xref: /openbmc/linux/drivers/infiniband/hw/qedr/main.c (revision 5abbeebd)
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, u32 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 					    u32 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, u32 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, u32 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->gid_tbl_len = 1;
114 	immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
115 	immutable->max_mad_size = 0;
116 
117 	return 0;
118 }
119 
120 /* QEDR sysfs interface */
121 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
122 			   char *buf)
123 {
124 	struct qedr_dev *dev =
125 		rdma_device_to_drv_device(device, struct qedr_dev, ibdev);
126 
127 	return sysfs_emit(buf, "0x%x\n", dev->attr.hw_ver);
128 }
129 static DEVICE_ATTR_RO(hw_rev);
130 
131 static ssize_t hca_type_show(struct device *device,
132 			     struct device_attribute *attr, char *buf)
133 {
134 	struct qedr_dev *dev =
135 		rdma_device_to_drv_device(device, struct qedr_dev, ibdev);
136 
137 	return sysfs_emit(buf, "FastLinQ QL%x %s\n", dev->pdev->device,
138 			  rdma_protocol_iwarp(&dev->ibdev, 1) ? "iWARP" :
139 								"RoCE");
140 }
141 static DEVICE_ATTR_RO(hca_type);
142 
143 static struct attribute *qedr_attributes[] = {
144 	&dev_attr_hw_rev.attr,
145 	&dev_attr_hca_type.attr,
146 	NULL
147 };
148 
149 static const struct attribute_group qedr_attr_group = {
150 	.attrs = qedr_attributes,
151 };
152 
153 static const struct ib_device_ops qedr_iw_dev_ops = {
154 	.get_port_immutable = qedr_iw_port_immutable,
155 	.iw_accept = qedr_iw_accept,
156 	.iw_add_ref = qedr_iw_qp_add_ref,
157 	.iw_connect = qedr_iw_connect,
158 	.iw_create_listen = qedr_iw_create_listen,
159 	.iw_destroy_listen = qedr_iw_destroy_listen,
160 	.iw_get_qp = qedr_iw_get_qp,
161 	.iw_reject = qedr_iw_reject,
162 	.iw_rem_ref = qedr_iw_qp_rem_ref,
163 	.query_gid = qedr_iw_query_gid,
164 };
165 
166 static int qedr_iw_register_device(struct qedr_dev *dev)
167 {
168 	dev->ibdev.node_type = RDMA_NODE_RNIC;
169 
170 	ib_set_device_ops(&dev->ibdev, &qedr_iw_dev_ops);
171 
172 	memcpy(dev->ibdev.iw_ifname,
173 	       dev->ndev->name, sizeof(dev->ibdev.iw_ifname));
174 
175 	return 0;
176 }
177 
178 static const struct ib_device_ops qedr_roce_dev_ops = {
179 	.alloc_xrcd = qedr_alloc_xrcd,
180 	.dealloc_xrcd = qedr_dealloc_xrcd,
181 	.get_port_immutable = qedr_roce_port_immutable,
182 	.query_pkey = qedr_query_pkey,
183 };
184 
185 static void qedr_roce_register_device(struct qedr_dev *dev)
186 {
187 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
188 
189 	ib_set_device_ops(&dev->ibdev, &qedr_roce_dev_ops);
190 }
191 
192 static const struct ib_device_ops qedr_dev_ops = {
193 	.owner = THIS_MODULE,
194 	.driver_id = RDMA_DRIVER_QEDR,
195 	.uverbs_abi_ver = QEDR_ABI_VERSION,
196 
197 	.alloc_mr = qedr_alloc_mr,
198 	.alloc_pd = qedr_alloc_pd,
199 	.alloc_ucontext = qedr_alloc_ucontext,
200 	.create_ah = qedr_create_ah,
201 	.create_cq = qedr_create_cq,
202 	.create_qp = qedr_create_qp,
203 	.create_srq = qedr_create_srq,
204 	.dealloc_pd = qedr_dealloc_pd,
205 	.dealloc_ucontext = qedr_dealloc_ucontext,
206 	.dereg_mr = qedr_dereg_mr,
207 	.destroy_ah = qedr_destroy_ah,
208 	.destroy_cq = qedr_destroy_cq,
209 	.destroy_qp = qedr_destroy_qp,
210 	.destroy_srq = qedr_destroy_srq,
211 	.device_group = &qedr_attr_group,
212 	.get_dev_fw_str = qedr_get_dev_fw_str,
213 	.get_dma_mr = qedr_get_dma_mr,
214 	.get_link_layer = qedr_link_layer,
