xref: /linux/drivers/infiniband/hw/cxgb4/device.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 ret = 0;
319 	int count = 1;
320 
321 	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
322 	if (!qpd) {
323 		ret = -ENOMEM;
324 		goto out;
325 	}
326 	qpd->devp = inode->i_private;
327 	qpd->pos = 0;
328 
329 	spin_lock_irq(&qpd->devp->lock);
330 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
331 	spin_unlock_irq(&qpd->devp->lock);
332 
333 	qpd->bufsize = count * 128;
334 	qpd->buf = vmalloc(qpd->bufsize);
335 	if (!qpd->buf) {
336 		ret = -ENOMEM;
337 		goto err1;
338 	}
339 
340 	spin_lock_irq(&qpd->devp->lock);
341 	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
342 	spin_unlock_irq(&qpd->devp->lock);
343 
344 	qpd->buf[qpd->pos++] = 0;
345 	file->private_data = qpd;
346 	goto out;
347 err1:
348 	kfree(qpd);
349 out:
350 	return ret;
351 }
352 
353 static const struct file_operations qp_debugfs_fops = {
354 	.owner   = THIS_MODULE,
355 	.open    = qp_open,
356 	.release = qp_release,
357 	.read    = debugfs_read,
358 	.llseek  = default_llseek,
359 };
360 
361 static int dump_stag(int id, void *p, void *data)
362 {
363 	struct c4iw_debugfs_data *stagd = data;
364 	int space;
365 	int cc;
366 	struct fw_ri_tpte tpte;
367 	int ret;
368 
369 	space = stagd->bufsize - stagd->pos - 1;
370 	if (space == 0)
371 		return 1;
372 
373 	ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
374 			      (__be32 *)&tpte);
375 	if (ret) {
376 		dev_err(&stagd->devp->rdev.lldi.pdev->dev,
377 			"%s cxgb4_read_tpte err %d\n", __func__, ret);
378 		return ret;
379 	}
380 	cc = snprintf(stagd->buf + stagd->pos, space,
381 		      "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
382 		      "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
383 		      (u32)id<<8,
384 		      FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
385 		      FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
386 		      FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
387 		      FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
388 		      FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
389 		      FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
390 		      ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
391 		      ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
392 	if (cc < space)
393 		stagd->pos += cc;
394 	return 0;
395 }
396 
397 static int stag_release(struct inode *inode, struct file *file)
398 {
399 	struct c4iw_debugfs_data *stagd = file->private_data;
400 	if (!stagd) {
401 		printk(KERN_INFO "%s null stagd?\n", __func__);
402 		return 0;
403 	}
404 	vfree(stagd->buf);
405 	kfree(stagd);
406 	return 0;
407 }
408 
409 static int stag_open(struct inode *inode, struct file *file)
410 {
411 	struct c4iw_debugfs_data *stagd;
412 	int ret = 0;
413 	int count = 1;
414 
415 	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
416 	if (!stagd) {
417 		ret = -ENOMEM;
418 		goto out;
419 	}
420 	stagd->devp = inode->i_private;
421 	stagd->pos = 0;
422 
423 	spin_lock_irq(&stagd->devp->lock);
424 	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
425 	spin_unlock_irq(&stagd->devp->lock);
426 
427 	stagd->bufsize = count * 256;
428 	stagd->buf = vmalloc(stagd->bufsize);
429 	if (!stagd->buf) {
430 		ret = -ENOMEM;
431 		goto err1;
432 	}
433 
434 	spin_lock_irq(&stagd->devp->lock);
435 	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
436 	spin_unlock_irq(&stagd->devp->lock);
437 
438 	stagd->buf[stagd->pos++] = 0;
439 	file->private_data = stagd;
440 	goto out;
441 err1:
442 	kfree(stagd);
443 out:
444 	return ret;
445 }
446 
447 static const struct file_operations stag_debugfs_fops = {
448 	.owner   = THIS_MODULE,
449 	.open    = stag_open,
450 	.release = stag_release,
451 	.read    = debugfs_read,
452 	.llseek  = default_llseek,
453 };
454 
455 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
456 
457 static int stats_show(struct seq_file *seq, void *v)
458 {
459 	struct c4iw_dev *dev = seq->private;
460 
461 	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
462 		   "Max", "Fail");
463 	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
464 			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
465 			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
466 	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
467 			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
468 			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
469 	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
470 			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
471 			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
472 	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
473 			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
474 			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
475 	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
476 			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
477 			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
