xref: /linux/drivers/infiniband/hw/cxgb4/device.c (revision 95e9fd10f06cb5642028b6b851e32b8c8afb4571)
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 
37 #include <rdma/ib_verbs.h>
38 
39 #include "iw_cxgb4.h"
40 
41 #define DRV_VERSION "0.1"
42 
43 MODULE_AUTHOR("Steve Wise");
44 MODULE_DESCRIPTION("Chelsio T4 RDMA Driver");
45 MODULE_LICENSE("Dual BSD/GPL");
46 MODULE_VERSION(DRV_VERSION);
47 
48 struct uld_ctx {
49 	struct list_head entry;
50 	struct cxgb4_lld_info lldi;
51 	struct c4iw_dev *dev;
52 };
53 
54 static LIST_HEAD(uld_ctx_list);
55 static DEFINE_MUTEX(dev_mutex);
56 
57 static struct dentry *c4iw_debugfs_root;
58 
59 struct c4iw_debugfs_data {
60 	struct c4iw_dev *devp;
61 	char *buf;
62 	int bufsize;
63 	int pos;
64 };
65 
66 static int count_idrs(int id, void *p, void *data)
67 {
68 	int *countp = data;
69 
70 	*countp = *countp + 1;
71 	return 0;
72 }
73 
74 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
75 			    loff_t *ppos)
76 {
77 	struct c4iw_debugfs_data *d = file->private_data;
78 
79 	return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
80 }
81 
82 static int dump_qp(int id, void *p, void *data)
83 {
84 	struct c4iw_qp *qp = p;
85 	struct c4iw_debugfs_data *qpd = data;
86 	int space;
87 	int cc;
88 
89 	if (id != qp->wq.sq.qid)
90 		return 0;
91 
92 	space = qpd->bufsize - qpd->pos - 1;
93 	if (space == 0)
94 		return 1;
95 
96 	if (qp->ep)
97 		cc = snprintf(qpd->buf + qpd->pos, space,
98 			     "qp sq id %u rq id %u state %u onchip %u "
99 			     "ep tid %u state %u %pI4:%u->%pI4:%u\n",
100 			     qp->wq.sq.qid, qp->wq.rq.qid, (int)qp->attr.state,
101 			     qp->wq.sq.flags & T4_SQ_ONCHIP,
102 			     qp->ep->hwtid, (int)qp->ep->com.state,
103 			     &qp->ep->com.local_addr.sin_addr.s_addr,
104 			     ntohs(qp->ep->com.local_addr.sin_port),
105 			     &qp->ep->com.remote_addr.sin_addr.s_addr,
106 			     ntohs(qp->ep->com.remote_addr.sin_port));
107 	else
108 		cc = snprintf(qpd->buf + qpd->pos, space,
109 			     "qp sq id %u rq id %u state %u onchip %u\n",
110 			      qp->wq.sq.qid, qp->wq.rq.qid,
111 			      (int)qp->attr.state,
112 			      qp->wq.sq.flags & T4_SQ_ONCHIP);
113 	if (cc < space)
114 		qpd->pos += cc;
115 	return 0;
116 }
117 
118 static int qp_release(struct inode *inode, struct file *file)
119 {
120 	struct c4iw_debugfs_data *qpd = file->private_data;
121 	if (!qpd) {
122 		printk(KERN_INFO "%s null qpd?\n", __func__);
123 		return 0;
124 	}
125 	vfree(qpd->buf);
126 	kfree(qpd);
127 	return 0;
128 }
129 
130 static int qp_open(struct inode *inode, struct file *file)
131 {
132 	struct c4iw_debugfs_data *qpd;
133 	int ret = 0;
134 	int count = 1;
135 
136 	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
137 	if (!qpd) {
138 		ret = -ENOMEM;
139 		goto out;
140 	}
141 	qpd->devp = inode->i_private;
142 	qpd->pos = 0;
143 
144 	spin_lock_irq(&qpd->devp->lock);
145 	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
146 	spin_unlock_irq(&qpd->devp->lock);
147 
148 	qpd->bufsize = count * 128;
149 	qpd->buf = vmalloc(qpd->bufsize);
150 	if (!qpd->buf) {
151 		ret = -ENOMEM;
152 		goto err1;
153 	}
154 
155 	spin_lock_irq(&qpd->devp->lock);
156 	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
157 	spin_unlock_irq(&qpd->devp->lock);
158 
159 	qpd->buf[qpd->pos++] = 0;
160 	file->private_data = qpd;
161 	goto out;
162 err1:
163 	kfree(qpd);
164 out:
165 	return ret;
166 }
167 
168 static const struct file_operations qp_debugfs_fops = {
169 	.owner   = THIS_MODULE,
