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
  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 int c4iw_wr_log = 0;
  59 module_param(c4iw_wr_log, int, 0444);
  60 MODULE_PARM_DESC(c4iw_wr_log, "Enables logging of work request timing data.");
  61
  62 static int c4iw_wr_log_size_order = 12;
  63 module_param(c4iw_wr_log_size_order, int, 0444);
  64 MODULE_PARM_DESC(c4iw_wr_log_size_order,
  65                  "Number of entries (log2) in the work request timing log.");
  66
  67 struct uld_ctx {
  68         struct list_head entry;
  69         struct cxgb4_lld_info lldi;
  70         struct c4iw_dev *dev;
  71 };
  72
  73 static LIST_HEAD(uld_ctx_list);
  74 static DEFINE_MUTEX(dev_mutex);
  75
  76 #define DB_FC_RESUME_SIZE 64
  77 #define DB_FC_RESUME_DELAY 1
  78 #define DB_FC_DRAIN_THRESH 0
  79
  80 static struct dentry *c4iw_debugfs_root;
  81
  82 struct c4iw_debugfs_data {
  83         struct c4iw_dev *devp;
  84         char *buf;
  85         int bufsize;
  86         int pos;
  87 };
  88
  89 static int count_idrs(int id, void *p, void *data)
  90 {
  91         int *countp = data;
  92
  93         *countp = *countp + 1;
  94         return 0;
  95 }
  96
  97 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
  98                             loff_t *ppos)
  99 {
 100         struct c4iw_debugfs_data *d = file->private_data;
 101
 102         return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
 103 }
 104
 105 void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
 106 {
 107         struct wr_log_entry le;
 108         int idx;
 109
 110         if (!wq->rdev->wr_log)
 111                 return;
 112
 113         idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
 114                 (wq->rdev->wr_log_size - 1);
 115         le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
 116         getnstimeofday(&le.poll_host_ts);
 117         le.valid = 1;
 118         le.cqe_sge_ts = CQE_TS(cqe);
 119         if (SQ_TYPE(cqe)) {
 120                 le.qid = wq->sq.qid;
 121                 le.opcode = CQE_OPCODE(cqe);
 122                 le.post_host_ts = wq->sq.sw_sq[wq->sq.cidx].host_ts;
 123                 le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
 124                 le.wr_id = CQE_WRID_SQ_IDX(cqe);
 125         } else {
 126                 le.qid = wq->rq.qid;
 127                 le.opcode = FW_RI_RECEIVE;
 128                 le.post_host_ts = wq->rq.sw_rq[wq->rq.cidx].host_ts;
 129                 le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
 130                 le.wr_id = CQE_WRID_MSN(cqe);
 131         }
 132         wq->rdev->wr_log[idx] = le;
 133 }
 134
 135 static int wr_log_show(struct seq_file *seq, void *v)
 136 {
 137         struct c4iw_dev *dev = seq->private;
 138         struct timespec prev_ts = {0, 0};
 139         struct wr_log_entry *lep;
 140         int prev_ts_set = 0;
 141         int idx, end;
 142
 143 #define ts2ns(ts) div64_u64((ts) * dev->rdev.lldi.cclk_ps, 1000)
 144
 145         idx = atomic_read(&dev->rdev.wr_log_idx) &
 146                 (dev->rdev.wr_log_size - 1);
 147         end = idx - 1;
 148         if (end < 0)
 149                 end = dev->rdev.wr_log_size - 1;
 150         lep = &dev->rdev.wr_log[idx];
 151         while (idx != end) {
 152                 if (lep->valid) {
 153                         if (!prev_ts_set) {
 154                                 prev_ts_set = 1;
 155                                 prev_ts = lep->poll_host_ts;
 156                         }
 157                         seq_printf(seq, "%04u: sec %lu nsec %lu qid %u opcode "
 158                                    "%u %s 0x%x host_wr_delta sec %lu nsec %lu "
 159                                    "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
 160                                    "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
 161                                    "cqe_poll_delta_ns %llu\n",
 162                                    idx,
 163                                    timespec_sub(lep->poll_host_ts,
 164                                                 prev_ts).tv_sec,
 165                                    timespec_sub(lep->poll_host_ts,
 166                                                 prev_ts).tv_nsec,
 167                                    lep->qid, lep->opcode,
 168                                    lep->opcode == FW_RI_RECEIVE ?
 169                                                         "msn" : "wrid",
 170                                    lep->wr_id,
 171                                    timespec_sub(lep->poll_host_ts,
 172                                                 lep->post_host_ts).tv_sec,
 173                                    timespec_sub(lep->poll_host_ts,
 174                                                 lep->post_host_ts).tv_nsec,
 175                                    lep->post_sge_ts, lep->cqe_sge_ts,
 176                                    lep->poll_sge_ts,
 177                                    ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
 178                                    ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
 179                         prev_ts = lep->poll_host_ts;
 180                 }
 181                 idx++;
 182                 if (idx > (dev->rdev.wr_log_size - 1))
 183                         idx = 0;
 184                 lep = &dev->rdev.wr_log[idx];
 185         }
 186 #undef ts2ns
 187         return 0;
 188 }
 189
 190 static int wr_log_open(struct inode *inode, struct file *file)
 191 {
 192         return single_open(file, wr_log_show, inode->i_private);
 193 }
 194
 195 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
 196                             size_t count, loff_t *pos)
 197 {
 198         struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
 199         int i;
 200
 201         if (dev->rdev.wr_log)
 202                 for (i = 0; i < dev->rdev.wr_log_size; i++)
 203                         dev->rdev.wr_log[i].valid = 0;
 204         return count;
 205 }
 206
 207 static const struct file_operations wr_log_debugfs_fops = {
 208         .owner   = THIS_MODULE,
 209         .open    = wr_log_open,
 210         .release = single_release,
 211         .read    = seq_read,
 212         .llseek  = seq_lseek,
 213         .write   = wr_log_clear,
 214 };
 215
 216 static struct sockaddr_in zero_sin = {
 217         .sin_family = AF_INET,
 218 };
 219
 220 static struct sockaddr_in6 zero_sin6 = {
 221         .sin6_family = AF_INET6,
 222 };
 223
 224 static void set_ep_sin_addrs(struct c4iw_ep *ep,
 225                              struct sockaddr_in **lsin,
 226                              struct sockaddr_in **rsin,
 227                              struct sockaddr_in **m_lsin,
 228                              struct sockaddr_in **m_rsin)
 229 {
 230         struct iw_cm_id *id = ep->com.cm_id;
 231
 232         *lsin = (struct sockaddr_in *)&ep->com.local_addr;
 233         *rsin = (struct sockaddr_in *)&ep->com.remote_addr;
 234         if (id) {
 235                 *m_lsin = (struct sockaddr_in *)&id->m_local_addr;
 236                 *m_rsin = (struct sockaddr_in *)&id->m_remote_addr;
 237         } else {
 238                 *m_lsin = &zero_sin;
 239                 *m_rsin = &zero_sin;
 240         }
 241 }
 242
 243 static void set_ep_sin6_addrs(struct c4iw_ep *ep,
 244                               struct sockaddr_in6 **lsin6,
 245                               struct sockaddr_in6 **rsin6,
 246                               struct sockaddr_in6 **m_lsin6,
 247                               struct sockaddr_in6 **m_rsin6)
 248 {
 249         struct iw_cm_id *id = ep->com.cm_id;
 250
