xref: /openbmc/linux/drivers/infiniband/hw/irdma/hw.c (revision 3b73c45e)
1 // SPDX-License-Identifier: GPL-2.0 or Linux-OpenIB
2 /* Copyright (c) 2015 - 2021 Intel Corporation */
3 #include "main.h"
4 
5 static struct irdma_rsrc_limits rsrc_limits_table[] = {
6 	[0] = {
7 		.qplimit = SZ_128,
8 	},
9 	[1] = {
10 		.qplimit = SZ_1K,
11 	},
12 	[2] = {
13 		.qplimit = SZ_2K,
14 	},
15 	[3] = {
16 		.qplimit = SZ_4K,
17 	},
18 	[4] = {
19 		.qplimit = SZ_16K,
20 	},
21 	[5] = {
22 		.qplimit = SZ_64K,
23 	},
24 	[6] = {
25 		.qplimit = SZ_128K,
26 	},
27 	[7] = {
28 		.qplimit = SZ_256K,
29 	},
30 };
31 
32 /* types of hmc objects */
33 static enum irdma_hmc_rsrc_type iw_hmc_obj_types[] = {
34 	IRDMA_HMC_IW_QP,
35 	IRDMA_HMC_IW_CQ,
36 	IRDMA_HMC_IW_HTE,
37 	IRDMA_HMC_IW_ARP,
38 	IRDMA_HMC_IW_APBVT_ENTRY,
39 	IRDMA_HMC_IW_MR,
40 	IRDMA_HMC_IW_XF,
41 	IRDMA_HMC_IW_XFFL,
42 	IRDMA_HMC_IW_Q1,
43 	IRDMA_HMC_IW_Q1FL,
44 	IRDMA_HMC_IW_TIMER,
45 	IRDMA_HMC_IW_FSIMC,
46 	IRDMA_HMC_IW_FSIAV,
47 	IRDMA_HMC_IW_RRF,
48 	IRDMA_HMC_IW_RRFFL,
49 	IRDMA_HMC_IW_HDR,
50 	IRDMA_HMC_IW_MD,
51 	IRDMA_HMC_IW_OOISC,
52 	IRDMA_HMC_IW_OOISCFFL,
53 };
54 
55 /**
56  * irdma_iwarp_ce_handler - handle iwarp completions
57  * @iwcq: iwarp cq receiving event
58  */
59 static void irdma_iwarp_ce_handler(struct irdma_sc_cq *iwcq)
60 {
61 	struct irdma_cq *cq = iwcq->back_cq;
62 
63 	if (!cq->user_mode)
64 		atomic_set(&cq->armed, 0);
65 	if (cq->ibcq.comp_handler)
66 		cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
67 }
68 
69 /**
70  * irdma_puda_ce_handler - handle puda completion events
71  * @rf: RDMA PCI function
72  * @cq: puda completion q for event
73  */
74 static void irdma_puda_ce_handler(struct irdma_pci_f *rf,
75 				  struct irdma_sc_cq *cq)
76 {
77 	struct irdma_sc_dev *dev = &rf->sc_dev;
78 	u32 compl_error;
79 	int status;
80 
81 	do {
82 		status = irdma_puda_poll_cmpl(dev, cq, &compl_error);
83 		if (status == -ENOENT)
84 			break;
85 		if (status) {
86 			ibdev_dbg(to_ibdev(dev), "ERR: puda status = %d\n", status);
87 			break;
88 		}
89 		if (compl_error) {
90 			ibdev_dbg(to_ibdev(dev), "ERR: puda compl_err  =0x%x\n",
91 				  compl_error);
92 			break;
93 		}
94 	} while (1);
95 
96 	irdma_sc_ccq_arm(cq);
97 }
98 
99 /**
100  * irdma_process_ceq - handle ceq for completions
101  * @rf: RDMA PCI function
102  * @ceq: ceq having cq for completion
103  */
104 static void irdma_process_ceq(struct irdma_pci_f *rf, struct irdma_ceq *ceq)
105 {
106 	struct irdma_sc_dev *dev = &rf->sc_dev;
107 	struct irdma_sc_ceq *sc_ceq;
108 	struct irdma_sc_cq *cq;
109 	unsigned long flags;
110 
111 	sc_ceq = &ceq->sc_ceq;
112 	do {
113 		spin_lock_irqsave(&ceq->ce_lock, flags);
114 		cq = irdma_sc_process_ceq(dev, sc_ceq);
115 		if (!cq) {
116 			spin_unlock_irqrestore(&ceq->ce_lock, flags);
117 			break;
118 		}
119 
120 		if (cq->cq_type == IRDMA_CQ_TYPE_IWARP)
121 			irdma_iwarp_ce_handler(cq);
122 
123 		spin_unlock_irqrestore(&ceq->ce_lock, flags);
124 
125 		if (cq->cq_type == IRDMA_CQ_TYPE_CQP)
126 			queue_work(rf->cqp_cmpl_wq, &rf->cqp_cmpl_work);
127 		else if (cq->cq_type == IRDMA_CQ_TYPE_ILQ ||
128 			 cq->cq_type == IRDMA_CQ_TYPE_IEQ)
129 			irdma_puda_ce_handler(rf, cq);
130 	} while (1);
131 }
132 
133 static void irdma_set_flush_fields(struct irdma_sc_qp *qp,
134 				   struct irdma_aeqe_info *info)
135 {
136 	qp->sq_flush_code = info->sq;
137 	qp->rq_flush_code = info->rq;
138 	qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
139 
140 	switch (info->ae_id) {
141 	case IRDMA_AE_AMP_UNALLOCATED_STAG:
142 	case IRDMA_AE_AMP_BOUNDS_VIOLATION:
143 	case IRDMA_AE_AMP_INVALID_STAG:
144 		qp->event_type = IRDMA_QP_EVENT_ACCESS_ERR;
145 		fallthrough;
146 	case IRDMA_AE_AMP_BAD_PD:
147 	case IRDMA_AE_UDA_XMIT_BAD_PD:
148 		qp->flush_code = FLUSH_PROT_ERR;
149 		break;
150 	case IRDMA_AE_AMP_BAD_QP:
151 	case IRDMA_AE_WQE_UNEXPECTED_OPCODE:
152 		qp->flush_code = FLUSH_LOC_QP_OP_ERR;
153 		break;
154 	case IRDMA_AE_AMP_BAD_STAG_KEY:
155 	case IRDMA_AE_AMP_BAD_STAG_INDEX:
156 	case IRDMA_AE_AMP_TO_WRAP:
157 	case IRDMA_AE_AMP_RIGHTS_VIOLATION:
158 	case IRDMA_AE_AMP_INVALIDATE_NO_REMOTE_ACCESS_RIGHTS:
159 	case IRDMA_AE_PRIV_OPERATION_DENIED:
160 	case IRDMA_AE_IB_INVALID_REQUEST:
161 	case IRDMA_AE_IB_REMOTE_ACCESS_ERROR:
162 		qp->flush_code = FLUSH_REM_ACCESS_ERR;
163 		qp->event_type = IRDMA_QP_EVENT_ACCESS_ERR;
164 		break;
165 	case IRDMA_AE_LLP_SEGMENT_TOO_SMALL:
166 	case IRDMA_AE_DDP_UBE_DDP_MESSAGE_TOO_LONG_FOR_AVAILABLE_BUFFER:
167 	case IRDMA_AE_UDA_XMIT_DGRAM_TOO_LONG:
168 	case IRDMA_AE_UDA_XMIT_DGRAM_TOO_SHORT:
169 	case IRDMA_AE_UDA_L4LEN_INVALID:
170 	case IRDMA_AE_ROCE_RSP_LENGTH_ERROR:
171 		qp->flush_code = FLUSH_LOC_LEN_ERR;
172 		break;
173 	case IRDMA_AE_LCE_QP_CATASTROPHIC:
174 		qp->flush_code = FLUSH_FATAL_ERR;
175 		break;
176 	case IRDMA_AE_DDP_UBE_INVALID_MO:
177 	case IRDMA_AE_IB_RREQ_AND_Q1_FULL:
178 	case IRDMA_AE_LLP_RECEIVED_MPA_CRC_ERROR:
179 		qp->flush_code = FLUSH_GENERAL_ERR;
180 		break;
181 	case IRDMA_AE_LLP_TOO_MANY_RETRIES:
182 		qp->flush_code = FLUSH_RETRY_EXC_ERR;
183 		break;
184 	case IRDMA_AE_AMP_MWBIND_INVALID_RIGHTS:
185 	case IRDMA_AE_AMP_MWBIND_BIND_DISABLED:
186 	case IRDMA_AE_AMP_MWBIND_INVALID_BOUNDS:
187 		qp->flush_code = FLUSH_MW_BIND_ERR;
188 		break;
189 	case IRDMA_AE_IB_REMOTE_OP_ERROR:
190 		qp->flush_code = FLUSH_REM_OP_ERR;
191 		break;
192 	default:
193 		qp->flush_code = FLUSH_FATAL_ERR;
194 		break;
195 	}
196 }
197 
198 /**
199  * irdma_process_aeq - handle aeq events
200  * @rf: RDMA PCI function
201  */
202 static void irdma_process_aeq(struct irdma_pci_f *rf)
203 {
204 	struct irdma_sc_dev *dev = &rf->sc_dev;
205 	struct irdma_aeq *aeq = &rf->aeq;
206 	struct irdma_sc_aeq *sc_aeq = &aeq->sc_aeq;
207 	struct irdma_aeqe_info aeinfo;
208 	struct irdma_aeqe_info *info = &aeinfo;
209 	int ret;
210 	struct irdma_qp *iwqp = NULL;
211 	struct irdma_sc_cq *cq = NULL;
212 	struct irdma_cq *iwcq = NULL;
213 	struct irdma_sc_qp *qp = NULL;
214 	struct irdma_qp_host_ctx_info *ctx_info = NULL;
215 	struct irdma_device *iwdev = rf->iwdev;
216 	unsigned long flags;
217 
218 	u32 aeqcnt = 0;
219 
220 	if (!sc_aeq->size)
221 		return;
222 
223 	do {
224 		memset(info, 0, sizeof(*info));
225 		ret = irdma_sc_get_next_aeqe(sc_aeq, info);
226 		if (ret)
227 			break;
228 
229 		aeqcnt++;
230 		ibdev_dbg(&iwdev->ibdev,
231 			  "AEQ: ae_id = 0x%x bool qp=%d qp_id = %d tcp_state=%d iwarp_state=%d ae_src=%d\n",
232 			  info->ae_id, info->qp, info->qp_cq_id, info->tcp_state,
233 			  info->iwarp_state, info->ae_src);
234 
235 		if (info->qp) {
236 			spin_lock_irqsave(&rf->qptable_lock, flags);
237 			iwqp = rf->qp_table[info->qp_cq_id];
238 			if (!iwqp) {
239 				spin_unlock_irqrestore(&rf->qptable_lock,
240 						       flags);
241 				if (info->ae_id == IRDMA_AE_QP_SUSPEND_COMPLETE) {
242 					atomic_dec(&iwdev->vsi.qp_suspend_reqs);
243 					wake_up(&iwdev->suspend_wq);
244 					continue;
245 				}
246 				ibdev_dbg(&iwdev->ibdev, "AEQ: qp_id %d is already freed\n",
247 					  info->qp_cq_id);
248 				continue;
249 			}
250 			irdma_qp_add_ref(&iwqp->ibqp);
251 			spin_unlock_irqrestore(&rf->qptable_lock, flags);
252 			qp = &iwqp->sc_qp;
253 			spin_lock_irqsave(&iwqp->lock, flags);
254 			iwqp->hw_tcp_state = info->tcp_state;
255 			iwqp->hw_iwarp_state = info->iwarp_state;
256 			if (info->ae_id != IRDMA_AE_QP_SUSPEND_COMPLETE)
257 				iwqp->last_aeq = info->ae_id;
258 			spin_unlock_irqrestore(&iwqp->lock, flags);
259 			ctx_info = &iwqp->ctx_info;
260 		} else {
261 			if (info->ae_id != IRDMA_AE_CQ_OPERATION_ERROR)
262 				continue;
263 		}
264 
265 		switch (info->ae_id) {
266 			struct irdma_cm_node *cm_node;
267 		case IRDMA_AE_LLP_CONNECTION_ESTABLISHED:
268 			cm_node = iwqp->cm_node;
269 			if (cm_node->accept_pend) {
270 				atomic_dec(&cm_node->listener->pend_accepts_cnt);
271 				cm_node->accept_pend = 0;
272 			}
273 			iwqp->rts_ae_rcvd = 1;
274 			wake_up_interruptible(&iwqp->waitq);
275 			break;
276 		case IRDMA_AE_LLP_FIN_RECEIVED:
277 		case IRDMA_AE_RDMAP_ROE_BAD_LLP_CLOSE:
278 			if (qp->term_flags)
279 				break;
280 			if (atomic_inc_return(&iwqp->close_timer_started) == 1) {
281 				iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSE_WAIT;
282 				if (iwqp->hw_tcp_state == IRDMA_TCP_STATE_CLOSE_WAIT &&
283 				    iwqp->ibqp_state == IB_QPS_RTS) {
284 					irdma_next_iw_state(iwqp,
285 							    IRDMA_QP_STATE_CLOSING,
286 							    0, 0, 0);
287 					irdma_cm_disconn(iwqp);
288 				}
289 				irdma_schedule_cm_timer(iwqp->cm_node,
290 							(struct irdma_puda_buf *)iwqp,
291 							IRDMA_TIMER_TYPE_CLOSE,
292 							1, 0);
293 			}
294 			break;
295 		case IRDMA_AE_LLP_CLOSE_COMPLETE:
296 			if (qp->term_flags)
297 				irdma_terminate_done(qp, 0);
298 			else
299 				irdma_cm_disconn(iwqp);
300 			break;
301 		case IRDMA_AE_BAD_CLOSE:
302 		case IRDMA_AE_RESET_SENT:
303 			irdma_next_iw_state(iwqp, IRDMA_QP_STATE_ERROR, 1, 0,
304 					    0);
305 			irdma_cm_disconn(iwqp);
306 			break;
307 		case IRDMA_AE_LLP_CONNECTION_RESET:
308 			if (atomic_read(&iwqp->close_timer_started))
309 				break;
310 			irdma_cm_disconn(iwqp);
311 			break;
312 		case IRDMA_AE_QP_SUSPEND_COMPLETE:
313 			if (iwqp->iwdev->vsi.tc_change_pending) {
314 				atomic_dec(&iwqp->sc_qp.vsi->qp_suspend_reqs);
315 				wake_up(&iwqp->iwdev->suspend_wq);
316 			}
317 			break;
318 		case IRDMA_AE_TERMINATE_SENT:
319 			irdma_terminate_send_fin(qp);
320 			break;
321 		case IRDMA_AE_LLP_TERMINATE_RECEIVED:
322 			irdma_terminate_received(qp, info);
323 			break;
324 		case IRDMA_AE_CQ_OPERATION_ERROR:
325 			ibdev_err(&iwdev->ibdev,
326 				  "Processing an iWARP related AE for CQ misc = 0x%04X\n",
327 				  info->ae_id);
328 			cq = (struct irdma_sc_cq *)(unsigned long)
329 			     info->compl_ctx;
330 
331 			iwcq = cq->back_cq;
332 
333 			if (iwcq->ibcq.event_handler) {
334 				struct ib_event ibevent;
335 
336 				ibevent.device = iwcq->ibcq.device;
337 				ibevent.event = IB_EVENT_CQ_ERR;
338 				ibevent.element.cq = &iwcq->ibcq;
339 				iwcq->ibcq.event_handler(&ibevent,
340 							 iwcq->ibcq.cq_context);
341 			}
342 			break;
343 		case IRDMA_AE_RESET_NOT_SENT:
344 		case IRDMA_AE_LLP_DOUBT_REACHABILITY:
345 		case IRDMA_AE_RESOURCE_EXHAUSTION:
346 			break;
347 		case IRDMA_AE_PRIV_OPERATION_DENIED:
348 		case IRDMA_AE_STAG_ZERO_INVALID:
349 		case IRDMA_AE_IB_RREQ_AND_Q1_FULL:
350 		case IRDMA_AE_DDP_UBE_INVALID_DDP_VERSION:
351 		case IRDMA_AE_DDP_UBE_INVALID_MO:
352 		case IRDMA_AE_DDP_UBE_INVALID_QN:
353 		case IRDMA_AE_DDP_NO_L_BIT:
354 		case IRDMA_AE_RDMAP_ROE_INVALID_RDMAP_VERSION:
355 		case IRDMA_AE_RDMAP_ROE_UNEXPECTED_OPCODE:
356 		case IRDMA_AE_ROE_INVALID_RDMA_READ_REQUEST:
