xref: /openbmc/linux/drivers/infiniband/hw/qedr/main.c (revision b830f94f)
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 
43 #include <linux/qed/qed_chain.h>
44 #include <linux/qed/qed_if.h>
45 #include "qedr.h"
46 #include "verbs.h"
47 #include <rdma/qedr-abi.h>
48 #include "qedr_iw_cm.h"
49 
50 MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
51 MODULE_AUTHOR("QLogic Corporation");
52 MODULE_LICENSE("Dual BSD/GPL");
53 
54 #define QEDR_WQ_MULTIPLIER_DFT	(3)
55 
56 static void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
57 				   enum ib_event_type type)
58 {
59 	struct ib_event ibev;
60 
61 	ibev.device = &dev->ibdev;
62 	ibev.element.port_num = port_num;
63 	ibev.event = type;
64 
65 	ib_dispatch_event(&ibev);
66 }
67 
68 static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
69 					    u8 port_num)
70 {
71 	return IB_LINK_LAYER_ETHERNET;
72 }
73 
74 static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str)
75 {
76 	struct qedr_dev *qedr = get_qedr_dev(ibdev);
77 	u32 fw_ver = (u32)qedr->attr.fw_ver;
78 
79 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d. %d. %d. %d",
80 		 (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF,
81 		 (fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
82 }
83 
84 static int qedr_roce_port_immutable(struct ib_device *ibdev, u8 port_num,
85 				    struct ib_port_immutable *immutable)
86 {
87 	struct ib_port_attr attr;
88 	int err;
89 
90 	err = qedr_query_port(ibdev, port_num, &attr);
91 	if (err)
92 		return err;
93 
94 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
95 	immutable->gid_tbl_len = attr.gid_tbl_len;
96 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
97 	    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
98 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
99 
100 	return 0;
101 }
102 
103 static int qedr_iw_port_immutable(struct ib_device *ibdev, u8 port_num,
104 				  struct ib_port_immutable *immutable)
105 {
106 	struct ib_port_attr attr;
107 	int err;
108 
109 	err = qedr_query_port(ibdev, port_num, &attr);
110 	if (err)
111 		return err;
112 
113 	immutable->pkey_tbl_len = 1;
114 	immutable->gid_tbl_len = 1;
115 	immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
116 	immutable->max_mad_size = 0;
117 
118 	return 0;
119 }
120 
121 /* QEDR sysfs interface */
122 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
123 			   char *buf)
124 {
125 	struct qedr_dev *dev =
126 		rdma_device_to_drv_device(device, struct qedr_dev, ibdev);
127 
128 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
129 }
130 static DEVICE_ATTR_RO(hw_rev);
131 
132 static ssize_t hca_type_show(struct device *device,
133 			     struct device_attribute *attr, char *buf)
134 {
135 	return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
136 }
137 static DEVICE_ATTR_RO(hca_type);
138 
139 static struct attribute *qedr_attributes[] = {
140 	&dev_attr_hw_rev.attr,
141 	&dev_attr_hca_type.attr,
142 	NULL
143 };
144 
145 static const struct attribute_group qedr_attr_group = {
146 	.attrs = qedr_attributes,
147 };
148 
149 static const struct ib_device_ops qedr_iw_dev_ops = {
150 	.get_port_immutable = qedr_iw_port_immutable,
151 	.iw_accept = qedr_iw_accept,
152 	.iw_add_ref = qedr_iw_qp_add_ref,
153 	.iw_connect = qedr_iw_connect,
154 	.iw_create_listen = qedr_iw_create_listen,
155 	.iw_destroy_listen = qedr_iw_destroy_listen,