 251         *lsin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
 252         *rsin6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
 253         if (id) {
 254                 *m_lsin6 = (struct sockaddr_in6 *)&id->m_local_addr;
 255                 *m_rsin6 = (struct sockaddr_in6 *)&id->m_remote_addr;
 256         } else {
 257                 *m_lsin6 = &zero_sin6;
 258                 *m_rsin6 = &zero_sin6;
 259         }
 260 }
 261
 262 static int dump_qp(int id, void *p, void *data)
 263 {
 264         struct c4iw_qp *qp = p;
 265         struct c4iw_debugfs_data *qpd = data;
 266         int space;
 267         int cc;
 268
 269         if (id != qp->wq.sq.qid)
 270                 return 0;
 271
 272         space = qpd->bufsize - qpd->pos - 1;
 273         if (space == 0)
 274                 return 1;
 275
 276         if (qp->ep) {
 277                 struct c4iw_ep *ep = qp->ep;
 278
 279                 if (ep->com.local_addr.ss_family == AF_INET) {
 280                         struct sockaddr_in *lsin;
 281                         struct sockaddr_in *rsin;
 282                         struct sockaddr_in *m_lsin;
 283                         struct sockaddr_in *m_rsin;
 284
 285                         set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
 286                         cc = snprintf(qpd->buf + qpd->pos, space,
 287                                       "rc qp sq id %u rq id %u state %u "
 288                                       "onchip %u ep tid %u state %u "
 289                                       "%pI4:%u/%u->%pI4:%u/%u\n",
 290                                       qp->wq.sq.qid, qp->wq.rq.qid,
 291                                       (int)qp->attr.state,
 292                                       qp->wq.sq.flags & T4_SQ_ONCHIP,
 293                                       ep->hwtid, (int)ep->com.state,
 294                                       &lsin->sin_addr, ntohs(lsin->sin_port),
 295                                       ntohs(m_lsin->sin_port),
 296                                       &rsin->sin_addr, ntohs(rsin->sin_port),
 297                                       ntohs(m_rsin->sin_port));
 298                 } else {
 299                         struct sockaddr_in6 *lsin6;
 300                         struct sockaddr_in6 *rsin6;
 301                         struct sockaddr_in6 *m_lsin6;
 302                         struct sockaddr_in6 *m_rsin6;
 303
 304                         set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6,
 305                                           &m_rsin6);
 306                         cc = snprintf(qpd->buf + qpd->pos, space,
 307                                       "rc qp sq id %u rq id %u state %u "
 308                                       "onchip %u ep tid %u state %u "
 309                                       "%pI6:%u/%u->%pI6:%u/%u\n",
 310                                       qp->wq.sq.qid, qp->wq.rq.qid,
 311                                       (int)qp->attr.state,
 312                                       qp->wq.sq.flags & T4_SQ_ONCHIP,
 313                                       ep->hwtid, (int)ep->com.state,
 314                                       &lsin6->sin6_addr,
 315                                       ntohs(lsin6->sin6_port),
 316                                       ntohs(m_lsin6->sin6_port),
 317                                       &rsin6->sin6_addr,
 318                                       ntohs(rsin6->sin6_port),
 319                                       ntohs(m_rsin6->sin6_port));
 320                 }
 321         } else
 322                 cc = snprintf(qpd->buf + qpd->pos, space,
 323                              "qp sq id %u rq id %u state %u onchip %u\n",
 324                               qp->wq.sq.qid, qp->wq.rq.qid,
 325                               (int)qp->attr.state,
 326                               qp->wq.sq.flags & T4_SQ_ONCHIP);
 327         if (cc < space)
 328                 qpd->pos += cc;
 329         return 0;
 330 }
 331
 332 static int qp_release(struct inode *inode, struct file *file)
 333 {
 334         struct c4iw_debugfs_data *qpd = file->private_data;
 335         if (!qpd) {
 336                 pr_info("%s null qpd?\n", __func__);
 337                 return 0;
 338         }
 339         vfree(qpd->buf);
 340         kfree(qpd);
 341         return 0;
 342 }
 343
 344 static int qp_open(struct inode *inode, struct file *file)
 345 {
 346         struct c4iw_debugfs_data *qpd;
 347         int count = 1;
 348
 349         qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
 350         if (!qpd)
 351                 return -ENOMEM;
 352
 353         qpd->devp = inode->i_private;
 354         qpd->pos = 0;
 355
 356         spin_lock_irq(&qpd->devp->lock);
 357         idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
 358         spin_unlock_irq(&qpd->devp->lock);
 359
 360         qpd->bufsize = count * 180;
 361         qpd->buf = vmalloc(qpd->bufsize);
 362         if (!qpd->buf) {
 363                 kfree(qpd);
 364                 return -ENOMEM;
 365         }
 366
 367         spin_lock_irq(&qpd->devp->lock);
 368         idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
 369         spin_unlock_irq(&qpd->devp->lock);
 370
 371         qpd->buf[qpd->pos++] = 0;
 372         file->private_data = qpd;
 373         return 0;
 374 }
 375
 376 static const struct file_operations qp_debugfs_fops = {
 377         .owner   = THIS_MODULE,
 378         .open    = qp_open,
 379         .release = qp_release,
 380         .read    = debugfs_read,
 381         .llseek  = default_llseek,
 382 };
 383
 384 static int dump_stag(int id, void *p, void *data)
 385 {
 386         struct c4iw_debugfs_data *stagd = data;
 387         int space;
 388         int cc;
 389         struct fw_ri_tpte tpte;
 390         int ret;
 391
 392         space = stagd->bufsize - stagd->pos - 1;
 393         if (space == 0)
 394                 return 1;
 395
 396         ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
 397                               (__be32 *)&tpte);
 398         if (ret) {
 399                 dev_err(&stagd->devp->rdev.lldi.pdev->dev,
 400                         "%s cxgb4_read_tpte err %d\n", __func__, ret);
 401                 return ret;
 402         }
 403         cc = snprintf(stagd->buf + stagd->pos, space,
 404                       "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
 405                       "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
 406                       (u32)id<<8,
 407                       FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
 408                       FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
 409                       FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
 410                       FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
 411                       FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
 412                       FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
 413                       ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
 414                       ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
 415         if (cc < space)
 416                 stagd->pos += cc;
 417         return 0;
 418 }
 419
 420 static int stag_release(struct inode *inode, struct file *file)
 421 {
 422         struct c4iw_debugfs_data *stagd = file->private_data;
 423         if (!stagd) {
 424                 pr_info("%s null stagd?\n", __func__);
 425                 return 0;
 426         }
 427         vfree(stagd->buf);
 428         kfree(stagd);
 429         return 0;
 430 }
 431
 432 static int stag_open(struct inode *inode, struct file *file)