215 	.map_mr_sg = qedr_map_mr_sg,
216 	.mmap = qedr_mmap,
217 	.mmap_free = qedr_mmap_free,
218 	.modify_qp = qedr_modify_qp,
219 	.modify_srq = qedr_modify_srq,
220 	.poll_cq = qedr_poll_cq,
221 	.post_recv = qedr_post_recv,
222 	.post_send = qedr_post_send,
223 	.post_srq_recv = qedr_post_srq_recv,
224 	.process_mad = qedr_process_mad,
225 	.query_device = qedr_query_device,
226 	.query_port = qedr_query_port,
227 	.query_qp = qedr_query_qp,
228 	.query_srq = qedr_query_srq,
229 	.reg_user_mr = qedr_reg_user_mr,
230 	.req_notify_cq = qedr_arm_cq,
231 
232 	INIT_RDMA_OBJ_SIZE(ib_ah, qedr_ah, ibah),
233 	INIT_RDMA_OBJ_SIZE(ib_cq, qedr_cq, ibcq),
234 	INIT_RDMA_OBJ_SIZE(ib_pd, qedr_pd, ibpd),
235 	INIT_RDMA_OBJ_SIZE(ib_qp, qedr_qp, ibqp),
236 	INIT_RDMA_OBJ_SIZE(ib_srq, qedr_srq, ibsrq),
237 	INIT_RDMA_OBJ_SIZE(ib_xrcd, qedr_xrcd, ibxrcd),
238 	INIT_RDMA_OBJ_SIZE(ib_ucontext, qedr_ucontext, ibucontext),
239 };
240 
241 static int qedr_register_device(struct qedr_dev *dev)
242 {
243 	int rc;
244 
245 	dev->ibdev.node_guid = dev->attr.node_guid;
246 	memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
247 
248 	if (IS_IWARP(dev)) {
249 		rc = qedr_iw_register_device(dev);
250 		if (rc)
251 			return rc;
252 	} else {
253 		qedr_roce_register_device(dev);
254 	}
255 
256 	dev->ibdev.phys_port_cnt = 1;
257 	dev->ibdev.num_comp_vectors = dev->num_cnq;
258 	dev->ibdev.dev.parent = &dev->pdev->dev;
259 
260 	ib_set_device_ops(&dev->ibdev, &qedr_dev_ops);
261 
262 	rc = ib_device_set_netdev(&dev->ibdev, dev->ndev, 1);
263 	if (rc)
264 		return rc;
265 
266 	dma_set_max_seg_size(&dev->pdev->dev, UINT_MAX);
267 	return ib_register_device(&dev->ibdev, "qedr%d", &dev->pdev->dev);
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 *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 					     QED_SB_TYPE_CNQ);
302 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
303 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
304 	}
305 }
306 
307 static void qedr_free_resources(struct qedr_dev *dev)
308 {
309 	int i;
310 
311 	if (IS_IWARP(dev))
312 		destroy_workqueue(dev->iwarp_wq);
313 
314 	for (i = 0; i < dev->num_cnq; i++) {
315 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
316 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
317 	}
318 
319 	kfree(dev->cnq_array);
320 	kfree(dev->sb_array);
321 	kfree(dev->sgid_tbl);
322 }
323 
324 static int qedr_alloc_resources(struct qedr_dev *dev)
325 {
326 	struct qed_chain_init_params params = {
327 		.mode		= QED_CHAIN_MODE_PBL,
328 		.intended_use	= QED_CHAIN_USE_TO_CONSUME,
329 		.cnt_type	= QED_CHAIN_CNT_TYPE_U16,
330 		.elem_size	= sizeof(struct regpair *),
331 	};
332 	struct qedr_cnq *cnq;
333 	__le16 *cons_pi;
334 	int i, rc;
335 
336 	dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
337 				GFP_KERNEL);
338 	if (!dev->sgid_tbl)
339 		return -ENOMEM;
340 
341 	spin_lock_init(&dev->sgid_lock);
342 	xa_init_flags(&dev->srqs, XA_FLAGS_LOCK_IRQ);
343 
344 	if (IS_IWARP(dev)) {
345 		xa_init(&dev->qps);
346 		dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
347 		if (!dev->iwarp_wq) {
348 			rc = -ENOMEM;
349 			goto err1;
350 		}
351 	}
352 
353 	/* Allocate Status blocks for CNQ */
354 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
355 				GFP_KERNEL);
356 	if (!dev->sb_array) {
357 		rc = -ENOMEM;
358 		goto err_destroy_wq;
359 	}
360 
361 	dev->cnq_array = kcalloc(dev->num_cnq,
362 				 sizeof(*dev->cnq_array), GFP_KERNEL);
363 	if (!dev->cnq_array) {
364 		rc = -ENOMEM;