156 	.iw_get_qp = qedr_iw_get_qp,
157 	.iw_reject = qedr_iw_reject,
158 	.iw_rem_ref = qedr_iw_qp_rem_ref,
159 	.query_gid = qedr_iw_query_gid,
160 };
161 
162 static int qedr_iw_register_device(struct qedr_dev *dev)
163 {
164 	dev->ibdev.node_type = RDMA_NODE_RNIC;
165 
166 	ib_set_device_ops(&dev->ibdev, &qedr_iw_dev_ops);
167 
168 	memcpy(dev->ibdev.iw_ifname,
169 	       dev->ndev->name, sizeof(dev->ibdev.iw_ifname));
170 
171 	return 0;
172 }
173 
174 static const struct ib_device_ops qedr_roce_dev_ops = {
175 	.get_port_immutable = qedr_roce_port_immutable,
176 };
177 
178 static void qedr_roce_register_device(struct qedr_dev *dev)
179 {
180 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
181 
182 	ib_set_device_ops(&dev->ibdev, &qedr_roce_dev_ops);
183 }
184 
185 static const struct ib_device_ops qedr_dev_ops = {
186 	.owner = THIS_MODULE,
187 	.driver_id = RDMA_DRIVER_QEDR,
188 	.uverbs_abi_ver = QEDR_ABI_VERSION,
189 
190 	.alloc_mr = qedr_alloc_mr,
191 	.alloc_pd = qedr_alloc_pd,
192 	.alloc_ucontext = qedr_alloc_ucontext,
193 	.create_ah = qedr_create_ah,
194 	.create_cq = qedr_create_cq,
195 	.create_qp = qedr_create_qp,
196 	.create_srq = qedr_create_srq,
197 	.dealloc_pd = qedr_dealloc_pd,
198 	.dealloc_ucontext = qedr_dealloc_ucontext,
199 	.dereg_mr = qedr_dereg_mr,
200 	.destroy_ah = qedr_destroy_ah,
201 	.destroy_cq = qedr_destroy_cq,
202 	.destroy_qp = qedr_destroy_qp,
203 	.destroy_srq = qedr_destroy_srq,
204 	.get_dev_fw_str = qedr_get_dev_fw_str,
205 	.get_dma_mr = qedr_get_dma_mr,
206 	.get_link_layer = qedr_link_layer,
207 	.map_mr_sg = qedr_map_mr_sg,
208 	.mmap = qedr_mmap,
209 	.modify_port = qedr_modify_port,
210 	.modify_qp = qedr_modify_qp,
211 	.modify_srq = qedr_modify_srq,
212 	.poll_cq = qedr_poll_cq,
213 	.post_recv = qedr_post_recv,
214 	.post_send = qedr_post_send,
215 	.post_srq_recv = qedr_post_srq_recv,
216 	.process_mad = qedr_process_mad,
217 	.query_device = qedr_query_device,
218 	.query_pkey = qedr_query_pkey,
219 	.query_port = qedr_query_port,
220 	.query_qp = qedr_query_qp,
221 	.query_srq = qedr_query_srq,
222 	.reg_user_mr = qedr_reg_user_mr,
223 	.req_notify_cq = qedr_arm_cq,
224 	.resize_cq = qedr_resize_cq,
225 
226 	INIT_RDMA_OBJ_SIZE(ib_ah, qedr_ah, ibah),
227 	INIT_RDMA_OBJ_SIZE(ib_cq, qedr_cq, ibcq),
228 	INIT_RDMA_OBJ_SIZE(ib_pd, qedr_pd, ibpd),
229 	INIT_RDMA_OBJ_SIZE(ib_srq, qedr_srq, ibsrq),
230 	INIT_RDMA_OBJ_SIZE(ib_ucontext, qedr_ucontext, ibucontext),
231 };
232 
233 static int qedr_register_device(struct qedr_dev *dev)
234 {
235 	int rc;
236 
237 	dev->ibdev.node_guid = dev->attr.node_guid;
238 	memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
239 
240 	dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
241 				     QEDR_UVERBS(QUERY_DEVICE) |
242 				     QEDR_UVERBS(QUERY_PORT) |
243 				     QEDR_UVERBS(ALLOC_PD) |
244 				     QEDR_UVERBS(DEALLOC_PD) |
245 				     QEDR_UVERBS(CREATE_COMP_CHANNEL) |
246 				     QEDR_UVERBS(CREATE_CQ) |
247 				     QEDR_UVERBS(RESIZE_CQ) |
248 				     QEDR_UVERBS(DESTROY_CQ) |
249 				     QEDR_UVERBS(REQ_NOTIFY_CQ) |
250 				     QEDR_UVERBS(CREATE_QP) |
251 				     QEDR_UVERBS(MODIFY_QP) |
252 				     QEDR_UVERBS(QUERY_QP) |
253 				     QEDR_UVERBS(DESTROY_QP) |
254 				     QEDR_UVERBS(CREATE_SRQ) |
255 				     QEDR_UVERBS(DESTROY_SRQ) |
256 				     QEDR_UVERBS(QUERY_SRQ) |