357 		case IRDMA_AE_ROE_INVALID_RDMA_WRITE_OR_READ_RESP:
358 		case IRDMA_AE_INVALID_ARP_ENTRY:
359 		case IRDMA_AE_INVALID_TCP_OPTION_RCVD:
360 		case IRDMA_AE_STALE_ARP_ENTRY:
361 		case IRDMA_AE_LLP_RECEIVED_MPA_CRC_ERROR:
362 		case IRDMA_AE_LLP_SEGMENT_TOO_SMALL:
363 		case IRDMA_AE_LLP_SYN_RECEIVED:
364 		case IRDMA_AE_LLP_TOO_MANY_RETRIES:
365 		case IRDMA_AE_LCE_QP_CATASTROPHIC:
366 		case IRDMA_AE_LCE_FUNCTION_CATASTROPHIC:
367 		case IRDMA_AE_LCE_CQ_CATASTROPHIC:
368 		case IRDMA_AE_UDA_XMIT_DGRAM_TOO_LONG:
369 		default:
370 			ibdev_err(&iwdev->ibdev, "abnormal ae_id = 0x%x bool qp=%d qp_id = %d, ae_src=%d\n",
371 				  info->ae_id, info->qp, info->qp_cq_id, info->ae_src);
372 			if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
373 				ctx_info->roce_info->err_rq_idx_valid = info->rq;
374 				if (info->rq) {
375 					ctx_info->roce_info->err_rq_idx = info->wqe_idx;
376 					irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va,
377 								ctx_info);
378 				}
379 				irdma_set_flush_fields(qp, info);
380 				irdma_cm_disconn(iwqp);
381 				break;
382 			}
383 			ctx_info->iwarp_info->err_rq_idx_valid = info->rq;
384 			if (info->rq) {
385 				ctx_info->iwarp_info->err_rq_idx = info->wqe_idx;
386 				ctx_info->tcp_info_valid = false;
387 				ctx_info->iwarp_info_valid = true;
388 				irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va,
389 						   ctx_info);
390 			}
391 			if (iwqp->hw_iwarp_state != IRDMA_QP_STATE_RTS &&
392 			    iwqp->hw_iwarp_state != IRDMA_QP_STATE_TERMINATE) {
393 				irdma_next_iw_state(iwqp, IRDMA_QP_STATE_ERROR, 1, 0, 0);
394 				irdma_cm_disconn(iwqp);
395 			} else {
396 				irdma_terminate_connection(qp, info);
397 			}
398 			break;
399 		}
400 		if (info->qp)
401 			irdma_qp_rem_ref(&iwqp->ibqp);
402 	} while (1);
403 
404 	if (aeqcnt)
405 		irdma_sc_repost_aeq_entries(dev, aeqcnt);
406 }
407 
408 /**
409  * irdma_ena_intr - set up device interrupts
410  * @dev: hardware control device structure
411  * @msix_id: id of the interrupt to be enabled
412  */
413 static void irdma_ena_intr(struct irdma_sc_dev *dev, u32 msix_id)
414 {
415 	dev->irq_ops->irdma_en_irq(dev, msix_id);
416 }
417 
418 /**
419  * irdma_dpc - tasklet for aeq and ceq 0
420  * @t: tasklet_struct ptr
421  */
422 static void irdma_dpc(struct tasklet_struct *t)
423 {
424 	struct irdma_pci_f *rf = from_tasklet(rf, t, dpc_tasklet);
425 
426 	if (rf->msix_shared)
427 		irdma_process_ceq(rf, rf->ceqlist);
428 	irdma_process_aeq(rf);
429 	irdma_ena_intr(&rf->sc_dev, rf->iw_msixtbl[0].idx);
430 }
431 
432 /**
433  * irdma_ceq_dpc - dpc handler for CEQ
434  * @t: tasklet_struct ptr
435  */
436 static void irdma_ceq_dpc(struct tasklet_struct *t)
437 {
438 	struct irdma_ceq *iwceq = from_tasklet(iwceq, t, dpc_tasklet);
439 	struct irdma_pci_f *rf = iwceq->rf;
440 
441 	irdma_process_ceq(rf, iwceq);
442 	irdma_ena_intr(&rf->sc_dev, iwceq->msix_idx);
443 }
444 
445 /**
446  * irdma_save_msix_info - copy msix vector information to iwarp device
447  * @rf: RDMA PCI function
448  *
449  * Allocate iwdev msix table and copy the msix info to the table
450  * Return 0 if successful, otherwise return error
451  */
452 static int irdma_save_msix_info(struct irdma_pci_f *rf)
453 {
454 	struct irdma_qvlist_info *iw_qvlist;
455 	struct irdma_qv_info *iw_qvinfo;
456 	struct msix_entry *pmsix;
457 	u32 ceq_idx;
458 	u32 i;
459 	size_t size;
460 
461 	if (!rf->msix_count)
462 		return -EINVAL;
463 
464 	size = sizeof(struct irdma_msix_vector) * rf->msix_count;
465 	size += struct_size(iw_qvlist, qv_info, rf->msix_count);
466 	rf->iw_msixtbl = kzalloc(size, GFP_KERNEL);
467 	if (!rf->iw_msixtbl)
468 		return -ENOMEM;
469 
470 	rf->iw_qvlist = (struct irdma_qvlist_info *)
471 			(&rf->iw_msixtbl[rf->msix_count]);
472 	iw_qvlist = rf->iw_qvlist;
473 	iw_qvinfo = iw_qvlist->qv_info;
474 	iw_qvlist->num_vectors = rf->msix_count;
475 	if (rf->msix_count <= num_online_cpus())
476 		rf->msix_shared = true;
477 
478 	pmsix = rf->msix_entries;
479 	for (i = 0, ceq_idx = 0; i < rf->msix_count; i++, iw_qvinfo++) {
480 		rf->iw_msixtbl[i].idx = pmsix->entry;
481 		rf->iw_msixtbl[i].irq = pmsix->vector;
482 		rf->iw_msixtbl[i].cpu_affinity = ceq_idx;
483 		if (!i) {
484 			iw_qvinfo->aeq_idx = 0;
485 			if (rf->msix_shared)
486 				iw_qvinfo->ceq_idx = ceq_idx++;
487 			else
488 				iw_qvinfo->ceq_idx = IRDMA_Q_INVALID_IDX;
489 		} else {
490 			iw_qvinfo->aeq_idx = IRDMA_Q_INVALID_IDX;
491 			iw_qvinfo->ceq_idx = ceq_idx++;
492 		}
493 		iw_qvinfo->itr_idx = 3;
494 		iw_qvinfo->v_idx = rf->iw_msixtbl[i].idx;
495 		pmsix++;
496 	}
497 
498 	return 0;
499 }
500 
501 /**
502  * irdma_irq_handler - interrupt handler for aeq and ceq0
503  * @irq: Interrupt request number
504  * @data: RDMA PCI function
505  */
506 static irqreturn_t irdma_irq_handler(int irq, void *data)
507 {
508 	struct irdma_pci_f *rf = data;
509 
510 	tasklet_schedule(&rf->dpc_tasklet);
511 
512 	return IRQ_HANDLED;
513 }
514 
515 /**
516  * irdma_ceq_handler - interrupt handler for ceq
517  * @irq: interrupt request number
518  * @data: ceq pointer
519  */
520 static irqreturn_t irdma_ceq_handler(int irq, void *data)
521 {
522 	struct irdma_ceq *iwceq = data;
523 
524 	if (iwceq->irq != irq)
525 		ibdev_err(to_ibdev(&iwceq->rf->sc_dev), "expected irq = %d received irq = %d\n",
526 			  iwceq->irq, irq);
527 	tasklet_schedule(&iwceq->dpc_tasklet);
528 
529 	return IRQ_HANDLED;
530 }
531 
532 /**
533  * irdma_destroy_irq - destroy device interrupts
534  * @rf: RDMA PCI function
535  * @msix_vec: msix vector to disable irq
536  * @dev_id: parameter to pass to free_irq (used during irq setup)
537  *
538  * The function is called when destroying aeq/ceq
539  */
540 static void irdma_destroy_irq(struct irdma_pci_f *rf,
541 			      struct irdma_msix_vector *msix_vec, void *dev_id)
542 {
543 	struct irdma_sc_dev *dev = &rf->sc_dev;
544 
545 	dev->irq_ops->irdma_dis_irq(dev, msix_vec->idx);
546 	irq_update_affinity_hint(msix_vec->irq, NULL);
547 	free_irq(msix_vec->irq, dev_id);
548 }
549 
550 /**
551  * irdma_destroy_cqp  - destroy control qp
552  * @rf: RDMA PCI function
553  * @free_hwcqp: 1 if hw cqp should be freed
554  *
555  * Issue destroy cqp request and
556  * free the resources associated with the cqp
557  */
558 static void irdma_destroy_cqp(struct irdma_pci_f *rf, bool free_hwcqp)
559 {
560 	struct irdma_sc_dev *dev = &rf->sc_dev;
561 	struct irdma_cqp *cqp = &rf->cqp;
562 	int status = 0;
563 
564 	if (rf->cqp_cmpl_wq)
565 		destroy_workqueue(rf->cqp_cmpl_wq);
566 	if (free_hwcqp)
567 		status = irdma_sc_cqp_destroy(dev->cqp);
568 	if (status)
569 		ibdev_dbg(to_ibdev(dev), "ERR: Destroy CQP failed %d\n", status);
570 
571 	irdma_cleanup_pending_cqp_op(rf);
572 	dma_free_coherent(dev->hw->device, cqp->sq.size, cqp->sq.va,
573 			  cqp->sq.pa);
574 	cqp->sq.va = NULL;
575 	kfree(cqp->scratch_array);
576 	cqp->scratch_array = NULL;
577 	kfree(cqp->cqp_requests);
578 	cqp->cqp_requests = NULL;
579 }
580 
581 static void irdma_destroy_virt_aeq(struct irdma_pci_f *rf)
582 {
583 	struct irdma_aeq *aeq = &rf->aeq;
584 	u32 pg_cnt = DIV_ROUND_UP(aeq->mem.size, PAGE_SIZE);
585 	dma_addr_t *pg_arr = (dma_addr_t *)aeq->palloc.level1.addr;
586 
587 	irdma_unmap_vm_page_list(&rf->hw, pg_arr, pg_cnt);
588 	irdma_free_pble(rf->pble_rsrc, &aeq->palloc);
589 	vfree(aeq->mem.va);
590 }
591 
592 /**
593  * irdma_destroy_aeq - destroy aeq
594  * @rf: RDMA PCI function
595  *
596  * Issue a destroy aeq request and
597  * free the resources associated with the aeq
598  * The function is called during driver unload
599  */
600 static void irdma_destroy_aeq(struct irdma_pci_f *rf)
601 {
602 	struct irdma_sc_dev *dev = &rf->sc_dev;
603 	struct irdma_aeq *aeq = &rf->aeq;
604 	int status = -EBUSY;
605 
606 	if (!rf->msix_shared) {
607 		rf->sc_dev.irq_ops->irdma_cfg_aeq(&rf->sc_dev, rf->iw_msixtbl->idx, false);
608 		irdma_destroy_irq(rf, rf->iw_msixtbl, rf);
609 	}
610 	if (rf->reset)
611 		goto exit;
612 
613 	aeq->sc_aeq.size = 0;
614 	status = irdma_cqp_aeq_cmd(dev, &aeq->sc_aeq, IRDMA_OP_AEQ_DESTROY);
615 	if (status)
616 		ibdev_dbg(to_ibdev(dev), "ERR: Destroy AEQ failed %d\n", status);
617 
618 exit:
619 	if (aeq->virtual_map) {
620 		irdma_destroy_virt_aeq(rf);
621 	} else {
622 		dma_free_coherent(dev->hw->device, aeq->mem.size, aeq->mem.va,
623 				  aeq->mem.pa);
624 		aeq->mem.va = NULL;
625 	}
626 }
627 
628 /**
629  * irdma_destroy_ceq - destroy ceq
630  * @rf: RDMA PCI function
631  * @iwceq: ceq to be destroyed
632  *
633  * Issue a destroy ceq request and
634  * free the resources associated with the ceq
635  */
636 static void irdma_destroy_ceq(struct irdma_pci_f *rf, struct irdma_ceq *iwceq)
637 {
638 	struct irdma_sc_dev *dev = &rf->sc_dev;
639 	int status;
640 
641 	if (rf->reset)
642 		goto exit;
643 
644 	status = irdma_sc_ceq_destroy(&iwceq->sc_ceq, 0, 1);
645 	if (status) {
646 		ibdev_dbg(to_ibdev(dev), "ERR: CEQ destroy command failed %d\n", status);
647 		goto exit;
648 	}
649 
650 	status = irdma_sc_cceq_destroy_done(&iwceq->sc_ceq);
651 	if (status)
652 		ibdev_dbg(to_ibdev(dev), "ERR: CEQ destroy completion failed %d\n",
653 			  status);
654 exit:
655 	dma_free_coherent(dev->hw->device, iwceq->mem.size, iwceq->mem.va,
656 			  iwceq->mem.pa);
657 	iwceq->mem.va = NULL;
658 }
659 
660 /**
661  * irdma_del_ceq_0 - destroy ceq 0
662  * @rf: RDMA PCI function
663  *
664  * Disable the ceq 0 interrupt and destroy the ceq 0
665  */
666 static void irdma_del_ceq_0(struct irdma_pci_f *rf)
667 {
668 	struct irdma_ceq *iwceq = rf->ceqlist;
669 	struct irdma_msix_vector *msix_vec;
670 
671 	if (rf->msix_shared) {
672 		msix_vec = &rf->iw_msixtbl[0];
673 		rf->sc_dev.irq_ops->irdma_cfg_ceq(&rf->sc_dev,
674 						  msix_vec->ceq_id,
675 						  msix_vec->idx, false);
676 		irdma_destroy_irq(rf, msix_vec, rf);
677 	} else {
678 		msix_vec = &rf->iw_msixtbl[1];
679 		irdma_destroy_irq(rf, msix_vec, iwceq);
680 	}
681 
682 	irdma_destroy_ceq(rf, iwceq);
683 	rf->sc_dev.ceq_valid = false;
684 	rf->ceqs_count = 0;
685 }
686 
687 /**
688  * irdma_del_ceqs - destroy all ceq's except CEQ 0
689  * @rf: RDMA PCI function
690  *
691  * Go through all of the device ceq's, except 0, and for each
692  * ceq disable the ceq interrupt and destroy the ceq
693  */
694 static void irdma_del_ceqs(struct irdma_pci_f *rf)
695 {
696 	struct irdma_ceq *iwceq = &rf->ceqlist[1];
697 	struct irdma_msix_vector *msix_vec;
698 	u32 i = 0;
699 
700 	if (rf->msix_shared)
701 		msix_vec = &rf->iw_msixtbl[1];
702 	else
703 		msix_vec = &rf->iw_msixtbl[2];
704 
705 	for (i = 1; i < rf->ceqs_count; i++, msix_vec++, iwceq++) {
706 		rf->sc_dev.irq_ops->irdma_cfg_ceq(&rf->sc_dev, msix_vec->ceq_id,