 433 {
 434         struct c4iw_debugfs_data *stagd;
 435         int ret = 0;
 436         int count = 1;
 437
 438         stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
 439         if (!stagd) {
 440                 ret = -ENOMEM;
 441                 goto out;
 442         }
 443         stagd->devp = inode->i_private;
 444         stagd->pos = 0;
 445
 446         spin_lock_irq(&stagd->devp->lock);
 447         idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
 448         spin_unlock_irq(&stagd->devp->lock);
 449
 450         stagd->bufsize = count * 256;
 451         stagd->buf = vmalloc(stagd->bufsize);
 452         if (!stagd->buf) {
 453                 ret = -ENOMEM;
 454                 goto err1;
 455         }
 456
 457         spin_lock_irq(&stagd->devp->lock);
 458         idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
 459         spin_unlock_irq(&stagd->devp->lock);
 460
 461         stagd->buf[stagd->pos++] = 0;
 462         file->private_data = stagd;
 463         goto out;
 464 err1:
 465         kfree(stagd);
 466 out:
 467         return ret;
 468 }
 469
 470 static const struct file_operations stag_debugfs_fops = {
 471         .owner   = THIS_MODULE,
 472         .open    = stag_open,
 473         .release = stag_release,
 474         .read    = debugfs_read,
 475         .llseek  = default_llseek,
 476 };
 477
 478 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
 479
 480 static int stats_show(struct seq_file *seq, void *v)
 481 {
 482         struct c4iw_dev *dev = seq->private;
 483
 484         seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
 485                    "Max", "Fail");
 486         seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
 487                         dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
 488                         dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
 489         seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
 490                         dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
 491                         dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
 492         seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
 493                         dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
 494                         dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
 495         seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
 496                         dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
 497                         dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
 498         seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
 499                         dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
 500                         dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
 501         seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
 502                         dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
 503                         dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
 504         seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
 505         seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
 506         seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
 507         seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
 508                    db_state_str[dev->db_state],
 509                    dev->rdev.stats.db_state_transitions,
 510                    dev->rdev.stats.db_fc_interruptions);
 511         seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
 512         seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
 513                    dev->rdev.stats.act_ofld_conn_fails);
 514         seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
 515                    dev->rdev.stats.pas_ofld_conn_fails);
 516         seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
 517         seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
 518         return 0;
 519 }
 520
 521 static int stats_open(struct inode *inode, struct file *file)
 522 {
 523         return single_open(file, stats_show, inode->i_private);
 524 }
 525
 526 static ssize_t stats_clear(struct file *file, const char __user *buf,
 527                 size_t count, loff_t *pos)
 528 {
 529         struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
 530
 531         mutex_lock(&dev->rdev.stats.lock);
 532         dev->rdev.stats.pd.max = 0;
 533         dev->rdev.stats.pd.fail = 0;
 534         dev->rdev.stats.qid.max = 0;
 535         dev->rdev.stats.qid.fail = 0;
 536         dev->rdev.stats.stag.max = 0;
 537         dev->rdev.stats.stag.fail = 0;
 538         dev->rdev.stats.pbl.max = 0;
 539         dev->rdev.stats.pbl.fail = 0;
 540         dev->rdev.stats.rqt.max = 0;
 541         dev->rdev.stats.rqt.fail = 0;
 542         dev->rdev.stats.ocqp.max = 0;
 543         dev->rdev.stats.ocqp.fail = 0;
 544         dev->rdev.stats.db_full = 0;
 545         dev->rdev.stats.db_empty = 0;
 546         dev->rdev.stats.db_drop = 0;
 547         dev->rdev.stats.db_state_transitions = 0;
 548         dev->rdev.stats.tcam_full = 0;
 549         dev->rdev.stats.act_ofld_conn_fails = 0;
 550         dev->rdev.stats.pas_ofld_conn_fails = 0;
 551         mutex_unlock(&dev->rdev.stats.lock);
 552         return count;
 553 }
 554
 555 static const struct file_operations stats_debugfs_fops = {
 556         .owner   = THIS_MODULE,
 557         .open    = stats_open,
 558         .release = single_release,
 559         .read    = seq_read,
 560         .llseek  = seq_lseek,
 561         .write   = stats_clear,
 562 };
 563
 564 static int dump_ep(int id, void *p, void *data)
 565 {
 566         struct c4iw_ep *ep = p;
 567         struct c4iw_debugfs_data *epd = data;
 568         int space;
 569         int cc;
 570
 571         space = epd->bufsize - epd->pos - 1;
 572         if (space == 0)
 573                 return 1;
 574
 575         if (ep->com.local_addr.ss_family == AF_INET) {
 576                 struct sockaddr_in *lsin;
 577                 struct sockaddr_in *rsin;
 578                 struct sockaddr_in *m_lsin;
 579                 struct sockaddr_in *m_rsin;
 580
 581                 set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
 582                 cc = snprintf(epd->buf + epd->pos, space,
 583                               "ep %p cm_id %p qp %p state %d flags 0x%lx "
 584                               "history 0x%lx hwtid %d atid %d "
 585                               "conn_na %u abort_na %u "
 586                               "%pI4:%d/%d <-> %pI4:%d/%d\n",
 587                               ep, ep->com.cm_id, ep->com.qp,
 588                               (int)ep->com.state, ep->com.flags,
 589                               ep->com.history, ep->hwtid, ep->atid,
 590                               ep->stats.connect_neg_adv,
 591                               ep->stats.abort_neg_adv,
 592                               &lsin->sin_addr, ntohs(lsin->sin_port),
 593                               ntohs(m_lsin->sin_port),
 594                               &rsin->sin_addr, ntohs(rsin->sin_port),
 595                               ntohs(m_rsin->sin_port));
 596         } else {
 597                 struct sockaddr_in6 *lsin6;
 598                 struct sockaddr_in6 *rsin6;
 599                 struct sockaddr_in6 *m_lsin6;
 600                 struct sockaddr_in6 *m_rsin6;
 601
 602                 set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6, &m_rsin6);
 603                 cc = snprintf(epd->buf + epd->pos, space,
 604                               "ep %p cm_id %p qp %p state %d flags 0x%lx "
 605                               "history 0x%lx hwtid %d atid %d "