170 	.open    = qp_open,
171 	.release = qp_release,
172 	.read    = debugfs_read,
173 	.llseek  = default_llseek,
174 };
175 
176 static int dump_stag(int id, void *p, void *data)
177 {
178 	struct c4iw_debugfs_data *stagd = data;
179 	int space;
180 	int cc;
181 
182 	space = stagd->bufsize - stagd->pos - 1;
183 	if (space == 0)
184 		return 1;
185 
186 	cc = snprintf(stagd->buf + stagd->pos, space, "0x%x\n", id<<8);
187 	if (cc < space)
188 		stagd->pos += cc;
189 	return 0;
190 }
191 
192 static int stag_release(struct inode *inode, struct file *file)
193 {
194 	struct c4iw_debugfs_data *stagd = file->private_data;
195 	if (!stagd) {
196 		printk(KERN_INFO "%s null stagd?\n", __func__);
197 		return 0;
198 	}
199 	kfree(stagd->buf);
200 	kfree(stagd);
201 	return 0;
202 }
203 
204 static int stag_open(struct inode *inode, struct file *file)
205 {
206 	struct c4iw_debugfs_data *stagd;
207 	int ret = 0;
208 	int count = 1;
209 
210 	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
211 	if (!stagd) {
212 		ret = -ENOMEM;
213 		goto out;
214 	}
215 	stagd->devp = inode->i_private;
216 	stagd->pos = 0;
217 
218 	spin_lock_irq(&stagd->devp->lock);
219 	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
220 	spin_unlock_irq(&stagd->devp->lock);
221 
222 	stagd->bufsize = count * sizeof("0x12345678\n");
223 	stagd->buf = kmalloc(stagd->bufsize, GFP_KERNEL);
224 	if (!stagd->buf) {
225 		ret = -ENOMEM;
226 		goto err1;
227 	}
228 
229 	spin_lock_irq(&stagd->devp->lock);
230 	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
231 	spin_unlock_irq(&stagd->devp->lock);
232 
233 	stagd->buf[stagd->pos++] = 0;
234 	file->private_data = stagd;
235 	goto out;
236 err1:
237 	kfree(stagd);
238 out:
239 	return ret;
240 }
241 
242 static const struct file_operations stag_debugfs_fops = {
243 	.owner   = THIS_MODULE,
244 	.open    = stag_open,
245 	.release = stag_release,
246 	.read    = debugfs_read,
247 	.llseek  = default_llseek,
248 };
249 
250 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY"};
251 
252 static int stats_show(struct seq_file *seq, void *v)
253 {
254 	struct c4iw_dev *dev = seq->private;
255 
256 	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
257 		   "Max", "Fail");
258 	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
259 			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
260 			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
261 	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
262 			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
263 			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
264 	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
265 			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
266 			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
267 	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
268 			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
269 			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
270 	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
271 			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
272 			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
273 	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
274 			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
275 			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
276 	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
277 	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
278 	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
