Merge tag 'for-linus-unmerged' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma...
[muen/linux.git] / 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 static LIST_HEAD(uld_ctx_list);
68 static DEFINE_MUTEX(dev_mutex);
69 static struct workqueue_struct *reg_workq;
70
71 #define DB_FC_RESUME_SIZE 64
72 #define DB_FC_RESUME_DELAY 1
73 #define DB_FC_DRAIN_THRESH 0
74
75 static struct dentry *c4iw_debugfs_root;
76
77 struct c4iw_debugfs_data {
78         struct c4iw_dev *devp;
79         char *buf;
80         int bufsize;
81         int pos;
82 };
83
84 static int count_idrs(int id, void *p, void *data)
85 {
86         int *countp = data;
87
88         *countp = *countp + 1;
89         return 0;
90 }
91
92 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
93                             loff_t *ppos)
94 {
95         struct c4iw_debugfs_data *d = file->private_data;
96
97         return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
98 }
99
100 void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
101 {
102         struct wr_log_entry le;
103         int idx;
104
105         if (!wq->rdev->wr_log)
106                 return;
107
108         idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
109                 (wq->rdev->wr_log_size - 1);
110         le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
111         le.poll_host_time = ktime_get();
112         le.valid = 1;
113         le.cqe_sge_ts = CQE_TS(cqe);
114         if (SQ_TYPE(cqe)) {
115                 le.qid = wq->sq.qid;
116                 le.opcode = CQE_OPCODE(cqe);
117                 le.post_host_time = wq->sq.sw_sq[wq->sq.cidx].host_time;
118                 le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
119                 le.wr_id = CQE_WRID_SQ_IDX(cqe);
120         } else {
121                 le.qid = wq->rq.qid;
122                 le.opcode = FW_RI_RECEIVE;
123                 le.post_host_time = wq->rq.sw_rq[wq->rq.cidx].host_time;
124                 le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
125                 le.wr_id = CQE_WRID_MSN(cqe);
126         }
127         wq->rdev->wr_log[idx] = le;
128 }
129
130 static int wr_log_show(struct seq_file *seq, void *v)
131 {
132         struct c4iw_dev *dev = seq->private;
133         ktime_t prev_time;
134         struct wr_log_entry *lep;
135         int prev_time_set = 0;
136         int idx, end;
137
138 #define ts2ns(ts) div64_u64((ts) * dev->rdev.lldi.cclk_ps, 1000)
139
140         idx = atomic_read(&dev->rdev.wr_log_idx) &
141                 (dev->rdev.wr_log_size - 1);
142         end = idx - 1;
143         if (end < 0)
144                 end = dev->rdev.wr_log_size - 1;
145         lep = &dev->rdev.wr_log[idx];
146         while (idx != end) {
147                 if (lep->valid) {
148                         if (!prev_time_set) {
149                                 prev_time_set = 1;
150                                 prev_time = lep->poll_host_time;
151                         }
152                         seq_printf(seq, "%04u: nsec %llu qid %u opcode "
153                                    "%u %s 0x%x host_wr_delta nsec %llu "
154                                    "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
155                                    "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
156                                    "cqe_poll_delta_ns %llu\n",
157                                    idx,
158                                    ktime_to_ns(ktime_sub(lep->poll_host_time,
159                                                          prev_time)),
160                                    lep->qid, lep->opcode,
161                                    lep->opcode == FW_RI_RECEIVE ?
162                                                         "msn" : "wrid",
163                                    lep->wr_id,
164                                    ktime_to_ns(ktime_sub(lep->poll_host_time,
165                                                          lep->post_host_time)),
166                                    lep->post_sge_ts, lep->cqe_sge_ts,
167                                    lep->poll_sge_ts,
168                                    ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
169                                    ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
170                         prev_time = lep->poll_host_time;
171                 }
172                 idx++;
173                 if (idx > (dev->rdev.wr_log_size - 1))
174                         idx = 0;
175                 lep = &dev->rdev.wr_log[idx];
176         }
177 #undef ts2ns
178         return 0;
179 }
180
181 static int wr_log_open(struct inode *inode, struct file *file)
182 {
183         return single_open(file, wr_log_show, inode->i_private);
184 }
185
186 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
187                             size_t count, loff_t *pos)
188 {
189         struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
190         int i;
191
192         if (dev->rdev.wr_log)
193                 for (i = 0; i < dev->rdev.wr_log_size; i++)
194                         dev->rdev.wr_log[i].valid = 0;
195         return count;
196 }
197
198 static const struct file_operations wr_log_debugfs_fops = {
199         .owner   = THIS_MODULE,
200         .open    = wr_log_open,
201         .release = single_release,
202         .read    = seq_read,
203         .llseek  = seq_lseek,
204         .write   = wr_log_clear,
205 };
206
207 static struct sockaddr_in zero_sin = {
208         .sin_family = AF_INET,
209 };
210
211 static struct sockaddr_in6 zero_sin6 = {
212         .sin6_family = AF_INET6,
213 };
214
215 static void set_ep_sin_addrs(struct c4iw_ep *ep,
216                              struct sockaddr_in **lsin,
217                              struct sockaddr_in **rsin,
218                              struct sockaddr_in **m_lsin,
219                              struct sockaddr_in **m_rsin)
220 {
221         struct iw_cm_id *id = ep->com.cm_id;
222
223         *m_lsin = (struct sockaddr_in *)&ep->com.local_addr;
224         *m_rsin = (struct sockaddr_in *)&ep->com.remote_addr;
225         if (id) {
226                 *lsin = (struct sockaddr_in *)&id->local_addr;
227                 *rsin = (struct sockaddr_in *)&id->remote_addr;
228         } else {
229                 *lsin = &zero_sin;
230                 *rsin = &zero_sin;
231         }
232 }
233
234 static void set_ep_sin6_addrs(struct c4iw_ep *ep,
235                               struct sockaddr_in6 **lsin6,
236                               struct sockaddr_in6 **rsin6,
237                               struct sockaddr_in6 **m_lsin6,
238                               struct sockaddr_in6 **m_rsin6)
239 {
240         struct iw_cm_id *id = ep->com.cm_id;
241
242         *m_lsin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
243         *m_rsin6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
244         if (id) {
245                 *lsin6 = (struct sockaddr_in6 *)&id->local_addr;
246                 *rsin6 = (struct sockaddr_in6 *)&id->remote_addr;
247         } else {
248                 *lsin6 = &zero_sin6;
249                 *rsin6 = &zero_sin6;
250         }
251 }
252
253 static int dump_qp(int id, void *p, void *data)
254 {
255         struct c4iw_qp *qp = p;
256         struct c4iw_debugfs_data *qpd = data;
257         int space;
258         int cc;
259
260         if (id != qp->wq.sq.qid)
261                 return 0;
262
263         space = qpd->bufsize - qpd->pos - 1;
264         if (space == 0)
265                 return 1;
266