 606                               "conn_na %u abort_na %u "
 607                               "%pI6:%d/%d <-> %pI6:%d/%d\n",
 608                               ep, ep->com.cm_id, ep->com.qp,
 609                               (int)ep->com.state, ep->com.flags,
 610                               ep->com.history, ep->hwtid, ep->atid,
 611                               ep->stats.connect_neg_adv,
 612                               ep->stats.abort_neg_adv,
 613                               &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
 614                               ntohs(m_lsin6->sin6_port),
 615                               &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
 616                               ntohs(m_rsin6->sin6_port));
 617         }
 618         if (cc < space)
 619                 epd->pos += cc;
 620         return 0;
 621 }
 622
 623 static int dump_listen_ep(int id, void *p, void *data)
 624 {
 625         struct c4iw_listen_ep *ep = p;
 626         struct c4iw_debugfs_data *epd = data;
 627         int space;
 628         int cc;
 629
 630         space = epd->bufsize - epd->pos - 1;
 631         if (space == 0)
 632                 return 1;
 633
 634         if (ep->com.local_addr.ss_family == AF_INET) {
 635                 struct sockaddr_in *lsin = (struct sockaddr_in *)
 636                         &ep->com.cm_id->local_addr;
 637                 struct sockaddr_in *m_lsin = (struct sockaddr_in *)
 638                         &ep->com.cm_id->m_local_addr;
 639
 640                 cc = snprintf(epd->buf + epd->pos, space,
 641                               "ep %p cm_id %p state %d flags 0x%lx stid %d "
 642                               "backlog %d %pI4:%d/%d\n",
 643                               ep, ep->com.cm_id, (int)ep->com.state,
 644                               ep->com.flags, ep->stid, ep->backlog,
 645                               &lsin->sin_addr, ntohs(lsin->sin_port),
 646                               ntohs(m_lsin->sin_port));
 647         } else {
 648                 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
 649                         &ep->com.cm_id->local_addr;
 650                 struct sockaddr_in6 *m_lsin6 = (struct sockaddr_in6 *)
 651                         &ep->com.cm_id->m_local_addr;
 652
 653                 cc = snprintf(epd->buf + epd->pos, space,
 654                               "ep %p cm_id %p state %d flags 0x%lx stid %d "
 655                               "backlog %d %pI6:%d/%d\n",
 656                               ep, ep->com.cm_id, (int)ep->com.state,
 657                               ep->com.flags, ep->stid, ep->backlog,
 658                               &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
 659                               ntohs(m_lsin6->sin6_port));
 660         }
 661         if (cc < space)
 662                 epd->pos += cc;
 663         return 0;
 664 }
 665
 666 static int ep_release(struct inode *inode, struct file *file)
 667 {
 668         struct c4iw_debugfs_data *epd = file->private_data;
 669         if (!epd) {
 670                 pr_info("%s null qpd?\n", __func__);
 671                 return 0;
 672         }
 673         vfree(epd->buf);
 674         kfree(epd);
 675         return 0;
 676 }
 677
 678 static int ep_open(struct inode *inode, struct file *file)
 679 {
 680         struct c4iw_debugfs_data *epd;
 681         int ret = 0;
 682         int count = 1;
 683
 684         epd = kmalloc(sizeof(*epd), GFP_KERNEL);
 685         if (!epd) {
 686                 ret = -ENOMEM;
 687                 goto out;
 688         }
 689         epd->devp = inode->i_private;
 690         epd->pos = 0;
 691
 692         spin_lock_irq(&epd->devp->lock);
 693         idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
 694         idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
 695         idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
 696         spin_unlock_irq(&epd->devp->lock);
 697
 698         epd->bufsize = count * 240;
 699         epd->buf = vmalloc(epd->bufsize);
 700         if (!epd->buf) {
 701                 ret = -ENOMEM;
 702                 goto err1;
 703         }
 704
 705         spin_lock_irq(&epd->devp->lock);
 706         idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
 707         idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
 708         idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
 709         spin_unlock_irq(&epd->devp->lock);
 710
 711         file->private_data = epd;
 712         goto out;
 713 err1:
 714         kfree(epd);
 715 out:
 716         return ret;
 717 }
 718
 719 static const struct file_operations ep_debugfs_fops = {
 720         .owner   = THIS_MODULE,
 721         .open    = ep_open,
 722         .release = ep_release,
 723         .read    = debugfs_read,
 724 };
 725
 726 static int setup_debugfs(struct c4iw_dev *devp)
 727 {
 728         if (!devp->debugfs_root)
 729                 return -1;
 730
 731         debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
 732                                  (void *)devp, &qp_debugfs_fops, 4096);
 733
 734         debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
 735                                  (void *)devp, &stag_debugfs_fops, 4096);
 736
 737         debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
 738                                  (void *)devp, &stats_debugfs_fops, 4096);
 739
 740         debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
 741                                  (void *)devp, &ep_debugfs_fops, 4096);
 742
 743         if (c4iw_wr_log)
 744                 debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
 745                                          (void *)devp, &wr_log_debugfs_fops, 4096);
 746         return 0;
 747 }
 748
 749 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
 750                                struct c4iw_dev_ucontext *uctx)
 751 {
 752         struct list_head *pos, *nxt;
 753         struct c4iw_qid_list *entry;
 754
 755         mutex_lock(&uctx->lock);
 756         list_for_each_safe(pos, nxt, &uctx->qpids) {
 757                 entry = list_entry(pos, struct c4iw_qid_list, entry);
 758                 list_del_init(&entry->entry);
 759                 if (!(entry->qid & rdev->qpmask)) {
 760                         c4iw_put_resource(&rdev->resource.qid_table,
 761                                           entry->qid);
 762                         mutex_lock(&rdev->stats.lock);
 763                         rdev->stats.qid.cur -= rdev->qpmask + 1;
 764                         mutex_unlock(&rdev->stats.lock);
 765                 }
 766                 kfree(entry);
 767         }
 768
 769         list_for_each_safe(pos, nxt, &uctx->cqids) {
 770                 entry = list_entry(pos, struct c4iw_qid_list, entry);
 771                 list_del_init(&entry->entry);
 772                 kfree(entry);
 773         }
 774         mutex_unlock(&uctx->lock);
 775 }
 776
 777 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
 778                             struct c4iw_dev_ucontext *uctx)
 779 {
 780         INIT_LIST_HEAD(&uctx->qpids);
 781         INIT_LIST_HEAD(&uctx->cqids);
 782         mutex_init(&uctx->lock);
 783 }
 784
 785 /* Caller takes care of locking if needed */
 786 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
 787 {
 788         int err;
 789
 790         c4iw_init_dev_ucontext(rdev, &rdev->uctx);
 791
 792         /*
 793          * This implementation assumes udb_density == ucq_density!  Eventually
 794          * we might need to support this but for now fail the open. Also the
 795          * cqid and qpid range must match for now.
