xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/device.c (revision 812f77b749a8ae11f58dacf0d3ed65e7ede47458)
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 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 	getnstimeofday(&le.poll_host_ts);
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_ts = wq->sq.sw_sq[wq->sq.cidx].host_ts;
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_ts = wq->rq.sw_rq[wq->rq.cidx].host_ts;
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 	struct timespec prev_ts = {0, 0};
134 	struct wr_log_entry *lep;
135 	int prev_ts_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_ts_set) {
149 				prev_ts_set = 1;
150 				prev_ts = lep->poll_host_ts;
151 			}
152 			seq_printf(seq, "%04u: sec %lu nsec %lu qid %u opcode "
153 				   "%u %s 0x%x host_wr_delta sec %lu nsec %lu "
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 				   timespec_sub(lep->poll_host_ts,
159 						prev_ts).tv_sec,
160 				   timespec_sub(lep->poll_host_ts,
161 						prev_ts).tv_nsec,
162 				   lep->qid, lep->opcode,
163 				   lep->opcode == FW_RI_RECEIVE ?
164 							"msn" : "wrid",
165 				   lep->wr_id,
166 				   timespec_sub(lep->poll_host_ts,
167 						lep->post_host_ts).tv_sec,
168 				   timespec_sub(lep->poll_host_ts,
169 						lep->post_host_ts).tv_nsec,
170 				   lep->post_sge_ts, lep->cqe_sge_ts,
171 				   lep->poll_sge_ts,
172 				   ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
173 				   ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
174 			prev_ts = lep->poll_host_ts;
175 		}
176 		idx++;
177 		if (idx > (dev->rdev.wr_log_size - 1))
178 			idx = 0;
179 		lep = &dev->rdev.wr_log[idx];
180 	}
181 #undef ts2ns
182 	return 0;
183 }
184 
185 static int wr_log_open(struct inode *inode, struct file *file)
186 {
187 	return single_open(file, wr_log_show, inode->i_private);
188 }
189 
190 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
191 			    size_t count, loff_t *pos)
192 {
193 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
194 	int i;
195 
196 	if (dev->rdev.wr_log)
197 		for (i = 0; i < dev->rdev.wr_log_size; i++)
198 			dev->rdev.wr_log[i].valid = 0;
199 	return count;
200 }
201 
202 static const struct file_operations wr_log_debugfs_fops = {
203 	.owner   = THIS_MODULE,
204 	.open    = wr_log_open,
205 	.release = single_release,
206 	.read    = seq_read,
207 	.llseek  = seq_lseek,
208 	.write   = wr_log_clear,
209 };
210 
211 static struct sockaddr_in zero_sin = {
212 	.sin_family = AF_INET,
213 };
214 
215 static struct sockaddr_in6 zero_sin6 = {
216 	.sin6_family = AF_INET6,
217 };
218 
219 static void set_ep_sin_addrs(struct c4iw_ep *ep,
220 			     struct sockaddr_in **lsin,
221 			     struct sockaddr_in **rsin,
222 			     struct sockaddr_in **m_lsin,
223 			     struct sockaddr_in **m_rsin)
224 {
225 	struct iw_cm_id *id = ep->com.cm_id;
226 
227 	*lsin = (struct sockaddr_in *)&ep->com.local_addr;
228 	*rsin = (struct sockaddr_in *)&ep->com.remote_addr;
229 	if (id) {
230 		*m_lsin = (struct sockaddr_in *)&id->m_local_addr;
231 		*m_rsin = (struct sockaddr_in *)&id->m_remote_addr;
232 	} else {
233 		*m_lsin = &zero_sin;
234 		*m_rsin = &zero_sin;
235 	}
236 }
237 
238 static void set_ep_sin6_addrs(struct c4iw_ep *ep,
239 			      struct sockaddr_in6 **lsin6,
240 			      struct sockaddr_in6 **rsin6,
241 			      struct sockaddr_in6 **m_lsin6,
242 			      struct sockaddr_in6 **m_rsin6)
243 {
244 	struct iw_cm_id *id = ep->com.cm_id;
245 
246 	*lsin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
247 	*rsin6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
248 	if (id) {
249 		*m_lsin6 = (struct sockaddr_in6 *)&id->m_local_addr;
250 		*m_rsin6 = (struct sockaddr_in6 *)&id->m_remote_addr;
251 	} else {
252 		*m_lsin6 = &zero_sin6;
253 		*m_rsin6 = &zero_sin6;
254 	}
255 }
256 
257 static int dump_qp(int id, void *p, void *data)
258 {
259 	struct c4iw_qp *qp = p;
260 	struct c4iw_debugfs_data *qpd = data;
261 	int space;
262 	int cc;
263 
264 	if (id != qp->wq.sq.qid)
265 		return 0;
266 
267 	space = qpd->bufsize - qpd->pos - 1;
268 	if (space == 0)
269 		return 1;
270 
271 	if (qp->ep) {
272 		struct c4iw_ep *ep = qp->ep;
273 
274 		if (ep->com.local_addr.ss_family == AF_INET) {
275 			struct sockaddr_in *lsin;
276 			struct sockaddr_in *rsin;
277 			struct sockaddr_in *m_lsin;
278 			struct sockaddr_in *m_rsin;
279 
280 			set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
281 			cc = snprintf(qpd->buf + qpd->pos, space,
282 				      "rc qp sq id %u rq id %u state %u "
283 				      "onchip %u ep tid %u state %u "
284 				      "%pI4:%u/%u->%pI4:%u/%u\n",
285 				      qp->wq.sq.qid, qp->wq.rq.qid,
286 				      (int)qp->attr.state,
287 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
288 				      ep->hwtid, (int)ep->com.state,
289 				      &lsin->sin_addr, ntohs(lsin->sin_port),
290 				      ntohs(m_lsin->sin_port),
291 				      &rsin->sin_addr, ntohs(rsin->sin_port),
292 				      ntohs(m_rsin->sin_port));
293 		} else {
294 			struct sockaddr_in6 *lsin6;
295 			struct sockaddr_in6 *rsin6;
296 			struct sockaddr_in6 *m_lsin6;
297 			struct sockaddr_in6 *m_rsin6;
298 
299 			set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6,
300 					  &m_rsin6);
301 			cc = snprintf(qpd->buf + qpd->pos, space,
302 				      "rc qp sq id %u rq id %u state %u "
303 				      "onchip %u ep tid %u state %u "
304 				      "%pI6:%u/%u->%pI6:%u/%u\n",
305 				      qp->wq.sq.qid, qp->wq.rq.qid,
306 				      (int)qp->attr.state,
307 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
308 				      ep->hwtid, (int)ep->com.state,
