xref: /openbmc/linux/drivers/net/ethernet/qlogic/qed/qed_roce.c (revision e983940270f10fe8551baf0098be76ea478294a3)
1 /* QLogic qed 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/types.h>
33 #include <asm/byteorder.h>
34 #include <linux/bitops.h>
35 #include <linux/delay.h>
36 #include <linux/dma-mapping.h>
37 #include <linux/errno.h>
38 #include <linux/etherdevice.h>
39 #include <linux/if_ether.h>
40 #include <linux/if_vlan.h>
41 #include <linux/io.h>
42 #include <linux/ip.h>
43 #include <linux/ipv6.h>
44 #include <linux/kernel.h>
45 #include <linux/list.h>
46 #include <linux/module.h>
47 #include <linux/mutex.h>
48 #include <linux/pci.h>
49 #include <linux/slab.h>
50 #include <linux/spinlock.h>
51 #include <linux/string.h>
52 #include <linux/tcp.h>
53 #include <linux/bitops.h>
54 #include <linux/qed/qed_roce_if.h>
55 #include <linux/qed/qed_roce_if.h>
56 #include "qed.h"
57 #include "qed_cxt.h"
58 #include "qed_hsi.h"
59 #include "qed_hw.h"
60 #include "qed_init_ops.h"
61 #include "qed_int.h"
62 #include "qed_ll2.h"
63 #include "qed_mcp.h"
64 #include "qed_reg_addr.h"
65 #include "qed_sp.h"
66 #include "qed_roce.h"
67 #include "qed_ll2.h"
68 
69 void qed_async_roce_event(struct qed_hwfn *p_hwfn,
70 			  struct event_ring_entry *p_eqe)
71 {
72 	struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
73 
74 	p_rdma_info->events.affiliated_event(p_rdma_info->events.context,
75 					     p_eqe->opcode, &p_eqe->data);
76 }
77 
78 static int qed_rdma_bmap_alloc(struct qed_hwfn *p_hwfn,
79 			       struct qed_bmap *bmap, u32 max_count)
80 {
81 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "max_count = %08x\n", max_count);
82 
83 	bmap->max_count = max_count;
84 
85 	bmap->bitmap = kzalloc(BITS_TO_LONGS(max_count) * sizeof(long),
86 			       GFP_KERNEL);
87 	if (!bmap->bitmap) {
88 		DP_NOTICE(p_hwfn,
89 			  "qed bmap alloc failed: cannot allocate memory (bitmap)\n");
90 		return -ENOMEM;
91 	}
92 
93 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocated bitmap %p\n",
94 		   bmap->bitmap);
95 	return 0;
96 }
97 
98 static int qed_rdma_bmap_alloc_id(struct qed_hwfn *p_hwfn,
99 				  struct qed_bmap *bmap, u32 *id_num)
100 {
101 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "bmap = %p\n", bmap);
102 
103 	*id_num = find_first_zero_bit(bmap->bitmap, bmap->max_count);
104 
105 	if (*id_num >= bmap->max_count) {
106 		DP_NOTICE(p_hwfn, "no id available max_count=%d\n",
107 			  bmap->max_count);
108 		return -EINVAL;
109 	}
110 
111 	__set_bit(*id_num, bmap->bitmap);
112 
113 	return 0;
114 }
115 
116 static void qed_bmap_release_id(struct qed_hwfn *p_hwfn,
117 				struct qed_bmap *bmap, u32 id_num)
118 {
119 	bool b_acquired;
120 
121 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "id_num = %08x", id_num);
122 	if (id_num >= bmap->max_count)
123 		return;
124 
125 	b_acquired = test_and_clear_bit(id_num, bmap->bitmap);
126 	if (!b_acquired) {
127 		DP_NOTICE(p_hwfn, "ID %d already released\n", id_num);
128 		return;
129 	}
130 }
131 
132 u32 qed_rdma_get_sb_id(void *p_hwfn, u32 rel_sb_id)
133 {
134 	/* First sb id for RoCE is after all the l2 sb */
135 	return FEAT_NUM((struct qed_hwfn *)p_hwfn, QED_PF_L2_QUE) + rel_sb_id;
136 }
137 
138 u32 qed_rdma_query_cau_timer_res(void *rdma_cxt)
139 {
140 	return QED_CAU_DEF_RX_TIMER_RES;
141 }
142 
143 static int qed_rdma_alloc(struct qed_hwfn *p_hwfn,
144 			  struct qed_ptt *p_ptt,
145 			  struct qed_rdma_start_in_params *params)
146 {
147 	struct qed_rdma_info *p_rdma_info;
148 	u32 num_cons, num_tasks;
149 	int rc = -ENOMEM;
150 
151 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocating RDMA\n");
152 
153 	/* Allocate a struct with current pf rdma info */
154 	p_rdma_info = kzalloc(sizeof(*p_rdma_info), GFP_KERNEL);
155 	if (!p_rdma_info) {
156 		DP_NOTICE(p_hwfn,
157 			  "qed rdma alloc failed: cannot allocate memory (rdma info). rc = %d\n",
158 			  rc);
159 		return rc;
160 	}
161 
162 	p_hwfn->p_rdma_info = p_rdma_info;
163 	p_rdma_info->proto = PROTOCOLID_ROCE;
164 
165 	num_cons = qed_cxt_get_proto_cid_count(p_hwfn, p_rdma_info->proto, 0);
166 
167 	p_rdma_info->num_qps = num_cons / 2;
168 
169 	num_tasks = qed_cxt_get_proto_tid_count(p_hwfn, PROTOCOLID_ROCE);
170 
171 	/* Each MR uses a single task */
172 	p_rdma_info->num_mrs = num_tasks;
173 
174 	/* Queue zone lines are shared between RoCE and L2 in such a way that
175 	 * they can be used by each without obstructing the other.
176 	 */
177 	p_rdma_info->queue_zone_base = (u16)FEAT_NUM(p_hwfn, QED_L2_QUEUE);
178 
179 	/* Allocate a struct with device params and fill it */
180 	p_rdma_info->dev = kzalloc(sizeof(*p_rdma_info->dev), GFP_KERNEL);
181 	if (!p_rdma_info->dev) {
182 		DP_NOTICE(p_hwfn,
183 			  "qed rdma alloc failed: cannot allocate memory (rdma info dev). rc = %d\n",
184 			  rc);
185 		goto free_rdma_info;
186 	}
187 
188 	/* Allocate a struct with port params and fill it */
189 	p_rdma_info->port = kzalloc(sizeof(*p_rdma_info->port), GFP_KERNEL);
190 	if (!p_rdma_info->port) {
191 		DP_NOTICE(p_hwfn,
192 			  "qed rdma alloc failed: cannot allocate memory (rdma info port). rc = %d\n",
193 			  rc);
194 		goto free_rdma_dev;
195 	}
196 
197 	/* Allocate bit map for pd's */
198 	rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->pd_map, RDMA_MAX_PDS);
199 	if (rc) {
200 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
201 			   "Failed to allocate pd_map, rc = %d\n",
202 			   rc);
203 		goto free_rdma_port;
204 	}
205 
206 	/* Allocate DPI bitmap */
207 	rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->dpi_map,
208 				 p_hwfn->dpi_count);
209 	if (rc) {
210 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
211 			   "Failed to allocate DPI bitmap, rc = %d\n", rc);
212 		goto free_pd_map;
213 	}
214 
215 	/* Allocate bitmap for cq's. The maximum number of CQs is bounded to
216 	 * twice the number of QPs.
217 	 */
218 	rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cq_map,
219 				 p_rdma_info->num_qps * 2);
220 	if (rc) {
221 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
222 			   "Failed to allocate cq bitmap, rc = %d\n", rc);
223 		goto free_dpi_map;
224 	}
225 
226 	/* Allocate bitmap for toggle bit for cq icids
227 	 * We toggle the bit every time we create or resize cq for a given icid.
228 	 * The maximum number of CQs is bounded to  twice the number of QPs.
229 	 */
230 	rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->toggle_bits,
231 				 p_rdma_info->num_qps * 2);
232 	if (rc) {
233 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
234 			   "Failed to allocate toogle bits, rc = %d\n", rc);
235 		goto free_cq_map;
236 	}
237 
238 	/* Allocate bitmap for itids */
239 	rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->tid_map,
240 				 p_rdma_info->num_mrs);
241 	if (rc) {
242 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
243 			   "Failed to allocate itids bitmaps, rc = %d\n", rc);
244 		goto free_toggle_map;
245 	}
246 
247 	/* Allocate bitmap for cids used for qps. */
248 	rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cid_map, num_cons);
249 	if (rc) {
250 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
251 			   "Failed to allocate cid bitmap, rc = %d\n", rc);
252 		goto free_tid_map;
253 	}
254 
255 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocation successful\n");
256 	return 0;
257 
258 free_tid_map:
259 	kfree(p_rdma_info->tid_map.bitmap);
260 free_toggle_map:
261 	kfree(p_rdma_info->toggle_bits.bitmap);
262 free_cq_map:
263 	kfree(p_rdma_info->cq_map.bitmap);
264 free_dpi_map:
265 	kfree(p_rdma_info->dpi_map.bitmap);
266 free_pd_map:
267 	kfree(p_rdma_info->pd_map.bitmap);
268 free_rdma_port:
269 	kfree(p_rdma_info->port);
270 free_rdma_dev:
271 	kfree(p_rdma_info->dev);
272 free_rdma_info:
273 	kfree(p_rdma_info);
274 
275 	return rc;
276 }
277 
278 void qed_rdma_resc_free(struct qed_hwfn *p_hwfn)
279 {
280 	struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
281 
282 	kfree(p_rdma_info->cid_map.bitmap);
283 	kfree(p_rdma_info->tid_map.bitmap);
284 	kfree(p_rdma_info->toggle_bits.bitmap);
285 	kfree(p_rdma_info->cq_map.bitmap);
286 	kfree(p_rdma_info->dpi_map.bitmap);
287 	kfree(p_rdma_info->pd_map.bitmap);
288 
289 	kfree(p_rdma_info->port);
290 	kfree(p_rdma_info->dev);
291 
292 	kfree(p_rdma_info);
293 }
294 
295 static void qed_rdma_free(struct qed_hwfn *p_hwfn)
296 {
297 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Freeing RDMA\n");
298 
299 	qed_rdma_resc_free(p_hwfn);
300 }
301 
302 static void qed_rdma_get_guid(struct qed_hwfn *p_hwfn, u8 *guid)
303 {
304 	guid[0] = p_hwfn->hw_info.hw_mac_addr[0] ^ 2;
305 	guid[1] = p_hwfn->hw_info.hw_mac_addr[1];
306 	guid[2] = p_hwfn->hw_info.hw_mac_addr[2];
307 	guid[3] = 0xff;
308 	guid[4] = 0xfe;
309 	guid[5] = p_hwfn->hw_info.hw_mac_addr[3];
310 	guid[6] = p_hwfn->hw_info.hw_mac_addr[4];
311 	guid[7] = p_hwfn->hw_info.hw_mac_addr[5];
312 }
313 
314 static void qed_rdma_init_events(struct qed_hwfn *p_hwfn,
315 				 struct qed_rdma_start_in_params *params)
316 {
317 	struct qed_rdma_events *events;
318 
319 	events = &p_hwfn->p_rdma_info->events;
320 
321 	events->unaffiliated_event = params->events->unaffiliated_event;
322 	events->affiliated_event = params->events->affiliated_event;
323 	events->context = params->events->context;
324 }
325 
326 static void qed_rdma_init_devinfo(struct qed_hwfn *p_hwfn,
327 				  struct qed_rdma_start_in_params *params)
328 {
329 	struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
330 	struct qed_dev *cdev = p_hwfn->cdev;
331 	u32 pci_status_control;
332 	u32 num_qps;
333 
334 	/* Vendor specific information */
335 	dev->vendor_id = cdev->vendor_id;
336 	dev->vendor_part_id = cdev->device_id;
337 	dev->hw_ver = 0;
338 	dev->fw_ver = (FW_MAJOR_VERSION << 24) | (FW_MINOR_VERSION << 16) |
339 		      (FW_REVISION_VERSION << 8) | (FW_ENGINEERING_VERSION);
340 
341 	qed_rdma_get_guid(p_hwfn, (u8 *)&dev->sys_image_guid);
342 	dev->node_guid = dev->sys_image_guid;
343 
344 	dev->max_sge = min_t(u32, RDMA_MAX_SGE_PER_SQ_WQE,
345 			     RDMA_MAX_SGE_PER_RQ_WQE);
346 
347 	if (cdev->rdma_max_sge)
348 		dev->max_sge = min_t(u32, cdev->rdma_max_sge, dev->max_sge);
349 
350 	dev->max_inline = ROCE_REQ_MAX_INLINE_DATA_SIZE;
351 
352 	dev->max_inline = (cdev->rdma_max_inline) ?
353 			  min_t(u32, cdev->rdma_max_inline, dev->max_inline) :
354 			  dev->max_inline;
355 
356 	dev->max_wqe = QED_RDMA_MAX_WQE;
357 	dev->max_cnq = (u8)FEAT_NUM(p_hwfn, QED_RDMA_CNQ);
358 
359 	/* The number of QPs may be higher than QED_ROCE_MAX_QPS, because
360 	 * it is up-aligned to 16 and then to ILT page size within qed cxt.
361 	 * This is OK in terms of ILT but we don't want to configure the FW
362 	 * above its abilities
363 	 */
364 	num_qps = ROCE_MAX_QPS;
365 	num_qps = min_t(u64, num_qps, p_hwfn->p_rdma_info->num_qps);
366 	dev->max_qp = num_qps;
367 
368 	/* CQs uses the same icids that QPs use hence they are limited by the
369 	 * number of icids. There are two icids per QP.
370 	 */
371 	dev->max_cq = num_qps * 2;
372 
373 	/* The number of mrs is smaller by 1 since the first is reserved */
374 	dev->max_mr = p_hwfn->p_rdma_info->num_mrs - 1;
375 	dev->max_mr_size = QED_RDMA_MAX_MR_SIZE;
376 
377 	/* The maximum CQE capacity per CQ supported.