478 	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
479 			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
480 			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
481 	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
482 	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
483 	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
484 	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
485 		   db_state_str[dev->db_state],
486 		   dev->rdev.stats.db_state_transitions,
487 		   dev->rdev.stats.db_fc_interruptions);
488 	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
489 	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
490 		   dev->rdev.stats.act_ofld_conn_fails);
491 	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
492 		   dev->rdev.stats.pas_ofld_conn_fails);
493 	seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
494 	seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
495 	return 0;
496 }
497 
498 static int stats_open(struct inode *inode, struct file *file)
499 {
500 	return single_open(file, stats_show, inode->i_private);
501 }
502 
503 static ssize_t stats_clear(struct file *file, const char __user *buf,
504 		size_t count, loff_t *pos)
505 {
506 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
507 
508 	mutex_lock(&dev->rdev.stats.lock);
509 	dev->rdev.stats.pd.max = 0;
510 	dev->rdev.stats.pd.fail = 0;
511 	dev->rdev.stats.qid.max = 0;
512 	dev->rdev.stats.qid.fail = 0;
513 	dev->rdev.stats.stag.max = 0;
514 	dev->rdev.stats.stag.fail = 0;
515 	dev->rdev.stats.pbl.max = 0;
516 	dev->rdev.stats.pbl.fail = 0;
517 	dev->rdev.stats.rqt.max = 0;
518 	dev->rdev.stats.rqt.fail = 0;
519 	dev->rdev.stats.ocqp.max = 0;
520 	dev->rdev.stats.ocqp.fail = 0;
521 	dev->rdev.stats.db_full = 0;
522 	dev->rdev.stats.db_empty = 0;
523 	dev->rdev.stats.db_drop = 0;
524 	dev->rdev.stats.db_state_transitions = 0;
525 	dev->rdev.stats.tcam_full = 0;
526 	dev->rdev.stats.act_ofld_conn_fails = 0;
527 	dev->rdev.stats.pas_ofld_conn_fails = 0;
528 	mutex_unlock(&dev->rdev.stats.lock);
529 	return count;
530 }
531 
532 static const struct file_operations stats_debugfs_fops = {
533 	.owner   = THIS_MODULE,
534 	.open    = stats_open,
535 	.release = single_release,
536 	.read    = seq_read,
537 	.llseek  = seq_lseek,
538 	.write   = stats_clear,
539 };
540 
541 static int dump_ep(int id, void *p, void *data)
542 {
543 	struct c4iw_ep *ep = p;
544 	struct c4iw_debugfs_data *epd = data;
545 	int space;
546 	int cc;
547 
548 	space = epd->bufsize - epd->pos - 1;
549 	if (space == 0)
550 		return 1;
551 
552 	if (ep->com.local_addr.ss_family == AF_INET) {
553 		struct sockaddr_in *lsin = (struct sockaddr_in *)
554 			&ep->com.local_addr;
555 		struct sockaddr_in *rsin = (struct sockaddr_in *)
556 			&ep->com.remote_addr;
557 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
558 			&ep->com.mapped_local_addr;
559 		struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
560 			&ep->com.mapped_remote_addr;
561 
562 		cc = snprintf(epd->buf + epd->pos, space,
563 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
564 			      "history 0x%lx hwtid %d atid %d "
565 			      "conn_na %u abort_na %u "
566 			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
567 			      ep, ep->com.cm_id, ep->com.qp,
568 			      (int)ep->com.state, ep->com.flags,
569 			      ep->com.history, ep->hwtid, ep->atid,
570 			      ep->stats.connect_neg_adv,
571 			      ep->stats.abort_neg_adv,
572 			      &lsin->sin_addr, ntohs(lsin->sin_port),
573 			      ntohs(mapped_lsin->sin_port),
574 			      &rsin->sin_addr, ntohs(rsin->sin_port),
575 			      ntohs(mapped_rsin->sin_port));
576 	} else {
577 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
578 			&ep->com.local_addr;
579 		struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
580 			&ep->com.remote_addr;
581 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
582 			&ep->com.mapped_local_addr;
583 		struct sockaddr_in6 *mapped_rsin6 = (struct sockaddr_in6 *)
584 			&ep->com.mapped_remote_addr;
585 
586 		cc = snprintf(epd->buf + epd->pos, space,
587 			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
588 			      "history 0x%lx hwtid %d atid %d "
589 			      "conn_na %u abort_na %u "
590 			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
591 			      ep, ep->com.cm_id, ep->com.qp,
592 			      (int)ep->com.state, ep->com.flags,
593 			      ep->com.history, ep->hwtid, ep->atid,
594 			      ep->stats.connect_neg_adv,
595 			      ep->stats.abort_neg_adv,
596 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
597 			      ntohs(mapped_lsin6->sin6_port),
598 			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
599 			      ntohs(mapped_rsin6->sin6_port));
600 	}
601 	if (cc < space)
602 		epd->pos += cc;
603 	return 0;
604 }
605 
606 static int dump_listen_ep(int id, void *p, void *data)
607 {
608 	struct c4iw_listen_ep *ep = p;
609 	struct c4iw_debugfs_data *epd = data;
610 	int space;
611 	int cc;
612 
613 	space = epd->bufsize - epd->pos - 1;
