xref: /openbmc/linux/drivers/infiniband/hw/qedr/main.c (revision b8b350af)
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 	.resize_cq = qedr_resize_cq,
232 
233 	INIT_RDMA_OBJ_SIZE(ib_ah, qedr_ah, ibah),
234 	INIT_RDMA_OBJ_SIZE(ib_cq, qedr_cq, ibcq),
235 	INIT_RDMA_OBJ_SIZE(ib_pd, qedr_pd, ibpd),
236 	INIT_RDMA_OBJ_SIZE(ib_qp, qedr_qp, ibqp),
237 	INIT_RDMA_OBJ_SIZE(ib_srq, qedr_srq, ibsrq),
238 	INIT_RDMA_OBJ_SIZE(ib_xrcd, qedr_xrcd, ibxrcd),
239 	INIT_RDMA_OBJ_SIZE(ib_ucontext, qedr_ucontext, ibucontext),
240 };
241 
242 static int qedr_register_device(struct qedr_dev *dev)
243 {
244 	int rc;
245 
246 	dev->ibdev.node_guid = dev->attr.node_guid;
247 	memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
248 
249 	if (IS_IWARP(dev)) {
250 		rc = qedr_iw_register_device(dev);
251 		if (rc)
252 			return rc;
253 	} else {
254 		qedr_roce_register_device(dev);
255 	}
256 
257 	dev->ibdev.phys_port_cnt = 1;
258 	dev->ibdev.num_comp_vectors = dev->num_cnq;
259 	dev->ibdev.dev.parent = &dev->pdev->dev;
260 
261 	ib_set_device_ops(&dev->ibdev, &qedr_dev_ops);
262 
263 	rc = ib_device_set_netdev(&dev->ibdev, dev->ndev, 1);
264 	if (rc)
265 		return rc;
266 
267 	dma_set_max_seg_size(&dev->pdev->dev, UINT_MAX);
268 	return ib_register_device(&dev->ibdev, "qedr%d", &dev->pdev->dev);
269 }
270 
271 /* This function allocates fast-path status block memory */
272 static int qedr_alloc_mem_sb(struct qedr_dev *dev,
273 			     struct qed_sb_info *sb_info, u16 sb_id)
274 {
275 	struct status_block_e4 *sb_virt;
276 	dma_addr_t sb_phys;
277 	int rc;
278 
279 	sb_virt = dma_alloc_coherent(&dev->pdev->dev,
280 				     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
281 	if (!sb_virt)
282 		return -ENOMEM;
283 
284 	rc = dev->ops->common->sb_init(dev->cdev, sb_info,
285 				       sb_virt, sb_phys, sb_id,
286 				       QED_SB_TYPE_CNQ);
287 	if (rc) {
288 		pr_err("Status block initialization failed\n");
289 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
290 				  sb_virt, sb_phys);
291 		return rc;
292 	}
293 
294 	return 0;
295 }
296 
297 static void qedr_free_mem_sb(struct qedr_dev *dev,
298 			     struct qed_sb_info *sb_info, int sb_id)
299 {
300 	if (sb_info->sb_virt) {
301 		dev->ops->common->sb_release(dev->cdev, sb_info, sb_id,
302 					     QED_SB_TYPE_CNQ);
303 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
304 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
305 	}
306 }
307 
308 static void qedr_free_resources(struct qedr_dev *dev)
309 {
310 	int i;
311 
312 	if (IS_IWARP(dev))
313 		destroy_workqueue(dev->iwarp_wq);
314 
315 	for (i = 0; i < dev->num_cnq; i++) {
316 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
317 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
318 	}
319 
320 	kfree(dev->cnq_array);
321 	kfree(dev->sb_array);
322 	kfree(dev->sgid_tbl);
323 }
324 
325 static int qedr_alloc_resources(struct qedr_dev *dev)
326 {
327 	struct qed_chain_init_params params = {
328 		.mode		= QED_CHAIN_MODE_PBL,
329 		.intended_use	= QED_CHAIN_USE_TO_CONSUME,
330 		.cnt_type	= QED_CHAIN_CNT_TYPE_U16,
331 		.elem_size	= sizeof(struct regpair *),
332 	};
333 	struct qedr_cnq *cnq;
334 	__le16 *cons_pi;
335 	int i, rc;
336 
337 	dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
338 				GFP_KERNEL);
339 	if (!dev->sgid_tbl)
340 		return -ENOMEM;
341 
342 	spin_lock_init(&dev->sgid_lock);
343 	xa_init_flags(&dev->srqs, XA_FLAGS_LOCK_IRQ);
344 
345 	if (IS_IWARP(dev)) {
346 		xa_init(&dev->qps);
347 		dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
348 	}
349 
350 	/* Allocate Status blocks for CNQ */
351 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
