Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
[linux-2.6-microblaze.git] / drivers / scsi / lpfc / lpfc_nvme.c
1 /*******************************************************************
2  * This file is part of the Emulex Linux Device Driver for         *
3  * Fibre Channel Host Bus Adapters.                                *
4  * Copyright (C) 2017-2021 Broadcom. All Rights Reserved. The term *
5  * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.  *
6  * Copyright (C) 2004-2016 Emulex.  All rights reserved.           *
7  * EMULEX and SLI are trademarks of Emulex.                        *
8  * www.broadcom.com                                                *
9  * Portions Copyright (C) 2004-2005 Christoph Hellwig              *
10  *                                                                 *
11  * This program is free software; you can redistribute it and/or   *
12  * modify it under the terms of version 2 of the GNU General       *
13  * Public License as published by the Free Software Foundation.    *
14  * This program is distributed in the hope that it will be useful. *
15  * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
16  * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
17  * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
18  * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
19  * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
20  * more details, a copy of which can be found in the file COPYING  *
21  * included with this package.                                     *
22  ********************************************************************/
23 #include <linux/pci.h>
24 #include <linux/slab.h>
25 #include <linux/interrupt.h>
26 #include <linux/delay.h>
27 #include <asm/unaligned.h>
28 #include <linux/crc-t10dif.h>
29 #include <net/checksum.h>
30
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_tcq.h>
36 #include <scsi/scsi_transport_fc.h>
37 #include <scsi/fc/fc_fs.h>
38
39 #include "lpfc_version.h"
40 #include "lpfc_hw4.h"
41 #include "lpfc_hw.h"
42 #include "lpfc_sli.h"
43 #include "lpfc_sli4.h"
44 #include "lpfc_nl.h"
45 #include "lpfc_disc.h"
46 #include "lpfc.h"
47 #include "lpfc_nvme.h"
48 #include "lpfc_scsi.h"
49 #include "lpfc_logmsg.h"
50 #include "lpfc_crtn.h"
51 #include "lpfc_vport.h"
52 #include "lpfc_debugfs.h"
53
54 /* NVME initiator-based functions */
55
56 static struct lpfc_io_buf *
57 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
58                   int idx, int expedite);
59
60 static void
61 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_io_buf *);
62
63 static struct nvme_fc_port_template lpfc_nvme_template;
64
65 /**
66  * lpfc_nvme_create_queue -
67  * @pnvme_lport: Transport localport that LS is to be issued from
68  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
69  * @qsize: Size of the queue in bytes
70  * @handle: An opaque driver handle used in follow-up calls.
71  *
72  * Driver registers this routine to preallocate and initialize any
73  * internal data structures to bind the @qidx to its internal IO queues.
74  * A hardware queue maps (qidx) to a specific driver MSI-X vector/EQ/CQ/WQ.
75  *
76  * Return value :
77  *   0 - Success
78  *   -EINVAL - Unsupported input value.
79  *   -ENOMEM - Could not alloc necessary memory
80  **/
81 static int
82 lpfc_nvme_create_queue(struct nvme_fc_local_port *pnvme_lport,
83                        unsigned int qidx, u16 qsize,
84                        void **handle)
85 {
86         struct lpfc_nvme_lport *lport;
87         struct lpfc_vport *vport;
88         struct lpfc_nvme_qhandle *qhandle;
89         char *str;
90
91         if (!pnvme_lport->private)
92                 return -ENOMEM;
93
94         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
95         vport = lport->vport;
96         qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL);
97         if (qhandle == NULL)
98                 return -ENOMEM;
99
100         qhandle->cpu_id = raw_smp_processor_id();
101         qhandle->qidx = qidx;
102         /*
103          * NVME qidx == 0 is the admin queue, so both admin queue
104          * and first IO queue will use MSI-X vector and associated
105          * EQ/CQ/WQ at index 0. After that they are sequentially assigned.