309 				      &lsin6->sin6_addr,
310 				      ntohs(lsin6->sin6_port),
311 				      ntohs(m_lsin6->sin6_port),
312 				      &rsin6->sin6_addr,
313 				      ntohs(rsin6->sin6_port),
314 				      ntohs(m_rsin6->sin6_port));
315 		}
316 	} else
317 		cc = snprintf(qpd->buf + qpd->pos, space,
318 			     "qp sq id %u rq id %u state %u onchip %u\n",
319 			      qp->wq.sq.qid, qp->wq.rq.qid,
320 			      (int)qp->attr.state,
321 			      qp->wq.sq.flags & T4_SQ_ONCHIP);
322 	if (cc < space)
323 		qpd->pos += cc;
324 	return 0;
325 }
326 
327 static int qp_release(struct inode *inode, struct file *file)
328 {
329 	struct c4iw_debugfs_data *qpd = file->private_data;
330 	if (!qpd) {
331 		pr_info("%s null qpd?\n", __func__);
332 		return 0;
333 	}
334 	vfree(qpd->buf);
335 	kfree(qpd);
336 	return 0;
337 }
338 
339 static int qp_open(struct inode *inode, struct file *file)
340 {
341 	struct c4iw_debugfs_data *qpd;
342 	int count = 1;
343 
344 	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
345 	if (!qpd)
346 		return -ENOMEM;
347 
348 	qpd->devp = inode->i_private;
349 	qpd->pos = 0;
350 
351 	spin_lock_irq(&qpd->devp->lock);
352 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
353 	spin_unlock_irq(&qpd->devp->lock);
354 
355 	qpd->bufsize = count * 180;
356 	qpd->buf = vmalloc(qpd->bufsize);
357 	if (!qpd->buf) {
358 		kfree(qpd);
359 		return -ENOMEM;
360 	}
361 
362 	spin_lock_irq(&qpd->devp->lock);
363 	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
364 	spin_unlock_irq(&qpd->devp->lock);
365 
366 	qpd->buf[qpd->pos++] = 0;
367 	file->private_data = qpd;
368 	return 0;
369 }
370 
371 static const struct file_operations qp_debugfs_fops = {
372 	.owner   = THIS_MODULE,
373 	.open    = qp_open,
374 	.release = qp_release,
375 	.read    = debugfs_read,
376 	.llseek  = default_llseek,
377 };
378 
379 static int dump_stag(int id, void *p, void *data)
380 {
381 	struct c4iw_debugfs_data *stagd = data;
382 	int space;
383 	int cc;
384 	struct fw_ri_tpte tpte;
385 	int ret;
386 
387 	space = stagd->bufsize - stagd->pos - 1;
388 	if (space == 0)
389 		return 1;
390 
391 	ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
392 			      (__be32 *)&tpte);
393 	if (ret) {
394 		dev_err(&stagd->devp->rdev.lldi.pdev->dev,
395 			"%s cxgb4_read_tpte err %d\n", __func__, ret);
396 		return ret;
397 	}
398 	cc = snprintf(stagd->buf + stagd->pos, space,
399 		      "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
400 		      "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
401 		      (u32)id<<8,
402 		      FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
403 		      FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
404 		      FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
405 		      FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
406 		      FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
407 		      FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
408 		      ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
409 		      ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
410 	if (cc < space)
411 		stagd->pos += cc;
412 	return 0;
413 }
414 
415 static int stag_release(struct inode *inode, struct file *file)
416 {
417 	struct c4iw_debugfs_data *stagd = file->private_data;
418 	if (!stagd) {
419 		pr_info("%s null stagd?\n", __func__);
420 		return 0;
421 	}
422 	vfree(stagd->buf);
423 	kfree(stagd);
424 	return 0;
425 }
426 
427 static int stag_open(struct inode *inode, struct file *file)
428 {
429 	struct c4iw_debugfs_data *stagd;
430 	int ret = 0;
431 	int count = 1;
432 
433 	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
434 	if (!stagd) {
435 		ret = -ENOMEM;
436 		goto out;
437 	}
438 	stagd->devp = inode->i_private;
439 	stagd->pos = 0;
440 
441 	spin_lock_irq(&stagd->devp->lock);
442 	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
443 	spin_unlock_irq(&stagd->devp->lock);
444 
445 	stagd->bufsize = count * 256;
446 	stagd->buf = vmalloc(stagd->bufsize);
447 	if (!stagd->buf) {
448 		ret = -ENOMEM;
449 		goto err1;
450 	}
451 
452 	spin_lock_irq(&stagd->devp->lock);
453 	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
454 	spin_unlock_irq(&stagd->devp->lock);
455 
456 	stagd->buf[stagd->pos++] = 0;
457 	file->private_data = stagd;
458 	goto out;
459 err1:
460 	kfree(stagd);
461 out:
462 	return ret;
463 }
464 
465 static const struct file_operations stag_debugfs_fops = {
466 	.owner   = THIS_MODULE,
467 	.open    = stag_open,
468 	.release = stag_release,
469 	.read    = debugfs_read,
470 	.llseek  = default_llseek,
471 };
472 
473 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
474 
475 static int stats_show(struct seq_file *seq, void *v)
476 {
477 	struct c4iw_dev *dev = seq->private;
478 
479 	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
480 		   "Max", "Fail");
481 	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
482 			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
483 			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
484 	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
485 			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
486 			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
487 	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
488 			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
489 			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
490 	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
491 			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
492 			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
493 	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