257 				     QEDR_UVERBS(MODIFY_SRQ) |
258 				     QEDR_UVERBS(POST_SRQ_RECV) |
259 				     QEDR_UVERBS(REG_MR) |
260 				     QEDR_UVERBS(DEREG_MR) |
261 				     QEDR_UVERBS(POLL_CQ) |
262 				     QEDR_UVERBS(POST_SEND) |
263 				     QEDR_UVERBS(POST_RECV);
264 
265 	if (IS_IWARP(dev)) {
266 		rc = qedr_iw_register_device(dev);
267 		if (rc)
268 			return rc;
269 	} else {
270 		qedr_roce_register_device(dev);
271 	}
272 
273 	dev->ibdev.phys_port_cnt = 1;
274 	dev->ibdev.num_comp_vectors = dev->num_cnq;
275 	dev->ibdev.dev.parent = &dev->pdev->dev;
276 
277 	rdma_set_device_sysfs_group(&dev->ibdev, &qedr_attr_group);
278 	ib_set_device_ops(&dev->ibdev, &qedr_dev_ops);
279 
280 	rc = ib_device_set_netdev(&dev->ibdev, dev->ndev, 1);
281 	if (rc)
282 		return rc;
283 
284 	return ib_register_device(&dev->ibdev, "qedr%d");
285 }
286 
287 /* This function allocates fast-path status block memory */
288 static int qedr_alloc_mem_sb(struct qedr_dev *dev,
289 			     struct qed_sb_info *sb_info, u16 sb_id)
290 {
291 	struct status_block_e4 *sb_virt;
292 	dma_addr_t sb_phys;
293 	int rc;
294 
295 	sb_virt = dma_alloc_coherent(&dev->pdev->dev,
296 				     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
297 	if (!sb_virt)
298 		return -ENOMEM;
299 
300 	rc = dev->ops->common->sb_init(dev->cdev, sb_info,
301 				       sb_virt, sb_phys, sb_id,
302 				       QED_SB_TYPE_CNQ);
303 	if (rc) {
304 		pr_err("Status block initialization failed\n");
305 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
306 				  sb_virt, sb_phys);
307 		return rc;
308 	}
309 
310 	return 0;
311 }
312 
313 static void qedr_free_mem_sb(struct qedr_dev *dev,
314 			     struct qed_sb_info *sb_info, int sb_id)
315 {
316 	if (sb_info->sb_virt) {
317 		dev->ops->common->sb_release(dev->cdev, sb_info, sb_id,
318 					     QED_SB_TYPE_CNQ);
319 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
320 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
321 	}
322 }
323 
324 static void qedr_free_resources(struct qedr_dev *dev)
325 {
326 	int i;
327 
328 	if (IS_IWARP(dev))
329 		destroy_workqueue(dev->iwarp_wq);
330 
331 	for (i = 0; i < dev->num_cnq; i++) {
332 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
333 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
334 	}
335 
336 	kfree(dev->cnq_array);
337 	kfree(dev->sb_array);
338 	kfree(dev->sgid_tbl);
339 }
340 
341 static int qedr_alloc_resources(struct qedr_dev *dev)
342 {
343 	struct qedr_cnq *cnq;
344 	__le16 *cons_pi;
345 	u16 n_entries;
346 	int i, rc;
347 
348 	dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
349 				GFP_KERNEL);
350 	if (!dev->sgid_tbl)
351 		return -ENOMEM;
352 
353 	spin_lock_init(&dev->sgid_lock);
354 
355 	if (IS_IWARP(dev)) {
356 		xa_init_flags(&dev->qps, XA_FLAGS_LOCK_IRQ);
357 		dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
358 	}
359 
360 	/* Allocate Status blocks for CNQ */
361 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
362 				GFP_KERNEL);
363 	if (!dev->sb_array) {
364 		rc = -ENOMEM;
365 		goto err1;
366 	}
367 
368 	dev->cnq_array = kcalloc(dev->num_cnq,
369 				 sizeof(*dev->cnq_array), GFP_KERNEL);
370 	if (!dev->cnq_array) {
371 		rc = -ENOMEM;
372 		goto err2;
373 	}
374 
375 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
376 
377 	/* Allocate CNQ PBLs */
378 	n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