707 						  msix_vec->idx, false);
708 		irdma_destroy_irq(rf, msix_vec, iwceq);
709 		irdma_cqp_ceq_cmd(&rf->sc_dev, &iwceq->sc_ceq,
710 				  IRDMA_OP_CEQ_DESTROY);
711 		dma_free_coherent(rf->sc_dev.hw->device, iwceq->mem.size,
712 				  iwceq->mem.va, iwceq->mem.pa);
713 		iwceq->mem.va = NULL;
714 	}
715 	rf->ceqs_count = 1;
716 }
717 
718 /**
719  * irdma_destroy_ccq - destroy control cq
720  * @rf: RDMA PCI function
721  *
722  * Issue destroy ccq request and
723  * free the resources associated with the ccq
724  */
725 static void irdma_destroy_ccq(struct irdma_pci_f *rf)
726 {
727 	struct irdma_sc_dev *dev = &rf->sc_dev;
728 	struct irdma_ccq *ccq = &rf->ccq;
729 	int status = 0;
730 
731 	if (!rf->reset)
732 		status = irdma_sc_ccq_destroy(dev->ccq, 0, true);
733 	if (status)
734 		ibdev_dbg(to_ibdev(dev), "ERR: CCQ destroy failed %d\n", status);
735 	dma_free_coherent(dev->hw->device, ccq->mem_cq.size, ccq->mem_cq.va,
736 			  ccq->mem_cq.pa);
737 	ccq->mem_cq.va = NULL;
738 }
739 
740 /**
741  * irdma_close_hmc_objects_type - delete hmc objects of a given type
742  * @dev: iwarp device
743  * @obj_type: the hmc object type to be deleted
744  * @hmc_info: host memory info struct
745  * @privileged: permission to close HMC objects
746  * @reset: true if called before reset
747  */
748 static void irdma_close_hmc_objects_type(struct irdma_sc_dev *dev,
749 					 enum irdma_hmc_rsrc_type obj_type,
750 					 struct irdma_hmc_info *hmc_info,
751 					 bool privileged, bool reset)
752 {
753 	struct irdma_hmc_del_obj_info info = {};
754 
755 	info.hmc_info = hmc_info;
756 	info.rsrc_type = obj_type;
757 	info.count = hmc_info->hmc_obj[obj_type].cnt;
758 	info.privileged = privileged;
759 	if (irdma_sc_del_hmc_obj(dev, &info, reset))
760 		ibdev_dbg(to_ibdev(dev), "ERR: del HMC obj of type %d failed\n",
761 			  obj_type);
762 }
763 
764 /**
765  * irdma_del_hmc_objects - remove all device hmc objects
766  * @dev: iwarp device
767  * @hmc_info: hmc_info to free
768  * @privileged: permission to delete HMC objects
769  * @reset: true if called before reset
770  * @vers: hardware version
771  */
772 static void irdma_del_hmc_objects(struct irdma_sc_dev *dev,
773 				  struct irdma_hmc_info *hmc_info, bool privileged,
774 				  bool reset, enum irdma_vers vers)
775 {
776 	unsigned int i;
777 
778 	for (i = 0; i < IW_HMC_OBJ_TYPE_NUM; i++) {
779 		if (dev->hmc_info->hmc_obj[iw_hmc_obj_types[i]].cnt)
780 			irdma_close_hmc_objects_type(dev, iw_hmc_obj_types[i],
781 						     hmc_info, privileged, reset);
782 		if (vers == IRDMA_GEN_1 && i == IRDMA_HMC_IW_TIMER)
783 			break;
784 	}
785 }
786 
787 /**
788  * irdma_create_hmc_obj_type - create hmc object of a given type
789  * @dev: hardware control device structure
790  * @info: information for the hmc object to create
791  */
792 static int irdma_create_hmc_obj_type(struct irdma_sc_dev *dev,
793 				     struct irdma_hmc_create_obj_info *info)
794 {
795 	return irdma_sc_create_hmc_obj(dev, info);
796 }
797 
798 /**
799  * irdma_create_hmc_objs - create all hmc objects for the device
800  * @rf: RDMA PCI function
801  * @privileged: permission to create HMC objects
802  * @vers: HW version
803  *
804  * Create the device hmc objects and allocate hmc pages
805  * Return 0 if successful, otherwise clean up and return error
806  */
807 static int irdma_create_hmc_objs(struct irdma_pci_f *rf, bool privileged,
808 				 enum irdma_vers vers)
809 {
810 	struct irdma_sc_dev *dev = &rf->sc_dev;
811 	struct irdma_hmc_create_obj_info info = {};
812 	int i, status = 0;
813 
814 	info.hmc_info = dev->hmc_info;
815 	info.privileged = privileged;
816 	info.entry_type = rf->sd_type;
817 
818 	for (i = 0; i < IW_HMC_OBJ_TYPE_NUM; i++) {
819 		if (dev->hmc_info->hmc_obj[iw_hmc_obj_types[i]].cnt) {
820 			info.rsrc_type = iw_hmc_obj_types[i];
821 			info.count = dev->hmc_info->hmc_obj[info.rsrc_type].cnt;
822 			info.add_sd_cnt = 0;
823 			status = irdma_create_hmc_obj_type(dev, &info);
824 			if (status) {
825 				ibdev_dbg(to_ibdev(dev),
826 					  "ERR: create obj type %d status = %d\n",
827 					  iw_hmc_obj_types[i], status);
828 				break;
829 			}
830 		}
831 		if (vers == IRDMA_GEN_1 && i == IRDMA_HMC_IW_TIMER)
832 			break;
833 	}
834 
835 	if (!status)
836 		return irdma_sc_static_hmc_pages_allocated(dev->cqp, 0, dev->hmc_fn_id,
837 							   true, true);
838 
839 	while (i) {
840 		i--;
841 		/* destroy the hmc objects of a given type */
842 		if (dev->hmc_info->hmc_obj[iw_hmc_obj_types[i]].cnt)
843 			irdma_close_hmc_objects_type(dev, iw_hmc_obj_types[i],
844 						     dev->hmc_info, privileged,
845 						     false);
846 	}
847 
848 	return status;
849 }
850 
851 /**
852  * irdma_obj_aligned_mem - get aligned memory from device allocated memory
853  * @rf: RDMA PCI function
854  * @memptr: points to the memory addresses
855  * @size: size of memory needed
856  * @mask: mask for the aligned memory
857  *
858  * Get aligned memory of the requested size and
859  * update the memptr to point to the new aligned memory
860  * Return 0 if successful, otherwise return no memory error
861  */
862 static int irdma_obj_aligned_mem(struct irdma_pci_f *rf,
863 				 struct irdma_dma_mem *memptr, u32 size,
864 				 u32 mask)
865 {
866 	unsigned long va, newva;
867 	unsigned long extra;
868 
869 	va = (unsigned long)rf->obj_next.va;
870 	newva = va;
871 	if (mask)
872 		newva = ALIGN(va, (unsigned long)mask + 1ULL);
873 	extra = newva - va;
874 	memptr->va = (u8 *)va + extra;
875 	memptr->pa = rf->obj_next.pa + extra;
876 	memptr->size = size;
877 	if (((u8 *)memptr->va + size) > ((u8 *)rf->obj_mem.va + rf->obj_mem.size))
878 		return -ENOMEM;
879 
880 	rf->obj_next.va = (u8 *)memptr->va + size;
881 	rf->obj_next.pa = memptr->pa + size;
882 
883 	return 0;
884 }
885 
886 /**
887  * irdma_create_cqp - create control qp
888  * @rf: RDMA PCI function
889  *
890  * Return 0, if the cqp and all the resources associated with it
891  * are successfully created, otherwise return error
892  */
893 static int irdma_create_cqp(struct irdma_pci_f *rf)
894 {
895 	u32 sqsize = IRDMA_CQP_SW_SQSIZE_2048;
896 	struct irdma_dma_mem mem;
897 	struct irdma_sc_dev *dev = &rf->sc_dev;
898 	struct irdma_cqp_init_info cqp_init_info = {};
899 	struct irdma_cqp *cqp = &rf->cqp;
900 	u16 maj_err, min_err;
901 	int i, status;
902 
903 	cqp->cqp_requests = kcalloc(sqsize, sizeof(*cqp->cqp_requests), GFP_KERNEL);
904 	if (!cqp->cqp_requests)
905 		return -ENOMEM;
906 
907 	cqp->scratch_array = kcalloc(sqsize, sizeof(*cqp->scratch_array), GFP_KERNEL);
908 	if (!cqp->scratch_array) {
909 		kfree(cqp->cqp_requests);
910 		return -ENOMEM;
911 	}
912 
913 	dev->cqp = &cqp->sc_cqp;
914 	dev->cqp->dev = dev;
915 	cqp->sq.size = ALIGN(sizeof(struct irdma_cqp_sq_wqe) * sqsize,
916 			     IRDMA_CQP_ALIGNMENT);
917 	cqp->sq.va = dma_alloc_coherent(dev->hw->device, cqp->sq.size,
918 					&cqp->sq.pa, GFP_KERNEL);
919 	if (!cqp->sq.va) {
920 		kfree(cqp->scratch_array);
921 		kfree(cqp->cqp_requests);
922 		return -ENOMEM;
923 	}
924 
925 	status = irdma_obj_aligned_mem(rf, &mem, sizeof(struct irdma_cqp_ctx),
926 				       IRDMA_HOST_CTX_ALIGNMENT_M);
927 	if (status)
928 		goto exit;
929 
930 	dev->cqp->host_ctx_pa = mem.pa;
931 	dev->cqp->host_ctx = mem.va;
932 	/* populate the cqp init info */
933 	cqp_init_info.dev = dev;
934 	cqp_init_info.sq_size = sqsize;
935 	cqp_init_info.sq = cqp->sq.va;
936 	cqp_init_info.sq_pa = cqp->sq.pa;
937 	cqp_init_info.host_ctx_pa = mem.pa;
938 	cqp_init_info.host_ctx = mem.va;
939 	cqp_init_info.hmc_profile = rf->rsrc_profile;
940 	cqp_init_info.scratch_array = cqp->scratch_array;
941 	cqp_init_info.protocol_used = rf->protocol_used;
942 
943 	switch (rf->rdma_ver) {
944 	case IRDMA_GEN_1:
945 		cqp_init_info.hw_maj_ver = IRDMA_CQPHC_HW_MAJVER_GEN_1;
946 		break;
947 	case IRDMA_GEN_2:
948 		cqp_init_info.hw_maj_ver = IRDMA_CQPHC_HW_MAJVER_GEN_2;
949 		break;
950 	}
951 	status = irdma_sc_cqp_init(dev->cqp, &cqp_init_info);
952 	if (status) {
953 		ibdev_dbg(to_ibdev(dev), "ERR: cqp init status %d\n", status);
954 		goto exit;
955 	}
956 
957 	spin_lock_init(&cqp->req_lock);
958 	spin_lock_init(&cqp->compl_lock);
959 
960 	status = irdma_sc_cqp_create(dev->cqp, &maj_err, &min_err);
961 	if (status) {
962 		ibdev_dbg(to_ibdev(dev),
963 			  "ERR: cqp create failed - status %d maj_err %d min_err %d\n",
964 			  status, maj_err, min_err);
965 		goto exit;
966 	}
967 
968 	INIT_LIST_HEAD(&cqp->cqp_avail_reqs);
969 	INIT_LIST_HEAD(&cqp->cqp_pending_reqs);
970 
971 	/* init the waitqueue of the cqp_requests and add them to the list */
972 	for (i = 0; i < sqsize; i++) {
973 		init_waitqueue_head(&cqp->cqp_requests[i].waitq);
974 		list_add_tail(&cqp->cqp_requests[i].list, &cqp->cqp_avail_reqs);
975 	}
976 	init_waitqueue_head(&cqp->remove_wq);
977 	return 0;
978 
979 exit:
980 	irdma_destroy_cqp(rf, false);
981 
982 	return status;
983 }
984 
985 /**
986  * irdma_create_ccq - create control cq
987  * @rf: RDMA PCI function
988  *
989  * Return 0, if the ccq and the resources associated with it
990  * are successfully created, otherwise return error
991  */
992 static int irdma_create_ccq(struct irdma_pci_f *rf)
993 {
994 	struct irdma_sc_dev *dev = &rf->sc_dev;
995 	struct irdma_ccq_init_info info = {};
996 	struct irdma_ccq *ccq = &rf->ccq;
997 	int status;
998 
999 	dev->ccq = &ccq->sc_cq;
1000 	dev->ccq->dev = dev;
1001 	info.dev = dev;
1002 	ccq->shadow_area.size = sizeof(struct irdma_cq_shadow_area);
1003 	ccq->mem_cq.size = ALIGN(sizeof(struct irdma_cqe) * IW_CCQ_SIZE,
1004 				 IRDMA_CQ0_ALIGNMENT);
1005 	ccq->mem_cq.va = dma_alloc_coherent(dev->hw->device, ccq->mem_cq.size,
1006 					    &ccq->mem_cq.pa, GFP_KERNEL);
1007 	if (!ccq->mem_cq.va)
1008 		return -ENOMEM;
1009 
1010 	status = irdma_obj_aligned_mem(rf, &ccq->shadow_area,
1011 				       ccq->shadow_area.size,
1012 				       IRDMA_SHADOWAREA_M);
1013 	if (status)
1014 		goto exit;
1015 
1016 	ccq->sc_cq.back_cq = ccq;
1017 	/* populate the ccq init info */
1018 	info.cq_base = ccq->mem_cq.va;
1019 	info.cq_pa = ccq->mem_cq.pa;
1020 	info.num_elem = IW_CCQ_SIZE;
1021 	info.shadow_area = ccq->shadow_area.va;
1022 	info.shadow_area_pa = ccq->shadow_area.pa;
1023 	info.ceqe_mask = false;
1024 	info.ceq_id_valid = true;
1025 	info.shadow_read_threshold = 16;
1026 	info.vsi = &rf->default_vsi;
1027 	status = irdma_sc_ccq_init(dev->ccq, &info);
1028 	if (!status)
1029 		status = irdma_sc_ccq_create(dev->ccq, 0, true, true);
1030 exit:
1031 	if (status) {
1032 		dma_free_coherent(dev->hw->device, ccq->mem_cq.size,
1033 				  ccq->mem_cq.va, ccq->mem_cq.pa);
1034 		ccq->mem_cq.va = NULL;
1035 	}
1036 
1037 	return status;
1038 }
1039 
1040 /**
