xref: /openbmc/linux/drivers/infiniband/hw/qedr/main.c (revision 3a83e4e6)
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 	.get_port_immutable = qedr_roce_port_immutable,
181 	.query_pkey = qedr_query_pkey,
182 };
183 
184 static void qedr_roce_register_device(struct qedr_dev *dev)
185 {
186 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
187 
188 	ib_set_device_ops(&dev->ibdev, &qedr_roce_dev_ops);
189 }
190 
191 static const struct ib_device_ops qedr_dev_ops = {
192 	.owner = THIS_MODULE,
193 	.driver_id = RDMA_DRIVER_QEDR,
194 	.uverbs_abi_ver = QEDR_ABI_VERSION,
195 
196 	.alloc_mr = qedr_alloc_mr,
197 	.alloc_pd = qedr_alloc_pd,
198 	.alloc_ucontext = qedr_alloc_ucontext,
199 	.create_ah = qedr_create_ah,
200 	.create_cq = qedr_create_cq,
201 	.create_qp = qedr_create_qp,
202 	.create_srq = qedr_create_srq,
203 	.dealloc_pd = qedr_dealloc_pd,
204 	.dealloc_ucontext = qedr_dealloc_ucontext,
205 	.dereg_mr = qedr_dereg_mr,
206 	.destroy_ah = qedr_destroy_ah,
207 	.destroy_cq = qedr_destroy_cq,
208 	.destroy_qp = qedr_destroy_qp,
209 	.destroy_srq = qedr_destroy_srq,
210 	.get_dev_fw_str = qedr_get_dev_fw_str,
211 	.get_dma_mr = qedr_get_dma_mr,
212 	.get_link_layer = qedr_link_layer,
213 	.map_mr_sg = qedr_map_mr_sg,
214 	.mmap = qedr_mmap,
215 	.mmap_free = qedr_mmap_free,
216 	.modify_qp = qedr_modify_qp,
217 	.modify_srq = qedr_modify_srq,
218 	.poll_cq = qedr_poll_cq,
219 	.post_recv = qedr_post_recv,
220 	.post_send = qedr_post_send,
221 	.post_srq_recv = qedr_post_srq_recv,
222 	.process_mad = qedr_process_mad,
223 	.query_device = qedr_query_device,
224 	.query_port = qedr_query_port,
225 	.query_qp = qedr_query_qp,
226 	.query_srq = qedr_query_srq,
227 	.reg_user_mr = qedr_reg_user_mr,
228 	.req_notify_cq = qedr_arm_cq,
229 	.resize_cq = qedr_resize_cq,
230 
231 	INIT_RDMA_OBJ_SIZE(ib_ah, qedr_ah, ibah),
232 	INIT_RDMA_OBJ_SIZE(ib_cq, qedr_cq, ibcq),
233 	INIT_RDMA_OBJ_SIZE(ib_pd, qedr_pd, ibpd),
234 	INIT_RDMA_OBJ_SIZE(ib_srq, qedr_srq, ibsrq),
235 	INIT_RDMA_OBJ_SIZE(ib_ucontext, qedr_ucontext, ibucontext),
236 };
237 
238 static int qedr_register_device(struct qedr_dev *dev)
239 {
240 	int rc;
241 
242 	dev->ibdev.node_guid = dev->attr.node_guid;
243 	memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
244 
245 	dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
246 				     QEDR_UVERBS(QUERY_DEVICE) |
247 				     QEDR_UVERBS(QUERY_PORT) |
248 				     QEDR_UVERBS(ALLOC_PD) |
249 				     QEDR_UVERBS(DEALLOC_PD) |
250 				     QEDR_UVERBS(CREATE_COMP_CHANNEL) |
251 				     QEDR_UVERBS(CREATE_CQ) |
252 				     QEDR_UVERBS(RESIZE_CQ) |
253 				     QEDR_UVERBS(DESTROY_CQ) |
254 				     QEDR_UVERBS(REQ_NOTIFY_CQ) |
255 				     QEDR_UVERBS(CREATE_QP) |
256 				     QEDR_UVERBS(MODIFY_QP) |
257 				     QEDR_UVERBS(QUERY_QP) |
258 				     QEDR_UVERBS(DESTROY_QP) |
259 				     QEDR_UVERBS(CREATE_SRQ) |
260 				     QEDR_UVERBS(DESTROY_SRQ) |
261 				     QEDR_UVERBS(QUERY_SRQ) |
262 				     QEDR_UVERBS(MODIFY_SRQ) |
263 				     QEDR_UVERBS(POST_SRQ_RECV) |
264 				     QEDR_UVERBS(REG_MR) |
265 				     QEDR_UVERBS(DEREG_MR) |
266 				     QEDR_UVERBS(POLL_CQ) |
267 				     QEDR_UVERBS(POST_SEND) |
268 				     QEDR_UVERBS(POST_RECV);
269 
270 	if (IS_IWARP(dev)) {
271 		rc = qedr_iw_register_device(dev);
