xref: /openbmc/linux/drivers/infiniband/hw/qedr/main.c (revision 09bae3b6)
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 static int qedr_iw_register_device(struct qedr_dev *dev)
137 {
138 	dev->ibdev.node_type = RDMA_NODE_RNIC;
139 	dev->ibdev.query_gid = qedr_iw_query_gid;
140 
141 	dev->ibdev.get_port_immutable = qedr_iw_port_immutable;
142 
143 	dev->ibdev.iwcm = kzalloc(sizeof(*dev->ibdev.iwcm), GFP_KERNEL);
144 	if (!dev->ibdev.iwcm)
145 		return -ENOMEM;
146 
147 	dev->ibdev.iwcm->connect = qedr_iw_connect;
148 	dev->ibdev.iwcm->accept = qedr_iw_accept;
149 	dev->ibdev.iwcm->reject = qedr_iw_reject;
150 	dev->ibdev.iwcm->create_listen = qedr_iw_create_listen;
151 	dev->ibdev.iwcm->destroy_listen = qedr_iw_destroy_listen;
152 	dev->ibdev.iwcm->add_ref = qedr_iw_qp_add_ref;
153 	dev->ibdev.iwcm->rem_ref = qedr_iw_qp_rem_ref;
154 	dev->ibdev.iwcm->get_qp = qedr_iw_get_qp;
155 
156 	memcpy(dev->ibdev.iwcm->ifname,
157 	       dev->ndev->name, sizeof(dev->ibdev.iwcm->ifname));
158 
159 	return 0;
160 }
161 
162 static void qedr_roce_register_device(struct qedr_dev *dev)
163 {
164 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
165 
166 	dev->ibdev.get_port_immutable = qedr_roce_port_immutable;
167 }
168 
169 static int qedr_register_device(struct qedr_dev *dev)
170 {
171 	int rc;
172 
173 	strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
174 
175 	dev->ibdev.node_guid = dev->attr.node_guid;
176 	memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
177 	dev->ibdev.owner = THIS_MODULE;
178 	dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
179 
180 	dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
181 				     QEDR_UVERBS(QUERY_DEVICE) |
182 				     QEDR_UVERBS(QUERY_PORT) |
183 				     QEDR_UVERBS(ALLOC_PD) |
184 				     QEDR_UVERBS(DEALLOC_PD) |
185 				     QEDR_UVERBS(CREATE_COMP_CHANNEL) |
186 				     QEDR_UVERBS(CREATE_CQ) |
187 				     QEDR_UVERBS(RESIZE_CQ) |
188 				     QEDR_UVERBS(DESTROY_CQ) |
189 				     QEDR_UVERBS(REQ_NOTIFY_CQ) |
190 				     QEDR_UVERBS(CREATE_QP) |
191 				     QEDR_UVERBS(MODIFY_QP) |
192 				     QEDR_UVERBS(QUERY_QP) |
193 				     QEDR_UVERBS(DESTROY_QP) |
194 				     QEDR_UVERBS(CREATE_SRQ) |
195 				     QEDR_UVERBS(DESTROY_SRQ) |
196 				     QEDR_UVERBS(QUERY_SRQ) |
197 				     QEDR_UVERBS(MODIFY_SRQ) |
198 				     QEDR_UVERBS(POST_SRQ_RECV) |
199 				     QEDR_UVERBS(REG_MR) |
200 				     QEDR_UVERBS(DEREG_MR) |
201 				     QEDR_UVERBS(POLL_CQ) |
202 				     QEDR_UVERBS(POST_SEND) |
203 				     QEDR_UVERBS(POST_RECV);
204 
205 	if (IS_IWARP(dev)) {
206 		rc = qedr_iw_register_device(dev);
207 		if (rc)
208 			return rc;
209 	} else {
210 		qedr_roce_register_device(dev);
211 	}
212 
213 	dev->ibdev.phys_port_cnt = 1;
214 	dev->ibdev.num_comp_vectors = dev->num_cnq;
215 
216 	dev->ibdev.query_device = qedr_query_device;
217 	dev->ibdev.query_port = qedr_query_port;
218 	dev->ibdev.modify_port = qedr_modify_port;
219 
220 	dev->ibdev.alloc_ucontext = qedr_alloc_ucontext;
221 	dev->ibdev.dealloc_ucontext = qedr_dealloc_ucontext;
222 	dev->ibdev.mmap = qedr_mmap;
223 
224 	dev->ibdev.alloc_pd = qedr_alloc_pd;
225 	dev->ibdev.dealloc_pd = qedr_dealloc_pd;
226 
227 	dev->ibdev.create_cq = qedr_create_cq;
228 	dev->ibdev.destroy_cq = qedr_destroy_cq;
229 	dev->ibdev.resize_cq = qedr_resize_cq;
230 	dev->ibdev.req_notify_cq = qedr_arm_cq;
231 
232 	dev->ibdev.create_qp = qedr_create_qp;
233 	dev->ibdev.modify_qp = qedr_modify_qp;
234 	dev->ibdev.query_qp = qedr_query_qp;
235 	dev->ibdev.destroy_qp = qedr_destroy_qp;
236 
237 	dev->ibdev.create_srq = qedr_create_srq;
238 	dev->ibdev.destroy_srq = qedr_destroy_srq;
239 	dev->ibdev.modify_srq = qedr_modify_srq;
240 	dev->ibdev.query_srq = qedr_query_srq;
241 	dev->ibdev.post_srq_recv = qedr_post_srq_recv;
242 	dev->ibdev.query_pkey = qedr_query_pkey;