1041  * irdma_alloc_set_mac - set up a mac address table entry
1042  * @iwdev: irdma device
1043  *
1044  * Allocate a mac ip entry and add it to the hw table Return 0
1045  * if successful, otherwise return error
1046  */
1047 static int irdma_alloc_set_mac(struct irdma_device *iwdev)
1048 {
1049 	int status;
1050 
1051 	status = irdma_alloc_local_mac_entry(iwdev->rf,
1052 					     &iwdev->mac_ip_table_idx);
1053 	if (!status) {
1054 		status = irdma_add_local_mac_entry(iwdev->rf,
1055 						   (const u8 *)iwdev->netdev->dev_addr,
1056 						   (u8)iwdev->mac_ip_table_idx);
1057 		if (status)
1058 			irdma_del_local_mac_entry(iwdev->rf,
1059 						  (u8)iwdev->mac_ip_table_idx);
1060 	}
1061 	return status;
1062 }
1063 
1064 /**
1065  * irdma_cfg_ceq_vector - set up the msix interrupt vector for
1066  * ceq
1067  * @rf: RDMA PCI function
1068  * @iwceq: ceq associated with the vector
1069  * @ceq_id: the id number of the iwceq
1070  * @msix_vec: interrupt vector information
1071  *
1072  * Allocate interrupt resources and enable irq handling
1073  * Return 0 if successful, otherwise return error
1074  */
1075 static int irdma_cfg_ceq_vector(struct irdma_pci_f *rf, struct irdma_ceq *iwceq,
1076 				u32 ceq_id, struct irdma_msix_vector *msix_vec)
1077 {
1078 	int status;
1079 
1080 	if (rf->msix_shared && !ceq_id) {
1081 		tasklet_setup(&rf->dpc_tasklet, irdma_dpc);
1082 		status = request_irq(msix_vec->irq, irdma_irq_handler, 0,
1083 				     "AEQCEQ", rf);
1084 	} else {
1085 		tasklet_setup(&iwceq->dpc_tasklet, irdma_ceq_dpc);
1086 
1087 		status = request_irq(msix_vec->irq, irdma_ceq_handler, 0,
1088 				     "CEQ", iwceq);
1089 	}
1090 	cpumask_clear(&msix_vec->mask);
1091 	cpumask_set_cpu(msix_vec->cpu_affinity, &msix_vec->mask);
1092 	irq_update_affinity_hint(msix_vec->irq, &msix_vec->mask);
1093 	if (status) {
1094 		ibdev_dbg(&rf->iwdev->ibdev, "ERR: ceq irq config fail\n");
1095 		return status;
1096 	}
1097 
1098 	msix_vec->ceq_id = ceq_id;
1099 	rf->sc_dev.irq_ops->irdma_cfg_ceq(&rf->sc_dev, ceq_id, msix_vec->idx, true);
1100 
1101 	return 0;
1102 }
1103 
1104 /**
1105  * irdma_cfg_aeq_vector - set up the msix vector for aeq
1106  * @rf: RDMA PCI function
1107  *
1108  * Allocate interrupt resources and enable irq handling
1109  * Return 0 if successful, otherwise return error
1110  */
1111 static int irdma_cfg_aeq_vector(struct irdma_pci_f *rf)
1112 {
1113 	struct irdma_msix_vector *msix_vec = rf->iw_msixtbl;
1114 	u32 ret = 0;
1115 
1116 	if (!rf->msix_shared) {
1117 		tasklet_setup(&rf->dpc_tasklet, irdma_dpc);
1118 		ret = request_irq(msix_vec->irq, irdma_irq_handler, 0,
1119 				  "irdma", rf);
1120 	}
1121 	if (ret) {
1122 		ibdev_dbg(&rf->iwdev->ibdev, "ERR: aeq irq config fail\n");
1123 		return -EINVAL;
1124 	}
1125 
1126 	rf->sc_dev.irq_ops->irdma_cfg_aeq(&rf->sc_dev, msix_vec->idx, true);
1127 
1128 	return 0;
1129 }
1130 
1131 /**
1132  * irdma_create_ceq - create completion event queue
1133  * @rf: RDMA PCI function
1134  * @iwceq: pointer to the ceq resources to be created
1135  * @ceq_id: the id number of the iwceq
1136  * @vsi: SC vsi struct
1137  *
1138  * Return 0, if the ceq and the resources associated with it
1139  * are successfully created, otherwise return error
1140  */
1141 static int irdma_create_ceq(struct irdma_pci_f *rf, struct irdma_ceq *iwceq,
1142 			    u32 ceq_id, struct irdma_sc_vsi *vsi)
1143 {
1144 	int status;
1145 	struct irdma_ceq_init_info info = {};
1146 	struct irdma_sc_dev *dev = &rf->sc_dev;
1147 	u64 scratch;
1148 	u32 ceq_size;
1149 
1150 	info.ceq_id = ceq_id;
1151 	iwceq->rf = rf;
1152 	ceq_size = min(rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_CQ].cnt,
1153 		       dev->hw_attrs.max_hw_ceq_size);
1154 	iwceq->mem.size = ALIGN(sizeof(struct irdma_ceqe) * ceq_size,
1155 				IRDMA_CEQ_ALIGNMENT);
1156 	iwceq->mem.va = dma_alloc_coherent(dev->hw->device, iwceq->mem.size,
1157 					   &iwceq->mem.pa, GFP_KERNEL);
1158 	if (!iwceq->mem.va)
1159 		return -ENOMEM;
1160 
1161 	info.ceq_id = ceq_id;
1162 	info.ceqe_base = iwceq->mem.va;
1163 	info.ceqe_pa = iwceq->mem.pa;
1164 	info.elem_cnt = ceq_size;
1165 	iwceq->sc_ceq.ceq_id = ceq_id;
1166 	info.dev = dev;
1167 	info.vsi = vsi;
1168 	scratch = (uintptr_t)&rf->cqp.sc_cqp;
1169 	status = irdma_sc_ceq_init(&iwceq->sc_ceq, &info);
1170 	if (!status) {
1171 		if (dev->ceq_valid)
1172 			status = irdma_cqp_ceq_cmd(&rf->sc_dev, &iwceq->sc_ceq,
1173 						   IRDMA_OP_CEQ_CREATE);
1174 		else
1175 			status = irdma_sc_cceq_create(&iwceq->sc_ceq, scratch);
1176 	}
1177 
1178 	if (status) {
1179 		dma_free_coherent(dev->hw->device, iwceq->mem.size,
1180 				  iwceq->mem.va, iwceq->mem.pa);
1181 		iwceq->mem.va = NULL;
1182 	}
1183 
1184 	return status;
1185 }
1186 
1187 /**
1188  * irdma_setup_ceq_0 - create CEQ 0 and it's interrupt resource
1189  * @rf: RDMA PCI function
1190  *
1191  * Allocate a list for all device completion event queues
1192  * Create the ceq 0 and configure it's msix interrupt vector
1193  * Return 0, if successfully set up, otherwise return error
1194  */
1195 static int irdma_setup_ceq_0(struct irdma_pci_f *rf)
1196 {
1197 	struct irdma_ceq *iwceq;
1198 	struct irdma_msix_vector *msix_vec;
1199 	u32 i;
1200 	int status = 0;
1201 	u32 num_ceqs;
1202 
1203 	num_ceqs = min(rf->msix_count, rf->sc_dev.hmc_fpm_misc.max_ceqs);
1204 	rf->ceqlist = kcalloc(num_ceqs, sizeof(*rf->ceqlist), GFP_KERNEL);
1205 	if (!rf->ceqlist) {
1206 		status = -ENOMEM;
1207 		goto exit;
1208 	}
1209 
1210 	iwceq = &rf->ceqlist[0];
1211 	status = irdma_create_ceq(rf, iwceq, 0, &rf->default_vsi);
1212 	if (status) {
1213 		ibdev_dbg(&rf->iwdev->ibdev, "ERR: create ceq status = %d\n",
1214 			  status);
1215 		goto exit;
1216 	}
1217 
1218 	spin_lock_init(&iwceq->ce_lock);
1219 	i = rf->msix_shared ? 0 : 1;
1220 	msix_vec = &rf->iw_msixtbl[i];
1221 	iwceq->irq = msix_vec->irq;
1222 	iwceq->msix_idx = msix_vec->idx;
1223 	status = irdma_cfg_ceq_vector(rf, iwceq, 0, msix_vec);
1224 	if (status) {
1225 		irdma_destroy_ceq(rf, iwceq);
1226 		goto exit;
1227 	}
1228 
1229 	irdma_ena_intr(&rf->sc_dev, msix_vec->idx);
1230 	rf->ceqs_count++;
1231 
1232 exit:
1233 	if (status && !rf->ceqs_count) {
1234 		kfree(rf->ceqlist);
1235 		rf->ceqlist = NULL;
1236 		return status;
1237 	}
1238 	rf->sc_dev.ceq_valid = true;
1239 
1240 	return 0;
1241 }
1242 
1243 /**
1244  * irdma_setup_ceqs - manage the device ceq's and their interrupt resources
1245  * @rf: RDMA PCI function
1246  * @vsi: VSI structure for this CEQ
1247  *
1248  * Allocate a list for all device completion event queues
1249  * Create the ceq's and configure their msix interrupt vectors
1250  * Return 0, if ceqs are successfully set up, otherwise return error
1251  */
1252 static int irdma_setup_ceqs(struct irdma_pci_f *rf, struct irdma_sc_vsi *vsi)
1253 {
1254 	u32 i;
1255 	u32 ceq_id;
1256 	struct irdma_ceq *iwceq;
1257 	struct irdma_msix_vector *msix_vec;
1258 	int status;
1259 	u32 num_ceqs;
1260 
1261 	num_ceqs = min(rf->msix_count, rf->sc_dev.hmc_fpm_misc.max_ceqs);
1262 	i = (rf->msix_shared) ? 1 : 2;
1263 	for (ceq_id = 1; i < num_ceqs; i++, ceq_id++) {
1264 		iwceq = &rf->ceqlist[ceq_id];
1265 		status = irdma_create_ceq(rf, iwceq, ceq_id, vsi);
1266 		if (status) {
1267 			ibdev_dbg(&rf->iwdev->ibdev,
1268 				  "ERR: create ceq status = %d\n", status);
1269 			goto del_ceqs;
1270 		}
1271 		spin_lock_init(&iwceq->ce_lock);
1272 		msix_vec = &rf->iw_msixtbl[i];
1273 		iwceq->irq = msix_vec->irq;
1274 		iwceq->msix_idx = msix_vec->idx;
1275 		status = irdma_cfg_ceq_vector(rf, iwceq, ceq_id, msix_vec);
1276 		if (status) {
1277 			irdma_destroy_ceq(rf, iwceq);
1278 			goto del_ceqs;
1279 		}
1280 		irdma_ena_intr(&rf->sc_dev, msix_vec->idx);
1281 		rf->ceqs_count++;
1282 	}
1283 
1284 	return 0;
1285 
1286 del_ceqs:
1287 	irdma_del_ceqs(rf);
1288 
1289 	return status;
1290 }
1291 
1292 static int irdma_create_virt_aeq(struct irdma_pci_f *rf, u32 size)
1293 {
1294 	struct irdma_aeq *aeq = &rf->aeq;
1295 	dma_addr_t *pg_arr;
1296 	u32 pg_cnt;
1297 	int status;
1298 
1299 	if (rf->rdma_ver < IRDMA_GEN_2)
1300 		return -EOPNOTSUPP;
1301 
1302 	aeq->mem.size = sizeof(struct irdma_sc_aeqe) * size;
1303 	aeq->mem.va = vzalloc(aeq->mem.size);
1304 
1305 	if (!aeq->mem.va)
1306 		return -ENOMEM;
1307 
1308 	pg_cnt = DIV_ROUND_UP(aeq->mem.size, PAGE_SIZE);
1309 	status = irdma_get_pble(rf->pble_rsrc, &aeq->palloc, pg_cnt, true);
1310 	if (status) {
1311 		vfree(aeq->mem.va);
1312 		return status;
1313 	}
1314 
1315 	pg_arr = (dma_addr_t *)aeq->palloc.level1.addr;
1316 	status = irdma_map_vm_page_list(&rf->hw, aeq->mem.va, pg_arr, pg_cnt);
1317 	if (status) {
1318 		irdma_free_pble(rf->pble_rsrc, &aeq->palloc);
1319 		vfree(aeq->mem.va);
1320 		return status;
1321 	}
1322 
1323 	return 0;
1324 }
1325 
1326 /**
1327  * irdma_create_aeq - create async event queue
1328  * @rf: RDMA PCI function
1329  *
1330  * Return 0, if the aeq and the resources associated with it
1331  * are successfully created, otherwise return error
1332  */
1333 static int irdma_create_aeq(struct irdma_pci_f *rf)
1334 {
1335 	struct irdma_aeq_init_info info = {};
1336 	struct irdma_sc_dev *dev = &rf->sc_dev;
1337 	struct irdma_aeq *aeq = &rf->aeq;
1338 	struct irdma_hmc_info *hmc_info = rf->sc_dev.hmc_info;
1339 	u32 aeq_size;
1340 	u8 multiplier = (rf->protocol_used == IRDMA_IWARP_PROTOCOL_ONLY) ? 2 : 1;
1341 	int status;
1342 
1343 	aeq_size = multiplier * hmc_info->hmc_obj[IRDMA_HMC_IW_QP].cnt +
1344 		   hmc_info->hmc_obj[IRDMA_HMC_IW_CQ].cnt;
1345 	aeq_size = min(aeq_size, dev->hw_attrs.max_hw_aeq_size);
1346 
1347 	aeq->mem.size = ALIGN(sizeof(struct irdma_sc_aeqe) * aeq_size,
1348 			      IRDMA_AEQ_ALIGNMENT);
1349 	aeq->mem.va = dma_alloc_coherent(dev->hw->device, aeq->mem.size,
1350 					 &aeq->mem.pa,
1351 					 GFP_KERNEL | __GFP_NOWARN);
1352 	if (aeq->mem.va)
1353 		goto skip_virt_aeq;
1354 
1355 	/* physically mapped aeq failed. setup virtual aeq */
1356 	status = irdma_create_virt_aeq(rf, aeq_size);
1357 	if (status)
1358 		return status;
1359 
1360 	info.virtual_map = true;
1361 	aeq->virtual_map = info.virtual_map;
1362 	info.pbl_chunk_size = 1;
1363 	info.first_pm_pbl_idx = aeq->palloc.level1.idx;
1364 
1365 skip_virt_aeq:
1366 	info.aeqe_base = aeq->mem.va;
1367 	info.aeq_elem_pa = aeq->mem.pa;
1368 	info.elem_cnt = aeq_size;
1369 	info.dev = dev;
1370 	info.msix_idx = rf->iw_msixtbl->idx;
1371 	status = irdma_sc_aeq_init(&aeq->sc_aeq, &info);
1372 	if (status)
1373 		goto err;
1374 
1375 	status = irdma_cqp_aeq_cmd(dev, &aeq->sc_aeq, IRDMA_OP_AEQ_CREATE);
1376 	if (status)
1377 		goto err;
1378 
1379 	return 0;
1380 
1381 err:
1382 	if (aeq->virtual_map) {
1383 		irdma_destroy_virt_aeq(rf);
