xref: /openbmc/linux/drivers/infiniband/hw/qedr/main.c (revision f2d8e15b)
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 	}
348 
349 	/* Allocate Status blocks for CNQ */
350 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
351 				GFP_KERNEL);
352 	if (!dev->sb_array) {
353 		rc = -ENOMEM;
354 		goto err1;
355 	}
356 
357 	dev->cnq_array = kcalloc(dev->num_cnq,
358 				 sizeof(*dev->cnq_array), GFP_KERNEL);
359 	if (!dev->cnq_array) {
360 		rc = -ENOMEM;
361 		goto err2;
362 	}
363 
364 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
365 
366 	/* Allocate CNQ PBLs */
367 	params.num_elems = min_t(u32, QED_RDMA_MAX_CNQ_SIZE,
368 				 QEDR_ROCE_MAX_CNQ_SIZE);
369 
370 	for (i = 0; i < dev->num_cnq; i++) {
371 		cnq = &dev->cnq_array[i];
372 
373 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
374 				       dev->sb_start + i);
375 		if (rc)
376 			goto err3;
377 
378 		rc = dev->ops->common->chain_alloc(dev->cdev, &cnq->pbl,
379 						   &params);
380 		if (rc)
381 			goto err4;
382 
383 		cnq->dev = dev;
384 		cnq->sb = &dev->sb_array[i];
385 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
386 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
387 		cnq->index = i;
388 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
389 
390 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
391 			 i, qed_chain_get_cons_idx(&cnq->pbl));
392 	}
393 
394 	return 0;
395 err4:
396 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
397 err3:
398 	for (--i; i >= 0; i--) {
399 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
400 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
401 	}
402 	kfree(dev->cnq_array);
403 err2:
404 	kfree(dev->sb_array);
405 err1:
406 	kfree(dev->sgid_tbl);
407 	return rc;
408 }
409 
410 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
411 {
412 	int rc = pci_enable_atomic_ops_to_root(pdev,
413 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
414 
415 	if (rc) {
416 		dev->atomic_cap = IB_ATOMIC_NONE;
417 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
418 	} else {
419 		dev->atomic_cap = IB_ATOMIC_GLOB;
420 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
421 	}
422 }
423 
424 static const struct qed_rdma_ops *qed_ops;
425 
426 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
427 
428 static irqreturn_t qedr_irq_handler(int irq, void *handle)
429 {
430 	u16 hw_comp_cons, sw_comp_cons;
431 	struct qedr_cnq *cnq = handle;
432 	struct regpair *cq_handle;
433 	struct qedr_cq *cq;
434 
435 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
436 
437 	qed_sb_update_sb_idx(cnq->sb);
438 
439 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
440 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
441 
442 	/* Align protocol-index and chain reads */
443 	rmb();
444 
445 	while (sw_comp_cons != hw_comp_cons) {
446 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
447 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
448 				cq_handle->lo);
449 
450 		if (cq == NULL) {
451 			DP_ERR(cnq->dev,
452 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
453 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
454 			       hw_comp_cons);
455 
456 			break;
457 		}
458 
459 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
460 			DP_ERR(cnq->dev,
461 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
462 			       cq_handle->hi, cq_handle->lo, cq);
463 			break;
464 		}
465 
466 		cq->arm_flags = 0;
467 
468 		if (!cq->destroyed && cq->ibcq.comp_handler)
469 			(*cq->ibcq.comp_handler)
470 				(&cq->ibcq, cq->ibcq.cq_context);
471 
472 		/* The CQ's CNQ notification counter is checked before
473 		 * destroying the CQ in a busy-wait loop that waits for all of
474 		 * the CQ's CNQ interrupts to be processed. It is increased
475 		 * here, only after the completion handler, to ensure that the
476 		 * the handler is not running when the CQ is destroyed.
