xref: /openbmc/linux/drivers/infiniband/hw/qedr/main.c (revision e15a5365)
1 /* QLogic qedr NIC Driver
2  * Copyright (c) 2015-2016  QLogic Corporation
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and /or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <rdma/ib_verbs.h>
34 #include <rdma/ib_addr.h>
35 #include <rdma/ib_user_verbs.h>
36 #include <rdma/iw_cm.h>
37 #include <rdma/ib_mad.h>
38 #include <linux/netdevice.h>
39 #include <linux/iommu.h>
40 #include <linux/pci.h>
41 #include <net/addrconf.h>
42 
43 #include <linux/qed/qed_chain.h>
44 #include <linux/qed/qed_if.h>
45 #include "qedr.h"
46 #include "verbs.h"
47 #include <rdma/qedr-abi.h>
48 #include "qedr_iw_cm.h"
49 
50 MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
51 MODULE_AUTHOR("QLogic Corporation");
52 MODULE_LICENSE("Dual BSD/GPL");
53 
54 #define QEDR_WQ_MULTIPLIER_DFT	(3)
55 
56 static void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
57 				   enum ib_event_type type)
58 {
59 	struct ib_event ibev;
60 
61 	ibev.device = &dev->ibdev;
62 	ibev.element.port_num = port_num;
63 	ibev.event = type;
64 
65 	ib_dispatch_event(&ibev);
66 }
67 
68 static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
69 					    u8 port_num)
70 {
71 	return IB_LINK_LAYER_ETHERNET;
72 }
73 
74 static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str)
75 {
76 	struct qedr_dev *qedr = get_qedr_dev(ibdev);
77 	u32 fw_ver = (u32)qedr->attr.fw_ver;
78 
79 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d.%d",
80 		 (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF,
81 		 (fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
82 }
83 
84 static int qedr_roce_port_immutable(struct ib_device *ibdev, u8 port_num,
85 				    struct ib_port_immutable *immutable)
86 {
87 	struct ib_port_attr attr;
88 	int err;
89 
90 	err = qedr_query_port(ibdev, port_num, &attr);
91 	if (err)
92 		return err;
93 
94 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
95 	immutable->gid_tbl_len = attr.gid_tbl_len;
96 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
97 	    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
98 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
99 
100 	return 0;
101 }
102 
103 static int qedr_iw_port_immutable(struct ib_device *ibdev, u8 port_num,
104 				  struct ib_port_immutable *immutable)
105 {
106 	struct ib_port_attr attr;
107 	int err;
108 
109 	err = qedr_query_port(ibdev, port_num, &attr);
110 	if (err)
111 		return err;
112 
113 	immutable->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 scnprintf(buf, PAGE_SIZE, "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 scnprintf(buf, PAGE_SIZE, "FastLinQ QL%x %s\n",
138 			 dev->pdev->device,
139 			 rdma_protocol_iwarp(&dev->ibdev, 1) ?
140 			 "iWARP" : "RoCE");
141 }
142 static DEVICE_ATTR_RO(hca_type);
143 
144 static struct attribute *qedr_attributes[] = {
145 	&dev_attr_hw_rev.attr,
146 	&dev_attr_hca_type.attr,
147 	NULL
148 };
149 
150 static const struct attribute_group qedr_attr_group = {
151 	.attrs = qedr_attributes,
152 };
153 
154 static const struct ib_device_ops qedr_iw_dev_ops = {
155 	.get_port_immutable = qedr_iw_port_immutable,
156 	.iw_accept = qedr_iw_accept,
157 	.iw_add_ref = qedr_iw_qp_add_ref,
158 	.iw_connect = qedr_iw_connect,
159 	.iw_create_listen = qedr_iw_create_listen,
160 	.iw_destroy_listen = qedr_iw_destroy_listen,
161 	.iw_get_qp = qedr_iw_get_qp,
162 	.iw_reject = qedr_iw_reject,
163 	.iw_rem_ref = qedr_iw_qp_rem_ref,
