_transitions,
 470                    dev->rdev.stats.db_fc_interruptions);
 471         seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
 472         seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
 473                    dev->rdev.stats.act_ofld_conn_fails);
 474         seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
 475                    dev->rdev.stats.pas_ofld_conn_fails);
 476         seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
 477         seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
 478         return 0;
 479 }
 480
 481 static int stats_open(struct inode *inode, struct file *file)
 482 {
 483         return single_open(file, stats_show, inode->i_private);
 484 }
 485
 486 static ssize_t stats_clear(struct file *file, const char __user *buf,
 487                 size_t count, loff_t *pos)
 488 {
 489         struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
 490
 491         mutex_lock(&dev->rdev.stats.lock);
 492         dev->rdev.stats.pd.max = 0;
 493         dev->rdev.stats.pd.fail = 0;
 494         dev->rdev.stats.qid.max = 0;
 495         dev->rdev.stats.qid.fail = 0;
 496         dev->rdev.stats.stag.max = 0;
 497         dev->rdev.stats.stag.fail = 0;
 498         dev->rdev.stats.pbl.max = 0;
 499         dev->rdev.stats.pbl.fail = 0;
 500         dev->rdev.stats.rqt.max = 0;
 501         dev->rdev.stats.rqt.fail = 0;
 502         dev->rdev.stats.ocqp.max = 0;
 503         dev->rdev.stats.ocqp.fail = 0;
 504         dev->rdev.stats.db_full = 0;
 505         dev->rdev.stats.db_empty = 0;
 506         dev->rdev.stats.db_drop = 0;
 507         dev->rdev.stats.db_state_transitions = 0;
 508         dev->rdev.stats.tcam_full = 0;
 509         dev->rdev.stats.act_ofld_conn_fails = 0;
 510         dev->rdev.stats.pas_ofld_conn_fails = 0;
 511         mutex_unlock(&dev->rdev.stats.lock);
 512         return count;
 513 }
 514
 515 static const struct file_operations stats_debugfs_fops = {
 516         .owner   = THIS_MODULE,
 517         .open    = stats_open,
 518         .release = single_release,
 519         .read    = seq_read,
 520         .llseek  = seq_lseek,
 521         .write   = stats_clear,
 522 };
 523
 524 static int dump_ep(int id, void *p, void *data)
 525 {
 526         struct c4iw_ep *ep = p;
 527         struct c4iw_debugfs_data *epd = data;
 528         int space;
 529         int cc;
 530
 531         space = epd->bufsize - epd->pos - 1;
 532         if (space == 0)
 533                 return 1;
 534
 535         if (ep->com.local_addr.ss_family == AF_INET) {
 536                 struct sockaddr_in *lsin = (struct sockaddr_in *)
 537                         &ep->com.cm_id->local_addr;
 538                 struct sockaddr_in *rsin = (struct sockaddr_in *)
 539                         &ep->com.cm_id->remote_addr;
 540                 struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
 541                         &ep->com.cm_id->m_local_addr;
 542                 struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
 543                         &ep->com.cm_id->m_remote_addr;
 544
 545                 cc = snprintf(epd->buf + epd->pos, space,
 546                               "ep %p cm_id %p qp %p state %d flags 0x%lx "
 547                               "history 0x%lx hwtid %d atid %d "
 548                               "conn_na %u abort_na %u "
 549                               "%pI4:%d/%d <-> %pI4:%d/%d\n",
 550                               ep, ep->com.cm_id, ep->com.qp,
 551                               (int)ep->com.state, ep->com.flags,
 552                               ep->com.history, ep->hwtid, ep->atid,
 553                               ep->stats.connect_neg_adv,
 554                               ep->stats.abort_neg_adv,
 555                               &lsin->sin_addr, ntohs(lsin->sin_port),
 556                               ntohs(mapped_lsin->sin_port),
 557                               &rsin->sin_addr, ntohs(rsin->sin_port),
 558                               ntohs(mapped_rsin->sin_port));
 559         } else {
 560                 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
 561                         &ep->com.cm_id->local_addr;
 562                 struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
 563                         &ep->com.cm_id->remote_addr;
 564                 struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
 565                         &ep->com.cm_id->m_local_addr;
 566                 struct sockaddr_in6 *mapped_rsin6 = (struct sockaddr_in6 *)
 567                         &ep->com.cm_id->m_remote_addr;
 568
 569                 cc = snprintf(epd->buf + epd->pos, space,
 570                               "ep %p cm_id %p qp %p state %d flags 0x%lx "
 571                               "history 0x%lx hwtid %d atid %d "
 572                               "conn_na %u abort_na %u "
 573                               "%pI6:%d/%d <-> %pI6:%d/%d\n",
 574                               ep, ep->com.cm_id, ep->com.qp,
 575                               (int)ep->com.state, ep->com.flags,
 576                               ep->com.history, ep->hwtid, ep->atid,
 577                               ep->stats.connect_neg_adv,