378 	 * max number of cqes will be in two layer pbl,
379 	 * 8 is the pointer size in bytes
380 	 * 32 is the size of cq element in bytes
381 	 */
382 	if (params->cq_mode == QED_RDMA_CQ_MODE_32_BITS)
383 		dev->max_cqe = QED_RDMA_MAX_CQE_32_BIT;
384 	else
385 		dev->max_cqe = QED_RDMA_MAX_CQE_16_BIT;
386 
387 	dev->max_mw = 0;
388 	dev->max_fmr = QED_RDMA_MAX_FMR;
389 	dev->max_mr_mw_fmr_pbl = (PAGE_SIZE / 8) * (PAGE_SIZE / 8);
390 	dev->max_mr_mw_fmr_size = dev->max_mr_mw_fmr_pbl * PAGE_SIZE;
391 	dev->max_pkey = QED_RDMA_MAX_P_KEY;
392 
393 	dev->max_qp_resp_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
394 					  (RDMA_RESP_RD_ATOMIC_ELM_SIZE * 2);
395 	dev->max_qp_req_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
396 					 RDMA_REQ_RD_ATOMIC_ELM_SIZE;
397 	dev->max_dev_resp_rd_atomic_resc = dev->max_qp_resp_rd_atomic_resc *
398 					   p_hwfn->p_rdma_info->num_qps;
399 	dev->page_size_caps = QED_RDMA_PAGE_SIZE_CAPS;
400 	dev->dev_ack_delay = QED_RDMA_ACK_DELAY;
401 	dev->max_pd = RDMA_MAX_PDS;
402 	dev->max_ah = p_hwfn->p_rdma_info->num_qps;
403 	dev->max_stats_queues = (u8)RESC_NUM(p_hwfn, QED_RDMA_STATS_QUEUE);
404 
405 	/* Set capablities */
406 	dev->dev_caps = 0;
407 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RNR_NAK, 1);
408 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT, 1);
409 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT, 1);
410 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RESIZE_CQ, 1);
411 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_MEMORY_EXT, 1);
412 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_QUEUE_EXT, 1);
413 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ZBVA, 1);
414 	SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_LOCAL_INV_FENCE, 1);
415 
416 	/* Check atomic operations support in PCI configuration space. */
417 	pci_read_config_dword(cdev->pdev,
418 			      cdev->pdev->pcie_cap + PCI_EXP_DEVCTL2,
419 			      &pci_status_control);
420 
421 	if (pci_status_control & PCI_EXP_DEVCTL2_LTR_EN)
422 		SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ATOMIC_OP, 1);
423 }
424 
425 static void qed_rdma_init_port(struct qed_hwfn *p_hwfn)
426 {
427 	struct qed_rdma_port *port = p_hwfn->p_rdma_info->port;
428 	struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
429 
430 	port->port_state = p_hwfn->mcp_info->link_output.link_up ?
431 			   QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
432 
433 	port->max_msg_size = min_t(u64,
434 				   (dev->max_mr_mw_fmr_size *
435 				    p_hwfn->cdev->rdma_max_sge),
436 				   BIT(31));
437 
438 	port->pkey_bad_counter = 0;
439 }
440 
441 static int qed_rdma_init_hw(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
442 {
443 	u32 ll2_ethertype_en;
444 
445 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW\n");
446 	p_hwfn->b_rdma_enabled_in_prs = false;
447 
448 	qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
449 
450 	p_hwfn->rdma_prs_search_reg = PRS_REG_SEARCH_ROCE;
451 
452 	/* We delay writing to this reg until first cid is allocated. See
453 	 * qed_cxt_dynamic_ilt_alloc function for more details
454 	 */
455 	ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
456 	qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
457 	       (ll2_ethertype_en | 0x01));
458 
459 	if (qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_ROCE) % 2) {
460 		DP_NOTICE(p_hwfn, "The first RoCE's cid should be even\n");
461 		return -EINVAL;
462 	}
463 
464 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW - Done\n");
465 	return 0;
466 }
467 
468 static int qed_rdma_start_fw(struct qed_hwfn *p_hwfn,
469 			     struct qed_rdma_start_in_params *params,
470 			     struct qed_ptt *p_ptt)
471 {
472 	struct rdma_init_func_ramrod_data *p_ramrod;
473 	struct qed_rdma_cnq_params *p_cnq_pbl_list;
474 	struct rdma_init_func_hdr *p_params_header;
475 	struct rdma_cnq_params *p_cnq_params;
476 	struct qed_sp_init_data init_data;
477 	struct qed_spq_entry *p_ent;
478 	u32 cnq_id, sb_id;
479 	int rc;
480 
481 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Starting FW\n");
482 
483 	/* Save the number of cnqs for the function close ramrod */
484 	p_hwfn->p_rdma_info->num_cnqs = params->desired_cnq;
485 
486 	/* Get SPQ entry */
487 	memset(&init_data, 0, sizeof(init_data));
488 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
489 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
490 
491 	rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_INIT,
492 				 p_hwfn->p_rdma_info->proto, &init_data);
493 	if (rc)
494 		return rc;
495 
496 	p_ramrod = &p_ent->ramrod.roce_init_func.rdma;
497 
498 	p_params_header = &p_ramrod->params_header;
499 	p_params_header->cnq_start_offset = (u8)RESC_START(p_hwfn,
500 							   QED_RDMA_CNQ_RAM);
501 	p_params_header->num_cnqs = params->desired_cnq;
502 
503 	if (params->cq_mode == QED_RDMA_CQ_MODE_16_BITS)
504 		p_params_header->cq_ring_mode = 1;
505 	else
506 		p_params_header->cq_ring_mode = 0;
507 
508 	for (cnq_id = 0; cnq_id < params->desired_cnq; cnq_id++) {
509 		sb_id = qed_rdma_get_sb_id(p_hwfn, cnq_id);
510 		p_cnq_params = &p_ramrod->cnq_params[cnq_id];
511 		p_cnq_pbl_list = &params->cnq_pbl_list[cnq_id];
512 		p_cnq_params->sb_num =
513 			cpu_to_le16(p_hwfn->sbs_info[sb_id]->igu_sb_id);
514 
515 		p_cnq_params->sb_index = p_hwfn->pf_params.rdma_pf_params.gl_pi;
516 		p_cnq_params->num_pbl_pages = p_cnq_pbl_list->num_pbl_pages;
517 
518 		DMA_REGPAIR_LE(p_cnq_params->pbl_base_addr,
519 			       p_cnq_pbl_list->pbl_ptr);
520 
521 		/* we assume here that cnq_id and qz_offset are the same */
522 		p_cnq_params->queue_zone_num =
523 			cpu_to_le16(p_hwfn->p_rdma_info->queue_zone_base +
524 				    cnq_id);
525 	}
526 
527 	return qed_spq_post(p_hwfn, p_ent, NULL);
528 }
529 
530 static int qed_rdma_reserve_lkey(struct qed_hwfn *p_hwfn)
531 {
532 	struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
533 
534 	/* The first DPI is reserved for the Kernel */
535 	__set_bit(0, p_hwfn->p_rdma_info->dpi_map.bitmap);
536 
537 	/* Tid 0 will be used as the key for "reserved MR".
538 	 * The driver should allocate memory for it so it can be loaded but no
539 	 * ramrod should be passed on it.
540 	 */
541 	qed_rdma_alloc_tid(p_hwfn, &dev->reserved_lkey);
542 	if (dev->reserved_lkey != RDMA_RESERVED_LKEY) {
543 		DP_NOTICE(p_hwfn,
544 			  "Reserved lkey should be equal to RDMA_RESERVED_LKEY\n");
545 		return -EINVAL;
546 	}
547 
548 	return 0;
549 }
550 
551 static int qed_rdma_setup(struct qed_hwfn *p_hwfn,
552 			  struct qed_ptt *p_ptt,
553 			  struct qed_rdma_start_in_params *params)
554 {
555 	int rc;
556 
557 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA setup\n");
558 
559 	spin_lock_init(&p_hwfn->p_rdma_info->lock);
560 
561 	qed_rdma_init_devinfo(p_hwfn, params);
562 	qed_rdma_init_port(p_hwfn);
563 	qed_rdma_init_events(p_hwfn, params);
564 
565 	rc = qed_rdma_reserve_lkey(p_hwfn);
566 	if (rc)
567 		return rc;
568 
569 	rc = qed_rdma_init_hw(p_hwfn, p_ptt);
570 	if (rc)
571 		return rc;
572 
573 	return qed_rdma_start_fw(p_hwfn, params, p_ptt);
574 }
575 
576 int qed_rdma_stop(void *rdma_cxt)
577 {
578 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
579 	struct rdma_close_func_ramrod_data *p_ramrod;
580 	struct qed_sp_init_data init_data;
581 	struct qed_spq_entry *p_ent;
582 	struct qed_ptt *p_ptt;
583 	u32 ll2_ethertype_en;
584 	int rc = -EBUSY;
585 
586 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop\n");
587 
588 	p_ptt = qed_ptt_acquire(p_hwfn);
589 	if (!p_ptt) {
590 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Failed to acquire PTT\n");
591 		return rc;
592 	}
593 
594 	/* Disable RoCE search */
595 	qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 0);
596 	p_hwfn->b_rdma_enabled_in_prs = false;
597 
598 	qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
599 
600 	ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
601 
602 	qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
603 	       (ll2_ethertype_en & 0xFFFE));
604 
605 	qed_ptt_release(p_hwfn, p_ptt);
606 
607 	/* Get SPQ entry */
608 	memset(&init_data, 0, sizeof(init_data));
609 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
610 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
611 
612 	/* Stop RoCE */
613 	rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_CLOSE,
614 				 p_hwfn->p_rdma_info->proto, &init_data);
615 	if (rc)
616 		goto out;
617 
618 	p_ramrod = &p_ent->ramrod.rdma_close_func;
619 
620 	p_ramrod->num_cnqs = p_hwfn->p_rdma_info->num_cnqs;
621 	p_ramrod->cnq_start_offset = (u8)RESC_START(p_hwfn, QED_RDMA_CNQ_RAM);
622 
623 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
624 
625 out:
626 	qed_rdma_free(p_hwfn);
627 
628 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop done, rc = %d\n", rc);
629 	return rc;
630 }
631 
632 int qed_rdma_add_user(void *rdma_cxt,
633 		      struct qed_rdma_add_user_out_params *out_params)
634 {
635 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
636 	u32 dpi_start_offset;
637 	u32 returned_id = 0;
638 	int rc;
639 
640 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding User\n");
641 
642 	/* Allocate DPI */
643 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
644 	rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map,
645 				    &returned_id);
646 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
647 
648 	out_params->dpi = (u16)returned_id;
649 
650 	/* Calculate the corresponding DPI address */
651 	dpi_start_offset = p_hwfn->dpi_start_offset;
652 
653 	out_params->dpi_addr = (u64)((u8 __iomem *)p_hwfn->doorbells +
654 				     dpi_start_offset +
655 				     ((out_params->dpi) * p_hwfn->dpi_size));
656 
657 	out_params->dpi_phys_addr = p_hwfn->cdev->db_phys_addr +
658 				    dpi_start_offset +
659 				    ((out_params->dpi) * p_hwfn->dpi_size);
660 
661 	out_params->dpi_size = p_hwfn->dpi_size;
662 
663 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding user - done, rc = %d\n", rc);
664 	return rc;
665 }
666 
667 struct qed_rdma_port *qed_rdma_query_port(void *rdma_cxt)
668 {
669 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
670 	struct qed_rdma_port *p_port = p_hwfn->p_rdma_info->port;
671 
672 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA Query port\n");
673 
674 	/* Link may have changed */
675 	p_port->port_state = p_hwfn->mcp_info->link_output.link_up ?