267         if (qp->ep) {
268                 struct c4iw_ep *ep = qp->ep;
269
270                 if (ep->com.local_addr.ss_family == AF_INET) {
271                         struct sockaddr_in *lsin;
272                         struct sockaddr_in *rsin;
273                         struct sockaddr_in *m_lsin;
274                         struct sockaddr_in *m_rsin;
275
276                         set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
277                         cc = snprintf(qpd->buf + qpd->pos, space,
278                                       "rc qp sq id %u rq id %u state %u "
279                                       "onchip %u ep tid %u state %u "
280                                       "%pI4:%u/%u->%pI4:%u/%u\n",
281                                       qp->wq.sq.qid, qp->wq.rq.qid,
282                                       (int)qp->attr.state,
283                                       qp->wq.sq.flags & T4_SQ_ONCHIP,
284                                       ep->hwtid, (int)ep->com.state,
285                                       &lsin->sin_addr, ntohs(lsin->sin_port),
286                                       ntohs(m_lsin->sin_port),
287                                       &rsin->sin_addr, ntohs(rsin->sin_port),
288                                       ntohs(m_rsin->sin_port));
289                 } else {
290                         struct sockaddr_in6 *lsin6;
291                         struct sockaddr_in6 *rsin6;
292                         struct sockaddr_in6 *m_lsin6;
293                         struct sockaddr_in6 *m_rsin6;
294
295                         set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6,
296                                           &m_rsin6);
297                         cc = snprintf(qpd->buf + qpd->pos, space,
298                                       "rc qp sq id %u rq id %u state %u "
299                                       "onchip %u ep tid %u state %u "
300                                       "%pI6:%u/%u->%pI6:%u/%u\n",
301                                       qp->wq.sq.qid, qp->wq.rq.qid,
302                                       (int)qp->attr.state,
303                                       qp->wq.sq.flags & T4_SQ_ONCHIP,
304                                       ep->hwtid, (int)ep->com.state,
305                                       &lsin6->sin6_addr,
306                                       ntohs(lsin6->sin6_port),
307                                       ntohs(m_lsin6->sin6_port),
308                                       &rsin6->sin6_addr,
309                                       ntohs(rsin6->sin6_port),
310                                       ntohs(m_rsin6->sin6_port));
311                 }
312         } else
313                 cc = snprintf(qpd->buf + qpd->pos, space,
314                              "qp sq id %u rq id %u state %u onchip %u\n",
315                               qp->wq.sq.qid, qp->wq.rq.qid,
316                               (int)qp->attr.state,
317                               qp->wq.sq.flags & T4_SQ_ONCHIP);
318         if (cc < space)
319                 qpd->pos += cc;
320         return 0;
321 }
322
323 static int qp_release(struct inode *inode, struct file *file)
324 {
325         struct c4iw_debugfs_data *qpd = file->private_data;
326         if (!qpd) {
327                 pr_info("%s null qpd?\n", __func__);
328                 return 0;
329         }
330         vfree(qpd->buf);
331         kfree(qpd);
332         return 0;
333 }
334
335 static int qp_open(struct inode *inode, struct file *file)
336 {
337         struct c4iw_debugfs_data *qpd;
338         int count = 1;
339
340         qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
341         if (!qpd)
342                 return -ENOMEM;
343
344         qpd->devp = inode->i_private;
345         qpd->pos = 0;
346
347         spin_lock_irq(&qpd->devp->lock);
348         idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
349         spin_unlock_irq(&qpd->devp->lock);
350
351         qpd->bufsize = count * 180;
352         qpd->buf = vmalloc(qpd->bufsize);
353         if (!qpd->buf) {
354                 kfree(qpd);
355                 return -ENOMEM;
356         }
357
358         spin_lock_irq(&qpd->devp->lock);
359         idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
360         spin_unlock_irq(&qpd->devp->lock);
361
362         qpd->buf[qpd->pos++] = 0;
363         file->private_data = qpd;
364         return 0;
365 }
366
367 static const struct file_operations qp_debugfs_fops = {
368         .owner   = THIS_MODULE,
369         .open    = qp_open,
370         .release = qp_release,
371         .read    = debugfs_read,
372         .llseek  = default_llseek,
373 };
374
375 static int dump_stag(int id, void *p, void *data)
376 {
377         struct c4iw_debugfs_data *stagd = data;
378         int space;
379         int cc;
380         struct fw_ri_tpte tpte;
381         int ret;
382
383         space = stagd->bufsize - stagd->pos - 1;
384         if (space == 0)
385                 return 1;
386
387         ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
388                               (__be32 *)&tpte);
389         if (ret) {
390                 dev_err(&stagd->devp->rdev.lldi.pdev->dev,
391                         "%s cxgb4_read_tpte err %d\n", __func__, ret);
392                 return ret;
393         }
394         cc = snprintf(stagd->buf + stagd->pos, space,
395                       "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
396                       "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
397                       (u32)id<<8,
398                       FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
399                       FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
400                       FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
401                       FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
402                       FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
403                       FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
404                       ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
405                       ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
406         if (cc < space)
407                 stagd->pos += cc;
408         return 0;
409 }
410
411 static int stag_release(struct inode *inode, struct file *file)
412 {
413         struct c4iw_debugfs_data *stagd = file->private_data;
414         if (!stagd) {
415                 pr_info("%s null stagd?\n", __func__);
416                 return 0;
417         }
418         vfree(stagd->buf);
419         kfree(stagd);
420         return 0;
421 }
422
423 static int stag_open(struct inode *inode, struct file *file)
424 {
425         struct c4iw_debugfs_data *stagd;
426         int ret = 0;
427         int count = 1;
428
429         stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
430         if (!stagd) {
431                 ret = -ENOMEM;
432                 goto out;
433         }
434         stagd->devp = inode->i_private;
435         stagd->pos = 0;
436
437         spin_lock_irq(&stagd->devp->lock);
438         idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
439         spin_unlock_irq(&stagd->devp->lock);
440
441         stagd->bufsize = count * 256;
442         stagd->buf = vmalloc(stagd->bufsize);
443         if (!stagd->buf) {
444                 ret = -ENOMEM;
445                 goto err1;
446         }
447
448         spin_lock_irq(&stagd->devp->lock);
449         idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