494 			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
495 			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
496 	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
497 			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
498 			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
499 	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
500 	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
501 	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
502 	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
503 		   db_state_str[dev->db_state],
504 		   dev->rdev.stats.db_state_transitions,
505 		   dev->rdev.stats.db_fc_interruptions);
506 	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
507 	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
508 		   dev->rdev.stats.act_ofld_conn_fails);
509 	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
510 		   dev->rdev.stats.pas_ofld_conn_fails);
511 	seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
512 	seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
513 	return 0;
514 }
515 
516 static int stats_open(struct inode *inode, struct file *file)
517 {
518 	return single_open(file, stats_show, inode->i_private);
519 }
520 
521 static ssize_t stats_clear(struct file *file, const char __user *buf,
522 		size_t count, loff_t *pos)
523 {
524 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
525 
526 	mutex_lock(&dev->rdev.stats.lock);
527 	dev->rdev.stats.pd.max = 0;
528 	dev->rdev.stats.pd.fail = 0;
529 	dev->rdev.stats.qid.max = 0;
530 	dev->rdev.stats.qid.fail = 0;
531 	dev->rdev.stats.stag.max = 0;
532 	dev->rdev.stats.stag.fail = 0;
533 	dev->rdev.stats.pbl.max = 0;
534 	dev->rdev.stats.pbl.fail = 0;
535 	dev->rdev.stats.rqt.max = 0;
536 	dev->rdev.stats.rqt.fail = 0;
537 	dev->rdev.stats.ocqp.max = 0;
538 	dev->rdev.stats.ocqp.fail = 0;
539 	dev->rdev.stats.db_full = 0;
540 	dev->rdev.stats.db_empty = 0;
541 	dev->rdev.stats.db_drop = 0;
542 	dev->rdev.stats.db_state_transitions = 0;
543 	dev->rdev.stats.tcam_full = 0;
544 	dev->rdev.stats.act_ofld_conn_fails = 0;
545 	dev->rdev.stats.pas_ofld_conn_fails = 0;
546 	mutex_unlock(&dev->rdev.stats.lock);
547 	return count;
548 }
549 
550 static const struct file_operations stats_debugfs_fops = {
551 	.owner   = THIS_MODULE,
552 	.open    = stats_open,
553 	.release = single_release,
554 	.read    = seq_read,
555 	.llseek  = seq_lseek,
556 	.write   = stats_clear,
557 };
558 
559 static int dump_ep(int id, void *p, void *data)
560 {
561 	struct c4iw_ep *ep = p;
562 	struct c4iw_debugfs_data *epd = data;
563 	int space;
564 	int cc;
565 
566 	space = epd->bufsize - epd->pos - 1;
567 	if (space == 0)
568 		return 1;
569 
570 	if (ep->com.local_addr.ss_family == AF_INET) {
571 		struct sockaddr_in *lsin;
572 		struct sockaddr_in *rsin;
573 		struct sockaddr_in *m_lsin;
574 		struct sockaddr_in *m_rsin;
575 
576 		set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
577 		cc = snprintf(epd->buf + epd->pos, space,
578 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
579 			      "history 0x%lx hwtid %d atid %d "
580 			      "conn_na %u abort_na %u "
581 			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
582 			      ep, ep->com.cm_id, ep->com.qp,
583 			      (int)ep->com.state, ep->com.flags,
584 			      ep->com.history, ep->hwtid, ep->atid,
585 			      ep->stats.connect_neg_adv,
586 			      ep->stats.abort_neg_adv,
587 			      &lsin->sin_addr, ntohs(lsin->sin_port),
588 			      ntohs(m_lsin->sin_port),
589 			      &rsin->sin_addr, ntohs(rsin->sin_port),
590 			      ntohs(m_rsin->sin_port));
591 	} else {
592 		struct sockaddr_in6 *lsin6;
593 		struct sockaddr_in6 *rsin6;
594 		struct sockaddr_in6 *m_lsin6;
595 		struct sockaddr_in6 *m_rsin6;
596 
597 		set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6, &m_rsin6);
598 		cc = snprintf(epd->buf + epd->pos, space,
599 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
600 			      "history 0x%lx hwtid %d atid %d "
601 			      "conn_na %u abort_na %u "
602 			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
603 			      ep, ep->com.cm_id, ep->com.qp,
604 			      (int)ep->com.state, ep->com.flags,
605 			      ep->com.history, ep->hwtid, ep->atid,
606 			      ep->stats.connect_neg_adv,
607 			      ep->stats.abort_neg_adv,
608 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
609 			      ntohs(m_lsin6->sin6_port),
610 			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
611 			      ntohs(m_rsin6->sin6_port));
612 	}
613 	if (cc < space)
614 		epd->pos += cc;
615 	return 0;
616 }
617 
618 static int dump_listen_ep(int id, void *p, void *data)
619 {
620 	struct c4iw_listen_ep *ep = p;
621 	struct c4iw_debugfs_data *epd = data;
622 	int space;
623 	int cc;
624 
625 	space = epd->bufsize - epd->pos - 1;
626 	if (space == 0)
627 		return 1;
628 
629 	if (ep->com.local_addr.ss_family == AF_INET) {
630 		struct sockaddr_in *lsin = (struct sockaddr_in *)
631 			&ep->com.cm_id->local_addr;
632 		struct sockaddr_in *m_lsin = (struct sockaddr_in *)
633 			&ep->com.cm_id->m_local_addr;
634 
635 		cc = snprintf(epd->buf + epd->pos, space,
636 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
637 			      "backlog %d %pI4:%d/%d\n",
638 			      ep, ep->com.cm_id, (int)ep->com.state,
639 			      ep->com.flags, ep->stid, ep->backlog,
640 			      &lsin->sin_addr, ntohs(lsin->sin_port),
641 			      ntohs(m_lsin->sin_port));
642 	} else {
643 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
644 			&ep->com.cm_id->local_addr;
645 		struct sockaddr_in6 *m_lsin6 = (struct sockaddr_in6 *)