379 	for (i = 0; i < dev->num_cnq; i++) {
380 		cnq = &dev->cnq_array[i];
381 
382 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
383 				       dev->sb_start + i);
384 		if (rc)
385 			goto err3;
386 
387 		rc = dev->ops->common->chain_alloc(dev->cdev,
388 						   QED_CHAIN_USE_TO_CONSUME,
389 						   QED_CHAIN_MODE_PBL,
390 						   QED_CHAIN_CNT_TYPE_U16,
391 						   n_entries,
392 						   sizeof(struct regpair *),
393 						   &cnq->pbl, NULL);
394 		if (rc)
395 			goto err4;
396 
397 		cnq->dev = dev;
398 		cnq->sb = &dev->sb_array[i];
399 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
400 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
401 		cnq->index = i;
402 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
403 
404 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
405 			 i, qed_chain_get_cons_idx(&cnq->pbl));
406 	}
407 
408 	return 0;
409 err4:
410 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
411 err3:
412 	for (--i; i >= 0; i--) {
413 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
414 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
415 	}
416 	kfree(dev->cnq_array);
417 err2:
418 	kfree(dev->sb_array);
419 err1:
420 	kfree(dev->sgid_tbl);
421 	return rc;
422 }
423 
424 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
425 {
426 	int rc = pci_enable_atomic_ops_to_root(pdev,
427 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
428 
429 	if (rc) {
430 		dev->atomic_cap = IB_ATOMIC_NONE;
431 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
432 	} else {
433 		dev->atomic_cap = IB_ATOMIC_GLOB;
434 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
435 	}
436 }
437 
438 static const struct qed_rdma_ops *qed_ops;
439 
440 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
441 
442 static irqreturn_t qedr_irq_handler(int irq, void *handle)
443 {
444 	u16 hw_comp_cons, sw_comp_cons;
445 	struct qedr_cnq *cnq = handle;
446 	struct regpair *cq_handle;
447 	struct qedr_cq *cq;
448 
449 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
450 
451 	qed_sb_update_sb_idx(cnq->sb);
452 
453 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
454 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
455 
456 	/* Align protocol-index and chain reads */
457 	rmb();
458 
459 	while (sw_comp_cons != hw_comp_cons) {
460 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
461 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
462 				cq_handle->lo);
463 
464 		if (cq == NULL) {
465 			DP_ERR(cnq->dev,
466 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
467 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
468 			       hw_comp_cons);
469 
470 			break;
471 		}
472 
473 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
474 			DP_ERR(cnq->dev,
475 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
476 			       cq_handle->hi, cq_handle->lo, cq);
477 			break;
478 		}
479 
480 		cq->arm_flags = 0;
481 
482 		if (!cq->destroyed && cq->ibcq.comp_handler)
483 			(*cq->ibcq.comp_handler)
484 				(&cq->ibcq, cq->ibcq.cq_context);
485 
486 		/* The CQ's CNQ notification counter is checked before
487 		 * destroying the CQ in a busy-wait loop that waits for all of
488 		 * the CQ's CNQ interrupts to be processed. It is increased
489 		 * here, only after the completion handler, to ensure that the
490 		 * the handler is not running when the CQ is destroyed.