 578                               ep->stats.abort_neg_adv,
 579                               &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
 580                               ntohs(mapped_lsin6->sin6_port),
 581                               &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
 582                               ntohs(mapped_rsin6->sin6_port));
 583         }
 584         if (cc < space)
 585                 epd->pos += cc;
 586         return 0;
 587 }
 588
 589 static int dump_listen_ep(int id, void *p, void *data)
 590 {
 591         struct c4iw_listen_ep *ep = p;
 592         struct c4iw_debugfs_data *epd = data;
 593         int space;
 594         int cc;
 595
 596         space = epd->bufsize - epd->pos - 1;
 597         if (space == 0)
 598                 return 1;
 599
 600         if (ep->com.local_addr.ss_family == AF_INET) {
 601                 struct sockaddr_in *lsin = (struct sockaddr_in *)
 602                         &ep->com.cm_id->local_addr;
 603                 struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
 604                         &ep->com.cm_id->m_local_addr;
 605
 606                 cc = snprintf(epd->buf + epd->pos, space,
 607                               "ep %p cm_id %p state %d flags 0x%lx stid %d "
 608                               "backlog %d %pI4:%d/%d\n",
 609                               ep, ep->com.cm_id, (int)ep->com.state,
 610                               ep->com.flags, ep->stid, ep->backlog,
 611                               &lsin->sin_addr, ntohs(lsin->sin_port),
 612                               ntohs(mapped_lsin->sin_port));
 613         } else {
 614                 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
 615                         &ep->com.cm_id->local_addr;
 616                 struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
 617                         &ep->com.cm_id->m_local_addr;
 618
 619                 cc = snprintf(epd->buf + epd->pos, space,
 620                               "ep %p cm_id %p state %d flags 0x%lx stid %d "
 621                               "backlog %d %pI6:%d/%d\n",
 622                               ep, ep->com.cm_id, (int)ep->com.state,
 623                               ep->com.flags, ep->stid, ep->backlog,
 624                               &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
 625                               ntohs(mapped_lsin6->sin6_port));
 626         }
 627         if (cc < space)
 628                 epd->pos += cc;
 629         return 0;
 630 }
 631
 632 static int ep_release(struct inode *inode, struct file *file)
 633 {
 634         struct c4iw_debugfs_data *epd = file->private_data;
 635         if (!epd) {
 636                 pr_info("%s null qpd?\n", __func__);
 637                 return 0;
 638         }
 639         vfree(epd->buf);
 640         kfree(epd);
 641         return 0;
 642 }
 643
 644 static int ep_open(struct inode *inode, struct file *file)
 645 {
 646         struct c4iw_debugfs_data *epd;
 647         int ret = 0;
 648         int count = 1;
 649
 650         epd = kmalloc(sizeof(*epd), GFP_KERNEL);
 651         if (!epd) {
 652                 ret = -ENOMEM;
 653                 goto out;
 654         }
 655         epd->devp = inode->i_private;
 656         epd->pos = 0;
 657
 658         spin_lock_irq(&epd->devp->lock);
 659         idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
 660         idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
 661         idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
 662         spin_unlock_irq(&epd->devp->lock);
 663
 664         epd->bufsize = count * 240;
 665         epd->buf = vmalloc(epd->bufsize);
 666         if (!epd->buf) {
 667                 ret = -ENOMEM;
 668                 goto err1;
 669         }
 670
 671         spin_lock_irq(&epd->devp->lock);
 672         idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
 673         idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
 674         idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
 675         spin_unlock_irq(&epd->devp->lock);
 676
 677         file->private_data = epd;
 678         goto out;
 679 err1:
 680         kfree(epd);
 681 out:
 682         return ret;
 683 }
 684
 685 static const struct file_operations ep_debugfs_fops = {
 686         .owner   = THIS_MODULE,
 687         .open    = ep_open,
 688         .release = ep_release,
 689         .read    = debugfs_read,
 690 };
 691
 692 static int setup_debugfs(struct c4iw_dev *devp)
 693 {
 694         if (!devp->debugfs_root)
 695                 return -1;
 696
 697         debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
 698                                  (void *)devp, &qp_debugfs_fops, 4096);
 699
 700         debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
 701                                  (void *)devp, &stag_debugfs_fops, 4096);
 702
 703         debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
 704                                  (void *)devp, &stats_debugfs_fops, 4096);
 705
 706         debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
