Linux 6.9-rc1
[linux-2.6-microblaze.git] / drivers / infiniband / hw / irdma / hw.c
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_PBLE,
45         IRDMA_HMC_IW_TIMER,
46         IRDMA_HMC_IW_FSIMC,
47         IRDMA_HMC_IW_FSIAV,
48         IRDMA_HMC_IW_RRF,
49         IRDMA_HMC_IW_RRFFL,
50         IRDMA_HMC_IW_HDR,
51         IRDMA_HMC_IW_MD,
52         IRDMA_HMC_IW_OOISC,
53         IRDMA_HMC_IW_OOISCFFL,
54 };
55
56 /**
57  * irdma_iwarp_ce_handler - handle iwarp completions
58  * @iwcq: iwarp cq receiving event
59  */
60 static void irdma_iwarp_ce_handler(struct irdma_sc_cq *iwcq)
61 {
62         struct irdma_cq *cq = iwcq->back_cq;
63
64         if (!cq->user_mode)
65                 atomic_set(&cq->armed, 0);
66         if (cq->ibcq.comp_handler)
67                 cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
68 }
69
70 /**
71  * irdma_puda_ce_handler - handle puda completion events
72  * @rf: RDMA PCI function
73  * @cq: puda completion q for event
74  */
75 static void irdma_puda_ce_handler(struct irdma_pci_f *rf,
76                                   struct irdma_sc_cq *cq)
77 {
78         struct irdma_sc_dev *dev = &rf->sc_dev;
79         u32 compl_error;
80         int status;
81
82         do {
83                 status = irdma_puda_poll_cmpl(dev, cq, &compl_error);
84                 if (status == -ENOENT)
85                         break;
86                 if (status) {
87                         ibdev_dbg(to_ibdev(dev), "ERR: puda status = %d\n", status);
88                         break;
89                 }
90                 if (compl_error) {
91                         ibdev_dbg(to_ibdev(dev), "ERR: puda compl_err  =0x%x\n",
92                                   compl_error);
93                         break;
94                 }
95         } while (1);
96
97         irdma_sc_ccq_arm(cq);
98 }
99
100 /**
101  * irdma_process_ceq - handle ceq for completions
102  * @rf: RDMA PCI function
103  * @ceq: ceq having cq for completion
104  */
105 static void irdma_process_ceq(struct irdma_pci_f *rf, struct irdma_ceq *ceq)
106 {
107         struct irdma_sc_dev *dev = &rf->sc_dev;
108         struct irdma_sc_ceq *sc_ceq;
109         struct irdma_sc_cq *cq;
110         unsigned long flags;
111
112         sc_ceq = &ceq->sc_ceq;
113         do {
114                 spin_lock_irqsave(&ceq->ce_lock, flags);
115                 cq = irdma_sc_process_ceq(dev, sc_ceq);
116                 if (!cq) {
117                         spin_unlock_irqrestore(&ceq->ce_lock, flags);
118                         break;
119                 }
120
121                 if (cq->cq_type == IRDMA_CQ_TYPE_IWARP)
122                         irdma_iwarp_ce_handler(cq);
123
124                 spin_unlock_irqrestore(&ceq->ce_lock, flags);
125
126                 if (cq->cq_type == IRDMA_CQ_TYPE_CQP)
127                         queue_work(rf->cqp_cmpl_wq, &rf->cqp_cmpl_work);
128                 else if (cq->cq_type == IRDMA_CQ_TYPE_ILQ ||
129                          cq->cq_type == IRDMA_CQ_TYPE_IEQ)
130                         irdma_puda_ce_handler(rf, cq);
131         } while (1);
132 }
133
134 static void irdma_set_flush_fields(struct irdma_sc_qp *qp,
135                                    struct irdma_aeqe_info *info)
136 {
137         qp->sq_flush_code = info->sq;
138         qp->rq_flush_code = info->rq;
139         qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
140
141         switch (info->ae_id) {
142         case IRDMA_AE_AMP_BOUNDS_VIOLATION:
143         case IRDMA_AE_AMP_INVALID_STAG:
144         case IRDMA_AE_AMP_RIGHTS_VIOLATION:
145         case IRDMA_AE_AMP_UNALLOCATED_STAG:
146         case IRDMA_AE_AMP_BAD_PD:
147         case IRDMA_AE_AMP_BAD_QP:
148         case IRDMA_AE_AMP_BAD_STAG_KEY:
149         case IRDMA_AE_AMP_BAD_STAG_INDEX:
150         case IRDMA_AE_AMP_TO_WRAP:
151         case IRDMA_AE_PRIV_OPERATION_DENIED:
152                 qp->flush_code = FLUSH_PROT_ERR;
153                 qp->event_type = IRDMA_QP_EVENT_ACCESS_ERR;
154                 break;
155         case IRDMA_AE_UDA_XMIT_BAD_PD:
156         case IRDMA_AE_WQE_UNEXPECTED_OPCODE:
157                 qp->flush_code = FLUSH_LOC_QP_OP_ERR;
158                 qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
159                 break;
160         case IRDMA_AE_UDA_XMIT_DGRAM_TOO_LONG:
161         case IRDMA_AE_UDA_XMIT_DGRAM_TOO_SHORT:
162         case IRDMA_AE_UDA_L4LEN_INVALID:
163         case IRDMA_AE_DDP_UBE_INVALID_MO:
164         case IRDMA_AE_DDP_UBE_DDP_MESSAGE_TOO_LONG_FOR_AVAILABLE_BUFFER:
165                 qp->flush_code = FLUSH_LOC_LEN_ERR;
166                 qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
167                 break;
168         case IRDMA_AE_AMP_INVALIDATE_NO_REMOTE_ACCESS_RIGHTS:
169         case IRDMA_AE_IB_REMOTE_ACCESS_ERROR:
170                 qp->flush_code = FLUSH_REM_ACCESS_ERR;
171                 qp->event_type = IRDMA_QP_EVENT_ACCESS_ERR;
172                 break;
173         case IRDMA_AE_LLP_SEGMENT_TOO_SMALL:
174         case IRDMA_AE_LLP_RECEIVED_MPA_CRC_ERROR:
175         case IRDMA_AE_ROCE_RSP_LENGTH_ERROR:
176         case IRDMA_AE_IB_REMOTE_OP_ERROR:
177                 qp->flush_code = FLUSH_REM_OP_ERR;
178                 qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
179                 break;
180         case IRDMA_AE_LCE_QP_CATASTROPHIC:
181                 qp->flush_code = FLUSH_FATAL_ERR;
182                 qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
183                 break;
184         case IRDMA_AE_IB_RREQ_AND_Q1_FULL:
185                 qp->flush_code = FLUSH_GENERAL_ERR;
186                 break;
187         case IRDMA_AE_LLP_TOO_MANY_RETRIES:
188                 qp->flush_code = FLUSH_RETRY_EXC_ERR;
189                 qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
190                 break;
191         case IRDMA_AE_AMP_MWBIND_INVALID_RIGHTS:
192         case IRDMA_AE_AMP_MWBIND_BIND_DISABLED:
193         case IRDMA_AE_AMP_MWBIND_INVALID_BOUNDS:
194         case IRDMA_AE_AMP_MWBIND_VALID_STAG:
195                 qp->flush_code = FLUSH_MW_BIND_ERR;
196                 qp->event_type = IRDMA_QP_EVENT_ACCESS_ERR;
197                 break;
198         case IRDMA_AE_IB_INVALID_REQUEST:
199                 qp->flush_code = FLUSH_REM_INV_REQ_ERR;
200                 qp->event_type = IRDMA_QP_EVENT_REQ_ERR;
201                 break;
202         default:
203                 qp->flush_code = FLUSH_GENERAL_ERR;
204                 qp->event_type = IRDMA_QP_EVENT_CATASTROPHIC;
205                 break;
206         }
207 }
208
209 /**
210  * irdma_process_aeq - handle aeq events
211  * @rf: RDMA PCI function
212  */
213 static void irdma_process_aeq(struct irdma_pci_f *rf)
214 {
215         struct irdma_sc_dev *dev = &rf->sc_dev;
216         struct irdma_aeq *aeq = &rf->aeq;
217         struct irdma_sc_aeq *sc_aeq = &aeq->sc_aeq;
218         struct irdma_aeqe_info aeinfo;
219         struct irdma_aeqe_info *info = &aeinfo;
220         int ret;
221         struct irdma_qp *iwqp = NULL;
222         struct irdma_cq *iwcq = NULL;
223         struct irdma_sc_qp *qp = NULL;
224         struct irdma_qp_host_ctx_info *ctx_info = NULL;
225         struct irdma_device *iwdev = rf->iwdev;
226         unsigned long flags;
227
228         u32 aeqcnt = 0;
229
230         if (!sc_aeq->size)
231                 return;
232
233         do {
234                 memset(info, 0, sizeof(*info));
235                 ret = irdma_sc_get_next_aeqe(sc_aeq, info);
236                 if (ret)
237                         break;
238
239                 aeqcnt++;
240                 ibdev_dbg(&iwdev->ibdev,
241                           "AEQ: ae_id = 0x%x bool qp=%d qp_id = %d tcp_state=%d iwarp_state=%d ae_src=%d\n",
242                           info->ae_id, info->qp, info->qp_cq_id, info->tcp_state,
243                           info->iwarp_state, info->ae_src);
244
245                 if (info->qp) {
246                         spin_lock_irqsave(&rf->qptable_lock, flags);
247                         iwqp = rf->qp_table[info->qp_cq_id];
248                         if (!iwqp) {
249                                 spin_unlock_irqrestore(&rf->qptable_lock,
250                                                        flags);
251                                 if (info->ae_id == IRDMA_AE_QP_SUSPEND_COMPLETE) {
252                                         atomic_dec(&iwdev->vsi.qp_suspend_reqs);
253                                         wake_up(&iwdev->suspend_wq);
254                                         continue;
255                                 }
256                                 ibdev_dbg(&iwdev->ibdev, "AEQ: qp_id %d is already freed\n",
257                                           info->qp_cq_id);
258                                 continue;
259                         }
260                         irdma_qp_add_ref(&iwqp->ibqp);
261                         spin_unlock_irqrestore(&rf->qptable_lock, flags);
262                         qp = &iwqp->sc_qp;
263                         spin_lock_irqsave(&iwqp->lock, flags);
264                         iwqp->hw_tcp_state = info->tcp_state;
265                         iwqp->hw_iwarp_state = info->iwarp_state;
266                         if (info->ae_id != IRDMA_AE_QP_SUSPEND_COMPLETE)
267                                 iwqp->last_aeq = info->ae_id;
268                         spin_unlock_irqrestore(&iwqp->lock, flags);
269                         ctx_info = &iwqp->ctx_info;
270                 } else {
271                         if (info->ae_id != IRDMA_AE_CQ_OPERATION_ERROR)
272                                 continue;
273                 }
274
275                 switch (info->ae_id) {
276                         struct irdma_cm_node *cm_node;
277                 case IRDMA_AE_LLP_CONNECTION_ESTABLISHED:
278                         cm_node = iwqp->cm_node;
279                         if (cm_node->accept_pend) {
280                                 atomic_dec(&cm_node->listener->pend_accepts_cnt);
281                                 cm_node->accept_pend = 0;
282                         }
283                         iwqp->rts_ae_rcvd = 1;
284                         wake_up_interruptible(&iwqp->waitq);
285                         break;
286                 case IRDMA_AE_LLP_FIN_RECEIVED:
287                 case IRDMA_AE_RDMAP_ROE_BAD_LLP_CLOSE:
288                         if (qp->term_flags)
289                                 break;
290                         if (atomic_inc_return(&iwqp->close_timer_started) == 1) {
291                                 iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSE_WAIT;
292                                 if (iwqp->hw_tcp_state == IRDMA_TCP_STATE_CLOSE_WAIT &&
293                                     iwqp->ibqp_state == IB_QPS_RTS) {
294                                         irdma_next_iw_state(iwqp,
295                                                             IRDMA_QP_STATE_CLOSING,
296                                                             0, 0, 0);
297                                         irdma_cm_disconn(iwqp);
298                                 }
299                                 irdma_schedule_cm_timer(iwqp->cm_node,
300                                                         (struct irdma_puda_buf *)iwqp,
301                                                         IRDMA_TIMER_TYPE_CLOSE,
302                                                         1, 0);
303                         }
304                         break;
305                 case IRDMA_AE_LLP_CLOSE_COMPLETE:
306                         if (qp->term_flags)
307                                 irdma_terminate_done(qp, 0);
308                         else
309                                 irdma_cm_disconn(iwqp);
310                         break;
311                 case IRDMA_AE_BAD_CLOSE:
312                 case IRDMA_AE_RESET_SENT:
313                         irdma_next_iw_state(iwqp, IRDMA_QP_STATE_ERROR, 1, 0,
314                                             0);
315                         irdma_cm_disconn(iwqp);
316                         break;
317                 case IRDMA_AE_LLP_CONNECTION_RESET:
318                         if (atomic_read(&iwqp->close_timer_started))
319                                 break;
320                         irdma_cm_disconn(iwqp);
321                         break;
322                 case IRDMA_AE_QP_SUSPEND_COMPLETE:
323                         if (iwqp->iwdev->vsi.tc_change_pending) {
324                                 if (!atomic_dec_return(&qp->vsi->qp_suspend_reqs))
325                                         wake_up(&iwqp->iwdev->suspend_wq);
326                         }
327                         if (iwqp->suspend_pending) {
328                                 iwqp->suspend_pending = false;
329                                 wake_up(&iwqp->iwdev->suspend_wq);
330                         }
331                         break;
332                 case IRDMA_AE_TERMINATE_SENT:
333                         irdma_terminate_send_fin(qp);
334                         break;
335                 case IRDMA_AE_LLP_TERMINATE_RECEIVED:
336                         irdma_terminate_received(qp, info);
337                         break;
338                 case IRDMA_AE_CQ_OPERATION_ERROR:
339                         ibdev_err(&iwdev->ibdev,
340                                   "Processing an iWARP related AE for CQ misc = 0x%04X\n",
341                                   info->ae_id);
342
343                         spin_lock_irqsave(&rf->cqtable_lock, flags);
344                         iwcq = rf->cq_table[info->qp_cq_id];
345                         if (!iwcq) {
346                                 spin_unlock_irqrestore(&rf->cqtable_lock,