106          */
107         if (qidx) {
108                 str = "IO ";  /* IO queue */
109                 qhandle->index = ((qidx - 1) %
110                         lpfc_nvme_template.max_hw_queues);
111         } else {
112                 str = "ADM";  /* Admin queue */
113                 qhandle->index = qidx;
114         }
115
116         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
117                          "6073 Binding %s HdwQueue %d  (cpu %d) to "
118                          "hdw_queue %d qhandle x%px\n", str,
119                          qidx, qhandle->cpu_id, qhandle->index, qhandle);
120         *handle = (void *)qhandle;
121         return 0;
122 }
123
124 /**
125  * lpfc_nvme_delete_queue -
126  * @pnvme_lport: Transport localport that LS is to be issued from
127  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
128  * @handle: An opaque driver handle from lpfc_nvme_create_queue
129  *
130  * Driver registers this routine to free
131  * any internal data structures to bind the @qidx to its internal
132  * IO queues.
133  *
134  * Return value :
135  *   0 - Success
136  *   TODO:  What are the failure codes.
137  **/
138 static void
139 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport,
140                        unsigned int qidx,
141                        void *handle)
142 {
143         struct lpfc_nvme_lport *lport;
144         struct lpfc_vport *vport;
145
146         if (!pnvme_lport->private)
147                 return;
148
149         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
150         vport = lport->vport;
151
152         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
153                         "6001 ENTER.  lpfc_pnvme x%px, qidx x%x qhandle x%px\n",
154                         lport, qidx, handle);
155         kfree(handle);
156 }
157
158 static void
159 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
160 {
161         struct lpfc_nvme_lport *lport = localport->private;
162
163         lpfc_printf_vlog(lport->vport, KERN_INFO, LOG_NVME,
164                          "6173 localport x%px delete complete\n",
165                          lport);
166
167         /* release any threads waiting for the unreg to complete */
168         if (lport->vport->localport)
169                 complete(lport->lport_unreg_cmp);
170 }
171
172 /* lpfc_nvme_remoteport_delete
173  *
174  * @remoteport: Pointer to an nvme transport remoteport instance.
175  *
176  * This is a template downcall.  NVME transport calls this function
177  * when it has completed the unregistration of a previously
178  * registered remoteport.
179  *
180  * Return value :
181  * None
182  */
183 static void
184 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
185 {
186         struct lpfc_nvme_rport *rport = remoteport->private;
187         struct lpfc_vport *vport;
188         struct lpfc_nodelist *ndlp;
189         u32 fc4_xpt_flags;
190
191         ndlp = rport->ndlp;
192         if (!ndlp) {
193                 pr_err("**** %s: NULL ndlp on rport x%px remoteport x%px\n",
194                        __func__, rport, remoteport);
195                 goto rport_err;
196         }
197
198         vport = ndlp->vport;
199         if (!vport) {
200                 pr_err("**** %s: Null vport on ndlp x%px, ste x%x rport x%px\n",
201                        __func__, ndlp, ndlp->nlp_state, rport);
202                 goto rport_err;
203         }
204
205         fc4_xpt_flags = NVME_XPT_REGD | SCSI_XPT_REGD;
206
207         /* Remove this rport from the lport's list - memory is owned by the
208          * transport. Remove the ndlp reference for the NVME transport before
209          * calling state machine to remove the node.
210          */
211         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
212                         "6146 remoteport delete of remoteport x%px\n",
213                         remoteport);
214         spin_lock_irq(&ndlp->lock);
215
216         /* The register rebind might have occurred before the delete
217          * downcall.  Guard against this race.
218          */
219         if (ndlp->fc4_xpt_flags & NLP_WAIT_FOR_UNREG)
220                 ndlp->fc4_xpt_flags &= ~(NLP_WAIT_FOR_UNREG | NVME_XPT_REGD);
221
222         spin_unlock_irq(&ndlp->lock);
223
224         /* On a devloss timeout event, one more put is executed provided the
225          * NVME and SCSI rport unregister requests are complete.  If the vport
226          * is unloading, this extra put is executed by lpfc_drop_node.
227          */
228         if (!(ndlp->fc4_xpt_flags & fc4_xpt_flags))
229                 lpfc_disc_state_machine(vport, ndlp, NULL, NLP_EVT_DEVICE_RM);
230
231  rport_err:
232         return;
233 }
234
235 /**
236  * lpfc_nvme_handle_lsreq - Process an unsolicited NVME LS request
237  * @phba: pointer to lpfc hba data structure.
