xref: /linux/drivers/infiniband/hw/qedr/main.c (revision bd628c1bed7902ec1f24ba0fe70758949146abbe)
1 /* QLogic qedr NIC Driver
2  * Copyright (c) 2015-2016  QLogic Corporation
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 <rdma/ib_verbs.h>
34 #include <rdma/ib_addr.h>
35 #include <rdma/ib_user_verbs.h>
36 #include <rdma/iw_cm.h>
37 #include <rdma/ib_mad.h>
38 #include <linux/netdevice.h>
39 #include <linux/iommu.h>
40 #include <linux/pci.h>
41 #include <net/addrconf.h>
42 #include <linux/idr.h>
43 
44 #include <linux/qed/qed_chain.h>
45 #include <linux/qed/qed_if.h>
46 #include "qedr.h"
47 #include "verbs.h"
48 #include <rdma/qedr-abi.h>
49 #include "qedr_iw_cm.h"
50 
51 MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
52 MODULE_AUTHOR("QLogic Corporation");
53 MODULE_LICENSE("Dual BSD/GPL");
54 
55 #define QEDR_WQ_MULTIPLIER_DFT	(3)
56 
57 static void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
58 				   enum ib_event_type type)
59 {
60 	struct ib_event ibev;
61 
62 	ibev.device = &dev->ibdev;
63 	ibev.element.port_num = port_num;
64 	ibev.event = type;
65 
66 	ib_dispatch_event(&ibev);
67 }
68 
69 static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
70 					    u8 port_num)
71 {
72 	return IB_LINK_LAYER_ETHERNET;
73 }
74 
75 static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str)
76 {
77 	struct qedr_dev *qedr = get_qedr_dev(ibdev);
78 	u32 fw_ver = (u32)qedr->attr.fw_ver;
79 
80 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d. %d. %d. %d",
81 		 (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF,
82 		 (fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
83 }
84 
85 static struct net_device *qedr_get_netdev(struct ib_device *dev, u8 port_num)
86 {
87 	struct qedr_dev *qdev;
88 
89 	qdev = get_qedr_dev(dev);
90 	dev_hold(qdev->ndev);
91 
92 	/* The HW vendor's device driver must guarantee
93 	 * that this function returns NULL before the net device has finished
94 	 * NETDEV_UNREGISTER state.
95 	 */
96 	return qdev->ndev;
97 }
98 
99 static int qedr_roce_port_immutable(struct ib_device *ibdev, u8 port_num,
100 				    struct ib_port_immutable *immutable)
101 {
102 	struct ib_port_attr attr;
103 	int err;
104 
105 	err = qedr_query_port(ibdev, port_num, &attr);
106 	if (err)
107 		return err;
108 
109 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
110 	immutable->gid_tbl_len = attr.gid_tbl_len;
111 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
112 	    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
113 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
114 
115 	return 0;
116 }
117 
118 static int qedr_iw_port_immutable(struct ib_device *ibdev, u8 port_num,
119 				  struct ib_port_immutable *immutable)
120 {
121 	struct ib_port_attr attr;
122 	int err;
123 
124 	err = qedr_query_port(ibdev, port_num, &attr);
125 	if (err)
126 		return err;
127 
128 	immutable->pkey_tbl_len = 1;
129 	immutable->gid_tbl_len = 1;
130 	immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
131 	immutable->max_mad_size = 0;
132 
133 	return 0;
134 }
135 
136 /* QEDR sysfs interface */
137 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
138 			   char *buf)
139 {
140 	struct qedr_dev *dev = dev_get_drvdata(device);
141 
142 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
143 }
144 static DEVICE_ATTR_RO(hw_rev);
145 
146 static ssize_t hca_type_show(struct device *device,
147 			     struct device_attribute *attr, char *buf)
148 {
149 	return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
150 }
151 static DEVICE_ATTR_RO(hca_type);