614 	if (space == 0)
615 		return 1;
616 
617 	if (ep->com.local_addr.ss_family == AF_INET) {
618 		struct sockaddr_in *lsin = (struct sockaddr_in *)
619 			&ep->com.local_addr;
620 		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
621 			&ep->com.mapped_local_addr;
622 
623 		cc = snprintf(epd->buf + epd->pos, space,
624 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
625 			      "backlog %d %pI4:%d/%d\n",
626 			      ep, ep->com.cm_id, (int)ep->com.state,
627 			      ep->com.flags, ep->stid, ep->backlog,
628 			      &lsin->sin_addr, ntohs(lsin->sin_port),
629 			      ntohs(mapped_lsin->sin_port));
630 	} else {
631 		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
632 			&ep->com.local_addr;
633 		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
634 			&ep->com.mapped_local_addr;
635 
636 		cc = snprintf(epd->buf + epd->pos, space,
637 			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
638 			      "backlog %d %pI6:%d/%d\n",
639 			      ep, ep->com.cm_id, (int)ep->com.state,
640 			      ep->com.flags, ep->stid, ep->backlog,
641 			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
642 			      ntohs(mapped_lsin6->sin6_port));
643 	}
644 	if (cc < space)
645 		epd->pos += cc;
646 	return 0;
647 }
648 
649 static int ep_release(struct inode *inode, struct file *file)
650 {
651 	struct c4iw_debugfs_data *epd = file->private_data;
652 	if (!epd) {
653 		pr_info("%s null qpd?\n", __func__);
654 		return 0;
655 	}
656 	vfree(epd->buf);
657 	kfree(epd);
658 	return 0;
659 }
660 
661 static int ep_open(struct inode *inode, struct file *file)
662 {
663 	struct c4iw_debugfs_data *epd;
664 	int ret = 0;
665 	int count = 1;
666 
667 	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
668 	if (!epd) {
669 		ret = -ENOMEM;
670 		goto out;
671 	}
672 	epd->devp = inode->i_private;
673 	epd->pos = 0;
674 
675 	spin_lock_irq(&epd->devp->lock);
676 	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
677 	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
678 	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
679 	spin_unlock_irq(&epd->devp->lock);
680 
681 	epd->bufsize = count * 240;
682 	epd->buf = vmalloc(epd->bufsize);
683 	if (!epd->buf) {
684 		ret = -ENOMEM;
685 		goto err1;
686 	}
687 
688 	spin_lock_irq(&epd->devp->lock);
689 	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
690 	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
691 	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
692 	spin_unlock_irq(&epd->devp->lock);
693 
694 	file->private_data = epd;
695 	goto out;
696 err1:
697 	kfree(epd);
698 out:
699 	return ret;
700 }
701 
702 static const struct file_operations ep_debugfs_fops = {
703 	.owner   = THIS_MODULE,
704 	.open    = ep_open,
705 	.release = ep_release,
706 	.read    = debugfs_read,
707 };
708 
709 static int setup_debugfs(struct c4iw_dev *devp)
710 {
711 	if (!devp->debugfs_root)
712 		return -1;
713 
714 	debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
715 				 (void *)devp, &qp_debugfs_fops, 4096);
716 
717 	debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
718 				 (void *)devp, &stag_debugfs_fops, 4096);
719 
720 	debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
721 				 (void *)devp, &stats_debugfs_fops, 4096);
722 
723 	debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
724 				 (void *)devp, &ep_debugfs_fops, 4096);
725 
726 	if (c4iw_wr_log)
727 		debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
728 					 (void *)devp, &wr_log_debugfs_fops, 4096);
729 	return 0;
730 }
731 
732 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
733 			       struct c4iw_dev_ucontext *uctx)
734 {
735 	struct list_head *pos, *nxt;
736 	struct c4iw_qid_list *entry;
737 
738 	mutex_lock(&uctx->lock);
739 	list_for_each_safe(pos, nxt, &uctx->qpids) {
740 		entry = list_entry(pos, struct c4iw_qid_list, entry);
741 		list_del_init(&entry->entry);
742 		if (!(entry->qid & rdev->qpmask)) {
743 			c4iw_put_resource(&rdev->resource.qid_table,
744 					  entry->qid);
745 			mutex_lock(&rdev->stats.lock);
746 			rdev->stats.qid.cur -= rdev->qpmask + 1;
747 			mutex_unlock(&rdev->stats.lock);
748 		}
749 		kfree(entry);
750 	}
751 
752 	list_for_each_safe(pos, nxt, &uctx->qpids) {
753 		entry = list_entry(pos, struct c4iw_qid_list, entry);
754 		list_del_init(&entry->entry);
755 		kfree(entry);
756 	}
757 	mutex_unlock(&uctx->lock);
758 }
759 
760 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
761 			    struct c4iw_dev_ucontext *uctx)
762 {
763 	INIT_LIST_HEAD(&uctx->qpids);
764 	INIT_LIST_HEAD(&uctx->cqids);
765 	mutex_init(&uctx->lock);
766 }
767 
768 /* Caller takes care of locking if needed */
769 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
770 {
771 	int err;
772 
773 	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
774 
775 	/*
776 	 * This implementation assumes udb_density == ucq_density!  Eventually
777 	 * we might need to support this but for now fail the open. Also the
778 	 * cqid and qpid range must match for now.