243 
244 	dev->ibdev.create_ah = qedr_create_ah;
245 	dev->ibdev.destroy_ah = qedr_destroy_ah;
246 
247 	dev->ibdev.get_dma_mr = qedr_get_dma_mr;
248 	dev->ibdev.dereg_mr = qedr_dereg_mr;
249 	dev->ibdev.reg_user_mr = qedr_reg_user_mr;
250 	dev->ibdev.alloc_mr = qedr_alloc_mr;
251 	dev->ibdev.map_mr_sg = qedr_map_mr_sg;
252 
253 	dev->ibdev.poll_cq = qedr_poll_cq;
254 	dev->ibdev.post_send = qedr_post_send;
255 	dev->ibdev.post_recv = qedr_post_recv;
256 
257 	dev->ibdev.process_mad = qedr_process_mad;
258 
259 	dev->ibdev.get_netdev = qedr_get_netdev;
260 
261 	dev->ibdev.dev.parent = &dev->pdev->dev;
262 
263 	dev->ibdev.get_link_layer = qedr_link_layer;
264 	dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
265 
266 	dev->ibdev.driver_id = RDMA_DRIVER_QEDR;
267 	return ib_register_device(&dev->ibdev, NULL);
268 }
269 
270 /* This function allocates fast-path status block memory */
271 static int qedr_alloc_mem_sb(struct qedr_dev *dev,
272 			     struct qed_sb_info *sb_info, u16 sb_id)
273 {
274 	struct status_block_e4 *sb_virt;
275 	dma_addr_t sb_phys;
276 	int rc;
277 
278 	sb_virt = dma_alloc_coherent(&dev->pdev->dev,
279 				     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
280 	if (!sb_virt)
281 		return -ENOMEM;
282 
283 	rc = dev->ops->common->sb_init(dev->cdev, sb_info,
284 				       sb_virt, sb_phys, sb_id,
285 				       QED_SB_TYPE_CNQ);
286 	if (rc) {
287 		pr_err("Status block initialization failed\n");
288 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
289 				  sb_virt, sb_phys);
290 		return rc;
291 	}
292 
293 	return 0;
294 }
295 
296 static void qedr_free_mem_sb(struct qedr_dev *dev,
297 			     struct qed_sb_info *sb_info, int sb_id)
298 {
299 	if (sb_info->sb_virt) {
300 		dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
301 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
302 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
303 	}
304 }
305 
306 static void qedr_free_resources(struct qedr_dev *dev)
307 {
308 	int i;
309 
310 	if (IS_IWARP(dev))
311 		destroy_workqueue(dev->iwarp_wq);
312 
313 	for (i = 0; i < dev->num_cnq; i++) {
314 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
315 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
316 	}
317 
318 	kfree(dev->cnq_array);
319 	kfree(dev->sb_array);
320 	kfree(dev->sgid_tbl);
321 }
322 
323 static int qedr_alloc_resources(struct qedr_dev *dev)
324 {
325 	struct qedr_cnq *cnq;
326 	__le16 *cons_pi;
327 	u16 n_entries;
328 	int i, rc;
329 
330 	dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
331 				GFP_KERNEL);
332 	if (!dev->sgid_tbl)
333 		return -ENOMEM;
334 
335 	spin_lock_init(&dev->sgid_lock);
336 
337 	if (IS_IWARP(dev)) {
338 		spin_lock_init(&dev->qpidr.idr_lock);
339 		idr_init(&dev->qpidr.idr);
340 		dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
341 	}
342 
343 	/* Allocate Status blocks for CNQ */
344 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
345 				GFP_KERNEL);
346 	if (!dev->sb_array) {
347 		rc = -ENOMEM;
348 		goto err1;
349 	}
350 
351 	dev->cnq_array = kcalloc(dev->num_cnq,
352 				 sizeof(*dev->cnq_array), GFP_KERNEL);
353 	if (!dev->cnq_array) {
354 		rc = -ENOMEM;
355 		goto err2;
356 	}
357 
358 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
359 
360 	/* Allocate CNQ PBLs */
361 	n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
362 	for (i = 0; i < dev->num_cnq; i++) {
363 		cnq = &dev->cnq_array[i];
364 
365 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
366 				       dev->sb_start + i);
367 		if (rc)
368 			goto err3;
369 
370 		rc = dev->ops->common->chain_alloc(dev->cdev,
371 						   QED_CHAIN_USE_TO_CONSUME,
372 						   QED_CHAIN_MODE_PBL,
373 						   QED_CHAIN_CNT_TYPE_U16,
374 						   n_entries,
375 						   sizeof(struct regpair *),