272 		if (rc)
273 			return rc;
274 	} else {
275 		qedr_roce_register_device(dev);
276 	}
277 
278 	dev->ibdev.phys_port_cnt = 1;
279 	dev->ibdev.num_comp_vectors = dev->num_cnq;
280 	dev->ibdev.dev.parent = &dev->pdev->dev;
281 
282 	rdma_set_device_sysfs_group(&dev->ibdev, &qedr_attr_group);
283 	ib_set_device_ops(&dev->ibdev, &qedr_dev_ops);
284 
285 	rc = ib_device_set_netdev(&dev->ibdev, dev->ndev, 1);
286 	if (rc)
287 		return rc;
288 
289 	return ib_register_device(&dev->ibdev, "qedr%d");
290 }
291 
292 /* This function allocates fast-path status block memory */
293 static int qedr_alloc_mem_sb(struct qedr_dev *dev,
294 			     struct qed_sb_info *sb_info, u16 sb_id)
295 {
296 	struct status_block_e4 *sb_virt;
297 	dma_addr_t sb_phys;
298 	int rc;
299 
300 	sb_virt = dma_alloc_coherent(&dev->pdev->dev,
301 				     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
302 	if (!sb_virt)
303 		return -ENOMEM;
304 
305 	rc = dev->ops->common->sb_init(dev->cdev, sb_info,
306 				       sb_virt, sb_phys, sb_id,
307 				       QED_SB_TYPE_CNQ);
308 	if (rc) {
309 		pr_err("Status block initialization failed\n");
310 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
311 				  sb_virt, sb_phys);
312 		return rc;
313 	}
314 
315 	return 0;
316 }
317 
318 static void qedr_free_mem_sb(struct qedr_dev *dev,
319 			     struct qed_sb_info *sb_info, int sb_id)
320 {
321 	if (sb_info->sb_virt) {
322 		dev->ops->common->sb_release(dev->cdev, sb_info, sb_id,
323 					     QED_SB_TYPE_CNQ);
324 		dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
325 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
326 	}
327 }
328 
329 static void qedr_free_resources(struct qedr_dev *dev)
330 {
331 	int i;
332 
333 	if (IS_IWARP(dev))
334 		destroy_workqueue(dev->iwarp_wq);
335 
336 	for (i = 0; i < dev->num_cnq; i++) {
337 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
338 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
339 	}
340 
341 	kfree(dev->cnq_array);
342 	kfree(dev->sb_array);
343 	kfree(dev->sgid_tbl);
344 }
345 
346 static int qedr_alloc_resources(struct qedr_dev *dev)
347 {
348 	struct qed_chain_init_params params = {
349 		.mode		= QED_CHAIN_MODE_PBL,
350 		.intended_use	= QED_CHAIN_USE_TO_CONSUME,
351 		.cnt_type	= QED_CHAIN_CNT_TYPE_U16,
352 		.elem_size	= sizeof(struct regpair *),
353 	};
354 	struct qedr_cnq *cnq;
355 	__le16 *cons_pi;
356 	int i, rc;
357 
358 	dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
359 				GFP_KERNEL);
360 	if (!dev->sgid_tbl)
361 		return -ENOMEM;
362 
363 	spin_lock_init(&dev->sgid_lock);
364 	xa_init_flags(&dev->srqs, XA_FLAGS_LOCK_IRQ);
365 
366 	if (IS_IWARP(dev)) {
367 		xa_init(&dev->qps);
368 		dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
369 	}
370 
371 	/* Allocate Status blocks for CNQ */
372 	dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
373 				GFP_KERNEL);
374 	if (!dev->sb_array) {
375 		rc = -ENOMEM;
376 		goto err1;
377 	}
378 
379 	dev->cnq_array = kcalloc(dev->num_cnq,
380 				 sizeof(*dev->cnq_array), GFP_KERNEL);
381 	if (!dev->cnq_array) {
382 		rc = -ENOMEM;
383 		goto err2;
384 	}
385 
386 	dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
387 
388 	/* Allocate CNQ PBLs */
389 	params.num_elems = min_t(u32, QED_RDMA_MAX_CNQ_SIZE,