 796          */
 797         if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
 798                 pr_err("%s: unsupported udb/ucq densities %u/%u\n",
 799                        pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
 800                        rdev->lldi.ucq_density);
 801                 return -EINVAL;
 802         }
 803         if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
 804             rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
 805                 pr_err("%s: unsupported qp and cq id ranges qp start %u size %u cq start %u size %u\n",
 806                        pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
 807                        rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
 808                        rdev->lldi.vr->cq.size);
 809                 return -EINVAL;
 810         }
 811
 812         rdev->qpmask = rdev->lldi.udb_density - 1;
 813         rdev->cqmask = rdev->lldi.ucq_density - 1;
 814         pr_debug("%s dev %s stag start 0x%0x size 0x%0x num stags %d pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x qp qid start %u size %u cq qid start %u size %u\n",
 815                  __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
 816                  rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
 817                  rdev->lldi.vr->pbl.start,
 818                  rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
 819                  rdev->lldi.vr->rq.size,
 820                  rdev->lldi.vr->qp.start,
 821                  rdev->lldi.vr->qp.size,
 822                  rdev->lldi.vr->cq.start,
 823                  rdev->lldi.vr->cq.size);
 824         pr_debug("udb %pR db_reg %p gts_reg %p qpmask 0x%x cqmask 0x%x\n",
 825                  &rdev->lldi.pdev->resource[2],
 826                  rdev->lldi.db_reg, rdev->lldi.gts_reg,
 827                  rdev->qpmask, rdev->cqmask);
 828
 829         if (c4iw_num_stags(rdev) == 0)
 830                 return -EINVAL;
 831
 832         rdev->stats.pd.total = T4_MAX_NUM_PD;
 833         rdev->stats.stag.total = rdev->lldi.vr->stag.size;
 834         rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
 835         rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
 836         rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
 837         rdev->stats.qid.total = rdev->lldi.vr->qp.size;
 838
 839         err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
 840         if (err) {
 841                 pr_err("error %d initializing resources\n", err);
 842                 return err;
 843         }
 844         err = c4iw_pblpool_create(rdev);
 845         if (err) {
 846                 pr_err("error %d initializing pbl pool\n", err);
 847                 goto destroy_resource;
 848         }
 849         err = c4iw_rqtpool_create(rdev);
 850         if (err) {
 851                 pr_err("error %d initializing rqt pool\n", err);
 852                 goto destroy_pblpool;
 853         }
 854         err = c4iw_ocqp_pool_create(rdev);
 855         if (err) {
 856                 pr_err("error %d initializing ocqp pool\n", err);
 857                 goto destroy_rqtpool;
 858         }
 859         rdev->status_page = (struct t4_dev_status_page *)
 860                             __get_free_page(GFP_KERNEL);
 861         if (!rdev->status_page) {
 862                 err = -ENOMEM;
 863                 goto destroy_ocqp_pool;
 864         }
 865         rdev->status_page->qp_start = rdev->lldi.vr->qp.start;
 866         rdev->status_page->qp_size = rdev->lldi.vr->qp.size;
 867         rdev->status_page->cq_start = rdev->lldi.vr->cq.start;
 868         rdev->status_page->cq_size = rdev->lldi.vr->cq.size;
 869
 870         if (c4iw_wr_log) {
 871                 rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
 872                                        sizeof(*rdev->wr_log), GFP_KERNEL);
 873                 if (rdev->wr_log) {
 874                         rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
 875                         atomic_set(&rdev->wr_log_idx, 0);
 876                 }
 877         }
 878
 879         rdev->free_workq = create_singlethread_workqueue("iw_cxgb4_free");
 880         if (!rdev->free_workq) {
 881                 err = -ENOMEM;
 882                 goto err_free_status_page_and_wr_log;
 883         }
 884
 885         rdev->status_page->db_off = 0;
 886
 887         init_completion(&rdev->rqt_compl);
 888         init_completion(&rdev->pbl_compl);
 889         kref_init(&rdev->rqt_kref);
 890         kref_init(&rdev->pbl_kref);
 891
 892         return 0;
 893 err_free_status_page_and_wr_log:
 894         if (c4iw_wr_log && rdev->wr_log)
 895                 kfree(rdev->wr_log);
 896         free_page((unsigned long)rdev->status_page);
 897 destroy_ocqp_pool:
 898         c4iw_ocqp_pool_destroy(rdev);
 899 destroy_rqtpool:
 900         c4iw_rqtpool_destroy(rdev);
 901 destroy_pblpool:
 902         c4iw_pblpool_destroy(rdev);
 903 destroy_resource:
 904         c4iw_destroy_resource(&rdev->resource);
 905         return err;
 906 }
 907
 908 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
 909 {
 910         kfree(rdev->wr_log);
 911         c4iw_release_dev_ucontext(rdev, &rdev->uctx);
 912         free_page((unsigned long)rdev->status_page);
 913         c4iw_pblpool_destroy(rdev);
 914         c4iw_rqtpool_destroy(rdev);
 915         wait_for_completion(&rdev->pbl_compl);
 916         wait_for_completion(&rdev->rqt_compl);
 917         c4iw_ocqp_pool_destroy(rdev);
 918         destroy_workqueue(rdev->free_workq);
 919         c4iw_destroy_resource(&rdev->resource);
 920 }
 921
 922 static void c4iw_dealloc(struct uld_ctx *ctx)
 923 {
 924         c4iw_rdev_close(&ctx->dev->rdev);
 925         WARN_ON_ONCE(!idr_is_empty(&ctx->dev->cqidr));
 926         idr_destroy(&ctx->dev->cqidr);
 927         WARN_ON_ONCE(!idr_is_empty(&ctx->dev->qpidr));
 928         idr_destroy(&ctx->dev->qpidr);
 929         WARN_ON_ONCE(!idr_is_empty(&ctx->dev->mmidr));
 930         idr_destroy(&ctx->dev->mmidr);
 931         wait_event(ctx->dev->wait, idr_is_empty(&ctx->dev->hwtid_idr));
 932         idr_destroy(&ctx->dev->hwtid_idr);
 933         idr_destroy(&ctx->dev->stid_idr);
 934         idr_destroy(&ctx->dev->atid_idr);
 935         if (ctx->dev->rdev.bar2_kva)
 936                 iounmap(ctx->dev->rdev.bar2_kva);
 937         if (ctx->dev->rdev.oc_mw_kva)
 938                 iounmap(ctx->dev->rdev.oc_mw_kva);
 939         ib_dealloc_device(&ctx->dev->ibdev);
 940         ctx->dev = NULL;
 941 }
 942
 943 static void c4iw_remove(struct uld_ctx *ctx)
 944 {
 945         pr_debug("%s c4iw_dev %p\n", __func__,  ctx->dev);
 946         c4iw_unregister_device(ctx->dev);
 947         c4iw_dealloc(ctx);
 948 }
 949
 950 static int rdma_supported(const struct cxgb4_lld_info *infop)
 951 {
 952         return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
 953                infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
 954                infop->vr->cq.size > 0;
 955 }
 956
 957 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
 958 {
 959         struct c4iw_dev *devp;
 960         int ret;
 961
 962         if (!rdma_supported(infop)) {
 963                 pr_info("%s: RDMA not supported on this device\n",
 964                         pci_name(infop->pdev));
 965                 return ERR_PTR(-ENOSYS);
 966         }
 967         if (!ocqp_supported(infop))
 968                 pr_info("%s: On-Chip Queues not supported on this device\n",
 969                         pci_name(infop->pdev));
 970
 971         devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
 972         if (!devp) {
 973                 pr_err("Cannot allocate ib device\n");
 974                 return ERR_PTR(-ENOMEM);
 975         }
 976         devp->rdev.lldi = *infop;
 977
 978         /* init various hw-queue params based on lld info */
 979         pr_debug("%s: Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
 980                  __func__, devp->rdev.lldi.sge_ingpadboundary,
 981                  devp->rdev.lldi.sge_egrstatuspagesize);
 982
 983         devp->rdev.hw_queue.t4_eq_status_entries =
 984                 devp->rdev.lldi.sge_egrstatuspagesize / 64;
 985         devp->rdev.hw_queue.t4_max_eq_size = 65520;
 986         devp->rdev.hw_queue.t4_max_iq_size = 65520;
 987         devp->rdev.hw_queue.t4_max_rq_size = 8192 -
 988                 devp->rdev.hw_queue.t4_eq_status_entries - 1;
 989         devp->rdev.hw_queue.t4_max_sq_size =
 990                 devp->rdev.hw_queue.t4_max_eq_size -
 991                 devp->rdev.hw_queue.t4_eq_status_entries - 1;
 992         devp->rdev.hw_queue.t4_max_qp_depth =
 993                 devp->rdev.hw_queue.t4_max_rq_size;
 994         devp->rdev.hw_queue.t4_max_cq_depth =
 995                 devp->rdev.hw_queue.t4_max_iq_size - 2;
 996         devp->rdev.hw_queue.t4_stat_len =
 997                 devp->rdev.lldi.sge_egrstatuspagesize;
 998
 999         /*
1000          * For T5/T6 devices, we map all of BAR2 with WC.