376 						   &cnq->pbl, NULL);
377 		if (rc)
378 			goto err4;
379 
380 		cnq->dev = dev;
381 		cnq->sb = &dev->sb_array[i];
382 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
383 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
384 		cnq->index = i;
385 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
386 
387 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
388 			 i, qed_chain_get_cons_idx(&cnq->pbl));
389 	}
390 
391 	return 0;
392 err4:
393 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
394 err3:
395 	for (--i; i >= 0; i--) {
396 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
397 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
398 	}
399 	kfree(dev->cnq_array);
400 err2:
401 	kfree(dev->sb_array);
402 err1:
403 	kfree(dev->sgid_tbl);
404 	return rc;
405 }
406 
407 /* QEDR sysfs interface */
408 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
409 			char *buf)
410 {
411 	struct qedr_dev *dev = dev_get_drvdata(device);
412 
413 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
414 }
415 
416 static ssize_t show_hca_type(struct device *device,
417 			     struct device_attribute *attr, char *buf)
418 {
419 	return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
420 }
421 
422 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
423 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
424 
425 static struct device_attribute *qedr_attributes[] = {
426 	&dev_attr_hw_rev,
427 	&dev_attr_hca_type
428 };
429 
430 static void qedr_remove_sysfiles(struct qedr_dev *dev)
431 {
432 	int i;
433 
434 	for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
435 		device_remove_file(&dev->ibdev.dev, qedr_attributes[i]);
436 }
437 
438 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
439 {
440 	int rc = pci_enable_atomic_ops_to_root(pdev,
441 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
442 
443 	if (rc) {
444 		dev->atomic_cap = IB_ATOMIC_NONE;
445 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
446 	} else {
447 		dev->atomic_cap = IB_ATOMIC_GLOB;
448 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
449 	}
450 }
451 
452 static const struct qed_rdma_ops *qed_ops;
453 
454 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
455 
456 static irqreturn_t qedr_irq_handler(int irq, void *handle)
457 {
458 	u16 hw_comp_cons, sw_comp_cons;
459 	struct qedr_cnq *cnq = handle;
460 	struct regpair *cq_handle;
461 	struct qedr_cq *cq;
462 
463 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
464 
465 	qed_sb_update_sb_idx(cnq->sb);
466 
467 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
468 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
469 
470 	/* Align protocol-index and chain reads */
471 	rmb();
472 
473 	while (sw_comp_cons != hw_comp_cons) {
474 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
475 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
476 				cq_handle->lo);
477 
478 		if (cq == NULL) {
479 			DP_ERR(cnq->dev,
480 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
481 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
482 			       hw_comp_cons);
483 
484 			break;
485 		}
486 
487 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
488 			DP_ERR(cnq->dev,
489 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
490 			       cq_handle->hi, cq_handle->lo, cq);
491 			break;
492 		}
493 
494 		cq->arm_flags = 0;
495 
496 		if (!cq->destroyed && cq->ibcq.comp_handler)
497 			(*cq->ibcq.comp_handler)
498 				(&cq->ibcq, cq->ibcq.cq_context);
499 
500 		/* The CQ's CNQ notification counter is checked before
501 		 * destroying the CQ in a busy-wait loop that waits for all of
502 		 * the CQ's CNQ interrupts to be processed. It is increased
503 		 * here, only after the completion handler, to ensure that the
504 		 * the handler is not running when the CQ is destroyed.