390 				 QEDR_ROCE_MAX_CNQ_SIZE);
391 
392 	for (i = 0; i < dev->num_cnq; i++) {
393 		cnq = &dev->cnq_array[i];
394 
395 		rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
396 				       dev->sb_start + i);
397 		if (rc)
398 			goto err3;
399 
400 		rc = dev->ops->common->chain_alloc(dev->cdev, &cnq->pbl,
401 						   &params);
402 		if (rc)
403 			goto err4;
404 
405 		cnq->dev = dev;
406 		cnq->sb = &dev->sb_array[i];
407 		cons_pi = dev->sb_array[i].sb_virt->pi_array;
408 		cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
409 		cnq->index = i;
410 		sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
411 
412 		DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
413 			 i, qed_chain_get_cons_idx(&cnq->pbl));
414 	}
415 
416 	return 0;
417 err4:
418 	qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
419 err3:
420 	for (--i; i >= 0; i--) {
421 		dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
422 		qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
423 	}
424 	kfree(dev->cnq_array);
425 err2:
426 	kfree(dev->sb_array);
427 err1:
428 	kfree(dev->sgid_tbl);
429 	return rc;
430 }
431 
432 static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
433 {
434 	int rc = pci_enable_atomic_ops_to_root(pdev,
435 					       PCI_EXP_DEVCAP2_ATOMIC_COMP64);
436 
437 	if (rc) {
438 		dev->atomic_cap = IB_ATOMIC_NONE;
439 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
440 	} else {
441 		dev->atomic_cap = IB_ATOMIC_GLOB;
442 		DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
443 	}
444 }
445 
446 static const struct qed_rdma_ops *qed_ops;
447 
448 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
449 
450 static irqreturn_t qedr_irq_handler(int irq, void *handle)
451 {
452 	u16 hw_comp_cons, sw_comp_cons;
453 	struct qedr_cnq *cnq = handle;
454 	struct regpair *cq_handle;
455 	struct qedr_cq *cq;
456 
457 	qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
458 
459 	qed_sb_update_sb_idx(cnq->sb);
460 
461 	hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
462 	sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
463 
464 	/* Align protocol-index and chain reads */
465 	rmb();
466 
467 	while (sw_comp_cons != hw_comp_cons) {
468 		cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
469 		cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
470 				cq_handle->lo);
471 
472 		if (cq == NULL) {
473 			DP_ERR(cnq->dev,
474 			       "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
475 			       cq_handle->hi, cq_handle->lo, sw_comp_cons,
476 			       hw_comp_cons);
477 
478 			break;
479 		}
480 
481 		if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
482 			DP_ERR(cnq->dev,
483 			       "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
484 			       cq_handle->hi, cq_handle->lo, cq);
485 			break;
486 		}
487 
488 		cq->arm_flags = 0;
489 
490 		if (!cq->destroyed && cq->ibcq.comp_handler)
491 			(*cq->ibcq.comp_handler)
492 				(&cq->ibcq, cq->ibcq.cq_context);
493 
494 		/* The CQ's CNQ notification counter is checked before
495 		 * destroying the CQ in a busy-wait loop that waits for all of
496 		 * the CQ's CNQ interrupts to be processed. It is increased
497 		 * here, only after the completion handler, to ensure that the
498 		 * the handler is not running when the CQ is destroyed.
