1 /*
2  * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *	  copyright notice, this list of conditions and the following
16  *	  disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *	  copyright notice, this list of conditions and the following
20  *	  disclaimer in the documentation and/or other materials
21  *	  provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/moduleparam.h>
34 #include <linux/debugfs.h>
35 #include <linux/vmalloc.h>
36 #include <linux/math64.h>
37 
38 #include <rdma/ib_verbs.h>
39 
40 #include "iw_cxgb4.h"
41 
42 #define DRV_VERSION "0.1"
43 
44 MODULE_AUTHOR("Steve Wise");
45 MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver");
46 MODULE_LICENSE("Dual BSD/GPL");
47 MODULE_VERSION(DRV_VERSION);
48 
49 static int allow_db_fc_on_t5;
50 module_param(allow_db_fc_on_t5, int, 0644);
51 MODULE_PARM_DESC(allow_db_fc_on_t5,
52 		 "Allow DB Flow Control on T5 (default = 0)");
53 
54 static int allow_db_coalescing_on_t5;
55 module_param(allow_db_coalescing_on_t5, int, 0644);
56 MODULE_PARM_DESC(allow_db_coalescing_on_t5,
57 		 "Allow DB Coalescing on T5 (default = 0)");
58 
59 int c4iw_wr_log = 0;
60 module_param(c4iw_wr_log, int, 0444);
61 MODULE_PARM_DESC(c4iw_wr_log, "Enables logging of work request timing data.");
62 
63 static int c4iw_wr_log_size_order = 12;
64 module_param(c4iw_wr_log_size_order, int, 0444);
65 MODULE_PARM_DESC(c4iw_wr_log_size_order,
66 		 "Number of entries (log2) in the work request timing log.");
67 
68 struct uld_ctx {
69 	struct list_head entry;
70 	struct cxgb4_lld_info lldi;
71 	struct c4iw_dev *dev;
72 };
73 
74 static LIST_HEAD(uld_ctx_list);
75 static DEFINE_MUTEX(dev_mutex);
76 
77 #define DB_FC_RESUME_SIZE 64
78 #define DB_FC_RESUME_DELAY 1
79 #define DB_FC_DRAIN_THRESH 0
80 
81 static struct dentry *c4iw_debugfs_root;
82 
83 struct c4iw_debugfs_data {
84 	struct c4iw_dev *devp;
85 	char *buf;
86 	int bufsize;
87 	int pos;
88 };
89 
90 /* registered cxgb4 netlink callbacks */
91 static struct ibnl_client_cbs c4iw_nl_cb_table[] = {
92 	[RDMA_NL_IWPM_REG_PID] = {.dump = iwpm_register_pid_cb},
93 	[RDMA_NL_IWPM_ADD_MAPPING] = {.dump = iwpm_add_mapping_cb},
94 	[RDMA_NL_IWPM_QUERY_MAPPING] = {.dump = iwpm_add_and_query_mapping_cb},
95 	[RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb},
96 	[RDMA_NL_IWPM_REMOTE_INFO] = {.dump = iwpm_remote_info_cb},
97 	[RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb},
98 	[RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}
99 };
100 
101 static int count_idrs(int id, void *p, void *data)
102 {
103 	int *countp = data;
104 
105 	*countp = *countp + 1;
106 	return 0;
107 }
108 
109 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
110 			    loff_t *ppos)
111 {
112 	struct c4iw_debugfs_data *d = file->private_data;
113 
114 	return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
115 }
116 
117 void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
118 {
119 	struct wr_log_entry le;
120 	int idx;
121 
122 	if (!wq->rdev->wr_log)
123 		return;
124 
125 	idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
126 		(wq->rdev->wr_log_size - 1);
127 	le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
128 	getnstimeofday(&le.poll_host_ts);
129 	le.valid = 1;
130 	le.cqe_sge_ts = CQE_TS(cqe);
131 	if (SQ_TYPE(cqe)) {
132 		le.qid = wq->sq.qid;
133 		le.opcode = CQE_OPCODE(cqe);
134 		le.post_host_ts = wq->sq.sw_sq[wq->sq.cidx].host_ts;
135 		le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
136 		le.wr_id = CQE_WRID_SQ_IDX(cqe);
137 	} else {
138 		le.qid = wq->rq.qid;
139 		le.opcode = FW_RI_RECEIVE;
140 		le.post_host_ts = wq->rq.sw_rq[wq->rq.cidx].host_ts;
141 		le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
142 		le.wr_id = CQE_WRID_MSN(cqe);
143 	}
144 	wq->rdev->wr_log[idx] = le;
145 }
146 
147 static int wr_log_show(struct seq_file *seq, void *v)
148 {
149 	struct c4iw_dev *dev = seq->private;
150 	struct timespec prev_ts = {0, 0};
151 	struct wr_log_entry *lep;
152 	int prev_ts_set = 0;
153 	int idx, end;
154 
155 #define ts2ns(ts) div64_u64((ts) * dev->rdev.lldi.cclk_ps, 1000)
156 
157 	idx = atomic_read(&dev->rdev.wr_log_idx) &
158 		(dev->rdev.wr_log_size - 1);
159 	end = idx - 1;
160 	if (end < 0)
161 		end = dev->rdev.wr_log_size - 1;
162 	lep = &dev->rdev.wr_log[idx];
163 	while (idx != end) {
164 		if (lep->valid) {
165 			if (!prev_ts_set) {
166 				prev_ts_set = 1;
167 				prev_ts = lep->poll_host_ts;
168 			}
169 			seq_printf(seq, "%04u: sec %lu nsec %lu qid %u opcode "
170 				   "%u %s 0x%x host_wr_delta sec %lu nsec %lu "
171 				   "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
172 				   "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
173 				   "cqe_poll_delta_ns %llu\n",
174 				   idx,
175 				   timespec_sub(lep->poll_host_ts,
176 						prev_ts).tv_sec,
177 				   timespec_sub(lep->poll_host_ts,
178 						prev_ts).tv_nsec,
179 				   lep->qid, lep->opcode,
180 				   lep->opcode == FW_RI_RECEIVE ?