279 	seq_printf(seq, " DB State: %s Transitions %llu\n",
280 		   db_state_str[dev->db_state],
281 		   dev->rdev.stats.db_state_transitions);
282 	return 0;
283 }
284 
285 static int stats_open(struct inode *inode, struct file *file)
286 {
287 	return single_open(file, stats_show, inode->i_private);
288 }
289 
290 static ssize_t stats_clear(struct file *file, const char __user *buf,
291 		size_t count, loff_t *pos)
292 {
293 	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
294 
295 	mutex_lock(&dev->rdev.stats.lock);
296 	dev->rdev.stats.pd.max = 0;
297 	dev->rdev.stats.pd.fail = 0;
298 	dev->rdev.stats.qid.max = 0;
299 	dev->rdev.stats.qid.fail = 0;
300 	dev->rdev.stats.stag.max = 0;
301 	dev->rdev.stats.stag.fail = 0;
302 	dev->rdev.stats.pbl.max = 0;
303 	dev->rdev.stats.pbl.fail = 0;
304 	dev->rdev.stats.rqt.max = 0;
305 	dev->rdev.stats.rqt.fail = 0;
306 	dev->rdev.stats.ocqp.max = 0;
307 	dev->rdev.stats.ocqp.fail = 0;
308 	dev->rdev.stats.db_full = 0;
309 	dev->rdev.stats.db_empty = 0;
310 	dev->rdev.stats.db_drop = 0;
311 	dev->rdev.stats.db_state_transitions = 0;
312 	mutex_unlock(&dev->rdev.stats.lock);
313 	return count;
314 }
315 
316 static const struct file_operations stats_debugfs_fops = {
317 	.owner   = THIS_MODULE,
318 	.open    = stats_open,
319 	.release = single_release,
320 	.read    = seq_read,
321 	.llseek  = seq_lseek,
322 	.write   = stats_clear,
323 };
324 
325 static int setup_debugfs(struct c4iw_dev *devp)
326 {
327 	struct dentry *de;
328 
329 	if (!devp->debugfs_root)
330 		return -1;
331 
332 	de = debugfs_create_file("qps", S_IWUSR, devp->debugfs_root,
333 				 (void *)devp, &qp_debugfs_fops);
334 	if (de && de->d_inode)
335 		de->d_inode->i_size = 4096;
336 
337 	de = debugfs_create_file("stags", S_IWUSR, devp->debugfs_root,
338 				 (void *)devp, &stag_debugfs_fops);
339 	if (de && de->d_inode)
340 		de->d_inode->i_size = 4096;
341 
342 	de = debugfs_create_file("stats", S_IWUSR, devp->debugfs_root,
343 			(void *)devp, &stats_debugfs_fops);
344 	if (de && de->d_inode)
345 		de->d_inode->i_size = 4096;
346 
347 	return 0;
348 }
349 
350 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
351 			       struct c4iw_dev_ucontext *uctx)
352 {
353 	struct list_head *pos, *nxt;
354 	struct c4iw_qid_list *entry;
355 
356 	mutex_lock(&uctx->lock);
357 	list_for_each_safe(pos, nxt, &uctx->qpids) {
358 		entry = list_entry(pos, struct c4iw_qid_list, entry);
359 		list_del_init(&entry->entry);
360 		if (!(entry->qid & rdev->qpmask)) {
361 			c4iw_put_resource(&rdev->resource.qid_table,
362 					  entry->qid);
363 			mutex_lock(&rdev->stats.lock);
364 			rdev->stats.qid.cur -= rdev->qpmask + 1;
365 			mutex_unlock(&rdev->stats.lock);
366 		}
367 		kfree(entry);
368 	}
369 
370 	list_for_each_safe(pos, nxt, &uctx->qpids) {
371 		entry = list_entry(pos, struct c4iw_qid_list, entry);
372 		list_del_init(&entry->entry);
373 		kfree(entry);
374 	}
375 	mutex_unlock(&uctx->lock);
376 }
377 
378 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
379 			    struct c4iw_dev_ucontext *uctx)
380 {
381 	INIT_LIST_HEAD(&uctx->qpids);
382 	INIT_LIST_HEAD(&uctx->cqids);
383 	mutex_init(&uctx->lock);
384 }
385 
386 /* Caller takes care of locking if needed */
387 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
388 {
389 	int err;
390 
391 	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
392 
393 	/*
394 	 * qpshift is the number of bits to shift the qpid left in order
395 	 * to get the correct address of the doorbell for that qp.