238  * @axchg: pointer to exchange context for the NVME LS request
239  *
240  * This routine is used for processing an asychronously received NVME LS
241  * request. Any remaining validation is done and the LS is then forwarded
242  * to the nvme-fc transport via nvme_fc_rcv_ls_req().
243  *
244  * The calling sequence should be: nvme_fc_rcv_ls_req() -> (processing)
245  * -> lpfc_nvme_xmt_ls_rsp/cmp -> req->done.
246  * __lpfc_nvme_xmt_ls_rsp_cmp should free the allocated axchg.
247  *
248  * Returns 0 if LS was handled and delivered to the transport
249  * Returns 1 if LS failed to be handled and should be dropped
250  */
251 int
252 lpfc_nvme_handle_lsreq(struct lpfc_hba *phba,
253                         struct lpfc_async_xchg_ctx *axchg)
254 {
255 #if (IS_ENABLED(CONFIG_NVME_FC))
256         struct lpfc_vport *vport;
257         struct lpfc_nvme_rport *lpfc_rport;
258         struct nvme_fc_remote_port *remoteport;
259         struct lpfc_nvme_lport *lport;
260         uint32_t *payload = axchg->payload;
261         int rc;
262
263         vport = axchg->ndlp->vport;
264         lpfc_rport = axchg->ndlp->nrport;
265         if (!lpfc_rport)
266                 return -EINVAL;
267
268         remoteport = lpfc_rport->remoteport;
269         if (!vport->localport)
270                 return -EINVAL;
271
272         lport = vport->localport->private;
273         if (!lport)
274                 return -EINVAL;
275
276         rc = nvme_fc_rcv_ls_req(remoteport, &axchg->ls_rsp, axchg->payload,
277                                 axchg->size);
278
279         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_DISC,
280                         "6205 NVME Unsol rcv: sz %d rc %d: %08x %08x %08x "
281                         "%08x %08x %08x\n",
282                         axchg->size, rc,
283                         *payload, *(payload+1), *(payload+2),
284                         *(payload+3), *(payload+4), *(payload+5));
285
286         if (!rc)
287                 return 0;
288 #endif
289         return 1;
290 }
291
292 /**
293  * __lpfc_nvme_ls_req_cmp - Generic completion handler for a NVME
294  *        LS request.
295  * @phba: Pointer to HBA context object
296  * @vport: The local port that issued the LS
297  * @cmdwqe: Pointer to driver command WQE object.
298  * @wcqe: Pointer to driver response CQE object.
299  *
300  * This function is the generic completion handler for NVME LS requests.
301  * The function updates any states and statistics, calls the transport
302  * ls_req done() routine, then tears down the command and buffers used
303  * for the LS request.
304  **/
305 void
306 __lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba,  struct lpfc_vport *vport,
307                         struct lpfc_iocbq *cmdwqe,
308                         struct lpfc_wcqe_complete *wcqe)
309 {
310         struct nvmefc_ls_req *pnvme_lsreq;
311         struct lpfc_dmabuf *buf_ptr;
312         struct lpfc_nodelist *ndlp;
313         uint32_t status;
314
315         pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2;
316         ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
317         status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
318
319         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
320                          "6047 NVMEx LS REQ x%px cmpl DID %x Xri: %x "
321                          "status %x reason x%x cmd:x%px lsreg:x%px bmp:x%px "
322                          "ndlp:x%px\n",
323                          pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
324                          cmdwqe->sli4_xritag, status,
325                          (wcqe->parameter & 0xffff),
326                          cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
327
328         lpfc_nvmeio_data(phba, "NVMEx LS CMPL: xri x%x stat x%x parm x%x\n",
329                          cmdwqe->sli4_xritag, status, wcqe->parameter);
330
331         if (cmdwqe->context3) {
332                 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
333                 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
334                 kfree(buf_ptr);
335                 cmdwqe->context3 = NULL;
336         }
337         if (pnvme_lsreq->done)
338                 pnvme_lsreq->done(pnvme_lsreq, status);
339         else
340                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
341                                  "6046 NVMEx cmpl without done call back? "
342                                  "Data x%px DID %x Xri: %x status %x\n",
343                                 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
344                                 cmdwqe->sli4_xritag, status);
345         if (ndlp) {
346                 lpfc_nlp_put(ndlp);
347                 cmdwqe->context1 = NULL;