499 		 */
500 		cq->cnq_notif++;
501 
502 		sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
503 
504 		cnq->n_comp++;
505 	}
506 
507 	qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
508 				      sw_comp_cons);
509 
510 	qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
511 
512 	return IRQ_HANDLED;
513 }
514 
515 static void qedr_sync_free_irqs(struct qedr_dev *dev)
516 {
517 	u32 vector;
518 	u16 idx;
519 	int i;
520 
521 	for (i = 0; i < dev->int_info.used_cnt; i++) {
522 		if (dev->int_info.msix_cnt) {
523 			idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
524 			vector = dev->int_info.msix[idx].vector;
525 			synchronize_irq(vector);
526 			free_irq(vector, &dev->cnq_array[i]);
527 		}
528 	}
529 
530 	dev->int_info.used_cnt = 0;
531 }
532 
533 static int qedr_req_msix_irqs(struct qedr_dev *dev)
534 {
535 	int i, rc = 0;
536 	u16 idx;
537 
538 	if (dev->num_cnq > dev->int_info.msix_cnt) {
539 		DP_ERR(dev,
540 		       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
541 		       dev->num_cnq, dev->int_info.msix_cnt);
542 		return -EINVAL;
543 	}
544 
545 	for (i = 0; i < dev->num_cnq; i++) {
546 		idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
547 		rc = request_irq(dev->int_info.msix[idx].vector,
548 				 qedr_irq_handler, 0, dev->cnq_array[i].name,
549 				 &dev->cnq_array[i]);
550 		if (rc) {
551 			DP_ERR(dev, "Request cnq %d irq failed\n", i);
552 			qedr_sync_free_irqs(dev);
553 		} else {
554 			DP_DEBUG(dev, QEDR_MSG_INIT,
555 				 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
556 				 dev->cnq_array[i].name, i,
557 				 &dev->cnq_array[i]);
558 			dev->int_info.used_cnt++;
559 		}
560 	}
561 
562 	return rc;
563 }
564 
565 static int qedr_setup_irqs(struct qedr_dev *dev)
566 {
567 	int rc;
568 
569 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
570 
571 	/* Learn Interrupt configuration */
572 	rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
573 	if (rc < 0)
574 		return rc;
575 
576 	rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
577 	if (rc) {
578 		DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
579 		return rc;
580 	}
581 
582 	if (dev->int_info.msix_cnt) {
583 		DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
584 			 dev->int_info.msix_cnt);
585 		rc = qedr_req_msix_irqs(dev);
586 		if (rc)
587 			return rc;
588 	}
589 
590 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
591 
592 	return 0;
593 }
594 
595 static int qedr_set_device_attr(struct qedr_dev *dev)
596 {
597 	struct qed_rdma_device *qed_attr;
598 	struct qedr_device_attr *attr;
599 	u32 page_size;
600 
601 	/* Part 1 - query core capabilities */
602 	qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
603 
604 	/* Part 2 - check capabilities */
605 	page_size = ~dev->attr.page_size_caps + 1;
606 	if (page_size > PAGE_SIZE) {
607 		DP_ERR(dev,
608 		       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
609 		       PAGE_SIZE, page_size);
610 		return -ENODEV;
611 	}
612 
613 	/* Part 3 - copy and update capabilities */
614 	attr = &dev->attr;
615 	attr->vendor_id = qed_attr->vendor_id;
616 	attr->vendor_part_id = qed_attr->vendor_part_id;
617 	attr->hw_ver = qed_attr->hw_ver;
618 	attr->fw_ver = qed_attr->fw_ver;
619 	attr->node_guid = qed_attr->node_guid;
620 	attr->sys_image_guid = qed_attr->sys_image_guid;
621 	attr->max_cnq = qed_attr->max_cnq;
622 	attr->max_sge = qed_attr->max_sge;
623 	attr->max_inline = qed_attr->max_inline;
624 	attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
625 	attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
626 	attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
627 	attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
628 	attr->max_dev_resp_rd_atomic_resc =
629 	    qed_attr->max_dev_resp_rd_atomic_resc;
630 	attr->max_cq = qed_attr->max_cq;
631 	attr->max_qp = qed_attr->max_qp;
632 	attr->max_mr = qed_attr->max_mr;
633 	attr->max_mr_size = qed_attr->max_mr_size;
634 	attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
635 	attr->max_mw = qed_attr->max_mw;
636 	attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
637 	attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
638 	attr->max_pd = qed_attr->max_pd;
639 	attr->max_ah = qed_attr->max_ah;
640 	attr->max_pkey = qed_attr->max_pkey;
641 	attr->max_srq = qed_attr->max_srq;
642 	attr->max_srq_wr = qed_attr->max_srq_wr;
643 	attr->dev_caps = qed_attr->dev_caps;