352 				GFP_KERNEL);
353 	if (!dev->sb_array) {
354 		rc = -ENOMEM;
355 		goto err1;
356 	}
357 
358 	dev->cnq_array = kcalloc(dev->num_cnq,
359 				 sizeof(*dev->cnq_array), GFP_KERNEL);
360 	if (!dev->cnq_array) {
361 		rc = -ENOMEM;
362 		goto err2;
363 	}
364 
365 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
366 
367 	/* Allocate CNQ PBLs */
368 	params.num_elems = min_t(u32, QED_RDMA_MAX_CNQ_SIZE,
369 				 QEDR_ROCE_MAX_CNQ_SIZE);
370 
371 	for (i = 0; i < dev->num_cnq; i++) {
372 		cnq = &dev->cnq_array[i];
373 
374 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
375 				       dev->sb_start + i);
376 		if (rc)
377 			goto err3;
378 
379 		rc = dev->ops->common->chain_alloc(dev->cdev, &cnq->pbl,
380 						   &params);
381 		if (rc)
382 			goto err4;
383 
384 		cnq->dev = dev;
385 		cnq->sb = &dev->sb_array[i];
386 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
387 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
388 		cnq->index = i;
389 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
390 
391 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
392 			 i, qed_chain_get_cons_idx(&cnq->pbl));
393 	}
394 
395 	return 0;
396 err4:
397 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
398 err3:
399 	for (--i; i >= 0; i--) {
400 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
401 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
402 	}
403 	kfree(dev->cnq_array);
404 err2:
405 	kfree(dev->sb_array);
406 err1:
407 	kfree(dev->sgid_tbl);
408 	return rc;
409 }
410 
411 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
412 {
413 	int rc = pci_enable_atomic_ops_to_root(pdev,
414 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
415 
416 	if (rc) {
417 		dev->atomic_cap = IB_ATOMIC_NONE;
418 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
419 	} else {
420 		dev->atomic_cap = IB_ATOMIC_GLOB;
421 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
422 	}
423 }
424 
425 static const struct qed_rdma_ops *qed_ops;
426 
427 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
428 
429 static irqreturn_t qedr_irq_handler(int irq, void *handle)
430 {
431 	u16 hw_comp_cons, sw_comp_cons;
432 	struct qedr_cnq *cnq = handle;
433 	struct regpair *cq_handle;
434 	struct qedr_cq *cq;
435 
436 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
437 
438 	qed_sb_update_sb_idx(cnq->sb);
439 
440 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
441 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
442 
443 	/* Align protocol-index and chain reads */
444 	rmb();
445 
446 	while (sw_comp_cons != hw_comp_cons) {
447 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
448 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
449 				cq_handle->lo);
450 
451 		if (cq == NULL) {
452 			DP_ERR(cnq->dev,
453 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
454 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
455 			       hw_comp_cons);
456 
457 			break;
458 		}
459 
460 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
461 			DP_ERR(cnq->dev,
462 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
463 			       cq_handle->hi, cq_handle->lo, cq);
464 			break;
465 		}
466 
467 		cq->arm_flags = 0;
468 
469 		if (!cq->destroyed && cq->ibcq.comp_handler)
470 			(*cq->ibcq.comp_handler)
471 				(&cq->ibcq, cq->ibcq.cq_context);
472 
473 		/* The CQ's CNQ notification counter is checked before
474 		 * destroying the CQ in a busy-wait loop that waits for all of
475 		 * the CQ's CNQ interrupts to be processed. It is increased
476 		 * here, only after the completion handler, to ensure that the
477 		 * the handler is not running when the CQ is destroyed.