164 	.query_gid = qedr_iw_query_gid,
165 };
166 
167 static int qedr_iw_register_device(struct qedr_dev *dev)
168 {
169 	dev->ibdev.node_type = RDMA_NODE_RNIC;
170 
171 	ib_set_device_ops(&dev->ibdev, &qedr_iw_dev_ops);
172 
173 	memcpy(dev->ibdev.iw_ifname,
174 	       dev->ndev->name, sizeof(dev->ibdev.iw_ifname));
175 
176 	return 0;
177 }
178 
179 static const struct ib_device_ops qedr_roce_dev_ops = {
180 	.alloc_xrcd = qedr_alloc_xrcd,
181 	.dealloc_xrcd = qedr_dealloc_xrcd,
182 	.get_port_immutable = qedr_roce_port_immutable,
183 	.query_pkey = qedr_query_pkey,
184 };
185 
186 static void qedr_roce_register_device(struct qedr_dev *dev)
187 {
188 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
189 
190 	ib_set_device_ops(&dev->ibdev, &qedr_roce_dev_ops);
191 
192 	dev->ibdev.uverbs_cmd_mask |= QEDR_UVERBS(OPEN_XRCD) |
193 		QEDR_UVERBS(CLOSE_XRCD) |
194 		QEDR_UVERBS(CREATE_XSRQ);
195 }
196 
197 static const struct ib_device_ops qedr_dev_ops = {
198 	.owner = THIS_MODULE,
199 	.driver_id = RDMA_DRIVER_QEDR,
200 	.uverbs_abi_ver = QEDR_ABI_VERSION,
201 
202 	.alloc_mr = qedr_alloc_mr,
203 	.alloc_pd = qedr_alloc_pd,
204 	.alloc_ucontext = qedr_alloc_ucontext,
205 	.create_ah = qedr_create_ah,
206 	.create_cq = qedr_create_cq,
207 	.create_qp = qedr_create_qp,
208 	.create_srq = qedr_create_srq,
209 	.dealloc_pd = qedr_dealloc_pd,
210 	.dealloc_ucontext = qedr_dealloc_ucontext,
211 	.dereg_mr = qedr_dereg_mr,
212 	.destroy_ah = qedr_destroy_ah,
213 	.destroy_cq = qedr_destroy_cq,
214 	.destroy_qp = qedr_destroy_qp,
215 	.destroy_srq = qedr_destroy_srq,
216 	.get_dev_fw_str = qedr_get_dev_fw_str,
217 	.get_dma_mr = qedr_get_dma_mr,
218 	.get_link_layer = qedr_link_layer,
219 	.map_mr_sg = qedr_map_mr_sg,
220 	.mmap = qedr_mmap,
221 	.mmap_free = qedr_mmap_free,
222 	.modify_qp = qedr_modify_qp,
223 	.modify_srq = qedr_modify_srq,
224 	.poll_cq = qedr_poll_cq,
225 	.post_recv = qedr_post_recv,
226 	.post_send = qedr_post_send,
227 	.post_srq_recv = qedr_post_srq_recv,
228 	.process_mad = qedr_process_mad,
229 	.query_device = qedr_query_device,
230 	.query_port = qedr_query_port,
231 	.query_qp = qedr_query_qp,
232 	.query_srq = qedr_query_srq,
233 	.reg_user_mr = qedr_reg_user_mr,
234 	.req_notify_cq = qedr_arm_cq,
235 	.resize_cq = qedr_resize_cq,
236 
237 	INIT_RDMA_OBJ_SIZE(ib_ah, qedr_ah, ibah),
238 	INIT_RDMA_OBJ_SIZE(ib_cq, qedr_cq, ibcq),
239 	INIT_RDMA_OBJ_SIZE(ib_pd, qedr_pd, ibpd),
240 	INIT_RDMA_OBJ_SIZE(ib_srq, qedr_srq, ibsrq),
241 	INIT_RDMA_OBJ_SIZE(ib_xrcd, qedr_xrcd, ibxrcd),
242 	INIT_RDMA_OBJ_SIZE(ib_ucontext, qedr_ucontext, ibucontext),
243 };
244 
245 static int qedr_register_device(struct qedr_dev *dev)
246 {
247 	int rc;
248 
249 	dev->ibdev.node_guid = dev->attr.node_guid;
250 	memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
251 
252 	dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
253 				     QEDR_UVERBS(QUERY_DEVICE) |
254 				     QEDR_UVERBS(QUERY_PORT) |
255 				     QEDR_UVERBS(ALLOC_PD) |
256 				     QEDR_UVERBS(DEALLOC_PD) |
257 				     QEDR_UVERBS(CREATE_COMP_CHANNEL) |
258 				     QEDR_UVERBS(CREATE_CQ) |
259 				     QEDR_UVERBS(RESIZE_CQ) |
260 				     QEDR_UVERBS(DESTROY_CQ) |
261 				     QEDR_UVERBS(REQ_NOTIFY_CQ) |
262 				     QEDR_UVERBS(CREATE_QP) |
263 				     QEDR_UVERBS(MODIFY_QP) |
264 				     QEDR_UVERBS(QUERY_QP) |
265 				     QEDR_UVERBS(DESTROY_QP) |
266 				     QEDR_UVERBS(CREATE_SRQ) |
267 				     QEDR_UVERBS(DESTROY_SRQ) |