181 							"msn" : "wrid",
182 				   lep->wr_id,
183 				   timespec_sub(lep->poll_host_ts,
184 						lep->post_host_ts).tv_sec,
185 				   timespec_sub(lep->poll_host_ts,
186 						lep->post_host_ts).tv_nsec,
187 				   lep->post_sge_ts, lep->cqe_sge_ts,
188 				   lep->poll_sge_ts,
189 				   ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
190 				   ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
191 			prev_ts = lep->poll_host_ts;
192 		}
193 		idx++;
194 		if (idx > (dev->rdev.wr_log_size - 1))
195 			idx = 0;
196 		lep = &dev->rdev.wr_log[idx];
197 	}
198 #undef ts2ns
199 	return 0;
200 }
201 
202 static int wr_log_open(struct inode *inode, struct file *file)
203 {
204 	return single_open(file, wr_log_show, inode->i_private);
205 }
206 
207 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
208 			    size_t count, loff_t *pos)
209 {
210 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
211 	int i;
212 
213 	if (dev->rdev.wr_log)
214 		for (i = 0; i < dev->rdev.wr_log_size; i++)
215 			dev->rdev.wr_log[i].valid = 0;
216 	return count;
217 }
218 
219 static const struct file_operations wr_log_debugfs_fops = {
220 	.owner   = THIS_MODULE,
221 	.open    = wr_log_open,
222 	.release = single_release,
223 	.read    = seq_read,
224 	.llseek  = seq_lseek,
225 	.write   = wr_log_clear,
226 };
227 
228 static int dump_qp(int id, void *p, void *data)
229 {
230 	struct c4iw_qp *qp = p;
231 	struct c4iw_debugfs_data *qpd = data;
232 	int space;
233 	int cc;
234 
235 	if (id != qp->wq.sq.qid)
236 		return 0;
237 
238 	space = qpd->bufsize - qpd->pos - 1;
239 	if (space == 0)
240 		return 1;
241 
242 	if (qp->ep) {
243 		if (qp->ep->com.local_addr.ss_family == AF_INET) {
244 			struct sockaddr_in *lsin = (struct sockaddr_in *)
245 				&qp->ep->com.local_addr;
246 			struct sockaddr_in *rsin = (struct sockaddr_in *)
247 				&qp->ep->com.remote_addr;
248 			struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
249 				&qp->ep->com.mapped_local_addr;
250 			struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
251 				&qp->ep->com.mapped_remote_addr;
252 
253 			cc = snprintf(qpd->buf + qpd->pos, space,
254 				      "rc qp sq id %u rq id %u state %u "
255 				      "onchip %u ep tid %u state %u "
256 				      "%pI4:%u/%u->%pI4:%u/%u\n",
257 				      qp->wq.sq.qid, qp->wq.rq.qid,
258 				      (int)qp->attr.state,
259 				      qp->wq.sq.flags & T4_SQ_ONCHIP,
260 				      qp->ep->hwtid, (int)qp->ep->com.state,
261 				      &lsin->sin_addr, ntohs(lsin->sin_port),
262 				      ntohs(mapped_lsin->sin_port),
263 				      &rsin->sin_addr, ntohs(rsin->sin_port),
264 				      ntohs(mapped_rsin->sin_port));
265 		} else {
266 			struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
267 				&qp->ep->com.local_addr;
268 			struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
269 				&qp->ep->com.remote_addr;
270 			struct sockaddr_in6 *mapped_lsin6 =
271 				(struct sockaddr_in6 *)
272 				&qp->ep->com.mapped_local_addr;
273 			struct sockaddr_in6 *mapped_rsin6 =
274 				(struct sockaddr_in6 *)
275 				&qp->ep->com.mapped_remote_addr;
276 
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 				      "%pI6:%u/%u->%pI6:%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 				      qp->ep->hwtid, (int)qp->ep->com.state,
285 				      &lsin6->sin6_addr,
286 				      ntohs(lsin6->sin6_port),
287 				      ntohs(mapped_lsin6->sin6_port),
288 				      &rsin6->sin6_addr,
289 				      ntohs(rsin6->sin6_port),
290 				      ntohs(mapped_rsin6->sin6_port));
291 		}
292 	} else
293 		cc = snprintf(qpd->buf + qpd->pos, space,
294 			     "qp sq id %u rq id %u state %u onchip %u\n",
295 			      qp->wq.sq.qid, qp->wq.rq.qid,
296 			      (int)qp->attr.state,
297 			      qp->wq.sq.flags & T4_SQ_ONCHIP);
298 	if (cc < space)
299 		qpd->pos += cc;
300 	return 0;
301 }
302 
303 static int qp_release(struct inode *inode, struct file *file)
304 {
305 	struct c4iw_debugfs_data *qpd = file->private_data;
306 	if (!qpd) {
307 		printk(KERN_INFO "%s null qpd?\n", __func__);
308 		return 0;
309 	}
310 	vfree(qpd->buf);
311 	kfree(qpd);
312 	return 0;
313 }
314 
315 static int qp_open(struct inode *inode, struct file *file)
316 {
317 	struct c4iw_debugfs_data *qpd;
318 	int count = 1;
319 
320 	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
321 	if (!qpd)
322 		return -ENOMEM;
323 
324 	qpd->devp = inode->i_private;
325 	qpd->pos = 0;
326 
327 	spin_lock_irq(&qpd->devp->lock);
328 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
329 	spin_unlock_irq(&qpd->devp->lock);
330 
331 	qpd->bufsize = count * 128;
332 	qpd->buf = vmalloc(qpd->bufsize);
333 	if (!qpd->buf) {
334 		kfree(qpd);
335 		return -ENOMEM;
336 	}
337 
338 	spin_lock_irq(&qpd->devp->lock);
339 	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
340 	spin_unlock_irq(&qpd->devp->lock);
341 
342 	qpd->buf[qpd->pos++] = 0;
343 	file->private_data = qpd;
344 	return 0;
345 }
346 
347 static const struct file_operations qp_debugfs_fops = {
348 	.owner   = THIS_MODULE,
349 	.open    = qp_open,
350 	.release = qp_release,
351 	.read    = debugfs_read,
352 	.llseek  = default_llseek,
353 };
354 
355 static int dump_stag(int id, void *p, void *data)
356 {
357 	struct c4iw_debugfs_data *stagd = data;
358 	int space;
359 	int cc;
360 	struct fw_ri_tpte tpte;
361 	int ret;
362 
363 	space = stagd->bufsize - stagd->pos - 1;
364 	if (space == 0)
365 		return 1;
366 
367 	ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
368 			      (__be32 *)&tpte);
369 	if (ret) {
370 		dev_err(&stagd->devp->rdev.lldi.pdev->dev,
371 			"%s cxgb4_read_tpte err %d\n", __func__, ret);
372 		return ret;
373 	}
374 	cc = snprintf(stagd->buf + stagd->pos, space,
375 		      "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
376 		      "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
377 		      (u32)id<<8,
378 		      FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
379 		      FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
380 		      FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
381 		      FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
382 		      FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
383 		      FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
384 		      ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
385 		      ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