779 	 */
780 	if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
781 		pr_err(MOD "%s: unsupported udb/ucq densities %u/%u\n",
782 		       pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
783 		       rdev->lldi.ucq_density);
784 		err = -EINVAL;
785 		goto err1;
786 	}
787 	if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
788 	    rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
789 		pr_err(MOD "%s: unsupported qp and cq id ranges "
790 		       "qp start %u size %u cq start %u size %u\n",
791 		       pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
792 		       rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
793 		       rdev->lldi.vr->cq.size);
794 		err = -EINVAL;
795 		goto err1;
796 	}
797 
798 	rdev->qpmask = rdev->lldi.udb_density - 1;
799 	rdev->cqmask = rdev->lldi.ucq_density - 1;
800 	PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
801 	     "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
802 	     "qp qid start %u size %u cq qid start %u size %u\n",
803 	     __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
804 	     rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
805 	     rdev->lldi.vr->pbl.start,
806 	     rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
807 	     rdev->lldi.vr->rq.size,
808 	     rdev->lldi.vr->qp.start,
809 	     rdev->lldi.vr->qp.size,
810 	     rdev->lldi.vr->cq.start,
811 	     rdev->lldi.vr->cq.size);
812 	PDBG("udb len 0x%x udb base %p db_reg %p gts_reg %p "
813 	     "qpmask 0x%x cqmask 0x%x\n",
814 	     (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
815 	     (void *)pci_resource_start(rdev->lldi.pdev, 2),
816 	     rdev->lldi.db_reg, rdev->lldi.gts_reg,
817 	     rdev->qpmask, rdev->cqmask);
818 
819 	if (c4iw_num_stags(rdev) == 0) {
820 		err = -EINVAL;
821 		goto err1;
822 	}
823 
824 	rdev->stats.pd.total = T4_MAX_NUM_PD;
825 	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
826 	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
827 	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
828 	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
829 	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
830 
831 	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
832 	if (err) {
833 		printk(KERN_ERR MOD "error %d initializing resources\n", err);
834 		goto err1;
835 	}
836 	err = c4iw_pblpool_create(rdev);
837 	if (err) {
838 		printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
839 		goto err2;
840 	}
841 	err = c4iw_rqtpool_create(rdev);
842 	if (err) {
843 		printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
844 		goto err3;
845 	}
846 	err = c4iw_ocqp_pool_create(rdev);
847 	if (err) {
848 		printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
849 		goto err4;
850 	}
851 	rdev->status_page = (struct t4_dev_status_page *)
852 			    __get_free_page(GFP_KERNEL);
853 	if (!rdev->status_page) {
854 		pr_err(MOD "error allocating status page\n");
855 		goto err4;
856 	}
857 
858 	if (c4iw_wr_log) {
859 		rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
860 				       sizeof(*rdev->wr_log), GFP_KERNEL);
861 		if (rdev->wr_log) {
862 			rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
863 			atomic_set(&rdev->wr_log_idx, 0);
864 		} else {
865 			pr_err(MOD "error allocating wr_log. Logging disabled\n");
866 		}
867 	}
868 
869 	rdev->status_page->db_off = 0;
870 
871 	return 0;
872 err4:
873 	c4iw_rqtpool_destroy(rdev);
874 err3:
875 	c4iw_pblpool_destroy(rdev);
876 err2:
877 	c4iw_destroy_resource(&rdev->resource);
878 err1:
879 	return err;
880 }
881 
882 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
883 {
884 	kfree(rdev->wr_log);
885 	free_page((unsigned long)rdev->status_page);
886 	c4iw_pblpool_destroy(rdev);
887 	c4iw_rqtpool_destroy(rdev);
888 	c4iw_destroy_resource(&rdev->resource);
889 }
890 
891 static void c4iw_dealloc(struct uld_ctx *ctx)
892 {
893 	c4iw_rdev_close(&ctx->dev->rdev);
894 	idr_destroy(&ctx->dev->cqidr);
895 	idr_destroy(&ctx->dev->qpidr);
896 	idr_destroy(&ctx->dev->mmidr);
897 	idr_destroy(&ctx->dev->hwtid_idr);
898 	idr_destroy(&ctx->dev->stid_idr);
899 	idr_destroy(&ctx->dev->atid_idr);
900 	if (ctx->dev->rdev.bar2_kva)
901 		iounmap(ctx->dev->rdev.bar2_kva);
902 	if (ctx->dev->rdev.oc_mw_kva)
903 		iounmap(ctx->dev->rdev.oc_mw_kva);
904 	ib_dealloc_device(&ctx->dev->ibdev);
905 	ctx->dev = NULL;
906 }
907 
908 static void c4iw_remove(struct uld_ctx *ctx)
909 {
910 	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
911 	c4iw_unregister_device(ctx->dev);
912 	c4iw_dealloc(ctx);
913 }
914 
915 static int rdma_supported(const struct cxgb4_lld_info *infop)
916 {
917 	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
918 	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
919 	       infop->vr->cq.size > 0;
920 }
921 
922 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
923 {
924 	struct c4iw_dev *devp;
925 	int ret;
926 
927 	if (!rdma_supported(infop)) {
928 		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
929 		       pci_name(infop->pdev));
930 		return ERR_PTR(-ENOSYS);
931 	}
932 	if (!ocqp_supported(infop))
933 		pr_info("%s: On-Chip Queues not supported on this device.\n",
934 			pci_name(infop->pdev));
935 
936 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
937 	if (!devp) {
938 		printk(KERN_ERR MOD "Cannot allocate ib device\n");
939 		return ERR_PTR(-ENOMEM);
940 	}
941 	devp->rdev.lldi = *infop;
942 
943 	/* init various hw-queue params based on lld info */
944 	PDBG("%s: Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
945 	     __func__, devp->rdev.lldi.sge_ingpadboundary,
946 	     devp->rdev.lldi.sge_egrstatuspagesize);
947 
948 	devp->rdev.hw_queue.t4_eq_status_entries =
949 		devp->rdev.lldi.sge_ingpadboundary > 64 ? 2 : 1;
950 	devp->rdev.hw_queue.t4_max_eq_size = 65520;
951 	devp->rdev.hw_queue.t4_max_iq_size = 65520;
952 	devp->rdev.hw_queue.t4_max_rq_size = 8192 -
953 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
954 	devp->rdev.hw_queue.t4_max_sq_size =
955 		devp->rdev.hw_queue.t4_max_eq_size -
956 		devp->rdev.hw_queue.t4_eq_status_entries - 1;
957 	devp->rdev.hw_queue.t4_max_qp_depth =
958 		devp->rdev.hw_queue.t4_max_rq_size;
959 	devp->rdev.hw_queue.t4_max_cq_depth =
960 		devp->rdev.hw_queue.t4_max_iq_size - 2;
961 	devp->rdev.hw_queue.t4_stat_len =
962 		devp->rdev.lldi.sge_egrstatuspagesize;
963 
964 	/*
965 	 * For T5 devices, we map all of BAR2 with WC.