491 		 */
492 		cq->cnq_notif++;
493 
494 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
495 
496 		cnq->n_comp++;
497 	}
498 
499 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
500 				      sw_comp_cons);
501 
502 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
503 
504 	return IRQ_HANDLED;
505 }
506 
507 static void qedr_sync_free_irqs(struct qedr_dev *dev)
508 {
509 	u32 vector;
510 	u16 idx;
511 	int i;
512 
513 	for (i = 0; i < dev->int_info.used_cnt; i++) {
514 		if (dev->int_info.msix_cnt) {
515 			idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
516 			vector = dev->int_info.msix[idx].vector;
517 			synchronize_irq(vector);
518 			free_irq(vector, &dev->cnq_array[i]);
519 		}
520 	}
521 
522 	dev->int_info.used_cnt = 0;
523 }
524 
525 static int qedr_req_msix_irqs(struct qedr_dev *dev)
526 {
527 	int i, rc = 0;
528 	u16 idx;
529 
530 	if (dev->num_cnq > dev->int_info.msix_cnt) {
531 		DP_ERR(dev,
532 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
533 		       dev->num_cnq, dev->int_info.msix_cnt);
534 		return -EINVAL;
535 	}
536 
537 	for (i = 0; i < dev->num_cnq; i++) {
538 		idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
539 		rc = request_irq(dev->int_info.msix[idx].vector,
540 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
541 				 &dev->cnq_array[i]);
542 		if (rc) {
543 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
544 			qedr_sync_free_irqs(dev);
545 		} else {
546 			DP_DEBUG(dev, QEDR_MSG_INIT,
547 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
548 				 dev->cnq_array[i].name, i,
549 				 &dev->cnq_array[i]);
550 			dev->int_info.used_cnt++;
551 		}
552 	}
553 
554 	return rc;
555 }
556 
557 static int qedr_setup_irqs(struct qedr_dev *dev)
558 {
559 	int rc;
560 
561 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
562 
563 	/* Learn Interrupt configuration */
564 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
565 	if (rc < 0)
566 		return rc;
567 
568 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
569 	if (rc) {
570 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
571 		return rc;
572 	}
573 
574 	if (dev->int_info.msix_cnt) {
575 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
576 			 dev->int_info.msix_cnt);
577 		rc = qedr_req_msix_irqs(dev);
578 		if (rc)
579 			return rc;
580 	}
581 
582 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
583 
584 	return 0;
585 }
586 
587 static int qedr_set_device_attr(struct qedr_dev *dev)
588 {
589 	struct qed_rdma_device *qed_attr;
590 	struct qedr_device_attr *attr;
591 	u32 page_size;
592 
593 	/* Part 1 - query core capabilities */
594 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
595 
596 	/* Part 2 - check capabilities */
597 	page_size = ~dev->attr.page_size_caps + 1;
598 	if (page_size > PAGE_SIZE) {
599 		DP_ERR(dev,
600 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
601 		       PAGE_SIZE, page_size);
602 		return -ENODEV;
603 	}
604 
605 	/* Part 3 - copy and update capabilities */
606 	attr = &dev->attr;
607 	attr->vendor_id = qed_attr->vendor_id;
608 	attr->vendor_part_id = qed_attr->vendor_part_id;
609 	attr->hw_ver = qed_attr->hw_ver;
610 	attr->fw_ver = qed_attr->fw_ver;
611 	attr->node_guid = qed_attr->node_guid;
612 	attr->sys_image_guid = qed_attr->sys_image_guid;
613 	attr->max_cnq = qed_attr->max_cnq;
614 	attr->max_sge = qed_attr->max_sge;
615 	attr->max_inline = qed_attr->max_inline;
616 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
617 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
618 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
619 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
620 	attr->max_dev_resp_rd_atomic_resc =
621 	    qed_attr->max_dev_resp_rd_atomic_resc;
622 	attr->max_cq = qed_attr->max_cq;
623 	attr->max_qp = qed_attr->max_qp;
624 	attr->max_mr = qed_attr->max_mr;
625 	attr->max_mr_size = qed_attr->max_mr_size;
626 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
627 	attr->max_mw = qed_attr->max_mw;
628 	attr->max_fmr = qed_attr->max_fmr;
629 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
630 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
631 	attr->max_pd = qed_attr->max_pd;
632 	attr->max_ah = qed_attr->max_ah;