1001          * For T4 devices with onchip qp mem, we map only that part
1002          * of BAR2 with WC.
1003          */
1004         devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
1005         if (!is_t4(devp->rdev.lldi.adapter_type)) {
1006                 devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
1007                         pci_resource_len(devp->rdev.lldi.pdev, 2));
1008                 if (!devp->rdev.bar2_kva) {
1009                         pr_err("Unable to ioremap BAR2\n");
1010                         ib_dealloc_device(&devp->ibdev);
1011                         return ERR_PTR(-EINVAL);
1012                 }
1013         } else if (ocqp_supported(infop)) {
1014                 devp->rdev.oc_mw_pa =
1015                         pci_resource_start(devp->rdev.lldi.pdev, 2) +
1016                         pci_resource_len(devp->rdev.lldi.pdev, 2) -
1017                         roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
1018                 devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
1019                         devp->rdev.lldi.vr->ocq.size);
1020                 if (!devp->rdev.oc_mw_kva) {
1021                         pr_err("Unable to ioremap onchip mem\n");
1022                         ib_dealloc_device(&devp->ibdev);
1023                         return ERR_PTR(-EINVAL);
1024                 }
1025         }
1026
1027         pr_debug("ocq memory: hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
1028                  devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
1029                  devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
1030
1031         ret = c4iw_rdev_open(&devp->rdev);
1032         if (ret) {
1033                 pr_err("Unable to open CXIO rdev err %d\n", ret);
1034                 ib_dealloc_device(&devp->ibdev);
1035                 return ERR_PTR(ret);
1036         }
1037
1038         idr_init(&devp->cqidr);
1039         idr_init(&devp->qpidr);
1040         idr_init(&devp->mmidr);
1041         idr_init(&devp->hwtid_idr);
1042         idr_init(&devp->stid_idr);
1043         idr_init(&devp->atid_idr);
1044         spin_lock_init(&devp->lock);
1045         mutex_init(&devp->rdev.stats.lock);
1046         mutex_init(&devp->db_mutex);
1047         INIT_LIST_HEAD(&devp->db_fc_list);
1048         init_waitqueue_head(&devp->wait);
1049         devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1050
1051         if (c4iw_debugfs_root) {
1052                 devp->debugfs_root = debugfs_create_dir(
1053                                         pci_name(devp->rdev.lldi.pdev),
1054                                         c4iw_debugfs_root);
1055                 setup_debugfs(devp);
1056         }
1057
1058
1059         return devp;
1060 }
1061
1062 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1063 {
1064         struct uld_ctx *ctx;
1065         static int vers_printed;
1066         int i;
1067
1068         if (!vers_printed++)
1069                 pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1070                         DRV_VERSION);
1071
1072         ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1073         if (!ctx) {
1074                 ctx = ERR_PTR(-ENOMEM);
1075                 goto out;
1076         }
1077         ctx->lldi = *infop;
1078
1079         pr_debug("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1080                  __func__, pci_name(ctx->lldi.pdev),
1081                  ctx->lldi.nchan, ctx->lldi.nrxq,
1082                  ctx->lldi.ntxq, ctx->lldi.nports);
1083
1084         mutex_lock(&dev_mutex);
1085         list_add_tail(&ctx->entry, &uld_ctx_list);
1086         mutex_unlock(&dev_mutex);
1087
1088         for (i = 0; i < ctx->lldi.nrxq; i++)
1089                 pr_debug("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1090 out:
1091         return ctx;
1092 }
1093
1094 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1095                                                  const __be64 *rsp,
1096                                                  u32 pktshift)
1097 {
1098         struct sk_buff *skb;
1099
1100         /*
1101          * Allocate space for cpl_pass_accept_req which will be synthesized by
1102          * driver. Once the driver synthesizes the request the skb will go
1103          * through the regular cpl_pass_accept_req processing.
1104          * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1105          * cpl_rx_pkt.