348         }
349         lpfc_sli_release_iocbq(phba, cmdwqe);
350 }
351
352 static void
353 lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
354                        struct lpfc_wcqe_complete *wcqe)
355 {
356         struct lpfc_vport *vport = cmdwqe->vport;
357         struct lpfc_nvme_lport *lport;
358         uint32_t status;
359
360         status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
361
362         if (vport->localport) {
363                 lport = (struct lpfc_nvme_lport *)vport->localport->private;
364                 if (lport) {
365                         atomic_inc(&lport->fc4NvmeLsCmpls);
366                         if (status) {
367                                 if (bf_get(lpfc_wcqe_c_xb, wcqe))
368                                         atomic_inc(&lport->cmpl_ls_xb);
369                                 atomic_inc(&lport->cmpl_ls_err);
370                         }
371                 }
372         }
373
374         __lpfc_nvme_ls_req_cmp(phba, vport, cmdwqe, wcqe);
375 }
376
377 static int
378 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
379                   struct lpfc_dmabuf *inp,
380                   struct nvmefc_ls_req *pnvme_lsreq,
381                   void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
382                                struct lpfc_wcqe_complete *),
383                   struct lpfc_nodelist *ndlp, uint32_t num_entry,
384                   uint32_t tmo, uint8_t retry)
385 {
386         struct lpfc_hba *phba = vport->phba;
387         union lpfc_wqe128 *wqe;
388         struct lpfc_iocbq *genwqe;
389         struct ulp_bde64 *bpl;
390         struct ulp_bde64 bde;
391         int i, rc, xmit_len, first_len;
392
393         /* Allocate buffer for  command WQE */
394         genwqe = lpfc_sli_get_iocbq(phba);
395         if (genwqe == NULL)
396                 return 1;
397
398         wqe = &genwqe->wqe;
399         /* Initialize only 64 bytes */
400         memset(wqe, 0, sizeof(union lpfc_wqe));
401
402         genwqe->context3 = (uint8_t *)bmp;
403         genwqe->iocb_flag |= LPFC_IO_NVME_LS;
404
405         /* Save for completion so we can release these resources */
406         genwqe->context1 = lpfc_nlp_get(ndlp);
407         if (!genwqe->context1) {
408                 dev_warn(&phba->pcidev->dev,
409                          "Warning: Failed node ref, not sending LS_REQ\n");
410                 lpfc_sli_release_iocbq(phba, genwqe);
411                 return 1;
412         }
413
414         genwqe->context2 = (uint8_t *)pnvme_lsreq;
415         /* Fill in payload, bp points to frame payload */
416
417         if (!tmo)
418                 /* FC spec states we need 3 * ratov for CT requests */
419                 tmo = (3 * phba->fc_ratov);
420
421         /* For this command calculate the xmit length of the request bde. */
422         xmit_len = 0;
423         first_len = 0;
424         bpl = (struct ulp_bde64 *)bmp->virt;
425         for (i = 0; i < num_entry; i++) {
426                 bde.tus.w = bpl[i].tus.w;
427                 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
428                         break;
429                 xmit_len += bde.tus.f.bdeSize;
430                 if (i == 0)
431                         first_len = xmit_len;
432         }
433
434         genwqe->rsvd2 = num_entry;
435         genwqe->hba_wqidx = 0;
436
437         /* Words 0 - 2 */
438         wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
439         wqe->generic.bde.tus.f.bdeSize = first_len;
440         wqe->generic.bde.addrLow = bpl[0].addrLow;
441         wqe->generic.bde.addrHigh = bpl[0].addrHigh;
442
443         /* Word 3 */
444         wqe->gen_req.request_payload_len = first_len;
445
446         /* Word 4 */
447
448         /* Word 5 */
449         bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
450         bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
451         bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
452         bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
453         bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
454
455         /* Word 6 */
456         bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
457                phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
458         bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
459
460         /* Word 7 */
461         bf_set(wqe_tmo, &wqe->gen_req.wqe_com, tmo);
462         bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
463         bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
464         bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
465
466         /* Word 8 */
467         wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
468