347                                                        flags);
348                                 ibdev_dbg(to_ibdev(dev),
349                                           "cq_id %d is already freed\n", info->qp_cq_id);
350                                 continue;
351                         }
352                         irdma_cq_add_ref(&iwcq->ibcq);
353                         spin_unlock_irqrestore(&rf->cqtable_lock, flags);
354
355                         if (iwcq->ibcq.event_handler) {
356                                 struct ib_event ibevent;
357
358                                 ibevent.device = iwcq->ibcq.device;
359                                 ibevent.event = IB_EVENT_CQ_ERR;
360                                 ibevent.element.cq = &iwcq->ibcq;
361                                 iwcq->ibcq.event_handler(&ibevent,
362                                                          iwcq->ibcq.cq_context);
363                         }
364                         irdma_cq_rem_ref(&iwcq->ibcq);
365                         break;
366                 case IRDMA_AE_RESET_NOT_SENT:
367                 case IRDMA_AE_LLP_DOUBT_REACHABILITY:
368                 case IRDMA_AE_RESOURCE_EXHAUSTION:
369                         break;
370                 case IRDMA_AE_PRIV_OPERATION_DENIED:
371                 case IRDMA_AE_STAG_ZERO_INVALID:
372                 case IRDMA_AE_IB_RREQ_AND_Q1_FULL:
373                 case IRDMA_AE_DDP_UBE_INVALID_DDP_VERSION:
374                 case IRDMA_AE_DDP_UBE_INVALID_MO:
375                 case IRDMA_AE_DDP_UBE_INVALID_QN:
376                 case IRDMA_AE_DDP_NO_L_BIT:
377                 case IRDMA_AE_RDMAP_ROE_INVALID_RDMAP_VERSION:
378                 case IRDMA_AE_RDMAP_ROE_UNEXPECTED_OPCODE:
379                 case IRDMA_AE_ROE_INVALID_RDMA_READ_REQUEST:
380                 case IRDMA_AE_ROE_INVALID_RDMA_WRITE_OR_READ_RESP:
381                 case IRDMA_AE_INVALID_ARP_ENTRY:
382                 case IRDMA_AE_INVALID_TCP_OPTION_RCVD:
383                 case IRDMA_AE_STALE_ARP_ENTRY:
384                 case IRDMA_AE_LLP_RECEIVED_MPA_CRC_ERROR:
385                 case IRDMA_AE_LLP_SEGMENT_TOO_SMALL:
386                 case IRDMA_AE_LLP_SYN_RECEIVED:
387                 case IRDMA_AE_LLP_TOO_MANY_RETRIES:
388                 case IRDMA_AE_LCE_QP_CATASTROPHIC:
389                 case IRDMA_AE_LCE_FUNCTION_CATASTROPHIC:
390                 case IRDMA_AE_LLP_TOO_MANY_RNRS:
391                 case IRDMA_AE_LCE_CQ_CATASTROPHIC:
392                 case IRDMA_AE_UDA_XMIT_DGRAM_TOO_LONG:
393                 default:
394                         ibdev_err(&iwdev->ibdev, "abnormal ae_id = 0x%x bool qp=%d qp_id = %d, ae_src=%d\n",
395                                   info->ae_id, info->qp, info->qp_cq_id, info->ae_src);
396                         if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
397                                 ctx_info->roce_info->err_rq_idx_valid = info->rq;
398                                 if (info->rq) {
399                                         ctx_info->roce_info->err_rq_idx = info->wqe_idx;
400                                         irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va,
401                                                                 ctx_info);
402                                 }
403                                 irdma_set_flush_fields(qp, info);
404                                 irdma_cm_disconn(iwqp);
405                                 break;
406                         }
407                         ctx_info->iwarp_info->err_rq_idx_valid = info->rq;
408                         if (info->rq) {
409                                 ctx_info->iwarp_info->err_rq_idx = info->wqe_idx;
410                                 ctx_info->tcp_info_valid = false;
411                                 ctx_info->iwarp_info_valid = true;
412                                 irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va,
413                                                    ctx_info);
414                         }
415                         if (iwqp->hw_iwarp_state != IRDMA_QP_STATE_RTS &&
416                             iwqp->hw_iwarp_state != IRDMA_QP_STATE_TERMINATE) {
417                                 irdma_next_iw_state(iwqp, IRDMA_QP_STATE_ERROR, 1, 0, 0);
418                                 irdma_cm_disconn(iwqp);
419                         } else {
420                                 irdma_terminate_connection(qp, info);
421                         }
422                         break;
423                 }
424                 if (info->qp)
425                         irdma_qp_rem_ref(&iwqp->ibqp);
426         } while (1);
427
428         if (aeqcnt)
429                 irdma_sc_repost_aeq_entries(dev, aeqcnt);
430 }
431
432 /**
433  * irdma_ena_intr - set up device interrupts
434  * @dev: hardware control device structure
435  * @msix_id: id of the interrupt to be enabled
436  */
437 static void irdma_ena_intr(struct irdma_sc_dev *dev, u32 msix_id)
438 {
439         dev->irq_ops->irdma_en_irq(dev, msix_id);
440 }
441
442 /**
443  * irdma_dpc - tasklet for aeq and ceq 0
444  * @t: tasklet_struct ptr
445  */
446 static void irdma_dpc(struct tasklet_struct *t)
447 {
448         struct irdma_pci_f *rf = from_tasklet(rf, t, dpc_tasklet);
449
450         if (rf->msix_shared)
451                 irdma_process_ceq(rf, rf->ceqlist);
452         irdma_process_aeq(rf);
453         irdma_ena_intr(&rf->sc_dev, rf->iw_msixtbl[0].idx);
454 }
455
456 /**
457  * irdma_ceq_dpc - dpc handler for CEQ
458  * @t: tasklet_struct ptr
459  */
460 static void irdma_ceq_dpc(struct tasklet_struct *t)
461 {
462         struct irdma_ceq *iwceq = from_tasklet(iwceq, t, dpc_tasklet);
463         struct irdma_pci_f *rf = iwceq->rf;
464
465         irdma_process_ceq(rf, iwceq);
466         irdma_ena_intr(&rf->sc_dev, iwceq->msix_idx);
467 }
468
469 /**
470  * irdma_save_msix_info - copy msix vector information to iwarp device
471  * @rf: RDMA PCI function
472  *
473  * Allocate iwdev msix table and copy the msix info to the table
474  * Return 0 if successful, otherwise return error
475  */
476 static int irdma_save_msix_info(struct irdma_pci_f *rf)
477 {
478         struct irdma_qvlist_info *iw_qvlist;
479         struct irdma_qv_info *iw_qvinfo;
480         struct msix_entry *pmsix;
481         u32 ceq_idx;
482         u32 i;
483         size_t size;
484
485         if (!rf->msix_count)
486                 return -EINVAL;
487
488         size = sizeof(struct irdma_msix_vector) * rf->msix_count;
489         size += struct_size(iw_qvlist, qv_info, rf->msix_count);
490         rf->iw_msixtbl = kzalloc(size, GFP_KERNEL);
491         if (!rf->iw_msixtbl)
492                 return -ENOMEM;
493
494         rf->iw_qvlist = (struct irdma_qvlist_info *)
495                         (&rf->iw_msixtbl[rf->msix_count]);
496         iw_qvlist = rf->iw_qvlist;
497         iw_qvinfo = iw_qvlist->qv_info;
498         iw_qvlist->num_vectors = rf->msix_count;
499         if (rf->msix_count <= num_online_cpus())
500                 rf->msix_shared = true;
501         else if (rf->msix_count > num_online_cpus() + 1)
502                 rf->msix_count = num_online_cpus() + 1;
503
504         pmsix = rf->msix_entries;
505         for (i = 0, ceq_idx = 0; i < rf->msix_count; i++, iw_qvinfo++) {
506                 rf->iw_msixtbl[i].idx = pmsix->entry;
507                 rf->iw_msixtbl[i].irq = pmsix->vector;
508                 rf->iw_msixtbl[i].cpu_affinity = ceq_idx;
509                 if (!i) {
510                         iw_qvinfo->aeq_idx = 0;
511                         if (rf->msix_shared)
512                                 iw_qvinfo->ceq_idx = ceq_idx++;
513                         else
514                                 iw_qvinfo->ceq_idx = IRDMA_Q_INVALID_IDX;
515                 } else {
516                         iw_qvinfo->aeq_idx = IRDMA_Q_INVALID_IDX;
517                         iw_qvinfo->ceq_idx = ceq_idx++;
518                 }
519                 iw_qvinfo->itr_idx = 3;
520                 iw_qvinfo->v_idx = rf->iw_msixtbl[i].idx;
521                 pmsix++;
522         }
523
524         return 0;
525 }
526
527 /**
528  * irdma_irq_handler - interrupt handler for aeq and ceq0
529  * @irq: Interrupt request number
530  * @data: RDMA PCI function
531  */
532 static irqreturn_t irdma_irq_handler(int irq, void *data)
533 {
534         struct irdma_pci_f *rf = data;
535
536         tasklet_schedule(&rf->dpc_tasklet);
537
538         return IRQ_HANDLED;
539 }
540
541 /**
542  * irdma_ceq_handler - interrupt handler for ceq
543  * @irq: interrupt request number
544  * @data: ceq pointer
545  */
546 static irqreturn_t irdma_ceq_handler(int irq, void *data)
547 {
548         struct irdma_ceq *iwceq = data;
549
550         if (iwceq->irq != irq)
551                 ibdev_err(to_ibdev(&iwceq->rf->sc_dev), "expected irq = %d received irq = %d\n",
552                           iwceq->irq, irq);
553         tasklet_schedule(&iwceq->dpc_tasklet);
554
555         return IRQ_HANDLED;
556 }
557
558 /**
559  * irdma_destroy_irq - destroy device interrupts
560  * @rf: RDMA PCI function
561  * @msix_vec: msix vector to disable irq
562  * @dev_id: parameter to pass to free_irq (used during irq setup)
563  *
564  * The function is called when destroying aeq/ceq
565  */
566 static void irdma_destroy_irq(struct irdma_pci_f *rf,
567                               struct irdma_msix_vector *msix_vec, void *dev_id)
568 {
569         struct irdma_sc_dev *dev = &rf->sc_dev;
570
571         dev->irq_ops->irdma_dis_irq(dev, msix_vec->idx);
572         irq_update_affinity_hint(msix_vec->irq, NULL);
573         free_irq(msix_vec->irq, dev_id);
574         if (rf == dev_id) {
575                 tasklet_kill(&rf->dpc_tasklet);
576         } else {
577                 struct irdma_ceq *iwceq = (struct irdma_ceq *)dev_id;
578
579                 tasklet_kill(&iwceq->dpc_tasklet);
580         }
581 }
582
583 /**
584  * irdma_destroy_cqp  - destroy control qp
585  * @rf: RDMA PCI function
586  *
587  * Issue destroy cqp request and
588  * free the resources associated with the cqp
589  */
590 static void irdma_destroy_cqp(struct irdma_pci_f *rf)
591 {
592         struct irdma_sc_dev *dev = &rf->sc_dev;
593         struct irdma_cqp *cqp = &rf->cqp;
594         int status = 0;
595
596         status = irdma_sc_cqp_destroy(dev->cqp);
597         if (status)
598                 ibdev_dbg(to_ibdev(dev), "ERR: Destroy CQP failed %d\n", status);
599
600         irdma_cleanup_pending_cqp_op(rf);
601         dma_free_coherent(dev->hw->device, cqp->sq.size, cqp->sq.va,
602                           cqp->sq.pa);
603         cqp->sq.va = NULL;
604         kfree(cqp->scratch_array);
605         cqp->scratch_array = NULL;
606         kfree(cqp->cqp_requests);
607         cqp->cqp_requests = NULL;
608 }
609
610 static void irdma_destroy_virt_aeq(struct irdma_pci_f *rf)
611 {
612         struct irdma_aeq *aeq = &rf->aeq;
613         u32 pg_cnt = DIV_ROUND_UP(aeq->mem.size, PAGE_SIZE);
614         dma_addr_t *pg_arr = (dma_addr_t *)aeq->palloc.level1.addr;
615
616         irdma_unmap_vm_page_list(&rf->hw, pg_arr, pg_cnt);
617         irdma_free_pble(rf->pble_rsrc, &aeq->palloc);
618         vfree(aeq->mem.va);
619 }
620
621 /**
622  * irdma_destroy_aeq - destroy aeq
623  * @rf: RDMA PCI function
624  *
625  * Issue a destroy aeq request and
626  * free the resources associated with the aeq
627  * The function is called during driver unload
628  */
629 static void irdma_destroy_aeq(struct irdma_pci_f *rf)
630 {
631         struct irdma_sc_dev *dev = &rf->sc_dev;
632         struct irdma_aeq *aeq = &rf->aeq;
633         int status = -EBUSY;
634
635         if (!rf->msix_shared) {
636                 rf->sc_dev.irq_ops->irdma_cfg_aeq(&rf->sc_dev, rf->iw_msixtbl->idx, false);
637                 irdma_destroy_irq(rf, rf->iw_msixtbl, rf);
638         }
639         if (rf->reset)
640                 goto exit;
641
642         aeq->sc_aeq.size = 0;
643         status = irdma_cqp_aeq_cmd(dev, &aeq->sc_aeq, IRDMA_OP_AEQ_DESTROY);
644         if (status)
645                 ibdev_dbg(to_ibdev(dev), "ERR: Destroy AEQ failed %d\n", status);
646
647 exit:
648         if (aeq->virtual_map) {
649                 irdma_destroy_virt_aeq(rf);
650         } else {
651                 dma_free_coherent(dev->hw->device, aeq->mem.size, aeq->mem.va,
652                                   aeq->mem.pa);
653                 aeq->mem.va = NULL;
654         }
655 }
656
657 /**
658  * irdma_destroy_ceq - destroy ceq
659  * @rf: RDMA PCI function
660  * @iwceq: ceq to be destroyed
661  *
662  * Issue a destroy ceq request and
663  * free the resources associated with the ceq
664  */
665 static void irdma_destroy_ceq(struct irdma_pci_f *rf, struct irdma_ceq *iwceq)
666 {
667         struct irdma_sc_dev *dev = &rf->sc_dev;
668         int status;
669
670         if (rf->reset)
671                 goto exit;
672
673         status = irdma_sc_ceq_destroy(&iwceq->sc_ceq, 0, 1);
674         if (status) {
675                 ibdev_dbg(to_ibdev(dev), "ERR: CEQ destroy command failed %d\n", status);
676                 goto exit;
677         }
678
679         status = irdma_sc_cceq_destroy_done(&iwceq->sc_ceq);
680         if (status)
681                 ibdev_dbg(to_ibdev(dev), "ERR: CEQ destroy completion failed %d\n",
682                           status);
683 exit:
684         dma_free_coherent(dev->hw->device, iwceq->mem.size, iwceq->mem.va,
685                           iwceq->mem.pa);
686         iwceq->mem.va = NULL;
687 }
688
689 /**
690  * irdma_del_ceq_0 - destroy ceq 0
691  * @rf: RDMA PCI function
692  *
693  * Disable the ceq 0 interrupt and destroy the ceq 0
694  */
695 static void irdma_del_ceq_0(struct irdma_pci_f *rf)
696 {
697         struct irdma_ceq *iwceq = rf->ceqlist;