505 		 */
506 		cq->cnq_notif++;
507 
508 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
509 
510 		cnq->n_comp++;
511 	}
512 
513 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
514 				      sw_comp_cons);
515 
516 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
517 
518 	return IRQ_HANDLED;
519 }
520 
521 static void qedr_sync_free_irqs(struct qedr_dev *dev)
522 {
523 	u32 vector;
524 	int i;
525 
526 	for (i = 0; i < dev->int_info.used_cnt; i++) {
527 		if (dev->int_info.msix_cnt) {
528 			vector = dev->int_info.msix[i * dev->num_hwfns].vector;
529 			synchronize_irq(vector);
530 			free_irq(vector, &dev->cnq_array[i]);
531 		}
532 	}
533 
534 	dev->int_info.used_cnt = 0;
535 }
536 
537 static int qedr_req_msix_irqs(struct qedr_dev *dev)
538 {
539 	int i, rc = 0;
540 
541 	if (dev->num_cnq > dev->int_info.msix_cnt) {
542 		DP_ERR(dev,
543 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
544 		       dev->num_cnq, dev->int_info.msix_cnt);
545 		return -EINVAL;
546 	}
547 
548 	for (i = 0; i < dev->num_cnq; i++) {
549 		rc = request_irq(dev->int_info.msix[i * dev->num_hwfns].vector,
550 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
551 				 &dev->cnq_array[i]);
552 		if (rc) {
553 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
554 			qedr_sync_free_irqs(dev);
555 		} else {
556 			DP_DEBUG(dev, QEDR_MSG_INIT,
557 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
558 				 dev->cnq_array[i].name, i,
559 				 &dev->cnq_array[i]);
560 			dev->int_info.used_cnt++;
561 		}
562 	}
563 
564 	return rc;
565 }
566 
567 static int qedr_setup_irqs(struct qedr_dev *dev)
568 {
569 	int rc;
570 
571 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
572 
573 	/* Learn Interrupt configuration */
574 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
575 	if (rc < 0)
576 		return rc;
577 
578 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
579 	if (rc) {
580 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
581 		return rc;
582 	}
583 
584 	if (dev->int_info.msix_cnt) {
585 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
586 			 dev->int_info.msix_cnt);
587 		rc = qedr_req_msix_irqs(dev);
588 		if (rc)
589 			return rc;
590 	}
591 
592 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
593 
594 	return 0;
595 }
596 
597 static int qedr_set_device_attr(struct qedr_dev *dev)
598 {
599 	struct qed_rdma_device *qed_attr;
600 	struct qedr_device_attr *attr;
601 	u32 page_size;
602 
603 	/* Part 1 - query core capabilities */
604 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
605 
606 	/* Part 2 - check capabilities */
607 	page_size = ~dev->attr.page_size_caps + 1;
608 	if (page_size > PAGE_SIZE) {
609 		DP_ERR(dev,
610 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
611 		       PAGE_SIZE, page_size);
612 		return -ENODEV;
613 	}
614 
615 	/* Part 3 - copy and update capabilities */
616 	attr = &dev->attr;
617 	attr->vendor_id = qed_attr->vendor_id;
618 	attr->vendor_part_id = qed_attr->vendor_part_id;
619 	attr->hw_ver = qed_attr->hw_ver;
620 	attr->fw_ver = qed_attr->fw_ver;
621 	attr->node_guid = qed_attr->node_guid;
622 	attr->sys_image_guid = qed_attr->sys_image_guid;
623 	attr->max_cnq = qed_attr->max_cnq;
624 	attr->max_sge = qed_attr->max_sge;
625 	attr->max_inline = qed_attr->max_inline;
626 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
627 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
628 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
629 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
630 	attr->max_dev_resp_rd_atomic_resc =
631 	    qed_attr->max_dev_resp_rd_atomic_resc;
632 	attr->max_cq = qed_attr->max_cq;
633 	attr->max_qp = qed_attr->max_qp;
634 	attr->max_mr = qed_attr->max_mr;
635 	attr->max_mr_size = qed_attr->max_mr_size;
636 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
637 	attr->max_mw = qed_attr->max_mw;
638 	attr->max_fmr = qed_attr->max_fmr;
639 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
640 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
641 	attr->max_pd = qed_attr->max_pd;
642 	attr->max_ah = qed_attr->max_ah;
643 	attr->max_pkey = qed_attr->max_pkey;
644 	attr->max_srq = qed_attr->max_srq;
645 	attr->max_srq_wr = qed_attr->max_srq_wr;
646 	attr->dev_caps = qed_attr->dev_caps;
647 	attr->page_size_caps = qed_attr->page_size_caps;