 707                                  (void *)devp, &ep_debugfs_fops, 4096);
 708
 709         if (c4iw_wr_log)
 710                 debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
 711                                          (void *)devp, &wr_log_debugfs_fops, 4096);
 712         return 0;
 713 }
 714
 715 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
 716                                struct c4iw_dev_ucontext *uctx)
 717 {
 718         struct list_head *pos, *nxt;
 719         struct c4iw_qid_list *entry;
 720
 721         mutex_lock(&uctx->lock);
 722         list_for_each_safe(pos, nxt, &uctx->qpids) {
 723                 entry = list_entry(pos, struct c4iw_qid_list, entry);
 724                 list_del_init(&entry->entry);
 725                 if (!(entry->qid & rdev->qpmask)) {
 726                         c4iw_put_resource(&rdev->resource.qid_table,
 727                                           entry->qid);
 728                         mutex_lock(&rdev->stats.lock);
 729                         rdev->stats.qid.cur -= rdev->qpmask + 1;
 730                         mutex_unlock(&rdev->stats.lock);
 731                 }
 732                 kfree(entry);
 733         }
 734
 735         list_for_each_safe(pos, nxt, &uctx->qpids) {
 736                 entry = list_entry(pos, struct c4iw_qid_list, entry);
 737                 list_del_init(&entry->entry);
 738                 kfree(entry);
 739         }
 740         mutex_unlock(&uctx->lock);
 741 }
 742
 743 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
 744                             struct c4iw_dev_ucontext *uctx)
 745 {
 746         INIT_LIST_HEAD(&uctx->qpids);
 747         INIT_LIST_HEAD(&uctx->cqids);
 748         mutex_init(&uctx->lock);
 749 }
 750
 751 /* Caller takes care of locking if needed */
 752 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
 753 {
 754         int err;
 755
 756         c4iw_init_dev_ucontext(rdev, &rdev->uctx);
 757
 758         /*
 759          * This implementation assumes udb_density == ucq_density!  Eventually
 760          * we might need to support this but for now fail the open. Also the
 761          * cqid and qpid range must match for now.
 762          */
 763         if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
 764                 pr_err(MOD "%s: unsupported udb/ucq densities %u/%u\n",
 765                        pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
 766                        rdev->lldi.ucq_density);
 767                 return -EINVAL;
 768         }
 769         if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
 770             rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
 771                 pr_err(MOD "%s: unsupported qp and cq id ranges "
 772                        "qp start %u size %u cq start %u size %u\n",
 773                        pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
 774                        rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
 775                        rdev->lldi.vr->cq.size);
 776                 return -EINVAL;
 777         }
 778
 779         rdev->qpmask = rdev->lldi.udb_density - 1;
 780         rdev->cqmask = rdev->lldi.ucq_density - 1;
 781         PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
 782              "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
 783              "qp qid start %u size %u cq qid start %u size %u\n",
 784              __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
 785              rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
 786              rdev->lldi.vr->pbl.start,
 787              rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
 788              rdev->lldi.vr->rq.size,
 789              rdev->lldi.vr->qp.start,
 790              rdev->lldi.vr->qp.size,
 791              rdev->lldi.vr->cq.start,
 792              rdev->lldi.vr->cq.size);
 793         PDBG("udb %pR db_reg %p gts_reg %p "
 794              "qpmask 0x%x cqmask 0x%x\n",
 795                 &rdev->lldi.pdev->resource[2],
 796              rdev->lldi.db_reg, rdev->lldi.gts_reg,
 797              rdev->qpmask, rdev->cqmask);
 798
 799         if (c4iw_num_stags(rdev) == 0)
 800                 return -EINVAL;
 801
 802         rdev->stats.pd.total = T4_MAX_NUM_PD;
 803         rdev->stats.stag.total = rdev->lldi.vr->stag.size;
 804         rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
 805         rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
 806         rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
 807         rdev->stats.qid.total = rdev->lldi.vr->qp.size;
 808
 809         err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
 810         if (err) {
 811                 printk(KERN_ERR MOD "error %d initializing resources\n", err);
 812                 return err;
 813         }
 814         err = c4iw_pblpool_create(rdev);
 815         if (err) {
 816                 printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
 817                 goto destroy_resource;
 818         }
 819         err = c4iw_rqtpool_create(rdev);
 820         if (err) {
 821                 printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
 822                 goto destroy_pblpool;