676 			     QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
677 
678 	p_port->link_speed = p_hwfn->mcp_info->link_output.speed;
679 
680 	return p_port;
681 }
682 
683 struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
684 {
685 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
686 
687 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Query device\n");
688 
689 	/* Return struct with device parameters */
690 	return p_hwfn->p_rdma_info->dev;
691 }
692 
693 void qed_rdma_free_tid(void *rdma_cxt, u32 itid)
694 {
695 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
696 
697 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
698 
699 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
700 	qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->tid_map, itid);
701 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
702 }
703 
704 int qed_rdma_alloc_tid(void *rdma_cxt, u32 *itid)
705 {
706 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
707 	int rc;
708 
709 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID\n");
710 
711 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
712 	rc = qed_rdma_bmap_alloc_id(p_hwfn,
713 				    &p_hwfn->p_rdma_info->tid_map, itid);
714 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
715 	if (rc)
716 		goto out;
717 
718 	rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_TASK, *itid);
719 out:
720 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID - done, rc = %d\n", rc);
721 	return rc;
722 }
723 
724 void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 qz_offset, u16 prod)
725 {
726 	struct qed_hwfn *p_hwfn;
727 	u16 qz_num;
728 	u32 addr;
729 
730 	p_hwfn = (struct qed_hwfn *)rdma_cxt;
731 	qz_num = p_hwfn->p_rdma_info->queue_zone_base + qz_offset;
732 	addr = GTT_BAR0_MAP_REG_USDM_RAM +
733 	       USTORM_COMMON_QUEUE_CONS_OFFSET(qz_num);
734 
735 	REG_WR16(p_hwfn, addr, prod);
736 
737 	/* keep prod updates ordered */
738 	wmb();
739 }
740 
741 static int qed_fill_rdma_dev_info(struct qed_dev *cdev,
742 				  struct qed_dev_rdma_info *info)
743 {
744 	memset(info, 0, sizeof(*info));
745 
746 	info->rdma_type = QED_RDMA_TYPE_ROCE;
747 
748 	qed_fill_dev_info(cdev, &info->common);
749 
750 	return 0;
751 }
752 
753 static int qed_rdma_get_sb_start(struct qed_dev *cdev)
754 {
755 	int feat_num;
756 
757 	if (cdev->num_hwfns > 1)
758 		feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE);
759 	else
760 		feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE) *
761 			   cdev->num_hwfns;
762 
763 	return feat_num;
764 }
765 
766 static int qed_rdma_get_min_cnq_msix(struct qed_dev *cdev)
767 {
768 	int n_cnq = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_RDMA_CNQ);
769 	int n_msix = cdev->int_params.rdma_msix_cnt;
770 
771 	return min_t(int, n_cnq, n_msix);
772 }
773 
774 static int qed_rdma_set_int(struct qed_dev *cdev, u16 cnt)
775 {
776 	int limit = 0;
777 
778 	/* Mark the fastpath as free/used */
779 	cdev->int_params.fp_initialized = cnt ? true : false;
780 
781 	if (cdev->int_params.out.int_mode != QED_INT_MODE_MSIX) {
782 		DP_ERR(cdev,
783 		       "qed roce supports only MSI-X interrupts (detected %d).\n",
784 		       cdev->int_params.out.int_mode);
785 		return -EINVAL;
786 	} else if (cdev->int_params.fp_msix_cnt) {
787 		limit = cdev->int_params.rdma_msix_cnt;
788 	}
789 
790 	if (!limit)
791 		return -ENOMEM;
792 
793 	return min_t(int, cnt, limit);
794 }
795 
796 static int qed_rdma_get_int(struct qed_dev *cdev, struct qed_int_info *info)
797 {
798 	memset(info, 0, sizeof(*info));
799 
800 	if (!cdev->int_params.fp_initialized) {
801 		DP_INFO(cdev,
802 			"Protocol driver requested interrupt information, but its support is not yet configured\n");
803 		return -EINVAL;
804 	}
805 
806 	if (cdev->int_params.out.int_mode == QED_INT_MODE_MSIX) {
807 		int msix_base = cdev->int_params.rdma_msix_base;
808 
809 		info->msix_cnt = cdev->int_params.rdma_msix_cnt;
810 		info->msix = &cdev->int_params.msix_table[msix_base];
811 
812 		DP_VERBOSE(cdev, QED_MSG_RDMA, "msix_cnt = %d msix_base=%d\n",
813 			   info->msix_cnt, msix_base);
814 	}
815 
816 	return 0;
817 }
818 
819 int qed_rdma_alloc_pd(void *rdma_cxt, u16 *pd)
820 {
821 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
822 	u32 returned_id;
823 	int rc;
824 
825 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Alloc PD\n");
826 
827 	/* Allocates an unused protection domain */
828 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
829 	rc = qed_rdma_bmap_alloc_id(p_hwfn,
830 				    &p_hwfn->p_rdma_info->pd_map, &returned_id);
831 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
832 
833 	*pd = (u16)returned_id;
834 
835 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Alloc PD - done, rc = %d\n", rc);
836 	return rc;
837 }
838 
839 void qed_rdma_free_pd(void *rdma_cxt, u16 pd)
840 {
841 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
842 
843 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "pd = %08x\n", pd);
844 
845 	/* Returns a previously allocated protection domain for reuse */
846 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
847 	qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->pd_map, pd);
848 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
849 }
850 
851 static enum qed_rdma_toggle_bit
852 qed_rdma_toggle_bit_create_resize_cq(struct qed_hwfn *p_hwfn, u16 icid)
853 {
854 	struct qed_rdma_info *p_info = p_hwfn->p_rdma_info;
855 	enum qed_rdma_toggle_bit toggle_bit;
856 	u32 bmap_id;
857 
858 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", icid);
859 
860 	/* the function toggle the bit that is related to a given icid
861 	 * and returns the new toggle bit's value
862 	 */
863 	bmap_id = icid - qed_cxt_get_proto_cid_start(p_hwfn, p_info->proto);
864 
865 	spin_lock_bh(&p_info->lock);
866 	toggle_bit = !test_and_change_bit(bmap_id,
867 					  p_info->toggle_bits.bitmap);
868 	spin_unlock_bh(&p_info->lock);
869 
870 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QED_RDMA_TOGGLE_BIT_= %d\n",
871 		   toggle_bit);
872 
873 	return toggle_bit;
874 }
875 
876 int qed_rdma_create_cq(void *rdma_cxt,
877 		       struct qed_rdma_create_cq_in_params *params, u16 *icid)
878 {
879 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
880 	struct qed_rdma_info *p_info = p_hwfn->p_rdma_info;
881 	struct rdma_create_cq_ramrod_data *p_ramrod;
882 	enum qed_rdma_toggle_bit toggle_bit;
883 	struct qed_sp_init_data init_data;
884 	struct qed_spq_entry *p_ent;
885 	u32 returned_id, start_cid;
886 	int rc;
887 
888 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "cq_handle = %08x%08x\n",
889 		   params->cq_handle_hi, params->cq_handle_lo);
890 
891 	/* Allocate icid */
892 	spin_lock_bh(&p_info->lock);
893 	rc = qed_rdma_bmap_alloc_id(p_hwfn,
894 				    &p_info->cq_map, &returned_id);
895 	spin_unlock_bh(&p_info->lock);
896 
897 	if (rc) {
898 		DP_NOTICE(p_hwfn, "Can't create CQ, rc = %d\n", rc);
899 		return rc;
900 	}
901 
902 	start_cid = qed_cxt_get_proto_cid_start(p_hwfn,
903 						p_info->proto);
904 	*icid = returned_id + start_cid;
905 
906 	/* Check if icid requires a page allocation */
907 	rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, *icid);
908 	if (rc)
909 		goto err;
910 
911 	/* Get SPQ entry */
912 	memset(&init_data, 0, sizeof(init_data));
913 	init_data.cid = *icid;
914 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
915 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
916 
917 	/* Send create CQ ramrod */
918 	rc = qed_sp_init_request(p_hwfn, &p_ent,
919 				 RDMA_RAMROD_CREATE_CQ,
920 				 p_info->proto, &init_data);
921 	if (rc)
922 		goto err;
923 
924 	p_ramrod = &p_ent->ramrod.rdma_create_cq;
925 
926 	p_ramrod->cq_handle.hi = cpu_to_le32(params->cq_handle_hi);
927 	p_ramrod->cq_handle.lo = cpu_to_le32(params->cq_handle_lo);
928 	p_ramrod->dpi = cpu_to_le16(params->dpi);
929 	p_ramrod->is_two_level_pbl = params->pbl_two_level;
930 	p_ramrod->max_cqes = cpu_to_le32(params->cq_size);
931 	DMA_REGPAIR_LE(p_ramrod->pbl_addr, params->pbl_ptr);
932 	p_ramrod->pbl_num_pages = cpu_to_le16(params->pbl_num_pages);
933 	p_ramrod->cnq_id = (u8)RESC_START(p_hwfn, QED_RDMA_CNQ_RAM) +
934 			   params->cnq_id;
935 	p_ramrod->int_timeout = params->int_timeout;
936 
937 	/* toggle the bit for every resize or create cq for a given icid */
938 	toggle_bit = qed_rdma_toggle_bit_create_resize_cq(p_hwfn, *icid);
939 
940 	p_ramrod->toggle_bit = toggle_bit;
941 
942 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
943 	if (rc) {
944 		/* restore toggle bit */
945 		qed_rdma_toggle_bit_create_resize_cq(p_hwfn, *icid);
946 		goto err;
947 	}
948 
949 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Created CQ, rc = %d\n", rc);
950 	return rc;
951 
952 err:
953 	/* release allocated icid */
954 	qed_bmap_release_id(p_hwfn, &p_info->cq_map, returned_id);
955 	DP_NOTICE(p_hwfn, "Create CQ failed, rc = %d\n", rc);
956 
957 	return rc;
958 }
959 
960 int qed_rdma_resize_cq(void *rdma_cxt,
961 		       struct qed_rdma_resize_cq_in_params *in_params,
962 		       struct qed_rdma_resize_cq_out_params *out_params)
963 {
964 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
965 	struct rdma_resize_cq_output_params *p_ramrod_res;
966 	struct rdma_resize_cq_ramrod_data *p_ramrod;
967 	enum qed_rdma_toggle_bit toggle_bit;
968 	struct qed_sp_init_data init_data;
969 	struct qed_spq_entry *p_ent;
970 	dma_addr_t ramrod_res_phys;
971 	u8 fw_return_code;
972 	int rc = -ENOMEM;
973 
974 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", in_params->icid);
975 
976 	p_ramrod_res =
977 	    (struct rdma_resize_cq_output_params *)
978 	    dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
979 			       sizeof(struct rdma_resize_cq_output_params),
980 			       &ramrod_res_phys, GFP_KERNEL);
981 	if (!p_ramrod_res) {
982 		DP_NOTICE(p_hwfn,
983 			  "qed resize cq failed: cannot allocate memory (ramrod)\n");
984 		return rc;
985 	}
986 
987 	/* Get SPQ entry */
988 	memset(&init_data, 0, sizeof(init_data));
989 	init_data.cid = in_params->icid;
990 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
991 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
992 
993 	rc = qed_sp_init_request(p_hwfn, &p_ent,
994 				 RDMA_RAMROD_RESIZE_CQ,
995 				 p_hwfn->p_rdma_info->proto, &init_data);
996 	if (rc)
997 		goto err;
998 
999 	p_ramrod = &p_ent->ramrod.rdma_resize_cq;
1000 
1001 	p_ramrod->flags = 0;
1002 
1003 	/* toggle the bit for every resize or create cq for a given icid */
1004 	toggle_bit = qed_rdma_toggle_bit_create_resize_cq(p_hwfn,
1005 							  in_params->icid);
1006 
1007 	SET_FIELD(p_ramrod->flags,
1008 		  RDMA_RESIZE_CQ_RAMROD_DATA_TOGGLE_BIT, toggle_bit);
1009 
1010 	SET_FIELD(p_ramrod->flags,
1011 		  RDMA_RESIZE_CQ_RAMROD_DATA_IS_TWO_LEVEL_PBL,
1012 		  in_params->pbl_two_level);
1013 
1014 	p_ramrod->pbl_log_page_size = in_params->pbl_page_size_log - 12;
1015 	p_ramrod->pbl_num_pages = cpu_to_le16(in_params->pbl_num_pages);
1016 	p_ramrod->max_cqes = cpu_to_le32(in_params->cq_size);
1017 	DMA_REGPAIR_LE(p_ramrod->pbl_addr, in_params->pbl_ptr);
1018 	DMA_REGPAIR_LE(p_ramrod->output_params_addr, ramrod_res_phys);
1019 
1020 	rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
1021 	if (rc)
1022 		goto err;
1023 
1024 	if (fw_return_code != RDMA_RETURN_OK) {
1025 		DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
1026 		rc = -EINVAL;
1027 		goto err;
1028 	}
1029 
1030 	out_params->prod = le32_to_cpu(p_ramrod_res->old_cq_prod);
1031 	out_params->cons = le32_to_cpu(p_ramrod_res->old_cq_cons);
1032 
1033 	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1034 			  sizeof(struct rdma_resize_cq_output_params),
1035 			  p_ramrod_res, ramrod_res_phys);
1036 
1037 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Resized CQ, rc = %d\n", rc);
1038 
1039 	return rc;
1040 
1041 err:	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1042 			  sizeof(struct rdma_resize_cq_output_params),
1043 			  p_ramrod_res, ramrod_res_phys);
1044 	DP_NOTICE(p_hwfn, "Resized CQ, Failed - rc = %d\n", rc);
1045 
1046 	return rc;
1047 }
1048 
1049 int qed_rdma_destroy_cq(void *rdma_cxt,
1050 			struct qed_rdma_destroy_cq_in_params *in_params,
1051 			struct qed_rdma_destroy_cq_out_params *out_params)
1052 {
1053 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
1054 	struct rdma_destroy_cq_output_params *p_ramrod_res;
1055 	struct rdma_destroy_cq_ramrod_data *p_ramrod;
1056 	struct qed_sp_init_data init_data;
1057 	struct qed_spq_entry *p_ent;
1058 	dma_addr_t ramrod_res_phys;
1059 	int rc = -ENOMEM;
1060 
1061 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", in_params->icid);
1062 
1063 	p_ramrod_res =
1064 	    (struct rdma_destroy_cq_output_params *)
1065 	    dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1066 			       sizeof(struct rdma_destroy_cq_output_params),
1067 			       &ramrod_res_phys, GFP_KERNEL);
1068 	if (!p_ramrod_res) {
1069 		DP_NOTICE(p_hwfn,
1070 			  "qed destroy cq failed: cannot allocate memory (ramrod)\n");
1071 		return rc;
1072 	}
1073 
1074 	/* Get SPQ entry */
1075 	memset(&init_data, 0, sizeof(init_data));
1076 	init_data.cid = in_params->icid;
1077 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1078 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1079 
1080 	/* Send destroy CQ ramrod */
1081 	rc = qed_sp_init_request(p_hwfn, &p_ent,
1082 				 RDMA_RAMROD_DESTROY_CQ,
1083 				 p_hwfn->p_rdma_info->proto, &init_data);
1084 	if (rc)
1085 		goto err;
1086 
1087 	p_ramrod = &p_ent->ramrod.rdma_destroy_cq;
1088 	DMA_REGPAIR_LE(p_ramrod->output_params_addr, ramrod_res_phys);
1089 
1090 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1091 	if (rc)
1092 		goto err;
1093 
1094 	out_params->num_cq_notif = le16_to_cpu(p_ramrod_res->cnq_num);
1095 
1096 	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1097 			  sizeof(struct rdma_destroy_cq_output_params),
1098 			  p_ramrod_res, ramrod_res_phys);
1099 
1100 	/* Free icid */
1101 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
1102 
1103 	qed_bmap_release_id(p_hwfn,
1104 			    &p_hwfn->p_rdma_info->cq_map,
1105 			    (in_params->icid -
1106 			     qed_cxt_get_proto_cid_start(p_hwfn,
1107 							 p_hwfn->
1108 							 p_rdma_info->proto)));
1109 
1110 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
1111 
1112 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroyed CQ, rc = %d\n", rc);
1113 	return rc;
1114 
1115 err:	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1116 			  sizeof(struct rdma_destroy_cq_output_params),
1117 			  p_ramrod_res, ramrod_res_phys);
1118 
1119 	return rc;
1120 }
1121 
1122 static void qed_rdma_set_fw_mac(u16 *p_fw_mac, u8 *p_qed_mac)
1123 {
1124 	p_fw_mac[0] = cpu_to_le16((p_qed_mac[0] << 8) + p_qed_mac[1]);
1125 	p_fw_mac[1] = cpu_to_le16((p_qed_mac[2] << 8) + p_qed_mac[3]);
1126 	p_fw_mac[2] = cpu_to_le16((p_qed_mac[4] << 8) + p_qed_mac[5]);
1127 }
1128 
1129 static void qed_rdma_copy_gids(struct qed_rdma_qp *qp, __le32 *src_gid,
1130 			       __le32 *dst_gid)
1131 {
1132 	u32 i;
1133 
1134 	if (qp->roce_mode == ROCE_V2_IPV4) {
1135 		/* The IPv4 addresses shall be aligned to the highest word.
1136 		 * The lower words must be zero.