478 		 */
479 		cq->cnq_notif++;
480 
481 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
482 
483 		cnq->n_comp++;
484 	}
485 
486 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
487 				      sw_comp_cons);
488 
489 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
490 
491 	return IRQ_HANDLED;
492 }
493 
494 static void qedr_sync_free_irqs(struct qedr_dev *dev)
495 {
496 	u32 vector;
497 	u16 idx;
498 	int i;
499 
500 	for (i = 0; i < dev->int_info.used_cnt; i++) {
501 		if (dev->int_info.msix_cnt) {
502 			idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
503 			vector = dev->int_info.msix[idx].vector;
504 			synchronize_irq(vector);
505 			free_irq(vector, &dev->cnq_array[i]);
506 		}
507 	}
508 
509 	dev->int_info.used_cnt = 0;
510 }
511 
512 static int qedr_req_msix_irqs(struct qedr_dev *dev)
513 {
514 	int i, rc = 0;
515 	u16 idx;
516 
517 	if (dev->num_cnq > dev->int_info.msix_cnt) {
518 		DP_ERR(dev,
519 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
520 		       dev->num_cnq, dev->int_info.msix_cnt);
521 		return -EINVAL;
522 	}
523 
524 	for (i = 0; i < dev->num_cnq; i++) {
525 		idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
526 		rc = request_irq(dev->int_info.msix[idx].vector,
527 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
528 				 &dev->cnq_array[i]);
529 		if (rc) {
530 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
531 			qedr_sync_free_irqs(dev);
532 		} else {
533 			DP_DEBUG(dev, QEDR_MSG_INIT,
534 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
535 				 dev->cnq_array[i].name, i,
536 				 &dev->cnq_array[i]);
537 			dev->int_info.used_cnt++;
538 		}
539 	}
540 
541 	return rc;
542 }
543 
544 static int qedr_setup_irqs(struct qedr_dev *dev)
545 {
546 	int rc;
547 
548 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
549 
550 	/* Learn Interrupt configuration */
551 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
552 	if (rc < 0)
553 		return rc;
554 
555 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
556 	if (rc) {
557 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
558 		return rc;
559 	}
560 
561 	if (dev->int_info.msix_cnt) {
562 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
563 			 dev->int_info.msix_cnt);
564 		rc = qedr_req_msix_irqs(dev);
565 		if (rc)
566 			return rc;
567 	}
568 
569 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
570 
571 	return 0;
572 }
573 
574 static int qedr_set_device_attr(struct qedr_dev *dev)
575 {
576 	struct qed_rdma_device *qed_attr;
577 	struct qedr_device_attr *attr;
578 	u32 page_size;
579 
580 	/* Part 1 - query core capabilities */
581 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
582 
583 	/* Part 2 - check capabilities */
584 	page_size = ~qed_attr->page_size_caps + 1;
585 	if (page_size > PAGE_SIZE) {
586 		DP_ERR(dev,
587 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
588 		       PAGE_SIZE, page_size);
589 		return -ENODEV;
590 	}
591 
592 	/* Part 3 - copy and update capabilities */
593 	attr = &dev->attr;
594 	attr->vendor_id = qed_attr->vendor_id;
595 	attr->vendor_part_id = qed_attr->vendor_part_id;
596 	attr->hw_ver = qed_attr->hw_ver;
597 	attr->fw_ver = qed_attr->fw_ver;
598 	attr->node_guid = qed_attr->node_guid;
599 	attr->sys_image_guid = qed_attr->sys_image_guid;
600 	attr->max_cnq = qed_attr->max_cnq;
601 	attr->max_sge = qed_attr->max_sge;
602 	attr->max_inline = qed_attr->max_inline;
603 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
604 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
605 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
606 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
607 	attr->max_dev_resp_rd_atomic_resc =
608 	    qed_attr->max_dev_resp_rd_atomic_resc;
609 	attr->max_cq = qed_attr->max_cq;
610 	attr->max_qp = qed_attr->max_qp;
611 	attr->max_mr = qed_attr->max_mr;
612 	attr->max_mr_size = qed_attr->max_mr_size;
613 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
614 	attr->max_mw = qed_attr->max_mw;
615 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
616 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
617 	attr->max_pd = qed_attr->max_pd;
618 	attr->max_ah = qed_attr->max_ah;
619 	attr->max_pkey = qed_attr->max_pkey;
620 	attr->max_srq = qed_attr->max_srq;
621 	attr->max_srq_wr = qed_attr->max_srq_wr;
622 	attr->dev_caps = qed_attr->dev_caps;
623 	attr->page_size_caps = qed_attr->page_size_caps;
624 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
625 	attr->reserved_lkey = qed_attr->reserved_lkey;