450         spin_unlock_irq(&stagd->devp->lock);
451
452         stagd->buf[stagd->pos++] = 0;
453         file->private_data = stagd;
454         goto out;
455 err1:
456         kfree(stagd);
457 out:
458         return ret;
459 }
460
461 static const struct file_operations stag_debugfs_fops = {
462         .owner   = THIS_MODULE,
463         .open    = stag_open,
464         .release = stag_release,
465         .read    = debugfs_read,
466         .llseek  = default_llseek,
467 };
468
469 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
470
471 static int stats_show(struct seq_file *seq, void *v)
472 {
473         struct c4iw_dev *dev = seq->private;
474
475         seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
476                    "Max", "Fail");
477         seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
478                         dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
479                         dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
480         seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
481                         dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
482                         dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
483         seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
484                         dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
485                         dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
486         seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
487                         dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
488                         dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
489         seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
490                         dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
491                         dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
492         seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
493                         dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
494                         dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
495         seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
496         seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
497         seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
498         seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
499                    db_state_str[dev->db_state],
500                    dev->rdev.stats.db_state_transitions,
501                    dev->rdev.stats.db_fc_interruptions);
502         seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
503         seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
504                    dev->rdev.stats.act_ofld_conn_fails);
505         seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
506                    dev->rdev.stats.pas_ofld_conn_fails);
507         seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
508         seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
509         return 0;
510 }
511
512 static int stats_open(struct inode *inode, struct file *file)
513 {
514         return single_open(file, stats_show, inode->i_private);
515 }
516
517 static ssize_t stats_clear(struct file *file, const char __user *buf,
518                 size_t count, loff_t *pos)
519 {
520         struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
521
522         mutex_lock(&dev->rdev.stats.lock);
523         dev->rdev.stats.pd.max = 0;
524         dev->rdev.stats.pd.fail = 0;
525         dev->rdev.stats.qid.max = 0;
526         dev->rdev.stats.qid.fail = 0;
527         dev->rdev.stats.stag.max = 0;
528         dev->rdev.stats.stag.fail = 0;
529         dev->rdev.stats.pbl.max = 0;
530         dev->rdev.stats.pbl.fail = 0;
531         dev->rdev.stats.rqt.max = 0;
532         dev->rdev.stats.rqt.fail = 0;
533         dev->rdev.stats.ocqp.max = 0;
534         dev->rdev.stats.ocqp.fail = 0;
535         dev->rdev.stats.db_full = 0;
536         dev->rdev.stats.db_empty = 0;
537         dev->rdev.stats.db_drop = 0;
538         dev->rdev.stats.db_state_transitions = 0;
539         dev->rdev.stats.tcam_full = 0;
540         dev->rdev.stats.act_ofld_conn_fails = 0;
541         dev->rdev.stats.pas_ofld_conn_fails = 0;
542         mutex_unlock(&dev->rdev.stats.lock);
543         return count;
544 }
545
546 static const struct file_operations stats_debugfs_fops = {
547         .owner   = THIS_MODULE,
548         .open    = stats_open,
549         .release = single_release,
550         .read    = seq_read,
551         .llseek  = seq_lseek,
552         .write   = stats_clear,
553 };
554
555 static int dump_ep(int id, void *p, void *data)
556 {
557         struct c4iw_ep *ep = p;
558         struct c4iw_debugfs_data *epd = data;
559         int space;
560         int cc;
561
562         space = epd->bufsize - epd->pos - 1;
563         if (space == 0)
564                 return 1;
565
566         if (ep->com.local_addr.ss_family == AF_INET) {
567                 struct sockaddr_in *lsin;
568                 struct sockaddr_in *rsin;
569                 struct sockaddr_in *m_lsin;
570                 struct sockaddr_in *m_rsin;
571
572                 set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
573                 cc = snprintf(epd->buf + epd->pos, space,
574                               "ep %p cm_id %p qp %p state %d flags 0x%lx "
575                               "history 0x%lx hwtid %d atid %d "
576                               "conn_na %u abort_na %u "
577                               "%pI4:%d/%d <-> %pI4:%d/%d\n",
578                               ep, ep->com.cm_id, ep->com.qp,
579                               (int)ep->com.state, ep->com.flags,
580                               ep->com.history, ep->hwtid, ep->atid,
581                               ep->stats.connect_neg_adv,
582                               ep->stats.abort_neg_adv,
583                               &lsin->sin_addr, ntohs(lsin->sin_port),
584                               ntohs(m_lsin->sin_port),
585                               &rsin->sin_addr, ntohs(rsin->sin_port),
586                               ntohs(m_rsin->sin_port));
587         } else {
588                 struct sockaddr_in6 *lsin6;
589                 struct sockaddr_in6 *rsin6;
590                 struct sockaddr_in6 *m_lsin6;
591                 struct sockaddr_in6 *m_rsin6;
592
593                 set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6, &m_rsin6);
594                 cc = snprintf(epd->buf + epd->pos, space,
595                               "ep %p cm_id %p qp %p state %d flags 0x%lx "
596                               "history 0x%lx hwtid %d atid %d "
597                               "conn_na %u abort_na %u "
598                               "%pI6:%d/%d <-> %pI6:%d/%d\n",
599                               ep, ep->com.cm_id, ep->com.qp,
600                               (int)ep->com.state, ep->com.flags,
601                               ep->com.history, ep->hwtid, ep->atid,
602                               ep->stats.connect_neg_adv,
603                               ep->stats.abort_neg_adv,
604                               &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
605                               ntohs(m_lsin6->sin6_port),
606                               &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
607                               ntohs(m_rsin6->sin6_port));
608         }
609         if (cc < space)
610                 epd->pos += cc;
611         return 0;
612 }
613