1137 		 */
1138 		memset(src_gid, 0, sizeof(union qed_gid));
1139 		memset(dst_gid, 0, sizeof(union qed_gid));
1140 		src_gid[3] = cpu_to_le32(qp->sgid.ipv4_addr);
1141 		dst_gid[3] = cpu_to_le32(qp->dgid.ipv4_addr);
1142 	} else {
1143 		/* GIDs and IPv6 addresses coincide in location and size */
1144 		for (i = 0; i < ARRAY_SIZE(qp->sgid.dwords); i++) {
1145 			src_gid[i] = cpu_to_le32(qp->sgid.dwords[i]);
1146 			dst_gid[i] = cpu_to_le32(qp->dgid.dwords[i]);
1147 		}
1148 	}
1149 }
1150 
1151 static enum roce_flavor qed_roce_mode_to_flavor(enum roce_mode roce_mode)
1152 {
1153 	enum roce_flavor flavor;
1154 
1155 	switch (roce_mode) {
1156 	case ROCE_V1:
1157 		flavor = PLAIN_ROCE;
1158 		break;
1159 	case ROCE_V2_IPV4:
1160 		flavor = RROCE_IPV4;
1161 		break;
1162 	case ROCE_V2_IPV6:
1163 		flavor = ROCE_V2_IPV6;
1164 		break;
1165 	default:
1166 		flavor = MAX_ROCE_MODE;
1167 		break;
1168 	}
1169 	return flavor;
1170 }
1171 
1172 int qed_roce_alloc_cid(struct qed_hwfn *p_hwfn, u16 *cid)
1173 {
1174 	struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
1175 	u32 responder_icid;
1176 	u32 requester_icid;
1177 	int rc;
1178 
1179 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
1180 	rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_rdma_info->cid_map,
1181 				    &responder_icid);
1182 	if (rc) {
1183 		spin_unlock_bh(&p_rdma_info->lock);
1184 		return rc;
1185 	}
1186 
1187 	rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_rdma_info->cid_map,
1188 				    &requester_icid);
1189 
1190 	spin_unlock_bh(&p_rdma_info->lock);
1191 	if (rc)
1192 		goto err;
1193 
1194 	/* the two icid's should be adjacent */
1195 	if ((requester_icid - responder_icid) != 1) {
1196 		DP_NOTICE(p_hwfn, "Failed to allocate two adjacent qp's'\n");
1197 		rc = -EINVAL;
1198 		goto err;
1199 	}
1200 
1201 	responder_icid += qed_cxt_get_proto_cid_start(p_hwfn,
1202 						      p_rdma_info->proto);
1203 	requester_icid += qed_cxt_get_proto_cid_start(p_hwfn,
1204 						      p_rdma_info->proto);
1205 
1206 	/* If these icids require a new ILT line allocate DMA-able context for
1207 	 * an ILT page
1208 	 */
1209 	rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, responder_icid);
1210 	if (rc)
1211 		goto err;
1212 
1213 	rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, requester_icid);
1214 	if (rc)
1215 		goto err;
1216 
1217 	*cid = (u16)responder_icid;
1218 	return rc;
1219 
1220 err:
1221 	spin_lock_bh(&p_rdma_info->lock);
1222 	qed_bmap_release_id(p_hwfn, &p_rdma_info->cid_map, responder_icid);
1223 	qed_bmap_release_id(p_hwfn, &p_rdma_info->cid_map, requester_icid);
1224 
1225 	spin_unlock_bh(&p_rdma_info->lock);
1226 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1227 		   "Allocate CID - failed, rc = %d\n", rc);
1228 	return rc;
1229 }
1230 
1231 static int qed_roce_sp_create_responder(struct qed_hwfn *p_hwfn,
1232 					struct qed_rdma_qp *qp)
1233 {
1234 	struct roce_create_qp_resp_ramrod_data *p_ramrod;
1235 	struct qed_sp_init_data init_data;
1236 	union qed_qm_pq_params qm_params;
1237 	enum roce_flavor roce_flavor;
1238 	struct qed_spq_entry *p_ent;
1239 	u16 physical_queue0 = 0;
1240 	int rc;
1241 
1242 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
1243 
1244 	/* Allocate DMA-able memory for IRQ */
1245 	qp->irq_num_pages = 1;
1246 	qp->irq = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1247 				     RDMA_RING_PAGE_SIZE,
1248 				     &qp->irq_phys_addr, GFP_KERNEL);
1249 	if (!qp->irq) {
1250 		rc = -ENOMEM;
1251 		DP_NOTICE(p_hwfn,
1252 			  "qed create responder failed: cannot allocate memory (irq). rc = %d\n",
1253 			  rc);
1254 		return rc;
1255 	}
1256 
1257 	/* Get SPQ entry */
1258 	memset(&init_data, 0, sizeof(init_data));
1259 	init_data.cid = qp->icid;
1260 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1261 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1262 
1263 	rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_CREATE_QP,
1264 				 PROTOCOLID_ROCE, &init_data);
1265 	if (rc)
1266 		goto err;
1267 
1268 	p_ramrod = &p_ent->ramrod.roce_create_qp_resp;
1269 
1270 	p_ramrod->flags = 0;
1271 
1272 	roce_flavor = qed_roce_mode_to_flavor(qp->roce_mode);
1273 	SET_FIELD(p_ramrod->flags,
1274 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_ROCE_FLAVOR, roce_flavor);
1275 
1276 	SET_FIELD(p_ramrod->flags,
1277 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_RDMA_RD_EN,
1278 		  qp->incoming_rdma_read_en);
1279 
1280 	SET_FIELD(p_ramrod->flags,
1281 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_RDMA_WR_EN,
1282 		  qp->incoming_rdma_write_en);
1283 
1284 	SET_FIELD(p_ramrod->flags,
1285 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_ATOMIC_EN,
1286 		  qp->incoming_atomic_en);
1287 
1288 	SET_FIELD(p_ramrod->flags,
1289 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_E2E_FLOW_CONTROL_EN,
1290 		  qp->e2e_flow_control_en);
1291 
1292 	SET_FIELD(p_ramrod->flags,
1293 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_SRQ_FLG, qp->use_srq);
1294 
1295 	SET_FIELD(p_ramrod->flags,
1296 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_RESERVED_KEY_EN,
1297 		  qp->fmr_and_reserved_lkey);
1298 
1299 	SET_FIELD(p_ramrod->flags,
1300 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER,
1301 		  qp->min_rnr_nak_timer);
1302 
1303 	p_ramrod->max_ird = qp->max_rd_atomic_resp;
1304 	p_ramrod->traffic_class = qp->traffic_class_tos;
1305 	p_ramrod->hop_limit = qp->hop_limit_ttl;
1306 	p_ramrod->irq_num_pages = qp->irq_num_pages;
1307 	p_ramrod->p_key = cpu_to_le16(qp->pkey);
1308 	p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
1309 	p_ramrod->dst_qp_id = cpu_to_le32(qp->dest_qp);
1310 	p_ramrod->mtu = cpu_to_le16(qp->mtu);
1311 	p_ramrod->initial_psn = cpu_to_le32(qp->rq_psn);
1312 	p_ramrod->pd = cpu_to_le16(qp->pd);
1313 	p_ramrod->rq_num_pages = cpu_to_le16(qp->rq_num_pages);
1314 	DMA_REGPAIR_LE(p_ramrod->rq_pbl_addr, qp->rq_pbl_ptr);
1315 	DMA_REGPAIR_LE(p_ramrod->irq_pbl_addr, qp->irq_phys_addr);
1316 	qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
1317 	p_ramrod->qp_handle_for_async.hi = cpu_to_le32(qp->qp_handle_async.hi);
1318 	p_ramrod->qp_handle_for_async.lo = cpu_to_le32(qp->qp_handle_async.lo);
1319 	p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
1320 	p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
1321 	p_ramrod->stats_counter_id = p_hwfn->rel_pf_id;
1322 	p_ramrod->cq_cid = cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) |
1323 				       qp->rq_cq_id);
1324 
1325 	memset(&qm_params, 0, sizeof(qm_params));
1326 	qm_params.roce.qpid = qp->icid >> 1;
1327 	physical_queue0 = qed_get_qm_pq(p_hwfn, PROTOCOLID_ROCE, &qm_params);
1328 
1329 	p_ramrod->physical_queue0 = cpu_to_le16(physical_queue0);
1330 	p_ramrod->dpi = cpu_to_le16(qp->dpi);
1331 
1332 	qed_rdma_set_fw_mac(p_ramrod->remote_mac_addr, qp->remote_mac_addr);
1333 	qed_rdma_set_fw_mac(p_ramrod->local_mac_addr, qp->local_mac_addr);
1334 
1335 	p_ramrod->udp_src_port = qp->udp_src_port;
1336 	p_ramrod->vlan_id = cpu_to_le16(qp->vlan_id);
1337 	p_ramrod->srq_id.srq_idx = cpu_to_le16(qp->srq_id);
1338 	p_ramrod->srq_id.opaque_fid = cpu_to_le16(p_hwfn->hw_info.opaque_fid);
1339 
1340 	p_ramrod->stats_counter_id = RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) +
1341 				     qp->stats_queue;
1342 
1343 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1344 
1345 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d physical_queue0 = 0x%x\n",
1346 		   rc, physical_queue0);
1347 
1348 	if (rc)
1349 		goto err;
1350 
1351 	qp->resp_offloaded = true;
1352 
1353 	return rc;
1354 
1355 err:
1356 	DP_NOTICE(p_hwfn, "create responder - failed, rc = %d\n", rc);
1357 	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1358 			  qp->irq_num_pages * RDMA_RING_PAGE_SIZE,
1359 			  qp->irq, qp->irq_phys_addr);
1360 
1361 	return rc;
1362 }
1363 
1364 static int qed_roce_sp_create_requester(struct qed_hwfn *p_hwfn,
1365 					struct qed_rdma_qp *qp)
1366 {
1367 	struct roce_create_qp_req_ramrod_data *p_ramrod;
1368 	struct qed_sp_init_data init_data;
1369 	union qed_qm_pq_params qm_params;
1370 	enum roce_flavor roce_flavor;
1371 	struct qed_spq_entry *p_ent;
1372 	u16 physical_queue0 = 0;
1373 	int rc;
1374 
1375 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
1376 
1377 	/* Allocate DMA-able memory for ORQ */
1378 	qp->orq_num_pages = 1;
1379 	qp->orq = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1380 				     RDMA_RING_PAGE_SIZE,
1381 				     &qp->orq_phys_addr, GFP_KERNEL);
1382 	if (!qp->orq) {
1383 		rc = -ENOMEM;
1384 		DP_NOTICE(p_hwfn,
1385 			  "qed create requester failed: cannot allocate memory (orq). rc = %d\n",
1386 			  rc);
1387 		return rc;
1388 	}
1389 
1390 	/* Get SPQ entry */
1391 	memset(&init_data, 0, sizeof(init_data));
1392 	init_data.cid = qp->icid + 1;
1393 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1394 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1395 
1396 	rc = qed_sp_init_request(p_hwfn, &p_ent,
1397 				 ROCE_RAMROD_CREATE_QP,
1398 				 PROTOCOLID_ROCE, &init_data);
1399 	if (rc)
1400 		goto err;
1401 
1402 	p_ramrod = &p_ent->ramrod.roce_create_qp_req;
1403 
1404 	p_ramrod->flags = 0;
1405 
1406 	roce_flavor = qed_roce_mode_to_flavor(qp->roce_mode);
1407 	SET_FIELD(p_ramrod->flags,
1408 		  ROCE_CREATE_QP_REQ_RAMROD_DATA_ROCE_FLAVOR, roce_flavor);
1409 
1410 	SET_FIELD(p_ramrod->flags,
1411 		  ROCE_CREATE_QP_REQ_RAMROD_DATA_FMR_AND_RESERVED_EN,
1412 		  qp->fmr_and_reserved_lkey);
1413 
1414 	SET_FIELD(p_ramrod->flags,
1415 		  ROCE_CREATE_QP_REQ_RAMROD_DATA_SIGNALED_COMP, qp->signal_all);
1416 
1417 	SET_FIELD(p_ramrod->flags,
1418 		  ROCE_CREATE_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT, qp->retry_cnt);
1419 
1420 	SET_FIELD(p_ramrod->flags,
1421 		  ROCE_CREATE_QP_REQ_RAMROD_DATA_RNR_NAK_CNT,
1422 		  qp->rnr_retry_cnt);
1423 
1424 	p_ramrod->max_ord = qp->max_rd_atomic_req;
1425 	p_ramrod->traffic_class = qp->traffic_class_tos;
1426 	p_ramrod->hop_limit = qp->hop_limit_ttl;
1427 	p_ramrod->orq_num_pages = qp->orq_num_pages;
1428 	p_ramrod->p_key = cpu_to_le16(qp->pkey);
1429 	p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
1430 	p_ramrod->dst_qp_id = cpu_to_le32(qp->dest_qp);
1431 	p_ramrod->ack_timeout_val = cpu_to_le32(qp->ack_timeout);
1432 	p_ramrod->mtu = cpu_to_le16(qp->mtu);
1433 	p_ramrod->initial_psn = cpu_to_le32(qp->sq_psn);
1434 	p_ramrod->pd = cpu_to_le16(qp->pd);
1435 	p_ramrod->sq_num_pages = cpu_to_le16(qp->sq_num_pages);
1436 	DMA_REGPAIR_LE(p_ramrod->sq_pbl_addr, qp->sq_pbl_ptr);
1437 	DMA_REGPAIR_LE(p_ramrod->orq_pbl_addr, qp->orq_phys_addr);
1438 	qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
1439 	p_ramrod->qp_handle_for_async.hi = cpu_to_le32(qp->qp_handle_async.hi);
1440 	p_ramrod->qp_handle_for_async.lo = cpu_to_le32(qp->qp_handle_async.lo);
1441 	p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
1442 	p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
1443 	p_ramrod->stats_counter_id = p_hwfn->rel_pf_id;
1444 	p_ramrod->cq_cid = cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) |
1445 				       qp->sq_cq_id);
1446 
1447 	memset(&qm_params, 0, sizeof(qm_params));
1448 	qm_params.roce.qpid = qp->icid >> 1;
1449 	physical_queue0 = qed_get_qm_pq(p_hwfn, PROTOCOLID_ROCE, &qm_params);
1450 
1451 	p_ramrod->physical_queue0 = cpu_to_le16(physical_queue0);
1452 	p_ramrod->dpi = cpu_to_le16(qp->dpi);
1453 
1454 	qed_rdma_set_fw_mac(p_ramrod->remote_mac_addr, qp->remote_mac_addr);
1455 	qed_rdma_set_fw_mac(p_ramrod->local_mac_addr, qp->local_mac_addr);
1456 
1457 	p_ramrod->udp_src_port = qp->udp_src_port;
1458 	p_ramrod->vlan_id = cpu_to_le16(qp->vlan_id);
1459 	p_ramrod->stats_counter_id = RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) +
1460 				     qp->stats_queue;
1461 
1462 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1463 
1464 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
1465 
1466 	if (rc)
1467 		goto err;
1468 
1469 	qp->req_offloaded = true;
1470 
1471 	return rc;
1472 
1473 err:
1474 	DP_NOTICE(p_hwfn, "Create requested - failed, rc = %d\n", rc);
1475 	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1476 			  qp->orq_num_pages * RDMA_RING_PAGE_SIZE,
1477 			  qp->orq, qp->orq_phys_addr);
1478 	return rc;
1479 }
1480 
1481 static int qed_roce_sp_modify_responder(struct qed_hwfn *p_hwfn,
1482 					struct qed_rdma_qp *qp,
1483 					bool move_to_err, u32 modify_flags)
1484 {
1485 	struct roce_modify_qp_resp_ramrod_data *p_ramrod;
1486 	struct qed_sp_init_data init_data;
1487 	struct qed_spq_entry *p_ent;
1488 	int rc;
1489 
1490 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
1491 
1492 	if (move_to_err && !qp->resp_offloaded)
1493 		return 0;
1494 
1495 	/* Get SPQ entry */