626 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
627 	attr->max_stats_queues = qed_attr->max_stats_queues;
628 
629 	return 0;
630 }
631 
632 static void qedr_unaffiliated_event(void *context, u8 event_code)
633 {
634 	pr_err("unaffiliated event not implemented yet\n");
635 }
636 
637 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
638 {
639 #define EVENT_TYPE_NOT_DEFINED	0
640 #define EVENT_TYPE_CQ		1
641 #define EVENT_TYPE_QP		2
642 #define EVENT_TYPE_SRQ		3
643 	struct qedr_dev *dev = (struct qedr_dev *)context;
644 	struct regpair *async_handle = (struct regpair *)fw_handle;
645 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
646 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
647 	struct ib_event event;
648 	struct ib_srq *ibsrq;
649 	struct qedr_srq *srq;
650 	unsigned long flags;
651 	struct ib_cq *ibcq;
652 	struct ib_qp *ibqp;
653 	struct qedr_cq *cq;
654 	struct qedr_qp *qp;
655 	u16 srq_id;
656 
657 	if (IS_ROCE(dev)) {
658 		switch (e_code) {
659 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
660 			event.event = IB_EVENT_CQ_ERR;
661 			event_type = EVENT_TYPE_CQ;
662 			break;
663 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
664 			event.event = IB_EVENT_SQ_DRAINED;
665 			event_type = EVENT_TYPE_QP;
666 			break;
667 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
668 			event.event = IB_EVENT_QP_FATAL;
669 			event_type = EVENT_TYPE_QP;
670 			break;
671 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
672 			event.event = IB_EVENT_QP_REQ_ERR;
673 			event_type = EVENT_TYPE_QP;
674 			break;
675 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
676 			event.event = IB_EVENT_QP_ACCESS_ERR;
677 			event_type = EVENT_TYPE_QP;
678 			break;
679 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
680 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
681 			event_type = EVENT_TYPE_SRQ;
682 			break;
683 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
684 			event.event = IB_EVENT_SRQ_ERR;
685 			event_type = EVENT_TYPE_SRQ;
686 			break;
687 		case ROCE_ASYNC_EVENT_XRC_DOMAIN_ERR:
688 			event.event = IB_EVENT_QP_ACCESS_ERR;
689 			event_type = EVENT_TYPE_QP;
690 			break;
691 		case ROCE_ASYNC_EVENT_INVALID_XRCETH_ERR:
692 			event.event = IB_EVENT_QP_ACCESS_ERR;
693 			event_type = EVENT_TYPE_QP;
694 			break;
695 		case ROCE_ASYNC_EVENT_XRC_SRQ_CATASTROPHIC_ERR:
696 			event.event = IB_EVENT_CQ_ERR;
697 			event_type = EVENT_TYPE_CQ;
698 			break;
699 		default:
700 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
701 			       e_code, roce_handle64);
702 		}
703 	} else {
704 		switch (e_code) {
705 		case QED_IWARP_EVENT_SRQ_LIMIT:
706 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
707 			event_type = EVENT_TYPE_SRQ;
708 			break;
709 		case QED_IWARP_EVENT_SRQ_EMPTY:
710 			event.event = IB_EVENT_SRQ_ERR;
711 			event_type = EVENT_TYPE_SRQ;
712 			break;
713 		default:
714 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
715 		       roce_handle64);
716 		}
717 	}
718 	switch (event_type) {
719 	case EVENT_TYPE_CQ:
720 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
721 		if (cq) {
722 			ibcq = &cq->ibcq;
723 			if (ibcq->event_handler) {
724 				event.device = ibcq->device;
725 				event.element.cq = ibcq;
726 				ibcq->event_handler(&event, ibcq->cq_context);
727 			}
728 		} else {
729 			WARN(1,
730 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
731 			     roce_handle64);
732 		}
733 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
734 		break;
735 	case EVENT_TYPE_QP:
736 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
737 		if (qp) {
738 			ibqp = &qp->ibqp;
739 			if (ibqp->event_handler) {
740 				event.device = ibqp->device;
741 				event.element.qp = ibqp;
742 				ibqp->event_handler(&event, ibqp->qp_context);
743 			}
744 		} else {
745 			WARN(1,
746 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
747 			     roce_handle64);
748 		}
749 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
750 		break;
751 	case EVENT_TYPE_SRQ:
752 		srq_id = (u16)roce_handle64;
753 		xa_lock_irqsave(&dev->srqs, flags);
754 		srq = xa_load(&dev->srqs, srq_id);
755 		if (srq) {
756 			ibsrq = &srq->ibsrq;
757 			if (ibsrq->event_handler) {
758 				event.device = ibsrq->device;
759 				event.element.srq = ibsrq;
760 				ibsrq->event_handler(&event,
761 						     ibsrq->srq_context);
762 			}
763 		} else {
764 			DP_NOTICE(dev,
765 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
766 				  roce_handle64);
767 		}
768 		xa_unlock_irqrestore(&dev->srqs, flags);