698         struct irdma_msix_vector *msix_vec;
699
700         if (rf->msix_shared) {
701                 msix_vec = &rf->iw_msixtbl[0];
702                 rf->sc_dev.irq_ops->irdma_cfg_ceq(&rf->sc_dev,
703                                                   msix_vec->ceq_id,
704                                                   msix_vec->idx, false);
705                 irdma_destroy_irq(rf, msix_vec, rf);
706         } else {
707                 msix_vec = &rf->iw_msixtbl[1];
708                 irdma_destroy_irq(rf, msix_vec, iwceq);
709         }
710
711         irdma_destroy_ceq(rf, iwceq);
712         rf->sc_dev.ceq_valid = false;
713         rf->ceqs_count = 0;
714 }
715
716 /**
717  * irdma_del_ceqs - destroy all ceq's except CEQ 0
718  * @rf: RDMA PCI function
719  *
720  * Go through all of the device ceq's, except 0, and for each
721  * ceq disable the ceq interrupt and destroy the ceq
722  */
723 static void irdma_del_ceqs(struct irdma_pci_f *rf)
724 {
725         struct irdma_ceq *iwceq = &rf->ceqlist[1];
726         struct irdma_msix_vector *msix_vec;
727         u32 i = 0;
728
729         if (rf->msix_shared)
730                 msix_vec = &rf->iw_msixtbl[1];
731         else
732                 msix_vec = &rf->iw_msixtbl[2];
733
734         for (i = 1; i < rf->ceqs_count; i++, msix_vec++, iwceq++) {
735                 rf->sc_dev.irq_ops->irdma_cfg_ceq(&rf->sc_dev, msix_vec->ceq_id,
736                                                   msix_vec->idx, false);
737                 irdma_destroy_irq(rf, msix_vec, iwceq);
738                 irdma_cqp_ceq_cmd(&rf->sc_dev, &iwceq->sc_ceq,
739                                   IRDMA_OP_CEQ_DESTROY);
740                 dma_free_coherent(rf->sc_dev.hw->device, iwceq->mem.size,
741                                   iwceq->mem.va, iwceq->mem.pa);
742                 iwceq->mem.va = NULL;
743         }
744         rf->ceqs_count = 1;
745 }
746
747 /**
748  * irdma_destroy_ccq - destroy control cq
749  * @rf: RDMA PCI function
750  *
751  * Issue destroy ccq request and
752  * free the resources associated with the ccq
753  */
754 static void irdma_destroy_ccq(struct irdma_pci_f *rf)
755 {
756         struct irdma_sc_dev *dev = &rf->sc_dev;
757         struct irdma_ccq *ccq = &rf->ccq;
758         int status = 0;
759
760         if (rf->cqp_cmpl_wq)
761                 destroy_workqueue(rf->cqp_cmpl_wq);
762
763         if (!rf->reset)
764                 status = irdma_sc_ccq_destroy(dev->ccq, 0, true);
765         if (status)
766                 ibdev_dbg(to_ibdev(dev), "ERR: CCQ destroy failed %d\n", status);
767         dma_free_coherent(dev->hw->device, ccq->mem_cq.size, ccq->mem_cq.va,
768                           ccq->mem_cq.pa);
769         ccq->mem_cq.va = NULL;
770 }
771
772 /**
773  * irdma_close_hmc_objects_type - delete hmc objects of a given type
774  * @dev: iwarp device
775  * @obj_type: the hmc object type to be deleted
776  * @hmc_info: host memory info struct
777  * @privileged: permission to close HMC objects
778  * @reset: true if called before reset
779  */
780 static void irdma_close_hmc_objects_type(struct irdma_sc_dev *dev,
781                                          enum irdma_hmc_rsrc_type obj_type,
782                                          struct irdma_hmc_info *hmc_info,
783                                          bool privileged, bool reset)
784 {
785         struct irdma_hmc_del_obj_info info = {};
786
787         info.hmc_info = hmc_info;
788         info.rsrc_type = obj_type;
789         info.count = hmc_info->hmc_obj[obj_type].cnt;
790         info.privileged = privileged;
791         if (irdma_sc_del_hmc_obj(dev, &info, reset))
792                 ibdev_dbg(to_ibdev(dev), "ERR: del HMC obj of type %d failed\n",
793                           obj_type);
794 }
795
796 /**
797  * irdma_del_hmc_objects - remove all device hmc objects
798  * @dev: iwarp device
799  * @hmc_info: hmc_info to free
800  * @privileged: permission to delete HMC objects
801  * @reset: true if called before reset
802  * @vers: hardware version
803  */
804 static void irdma_del_hmc_objects(struct irdma_sc_dev *dev,
805                                   struct irdma_hmc_info *hmc_info, bool privileged,
806                                   bool reset, enum irdma_vers vers)
807 {
808         unsigned int i;
809
810         for (i = 0; i < IW_HMC_OBJ_TYPE_NUM; i++) {
811                 if (dev->hmc_info->hmc_obj[iw_hmc_obj_types[i]].cnt)
812                         irdma_close_hmc_objects_type(dev, iw_hmc_obj_types[i],
813                                                      hmc_info, privileged, reset);
814                 if (vers == IRDMA_GEN_1 && i == IRDMA_HMC_IW_TIMER)
815                         break;
816         }
817 }
818
819 /**
820  * irdma_create_hmc_obj_type - create hmc object of a given type
821  * @dev: hardware control device structure
822  * @info: information for the hmc object to create
823  */
824 static int irdma_create_hmc_obj_type(struct irdma_sc_dev *dev,
825                                      struct irdma_hmc_create_obj_info *info)
826 {
827         return irdma_sc_create_hmc_obj(dev, info);
828 }
829
830 /**
831  * irdma_create_hmc_objs - create all hmc objects for the device
832  * @rf: RDMA PCI function
833  * @privileged: permission to create HMC objects
834  * @vers: HW version
835  *
836  * Create the device hmc objects and allocate hmc pages
837  * Return 0 if successful, otherwise clean up and return error
838  */
839 static int irdma_create_hmc_objs(struct irdma_pci_f *rf, bool privileged,
840                                  enum irdma_vers vers)
841 {
842         struct irdma_sc_dev *dev = &rf->sc_dev;
843         struct irdma_hmc_create_obj_info info = {};
844         int i, status = 0;
845
846         info.hmc_info = dev->hmc_info;
847         info.privileged = privileged;
848         info.entry_type = rf->sd_type;
849
850         for (i = 0; i < IW_HMC_OBJ_TYPE_NUM; i++) {
851                 if (iw_hmc_obj_types[i] == IRDMA_HMC_IW_PBLE)
852                         continue;
853                 if (dev->hmc_info->hmc_obj[iw_hmc_obj_types[i]].cnt) {
854                         info.rsrc_type = iw_hmc_obj_types[i];
855                         info.count = dev->hmc_info->hmc_obj[info.rsrc_type].cnt;
856                         info.add_sd_cnt = 0;
857                         status = irdma_create_hmc_obj_type(dev, &info);
858                         if (status) {
859                                 ibdev_dbg(to_ibdev(dev),
860                                           "ERR: create obj type %d status = %d\n",
861                                           iw_hmc_obj_types[i], status);
862                                 break;
863                         }
864                 }
865                 if (vers == IRDMA_GEN_1 && i == IRDMA_HMC_IW_TIMER)
866                         break;
867         }
868
869         if (!status)
870                 return irdma_sc_static_hmc_pages_allocated(dev->cqp, 0, dev->hmc_fn_id,
871                                                            true, true);
872
873         while (i) {
874                 i--;
875                 /* destroy the hmc objects of a given type */
876                 if (dev->hmc_info->hmc_obj[iw_hmc_obj_types[i]].cnt)
877                         irdma_close_hmc_objects_type(dev, iw_hmc_obj_types[i],
878                                                      dev->hmc_info, privileged,
879                                                      false);
880         }
881
882         return status;
883 }
884
885 /**
886  * irdma_obj_aligned_mem - get aligned memory from device allocated memory
887  * @rf: RDMA PCI function
888  * @memptr: points to the memory addresses
889  * @size: size of memory needed
890  * @mask: mask for the aligned memory
891  *
892  * Get aligned memory of the requested size and
893  * update the memptr to point to the new aligned memory
894  * Return 0 if successful, otherwise return no memory error
895  */
896 static int irdma_obj_aligned_mem(struct irdma_pci_f *rf,
897                                  struct irdma_dma_mem *memptr, u32 size,
898                                  u32 mask)
899 {
900         unsigned long va, newva;
901         unsigned long extra;
902
903         va = (unsigned long)rf->obj_next.va;
904         newva = va;
905         if (mask)
906                 newva = ALIGN(va, (unsigned long)mask + 1ULL);
907         extra = newva - va;
908         memptr->va = (u8 *)va + extra;
909         memptr->pa = rf->obj_next.pa + extra;
910         memptr->size = size;
911         if (((u8 *)memptr->va + size) > ((u8 *)rf->obj_mem.va + rf->obj_mem.size))
912                 return -ENOMEM;
913
914         rf->obj_next.va = (u8 *)memptr->va + size;
915         rf->obj_next.pa = memptr->pa + size;
916
917         return 0;
918 }
919
920 /**
921  * irdma_create_cqp - create control qp
922  * @rf: RDMA PCI function
923  *
924  * Return 0, if the cqp and all the resources associated with it
925  * are successfully created, otherwise return error
926  */
927 static int irdma_create_cqp(struct irdma_pci_f *rf)
928 {
929         u32 sqsize = IRDMA_CQP_SW_SQSIZE_2048;
930         struct irdma_dma_mem mem;
931         struct irdma_sc_dev *dev = &rf->sc_dev;
932         struct irdma_cqp_init_info cqp_init_info = {};
933         struct irdma_cqp *cqp = &rf->cqp;
934         u16 maj_err, min_err;
935         int i, status;
936
937         cqp->cqp_requests = kcalloc(sqsize, sizeof(*cqp->cqp_requests), GFP_KERNEL);
938         if (!cqp->cqp_requests)
939                 return -ENOMEM;
940
941         cqp->scratch_array = kcalloc(sqsize, sizeof(*cqp->scratch_array), GFP_KERNEL);
942         if (!cqp->scratch_array) {
943                 status = -ENOMEM;
944                 goto err_scratch;
945         }
946
947         dev->cqp = &cqp->sc_cqp;
948         dev->cqp->dev = dev;
949         cqp->sq.size = ALIGN(sizeof(struct irdma_cqp_sq_wqe) * sqsize,
950                              IRDMA_CQP_ALIGNMENT);
951         cqp->sq.va = dma_alloc_coherent(dev->hw->device, cqp->sq.size,
952                                         &cqp->sq.pa, GFP_KERNEL);
953         if (!cqp->sq.va) {
954                 status = -ENOMEM;
955                 goto err_sq;
956         }
957
958         status = irdma_obj_aligned_mem(rf, &mem, sizeof(struct irdma_cqp_ctx),
959                                        IRDMA_HOST_CTX_ALIGNMENT_M);
960         if (status)
961                 goto err_ctx;
962
963         dev->cqp->host_ctx_pa = mem.pa;
964         dev->cqp->host_ctx = mem.va;
965         /* populate the cqp init info */
966         cqp_init_info.dev = dev;
967         cqp_init_info.sq_size = sqsize;
968         cqp_init_info.sq = cqp->sq.va;
969         cqp_init_info.sq_pa = cqp->sq.pa;
970         cqp_init_info.host_ctx_pa = mem.pa;
971         cqp_init_info.host_ctx = mem.va;
972         cqp_init_info.hmc_profile = rf->rsrc_profile;
973         cqp_init_info.scratch_array = cqp->scratch_array;
974         cqp_init_info.protocol_used = rf->protocol_used;
975
976         switch (rf->rdma_ver) {
977         case IRDMA_GEN_1:
978                 cqp_init_info.hw_maj_ver = IRDMA_CQPHC_HW_MAJVER_GEN_1;
979                 break;
980         case IRDMA_GEN_2:
981                 cqp_init_info.hw_maj_ver = IRDMA_CQPHC_HW_MAJVER_GEN_2;
982                 break;
983         }
984         status = irdma_sc_cqp_init(dev->cqp, &cqp_init_info);
985         if (status) {
986                 ibdev_dbg(to_ibdev(dev), "ERR: cqp init status %d\n", status);
987                 goto err_ctx;
988         }
989
990         spin_lock_init(&cqp->req_lock);
991         spin_lock_init(&cqp->compl_lock);
992
993         status = irdma_sc_cqp_create(dev->cqp, &maj_err, &min_err);
994         if (status) {
995                 ibdev_dbg(to_ibdev(dev),
996                           "ERR: cqp create failed - status %d maj_err %d min_err %d\n",
997                           status, maj_err, min_err);
998                 goto err_ctx;
999         }
1000
1001         INIT_LIST_HEAD(&cqp->cqp_avail_reqs);
1002         INIT_LIST_HEAD(&cqp->cqp_pending_reqs);
1003
1004         /* init the waitqueue of the cqp_requests and add them to the list */
1005         for (i = 0; i < sqsize; i++) {
1006                 init_waitqueue_head(&cqp->cqp_requests[i].waitq);
1007                 list_add_tail(&cqp->cqp_requests[i].list, &cqp->cqp_avail_reqs);
1008         }
1009         init_waitqueue_head(&cqp->remove_wq);
1010         return 0;
1011
1012 err_ctx:
1013         dma_free_coherent(dev->hw->device, cqp->sq.size,
1014                           cqp->sq.va, cqp->sq.pa);
1015         cqp->sq.va = NULL;
1016 err_sq:
1017         kfree(cqp->scratch_array);
1018         cqp->scratch_array = NULL;
1019 err_scratch:
1020         kfree(cqp->cqp_requests);
1021         cqp->cqp_requests = NULL;
1022
1023         return status;
1024 }
1025
1026 /**
1027  * irdma_create_ccq - create control cq
1028  * @rf: RDMA PCI function
1029  *
1030  * Return 0, if the ccq and the resources associated with it
1031  * are successfully created, otherwise return error
1032  */
1033 static int irdma_create_ccq(struct irdma_pci_f *rf)
1034 {
1035         struct irdma_sc_dev *dev = &rf->sc_dev;
1036         struct irdma_ccq_init_info info = {};
1037         struct irdma_ccq *ccq = &rf->ccq;
1038         int status;
1039
1040         dev->ccq = &ccq->sc_cq;
1041         dev->ccq->dev = dev;
1042         info.dev = dev;
1043         ccq->shadow_area.size = sizeof(struct irdma_cq_shadow_area);
1044         ccq->mem_cq.size = ALIGN(sizeof(struct irdma_cqe) * IW_CCQ_SIZE,
1045                                  IRDMA_CQ0_ALIGNMENT);
1046         ccq->mem_cq.va = dma_alloc_coherent(dev->hw->device, ccq->mem_cq.size,
1047                                             &ccq->mem_cq.pa, GFP_KERNEL);