1384 	} else {
1385 		dma_free_coherent(dev->hw->device, aeq->mem.size, aeq->mem.va,
1386 				  aeq->mem.pa);
1387 		aeq->mem.va = NULL;
1388 	}
1389 
1390 	return status;
1391 }
1392 
1393 /**
1394  * irdma_setup_aeq - set up the device aeq
1395  * @rf: RDMA PCI function
1396  *
1397  * Create the aeq and configure its msix interrupt vector
1398  * Return 0 if successful, otherwise return error
1399  */
1400 static int irdma_setup_aeq(struct irdma_pci_f *rf)
1401 {
1402 	struct irdma_sc_dev *dev = &rf->sc_dev;
1403 	int status;
1404 
1405 	status = irdma_create_aeq(rf);
1406 	if (status)
1407 		return status;
1408 
1409 	status = irdma_cfg_aeq_vector(rf);
1410 	if (status) {
1411 		irdma_destroy_aeq(rf);
1412 		return status;
1413 	}
1414 
1415 	if (!rf->msix_shared)
1416 		irdma_ena_intr(dev, rf->iw_msixtbl[0].idx);
1417 
1418 	return 0;
1419 }
1420 
1421 /**
1422  * irdma_initialize_ilq - create iwarp local queue for cm
1423  * @iwdev: irdma device
1424  *
1425  * Return 0 if successful, otherwise return error
1426  */
1427 static int irdma_initialize_ilq(struct irdma_device *iwdev)
1428 {
1429 	struct irdma_puda_rsrc_info info = {};
1430 	int status;
1431 
1432 	info.type = IRDMA_PUDA_RSRC_TYPE_ILQ;
1433 	info.cq_id = 1;
1434 	info.qp_id = 1;
1435 	info.count = 1;
1436 	info.pd_id = 1;
1437 	info.abi_ver = IRDMA_ABI_VER;
1438 	info.sq_size = min(iwdev->rf->max_qp / 2, (u32)32768);
1439 	info.rq_size = info.sq_size;
1440 	info.buf_size = 1024;
1441 	info.tx_buf_cnt = 2 * info.sq_size;
1442 	info.receive = irdma_receive_ilq;
1443 	info.xmit_complete = irdma_free_sqbuf;
1444 	status = irdma_puda_create_rsrc(&iwdev->vsi, &info);
1445 	if (status)
1446 		ibdev_dbg(&iwdev->ibdev, "ERR: ilq create fail\n");
1447 
1448 	return status;
1449 }
1450 
1451 /**
1452  * irdma_initialize_ieq - create iwarp exception queue
1453  * @iwdev: irdma device
1454  *
1455  * Return 0 if successful, otherwise return error
1456  */
1457 static int irdma_initialize_ieq(struct irdma_device *iwdev)
1458 {
1459 	struct irdma_puda_rsrc_info info = {};
1460 	int status;
1461 
1462 	info.type = IRDMA_PUDA_RSRC_TYPE_IEQ;
1463 	info.cq_id = 2;
1464 	info.qp_id = iwdev->vsi.exception_lan_q;
1465 	info.count = 1;
1466 	info.pd_id = 2;
1467 	info.abi_ver = IRDMA_ABI_VER;
1468 	info.sq_size = min(iwdev->rf->max_qp / 2, (u32)32768);
1469 	info.rq_size = info.sq_size;
1470 	info.buf_size = iwdev->vsi.mtu + IRDMA_IPV4_PAD;
1471 	info.tx_buf_cnt = 4096;
1472 	status = irdma_puda_create_rsrc(&iwdev->vsi, &info);
1473 	if (status)
1474 		ibdev_dbg(&iwdev->ibdev, "ERR: ieq create fail\n");
1475 
1476 	return status;
1477 }
1478 
1479 /**
1480  * irdma_reinitialize_ieq - destroy and re-create ieq
1481  * @vsi: VSI structure
1482  */
1483 void irdma_reinitialize_ieq(struct irdma_sc_vsi *vsi)
1484 {
1485 	struct irdma_device *iwdev = vsi->back_vsi;
1486 	struct irdma_pci_f *rf = iwdev->rf;
1487 
1488 	irdma_puda_dele_rsrc(vsi, IRDMA_PUDA_RSRC_TYPE_IEQ, false);
1489 	if (irdma_initialize_ieq(iwdev)) {
1490 		iwdev->rf->reset = true;
1491 		rf->gen_ops.request_reset(rf);
1492 	}
1493 }
1494 
1495 /**
1496  * irdma_hmc_setup - create hmc objects for the device
1497  * @rf: RDMA PCI function
1498  *
1499  * Set up the device private memory space for the number and size of
1500  * the hmc objects and create the objects
1501  * Return 0 if successful, otherwise return error
1502  */
1503 static int irdma_hmc_setup(struct irdma_pci_f *rf)
1504 {
1505 	int status;
1506 	u32 qpcnt;
1507 
1508 	qpcnt = rsrc_limits_table[rf->limits_sel].qplimit;
1509 
1510 	rf->sd_type = IRDMA_SD_TYPE_DIRECT;
1511 	status = irdma_cfg_fpm_val(&rf->sc_dev, qpcnt);
1512 	if (status)
1513 		return status;
1514 
1515 	status = irdma_create_hmc_objs(rf, true, rf->rdma_ver);
1516 
1517 	return status;
1518 }
1519 
1520 /**
1521  * irdma_del_init_mem - deallocate memory resources
1522  * @rf: RDMA PCI function
1523  */
1524 static void irdma_del_init_mem(struct irdma_pci_f *rf)
1525 {
1526 	struct irdma_sc_dev *dev = &rf->sc_dev;
1527 
1528 	kfree(dev->hmc_info->sd_table.sd_entry);
1529 	dev->hmc_info->sd_table.sd_entry = NULL;
1530 	kfree(rf->mem_rsrc);
1531 	rf->mem_rsrc = NULL;
1532 	dma_free_coherent(rf->hw.device, rf->obj_mem.size, rf->obj_mem.va,
1533 			  rf->obj_mem.pa);
1534 	rf->obj_mem.va = NULL;
1535 	if (rf->rdma_ver != IRDMA_GEN_1) {
1536 		bitmap_free(rf->allocated_ws_nodes);
1537 		rf->allocated_ws_nodes = NULL;
1538 	}
1539 	kfree(rf->ceqlist);
1540 	rf->ceqlist = NULL;
1541 	kfree(rf->iw_msixtbl);
1542 	rf->iw_msixtbl = NULL;
1543 	kfree(rf->hmc_info_mem);
1544 	rf->hmc_info_mem = NULL;
1545 }
1546 
1547 /**
1548  * irdma_initialize_dev - initialize device
1549  * @rf: RDMA PCI function
1550  *
1551  * Allocate memory for the hmc objects and initialize iwdev
1552  * Return 0 if successful, otherwise clean up the resources
1553  * and return error
1554  */
1555 static int irdma_initialize_dev(struct irdma_pci_f *rf)
1556 {
1557 	int status;
1558 	struct irdma_sc_dev *dev = &rf->sc_dev;
1559 	struct irdma_device_init_info info = {};
1560 	struct irdma_dma_mem mem;
1561 	u32 size;
1562 
1563 	size = sizeof(struct irdma_hmc_pble_rsrc) +
1564 	       sizeof(struct irdma_hmc_info) +
1565 	       (sizeof(struct irdma_hmc_obj_info) * IRDMA_HMC_IW_MAX);
1566 
1567 	rf->hmc_info_mem = kzalloc(size, GFP_KERNEL);
1568 	if (!rf->hmc_info_mem)
1569 		return -ENOMEM;
1570 
1571 	rf->pble_rsrc = (struct irdma_hmc_pble_rsrc *)rf->hmc_info_mem;
1572 	dev->hmc_info = &rf->hw.hmc;
1573 	dev->hmc_info->hmc_obj = (struct irdma_hmc_obj_info *)
1574 				 (rf->pble_rsrc + 1);
1575 
1576 	status = irdma_obj_aligned_mem(rf, &mem, IRDMA_QUERY_FPM_BUF_SIZE,
1577 				       IRDMA_FPM_QUERY_BUF_ALIGNMENT_M);
1578 	if (status)
1579 		goto error;
1580 
1581 	info.fpm_query_buf_pa = mem.pa;
1582 	info.fpm_query_buf = mem.va;
1583 
1584 	status = irdma_obj_aligned_mem(rf, &mem, IRDMA_COMMIT_FPM_BUF_SIZE,
1585 				       IRDMA_FPM_COMMIT_BUF_ALIGNMENT_M);
1586 	if (status)
1587 		goto error;
1588 
1589 	info.fpm_commit_buf_pa = mem.pa;
1590 	info.fpm_commit_buf = mem.va;
1591 
1592 	info.bar0 = rf->hw.hw_addr;
1593 	info.hmc_fn_id = rf->pf_id;
1594 	info.hw = &rf->hw;
1595 	status = irdma_sc_dev_init(rf->rdma_ver, &rf->sc_dev, &info);
1596 	if (status)
1597 		goto error;
1598 
1599 	return status;
1600 error:
1601 	kfree(rf->hmc_info_mem);
1602 	rf->hmc_info_mem = NULL;
1603 
1604 	return status;
1605 }
1606 
1607 /**
1608  * irdma_rt_deinit_hw - clean up the irdma device resources
1609  * @iwdev: irdma device
1610  *
1611  * remove the mac ip entry and ipv4/ipv6 addresses, destroy the
1612  * device queues and free the pble and the hmc objects
1613  */
1614 void irdma_rt_deinit_hw(struct irdma_device *iwdev)
1615 {
1616 	ibdev_dbg(&iwdev->ibdev, "INIT: state = %d\n", iwdev->init_state);
1617 
1618 	switch (iwdev->init_state) {
1619 	case IP_ADDR_REGISTERED:
1620 		if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_1)
1621 			irdma_del_local_mac_entry(iwdev->rf,
1622 						  (u8)iwdev->mac_ip_table_idx);
1623 		fallthrough;
1624 	case AEQ_CREATED:
1625 	case PBLE_CHUNK_MEM:
1626 	case CEQS_CREATED:
1627 	case IEQ_CREATED:
1628 		if (!iwdev->roce_mode)
1629 			irdma_puda_dele_rsrc(&iwdev->vsi, IRDMA_PUDA_RSRC_TYPE_IEQ,
1630 					     iwdev->rf->reset);
1631 		fallthrough;
1632 	case ILQ_CREATED:
1633 		if (!iwdev->roce_mode)
1634 			irdma_puda_dele_rsrc(&iwdev->vsi,
1635 					     IRDMA_PUDA_RSRC_TYPE_ILQ,
1636 					     iwdev->rf->reset);
1637 		break;
1638 	default:
1639 		ibdev_warn(&iwdev->ibdev, "bad init_state = %d\n", iwdev->init_state);
1640 		break;
1641 	}
1642 
1643 	irdma_cleanup_cm_core(&iwdev->cm_core);
1644 	if (iwdev->vsi.pestat) {
1645 		irdma_vsi_stats_free(&iwdev->vsi);
1646 		kfree(iwdev->vsi.pestat);
1647 	}
1648 	if (iwdev->cleanup_wq)
1649 		destroy_workqueue(iwdev->cleanup_wq);
1650 }
1651 
1652 static int irdma_setup_init_state(struct irdma_pci_f *rf)
1653 {
1654 	int status;
1655 
1656 	status = irdma_save_msix_info(rf);
1657 	if (status)
1658 		return status;
1659 
1660 	rf->hw.device = &rf->pcidev->dev;
1661 	rf->obj_mem.size = ALIGN(8192, IRDMA_HW_PAGE_SIZE);
1662 	rf->obj_mem.va = dma_alloc_coherent(rf->hw.device, rf->obj_mem.size,
1663 					    &rf->obj_mem.pa, GFP_KERNEL);
1664 	if (!rf->obj_mem.va) {
1665 		status = -ENOMEM;
1666 		goto clean_msixtbl;
1667 	}
1668 
1669 	rf->obj_next = rf->obj_mem;
1670 	status = irdma_initialize_dev(rf);
1671 	if (status)
1672 		goto clean_obj_mem;
1673 
1674 	return 0;
1675 
1676 clean_obj_mem:
1677 	dma_free_coherent(rf->hw.device, rf->obj_mem.size, rf->obj_mem.va,
1678 			  rf->obj_mem.pa);
1679 	rf->obj_mem.va = NULL;
1680 clean_msixtbl:
1681 	kfree(rf->iw_msixtbl);
1682 	rf->iw_msixtbl = NULL;
1683 	return status;
1684 }
1685 
1686 /**
1687  * irdma_get_used_rsrc - determine resources used internally
1688  * @iwdev: irdma device
1689  *
1690  * Called at the end of open to get all internal allocations
1691  */
1692 static void irdma_get_used_rsrc(struct irdma_device *iwdev)
1693 {
1694 	iwdev->rf->used_pds = find_first_zero_bit(iwdev->rf->allocated_pds,
1695 						 iwdev->rf->max_pd);
1696 	iwdev->rf->used_qps = find_first_zero_bit(iwdev->rf->allocated_qps,
1697 						 iwdev->rf->max_qp);
1698 	iwdev->rf->used_cqs = find_first_zero_bit(iwdev->rf->allocated_cqs,
1699 						 iwdev->rf->max_cq);
1700 	iwdev->rf->used_mrs = find_first_zero_bit(iwdev->rf->allocated_mrs,
1701 						 iwdev->rf->max_mr);
1702 }
1703 
1704 void irdma_ctrl_deinit_hw(struct irdma_pci_f *rf)
1705 {
1706 	enum init_completion_state state = rf->init_state;
1707 
1708 	rf->init_state = INVALID_STATE;
1709 	if (rf->rsrc_created) {
1710 		irdma_destroy_aeq(rf);
1711 		irdma_destroy_pble_prm(rf->pble_rsrc);
1712 		irdma_del_ceqs(rf);
1713 		rf->rsrc_created = false;
1714 	}
1715 	switch (state) {
1716 	case CEQ0_CREATED:
1717 		irdma_del_ceq_0(rf);
1718 		fallthrough;
1719 	case CCQ_CREATED:
1720 		irdma_destroy_ccq(rf);
1721 		fallthrough;
1722 	case HW_RSRC_INITIALIZED:
1723 	case HMC_OBJS_CREATED:
1724 		irdma_del_hmc_objects(&rf->sc_dev, rf->sc_dev.hmc_info, true,
1725 				      rf->reset, rf->rdma_ver);
1726 		fallthrough;
1727 	case CQP_CREATED:
1728 		irdma_destroy_cqp(rf, true);
1729 		fallthrough;
1730 	case INITIAL_STATE:
1731 		irdma_del_init_mem(rf);
1732 		break;
1733 	case INVALID_STATE:
1734 	default:
1735 		ibdev_warn(&rf->iwdev->ibdev, "bad init_state = %d\n", rf->init_state);
1736 		break;
1737 	}
1738 }
1739 
1740 /**
1741  * irdma_rt_init_hw - Initializes runtime portion of HW
1742  * @iwdev: irdma device
1743  * @l2params: qos, tc, mtu info from netdev driver
1744  *
1745  * Create device queues ILQ, IEQ, CEQs and PBLEs. Setup irdma
1746  * device resource objects.