769 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
770 		break;
771 	default:
772 		break;
773 	}
774 }
775 
776 static int qedr_init_hw(struct qedr_dev *dev)
777 {
778 	struct qed_rdma_add_user_out_params out_params;
779 	struct qed_rdma_start_in_params *in_params;
780 	struct qed_rdma_cnq_params *cur_pbl;
781 	struct qed_rdma_events events;
782 	dma_addr_t p_phys_table;
783 	u32 page_cnt;
784 	int rc = 0;
785 	int i;
786 
787 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
788 	if (!in_params) {
789 		rc = -ENOMEM;
790 		goto out;
791 	}
792 
793 	in_params->desired_cnq = dev->num_cnq;
794 	for (i = 0; i < dev->num_cnq; i++) {
795 		cur_pbl = &in_params->cnq_pbl_list[i];
796 
797 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
798 		cur_pbl->num_pbl_pages = page_cnt;
799 
800 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
801 		cur_pbl->pbl_ptr = (u64)p_phys_table;
802 	}
803 
804 	events.affiliated_event = qedr_affiliated_event;
805 	events.unaffiliated_event = qedr_unaffiliated_event;
806 	events.context = dev;
807 
808 	in_params->events = &events;
809 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
810 	in_params->max_mtu = dev->ndev->mtu;
811 	dev->iwarp_max_mtu = dev->ndev->mtu;
812 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
813 
814 	rc = dev->ops->rdma_init(dev->cdev, in_params);
815 	if (rc)
816 		goto out;
817 
818 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
819 	if (rc)
820 		goto out;
821 
822 	dev->db_addr = out_params.dpi_addr;
823 	dev->db_phys_addr = out_params.dpi_phys_addr;
824 	dev->db_size = out_params.dpi_size;
825 	dev->dpi = out_params.dpi;
826 
827 	rc = qedr_set_device_attr(dev);
828 out:
829 	kfree(in_params);
830 	if (rc)
831 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
832 
833 	return rc;
834 }
835 
836 static void qedr_stop_hw(struct qedr_dev *dev)
837 {
838 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
839 	dev->ops->rdma_stop(dev->rdma_ctx);
840 }
841 
842 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
843 				 struct net_device *ndev)
844 {
845 	struct qed_dev_rdma_info dev_info;
846 	struct qedr_dev *dev;
847 	int rc = 0;
848 
849 	dev = ib_alloc_device(qedr_dev, ibdev);
850 	if (!dev) {
851 		pr_err("Unable to allocate ib device\n");
852 		return NULL;
853 	}
854 
855 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
856 
857 	dev->pdev = pdev;
858 	dev->ndev = ndev;
859 	dev->cdev = cdev;
860 
861 	qed_ops = qed_get_rdma_ops();
862 	if (!qed_ops) {
863 		DP_ERR(dev, "Failed to get qed roce operations\n");
864 		goto init_err;
865 	}
866 
867 	dev->ops = qed_ops;
868 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
869 	if (rc)
870 		goto init_err;
871 
872 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
873 	dev->rdma_type = dev_info.rdma_type;
874 	dev->num_hwfns = dev_info.common.num_hwfns;
875 
876 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev)) {
877 		rc = dev->ops->iwarp_set_engine_affin(cdev, false);
878 		if (rc) {
879 			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");
880 			goto init_err;
881 		}
882 	}
883 	dev->affin_hwfn_idx = dev->ops->common->get_affin_hwfn_idx(cdev);
884 
885 	dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
886 
887 	dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
888 	if (!dev->num_cnq) {
889 		DP_ERR(dev, "Failed. At least one CNQ is required.\n");
890 		rc = -ENOMEM;
891 		goto init_err;
892 	}
893 
894 	dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
895 
896 	qedr_pci_set_atomic(dev, pdev);
897 
898 	rc = qedr_alloc_resources(dev);
899 	if (rc)
900 		goto init_err;
901 
902 	rc = qedr_init_hw(dev);
903 	if (rc)
904 		goto alloc_err;
905 
906 	rc = qedr_setup_irqs(dev);
907 	if (rc)
908 		goto irq_err;
909 
910 	rc = qedr_register_device(dev);
911 	if (rc) {
912 		DP_ERR(dev, "Unable to allocate register device\n");
913 		goto reg_err;
914 	}
915 
916 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
917 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
918 
919 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
920 	return dev;
921 
922 reg_err:
923 	qedr_sync_free_irqs(dev);
924 irq_err:
925 	qedr_stop_hw(dev);
926 alloc_err:
927 	qedr_free_resources(dev);
928 init_err:
929 	ib_dealloc_device(&dev->ibdev);
930 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
931 
932 	return NULL;
933 }
934 
935 static void qedr_remove(struct qedr_dev *dev)
936 {
937 	/* First unregister with stack to stop all the active traffic
938 	 * of the registered clients.