1048         if (!ccq->mem_cq.va)
1049                 return -ENOMEM;
1050
1051         status = irdma_obj_aligned_mem(rf, &ccq->shadow_area,
1052                                        ccq->shadow_area.size,
1053                                        IRDMA_SHADOWAREA_M);
1054         if (status)
1055                 goto exit;
1056
1057         ccq->sc_cq.back_cq = ccq;
1058         /* populate the ccq init info */
1059         info.cq_base = ccq->mem_cq.va;
1060         info.cq_pa = ccq->mem_cq.pa;
1061         info.num_elem = IW_CCQ_SIZE;
1062         info.shadow_area = ccq->shadow_area.va;
1063         info.shadow_area_pa = ccq->shadow_area.pa;
1064         info.ceqe_mask = false;
1065         info.ceq_id_valid = true;
1066         info.shadow_read_threshold = 16;
1067         info.vsi = &rf->default_vsi;
1068         status = irdma_sc_ccq_init(dev->ccq, &info);
1069         if (!status)
1070                 status = irdma_sc_ccq_create(dev->ccq, 0, true, true);
1071 exit:
1072         if (status) {
1073                 dma_free_coherent(dev->hw->device, ccq->mem_cq.size,
1074                                   ccq->mem_cq.va, ccq->mem_cq.pa);
1075                 ccq->mem_cq.va = NULL;
1076         }
1077
1078         return status;
1079 }
1080
1081 /**
1082  * irdma_alloc_set_mac - set up a mac address table entry
1083  * @iwdev: irdma device
1084  *
1085  * Allocate a mac ip entry and add it to the hw table Return 0
1086  * if successful, otherwise return error
1087  */
1088 static int irdma_alloc_set_mac(struct irdma_device *iwdev)
1089 {
1090         int status;
1091
1092         status = irdma_alloc_local_mac_entry(iwdev->rf,
1093                                              &iwdev->mac_ip_table_idx);
1094         if (!status) {
1095                 status = irdma_add_local_mac_entry(iwdev->rf,
1096                                                    (const u8 *)iwdev->netdev->dev_addr,
1097                                                    (u8)iwdev->mac_ip_table_idx);
1098                 if (status)
1099                         irdma_del_local_mac_entry(iwdev->rf,
1100                                                   (u8)iwdev->mac_ip_table_idx);
1101         }
1102         return status;
1103 }
1104
1105 /**
1106  * irdma_cfg_ceq_vector - set up the msix interrupt vector for
1107  * ceq
1108  * @rf: RDMA PCI function
1109  * @iwceq: ceq associated with the vector
1110  * @ceq_id: the id number of the iwceq
1111  * @msix_vec: interrupt vector information
1112  *
1113  * Allocate interrupt resources and enable irq handling
1114  * Return 0 if successful, otherwise return error
1115  */
1116 static int irdma_cfg_ceq_vector(struct irdma_pci_f *rf, struct irdma_ceq *iwceq,
1117                                 u32 ceq_id, struct irdma_msix_vector *msix_vec)
1118 {
1119         int status;
1120
1121         if (rf->msix_shared && !ceq_id) {
1122                 snprintf(msix_vec->name, sizeof(msix_vec->name) - 1,
1123                          "irdma-%s-AEQCEQ-0", dev_name(&rf->pcidev->dev));
1124                 tasklet_setup(&rf->dpc_tasklet, irdma_dpc);
1125                 status = request_irq(msix_vec->irq, irdma_irq_handler, 0,
1126                                      msix_vec->name, rf);
1127         } else {
1128                 snprintf(msix_vec->name, sizeof(msix_vec->name) - 1,
1129                          "irdma-%s-CEQ-%d",
1130                          dev_name(&rf->pcidev->dev), ceq_id);
1131                 tasklet_setup(&iwceq->dpc_tasklet, irdma_ceq_dpc);
1132
1133                 status = request_irq(msix_vec->irq, irdma_ceq_handler, 0,
1134                                      msix_vec->name, iwceq);
1135         }
1136         cpumask_clear(&msix_vec->mask);
1137         cpumask_set_cpu(msix_vec->cpu_affinity, &msix_vec->mask);
1138         irq_update_affinity_hint(msix_vec->irq, &msix_vec->mask);
1139         if (status) {
1140                 ibdev_dbg(&rf->iwdev->ibdev, "ERR: ceq irq config fail\n");
1141                 return status;
1142         }
1143
1144         msix_vec->ceq_id = ceq_id;
1145         rf->sc_dev.irq_ops->irdma_cfg_ceq(&rf->sc_dev, ceq_id, msix_vec->idx, true);
1146
1147         return 0;
1148 }
1149
1150 /**
1151  * irdma_cfg_aeq_vector - set up the msix vector for aeq
1152  * @rf: RDMA PCI function
1153  *
1154  * Allocate interrupt resources and enable irq handling
1155  * Return 0 if successful, otherwise return error
1156  */
1157 static int irdma_cfg_aeq_vector(struct irdma_pci_f *rf)
1158 {
1159         struct irdma_msix_vector *msix_vec = rf->iw_msixtbl;
1160         u32 ret = 0;
1161
1162         if (!rf->msix_shared) {
1163                 snprintf(msix_vec->name, sizeof(msix_vec->name) - 1,
1164                          "irdma-%s-AEQ", dev_name(&rf->pcidev->dev));
1165                 tasklet_setup(&rf->dpc_tasklet, irdma_dpc);
1166                 ret = request_irq(msix_vec->irq, irdma_irq_handler, 0,
1167                                   msix_vec->name, rf);
1168         }
1169         if (ret) {
1170                 ibdev_dbg(&rf->iwdev->ibdev, "ERR: aeq irq config fail\n");
1171                 return -EINVAL;
1172         }
1173
1174         rf->sc_dev.irq_ops->irdma_cfg_aeq(&rf->sc_dev, msix_vec->idx, true);
1175
1176         return 0;
1177 }
1178
1179 /**
1180  * irdma_create_ceq - create completion event queue
1181  * @rf: RDMA PCI function
1182  * @iwceq: pointer to the ceq resources to be created
1183  * @ceq_id: the id number of the iwceq
1184  * @vsi: SC vsi struct
1185  *
1186  * Return 0, if the ceq and the resources associated with it
1187  * are successfully created, otherwise return error
1188  */
1189 static int irdma_create_ceq(struct irdma_pci_f *rf, struct irdma_ceq *iwceq,
1190                             u32 ceq_id, struct irdma_sc_vsi *vsi)
1191 {
1192         int status;
1193         struct irdma_ceq_init_info info = {};
1194         struct irdma_sc_dev *dev = &rf->sc_dev;
1195         u32 ceq_size;
1196
1197         info.ceq_id = ceq_id;
1198         iwceq->rf = rf;
1199         ceq_size = min(rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_CQ].cnt,
1200                        dev->hw_attrs.max_hw_ceq_size);
1201         iwceq->mem.size = ALIGN(sizeof(struct irdma_ceqe) * ceq_size,
1202                                 IRDMA_CEQ_ALIGNMENT);
1203         iwceq->mem.va = dma_alloc_coherent(dev->hw->device, iwceq->mem.size,
1204                                            &iwceq->mem.pa, GFP_KERNEL);
1205         if (!iwceq->mem.va)
1206                 return -ENOMEM;
1207
1208         info.ceq_id = ceq_id;
1209         info.ceqe_base = iwceq->mem.va;
1210         info.ceqe_pa = iwceq->mem.pa;
1211         info.elem_cnt = ceq_size;
1212         iwceq->sc_ceq.ceq_id = ceq_id;
1213         info.dev = dev;
1214         info.vsi = vsi;
1215         status = irdma_sc_ceq_init(&iwceq->sc_ceq, &info);
1216         if (!status) {
1217                 if (dev->ceq_valid)
1218                         status = irdma_cqp_ceq_cmd(&rf->sc_dev, &iwceq->sc_ceq,
1219                                                    IRDMA_OP_CEQ_CREATE);
1220                 else
1221                         status = irdma_sc_cceq_create(&iwceq->sc_ceq, 0);
1222         }
1223
1224         if (status) {
1225                 dma_free_coherent(dev->hw->device, iwceq->mem.size,
1226                                   iwceq->mem.va, iwceq->mem.pa);
1227                 iwceq->mem.va = NULL;
1228         }
1229
1230         return status;
1231 }
1232
1233 /**
1234  * irdma_setup_ceq_0 - create CEQ 0 and it's interrupt resource
1235  * @rf: RDMA PCI function
1236  *
1237  * Allocate a list for all device completion event queues
1238  * Create the ceq 0 and configure it's msix interrupt vector
1239  * Return 0, if successfully set up, otherwise return error
1240  */
1241 static int irdma_setup_ceq_0(struct irdma_pci_f *rf)
1242 {
1243         struct irdma_ceq *iwceq;
1244         struct irdma_msix_vector *msix_vec;
1245         u32 i;
1246         int status = 0;
1247         u32 num_ceqs;
1248
1249         num_ceqs = min(rf->msix_count, rf->sc_dev.hmc_fpm_misc.max_ceqs);
1250         rf->ceqlist = kcalloc(num_ceqs, sizeof(*rf->ceqlist), GFP_KERNEL);
1251         if (!rf->ceqlist) {
1252                 status = -ENOMEM;
1253                 goto exit;
1254         }
1255
1256         iwceq = &rf->ceqlist[0];
1257         status = irdma_create_ceq(rf, iwceq, 0, &rf->default_vsi);
1258         if (status) {
1259                 ibdev_dbg(&rf->iwdev->ibdev, "ERR: create ceq status = %d\n",
1260                           status);
1261                 goto exit;
1262         }
1263
1264         spin_lock_init(&iwceq->ce_lock);
1265         i = rf->msix_shared ? 0 : 1;
1266         msix_vec = &rf->iw_msixtbl[i];
1267         iwceq->irq = msix_vec->irq;
1268         iwceq->msix_idx = msix_vec->idx;
1269         status = irdma_cfg_ceq_vector(rf, iwceq, 0, msix_vec);
1270         if (status) {
1271                 irdma_destroy_ceq(rf, iwceq);
1272                 goto exit;
1273         }
1274
1275         irdma_ena_intr(&rf->sc_dev, msix_vec->idx);
1276         rf->ceqs_count++;
1277
1278 exit:
1279         if (status && !rf->ceqs_count) {
1280                 kfree(rf->ceqlist);
1281                 rf->ceqlist = NULL;
1282                 return status;
1283         }
1284         rf->sc_dev.ceq_valid = true;
1285
1286         return 0;
1287 }
1288
1289 /**
1290  * irdma_setup_ceqs - manage the device ceq's and their interrupt resources
1291  * @rf: RDMA PCI function
1292  * @vsi: VSI structure for this CEQ
1293  *
1294  * Allocate a list for all device completion event queues
1295  * Create the ceq's and configure their msix interrupt vectors
1296  * Return 0, if ceqs are successfully set up, otherwise return error
1297  */
1298 static int irdma_setup_ceqs(struct irdma_pci_f *rf, struct irdma_sc_vsi *vsi)
1299 {
1300         u32 i;
1301         u32 ceq_id;
1302         struct irdma_ceq *iwceq;
1303         struct irdma_msix_vector *msix_vec;
1304         int status;
1305         u32 num_ceqs;
1306
1307         num_ceqs = min(rf->msix_count, rf->sc_dev.hmc_fpm_misc.max_ceqs);
1308         i = (rf->msix_shared) ? 1 : 2;
1309         for (ceq_id = 1; i < num_ceqs; i++, ceq_id++) {
1310                 iwceq = &rf->ceqlist[ceq_id];
1311                 status = irdma_create_ceq(rf, iwceq, ceq_id, vsi);
1312                 if (status) {
1313                         ibdev_dbg(&rf->iwdev->ibdev,
1314                                   "ERR: create ceq status = %d\n", status);
1315                         goto del_ceqs;
1316                 }
1317                 spin_lock_init(&iwceq->ce_lock);
1318                 msix_vec = &rf->iw_msixtbl[i];
1319                 iwceq->irq = msix_vec->irq;
1320                 iwceq->msix_idx = msix_vec->idx;
1321                 status = irdma_cfg_ceq_vector(rf, iwceq, ceq_id, msix_vec);
1322                 if (status) {
1323                         irdma_destroy_ceq(rf, iwceq);
1324                         goto del_ceqs;
1325                 }
1326                 irdma_ena_intr(&rf->sc_dev, msix_vec->idx);
1327                 rf->ceqs_count++;
1328         }
1329
1330         return 0;
1331
1332 del_ceqs:
1333         irdma_del_ceqs(rf);
1334
1335         return status;
1336 }
1337
1338 static int irdma_create_virt_aeq(struct irdma_pci_f *rf, u32 size)
1339 {
1340         struct irdma_aeq *aeq = &rf->aeq;
1341         dma_addr_t *pg_arr;
1342         u32 pg_cnt;
1343         int status;
1344
1345         if (rf->rdma_ver < IRDMA_GEN_2)
1346                 return -EOPNOTSUPP;
1347
1348         aeq->mem.size = sizeof(struct irdma_sc_aeqe) * size;
1349         aeq->mem.va = vzalloc(aeq->mem.size);
1350
1351         if (!aeq->mem.va)
1352                 return -ENOMEM;
1353
1354         pg_cnt = DIV_ROUND_UP(aeq->mem.size, PAGE_SIZE);
1355         status = irdma_get_pble(rf->pble_rsrc, &aeq->palloc, pg_cnt, true);
1356         if (status) {
1357                 vfree(aeq->mem.va);
1358                 return status;
1359         }
1360
1361         pg_arr = (dma_addr_t *)aeq->palloc.level1.addr;
1362         status = irdma_map_vm_page_list(&rf->hw, aeq->mem.va, pg_arr, pg_cnt);
1363         if (status) {
1364                 irdma_free_pble(rf->pble_rsrc, &aeq->palloc);
1365                 vfree(aeq->mem.va);
1366                 return status;
1367         }
1368
1369         return 0;
1370 }
1371
1372 /**
1373  * irdma_create_aeq - create async event queue
1374  * @rf: RDMA PCI function
1375  *
1376  * Return 0, if the aeq and the resources associated with it
1377  * are successfully created, otherwise return error
1378  */
1379 static int irdma_create_aeq(struct irdma_pci_f *rf)
1380 {
1381         struct irdma_aeq_init_info info = {};
1382         struct irdma_sc_dev *dev = &rf->sc_dev;
1383         struct irdma_aeq *aeq = &rf->aeq;
1384         struct irdma_hmc_info *hmc_info = rf->sc_dev.hmc_info;
1385         u32 aeq_size;
1386         u8 multiplier = (rf->protocol_used == IRDMA_IWARP_PROTOCOL_ONLY) ? 2 : 1;
1387         int status;
1388
1389         aeq_size = multiplier * hmc_info->hmc_obj[IRDMA_HMC_IW_QP].cnt +
1390                    hmc_info->hmc_obj[IRDMA_HMC_IW_CQ].cnt;
1391         aeq_size = min(aeq_size, dev->hw_attrs.max_hw_aeq_size);
1392
1393         aeq->mem.size = ALIGN(sizeof(struct irdma_sc_aeqe) * aeq_size,
1394                               IRDMA_AEQ_ALIGNMENT);
1395         aeq->mem.va = dma_alloc_coherent(dev->hw->device, aeq->mem.size,
1396                                          &aeq->mem.pa,
1397                                          GFP_KERNEL | __GFP_NOWARN);
1398         if (aeq->mem.va)
1399                 goto skip_virt_aeq;
1400
1401         /* physically mapped aeq failed. setup virtual aeq */