644 	attr->page_size_caps = qed_attr->page_size_caps;
645 	attr->dev_ack_delay = qed_attr->dev_ack_delay;
646 	attr->reserved_lkey = qed_attr->reserved_lkey;
647 	attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
648 	attr->max_stats_queues = qed_attr->max_stats_queues;
649 
650 	return 0;
651 }
652 
653 static void qedr_unaffiliated_event(void *context, u8 event_code)
654 {
655 	pr_err("unaffiliated event not implemented yet\n");
656 }
657 
658 static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
659 {
660 #define EVENT_TYPE_NOT_DEFINED	0
661 #define EVENT_TYPE_CQ		1
662 #define EVENT_TYPE_QP		2
663 #define EVENT_TYPE_SRQ		3
664 	struct qedr_dev *dev = (struct qedr_dev *)context;
665 	struct regpair *async_handle = (struct regpair *)fw_handle;
666 	u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
667 	u8 event_type = EVENT_TYPE_NOT_DEFINED;
668 	struct ib_event event;
669 	struct ib_srq *ibsrq;
670 	struct qedr_srq *srq;
671 	unsigned long flags;
672 	struct ib_cq *ibcq;
673 	struct ib_qp *ibqp;
674 	struct qedr_cq *cq;
675 	struct qedr_qp *qp;
676 	u16 srq_id;
677 
678 	if (IS_ROCE(dev)) {
679 		switch (e_code) {
680 		case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
681 			event.event = IB_EVENT_CQ_ERR;
682 			event_type = EVENT_TYPE_CQ;
683 			break;
684 		case ROCE_ASYNC_EVENT_SQ_DRAINED:
685 			event.event = IB_EVENT_SQ_DRAINED;
686 			event_type = EVENT_TYPE_QP;
687 			break;
688 		case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
689 			event.event = IB_EVENT_QP_FATAL;
690 			event_type = EVENT_TYPE_QP;
691 			break;
692 		case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
693 			event.event = IB_EVENT_QP_REQ_ERR;
694 			event_type = EVENT_TYPE_QP;
695 			break;
696 		case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
697 			event.event = IB_EVENT_QP_ACCESS_ERR;
698 			event_type = EVENT_TYPE_QP;
699 			break;
700 		case ROCE_ASYNC_EVENT_SRQ_LIMIT:
701 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
702 			event_type = EVENT_TYPE_SRQ;
703 			break;
704 		case ROCE_ASYNC_EVENT_SRQ_EMPTY:
705 			event.event = IB_EVENT_SRQ_ERR;
706 			event_type = EVENT_TYPE_SRQ;
707 			break;
708 		default:
709 			DP_ERR(dev, "unsupported event %d on handle=%llx\n",
710 			       e_code, roce_handle64);
711 		}
712 	} else {
713 		switch (e_code) {
714 		case QED_IWARP_EVENT_SRQ_LIMIT:
715 			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
716 			event_type = EVENT_TYPE_SRQ;
717 			break;
718 		case QED_IWARP_EVENT_SRQ_EMPTY:
719 			event.event = IB_EVENT_SRQ_ERR;
720 			event_type = EVENT_TYPE_SRQ;
721 			break;
722 		default:
723 		DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
724 		       roce_handle64);
725 		}
726 	}
727 	switch (event_type) {
728 	case EVENT_TYPE_CQ:
729 		cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
730 		if (cq) {
731 			ibcq = &cq->ibcq;
732 			if (ibcq->event_handler) {
733 				event.device = ibcq->device;
734 				event.element.cq = ibcq;
735 				ibcq->event_handler(&event, ibcq->cq_context);
736 			}
737 		} else {
738 			WARN(1,
739 			     "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
740 			     roce_handle64);
741 		}
742 		DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
743 		break;
744 	case EVENT_TYPE_QP:
745 		qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
746 		if (qp) {
747 			ibqp = &qp->ibqp;
748 			if (ibqp->event_handler) {
749 				event.device = ibqp->device;
750 				event.element.qp = ibqp;
751 				ibqp->event_handler(&event, ibqp->qp_context);
752 			}
753 		} else {
754 			WARN(1,
755 			     "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
756 			     roce_handle64);
757 		}
758 		DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
759 		break;
760 	case EVENT_TYPE_SRQ:
761 		srq_id = (u16)roce_handle64;
762 		xa_lock_irqsave(&dev->srqs, flags);
763 		srq = xa_load(&dev->srqs, srq_id);
764 		if (srq) {
765 			ibsrq = &srq->ibsrq;
766 			if (ibsrq->event_handler) {
767 				event.device = ibsrq->device;
768 				event.element.srq = ibsrq;
769 				ibsrq->event_handler(&event,
770 						     ibsrq->srq_context);
771 			}
772 		} else {
773 			DP_NOTICE(dev,
774 				  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
775 				  roce_handle64);
776 		}
777 		xa_unlock_irqrestore(&dev->srqs, flags);
778 		DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
779 	default:
780 		break;
781 	}
782 }
783 