396 	 */
397 	rdev->qpshift = PAGE_SHIFT - ilog2(rdev->lldi.udb_density);
398 	rdev->qpmask = rdev->lldi.udb_density - 1;
399 	rdev->cqshift = PAGE_SHIFT - ilog2(rdev->lldi.ucq_density);
400 	rdev->cqmask = rdev->lldi.ucq_density - 1;
401 	PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
402 	     "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
403 	     "qp qid start %u size %u cq qid start %u size %u\n",
404 	     __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
405 	     rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
406 	     rdev->lldi.vr->pbl.start,
407 	     rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
408 	     rdev->lldi.vr->rq.size,
409 	     rdev->lldi.vr->qp.start,
410 	     rdev->lldi.vr->qp.size,
411 	     rdev->lldi.vr->cq.start,
412 	     rdev->lldi.vr->cq.size);
413 	PDBG("udb len 0x%x udb base %p db_reg %p gts_reg %p qpshift %lu "
414 	     "qpmask 0x%x cqshift %lu cqmask 0x%x\n",
415 	     (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
416 	     (void *)pci_resource_start(rdev->lldi.pdev, 2),
417 	     rdev->lldi.db_reg,
418 	     rdev->lldi.gts_reg,
419 	     rdev->qpshift, rdev->qpmask,
420 	     rdev->cqshift, rdev->cqmask);
421 
422 	if (c4iw_num_stags(rdev) == 0) {
423 		err = -EINVAL;
424 		goto err1;
425 	}
426 
427 	rdev->stats.pd.total = T4_MAX_NUM_PD;
428 	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
429 	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
430 	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
431 	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
432 	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
433 
434 	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
435 	if (err) {
436 		printk(KERN_ERR MOD "error %d initializing resources\n", err);
437 		goto err1;
438 	}
439 	err = c4iw_pblpool_create(rdev);
440 	if (err) {
441 		printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
442 		goto err2;
443 	}
444 	err = c4iw_rqtpool_create(rdev);
445 	if (err) {
446 		printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
447 		goto err3;
448 	}
449 	err = c4iw_ocqp_pool_create(rdev);
450 	if (err) {
451 		printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
452 		goto err4;
453 	}
454 	return 0;
455 err4:
456 	c4iw_rqtpool_destroy(rdev);
457 err3:
458 	c4iw_pblpool_destroy(rdev);
459 err2:
460 	c4iw_destroy_resource(&rdev->resource);
461 err1:
462 	return err;
463 }
464 
465 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
466 {
467 	c4iw_pblpool_destroy(rdev);
468 	c4iw_rqtpool_destroy(rdev);
469 	c4iw_destroy_resource(&rdev->resource);
470 }
471 
472 static void c4iw_dealloc(struct uld_ctx *ctx)
473 {
474 	c4iw_rdev_close(&ctx->dev->rdev);
475 	idr_destroy(&ctx->dev->cqidr);
476 	idr_destroy(&ctx->dev->qpidr);
477 	idr_destroy(&ctx->dev->mmidr);
478 	iounmap(ctx->dev->rdev.oc_mw_kva);
479 	ib_dealloc_device(&ctx->dev->ibdev);
480 	ctx->dev = NULL;
481 }
482 
483 static void c4iw_remove(struct uld_ctx *ctx)
484 {
485 	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
486 	c4iw_unregister_device(ctx->dev);
487 	c4iw_dealloc(ctx);
488 }
489 
490 static int rdma_supported(const struct cxgb4_lld_info *infop)
491 {
492 	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