 823         }
 824         err = c4iw_ocqp_pool_create(rdev);
 825         if (err) {
 826                 printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
 827                 goto destroy_rqtpool;
 828         }
 829         rdev->status_page = (struct t4_dev_status_page *)
 830                             __get_free_page(GFP_KERNEL);
 831         if (!rdev->status_page)
 832                 goto destroy_ocqp_pool;
 833         rdev->status_page->qp_start = rdev->lldi.vr->qp.start;
 834         rdev->status_page->qp_size = rdev->lldi.vr->qp.size;
 835         rdev->status_page->cq_start = rdev->lldi.vr->cq.start;
 836         rdev->status_page->cq_size = rdev->lldi.vr->cq.size;
 837
 838         if (c4iw_wr_log) {
 839                 rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
 840                                        sizeof(*rdev->wr_log), GFP_KERNEL);
 841                 if (rdev->wr_log) {
 842                         rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
 843                         atomic_set(&rdev->wr_log_idx, 0);
 844                 } else {
 845                         pr_err(MOD "error allocating wr_log. Logging disabled\n");
 846                 }
 847         }
 848
 849         rdev->status_page->db_off = 0;
 850
 851         return 0;
 852 destroy_ocqp_pool:
 853         c4iw_ocqp_pool_destroy(rdev);
 854 destroy_rqtpool:
 855         c4iw_rqtpool_destroy(rdev);
 856 destroy_pblpool:
 857         c4iw_pblpool_destroy(rdev);
 858 destroy_resource:
 859         c4iw_destroy_resource(&rdev->resource);
 860         return err;
 861 }
 862
 863 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
 864 {
 865         kfree(rdev->wr_log);
 866         free_page((unsigned long)rdev->status_page);
 867         c4iw_pblpool_destroy(rdev);
 868         c4iw_rqtpool_destroy(rdev);
 869         c4iw_destroy_resource(&rdev->resource);
 870 }
 871
 872 static void c4iw_dealloc(struct uld_ctx *ctx)
 873 {
 874         c4iw_rdev_close(&ctx->dev->rdev);
 875         idr_destroy(&ctx->dev->cqidr);
 876         idr_destroy(&ctx->dev->qpidr);
 877         idr_destroy(&ctx->dev->mmidr);
 878         idr_destroy(&ctx->dev->hwtid_idr);
 879         idr_destroy(&ctx->dev->stid_idr);
 880         idr_destroy(&ctx->dev->atid_idr);
 881         if (ctx->dev->rdev.bar2_kva)
 882                 iounmap(ctx->dev->rdev.bar2_kva);
 883         if (ctx->dev->rdev.oc_mw_kva)
 884                 iounmap(ctx->dev->rdev.oc_mw_kva);
 885         ib_dealloc_device(&ctx->dev->ibdev);
 886         ctx->dev = NULL;
 887 }
 888
 889 static void c4iw_remove(struct uld_ctx *ctx)
 890 {
 891         PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
 892         c4iw_unregister_device(ctx->dev);
 893         c4iw_dealloc(ctx);
 894 }
 895
 896 static int rdma_supported(const struct cxgb4_lld_info *infop)
 897 {
 898         return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
 899                infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
 900                infop->vr->cq.size > 0;
 901 }
 902
 903 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
 904 {
 905         struct c4iw_dev *devp;
 906         int ret;
 907
 908         if (!rdma_supported(infop)) {
 909                 printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
 910                        pci_name(infop->pdev));
 911                 return ERR_PTR(-ENOSYS);
 912         }
 913         if (!ocqp_supported(infop))
 914                 pr_info("%s: On-Chip Queues not supported on this device.\n",
 915                         pci_name(infop->pdev));
 916
 917         devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
 918         if (!devp) {
 919                 printk(KERN_ERR MOD "Cannot allocate ib device\n");
 920                 return ERR_PTR(-ENOMEM);
 921         }
 922         devp->rdev.lldi = *infop;
 923
 924         /* init various hw-queue params based on lld info */
 925         PDBG("%s: Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
 926              __func__, devp->rdev.lldi.sge_ingpadboundary,
 927              devp->rdev.lldi.sge_egrstatuspagesize);
 928
 929         devp->rdev.hw_queue.t4_eq_status_entries =
 930                 devp->rdev.lldi.sge_ingpadboundary > 64 ? 2 : 1;
 931         devp->rdev.hw_queue.t4_max_eq_size = 65520;
 932         devp->rdev.hw_queue.t4_max_iq_size = 65520;
 933         devp->rdev.hw_queue.t4_max_rq_size = 8192 -
 934                 devp->rdev.hw_queue.t4_eq_status_entries - 1;
 935         devp->rdev.hw_queue.t4_max_sq_size =
 936                 devp->rdev.hw_queue.t4_max_eq_size -
 937                 devp->rdev.hw_queue.t4_eq_status_entries - 1;
 938         devp->rdev.hw_queue.t4_max_qp_depth =
 939                 devp->rdev.hw_queue.t4_max_rq_size;
 940         devp->rdev.hw_queue.t4_max_cq_depth =
 941                 devp->rdev.hw_queue.t4_max_iq_size - 2;
 942         devp->rdev.hw_queue.t4_stat_len =
 943                 devp->rdev.lldi.sge_egrstatuspagesize;
 944
 945         /*
 946          * For T5/T6 devices, we map all of BAR2 with WC.
 947          * For T4 devices with onchip qp mem, we map only that part
 948          * of BAR2 with WC.