648 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
649 	attr->reserved_lkey = qed_attr->reserved_lkey;
650 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
651 	attr->max_stats_queues = qed_attr->max_stats_queues;
652 
653 	return 0;
654 }
655 
656 static void qedr_unaffiliated_event(void *context, u8 event_code)
657 {
658 	pr_err("unaffiliated event not implemented yet\n");
659 }
660 
661 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
662 {
663 #define EVENT_TYPE_NOT_DEFINED	0
664 #define EVENT_TYPE_CQ		1
665 #define EVENT_TYPE_QP		2
666 #define EVENT_TYPE_SRQ		3
667 	struct qedr_dev *dev = (struct qedr_dev *)context;
668 	struct regpair *async_handle = (struct regpair *)fw_handle;
669 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
670 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
671 	struct ib_event event;
672 	struct ib_srq *ibsrq;
673 	struct qedr_srq *srq;
674 	unsigned long flags;
675 	struct ib_cq *ibcq;
676 	struct ib_qp *ibqp;
677 	struct qedr_cq *cq;
678 	struct qedr_qp *qp;
679 	u16 srq_id;
680 
681 	if (IS_ROCE(dev)) {
682 		switch (e_code) {
683 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
684 			event.event = IB_EVENT_CQ_ERR;
685 			event_type = EVENT_TYPE_CQ;
686 			break;
687 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
688 			event.event = IB_EVENT_SQ_DRAINED;
689 			event_type = EVENT_TYPE_QP;
690 			break;
691 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
692 			event.event = IB_EVENT_QP_FATAL;
693 			event_type = EVENT_TYPE_QP;
694 			break;
695 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
696 			event.event = IB_EVENT_QP_REQ_ERR;
697 			event_type = EVENT_TYPE_QP;
698 			break;
699 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
700 			event.event = IB_EVENT_QP_ACCESS_ERR;
701 			event_type = EVENT_TYPE_QP;
702 			break;
703 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
704 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
705 			event_type = EVENT_TYPE_SRQ;
706 			break;
707 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
708 			event.event = IB_EVENT_SRQ_ERR;
709 			event_type = EVENT_TYPE_SRQ;
710 			break;
711 		default:
712 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
713 			       e_code, roce_handle64);
714 		}
715 	} else {
716 		switch (e_code) {
717 		case QED_IWARP_EVENT_SRQ_LIMIT:
718 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
719 			event_type = EVENT_TYPE_SRQ;
720 			break;
721 		case QED_IWARP_EVENT_SRQ_EMPTY:
722 			event.event = IB_EVENT_SRQ_ERR;
723 			event_type = EVENT_TYPE_SRQ;
724 			break;
725 		default:
726 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
727 		       roce_handle64);
728 		}
729 	}
730 	switch (event_type) {
731 	case EVENT_TYPE_CQ:
732 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
733 		if (cq) {
734 			ibcq = &cq->ibcq;
735 			if (ibcq->event_handler) {
736 				event.device = ibcq->device;
737 				event.element.cq = ibcq;
738 				ibcq->event_handler(&event, ibcq->cq_context);
739 			}
740 		} else {
741 			WARN(1,
742 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
743 			     roce_handle64);
744 		}
745 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
746 		break;
747 	case EVENT_TYPE_QP:
748 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
749 		if (qp) {
750 			ibqp = &qp->ibqp;
751 			if (ibqp->event_handler) {
752 				event.device = ibqp->device;
753 				event.element.qp = ibqp;
754 				ibqp->event_handler(&event, ibqp->qp_context);
755 			}
756 		} else {
757 			WARN(1,
758 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
759 			     roce_handle64);
760 		}
761 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
762 		break;
763 	case EVENT_TYPE_SRQ:
764 		srq_id = (u16)roce_handle64;
765 		spin_lock_irqsave(&dev->srqidr.idr_lock, flags);
766 		srq = idr_find(&dev->srqidr.idr, srq_id);
767 		if (srq) {
768 			ibsrq = &srq->ibsrq;
769 			if (ibsrq->event_handler) {
770 				event.device = ibsrq->device;
771 				event.element.srq = ibsrq;
772 				ibsrq->event_handler(&event,
773 						     ibsrq->srq_context);
774 			}
775 		} else {
776 			DP_NOTICE(dev,
777 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
778 				  roce_handle64);
779 		}
780 		spin_unlock_irqrestore(&dev->srqidr.idr_lock, flags);
781 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
782 	default:
783 		break;
784 	}
785 }