1747  */
1748 int irdma_rt_init_hw(struct irdma_device *iwdev,
1749 		     struct irdma_l2params *l2params)
1750 {
1751 	struct irdma_pci_f *rf = iwdev->rf;
1752 	struct irdma_sc_dev *dev = &rf->sc_dev;
1753 	struct irdma_vsi_init_info vsi_info = {};
1754 	struct irdma_vsi_stats_info stats_info = {};
1755 	int status;
1756 
1757 	vsi_info.dev = dev;
1758 	vsi_info.back_vsi = iwdev;
1759 	vsi_info.params = l2params;
1760 	vsi_info.pf_data_vsi_num = iwdev->vsi_num;
1761 	vsi_info.register_qset = rf->gen_ops.register_qset;
1762 	vsi_info.unregister_qset = rf->gen_ops.unregister_qset;
1763 	vsi_info.exception_lan_q = 2;
1764 	irdma_sc_vsi_init(&iwdev->vsi, &vsi_info);
1765 
1766 	status = irdma_setup_cm_core(iwdev, rf->rdma_ver);
1767 	if (status)
1768 		return status;
1769 
1770 	stats_info.pestat = kzalloc(sizeof(*stats_info.pestat), GFP_KERNEL);
1771 	if (!stats_info.pestat) {
1772 		irdma_cleanup_cm_core(&iwdev->cm_core);
1773 		return -ENOMEM;
1774 	}
1775 	stats_info.fcn_id = dev->hmc_fn_id;
1776 	status = irdma_vsi_stats_init(&iwdev->vsi, &stats_info);
1777 	if (status) {
1778 		irdma_cleanup_cm_core(&iwdev->cm_core);
1779 		kfree(stats_info.pestat);
1780 		return status;
1781 	}
1782 
1783 	do {
1784 		if (!iwdev->roce_mode) {
1785 			status = irdma_initialize_ilq(iwdev);
1786 			if (status)
1787 				break;
1788 			iwdev->init_state = ILQ_CREATED;
1789 			status = irdma_initialize_ieq(iwdev);
1790 			if (status)
1791 				break;
1792 			iwdev->init_state = IEQ_CREATED;
1793 		}
1794 		if (!rf->rsrc_created) {
1795 			status = irdma_setup_ceqs(rf, &iwdev->vsi);
1796 			if (status)
1797 				break;
1798 
1799 			iwdev->init_state = CEQS_CREATED;
1800 
1801 			status = irdma_hmc_init_pble(&rf->sc_dev,
1802 						     rf->pble_rsrc);
1803 			if (status) {
1804 				irdma_del_ceqs(rf);
1805 				break;
1806 			}
1807 
1808 			iwdev->init_state = PBLE_CHUNK_MEM;
1809 
1810 			status = irdma_setup_aeq(rf);
1811 			if (status) {
1812 				irdma_destroy_pble_prm(rf->pble_rsrc);
1813 				irdma_del_ceqs(rf);
1814 				break;
1815 			}
1816 			iwdev->init_state = AEQ_CREATED;
1817 			rf->rsrc_created = true;
1818 		}
1819 
1820 		if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_1)
1821 			irdma_alloc_set_mac(iwdev);
1822 		irdma_add_ip(iwdev);
1823 		iwdev->init_state = IP_ADDR_REGISTERED;
1824 
1825 		/* handles asynch cleanup tasks - disconnect CM , free qp,
1826 		 * free cq bufs
1827 		 */
1828 		iwdev->cleanup_wq = alloc_workqueue("irdma-cleanup-wq",
1829 					WQ_UNBOUND, WQ_UNBOUND_MAX_ACTIVE);
1830 		if (!iwdev->cleanup_wq)
1831 			return -ENOMEM;
1832 		irdma_get_used_rsrc(iwdev);
1833 		init_waitqueue_head(&iwdev->suspend_wq);
1834 
1835 		return 0;
1836 	} while (0);
1837 
1838 	dev_err(&rf->pcidev->dev, "HW runtime init FAIL status = %d last cmpl = %d\n",
1839 		status, iwdev->init_state);
1840 	irdma_rt_deinit_hw(iwdev);
1841 
1842 	return status;
1843 }
1844 
1845 /**
1846  * irdma_ctrl_init_hw - Initializes control portion of HW
1847  * @rf: RDMA PCI function
1848  *
1849  * Create admin queues, HMC obejcts and RF resource objects
1850  */
1851 int irdma_ctrl_init_hw(struct irdma_pci_f *rf)
1852 {
1853 	struct irdma_sc_dev *dev = &rf->sc_dev;
1854 	int status;
1855 	do {
1856 		status = irdma_setup_init_state(rf);
1857 		if (status)
1858 			break;
1859 		rf->init_state = INITIAL_STATE;
1860 
1861 		status = irdma_create_cqp(rf);
1862 		if (status)
1863 			break;
1864 		rf->init_state = CQP_CREATED;
1865 
1866 		status = irdma_hmc_setup(rf);
1867 		if (status)
1868 			break;
1869 		rf->init_state = HMC_OBJS_CREATED;
1870 
1871 		status = irdma_initialize_hw_rsrc(rf);
1872 		if (status)
1873 			break;
1874 		rf->init_state = HW_RSRC_INITIALIZED;
1875 
1876 		status = irdma_create_ccq(rf);
1877 		if (status)
1878 			break;
1879 		rf->init_state = CCQ_CREATED;
1880 
1881 		dev->feature_info[IRDMA_FEATURE_FW_INFO] = IRDMA_FW_VER_DEFAULT;
1882 		if (rf->rdma_ver != IRDMA_GEN_1) {
1883 			status = irdma_get_rdma_features(dev);
1884 			if (status)
1885 				break;
1886 		}
1887 
1888 		status = irdma_setup_ceq_0(rf);
1889 		if (status)
1890 			break;
1891 		rf->init_state = CEQ0_CREATED;
1892 		/* Handles processing of CQP completions */
1893 		rf->cqp_cmpl_wq = alloc_ordered_workqueue("cqp_cmpl_wq",
1894 						WQ_HIGHPRI | WQ_UNBOUND);
1895 		if (!rf->cqp_cmpl_wq) {
1896 			status = -ENOMEM;
1897 			break;
1898 		}
1899 		INIT_WORK(&rf->cqp_cmpl_work, cqp_compl_worker);
1900 		irdma_sc_ccq_arm(dev->ccq);
1901 		return 0;
1902 	} while (0);
1903 
1904 	dev_err(&rf->pcidev->dev, "IRDMA hardware initialization FAILED init_state=%d status=%d\n",
1905 		rf->init_state, status);
1906 	irdma_ctrl_deinit_hw(rf);
1907 	return status;
1908 }
1909 
1910 /**
1911  * irdma_set_hw_rsrc - set hw memory resources.
1912  * @rf: RDMA PCI function
1913  */
1914 static void irdma_set_hw_rsrc(struct irdma_pci_f *rf)
1915 {
1916 	rf->allocated_qps = (void *)(rf->mem_rsrc +
1917 		   (sizeof(struct irdma_arp_entry) * rf->arp_table_size));
1918 	rf->allocated_cqs = &rf->allocated_qps[BITS_TO_LONGS(rf->max_qp)];
1919 	rf->allocated_mrs = &rf->allocated_cqs[BITS_TO_LONGS(rf->max_cq)];
1920 	rf->allocated_pds = &rf->allocated_mrs[BITS_TO_LONGS(rf->max_mr)];
1921 	rf->allocated_ahs = &rf->allocated_pds[BITS_TO_LONGS(rf->max_pd)];
1922 	rf->allocated_mcgs = &rf->allocated_ahs[BITS_TO_LONGS(rf->max_ah)];
1923 	rf->allocated_arps = &rf->allocated_mcgs[BITS_TO_LONGS(rf->max_mcg)];
1924 	rf->qp_table = (struct irdma_qp **)
1925 		(&rf->allocated_arps[BITS_TO_LONGS(rf->arp_table_size)]);
1926 
1927 	spin_lock_init(&rf->rsrc_lock);
1928 	spin_lock_init(&rf->arp_lock);
1929 	spin_lock_init(&rf->qptable_lock);
1930 	spin_lock_init(&rf->qh_list_lock);
1931 }
1932 
1933 /**
1934  * irdma_calc_mem_rsrc_size - calculate memory resources size.
1935  * @rf: RDMA PCI function
1936  */
1937 static u32 irdma_calc_mem_rsrc_size(struct irdma_pci_f *rf)
1938 {
1939 	u32 rsrc_size;
1940 
1941 	rsrc_size = sizeof(struct irdma_arp_entry) * rf->arp_table_size;
1942 	rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_qp);
1943 	rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_mr);
1944 	rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_cq);
1945 	rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_pd);
1946 	rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->arp_table_size);
1947 	rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_ah);
1948 	rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_mcg);
1949 	rsrc_size += sizeof(struct irdma_qp **) * rf->max_qp;
1950 
1951 	return rsrc_size;
1952 }
1953 
1954 /**
1955  * irdma_initialize_hw_rsrc - initialize hw resource tracking array
1956  * @rf: RDMA PCI function
1957  */
1958 u32 irdma_initialize_hw_rsrc(struct irdma_pci_f *rf)
1959 {
1960 	u32 rsrc_size;
1961 	u32 mrdrvbits;
1962 	u32 ret;
1963 
1964 	if (rf->rdma_ver != IRDMA_GEN_1) {
1965 		rf->allocated_ws_nodes = bitmap_zalloc(IRDMA_MAX_WS_NODES,
1966 						       GFP_KERNEL);
1967 		if (!rf->allocated_ws_nodes)
1968 			return -ENOMEM;
1969 
1970 		set_bit(0, rf->allocated_ws_nodes);
1971 		rf->max_ws_node_id = IRDMA_MAX_WS_NODES;
1972 	}
1973 	rf->max_cqe = rf->sc_dev.hw_attrs.uk_attrs.max_hw_cq_size;
1974 	rf->max_qp = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_QP].cnt;
1975 	rf->max_mr = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_MR].cnt;
1976 	rf->max_cq = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_CQ].cnt;
1977 	rf->max_pd = rf->sc_dev.hw_attrs.max_hw_pds;
1978 	rf->arp_table_size = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_ARP].cnt;
1979 	rf->max_ah = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_FSIAV].cnt;
1980 	rf->max_mcg = rf->max_qp;
1981 
1982 	rsrc_size = irdma_calc_mem_rsrc_size(rf);
1983 	rf->mem_rsrc = kzalloc(rsrc_size, GFP_KERNEL);
1984 	if (!rf->mem_rsrc) {
1985 		ret = -ENOMEM;
1986 		goto mem_rsrc_kzalloc_fail;
1987 	}
1988 
1989 	rf->arp_table = (struct irdma_arp_entry *)rf->mem_rsrc;
1990 
1991 	irdma_set_hw_rsrc(rf);
1992 
1993 	set_bit(0, rf->allocated_mrs);
1994 	set_bit(0, rf->allocated_qps);
1995 	set_bit(0, rf->allocated_cqs);
1996 	set_bit(0, rf->allocated_pds);
1997 	set_bit(0, rf->allocated_arps);
1998 	set_bit(0, rf->allocated_ahs);
1999 	set_bit(0, rf->allocated_mcgs);
2000 	set_bit(2, rf->allocated_qps); /* qp 2 IEQ */
2001 	set_bit(1, rf->allocated_qps); /* qp 1 ILQ */
2002 	set_bit(1, rf->allocated_cqs);
2003 	set_bit(1, rf->allocated_pds);
2004 	set_bit(2, rf->allocated_cqs);
2005 	set_bit(2, rf->allocated_pds);
2006 
2007 	INIT_LIST_HEAD(&rf->mc_qht_list.list);
2008 	/* stag index mask has a minimum of 14 bits */
2009 	mrdrvbits = 24 - max(get_count_order(rf->max_mr), 14);
2010 	rf->mr_stagmask = ~(((1 << mrdrvbits) - 1) << (32 - mrdrvbits));
2011 
2012 	return 0;
2013 
2014 mem_rsrc_kzalloc_fail:
2015 	bitmap_free(rf->allocated_ws_nodes);
2016 	rf->allocated_ws_nodes = NULL;
2017 
2018 	return ret;
2019 }
2020 
2021 /**
2022  * irdma_cqp_ce_handler - handle cqp completions
2023  * @rf: RDMA PCI function
2024  * @cq: cq for cqp completions
2025  */
2026 void irdma_cqp_ce_handler(struct irdma_pci_f *rf, struct irdma_sc_cq *cq)
2027 {
2028 	struct irdma_cqp_request *cqp_request;
2029 	struct irdma_sc_dev *dev = &rf->sc_dev;
2030 	u32 cqe_count = 0;
2031 	struct irdma_ccq_cqe_info info;
2032 	unsigned long flags;
2033 	int ret;
2034 
2035 	do {
2036 		memset(&info, 0, sizeof(info));
2037 		spin_lock_irqsave(&rf->cqp.compl_lock, flags);
2038 		ret = irdma_sc_ccq_get_cqe_info(cq, &info);
2039 		spin_unlock_irqrestore(&rf->cqp.compl_lock, flags);
2040 		if (ret)
2041 			break;
2042 
2043 		cqp_request = (struct irdma_cqp_request *)
2044 			      (unsigned long)info.scratch;