268 				     QEDR_UVERBS(QUERY_SRQ) |
269 				     QEDR_UVERBS(MODIFY_SRQ) |
270 				     QEDR_UVERBS(POST_SRQ_RECV) |
271 				     QEDR_UVERBS(REG_MR) |
272 				     QEDR_UVERBS(DEREG_MR) |
273 				     QEDR_UVERBS(POLL_CQ) |
274 				     QEDR_UVERBS(POST_SEND) |
275 				     QEDR_UVERBS(POST_RECV);
276 
277 	if (IS_IWARP(dev)) {
278 		rc = qedr_iw_register_device(dev);
279 		if (rc)
280 			return rc;
281 	} else {
282 		qedr_roce_register_device(dev);
283 	}
284 
285 	dev->ibdev.phys_port_cnt = 1;
286 	dev->ibdev.num_comp_vectors = dev->num_cnq;
287 	dev->ibdev.dev.parent = &dev->pdev->dev;
288 
289 	rdma_set_device_sysfs_group(&dev->ibdev, &qedr_attr_group);
290 	ib_set_device_ops(&dev->ibdev, &qedr_dev_ops);
291 
292 	rc = ib_device_set_netdev(&dev->ibdev, dev->ndev, 1);
293 	if (rc)
294 		return rc;
295 
296 	dma_set_max_seg_size(&dev->pdev->dev, UINT_MAX);
297 	return ib_register_device(&dev->ibdev, "qedr%d", &dev->pdev->dev);
298 }
299 
300 /* This function allocates fast-path status block memory */
301 static int qedr_alloc_mem_sb(struct qedr_dev *dev,
302 			     struct qed_sb_info *sb_info, u16 sb_id)
303 {
304 	struct status_block_e4 *sb_virt;
305 	dma_addr_t sb_phys;
306 	int rc;
307 
308 	sb_virt = dma_alloc_coherent(&dev->pdev->dev,
309 				     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
310 	if (!sb_virt)
311 		return -ENOMEM;
312 
313 	rc = dev->ops->common->sb_init(dev->cdev, sb_info,
314 				       sb_virt, sb_phys, sb_id,
315 				       QED_SB_TYPE_CNQ);
316 	if (rc) {
317 		pr_err("Status block initialization failed\n");
318 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
319 				  sb_virt, sb_phys);
320 		return rc;
321 	}
322 
323 	return 0;
324 }
325 
326 static void qedr_free_mem_sb(struct qedr_dev *dev,
327 			     struct qed_sb_info *sb_info, int sb_id)
328 {
329 	if (sb_info->sb_virt) {
330 		dev->ops->common->sb_release(dev->cdev, sb_info, sb_id,
331 					     QED_SB_TYPE_CNQ);
332 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
333 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
334 	}
335 }
336 
337 static void qedr_free_resources(struct qedr_dev *dev)
338 {
339 	int i;
340 
341 	if (IS_IWARP(dev))
342 		destroy_workqueue(dev->iwarp_wq);
343 
344 	for (i = 0; i < dev->num_cnq; i++) {
345 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
346 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
347 	}
348 
349 	kfree(dev->cnq_array);
350 	kfree(dev->sb_array);
351 	kfree(dev->sgid_tbl);
352 }
353 
354 static int qedr_alloc_resources(struct qedr_dev *dev)
355 {
356 	struct qed_chain_init_params params = {
357 		.mode		= QED_CHAIN_MODE_PBL,
358 		.intended_use	= QED_CHAIN_USE_TO_CONSUME,
359 		.cnt_type	= QED_CHAIN_CNT_TYPE_U16,
360 		.elem_size	= sizeof(struct regpair *),
361 	};
362 	struct qedr_cnq *cnq;
363 	__le16 *cons_pi;
364 	int i, rc;
365 
366 	dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
367 				GFP_KERNEL);
368 	if (!dev->sgid_tbl)
369 		return -ENOMEM;
370 
371 	spin_lock_init(&dev->sgid_lock);
372 	xa_init_flags(&dev->srqs, XA_FLAGS_LOCK_IRQ);
373 
374 	if (IS_IWARP(dev)) {
375 		xa_init(&dev->qps);
376 		dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
377 	}
378 
379 	/* Allocate Status blocks for CNQ */
380 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
381 				GFP_KERNEL);
382 	if (!dev->sb_array) {
383 		rc = -ENOMEM;
384 		goto err1;
385 	}
386 
387 	dev->cnq_array = kcalloc(dev->num_cnq,
388 				 sizeof(*dev->cnq_array), GFP_KERNEL);