152 
153 static struct attribute *qedr_attributes[] = {
154 	&dev_attr_hw_rev.attr,
155 	&dev_attr_hca_type.attr,
156 	NULL
157 };
158 
159 static const struct attribute_group qedr_attr_group = {
160 	.attrs = qedr_attributes,
161 };
162 
163 static const struct ib_device_ops qedr_iw_dev_ops = {
164 	.get_port_immutable = qedr_iw_port_immutable,
165 	.query_gid = qedr_iw_query_gid,
166 };
167 
168 static int qedr_iw_register_device(struct qedr_dev *dev)
169 {
170 	dev->ibdev.node_type = RDMA_NODE_RNIC;
171 
172 	ib_set_device_ops(&dev->ibdev, &qedr_iw_dev_ops);
173 
174 	dev->ibdev.iwcm = kzalloc(sizeof(*dev->ibdev.iwcm), GFP_KERNEL);
175 	if (!dev->ibdev.iwcm)
176 		return -ENOMEM;
177 
178 	dev->ibdev.iwcm->connect = qedr_iw_connect;
179 	dev->ibdev.iwcm->accept = qedr_iw_accept;
180 	dev->ibdev.iwcm->reject = qedr_iw_reject;
181 	dev->ibdev.iwcm->create_listen = qedr_iw_create_listen;
182 	dev->ibdev.iwcm->destroy_listen = qedr_iw_destroy_listen;
183 	dev->ibdev.iwcm->add_ref = qedr_iw_qp_add_ref;
184 	dev->ibdev.iwcm->rem_ref = qedr_iw_qp_rem_ref;
185 	dev->ibdev.iwcm->get_qp = qedr_iw_get_qp;
186 
187 	memcpy(dev->ibdev.iwcm->ifname,
188 	       dev->ndev->name, sizeof(dev->ibdev.iwcm->ifname));
189 
190 	return 0;
191 }
192 
193 static const struct ib_device_ops qedr_roce_dev_ops = {
194 	.get_port_immutable = qedr_roce_port_immutable,
195 };
196 
197 static void qedr_roce_register_device(struct qedr_dev *dev)
198 {
199 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
200 
201 	ib_set_device_ops(&dev->ibdev, &qedr_roce_dev_ops);
202 }
203 
204 static const struct ib_device_ops qedr_dev_ops = {
205 	.alloc_mr = qedr_alloc_mr,
206 	.alloc_pd = qedr_alloc_pd,
207 	.alloc_ucontext = qedr_alloc_ucontext,
208 	.create_ah = qedr_create_ah,
209 	.create_cq = qedr_create_cq,
210 	.create_qp = qedr_create_qp,
211 	.create_srq = qedr_create_srq,
212 	.dealloc_pd = qedr_dealloc_pd,
213 	.dealloc_ucontext = qedr_dealloc_ucontext,
214 	.dereg_mr = qedr_dereg_mr,
215 	.destroy_ah = qedr_destroy_ah,
216 	.destroy_cq = qedr_destroy_cq,
217 	.destroy_qp = qedr_destroy_qp,
218 	.destroy_srq = qedr_destroy_srq,
219 	.get_dev_fw_str = qedr_get_dev_fw_str,
220 	.get_dma_mr = qedr_get_dma_mr,
221 	.get_link_layer = qedr_link_layer,
222 	.get_netdev = qedr_get_netdev,
223 	.map_mr_sg = qedr_map_mr_sg,
224 	.mmap = qedr_mmap,
225 	.modify_port = qedr_modify_port,
226 	.modify_qp = qedr_modify_qp,
227 	.modify_srq = qedr_modify_srq,
228 	.poll_cq = qedr_poll_cq,
229 	.post_recv = qedr_post_recv,
230 	.post_send = qedr_post_send,
231 	.post_srq_recv = qedr_post_srq_recv,
232 	.process_mad = qedr_process_mad,
233 	.query_device = qedr_query_device,
234 	.query_pkey = qedr_query_pkey,
235 	.query_port = qedr_query_port,
236 	.query_qp = qedr_query_qp,
237 	.query_srq = qedr_query_srq,
238 	.reg_user_mr = qedr_reg_user_mr,
239 	.req_notify_cq = qedr_arm_cq,
240 	.resize_cq = qedr_resize_cq,
241 };
242 
243 static int qedr_register_device(struct qedr_dev *dev)
244 {
245 	int rc;
246 
247 	dev->ibdev.node_guid = dev->attr.node_guid;
248 	memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
249 	dev->ibdev.owner = THIS_MODULE;
250 	dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
251 
252 	dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
253 				     QEDR_UVERBS(QUERY_DEVICE) |
254 				     QEDR_UVERBS(QUERY_PORT) |
255 				     QEDR_UVERBS(ALLOC_PD) |
256 				     QEDR_UVERBS(DEALLOC_PD) |
257 				     QEDR_UVERBS(CREATE_COMP_CHANNEL) |
258 				     QEDR_UVERBS(CREATE_CQ) |