966 	 * For T4 devices with onchip qp mem, we map only that part
967 	 * of BAR2 with WC.
968 	 */
969 	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
970 	if (is_t5(devp->rdev.lldi.adapter_type)) {
971 		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
972 			pci_resource_len(devp->rdev.lldi.pdev, 2));
973 		if (!devp->rdev.bar2_kva) {
974 			pr_err(MOD "Unable to ioremap BAR2\n");
975 			ib_dealloc_device(&devp->ibdev);
976 			return ERR_PTR(-EINVAL);
977 		}
978 	} else if (ocqp_supported(infop)) {
979 		devp->rdev.oc_mw_pa =
980 			pci_resource_start(devp->rdev.lldi.pdev, 2) +
981 			pci_resource_len(devp->rdev.lldi.pdev, 2) -
982 			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
983 		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
984 			devp->rdev.lldi.vr->ocq.size);
985 		if (!devp->rdev.oc_mw_kva) {
986 			pr_err(MOD "Unable to ioremap onchip mem\n");
987 			ib_dealloc_device(&devp->ibdev);
988 			return ERR_PTR(-EINVAL);
989 		}
990 	}
991 
992 	PDBG(KERN_INFO MOD "ocq memory: "
993 	       "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
994 	       devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
995 	       devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
996 
997 	ret = c4iw_rdev_open(&devp->rdev);
998 	if (ret) {
999 		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
1000 		ib_dealloc_device(&devp->ibdev);
1001 		return ERR_PTR(ret);
1002 	}
1003 
1004 	idr_init(&devp->cqidr);
1005 	idr_init(&devp->qpidr);
1006 	idr_init(&devp->mmidr);
1007 	idr_init(&devp->hwtid_idr);
1008 	idr_init(&devp->stid_idr);
1009 	idr_init(&devp->atid_idr);
1010 	spin_lock_init(&devp->lock);
1011 	mutex_init(&devp->rdev.stats.lock);
1012 	mutex_init(&devp->db_mutex);
1013 	INIT_LIST_HEAD(&devp->db_fc_list);
1014 	devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1015 
1016 	if (c4iw_debugfs_root) {
1017 		devp->debugfs_root = debugfs_create_dir(
1018 					pci_name(devp->rdev.lldi.pdev),
1019 					c4iw_debugfs_root);
1020 		setup_debugfs(devp);
1021 	}
1022 
1023 
1024 	return devp;
1025 }
1026 
1027 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1028 {
1029 	struct uld_ctx *ctx;
1030 	static int vers_printed;
1031 	int i;
1032 
1033 	if (!vers_printed++)
1034 		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1035 			DRV_VERSION);
1036 
1037 	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1038 	if (!ctx) {
1039 		ctx = ERR_PTR(-ENOMEM);
1040 		goto out;
1041 	}
1042 	ctx->lldi = *infop;
1043 
1044 	PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1045 	     __func__, pci_name(ctx->lldi.pdev),
1046 	     ctx->lldi.nchan, ctx->lldi.nrxq,
1047 	     ctx->lldi.ntxq, ctx->lldi.nports);
1048 
1049 	mutex_lock(&dev_mutex);
1050 	list_add_tail(&ctx->entry, &uld_ctx_list);
1051 	mutex_unlock(&dev_mutex);
1052 
1053 	for (i = 0; i < ctx->lldi.nrxq; i++)
1054 		PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1055 out:
1056 	return ctx;
1057 }
1058 
1059 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1060 						 const __be64 *rsp,
1061 						 u32 pktshift)
1062 {
1063 	struct sk_buff *skb;
1064 
1065 	/*
1066 	 * Allocate space for cpl_pass_accept_req which will be synthesized by
1067 	 * driver. Once the driver synthesizes the request the skb will go
1068 	 * through the regular cpl_pass_accept_req processing.
1069 	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1070 	 * cpl_rx_pkt.