477 		 */
478 		cq->cnq_notif++;
479 
480 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
481 
482 		cnq->n_comp++;
483 	}
484 
485 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
486 				      sw_comp_cons);
487 
488 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
489 
490 	return IRQ_HANDLED;
491 }
492 
493 static void qedr_sync_free_irqs(struct qedr_dev *dev)
494 {
495 	u32 vector;
496 	u16 idx;
497 	int i;
498 
499 	for (i = 0; i < dev->int_info.used_cnt; i++) {
500 		if (dev->int_info.msix_cnt) {
501 			idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
502 			vector = dev->int_info.msix[idx].vector;
503 			free_irq(vector, &dev->cnq_array[i]);
504 		}
505 	}
506 
507 	dev->int_info.used_cnt = 0;
508 }
509 
510 static int qedr_req_msix_irqs(struct qedr_dev *dev)
511 {
512 	int i, rc = 0;
513 	u16 idx;
514 
515 	if (dev->num_cnq > dev->int_info.msix_cnt) {
516 		DP_ERR(dev,
517 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
518 		       dev->num_cnq, dev->int_info.msix_cnt);
519 		return -EINVAL;
520 	}
521 
522 	for (i = 0; i < dev->num_cnq; i++) {
523 		idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
524 		rc = request_irq(dev->int_info.msix[idx].vector,
525 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
526 				 &dev->cnq_array[i]);
527 		if (rc) {
528 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
529 			qedr_sync_free_irqs(dev);
530 		} else {
531 			DP_DEBUG(dev, QEDR_MSG_INIT,
532 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
533 				 dev->cnq_array[i].name, i,
534 				 &dev->cnq_array[i]);
535 			dev->int_info.used_cnt++;
536 		}
537 	}
538 
539 	return rc;
540 }
541 
542 static int qedr_setup_irqs(struct qedr_dev *dev)
543 {
544 	int rc;
545 
546 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
547 
548 	/* Learn Interrupt configuration */
549 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
550 	if (rc < 0)
551 		return rc;
552 
553 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
554 	if (rc) {
555 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
556 		return rc;
557 	}
558 
559 	if (dev->int_info.msix_cnt) {
560 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
561 			 dev->int_info.msix_cnt);
562 		rc = qedr_req_msix_irqs(dev);
563 		if (rc)
564 			return rc;
565 	}
566 
567 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
568 
569 	return 0;
570 }
571 
572 static int qedr_set_device_attr(struct qedr_dev *dev)
573 {
574 	struct qed_rdma_device *qed_attr;
575 	struct qedr_device_attr *attr;
576 	u32 page_size;
577 
578 	/* Part 1 - query core capabilities */
579 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
580 
581 	/* Part 2 - check capabilities */
582 	page_size = ~qed_attr->page_size_caps + 1;
583 	if (page_size > PAGE_SIZE) {
584 		DP_ERR(dev,
585 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
586 		       PAGE_SIZE, page_size);
587 		return -ENODEV;
588 	}
589 
590 	/* Part 3 - copy and update capabilities */
591 	attr = &dev->attr;
592 	attr->vendor_id = qed_attr->vendor_id;
593 	attr->vendor_part_id = qed_attr->vendor_part_id;
594 	attr->hw_ver = qed_attr->hw_ver;
595 	attr->fw_ver = qed_attr->fw_ver;
596 	attr->node_guid = qed_attr->node_guid;
597 	attr->sys_image_guid = qed_attr->sys_image_guid;
598 	attr->max_cnq = qed_attr->max_cnq;
599 	attr->max_sge = qed_attr->max_sge;
600 	attr->max_inline = qed_attr->max_inline;
601 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
602 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
603 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
604 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
605 	attr->max_dev_resp_rd_atomic_resc =
606 	    qed_attr->max_dev_resp_rd_atomic_resc;
607 	attr->max_cq = qed_attr->max_cq;
608 	attr->max_qp = qed_attr->max_qp;
609 	attr->max_mr = qed_attr->max_mr;
610 	attr->max_mr_size = qed_attr->max_mr_size;
611 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
612 	attr->max_mw = qed_attr->max_mw;
613 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
614 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
615 	attr->max_pd = qed_attr->max_pd;
616 	attr->max_ah = qed_attr->max_ah;
617 	attr->max_pkey = qed_attr->max_pkey;
618 	attr->max_srq = qed_attr->max_srq;
619 	attr->max_srq_wr = qed_attr->max_srq_wr;
620 	attr->dev_caps = qed_attr->dev_caps;
621 	attr->page_size_caps = qed_attr->page_size_caps;
622 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
623 	attr->reserved_lkey = qed_attr->reserved_lkey;
624 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