389 	if (!dev->cnq_array) {
390 		rc = -ENOMEM;
391 		goto err2;
392 	}
393 
394 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
395 
396 	/* Allocate CNQ PBLs */
397 	params.num_elems = min_t(u32, QED_RDMA_MAX_CNQ_SIZE,
398 				 QEDR_ROCE_MAX_CNQ_SIZE);
399 
400 	for (i = 0; i < dev->num_cnq; i++) {
401 		cnq = &dev->cnq_array[i];
402 
403 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
404 				       dev->sb_start + i);
405 		if (rc)
406 			goto err3;
407 
408 		rc = dev->ops->common->chain_alloc(dev->cdev, &cnq->pbl,
409 						   &params);
410 		if (rc)
411 			goto err4;
412 
413 		cnq->dev = dev;
414 		cnq->sb = &dev->sb_array[i];
415 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
416 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
417 		cnq->index = i;
418 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
419 
420 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
421 			 i, qed_chain_get_cons_idx(&cnq->pbl));
422 	}
423 
424 	return 0;
425 err4:
426 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
427 err3:
428 	for (--i; i >= 0; i--) {
429 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
430 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
431 	}
432 	kfree(dev->cnq_array);
433 err2:
434 	kfree(dev->sb_array);
435 err1:
436 	kfree(dev->sgid_tbl);
437 	return rc;
438 }
439 
440 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
441 {
442 	int rc = pci_enable_atomic_ops_to_root(pdev,
443 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
444 
445 	if (rc) {
446 		dev->atomic_cap = IB_ATOMIC_NONE;
447 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
448 	} else {
449 		dev->atomic_cap = IB_ATOMIC_GLOB;
450 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
451 	}
452 }
453 
454 static const struct qed_rdma_ops *qed_ops;
455 
456 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
457 
458 static irqreturn_t qedr_irq_handler(int irq, void *handle)
459 {
460 	u16 hw_comp_cons, sw_comp_cons;
461 	struct qedr_cnq *cnq = handle;
462 	struct regpair *cq_handle;
463 	struct qedr_cq *cq;
464 
465 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
466 
467 	qed_sb_update_sb_idx(cnq->sb);
468 
469 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
470 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
471 
472 	/* Align protocol-index and chain reads */
473 	rmb();
474 
475 	while (sw_comp_cons != hw_comp_cons) {
476 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
477 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
478 				cq_handle->lo);
479 
480 		if (cq == NULL) {
481 			DP_ERR(cnq->dev,
482 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
483 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
484 			       hw_comp_cons);
485 
486 			break;
487 		}
488 
489 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
490 			DP_ERR(cnq->dev,
491 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
492 			       cq_handle->hi, cq_handle->lo, cq);
493 			break;
494 		}
495 
496 		cq->arm_flags = 0;
497 
498 		if (!cq->destroyed && cq->ibcq.comp_handler)
499 			(*cq->ibcq.comp_handler)
500 				(&cq->ibcq, cq->ibcq.cq_context);
501 
502 		/* The CQ's CNQ notification counter is checked before
503 		 * destroying the CQ in a busy-wait loop that waits for all of
504 		 * the CQ's CNQ interrupts to be processed. It is increased
505 		 * here, only after the completion handler, to ensure that the
506 		 * the handler is not running when the CQ is destroyed.