493 	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
494 	       infop->vr->cq.size > 0 && infop->vr->ocq.size > 0;
495 }
496 
497 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
498 {
499 	struct c4iw_dev *devp;
500 	int ret;
501 
502 	if (!rdma_supported(infop)) {
503 		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
504 		       pci_name(infop->pdev));
505 		return ERR_PTR(-ENOSYS);
506 	}
507 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
508 	if (!devp) {
509 		printk(KERN_ERR MOD "Cannot allocate ib device\n");
510 		return ERR_PTR(-ENOMEM);
511 	}
512 	devp->rdev.lldi = *infop;
513 
514 	devp->rdev.oc_mw_pa = pci_resource_start(devp->rdev.lldi.pdev, 2) +
515 		(pci_resource_len(devp->rdev.lldi.pdev, 2) -
516 		 roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size));
517 	devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
518 					       devp->rdev.lldi.vr->ocq.size);
519 
520 	PDBG(KERN_INFO MOD "ocq memory: "
521 	       "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
522 	       devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
523 	       devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
524 
525 	ret = c4iw_rdev_open(&devp->rdev);
526 	if (ret) {
527 		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
528 		ib_dealloc_device(&devp->ibdev);
529 		return ERR_PTR(ret);
530 	}
531 
532 	idr_init(&devp->cqidr);
533 	idr_init(&devp->qpidr);
534 	idr_init(&devp->mmidr);
535 	spin_lock_init(&devp->lock);
536 	mutex_init(&devp->rdev.stats.lock);
537 	mutex_init(&devp->db_mutex);
538 
539 	if (c4iw_debugfs_root) {
540 		devp->debugfs_root = debugfs_create_dir(
541 					pci_name(devp->rdev.lldi.pdev),
542 					c4iw_debugfs_root);
543 		setup_debugfs(devp);
544 	}
545 	return devp;
546 }
547 
548 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
549 {
550 	struct uld_ctx *ctx;
551 	static int vers_printed;
552 	int i;
553 
554 	if (!vers_printed++)
555 		printk(KERN_INFO MOD "Chelsio T4 RDMA Driver - version %s\n",
556 		       DRV_VERSION);
557 
558 	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
559 	if (!ctx) {
560 		ctx = ERR_PTR(-ENOMEM);
561 		goto out;
562 	}
563 	ctx->lldi = *infop;
564 
565 	PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
566 	     __func__, pci_name(ctx->lldi.pdev),
567 	     ctx->lldi.nchan, ctx->lldi.nrxq,
568 	     ctx->lldi.ntxq, ctx->lldi.nports);
569 
570 	mutex_lock(&dev_mutex);
571 	list_add_tail(&ctx->entry, &uld_ctx_list);
572 	mutex_unlock(&dev_mutex);
573 
574 	for (i = 0; i < ctx->lldi.nrxq; i++)
575 		PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
576 out:
577 	return ctx;
578 }
579 
580 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
581 			const struct pkt_gl *gl)
582 {
583 	struct uld_ctx *ctx = handle;
584 	struct c4iw_dev *dev = ctx->dev;
585 	struct sk_buff *skb;
586 	const struct cpl_act_establish *rpl;
587 	unsigned int opcode;
588 
589 	if (gl == NULL) {
590 		/* omit RSS and rsp_ctrl at end of descriptor */
591 		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
592 
593 		skb = alloc_skb(256, GFP_ATOMIC);
594 		if (!skb)
595 			goto nomem;
596 		__skb_put(skb, len);
597 		skb_copy_to_linear_data(skb, &rsp[1], len);
598 	} else if (gl == CXGB4_MSG_AN) {