365 		goto err2;
366 	}
367 
368 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
369 
370 	/* Allocate CNQ PBLs */
371 	params.num_elems = min_t(u32, QED_RDMA_MAX_CNQ_SIZE,
372 				 QEDR_ROCE_MAX_CNQ_SIZE);
373 
374 	for (i = 0; i < dev->num_cnq; i++) {
375 		cnq = &dev->cnq_array[i];
376 
377 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
378 				       dev->sb_start + i);
379 		if (rc)
380 			goto err3;
381 
382 		rc = dev->ops->common->chain_alloc(dev->cdev, &cnq->pbl,
383 						   &params);
384 		if (rc)
385 			goto err4;
386 
387 		cnq->dev = dev;
388 		cnq->sb = &dev->sb_array[i];
389 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
390 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
391 		cnq->index = i;
392 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
393 
394 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
395 			 i, qed_chain_get_cons_idx(&cnq->pbl));
396 	}
397 
398 	return 0;
399 err4:
400 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
401 err3:
402 	for (--i; i >= 0; i--) {
403 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
404 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
405 	}
406 	kfree(dev->cnq_array);
407 err2:
408 	kfree(dev->sb_array);
409 err_destroy_wq:
410 	if (IS_IWARP(dev))
411 		destroy_workqueue(dev->iwarp_wq);
412 err1:
413 	kfree(dev->sgid_tbl);
414 	return rc;
415 }
416 
417 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
418 {
419 	int rc = pci_enable_atomic_ops_to_root(pdev,
420 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
421 
422 	if (rc) {
423 		dev->atomic_cap = IB_ATOMIC_NONE;
424 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
425 	} else {
426 		dev->atomic_cap = IB_ATOMIC_GLOB;
427 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
428 	}
429 }
430 
431 static const struct qed_rdma_ops *qed_ops;
432 
433 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
434 
435 static irqreturn_t qedr_irq_handler(int irq, void *handle)
436 {
437 	u16 hw_comp_cons, sw_comp_cons;
438 	struct qedr_cnq *cnq = handle;
439 	struct regpair *cq_handle;
440 	struct qedr_cq *cq;
441 
442 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
443 
444 	qed_sb_update_sb_idx(cnq->sb);
445 
446 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
447 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
448 
449 	/* Align protocol-index and chain reads */
450 	rmb();
451 
452 	while (sw_comp_cons != hw_comp_cons) {
453 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
454 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
455 				cq_handle->lo);
456 
457 		if (cq == NULL) {
458 			DP_ERR(cnq->dev,
459 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
460 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
461 			       hw_comp_cons);
462 
463 			break;
464 		}
465 
466 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
467 			DP_ERR(cnq->dev,
468 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
469 			       cq_handle->hi, cq_handle->lo, cq);
470 			break;
471 		}
472 
473 		cq->arm_flags = 0;
474 
475 		if (!cq->destroyed && cq->ibcq.comp_handler)
476 			(*cq->ibcq.comp_handler)
477 				(&cq->ibcq, cq->ibcq.cq_context);
478 
479 		/* The CQ's CNQ notification counter is checked before
480 		 * destroying the CQ in a busy-wait loop that waits for all of
481 		 * the CQ's CNQ interrupts to be processed. It is increased
482 		 * here, only after the completion handler, to ensure that
483 		 * the handler is not running when the CQ is destroyed.