786 
787 static int qedr_init_hw(struct qedr_dev *dev)
788 {
789 	struct qed_rdma_add_user_out_params out_params;
790 	struct qed_rdma_start_in_params *in_params;
791 	struct qed_rdma_cnq_params *cur_pbl;
792 	struct qed_rdma_events events;
793 	dma_addr_t p_phys_table;
794 	u32 page_cnt;
795 	int rc = 0;
796 	int i;
797 
798 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
799 	if (!in_params) {
800 		rc = -ENOMEM;
801 		goto out;
802 	}
803 
804 	in_params->desired_cnq = dev->num_cnq;
805 	for (i = 0; i < dev->num_cnq; i++) {
806 		cur_pbl = &in_params->cnq_pbl_list[i];
807 
808 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
809 		cur_pbl->num_pbl_pages = page_cnt;
810 
811 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
812 		cur_pbl->pbl_ptr = (u64)p_phys_table;
813 	}
814 
815 	events.affiliated_event = qedr_affiliated_event;
816 	events.unaffiliated_event = qedr_unaffiliated_event;
817 	events.context = dev;
818 
819 	in_params->events = &events;
820 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
821 	in_params->max_mtu = dev->ndev->mtu;
822 	dev->iwarp_max_mtu = dev->ndev->mtu;
823 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
824 
825 	rc = dev->ops->rdma_init(dev->cdev, in_params);
826 	if (rc)
827 		goto out;
828 
829 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
830 	if (rc)
831 		goto out;
832 
833 	dev->db_addr = (void __iomem *)(uintptr_t)out_params.dpi_addr;
834 	dev->db_phys_addr = out_params.dpi_phys_addr;
835 	dev->db_size = out_params.dpi_size;
836 	dev->dpi = out_params.dpi;
837 
838 	rc = qedr_set_device_attr(dev);
839 out:
840 	kfree(in_params);
841 	if (rc)
842 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
843 
844 	return rc;
845 }
846 
847 static void qedr_stop_hw(struct qedr_dev *dev)
848 {
849 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
850 	dev->ops->rdma_stop(dev->rdma_ctx);
851 }
852 
853 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
854 				 struct net_device *ndev)
855 {
856 	struct qed_dev_rdma_info dev_info;
857 	struct qedr_dev *dev;
858 	int rc = 0, i;
859 
860 	dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
861 	if (!dev) {
862 		pr_err("Unable to allocate ib device\n");
863 		return NULL;
864 	}
865 
866 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
867 
868 	dev->pdev = pdev;
869 	dev->ndev = ndev;
870 	dev->cdev = cdev;
871 
872 	qed_ops = qed_get_rdma_ops();
873 	if (!qed_ops) {
874 		DP_ERR(dev, "Failed to get qed roce operations\n");
875 		goto init_err;
876 	}
877 
878 	dev->ops = qed_ops;
879 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
880 	if (rc)
881 		goto init_err;
882 
883 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
884 	dev->rdma_type = dev_info.rdma_type;
885 	dev->num_hwfns = dev_info.common.num_hwfns;
886 	dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
887 
888 	dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
889 	if (!dev->num_cnq) {
890 		DP_ERR(dev, "Failed. At least one CNQ is required.\n");
891 		rc = -ENOMEM;
892 		goto init_err;
893 	}
894 
895 	dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
896 
897 	qedr_pci_set_atomic(dev, pdev);
898 
899 	rc = qedr_alloc_resources(dev);
900 	if (rc)
901 		goto init_err;
902 
903 	rc = qedr_init_hw(dev);
904 	if (rc)
905 		goto alloc_err;
906 
907 	rc = qedr_setup_irqs(dev);
908 	if (rc)
909 		goto irq_err;
910 
911 	rc = qedr_register_device(dev);
912 	if (rc) {
913 		DP_ERR(dev, "Unable to allocate register device\n");
914 		goto reg_err;
915 	}
916 
917 	for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
918 		if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
919 			goto sysfs_err;
920 
921 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
922 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
923 
924 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
925 	return dev;
926 
927 sysfs_err:
928 	ib_unregister_device(&dev->ibdev);
929 reg_err:
930 	qedr_sync_free_irqs(dev);
931 irq_err:
932 	qedr_stop_hw(dev);
933 alloc_err:
934 	qedr_free_resources(dev);
935 init_err:
936 	ib_dealloc_device(&dev->ibdev);
937 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
938 
939 	return NULL;
940 }
941 
942 static void qedr_remove(struct qedr_dev *dev)
943 {
944 	/* First unregister with stack to stop all the active traffic
945 	 * of the registered clients.