599 		const struct rsp_ctrl *rc = (void *)rsp;
600 
601 		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
602 		c4iw_ev_handler(dev, qid);
603 		return 0;
604 	} else {
605 		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
606 		if (unlikely(!skb))
607 			goto nomem;
608 	}
609 
610 	rpl = cplhdr(skb);
611 	opcode = rpl->ot.opcode;
612 
613 	if (c4iw_handlers[opcode])
614 		c4iw_handlers[opcode](dev, skb);
615 	else
616 		printk(KERN_INFO "%s no handler opcode 0x%x...\n", __func__,
617 		       opcode);
618 
619 	return 0;
620 nomem:
621 	return -1;
622 }
623 
624 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
625 {
626 	struct uld_ctx *ctx = handle;
627 
628 	PDBG("%s new_state %u\n", __func__, new_state);
629 	switch (new_state) {
630 	case CXGB4_STATE_UP:
631 		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
632 		if (!ctx->dev) {
633 			int ret;
634 
635 			ctx->dev = c4iw_alloc(&ctx->lldi);
636 			if (IS_ERR(ctx->dev)) {
637 				printk(KERN_ERR MOD
638 				       "%s: initialization failed: %ld\n",
639 				       pci_name(ctx->lldi.pdev),
640 				       PTR_ERR(ctx->dev));
641 				ctx->dev = NULL;
642 				break;
643 			}
644 			ret = c4iw_register_device(ctx->dev);
645 			if (ret) {
646 				printk(KERN_ERR MOD
647 				       "%s: RDMA registration failed: %d\n",
648 				       pci_name(ctx->lldi.pdev), ret);
649 				c4iw_dealloc(ctx);
650 			}
651 		}
652 		break;
653 	case CXGB4_STATE_DOWN:
654 		printk(KERN_INFO MOD "%s: Down\n",
655 		       pci_name(ctx->lldi.pdev));
656 		if (ctx->dev)
657 			c4iw_remove(ctx);
658 		break;
659 	case CXGB4_STATE_START_RECOVERY:
660 		printk(KERN_INFO MOD "%s: Fatal Error\n",
661 		       pci_name(ctx->lldi.pdev));
662 		if (ctx->dev) {
663 			struct ib_event event;
664 
665 			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
666 			memset(&event, 0, sizeof event);
667 			event.event  = IB_EVENT_DEVICE_FATAL;
668 			event.device = &ctx->dev->ibdev;
669 			ib_dispatch_event(&event);
670 			c4iw_remove(ctx);
671 		}
672 		break;
673 	case CXGB4_STATE_DETACH:
674 		printk(KERN_INFO MOD "%s: Detach\n",
675 		       pci_name(ctx->lldi.pdev));
676 		if (ctx->dev)
677 			c4iw_remove(ctx);
678 		break;
679 	}
680 	return 0;
681 }
682 
683 static int disable_qp_db(int id, void *p, void *data)
684 {
685 	struct c4iw_qp *qp = p;
686 
687 	t4_disable_wq_db(&qp->wq);
688 	return 0;
689 }
690 
691 static void stop_queues(struct uld_ctx *ctx)
692 {
693 	spin_lock_irq(&ctx->dev->lock);
694 	if (ctx->dev->db_state == NORMAL) {
695 		ctx->dev->rdev.stats.db_state_transitions++;
696 		ctx->dev->db_state = FLOW_CONTROL;
697 		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
698 	}
699 	spin_unlock_irq(&ctx->dev->lock);
700 }
701 
702 static int enable_qp_db(int id, void *p, void *data)
703 {
704 	struct c4iw_qp *qp = p;
705 
706 	t4_enable_wq_db(&qp->wq);
707 	return 0;
708 }
709 
710 static void resume_queues(struct uld_ctx *ctx)
711 {
712 	spin_lock_irq(&ctx->dev->lock);
713 	if (ctx->dev->qpcnt <= db_fc_threshold &&
714 	    ctx->dev->db_state == FLOW_CONTROL) {
715 		ctx->dev->db_state = NORMAL;
716 		ctx->dev->rdev.stats.db_state_transitions++;
717 		idr_for_each(&ctx->dev->qpidr, enable_qp_db, NULL);
718 	}
719 	spin_unlock_irq(&ctx->dev->lock);
720 }