259 				     QEDR_UVERBS(RESIZE_CQ) |
260 				     QEDR_UVERBS(DESTROY_CQ) |
261 				     QEDR_UVERBS(REQ_NOTIFY_CQ) |
262 				     QEDR_UVERBS(CREATE_QP) |
263 				     QEDR_UVERBS(MODIFY_QP) |
264 				     QEDR_UVERBS(QUERY_QP) |
265 				     QEDR_UVERBS(DESTROY_QP) |
266 				     QEDR_UVERBS(CREATE_SRQ) |
267 				     QEDR_UVERBS(DESTROY_SRQ) |
268 				     QEDR_UVERBS(QUERY_SRQ) |
269 				     QEDR_UVERBS(MODIFY_SRQ) |
270 				     QEDR_UVERBS(POST_SRQ_RECV) |
271 				     QEDR_UVERBS(REG_MR) |
272 				     QEDR_UVERBS(DEREG_MR) |
273 				     QEDR_UVERBS(POLL_CQ) |
274 				     QEDR_UVERBS(POST_SEND) |
275 				     QEDR_UVERBS(POST_RECV);
276 
277 	if (IS_IWARP(dev)) {
278 		rc = qedr_iw_register_device(dev);
279 		if (rc)
280 			return rc;
281 	} else {
282 		qedr_roce_register_device(dev);
283 	}
284 
285 	dev->ibdev.phys_port_cnt = 1;
286 	dev->ibdev.num_comp_vectors = dev->num_cnq;
287 	dev->ibdev.dev.parent = &dev->pdev->dev;
288 
289 	rdma_set_device_sysfs_group(&dev->ibdev, &qedr_attr_group);
290 	ib_set_device_ops(&dev->ibdev, &qedr_dev_ops);
291 
292 	dev->ibdev.driver_id = RDMA_DRIVER_QEDR;
293 	return ib_register_device(&dev->ibdev, "qedr%d", NULL);
294 }
295 
296 /* This function allocates fast-path status block memory */
297 static int qedr_alloc_mem_sb(struct qedr_dev *dev,
298 			     struct qed_sb_info *sb_info, u16 sb_id)
299 {
300 	struct status_block_e4 *sb_virt;
301 	dma_addr_t sb_phys;
302 	int rc;
303 
304 	sb_virt = dma_alloc_coherent(&dev->pdev->dev,
305 				     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
306 	if (!sb_virt)
307 		return -ENOMEM;
308 
309 	rc = dev->ops->common->sb_init(dev->cdev, sb_info,
310 				       sb_virt, sb_phys, sb_id,
311 				       QED_SB_TYPE_CNQ);
312 	if (rc) {
313 		pr_err("Status block initialization failed\n");
314 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
315 				  sb_virt, sb_phys);
316 		return rc;
317 	}
318 
319 	return 0;
320 }
321 
322 static void qedr_free_mem_sb(struct qedr_dev *dev,
323 			     struct qed_sb_info *sb_info, int sb_id)
324 {
325 	if (sb_info->sb_virt) {
326 		dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
327 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
328 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
329 	}
330 }
331 
332 static void qedr_free_resources(struct qedr_dev *dev)
333 {
334 	int i;
335 
336 	if (IS_IWARP(dev))
337 		destroy_workqueue(dev->iwarp_wq);
338 
339 	for (i = 0; i < dev->num_cnq; i++) {
340 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
341 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
342 	}
343 
344 	kfree(dev->cnq_array);
345 	kfree(dev->sb_array);
346 	kfree(dev->sgid_tbl);
347 }
348 
349 static int qedr_alloc_resources(struct qedr_dev *dev)
350 {
351 	struct qedr_cnq *cnq;
352 	__le16 *cons_pi;
353 	u16 n_entries;
354 	int i, rc;
355 
356 	dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
357 				GFP_KERNEL);
358 	if (!dev->sgid_tbl)
359 		return -ENOMEM;
360 
361 	spin_lock_init(&dev->sgid_lock);
362 
363 	if (IS_IWARP(dev)) {
364 		spin_lock_init(&dev->qpidr.idr_lock);
365 		idr_init(&dev->qpidr.idr);
366 		dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
367 	}
368 
369 	/* Allocate Status blocks for CNQ */
370 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
371 				GFP_KERNEL);
372 	if (!dev->sb_array) {
373 		rc = -ENOMEM;
374 		goto err1;
375 	}
376 
377 	dev->cnq_array = kcalloc(dev->num_cnq,
378 				 sizeof(*dev->cnq_array), GFP_KERNEL);
379 	if (!dev->cnq_array) {
380 		rc = -ENOMEM;
381 		goto err2;