 949          */
 950         devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
 951         if (!is_t4(devp->rdev.lldi.adapter_type)) {
 952                 devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
 953                         pci_resource_len(devp->rdev.lldi.pdev, 2));
 954                 if (!devp->rdev.bar2_kva) {
 955                         pr_err(MOD "Unable to ioremap BAR2\n");
 956                         ib_dealloc_device(&devp->ibdev);
 957                         return ERR_PTR(-EINVAL);
 958                 }
 959         } else if (ocqp_supported(infop)) {
 960                 devp->rdev.oc_mw_pa =
 961                         pci_resource_start(devp->rdev.lldi.pdev, 2) +
 962                         pci_resource_len(devp->rdev.lldi.pdev, 2) -
 963                         roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
 964                 devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
 965                         devp->rdev.lldi.vr->ocq.size);
 966                 if (!devp->rdev.oc_mw_kva) {
 967                         pr_err(MOD "Unable to ioremap onchip mem\n");
 968                         ib_dealloc_device(&devp->ibdev);
 969                         return ERR_PTR(-EINVAL);
 970                 }
 971         }
 972
 973         PDBG(KERN_INFO MOD "ocq memory: "
 974                "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
 975                devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
 976                devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
 977
 978         ret = c4iw_rdev_open(&devp->rdev);
 979         if (ret) {
 980                 printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
 981                 ib_dealloc_device(&devp->ibdev);
 982                 return ERR_PTR(ret);
 983         }
 984
 985         idr_init(&devp->cqidr);
 986         idr_init(&devp->qpidr);
 987         idr_init(&devp->mmidr);
 988         idr_init(&devp->hwtid_idr);
 989         idr_init(&devp->stid_idr);
 990         idr_init(&devp->atid_idr);
 991         spin_lock_init(&devp->lock);
 992         mutex_init(&devp->rdev.stats.lock);
 993         mutex_init(&devp->db_mutex);
 994         INIT_LIST_HEAD(&devp->db_fc_list);
 995         devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
 996
 997         if (c4iw_debugfs_root) {
 998                 devp->debugfs_root = debugfs_create_dir(
 999                                         pci_name(devp->rdev.lldi.pdev),
1000                                         c4iw_debugfs_root);
1001                 setup_debugfs(devp);
1002         }
1003
1004
1005         return devp;
1006 }
1007
1008 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1009 {
1010         struct uld_ctx *ctx;
1011         static int vers_printed;
1012         int i;
1013
1014         if (!vers_printed++)
1015                 pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1016                         DRV_VERSION);
1017
1018         ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1019         if (!ctx) {
1020                 ctx = ERR_PTR(-ENOMEM);
1021                 goto out;
1022         }
1023         ctx->lldi = *infop;
1024
1025         PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1026              __func__, pci_name(ctx->lldi.pdev),
1027              ctx->lldi.nchan, ctx->lldi.nrxq,
1028              ctx->lldi.ntxq, ctx->lldi.nports);
1029
1030         mutex_lock(&dev_mutex);
1031         list_add_tail(&ctx->entry, &uld_ctx_list);
1032         mutex_unlock(&dev_mutex);
1033
1034         for (i = 0; i < ctx->lldi.nrxq; i++)
1035                 PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1036 out:
1037         return ctx;
1038 }
1039
1040 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1041                                                  const __be64 *rsp,
1042                                                  u32 pktshift)
1043 {
1044         struct sk_buff *skb;
1045
1046         /*
1047          * Allocate space for cpl_pass_accept_req which will be synthesized by
1048          * driver. Once the driver synthesizes the request the skb will go
1049          * through the regular cpl_pass_accept_req processing.
1050          * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1051          * cpl_rx_pkt.