721 
722 struct qp_list {
723 	unsigned idx;
724 	struct c4iw_qp **qps;
725 };
726 
727 static int add_and_ref_qp(int id, void *p, void *data)
728 {
729 	struct qp_list *qp_listp = data;
730 	struct c4iw_qp *qp = p;
731 
732 	c4iw_qp_add_ref(&qp->ibqp);
733 	qp_listp->qps[qp_listp->idx++] = qp;
734 	return 0;
735 }
736 
737 static int count_qps(int id, void *p, void *data)
738 {
739 	unsigned *countp = data;
740 	(*countp)++;
741 	return 0;
742 }
743 
744 static void deref_qps(struct qp_list qp_list)
745 {
746 	int idx;
747 
748 	for (idx = 0; idx < qp_list.idx; idx++)
749 		c4iw_qp_rem_ref(&qp_list.qps[idx]->ibqp);
750 }
751 
752 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
753 {
754 	int idx;
755 	int ret;
756 
757 	for (idx = 0; idx < qp_list->idx; idx++) {
758 		struct c4iw_qp *qp = qp_list->qps[idx];
759 
760 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
761 					  qp->wq.sq.qid,
762 					  t4_sq_host_wq_pidx(&qp->wq),
763 					  t4_sq_wq_size(&qp->wq));
764 		if (ret) {
765 			printk(KERN_ERR MOD "%s: Fatal error - "
766 			       "DB overflow recovery failed - "
767 			       "error syncing SQ qid %u\n",
768 			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
769 			return;
770 		}
771 
772 		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
773 					  qp->wq.rq.qid,
774 					  t4_rq_host_wq_pidx(&qp->wq),
775 					  t4_rq_wq_size(&qp->wq));
776 
777 		if (ret) {
778 			printk(KERN_ERR MOD "%s: Fatal error - "
779 			       "DB overflow recovery failed - "
780 			       "error syncing RQ qid %u\n",
781 			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
782 			return;
783 		}
784 
785 		/* Wait for the dbfifo to drain */
786 		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
787 			set_current_state(TASK_UNINTERRUPTIBLE);
788 			schedule_timeout(usecs_to_jiffies(10));
789 		}
790 	}
791 }
792 
793 static void recover_queues(struct uld_ctx *ctx)
794 {
795 	int count = 0;
796 	struct qp_list qp_list;
797 	int ret;
798 
799 	/* lock out kernel db ringers */
800 	mutex_lock(&ctx->dev->db_mutex);
801 
802 	/* put all queues in to recovery mode */
803 	spin_lock_irq(&ctx->dev->lock);
804 	ctx->dev->db_state = RECOVERY;
805 	ctx->dev->rdev.stats.db_state_transitions++;
806 	idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
807 	spin_unlock_irq(&ctx->dev->lock);
808 
809 	/* slow everybody down */
810 	set_current_state(TASK_UNINTERRUPTIBLE);
811 	schedule_timeout(usecs_to_jiffies(1000));
812 
813 	/* Wait for the dbfifo to completely drain. */
814 	while (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1) > 0) {
815 		set_current_state(TASK_UNINTERRUPTIBLE);
816 		schedule_timeout(usecs_to_jiffies(10));
817 	}
818 
819 	/* flush the SGE contexts */
820 	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
821 	if (ret) {
822 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
823 		       pci_name(ctx->lldi.pdev));
824 		goto out;
825 	}
826 
827 	/* Count active queues so we can build a list of queues to recover */
828 	spin_lock_irq(&ctx->dev->lock);
829 	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
830 
831 	qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
832 	if (!qp_list.qps) {
833 		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