386 	if (cc < space)
387 		stagd->pos += cc;
388 	return 0;
389 }
390 
391 static int stag_release(struct inode *inode, struct file *file)
392 {
393 	struct c4iw_debugfs_data *stagd = file->private_data;
394 	if (!stagd) {
395 		printk(KERN_INFO "%s null stagd?\n", __func__);
396 		return 0;
397 	}
398 	vfree(stagd->buf);
399 	kfree(stagd);
400 	return 0;
401 }
402 
403 static int stag_open(struct inode *inode, struct file *file)
404 {
405 	struct c4iw_debugfs_data *stagd;
406 	int ret = 0;
407 	int count = 1;
408 
409 	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
410 	if (!stagd) {
411 		ret = -ENOMEM;
412 		goto out;
413 	}
414 	stagd->devp = inode->i_private;
415 	stagd->pos = 0;
416 
417 	spin_lock_irq(&stagd->devp->lock);
418 	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
419 	spin_unlock_irq(&stagd->devp->lock);
420 
421 	stagd->bufsize = count * 256;
422 	stagd->buf = vmalloc(stagd->bufsize);
423 	if (!stagd->buf) {
424 		ret = -ENOMEM;
425 		goto err1;
426 	}
427 
428 	spin_lock_irq(&stagd->devp->lock);
429 	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
430 	spin_unlock_irq(&stagd->devp->lock);
431 
432 	stagd->buf[stagd->pos++] = 0;
433 	file->private_data = stagd;
434 	goto out;
435 err1:
436 	kfree(stagd);
437 out:
438 	return ret;
439 }
440 
441 static const struct file_operations stag_debugfs_fops = {
442 	.owner   = THIS_MODULE,
443 	.open    = stag_open,
444 	.release = stag_release,
445 	.read    = debugfs_read,
446 	.llseek  = default_llseek,
447 };
448 
449 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
450 
451 static int stats_show(struct seq_file *seq, void *v)
452 {
453 	struct c4iw_dev *dev = seq->private;
454 
455 	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
456 		   "Max", "Fail");
457 	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
458 			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
459 			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
460 	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
461 			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
462 			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
463 	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
464 			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
465 			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
466 	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
467 			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
468 			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
469 	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
470 			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
471 			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
472 	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
473 			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
474 			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
475 	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
476 	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
477 	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
478 	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
479 		   db_state_str[dev->db_state],
480 		   dev->rdev.stats.db_state_transitions,
481 		   dev->rdev.stats.db_fc_interruptions);
482 	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
483 	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
484 		   dev->rdev.stats.act_ofld_conn_fails);
485 	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
486 		   dev->rdev.stats.pas_ofld_conn_fails);
487 	seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
488 	seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
489 	return 0;
490 }
491 
492 static int stats_open(struct inode *inode, struct file *file)
493 {
494 	return single_open(file, stats_show, inode->i_private);
495 }
496 
497 static ssize_t stats_clear(struct file *file, const char __user *buf,
498 		size_t count, loff_t *pos)
499 {
500 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
501 
502 	mutex_lock(&dev->rdev.stats.lock);
503 	dev->rdev.stats.pd.max = 0;
504 	dev->rdev.stats.pd.fail = 0;
505 	dev->rdev.stats.qid.max = 0;
506 	dev->rdev.stats.qid.fail = 0;
507 	dev->rdev.stats.stag.max = 0;
508 	dev->rdev.stats.stag.fail = 0;
509 	dev->rdev.stats.pbl.max = 0;
510 	dev->rdev.stats.pbl.fail = 0;
511 	dev->rdev.stats.rqt.max = 0;
512 	dev->rdev.stats.rqt.fail = 0;
513 	dev->rdev.stats.ocqp.max = 0;
514 	dev->rdev.stats.ocqp.fail = 0;
515 	dev->rdev.stats.db_full = 0;
516 	dev->rdev.stats.db_empty = 0;
517 	dev->rdev.stats.db_drop = 0;
518 	dev->rdev.stats.db_state_transitions = 0;
519 	dev->rdev.stats.tcam_full = 0;
520 	dev->rdev.stats.act_ofld_conn_fails = 0;
521 	dev->rdev.stats.pas_ofld_conn_fails = 0;
522 	mutex_unlock(&dev->rdev.stats.lock);
523 	return count;
524 }
525 
526 static const struct file_operations stats_debugfs_fops = {
527 	.owner   = THIS_MODULE,
528 	.open    = stats_open,
529 	.release = single_release,
530 	.read    = seq_read,
531 	.llseek  = seq_lseek,
532 	.write   = stats_clear,
533 };
534 
535 static int dump_ep(int id, void *p, void *data)
536 {
537 	struct c4iw_ep *ep = p;
538 	struct c4iw_debugfs_data *epd = data;
539 	int space;
540 	int cc;
541 
542 	space = epd->bufsize - epd->pos - 1;
543 	if (space == 0)
544 		return 1;
545 
546 	if (ep->com.local_addr.ss_family == AF_INET) {
547 		struct sockaddr_in *lsin = (struct sockaddr_in *)
548 			&ep->com.local_addr;
549 		struct sockaddr_in *rsin = (struct sockaddr_in *)
550 			&ep->com.remote_addr;
551 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
552 			&ep->com.mapped_local_addr;
553 		struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
554 			&ep->com.mapped_remote_addr;
555 
556 		cc = snprintf(epd->buf + epd->pos, space,
557 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
558 			      "history 0x%lx hwtid %d atid %d "
559 			      "conn_na %u abort_na %u "
560 			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
561 			      ep, ep->com.cm_id, ep->com.qp,
562 			      (int)ep->com.state, ep->com.flags,
563 			      ep->com.history, ep->hwtid, ep->atid,
564 			      ep->stats.connect_neg_adv,
565 			      ep->stats.abort_neg_adv,
566 			      &lsin->sin_addr, ntohs(lsin->sin_port),
567 			      ntohs(mapped_lsin->sin_port),
568 			      &rsin->sin_addr, ntohs(rsin->sin_port),
569 			      ntohs(mapped_rsin->sin_port));
570 	} else {