382 	}
383 
384 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
385 
386 	/* Allocate CNQ PBLs */
387 	n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
388 	for (i = 0; i < dev->num_cnq; i++) {
389 		cnq = &dev->cnq_array[i];
390 
391 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
392 				       dev->sb_start + i);
393 		if (rc)
394 			goto err3;
395 
396 		rc = dev->ops->common->chain_alloc(dev->cdev,
397 						   QED_CHAIN_USE_TO_CONSUME,
398 						   QED_CHAIN_MODE_PBL,
399 						   QED_CHAIN_CNT_TYPE_U16,
400 						   n_entries,
401 						   sizeof(struct regpair *),
402 						   &cnq->pbl, NULL);
403 		if (rc)
404 			goto err4;
405 
406 		cnq->dev = dev;
407 		cnq->sb = &dev->sb_array[i];
408 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
409 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
410 		cnq->index = i;
411 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
412 
413 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
414 			 i, qed_chain_get_cons_idx(&cnq->pbl));
415 	}
416 
417 	return 0;
418 err4:
419 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
420 err3:
421 	for (--i; i >= 0; i--) {
422 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
423 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
424 	}
425 	kfree(dev->cnq_array);
426 err2:
427 	kfree(dev->sb_array);
428 err1:
429 	kfree(dev->sgid_tbl);
430 	return rc;
431 }
432 
433 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
434 {
435 	int rc = pci_enable_atomic_ops_to_root(pdev,
436 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
437 
438 	if (rc) {
439 		dev->atomic_cap = IB_ATOMIC_NONE;
440 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
441 	} else {
442 		dev->atomic_cap = IB_ATOMIC_GLOB;
443 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
444 	}
445 }
446 
447 static const struct qed_rdma_ops *qed_ops;
448 
449 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
450 
451 static irqreturn_t qedr_irq_handler(int irq, void *handle)
452 {
453 	u16 hw_comp_cons, sw_comp_cons;
454 	struct qedr_cnq *cnq = handle;
455 	struct regpair *cq_handle;
456 	struct qedr_cq *cq;
457 
458 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
459 
460 	qed_sb_update_sb_idx(cnq->sb);
461 
462 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
463 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
464 
465 	/* Align protocol-index and chain reads */
466 	rmb();
467 
468 	while (sw_comp_cons != hw_comp_cons) {
469 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
470 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
471 				cq_handle->lo);
472 
473 		if (cq == NULL) {
474 			DP_ERR(cnq->dev,
475 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
476 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
477 			       hw_comp_cons);
478 
479 			break;
480 		}
481 
482 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
483 			DP_ERR(cnq->dev,
484 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
485 			       cq_handle->hi, cq_handle->lo, cq);
486 			break;
487 		}
488 
489 		cq->arm_flags = 0;
490 
491 		if (!cq->destroyed && cq->ibcq.comp_handler)
492 			(*cq->ibcq.comp_handler)
493 				(&cq->ibcq, cq->ibcq.cq_context);
494 
495 		/* The CQ's CNQ notification counter is checked before
496 		 * destroying the CQ in a busy-wait loop that waits for all of
497 		 * the CQ's CNQ interrupts to be processed. It is increased
498 		 * here, only after the completion handler, to ensure that the
499 		 * the handler is not running when the CQ is destroyed.