1496 	memset(&init_data, 0, sizeof(init_data));
1497 	init_data.cid = qp->icid;
1498 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1499 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1500 
1501 	rc = qed_sp_init_request(p_hwfn, &p_ent,
1502 				 ROCE_EVENT_MODIFY_QP,
1503 				 PROTOCOLID_ROCE, &init_data);
1504 	if (rc) {
1505 		DP_NOTICE(p_hwfn, "rc = %d\n", rc);
1506 		return rc;
1507 	}
1508 
1509 	p_ramrod = &p_ent->ramrod.roce_modify_qp_resp;
1510 
1511 	p_ramrod->flags = 0;
1512 
1513 	SET_FIELD(p_ramrod->flags,
1514 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_MOVE_TO_ERR_FLG, move_to_err);
1515 
1516 	SET_FIELD(p_ramrod->flags,
1517 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_RD_EN,
1518 		  qp->incoming_rdma_read_en);
1519 
1520 	SET_FIELD(p_ramrod->flags,
1521 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_WR_EN,
1522 		  qp->incoming_rdma_write_en);
1523 
1524 	SET_FIELD(p_ramrod->flags,
1525 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_ATOMIC_EN,
1526 		  qp->incoming_atomic_en);
1527 
1528 	SET_FIELD(p_ramrod->flags,
1529 		  ROCE_CREATE_QP_RESP_RAMROD_DATA_E2E_FLOW_CONTROL_EN,
1530 		  qp->e2e_flow_control_en);
1531 
1532 	SET_FIELD(p_ramrod->flags,
1533 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_OPS_EN_FLG,
1534 		  GET_FIELD(modify_flags,
1535 			    QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN));
1536 
1537 	SET_FIELD(p_ramrod->flags,
1538 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_P_KEY_FLG,
1539 		  GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY));
1540 
1541 	SET_FIELD(p_ramrod->flags,
1542 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_ADDRESS_VECTOR_FLG,
1543 		  GET_FIELD(modify_flags,
1544 			    QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR));
1545 
1546 	SET_FIELD(p_ramrod->flags,
1547 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_MAX_IRD_FLG,
1548 		  GET_FIELD(modify_flags,
1549 			    QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP));
1550 
1551 	SET_FIELD(p_ramrod->flags,
1552 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER_FLG,
1553 		  GET_FIELD(modify_flags,
1554 			    QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER));
1555 
1556 	p_ramrod->fields = 0;
1557 	SET_FIELD(p_ramrod->fields,
1558 		  ROCE_MODIFY_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER,
1559 		  qp->min_rnr_nak_timer);
1560 
1561 	p_ramrod->max_ird = qp->max_rd_atomic_resp;
1562 	p_ramrod->traffic_class = qp->traffic_class_tos;
1563 	p_ramrod->hop_limit = qp->hop_limit_ttl;
1564 	p_ramrod->p_key = cpu_to_le16(qp->pkey);
1565 	p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
1566 	p_ramrod->mtu = cpu_to_le16(qp->mtu);
1567 	qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
1568 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1569 
1570 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify responder, rc = %d\n", rc);
1571 	return rc;
1572 }
1573 
1574 static int qed_roce_sp_modify_requester(struct qed_hwfn *p_hwfn,
1575 					struct qed_rdma_qp *qp,
1576 					bool move_to_sqd,
1577 					bool move_to_err, u32 modify_flags)
1578 {
1579 	struct roce_modify_qp_req_ramrod_data *p_ramrod;
1580 	struct qed_sp_init_data init_data;
1581 	struct qed_spq_entry *p_ent;
1582 	int rc;
1583 
1584 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
1585 
1586 	if (move_to_err && !(qp->req_offloaded))
1587 		return 0;
1588 
1589 	/* Get SPQ entry */
1590 	memset(&init_data, 0, sizeof(init_data));
1591 	init_data.cid = qp->icid + 1;
1592 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1593 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1594 
1595 	rc = qed_sp_init_request(p_hwfn, &p_ent,
1596 				 ROCE_EVENT_MODIFY_QP,
1597 				 PROTOCOLID_ROCE, &init_data);
1598 	if (rc) {
1599 		DP_NOTICE(p_hwfn, "rc = %d\n", rc);
1600 		return rc;
1601 	}
1602 
1603 	p_ramrod = &p_ent->ramrod.roce_modify_qp_req;
1604 
1605 	p_ramrod->flags = 0;
1606 
1607 	SET_FIELD(p_ramrod->flags,
1608 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_MOVE_TO_ERR_FLG, move_to_err);
1609 
1610 	SET_FIELD(p_ramrod->flags,
1611 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_MOVE_TO_SQD_FLG, move_to_sqd);
1612 
1613 	SET_FIELD(p_ramrod->flags,
1614 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_EN_SQD_ASYNC_NOTIFY,
1615 		  qp->sqd_async);
1616 
1617 	SET_FIELD(p_ramrod->flags,
1618 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_P_KEY_FLG,
1619 		  GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY));
1620 
1621 	SET_FIELD(p_ramrod->flags,
1622 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_ADDRESS_VECTOR_FLG,
1623 		  GET_FIELD(modify_flags,
1624 			    QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR));
1625 
1626 	SET_FIELD(p_ramrod->flags,
1627 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_MAX_ORD_FLG,
1628 		  GET_FIELD(modify_flags,
1629 			    QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ));
1630 
1631 	SET_FIELD(p_ramrod->flags,
1632 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_RNR_NAK_CNT_FLG,
1633 		  GET_FIELD(modify_flags,
1634 			    QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT));
1635 
1636 	SET_FIELD(p_ramrod->flags,
1637 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT_FLG,
1638 		  GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_RETRY_CNT));
1639 
1640 	SET_FIELD(p_ramrod->flags,
1641 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_ACK_TIMEOUT_FLG,
1642 		  GET_FIELD(modify_flags,
1643 			    QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT));
1644 
1645 	p_ramrod->fields = 0;
1646 	SET_FIELD(p_ramrod->fields,
1647 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT, qp->retry_cnt);
1648 
1649 	SET_FIELD(p_ramrod->fields,
1650 		  ROCE_MODIFY_QP_REQ_RAMROD_DATA_RNR_NAK_CNT,
1651 		  qp->rnr_retry_cnt);
1652 
1653 	p_ramrod->max_ord = qp->max_rd_atomic_req;
1654 	p_ramrod->traffic_class = qp->traffic_class_tos;
1655 	p_ramrod->hop_limit = qp->hop_limit_ttl;
1656 	p_ramrod->p_key = cpu_to_le16(qp->pkey);
1657 	p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
1658 	p_ramrod->ack_timeout_val = cpu_to_le32(qp->ack_timeout);
1659 	p_ramrod->mtu = cpu_to_le16(qp->mtu);
1660 	qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
1661 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1662 
1663 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify requester, rc = %d\n", rc);
1664 	return rc;
1665 }
1666 
1667 static int qed_roce_sp_destroy_qp_responder(struct qed_hwfn *p_hwfn,
1668 					    struct qed_rdma_qp *qp,
1669 					    u32 *num_invalidated_mw)
1670 {
1671 	struct roce_destroy_qp_resp_output_params *p_ramrod_res;
1672 	struct roce_destroy_qp_resp_ramrod_data *p_ramrod;
1673 	struct qed_sp_init_data init_data;
1674 	struct qed_spq_entry *p_ent;
1675 	dma_addr_t ramrod_res_phys;
1676 	int rc;
1677 
1678 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
1679 
1680 	if (!qp->resp_offloaded)
1681 		return 0;
1682 
1683 	/* Get SPQ entry */
1684 	memset(&init_data, 0, sizeof(init_data));
1685 	init_data.cid = qp->icid;
1686 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1687 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1688 
1689 	rc = qed_sp_init_request(p_hwfn, &p_ent,
1690 				 ROCE_RAMROD_DESTROY_QP,
1691 				 PROTOCOLID_ROCE, &init_data);
1692 	if (rc)
1693 		return rc;
1694 
1695 	p_ramrod = &p_ent->ramrod.roce_destroy_qp_resp;
1696 
1697 	p_ramrod_res = (struct roce_destroy_qp_resp_output_params *)
1698 	    dma_alloc_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_ramrod_res),
1699 			       &ramrod_res_phys, GFP_KERNEL);
1700 
1701 	if (!p_ramrod_res) {
1702 		rc = -ENOMEM;
1703 		DP_NOTICE(p_hwfn,
1704 			  "qed destroy responder failed: cannot allocate memory (ramrod). rc = %d\n",
1705 			  rc);
1706 		return rc;
1707 	}
1708 
1709 	DMA_REGPAIR_LE(p_ramrod->output_params_addr, ramrod_res_phys);
1710 
1711 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1712 	if (rc)
1713 		goto err;
1714 
1715 	*num_invalidated_mw = le32_to_cpu(p_ramrod_res->num_invalidated_mw);
1716 
1717 	/* Free IRQ - only if ramrod succeeded, in case FW is still using it */
1718 	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1719 			  qp->irq_num_pages * RDMA_RING_PAGE_SIZE,
1720 			  qp->irq, qp->irq_phys_addr);
1721 
1722 	qp->resp_offloaded = false;
1723 
1724 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroy responder, rc = %d\n", rc);
1725 
1726 err:
1727 	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1728 			  sizeof(struct roce_destroy_qp_resp_output_params),
1729 			  p_ramrod_res, ramrod_res_phys);
1730 
1731 	return rc;
1732 }
1733 
1734 static int qed_roce_sp_destroy_qp_requester(struct qed_hwfn *p_hwfn,
1735 					    struct qed_rdma_qp *qp,
1736 					    u32 *num_bound_mw)
1737 {
1738 	struct roce_destroy_qp_req_output_params *p_ramrod_res;
1739 	struct roce_destroy_qp_req_ramrod_data *p_ramrod;
1740 	struct qed_sp_init_data init_data;
1741 	struct qed_spq_entry *p_ent;
1742 	dma_addr_t ramrod_res_phys;
1743 	int rc = -ENOMEM;
1744 
1745 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
1746 
1747 	if (!qp->req_offloaded)
1748 		return 0;
1749 
1750 	p_ramrod_res = (struct roce_destroy_qp_req_output_params *)
1751 		       dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1752 					  sizeof(*p_ramrod_res),
1753 					  &ramrod_res_phys, GFP_KERNEL);
1754 	if (!p_ramrod_res) {
1755 		DP_NOTICE(p_hwfn,
1756 			  "qed destroy requester failed: cannot allocate memory (ramrod)\n");
1757 		return rc;
1758 	}
1759 
1760 	/* Get SPQ entry */
1761 	memset(&init_data, 0, sizeof(init_data));
1762 	init_data.cid = qp->icid + 1;
1763 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1764 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1765 
1766 	rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_DESTROY_QP,
1767 				 PROTOCOLID_ROCE, &init_data);
1768 	if (rc)
1769 		goto err;
1770 
1771 	p_ramrod = &p_ent->ramrod.roce_destroy_qp_req;
1772 	DMA_REGPAIR_LE(p_ramrod->output_params_addr, ramrod_res_phys);
1773 
1774 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1775 	if (rc)
1776 		goto err;
1777 
1778 	*num_bound_mw = le32_to_cpu(p_ramrod_res->num_bound_mw);
1779 
1780 	/* Free ORQ - only if ramrod succeeded, in case FW is still using it */
1781 	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1782 			  qp->orq_num_pages * RDMA_RING_PAGE_SIZE,
1783 			  qp->orq, qp->orq_phys_addr);
1784 
1785 	qp->req_offloaded = false;
1786 
1787 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroy requester, rc = %d\n", rc);
1788 
1789 err:
1790 	dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_ramrod_res),
1791 			  p_ramrod_res, ramrod_res_phys);
1792 
1793 	return rc;
1794 }
1795 
1796 int qed_roce_query_qp(struct qed_hwfn *p_hwfn,
1797 		      struct qed_rdma_qp *qp,
1798 		      struct qed_rdma_query_qp_out_params *out_params)
1799 {
1800 	struct roce_query_qp_resp_output_params *p_resp_ramrod_res;
1801 	struct roce_query_qp_req_output_params *p_req_ramrod_res;
1802 	struct roce_query_qp_resp_ramrod_data *p_resp_ramrod;
1803 	struct roce_query_qp_req_ramrod_data *p_req_ramrod;
1804 	struct qed_sp_init_data init_data;
1805 	dma_addr_t resp_ramrod_res_phys;
1806 	dma_addr_t req_ramrod_res_phys;
1807 	struct qed_spq_entry *p_ent;
1808 	bool rq_err_state;
1809 	bool sq_err_state;
1810 	bool sq_draining;
1811 	int rc = -ENOMEM;
1812 
1813 	if ((!(qp->resp_offloaded)) && (!(qp->req_offloaded))) {
1814 		/* We can't send ramrod to the fw since this qp wasn't offloaded
1815 		 * to the fw yet
1816 		 */
1817 		out_params->draining = false;
1818 		out_params->rq_psn = qp->rq_psn;
1819 		out_params->sq_psn = qp->sq_psn;
1820 		out_params->state = qp->cur_state;
1821 
1822 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "No QPs as no offload\n");
1823 		return 0;
1824 	}
1825 
1826 	if (!(qp->resp_offloaded)) {
1827 		DP_NOTICE(p_hwfn,
1828 			  "The responder's qp should be offloded before requester's\n");
1829 		return -EINVAL;
1830 	}
1831 
1832 	/* Send a query responder ramrod to FW to get RQ-PSN and state */
1833 	p_resp_ramrod_res = (struct roce_query_qp_resp_output_params *)
1834 	    dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1835 			       sizeof(*p_resp_ramrod_res),
1836 			       &resp_ramrod_res_phys, GFP_KERNEL);
1837 	if (!p_resp_ramrod_res) {
1838 		DP_NOTICE(p_hwfn,
1839 			  "qed query qp failed: cannot allocate memory (ramrod)\n");
1840 		return rc;
1841 	}
1842 
1843 	/* Get SPQ entry */
1844 	memset(&init_data, 0, sizeof(init_data));
1845 	init_data.cid = qp->icid;
1846 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
1847 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