939 	 */
940 	ib_unregister_device(&dev->ibdev);
941 
942 	qedr_stop_hw(dev);
943 	qedr_sync_free_irqs(dev);
944 	qedr_free_resources(dev);
945 
946 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev))
947 		dev->ops->iwarp_set_engine_affin(dev->cdev, true);
948 
949 	ib_dealloc_device(&dev->ibdev);
950 }
951 
952 static void qedr_close(struct qedr_dev *dev)
953 {
954 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
955 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
956 }
957 
958 static void qedr_shutdown(struct qedr_dev *dev)
959 {
960 	qedr_close(dev);
961 	qedr_remove(dev);
962 }
963 
964 static void qedr_open(struct qedr_dev *dev)
965 {
966 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
967 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
968 }
969 
970 static void qedr_mac_address_change(struct qedr_dev *dev)
971 {
972 	union ib_gid *sgid = &dev->sgid_tbl[0];
973 	u8 guid[8], mac_addr[6];
974 	int rc;
975 
976 	/* Update SGID */
977 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
978 	guid[0] = mac_addr[0] ^ 2;
979 	guid[1] = mac_addr[1];
980 	guid[2] = mac_addr[2];
981 	guid[3] = 0xff;
982 	guid[4] = 0xfe;
983 	guid[5] = mac_addr[3];
984 	guid[6] = mac_addr[4];
985 	guid[7] = mac_addr[5];
986 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
987 	memcpy(&sgid->raw[8], guid, sizeof(guid));
988 
989 	/* Update LL2 */
990 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
991 					  dev->gsi_ll2_mac_address,
992 					  dev->ndev->dev_addr);
993 
994 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
995 
996 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
997 
998 	if (rc)
999 		DP_ERR(dev, "Error updating mac filter\n");
1000 }
1001 
1002 /* event handling via NIC driver ensures that all the NIC specific
1003  * initialization done before RoCE driver notifies
1004  * event to stack.
1005  */
1006 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1007 {
1008 	switch (event) {
1009 	case QEDE_UP:
1010 		qedr_open(dev);
1011 		break;
1012 	case QEDE_DOWN:
1013 		qedr_close(dev);
1014 		break;
1015 	case QEDE_CLOSE:
1016 		qedr_shutdown(dev);
1017 		break;
1018 	case QEDE_CHANGE_ADDR:
1019 		qedr_mac_address_change(dev);
1020 		break;
1021 	case QEDE_CHANGE_MTU:
1022 		if (rdma_protocol_iwarp(&dev->ibdev, 1))
1023 			if (dev->ndev->mtu != dev->iwarp_max_mtu)
1024 				DP_NOTICE(dev,
1025 					  "Mtu was changed from %d to %d. This will not take affect for iWARP until qedr is reloaded\n",
1026 					  dev->iwarp_max_mtu, dev->ndev->mtu);
1027 		break;
1028 	default:
1029 		pr_err("Event not supported\n");
1030 	}
1031 }
1032 
1033 static struct qedr_driver qedr_drv = {
1034 	.name = "qedr_driver",
1035 	.add = qedr_add,
1036 	.remove = qedr_remove,
1037 	.notify = qedr_notify,
1038 };
1039 
1040 static int __init qedr_init_module(void)
1041 {
1042 	return qede_rdma_register_driver(&qedr_drv);
1043 }
1044 
1045 static void __exit qedr_exit_module(void)
1046 {
1047 	qede_rdma_unregister_driver(&qedr_drv);
1048 }
1049 
1050 module_init(qedr_init_module);
1051 module_exit(qedr_exit_module);
1052