614 static int dump_listen_ep(int id, void *p, void *data)
615 {
616         struct c4iw_listen_ep *ep = p;
617         struct c4iw_debugfs_data *epd = data;
618         int space;
619         int cc;
620
621         space = epd->bufsize - epd->pos - 1;
622         if (space == 0)
623                 return 1;
624
625         if (ep->com.local_addr.ss_family == AF_INET) {
626                 struct sockaddr_in *lsin = (struct sockaddr_in *)
627                         &ep->com.cm_id->local_addr;
628                 struct sockaddr_in *m_lsin = (struct sockaddr_in *)
629                         &ep->com.cm_id->m_local_addr;
630
631                 cc = snprintf(epd->buf + epd->pos, space,
632                               "ep %p cm_id %p state %d flags 0x%lx stid %d "
633                               "backlog %d %pI4:%d/%d\n",
634                               ep, ep->com.cm_id, (int)ep->com.state,
635                               ep->com.flags, ep->stid, ep->backlog,
636                               &lsin->sin_addr, ntohs(lsin->sin_port),
637                               ntohs(m_lsin->sin_port));
638         } else {
639                 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
640                         &ep->com.cm_id->local_addr;
641                 struct sockaddr_in6 *m_lsin6 = (struct sockaddr_in6 *)
642                         &ep->com.cm_id->m_local_addr;
643
644                 cc = snprintf(epd->buf + epd->pos, space,
645                               "ep %p cm_id %p state %d flags 0x%lx stid %d "
646                               "backlog %d %pI6:%d/%d\n",
647                               ep, ep->com.cm_id, (int)ep->com.state,
648                               ep->com.flags, ep->stid, ep->backlog,
649                               &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
650                               ntohs(m_lsin6->sin6_port));
651         }
652         if (cc < space)
653                 epd->pos += cc;
654         return 0;
655 }
656
657 static int ep_release(struct inode *inode, struct file *file)
658 {
659         struct c4iw_debugfs_data *epd = file->private_data;
660         if (!epd) {
661                 pr_info("%s null qpd?\n", __func__);
662                 return 0;
663         }
664         vfree(epd->buf);
665         kfree(epd);
666         return 0;
667 }
668
669 static int ep_open(struct inode *inode, struct file *file)
670 {
671         struct c4iw_debugfs_data *epd;
672         int ret = 0;
673         int count = 1;
674
675         epd = kmalloc(sizeof(*epd), GFP_KERNEL);
676         if (!epd) {
677                 ret = -ENOMEM;
678                 goto out;
679         }
680         epd->devp = inode->i_private;
681         epd->pos = 0;
682
683         spin_lock_irq(&epd->devp->lock);
684         idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
685         idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
686         idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
687         spin_unlock_irq(&epd->devp->lock);
688
689         epd->bufsize = count * 240;
690         epd->buf = vmalloc(epd->bufsize);
691         if (!epd->buf) {
692                 ret = -ENOMEM;
693                 goto err1;
694         }
695
696         spin_lock_irq(&epd->devp->lock);
697         idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
698         idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
699         idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
700         spin_unlock_irq(&epd->devp->lock);
701
702         file->private_data = epd;
703         goto out;
704 err1:
705         kfree(epd);
706 out:
707         return ret;
708 }
709
710 static const struct file_operations ep_debugfs_fops = {
711         .owner   = THIS_MODULE,
712         .open    = ep_open,
713         .release = ep_release,
714         .read    = debugfs_read,
715 };
716
717 static int setup_debugfs(struct c4iw_dev *devp)
718 {
719         if (!devp->debugfs_root)
720                 return -1;
721
722         debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
723                                  (void *)devp, &qp_debugfs_fops, 4096);
724
725         debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
726                                  (void *)devp, &stag_debugfs_fops, 4096);
727
728         debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
729                                  (void *)devp, &stats_debugfs_fops, 4096);
730
731         debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
732                                  (void *)devp, &ep_debugfs_fops, 4096);
733
734         if (c4iw_wr_log)
735                 debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
736                                          (void *)devp, &wr_log_debugfs_fops, 4096);
737         return 0;
738 }
739
740 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
741                                struct c4iw_dev_ucontext *uctx)
742 {
743         struct list_head *pos, *nxt;
744         struct c4iw_qid_list *entry;
745
746         mutex_lock(&uctx->lock);
747         list_for_each_safe(pos, nxt, &uctx->qpids) {
748                 entry = list_entry(pos, struct c4iw_qid_list, entry);
749                 list_del_init(&entry->entry);
750                 if (!(entry->qid & rdev->qpmask)) {
751                         c4iw_put_resource(&rdev->resource.qid_table,
752                                           entry->qid);
753                         mutex_lock(&rdev->stats.lock);
754                         rdev->stats.qid.cur -= rdev->qpmask + 1;
755                         mutex_unlock(&rdev->stats.lock);
756                 }
757                 kfree(entry);
758         }
759
760         list_for_each_safe(pos, nxt, &uctx->cqids) {
761                 entry = list_entry(pos, struct c4iw_qid_list, entry);
762                 list_del_init(&entry->entry);
763                 kfree(entry);
764         }
765         mutex_unlock(&uctx->lock);
766 }
767
768 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
769                             struct c4iw_dev_ucontext *uctx)
770 {
771         INIT_LIST_HEAD(&uctx->qpids);
772         INIT_LIST_HEAD(&uctx->cqids);
773         mutex_init(&uctx->lock);
774 }
775
776 /* Caller takes care of locking if needed */
777 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
778 {
779         int err;
780
781         c4iw_init_dev_ucontext(rdev, &rdev->uctx);
782
783         /*
784          * This implementation assumes udb_density == ucq_density!  Eventually
785          * we might need to support this but for now fail the open. Also the
786          * cqid and qpid range must match for now.
787          */
788         if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
789                 pr_err("%s: unsupported udb/ucq densities %u/%u\n",
790                        pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
791                        rdev->lldi.ucq_density);
792                 return -EINVAL;
793         }
794         if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
795             rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
796                 pr_err("%s: unsupported qp and cq id ranges qp start %u size %u cq start %u size %u\n",
797                        pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
798                        rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
799                        rdev->lldi.vr->cq.size);
800                 return -EINVAL;
801         }
802
803         rdev->qpmask = rdev->lldi.udb_density - 1;
804         rdev->cqmask = rdev->lldi.ucq_density - 1;