1402         status = irdma_create_virt_aeq(rf, aeq_size);
1403         if (status)
1404                 return status;
1405
1406         info.virtual_map = true;
1407         aeq->virtual_map = info.virtual_map;
1408         info.pbl_chunk_size = 1;
1409         info.first_pm_pbl_idx = aeq->palloc.level1.idx;
1410
1411 skip_virt_aeq:
1412         info.aeqe_base = aeq->mem.va;
1413         info.aeq_elem_pa = aeq->mem.pa;
1414         info.elem_cnt = aeq_size;
1415         info.dev = dev;
1416         info.msix_idx = rf->iw_msixtbl->idx;
1417         status = irdma_sc_aeq_init(&aeq->sc_aeq, &info);
1418         if (status)
1419                 goto err;
1420
1421         status = irdma_cqp_aeq_cmd(dev, &aeq->sc_aeq, IRDMA_OP_AEQ_CREATE);
1422         if (status)
1423                 goto err;
1424
1425         return 0;
1426
1427 err:
1428         if (aeq->virtual_map) {
1429                 irdma_destroy_virt_aeq(rf);
1430         } else {
1431                 dma_free_coherent(dev->hw->device, aeq->mem.size, aeq->mem.va,
1432                                   aeq->mem.pa);
1433                 aeq->mem.va = NULL;
1434         }
1435
1436         return status;
1437 }
1438
1439 /**
1440  * irdma_setup_aeq - set up the device aeq
1441  * @rf: RDMA PCI function
1442  *
1443  * Create the aeq and configure its msix interrupt vector
1444  * Return 0 if successful, otherwise return error
1445  */
1446 static int irdma_setup_aeq(struct irdma_pci_f *rf)
1447 {
1448         struct irdma_sc_dev *dev = &rf->sc_dev;
1449         int status;
1450
1451         status = irdma_create_aeq(rf);
1452         if (status)
1453                 return status;
1454
1455         status = irdma_cfg_aeq_vector(rf);
1456         if (status) {
1457                 irdma_destroy_aeq(rf);
1458                 return status;
1459         }
1460
1461         if (!rf->msix_shared)
1462                 irdma_ena_intr(dev, rf->iw_msixtbl[0].idx);
1463
1464         return 0;
1465 }
1466
1467 /**
1468  * irdma_initialize_ilq - create iwarp local queue for cm
1469  * @iwdev: irdma device
1470  *
1471  * Return 0 if successful, otherwise return error
1472  */
1473 static int irdma_initialize_ilq(struct irdma_device *iwdev)
1474 {
1475         struct irdma_puda_rsrc_info info = {};
1476         int status;
1477
1478         info.type = IRDMA_PUDA_RSRC_TYPE_ILQ;
1479         info.cq_id = 1;
1480         info.qp_id = 1;
1481         info.count = 1;
1482         info.pd_id = 1;
1483         info.abi_ver = IRDMA_ABI_VER;
1484         info.sq_size = min(iwdev->rf->max_qp / 2, (u32)32768);
1485         info.rq_size = info.sq_size;
1486         info.buf_size = 1024;
1487         info.tx_buf_cnt = 2 * info.sq_size;
1488         info.receive = irdma_receive_ilq;
1489         info.xmit_complete = irdma_free_sqbuf;
1490         status = irdma_puda_create_rsrc(&iwdev->vsi, &info);
1491         if (status)
1492                 ibdev_dbg(&iwdev->ibdev, "ERR: ilq create fail\n");
1493
1494         return status;
1495 }
1496
1497 /**
1498  * irdma_initialize_ieq - create iwarp exception queue
1499  * @iwdev: irdma device
1500  *
1501  * Return 0 if successful, otherwise return error
1502  */
1503 static int irdma_initialize_ieq(struct irdma_device *iwdev)
1504 {
1505         struct irdma_puda_rsrc_info info = {};
1506         int status;
1507
1508         info.type = IRDMA_PUDA_RSRC_TYPE_IEQ;
1509         info.cq_id = 2;
1510         info.qp_id = iwdev->vsi.exception_lan_q;
1511         info.count = 1;
1512         info.pd_id = 2;
1513         info.abi_ver = IRDMA_ABI_VER;
1514         info.sq_size = min(iwdev->rf->max_qp / 2, (u32)32768);
1515         info.rq_size = info.sq_size;
1516         info.buf_size = iwdev->vsi.mtu + IRDMA_IPV4_PAD;
1517         info.tx_buf_cnt = 4096;
1518         status = irdma_puda_create_rsrc(&iwdev->vsi, &info);
1519         if (status)
1520                 ibdev_dbg(&iwdev->ibdev, "ERR: ieq create fail\n");
1521
1522         return status;
1523 }
1524
1525 /**
1526  * irdma_reinitialize_ieq - destroy and re-create ieq
1527  * @vsi: VSI structure
1528  */
1529 void irdma_reinitialize_ieq(struct irdma_sc_vsi *vsi)
1530 {
1531         struct irdma_device *iwdev = vsi->back_vsi;
1532         struct irdma_pci_f *rf = iwdev->rf;
1533
1534         irdma_puda_dele_rsrc(vsi, IRDMA_PUDA_RSRC_TYPE_IEQ, false);
1535         if (irdma_initialize_ieq(iwdev)) {
1536                 iwdev->rf->reset = true;
1537                 rf->gen_ops.request_reset(rf);
1538         }
1539 }
1540
1541 /**
1542  * irdma_hmc_setup - create hmc objects for the device
1543  * @rf: RDMA PCI function
1544  *
1545  * Set up the device private memory space for the number and size of
1546  * the hmc objects and create the objects
1547  * Return 0 if successful, otherwise return error
1548  */
1549 static int irdma_hmc_setup(struct irdma_pci_f *rf)
1550 {
1551         int status;
1552         u32 qpcnt;
1553
1554         qpcnt = rsrc_limits_table[rf->limits_sel].qplimit;
1555
1556         rf->sd_type = IRDMA_SD_TYPE_DIRECT;
1557         status = irdma_cfg_fpm_val(&rf->sc_dev, qpcnt);
1558         if (status)
1559                 return status;
1560
1561         status = irdma_create_hmc_objs(rf, true, rf->rdma_ver);
1562
1563         return status;
1564 }
1565
1566 /**
1567  * irdma_del_init_mem - deallocate memory resources
1568  * @rf: RDMA PCI function
1569  */
1570 static void irdma_del_init_mem(struct irdma_pci_f *rf)
1571 {
1572         struct irdma_sc_dev *dev = &rf->sc_dev;
1573
1574         kfree(dev->hmc_info->sd_table.sd_entry);
1575         dev->hmc_info->sd_table.sd_entry = NULL;
1576         vfree(rf->mem_rsrc);
1577         rf->mem_rsrc = NULL;
1578         dma_free_coherent(rf->hw.device, rf->obj_mem.size, rf->obj_mem.va,
1579                           rf->obj_mem.pa);
1580         rf->obj_mem.va = NULL;
1581         if (rf->rdma_ver != IRDMA_GEN_1) {
1582                 bitmap_free(rf->allocated_ws_nodes);
1583                 rf->allocated_ws_nodes = NULL;
1584         }
1585         kfree(rf->ceqlist);
1586         rf->ceqlist = NULL;
1587         kfree(rf->iw_msixtbl);
1588         rf->iw_msixtbl = NULL;
1589         kfree(rf->hmc_info_mem);
1590         rf->hmc_info_mem = NULL;
1591 }
1592
1593 /**
1594  * irdma_initialize_dev - initialize device
1595  * @rf: RDMA PCI function
1596  *
1597  * Allocate memory for the hmc objects and initialize iwdev
1598  * Return 0 if successful, otherwise clean up the resources
1599  * and return error
1600  */
1601 static int irdma_initialize_dev(struct irdma_pci_f *rf)
1602 {
1603         int status;
1604         struct irdma_sc_dev *dev = &rf->sc_dev;
1605         struct irdma_device_init_info info = {};
1606         struct irdma_dma_mem mem;
1607         u32 size;
1608
1609         size = sizeof(struct irdma_hmc_pble_rsrc) +
1610                sizeof(struct irdma_hmc_info) +
1611                (sizeof(struct irdma_hmc_obj_info) * IRDMA_HMC_IW_MAX);
1612
1613         rf->hmc_info_mem = kzalloc(size, GFP_KERNEL);
1614         if (!rf->hmc_info_mem)
1615                 return -ENOMEM;
1616
1617         rf->pble_rsrc = (struct irdma_hmc_pble_rsrc *)rf->hmc_info_mem;
1618         dev->hmc_info = &rf->hw.hmc;
1619         dev->hmc_info->hmc_obj = (struct irdma_hmc_obj_info *)
1620                                  (rf->pble_rsrc + 1);
1621
1622         status = irdma_obj_aligned_mem(rf, &mem, IRDMA_QUERY_FPM_BUF_SIZE,
1623                                        IRDMA_FPM_QUERY_BUF_ALIGNMENT_M);
1624         if (status)
1625                 goto error;
1626
1627         info.fpm_query_buf_pa = mem.pa;
1628         info.fpm_query_buf = mem.va;
1629
1630         status = irdma_obj_aligned_mem(rf, &mem, IRDMA_COMMIT_FPM_BUF_SIZE,
1631                                        IRDMA_FPM_COMMIT_BUF_ALIGNMENT_M);
1632         if (status)
1633                 goto error;
1634
1635         info.fpm_commit_buf_pa = mem.pa;
1636         info.fpm_commit_buf = mem.va;
1637
1638         info.bar0 = rf->hw.hw_addr;
1639         info.hmc_fn_id = rf->pf_id;
1640         info.hw = &rf->hw;
1641         status = irdma_sc_dev_init(rf->rdma_ver, &rf->sc_dev, &info);
1642         if (status)
1643                 goto error;
1644
1645         return status;
1646 error:
1647         kfree(rf->hmc_info_mem);
1648         rf->hmc_info_mem = NULL;
1649
1650         return status;
1651 }
1652
1653 /**
1654  * irdma_rt_deinit_hw - clean up the irdma device resources
1655  * @iwdev: irdma device
1656  *
1657  * remove the mac ip entry and ipv4/ipv6 addresses, destroy the
1658  * device queues and free the pble and the hmc objects
1659  */
1660 void irdma_rt_deinit_hw(struct irdma_device *iwdev)
1661 {
1662         ibdev_dbg(&iwdev->ibdev, "INIT: state = %d\n", iwdev->init_state);
1663
1664         switch (iwdev->init_state) {
1665         case IP_ADDR_REGISTERED:
1666                 if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_1)
1667                         irdma_del_local_mac_entry(iwdev->rf,
1668                                                   (u8)iwdev->mac_ip_table_idx);
1669                 fallthrough;
1670         case AEQ_CREATED:
1671         case PBLE_CHUNK_MEM:
1672         case CEQS_CREATED:
1673         case IEQ_CREATED:
1674                 if (!iwdev->roce_mode)
1675                         irdma_puda_dele_rsrc(&iwdev->vsi, IRDMA_PUDA_RSRC_TYPE_IEQ,
1676                                              iwdev->rf->reset);
1677                 fallthrough;
1678         case ILQ_CREATED:
1679                 if (!iwdev->roce_mode)
1680                         irdma_puda_dele_rsrc(&iwdev->vsi,
1681                                              IRDMA_PUDA_RSRC_TYPE_ILQ,
1682                                              iwdev->rf->reset);
1683                 break;
1684         default:
1685                 ibdev_warn(&iwdev->ibdev, "bad init_state = %d\n", iwdev->init_state);
1686                 break;
1687         }
1688
1689         irdma_cleanup_cm_core(&iwdev->cm_core);
1690         if (iwdev->vsi.pestat) {
1691                 irdma_vsi_stats_free(&iwdev->vsi);
1692                 kfree(iwdev->vsi.pestat);
1693         }
1694         if (iwdev->cleanup_wq)
1695                 destroy_workqueue(iwdev->cleanup_wq);
1696 }
1697
1698 static int irdma_setup_init_state(struct irdma_pci_f *rf)
1699 {
1700         int status;
1701
1702         status = irdma_save_msix_info(rf);
1703         if (status)
1704                 return status;
1705
1706         rf->hw.device = &rf->pcidev->dev;
1707         rf->obj_mem.size = ALIGN(8192, IRDMA_HW_PAGE_SIZE);
1708         rf->obj_mem.va = dma_alloc_coherent(rf->hw.device, rf->obj_mem.size,
1709                                             &rf->obj_mem.pa, GFP_KERNEL);
1710         if (!rf->obj_mem.va) {
1711                 status = -ENOMEM;
1712                 goto clean_msixtbl;
1713         }
1714
1715         rf->obj_next = rf->obj_mem;
1716         status = irdma_initialize_dev(rf);
1717         if (status)
1718                 goto clean_obj_mem;
1719
1720         return 0;
1721
1722 clean_obj_mem:
1723         dma_free_coherent(rf->hw.device, rf->obj_mem.size, rf->obj_mem.va,
1724                           rf->obj_mem.pa);
1725         rf->obj_mem.va = NULL;
1726 clean_msixtbl:
1727         kfree(rf->iw_msixtbl);
1728         rf->iw_msixtbl = NULL;
1729         return status;
1730 }
1731
1732 /**
1733  * irdma_get_used_rsrc - determine resources used internally
1734  * @iwdev: irdma device
1735  *
1736  * Called at the end of open to get all internal allocations
1737  */
1738 static void irdma_get_used_rsrc(struct irdma_device *iwdev)
1739 {
1740         iwdev->rf->used_pds = find_first_zero_bit(iwdev->rf->allocated_pds,
1741                                                  iwdev->rf->max_pd);
1742         iwdev->rf->used_qps = find_first_zero_bit(iwdev->rf->allocated_qps,
1743                                                  iwdev->rf->max_qp);
1744         iwdev->rf->used_cqs = find_first_zero_bit(iwdev->rf->allocated_cqs,
1745                                                  iwdev->rf->max_cq);
1746         iwdev->rf->used_mrs = find_first_zero_bit(iwdev->rf->allocated_mrs,
1747                                                  iwdev->rf->max_mr);
1748 }
1749
1750 void irdma_ctrl_deinit_hw(struct irdma_pci_f *rf)
1751 {
1752         enum init_completion_state state = rf->init_state;
1753
1754         rf->init_state = INVALID_STATE;
1755         if (rf->rsrc_created) {
1756                 irdma_destroy_aeq(rf);
1757                 irdma_destroy_pble_prm(rf->pble_rsrc);
1758                 irdma_del_ceqs(rf);
1759                 rf->rsrc_created = false;
1760         }
1761         switch (state) {
1762         case CEQ0_CREATED:
1763                 irdma_del_ceq_0(rf);
1764                 fallthrough;
1765         case CCQ_CREATED:
1766                 irdma_destroy_ccq(rf);
1767                 fallthrough;
1768         case HW_RSRC_INITIALIZED:
1769         case HMC_OBJS_CREATED:
1770                 irdma_del_hmc_objects(&rf->sc_dev, rf->sc_dev.hmc_info, true,
1771                                       rf->reset, rf->rdma_ver);
1772                 fallthrough;
1773         case CQP_CREATED:
1774                 irdma_destroy_cqp(rf);
1775                 fallthrough;
1776         case INITIAL_STATE:
1777                 irdma_del_init_mem(rf);
1778                 break;
1779         case INVALID_STATE:
1780         default:
1781                 ibdev_warn(&rf->iwdev->ibdev, "bad init_state = %d\n", rf->init_state);
1782                 break;
1783         }
1784 }
1785
1786 /**
1787  * irdma_rt_init_hw - Initializes runtime portion of HW
1788  * @iwdev: irdma device
1789  * @l2params: qos, tc, mtu info from netdev driver
1790  *
1791  * Create device queues ILQ, IEQ, CEQs and PBLEs. Setup irdma
1792  * device resource objects.