500 		 */
501 		cq->cnq_notif++;
502 
503 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
504 
505 		cnq->n_comp++;
506 	}
507 
508 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
509 				      sw_comp_cons);
510 
511 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
512 
513 	return IRQ_HANDLED;
514 }
515 
516 static void qedr_sync_free_irqs(struct qedr_dev *dev)
517 {
518 	u32 vector;
519 	int i;
520 
521 	for (i = 0; i < dev->int_info.used_cnt; i++) {
522 		if (dev->int_info.msix_cnt) {
523 			vector = dev->int_info.msix[i * dev->num_hwfns].vector;
524 			synchronize_irq(vector);
525 			free_irq(vector, &dev->cnq_array[i]);
526 		}
527 	}
528 
529 	dev->int_info.used_cnt = 0;
530 }
531 
532 static int qedr_req_msix_irqs(struct qedr_dev *dev)
533 {
534 	int i, rc = 0;
535 
536 	if (dev->num_cnq > dev->int_info.msix_cnt) {
537 		DP_ERR(dev,
538 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
539 		       dev->num_cnq, dev->int_info.msix_cnt);
540 		return -EINVAL;
541 	}
542 
543 	for (i = 0; i < dev->num_cnq; i++) {
544 		rc = request_irq(dev->int_info.msix[i * dev->num_hwfns].vector,
545 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
546 				 &dev->cnq_array[i]);
547 		if (rc) {
548 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
549 			qedr_sync_free_irqs(dev);
550 		} else {
551 			DP_DEBUG(dev, QEDR_MSG_INIT,
552 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
553 				 dev->cnq_array[i].name, i,
554 				 &dev->cnq_array[i]);
555 			dev->int_info.used_cnt++;
556 		}
557 	}
558 
559 	return rc;
560 }
561 
562 static int qedr_setup_irqs(struct qedr_dev *dev)
563 {
564 	int rc;
565 
566 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
567 
568 	/* Learn Interrupt configuration */
569 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
570 	if (rc < 0)
571 		return rc;
572 
573 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
574 	if (rc) {
575 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
576 		return rc;
577 	}
578 
579 	if (dev->int_info.msix_cnt) {
580 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
581 			 dev->int_info.msix_cnt);
582 		rc = qedr_req_msix_irqs(dev);
583 		if (rc)
584 			return rc;
585 	}
586 
587 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
588 
589 	return 0;
590 }
591 
592 static int qedr_set_device_attr(struct qedr_dev *dev)
593 {
594 	struct qed_rdma_device *qed_attr;
595 	struct qedr_device_attr *attr;
596 	u32 page_size;
597 
598 	/* Part 1 - query core capabilities */
599 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
600 
601 	/* Part 2 - check capabilities */
602 	page_size = ~dev->attr.page_size_caps + 1;
603 	if (page_size > PAGE_SIZE) {
604 		DP_ERR(dev,
605 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
606 		       PAGE_SIZE, page_size);
607 		return -ENODEV;
608 	}
609 
610 	/* Part 3 - copy and update capabilities */
611 	attr = &dev->attr;
612 	attr->vendor_id = qed_attr->vendor_id;
613 	attr->vendor_part_id = qed_attr->vendor_part_id;
614 	attr->hw_ver = qed_attr->hw_ver;
615 	attr->fw_ver = qed_attr->fw_ver;
616 	attr->node_guid = qed_attr->node_guid;
617 	attr->sys_image_guid = qed_attr->sys_image_guid;
618 	attr->max_cnq = qed_attr->max_cnq;
619 	attr->max_sge = qed_attr->max_sge;
620 	attr->max_inline = qed_attr->max_inline;
621 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
622 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
623 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
624 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
625 	attr->max_dev_resp_rd_atomic_resc =
626 	    qed_attr->max_dev_resp_rd_atomic_resc;
627 	attr->max_cq = qed_attr->max_cq;
628 	attr->max_qp = qed_attr->max_qp;
629 	attr->max_mr = qed_attr->max_mr;
630 	attr->max_mr_size = qed_attr->max_mr_size;
631 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
632 	attr->max_mw = qed_attr->max_mw;
633 	attr->max_fmr = qed_attr->max_fmr;
634 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
635 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
636 	attr->max_pd = qed_attr->max_pd;
637 	attr->max_ah = qed_attr->max_ah;
638 	attr->max_pkey = qed_attr->max_pkey;
639 	attr->max_srq = qed_attr->max_srq;
640 	attr->max_srq_wr = qed_attr->max_srq_wr;
641 	attr->dev_caps = qed_attr->dev_caps;