805         pr_debug("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",
806                  pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
807                  rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
808                  rdev->lldi.vr->pbl.start,
809                  rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
810                  rdev->lldi.vr->rq.size,
811                  rdev->lldi.vr->qp.start,
812                  rdev->lldi.vr->qp.size,
813                  rdev->lldi.vr->cq.start,
814                  rdev->lldi.vr->cq.size);
815         pr_debug("udb %pR db_reg %p gts_reg %p qpmask 0x%x cqmask 0x%x\n",
816                  &rdev->lldi.pdev->resource[2],
817                  rdev->lldi.db_reg, rdev->lldi.gts_reg,
818                  rdev->qpmask, rdev->cqmask);
819
820         if (c4iw_num_stags(rdev) == 0)
821                 return -EINVAL;
822
823         rdev->stats.pd.total = T4_MAX_NUM_PD;
824         rdev->stats.stag.total = rdev->lldi.vr->stag.size;
825         rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
826         rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
827         rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
828         rdev->stats.qid.total = rdev->lldi.vr->qp.size;
829
830         err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
831         if (err) {
832                 pr_err("error %d initializing resources\n", err);
833                 return err;
834         }
835         err = c4iw_pblpool_create(rdev);
836         if (err) {
837                 pr_err("error %d initializing pbl pool\n", err);
838                 goto destroy_resource;
839         }
840         err = c4iw_rqtpool_create(rdev);
841         if (err) {
842                 pr_err("error %d initializing rqt pool\n", err);
843                 goto destroy_pblpool;
844         }
845         err = c4iw_ocqp_pool_create(rdev);
846         if (err) {
847                 pr_err("error %d initializing ocqp pool\n", err);
848                 goto destroy_rqtpool;
849         }
850         rdev->status_page = (struct t4_dev_status_page *)
851                             __get_free_page(GFP_KERNEL);
852         if (!rdev->status_page) {
853                 err = -ENOMEM;
854                 goto destroy_ocqp_pool;
855         }
856         rdev->status_page->qp_start = rdev->lldi.vr->qp.start;
857         rdev->status_page->qp_size = rdev->lldi.vr->qp.size;
858         rdev->status_page->cq_start = rdev->lldi.vr->cq.start;
859         rdev->status_page->cq_size = rdev->lldi.vr->cq.size;
860
861         if (c4iw_wr_log) {
862                 rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
863                                        sizeof(*rdev->wr_log), GFP_KERNEL);
864                 if (rdev->wr_log) {
865                         rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
866                         atomic_set(&rdev->wr_log_idx, 0);
867                 }
868         }
869
870         rdev->free_workq = create_singlethread_workqueue("iw_cxgb4_free");
871         if (!rdev->free_workq) {
872                 err = -ENOMEM;
873                 goto err_free_status_page_and_wr_log;
874         }
875
876         rdev->status_page->db_off = 0;
877
878         return 0;
879 err_free_status_page_and_wr_log:
880         if (c4iw_wr_log && rdev->wr_log)
881                 kfree(rdev->wr_log);
882         free_page((unsigned long)rdev->status_page);
883 destroy_ocqp_pool:
884         c4iw_ocqp_pool_destroy(rdev);
885 destroy_rqtpool:
886         c4iw_rqtpool_destroy(rdev);
887 destroy_pblpool:
888         c4iw_pblpool_destroy(rdev);
889 destroy_resource:
890         c4iw_destroy_resource(&rdev->resource);
891         return err;
892 }
893
894 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
895 {
896         destroy_workqueue(rdev->free_workq);
897         kfree(rdev->wr_log);
898         c4iw_release_dev_ucontext(rdev, &rdev->uctx);
899         free_page((unsigned long)rdev->status_page);
900         c4iw_pblpool_destroy(rdev);
901         c4iw_rqtpool_destroy(rdev);
902         c4iw_ocqp_pool_destroy(rdev);
903         c4iw_destroy_resource(&rdev->resource);
904 }
905
906 void c4iw_dealloc(struct uld_ctx *ctx)
907 {
908         c4iw_rdev_close(&ctx->dev->rdev);
909         WARN_ON_ONCE(!idr_is_empty(&ctx->dev->cqidr));
910         idr_destroy(&ctx->dev->cqidr);
911         WARN_ON_ONCE(!idr_is_empty(&ctx->dev->qpidr));
912         idr_destroy(&ctx->dev->qpidr);
913         WARN_ON_ONCE(!idr_is_empty(&ctx->dev->mmidr));
914         idr_destroy(&ctx->dev->mmidr);
915         wait_event(ctx->dev->wait, idr_is_empty(&ctx->dev->hwtid_idr));
916         idr_destroy(&ctx->dev->hwtid_idr);
917         idr_destroy(&ctx->dev->stid_idr);
918         idr_destroy(&ctx->dev->atid_idr);
919         if (ctx->dev->rdev.bar2_kva)
920                 iounmap(ctx->dev->rdev.bar2_kva);
921         if (ctx->dev->rdev.oc_mw_kva)
922                 iounmap(ctx->dev->rdev.oc_mw_kva);
923         ib_dealloc_device(&ctx->dev->ibdev);
924         ctx->dev = NULL;
925 }
926
927 static void c4iw_remove(struct uld_ctx *ctx)
928 {
929         pr_debug("c4iw_dev %p\n", ctx->dev);
930         c4iw_unregister_device(ctx->dev);
931         c4iw_dealloc(ctx);
932 }
933
934 static int rdma_supported(const struct cxgb4_lld_info *infop)
935 {
936         return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
937                infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
938                infop->vr->cq.size > 0;
939 }
940
941 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
942 {
943         struct c4iw_dev *devp;
944         int ret;
945
946         if (!rdma_supported(infop)) {
947                 pr_info("%s: RDMA not supported on this device\n",
948                         pci_name(infop->pdev));
949                 return ERR_PTR(-ENOSYS);
950         }
951         if (!ocqp_supported(infop))
952                 pr_info("%s: On-Chip Queues not supported on this device\n",
953                         pci_name(infop->pdev));
954
955         devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
956         if (!devp) {
957                 pr_err("Cannot allocate ib device\n");
958                 return ERR_PTR(-ENOMEM);
959         }
960         devp->rdev.lldi = *infop;
961
962         /* init various hw-queue params based on lld info */
963         pr_debug("Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
964                  devp->rdev.lldi.sge_ingpadboundary,
965                  devp->rdev.lldi.sge_egrstatuspagesize);
966
967         devp->rdev.hw_queue.t4_eq_status_entries =
968                 devp->rdev.lldi.sge_egrstatuspagesize / 64;
969         devp->rdev.hw_queue.t4_max_eq_size = 65520;
970         devp->rdev.hw_queue.t4_max_iq_size = 65520;
971         devp->rdev.hw_queue.t4_max_rq_size = 8192 -
972                 devp->rdev.hw_queue.t4_eq_status_entries - 1;
973         devp->rdev.hw_queue.t4_max_sq_size =
974                 devp->rdev.hw_queue.t4_max_eq_size -
975                 devp->rdev.hw_queue.t4_eq_status_entries - 1;
976         devp->rdev.hw_queue.t4_max_qp_depth =
977                 devp->rdev.hw_queue.t4_max_rq_size;
978         devp->rdev.hw_queue.t4_max_cq_depth =
979                 devp->rdev.hw_queue.t4_max_iq_size - 2;
980         devp->rdev.hw_queue.t4_stat_len =
981                 devp->rdev.lldi.sge_egrstatuspagesize;
982
983         /*
984          * For T5/T6 devices, we map all of BAR2 with WC.
985          * For T4 devices with onchip qp mem, we map only that part
986          * of BAR2 with WC.