484 		 */
485 		cq->cnq_notif++;
486 
487 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
488 
489 		cnq->n_comp++;
490 	}
491 
492 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
493 				      sw_comp_cons);
494 
495 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
496 
497 	return IRQ_HANDLED;
498 }
499 
500 static void qedr_sync_free_irqs(struct qedr_dev *dev)
501 {
502 	u32 vector;
503 	u16 idx;
504 	int i;
505 
506 	for (i = 0; i < dev->int_info.used_cnt; i++) {
507 		if (dev->int_info.msix_cnt) {
508 			idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
509 			vector = dev->int_info.msix[idx].vector;
510 			free_irq(vector, &dev->cnq_array[i]);
511 		}
512 	}
513 
514 	dev->int_info.used_cnt = 0;
515 }
516 
517 static int qedr_req_msix_irqs(struct qedr_dev *dev)
518 {
519 	int i, rc = 0;
520 	u16 idx;
521 
522 	if (dev->num_cnq > dev->int_info.msix_cnt) {
523 		DP_ERR(dev,
524 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
525 		       dev->num_cnq, dev->int_info.msix_cnt);
526 		return -EINVAL;
527 	}
528 
529 	for (i = 0; i < dev->num_cnq; i++) {
530 		idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
531 		rc = request_irq(dev->int_info.msix[idx].vector,
532 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
533 				 &dev->cnq_array[i]);
534 		if (rc) {
535 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
536 			qedr_sync_free_irqs(dev);
537 		} else {
538 			DP_DEBUG(dev, QEDR_MSG_INIT,
539 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
540 				 dev->cnq_array[i].name, i,
541 				 &dev->cnq_array[i]);
542 			dev->int_info.used_cnt++;
543 		}
544 	}
545 
546 	return rc;
547 }
548 
549 static int qedr_setup_irqs(struct qedr_dev *dev)
550 {
551 	int rc;
552 
553 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
554 
555 	/* Learn Interrupt configuration */
556 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
557 	if (rc < 0)
558 		return rc;
559 
560 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
561 	if (rc) {
562 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
563 		return rc;
564 	}
565 
566 	if (dev->int_info.msix_cnt) {
567 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
568 			 dev->int_info.msix_cnt);
569 		rc = qedr_req_msix_irqs(dev);
570 		if (rc)
571 			return rc;
572 	}
573 
574 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
575 
576 	return 0;
577 }
578 
579 static int qedr_set_device_attr(struct qedr_dev *dev)
580 {
581 	struct qed_rdma_device *qed_attr;
582 	struct qedr_device_attr *attr;
583 	u32 page_size;
584 
585 	/* Part 1 - query core capabilities */
586 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
587 
588 	/* Part 2 - check capabilities */
589 	page_size = ~qed_attr->page_size_caps + 1;
590 	if (page_size > PAGE_SIZE) {
591 		DP_ERR(dev,
592 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
593 		       PAGE_SIZE, page_size);
594 		return -ENODEV;
595 	}
596 
597 	/* Part 3 - copy and update capabilities */
598 	attr = &dev->attr;
599 	attr->vendor_id = qed_attr->vendor_id;
600 	attr->vendor_part_id = qed_attr->vendor_part_id;
601 	attr->hw_ver = qed_attr->hw_ver;
602 	attr->fw_ver = qed_attr->fw_ver;
603 	attr->node_guid = qed_attr->node_guid;
604 	attr->sys_image_guid = qed_attr->sys_image_guid;
605 	attr->max_cnq = qed_attr->max_cnq;
606 	attr->max_sge = qed_attr->max_sge;
607 	attr->max_inline = qed_attr->max_inline;
608 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
609 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
610 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
611 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
612 	attr->max_dev_resp_rd_atomic_resc =
613 	    qed_attr->max_dev_resp_rd_atomic_resc;
614 	attr->max_cq = qed_attr->max_cq;
615 	attr->max_qp = qed_attr->max_qp;
616 	attr->max_mr = qed_attr->max_mr;
617 	attr->max_mr_size = qed_attr->max_mr_size;
618 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
619 	attr->max_mw = qed_attr->max_mw;
620 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
621 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
622 	attr->max_pd = qed_attr->max_pd;
623 	attr->max_ah = qed_attr->max_ah;
624 	attr->max_pkey = qed_attr->max_pkey;
625 	attr->max_srq = qed_attr->max_srq;
626 	attr->max_srq_wr = qed_attr->max_srq_wr;
627 	attr->dev_caps = qed_attr->dev_caps;
628 	attr->page_size_caps = qed_attr->page_size_caps;
629 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
630 	attr->reserved_lkey = qed_attr->reserved_lkey;