1071 	 */
1072 	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1073 			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1074 	if (unlikely(!skb))
1075 		return NULL;
1076 
1077 	 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1078 		   sizeof(struct rss_header) - pktshift);
1079 
1080 	/*
1081 	 * This skb will contain:
1082 	 *   rss_header from the rspq descriptor (1 flit)
1083 	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
1084 	 *   space for the difference between the size of an
1085 	 *      rx_pkt and pass_accept_req cpl (1 flit)
1086 	 *   the packet data from the gl
1087 	 */
1088 	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1089 				sizeof(struct rss_header));
1090 	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1091 				       sizeof(struct cpl_pass_accept_req),
1092 				       gl->va + pktshift,
1093 				       gl->tot_len - pktshift);
1094 	return skb;
1095 }
1096 
1097 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1098 			   const __be64 *rsp)
1099 {
1100 	unsigned int opcode = *(u8 *)rsp;
1101 	struct sk_buff *skb;
1102 
1103 	if (opcode != CPL_RX_PKT)
1104 		goto out;
1105 
1106 	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1107 	if (skb == NULL)
1108 		goto out;
1109 
1110 	if (c4iw_handlers[opcode] == NULL) {
1111 		pr_info("%s no handler opcode 0x%x...\n", __func__,
1112 		       opcode);
1113 		kfree_skb(skb);
1114 		goto out;
1115 	}
1116 	c4iw_handlers[opcode](dev, skb);
1117 	return 1;
1118 out:
1119 	return 0;
1120 }
1121 
1122 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1123 			const struct pkt_gl *gl)
1124 {
1125 	struct uld_ctx *ctx = handle;
1126 	struct c4iw_dev *dev = ctx->dev;
1127 	struct sk_buff *skb;
1128 	u8 opcode;
1129 
1130 	if (gl == NULL) {
1131 		/* omit RSS and rsp_ctrl at end of descriptor */
1132 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1133 
1134 		skb = alloc_skb(256, GFP_ATOMIC);
1135 		if (!skb)
1136 			goto nomem;
1137 		__skb_put(skb, len);
1138 		skb_copy_to_linear_data(skb, &rsp[1], len);
1139 	} else if (gl == CXGB4_MSG_AN) {
1140 		const struct rsp_ctrl *rc = (void *)rsp;
1141 
1142 		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1143 		c4iw_ev_handler(dev, qid);
1144 		return 0;
1145 	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1146 		if (recv_rx_pkt(dev, gl, rsp))
1147 			return 0;
1148 
1149 		pr_info("%s: unexpected FL contents at %p, " \
1150 		       "RSS %#llx, FL %#llx, len %u\n",
1151 		       pci_name(ctx->lldi.pdev), gl->va,
1152 		       (unsigned long long)be64_to_cpu(*rsp),
1153 		       (unsigned long long)be64_to_cpu(
1154 		       *(__force __be64 *)gl->va),
1155 		       gl->tot_len);
1156 
1157 		return 0;
1158 	} else {
1159 		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1160 		if (unlikely(!skb))
1161 			goto nomem;
1162 	}
1163 
1164 	opcode = *(u8 *)rsp;
1165 	if (c4iw_handlers[opcode]) {
1166 		c4iw_handlers[opcode](dev, skb);
1167 	} else {
1168 		pr_info("%s no handler opcode 0x%x...\n", __func__,
1169 		       opcode);
1170 		kfree_skb(skb);
1171 	}
1172 
1173 	return 0;
1174 nomem:
1175 	return -1;
1176 }
1177 
1178 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1179 {
1180 	struct uld_ctx *ctx = handle;
1181 
1182 	PDBG("%s new_state %u\n", __func__, new_state);
1183 	switch (new_state) {
1184 	case CXGB4_STATE_UP:
1185 		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
1186 		if (!ctx->dev) {
1187 			int ret;
1188 
1189 			ctx->dev = c4iw_alloc(&ctx->lldi);
1190 			if (IS_ERR(ctx->dev)) {
1191 				printk(KERN_ERR MOD
1192 				       "%s: initialization failed: %ld\n",
1193 				       pci_name(ctx->lldi.pdev),
1194 				       PTR_ERR(ctx->dev));
1195 				ctx->dev = NULL;
1196 				break;
1197 			}
1198 			ret = c4iw_register_device(ctx->dev);
1199 			if (ret) {
1200 				printk(KERN_ERR MOD
1201 				       "%s: RDMA registration failed: %d\n",
1202 				       pci_name(ctx->lldi.pdev), ret);
1203 				c4iw_dealloc(ctx);
1204 			}
1205 		}
1206 		break;
1207 	case CXGB4_STATE_DOWN:
1208 		printk(KERN_INFO MOD "%s: Down\n",
1209 		       pci_name(ctx->lldi.pdev));
1210 		if (ctx->dev)
1211 			c4iw_remove(ctx);
1212 		break;
1213 	case CXGB4_STATE_START_RECOVERY:
1214 		printk(KERN_INFO MOD "%s: Fatal Error\n",
1215 		       pci_name(ctx->lldi.pdev));
1216 		if (ctx->dev) {
1217 			struct ib_event event;
1218 
1219 			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1220 			memset(&event, 0, sizeof event);
1221 			event.event  = IB_EVENT_DEVICE_FATAL;
1222 			event.device = &ctx->dev->ibdev;
1223 			ib_dispatch_event(&event);
1224 			c4iw_remove(ctx);
1225 		}
1226 		break;
1227 	case CXGB4_STATE_DETACH:
1228 		printk(KERN_INFO MOD "%s: Detach\n",
1229 		       pci_name(ctx->lldi.pdev));
1230 		if (ctx->dev)