987          */
988         devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
989         if (!is_t4(devp->rdev.lldi.adapter_type)) {
990                 devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
991                         pci_resource_len(devp->rdev.lldi.pdev, 2));
992                 if (!devp->rdev.bar2_kva) {
993                         pr_err("Unable to ioremap BAR2\n");
994                         ib_dealloc_device(&devp->ibdev);
995                         return ERR_PTR(-EINVAL);
996                 }
997         } else if (ocqp_supported(infop)) {
998                 devp->rdev.oc_mw_pa =
999                         pci_resource_start(devp->rdev.lldi.pdev, 2) +
1000                         pci_resource_len(devp->rdev.lldi.pdev, 2) -
1001                         roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
1002                 devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
1003                         devp->rdev.lldi.vr->ocq.size);
1004                 if (!devp->rdev.oc_mw_kva) {
1005                         pr_err("Unable to ioremap onchip mem\n");
1006                         ib_dealloc_device(&devp->ibdev);
1007                         return ERR_PTR(-EINVAL);
1008                 }
1009         }
1010
1011         pr_debug("ocq memory: hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
1012                  devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
1013                  devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
1014
1015         ret = c4iw_rdev_open(&devp->rdev);
1016         if (ret) {
1017                 pr_err("Unable to open CXIO rdev err %d\n", ret);
1018                 ib_dealloc_device(&devp->ibdev);
1019                 return ERR_PTR(ret);
1020         }
1021
1022         idr_init(&devp->cqidr);
1023         idr_init(&devp->qpidr);
1024         idr_init(&devp->mmidr);
1025         idr_init(&devp->hwtid_idr);
1026         idr_init(&devp->stid_idr);
1027         idr_init(&devp->atid_idr);
1028         spin_lock_init(&devp->lock);
1029         mutex_init(&devp->rdev.stats.lock);
1030         mutex_init(&devp->db_mutex);
1031         INIT_LIST_HEAD(&devp->db_fc_list);
1032         init_waitqueue_head(&devp->wait);
1033         devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1034
1035         if (c4iw_debugfs_root) {
1036                 devp->debugfs_root = debugfs_create_dir(
1037                                         pci_name(devp->rdev.lldi.pdev),
1038                                         c4iw_debugfs_root);
1039                 setup_debugfs(devp);
1040         }
1041
1042
1043         return devp;
1044 }
1045
1046 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1047 {
1048         struct uld_ctx *ctx;
1049         static int vers_printed;
1050         int i;
1051
1052         if (!vers_printed++)
1053                 pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1054                         DRV_VERSION);
1055
1056         ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1057         if (!ctx) {
1058                 ctx = ERR_PTR(-ENOMEM);
1059                 goto out;
1060         }
1061         ctx->lldi = *infop;
1062
1063         pr_debug("found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1064                  pci_name(ctx->lldi.pdev),
1065                  ctx->lldi.nchan, ctx->lldi.nrxq,
1066                  ctx->lldi.ntxq, ctx->lldi.nports);
1067
1068         mutex_lock(&dev_mutex);
1069         list_add_tail(&ctx->entry, &uld_ctx_list);
1070         mutex_unlock(&dev_mutex);
1071
1072         for (i = 0; i < ctx->lldi.nrxq; i++)
1073                 pr_debug("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1074 out:
1075         return ctx;
1076 }
1077
1078 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1079                                                  const __be64 *rsp,
1080                                                  u32 pktshift)
1081 {
1082         struct sk_buff *skb;
1083
1084         /*
1085          * Allocate space for cpl_pass_accept_req which will be synthesized by
1086          * driver. Once the driver synthesizes the request the skb will go
1087          * through the regular cpl_pass_accept_req processing.
1088          * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1089          * cpl_rx_pkt.
1090          */
1091         skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1092                         sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1093         if (unlikely(!skb))
1094                 return NULL;
1095
1096         __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1097                   sizeof(struct rss_header) - pktshift);
1098
1099         /*
1100          * This skb will contain:
1101          *   rss_header from the rspq descriptor (1 flit)
1102          *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1103          *   space for the difference between the size of an
1104          *      rx_pkt and pass_accept_req cpl (1 flit)
1105          *   the packet data from the gl
1106          */
1107         skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1108                                 sizeof(struct rss_header));
1109         skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1110                                        sizeof(struct cpl_pass_accept_req),
1111                                        gl->va + pktshift,
1112                                        gl->tot_len - pktshift);
1113         return skb;
1114 }
1115
1116 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1117                            const __be64 *rsp)
1118 {
1119         unsigned int opcode = *(u8 *)rsp;
1120         struct sk_buff *skb;
1121
1122         if (opcode != CPL_RX_PKT)
1123                 goto out;
1124
1125         skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1126         if (skb == NULL)
1127                 goto out;
1128
1129         if (c4iw_handlers[opcode] == NULL) {
1130                 pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1131                 kfree_skb(skb);
1132                 goto out;
1133         }
1134         c4iw_handlers[opcode](dev, skb);
1135         return 1;
1136 out:
1137         return 0;
1138 }
1139
1140 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1141                         const struct pkt_gl *gl)
1142 {
1143         struct uld_ctx *ctx = handle;
1144         struct c4iw_dev *dev = ctx->dev;
1145         struct sk_buff *skb;
1146         u8 opcode;
1147
1148         if (gl == NULL) {
1149                 /* omit RSS and rsp_ctrl at end of descriptor */
1150                 unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1151
1152                 skb = alloc_skb(256, GFP_ATOMIC);
1153                 if (!skb)
1154                         goto nomem;
1155                 __skb_put(skb, len);
1156                 skb_copy_to_linear_data(skb, &rsp[1], len);
1157         } else if (gl == CXGB4_MSG_AN) {
1158                 const struct rsp_ctrl *rc = (void *)rsp;
1159
1160                 u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1161                 c4iw_ev_handler(dev, qid);
1162                 return 0;
1163         } else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1164                 if (recv_rx_pkt(dev, gl, rsp))
1165                         return 0;
1166
1167                 pr_info("%s: unexpected FL contents at %p, RSS %#llx, FL %#llx, len %u\n",
1168                         pci_name(ctx->lldi.pdev), gl->va,
1169                         be64_to_cpu(*rsp),