646 			&ep->com.cm_id->m_local_addr;
647 
648 		cc = snprintf(epd->buf + epd->pos, space,
649 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
650 			      "backlog %d %pI6:%d/%d\n",
651 			      ep, ep->com.cm_id, (int)ep->com.state,
652 			      ep->com.flags, ep->stid, ep->backlog,
653 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
654 			      ntohs(m_lsin6->sin6_port));
655 	}
656 	if (cc < space)
657 		epd->pos += cc;
658 	return 0;
659 }
660 
661 static int ep_release(struct inode *inode, struct file *file)
662 {
663 	struct c4iw_debugfs_data *epd = file->private_data;
664 	if (!epd) {
665 		pr_info("%s null qpd?\n", __func__);
666 		return 0;
667 	}
668 	vfree(epd->buf);
669 	kfree(epd);
670 	return 0;
671 }
672 
673 static int ep_open(struct inode *inode, struct file *file)
674 {
675 	struct c4iw_debugfs_data *epd;
676 	int ret = 0;
677 	int count = 1;
678 
679 	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
680 	if (!epd) {
681 		ret = -ENOMEM;
682 		goto out;
683 	}
684 	epd->devp = inode->i_private;
685 	epd->pos = 0;
686 
687 	spin_lock_irq(&epd->devp->lock);
688 	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
689 	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
690 	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
691 	spin_unlock_irq(&epd->devp->lock);
692 
693 	epd->bufsize = count * 240;
694 	epd->buf = vmalloc(epd->bufsize);
695 	if (!epd->buf) {
696 		ret = -ENOMEM;
697 		goto err1;
698 	}
699 
700 	spin_lock_irq(&epd->devp->lock);
701 	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
702 	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
703 	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
704 	spin_unlock_irq(&epd->devp->lock);
705 
706 	file->private_data = epd;
707 	goto out;
708 err1:
709 	kfree(epd);
710 out:
711 	return ret;
712 }
713 
714 static const struct file_operations ep_debugfs_fops = {
715 	.owner   = THIS_MODULE,
716 	.open    = ep_open,
717 	.release = ep_release,
718 	.read    = debugfs_read,
719 };
720 
721 static int setup_debugfs(struct c4iw_dev *devp)
722 {
723 	if (!devp->debugfs_root)
724 		return -1;
725 
726 	debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
727 				 (void *)devp, &qp_debugfs_fops, 4096);
728 
729 	debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
730 				 (void *)devp, &stag_debugfs_fops, 4096);
731 
732 	debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
733 				 (void *)devp, &stats_debugfs_fops, 4096);
734 
735 	debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
736 				 (void *)devp, &ep_debugfs_fops, 4096);
737 
738 	if (c4iw_wr_log)
739 		debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
740 					 (void *)devp, &wr_log_debugfs_fops, 4096);
741 	return 0;
742 }
743 
744 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
745 			       struct c4iw_dev_ucontext *uctx)
746 {
747 	struct list_head *pos, *nxt;
748 	struct c4iw_qid_list *entry;
749 
750 	mutex_lock(&uctx->lock);
751 	list_for_each_safe(pos, nxt, &uctx->qpids) {
752 		entry = list_entry(pos, struct c4iw_qid_list, entry);
753 		list_del_init(&entry->entry);
754 		if (!(entry->qid & rdev->qpmask)) {
755 			c4iw_put_resource(&rdev->resource.qid_table,
756 					  entry->qid);
757 			mutex_lock(&rdev->stats.lock);
758 			rdev->stats.qid.cur -= rdev->qpmask + 1;
759 			mutex_unlock(&rdev->stats.lock);
760 		}
761 		kfree(entry);
762 	}
763 
764 	list_for_each_safe(pos, nxt, &uctx->cqids) {
765 		entry = list_entry(pos, struct c4iw_qid_list, entry);
766 		list_del_init(&entry->entry);
767 		kfree(entry);
768 	}
769 	mutex_unlock(&uctx->lock);
770 }
771 
772 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
773 			    struct c4iw_dev_ucontext *uctx)
774 {
775 	INIT_LIST_HEAD(&uctx->qpids);
776 	INIT_LIST_HEAD(&uctx->cqids);
777 	mutex_init(&uctx->lock);
778 }
779 
780 /* Caller takes care of locking if needed */
781 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
782 {
783 	int err;
784 
785 	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
786 
787 	/*
788 	 * This implementation assumes udb_density == ucq_density!  Eventually
789 	 * we might need to support this but for now fail the open. Also the
790 	 * cqid and qpid range must match for now.
791 	 */
792 	if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
793 		pr_err("%s: unsupported udb/ucq densities %u/%u\n",
794 		       pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
795 		       rdev->lldi.ucq_density);
796 		return -EINVAL;
797 	}
798 	if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
799 	    rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
800 		pr_err("%s: unsupported qp and cq id ranges qp start %u size %u cq start %u size %u\n",
801 		       pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
802 		       rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
803 		       rdev->lldi.vr->cq.size);
804 		return -EINVAL;
805 	}
806 
807 	rdev->qpmask = rdev->lldi.udb_density - 1;
808 	rdev->cqmask = rdev->lldi.ucq_density - 1;
809 	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",
810 		 pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
811 		 rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
812 		 rdev->lldi.vr->pbl.start,
813 		 rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
814 		 rdev->lldi.vr->rq.size,
815 		 rdev->lldi.vr->qp.start,
816 		 rdev->lldi.vr->qp.size,
817 		 rdev->lldi.vr->cq.start,
818 		 rdev->lldi.vr->cq.size);
819 	pr_debug("udb %pR db_reg %p gts_reg %p qpmask 0x%x cqmask 0x%x\n",
820 		 &rdev->lldi.pdev->resource[2],
821 		 rdev->lldi.db_reg, rdev->lldi.gts_reg,