642 	attr->page_size_caps = qed_attr->page_size_caps;
643 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
644 	attr->reserved_lkey = qed_attr->reserved_lkey;
645 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
646 	attr->max_stats_queues = qed_attr->max_stats_queues;
647 
648 	return 0;
649 }
650 
651 static void qedr_unaffiliated_event(void *context, u8 event_code)
652 {
653 	pr_err("unaffiliated event not implemented yet\n");
654 }
655 
656 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
657 {
658 #define EVENT_TYPE_NOT_DEFINED	0
659 #define EVENT_TYPE_CQ		1
660 #define EVENT_TYPE_QP		2
661 #define EVENT_TYPE_SRQ		3
662 	struct qedr_dev *dev = (struct qedr_dev *)context;
663 	struct regpair *async_handle = (struct regpair *)fw_handle;
664 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
665 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
666 	struct ib_event event;
667 	struct ib_srq *ibsrq;
668 	struct qedr_srq *srq;
669 	unsigned long flags;
670 	struct ib_cq *ibcq;
671 	struct ib_qp *ibqp;
672 	struct qedr_cq *cq;
673 	struct qedr_qp *qp;
674 	u16 srq_id;
675 
676 	if (IS_ROCE(dev)) {
677 		switch (e_code) {
678 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
679 			event.event = IB_EVENT_CQ_ERR;
680 			event_type = EVENT_TYPE_CQ;
681 			break;
682 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
683 			event.event = IB_EVENT_SQ_DRAINED;
684 			event_type = EVENT_TYPE_QP;
685 			break;
686 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
687 			event.event = IB_EVENT_QP_FATAL;
688 			event_type = EVENT_TYPE_QP;
689 			break;
690 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
691 			event.event = IB_EVENT_QP_REQ_ERR;
692 			event_type = EVENT_TYPE_QP;
693 			break;
694 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
695 			event.event = IB_EVENT_QP_ACCESS_ERR;
696 			event_type = EVENT_TYPE_QP;
697 			break;
698 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
699 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
700 			event_type = EVENT_TYPE_SRQ;
701 			break;
702 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
703 			event.event = IB_EVENT_SRQ_ERR;
704 			event_type = EVENT_TYPE_SRQ;
705 			break;
706 		default:
707 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
708 			       e_code, roce_handle64);
709 		}
710 	} else {
711 		switch (e_code) {
712 		case QED_IWARP_EVENT_SRQ_LIMIT:
713 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
714 			event_type = EVENT_TYPE_SRQ;
715 			break;
716 		case QED_IWARP_EVENT_SRQ_EMPTY:
717 			event.event = IB_EVENT_SRQ_ERR;
718 			event_type = EVENT_TYPE_SRQ;
719 			break;
720 		default:
721 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
722 		       roce_handle64);
723 		}
724 	}
725 	switch (event_type) {
726 	case EVENT_TYPE_CQ:
727 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
728 		if (cq) {
729 			ibcq = &cq->ibcq;
730 			if (ibcq->event_handler) {
731 				event.device = ibcq->device;
732 				event.element.cq = ibcq;
733 				ibcq->event_handler(&event, ibcq->cq_context);
734 			}
735 		} else {
736 			WARN(1,
737 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
738 			     roce_handle64);
739 		}
740 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
741 		break;
742 	case EVENT_TYPE_QP:
743 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
744 		if (qp) {
745 			ibqp = &qp->ibqp;
746 			if (ibqp->event_handler) {
747 				event.device = ibqp->device;
748 				event.element.qp = ibqp;
749 				ibqp->event_handler(&event, ibqp->qp_context);
750 			}
751 		} else {
752 			WARN(1,
753 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
754 			     roce_handle64);
755 		}
756 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
757 		break;
758 	case EVENT_TYPE_SRQ:
759 		srq_id = (u16)roce_handle64;
760 		spin_lock_irqsave(&dev->srqidr.idr_lock, flags);
761 		srq = idr_find(&dev->srqidr.idr, srq_id);
762 		if (srq) {
763 			ibsrq = &srq->ibsrq;
764 			if (ibsrq->event_handler) {
765 				event.device = ibsrq->device;
766 				event.element.srq = ibsrq;
767 				ibsrq->event_handler(&event,
768 						     ibsrq->srq_context);
769 			}
770 		} else {
771 			DP_NOTICE(dev,
772 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
773 				  roce_handle64);
774 		}