571 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
572 			&ep->com.local_addr;
573 		struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
574 			&ep->com.remote_addr;
575 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
576 			&ep->com.mapped_local_addr;
577 		struct sockaddr_in6 *mapped_rsin6 = (struct sockaddr_in6 *)
578 			&ep->com.mapped_remote_addr;
579 
580 		cc = snprintf(epd->buf + epd->pos, space,
581 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
582 			      "history 0x%lx hwtid %d atid %d "
583 			      "conn_na %u abort_na %u "
584 			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
585 			      ep, ep->com.cm_id, ep->com.qp,
586 			      (int)ep->com.state, ep->com.flags,
587 			      ep->com.history, ep->hwtid, ep->atid,
588 			      ep->stats.connect_neg_adv,
589 			      ep->stats.abort_neg_adv,
590 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
591 			      ntohs(mapped_lsin6->sin6_port),
592 			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
593 			      ntohs(mapped_rsin6->sin6_port));
594 	}
595 	if (cc < space)
596 		epd->pos += cc;
597 	return 0;
598 }
599 
600 static int dump_listen_ep(int id, void *p, void *data)
601 {
602 	struct c4iw_listen_ep *ep = p;
603 	struct c4iw_debugfs_data *epd = data;
604 	int space;
605 	int cc;
606 
607 	space = epd->bufsize - epd->pos - 1;
608 	if (space == 0)
609 		return 1;
610 
611 	if (ep->com.local_addr.ss_family == AF_INET) {
612 		struct sockaddr_in *lsin = (struct sockaddr_in *)
613 			&ep->com.local_addr;
614 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
615 			&ep->com.mapped_local_addr;
616 
617 		cc = snprintf(epd->buf + epd->pos, space,
618 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
619 			      "backlog %d %pI4:%d/%d\n",
620 			      ep, ep->com.cm_id, (int)ep->com.state,
621 			      ep->com.flags, ep->stid, ep->backlog,
622 			      &lsin->sin_addr, ntohs(lsin->sin_port),
623 			      ntohs(mapped_lsin->sin_port));
624 	} else {
625 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
626 			&ep->com.local_addr;
627 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
628 			&ep->com.mapped_local_addr;
629 
630 		cc = snprintf(epd->buf + epd->pos, space,
631 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
632 			      "backlog %d %pI6:%d/%d\n",
633 			      ep, ep->com.cm_id, (int)ep->com.state,
634 			      ep->com.flags, ep->stid, ep->backlog,
635 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
636 			      ntohs(mapped_lsin6->sin6_port));
637 	}
638 	if (cc < space)
639 		epd->pos += cc;
640 	return 0;
641 }
642 
643 static int ep_release(struct inode *inode, struct file *file)
644 {
645 	struct c4iw_debugfs_data *epd = file->private_data;
646 	if (!epd) {
647 		pr_info("%s null qpd?\n", __func__);
648 		return 0;
649 	}
650 	vfree(epd->buf);
651 	kfree(epd);
652 	return 0;
653 }
654 
655 static int ep_open(struct inode *inode, struct file *file)
656 {
657 	struct c4iw_debugfs_data *epd;
658 	int ret = 0;
659 	int count = 1;
660 
661 	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
662 	if (!epd) {
663 		ret = -ENOMEM;
664 		goto out;
665 	}
666 	epd->devp = inode->i_private;
667 	epd->pos = 0;
668 
669 	spin_lock_irq(&epd->devp->lock);
670 	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
671 	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
672 	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
673 	spin_unlock_irq(&epd->devp->lock);
674 
675 	epd->bufsize = count * 240;
676 	epd->buf = vmalloc(epd->bufsize);
677 	if (!epd->buf) {
678 		ret = -ENOMEM;
679 		goto err1;
680 	}
681 
682 	spin_lock_irq(&epd->devp->lock);
683 	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
684 	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
685 	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
686 	spin_unlock_irq(&epd->devp->lock);
687 
688 	file->private_data = epd;
689 	goto out;
690 err1:
691 	kfree(epd);
692 out:
693 	return ret;
694 }
695 
696 static const struct file_operations ep_debugfs_fops = {
697 	.owner   = THIS_MODULE,
698 	.open    = ep_open,
699 	.release = ep_release,
700 	.read    = debugfs_read,
701 };
702 
703 static int setup_debugfs(struct c4iw_dev *devp)
704 {
705 	if (!devp->debugfs_root)
706 		return -1;
707 
708 	debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
709 				 (void *)devp, &qp_debugfs_fops, 4096);
710 
711 	debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
712 				 (void *)devp, &stag_debugfs_fops, 4096);
713 
714 	debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
715 				 (void *)devp, &stats_debugfs_fops, 4096);
716 
717 	debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
718 				 (void *)devp, &ep_debugfs_fops, 4096);
719 
720 	if (c4iw_wr_log)
721 		debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
722 					 (void *)devp, &wr_log_debugfs_fops, 4096);
723 	return 0;
724 }
725 
726 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
727 			       struct c4iw_dev_ucontext *uctx)
728 {
729 	struct list_head *pos, *nxt;
730 	struct c4iw_qid_list *entry;
731 
732 	mutex_lock(&uctx->lock);
733 	list_for_each_safe(pos, nxt, &uctx->qpids) {
734 		entry = list_entry(pos, struct c4iw_qid_list, entry);
735 		list_del_init(&entry->entry);
736 		if (!(entry->qid & rdev->qpmask)) {
737 			c4iw_put_resource(&rdev->resource.qid_table,
738 					  entry->qid);
739 			mutex_lock(&rdev->stats.lock);
740 			rdev->stats.qid.cur -= rdev->qpmask + 1;
741 			mutex_unlock(&rdev->stats.lock);
742 		}
743 		kfree(entry);
744 	}
745 
746 	list_for_each_safe(pos, nxt, &uctx->qpids) {
747 		entry = list_entry(pos, struct c4iw_qid_list, entry);
748 		list_del_init(&entry->entry);
749 		kfree(entry);
750 	}
751 	mutex_unlock(&uctx->lock);
752 }
753 
754 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
755 			    struct c4iw_dev_ucontext *uctx)
756 {
757 	INIT_LIST_HEAD(&uctx->qpids);
758 	INIT_LIST_HEAD(&uctx->cqids);
759 	mutex_init(&uctx->lock);
760 }
761 
762 /* Caller takes care of locking if needed */
763 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
764 {
765 	int err;
766 
767 	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
768 
769 	/*
770 	 * This implementation assumes udb_density == ucq_density!  Eventually
771 	 * we might need to support this but for now fail the open. Also the
772 	 * cqid and qpid range must match for now.