2045 		if (info.error && irdma_cqp_crit_err(dev, cqp_request->info.cqp_cmd,
2046 						     info.maj_err_code,
2047 						     info.min_err_code))
2048 			ibdev_err(&rf->iwdev->ibdev, "cqp opcode = 0x%x maj_err_code = 0x%x min_err_code = 0x%x\n",
2049 				  info.op_code, info.maj_err_code, info.min_err_code);
2050 		if (cqp_request) {
2051 			cqp_request->compl_info.maj_err_code = info.maj_err_code;
2052 			cqp_request->compl_info.min_err_code = info.min_err_code;
2053 			cqp_request->compl_info.op_ret_val = info.op_ret_val;
2054 			cqp_request->compl_info.error = info.error;
2055 
2056 			if (cqp_request->waiting) {
2057 				cqp_request->request_done = true;
2058 				wake_up(&cqp_request->waitq);
2059 				irdma_put_cqp_request(&rf->cqp, cqp_request);
2060 			} else {
2061 				if (cqp_request->callback_fcn)
2062 					cqp_request->callback_fcn(cqp_request);
2063 				irdma_put_cqp_request(&rf->cqp, cqp_request);
2064 			}
2065 		}
2066 
2067 		cqe_count++;
2068 	} while (1);
2069 
2070 	if (cqe_count) {
2071 		irdma_process_bh(dev);
2072 		irdma_sc_ccq_arm(cq);
2073 	}
2074 }
2075 
2076 /**
2077  * cqp_compl_worker - Handle cqp completions
2078  * @work: Pointer to work structure
2079  */
2080 void cqp_compl_worker(struct work_struct *work)
2081 {
2082 	struct irdma_pci_f *rf = container_of(work, struct irdma_pci_f,
2083 					      cqp_cmpl_work);
2084 	struct irdma_sc_cq *cq = &rf->ccq.sc_cq;
2085 
2086 	irdma_cqp_ce_handler(rf, cq);
2087 }
2088 
2089 /**
2090  * irdma_lookup_apbvt_entry - lookup hash table for an existing apbvt entry corresponding to port
2091  * @cm_core: cm's core
2092  * @port: port to identify apbvt entry
2093  */
2094 static struct irdma_apbvt_entry *irdma_lookup_apbvt_entry(struct irdma_cm_core *cm_core,
2095 							  u16 port)
2096 {
2097 	struct irdma_apbvt_entry *entry;
2098 
2099 	hash_for_each_possible(cm_core->apbvt_hash_tbl, entry, hlist, port) {
2100 		if (entry->port == port) {
2101 			entry->use_cnt++;
2102 			return entry;
2103 		}
2104 	}
2105 
2106 	return NULL;
2107 }
2108 
2109 /**
2110  * irdma_next_iw_state - modify qp state
2111  * @iwqp: iwarp qp to modify
2112  * @state: next state for qp
2113  * @del_hash: del hash
2114  * @term: term message
2115  * @termlen: length of term message
2116  */
2117 void irdma_next_iw_state(struct irdma_qp *iwqp, u8 state, u8 del_hash, u8 term,
2118 			 u8 termlen)
2119 {
2120 	struct irdma_modify_qp_info info = {};
2121 
2122 	info.next_iwarp_state = state;
2123 	info.remove_hash_idx = del_hash;
2124 	info.cq_num_valid = true;
2125 	info.arp_cache_idx_valid = true;
2126 	info.dont_send_term = true;
2127 	info.dont_send_fin = true;
2128 	info.termlen = termlen;
2129 
2130 	if (term & IRDMAQP_TERM_SEND_TERM_ONLY)
2131 		info.dont_send_term = false;
2132 	if (term & IRDMAQP_TERM_SEND_FIN_ONLY)
2133 		info.dont_send_fin = false;
2134 	if (iwqp->sc_qp.term_flags && state == IRDMA_QP_STATE_ERROR)
2135 		info.reset_tcp_conn = true;
2136 	iwqp->hw_iwarp_state = state;
2137 	irdma_hw_modify_qp(iwqp->iwdev, iwqp, &info, 0);
2138 	iwqp->iwarp_state = info.next_iwarp_state;
2139 }
2140 
2141 /**
2142  * irdma_del_local_mac_entry - remove a mac entry from the hw
2143  * table
2144  * @rf: RDMA PCI function
2145  * @idx: the index of the mac ip address to delete
2146  */
2147 void irdma_del_local_mac_entry(struct irdma_pci_f *rf, u16 idx)
2148 {
2149 	struct irdma_cqp *iwcqp = &rf->cqp;
2150 	struct irdma_cqp_request *cqp_request;
2151 	struct cqp_cmds_info *cqp_info;
2152 
2153 	cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, true);
2154 	if (!cqp_request)
2155 		return;
2156 
2157 	cqp_info = &cqp_request->info;
2158 	cqp_info->cqp_cmd = IRDMA_OP_DELETE_LOCAL_MAC_ENTRY;
2159 	cqp_info->post_sq = 1;
2160 	cqp_info->in.u.del_local_mac_entry.cqp = &iwcqp->sc_cqp;
2161 	cqp_info->in.u.del_local_mac_entry.scratch = (uintptr_t)cqp_request;
2162 	cqp_info->in.u.del_local_mac_entry.entry_idx = idx;
2163 	cqp_info->in.u.del_local_mac_entry.ignore_ref_count = 0;
2164 
2165 	irdma_handle_cqp_op(rf, cqp_request);
2166 	irdma_put_cqp_request(iwcqp, cqp_request);
2167 }
2168 
2169 /**
2170  * irdma_add_local_mac_entry - add a mac ip address entry to the
2171  * hw table
2172  * @rf: RDMA PCI function
2173  * @mac_addr: pointer to mac address
2174  * @idx: the index of the mac ip address to add
2175  */
2176 int irdma_add_local_mac_entry(struct irdma_pci_f *rf, const u8 *mac_addr, u16 idx)
2177 {
2178 	struct irdma_local_mac_entry_info *info;
2179 	struct irdma_cqp *iwcqp = &rf->cqp;
2180 	struct irdma_cqp_request *cqp_request;
2181 	struct cqp_cmds_info *cqp_info;
2182 	int status;
2183 
2184 	cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, true);
2185 	if (!cqp_request)
2186 		return -ENOMEM;
2187 
2188 	cqp_info = &cqp_request->info;
2189 	cqp_info->post_sq = 1;
2190 	info = &cqp_info->in.u.add_local_mac_entry.info;
2191 	ether_addr_copy(info->mac_addr, mac_addr);
2192 	info->entry_idx = idx;
2193 	cqp_info->in.u.add_local_mac_entry.scratch = (uintptr_t)cqp_request;
2194 	cqp_info->cqp_cmd = IRDMA_OP_ADD_LOCAL_MAC_ENTRY;
2195 	cqp_info->in.u.add_local_mac_entry.cqp = &iwcqp->sc_cqp;
2196 	cqp_info->in.u.add_local_mac_entry.scratch = (uintptr_t)cqp_request;
2197 
2198 	status = irdma_handle_cqp_op(rf, cqp_request);
2199 	irdma_put_cqp_request(iwcqp, cqp_request);
2200 
2201 	return status;
2202 }
2203 
2204 /**
2205  * irdma_alloc_local_mac_entry - allocate a mac entry
2206  * @rf: RDMA PCI function
2207  * @mac_tbl_idx: the index of the new mac address
2208  *
2209  * Allocate a mac address entry and update the mac_tbl_idx
2210  * to hold the index of the newly created mac address
2211  * Return 0 if successful, otherwise return error
2212  */
2213 int irdma_alloc_local_mac_entry(struct irdma_pci_f *rf, u16 *mac_tbl_idx)
2214 {
2215 	struct irdma_cqp *iwcqp = &rf->cqp;
2216 	struct irdma_cqp_request *cqp_request;
2217 	struct cqp_cmds_info *cqp_info;
2218 	int status = 0;
2219 
2220 	cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, true);
2221 	if (!cqp_request)
2222 		return -ENOMEM;
2223 
2224 	cqp_info = &cqp_request->info;
2225 	cqp_info->cqp_cmd = IRDMA_OP_ALLOC_LOCAL_MAC_ENTRY;
2226 	cqp_info->post_sq = 1;
2227 	cqp_info->in.u.alloc_local_mac_entry.cqp = &iwcqp->sc_cqp;
2228 	cqp_info->in.u.alloc_local_mac_entry.scratch = (uintptr_t)cqp_request;
2229 	status = irdma_handle_cqp_op(rf, cqp_request);
2230 	if (!status)
2231 		*mac_tbl_idx = (u16)cqp_request->compl_info.op_ret_val;
2232 
2233 	irdma_put_cqp_request(iwcqp, cqp_request);
2234 
2235 	return status;
2236 }
2237 
2238 /**
2239  * irdma_cqp_manage_apbvt_cmd - send cqp command manage apbvt
2240  * @iwdev: irdma device
2241  * @accel_local_port: port for apbvt
2242  * @add_port: add ordelete port
2243  */
2244 static int irdma_cqp_manage_apbvt_cmd(struct irdma_device *iwdev,
2245 				      u16 accel_local_port, bool add_port)
2246 {
2247 	struct irdma_apbvt_info *info;
2248 	struct irdma_cqp_request *cqp_request;
2249 	struct cqp_cmds_info *cqp_info;
2250 	int status;
2251 
2252 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, add_port);
2253 	if (!cqp_request)
2254 		return -ENOMEM;
2255 
2256 	cqp_info = &cqp_request->info;
2257 	info = &cqp_info->in.u.manage_apbvt_entry.info;
2258 	memset(info, 0, sizeof(*info));
2259 	info->add = add_port;
2260 	info->port = accel_local_port;
2261 	cqp_info->cqp_cmd = IRDMA_OP_MANAGE_APBVT_ENTRY;
2262 	cqp_info->post_sq = 1;
2263 	cqp_info->in.u.manage_apbvt_entry.cqp = &iwdev->rf->cqp.sc_cqp;
2264 	cqp_info->in.u.manage_apbvt_entry.scratch = (uintptr_t)cqp_request;
2265 	ibdev_dbg(&iwdev->ibdev, "DEV: %s: port=0x%04x\n",
2266 		  (!add_port) ? "DELETE" : "ADD", accel_local_port);
2267 
2268 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2269 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2270 
2271 	return status;
2272 }
2273 
2274 /**
2275  * irdma_add_apbvt - add tcp port to HW apbvt table
2276  * @iwdev: irdma device
2277  * @port: port for apbvt
2278  */
2279 struct irdma_apbvt_entry *irdma_add_apbvt(struct irdma_device *iwdev, u16 port)
2280 {
2281 	struct irdma_cm_core *cm_core = &iwdev->cm_core;
2282 	struct irdma_apbvt_entry *entry;
2283 	unsigned long flags;
2284 
2285 	spin_lock_irqsave(&cm_core->apbvt_lock, flags);
2286 	entry = irdma_lookup_apbvt_entry(cm_core, port);
2287 	if (entry) {
2288 		spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2289 		return entry;
2290 	}
2291 
2292 	entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
2293 	if (!entry) {
2294 		spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2295 		return NULL;
2296 	}
2297 
2298 	entry->port = port;
2299 	entry->use_cnt = 1;
2300 	hash_add(cm_core->apbvt_hash_tbl, &entry->hlist, entry->port);
2301 	spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2302 
2303 	if (irdma_cqp_manage_apbvt_cmd(iwdev, port, true)) {
2304 		kfree(entry);
2305 		return NULL;
2306 	}
2307 
2308 	return entry;
2309 }
2310 
2311 /**
2312  * irdma_del_apbvt - delete tcp port from HW apbvt table
2313  * @iwdev: irdma device
2314  * @entry: apbvt entry object
2315  */
2316 void irdma_del_apbvt(struct irdma_device *iwdev,
2317 		     struct irdma_apbvt_entry *entry)
2318 {
2319 	struct irdma_cm_core *cm_core = &iwdev->cm_core;
2320 	unsigned long flags;
2321 
2322 	spin_lock_irqsave(&cm_core->apbvt_lock, flags);
2323 	if (--entry->use_cnt) {
2324 		spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2325 		return;
2326 	}
2327 
2328 	hash_del(&entry->hlist);
2329 	/* apbvt_lock is held across CQP delete APBVT OP (non-waiting) to
2330 	 * protect against race where add APBVT CQP can race ahead of the delete
2331 	 * APBVT for same port.