631 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
632 	attr->max_stats_queues = qed_attr->max_stats_queues;
633 
634 	return 0;
635 }
636 
637 static void qedr_unaffiliated_event(void *context, u8 event_code)
638 {
639 	pr_err("unaffiliated event not implemented yet\n");
640 }
641 
642 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
643 {
644 #define EVENT_TYPE_NOT_DEFINED	0
645 #define EVENT_TYPE_CQ		1
646 #define EVENT_TYPE_QP		2
647 #define EVENT_TYPE_SRQ		3
648 	struct qedr_dev *dev = (struct qedr_dev *)context;
649 	struct regpair *async_handle = (struct regpair *)fw_handle;
650 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
651 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
652 	struct ib_event event;
653 	struct ib_srq *ibsrq;
654 	struct qedr_srq *srq;
655 	unsigned long flags;
656 	struct ib_cq *ibcq;
657 	struct ib_qp *ibqp;
658 	struct qedr_cq *cq;
659 	struct qedr_qp *qp;
660 	u16 srq_id;
661 
662 	if (IS_ROCE(dev)) {
663 		switch (e_code) {
664 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
665 			event.event = IB_EVENT_CQ_ERR;
666 			event_type = EVENT_TYPE_CQ;
667 			break;
668 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
669 			event.event = IB_EVENT_SQ_DRAINED;
670 			event_type = EVENT_TYPE_QP;
671 			break;
672 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
673 			event.event = IB_EVENT_QP_FATAL;
674 			event_type = EVENT_TYPE_QP;
675 			break;
676 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
677 			event.event = IB_EVENT_QP_REQ_ERR;
678 			event_type = EVENT_TYPE_QP;
679 			break;
680 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
681 			event.event = IB_EVENT_QP_ACCESS_ERR;
682 			event_type = EVENT_TYPE_QP;
683 			break;
684 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
685 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
686 			event_type = EVENT_TYPE_SRQ;
687 			break;
688 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
689 			event.event = IB_EVENT_SRQ_ERR;
690 			event_type = EVENT_TYPE_SRQ;
691 			break;
692 		case ROCE_ASYNC_EVENT_XRC_DOMAIN_ERR:
693 			event.event = IB_EVENT_QP_ACCESS_ERR;
694 			event_type = EVENT_TYPE_QP;
695 			break;
696 		case ROCE_ASYNC_EVENT_INVALID_XRCETH_ERR:
697 			event.event = IB_EVENT_QP_ACCESS_ERR;
698 			event_type = EVENT_TYPE_QP;
699 			break;
700 		case ROCE_ASYNC_EVENT_XRC_SRQ_CATASTROPHIC_ERR:
701 			event.event = IB_EVENT_CQ_ERR;
702 			event_type = EVENT_TYPE_CQ;
703 			break;
704 		default:
705 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
706 			       e_code, roce_handle64);
707 		}
708 	} else {
709 		switch (e_code) {
710 		case QED_IWARP_EVENT_SRQ_LIMIT:
711 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
712 			event_type = EVENT_TYPE_SRQ;
713 			break;
714 		case QED_IWARP_EVENT_SRQ_EMPTY:
715 			event.event = IB_EVENT_SRQ_ERR;
716 			event_type = EVENT_TYPE_SRQ;
717 			break;
718 		default:
719 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
720 		       roce_handle64);
721 		}
722 	}
723 	switch (event_type) {
724 	case EVENT_TYPE_CQ:
725 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
726 		if (cq) {
727 			ibcq = &cq->ibcq;
728 			if (ibcq->event_handler) {
729 				event.device = ibcq->device;
730 				event.element.cq = ibcq;
731 				ibcq->event_handler(&event, ibcq->cq_context);
732 			}
733 		} else {
734 			WARN(1,
735 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
736 			     roce_handle64);
737 		}
738 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
739 		break;
740 	case EVENT_TYPE_QP:
741 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
742 		if (qp) {
743 			ibqp = &qp->ibqp;
744 			if (ibqp->event_handler) {
745 				event.device = ibqp->device;
746 				event.element.qp = ibqp;
747 				ibqp->event_handler(&event, ibqp->qp_context);
748 			}
749 		} else {
750 			WARN(1,
751 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
752 			     roce_handle64);
753 		}
754 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
755 		break;
756 	case EVENT_TYPE_SRQ:
757 		srq_id = (u16)roce_handle64;
758 		xa_lock_irqsave(&dev->srqs, flags);
759 		srq = xa_load(&dev->srqs, srq_id);
760 		if (srq) {
761 			ibsrq = &srq->ibsrq;
762 			if (ibsrq->event_handler) {
763 				event.device = ibsrq->device;
764 				event.element.srq = ibsrq;
765 				ibsrq->event_handler(&event,
766 						     ibsrq->srq_context);
767 			}
768 		} else {
769 			DP_NOTICE(dev,
770 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