507 		 */
508 		cq->cnq_notif++;
509 
510 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
511 
512 		cnq->n_comp++;
513 	}
514 
515 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
516 				      sw_comp_cons);
517 
518 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
519 
520 	return IRQ_HANDLED;
521 }
522 
523 static void qedr_sync_free_irqs(struct qedr_dev *dev)
524 {
525 	u32 vector;
526 	u16 idx;
527 	int i;
528 
529 	for (i = 0; i < dev->int_info.used_cnt; i++) {
530 		if (dev->int_info.msix_cnt) {
531 			idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
532 			vector = dev->int_info.msix[idx].vector;
533 			synchronize_irq(vector);
534 			free_irq(vector, &dev->cnq_array[i]);
535 		}
536 	}
537 
538 	dev->int_info.used_cnt = 0;
539 }
540 
541 static int qedr_req_msix_irqs(struct qedr_dev *dev)
542 {
543 	int i, rc = 0;
544 	u16 idx;
545 
546 	if (dev->num_cnq > dev->int_info.msix_cnt) {
547 		DP_ERR(dev,
548 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
549 		       dev->num_cnq, dev->int_info.msix_cnt);
550 		return -EINVAL;
551 	}
552 
553 	for (i = 0; i < dev->num_cnq; i++) {
554 		idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
555 		rc = request_irq(dev->int_info.msix[idx].vector,
556 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
557 				 &dev->cnq_array[i]);
558 		if (rc) {
559 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
560 			qedr_sync_free_irqs(dev);
561 		} else {
562 			DP_DEBUG(dev, QEDR_MSG_INIT,
563 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
564 				 dev->cnq_array[i].name, i,
565 				 &dev->cnq_array[i]);
566 			dev->int_info.used_cnt++;
567 		}
568 	}
569 
570 	return rc;
571 }
572 
573 static int qedr_setup_irqs(struct qedr_dev *dev)
574 {
575 	int rc;
576 
577 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
578 
579 	/* Learn Interrupt configuration */
580 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
581 	if (rc < 0)
582 		return rc;
583 
584 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
585 	if (rc) {
586 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
587 		return rc;
588 	}
589 
590 	if (dev->int_info.msix_cnt) {
591 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
592 			 dev->int_info.msix_cnt);
593 		rc = qedr_req_msix_irqs(dev);
594 		if (rc)
595 			return rc;
596 	}
597 
598 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
599 
600 	return 0;
601 }
602 
603 static int qedr_set_device_attr(struct qedr_dev *dev)
604 {
605 	struct qed_rdma_device *qed_attr;
606 	struct qedr_device_attr *attr;
607 	u32 page_size;
608 
609 	/* Part 1 - query core capabilities */
610 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
611 
612 	/* Part 2 - check capabilities */
613 	page_size = ~qed_attr->page_size_caps + 1;
614 	if (page_size > PAGE_SIZE) {
615 		DP_ERR(dev,
616 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
617 		       PAGE_SIZE, page_size);
618 		return -ENODEV;
619 	}
620 
621 	/* Part 3 - copy and update capabilities */
622 	attr = &dev->attr;
623 	attr->vendor_id = qed_attr->vendor_id;
624 	attr->vendor_part_id = qed_attr->vendor_part_id;
625 	attr->hw_ver = qed_attr->hw_ver;
626 	attr->fw_ver = qed_attr->fw_ver;
627 	attr->node_guid = qed_attr->node_guid;
628 	attr->sys_image_guid = qed_attr->sys_image_guid;
629 	attr->max_cnq = qed_attr->max_cnq;
630 	attr->max_sge = qed_attr->max_sge;
631 	attr->max_inline = qed_attr->max_inline;
632 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
633 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
634 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
635 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
636 	attr->max_dev_resp_rd_atomic_resc =
637 	    qed_attr->max_dev_resp_rd_atomic_resc;
638 	attr->max_cq = qed_attr->max_cq;
639 	attr->max_qp = qed_attr->max_qp;
640 	attr->max_mr = qed_attr->max_mr;
641 	attr->max_mr_size = qed_attr->max_mr_size;
642 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
643 	attr->max_mw = qed_attr->max_mw;
644 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
645 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
646 	attr->max_pd = qed_attr->max_pd;
647 	attr->max_ah = qed_attr->max_ah;
648 	attr->max_pkey = qed_attr->max_pkey;
649 	attr->max_srq = qed_attr->max_srq;
650 	attr->max_srq_wr = qed_attr->max_srq_wr;
651 	attr->dev_caps = qed_attr->dev_caps;
652 	attr->page_size_caps = qed_attr->page_size_caps;
653 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
654 	attr->reserved_lkey = qed_attr->reserved_lkey;