625 	attr->max_stats_queues = qed_attr->max_stats_queues;
626 
627 	return 0;
628 }
629 
630 static void qedr_unaffiliated_event(void *context, u8 event_code)
631 {
632 	pr_err("unaffiliated event not implemented yet\n");
633 }
634 
635 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
636 {
637 #define EVENT_TYPE_NOT_DEFINED	0
638 #define EVENT_TYPE_CQ		1
639 #define EVENT_TYPE_QP		2
640 #define EVENT_TYPE_SRQ		3
641 	struct qedr_dev *dev = (struct qedr_dev *)context;
642 	struct regpair *async_handle = (struct regpair *)fw_handle;
643 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
644 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
645 	struct ib_event event;
646 	struct ib_srq *ibsrq;
647 	struct qedr_srq *srq;
648 	unsigned long flags;
649 	struct ib_cq *ibcq;
650 	struct ib_qp *ibqp;
651 	struct qedr_cq *cq;
652 	struct qedr_qp *qp;
653 	u16 srq_id;
654 
655 	if (IS_ROCE(dev)) {
656 		switch (e_code) {
657 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
658 			event.event = IB_EVENT_CQ_ERR;
659 			event_type = EVENT_TYPE_CQ;
660 			break;
661 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
662 			event.event = IB_EVENT_SQ_DRAINED;
663 			event_type = EVENT_TYPE_QP;
664 			break;
665 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
666 			event.event = IB_EVENT_QP_FATAL;
667 			event_type = EVENT_TYPE_QP;
668 			break;
669 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
670 			event.event = IB_EVENT_QP_REQ_ERR;
671 			event_type = EVENT_TYPE_QP;
672 			break;
673 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
674 			event.event = IB_EVENT_QP_ACCESS_ERR;
675 			event_type = EVENT_TYPE_QP;
676 			break;
677 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
678 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
679 			event_type = EVENT_TYPE_SRQ;
680 			break;
681 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
682 			event.event = IB_EVENT_SRQ_ERR;
683 			event_type = EVENT_TYPE_SRQ;
684 			break;
685 		case ROCE_ASYNC_EVENT_XRC_DOMAIN_ERR:
686 			event.event = IB_EVENT_QP_ACCESS_ERR;
687 			event_type = EVENT_TYPE_QP;
688 			break;
689 		case ROCE_ASYNC_EVENT_INVALID_XRCETH_ERR:
690 			event.event = IB_EVENT_QP_ACCESS_ERR;
691 			event_type = EVENT_TYPE_QP;
692 			break;
693 		case ROCE_ASYNC_EVENT_XRC_SRQ_CATASTROPHIC_ERR:
694 			event.event = IB_EVENT_CQ_ERR;
695 			event_type = EVENT_TYPE_CQ;
696 			break;
697 		default:
698 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
699 			       e_code, roce_handle64);
700 		}
701 	} else {
702 		switch (e_code) {
703 		case QED_IWARP_EVENT_SRQ_LIMIT:
704 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
705 			event_type = EVENT_TYPE_SRQ;
706 			break;
707 		case QED_IWARP_EVENT_SRQ_EMPTY:
708 			event.event = IB_EVENT_SRQ_ERR;
709 			event_type = EVENT_TYPE_SRQ;
710 			break;
711 		default:
712 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
713 		       roce_handle64);
714 		}
715 	}
716 	switch (event_type) {
717 	case EVENT_TYPE_CQ:
718 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
719 		if (cq) {
720 			ibcq = &cq->ibcq;
721 			if (ibcq->event_handler) {
722 				event.device = ibcq->device;
723 				event.element.cq = ibcq;
724 				ibcq->event_handler(&event, ibcq->cq_context);
725 			}
726 		} else {
727 			WARN(1,
728 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
729 			     roce_handle64);
730 		}
731 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
732 		break;
733 	case EVENT_TYPE_QP:
734 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
735 		if (qp) {
736 			ibqp = &qp->ibqp;
737 			if (ibqp->event_handler) {
738 				event.device = ibqp->device;
739 				event.element.qp = ibqp;
740 				ibqp->event_handler(&event, ibqp->qp_context);
741 			}
742 		} else {
743 			WARN(1,
744 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
745 			     roce_handle64);
746 		}
747 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
748 		break;
749 	case EVENT_TYPE_SRQ:
750 		srq_id = (u16)roce_handle64;
751 		xa_lock_irqsave(&dev->srqs, flags);
752 		srq = xa_load(&dev->srqs, srq_id);
753 		if (srq) {
754 			ibsrq = &srq->ibsrq;
755 			if (ibsrq->event_handler) {
756 				event.device = ibsrq->device;
757 				event.element.srq = ibsrq;
758 				ibsrq->event_handler(&event,
759 						     ibsrq->srq_context);
760 			}
761 		} else {
762 			DP_NOTICE(dev,
763 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
764 				  roce_handle64);
765 		}
766 		xa_unlock_irqrestore(&dev->srqs, flags);