822 		 rdev->qpmask, rdev->cqmask);
823 
824 	if (c4iw_num_stags(rdev) == 0)
825 		return -EINVAL;
826 
827 	rdev->stats.pd.total = T4_MAX_NUM_PD;
828 	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
829 	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
830 	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
831 	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
832 	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
833 
834 	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
835 	if (err) {
836 		pr_err("error %d initializing resources\n", err);
837 		return err;
838 	}
839 	err = c4iw_pblpool_create(rdev);
840 	if (err) {
841 		pr_err("error %d initializing pbl pool\n", err);
842 		goto destroy_resource;
843 	}
844 	err = c4iw_rqtpool_create(rdev);
845 	if (err) {
846 		pr_err("error %d initializing rqt pool\n", err);
847 		goto destroy_pblpool;
848 	}
849 	err = c4iw_ocqp_pool_create(rdev);
850 	if (err) {
851 		pr_err("error %d initializing ocqp pool\n", err);
852 		goto destroy_rqtpool;
853 	}
854 	rdev->status_page = (struct t4_dev_status_page *)
855 			    __get_free_page(GFP_KERNEL);
856 	if (!rdev->status_page) {
857 		err = -ENOMEM;
858 		goto destroy_ocqp_pool;
859 	}
860 	rdev->status_page->qp_start = rdev->lldi.vr->qp.start;
861 	rdev->status_page->qp_size = rdev->lldi.vr->qp.size;
862 	rdev->status_page->cq_start = rdev->lldi.vr->cq.start;
863 	rdev->status_page->cq_size = rdev->lldi.vr->cq.size;
864 
865 	if (c4iw_wr_log) {
866 		rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
867 				       sizeof(*rdev->wr_log), GFP_KERNEL);
868 		if (rdev->wr_log) {
869 			rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
870 			atomic_set(&rdev->wr_log_idx, 0);
871 		}
872 	}
873 
874 	rdev->free_workq = create_singlethread_workqueue("iw_cxgb4_free");
875 	if (!rdev->free_workq) {
876 		err = -ENOMEM;
877 		goto err_free_status_page_and_wr_log;
878 	}
879 
880 	rdev->status_page->db_off = 0;
881 
882 	return 0;
883 err_free_status_page_and_wr_log:
884 	if (c4iw_wr_log && rdev->wr_log)
885 		kfree(rdev->wr_log);
886 	free_page((unsigned long)rdev->status_page);
887 destroy_ocqp_pool:
888 	c4iw_ocqp_pool_destroy(rdev);
889 destroy_rqtpool:
890 	c4iw_rqtpool_destroy(rdev);
891 destroy_pblpool:
892 	c4iw_pblpool_destroy(rdev);
893 destroy_resource:
894 	c4iw_destroy_resource(&rdev->resource);
895 	return err;
896 }
897 
898 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
899 {
900 	destroy_workqueue(rdev->free_workq);
901 	kfree(rdev->wr_log);
902 	c4iw_release_dev_ucontext(rdev, &rdev->uctx);
903 	free_page((unsigned long)rdev->status_page);
904 	c4iw_pblpool_destroy(rdev);
905 	c4iw_rqtpool_destroy(rdev);
906 	c4iw_ocqp_pool_destroy(rdev);
907 	c4iw_destroy_resource(&rdev->resource);
908 }
909 
910 void c4iw_dealloc(struct uld_ctx *ctx)
911 {
912 	c4iw_rdev_close(&ctx->dev->rdev);
913 	WARN_ON_ONCE(!idr_is_empty(&ctx->dev->cqidr));
914 	idr_destroy(&ctx->dev->cqidr);
915 	WARN_ON_ONCE(!idr_is_empty(&ctx->dev->qpidr));
916 	idr_destroy(&ctx->dev->qpidr);
917 	WARN_ON_ONCE(!idr_is_empty(&ctx->dev->mmidr));
918 	idr_destroy(&ctx->dev->mmidr);
919 	wait_event(ctx->dev->wait, idr_is_empty(&ctx->dev->hwtid_idr));
920 	idr_destroy(&ctx->dev->hwtid_idr);
921 	idr_destroy(&ctx->dev->stid_idr);
922 	idr_destroy(&ctx->dev->atid_idr);
923 	if (ctx->dev->rdev.bar2_kva)
924 		iounmap(ctx->dev->rdev.bar2_kva);
925 	if (ctx->dev->rdev.oc_mw_kva)
926 		iounmap(ctx->dev->rdev.oc_mw_kva);
927 	ib_dealloc_device(&ctx->dev->ibdev);
928 	ctx->dev = NULL;
929 }
930 
931 static void c4iw_remove(struct uld_ctx *ctx)
932 {
933 	pr_debug("c4iw_dev %p\n", ctx->dev);
934 	c4iw_unregister_device(ctx->dev);
935 	c4iw_dealloc(ctx);
936 }
937 
938 static int rdma_supported(const struct cxgb4_lld_info *infop)
939 {
940 	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
941 	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
942 	       infop->vr->cq.size > 0;
943 }
944 
945 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
946 {
947 	struct c4iw_dev *devp;
948 	int ret;
949 
950 	if (!rdma_supported(infop)) {
951 		pr_info("%s: RDMA not supported on this device\n",
952 			pci_name(infop->pdev));
953 		return ERR_PTR(-ENOSYS);
954 	}
955 	if (!ocqp_supported(infop))
956 		pr_info("%s: On-Chip Queues not supported on this device\n",
957 			pci_name(infop->pdev));
958 
959 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
960 	if (!devp) {
961 		pr_err("Cannot allocate ib device\n");
962 		return ERR_PTR(-ENOMEM);
963 	}
964 	devp->rdev.lldi = *infop;
965 
966 	/* init various hw-queue params based on lld info */
967 	pr_debug("Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
968 		 devp->rdev.lldi.sge_ingpadboundary,
969 		 devp->rdev.lldi.sge_egrstatuspagesize);
970 
971 	devp->rdev.hw_queue.t4_eq_status_entries =
972 		devp->rdev.lldi.sge_egrstatuspagesize / 64;
973 	devp->rdev.hw_queue.t4_max_eq_size = 65520;
974 	devp->rdev.hw_queue.t4_max_iq_size = 65520;
975 	devp->rdev.hw_queue.t4_max_rq_size = 8192 -
976 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
977 	devp->rdev.hw_queue.t4_max_sq_size =
978 		devp->rdev.hw_queue.t4_max_eq_size -
979 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
980 	devp->rdev.hw_queue.t4_max_qp_depth =
981 		devp->rdev.hw_queue.t4_max_rq_size;
982 	devp->rdev.hw_queue.t4_max_cq_depth =
983 		devp->rdev.hw_queue.t4_max_iq_size - 2;
984 	devp->rdev.hw_queue.t4_stat_len =
985 		devp->rdev.lldi.sge_egrstatuspagesize;
986 
987 	/*
988 	 * For T5/T6 devices, we map all of BAR2 with WC.
989 	 * For T4 devices with onchip qp mem, we map only that part
990 	 * of BAR2 with WC.