946 	 */
947 	qedr_remove_sysfiles(dev);
948 	ib_unregister_device(&dev->ibdev);
949 
950 	qedr_stop_hw(dev);
951 	qedr_sync_free_irqs(dev);
952 	qedr_free_resources(dev);
953 	ib_dealloc_device(&dev->ibdev);
954 }
955 
956 static void qedr_close(struct qedr_dev *dev)
957 {
958 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
959 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
960 }
961 
962 static void qedr_shutdown(struct qedr_dev *dev)
963 {
964 	qedr_close(dev);
965 	qedr_remove(dev);
966 }
967 
968 static void qedr_open(struct qedr_dev *dev)
969 {
970 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
971 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
972 }
973 
974 static void qedr_mac_address_change(struct qedr_dev *dev)
975 {
976 	union ib_gid *sgid = &dev->sgid_tbl[0];
977 	u8 guid[8], mac_addr[6];
978 	int rc;
979 
980 	/* Update SGID */
981 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
982 	guid[0] = mac_addr[0] ^ 2;
983 	guid[1] = mac_addr[1];
984 	guid[2] = mac_addr[2];
985 	guid[3] = 0xff;
986 	guid[4] = 0xfe;
987 	guid[5] = mac_addr[3];
988 	guid[6] = mac_addr[4];
989 	guid[7] = mac_addr[5];
990 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
991 	memcpy(&sgid->raw[8], guid, sizeof(guid));
992 
993 	/* Update LL2 */
994 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
995 					  dev->gsi_ll2_mac_address,
996 					  dev->ndev->dev_addr);
997 
998 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
999 
1000 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
1001 
1002 	if (rc)
1003 		DP_ERR(dev, "Error updating mac filter\n");
1004 }
1005 
1006 /* event handling via NIC driver ensures that all the NIC specific
1007  * initialization done before RoCE driver notifies
1008  * event to stack.
1009  */
1010 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1011 {
1012 	switch (event) {
1013 	case QEDE_UP:
1014 		qedr_open(dev);
1015 		break;
1016 	case QEDE_DOWN:
1017 		qedr_close(dev);
1018 		break;
1019 	case QEDE_CLOSE:
1020 		qedr_shutdown(dev);
1021 		break;
1022 	case QEDE_CHANGE_ADDR:
1023 		qedr_mac_address_change(dev);
1024 		break;
1025 	default:
1026 		pr_err("Event not supported\n");
1027 	}
1028 }
1029 
1030 static struct qedr_driver qedr_drv = {
1031 	.name = "qedr_driver",
1032 	.add = qedr_add,
1033 	.remove = qedr_remove,
1034 	.notify = qedr_notify,
1035 };
1036 
1037 static int __init qedr_init_module(void)
1038 {
1039 	return qede_rdma_register_driver(&qedr_drv);
1040 }
1041 
1042 static void __exit qedr_exit_module(void)
1043 {
1044 	qede_rdma_unregister_driver(&qedr_drv);
1045 }
1046 
1047 module_init(qedr_init_module);
1048 module_exit(qedr_exit_module);
1049