633 	attr->max_pkey = qed_attr->max_pkey;
634 	attr->max_srq = qed_attr->max_srq;
635 	attr->max_srq_wr = qed_attr->max_srq_wr;
636 	attr->dev_caps = qed_attr->dev_caps;
637 	attr->page_size_caps = qed_attr->page_size_caps;
638 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
639 	attr->reserved_lkey = qed_attr->reserved_lkey;
640 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
641 	attr->max_stats_queues = qed_attr->max_stats_queues;
642 
643 	return 0;
644 }
645 
646 static void qedr_unaffiliated_event(void *context, u8 event_code)
647 {
648 	pr_err("unaffiliated event not implemented yet\n");
649 }
650 
651 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
652 {
653 #define EVENT_TYPE_NOT_DEFINED	0
654 #define EVENT_TYPE_CQ		1
655 #define EVENT_TYPE_QP		2
656 #define EVENT_TYPE_SRQ		3
657 	struct qedr_dev *dev = (struct qedr_dev *)context;
658 	struct regpair *async_handle = (struct regpair *)fw_handle;
659 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
660 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
661 	struct ib_event event;
662 	struct ib_srq *ibsrq;
663 	struct qedr_srq *srq;
664 	unsigned long flags;
665 	struct ib_cq *ibcq;
666 	struct ib_qp *ibqp;
667 	struct qedr_cq *cq;
668 	struct qedr_qp *qp;
669 	u16 srq_id;
670 
671 	if (IS_ROCE(dev)) {
672 		switch (e_code) {
673 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
674 			event.event = IB_EVENT_CQ_ERR;
675 			event_type = EVENT_TYPE_CQ;
676 			break;
677 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
678 			event.event = IB_EVENT_SQ_DRAINED;
679 			event_type = EVENT_TYPE_QP;
680 			break;
681 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
682 			event.event = IB_EVENT_QP_FATAL;
683 			event_type = EVENT_TYPE_QP;
684 			break;
685 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
686 			event.event = IB_EVENT_QP_REQ_ERR;
687 			event_type = EVENT_TYPE_QP;
688 			break;
689 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
690 			event.event = IB_EVENT_QP_ACCESS_ERR;
691 			event_type = EVENT_TYPE_QP;
692 			break;
693 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
694 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
695 			event_type = EVENT_TYPE_SRQ;
696 			break;
697 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
698 			event.event = IB_EVENT_SRQ_ERR;
699 			event_type = EVENT_TYPE_SRQ;
700 			break;
701 		default:
702 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
703 			       e_code, roce_handle64);
704 		}
705 	} else {
706 		switch (e_code) {
707 		case QED_IWARP_EVENT_SRQ_LIMIT:
708 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
709 			event_type = EVENT_TYPE_SRQ;
710 			break;
711 		case QED_IWARP_EVENT_SRQ_EMPTY:
712 			event.event = IB_EVENT_SRQ_ERR;
713 			event_type = EVENT_TYPE_SRQ;
714 			break;
715 		default:
716 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
717 		       roce_handle64);
718 		}
719 	}
720 	switch (event_type) {
721 	case EVENT_TYPE_CQ:
722 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
723 		if (cq) {
724 			ibcq = &cq->ibcq;
725 			if (ibcq->event_handler) {
726 				event.device = ibcq->device;
727 				event.element.cq = ibcq;
728 				ibcq->event_handler(&event, ibcq->cq_context);
729 			}
730 		} else {
731 			WARN(1,
732 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
733 			     roce_handle64);
734 		}
735 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
736 		break;
737 	case EVENT_TYPE_QP:
738 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
739 		if (qp) {
740 			ibqp = &qp->ibqp;
741 			if (ibqp->event_handler) {
742 				event.device = ibqp->device;
743 				event.element.qp = ibqp;
744 				ibqp->event_handler(&event, ibqp->qp_context);
745 			}
746 		} else {
747 			WARN(1,
748 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
749 			     roce_handle64);
750 		}
751 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
752 		break;
753 	case EVENT_TYPE_SRQ:
754 		srq_id = (u16)roce_handle64;
755 		xa_lock_irqsave(&dev->srqs, flags);
756 		srq = xa_load(&dev->srqs, srq_id);
757 		if (srq) {
758 			ibsrq = &srq->ibsrq;
759 			if (ibsrq->event_handler) {
760 				event.device = ibsrq->device;
761 				event.element.srq = ibsrq;