784 static int qedr_init_hw(struct qedr_dev *dev)
785 {
786 	struct qed_rdma_add_user_out_params out_params;
787 	struct qed_rdma_start_in_params *in_params;
788 	struct qed_rdma_cnq_params *cur_pbl;
789 	struct qed_rdma_events events;
790 	dma_addr_t p_phys_table;
791 	u32 page_cnt;
792 	int rc = 0;
793 	int i;
794 
795 	in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
796 	if (!in_params) {
797 		rc = -ENOMEM;
798 		goto out;
799 	}
800 
801 	in_params->desired_cnq = dev->num_cnq;
802 	for (i = 0; i < dev->num_cnq; i++) {
803 		cur_pbl = &in_params->cnq_pbl_list[i];
804 
805 		page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
806 		cur_pbl->num_pbl_pages = page_cnt;
807 
808 		p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
809 		cur_pbl->pbl_ptr = (u64)p_phys_table;
810 	}
811 
812 	events.affiliated_event = qedr_affiliated_event;
813 	events.unaffiliated_event = qedr_unaffiliated_event;
814 	events.context = dev;
815 
816 	in_params->events = &events;
817 	in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
818 	in_params->max_mtu = dev->ndev->mtu;
819 	dev->iwarp_max_mtu = dev->ndev->mtu;
820 	ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
821 
822 	rc = dev->ops->rdma_init(dev->cdev, in_params);
823 	if (rc)
824 		goto out;
825 
826 	rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
827 	if (rc)
828 		goto out;
829 
830 	dev->db_addr = out_params.dpi_addr;
831 	dev->db_phys_addr = out_params.dpi_phys_addr;
832 	dev->db_size = out_params.dpi_size;
833 	dev->dpi = out_params.dpi;
834 
835 	rc = qedr_set_device_attr(dev);
836 out:
837 	kfree(in_params);
838 	if (rc)
839 		DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
840 
841 	return rc;
842 }
843 
844 static void qedr_stop_hw(struct qedr_dev *dev)
845 {
846 	dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
847 	dev->ops->rdma_stop(dev->rdma_ctx);
848 }
849 
850 static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
851 				 struct net_device *ndev)
852 {
853 	struct qed_dev_rdma_info dev_info;
854 	struct qedr_dev *dev;
855 	int rc = 0;
856 
857 	dev = ib_alloc_device(qedr_dev, ibdev);
858 	if (!dev) {
859 		pr_err("Unable to allocate ib device\n");
860 		return NULL;
861 	}
862 
863 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
864 
865 	dev->pdev = pdev;
866 	dev->ndev = ndev;
867 	dev->cdev = cdev;
868 
869 	qed_ops = qed_get_rdma_ops();
870 	if (!qed_ops) {
871 		DP_ERR(dev, "Failed to get qed roce operations\n");
872 		goto init_err;
873 	}
874 
875 	dev->ops = qed_ops;
876 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
877 	if (rc)
878 		goto init_err;
879 
880 	dev->user_dpm_enabled = dev_info.user_dpm_enabled;
881 	dev->rdma_type = dev_info.rdma_type;
882 	dev->num_hwfns = dev_info.common.num_hwfns;
883 
884 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev)) {
885 		rc = dev->ops->iwarp_set_engine_affin(cdev, false);
886 		if (rc) {
887 			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");
888 			goto init_err;
889 		}
890 	}
891 	dev->affin_hwfn_idx = dev->ops->common->get_affin_hwfn_idx(cdev);
892 
893 	dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
894 
895 	dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
896 	if (!dev->num_cnq) {
897 		DP_ERR(dev, "Failed. At least one CNQ is required.\n");
898 		rc = -ENOMEM;
899 		goto init_err;
900 	}
901 
902 	dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
903 
904 	qedr_pci_set_atomic(dev, pdev);
905 
906 	rc = qedr_alloc_resources(dev);
907 	if (rc)
908 		goto init_err;
909 
910 	rc = qedr_init_hw(dev);
911 	if (rc)
912 		goto alloc_err;
913 
914 	rc = qedr_setup_irqs(dev);
915 	if (rc)
916 		goto irq_err;
917 
918 	rc = qedr_register_device(dev);
919 	if (rc) {
920 		DP_ERR(dev, "Unable to allocate register device\n");
921 		goto reg_err;
922 	}
923 
924 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
925 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
926 
927 	DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
928 	return dev;
929 
930 reg_err:
931 	qedr_sync_free_irqs(dev);
932 irq_err:
933 	qedr_stop_hw(dev);
934 alloc_err:
935 	qedr_free_resources(dev);
936 init_err:
937 	ib_dealloc_device(&dev->ibdev);
938 	DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
939 
940 	return NULL;
941 }
942 
943 static void qedr_remove(struct qedr_dev *dev)
944 {
945 	/* First unregister with stack to stop all the active traffic
946 	 * of the registered clients.