1231 			c4iw_remove(ctx);
1232 		break;
1233 	}
1234 	return 0;
1235 }
1236 
1237 static int disable_qp_db(int id, void *p, void *data)
1238 {
1239 	struct c4iw_qp *qp = p;
1240 
1241 	t4_disable_wq_db(&qp->wq);
1242 	return 0;
1243 }
1244 
1245 static void stop_queues(struct uld_ctx *ctx)
1246 {
1247 	unsigned long flags;
1248 
1249 	spin_lock_irqsave(&ctx->dev->lock, flags);
1250 	ctx->dev->rdev.stats.db_state_transitions++;
1251 	ctx->dev->db_state = STOPPED;
1252 	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1253 		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1254 	else
1255 		ctx->dev->rdev.status_page->db_off = 1;
1256 	spin_unlock_irqrestore(&ctx->dev->lock, flags);
1257 }
1258 
1259 static int enable_qp_db(int id, void *p, void *data)
1260 {
1261 	struct c4iw_qp *qp = p;
1262 
1263 	t4_enable_wq_db(&qp->wq);
1264 	return 0;
1265 }
1266 
1267 static void resume_rc_qp(struct c4iw_qp *qp)
1268 {
1269 	spin_lock(&qp->lock);
1270 	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc,
1271 		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1272 	qp->wq.sq.wq_pidx_inc = 0;
1273 	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc,
1274 		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
1275 	qp->wq.rq.wq_pidx_inc = 0;
1276 	spin_unlock(&qp->lock);
1277 }
1278 
1279 static void resume_a_chunk(struct uld_ctx *ctx)
1280 {
1281 	int i;
1282 	struct c4iw_qp *qp;
1283 
1284 	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1285 		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1286 				      db_fc_entry);
1287 		list_del_init(&qp->db_fc_entry);
1288 		resume_rc_qp(qp);
1289 		if (list_empty(&ctx->dev->db_fc_list))
1290 			break;
1291 	}
1292 }
1293 
1294 static void resume_queues(struct uld_ctx *ctx)
1295 {
1296 	spin_lock_irq(&ctx->dev->lock);
1297 	if (ctx->dev->db_state != STOPPED)
1298 		goto out;
1299 	ctx->dev->db_state = FLOW_CONTROL;
1300 	while (1) {
1301 		if (list_empty(&ctx->dev->db_fc_list)) {
1302 			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1303 			ctx->dev->db_state = NORMAL;
1304 			ctx->dev->rdev.stats.db_state_transitions++;
1305 			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1306 				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1307 					     NULL);
1308 			} else {
1309 				ctx->dev->rdev.status_page->db_off = 0;
1310 			}
1311 			break;
1312 		} else {
1313 			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1314 			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1315 			       DB_FC_DRAIN_THRESH)) {
1316 				resume_a_chunk(ctx);
1317 			}
1318 			if (!list_empty(&ctx->dev->db_fc_list)) {
1319 				spin_unlock_irq(&ctx->dev->lock);
1320 				if (DB_FC_RESUME_DELAY) {
1321 					set_current_state(TASK_UNINTERRUPTIBLE);
1322 					schedule_timeout(DB_FC_RESUME_DELAY);
1323 				}
1324 				spin_lock_irq(&ctx->dev->lock);
1325 				if (ctx->dev->db_state != FLOW_CONTROL)
1326 					break;
1327 			}
1328 		}
1329 	}
1330 out:
1331 	if (ctx->dev->db_state != NORMAL)
1332 		ctx->dev->rdev.stats.db_fc_interruptions++;
1333 	spin_unlock_irq(&ctx->dev->lock);
1334 }
1335 
1336 struct qp_list {
1337 	unsigned idx;
1338 	struct c4iw_qp **qps;
1339 };
1340 
1341 static int add_and_ref_qp(int id, void *p, void *data)
1342 {
1343 	struct qp_list *qp_listp = data;
1344 	struct c4iw_qp *qp = p;
1345 
1346 	c4iw_qp_add_ref(&qp->ibqp);
1347 	qp_listp->qps[qp_listp->idx++] = qp;
1348 	return 0;
1349 }
1350 
1351 static int count_qps(int id, void *p, void *data)
1352 {
1353 	unsigned *countp = data;
1354 	(*countp)++;
1355 	return 0;
1356 }
1357 
1358 static void deref_qps(struct qp_list *qp_list)
1359 {
1360 	int idx;
1361 
1362 	for (idx = 0; idx < qp_list->idx; idx++)
1363 		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1364 }
1365 
1366 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1367 {
1368 	int idx;
1369 	int ret;
1370 
1371 	for (idx = 0; idx < qp_list->idx; idx++) {
1372 		struct c4iw_qp *qp = qp_list->qps[idx];
1373 
1374 		spin_lock_irq(&qp->rhp->lock);
1375 		spin_lock(&qp->lock);
1376 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1377 					  qp->wq.sq.qid,
1378 					  t4_sq_host_wq_pidx(&qp->wq),
1379 					  t4_sq_wq_size(&qp->wq));
1380 		if (ret) {
1381 			pr_err(MOD "%s: Fatal error - "
1382 			       "DB overflow recovery failed - "
1383 			       "error syncing SQ qid %u\n",
1384 			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1385 			spin_unlock(&qp->lock);
1386 			spin_unlock_irq(&qp->rhp->lock);
1387 			return;
1388 		}
1389 		qp->wq.sq.wq_pidx_inc = 0;
1390 
1391 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1392 					  qp->wq.rq.qid,
1393 					  t4_rq_host_wq_pidx(&qp->wq),
1394 					  t4_rq_wq_size(&qp->wq));
1395 
1396 		if (ret) {
1397 			pr_err(MOD "%s: Fatal error - "