1052          */
1053         skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1054                         sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1055         if (unlikely(!skb))
1056                 return NULL;
1057
1058          __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1059                    sizeof(struct rss_header) - pktshift);
1060
1061         /*
1062          * This skb will contain:
1063          *   rss_header from the rspq descriptor (1 flit)
1064          *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1065          *   space for the difference between the size of an
1066          *      rx_pkt and pass_accept_req cpl (1 flit)
1067          *   the packet data from the gl
1068          */
1069         skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1070                                 sizeof(struct rss_header));
1071         skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1072                                        sizeof(struct cpl_pass_accept_req),
1073                                        gl->va + pktshift,
1074                                        gl->tot_len - pktshift);
1075         return skb;
1076 }
1077
1078 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1079                            const __be64 *rsp)
1080 {
1081         unsigned int opcode = *(u8 *)rsp;
1082         struct sk_buff *skb;
1083
1084         if (opcode != CPL_RX_PKT)
1085                 goto out;
1086
1087         skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1088         if (skb == NULL)
1089                 goto out;
1090
1091         if (c4iw_handlers[opcode] == NULL) {
1092                 pr_info("%s no handler opcode 0x%x...\n", __func__,
1093                        opcode);
1094                 kfree_skb(skb);
1095                 goto out;
1096         }
1097         c4iw_handlers[opcode](dev, skb);
1098         return 1;
1099 out:
1100         return 0;
1101 }
1102
1103 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1104                         const struct pkt_gl *gl)
1105 {
1106         struct uld_ctx *ctx = handle;
1107         struct c4iw_dev *dev = ctx->dev;
1108         struct sk_buff *skb;
1109         u8 opcode;
1110
1111         if (gl == NULL) {
1112                 /* omit RSS and rsp_ctrl at end of descriptor */
1113                 unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1114
1115                 skb = alloc_skb(256, GFP_ATOMIC);
1116                 if (!skb)
1117                         goto nomem;
1118                 __skb_put(skb, len);
1119                 skb_copy_to_linear_data(skb, &rsp[1], len);
1120         } else if (gl == CXGB4_MSG_AN) {
1121                 const struct rsp_ctrl *rc = (void *)rsp;
1122
1123                 u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1124                 c4iw_ev_handler(dev, qid);
1125                 return 0;
1126         } else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1127                 if (recv_rx_pkt(dev, gl, rsp))
1128                         return 0;
1129
1130                 pr_info("%s: unexpected FL contents at %p, " \
1131                        "RSS %#llx, FL %#llx, len %u\n",
1132                        pci_name(ctx->lldi.pdev), gl->va,
1133                        (unsigned long long)be64_to_cpu(*rsp),
1134                        (unsigned long long)be64_to_cpu(
1135                        *(__force __be64 *)gl->va),
1136                        gl->tot_len);
1137
1138                 return 0;
1139         } else {
1140                 skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1141                 if (unlikely(!skb))
1142                         goto nomem;
1143         }
1144
1145         opcode = *(u8 *)rsp;
1146         if (c4iw_handlers[opcode]) {
1147                 c4iw_handlers[opcode](dev, skb);
1148         } else {
1149                 pr_info("%s no handler opcode 0x%x...\n", __func__,
1150                        opcode);
1151                 kfree_skb(skb);
1152         }
1153
1154         return 0;
1155 nomem:
1156         return -1;
1157 }
1158
1159 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1160 {
1161         struct uld_ctx *ctx = handle;
1162
1163         PDBG("%s new_state %u\n", __func__, new_state);
1164         switch (new_state) {
1165         case CXGB4_STATE_UP:
1166                 printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
1167                 if (!ctx->dev) {
1168                         int ret;
1169
1170                         ctx->dev = c4iw_alloc(&ctx->lldi);
1171                         if (IS_ERR(ctx->dev)) {
1172                                 printk(KERN_ERR MOD
1173                                        "%s: initialization failed: %ld\n",
1174                                        pci_name(ctx->lldi.pdev),
1175                                        PTR_ERR(ctx->dev));
1176                                 ctx->dev = NULL;
1177                                 break;
1178                         }
1179                         ret = c4iw_register_device(ctx->dev);
1180                         if (ret) {
1181                                 printk(KERN_ERR MOD
1182                                        "%s: RDMA registration failed: %d\n",
1183                                        pci_name(ctx->lldi.pdev), ret);
1184                                 c4iw_dealloc(ctx);
1185                         }
1186                 }
1187                 break;
1188         case CXGB4_STATE_DOWN:
1189                 printk(KERN_INFO MOD "%s: Down\n",
1190                        pci_name(ctx->lldi.pdev));
1191                 if (ctx->dev)
1192                         c4iw_remove(ctx);
1193                 break;
1194         case CXGB4_STATE_START_RECOVERY:
1195                 printk(KERN_INFO MOD "%s: Fatal Error\n",
1196                        pci_name(ctx->lldi.pdev));
1197                 if (ctx->dev) {
1198                         struct ib_event event;
1199
1200                         ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1201                         memset(&event, 0, sizeof event);