834 		       pci_name(ctx->lldi.pdev));
835 		spin_unlock_irq(&ctx->dev->lock);
836 		goto out;
837 	}
838 	qp_list.idx = 0;
839 
840 	/* add and ref each qp so it doesn't get freed */
841 	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
842 
843 	spin_unlock_irq(&ctx->dev->lock);
844 
845 	/* now traverse the list in a safe context to recover the db state*/
846 	recover_lost_dbs(ctx, &qp_list);
847 
848 	/* we're almost done!  deref the qps and clean up */
849 	deref_qps(qp_list);
850 	kfree(qp_list.qps);
851 
852 	/* Wait for the dbfifo to completely drain again */
853 	while (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1) > 0) {
854 		set_current_state(TASK_UNINTERRUPTIBLE);
855 		schedule_timeout(usecs_to_jiffies(10));
856 	}
857 
858 	/* resume the queues */
859 	spin_lock_irq(&ctx->dev->lock);
860 	if (ctx->dev->qpcnt > db_fc_threshold)
861 		ctx->dev->db_state = FLOW_CONTROL;
862 	else {
863 		ctx->dev->db_state = NORMAL;
864 		idr_for_each(&ctx->dev->qpidr, enable_qp_db, NULL);
865 	}
866 	ctx->dev->rdev.stats.db_state_transitions++;
867 	spin_unlock_irq(&ctx->dev->lock);
868 
869 out:
870 	/* start up kernel db ringers again */
871 	mutex_unlock(&ctx->dev->db_mutex);
872 }
873 
874 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
875 {
876 	struct uld_ctx *ctx = handle;
877 
878 	switch (control) {
879 	case CXGB4_CONTROL_DB_FULL:
880 		stop_queues(ctx);
881 		mutex_lock(&ctx->dev->rdev.stats.lock);
882 		ctx->dev->rdev.stats.db_full++;
883 		mutex_unlock(&ctx->dev->rdev.stats.lock);
884 		break;
885 	case CXGB4_CONTROL_DB_EMPTY:
886 		resume_queues(ctx);
887 		mutex_lock(&ctx->dev->rdev.stats.lock);
888 		ctx->dev->rdev.stats.db_empty++;
889 		mutex_unlock(&ctx->dev->rdev.stats.lock);
890 		break;
891 	case CXGB4_CONTROL_DB_DROP:
892 		recover_queues(ctx);
893 		mutex_lock(&ctx->dev->rdev.stats.lock);
894 		ctx->dev->rdev.stats.db_drop++;
895 		mutex_unlock(&ctx->dev->rdev.stats.lock);
896 		break;
897 	default:
898 		printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
899 		       pci_name(ctx->lldi.pdev), control);
900 		break;
901 	}
902 	return 0;
903 }
904 
905 static struct cxgb4_uld_info c4iw_uld_info = {
906 	.name = DRV_NAME,
907 	.add = c4iw_uld_add,
908 	.rx_handler = c4iw_uld_rx_handler,
909 	.state_change = c4iw_uld_state_change,
910 	.control = c4iw_uld_control,
911 };
912 
913 static int __init c4iw_init_module(void)
914 {
915 	int err;
916 
917 	err = c4iw_cm_init();
918 	if (err)
919 		return err;
920 
921 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
922 	if (!c4iw_debugfs_root)
923 		printk(KERN_WARNING MOD
924 		       "could not create debugfs entry, continuing\n");
925 
926 	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
927 
928 	return 0;
929 }
930 
931 static void __exit c4iw_exit_module(void)
932 {
933 	struct uld_ctx *ctx, *tmp;
934 
935 	mutex_lock(&dev_mutex);
936 	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
937 		if (ctx->dev)
938 			c4iw_remove(ctx);
939 		kfree(ctx);
940 	}
941 	mutex_unlock(&dev_mutex);
942 	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
943 	c4iw_cm_term();
944 	debugfs_remove_recursive(c4iw_debugfs_root);
945 }
946 
947 module_init(c4iw_init_module);
948 module_exit(c4iw_exit_module);
949