991 	 */
992 	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
993 	if (!is_t4(devp->rdev.lldi.adapter_type)) {
994 		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
995 			pci_resource_len(devp->rdev.lldi.pdev, 2));
996 		if (!devp->rdev.bar2_kva) {
997 			pr_err("Unable to ioremap BAR2\n");
998 			ib_dealloc_device(&devp->ibdev);
999 			return ERR_PTR(-EINVAL);
1000 		}
1001 	} else if (ocqp_supported(infop)) {
1002 		devp->rdev.oc_mw_pa =
1003 			pci_resource_start(devp->rdev.lldi.pdev, 2) +
1004 			pci_resource_len(devp->rdev.lldi.pdev, 2) -
1005 			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
1006 		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
1007 			devp->rdev.lldi.vr->ocq.size);
1008 		if (!devp->rdev.oc_mw_kva) {
1009 			pr_err("Unable to ioremap onchip mem\n");
1010 			ib_dealloc_device(&devp->ibdev);
1011 			return ERR_PTR(-EINVAL);
1012 		}
1013 	}
1014 
1015 	pr_debug("ocq memory: hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
1016 		 devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
1017 		 devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
1018 
1019 	ret = c4iw_rdev_open(&devp->rdev);
1020 	if (ret) {
1021 		pr_err("Unable to open CXIO rdev err %d\n", ret);
1022 		ib_dealloc_device(&devp->ibdev);
1023 		return ERR_PTR(ret);
1024 	}
1025 
1026 	idr_init(&devp->cqidr);
1027 	idr_init(&devp->qpidr);
1028 	idr_init(&devp->mmidr);
1029 	idr_init(&devp->hwtid_idr);
1030 	idr_init(&devp->stid_idr);
1031 	idr_init(&devp->atid_idr);
1032 	spin_lock_init(&devp->lock);
1033 	mutex_init(&devp->rdev.stats.lock);
1034 	mutex_init(&devp->db_mutex);
1035 	INIT_LIST_HEAD(&devp->db_fc_list);
1036 	init_waitqueue_head(&devp->wait);
1037 	devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1038 
1039 	if (c4iw_debugfs_root) {
1040 		devp->debugfs_root = debugfs_create_dir(
1041 					pci_name(devp->rdev.lldi.pdev),
1042 					c4iw_debugfs_root);
1043 		setup_debugfs(devp);
1044 	}
1045 
1046 
1047 	return devp;
1048 }
1049 
1050 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1051 {
1052 	struct uld_ctx *ctx;
1053 	static int vers_printed;
1054 	int i;
1055 
1056 	if (!vers_printed++)
1057 		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1058 			DRV_VERSION);
1059 
1060 	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1061 	if (!ctx) {
1062 		ctx = ERR_PTR(-ENOMEM);
1063 		goto out;
1064 	}
1065 	ctx->lldi = *infop;
1066 
1067 	pr_debug("found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1068 		 pci_name(ctx->lldi.pdev),
1069 		 ctx->lldi.nchan, ctx->lldi.nrxq,
1070 		 ctx->lldi.ntxq, ctx->lldi.nports);
1071 
1072 	mutex_lock(&dev_mutex);
1073 	list_add_tail(&ctx->entry, &uld_ctx_list);
1074 	mutex_unlock(&dev_mutex);
1075 
1076 	for (i = 0; i < ctx->lldi.nrxq; i++)
1077 		pr_debug("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1078 out:
1079 	return ctx;
1080 }
1081 
1082 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1083 						 const __be64 *rsp,
1084 						 u32 pktshift)
1085 {
1086 	struct sk_buff *skb;
1087 
1088 	/*
1089 	 * Allocate space for cpl_pass_accept_req which will be synthesized by
1090 	 * driver. Once the driver synthesizes the request the skb will go
1091 	 * through the regular cpl_pass_accept_req processing.
1092 	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1093 	 * cpl_rx_pkt.
1094 	 */
1095 	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1096 			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1097 	if (unlikely(!skb))
1098 		return NULL;
1099 
1100 	__skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1101 		  sizeof(struct rss_header) - pktshift);
1102 
1103 	/*
1104 	 * This skb will contain:
1105 	 *   rss_header from the rspq descriptor (1 flit)
1106 	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1107 	 *   space for the difference between the size of an
1108 	 *      rx_pkt and pass_accept_req cpl (1 flit)
1109 	 *   the packet data from the gl
1110 	 */
1111 	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1112 				sizeof(struct rss_header));
1113 	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1114 				       sizeof(struct cpl_pass_accept_req),
1115 				       gl->va + pktshift,
1116 				       gl->tot_len - pktshift);
1117 	return skb;
1118 }
1119 
1120 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1121 			   const __be64 *rsp)
1122 {
1123 	unsigned int opcode = *(u8 *)rsp;
1124 	struct sk_buff *skb;
1125 
1126 	if (opcode != CPL_RX_PKT)
1127 		goto out;
1128 
1129 	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1130 	if (skb == NULL)
1131 		goto out;
1132 
1133 	if (c4iw_handlers[opcode] == NULL) {
1134 		pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1135 		kfree_skb(skb);
1136 		goto out;
1137 	}
1138 	c4iw_handlers[opcode](dev, skb);
1139 	return 1;
1140 out:
1141 	return 0;
1142 }
1143 
1144 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1145 			const struct pkt_gl *gl)
1146 {
1147 	struct uld_ctx *ctx = handle;
1148 	struct c4iw_dev *dev = ctx->dev;
1149 	struct sk_buff *skb;
1150 	u8 opcode;
1151 
1152 	if (gl == NULL) {
1153 		/* omit RSS and rsp_ctrl at end of descriptor */
1154 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1155 
1156 		skb = alloc_skb(256, GFP_ATOMIC);
1157 		if (!skb)
1158 			goto nomem;
1159 		__skb_put(skb, len);
1160 		skb_copy_to_linear_data(skb, &rsp[1], len);
1161 	} else if (gl == CXGB4_MSG_AN) {
1162 		const struct rsp_ctrl *rc = (void *)rsp;
1163 
1164 		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1165 		c4iw_ev_handler(dev, qid);
1166 		return 0;
1167 	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1168 		if (recv_rx_pkt(dev, gl, rsp))
1169 			return 0;
1170 
1171 		pr_info("%s: unexpected FL contents at %p, RSS %#llx, FL %#llx, len %u\n",