762 				ibsrq->event_handler(&event,
763 						     ibsrq->srq_context);
764 			}
765 		} else {
766 			DP_NOTICE(dev,
767 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
768 				  roce_handle64);
769 		}
770 		xa_unlock_irqrestore(&dev->srqs, flags);
771 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
772 	default:
773 		break;
774 	}
775 }
776 
777 static int qedr_init_hw(struct qedr_dev *dev)
778 {
779 	struct qed_rdma_add_user_out_params out_params;
780 	struct qed_rdma_start_in_params *in_params;
781 	struct qed_rdma_cnq_params *cur_pbl;
782 	struct qed_rdma_events events;
783 	dma_addr_t p_phys_table;
784 	u32 page_cnt;
785 	int rc = 0;
786 	int i;
787 
788 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
789 	if (!in_params) {
790 		rc = -ENOMEM;
791 		goto out;
792 	}
793 
794 	in_params->desired_cnq = dev->num_cnq;
795 	for (i = 0; i < dev->num_cnq; i++) {
796 		cur_pbl = &in_params->cnq_pbl_list[i];
797 
798 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
799 		cur_pbl->num_pbl_pages = page_cnt;
800 
801 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
802 		cur_pbl->pbl_ptr = (u64)p_phys_table;
803 	}
804 
805 	events.affiliated_event = qedr_affiliated_event;
806 	events.unaffiliated_event = qedr_unaffiliated_event;
807 	events.context = dev;
808 
809 	in_params->events = &events;
810 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
811 	in_params->max_mtu = dev->ndev->mtu;
812 	dev->iwarp_max_mtu = dev->ndev->mtu;
813 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
814 
815 	rc = dev->ops->rdma_init(dev->cdev, in_params);
816 	if (rc)
817 		goto out;
818 
819 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
820 	if (rc)
821 		goto out;
822 
823 	dev->db_addr = (void __iomem *)(uintptr_t)out_params.dpi_addr;
824 	dev->db_phys_addr = out_params.dpi_phys_addr;
825 	dev->db_size = out_params.dpi_size;
826 	dev->dpi = out_params.dpi;
827 
828 	rc = qedr_set_device_attr(dev);
829 out:
830 	kfree(in_params);
831 	if (rc)
832 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
833 
834 	return rc;
835 }
836 
837 static void qedr_stop_hw(struct qedr_dev *dev)
838 {
839 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
840 	dev->ops->rdma_stop(dev->rdma_ctx);
841 }
842 
843 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
844 				 struct net_device *ndev)
845 {
846 	struct qed_dev_rdma_info dev_info;
847 	struct qedr_dev *dev;
848 	int rc = 0;
849 
850 	dev = ib_alloc_device(qedr_dev, ibdev);
851 	if (!dev) {
852 		pr_err("Unable to allocate ib device\n");
853 		return NULL;
854 	}
855 
856 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
857 
858 	dev->pdev = pdev;
859 	dev->ndev = ndev;
860 	dev->cdev = cdev;
861 
862 	qed_ops = qed_get_rdma_ops();
863 	if (!qed_ops) {
864 		DP_ERR(dev, "Failed to get qed roce operations\n");
865 		goto init_err;
866 	}
867 
868 	dev->ops = qed_ops;
869 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
870 	if (rc)
871 		goto init_err;
872 
873 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
874 	dev->rdma_type = dev_info.rdma_type;
875 	dev->num_hwfns = dev_info.common.num_hwfns;
876 
877 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev)) {
878 		rc = dev->ops->iwarp_set_engine_affin(cdev, false);
879 		if (rc) {
880 			DP_ERR(dev, "iWARP is disabled over a 100g device Enabling it may impact L2 performance. To enable it run devlink dev param set <dev> name iwarp_cmt value true cmode runtime\n");
881 			goto init_err;
882 		}
883 	}
884 	dev->affin_hwfn_idx = dev->ops->common->get_affin_hwfn_idx(cdev);
885 
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 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
918 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
919 
920 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
921 	return dev;
922 
923 reg_err:
924 	qedr_sync_free_irqs(dev);
925 irq_err:
926 	qedr_stop_hw(dev);
927 alloc_err:
928 	qedr_free_resources(dev);
929 init_err:
930 	ib_dealloc_device(&dev->ibdev);
931 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
932 
933 	return NULL;
934 }
935 
936 static void qedr_remove(struct qedr_dev *dev)
937 {
938 	/* First unregister with stack to stop all the active traffic
939 	 * of the registered clients.