771 				  roce_handle64);
772 		}
773 		xa_unlock_irqrestore(&dev->srqs, flags);
774 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
775 		break;
776 	default:
777 		break;
778 	}
779 }
780 
781 static int qedr_init_hw(struct qedr_dev *dev)
782 {
783 	struct qed_rdma_add_user_out_params out_params;
784 	struct qed_rdma_start_in_params *in_params;
785 	struct qed_rdma_cnq_params *cur_pbl;
786 	struct qed_rdma_events events;
787 	dma_addr_t p_phys_table;
788 	u32 page_cnt;
789 	int rc = 0;
790 	int i;
791 
792 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
793 	if (!in_params) {
794 		rc = -ENOMEM;
795 		goto out;
796 	}
797 
798 	in_params->desired_cnq = dev->num_cnq;
799 	for (i = 0; i < dev->num_cnq; i++) {
800 		cur_pbl = &in_params->cnq_pbl_list[i];
801 
802 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
803 		cur_pbl->num_pbl_pages = page_cnt;
804 
805 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
806 		cur_pbl->pbl_ptr = (u64)p_phys_table;
807 	}
808 
809 	events.affiliated_event = qedr_affiliated_event;
810 	events.unaffiliated_event = qedr_unaffiliated_event;
811 	events.context = dev;
812 
813 	in_params->events = &events;
814 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
815 	in_params->max_mtu = dev->ndev->mtu;
816 	dev->iwarp_max_mtu = dev->ndev->mtu;
817 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
818 
819 	rc = dev->ops->rdma_init(dev->cdev, in_params);
820 	if (rc)
821 		goto out;
822 
823 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
824 	if (rc)
825 		goto out;
826 
827 	dev->db_addr = out_params.dpi_addr;
828 	dev->db_phys_addr = out_params.dpi_phys_addr;
829 	dev->db_size = out_params.dpi_size;
830 	dev->dpi = out_params.dpi;
831 
832 	rc = qedr_set_device_attr(dev);
833 out:
834 	kfree(in_params);
835 	if (rc)
836 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
837 
838 	return rc;
839 }
840 
841 static void qedr_stop_hw(struct qedr_dev *dev)
842 {
843 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
844 	dev->ops->rdma_stop(dev->rdma_ctx);
845 }
846 
847 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
848 				 struct net_device *ndev)
849 {
850 	struct qed_dev_rdma_info dev_info;
851 	struct qedr_dev *dev;
852 	int rc = 0;
853 
854 	dev = ib_alloc_device(qedr_dev, ibdev);
855 	if (!dev) {
856 		pr_err("Unable to allocate ib device\n");
857 		return NULL;
858 	}
859 
860 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
861 
862 	dev->pdev = pdev;
863 	dev->ndev = ndev;
864 	dev->cdev = cdev;
865 
866 	qed_ops = qed_get_rdma_ops();
867 	if (!qed_ops) {
868 		DP_ERR(dev, "Failed to get qed roce operations\n");
869 		goto init_err;
870 	}
871 
872 	dev->ops = qed_ops;
873 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
874 	if (rc)
875 		goto init_err;
876 
877 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
878 	dev->rdma_type = dev_info.rdma_type;
879 	dev->num_hwfns = dev_info.common.num_hwfns;
880 
881 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev)) {
882 		rc = dev->ops->iwarp_set_engine_affin(cdev, false);
883 		if (rc) {
884 			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");
885 			goto init_err;
886 		}
887 	}
888 	dev->affin_hwfn_idx = dev->ops->common->get_affin_hwfn_idx(cdev);
889 
890 	dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
891 
892 	dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
893 	if (!dev->num_cnq) {
894 		DP_ERR(dev, "Failed. At least one CNQ is required.\n");
895 		rc = -ENOMEM;
896 		goto init_err;
897 	}
898 
899 	dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
900 
901 	qedr_pci_set_atomic(dev, pdev);
902 
903 	rc = qedr_alloc_resources(dev);
904 	if (rc)
905 		goto init_err;
906 
907 	rc = qedr_init_hw(dev);
908 	if (rc)
909 		goto alloc_err;
910 
911 	rc = qedr_setup_irqs(dev);
912 	if (rc)
913 		goto irq_err;
914 
915 	rc = qedr_register_device(dev);
916 	if (rc) {
917 		DP_ERR(dev, "Unable to allocate register device\n");
918 		goto reg_err;
919 	}
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 reg_err:
928 	qedr_sync_free_irqs(dev);
929 irq_err:
930 	qedr_stop_hw(dev);
931 alloc_err:
932 	qedr_free_resources(dev);
933 init_err:
934 	ib_dealloc_device(&dev->ibdev);
935 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
936 
937 	return NULL;
938 }
939 
940 static void qedr_remove(struct qedr_dev *dev)
941 {
942 	/* First unregister with stack to stop all the active traffic
943 	 * of the registered clients.