767 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
768 		break;
769 	default:
770 		break;
771 	}
772 }
773 
774 static int qedr_init_hw(struct qedr_dev *dev)
775 {
776 	struct qed_rdma_add_user_out_params out_params;
777 	struct qed_rdma_start_in_params *in_params;
778 	struct qed_rdma_cnq_params *cur_pbl;
779 	struct qed_rdma_events events;
780 	dma_addr_t p_phys_table;
781 	u32 page_cnt;
782 	int rc = 0;
783 	int i;
784 
785 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
786 	if (!in_params) {
787 		rc = -ENOMEM;
788 		goto out;
789 	}
790 
791 	in_params->desired_cnq = dev->num_cnq;
792 	for (i = 0; i < dev->num_cnq; i++) {
793 		cur_pbl = &in_params->cnq_pbl_list[i];
794 
795 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
796 		cur_pbl->num_pbl_pages = page_cnt;
797 
798 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
799 		cur_pbl->pbl_ptr = (u64)p_phys_table;
800 	}
801 
802 	events.affiliated_event = qedr_affiliated_event;
803 	events.unaffiliated_event = qedr_unaffiliated_event;
804 	events.context = dev;
805 
806 	in_params->events = &events;
807 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
808 	in_params->max_mtu = dev->ndev->mtu;
809 	dev->iwarp_max_mtu = dev->ndev->mtu;
810 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
811 
812 	rc = dev->ops->rdma_init(dev->cdev, in_params);
813 	if (rc)
814 		goto out;
815 
816 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
817 	if (rc)
818 		goto out;
819 
820 	dev->db_addr = out_params.dpi_addr;
821 	dev->db_phys_addr = out_params.dpi_phys_addr;
822 	dev->db_size = out_params.dpi_size;
823 	dev->dpi = out_params.dpi;
824 
825 	rc = qedr_set_device_attr(dev);
826 out:
827 	kfree(in_params);
828 	if (rc)
829 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
830 
831 	return rc;
832 }
833 
834 static void qedr_stop_hw(struct qedr_dev *dev)
835 {
836 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
837 	dev->ops->rdma_stop(dev->rdma_ctx);
838 }
839 
840 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
841 				 struct net_device *ndev)
842 {
843 	struct qed_dev_rdma_info dev_info;
844 	struct qedr_dev *dev;
845 	int rc = 0;
846 
847 	dev = ib_alloc_device(qedr_dev, ibdev);
848 	if (!dev) {
849 		pr_err("Unable to allocate ib device\n");
850 		return NULL;
851 	}
852 
853 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
854 
855 	dev->pdev = pdev;
856 	dev->ndev = ndev;
857 	dev->cdev = cdev;
858 
859 	qed_ops = qed_get_rdma_ops();
860 	if (!qed_ops) {
861 		DP_ERR(dev, "Failed to get qed roce operations\n");
862 		goto init_err;
863 	}
864 
865 	dev->ops = qed_ops;
866 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
867 	if (rc)
868 		goto init_err;
869 
870 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
871 	dev->rdma_type = dev_info.rdma_type;
872 	dev->num_hwfns = dev_info.common.num_hwfns;
873 
874 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev)) {
875 		rc = dev->ops->iwarp_set_engine_affin(cdev, false);
876 		if (rc) {
877 			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");
878 			goto init_err;
879 		}
880 	}
881 	dev->affin_hwfn_idx = dev->ops->common->get_affin_hwfn_idx(cdev);
882 
883 	dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
884 
885 	dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
886 	if (!dev->num_cnq) {
887 		DP_ERR(dev, "Failed. At least one CNQ is required.\n");
888 		rc = -ENOMEM;
889 		goto init_err;
890 	}
891 
892 	dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
893 
894 	qedr_pci_set_atomic(dev, pdev);
895 
896 	rc = qedr_alloc_resources(dev);
897 	if (rc)
898 		goto init_err;
899 
900 	rc = qedr_init_hw(dev);
901 	if (rc)
902 		goto alloc_err;
903 
904 	rc = qedr_setup_irqs(dev);
905 	if (rc)
906 		goto irq_err;
907 
908 	rc = qedr_register_device(dev);
909 	if (rc) {
910 		DP_ERR(dev, "Unable to allocate register device\n");
911 		goto reg_err;
912 	}
913 
914 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
915 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
916 
917 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
918 	return dev;
919 
920 reg_err:
921 	qedr_sync_free_irqs(dev);
922 irq_err:
923 	qedr_stop_hw(dev);
924 alloc_err:
925 	qedr_free_resources(dev);
926 init_err:
927 	ib_dealloc_device(&dev->ibdev);
928 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
929 
930 	return NULL;
931 }
932 
933 static void qedr_remove(struct qedr_dev *dev)
934 {
935 	/* First unregister with stack to stop all the active traffic
936 	 * of the registered clients.