773 	 */
774 	if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
775 		pr_err(MOD "%s: unsupported udb/ucq densities %u/%u\n",
776 		       pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
777 		       rdev->lldi.ucq_density);
778 		return -EINVAL;
779 	}
780 	if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
781 	    rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
782 		pr_err(MOD "%s: unsupported qp and cq id ranges "
783 		       "qp start %u size %u cq start %u size %u\n",
784 		       pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
785 		       rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
786 		       rdev->lldi.vr->cq.size);
787 		return -EINVAL;
788 	}
789 
790 	rdev->qpmask = rdev->lldi.udb_density - 1;
791 	rdev->cqmask = rdev->lldi.ucq_density - 1;
792 	PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
793 	     "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
794 	     "qp qid start %u size %u cq qid start %u size %u\n",
795 	     __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
796 	     rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
797 	     rdev->lldi.vr->pbl.start,
798 	     rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
799 	     rdev->lldi.vr->rq.size,
800 	     rdev->lldi.vr->qp.start,
801 	     rdev->lldi.vr->qp.size,
802 	     rdev->lldi.vr->cq.start,
803 	     rdev->lldi.vr->cq.size);
804 	PDBG("udb len 0x%x udb base %p db_reg %p gts_reg %p "
805 	     "qpmask 0x%x cqmask 0x%x\n",
806 	     (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
807 	     (void *)pci_resource_start(rdev->lldi.pdev, 2),
808 	     rdev->lldi.db_reg, rdev->lldi.gts_reg,
809 	     rdev->qpmask, rdev->cqmask);
810 
811 	if (c4iw_num_stags(rdev) == 0)
812 		return -EINVAL;
813 
814 	rdev->stats.pd.total = T4_MAX_NUM_PD;
815 	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
816 	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
817 	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
818 	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
819 	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
820 
821 	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
822 	if (err) {
823 		printk(KERN_ERR MOD "error %d initializing resources\n", err);
824 		return err;
825 	}
826 	err = c4iw_pblpool_create(rdev);
827 	if (err) {
828 		printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
829 		goto destroy_resource;
830 	}
831 	err = c4iw_rqtpool_create(rdev);
832 	if (err) {
833 		printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
834 		goto destroy_pblpool;
835 	}
836 	err = c4iw_ocqp_pool_create(rdev);
837 	if (err) {
838 		printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
839 		goto destroy_rqtpool;
840 	}
841 	rdev->status_page = (struct t4_dev_status_page *)
842 			    __get_free_page(GFP_KERNEL);
843 	if (!rdev->status_page)
844 		goto destroy_ocqp_pool;
845 	rdev->status_page->qp_start = rdev->lldi.vr->qp.start;
846 	rdev->status_page->qp_size = rdev->lldi.vr->qp.size;
847 	rdev->status_page->cq_start = rdev->lldi.vr->cq.start;
848 	rdev->status_page->cq_size = rdev->lldi.vr->cq.size;
849 
850 	if (c4iw_wr_log) {
851 		rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
852 				       sizeof(*rdev->wr_log), GFP_KERNEL);
853 		if (rdev->wr_log) {
854 			rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
855 			atomic_set(&rdev->wr_log_idx, 0);
856 		} else {
857 			pr_err(MOD "error allocating wr_log. Logging disabled\n");
858 		}
859 	}
860 
861 	rdev->status_page->db_off = 0;
862 
863 	return 0;
864 destroy_ocqp_pool:
865 	c4iw_ocqp_pool_destroy(rdev);
866 destroy_rqtpool:
867 	c4iw_rqtpool_destroy(rdev);
868 destroy_pblpool:
869 	c4iw_pblpool_destroy(rdev);
870 destroy_resource:
871 	c4iw_destroy_resource(&rdev->resource);
872 	return err;
873 }
874 
875 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
876 {
877 	kfree(rdev->wr_log);
878 	free_page((unsigned long)rdev->status_page);
879 	c4iw_pblpool_destroy(rdev);
880 	c4iw_rqtpool_destroy(rdev);
881 	c4iw_destroy_resource(&rdev->resource);
882 }
883 
884 static void c4iw_dealloc(struct uld_ctx *ctx)
885 {
886 	c4iw_rdev_close(&ctx->dev->rdev);
887 	idr_destroy(&ctx->dev->cqidr);
888 	idr_destroy(&ctx->dev->qpidr);
889 	idr_destroy(&ctx->dev->mmidr);
890 	idr_destroy(&ctx->dev->hwtid_idr);
891 	idr_destroy(&ctx->dev->stid_idr);
892 	idr_destroy(&ctx->dev->atid_idr);
893 	if (ctx->dev->rdev.bar2_kva)
894 		iounmap(ctx->dev->rdev.bar2_kva);
895 	if (ctx->dev->rdev.oc_mw_kva)
896 		iounmap(ctx->dev->rdev.oc_mw_kva);
897 	ib_dealloc_device(&ctx->dev->ibdev);
898 	ctx->dev = NULL;
899 }
900 
901 static void c4iw_remove(struct uld_ctx *ctx)
902 {
903 	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
904 	c4iw_unregister_device(ctx->dev);
905 	c4iw_dealloc(ctx);
906 }
907 
908 static int rdma_supported(const struct cxgb4_lld_info *infop)
909 {
910 	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
911 	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
912 	       infop->vr->cq.size > 0;
913 }
914 
915 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
916 {
917 	struct c4iw_dev *devp;
918 	int ret;
919 
920 	if (!rdma_supported(infop)) {
921 		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
922 		       pci_name(infop->pdev));
923 		return ERR_PTR(-ENOSYS);
924 	}
925 	if (!ocqp_supported(infop))
926 		pr_info("%s: On-Chip Queues not supported on this device.\n",
927 			pci_name(infop->pdev));
928 
929 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
930 	if (!devp) {
931 		printk(KERN_ERR MOD "Cannot allocate ib device\n");
932 		return ERR_PTR(-ENOMEM);
933 	}
934 	devp->rdev.lldi = *infop;
935 
936 	/* init various hw-queue params based on lld info */
937 	PDBG("%s: Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
938 	     __func__, devp->rdev.lldi.sge_ingpadboundary,
939 	     devp->rdev.lldi.sge_egrstatuspagesize);
940 
941 	devp->rdev.hw_queue.t4_eq_status_entries =
942 		devp->rdev.lldi.sge_ingpadboundary > 64 ? 2 : 1;
943 	devp->rdev.hw_queue.t4_max_eq_size = 65520;
944 	devp->rdev.hw_queue.t4_max_iq_size = 65520;
945 	devp->rdev.hw_queue.t4_max_rq_size = 8192 -
946 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
947 	devp->rdev.hw_queue.t4_max_sq_size =
948 		devp->rdev.hw_queue.t4_max_eq_size -
949 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
950 	devp->rdev.hw_queue.t4_max_qp_depth =
951 		devp->rdev.hw_queue.t4_max_rq_size;
952 	devp->rdev.hw_queue.t4_max_cq_depth =
953 		devp->rdev.hw_queue.t4_max_iq_size - 2;
954 	devp->rdev.hw_queue.t4_stat_len =
955 		devp->rdev.lldi.sge_egrstatuspagesize;
956 
957 	/*
958 	 * For T5/T6 devices, we map all of BAR2 with WC.
959 	 * For T4 devices with onchip qp mem, we map only that part
960 	 * of BAR2 with WC.