944 	 */
945 	ib_unregister_device(&dev->ibdev);
946 
947 	qedr_stop_hw(dev);
948 	qedr_sync_free_irqs(dev);
949 	qedr_free_resources(dev);
950 
951 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev))
952 		dev->ops->iwarp_set_engine_affin(dev->cdev, true);
953 
954 	ib_dealloc_device(&dev->ibdev);
955 }
956 
957 static void qedr_close(struct qedr_dev *dev)
958 {
959 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
960 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
961 }
962 
963 static void qedr_shutdown(struct qedr_dev *dev)
964 {
965 	qedr_close(dev);
966 	qedr_remove(dev);
967 }
968 
969 static void qedr_open(struct qedr_dev *dev)
970 {
971 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
972 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
973 }
974 
975 static void qedr_mac_address_change(struct qedr_dev *dev)
976 {
977 	union ib_gid *sgid = &dev->sgid_tbl[0];
978 	u8 guid[8], mac_addr[6];
979 	int rc;
980 
981 	/* Update SGID */
982 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
983 	guid[0] = mac_addr[0] ^ 2;
984 	guid[1] = mac_addr[1];
985 	guid[2] = mac_addr[2];
986 	guid[3] = 0xff;
987 	guid[4] = 0xfe;
988 	guid[5] = mac_addr[3];
989 	guid[6] = mac_addr[4];
990 	guid[7] = mac_addr[5];
991 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
992 	memcpy(&sgid->raw[8], guid, sizeof(guid));
993 
994 	/* Update LL2 */
995 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
996 					  dev->gsi_ll2_mac_address,
997 					  dev->ndev->dev_addr);
998 
999 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
1000 
1001 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
1002 
1003 	if (rc)
1004 		DP_ERR(dev, "Error updating mac filter\n");
1005 }
1006 
1007 /* event handling via NIC driver ensures that all the NIC specific
1008  * initialization done before RoCE driver notifies
1009  * event to stack.
1010  */
1011 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1012 {
1013 	switch (event) {
1014 	case QEDE_UP:
1015 		qedr_open(dev);
1016 		break;
1017 	case QEDE_DOWN:
1018 		qedr_close(dev);
1019 		break;
1020 	case QEDE_CLOSE:
1021 		qedr_shutdown(dev);
1022 		break;
1023 	case QEDE_CHANGE_ADDR:
1024 		qedr_mac_address_change(dev);
1025 		break;
1026 	case QEDE_CHANGE_MTU:
1027 		if (rdma_protocol_iwarp(&dev->ibdev, 1))
1028 			if (dev->ndev->mtu != dev->iwarp_max_mtu)
1029 				DP_NOTICE(dev,
1030 					  "Mtu was changed from %d to %d. This will not take affect for iWARP until qedr is reloaded\n",
1031 					  dev->iwarp_max_mtu, dev->ndev->mtu);
1032 		break;
1033 	default:
1034 		pr_err("Event not supported\n");
1035 	}
1036 }
1037 
1038 static struct qedr_driver qedr_drv = {
1039 	.name = "qedr_driver",
1040 	.add = qedr_add,
1041 	.remove = qedr_remove,
1042 	.notify = qedr_notify,
1043 };
1044 
1045 static int __init qedr_init_module(void)
1046 {
1047 	return qede_rdma_register_driver(&qedr_drv);
1048 }
1049 
1050 static void __exit qedr_exit_module(void)
1051 {
1052 	qede_rdma_unregister_driver(&qedr_drv);
1053 }
1054 
1055 module_init(qedr_init_module);
1056 module_exit(qedr_exit_module);
1057