1202                         event.event  = IB_EVENT_DEVICE_FATAL;
1203                         event.device = &ctx->dev->ibdev;
1204                         ib_dispatch_event(&event);
1205                         c4iw_remove(ctx);
1206                 }
1207                 break;
1208         case CXGB4_STATE_DETACH:
1209                 printk(KERN_INFO MOD "%s: Detach\n",
1210                        pci_name(ctx->lldi.pdev));
1211                 if (ctx->dev)
1212                         c4iw_remove(ctx);
1213                 break;
1214         }
1215         return 0;
1216 }
1217
1218 static int disable_qp_db(int id, void *p, void *data)
1219 {
1220         struct c4iw_qp *qp = p;
1221
1222         t4_disable_wq_db(&qp->wq);
1223         return 0;
1224 }
1225
1226 static void stop_queues(struct uld_ctx *ctx)
1227 {
1228         unsigned long flags;
1229
1230         spin_lock_irqsave(&ctx->dev->lock, flags);
1231         ctx->dev->rdev.stats.db_state_transitions++;
1232         ctx->dev->db_state = STOPPED;
1233         if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1234                 idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1235         else
1236                 ctx->dev->rdev.status_page->db_off = 1;
1237         spin_unlock_irqrestore(&ctx->dev->lock, flags);
1238 }
1239
1240 static int enable_qp_db(int id, void *p, void *data)
1241 {
1242         struct c4iw_qp *qp = p;
1243
1244         t4_enable_wq_db(&qp->wq);
1245         return 0;
1246 }
1247
1248 static void resume_rc_qp(struct c4iw_qp *qp)
1249 {
1250         spin_lock(&qp->lock);
1251         t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc, NULL);
1252         qp->wq.sq.wq_pidx_inc = 0;
1253         t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc, NULL);
1254         qp->wq.rq.wq_pidx_inc = 0;
1255         spin_unlock(&qp->lock);
1256 }
1257
1258 static void resume_a_chunk(struct uld_ctx *ctx)
1259 {
1260         int i;
1261         struct c4iw_qp *qp;
1262
1263         for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1264                 qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1265                                       db_fc_entry);
1266                 list_del_init(&qp->db_fc_entry);
1267                 resume_rc_qp(qp);
1268                 if (list_empty(&ctx->dev->db_fc_list))
1269                         break;
1270         }
1271 }
1272
1273 static void resume_queues(struct uld_ctx *ctx)
1274 {
1275         spin_lock_irq(&ctx->dev->lock);
1276         if (ctx->dev->db_state != STOPPED)
1277                 goto out;
1278         ctx->dev->db_state = FLOW_CONTROL;
1279         while (1) {
1280                 if (list_empty(&ctx->dev->db_fc_list)) {
1281                         WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1282                         ctx->dev->db_state = NORMAL;
1283                         ctx->dev->rdev.stats.db_state_transitions++;
1284                         if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1285                                 idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1286                                              NULL);
1287                         } else {
1288                                 ctx->dev->rdev.status_page->db_off = 0;
1289                         }
1290                         break;
1291                 } else {
1292                         if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1293                             < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1294                                DB_FC_DRAIN_THRESH)) {
1295                                 resume_a_chunk(ctx);
1296                         }
1297                         if (!list_empty(&ctx->dev->db_fc_list)) {
1298                                 spin_unlock_irq(&ctx->dev->lock);
1299                                 if (DB_FC_RESUME_DELAY) {
1300                                         set_current_state(TASK_UNINTERRUPTIBLE);
1301                                         schedule_timeout(DB_FC_RESUME_DELAY);
1302                                 }
1303                                 spin_lock_irq(&ctx->dev->lock);
1304                                 if (ctx->dev->db_state != FLOW_CONTROL)
1305                                         break;
1306                         }
1307                 }
1308         }
1309 out:
1310         if (ctx->dev->db_state != NORMAL)
1311                 ctx->dev->rdev.stats.db_fc_interruptions++;
1312         spin_unlock_irq(&ctx->dev->lock);
1313 }
1314
1315 struct qp_list {
1316         unsigned idx;
1317         struct c4iw_qp **qps;
1318 };
1319
1320 static int add_and_ref_qp(int id, void *p, void *data)
1321 {
1322         struct qp_list *qp_listp = data;
1323         struct c4iw_qp *qp = p;
1324
1325         c4iw_qp_add_ref(&qp->ibqp);
1326         qp_listp->qps[qp_listp->idx++] = qp;
1327         return 0;
1328 }
1329
1330 static int count_qps(int id, void *p, void *data)
1331 {
1332         unsigned *countp = data;
1333         (*countp)++;
1334         return 0;
1335 }
1336
1337 static void deref_qps(struct qp_list *qp_list)
1338 {
1339         int idx;
1340
1341         for (idx = 0; idx < qp_list->idx; idx++)
1342                 c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1343 }
1344
1345 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1346 {
1347         int idx;
1348         int ret;
1349
1350         for (idx = 0; idx < qp_list->idx; idx++) {
1351                 struct c4iw_qp *qp = qp_list->qps[idx];
1352
1353                 spin_lock_irq(&qp->rhp->lock);
1354                 spin_lock(&qp->lock);
1355                 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1356                                           qp->wq.sq.qid,
1357                                           t4_sq_host_wq_pidx(&qp->wq),
1358                                           t4_sq_wq_size(&qp->wq));
1359                 if (ret) {
1360                         pr_err(MOD "%s: Fatal error - "
1361                                "DB overflow recovery failed - "