655 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
656 	attr->max_stats_queues = qed_attr->max_stats_queues;
657 
658 	return 0;
659 }
660 
661 static void qedr_unaffiliated_event(void *context, u8 event_code)
662 {
663 	pr_err("unaffiliated event not implemented yet\n");
664 }
665 
666 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
667 {
668 #define EVENT_TYPE_NOT_DEFINED	0
669 #define EVENT_TYPE_CQ		1
670 #define EVENT_TYPE_QP		2
671 #define EVENT_TYPE_SRQ		3
672 	struct qedr_dev *dev = (struct qedr_dev *)context;
673 	struct regpair *async_handle = (struct regpair *)fw_handle;
674 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
675 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
676 	struct ib_event event;
677 	struct ib_srq *ibsrq;
678 	struct qedr_srq *srq;
679 	unsigned long flags;
680 	struct ib_cq *ibcq;
681 	struct ib_qp *ibqp;
682 	struct qedr_cq *cq;
683 	struct qedr_qp *qp;
684 	u16 srq_id;
685 
686 	if (IS_ROCE(dev)) {
687 		switch (e_code) {
688 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
689 			event.event = IB_EVENT_CQ_ERR;
690 			event_type = EVENT_TYPE_CQ;
691 			break;
692 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
693 			event.event = IB_EVENT_SQ_DRAINED;
694 			event_type = EVENT_TYPE_QP;
695 			break;
696 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
697 			event.event = IB_EVENT_QP_FATAL;
698 			event_type = EVENT_TYPE_QP;
699 			break;
700 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
701 			event.event = IB_EVENT_QP_REQ_ERR;
702 			event_type = EVENT_TYPE_QP;
703 			break;
704 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
705 			event.event = IB_EVENT_QP_ACCESS_ERR;
706 			event_type = EVENT_TYPE_QP;
707 			break;
708 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
709 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
710 			event_type = EVENT_TYPE_SRQ;
711 			break;
712 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
713 			event.event = IB_EVENT_SRQ_ERR;
714 			event_type = EVENT_TYPE_SRQ;
715 			break;
716 		case ROCE_ASYNC_EVENT_XRC_DOMAIN_ERR:
717 			event.event = IB_EVENT_QP_ACCESS_ERR;
718 			event_type = EVENT_TYPE_QP;
719 			break;
720 		case ROCE_ASYNC_EVENT_INVALID_XRCETH_ERR:
721 			event.event = IB_EVENT_QP_ACCESS_ERR;
722 			event_type = EVENT_TYPE_QP;
723 			break;
724 		case ROCE_ASYNC_EVENT_XRC_SRQ_CATASTROPHIC_ERR:
725 			event.event = IB_EVENT_CQ_ERR;
726 			event_type = EVENT_TYPE_CQ;
727 			break;
728 		default:
729 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
730 			       e_code, roce_handle64);
731 		}
732 	} else {
733 		switch (e_code) {
734 		case QED_IWARP_EVENT_SRQ_LIMIT:
735 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
736 			event_type = EVENT_TYPE_SRQ;
737 			break;
738 		case QED_IWARP_EVENT_SRQ_EMPTY:
739 			event.event = IB_EVENT_SRQ_ERR;
740 			event_type = EVENT_TYPE_SRQ;
741 			break;
742 		default:
743 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
744 		       roce_handle64);
745 		}
746 	}
747 	switch (event_type) {
748 	case EVENT_TYPE_CQ:
749 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
750 		if (cq) {
751 			ibcq = &cq->ibcq;
752 			if (ibcq->event_handler) {
753 				event.device = ibcq->device;
754 				event.element.cq = ibcq;
755 				ibcq->event_handler(&event, ibcq->cq_context);
756 			}
757 		} else {
758 			WARN(1,
759 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
760 			     roce_handle64);
761 		}
762 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
763 		break;
764 	case EVENT_TYPE_QP:
765 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
766 		if (qp) {
767 			ibqp = &qp->ibqp;
768 			if (ibqp->event_handler) {
769 				event.device = ibqp->device;
770 				event.element.qp = ibqp;
771 				ibqp->event_handler(&event, ibqp->qp_context);
772 			}
773 		} else {
774 			WARN(1,
775 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
776 			     roce_handle64);
777 		}
778 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
779 		break;
780 	case EVENT_TYPE_SRQ:
781 		srq_id = (u16)roce_handle64;
782 		xa_lock_irqsave(&dev->srqs, flags);
783 		srq = xa_load(&dev->srqs, srq_id);
784 		if (srq) {
785 			ibsrq = &srq->ibsrq;
786 			if (ibsrq->event_handler) {
787 				event.device = ibsrq->device;
788 				event.element.srq = ibsrq;
789 				ibsrq->event_handler(&event,
790 						     ibsrq->srq_context);
791 			}
792 		} else {
793 			DP_NOTICE(dev,
794 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
795 				  roce_handle64);
796 		}
797 		xa_unlock_irqrestore(&dev->srqs, flags);