1172 			pci_name(ctx->lldi.pdev), gl->va,
1173 			be64_to_cpu(*rsp),
1174 			be64_to_cpu(*(__force __be64 *)gl->va),
1175 			gl->tot_len);
1176 
1177 		return 0;
1178 	} else {
1179 		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1180 		if (unlikely(!skb))
1181 			goto nomem;
1182 	}
1183 
1184 	opcode = *(u8 *)rsp;
1185 	if (c4iw_handlers[opcode]) {
1186 		c4iw_handlers[opcode](dev, skb);
1187 	} else {
1188 		pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1189 		kfree_skb(skb);
1190 	}
1191 
1192 	return 0;
1193 nomem:
1194 	return -1;
1195 }
1196 
1197 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1198 {
1199 	struct uld_ctx *ctx = handle;
1200 
1201 	pr_debug("new_state %u\n", new_state);
1202 	switch (new_state) {
1203 	case CXGB4_STATE_UP:
1204 		pr_info("%s: Up\n", pci_name(ctx->lldi.pdev));
1205 		if (!ctx->dev) {
1206 			ctx->dev = c4iw_alloc(&ctx->lldi);
1207 			if (IS_ERR(ctx->dev)) {
1208 				pr_err("%s: initialization failed: %ld\n",
1209 				       pci_name(ctx->lldi.pdev),
1210 				       PTR_ERR(ctx->dev));
1211 				ctx->dev = NULL;
1212 				break;
1213 			}
1214 
1215 			INIT_WORK(&ctx->reg_work, c4iw_register_device);
1216 			queue_work(reg_workq, &ctx->reg_work);
1217 		}
1218 		break;
1219 	case CXGB4_STATE_DOWN:
1220 		pr_info("%s: Down\n", pci_name(ctx->lldi.pdev));
1221 		if (ctx->dev)
1222 			c4iw_remove(ctx);
1223 		break;
1224 	case CXGB4_STATE_START_RECOVERY:
1225 		pr_info("%s: Fatal Error\n", pci_name(ctx->lldi.pdev));
1226 		if (ctx->dev) {
1227 			struct ib_event event;
1228 
1229 			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1230 			memset(&event, 0, sizeof event);
1231 			event.event  = IB_EVENT_DEVICE_FATAL;
1232 			event.device = &ctx->dev->ibdev;
1233 			ib_dispatch_event(&event);
1234 			c4iw_remove(ctx);
1235 		}
1236 		break;
1237 	case CXGB4_STATE_DETACH:
1238 		pr_info("%s: Detach\n", pci_name(ctx->lldi.pdev));
1239 		if (ctx->dev)
1240 			c4iw_remove(ctx);
1241 		break;
1242 	}
1243 	return 0;
1244 }
1245 
1246 static int disable_qp_db(int id, void *p, void *data)
1247 {
1248 	struct c4iw_qp *qp = p;
1249 
1250 	t4_disable_wq_db(&qp->wq);
1251 	return 0;
1252 }
1253 
1254 static void stop_queues(struct uld_ctx *ctx)
1255 {
1256 	unsigned long flags;
1257 
1258 	spin_lock_irqsave(&ctx->dev->lock, flags);
1259 	ctx->dev->rdev.stats.db_state_transitions++;
1260 	ctx->dev->db_state = STOPPED;
1261 	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1262 		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1263 	else
1264 		ctx->dev->rdev.status_page->db_off = 1;
1265 	spin_unlock_irqrestore(&ctx->dev->lock, flags);
1266 }
1267 
1268 static int enable_qp_db(int id, void *p, void *data)
1269 {
1270 	struct c4iw_qp *qp = p;
1271 
1272 	t4_enable_wq_db(&qp->wq);
1273 	return 0;
1274 }
1275 
1276 static void resume_rc_qp(struct c4iw_qp *qp)
1277 {
1278 	spin_lock(&qp->lock);
1279 	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc, NULL);
1280 	qp->wq.sq.wq_pidx_inc = 0;
1281 	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc, NULL);
1282 	qp->wq.rq.wq_pidx_inc = 0;
1283 	spin_unlock(&qp->lock);
1284 }
1285 
1286 static void resume_a_chunk(struct uld_ctx *ctx)
1287 {
1288 	int i;
1289 	struct c4iw_qp *qp;
1290 
1291 	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1292 		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1293 				      db_fc_entry);
1294 		list_del_init(&qp->db_fc_entry);
1295 		resume_rc_qp(qp);
1296 		if (list_empty(&ctx->dev->db_fc_list))
1297 			break;
1298 	}
1299 }
1300 
1301 static void resume_queues(struct uld_ctx *ctx)
1302 {
1303 	spin_lock_irq(&ctx->dev->lock);
1304 	if (ctx->dev->db_state != STOPPED)
1305 		goto out;
1306 	ctx->dev->db_state = FLOW_CONTROL;
1307 	while (1) {
1308 		if (list_empty(&ctx->dev->db_fc_list)) {
1309 			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1310 			ctx->dev->db_state = NORMAL;
1311 			ctx->dev->rdev.stats.db_state_transitions++;
1312 			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1313 				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1314 					     NULL);
1315 			} else {
1316 				ctx->dev->rdev.status_page->db_off = 0;
1317 			}
1318 			break;
1319 		} else {
1320 			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1321 			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1322 			       DB_FC_DRAIN_THRESH)) {
1323 				resume_a_chunk(ctx);
1324 			}
1325 			if (!list_empty(&ctx->dev->db_fc_list)) {
1326 				spin_unlock_irq(&ctx->dev->lock);
1327 				if (DB_FC_RESUME_DELAY) {
1328 					set_current_state(TASK_UNINTERRUPTIBLE);
1329 					schedule_timeout(DB_FC_RESUME_DELAY);
1330 				}
1331 				spin_lock_irq(&ctx->dev->lock);
1332 				if (ctx->dev->db_state != FLOW_CONTROL)
1333 					break;
1334 			}
1335 		}
1336 	}
1337 out:
1338 	if (ctx->dev->db_state != NORMAL)
1339 		ctx->dev->rdev.stats.db_fc_interruptions++;
1340 	spin_unlock_irq(&ctx->dev->lock);
1341 }
1342 
1343 struct qp_list {
1344 	unsigned idx;
1345 	struct c4iw_qp **qps;
1346 };
1347 
1348 static int add_and_ref_qp(int id, void *p, void *data)
1349 {
1350 	struct qp_list *qp_listp = data;
1351 	struct c4iw_qp *qp = p;
1352 
1353 	c4iw_qp_add_ref(&qp->ibqp);
1354 	qp_listp->qps[qp_listp->idx++] = qp;
1355 	return 0;
1356 }
1357 
1358 static int count_qps(int id, void *p, void *data)
1359 {
1360 	unsigned *countp = data;
1361 	(*countp)++;
1362 	return 0;
1363 }
1364 
1365 static void deref_qps(struct qp_list *qp_list)
1366 {
1367 	int idx;
1368 
1369 	for (idx = 0; idx < qp_list->idx; idx++)
1370 		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1371 }
1372 
1373 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1374 {
1375 	int idx;
1376 	int ret;
1377 
1378 	for (idx = 0; idx < qp_list->idx; idx++) {
1379 		struct c4iw_qp *qp = qp_list->qps[idx];