1106          */
1107         skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1108                         sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1109         if (unlikely(!skb))
1110                 return NULL;
1111
1112          __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1113                    sizeof(struct rss_header) - pktshift);
1114
1115         /*
1116          * This skb will contain:
1117          *   rss_header from the rspq descriptor (1 flit)
1118          *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1119          *   space for the difference between the size of an
1120          *      rx_pkt and pass_accept_req cpl (1 flit)
1121          *   the packet data from the gl
1122          */
1123         skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1124                                 sizeof(struct rss_header));
1125         skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1126                                        sizeof(struct cpl_pass_accept_req),
1127                                        gl->va + pktshift,
1128                                        gl->tot_len - pktshift);
1129         return skb;
1130 }
1131
1132 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1133                            const __be64 *rsp)
1134 {
1135         unsigned int opcode = *(u8 *)rsp;
1136         struct sk_buff *skb;
1137
1138         if (opcode != CPL_RX_PKT)
1139                 goto out;
1140
1141         skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1142         if (skb == NULL)
1143                 goto out;
1144
1145         if (c4iw_handlers[opcode] == NULL) {
1146                 pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1147                 kfree_skb(skb);
1148                 goto out;
1149         }
1150         c4iw_handlers[opcode](dev, skb);
1151         return 1;
1152 out:
1153         return 0;
1154 }
1155
1156 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1157                         const struct pkt_gl *gl)
1158 {
1159         struct uld_ctx *ctx = handle;
1160         struct c4iw_dev *dev = ctx->dev;
1161         struct sk_buff *skb;
1162         u8 opcode;
1163
1164         if (gl == NULL) {
1165                 /* omit RSS and rsp_ctrl at end of descriptor */
1166                 unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1167
1168                 skb = alloc_skb(256, GFP_ATOMIC);
1169                 if (!skb)
1170                         goto nomem;
1171                 __skb_put(skb, len);
1172                 skb_copy_to_linear_data(skb, &rsp[1], len);
1173         } else if (gl == CXGB4_MSG_AN) {
1174                 const struct rsp_ctrl *rc = (void *)rsp;
1175
1176                 u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1177                 c4iw_ev_handler(dev, qid);
1178                 return 0;
1179         } else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1180                 if (recv_rx_pkt(dev, gl, rsp))
1181                         return 0;
1182
1183                 pr_info("%s: unexpected FL contents at %p, RSS %#llx, FL %#llx, len %u\n",
1184                         pci_name(ctx->lldi.pdev), gl->va,
1185                         be64_to_cpu(*rsp),
1186                         be64_to_cpu(*(__force __be64 *)gl->va),
1187                         gl->tot_len);
1188
1189                 return 0;
1190         } else {
1191                 skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1192                 if (unlikely(!skb))
1193                         goto nomem;
1194         }
1195
1196         opcode = *(u8 *)rsp;
1197         if (c4iw_handlers[opcode]) {
1198                 c4iw_handlers[opcode](dev, skb);
1199         } else {
1200                 pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1201                 kfree_skb(skb);
1202         }
1203
1204         return 0;
1205 nomem:
1206         return -1;
1207 }
1208
1209 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1210 {
1211         struct uld_ctx *ctx = handle;
1212
1213         pr_debug("%s new_state %u\n", __func__, new_state);
1214         switch (new_state) {
1215         case CXGB4_STATE_UP:
1216                 pr_info("%s: Up\n", pci_name(ctx->lldi.pdev));
1217                 if (!ctx->dev) {
1218                         int ret;
1219
1220                         ctx->dev = c4iw_alloc(&ctx->lldi);
1221                         if (IS_ERR(ctx->dev)) {
1222                                 pr_err("%s: initialization failed: %ld\n",
1223                                        pci_name(ctx->lldi.pdev),
1224                                        PTR_ERR(ctx->dev));
1225                                 ctx->dev = NULL;
1226                                 break;
1227                         }
1228                         ret = c4iw_register_device(ctx->dev);
1229                         if (ret) {
1230                                 pr_err("%s: RDMA registration failed: %d\n",
1231                                        pci_name(ctx->lldi.pdev), ret);
1232                                 c4iw_dealloc(ctx);
1233                         }
1234                 }
1235                 break;
1236         case CXGB4_STATE_DOWN:
1237                 pr_info("%s: Down\n", pci_name(ctx->lldi.pdev));
1238                 if (ctx->dev)
1239                         c4iw_remove(ctx);
1240                 break;
1241         case CXGB4_STATE_START_RECOVERY:
1242                 pr_info("%s: Fatal Error\n", pci_name(ctx->lldi.pdev));
1243                 if (ctx->dev) {
1244                         struct ib_event event;
1245
1246                         ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1247                         memset(&event, 0, sizeof event);
1248                         event.event  = IB_EVENT_DEVICE_FATAL;
1249                         event.device = &ctx->dev->ibdev;
1250                         ib_dispatch_event(&event);
1251                         c4iw_remove(ctx);
1252                 }
1253                 break;
1254         case CXGB4_STATE_DETACH:
1255                 pr_info("%s: Detach\n", pci_name(ctx->lldi.pdev));
1256                 if (ctx->dev)
1257                         c4iw_remove(ctx);
1258                 break;
1259         }
1260         return 0;
1261 }
1262
1263 static int disable_qp_db(int id, void *p, void *data)
1264 {
1265         struct c4iw_qp *qp = p;
1266
1267         t4_disable_wq_db(&qp->wq);
1268         return 0;
1269 }
1270
1271 static void stop_queues(struct uld_ctx *ctx)
1272 {
1273         unsigned long flags;
1274
1275         spin_lock_irqsave(&ctx->dev->lock, flags);
1276         ctx->dev->rdev.stats.db_state_transitions++;
1277         ctx->dev->db_state = STOPPED;
1278         if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1279                 idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1280         else
1281                 ctx->dev->rdev.status_page->db_off = 1;
1282         spin_unlock_irqrestore(&ctx->dev->lock, flags);
1283 }
1284
1285 static int enable_qp_db(int id, void *p, void *data)
1286 {
1287         struct c4iw_qp *qp = p;
1288
1289         t4_enable_wq_db(&qp->wq);
1290         return 0;
1291 }
1292
1293 static void resume_rc_qp(struct c4iw_qp *qp)
1294 {
1295         spin_lock(&qp->lock);
1296         t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc, NULL);
1297         qp->wq.sq.wq_pidx_inc = 0;
1298         t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc, NULL);
1299         qp->wq.rq.wq_pidx_inc = 0;
1300         spin_unlock(&qp->lock);
1301 }
1302
1303 static void resume_a_chunk(struct uld_ctx *ctx)
1304 {
1305         int i;
1306         struct c4iw_qp *qp;
1307
1308         for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1309                 qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1310                                       db_fc_entry);
1311                 list_del_init(&qp->db_fc_entry);
1312                 resume_rc_qp(qp);
1313                 if (list_empty(&ctx->dev->db_fc_list))
1314                         break;
1315         }
1316 }
1317
1318 static void resume_queues(struct uld_ctx *ctx)
1319 {
1320         spin_lock_irq(&ctx->dev->lock);
1321         if (ctx->dev->db_state != STOPPED)
1322                 goto out;
1323         ctx->dev->db_state = FLOW_CONTROL;
1324         while (1) {
1325                 if (list_empty(&ctx->dev->db_fc_list)) {
1326                         WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1327                         ctx->dev->db_state = NORMAL;
1328                         ctx->dev->rdev.stats.db_state_transitions++;
1329                         if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1330                                 idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1331                                              NULL);