798 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
799 	default:
800 		break;
801 	}
802 }
803 
804 static int qedr_init_hw(struct qedr_dev *dev)
805 {
806 	struct qed_rdma_add_user_out_params out_params;
807 	struct qed_rdma_start_in_params *in_params;
808 	struct qed_rdma_cnq_params *cur_pbl;
809 	struct qed_rdma_events events;
810 	dma_addr_t p_phys_table;
811 	u32 page_cnt;
812 	int rc = 0;
813 	int i;
814 
815 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
816 	if (!in_params) {
817 		rc = -ENOMEM;
818 		goto out;
819 	}
820 
821 	in_params->desired_cnq = dev->num_cnq;
822 	for (i = 0; i < dev->num_cnq; i++) {
823 		cur_pbl = &in_params->cnq_pbl_list[i];
824 
825 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
826 		cur_pbl->num_pbl_pages = page_cnt;
827 
828 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
829 		cur_pbl->pbl_ptr = (u64)p_phys_table;
830 	}
831 
832 	events.affiliated_event = qedr_affiliated_event;
833 	events.unaffiliated_event = qedr_unaffiliated_event;
834 	events.context = dev;
835 
836 	in_params->events = &events;
837 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
838 	in_params->max_mtu = dev->ndev->mtu;
839 	dev->iwarp_max_mtu = dev->ndev->mtu;
840 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
841 
842 	rc = dev->ops->rdma_init(dev->cdev, in_params);
843 	if (rc)
844 		goto out;
845 
846 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
847 	if (rc)
848 		goto out;
849 
850 	dev->db_addr = out_params.dpi_addr;
851 	dev->db_phys_addr = out_params.dpi_phys_addr;
852 	dev->db_size = out_params.dpi_size;
853 	dev->dpi = out_params.dpi;
854 
855 	rc = qedr_set_device_attr(dev);
856 out:
857 	kfree(in_params);
858 	if (rc)
859 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
860 
861 	return rc;
862 }
863 
864 static void qedr_stop_hw(struct qedr_dev *dev)
865 {
866 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
867 	dev->ops->rdma_stop(dev->rdma_ctx);
868 }
869 
870 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
871 				 struct net_device *ndev)
872 {
873 	struct qed_dev_rdma_info dev_info;
874 	struct qedr_dev *dev;
875 	int rc = 0;
876 
877 	dev = ib_alloc_device(qedr_dev, ibdev);
878 	if (!dev) {
879 		pr_err("Unable to allocate ib device\n");
880 		return NULL;
881 	}
882 
883 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
884 
885 	dev->pdev = pdev;
886 	dev->ndev = ndev;
887 	dev->cdev = cdev;
888 
889 	qed_ops = qed_get_rdma_ops();
890 	if (!qed_ops) {
891 		DP_ERR(dev, "Failed to get qed roce operations\n");
892 		goto init_err;
893 	}
894 
895 	dev->ops = qed_ops;
896 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
897 	if (rc)
898 		goto init_err;
899 
900 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
901 	dev->rdma_type = dev_info.rdma_type;
902 	dev->num_hwfns = dev_info.common.num_hwfns;
903 
904 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev)) {
905 		rc = dev->ops->iwarp_set_engine_affin(cdev, false);
906 		if (rc) {
907 			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");
908 			goto init_err;
909 		}
910 	}
911 	dev->affin_hwfn_idx = dev->ops->common->get_affin_hwfn_idx(cdev);
912 
913 	dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
914 
915 	dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
916 	if (!dev->num_cnq) {
917 		DP_ERR(dev, "Failed. At least one CNQ is required.\n");
918 		rc = -ENOMEM;
919 		goto init_err;
920 	}
921 
922 	dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
923 
924 	qedr_pci_set_atomic(dev, pdev);
925 
926 	rc = qedr_alloc_resources(dev);
927 	if (rc)
928 		goto init_err;
929 
930 	rc = qedr_init_hw(dev);
931 	if (rc)
932 		goto alloc_err;
933 
934 	rc = qedr_setup_irqs(dev);
935 	if (rc)
936 		goto irq_err;
937 
938 	rc = qedr_register_device(dev);
939 	if (rc) {
940 		DP_ERR(dev, "Unable to allocate register device\n");
941 		goto reg_err;
942 	}
943 
944 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
945 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
946 
947 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
948 	return dev;
949 
950 reg_err:
951 	qedr_sync_free_irqs(dev);
952 irq_err:
953 	qedr_stop_hw(dev);
954 alloc_err:
955 	qedr_free_resources(dev);
956 init_err:
957 	ib_dealloc_device(&dev->ibdev);
958 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
959 
960 	return NULL;
961 }
962 
963 static void qedr_remove(struct qedr_dev *dev)
964 {
965 	/* First unregister with stack to stop all the active traffic
966 	 * of the registered clients.