1362                                "error syncing SQ qid %u\n",
1363                                pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1364                         spin_unlock(&qp->lock);
1365                         spin_unlock_irq(&qp->rhp->lock);
1366                         return;
1367                 }
1368                 qp->wq.sq.wq_pidx_inc = 0;
1369
1370                 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1371                                           qp->wq.rq.qid,
1372                                           t4_rq_host_wq_pidx(&qp->wq),
1373                                           t4_rq_wq_size(&qp->wq));
1374
1375                 if (ret) {
1376                         pr_err(MOD "%s: Fatal error - "
1377                                "DB overflow recovery failed - "
1378                                "error syncing RQ qid %u\n",
1379                                pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1380                         spin_unlock(&qp->lock);
1381                         spin_unlock_irq(&qp->rhp->lock);
1382                         return;
1383                 }
1384                 qp->wq.rq.wq_pidx_inc = 0;
1385                 spin_unlock(&qp->lock);
1386                 spin_unlock_irq(&qp->rhp->lock);
1387
1388                 /* Wait for the dbfifo to drain */
1389                 while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1390                         set_current_state(TASK_UNINTERRUPTIBLE);
1391                         schedule_timeout(usecs_to_jiffies(10));
1392                 }
1393         }
1394 }
1395
1396 static void recover_queues(struct uld_ctx *ctx)
1397 {
1398         int count = 0;
1399         struct qp_list qp_list;
1400         int ret;
1401
1402         /* slow everybody down */
1403         set_current_state(TASK_UNINTERRUPTIBLE);
1404         schedule_timeout(usecs_to_jiffies(1000));
1405
1406         /* flush the SGE contexts */
1407         ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1408         if (ret) {
1409                 printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1410                        pci_name(ctx->lldi.pdev));
1411                 return;
1412         }
1413
1414         /* Count active queues so we can build a list of queues to recover */
1415         spin_lock_irq(&ctx->dev->lock);
1416         WARN_ON(ctx->dev->db_state != STOPPED);
1417         ctx->dev->db_state = RECOVERY;
1418         idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1419
1420         qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1421         if (!qp_list.qps) {
1422                 printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1423                        pci_name(ctx->lldi.pdev));
1424                 spin_unlock_irq(&ctx->dev->lock);
1425                 return;
1426         }
1427         qp_list.idx = 0;
1428
1429         /* add and ref each qp so it doesn't get freed */
1430         idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1431
1432         spin_unlock_irq(&ctx->dev->lock);
1433
1434         /* now traverse the list in a safe context to recover the db state*/
1435         recover_lost_dbs(ctx, &qp_list);
1436
1437         /* we're almost done!  deref the qps and clean up */
1438         deref_qps(&qp_list);
1439         kfree(qp_list.qps);
1440
1441         spin_lock_irq(&ctx->dev->lock);
1442         WARN_ON(ctx->dev->db_state != RECOVERY);
1443         ctx->dev->db_state = STOPPED;
1444         spin_unlock_irq(&ctx->dev->lock);
1445 }
1446
1447 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1448 {
1449         struct uld_ctx *ctx = handle;
1450
1451         switch (control) {
1452         case CXGB4_CONTROL_DB_FULL:
1453                 stop_queues(ctx);
1454                 ctx->dev->rdev.stats.db_full++;
1455                 break;
1456         case CXGB4_CONTROL_DB_EMPTY:
1457                 resume_queues(ctx);
1458                 mutex_lock(&ctx->dev->rdev.stats.lock);
1459                 ctx->dev->rdev.stats.db_empty++;
1460                 mutex_unlock(&ctx->dev->rdev.stats.lock);
1461                 break;
1462         case CXGB4_CONTROL_DB_DROP:
1463                 recover_queues(ctx);
1464                 mutex_lock(&ctx->dev->rdev.stats.lock);
1465                 ctx->dev->rdev.stats.db_drop++;
1466                 mutex_unlock(&ctx->dev->rdev.stats.lock);
1467                 break;
1468         default:
1469                 printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
1470                        pci_name(ctx->lldi.pdev), control);
1471                 break;
1472         }
1473         return 0;
1474 }
1475
1476 static struct cxgb4_uld_info c4iw_uld_info = {
1477         .name = DRV_NAME,
1478         .add = c4iw_uld_add,
1479         .rx_handler = c4iw_uld_rx_handler,
1480         .state_change = c4iw_uld_state_change,
1481         .control = c4iw_uld_control,
1482 };
1483
1484 static int __init c4iw_init_module(void)
1485 {
1486         int err;
1487
1488         err = c4iw_cm_init();
1489         if (err)
1490                 return err;
1491
1492         c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1493         if (!c4iw_debugfs_root)
1494                 printk(KERN_WARNING MOD
1495                        "could not create debugfs entry, continuing\n");
1496
1497         cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1498
1499         return 0;
1500 }
1501
1502 static void __exit c4iw_exit_module(void)
1503 {
1504         struct uld_ctx *ctx, *tmp;
1505
1506         mutex_lock(&dev_mutex);
1507         list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1508                 if (ctx->dev)
1509                         c4iw_remove(ctx);
1510                 kfree(ctx);
1511         }
1512         mutex_unlock(&dev_mutex);
1513         cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1514         c4iw_cm_term();
1515         debugfs_remove_recursive(c4iw_debugfs_root);
1516 }
1517
1518 module_init(c4iw_init_module);
1519 module_exit(c4iw_exit_module);