1848 	rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_QUERY_QP,
1849 				 PROTOCOLID_ROCE, &init_data);
1850 	if (rc)
1851 		goto err_resp;
1852 
1853 	p_resp_ramrod = &p_ent->ramrod.roce_query_qp_resp;
1854 	DMA_REGPAIR_LE(p_resp_ramrod->output_params_addr, resp_ramrod_res_phys);
1855 
1856 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1857 	if (rc)
1858 		goto err_resp;
1859 
1860 	dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_resp_ramrod_res),
1861 			  p_resp_ramrod_res, resp_ramrod_res_phys);
1862 
1863 	out_params->rq_psn = le32_to_cpu(p_resp_ramrod_res->psn);
1864 	rq_err_state = GET_FIELD(le32_to_cpu(p_resp_ramrod_res->err_flag),
1865 				 ROCE_QUERY_QP_RESP_OUTPUT_PARAMS_ERROR_FLG);
1866 
1867 	if (!(qp->req_offloaded)) {
1868 		/* Don't send query qp for the requester */
1869 		out_params->sq_psn = qp->sq_psn;
1870 		out_params->draining = false;
1871 
1872 		if (rq_err_state)
1873 			qp->cur_state = QED_ROCE_QP_STATE_ERR;
1874 
1875 		out_params->state = qp->cur_state;
1876 
1877 		return 0;
1878 	}
1879 
1880 	/* Send a query requester ramrod to FW to get SQ-PSN and state */
1881 	p_req_ramrod_res = (struct roce_query_qp_req_output_params *)
1882 			   dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1883 					      sizeof(*p_req_ramrod_res),
1884 					      &req_ramrod_res_phys,
1885 					      GFP_KERNEL);
1886 	if (!p_req_ramrod_res) {
1887 		rc = -ENOMEM;
1888 		DP_NOTICE(p_hwfn,
1889 			  "qed query qp failed: cannot allocate memory (ramrod)\n");
1890 		return rc;
1891 	}
1892 
1893 	/* Get SPQ entry */
1894 	init_data.cid = qp->icid + 1;
1895 	rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_QUERY_QP,
1896 				 PROTOCOLID_ROCE, &init_data);
1897 	if (rc)
1898 		goto err_req;
1899 
1900 	p_req_ramrod = &p_ent->ramrod.roce_query_qp_req;
1901 	DMA_REGPAIR_LE(p_req_ramrod->output_params_addr, req_ramrod_res_phys);
1902 
1903 	rc = qed_spq_post(p_hwfn, p_ent, NULL);
1904 	if (rc)
1905 		goto err_req;
1906 
1907 	dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_req_ramrod_res),
1908 			  p_req_ramrod_res, req_ramrod_res_phys);
1909 
1910 	out_params->sq_psn = le32_to_cpu(p_req_ramrod_res->psn);
1911 	sq_err_state = GET_FIELD(le32_to_cpu(p_req_ramrod_res->flags),
1912 				 ROCE_QUERY_QP_REQ_OUTPUT_PARAMS_ERR_FLG);
1913 	sq_draining =
1914 		GET_FIELD(le32_to_cpu(p_req_ramrod_res->flags),
1915 			  ROCE_QUERY_QP_REQ_OUTPUT_PARAMS_SQ_DRAINING_FLG);
1916 
1917 	out_params->draining = false;
1918 
1919 	if (rq_err_state)
1920 		qp->cur_state = QED_ROCE_QP_STATE_ERR;
1921 	else if (sq_err_state)
1922 		qp->cur_state = QED_ROCE_QP_STATE_SQE;
1923 	else if (sq_draining)
1924 		out_params->draining = true;
1925 	out_params->state = qp->cur_state;
1926 
1927 	return 0;
1928 
1929 err_req:
1930 	dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_req_ramrod_res),
1931 			  p_req_ramrod_res, req_ramrod_res_phys);
1932 	return rc;
1933 err_resp:
1934 	dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_resp_ramrod_res),
1935 			  p_resp_ramrod_res, resp_ramrod_res_phys);
1936 	return rc;
1937 }
1938 
1939 int qed_roce_destroy_qp(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
1940 {
1941 	u32 num_invalidated_mw = 0;
1942 	u32 num_bound_mw = 0;
1943 	u32 start_cid;
1944 	int rc;
1945 
1946 	/* Destroys the specified QP */
1947 	if ((qp->cur_state != QED_ROCE_QP_STATE_RESET) &&
1948 	    (qp->cur_state != QED_ROCE_QP_STATE_ERR) &&
1949 	    (qp->cur_state != QED_ROCE_QP_STATE_INIT)) {
1950 		DP_NOTICE(p_hwfn,
1951 			  "QP must be in error, reset or init state before destroying it\n");
1952 		return -EINVAL;
1953 	}
1954 
1955 	rc = qed_roce_sp_destroy_qp_responder(p_hwfn, qp, &num_invalidated_mw);
1956 	if (rc)
1957 		return rc;
1958 
1959 	/* Send destroy requester ramrod */
1960 	rc = qed_roce_sp_destroy_qp_requester(p_hwfn, qp, &num_bound_mw);
1961 	if (rc)
1962 		return rc;
1963 
1964 	if (num_invalidated_mw != num_bound_mw) {
1965 		DP_NOTICE(p_hwfn,
1966 			  "number of invalidate memory windows is different from bounded ones\n");
1967 		return -EINVAL;
1968 	}
1969 
1970 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
1971 
1972 	start_cid = qed_cxt_get_proto_cid_start(p_hwfn,
1973 						p_hwfn->p_rdma_info->proto);
1974 
1975 	/* Release responder's icid */
1976 	qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map,
1977 			    qp->icid - start_cid);
1978 
1979 	/* Release requester's icid */
1980 	qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map,
1981 			    qp->icid + 1 - start_cid);
1982 
1983 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
1984 
1985 	return 0;
1986 }
1987 
1988 int qed_rdma_query_qp(void *rdma_cxt,
1989 		      struct qed_rdma_qp *qp,
1990 		      struct qed_rdma_query_qp_out_params *out_params)
1991 {
1992 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
1993 	int rc;
1994 
1995 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
1996 
1997 	/* The following fields are filled in from qp and not FW as they can't
1998 	 * be modified by FW
1999 	 */
2000 	out_params->mtu = qp->mtu;
2001 	out_params->dest_qp = qp->dest_qp;
2002 	out_params->incoming_atomic_en = qp->incoming_atomic_en;
2003 	out_params->e2e_flow_control_en = qp->e2e_flow_control_en;
2004 	out_params->incoming_rdma_read_en = qp->incoming_rdma_read_en;
2005 	out_params->incoming_rdma_write_en = qp->incoming_rdma_write_en;
2006 	out_params->dgid = qp->dgid;
2007 	out_params->flow_label = qp->flow_label;
2008 	out_params->hop_limit_ttl = qp->hop_limit_ttl;
2009 	out_params->traffic_class_tos = qp->traffic_class_tos;
2010 	out_params->timeout = qp->ack_timeout;
2011 	out_params->rnr_retry = qp->rnr_retry_cnt;
2012 	out_params->retry_cnt = qp->retry_cnt;
2013 	out_params->min_rnr_nak_timer = qp->min_rnr_nak_timer;
2014 	out_params->pkey_index = 0;
2015 	out_params->max_rd_atomic = qp->max_rd_atomic_req;
2016 	out_params->max_dest_rd_atomic = qp->max_rd_atomic_resp;
2017 	out_params->sqd_async = qp->sqd_async;
2018 
2019 	rc = qed_roce_query_qp(p_hwfn, qp, out_params);
2020 
2021 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Query QP, rc = %d\n", rc);
2022 	return rc;
2023 }
2024 
2025 int qed_rdma_destroy_qp(void *rdma_cxt, struct qed_rdma_qp *qp)
2026 {
2027 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
2028 	int rc = 0;
2029 
2030 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
2031 
2032 	rc = qed_roce_destroy_qp(p_hwfn, qp);
2033 
2034 	/* free qp params struct */
2035 	kfree(qp);
2036 
2037 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP destroyed\n");
2038 	return rc;
2039 }
2040 
2041 struct qed_rdma_qp *
2042 qed_rdma_create_qp(void *rdma_cxt,
2043 		   struct qed_rdma_create_qp_in_params *in_params,
2044 		   struct qed_rdma_create_qp_out_params *out_params)
2045 {
2046 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
2047 	struct qed_rdma_qp *qp;
2048 	u8 max_stats_queues;
2049 	int rc;
2050 
2051 	if (!rdma_cxt || !in_params || !out_params || !p_hwfn->p_rdma_info) {
2052 		DP_ERR(p_hwfn->cdev,
2053 		       "qed roce create qp failed due to NULL entry (rdma_cxt=%p, in=%p, out=%p, roce_info=?\n",
2054 		       rdma_cxt, in_params, out_params);
2055 		return NULL;
2056 	}
2057 
2058 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2059 		   "qed rdma create qp called with qp_handle = %08x%08x\n",
2060 		   in_params->qp_handle_hi, in_params->qp_handle_lo);
2061 
2062 	/* Some sanity checks... */
2063 	max_stats_queues = p_hwfn->p_rdma_info->dev->max_stats_queues;
2064 	if (in_params->stats_queue >= max_stats_queues) {
2065 		DP_ERR(p_hwfn->cdev,
2066 		       "qed rdma create qp failed due to invalid statistics queue %d. maximum is %d\n",
2067 		       in_params->stats_queue, max_stats_queues);
2068 		return NULL;
2069 	}
2070 
2071 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
2072 	if (!qp) {
2073 		DP_NOTICE(p_hwfn, "Failed to allocate qed_rdma_qp\n");
2074 		return NULL;
2075 	}
2076 
2077 	rc = qed_roce_alloc_cid(p_hwfn, &qp->icid);
2078 	qp->qpid = ((0xFF << 16) | qp->icid);
2079 
2080 	DP_INFO(p_hwfn, "ROCE qpid=%x\n", qp->qpid);
2081 
2082 	if (rc) {
2083 		kfree(qp);
2084 		return NULL;
2085 	}
2086 
2087 	qp->cur_state = QED_ROCE_QP_STATE_RESET;
2088 	qp->qp_handle.hi = cpu_to_le32(in_params->qp_handle_hi);
2089 	qp->qp_handle.lo = cpu_to_le32(in_params->qp_handle_lo);
2090 	qp->qp_handle_async.hi = cpu_to_le32(in_params->qp_handle_async_hi);
2091 	qp->qp_handle_async.lo = cpu_to_le32(in_params->qp_handle_async_lo);
2092 	qp->use_srq = in_params->use_srq;
2093 	qp->signal_all = in_params->signal_all;
2094 	qp->fmr_and_reserved_lkey = in_params->fmr_and_reserved_lkey;
2095 	qp->pd = in_params->pd;
2096 	qp->dpi = in_params->dpi;
2097 	qp->sq_cq_id = in_params->sq_cq_id;
2098 	qp->sq_num_pages = in_params->sq_num_pages;
2099 	qp->sq_pbl_ptr = in_params->sq_pbl_ptr;
2100 	qp->rq_cq_id = in_params->rq_cq_id;
2101 	qp->rq_num_pages = in_params->rq_num_pages;
2102 	qp->rq_pbl_ptr = in_params->rq_pbl_ptr;
2103 	qp->srq_id = in_params->srq_id;
2104 	qp->req_offloaded = false;
2105 	qp->resp_offloaded = false;
2106 	qp->e2e_flow_control_en = qp->use_srq ? false : true;
2107 	qp->stats_queue = in_params->stats_queue;
2108 
2109 	out_params->icid = qp->icid;
2110 	out_params->qp_id = qp->qpid;
2111 
2112 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Create QP, rc = %d\n", rc);
2113 	return qp;
2114 }
2115 
2116 static int qed_roce_modify_qp(struct qed_hwfn *p_hwfn,
2117 			      struct qed_rdma_qp *qp,
2118 			      enum qed_roce_qp_state prev_state,
2119 			      struct qed_rdma_modify_qp_in_params *params)
2120 {
2121 	u32 num_invalidated_mw = 0, num_bound_mw = 0;
2122 	int rc = 0;
2123 
2124 	/* Perform additional operations according to the current state and the
2125 	 * next state
2126 	 */
2127 	if (((prev_state == QED_ROCE_QP_STATE_INIT) ||
2128 	     (prev_state == QED_ROCE_QP_STATE_RESET)) &&
2129 	    (qp->cur_state == QED_ROCE_QP_STATE_RTR)) {
2130 		/* Init->RTR or Reset->RTR */
2131 		rc = qed_roce_sp_create_responder(p_hwfn, qp);
2132 		return rc;
2133 	} else if ((prev_state == QED_ROCE_QP_STATE_RTR) &&
2134 		   (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
2135 		/* RTR-> RTS */
2136 		rc = qed_roce_sp_create_requester(p_hwfn, qp);
2137 		if (rc)
2138 			return rc;
2139 
2140 		/* Send modify responder ramrod */
2141 		rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
2142 						  params->modify_flags);
2143 		return rc;
2144 	} else if ((prev_state == QED_ROCE_QP_STATE_RTS) &&
2145 		   (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
2146 		/* RTS->RTS */
2147 		rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
2148 						  params->modify_flags);
2149 		if (rc)
2150 			return rc;
2151 
2152 		rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
2153 						  params->modify_flags);
2154 		return rc;
2155 	} else if ((prev_state == QED_ROCE_QP_STATE_RTS) &&
2156 		   (qp->cur_state == QED_ROCE_QP_STATE_SQD)) {
2157 		/* RTS->SQD */
2158 		rc = qed_roce_sp_modify_requester(p_hwfn, qp, true, false,
2159 						  params->modify_flags);
2160 		return rc;
2161 	} else if ((prev_state == QED_ROCE_QP_STATE_SQD) &&
2162 		   (qp->cur_state == QED_ROCE_QP_STATE_SQD)) {
2163 		/* SQD->SQD */
2164 		rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
2165 						  params->modify_flags);
2166 		if (rc)
2167 			return rc;
2168 
2169 		rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
2170 						  params->modify_flags);
2171 		return rc;
2172 	} else if ((prev_state == QED_ROCE_QP_STATE_SQD) &&
2173 		   (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
2174 		/* SQD->RTS */
2175 		rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
2176 						  params->modify_flags);
2177 		if (rc)
2178 			return rc;
2179 
2180 		rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
2181 						  params->modify_flags);
2182 
2183 		return rc;
2184 	} else if (qp->cur_state == QED_ROCE_QP_STATE_ERR ||
2185 		   qp->cur_state == QED_ROCE_QP_STATE_SQE) {
2186 		/* ->ERR */
2187 		rc = qed_roce_sp_modify_responder(p_hwfn, qp, true,
2188 						  params->modify_flags);
2189 		if (rc)
2190 			return rc;
2191 
2192 		rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, true,
2193 						  params->modify_flags);
2194 		return rc;
2195 	} else if (qp->cur_state == QED_ROCE_QP_STATE_RESET) {
2196 		/* Any state -> RESET */
2197 
2198 		rc = qed_roce_sp_destroy_qp_responder(p_hwfn, qp,
2199 						      &num_invalidated_mw);
2200 		if (rc)
2201 			return rc;
2202 
2203 		rc = qed_roce_sp_destroy_qp_requester(p_hwfn, qp,
2204 						      &num_bound_mw);
2205 
2206 		if (num_invalidated_mw != num_bound_mw) {
2207 			DP_NOTICE(p_hwfn,
2208 				  "number of invalidate memory windows is different from bounded ones\n");