1170                         be64_to_cpu(*(__force __be64 *)gl->va),
1171                         gl->tot_len);
1172
1173                 return 0;
1174         } else {
1175                 skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1176                 if (unlikely(!skb))
1177                         goto nomem;
1178         }
1179
1180         opcode = *(u8 *)rsp;
1181         if (c4iw_handlers[opcode]) {
1182                 c4iw_handlers[opcode](dev, skb);
1183         } else {
1184                 pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1185                 kfree_skb(skb);
1186         }
1187
1188         return 0;
1189 nomem:
1190         return -1;
1191 }
1192
1193 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1194 {
1195         struct uld_ctx *ctx = handle;
1196
1197         pr_debug("new_state %u\n", new_state);
1198         switch (new_state) {
1199         case CXGB4_STATE_UP:
1200                 pr_info("%s: Up\n", pci_name(ctx->lldi.pdev));
1201                 if (!ctx->dev) {
1202                         ctx->dev = c4iw_alloc(&ctx->lldi);
1203                         if (IS_ERR(ctx->dev)) {
1204                                 pr_err("%s: initialization failed: %ld\n",
1205                                        pci_name(ctx->lldi.pdev),
1206                                        PTR_ERR(ctx->dev));
1207                                 ctx->dev = NULL;
1208                                 break;
1209                         }
1210
1211                         INIT_WORK(&ctx->reg_work, c4iw_register_device);
1212                         queue_work(reg_workq, &ctx->reg_work);
1213                 }
1214                 break;
1215         case CXGB4_STATE_DOWN:
1216                 pr_info("%s: Down\n", pci_name(ctx->lldi.pdev));
1217                 if (ctx->dev)
1218                         c4iw_remove(ctx);
1219                 break;
1220         case CXGB4_STATE_FATAL_ERROR:
1221         case CXGB4_STATE_START_RECOVERY:
1222                 pr_info("%s: Fatal Error\n", pci_name(ctx->lldi.pdev));
1223                 if (ctx->dev) {
1224                         struct ib_event event;
1225
1226                         ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1227                         memset(&event, 0, sizeof event);
1228                         event.event  = IB_EVENT_DEVICE_FATAL;
1229                         event.device = &ctx->dev->ibdev;
1230                         ib_dispatch_event(&event);
1231                         c4iw_remove(ctx);
1232                 }
1233                 break;
1234         case CXGB4_STATE_DETACH:
1235                 pr_info("%s: Detach\n", pci_name(ctx->lldi.pdev));
1236                 if (ctx->dev)
1237                         c4iw_remove(ctx);
1238                 break;
1239         }
1240         return 0;
1241 }
1242
1243 static int disable_qp_db(int id, void *p, void *data)
1244 {
1245         struct c4iw_qp *qp = p;
1246
1247         t4_disable_wq_db(&qp->wq);
1248         return 0;
1249 }
1250
1251 static void stop_queues(struct uld_ctx *ctx)
1252 {
1253         unsigned long flags;
1254
1255         spin_lock_irqsave(&ctx->dev->lock, flags);
1256         ctx->dev->rdev.stats.db_state_transitions++;
1257         ctx->dev->db_state = STOPPED;
1258         if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1259                 idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1260         else
1261                 ctx->dev->rdev.status_page->db_off = 1;
1262         spin_unlock_irqrestore(&ctx->dev->lock, flags);
1263 }
1264
1265 static int enable_qp_db(int id, void *p, void *data)
1266 {
1267         struct c4iw_qp *qp = p;
1268
1269         t4_enable_wq_db(&qp->wq);
1270         return 0;
1271 }
1272
1273 static void resume_rc_qp(struct c4iw_qp *qp)
1274 {
1275         spin_lock(&qp->lock);
1276         t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc, NULL);
1277         qp->wq.sq.wq_pidx_inc = 0;
1278         t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc, NULL);
1279         qp->wq.rq.wq_pidx_inc = 0;
1280         spin_unlock(&qp->lock);
1281 }
1282
1283 static void resume_a_chunk(struct uld_ctx *ctx)
1284 {
1285         int i;
1286         struct c4iw_qp *qp;
1287
1288         for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1289                 qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1290                                       db_fc_entry);
1291                 list_del_init(&qp->db_fc_entry);
1292                 resume_rc_qp(qp);
1293                 if (list_empty(&ctx->dev->db_fc_list))
1294                         break;
1295         }
1296 }
1297
1298 static void resume_queues(struct uld_ctx *ctx)
1299 {
1300         spin_lock_irq(&ctx->dev->lock);
1301         if (ctx->dev->db_state != STOPPED)
1302                 goto out;
1303         ctx->dev->db_state = FLOW_CONTROL;
1304         while (1) {
1305                 if (list_empty(&ctx->dev->db_fc_list)) {
1306                         WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1307                         ctx->dev->db_state = NORMAL;
1308                         ctx->dev->rdev.stats.db_state_transitions++;
1309                         if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1310                                 idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1311                                              NULL);
1312                         } else {
1313                                 ctx->dev->rdev.status_page->db_off = 0;
1314                         }
1315                         break;
1316                 } else {
1317                         if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1318                             < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1319                                DB_FC_DRAIN_THRESH)) {
1320                                 resume_a_chunk(ctx);
1321                         }
1322                         if (!list_empty(&ctx->dev->db_fc_list)) {
1323                                 spin_unlock_irq(&ctx->dev->lock);
1324                                 if (DB_FC_RESUME_DELAY) {
1325                                         set_current_state(TASK_UNINTERRUPTIBLE);
1326                                         schedule_timeout(DB_FC_RESUME_DELAY);
1327                                 }
1328                                 spin_lock_irq(&ctx->dev->lock);
1329                                 if (ctx->dev->db_state != FLOW_CONTROL)
1330                                         break;
1331                         }
1332                 }
1333         }
1334 out:
1335         if (ctx->dev->db_state != NORMAL)
1336                 ctx->dev->rdev.stats.db_fc_interruptions++;
1337         spin_unlock_irq(&ctx->dev->lock);
1338 }
1339
1340 struct qp_list {
1341         unsigned idx;
1342         struct c4iw_qp **qps;
1343 };
1344
1345 static int add_and_ref_qp(int id, void *p, void *data)
1346 {
1347         struct qp_list *qp_listp = data;
1348         struct c4iw_qp *qp = p;
1349
1350         c4iw_qp_add_ref(&qp->ibqp);
1351         qp_listp->qps[qp_listp->idx++] = qp;
1352         return 0;
1353 }
1354
1355 static int count_qps(int id, void *p, void *data)
1356 {
1357         unsigned *countp = data;
1358         (*countp)++;
1359         return 0;
1360 }
1361
1362 static void deref_qps(struct qp_list *qp_list)
1363 {
1364         int idx;
1365
1366         for (idx = 0; idx < qp_list->idx; idx++)
1367                 c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1368 }
1369