469         /* Word 9 */
470         bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
471
472         /* Word 10 */
473         bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
474         bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
475         bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
476         bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
477         bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
478
479         /* Word 11 */
480         bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
481         bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
482
483
484         /* Issue GEN REQ WQE for NPORT <did> */
485         genwqe->wqe_cmpl = cmpl;
486         genwqe->iocb_cmpl = NULL;
487         genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
488         genwqe->vport = vport;
489         genwqe->retry = retry;
490
491         lpfc_nvmeio_data(phba, "NVME LS  XMIT: xri x%x iotag x%x to x%06x\n",
492                          genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
493
494         rc = lpfc_sli4_issue_wqe(phba, &phba->sli4_hba.hdwq[0], genwqe);
495         if (rc) {
496                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
497                                  "6045 Issue GEN REQ WQE to NPORT x%x "
498                                  "Data: x%x x%x  rc x%x\n",
499                                  ndlp->nlp_DID, genwqe->iotag,
500                                  vport->port_state, rc);
501                 lpfc_nlp_put(ndlp);
502                 lpfc_sli_release_iocbq(phba, genwqe);
503                 return 1;
504         }
505
506         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_ELS,
507                          "6050 Issue GEN REQ WQE to NPORT x%x "
508                          "Data: oxid: x%x state: x%x wq:x%px lsreq:x%px "
509                          "bmp:x%px xmit:%d 1st:%d\n",
510                          ndlp->nlp_DID, genwqe->sli4_xritag,
511                          vport->port_state,
512                          genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
513         return 0;
514 }
515
516
517 /**
518  * __lpfc_nvme_ls_req - Generic service routine to issue an NVME LS request
519  * @vport: The local port issuing the LS
520  * @ndlp: The remote port to send the LS to
521  * @pnvme_lsreq: Pointer to LS request structure from the transport
522  * @gen_req_cmp: Completion call-back
523  *
524  * Routine validates the ndlp, builds buffers and sends a GEN_REQUEST
525  * WQE to perform the LS operation.
526  *
527  * Return value :
528  *   0 - Success
529  *   non-zero: various error codes, in form of -Exxx
530  **/
531 int
532 __lpfc_nvme_ls_req(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
533                       struct nvmefc_ls_req *pnvme_lsreq,
534                       void (*gen_req_cmp)(struct lpfc_hba *phba,
535                                 struct lpfc_iocbq *cmdwqe,
536                                 struct lpfc_wcqe_complete *wcqe))
537 {
538         struct lpfc_dmabuf *bmp;
539         struct ulp_bde64 *bpl;
540         int ret;
541         uint16_t ntype, nstate;
542
543         if (!ndlp) {
544                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
545                                  "6051 NVMEx LS REQ: Bad NDLP x%px, Failing "
546                                  "LS Req\n",
547                                  ndlp);
548                 return -ENODEV;
549         }
550
551         ntype = ndlp->nlp_type;
552         nstate = ndlp->nlp_state;
553         if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
554             (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
555                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
556                                  "6088 NVMEx LS REQ: Fail DID x%06x not "
557                                  "ready for IO. Type x%x, State x%x\n",
558                                  ndlp->nlp_DID, ntype, nstate);
559                 return -ENODEV;
560         }
561
562         if (!vport->phba->sli4_hba.nvmels_wq)
563                 return -ENOMEM;
564
565         /*
566          * there are two dma buf in the request, actually there is one and
567          * the second one is just the start address + cmd size.
568          * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
569          * in a lpfc_dmabuf struct. When freeing we just free the wrapper
570          * because the nvem layer owns the data bufs.
571          * We do not have to break these packets open, we don't care what is
572          * in them. And we do not have to look at the resonse data, we only
573          * care that we got a response. All of the caring is going to happen
574          * in the nvme-fc layer.