1398 			       "DB overflow recovery failed - "
1399 			       "error syncing RQ qid %u\n",
1400 			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1401 			spin_unlock(&qp->lock);
1402 			spin_unlock_irq(&qp->rhp->lock);
1403 			return;
1404 		}
1405 		qp->wq.rq.wq_pidx_inc = 0;
1406 		spin_unlock(&qp->lock);
1407 		spin_unlock_irq(&qp->rhp->lock);
1408 
1409 		/* Wait for the dbfifo to drain */
1410 		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1411 			set_current_state(TASK_UNINTERRUPTIBLE);
1412 			schedule_timeout(usecs_to_jiffies(10));
1413 		}
1414 	}
1415 }
1416 
1417 static void recover_queues(struct uld_ctx *ctx)
1418 {
1419 	int count = 0;
1420 	struct qp_list qp_list;
1421 	int ret;
1422 
1423 	/* slow everybody down */
1424 	set_current_state(TASK_UNINTERRUPTIBLE);
1425 	schedule_timeout(usecs_to_jiffies(1000));
1426 
1427 	/* flush the SGE contexts */
1428 	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1429 	if (ret) {
1430 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1431 		       pci_name(ctx->lldi.pdev));
1432 		return;
1433 	}
1434 
1435 	/* Count active queues so we can build a list of queues to recover */
1436 	spin_lock_irq(&ctx->dev->lock);
1437 	WARN_ON(ctx->dev->db_state != STOPPED);
1438 	ctx->dev->db_state = RECOVERY;
1439 	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1440 
1441 	qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1442 	if (!qp_list.qps) {
1443 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
1444 		       pci_name(ctx->lldi.pdev));
1445 		spin_unlock_irq(&ctx->dev->lock);
1446 		return;
1447 	}
1448 	qp_list.idx = 0;
1449 
1450 	/* add and ref each qp so it doesn't get freed */
1451 	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1452 
1453 	spin_unlock_irq(&ctx->dev->lock);
1454 
1455 	/* now traverse the list in a safe context to recover the db state*/
1456 	recover_lost_dbs(ctx, &qp_list);
1457 
1458 	/* we're almost done!  deref the qps and clean up */
1459 	deref_qps(&qp_list);
1460 	kfree(qp_list.qps);
1461 
1462 	spin_lock_irq(&ctx->dev->lock);
1463 	WARN_ON(ctx->dev->db_state != RECOVERY);
1464 	ctx->dev->db_state = STOPPED;
1465 	spin_unlock_irq(&ctx->dev->lock);
1466 }
1467 
1468 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1469 {
1470 	struct uld_ctx *ctx = handle;
1471 
1472 	switch (control) {
1473 	case CXGB4_CONTROL_DB_FULL:
1474 		stop_queues(ctx);
1475 		ctx->dev->rdev.stats.db_full++;
1476 		break;
1477 	case CXGB4_CONTROL_DB_EMPTY:
1478 		resume_queues(ctx);
1479 		mutex_lock(&ctx->dev->rdev.stats.lock);
1480 		ctx->dev->rdev.stats.db_empty++;
1481 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1482 		break;
1483 	case CXGB4_CONTROL_DB_DROP:
1484 		recover_queues(ctx);
1485 		mutex_lock(&ctx->dev->rdev.stats.lock);
1486 		ctx->dev->rdev.stats.db_drop++;
1487 		mutex_unlock(&ctx->dev->rdev.stats.lock);
1488 		break;
1489 	default:
1490 		printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
1491 		       pci_name(ctx->lldi.pdev), control);
1492 		break;
1493 	}
1494 	return 0;
1495 }
1496 
1497 static struct cxgb4_uld_info c4iw_uld_info = {
1498 	.name = DRV_NAME,
1499 	.add = c4iw_uld_add,
1500 	.rx_handler = c4iw_uld_rx_handler,
1501 	.state_change = c4iw_uld_state_change,
1502 	.control = c4iw_uld_control,
1503 };
1504 
1505 static int __init c4iw_init_module(void)
1506 {
1507 	int err;
1508 
1509 	err = c4iw_cm_init();
1510 	if (err)
1511 		return err;
1512 
1513 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1514 	if (!c4iw_debugfs_root)
1515 		printk(KERN_WARNING MOD
1516 		       "could not create debugfs entry, continuing\n");
1517 
1518 	if (ibnl_add_client(RDMA_NL_C4IW, RDMA_NL_IWPM_NUM_OPS,
1519 			    c4iw_nl_cb_table))
1520 		pr_err("%s[%u]: Failed to add netlink callback\n"
1521 		       , __func__, __LINE__);
1522 
1523 	err = iwpm_init(RDMA_NL_C4IW);
1524 	if (err) {
1525 		pr_err("port mapper initialization failed with %d\n", err);
1526 		ibnl_remove_client(RDMA_NL_C4IW);
1527 		c4iw_cm_term();
1528 		debugfs_remove_recursive(c4iw_debugfs_root);
1529 		return err;
1530 	}
1531 
1532 	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1533 
1534 	return 0;
1535 }
1536 
1537 static void __exit c4iw_exit_module(void)
1538 {
1539 	struct uld_ctx *ctx, *tmp;
1540 
1541 	mutex_lock(&dev_mutex);
1542 	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1543 		if (ctx->dev)
1544 			c4iw_remove(ctx);
1545 		kfree(ctx);
1546 	}
1547 	mutex_unlock(&dev_mutex);
1548 	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1549 	iwpm_exit(RDMA_NL_C4IW);
1550 	ibnl_remove_client(RDMA_NL_C4IW);
1551 	c4iw_cm_term();
1552 	debugfs_remove_recursive(c4iw_debugfs_root);
1553 }
1554 
1555 module_init(c4iw_init_module);
1556 module_exit(c4iw_exit_module);
1557