1793  */
1794 int irdma_rt_init_hw(struct irdma_device *iwdev,
1795                      struct irdma_l2params *l2params)
1796 {
1797         struct irdma_pci_f *rf = iwdev->rf;
1798         struct irdma_sc_dev *dev = &rf->sc_dev;
1799         struct irdma_vsi_init_info vsi_info = {};
1800         struct irdma_vsi_stats_info stats_info = {};
1801         int status;
1802
1803         vsi_info.dev = dev;
1804         vsi_info.back_vsi = iwdev;
1805         vsi_info.params = l2params;
1806         vsi_info.pf_data_vsi_num = iwdev->vsi_num;
1807         vsi_info.register_qset = rf->gen_ops.register_qset;
1808         vsi_info.unregister_qset = rf->gen_ops.unregister_qset;
1809         vsi_info.exception_lan_q = 2;
1810         irdma_sc_vsi_init(&iwdev->vsi, &vsi_info);
1811
1812         status = irdma_setup_cm_core(iwdev, rf->rdma_ver);
1813         if (status)
1814                 return status;
1815
1816         stats_info.pestat = kzalloc(sizeof(*stats_info.pestat), GFP_KERNEL);
1817         if (!stats_info.pestat) {
1818                 irdma_cleanup_cm_core(&iwdev->cm_core);
1819                 return -ENOMEM;
1820         }
1821         stats_info.fcn_id = dev->hmc_fn_id;
1822         status = irdma_vsi_stats_init(&iwdev->vsi, &stats_info);
1823         if (status) {
1824                 irdma_cleanup_cm_core(&iwdev->cm_core);
1825                 kfree(stats_info.pestat);
1826                 return status;
1827         }
1828
1829         do {
1830                 if (!iwdev->roce_mode) {
1831                         status = irdma_initialize_ilq(iwdev);
1832                         if (status)
1833                                 break;
1834                         iwdev->init_state = ILQ_CREATED;
1835                         status = irdma_initialize_ieq(iwdev);
1836                         if (status)
1837                                 break;
1838                         iwdev->init_state = IEQ_CREATED;
1839                 }
1840                 if (!rf->rsrc_created) {
1841                         status = irdma_setup_ceqs(rf, &iwdev->vsi);
1842                         if (status)
1843                                 break;
1844
1845                         iwdev->init_state = CEQS_CREATED;
1846
1847                         status = irdma_hmc_init_pble(&rf->sc_dev,
1848                                                      rf->pble_rsrc);
1849                         if (status) {
1850                                 irdma_del_ceqs(rf);
1851                                 break;
1852                         }
1853
1854                         iwdev->init_state = PBLE_CHUNK_MEM;
1855
1856                         status = irdma_setup_aeq(rf);
1857                         if (status) {
1858                                 irdma_destroy_pble_prm(rf->pble_rsrc);
1859                                 irdma_del_ceqs(rf);
1860                                 break;
1861                         }
1862                         iwdev->init_state = AEQ_CREATED;
1863                         rf->rsrc_created = true;
1864                 }
1865
1866                 if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_1)
1867                         irdma_alloc_set_mac(iwdev);
1868                 irdma_add_ip(iwdev);
1869                 iwdev->init_state = IP_ADDR_REGISTERED;
1870
1871                 /* handles asynch cleanup tasks - disconnect CM , free qp,
1872                  * free cq bufs
1873                  */
1874                 iwdev->cleanup_wq = alloc_workqueue("irdma-cleanup-wq",
1875                                         WQ_UNBOUND, WQ_UNBOUND_MAX_ACTIVE);
1876                 if (!iwdev->cleanup_wq)
1877                         return -ENOMEM;
1878                 irdma_get_used_rsrc(iwdev);
1879                 init_waitqueue_head(&iwdev->suspend_wq);
1880
1881                 return 0;
1882         } while (0);
1883
1884         dev_err(&rf->pcidev->dev, "HW runtime init FAIL status = %d last cmpl = %d\n",
1885                 status, iwdev->init_state);
1886         irdma_rt_deinit_hw(iwdev);
1887
1888         return status;
1889 }
1890
1891 /**
1892  * irdma_ctrl_init_hw - Initializes control portion of HW
1893  * @rf: RDMA PCI function
1894  *
1895  * Create admin queues, HMC obejcts and RF resource objects
1896  */
1897 int irdma_ctrl_init_hw(struct irdma_pci_f *rf)
1898 {
1899         struct irdma_sc_dev *dev = &rf->sc_dev;
1900         int status;
1901         do {
1902                 status = irdma_setup_init_state(rf);
1903                 if (status)
1904                         break;
1905                 rf->init_state = INITIAL_STATE;
1906
1907                 status = irdma_create_cqp(rf);
1908                 if (status)
1909                         break;
1910                 rf->init_state = CQP_CREATED;
1911
1912                 status = irdma_hmc_setup(rf);
1913                 if (status)
1914                         break;
1915                 rf->init_state = HMC_OBJS_CREATED;
1916
1917                 status = irdma_initialize_hw_rsrc(rf);
1918                 if (status)
1919                         break;
1920                 rf->init_state = HW_RSRC_INITIALIZED;
1921
1922                 status = irdma_create_ccq(rf);
1923                 if (status)
1924                         break;
1925                 rf->init_state = CCQ_CREATED;
1926
1927                 dev->feature_info[IRDMA_FEATURE_FW_INFO] = IRDMA_FW_VER_DEFAULT;
1928                 if (rf->rdma_ver != IRDMA_GEN_1) {
1929                         status = irdma_get_rdma_features(dev);
1930                         if (status)
1931                                 break;
1932                 }
1933
1934                 status = irdma_setup_ceq_0(rf);
1935                 if (status)
1936                         break;
1937                 rf->init_state = CEQ0_CREATED;
1938                 /* Handles processing of CQP completions */
1939                 rf->cqp_cmpl_wq =
1940                         alloc_ordered_workqueue("cqp_cmpl_wq", WQ_HIGHPRI);
1941                 if (!rf->cqp_cmpl_wq) {
1942                         status = -ENOMEM;
1943                         break;
1944                 }
1945                 INIT_WORK(&rf->cqp_cmpl_work, cqp_compl_worker);
1946                 irdma_sc_ccq_arm(dev->ccq);
1947                 return 0;
1948         } while (0);
1949
1950         dev_err(&rf->pcidev->dev, "IRDMA hardware initialization FAILED init_state=%d status=%d\n",
1951                 rf->init_state, status);
1952         irdma_ctrl_deinit_hw(rf);
1953         return status;
1954 }
1955
1956 /**
1957  * irdma_set_hw_rsrc - set hw memory resources.
1958  * @rf: RDMA PCI function
1959  */
1960 static void irdma_set_hw_rsrc(struct irdma_pci_f *rf)
1961 {
1962         rf->allocated_qps = (void *)(rf->mem_rsrc +
1963                    (sizeof(struct irdma_arp_entry) * rf->arp_table_size));
1964         rf->allocated_cqs = &rf->allocated_qps[BITS_TO_LONGS(rf->max_qp)];
1965         rf->allocated_mrs = &rf->allocated_cqs[BITS_TO_LONGS(rf->max_cq)];
1966         rf->allocated_pds = &rf->allocated_mrs[BITS_TO_LONGS(rf->max_mr)];
1967         rf->allocated_ahs = &rf->allocated_pds[BITS_TO_LONGS(rf->max_pd)];
1968         rf->allocated_mcgs = &rf->allocated_ahs[BITS_TO_LONGS(rf->max_ah)];
1969         rf->allocated_arps = &rf->allocated_mcgs[BITS_TO_LONGS(rf->max_mcg)];
1970         rf->qp_table = (struct irdma_qp **)
1971                 (&rf->allocated_arps[BITS_TO_LONGS(rf->arp_table_size)]);
1972         rf->cq_table = (struct irdma_cq **)(&rf->qp_table[rf->max_qp]);
1973
1974         spin_lock_init(&rf->rsrc_lock);
1975         spin_lock_init(&rf->arp_lock);
1976         spin_lock_init(&rf->qptable_lock);
1977         spin_lock_init(&rf->cqtable_lock);
1978         spin_lock_init(&rf->qh_list_lock);
1979 }
1980
1981 /**
1982  * irdma_calc_mem_rsrc_size - calculate memory resources size.
1983  * @rf: RDMA PCI function
1984  */
1985 static u32 irdma_calc_mem_rsrc_size(struct irdma_pci_f *rf)
1986 {
1987         u32 rsrc_size;
1988
1989         rsrc_size = sizeof(struct irdma_arp_entry) * rf->arp_table_size;
1990         rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_qp);
1991         rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_mr);
1992         rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_cq);
1993         rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_pd);
1994         rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->arp_table_size);
1995         rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_ah);
1996         rsrc_size += sizeof(unsigned long) * BITS_TO_LONGS(rf->max_mcg);
1997         rsrc_size += sizeof(struct irdma_qp **) * rf->max_qp;
1998         rsrc_size += sizeof(struct irdma_cq **) * rf->max_cq;
1999
2000         return rsrc_size;
2001 }
2002
2003 /**
2004  * irdma_initialize_hw_rsrc - initialize hw resource tracking array
2005  * @rf: RDMA PCI function
2006  */
2007 u32 irdma_initialize_hw_rsrc(struct irdma_pci_f *rf)
2008 {
2009         u32 rsrc_size;
2010         u32 mrdrvbits;
2011         u32 ret;
2012
2013         if (rf->rdma_ver != IRDMA_GEN_1) {
2014                 rf->allocated_ws_nodes = bitmap_zalloc(IRDMA_MAX_WS_NODES,
2015                                                        GFP_KERNEL);
2016                 if (!rf->allocated_ws_nodes)
2017                         return -ENOMEM;
2018
2019                 set_bit(0, rf->allocated_ws_nodes);
2020                 rf->max_ws_node_id = IRDMA_MAX_WS_NODES;
2021         }
2022         rf->max_cqe = rf->sc_dev.hw_attrs.uk_attrs.max_hw_cq_size;
2023         rf->max_qp = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_QP].cnt;
2024         rf->max_mr = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_MR].cnt;
2025         rf->max_cq = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_CQ].cnt;
2026         rf->max_pd = rf->sc_dev.hw_attrs.max_hw_pds;
2027         rf->arp_table_size = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_ARP].cnt;
2028         rf->max_ah = rf->sc_dev.hmc_info->hmc_obj[IRDMA_HMC_IW_FSIAV].cnt;
2029         rf->max_mcg = rf->max_qp;
2030
2031         rsrc_size = irdma_calc_mem_rsrc_size(rf);
2032         rf->mem_rsrc = vzalloc(rsrc_size);
2033         if (!rf->mem_rsrc) {
2034                 ret = -ENOMEM;
2035                 goto mem_rsrc_vzalloc_fail;
2036         }
2037
2038         rf->arp_table = (struct irdma_arp_entry *)rf->mem_rsrc;
2039
2040         irdma_set_hw_rsrc(rf);
2041
2042         set_bit(0, rf->allocated_mrs);
2043         set_bit(0, rf->allocated_qps);
2044         set_bit(0, rf->allocated_cqs);
2045         set_bit(0, rf->allocated_pds);
2046         set_bit(0, rf->allocated_arps);
2047         set_bit(0, rf->allocated_ahs);
2048         set_bit(0, rf->allocated_mcgs);
2049         set_bit(2, rf->allocated_qps); /* qp 2 IEQ */
2050         set_bit(1, rf->allocated_qps); /* qp 1 ILQ */
2051         set_bit(1, rf->allocated_cqs);
2052         set_bit(1, rf->allocated_pds);
2053         set_bit(2, rf->allocated_cqs);
2054         set_bit(2, rf->allocated_pds);
2055
2056         INIT_LIST_HEAD(&rf->mc_qht_list.list);
2057         /* stag index mask has a minimum of 14 bits */
2058         mrdrvbits = 24 - max(get_count_order(rf->max_mr), 14);
2059         rf->mr_stagmask = ~(((1 << mrdrvbits) - 1) << (32 - mrdrvbits));
2060
2061         return 0;
2062
2063 mem_rsrc_vzalloc_fail:
2064         bitmap_free(rf->allocated_ws_nodes);
2065         rf->allocated_ws_nodes = NULL;
2066
2067         return ret;
2068 }
2069
2070 /**
2071  * irdma_cqp_ce_handler - handle cqp completions
2072  * @rf: RDMA PCI function
2073  * @cq: cq for cqp completions
2074  */
2075 void irdma_cqp_ce_handler(struct irdma_pci_f *rf, struct irdma_sc_cq *cq)
2076 {
2077         struct irdma_cqp_request *cqp_request;
2078         struct irdma_sc_dev *dev = &rf->sc_dev;
2079         u32 cqe_count = 0;
2080         struct irdma_ccq_cqe_info info;