2332 	 */
2333 	irdma_cqp_manage_apbvt_cmd(iwdev, entry->port, false);
2334 	kfree(entry);
2335 	spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2336 }
2337 
2338 /**
2339  * irdma_manage_arp_cache - manage hw arp cache
2340  * @rf: RDMA PCI function
2341  * @mac_addr: mac address ptr
2342  * @ip_addr: ip addr for arp cache
2343  * @ipv4: flag inicating IPv4
2344  * @action: add, delete or modify
2345  */
2346 void irdma_manage_arp_cache(struct irdma_pci_f *rf,
2347 			    const unsigned char *mac_addr,
2348 			    u32 *ip_addr, bool ipv4, u32 action)
2349 {
2350 	struct irdma_add_arp_cache_entry_info *info;
2351 	struct irdma_cqp_request *cqp_request;
2352 	struct cqp_cmds_info *cqp_info;
2353 	int arp_index;
2354 
2355 	arp_index = irdma_arp_table(rf, ip_addr, ipv4, mac_addr, action);
2356 	if (arp_index == -1)
2357 		return;
2358 
2359 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, false);
2360 	if (!cqp_request)
2361 		return;
2362 
2363 	cqp_info = &cqp_request->info;
2364 	if (action == IRDMA_ARP_ADD) {
2365 		cqp_info->cqp_cmd = IRDMA_OP_ADD_ARP_CACHE_ENTRY;
2366 		info = &cqp_info->in.u.add_arp_cache_entry.info;
2367 		memset(info, 0, sizeof(*info));
2368 		info->arp_index = (u16)arp_index;
2369 		info->permanent = true;
2370 		ether_addr_copy(info->mac_addr, mac_addr);
2371 		cqp_info->in.u.add_arp_cache_entry.scratch =
2372 			(uintptr_t)cqp_request;
2373 		cqp_info->in.u.add_arp_cache_entry.cqp = &rf->cqp.sc_cqp;
2374 	} else {
2375 		cqp_info->cqp_cmd = IRDMA_OP_DELETE_ARP_CACHE_ENTRY;
2376 		cqp_info->in.u.del_arp_cache_entry.scratch =
2377 			(uintptr_t)cqp_request;
2378 		cqp_info->in.u.del_arp_cache_entry.cqp = &rf->cqp.sc_cqp;
2379 		cqp_info->in.u.del_arp_cache_entry.arp_index = arp_index;
2380 	}
2381 
2382 	cqp_info->post_sq = 1;
2383 	irdma_handle_cqp_op(rf, cqp_request);
2384 	irdma_put_cqp_request(&rf->cqp, cqp_request);
2385 }
2386 
2387 /**
2388  * irdma_send_syn_cqp_callback - do syn/ack after qhash
2389  * @cqp_request: qhash cqp completion
2390  */
2391 static void irdma_send_syn_cqp_callback(struct irdma_cqp_request *cqp_request)
2392 {
2393 	struct irdma_cm_node *cm_node = cqp_request->param;
2394 
2395 	irdma_send_syn(cm_node, 1);
2396 	irdma_rem_ref_cm_node(cm_node);
2397 }
2398 
2399 /**
2400  * irdma_manage_qhash - add or modify qhash
2401  * @iwdev: irdma device
2402  * @cminfo: cm info for qhash
2403  * @etype: type (syn or quad)
2404  * @mtype: type of qhash
2405  * @cmnode: cmnode associated with connection
2406  * @wait: wait for completion
2407  */
2408 int irdma_manage_qhash(struct irdma_device *iwdev, struct irdma_cm_info *cminfo,
2409 		       enum irdma_quad_entry_type etype,
2410 		       enum irdma_quad_hash_manage_type mtype, void *cmnode,
2411 		       bool wait)
2412 {
2413 	struct irdma_qhash_table_info *info;
2414 	struct irdma_cqp *iwcqp = &iwdev->rf->cqp;
2415 	struct irdma_cqp_request *cqp_request;
2416 	struct cqp_cmds_info *cqp_info;
2417 	struct irdma_cm_node *cm_node = cmnode;
2418 	int status;
2419 
2420 	cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, wait);
2421 	if (!cqp_request)
2422 		return -ENOMEM;
2423 
2424 	cqp_info = &cqp_request->info;
2425 	info = &cqp_info->in.u.manage_qhash_table_entry.info;
2426 	memset(info, 0, sizeof(*info));
2427 	info->vsi = &iwdev->vsi;
2428 	info->manage = mtype;
2429 	info->entry_type = etype;
2430 	if (cminfo->vlan_id < VLAN_N_VID) {
2431 		info->vlan_valid = true;
2432 		info->vlan_id = cminfo->vlan_id;
2433 	} else {
2434 		info->vlan_valid = false;
2435 	}
2436 	info->ipv4_valid = cminfo->ipv4;
2437 	info->user_pri = cminfo->user_pri;
2438 	ether_addr_copy(info->mac_addr, iwdev->netdev->dev_addr);
2439 	info->qp_num = cminfo->qh_qpid;
2440 	info->dest_port = cminfo->loc_port;
2441 	info->dest_ip[0] = cminfo->loc_addr[0];
2442 	info->dest_ip[1] = cminfo->loc_addr[1];
2443 	info->dest_ip[2] = cminfo->loc_addr[2];
2444 	info->dest_ip[3] = cminfo->loc_addr[3];
2445 	if (etype == IRDMA_QHASH_TYPE_TCP_ESTABLISHED ||
2446 	    etype == IRDMA_QHASH_TYPE_UDP_UNICAST ||
2447 	    etype == IRDMA_QHASH_TYPE_UDP_MCAST ||
2448 	    etype == IRDMA_QHASH_TYPE_ROCE_MCAST ||
2449 	    etype == IRDMA_QHASH_TYPE_ROCEV2_HW) {
2450 		info->src_port = cminfo->rem_port;
2451 		info->src_ip[0] = cminfo->rem_addr[0];
2452 		info->src_ip[1] = cminfo->rem_addr[1];
2453 		info->src_ip[2] = cminfo->rem_addr[2];
2454 		info->src_ip[3] = cminfo->rem_addr[3];
2455 	}
2456 	if (cmnode) {
2457 		cqp_request->callback_fcn = irdma_send_syn_cqp_callback;
2458 		cqp_request->param = cmnode;
2459 		if (!wait)
2460 			refcount_inc(&cm_node->refcnt);
2461 	}
2462 	if (info->ipv4_valid)
2463 		ibdev_dbg(&iwdev->ibdev,
2464 			  "CM: %s caller: %pS loc_port=0x%04x rem_port=0x%04x loc_addr=%pI4 rem_addr=%pI4 mac=%pM, vlan_id=%d cm_node=%p\n",
2465 			  (!mtype) ? "DELETE" : "ADD",
2466 			  __builtin_return_address(0), info->dest_port,
2467 			  info->src_port, info->dest_ip, info->src_ip,
2468 			  info->mac_addr, cminfo->vlan_id,
2469 			  cmnode ? cmnode : NULL);
2470 	else
2471 		ibdev_dbg(&iwdev->ibdev,
2472 			  "CM: %s caller: %pS loc_port=0x%04x rem_port=0x%04x loc_addr=%pI6 rem_addr=%pI6 mac=%pM, vlan_id=%d cm_node=%p\n",
2473 			  (!mtype) ? "DELETE" : "ADD",
2474 			  __builtin_return_address(0), info->dest_port,
2475 			  info->src_port, info->dest_ip, info->src_ip,
2476 			  info->mac_addr, cminfo->vlan_id,
2477 			  cmnode ? cmnode : NULL);
2478 
2479 	cqp_info->in.u.manage_qhash_table_entry.cqp = &iwdev->rf->cqp.sc_cqp;
2480 	cqp_info->in.u.manage_qhash_table_entry.scratch = (uintptr_t)cqp_request;
2481 	cqp_info->cqp_cmd = IRDMA_OP_MANAGE_QHASH_TABLE_ENTRY;
2482 	cqp_info->post_sq = 1;
2483 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2484 	if (status && cm_node && !wait)
2485 		irdma_rem_ref_cm_node(cm_node);
2486 
2487 	irdma_put_cqp_request(iwcqp, cqp_request);
2488 
2489 	return status;
2490 }
2491 
2492 /**
2493  * irdma_hw_flush_wqes_callback - Check return code after flush
2494  * @cqp_request: qhash cqp completion
2495  */
2496 static void irdma_hw_flush_wqes_callback(struct irdma_cqp_request *cqp_request)
2497 {
2498 	struct irdma_qp_flush_info *hw_info;
2499 	struct irdma_sc_qp *qp;
2500 	struct irdma_qp *iwqp;
2501 	struct cqp_cmds_info *cqp_info;
2502 
2503 	cqp_info = &cqp_request->info;
2504 	hw_info = &cqp_info->in.u.qp_flush_wqes.info;
2505 	qp = cqp_info->in.u.qp_flush_wqes.qp;
2506 	iwqp = qp->qp_uk.back_qp;
2507 
2508 	if (cqp_request->compl_info.maj_err_code)
2509 		return;
2510 
2511 	if (hw_info->rq &&
2512 	    (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_SQ_WQE_FLUSHED ||
2513 	     cqp_request->compl_info.min_err_code == 0)) {
2514 		/* RQ WQE flush was requested but did not happen */
2515 		qp->qp_uk.rq_flush_complete = true;
2516 	}
2517 	if (hw_info->sq &&
2518 	    (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_RQ_WQE_FLUSHED ||
2519 	     cqp_request->compl_info.min_err_code == 0)) {
2520 		if (IRDMA_RING_MORE_WORK(qp->qp_uk.sq_ring)) {
2521 			ibdev_err(&iwqp->iwdev->ibdev, "Flush QP[%d] failed, SQ has more work",
2522 				  qp->qp_uk.qp_id);
2523 			irdma_ib_qp_event(iwqp, IRDMA_QP_EVENT_CATASTROPHIC);
2524 		}
2525 		qp->qp_uk.sq_flush_complete = true;
2526 	}
2527 }
2528 
2529 /**
2530  * irdma_hw_flush_wqes - flush qp's wqe
2531  * @rf: RDMA PCI function
2532  * @qp: hardware control qp
2533  * @info: info for flush
2534  * @wait: flag wait for completion
2535  */
2536 int irdma_hw_flush_wqes(struct irdma_pci_f *rf, struct irdma_sc_qp *qp,
2537 			struct irdma_qp_flush_info *info, bool wait)
2538 {
2539 	int status;
2540 	struct irdma_qp_flush_info *hw_info;
2541 	struct irdma_cqp_request *cqp_request;
2542 	struct cqp_cmds_info *cqp_info;
2543 	struct irdma_qp *iwqp = qp->qp_uk.back_qp;
2544 
2545 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, wait);
2546 	if (!cqp_request)
2547 		return -ENOMEM;
2548 
2549 	cqp_info = &cqp_request->info;
2550 	if (!wait)
2551 		cqp_request->callback_fcn = irdma_hw_flush_wqes_callback;
2552 	hw_info = &cqp_request->info.in.u.qp_flush_wqes.info;
2553 	memcpy(hw_info, info, sizeof(*hw_info));
2554 	cqp_info->cqp_cmd = IRDMA_OP_QP_FLUSH_WQES;
2555 	cqp_info->post_sq = 1;
2556 	cqp_info->in.u.qp_flush_wqes.qp = qp;
2557 	cqp_info->in.u.qp_flush_wqes.scratch = (uintptr_t)cqp_request;
2558 	status = irdma_handle_cqp_op(rf, cqp_request);
2559 	if (status) {
2560 		qp->qp_uk.sq_flush_complete = true;
2561 		qp->qp_uk.rq_flush_complete = true;
2562 		irdma_put_cqp_request(&rf->cqp, cqp_request);
2563 		return status;
2564 	}
2565 
2566 	if (!wait || cqp_request->compl_info.maj_err_code)
2567 		goto put_cqp;
2568 
2569 	if (info->rq) {
2570 		if (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_SQ_WQE_FLUSHED ||
2571 		    cqp_request->compl_info.min_err_code == 0) {
2572 			/* RQ WQE flush was requested but did not happen */
2573 			qp->qp_uk.rq_flush_complete = true;
2574 		}
2575 	}
2576 	if (info->sq) {
2577 		if (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_RQ_WQE_FLUSHED ||
2578 		    cqp_request->compl_info.min_err_code == 0) {
2579 			/*
2580 			 * Handling case where WQE is posted to empty SQ when
2581 			 * flush has not completed
2582 			 */
2583 			if (IRDMA_RING_MORE_WORK(qp->qp_uk.sq_ring)) {
2584 				struct irdma_cqp_request *new_req;
2585 
2586 				if (!qp->qp_uk.sq_flush_complete)
2587 					goto put_cqp;
2588 				qp->qp_uk.sq_flush_complete = false;
2589 				qp->flush_sq = false;
2590 
2591 				info->rq = false;
2592 				info->sq = true;
2593 				new_req = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2594 				if (!new_req) {
2595 					status = -ENOMEM;
2596 					goto put_cqp;
2597 				}
2598 				cqp_info = &new_req->info;
2599 				hw_info = &new_req->info.in.u.qp_flush_wqes.info;
2600 				memcpy(hw_info, info, sizeof(*hw_info));
2601 				cqp_info->cqp_cmd = IRDMA_OP_QP_FLUSH_WQES;
2602 				cqp_info->post_sq = 1;
2603 				cqp_info->in.u.qp_flush_wqes.qp = qp;
2604 				cqp_info->in.u.qp_flush_wqes.scratch = (uintptr_t)new_req;
2605 
2606 				status = irdma_handle_cqp_op(rf, new_req);
2607 				if (new_req->compl_info.maj_err_code ||
2608 				    new_req->compl_info.min_err_code != IRDMA_CQP_COMPL_SQ_WQE_FLUSHED ||
2609 				    status) {
2610 					ibdev_err(&iwqp->iwdev->ibdev, "fatal QP event: SQ in error but not flushed, qp: %d",
2611 						  iwqp->ibqp.qp_num);
2612 					qp->qp_uk.sq_flush_complete = false;
2613 					irdma_ib_qp_event(iwqp, IRDMA_QP_EVENT_CATASTROPHIC);
2614 				}
2615 				irdma_put_cqp_request(&rf->cqp, new_req);
2616 			} else {
2617 				/* SQ WQE flush was requested but did not happen */
2618 				qp->qp_uk.sq_flush_complete = true;
2619 			}
2620 		} else {
2621 			if (!IRDMA_RING_MORE_WORK(qp->qp_uk.sq_ring))
2622 				qp->qp_uk.sq_flush_complete = true;
2623 		}
2624 	}
2625 
2626 	ibdev_dbg(&rf->iwdev->ibdev,
2627 		  "VERBS: qp_id=%d qp_type=%d qpstate=%d ibqpstate=%d last_aeq=%d hw_iw_state=%d maj_err_code=%d min_err_code=%d\n",
2628 		  iwqp->ibqp.qp_num, rf->protocol_used, iwqp->iwarp_state,
2629 		  iwqp->ibqp_state, iwqp->last_aeq, iwqp->hw_iwarp_state,
2630 		  cqp_request->compl_info.maj_err_code,
2631 		  cqp_request->compl_info.min_err_code);
2632 put_cqp:
2633 	irdma_put_cqp_request(&rf->cqp, cqp_request);
2634 
2635 	return status;
2636 }
2637 
2638 /**
2639  * irdma_gen_ae - generate AE
2640  * @rf: RDMA PCI function
2641  * @qp: qp associated with AE
2642  * @info: info for ae
2643  * @wait: wait for completion
2644  */
2645 void irdma_gen_ae(struct irdma_pci_f *rf, struct irdma_sc_qp *qp,
2646 		  struct irdma_gen_ae_info *info, bool wait)
2647 {
2648 	struct irdma_gen_ae_info *ae_info;
2649 	struct irdma_cqp_request *cqp_request;
2650 	struct cqp_cmds_info *cqp_info;
2651 
2652 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, wait);
2653 	if (!cqp_request)
2654 		return;
2655 
2656 	cqp_info = &cqp_request->info;
2657 	ae_info = &cqp_request->info.in.u.gen_ae.info;
2658 	memcpy(ae_info, info, sizeof(*ae_info));
2659 	cqp_info->cqp_cmd = IRDMA_OP_GEN_AE;
2660 	cqp_info->post_sq = 1;
2661 	cqp_info->in.u.gen_ae.qp = qp;
2662 	cqp_info->in.u.gen_ae.scratch = (uintptr_t)cqp_request;
2663 
2664 	irdma_handle_cqp_op(rf, cqp_request);
2665 	irdma_put_cqp_request(&rf->cqp, cqp_request);
2666 }
2667 
2668 void irdma_flush_wqes(struct irdma_qp *iwqp, u32 flush_mask)
2669 {
2670 	struct irdma_qp_flush_info info = {};
2671 	struct irdma_pci_f *rf = iwqp->iwdev->rf;
2672 	u8 flush_code = iwqp->sc_qp.flush_code;
2673 
2674 	if (!(flush_mask & IRDMA_FLUSH_SQ) && !(flush_mask & IRDMA_FLUSH_RQ))
2675 		return;
2676 
2677 	/* Set flush info fields*/
2678 	info.sq = flush_mask & IRDMA_FLUSH_SQ;
2679 	info.rq = flush_mask & IRDMA_FLUSH_RQ;
2680 
2681 	/* Generate userflush errors in CQE */
2682 	info.sq_major_code = IRDMA_FLUSH_MAJOR_ERR;
2683 	info.sq_minor_code = FLUSH_GENERAL_ERR;
2684 	info.rq_major_code = IRDMA_FLUSH_MAJOR_ERR;
2685 	info.rq_minor_code = FLUSH_GENERAL_ERR;
2686 	info.userflushcode = true;
2687 
2688 	if (flush_mask & IRDMA_REFLUSH) {
2689 		if (info.sq)
2690 			iwqp->sc_qp.flush_sq = false;
2691 		if (info.rq)
2692 			iwqp->sc_qp.flush_rq = false;
2693 	} else {
2694 		if (flush_code) {
2695 			if (info.sq && iwqp->sc_qp.sq_flush_code)
2696 				info.sq_minor_code = flush_code;
2697 			if (info.rq && iwqp->sc_qp.rq_flush_code)
2698 				info.rq_minor_code = flush_code;
2699 		}
2700 		if (!iwqp->user_mode)
2701 			queue_delayed_work(iwqp->iwdev->cleanup_wq,
2702 					   &iwqp->dwork_flush,
2703 					   msecs_to_jiffies(IRDMA_FLUSH_DELAY_MS));
2704 	}
2705 
2706 	/* Issue flush */
2707 	(void)irdma_hw_flush_wqes(rf, &iwqp->sc_qp, &info,
2708 				  flush_mask & IRDMA_FLUSH_WAIT);
2709 	iwqp->flush_issued = true;
2710 }
2711