575          */
576
577         bmp = kmalloc(sizeof(*bmp), GFP_KERNEL);
578         if (!bmp) {
579                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
580                                  "6044 NVMEx LS REQ: Could not alloc LS buf "
581                                  "for DID %x\n",
582                                  ndlp->nlp_DID);
583                 return -ENOMEM;
584         }
585
586         bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
587         if (!bmp->virt) {
588                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
589                                  "6042 NVMEx LS REQ: Could not alloc mbuf "
590                                  "for DID %x\n",
591                                  ndlp->nlp_DID);
592                 kfree(bmp);
593                 return -ENOMEM;
594         }
595
596         INIT_LIST_HEAD(&bmp->list);
597
598         bpl = (struct ulp_bde64 *)bmp->virt;
599         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
600         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
601         bpl->tus.f.bdeFlags = 0;
602         bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
603         bpl->tus.w = le32_to_cpu(bpl->tus.w);
604         bpl++;
605
606         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
607         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
608         bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
609         bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
610         bpl->tus.w = le32_to_cpu(bpl->tus.w);
611
612         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
613                         "6149 NVMEx LS REQ: Issue to DID 0x%06x lsreq x%px, "
614                         "rqstlen:%d rsplen:%d %pad %pad\n",
615                         ndlp->nlp_DID, pnvme_lsreq, pnvme_lsreq->rqstlen,
616                         pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
617                         &pnvme_lsreq->rspdma);
618
619         ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
620                                 pnvme_lsreq, gen_req_cmp, ndlp, 2,
621                                 pnvme_lsreq->timeout, 0);
622         if (ret != WQE_SUCCESS) {
623                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
624                                  "6052 NVMEx REQ: EXIT. issue ls wqe failed "
625                                  "lsreq x%px Status %x DID %x\n",
626                                  pnvme_lsreq, ret, ndlp->nlp_DID);
627                 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
628                 kfree(bmp);
629                 return -EIO;
630         }
631
632         return 0;
633 }
634
635 /**
636  * lpfc_nvme_ls_req - Issue an NVME Link Service request
637  * @pnvme_lport: Transport localport that LS is to be issued from.
638  * @pnvme_rport: Transport remoteport that LS is to be sent to.
639  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
640  *
641  * Driver registers this routine to handle any link service request
642  * from the nvme_fc transport to a remote nvme-aware port.
643  *
644  * Return value :
645  *   0 - Success
646  *   non-zero: various error codes, in form of -Exxx
647  **/
648 static int
649 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
650                  struct nvme_fc_remote_port *pnvme_rport,
651                  struct nvmefc_ls_req *pnvme_lsreq)
652 {
653         struct lpfc_nvme_lport *lport;
654         struct lpfc_nvme_rport *rport;
655         struct lpfc_vport *vport;
656         int ret;
657
658         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
659         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
660         if (unlikely(!lport) || unlikely(!rport))
661                 return -EINVAL;
662
663         vport = lport->vport;
664         if (vport->load_flag & FC_UNLOADING)
665                 return -ENODEV;
666
667         atomic_inc(&lport->fc4NvmeLsRequests);
668
669         ret = __lpfc_nvme_ls_req(vport, rport->ndlp, pnvme_lsreq,
670                                  lpfc_nvme_ls_req_cmp);
671         if (ret)
672                 atomic_inc(&lport->xmt_ls_err);
673
674         return ret;
675 }
676
677 /**
678  * __lpfc_nvme_ls_abort - Generic service routine to abort a prior
679  *         NVME LS request
680  * @vport: The local port that issued the LS
681  * @ndlp: The remote port the LS was sent to
682  * @pnvme_lsreq: Pointer to LS request structure from the transport
683  *
684  * The driver validates the ndlp, looks for the LS, and aborts the
685  * LS if found.