2081         unsigned long flags;
2082         int ret;
2083
2084         do {
2085                 memset(&info, 0, sizeof(info));
2086                 spin_lock_irqsave(&rf->cqp.compl_lock, flags);
2087                 ret = irdma_sc_ccq_get_cqe_info(cq, &info);
2088                 spin_unlock_irqrestore(&rf->cqp.compl_lock, flags);
2089                 if (ret)
2090                         break;
2091
2092                 cqp_request = (struct irdma_cqp_request *)
2093                               (unsigned long)info.scratch;
2094                 if (info.error && irdma_cqp_crit_err(dev, cqp_request->info.cqp_cmd,
2095                                                      info.maj_err_code,
2096                                                      info.min_err_code))
2097                         ibdev_err(&rf->iwdev->ibdev, "cqp opcode = 0x%x maj_err_code = 0x%x min_err_code = 0x%x\n",
2098                                   info.op_code, info.maj_err_code, info.min_err_code);
2099                 if (cqp_request) {
2100                         cqp_request->compl_info.maj_err_code = info.maj_err_code;
2101                         cqp_request->compl_info.min_err_code = info.min_err_code;
2102                         cqp_request->compl_info.op_ret_val = info.op_ret_val;
2103                         cqp_request->compl_info.error = info.error;
2104
2105                         if (cqp_request->waiting) {
2106                                 WRITE_ONCE(cqp_request->request_done, true);
2107                                 wake_up(&cqp_request->waitq);
2108                                 irdma_put_cqp_request(&rf->cqp, cqp_request);
2109                         } else {
2110                                 if (cqp_request->callback_fcn)
2111                                         cqp_request->callback_fcn(cqp_request);
2112                                 irdma_put_cqp_request(&rf->cqp, cqp_request);
2113                         }
2114                 }
2115
2116                 cqe_count++;
2117         } while (1);
2118
2119         if (cqe_count) {
2120                 irdma_process_bh(dev);
2121                 irdma_sc_ccq_arm(cq);
2122         }
2123 }
2124
2125 /**
2126  * cqp_compl_worker - Handle cqp completions
2127  * @work: Pointer to work structure
2128  */
2129 void cqp_compl_worker(struct work_struct *work)
2130 {
2131         struct irdma_pci_f *rf = container_of(work, struct irdma_pci_f,
2132                                               cqp_cmpl_work);
2133         struct irdma_sc_cq *cq = &rf->ccq.sc_cq;
2134
2135         irdma_cqp_ce_handler(rf, cq);
2136 }
2137
2138 /**
2139  * irdma_lookup_apbvt_entry - lookup hash table for an existing apbvt entry corresponding to port
2140  * @cm_core: cm's core
2141  * @port: port to identify apbvt entry
2142  */
2143 static struct irdma_apbvt_entry *irdma_lookup_apbvt_entry(struct irdma_cm_core *cm_core,
2144                                                           u16 port)
2145 {
2146         struct irdma_apbvt_entry *entry;
2147
2148         hash_for_each_possible(cm_core->apbvt_hash_tbl, entry, hlist, port) {
2149                 if (entry->port == port) {
2150                         entry->use_cnt++;
2151                         return entry;
2152                 }
2153         }
2154
2155         return NULL;
2156 }
2157
2158 /**
2159  * irdma_next_iw_state - modify qp state
2160  * @iwqp: iwarp qp to modify
2161  * @state: next state for qp
2162  * @del_hash: del hash
2163  * @term: term message
2164  * @termlen: length of term message
2165  */
2166 void irdma_next_iw_state(struct irdma_qp *iwqp, u8 state, u8 del_hash, u8 term,
2167                          u8 termlen)
2168 {
2169         struct irdma_modify_qp_info info = {};
2170
2171         info.next_iwarp_state = state;
2172         info.remove_hash_idx = del_hash;
2173         info.cq_num_valid = true;
2174         info.arp_cache_idx_valid = true;
2175         info.dont_send_term = true;
2176         info.dont_send_fin = true;
2177         info.termlen = termlen;
2178
2179         if (term & IRDMAQP_TERM_SEND_TERM_ONLY)
2180                 info.dont_send_term = false;
2181         if (term & IRDMAQP_TERM_SEND_FIN_ONLY)
2182                 info.dont_send_fin = false;
2183         if (iwqp->sc_qp.term_flags && state == IRDMA_QP_STATE_ERROR)
2184                 info.reset_tcp_conn = true;
2185         iwqp->hw_iwarp_state = state;
2186         irdma_hw_modify_qp(iwqp->iwdev, iwqp, &info, 0);
2187         iwqp->iwarp_state = info.next_iwarp_state;
2188 }
2189
2190 /**
2191  * irdma_del_local_mac_entry - remove a mac entry from the hw
2192  * table
2193  * @rf: RDMA PCI function
2194  * @idx: the index of the mac ip address to delete
2195  */
2196 void irdma_del_local_mac_entry(struct irdma_pci_f *rf, u16 idx)
2197 {
2198         struct irdma_cqp *iwcqp = &rf->cqp;
2199         struct irdma_cqp_request *cqp_request;
2200         struct cqp_cmds_info *cqp_info;
2201
2202         cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, true);
2203         if (!cqp_request)
2204                 return;
2205
2206         cqp_info = &cqp_request->info;
2207         cqp_info->cqp_cmd = IRDMA_OP_DELETE_LOCAL_MAC_ENTRY;
2208         cqp_info->post_sq = 1;
2209         cqp_info->in.u.del_local_mac_entry.cqp = &iwcqp->sc_cqp;
2210         cqp_info->in.u.del_local_mac_entry.scratch = (uintptr_t)cqp_request;
2211         cqp_info->in.u.del_local_mac_entry.entry_idx = idx;
2212         cqp_info->in.u.del_local_mac_entry.ignore_ref_count = 0;
2213
2214         irdma_handle_cqp_op(rf, cqp_request);
2215         irdma_put_cqp_request(iwcqp, cqp_request);
2216 }
2217
2218 /**
2219  * irdma_add_local_mac_entry - add a mac ip address entry to the
2220  * hw table
2221  * @rf: RDMA PCI function
2222  * @mac_addr: pointer to mac address
2223  * @idx: the index of the mac ip address to add
2224  */
2225 int irdma_add_local_mac_entry(struct irdma_pci_f *rf, const u8 *mac_addr, u16 idx)
2226 {
2227         struct irdma_local_mac_entry_info *info;
2228         struct irdma_cqp *iwcqp = &rf->cqp;
2229         struct irdma_cqp_request *cqp_request;
2230         struct cqp_cmds_info *cqp_info;
2231         int status;
2232
2233         cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, true);
2234         if (!cqp_request)
2235                 return -ENOMEM;
2236
2237         cqp_info = &cqp_request->info;
2238         cqp_info->post_sq = 1;
2239         info = &cqp_info->in.u.add_local_mac_entry.info;
2240         ether_addr_copy(info->mac_addr, mac_addr);
2241         info->entry_idx = idx;
2242         cqp_info->in.u.add_local_mac_entry.scratch = (uintptr_t)cqp_request;
2243         cqp_info->cqp_cmd = IRDMA_OP_ADD_LOCAL_MAC_ENTRY;
2244         cqp_info->in.u.add_local_mac_entry.cqp = &iwcqp->sc_cqp;
2245         cqp_info->in.u.add_local_mac_entry.scratch = (uintptr_t)cqp_request;
2246
2247         status = irdma_handle_cqp_op(rf, cqp_request);
2248         irdma_put_cqp_request(iwcqp, cqp_request);
2249
2250         return status;
2251 }
2252
2253 /**
2254  * irdma_alloc_local_mac_entry - allocate a mac entry
2255  * @rf: RDMA PCI function
2256  * @mac_tbl_idx: the index of the new mac address
2257  *
2258  * Allocate a mac address entry and update the mac_tbl_idx
2259  * to hold the index of the newly created mac address
2260  * Return 0 if successful, otherwise return error
2261  */
2262 int irdma_alloc_local_mac_entry(struct irdma_pci_f *rf, u16 *mac_tbl_idx)
2263 {
2264         struct irdma_cqp *iwcqp = &rf->cqp;
2265         struct irdma_cqp_request *cqp_request;
2266         struct cqp_cmds_info *cqp_info;
2267         int status = 0;
2268
2269         cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, true);
2270         if (!cqp_request)
2271                 return -ENOMEM;
2272
2273         cqp_info = &cqp_request->info;
2274         cqp_info->cqp_cmd = IRDMA_OP_ALLOC_LOCAL_MAC_ENTRY;
2275         cqp_info->post_sq = 1;
2276         cqp_info->in.u.alloc_local_mac_entry.cqp = &iwcqp->sc_cqp;
2277         cqp_info->in.u.alloc_local_mac_entry.scratch = (uintptr_t)cqp_request;
2278         status = irdma_handle_cqp_op(rf, cqp_request);
2279         if (!status)
2280                 *mac_tbl_idx = (u16)cqp_request->compl_info.op_ret_val;
2281
2282         irdma_put_cqp_request(iwcqp, cqp_request);
2283
2284         return status;
2285 }
2286
2287 /**
2288  * irdma_cqp_manage_apbvt_cmd - send cqp command manage apbvt
2289  * @iwdev: irdma device
2290  * @accel_local_port: port for apbvt
2291  * @add_port: add ordelete port
2292  */
2293 static int irdma_cqp_manage_apbvt_cmd(struct irdma_device *iwdev,
2294                                       u16 accel_local_port, bool add_port)
2295 {
2296         struct irdma_apbvt_info *info;
2297         struct irdma_cqp_request *cqp_request;
2298         struct cqp_cmds_info *cqp_info;
2299         int status;
2300
2301         cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, add_port);
2302         if (!cqp_request)
2303                 return -ENOMEM;
2304
2305         cqp_info = &cqp_request->info;
2306         info = &cqp_info->in.u.manage_apbvt_entry.info;
2307         memset(info, 0, sizeof(*info));
2308         info->add = add_port;
2309         info->port = accel_local_port;
2310         cqp_info->cqp_cmd = IRDMA_OP_MANAGE_APBVT_ENTRY;
2311         cqp_info->post_sq = 1;
2312         cqp_info->in.u.manage_apbvt_entry.cqp = &iwdev->rf->cqp.sc_cqp;
2313         cqp_info->in.u.manage_apbvt_entry.scratch = (uintptr_t)cqp_request;
2314         ibdev_dbg(&iwdev->ibdev, "DEV: %s: port=0x%04x\n",
2315                   (!add_port) ? "DELETE" : "ADD", accel_local_port);
2316
2317         status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2318         irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2319
2320         return status;
2321 }
2322
2323 /**
2324  * irdma_add_apbvt - add tcp port to HW apbvt table
2325  * @iwdev: irdma device
2326  * @port: port for apbvt
2327  */
2328 struct irdma_apbvt_entry *irdma_add_apbvt(struct irdma_device *iwdev, u16 port)
2329 {
2330         struct irdma_cm_core *cm_core = &iwdev->cm_core;
2331         struct irdma_apbvt_entry *entry;
2332         unsigned long flags;
2333
2334         spin_lock_irqsave(&cm_core->apbvt_lock, flags);
2335         entry = irdma_lookup_apbvt_entry(cm_core, port);
2336         if (entry) {
2337                 spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2338                 return entry;
2339         }
2340
2341         entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
2342         if (!entry) {
2343                 spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2344                 return NULL;
2345         }
2346
2347         entry->port = port;
2348         entry->use_cnt = 1;
2349         hash_add(cm_core->apbvt_hash_tbl, &entry->hlist, entry->port);
2350         spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2351
2352         if (irdma_cqp_manage_apbvt_cmd(iwdev, port, true)) {
2353                 kfree(entry);
2354                 return NULL;
2355         }
2356
2357         return entry;
2358 }
2359
2360 /**
2361  * irdma_del_apbvt - delete tcp port from HW apbvt table
2362  * @iwdev: irdma device
2363  * @entry: apbvt entry object
2364  */
2365 void irdma_del_apbvt(struct irdma_device *iwdev,
2366                      struct irdma_apbvt_entry *entry)
2367 {
2368         struct irdma_cm_core *cm_core = &iwdev->cm_core;
2369         unsigned long flags;
2370
2371         spin_lock_irqsave(&cm_core->apbvt_lock, flags);
2372         if (--entry->use_cnt) {
2373                 spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2374                 return;
2375         }
2376
2377         hash_del(&entry->hlist);
2378         /* apbvt_lock is held across CQP delete APBVT OP (non-waiting) to
2379          * protect against race where add APBVT CQP can race ahead of the delete
2380          * APBVT for same port.