1332                         } else {
1333                                 ctx->dev->rdev.status_page->db_off = 0;
1334                         }
1335                         break;
1336                 } else {
1337                         if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1338                             < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1339                                DB_FC_DRAIN_THRESH)) {
1340                                 resume_a_chunk(ctx);
1341                         }
1342                         if (!list_empty(&ctx->dev->db_fc_list)) {
1343                                 spin_unlock_irq(&ctx->dev->lock);
1344                                 if (DB_FC_RESUME_DELAY) {
1345                                         set_current_state(TASK_UNINTERRUPTIBLE);
1346                                         schedule_timeout(DB_FC_RESUME_DELAY);
1347                                 }
1348                                 spin_lock_irq(&ctx->dev->lock);
1349                                 if (ctx->dev->db_state != FLOW_CONTROL)
1350                                         break;
1351                         }
1352                 }
1353         }
1354 out:
1355         if (ctx->dev->db_state != NORMAL)
1356                 ctx->dev->rdev.stats.db_fc_interruptions++;
1357         spin_unlock_irq(&ctx->dev->lock);
1358 }
1359
1360 struct qp_list {
1361         unsigned idx;
1362         struct c4iw_qp **qps;
1363 };
1364
1365 static int add_and_ref_qp(int id, void *p, void *data)
1366 {
1367         struct qp_list *qp_listp = data;
1368         struct c4iw_qp *qp = p;
1369
1370         c4iw_qp_add_ref(&qp->ibqp);
1371         qp_listp->qps[qp_listp->idx++] = qp;
1372         return 0;
1373 }
1374
1375 static int count_qps(int id, void *p, void *data)
1376 {
1377         unsigned *countp = data;
1378         (*countp)++;
1379         return 0;
1380 }
1381
1382 static void deref_qps(struct qp_list *qp_list)
1383 {
1384         int idx;
1385
1386         for (idx = 0; idx < qp_list->idx; idx++)
1387                 c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1388 }
1389
1390 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1391 {
1392         int idx;
1393         int ret;
1394
1395         for (idx = 0; idx < qp_list->idx; idx++) {
1396                 struct c4iw_qp *qp = qp_list->qps[idx];
1397
1398                 spin_lock_irq(&qp->rhp->lock);
1399                 spin_lock(&qp->lock);
1400                 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1401                                           qp->wq.sq.qid,
1402                                           t4_sq_host_wq_pidx(&qp->wq),
1403                                           t4_sq_wq_size(&qp->wq));
1404                 if (ret) {
1405                         pr_err("%s: Fatal error - DB overflow recovery failed - error syncing SQ qid %u\n",
1406                                pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1407                         spin_unlock(&qp->lock);
1408                         spin_unlock_irq(&qp->rhp->lock);
1409                         return;
1410                 }
1411                 qp->wq.sq.wq_pidx_inc = 0;
1412
1413                 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1414                                           qp->wq.rq.qid,
1415                                           t4_rq_host_wq_pidx(&qp->wq),
1416                                           t4_rq_wq_size(&qp->wq));
1417
1418                 if (ret) {
1419                         pr_err("%s: Fatal error - DB overflow recovery failed - error syncing RQ qid %u\n",
1420                                pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1421                         spin_unlock(&qp->lock);
1422                         spin_unlock_irq(&qp->rhp->lock);
1423                         return;
1424                 }
1425                 qp->wq.rq.wq_pidx_inc = 0;
1426                 spin_unlock(&qp->lock);
1427                 spin_unlock_irq(&qp->rhp->lock);
1428
1429                 /* Wait for the dbfifo to drain */
1430                 while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1431                         set_current_state(TASK_UNINTERRUPTIBLE);
1432                         schedule_timeout(usecs_to_jiffies(10));
1433                 }
1434         }
1435 }
1436
1437 static void recover_queues(struct uld_ctx *ctx)
1438 {
1439         int count = 0;
1440         struct qp_list qp_list;
1441         int ret;
1442
1443         /* slow everybody down */
1444         set_current_state(TASK_UNINTERRUPTIBLE);
1445         schedule_timeout(usecs_to_jiffies(1000));
1446
1447         /* flush the SGE contexts */
1448         ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1449         if (ret) {
1450                 pr_err("%s: Fatal error - DB overflow recovery failed\n",
1451                        pci_name(ctx->lldi.pdev));
1452                 return;
1453         }
1454
1455         /* Count active queues so we can build a list of queues to recover */
1456         spin_lock_irq(&ctx->dev->lock);
1457         WARN_ON(ctx->dev->db_state != STOPPED);
1458         ctx->dev->db_state = RECOVERY;
1459         idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1460
1461         qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1462         if (!qp_list.qps) {
1463                 spin_unlock_irq(&ctx->dev->lock);
1464                 return;
1465         }
1466         qp_list.idx = 0;
1467
1468         /* add and ref each qp so it doesn't get freed */
1469         idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1470
1471         spin_unlock_irq(&ctx->dev->lock);
1472
1473         /* now traverse the list in a safe context to recover the db state*/
1474         recover_lost_dbs(ctx, &qp_list);
1475
1476         /* we're almost done!  deref the qps and clean up */
1477         deref_qps(&qp_list);
1478         kfree(qp_list.qps);
1479
1480         spin_lock_irq(&ctx->dev->lock);
1481         WARN_ON(ctx->dev->db_state != RECOVERY);
1482         ctx->dev->db_state = STOPPED;
1483         spin_unlock_irq(&ctx->dev->lock);
1484 }
1485
1486 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1487 {
1488         struct uld_ctx *ctx = handle;
1489
1490         switch (control) {
1491         case CXGB4_CONTROL_DB_FULL:
1492                 stop_queues(ctx);
1493                 ctx->dev->rdev.stats.db_full++;
1494                 break;
1495         case CXGB4_CONTROL_DB_EMPTY:
1496                 resume_queues(ctx);
1497                 mutex_lock(&ctx->dev->rdev.stats.lock);
1498                 ctx->dev->rdev.stats.db_empty++;
1499                 mutex_unlock(&ctx->dev->rdev.stats.lock);
1500                 break;
1501         case CXGB4_CONTROL_DB_DROP:
1502                 recover_queues(ctx);
1503                 mutex_lock(&ctx->dev->rdev.stats.lock);
1504                 ctx->dev->rdev.stats.db_drop++;
1505                 mutex_unlock(&ctx->dev->rdev.stats.lock);
1506                 break;
1507         default:
1508                 pr_warn("%s: unknown control cmd %u\n",
1509                         pci_name(ctx->lldi.pdev), control);
1510                 break;
1511         }
1512         return 0;
1513 }
1514
1515 static struct cxgb4_uld_info c4iw_uld_info = {
1516         .name = DRV_NAME,
1517         .nrxq = MAX_ULD_QSETS,
1518         .ntxq = MAX_ULD_QSETS,
1519         .rxq_size = 511,
1520         .ciq = true,
1521         .lro = false,
1522         .add = c4iw_uld_add,
1523         .rx_handler = c4iw_uld_rx_handler,
1524         .state_change = c4iw_uld_state_change,
1525         .control = c4iw_uld_control,
1526 };
1527
1528 static int __init c4iw_init_module(void)
1529 {
1530         int err;
1531
1532         err = c4iw_cm_init();
1533         if (err)
1534                 return err;
1535
1536         c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1537         if (!c4iw_debugfs_root)
1538                 pr_warn("could not create debugfs entry, continuing\n");
1539
1540         cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1541
1542         return 0;
1543 }
1544
1545 static void __exit c4iw_exit_module(void)
1546 {
1547         struct uld_ctx *ctx, *tmp;
1548
1549         mutex_lock(&dev_mutex);
1550         list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1551                 if (ctx->dev)
1552                         c4iw_remove(ctx);
1553                 kfree(ctx);
1554         }
1555         mutex_unlock(&dev_mutex);
1556         cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1557         c4iw_cm_term();
1558         debugfs_remove_recursive(c4iw_debugfs_root);
1559 }
1560
1561 module_init(c4iw_init_module);
1562 module_exit(c4iw_exit_module);