775 		spin_unlock_irqrestore(&dev->srqidr.idr_lock, flags);
776 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
777 	default:
778 		break;
779 	}
780 }
781 
782 static int qedr_init_hw(struct qedr_dev *dev)
783 {
784 	struct qed_rdma_add_user_out_params out_params;
785 	struct qed_rdma_start_in_params *in_params;
786 	struct qed_rdma_cnq_params *cur_pbl;
787 	struct qed_rdma_events events;
788 	dma_addr_t p_phys_table;
789 	u32 page_cnt;
790 	int rc = 0;
791 	int i;
792 
793 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
794 	if (!in_params) {
795 		rc = -ENOMEM;
796 		goto out;
797 	}
798 
799 	in_params->desired_cnq = dev->num_cnq;
800 	for (i = 0; i < dev->num_cnq; i++) {
801 		cur_pbl = &in_params->cnq_pbl_list[i];
802 
803 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
804 		cur_pbl->num_pbl_pages = page_cnt;
805 
806 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
807 		cur_pbl->pbl_ptr = (u64)p_phys_table;
808 	}
809 
810 	events.affiliated_event = qedr_affiliated_event;
811 	events.unaffiliated_event = qedr_unaffiliated_event;
812 	events.context = dev;
813 
814 	in_params->events = &events;
815 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
816 	in_params->max_mtu = dev->ndev->mtu;
817 	dev->iwarp_max_mtu = dev->ndev->mtu;
818 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
819 
820 	rc = dev->ops->rdma_init(dev->cdev, in_params);
821 	if (rc)
822 		goto out;
823 
824 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
825 	if (rc)
826 		goto out;
827 
828 	dev->db_addr = (void __iomem *)(uintptr_t)out_params.dpi_addr;
829 	dev->db_phys_addr = out_params.dpi_phys_addr;
830 	dev->db_size = out_params.dpi_size;
831 	dev->dpi = out_params.dpi;
832 
833 	rc = qedr_set_device_attr(dev);
834 out:
835 	kfree(in_params);
836 	if (rc)
837 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
838 
839 	return rc;
840 }
841 
842 static void qedr_stop_hw(struct qedr_dev *dev)
843 {
844 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
845 	dev->ops->rdma_stop(dev->rdma_ctx);
846 }
847 
848 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
849 				 struct net_device *ndev)
850 {
851 	struct qed_dev_rdma_info dev_info;
852 	struct qedr_dev *dev;
853 	int rc = 0;
854 
855 	dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
856 	if (!dev) {
857 		pr_err("Unable to allocate ib device\n");
858 		return NULL;
859 	}
860 
861 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
862 
863 	dev->pdev = pdev;
864 	dev->ndev = ndev;
865 	dev->cdev = cdev;
866 
867 	qed_ops = qed_get_rdma_ops();
868 	if (!qed_ops) {
869 		DP_ERR(dev, "Failed to get qed roce operations\n");
870 		goto init_err;
871 	}
872 
873 	dev->ops = qed_ops;
874 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
875 	if (rc)
876 		goto init_err;
877 
878 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
879 	dev->rdma_type = dev_info.rdma_type;
880 	dev->num_hwfns = dev_info.common.num_hwfns;
881 	dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
882 
883 	dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
884 	if (!dev->num_cnq) {
885 		DP_ERR(dev, "Failed. At least one CNQ is required.\n");
886 		rc = -ENOMEM;
887 		goto init_err;
888 	}
889 
890 	dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
891 
892 	qedr_pci_set_atomic(dev, pdev);
893 
894 	rc = qedr_alloc_resources(dev);
895 	if (rc)
896 		goto init_err;
897 
898 	rc = qedr_init_hw(dev);
899 	if (rc)
900 		goto alloc_err;
901 
902 	rc = qedr_setup_irqs(dev);
903 	if (rc)
904 		goto irq_err;
905 
906 	rc = qedr_register_device(dev);
907 	if (rc) {
908 		DP_ERR(dev, "Unable to allocate register device\n");
909 		goto reg_err;
910 	}
911 
912 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
913 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
914 
915 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
916 	return dev;
917 
918 reg_err:
919 	qedr_sync_free_irqs(dev);
920 irq_err:
921 	qedr_stop_hw(dev);
922 alloc_err:
923 	qedr_free_resources(dev);
924 init_err:
925 	ib_dealloc_device(&dev->ibdev);
926 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
927 
928 	return NULL;
929 }
930 
931 static void qedr_remove(struct qedr_dev *dev)
932 {
933 	/* First unregister with stack to stop all the active traffic
934 	 * of the registered clients.