2381          */
2382         irdma_cqp_manage_apbvt_cmd(iwdev, entry->port, false);
2383         kfree(entry);
2384         spin_unlock_irqrestore(&cm_core->apbvt_lock, flags);
2385 }
2386
2387 /**
2388  * irdma_manage_arp_cache - manage hw arp cache
2389  * @rf: RDMA PCI function
2390  * @mac_addr: mac address ptr
2391  * @ip_addr: ip addr for arp cache
2392  * @ipv4: flag inicating IPv4
2393  * @action: add, delete or modify
2394  */
2395 void irdma_manage_arp_cache(struct irdma_pci_f *rf,
2396                             const unsigned char *mac_addr,
2397                             u32 *ip_addr, bool ipv4, u32 action)
2398 {
2399         struct irdma_add_arp_cache_entry_info *info;
2400         struct irdma_cqp_request *cqp_request;
2401         struct cqp_cmds_info *cqp_info;
2402         int arp_index;
2403
2404         arp_index = irdma_arp_table(rf, ip_addr, ipv4, mac_addr, action);
2405         if (arp_index == -1)
2406                 return;
2407
2408         cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, false);
2409         if (!cqp_request)
2410                 return;
2411
2412         cqp_info = &cqp_request->info;
2413         if (action == IRDMA_ARP_ADD) {
2414                 cqp_info->cqp_cmd = IRDMA_OP_ADD_ARP_CACHE_ENTRY;
2415                 info = &cqp_info->in.u.add_arp_cache_entry.info;
2416                 memset(info, 0, sizeof(*info));
2417                 info->arp_index = (u16)arp_index;
2418                 info->permanent = true;
2419                 ether_addr_copy(info->mac_addr, mac_addr);
2420                 cqp_info->in.u.add_arp_cache_entry.scratch =
2421                         (uintptr_t)cqp_request;
2422                 cqp_info->in.u.add_arp_cache_entry.cqp = &rf->cqp.sc_cqp;
2423         } else {
2424                 cqp_info->cqp_cmd = IRDMA_OP_DELETE_ARP_CACHE_ENTRY;
2425                 cqp_info->in.u.del_arp_cache_entry.scratch =
2426                         (uintptr_t)cqp_request;
2427                 cqp_info->in.u.del_arp_cache_entry.cqp = &rf->cqp.sc_cqp;
2428                 cqp_info->in.u.del_arp_cache_entry.arp_index = arp_index;
2429         }
2430
2431         cqp_info->post_sq = 1;
2432         irdma_handle_cqp_op(rf, cqp_request);
2433         irdma_put_cqp_request(&rf->cqp, cqp_request);
2434 }
2435
2436 /**
2437  * irdma_send_syn_cqp_callback - do syn/ack after qhash
2438  * @cqp_request: qhash cqp completion
2439  */
2440 static void irdma_send_syn_cqp_callback(struct irdma_cqp_request *cqp_request)
2441 {
2442         struct irdma_cm_node *cm_node = cqp_request->param;
2443
2444         irdma_send_syn(cm_node, 1);
2445         irdma_rem_ref_cm_node(cm_node);
2446 }
2447
2448 /**
2449  * irdma_manage_qhash - add or modify qhash
2450  * @iwdev: irdma device
2451  * @cminfo: cm info for qhash
2452  * @etype: type (syn or quad)
2453  * @mtype: type of qhash
2454  * @cmnode: cmnode associated with connection
2455  * @wait: wait for completion
2456  */
2457 int irdma_manage_qhash(struct irdma_device *iwdev, struct irdma_cm_info *cminfo,
2458                        enum irdma_quad_entry_type etype,
2459                        enum irdma_quad_hash_manage_type mtype, void *cmnode,
2460                        bool wait)
2461 {
2462         struct irdma_qhash_table_info *info;
2463         struct irdma_cqp *iwcqp = &iwdev->rf->cqp;
2464         struct irdma_cqp_request *cqp_request;
2465         struct cqp_cmds_info *cqp_info;
2466         struct irdma_cm_node *cm_node = cmnode;
2467         int status;
2468
2469         cqp_request = irdma_alloc_and_get_cqp_request(iwcqp, wait);
2470         if (!cqp_request)
2471                 return -ENOMEM;
2472
2473         cqp_info = &cqp_request->info;
2474         info = &cqp_info->in.u.manage_qhash_table_entry.info;
2475         memset(info, 0, sizeof(*info));
2476         info->vsi = &iwdev->vsi;
2477         info->manage = mtype;
2478         info->entry_type = etype;
2479         if (cminfo->vlan_id < VLAN_N_VID) {
2480                 info->vlan_valid = true;
2481                 info->vlan_id = cminfo->vlan_id;
2482         } else {
2483                 info->vlan_valid = false;
2484         }
2485         info->ipv4_valid = cminfo->ipv4;
2486         info->user_pri = cminfo->user_pri;
2487         ether_addr_copy(info->mac_addr, iwdev->netdev->dev_addr);
2488         info->qp_num = cminfo->qh_qpid;
2489         info->dest_port = cminfo->loc_port;
2490         info->dest_ip[0] = cminfo->loc_addr[0];
2491         info->dest_ip[1] = cminfo->loc_addr[1];
2492         info->dest_ip[2] = cminfo->loc_addr[2];
2493         info->dest_ip[3] = cminfo->loc_addr[3];
2494         if (etype == IRDMA_QHASH_TYPE_TCP_ESTABLISHED ||
2495             etype == IRDMA_QHASH_TYPE_UDP_UNICAST ||
2496             etype == IRDMA_QHASH_TYPE_UDP_MCAST ||
2497             etype == IRDMA_QHASH_TYPE_ROCE_MCAST ||
2498             etype == IRDMA_QHASH_TYPE_ROCEV2_HW) {
2499                 info->src_port = cminfo->rem_port;
2500                 info->src_ip[0] = cminfo->rem_addr[0];
2501                 info->src_ip[1] = cminfo->rem_addr[1];
2502                 info->src_ip[2] = cminfo->rem_addr[2];
2503                 info->src_ip[3] = cminfo->rem_addr[3];
2504         }
2505         if (cmnode) {
2506                 cqp_request->callback_fcn = irdma_send_syn_cqp_callback;
2507                 cqp_request->param = cmnode;
2508                 if (!wait)
2509                         refcount_inc(&cm_node->refcnt);
2510         }
2511         if (info->ipv4_valid)
2512                 ibdev_dbg(&iwdev->ibdev,
2513                           "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",
2514                           (!mtype) ? "DELETE" : "ADD",
2515                           __builtin_return_address(0), info->dest_port,
2516                           info->src_port, info->dest_ip, info->src_ip,
2517                           info->mac_addr, cminfo->vlan_id,
2518                           cmnode ? cmnode : NULL);
2519         else
2520                 ibdev_dbg(&iwdev->ibdev,
2521                           "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",
2522                           (!mtype) ? "DELETE" : "ADD",
2523                           __builtin_return_address(0), info->dest_port,
2524                           info->src_port, info->dest_ip, info->src_ip,
2525                           info->mac_addr, cminfo->vlan_id,
2526                           cmnode ? cmnode : NULL);
2527
2528         cqp_info->in.u.manage_qhash_table_entry.cqp = &iwdev->rf->cqp.sc_cqp;
2529         cqp_info->in.u.manage_qhash_table_entry.scratch = (uintptr_t)cqp_request;
2530         cqp_info->cqp_cmd = IRDMA_OP_MANAGE_QHASH_TABLE_ENTRY;
2531         cqp_info->post_sq = 1;
2532         status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2533         if (status && cm_node && !wait)
2534                 irdma_rem_ref_cm_node(cm_node);
2535
2536         irdma_put_cqp_request(iwcqp, cqp_request);
2537
2538         return status;
2539 }
2540
2541 /**
2542  * irdma_hw_flush_wqes_callback - Check return code after flush
2543  * @cqp_request: qhash cqp completion
2544  */
2545 static void irdma_hw_flush_wqes_callback(struct irdma_cqp_request *cqp_request)
2546 {
2547         struct irdma_qp_flush_info *hw_info;
2548         struct irdma_sc_qp *qp;
2549         struct irdma_qp *iwqp;
2550         struct cqp_cmds_info *cqp_info;
2551
2552         cqp_info = &cqp_request->info;
2553         hw_info = &cqp_info->in.u.qp_flush_wqes.info;
2554         qp = cqp_info->in.u.qp_flush_wqes.qp;
2555         iwqp = qp->qp_uk.back_qp;
2556
2557         if (cqp_request->compl_info.maj_err_code)
2558                 return;
2559
2560         if (hw_info->rq &&
2561             (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_SQ_WQE_FLUSHED ||
2562              cqp_request->compl_info.min_err_code == 0)) {
2563                 /* RQ WQE flush was requested but did not happen */
2564                 qp->qp_uk.rq_flush_complete = true;
2565         }
2566         if (hw_info->sq &&
2567             (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_RQ_WQE_FLUSHED ||
2568              cqp_request->compl_info.min_err_code == 0)) {
2569                 if (IRDMA_RING_MORE_WORK(qp->qp_uk.sq_ring)) {
2570                         ibdev_err(&iwqp->iwdev->ibdev, "Flush QP[%d] failed, SQ has more work",
2571                                   qp->qp_uk.qp_id);
2572                         irdma_ib_qp_event(iwqp, IRDMA_QP_EVENT_CATASTROPHIC);
2573                 }
2574                 qp->qp_uk.sq_flush_complete = true;
2575         }
2576 }
2577
2578 /**
2579  * irdma_hw_flush_wqes - flush qp's wqe
2580  * @rf: RDMA PCI function
2581  * @qp: hardware control qp
2582  * @info: info for flush
2583  * @wait: flag wait for completion
2584  */
2585 int irdma_hw_flush_wqes(struct irdma_pci_f *rf, struct irdma_sc_qp *qp,
2586                         struct irdma_qp_flush_info *info, bool wait)
2587 {
2588         int status;
2589         struct irdma_qp_flush_info *hw_info;
2590         struct irdma_cqp_request *cqp_request;
2591         struct cqp_cmds_info *cqp_info;
2592         struct irdma_qp *iwqp = qp->qp_uk.back_qp;
2593
2594         cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, wait);
2595         if (!cqp_request)
2596                 return -ENOMEM;
2597
2598         cqp_info = &cqp_request->info;
2599         if (!wait)
2600                 cqp_request->callback_fcn = irdma_hw_flush_wqes_callback;
2601         hw_info = &cqp_request->info.in.u.qp_flush_wqes.info;
2602         memcpy(hw_info, info, sizeof(*hw_info));
2603         cqp_info->cqp_cmd = IRDMA_OP_QP_FLUSH_WQES;
2604         cqp_info->post_sq = 1;
2605         cqp_info->in.u.qp_flush_wqes.qp = qp;
2606         cqp_info->in.u.qp_flush_wqes.scratch = (uintptr_t)cqp_request;
2607         status = irdma_handle_cqp_op(rf, cqp_request);
2608         if (status) {
2609                 qp->qp_uk.sq_flush_complete = true;
2610                 qp->qp_uk.rq_flush_complete = true;
2611                 irdma_put_cqp_request(&rf->cqp, cqp_request);
2612                 return status;
2613         }
2614
2615         if (!wait || cqp_request->compl_info.maj_err_code)
2616                 goto put_cqp;
2617
2618         if (info->rq) {
2619                 if (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_SQ_WQE_FLUSHED ||
2620                     cqp_request->compl_info.min_err_code == 0) {
2621                         /* RQ WQE flush was requested but did not happen */
2622                         qp->qp_uk.rq_flush_complete = true;
2623                 }
2624         }
2625         if (info->sq) {
2626                 if (cqp_request->compl_info.min_err_code == IRDMA_CQP_COMPL_RQ_WQE_FLUSHED ||
2627                     cqp_request->compl_info.min_err_code == 0) {
2628                         /*
2629                          * Handling case where WQE is posted to empty SQ when
2630                          * flush has not completed
2631                          */
2632                         if (IRDMA_RING_MORE_WORK(qp->qp_uk.sq_ring)) {
2633                                 struct irdma_cqp_request *new_req;
2634
2635                                 if (!qp->qp_uk.sq_flush_complete)
2636                                         goto put_cqp;
2637                                 qp->qp_uk.sq_flush_complete = false;
2638                                 qp->flush_sq = false;
2639
2640                                 info->rq = false;
2641                                 info->sq = true;
2642                                 new_req = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2643                                 if (!new_req) {
2644                                         status = -ENOMEM;
2645                                         goto put_cqp;
2646                                 }
2647                                 cqp_info = &new_req->info;
2648                                 hw_info = &new_req->info.in.u.qp_flush_wqes.info;
2649                                 memcpy(hw_info, info, sizeof(*hw_info));
2650                                 cqp_info->cqp_cmd = IRDMA_OP_QP_FLUSH_WQES;
2651                                 cqp_info->post_sq = 1;
2652                                 cqp_info->in.u.qp_flush_wqes.qp = qp;
2653                                 cqp_info->in.u.qp_flush_wqes.scratch = (uintptr_t)new_req;
2654
2655                                 status = irdma_handle_cqp_op(rf, new_req);
2656                                 if (new_req->compl_info.maj_err_code ||
2657                                     new_req->compl_info.min_err_code != IRDMA_CQP_COMPL_SQ_WQE_FLUSHED ||
2658                                     status) {
2659                                         ibdev_err(&iwqp->iwdev->ibdev, "fatal QP event: SQ in error but not flushed, qp: %d",
2660                                                   iwqp->ibqp.qp_num);
2661                                         qp->qp_uk.sq_flush_complete = false;
2662                                         irdma_ib_qp_event(iwqp, IRDMA_QP_EVENT_CATASTROPHIC);
2663                                 }
2664                                 irdma_put_cqp_request(&rf->cqp, new_req);
2665                         } else {
2666                                 /* SQ WQE flush was requested but did not happen */
2667                                 qp->qp_uk.sq_flush_complete = true;
2668                         }
2669                 } else {
2670                         if (!IRDMA_RING_MORE_WORK(qp->qp_uk.sq_ring))
2671                                 qp->qp_uk.sq_flush_complete = true;
2672                 }
2673         }
2674
2675         ibdev_dbg(&rf->iwdev->ibdev,
2676                   "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",
2677                   iwqp->ibqp.qp_num, rf->protocol_used, iwqp->iwarp_state,
2678                   iwqp->ibqp_state, iwqp->last_aeq, iwqp->hw_iwarp_state,
2679                   cqp_request->compl_info.maj_err_code,
2680                   cqp_request->compl_info.min_err_code);
2681 put_cqp:
2682         irdma_put_cqp_request(&rf->cqp, cqp_request);
2683
2684         return status;
2685 }
2686
2687 /**
2688  * irdma_gen_ae - generate AE
2689  * @rf: RDMA PCI function
2690  * @qp: qp associated with AE
2691  * @info: info for ae
2692  * @wait: wait for completion
2693  */
2694 void irdma_gen_ae(struct irdma_pci_f *rf, struct irdma_sc_qp *qp,
2695                   struct irdma_gen_ae_info *info, bool wait)
2696 {
2697         struct irdma_gen_ae_info *ae_info;
2698         struct irdma_cqp_request *cqp_request;
2699         struct cqp_cmds_info *cqp_info;
2700
2701         cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, wait);
2702         if (!cqp_request)
2703                 return;
2704
2705         cqp_info = &cqp_request->info;
2706         ae_info = &cqp_request->info.in.u.gen_ae.info;
2707         memcpy(ae_info, info, sizeof(*ae_info));
2708         cqp_info->cqp_cmd = IRDMA_OP_GEN_AE;
2709         cqp_info->post_sq = 1;
2710         cqp_info->in.u.gen_ae.qp = qp;
2711         cqp_info->in.u.gen_ae.scratch = (uintptr_t)cqp_request;
2712
2713         irdma_handle_cqp_op(rf, cqp_request);
2714         irdma_put_cqp_request(&rf->cqp, cqp_request);
2715 }
2716
2717 void irdma_flush_wqes(struct irdma_qp *iwqp, u32 flush_mask)
2718 {
2719         struct irdma_qp_flush_info info = {};
2720         struct irdma_pci_f *rf = iwqp->iwdev->rf;
2721         u8 flush_code = iwqp->sc_qp.flush_code;
2722
2723         if (!(flush_mask & IRDMA_FLUSH_SQ) && !(flush_mask & IRDMA_FLUSH_RQ))
2724                 return;
2725
2726         /* Set flush info fields*/
2727         info.sq = flush_mask & IRDMA_FLUSH_SQ;
2728         info.rq = flush_mask & IRDMA_FLUSH_RQ;
2729
2730         /* Generate userflush errors in CQE */
2731         info.sq_major_code = IRDMA_FLUSH_MAJOR_ERR;
2732         info.sq_minor_code = FLUSH_GENERAL_ERR;
2733         info.rq_major_code = IRDMA_FLUSH_MAJOR_ERR;
2734         info.rq_minor_code = FLUSH_GENERAL_ERR;
2735         info.userflushcode = true;
2736
2737         if (flush_mask & IRDMA_REFLUSH) {
2738                 if (info.sq)
2739                         iwqp->sc_qp.flush_sq = false;
2740                 if (info.rq)
2741                         iwqp->sc_qp.flush_rq = false;
2742         } else {
2743                 if (flush_code) {
2744                         if (info.sq && iwqp->sc_qp.sq_flush_code)
2745                                 info.sq_minor_code = flush_code;
2746                         if (info.rq && iwqp->sc_qp.rq_flush_code)
2747                                 info.rq_minor_code = flush_code;
2748                 }
2749                 if (!iwqp->user_mode)
2750                         queue_delayed_work(iwqp->iwdev->cleanup_wq,
2751                                            &iwqp->dwork_flush,
2752                                            msecs_to_jiffies(IRDMA_FLUSH_DELAY_MS));
2753         }
2754
2755         /* Issue flush */
2756         (void)irdma_hw_flush_wqes(rf, &iwqp->sc_qp, &info,
2757                                   flush_mask & IRDMA_FLUSH_WAIT);
2758         iwqp->flush_issued = true;
2759 }