947 	 */
948 	ib_unregister_device(&dev->ibdev);
949 
950 	qedr_stop_hw(dev);
951 	qedr_sync_free_irqs(dev);
952 	qedr_free_resources(dev);
953 
954 	if (IS_IWARP(dev) && QEDR_IS_CMT(dev))
955 		dev->ops->iwarp_set_engine_affin(dev->cdev, true);
956 
957 	ib_dealloc_device(&dev->ibdev);
958 }
959 
960 static void qedr_close(struct qedr_dev *dev)
961 {
962 	if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
963 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
964 }
965 
966 static void qedr_shutdown(struct qedr_dev *dev)
967 {
968 	qedr_close(dev);
969 	qedr_remove(dev);
970 }
971 
972 static void qedr_open(struct qedr_dev *dev)
973 {
974 	if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
975 		qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
976 }
977 
978 static void qedr_mac_address_change(struct qedr_dev *dev)
979 {
980 	union ib_gid *sgid = &dev->sgid_tbl[0];
981 	u8 guid[8], mac_addr[6];
982 	int rc;
983 
984 	/* Update SGID */
985 	ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
986 	guid[0] = mac_addr[0] ^ 2;
987 	guid[1] = mac_addr[1];
988 	guid[2] = mac_addr[2];
989 	guid[3] = 0xff;
990 	guid[4] = 0xfe;
991 	guid[5] = mac_addr[3];
992 	guid[6] = mac_addr[4];
993 	guid[7] = mac_addr[5];
994 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
995 	memcpy(&sgid->raw[8], guid, sizeof(guid));
996 
997 	/* Update LL2 */
998 	rc = dev->ops->ll2_set_mac_filter(dev->cdev,
999 					  dev->gsi_ll2_mac_address,
1000 					  dev->ndev->dev_addr);
1001 
1002 	ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
1003 
1004 	qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
1005 
1006 	if (rc)
1007 		DP_ERR(dev, "Error updating mac filter\n");
1008 }
1009 
1010 /* event handling via NIC driver ensures that all the NIC specific
1011  * initialization done before RoCE driver notifies
1012  * event to stack.
1013  */
1014 static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1015 {
1016 	switch (event) {
1017 	case QEDE_UP:
1018 		qedr_open(dev);
1019 		break;
1020 	case QEDE_DOWN:
1021 		qedr_close(dev);
1022 		break;
1023 	case QEDE_CLOSE:
1024 		qedr_shutdown(dev);
1025 		break;
1026 	case QEDE_CHANGE_ADDR:
1027 		qedr_mac_address_change(dev);
1028 		break;
1029 	default:
1030 		pr_err("Event not supported\n");
1031 	}
1032 }
1033 
1034 static struct qedr_driver qedr_drv = {
1035 	.name = "qedr_driver",
1036 	.add = qedr_add,
1037 	.remove = qedr_remove,
1038 	.notify = qedr_notify,
1039 };
1040 
1041 static int __init qedr_init_module(void)
1042 {
1043 	return qede_rdma_register_driver(&qedr_drv);
1044 }
1045 
1046 static void __exit qedr_exit_module(void)
1047 {
1048 	qede_rdma_unregister_driver(&qedr_drv);
1049 }
1050 
1051 module_init(qedr_init_module);
1052 module_exit(qedr_exit_module);
1053