937 	 */
938 	ib_unregister_device(&dev->ibdev);
939 
940 	qedr_stop_hw(dev);
941 	qedr_sync_free_irqs(dev);
942 	qedr_free_resources(dev);
943 
944 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev))
945 		dev->ops->iwarp_set_engine_affin(dev->cdev, true);
946 
947 	ib_dealloc_device(&dev->ibdev);
948 }
949 
950 static void qedr_close(struct qedr_dev *dev)
951 {
952 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
953 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
954 }
955 
956 static void qedr_shutdown(struct qedr_dev *dev)
957 {
958 	qedr_close(dev);
959 	qedr_remove(dev);
960 }
961 
962 static void qedr_open(struct qedr_dev *dev)
963 {
964 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
965 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
966 }
967 
968 static void qedr_mac_address_change(struct qedr_dev *dev)
969 {
970 	union ib_gid *sgid = &dev->sgid_tbl[0];
971 	u8 guid[8], mac_addr[6];
972 	int rc;
973 
974 	/* Update SGID */
975 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
976 	guid[0] = mac_addr[0] ^ 2;
977 	guid[1] = mac_addr[1];
978 	guid[2] = mac_addr[2];
979 	guid[3] = 0xff;
980 	guid[4] = 0xfe;
981 	guid[5] = mac_addr[3];
982 	guid[6] = mac_addr[4];
983 	guid[7] = mac_addr[5];
984 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
985 	memcpy(&sgid->raw[8], guid, sizeof(guid));
986 
987 	/* Update LL2 */
988 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
989 					  dev->gsi_ll2_mac_address,
990 					  dev->ndev->dev_addr);
991 
992 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
993 
994 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
995 
996 	if (rc)
997 		DP_ERR(dev, "Error updating mac filter\n");
998 }
999 
1000 /* event handling via NIC driver ensures that all the NIC specific
1001  * initialization done before RoCE driver notifies
1002  * event to stack.
1003  */
1004 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1005 {
1006 	switch (event) {
1007 	case QEDE_UP:
1008 		qedr_open(dev);
1009 		break;
1010 	case QEDE_DOWN:
1011 		qedr_close(dev);
1012 		break;
1013 	case QEDE_CLOSE:
1014 		qedr_shutdown(dev);
1015 		break;
1016 	case QEDE_CHANGE_ADDR:
1017 		qedr_mac_address_change(dev);
1018 		break;
1019 	case QEDE_CHANGE_MTU:
1020 		if (rdma_protocol_iwarp(&dev->ibdev, 1))
1021 			if (dev->ndev->mtu != dev->iwarp_max_mtu)
1022 				DP_NOTICE(dev,
1023 					  "Mtu was changed from %d to %d. This will not take affect for iWARP until qedr is reloaded\n",
1024 					  dev->iwarp_max_mtu, dev->ndev->mtu);
1025 		break;
1026 	default:
1027 		pr_err("Event not supported\n");
1028 	}
1029 }
1030 
1031 static struct qedr_driver qedr_drv = {
1032 	.name = "qedr_driver",
1033 	.add = qedr_add,
1034 	.remove = qedr_remove,
1035 	.notify = qedr_notify,
1036 };
1037 
1038 static int __init qedr_init_module(void)
1039 {
1040 	return qede_rdma_register_driver(&qedr_drv);
1041 }
1042 
1043 static void __exit qedr_exit_module(void)
1044 {
1045 	qede_rdma_unregister_driver(&qedr_drv);
1046 }
1047 
1048 module_init(qedr_init_module);
1049 module_exit(qedr_exit_module);
1050