1370 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1371 {
1372         int idx;
1373         int ret;
1374
1375         for (idx = 0; idx < qp_list->idx; idx++) {
1376                 struct c4iw_qp *qp = qp_list->qps[idx];
1377
1378                 spin_lock_irq(&qp->rhp->lock);
1379                 spin_lock(&qp->lock);
1380                 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1381                                           qp->wq.sq.qid,
1382                                           t4_sq_host_wq_pidx(&qp->wq),
1383                                           t4_sq_wq_size(&qp->wq));
1384                 if (ret) {
1385                         pr_err("%s: Fatal error - DB overflow recovery failed - error syncing SQ qid %u\n",
1386                                pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1387                         spin_unlock(&qp->lock);
1388                         spin_unlock_irq(&qp->rhp->lock);
1389                         return;
1390                 }
1391                 qp->wq.sq.wq_pidx_inc = 0;
1392
1393                 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1394                                           qp->wq.rq.qid,
1395                                           t4_rq_host_wq_pidx(&qp->wq),
1396                                           t4_rq_wq_size(&qp->wq));
1397
1398                 if (ret) {
1399                         pr_err("%s: Fatal error - DB overflow recovery failed - error syncing RQ qid %u\n",
1400                                pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1401                         spin_unlock(&qp->lock);
1402                         spin_unlock_irq(&qp->rhp->lock);
1403                         return;
1404                 }
1405                 qp->wq.rq.wq_pidx_inc = 0;
1406                 spin_unlock(&qp->lock);
1407                 spin_unlock_irq(&qp->rhp->lock);
1408
1409                 /* Wait for the dbfifo to drain */
1410                 while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1411                         set_current_state(TASK_UNINTERRUPTIBLE);
1412                         schedule_timeout(usecs_to_jiffies(10));
1413                 }
1414         }
1415 }
1416
1417 static void recover_queues(struct uld_ctx *ctx)
1418 {
1419         int count = 0;
1420         struct qp_list qp_list;
1421         int ret;
1422
1423         /* slow everybody down */
1424         set_current_state(TASK_UNINTERRUPTIBLE);
1425         schedule_timeout(usecs_to_jiffies(1000));
1426
1427         /* flush the SGE contexts */
1428         ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1429         if (ret) {
1430                 pr_err("%s: Fatal error - DB overflow recovery failed\n",
1431                        pci_name(ctx->lldi.pdev));
1432                 return;
1433         }
1434
1435         /* Count active queues so we can build a list of queues to recover */
1436         spin_lock_irq(&ctx->dev->lock);
1437         WARN_ON(ctx->dev->db_state != STOPPED);
1438         ctx->dev->db_state = RECOVERY;
1439         idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1440
1441         qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1442         if (!qp_list.qps) {
1443                 spin_unlock_irq(&ctx->dev->lock);
1444                 return;
1445         }
1446         qp_list.idx = 0;
1447
1448         /* add and ref each qp so it doesn't get freed */
1449         idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1450
1451         spin_unlock_irq(&ctx->dev->lock);
1452
1453         /* now traverse the list in a safe context to recover the db state*/
1454         recover_lost_dbs(ctx, &qp_list);
1455
1456         /* we're almost done!  deref the qps and clean up */
1457         deref_qps(&qp_list);
1458         kfree(qp_list.qps);
1459
1460         spin_lock_irq(&ctx->dev->lock);
1461         WARN_ON(ctx->dev->db_state != RECOVERY);
1462         ctx->dev->db_state = STOPPED;
1463         spin_unlock_irq(&ctx->dev->lock);
1464 }
1465
1466 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1467 {
1468         struct uld_ctx *ctx = handle;
1469
1470         switch (control) {
1471         case CXGB4_CONTROL_DB_FULL:
1472                 stop_queues(ctx);
1473                 ctx->dev->rdev.stats.db_full++;
1474                 break;
1475         case CXGB4_CONTROL_DB_EMPTY:
1476                 resume_queues(ctx);
1477                 mutex_lock(&ctx->dev->rdev.stats.lock);
1478                 ctx->dev->rdev.stats.db_empty++;
1479                 mutex_unlock(&ctx->dev->rdev.stats.lock);
1480                 break;
1481         case CXGB4_CONTROL_DB_DROP:
1482                 recover_queues(ctx);
1483                 mutex_lock(&ctx->dev->rdev.stats.lock);
1484                 ctx->dev->rdev.stats.db_drop++;
1485                 mutex_unlock(&ctx->dev->rdev.stats.lock);
1486                 break;
1487         default:
1488                 pr_warn("%s: unknown control cmd %u\n",
1489                         pci_name(ctx->lldi.pdev), control);
1490                 break;
1491         }
1492         return 0;
1493 }
1494
1495 static struct cxgb4_uld_info c4iw_uld_info = {
1496         .name = DRV_NAME,
1497         .nrxq = MAX_ULD_QSETS,
1498         .ntxq = MAX_ULD_QSETS,
1499         .rxq_size = 511,
1500         .ciq = true,
1501         .lro = false,
1502         .add = c4iw_uld_add,
1503         .rx_handler = c4iw_uld_rx_handler,
1504         .state_change = c4iw_uld_state_change,
1505         .control = c4iw_uld_control,
1506 };
1507
1508 void _c4iw_free_wr_wait(struct kref *kref)
1509 {
1510         struct c4iw_wr_wait *wr_waitp;
1511
1512         wr_waitp = container_of(kref, struct c4iw_wr_wait, kref);
1513         pr_debug("Free wr_wait %p\n", wr_waitp);
1514         kfree(wr_waitp);
1515 }
1516
1517 struct c4iw_wr_wait *c4iw_alloc_wr_wait(gfp_t gfp)
1518 {
1519         struct c4iw_wr_wait *wr_waitp;
1520
1521         wr_waitp = kzalloc(sizeof(*wr_waitp), gfp);
1522         if (wr_waitp) {
1523                 kref_init(&wr_waitp->kref);
1524                 pr_debug("wr_wait %p\n", wr_waitp);
1525         }
1526         return wr_waitp;
1527 }
1528
1529 static int __init c4iw_init_module(void)
1530 {
1531         int err;
1532
1533         err = c4iw_cm_init();
1534         if (err)
1535                 return err;
1536
1537         c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1538         if (!c4iw_debugfs_root)
1539                 pr_warn("could not create debugfs entry, continuing\n");
1540
1541         reg_workq = create_singlethread_workqueue("Register_iWARP_device");
1542         if (!reg_workq) {
1543                 pr_err("Failed creating workqueue to register iwarp device\n");
1544                 return -ENOMEM;
1545         }
1546
1547         cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1548
1549         return 0;
1550 }
1551
1552 static void __exit c4iw_exit_module(void)
1553 {
1554         struct uld_ctx *ctx, *tmp;
1555
1556         mutex_lock(&dev_mutex);
1557         list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1558                 if (ctx->dev)
1559                         c4iw_remove(ctx);
1560                 kfree(ctx);
1561         }
1562         mutex_unlock(&dev_mutex);
1563         flush_workqueue(reg_workq);
1564         destroy_workqueue(reg_workq);
1565         cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1566         c4iw_cm_term();
1567         debugfs_remove_recursive(c4iw_debugfs_root);
1568 }
1569
1570 module_init(c4iw_init_module);
1571 module_exit(c4iw_exit_module);