1380 
1381 		spin_lock_irq(&qp->rhp->lock);
1382 		spin_lock(&qp->lock);
1383 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1384 					  qp->wq.sq.qid,
1385 					  t4_sq_host_wq_pidx(&qp->wq),
1386 					  t4_sq_wq_size(&qp->wq));
1387 		if (ret) {
1388 			pr_err("%s: Fatal error - DB overflow recovery failed - error syncing SQ qid %u\n",
1389 			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1390 			spin_unlock(&qp->lock);
1391 			spin_unlock_irq(&qp->rhp->lock);
1392 			return;
1393 		}
1394 		qp->wq.sq.wq_pidx_inc = 0;
1395 
1396 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1397 					  qp->wq.rq.qid,
1398 					  t4_rq_host_wq_pidx(&qp->wq),
1399 					  t4_rq_wq_size(&qp->wq));
1400 
1401 		if (ret) {
1402 			pr_err("%s: Fatal error - DB overflow recovery failed - error syncing RQ qid %u\n",
1403 			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1404 			spin_unlock(&qp->lock);
1405 			spin_unlock_irq(&qp->rhp->lock);
1406 			return;
1407 		}
1408 		qp->wq.rq.wq_pidx_inc = 0;
1409 		spin_unlock(&qp->lock);
1410 		spin_unlock_irq(&qp->rhp->lock);
1411 
1412 		/* Wait for the dbfifo to drain */
1413 		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1414 			set_current_state(TASK_UNINTERRUPTIBLE);
1415 			schedule_timeout(usecs_to_jiffies(10));
1416 		}
1417 	}
1418 }
1419 
1420 static void recover_queues(struct uld_ctx *ctx)
1421 {
1422 	int count = 0;
1423 	struct qp_list qp_list;
1424 	int ret;
1425 
1426 	/* slow everybody down */
1427 	set_current_state(TASK_UNINTERRUPTIBLE);
1428 	schedule_timeout(usecs_to_jiffies(1000));
1429 
1430 	/* flush the SGE contexts */
1431 	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1432 	if (ret) {
1433 		pr_err("%s: Fatal error - DB overflow recovery failed\n",
1434 		       pci_name(ctx->lldi.pdev));
1435 		return;
1436 	}
1437 
1438 	/* Count active queues so we can build a list of queues to recover */
1439 	spin_lock_irq(&ctx->dev->lock);
1440 	WARN_ON(ctx->dev->db_state != STOPPED);
1441 	ctx->dev->db_state = RECOVERY;
1442 	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1443 
1444 	qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1445 	if (!qp_list.qps) {
1446 		spin_unlock_irq(&ctx->dev->lock);
1447 		return;
1448 	}
1449 	qp_list.idx = 0;
1450 
1451 	/* add and ref each qp so it doesn't get freed */
1452 	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1453 
1454 	spin_unlock_irq(&ctx->dev->lock);
1455 
1456 	/* now traverse the list in a safe context to recover the db state*/
1457 	recover_lost_dbs(ctx, &qp_list);
1458 
1459 	/* we're almost done!  deref the qps and clean up */
1460 	deref_qps(&qp_list);
1461 	kfree(qp_list.qps);
1462 
1463 	spin_lock_irq(&ctx->dev->lock);
1464 	WARN_ON(ctx->dev->db_state != RECOVERY);
1465 	ctx->dev->db_state = STOPPED;
1466 	spin_unlock_irq(&ctx->dev->lock);
1467 }
1468 
1469 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1470 {
1471 	struct uld_ctx *ctx = handle;
1472 
1473 	switch (control) {
1474 	case CXGB4_CONTROL_DB_FULL:
1475 		stop_queues(ctx);
1476 		ctx->dev->rdev.stats.db_full++;
1477 		break;
1478 	case CXGB4_CONTROL_DB_EMPTY:
1479 		resume_queues(ctx);
1480 		mutex_lock(&ctx->dev->rdev.stats.lock);
1481 		ctx->dev->rdev.stats.db_empty++;
1482 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1483 		break;
1484 	case CXGB4_CONTROL_DB_DROP:
1485 		recover_queues(ctx);
1486 		mutex_lock(&ctx->dev->rdev.stats.lock);
1487 		ctx->dev->rdev.stats.db_drop++;
1488 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1489 		break;
1490 	default:
1491 		pr_warn("%s: unknown control cmd %u\n",
1492 			pci_name(ctx->lldi.pdev), control);
1493 		break;
1494 	}
1495 	return 0;
1496 }
1497 
1498 static struct cxgb4_uld_info c4iw_uld_info = {
1499 	.name = DRV_NAME,
1500 	.nrxq = MAX_ULD_QSETS,
1501 	.ntxq = MAX_ULD_QSETS,
1502 	.rxq_size = 511,
1503 	.ciq = true,
1504 	.lro = false,
1505 	.add = c4iw_uld_add,
1506 	.rx_handler = c4iw_uld_rx_handler,
1507 	.state_change = c4iw_uld_state_change,
1508 	.control = c4iw_uld_control,
1509 };
1510 
1511 void _c4iw_free_wr_wait(struct kref *kref)
1512 {
1513 	struct c4iw_wr_wait *wr_waitp;
1514 
1515 	wr_waitp = container_of(kref, struct c4iw_wr_wait, kref);
1516 	pr_debug("Free wr_wait %p\n", wr_waitp);
1517 	kfree(wr_waitp);
1518 }
1519 
1520 struct c4iw_wr_wait *c4iw_alloc_wr_wait(gfp_t gfp)
1521 {
1522 	struct c4iw_wr_wait *wr_waitp;
1523 
1524 	wr_waitp = kzalloc(sizeof(*wr_waitp), gfp);
1525 	if (wr_waitp) {
1526 		kref_init(&wr_waitp->kref);
1527 		pr_debug("wr_wait %p\n", wr_waitp);
1528 	}
1529 	return wr_waitp;
1530 }
1531 
1532 static int __init c4iw_init_module(void)
1533 {
1534 	int err;
1535 
1536 	err = c4iw_cm_init();
1537 	if (err)
1538 		return err;
1539 
1540 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1541 	if (!c4iw_debugfs_root)
1542 		pr_warn("could not create debugfs entry, continuing\n");
1543 
1544 	reg_workq = create_singlethread_workqueue("Register_iWARP_device");
1545 	if (!reg_workq) {
1546 		pr_err("Failed creating workqueue to register iwarp device\n");
1547 		return -ENOMEM;
1548 	}
1549 
1550 	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1551 
1552 	return 0;
1553 }
1554 
1555 static void __exit c4iw_exit_module(void)
1556 {
1557 	struct uld_ctx *ctx, *tmp;
1558 
1559 	mutex_lock(&dev_mutex);
1560 	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1561 		if (ctx->dev)
1562 			c4iw_remove(ctx);
1563 		kfree(ctx);
1564 	}
1565 	mutex_unlock(&dev_mutex);
1566 	flush_workqueue(reg_workq);
1567 	destroy_workqueue(reg_workq);
1568 	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1569 	c4iw_cm_term();
1570 	debugfs_remove_recursive(c4iw_debugfs_root);
1571 }
1572 
1573 module_init(c4iw_init_module);
1574 module_exit(c4iw_exit_module);
1575