935 	 */
936 	ib_unregister_device(&dev->ibdev);
937 
938 	qedr_stop_hw(dev);
939 	qedr_sync_free_irqs(dev);
940 	qedr_free_resources(dev);
941 	ib_dealloc_device(&dev->ibdev);
942 }
943 
944 static void qedr_close(struct qedr_dev *dev)
945 {
946 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
947 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
948 }
949 
950 static void qedr_shutdown(struct qedr_dev *dev)
951 {
952 	qedr_close(dev);
953 	qedr_remove(dev);
954 }
955 
956 static void qedr_open(struct qedr_dev *dev)
957 {
958 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
959 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
960 }
961 
962 static void qedr_mac_address_change(struct qedr_dev *dev)
963 {
964 	union ib_gid *sgid = &dev->sgid_tbl[0];
965 	u8 guid[8], mac_addr[6];
966 	int rc;
967 
968 	/* Update SGID */
969 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
970 	guid[0] = mac_addr[0] ^ 2;
971 	guid[1] = mac_addr[1];
972 	guid[2] = mac_addr[2];
973 	guid[3] = 0xff;
974 	guid[4] = 0xfe;
975 	guid[5] = mac_addr[3];
976 	guid[6] = mac_addr[4];
977 	guid[7] = mac_addr[5];
978 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
979 	memcpy(&sgid->raw[8], guid, sizeof(guid));
980 
981 	/* Update LL2 */
982 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
983 					  dev->gsi_ll2_mac_address,
984 					  dev->ndev->dev_addr);
985 
986 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
987 
988 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
989 
990 	if (rc)
991 		DP_ERR(dev, "Error updating mac filter\n");
992 }
993 
994 /* event handling via NIC driver ensures that all the NIC specific
995  * initialization done before RoCE driver notifies
996  * event to stack.
997  */
998 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
999 {
1000 	switch (event) {
1001 	case QEDE_UP:
1002 		qedr_open(dev);
1003 		break;
1004 	case QEDE_DOWN:
1005 		qedr_close(dev);
1006 		break;
1007 	case QEDE_CLOSE:
1008 		qedr_shutdown(dev);
1009 		break;
1010 	case QEDE_CHANGE_ADDR:
1011 		qedr_mac_address_change(dev);
1012 		break;
1013 	default:
1014 		pr_err("Event not supported\n");
1015 	}
1016 }
1017 
1018 static struct qedr_driver qedr_drv = {
1019 	.name = "qedr_driver",
1020 	.add = qedr_add,
1021 	.remove = qedr_remove,
1022 	.notify = qedr_notify,
1023 };
1024 
1025 static int __init qedr_init_module(void)
1026 {
1027 	return qede_rdma_register_driver(&qedr_drv);
1028 }
1029 
1030 static void __exit qedr_exit_module(void)
1031 {
1032 	qede_rdma_unregister_driver(&qedr_drv);
1033 }
1034 
1035 module_init(qedr_init_module);
1036 module_exit(qedr_exit_module);
1037