961 	 */
962 	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
963 	if (!is_t4(devp->rdev.lldi.adapter_type)) {
964 		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
965 			pci_resource_len(devp->rdev.lldi.pdev, 2));
966 		if (!devp->rdev.bar2_kva) {
967 			pr_err(MOD "Unable to ioremap BAR2\n");
968 			ib_dealloc_device(&devp->ibdev);
969 			return ERR_PTR(-EINVAL);
970 		}
971 	} else if (ocqp_supported(infop)) {
972 		devp->rdev.oc_mw_pa =
973 			pci_resource_start(devp->rdev.lldi.pdev, 2) +
974 			pci_resource_len(devp->rdev.lldi.pdev, 2) -
975 			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
976 		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
977 			devp->rdev.lldi.vr->ocq.size);
978 		if (!devp->rdev.oc_mw_kva) {
979 			pr_err(MOD "Unable to ioremap onchip mem\n");
980 			ib_dealloc_device(&devp->ibdev);
981 			return ERR_PTR(-EINVAL);
982 		}
983 	}
984 
985 	PDBG(KERN_INFO MOD "ocq memory: "
986 	       "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
987 	       devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
988 	       devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
989 
990 	ret = c4iw_rdev_open(&devp->rdev);
991 	if (ret) {
992 		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
993 		ib_dealloc_device(&devp->ibdev);
994 		return ERR_PTR(ret);
995 	}
996 
997 	idr_init(&devp->cqidr);
998 	idr_init(&devp->qpidr);
999 	idr_init(&devp->mmidr);
1000 	idr_init(&devp->hwtid_idr);
1001 	idr_init(&devp->stid_idr);
1002 	idr_init(&devp->atid_idr);
1003 	spin_lock_init(&devp->lock);
1004 	mutex_init(&devp->rdev.stats.lock);
1005 	mutex_init(&devp->db_mutex);
1006 	INIT_LIST_HEAD(&devp->db_fc_list);
1007 	devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1008 
1009 	if (c4iw_debugfs_root) {
1010 		devp->debugfs_root = debugfs_create_dir(
1011 					pci_name(devp->rdev.lldi.pdev),
1012 					c4iw_debugfs_root);
1013 		setup_debugfs(devp);
1014 	}
1015 
1016 
1017 	return devp;
1018 }
1019 
1020 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1021 {
1022 	struct uld_ctx *ctx;
1023 	static int vers_printed;
1024 	int i;
1025 
1026 	if (!vers_printed++)
1027 		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1028 			DRV_VERSION);
1029 
1030 	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1031 	if (!ctx) {
1032 		ctx = ERR_PTR(-ENOMEM);
1033 		goto out;
1034 	}
1035 	ctx->lldi = *infop;
1036 
1037 	PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1038 	     __func__, pci_name(ctx->lldi.pdev),
1039 	     ctx->lldi.nchan, ctx->lldi.nrxq,
1040 	     ctx->lldi.ntxq, ctx->lldi.nports);
1041 
1042 	mutex_lock(&dev_mutex);
1043 	list_add_tail(&ctx->entry, &uld_ctx_list);
1044 	mutex_unlock(&dev_mutex);
1045 
1046 	for (i = 0; i < ctx->lldi.nrxq; i++)
1047 		PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1048 out:
1049 	return ctx;
1050 }
1051 
1052 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1053 						 const __be64 *rsp,
1054 						 u32 pktshift)
1055 {
1056 	struct sk_buff *skb;
1057 
1058 	/*
1059 	 * Allocate space for cpl_pass_accept_req which will be synthesized by
1060 	 * driver. Once the driver synthesizes the request the skb will go
1061 	 * through the regular cpl_pass_accept_req processing.
1062 	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1063 	 * cpl_rx_pkt.
1064 	 */
1065 	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1066 			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1067 	if (unlikely(!skb))
1068 		return NULL;
1069 
1070 	 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1071 		   sizeof(struct rss_header) - pktshift);
1072 
1073 	/*
1074 	 * This skb will contain:
1075 	 *   rss_header from the rspq descriptor (1 flit)
1076 	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1077 	 *   space for the difference between the size of an
1078 	 *      rx_pkt and pass_accept_req cpl (1 flit)
1079 	 *   the packet data from the gl
1080 	 */
1081 	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1082 				sizeof(struct rss_header));
1083 	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1084 				       sizeof(struct cpl_pass_accept_req),
1085 				       gl->va + pktshift,
1086 				       gl->tot_len - pktshift);
1087 	return skb;
1088 }
1089 
1090 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1091 			   const __be64 *rsp)
1092 {
1093 	unsigned int opcode = *(u8 *)rsp;
1094 	struct sk_buff *skb;
1095 
1096 	if (opcode != CPL_RX_PKT)
1097 		goto out;
1098 
1099 	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1100 	if (skb == NULL)
1101 		goto out;
1102 
1103 	if (c4iw_handlers[opcode] == NULL) {
1104 		pr_info("%s no handler opcode 0x%x...\n", __func__,
1105 		       opcode);
1106 		kfree_skb(skb);
1107 		goto out;
1108 	}
1109 	c4iw_handlers[opcode](dev, skb);
1110 	return 1;
1111 out:
1112 	return 0;
1113 }
1114 
1115 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1116 			const struct pkt_gl *gl)
1117 {
1118 	struct uld_ctx *ctx = handle;
1119 	struct c4iw_dev *dev = ctx->dev;
1120 	struct sk_buff *skb;
1121 	u8 opcode;
1122 
1123 	if (gl == NULL) {
1124 		/* omit RSS and rsp_ctrl at end of descriptor */
1125 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1126 
1127 		skb = alloc_skb(256, GFP_ATOMIC);
1128 		if (!skb)
1129 			goto nomem;
1130 		__skb_put(skb, len);
1131 		skb_copy_to_linear_data(skb, &rsp[1], len);
1132 	} else if (gl == CXGB4_MSG_AN) {
1133 		const struct rsp_ctrl *rc = (void *)rsp;
1134 
1135 		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1136 		c4iw_ev_handler(dev, qid);
1137 		return 0;
1138 	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1139 		if (recv_rx_pkt(dev, gl, rsp))
1140 			return 0;
1141 
1142 		pr_info("%s: unexpected FL contents at %p, " \
1143 		       "RSS %#llx, FL %#llx, len %u\n",
1144 		       pci_name(ctx->lldi.pdev), gl->va,