2209 			return -EINVAL;
2210 		}
2211 	} else {
2212 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "0\n");
2213 	}
2214 
2215 	return rc;
2216 }
2217 
2218 int qed_rdma_modify_qp(void *rdma_cxt,
2219 		       struct qed_rdma_qp *qp,
2220 		       struct qed_rdma_modify_qp_in_params *params)
2221 {
2222 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
2223 	enum qed_roce_qp_state prev_state;
2224 	int rc = 0;
2225 
2226 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x params->new_state=%d\n",
2227 		   qp->icid, params->new_state);
2228 
2229 	if (rc) {
2230 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
2231 		return rc;
2232 	}
2233 
2234 	if (GET_FIELD(params->modify_flags,
2235 		      QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN)) {
2236 		qp->incoming_rdma_read_en = params->incoming_rdma_read_en;
2237 		qp->incoming_rdma_write_en = params->incoming_rdma_write_en;
2238 		qp->incoming_atomic_en = params->incoming_atomic_en;
2239 	}
2240 
2241 	/* Update QP structure with the updated values */
2242 	if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_ROCE_MODE))
2243 		qp->roce_mode = params->roce_mode;
2244 	if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY))
2245 		qp->pkey = params->pkey;
2246 	if (GET_FIELD(params->modify_flags,
2247 		      QED_ROCE_MODIFY_QP_VALID_E2E_FLOW_CONTROL_EN))
2248 		qp->e2e_flow_control_en = params->e2e_flow_control_en;
2249 	if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_DEST_QP))
2250 		qp->dest_qp = params->dest_qp;
2251 	if (GET_FIELD(params->modify_flags,
2252 		      QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR)) {
2253 		/* Indicates that the following parameters have changed:
2254 		 * Traffic class, flow label, hop limit, source GID,
2255 		 * destination GID, loopback indicator
2256 		 */
2257 		qp->traffic_class_tos = params->traffic_class_tos;
2258 		qp->flow_label = params->flow_label;
2259 		qp->hop_limit_ttl = params->hop_limit_ttl;
2260 
2261 		qp->sgid = params->sgid;
2262 		qp->dgid = params->dgid;
2263 		qp->udp_src_port = 0;
2264 		qp->vlan_id = params->vlan_id;
2265 		qp->mtu = params->mtu;
2266 		qp->lb_indication = params->lb_indication;
2267 		memcpy((u8 *)&qp->remote_mac_addr[0],
2268 		       (u8 *)&params->remote_mac_addr[0], ETH_ALEN);
2269 		if (params->use_local_mac) {
2270 			memcpy((u8 *)&qp->local_mac_addr[0],
2271 			       (u8 *)&params->local_mac_addr[0], ETH_ALEN);
2272 		} else {
2273 			memcpy((u8 *)&qp->local_mac_addr[0],
2274 			       (u8 *)&p_hwfn->hw_info.hw_mac_addr, ETH_ALEN);
2275 		}
2276 	}
2277 	if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_RQ_PSN))
2278 		qp->rq_psn = params->rq_psn;
2279 	if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_SQ_PSN))
2280 		qp->sq_psn = params->sq_psn;
2281 	if (GET_FIELD(params->modify_flags,
2282 		      QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ))
2283 		qp->max_rd_atomic_req = params->max_rd_atomic_req;
2284 	if (GET_FIELD(params->modify_flags,
2285 		      QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP))
2286 		qp->max_rd_atomic_resp = params->max_rd_atomic_resp;
2287 	if (GET_FIELD(params->modify_flags,
2288 		      QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT))
2289 		qp->ack_timeout = params->ack_timeout;
2290 	if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_RETRY_CNT))
2291 		qp->retry_cnt = params->retry_cnt;
2292 	if (GET_FIELD(params->modify_flags,
2293 		      QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT))
2294 		qp->rnr_retry_cnt = params->rnr_retry_cnt;
2295 	if (GET_FIELD(params->modify_flags,
2296 		      QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER))
2297 		qp->min_rnr_nak_timer = params->min_rnr_nak_timer;
2298 
2299 	qp->sqd_async = params->sqd_async;
2300 
2301 	prev_state = qp->cur_state;
2302 	if (GET_FIELD(params->modify_flags,
2303 		      QED_RDMA_MODIFY_QP_VALID_NEW_STATE)) {
2304 		qp->cur_state = params->new_state;
2305 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "qp->cur_state=%d\n",
2306 			   qp->cur_state);
2307 	}
2308 
2309 	rc = qed_roce_modify_qp(p_hwfn, qp, prev_state, params);
2310 
2311 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify QP, rc = %d\n", rc);
2312 	return rc;
2313 }
2314 
2315 int qed_rdma_register_tid(void *rdma_cxt,
2316 			  struct qed_rdma_register_tid_in_params *params)
2317 {
2318 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
2319 	struct rdma_register_tid_ramrod_data *p_ramrod;
2320 	struct qed_sp_init_data init_data;
2321 	struct qed_spq_entry *p_ent;
2322 	enum rdma_tid_type tid_type;
2323 	u8 fw_return_code;
2324 	int rc;
2325 
2326 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", params->itid);
2327 
2328 	/* Get SPQ entry */
2329 	memset(&init_data, 0, sizeof(init_data));
2330 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
2331 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
2332 
2333 	rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_REGISTER_MR,
2334 				 p_hwfn->p_rdma_info->proto, &init_data);
2335 	if (rc) {
2336 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
2337 		return rc;
2338 	}
2339 
2340 	if (p_hwfn->p_rdma_info->last_tid < params->itid)
2341 		p_hwfn->p_rdma_info->last_tid = params->itid;
2342 
2343 	p_ramrod = &p_ent->ramrod.rdma_register_tid;
2344 
2345 	p_ramrod->flags = 0;
2346 	SET_FIELD(p_ramrod->flags,
2347 		  RDMA_REGISTER_TID_RAMROD_DATA_TWO_LEVEL_PBL,
2348 		  params->pbl_two_level);
2349 
2350 	SET_FIELD(p_ramrod->flags,
2351 		  RDMA_REGISTER_TID_RAMROD_DATA_ZERO_BASED, params->zbva);
2352 
2353 	SET_FIELD(p_ramrod->flags,
2354 		  RDMA_REGISTER_TID_RAMROD_DATA_PHY_MR, params->phy_mr);
2355 
2356 	/* Don't initialize D/C field, as it may override other bits. */
2357 	if (!(params->tid_type == QED_RDMA_TID_FMR) && !(params->dma_mr))
2358 		SET_FIELD(p_ramrod->flags,
2359 			  RDMA_REGISTER_TID_RAMROD_DATA_PAGE_SIZE_LOG,
2360 			  params->page_size_log - 12);
2361 
2362 	SET_FIELD(p_ramrod->flags,
2363 		  RDMA_REGISTER_TID_RAMROD_DATA_MAX_ID,
2364 		  p_hwfn->p_rdma_info->last_tid);
2365 
2366 	SET_FIELD(p_ramrod->flags,
2367 		  RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_READ,
2368 		  params->remote_read);
2369 
2370 	SET_FIELD(p_ramrod->flags,
2371 		  RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_WRITE,
2372 		  params->remote_write);
2373 
2374 	SET_FIELD(p_ramrod->flags,
2375 		  RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_ATOMIC,
2376 		  params->remote_atomic);
2377 
2378 	SET_FIELD(p_ramrod->flags,
2379 		  RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_WRITE,
2380 		  params->local_write);
2381 
2382 	SET_FIELD(p_ramrod->flags,
2383 		  RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_READ, params->local_read);
2384 
2385 	SET_FIELD(p_ramrod->flags,
2386 		  RDMA_REGISTER_TID_RAMROD_DATA_ENABLE_MW_BIND,
2387 		  params->mw_bind);
2388 
2389 	SET_FIELD(p_ramrod->flags1,
2390 		  RDMA_REGISTER_TID_RAMROD_DATA_PBL_PAGE_SIZE_LOG,
2391 		  params->pbl_page_size_log - 12);
2392 
2393 	SET_FIELD(p_ramrod->flags2,
2394 		  RDMA_REGISTER_TID_RAMROD_DATA_DMA_MR, params->dma_mr);
2395 
2396 	switch (params->tid_type) {
2397 	case QED_RDMA_TID_REGISTERED_MR:
2398 		tid_type = RDMA_TID_REGISTERED_MR;
2399 		break;
2400 	case QED_RDMA_TID_FMR:
2401 		tid_type = RDMA_TID_FMR;
2402 		break;
2403 	case QED_RDMA_TID_MW_TYPE1:
2404 		tid_type = RDMA_TID_MW_TYPE1;
2405 		break;
2406 	case QED_RDMA_TID_MW_TYPE2A:
2407 		tid_type = RDMA_TID_MW_TYPE2A;
2408 		break;
2409 	default:
2410 		rc = -EINVAL;
2411 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
2412 		return rc;
2413 	}
2414 	SET_FIELD(p_ramrod->flags1,
2415 		  RDMA_REGISTER_TID_RAMROD_DATA_TID_TYPE, tid_type);
2416 
2417 	p_ramrod->itid = cpu_to_le32(params->itid);
2418 	p_ramrod->key = params->key;
2419 	p_ramrod->pd = cpu_to_le16(params->pd);
2420 	p_ramrod->length_hi = (u8)(params->length >> 32);
2421 	p_ramrod->length_lo = DMA_LO_LE(params->length);
2422 	if (params->zbva) {
2423 		/* Lower 32 bits of the registered MR address.
2424 		 * In case of zero based MR, will hold FBO
2425 		 */
2426 		p_ramrod->va.hi = 0;
2427 		p_ramrod->va.lo = cpu_to_le32(params->fbo);
2428 	} else {
2429 		DMA_REGPAIR_LE(p_ramrod->va, params->vaddr);
2430 	}
2431 	DMA_REGPAIR_LE(p_ramrod->pbl_base, params->pbl_ptr);
2432 
2433 	/* DIF */
2434 	if (params->dif_enabled) {
2435 		SET_FIELD(p_ramrod->flags2,
2436 			  RDMA_REGISTER_TID_RAMROD_DATA_DIF_ON_HOST_FLG, 1);
2437 		DMA_REGPAIR_LE(p_ramrod->dif_error_addr,
2438 			       params->dif_error_addr);
2439 		DMA_REGPAIR_LE(p_ramrod->dif_runt_addr, params->dif_runt_addr);
2440 	}
2441 
2442 	rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
2443 
2444 	if (fw_return_code != RDMA_RETURN_OK) {
2445 		DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
2446 		return -EINVAL;
2447 	}
2448 
2449 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Register TID, rc = %d\n", rc);
2450 	return rc;
2451 }
2452 
2453 int qed_rdma_deregister_tid(void *rdma_cxt, u32 itid)
2454 {
2455 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
2456 	struct rdma_deregister_tid_ramrod_data *p_ramrod;
2457 	struct qed_sp_init_data init_data;
2458 	struct qed_spq_entry *p_ent;
2459 	struct qed_ptt *p_ptt;
2460 	u8 fw_return_code;
2461 	int rc;
2462 
2463 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
2464 
2465 	/* Get SPQ entry */
2466 	memset(&init_data, 0, sizeof(init_data));
2467 	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
2468 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
2469 
2470 	rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_DEREGISTER_MR,
2471 				 p_hwfn->p_rdma_info->proto, &init_data);
2472 	if (rc) {
2473 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
2474 		return rc;
2475 	}
2476 
2477 	p_ramrod = &p_ent->ramrod.rdma_deregister_tid;
2478 	p_ramrod->itid = cpu_to_le32(itid);
2479 
2480 	rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
2481 	if (rc) {
2482 		DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
2483 		return rc;
2484 	}
2485 
2486 	if (fw_return_code == RDMA_RETURN_DEREGISTER_MR_BAD_STATE_ERR) {
2487 		DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
2488 		return -EINVAL;
2489 	} else if (fw_return_code == RDMA_RETURN_NIG_DRAIN_REQ) {
2490 		/* Bit indicating that the TID is in use and a nig drain is
2491 		 * required before sending the ramrod again
2492 		 */
2493 		p_ptt = qed_ptt_acquire(p_hwfn);
2494 		if (!p_ptt) {
2495 			rc = -EBUSY;
2496 			DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2497 				   "Failed to acquire PTT\n");
2498 			return rc;
2499 		}
2500 
2501 		rc = qed_mcp_drain(p_hwfn, p_ptt);
2502 		if (rc) {
2503 			qed_ptt_release(p_hwfn, p_ptt);
2504 			DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2505 				   "Drain failed\n");
2506 			return rc;
2507 		}
2508 
2509 		qed_ptt_release(p_hwfn, p_ptt);
2510 
2511 		/* Resend the ramrod */
2512 		rc = qed_sp_init_request(p_hwfn, &p_ent,
2513 					 RDMA_RAMROD_DEREGISTER_MR,
2514 					 p_hwfn->p_rdma_info->proto,
2515 					 &init_data);
2516 		if (rc) {
2517 			DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2518 				   "Failed to init sp-element\n");
2519 			return rc;
2520 		}
2521 
2522 		rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
2523 		if (rc) {
2524 			DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2525 				   "Ramrod failed\n");
2526 			return rc;
2527 		}
2528 
2529 		if (fw_return_code != RDMA_RETURN_OK) {
2530 			DP_NOTICE(p_hwfn, "fw_return_code = %d\n",
2531 				  fw_return_code);
2532 			return rc;
2533 		}
2534 	}
2535 
2536 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "De-registered TID, rc = %d\n", rc);
2537 	return rc;
2538 }
2539 
2540 static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
2541 {
2542 	return QED_LEADING_HWFN(cdev);
2543 }
2544 
2545 static void qed_rdma_dpm_conf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
2546 {
2547 	u32 val;
2548 
2549 	val = (p_hwfn->dcbx_no_edpm || p_hwfn->db_bar_no_edpm) ? 0 : 1;
2550 
2551 	qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_DPM_ENABLE, val);
2552 	DP_VERBOSE(p_hwfn, (QED_MSG_DCB | QED_MSG_RDMA),
2553 		   "Changing DPM_EN state to %d (DCBX=%d, DB_BAR=%d)\n",
2554 		   val, p_hwfn->dcbx_no_edpm, p_hwfn->db_bar_no_edpm);
2555 }
2556 
2557 void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
2558 {
2559 	p_hwfn->db_bar_no_edpm = true;
2560 
2561 	qed_rdma_dpm_conf(p_hwfn, p_ptt);
2562 }
2563 
2564 int qed_rdma_start(void *rdma_cxt, struct qed_rdma_start_in_params *params)
2565 {
2566 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
2567 	struct qed_ptt *p_ptt;
2568 	int rc = -EBUSY;
2569 