967 	 */
968 	ib_unregister_device(&dev->ibdev);
969 
970 	qedr_stop_hw(dev);
971 	qedr_sync_free_irqs(dev);
972 	qedr_free_resources(dev);
973 
974 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev))
975 		dev->ops->iwarp_set_engine_affin(dev->cdev, true);
976 
977 	ib_dealloc_device(&dev->ibdev);
978 }
979 
980 static void qedr_close(struct qedr_dev *dev)
981 {
982 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
983 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
984 }
985 
986 static void qedr_shutdown(struct qedr_dev *dev)
987 {
988 	qedr_close(dev);
989 	qedr_remove(dev);
990 }
991 
992 static void qedr_open(struct qedr_dev *dev)
993 {
994 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
995 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
996 }
997 
998 static void qedr_mac_address_change(struct qedr_dev *dev)
999 {
1000 	union ib_gid *sgid = &dev->sgid_tbl[0];
1001 	u8 guid[8], mac_addr[6];
1002 	int rc;
1003 
1004 	/* Update SGID */
1005 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
1006 	guid[0] = mac_addr[0] ^ 2;
1007 	guid[1] = mac_addr[1];
1008 	guid[2] = mac_addr[2];
1009 	guid[3] = 0xff;
1010 	guid[4] = 0xfe;
1011 	guid[5] = mac_addr[3];
1012 	guid[6] = mac_addr[4];
1013 	guid[7] = mac_addr[5];
1014 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1015 	memcpy(&sgid->raw[8], guid, sizeof(guid));
1016 
1017 	/* Update LL2 */
1018 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
1019 					  dev->gsi_ll2_mac_address,
1020 					  dev->ndev->dev_addr);
1021 
1022 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
1023 
1024 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
1025 
1026 	if (rc)
1027 		DP_ERR(dev, "Error updating mac filter\n");
1028 }
1029 
1030 /* event handling via NIC driver ensures that all the NIC specific
1031  * initialization done before RoCE driver notifies
1032  * event to stack.
1033  */
1034 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1035 {
1036 	switch (event) {
1037 	case QEDE_UP:
1038 		qedr_open(dev);
1039 		break;
1040 	case QEDE_DOWN:
1041 		qedr_close(dev);
1042 		break;
1043 	case QEDE_CLOSE:
1044 		qedr_shutdown(dev);
1045 		break;
1046 	case QEDE_CHANGE_ADDR:
1047 		qedr_mac_address_change(dev);
1048 		break;
1049 	case QEDE_CHANGE_MTU:
1050 		if (rdma_protocol_iwarp(&dev->ibdev, 1))
1051 			if (dev->ndev->mtu != dev->iwarp_max_mtu)
1052 				DP_NOTICE(dev,
1053 					  "Mtu was changed from %d to %d. This will not take affect for iWARP until qedr is reloaded\n",
1054 					  dev->iwarp_max_mtu, dev->ndev->mtu);
1055 		break;
1056 	default:
1057 		pr_err("Event not supported\n");
1058 	}
1059 }
1060 
1061 static struct qedr_driver qedr_drv = {
1062 	.name = "qedr_driver",
1063 	.add = qedr_add,
1064 	.remove = qedr_remove,
1065 	.notify = qedr_notify,
1066 };
1067 
1068 static int __init qedr_init_module(void)
1069 {
1070 	return qede_rdma_register_driver(&qedr_drv);
1071 }
1072 
1073 static void __exit qedr_exit_module(void)
1074 {
1075 	qede_rdma_unregister_driver(&qedr_drv);
1076 }
1077 
1078 module_init(qedr_init_module);
1079 module_exit(qedr_exit_module);
1080