1145 		       (unsigned long long)be64_to_cpu(*rsp),
1146 		       (unsigned long long)be64_to_cpu(
1147 		       *(__force __be64 *)gl->va),
1148 		       gl->tot_len);
1149 
1150 		return 0;
1151 	} else {
1152 		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1153 		if (unlikely(!skb))
1154 			goto nomem;
1155 	}
1156 
1157 	opcode = *(u8 *)rsp;
1158 	if (c4iw_handlers[opcode]) {
1159 		c4iw_handlers[opcode](dev, skb);
1160 	} else {
1161 		pr_info("%s no handler opcode 0x%x...\n", __func__,
1162 		       opcode);
1163 		kfree_skb(skb);
1164 	}
1165 
1166 	return 0;
1167 nomem:
1168 	return -1;
1169 }
1170 
1171 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1172 {
1173 	struct uld_ctx *ctx = handle;
1174 
1175 	PDBG("%s new_state %u\n", __func__, new_state);
1176 	switch (new_state) {
1177 	case CXGB4_STATE_UP:
1178 		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
1179 		if (!ctx->dev) {
1180 			int ret;
1181 
1182 			ctx->dev = c4iw_alloc(&ctx->lldi);
1183 			if (IS_ERR(ctx->dev)) {
1184 				printk(KERN_ERR MOD
1185 				       "%s: initialization failed: %ld\n",
1186 				       pci_name(ctx->lldi.pdev),
1187 				       PTR_ERR(ctx->dev));
1188 				ctx->dev = NULL;
1189 				break;
1190 			}
1191 			ret = c4iw_register_device(ctx->dev);
1192 			if (ret) {
1193 				printk(KERN_ERR MOD
1194 				       "%s: RDMA registration failed: %d\n",
1195 				       pci_name(ctx->lldi.pdev), ret);
1196 				c4iw_dealloc(ctx);
1197 			}
1198 		}
1199 		break;
1200 	case CXGB4_STATE_DOWN:
1201 		printk(KERN_INFO MOD "%s: Down\n",
1202 		       pci_name(ctx->lldi.pdev));
1203 		if (ctx->dev)
1204 			c4iw_remove(ctx);
1205 		break;
1206 	case CXGB4_STATE_START_RECOVERY:
1207 		printk(KERN_INFO MOD "%s: Fatal Error\n",
1208 		       pci_name(ctx->lldi.pdev));
1209 		if (ctx->dev) {
1210 			struct ib_event event;
1211 
1212 			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1213 			memset(&event, 0, sizeof event);
1214 			event.event  = IB_EVENT_DEVICE_FATAL;
1215 			event.device = &ctx->dev->ibdev;
1216 			ib_dispatch_event(&event);
1217 			c4iw_remove(ctx);
1218 		}
1219 		break;
1220 	case CXGB4_STATE_DETACH:
1221 		printk(KERN_INFO MOD "%s: Detach\n",
1222 		       pci_name(ctx->lldi.pdev));
1223 		if (ctx->dev)
1224 			c4iw_remove(ctx);
1225 		break;
1226 	}
1227 	return 0;
1228 }
1229 
1230 static int disable_qp_db(int id, void *p, void *data)
1231 {
1232 	struct c4iw_qp *qp = p;
1233 
1234 	t4_disable_wq_db(&qp->wq);
1235 	return 0;
1236 }
1237 
1238 static void stop_queues(struct uld_ctx *ctx)
1239 {
1240 	unsigned long flags;
1241 
1242 	spin_lock_irqsave(&ctx->dev->lock, flags);
1243 	ctx->dev->rdev.stats.db_state_transitions++;
1244 	ctx->dev->db_state = STOPPED;
1245 	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1246 		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1247 	else
1248 		ctx->dev->rdev.status_page->db_off = 1;
1249 	spin_unlock_irqrestore(&ctx->dev->lock, flags);
1250 }
1251 
1252 static int enable_qp_db(int id, void *p, void *data)
1253 {
1254 	struct c4iw_qp *qp = p;
1255 
1256 	t4_enable_wq_db(&qp->wq);
1257 	return 0;
1258 }
1259 
1260 static void resume_rc_qp(struct c4iw_qp *qp)
1261 {
1262 	spin_lock(&qp->lock);
1263 	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc, NULL);
1264 	qp->wq.sq.wq_pidx_inc = 0;
1265 	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc, NULL);
1266 	qp->wq.rq.wq_pidx_inc = 0;
1267 	spin_unlock(&qp->lock);
1268 }
1269 
1270 static void resume_a_chunk(struct uld_ctx *ctx)
1271 {
1272 	int i;
1273 	struct c4iw_qp *qp;
1274 
1275 	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1276 		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1277 				      db_fc_entry);
1278 		list_del_init(&qp->db_fc_entry);
1279 		resume_rc_qp(qp);
1280 		if (list_empty(&ctx->dev->db_fc_list))
1281 			break;
1282 	}
1283 }
1284 
1285 static void resume_queues(struct uld_ctx *ctx)
1286 {
1287 	spin_lock_irq(&ctx->dev->lock);
1288 	if (ctx->dev->db_state != STOPPED)
1289 		goto out;
1290 	ctx->dev->db_state = FLOW_CONTROL;
1291 	while (1) {
1292 		if (list_empty(&ctx->dev->db_fc_list)) {
1293 			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1294 			ctx->dev->db_state = NORMAL;
1295 			ctx->dev->rdev.stats.db_state_transitions++;
1296 			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1297 				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1298 					     NULL);
1299 			} else {
1300 				ctx->dev->rdev.status_page->db_off = 0;
1301 			}
1302 			break;
1303 		} else {
1304 			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1305 			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1306 			       DB_FC_DRAIN_THRESH)) {
1307 				resume_a_chunk(ctx);
1308 			}
1309 			if (!list_empty(&ctx->dev->db_fc_list)) {
1310 				spin_unlock_irq(&ctx->dev->lock);
1311 				if (DB_FC_RESUME_DELAY) {
1312 					set_current_state(TASK_UNINTERRUPTIBLE);
1313 					schedule_timeout(DB_FC_RESUME_DELAY);
1314 				}
1315 				spin_lock_irq(&ctx->dev->lock);
1316 				if (ctx->dev->db_state != FLOW_CONTROL)
1317 					break;
1318 			}
1319 		}
1320 	}
1321 out:
1322 	if (ctx->dev->db_state != NORMAL)
1323 		ctx->dev->rdev.stats.db_fc_interruptions++;
1324 	spin_unlock_irq(&ctx->dev->lock);
1325 }
1326 
1327 struct qp_list {
1328 	unsigned idx;
1329 	struct c4iw_qp **qps;
1330 };
1331 
1332 static int add_and_ref_qp(int id, void *p, void *data)
1333 {
1334 	struct qp_list *qp_listp = data;
1335 	struct c4iw_qp *qp = p;
1336 
1337 	c4iw_qp_add_ref(&qp->ibqp);
1338 	qp_listp->qps[qp_listp->idx++] = qp;
1339 	return 0;
1340 }
1341 
1342 static int count_qps(int id, void *p, void *data)
1343 {
1344 	unsigned *countp = data;
1345 	(*countp)++;
1346 	return 0;
1347 }
1348 
1349 static void deref_qps(struct qp_list *qp_list)
1350 {
1351 	int idx;
1352 
1353 	for (idx = 0; idx < qp_list->idx; idx++)