2570 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2571 		   "desired_cnq = %08x\n", params->desired_cnq);
2572 
2573 	p_ptt = qed_ptt_acquire(p_hwfn);
2574 	if (!p_ptt)
2575 		goto err;
2576 
2577 	rc = qed_rdma_alloc(p_hwfn, p_ptt, params);
2578 	if (rc)
2579 		goto err1;
2580 
2581 	rc = qed_rdma_setup(p_hwfn, p_ptt, params);
2582 	if (rc)
2583 		goto err2;
2584 
2585 	qed_ptt_release(p_hwfn, p_ptt);
2586 
2587 	return rc;
2588 
2589 err2:
2590 	qed_rdma_free(p_hwfn);
2591 err1:
2592 	qed_ptt_release(p_hwfn, p_ptt);
2593 err:
2594 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA start - error, rc = %d\n", rc);
2595 	return rc;
2596 }
2597 
2598 static int qed_rdma_init(struct qed_dev *cdev,
2599 			 struct qed_rdma_start_in_params *params)
2600 {
2601 	return qed_rdma_start(QED_LEADING_HWFN(cdev), params);
2602 }
2603 
2604 void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
2605 {
2606 	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
2607 
2608 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "dpi = %08x\n", dpi);
2609 
2610 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
2611 	qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map, dpi);
2612 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
2613 }
2614 
2615 void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
2616 				     u8 connection_handle,
2617 				     void *cookie,
2618 				     dma_addr_t first_frag_addr,
2619 				     bool b_last_fragment, bool b_last_packet)
2620 {
2621 	struct qed_roce_ll2_packet *packet = cookie;
2622 	struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
2623 
2624 	roce_ll2->cbs.tx_cb(roce_ll2->cb_cookie, packet);
2625 }
2626 
2627 void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
2628 				    u8 connection_handle,
2629 				    void *cookie,
2630 				    dma_addr_t first_frag_addr,
2631 				    bool b_last_fragment, bool b_last_packet)
2632 {
2633 	qed_ll2b_complete_tx_gsi_packet(p_hwfn, connection_handle,
2634 					cookie, first_frag_addr,
2635 					b_last_fragment, b_last_packet);
2636 }
2637 
2638 void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
2639 				     u8 connection_handle,
2640 				     void *cookie,
2641 				     dma_addr_t rx_buf_addr,
2642 				     u16 data_length,
2643 				     u8 data_length_error,
2644 				     u16 parse_flags,
2645 				     u16 vlan,
2646 				     u32 src_mac_addr_hi,
2647 				     u16 src_mac_addr_lo, bool b_last_packet)
2648 {
2649 	struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
2650 	struct qed_roce_ll2_rx_params params;
2651 	struct qed_dev *cdev = p_hwfn->cdev;
2652 	struct qed_roce_ll2_packet pkt;
2653 
2654 	DP_VERBOSE(cdev,
2655 		   QED_MSG_LL2,
2656 		   "roce ll2 rx complete: bus_addr=%p, len=%d, data_len_err=%d\n",
2657 		   (void *)(uintptr_t)rx_buf_addr,
2658 		   data_length, data_length_error);
2659 
2660 	memset(&pkt, 0, sizeof(pkt));
2661 	pkt.n_seg = 1;
2662 	pkt.payload[0].baddr = rx_buf_addr;
2663 	pkt.payload[0].len = data_length;
2664 
2665 	memset(&params, 0, sizeof(params));
2666 	params.vlan_id = vlan;
2667 	*((u32 *)&params.smac[0]) = ntohl(src_mac_addr_hi);
2668 	*((u16 *)&params.smac[4]) = ntohs(src_mac_addr_lo);
2669 
2670 	if (data_length_error) {
2671 		DP_ERR(cdev,
2672 		       "roce ll2 rx complete: data length error %d, length=%d\n",
2673 		       data_length_error, data_length);
2674 		params.rc = -EINVAL;
2675 	}
2676 
2677 	roce_ll2->cbs.rx_cb(roce_ll2->cb_cookie, &pkt, &params);
2678 }
2679 
2680 static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
2681 				       u8 *old_mac_address,
2682 				       u8 *new_mac_address)
2683 {
2684 	struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
2685 	struct qed_ptt *p_ptt;
2686 	int rc = 0;
2687 
2688 	if (!hwfn->ll2 || hwfn->ll2->handle == QED_LL2_UNUSED_HANDLE) {
2689 		DP_ERR(cdev,
2690 		       "qed roce mac filter failed - roce_info/ll2 NULL\n");
2691 		return -EINVAL;
2692 	}
2693 
2694 	p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
2695 	if (!p_ptt) {
2696 		DP_ERR(cdev,
2697 		       "qed roce ll2 mac filter set: failed to acquire PTT\n");
2698 		return -EINVAL;
2699 	}
2700 
2701 	mutex_lock(&hwfn->ll2->lock);
2702 	if (old_mac_address)
2703 		qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
2704 					  old_mac_address);
2705 	if (new_mac_address)
2706 		rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
2707 					    new_mac_address);
2708 	mutex_unlock(&hwfn->ll2->lock);
2709 
2710 	qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
2711 
2712 	if (rc)
2713 		DP_ERR(cdev,
2714 		       "qed roce ll2 mac filter set: failed to add mac filter\n");
2715 
2716 	return rc;
2717 }
2718 
2719 static int qed_roce_ll2_start(struct qed_dev *cdev,
2720 			      struct qed_roce_ll2_params *params)
2721 {
2722 	struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
2723 	struct qed_roce_ll2_info *roce_ll2;
2724 	struct qed_ll2_info ll2_params;
2725 	int rc;
2726 
2727 	if (!params) {
2728 		DP_ERR(cdev, "qed roce ll2 start: failed due to NULL params\n");
2729 		return -EINVAL;
2730 	}
2731 	if (!params->cbs.tx_cb || !params->cbs.rx_cb) {
2732 		DP_ERR(cdev,
2733 		       "qed roce ll2 start: failed due to NULL tx/rx. tx_cb=%p, rx_cb=%p\n",
2734 		       params->cbs.tx_cb, params->cbs.rx_cb);
2735 		return -EINVAL;
2736 	}
2737 	if (!is_valid_ether_addr(params->mac_address)) {
2738 		DP_ERR(cdev,
2739 		       "qed roce ll2 start: failed due to invalid Ethernet address %pM\n",
2740 		       params->mac_address);
2741 		return -EINVAL;
2742 	}
2743 
2744 	/* Initialize */
2745 	roce_ll2 = kzalloc(sizeof(*roce_ll2), GFP_ATOMIC);
2746 	if (!roce_ll2) {
2747 		DP_ERR(cdev, "qed roce ll2 start: failed memory allocation\n");
2748 		return -ENOMEM;
2749 	}
2750 	memset(roce_ll2, 0, sizeof(*roce_ll2));
2751 	roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
2752 	roce_ll2->cbs = params->cbs;
2753 	roce_ll2->cb_cookie = params->cb_cookie;
2754 	mutex_init(&roce_ll2->lock);
2755 
2756 	memset(&ll2_params, 0, sizeof(ll2_params));
2757 	ll2_params.conn_type = QED_LL2_TYPE_ROCE;
2758 	ll2_params.mtu = params->mtu;
2759 	ll2_params.rx_drop_ttl0_flg = true;
2760 	ll2_params.rx_vlan_removal_en = false;
2761 	ll2_params.tx_dest = CORE_TX_DEST_NW;
2762 	ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
2763 	ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
2764 	ll2_params.gsi_enable = true;
2765 
2766 	rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
2767 					params->max_rx_buffers,
2768 					params->max_tx_buffers,
2769 					&roce_ll2->handle);
2770 	if (rc) {
2771 		DP_ERR(cdev,
2772 		       "qed roce ll2 start: failed to acquire LL2 connection (rc=%d)\n",
2773 		       rc);
2774 		goto err;
2775 	}
2776 
2777 	rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
2778 					  roce_ll2->handle);
2779 	if (rc) {
2780 		DP_ERR(cdev,
2781 		       "qed roce ll2 start: failed to establish LL2 connection (rc=%d)\n",
2782 		       rc);
2783 		goto err1;
2784 	}
2785 
2786 	hwfn->ll2 = roce_ll2;
2787 
2788 	rc = qed_roce_ll2_set_mac_filter(cdev, NULL, params->mac_address);
2789 	if (rc) {
2790 		hwfn->ll2 = NULL;
2791 		goto err2;
2792 	}
2793 	ether_addr_copy(roce_ll2->mac_address, params->mac_address);
2794 
2795 	return 0;
2796 
2797 err2:
2798 	qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
2799 err1:
2800 	qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
2801 err:
2802 	kfree(roce_ll2);
2803 	return rc;
2804 }
2805 
2806 static int qed_roce_ll2_stop(struct qed_dev *cdev)
2807 {
2808 	struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
2809 	struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
2810 	int rc;
2811 
2812 	if (!cdev) {
2813 		DP_ERR(cdev, "qed roce ll2 stop: invalid cdev\n");
2814 		return -EINVAL;
2815 	}
2816 
2817 	if (roce_ll2->handle == QED_LL2_UNUSED_HANDLE) {
2818 		DP_ERR(cdev, "qed roce ll2 stop: cannot stop an unused LL2\n");
2819 		return -EINVAL;
2820 	}
2821 
2822 	/* remove LL2 MAC address filter */
2823 	rc = qed_roce_ll2_set_mac_filter(cdev, roce_ll2->mac_address, NULL);
2824 	eth_zero_addr(roce_ll2->mac_address);
2825 
2826 	rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
2827 					  roce_ll2->handle);
2828 	if (rc)
2829 		DP_ERR(cdev,
2830 		       "qed roce ll2 stop: failed to terminate LL2 connection (rc=%d)\n",
2831 		       rc);
2832 
2833 	qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
2834 
2835 	roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
2836 
2837 	kfree(roce_ll2);
2838 
2839 	return rc;
2840 }
2841 
2842 static int qed_roce_ll2_tx(struct qed_dev *cdev,
2843 			   struct qed_roce_ll2_packet *pkt,
2844 			   struct qed_roce_ll2_tx_params *params)
2845 {
2846 	struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
2847 	struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
2848 	enum qed_ll2_roce_flavor_type qed_roce_flavor;
2849 	u8 flags = 0;
2850 	int rc;
2851 	int i;
2852 
2853 	if (!cdev || !pkt || !params) {
2854 		DP_ERR(cdev,
2855 		       "roce ll2 tx: failed tx because one of the following is NULL - drv=%p, pkt=%p, params=%p\n",
2856 		       cdev, pkt, params);
2857 		return -EINVAL;
2858 	}
2859 
2860 	qed_roce_flavor = (pkt->roce_mode == ROCE_V1) ? QED_LL2_ROCE
2861 						      : QED_LL2_RROCE;
2862 
2863 	if (pkt->roce_mode == ROCE_V2_IPV4)
2864 		flags |= BIT(CORE_TX_BD_FLAGS_IP_CSUM_SHIFT);
2865 
2866 	/* Tx header */
2867 	rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev), roce_ll2->handle,
2868 				       1 + pkt->n_seg, 0, flags, 0,
2869 				       qed_roce_flavor, pkt->header.baddr,
2870 				       pkt->header.len, pkt, 1);
2871 	if (rc) {
2872 		DP_ERR(cdev, "roce ll2 tx: header failed (rc=%d)\n", rc);
2873 		return QED_ROCE_TX_HEAD_FAILURE;
2874 	}
2875 
2876 	/* Tx payload */
2877 	for (i = 0; i < pkt->n_seg; i++) {
2878 		rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
2879 						       roce_ll2->handle,
2880 						       pkt->payload[i].baddr,
2881 						       pkt->payload[i].len);
2882 		if (rc) {
2883 			/* If failed not much to do here, partial packet has
2884 			 * been posted * we can't free memory, will need to wait
2885 			 * for completion
2886 			 */
2887 			DP_ERR(cdev,
2888 			       "roce ll2 tx: payload failed (rc=%d)\n", rc);
2889 			return QED_ROCE_TX_FRAG_FAILURE;
2890 		}
2891 	}
2892 
2893 	return 0;
2894 }
2895 
2896 static int qed_roce_ll2_post_rx_buffer(struct qed_dev *cdev,
2897 				       struct qed_roce_ll2_buffer *buf,
2898 				       u64 cookie, u8 notify_fw)
2899 {
2900 	return qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
2901 				      QED_LEADING_HWFN(cdev)->ll2->handle,
2902 				      buf->baddr, buf->len,
2903 				      (void *)(uintptr_t)cookie, notify_fw);
2904 }
2905 
2906 static int qed_roce_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
2907 {
2908 	struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
2909 	struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
2910 
2911 	return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
2912 				 roce_ll2->handle, stats);
2913 }
2914 
2915 static const struct qed_rdma_ops qed_rdma_ops_pass = {
2916 	.common = &qed_common_ops_pass,
2917 	.fill_dev_info = &qed_fill_rdma_dev_info,
2918 	.rdma_get_rdma_ctx = &qed_rdma_get_rdma_ctx,
2919 	.rdma_init = &qed_rdma_init,
2920 	.rdma_add_user = &qed_rdma_add_user,
2921 	.rdma_remove_user = &qed_rdma_remove_user,
2922 	.rdma_stop = &qed_rdma_stop,
2923 	.rdma_query_port = &qed_rdma_query_port,
2924 	.rdma_query_device = &qed_rdma_query_device,
2925 	.rdma_get_start_sb = &qed_rdma_get_sb_start,
2926 	.rdma_get_rdma_int = &qed_rdma_get_int,
2927 	.rdma_set_rdma_int = &qed_rdma_set_int,
2928 	.rdma_get_min_cnq_msix = &qed_rdma_get_min_cnq_msix,
2929 	.rdma_cnq_prod_update = &qed_rdma_cnq_prod_update,
2930 	.rdma_alloc_pd = &qed_rdma_alloc_pd,
2931 	.rdma_dealloc_pd = &qed_rdma_free_pd,
2932 	.rdma_create_cq = &qed_rdma_create_cq,
2933 	.rdma_destroy_cq = &qed_rdma_destroy_cq,
2934 	.rdma_create_qp = &qed_rdma_create_qp,
2935 	.rdma_modify_qp = &qed_rdma_modify_qp,
2936 	.rdma_query_qp = &qed_rdma_query_qp,
2937 	.rdma_destroy_qp = &qed_rdma_destroy_qp,
2938 	.rdma_alloc_tid = &qed_rdma_alloc_tid,
2939 	.rdma_free_tid = &qed_rdma_free_tid,
2940 	.rdma_register_tid = &qed_rdma_register_tid,
2941 	.rdma_deregister_tid = &qed_rdma_deregister_tid,
2942 	.roce_ll2_start = &qed_roce_ll2_start,
2943 	.roce_ll2_stop = &qed_roce_ll2_stop,
2944 	.roce_ll2_tx = &qed_roce_ll2_tx,
2945 	.roce_ll2_post_rx_buffer = &qed_roce_ll2_post_rx_buffer,
2946 	.roce_ll2_set_mac_filter = &qed_roce_ll2_set_mac_filter,
2947 	.roce_ll2_stats = &qed_roce_ll2_stats,
2948 };
2949 
2950 const struct qed_rdma_ops *qed_get_rdma_ops(void)
2951 {
2952 	return &qed_rdma_ops_pass;
2953 }
2954 EXPORT_SYMBOL(qed_get_rdma_ops);
2955