940 	 */
941 	ib_unregister_device(&dev->ibdev);
942 
943 	qedr_stop_hw(dev);
944 	qedr_sync_free_irqs(dev);
945 	qedr_free_resources(dev);
946 
947 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev))
948 		dev->ops->iwarp_set_engine_affin(dev->cdev, true);
949 
950 	ib_dealloc_device(&dev->ibdev);
951 }
952 
953 static void qedr_close(struct qedr_dev *dev)
954 {
955 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
956 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
957 }
958 
959 static void qedr_shutdown(struct qedr_dev *dev)
960 {
961 	qedr_close(dev);
962 	qedr_remove(dev);
963 }
964 
965 static void qedr_open(struct qedr_dev *dev)
966 {
967 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
968 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
969 }
970 
971 static void qedr_mac_address_change(struct qedr_dev *dev)
972 {
973 	union ib_gid *sgid = &dev->sgid_tbl[0];
974 	u8 guid[8], mac_addr[6];
975 	int rc;
976 
977 	/* Update SGID */
978 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
979 	guid[0] = mac_addr[0] ^ 2;
980 	guid[1] = mac_addr[1];
981 	guid[2] = mac_addr[2];
982 	guid[3] = 0xff;
983 	guid[4] = 0xfe;
984 	guid[5] = mac_addr[3];
985 	guid[6] = mac_addr[4];
986 	guid[7] = mac_addr[5];
987 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
988 	memcpy(&sgid->raw[8], guid, sizeof(guid));
989 
990 	/* Update LL2 */
991 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
992 					  dev->gsi_ll2_mac_address,
993 					  dev->ndev->dev_addr);
994 
995 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
996 
997 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
998 
999 	if (rc)
1000 		DP_ERR(dev, "Error updating mac filter\n");
1001 }
1002 
1003 /* event handling via NIC driver ensures that all the NIC specific
1004  * initialization done before RoCE driver notifies
1005  * event to stack.
1006  */
1007 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1008 {
1009 	switch (event) {
1010 	case QEDE_UP:
1011 		qedr_open(dev);
1012 		break;
1013 	case QEDE_DOWN:
1014 		qedr_close(dev);
1015 		break;
1016 	case QEDE_CLOSE:
1017 		qedr_shutdown(dev);
1018 		break;
1019 	case QEDE_CHANGE_ADDR:
1020 		qedr_mac_address_change(dev);
1021 		break;
1022 	default:
1023 		pr_err("Event not supported\n");
1024 	}
1025 }
1026 
1027 static struct qedr_driver qedr_drv = {
1028 	.name = "qedr_driver",
1029 	.add = qedr_add,
1030 	.remove = qedr_remove,
1031 	.notify = qedr_notify,
1032 };
1033 
1034 static int __init qedr_init_module(void)
1035 {
1036 	return qede_rdma_register_driver(&qedr_drv);
1037 }
1038 
1039 static void __exit qedr_exit_module(void)
1040 {
1041 	qede_rdma_unregister_driver(&qedr_drv);
1042 }
1043 
1044 module_init(qedr_init_module);
1045 module_exit(qedr_exit_module);
1046