1354 		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1355 }
1356 
1357 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1358 {
1359 	int idx;
1360 	int ret;
1361 
1362 	for (idx = 0; idx < qp_list->idx; idx++) {
1363 		struct c4iw_qp *qp = qp_list->qps[idx];
1364 
1365 		spin_lock_irq(&qp->rhp->lock);
1366 		spin_lock(&qp->lock);
1367 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1368 					  qp->wq.sq.qid,
1369 					  t4_sq_host_wq_pidx(&qp->wq),
1370 					  t4_sq_wq_size(&qp->wq));
1371 		if (ret) {
1372 			pr_err(MOD "%s: Fatal error - "
1373 			       "DB overflow recovery failed - "
1374 			       "error syncing SQ qid %u\n",
1375 			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1376 			spin_unlock(&qp->lock);
1377 			spin_unlock_irq(&qp->rhp->lock);
1378 			return;
1379 		}
1380 		qp->wq.sq.wq_pidx_inc = 0;
1381 
1382 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1383 					  qp->wq.rq.qid,
1384 					  t4_rq_host_wq_pidx(&qp->wq),
1385 					  t4_rq_wq_size(&qp->wq));
1386 
1387 		if (ret) {
1388 			pr_err(MOD "%s: Fatal error - "
1389 			       "DB overflow recovery failed - "
1390 			       "error syncing RQ qid %u\n",
1391 			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1392 			spin_unlock(&qp->lock);
1393 			spin_unlock_irq(&qp->rhp->lock);
1394 			return;
1395 		}
1396 		qp->wq.rq.wq_pidx_inc = 0;
1397 		spin_unlock(&qp->lock);
1398 		spin_unlock_irq(&qp->rhp->lock);
1399 
1400 		/* Wait for the dbfifo to drain */
1401 		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1402 			set_current_state(TASK_UNINTERRUPTIBLE);
1403 			schedule_timeout(usecs_to_jiffies(10));
1404 		}
1405 	}
1406 }
1407 
1408 static void recover_queues(struct uld_ctx *ctx)
1409 {
1410 	int count = 0;
1411 	struct qp_list qp_list;
1412 	int ret;
1413 
1414 	/* slow everybody down */
1415 	set_current_state(TASK_UNINTERRUPTIBLE);
1416 	schedule_timeout(usecs_to_jiffies(1000));
1417 
1418 	/* flush the SGE contexts */
1419 	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1420 	if (ret) {
1421 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1422 		       pci_name(ctx->lldi.pdev));
1423 		return;
1424 	}
1425 
1426 	/* Count active queues so we can build a list of queues to recover */
1427 	spin_lock_irq(&ctx->dev->lock);
1428 	WARN_ON(ctx->dev->db_state != STOPPED);
1429 	ctx->dev->db_state = RECOVERY;
1430 	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1431 
1432 	qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1433 	if (!qp_list.qps) {
1434 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1435 		       pci_name(ctx->lldi.pdev));
1436 		spin_unlock_irq(&ctx->dev->lock);
1437 		return;
1438 	}
1439 	qp_list.idx = 0;
1440 
1441 	/* add and ref each qp so it doesn't get freed */
1442 	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1443 
1444 	spin_unlock_irq(&ctx->dev->lock);
1445 
1446 	/* now traverse the list in a safe context to recover the db state*/
1447 	recover_lost_dbs(ctx, &qp_list);
1448 
1449 	/* we're almost done!  deref the qps and clean up */
1450 	deref_qps(&qp_list);
1451 	kfree(qp_list.qps);
1452 
1453 	spin_lock_irq(&ctx->dev->lock);
1454 	WARN_ON(ctx->dev->db_state != RECOVERY);
1455 	ctx->dev->db_state = STOPPED;
1456 	spin_unlock_irq(&ctx->dev->lock);
1457 }
1458 
1459 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1460 {
1461 	struct uld_ctx *ctx = handle;
1462 
1463 	switch (control) {
1464 	case CXGB4_CONTROL_DB_FULL:
1465 		stop_queues(ctx);
1466 		ctx->dev->rdev.stats.db_full++;
1467 		break;
1468 	case CXGB4_CONTROL_DB_EMPTY:
1469 		resume_queues(ctx);
1470 		mutex_lock(&ctx->dev->rdev.stats.lock);
1471 		ctx->dev->rdev.stats.db_empty++;
1472 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1473 		break;
1474 	case CXGB4_CONTROL_DB_DROP:
1475 		recover_queues(ctx);
1476 		mutex_lock(&ctx->dev->rdev.stats.lock);
1477 		ctx->dev->rdev.stats.db_drop++;
1478 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1479 		break;
1480 	default:
1481 		printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
1482 		       pci_name(ctx->lldi.pdev), control);
1483 		break;
1484 	}
1485 	return 0;
1486 }
1487 
1488 static struct cxgb4_uld_info c4iw_uld_info = {
1489 	.name = DRV_NAME,
1490 	.add = c4iw_uld_add,
1491 	.rx_handler = c4iw_uld_rx_handler,
1492 	.state_change = c4iw_uld_state_change,
1493 	.control = c4iw_uld_control,
1494 };
1495 
1496 static int __init c4iw_init_module(void)
1497 {
1498 	int err;
1499 
1500 	err = c4iw_cm_init();
1501 	if (err)
1502 		return err;
1503 
1504 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1505 	if (!c4iw_debugfs_root)
1506 		printk(KERN_WARNING MOD
1507 		       "could not create debugfs entry, continuing\n");
1508 
1509 	if (ibnl_add_client(RDMA_NL_C4IW, RDMA_NL_IWPM_NUM_OPS,
1510 			    c4iw_nl_cb_table))
1511 		pr_err("%s[%u]: Failed to add netlink callback\n"
1512 		       , __func__, __LINE__);
1513 
1514 	err = iwpm_init(RDMA_NL_C4IW);
1515 	if (err) {
1516 		pr_err("port mapper initialization failed with %d\n", err);
1517 		ibnl_remove_client(RDMA_NL_C4IW);
1518 		c4iw_cm_term();
1519 		debugfs_remove_recursive(c4iw_debugfs_root);
1520 		return err;
1521 	}
1522 
1523 	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1524 
1525 	return 0;
1526 }
1527 
1528 static void __exit c4iw_exit_module(void)
1529 {
1530 	struct uld_ctx *ctx, *tmp;
1531 
1532 	mutex_lock(&dev_mutex);
1533 	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1534 		if (ctx->dev)
1535 			c4iw_remove(ctx);
1536 		kfree(ctx);
1537 	}
1538 	mutex_unlock(&dev_mutex);
1539 	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1540 	iwpm_exit(RDMA_NL_C4IW);
1541 	ibnl_remove_client(RDMA_NL_C4IW);
1542 	c4iw_cm_term();
1543 	debugfs_remove_recursive(c4iw_debugfs_root);
1544 }
1545 
1546 module_init(c4iw_init_module);
1547 module_exit(c4iw_exit_module);
1548