686  *
687  * Returns:
688  * 0 : if LS found and aborted
689  * non-zero: various error conditions in form -Exxx
690  **/
691 int
692 __lpfc_nvme_ls_abort(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
693                         struct nvmefc_ls_req *pnvme_lsreq)
694 {
695         struct lpfc_hba *phba = vport->phba;
696         struct lpfc_sli_ring *pring;
697         struct lpfc_iocbq *wqe, *next_wqe;
698         bool foundit = false;
699
700         if (!ndlp) {
701                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
702                                 "6049 NVMEx LS REQ Abort: Bad NDLP x%px DID "
703                                 "x%06x, Failing LS Req\n",
704                                 ndlp, ndlp ? ndlp->nlp_DID : 0);
705                 return -EINVAL;
706         }
707
708         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
709                          "6040 NVMEx LS REQ Abort: Issue LS_ABORT for lsreq "
710                          "x%px rqstlen:%d rsplen:%d %pad %pad\n",
711                          pnvme_lsreq, pnvme_lsreq->rqstlen,
712                          pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
713                          &pnvme_lsreq->rspdma);
714
715         /*
716          * Lock the ELS ring txcmplq and look for the wqe that matches
717          * this ELS. If found, issue an abort on the wqe.
718          */
719         pring = phba->sli4_hba.nvmels_wq->pring;
720         spin_lock_irq(&phba->hbalock);
721         spin_lock(&pring->ring_lock);
722         list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
723                 if (wqe->context2 == pnvme_lsreq) {
724                         wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
725                         foundit = true;
726                         break;
727                 }
728         }
729         spin_unlock(&pring->ring_lock);
730
731         if (foundit)
732                 lpfc_sli_issue_abort_iotag(phba, pring, wqe, NULL);
733         spin_unlock_irq(&phba->hbalock);
734
735         if (foundit)
736                 return 0;
737
738         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
739                          "6213 NVMEx LS REQ Abort: Unable to locate req x%px\n",
740                          pnvme_lsreq);
741         return -EINVAL;
742 }
743
744 static int
745 lpfc_nvme_xmt_ls_rsp(struct nvme_fc_local_port *localport,
746                      struct nvme_fc_remote_port *remoteport,
747                      struct nvmefc_ls_rsp *ls_rsp)
748 {
749         struct lpfc_async_xchg_ctx *axchg =
750                 container_of(ls_rsp, struct lpfc_async_xchg_ctx, ls_rsp);
751         struct lpfc_nvme_lport *lport;
752         int rc;
753
754         if (axchg->phba->pport->load_flag & FC_UNLOADING)
755                 return -ENODEV;
756
757         lport = (struct lpfc_nvme_lport *)localport->private;
758
759         rc = __lpfc_nvme_xmt_ls_rsp(axchg, ls_rsp, __lpfc_nvme_xmt_ls_rsp_cmp);
760
761         if (rc) {
762                 /*
763                  * unless the failure is due to having already sent
764                  * the response, an abort will be generated for the
765                  * exchange if the rsp can't be sent.
766                  */
767                 if (rc != -EALREADY)
768                         atomic_inc(&lport->xmt_ls_abort);
769                 return rc;
770         }
771
772         return 0;
773 }
774
775 /**
776  * lpfc_nvme_ls_abort - Abort a prior NVME LS request
777  * @pnvme_lport: Transport localport that LS is to be issued from.
778  * @pnvme_rport: Transport remoteport that LS is to be sent to.
779  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
780  *
781  * Driver registers this routine to abort a NVME LS request that is
782  * in progress (from the transports perspective).
783  **/
784 static void
785 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
786                    struct nvme_fc_remote_port *pnvme_rport,
787                    struct nvmefc_ls_req *pnvme_lsreq)
788 {
789         struct lpfc_nvme_lport *lport;
790         struct lpfc_vport *vport;
791         struct lpfc_nodelist *ndlp;
792         int ret;
793
794         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
795         if (unlikely(!lport))
796                 return;
797         vport = lport->vport;
798
799         if (vport->load_flag & FC_UNLOADING)
800                 return;
801
802         ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
803
804         ret = __lpfc_nvme_ls_abort(vport, ndlp, pnvme_lsreq);
805         if (!ret)
806                 atomic_inc(&lport->xmt_ls_abort);
807 }
808
809 /* Fix up the existing sgls for NVME IO. */
810 static inline void
811 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
812                        struct lpfc_io_buf *lpfc_ncmd,
813                        struct nvmefc_fcp_req *nCmd)
814 {
815         struct lpfc_hba  *phba = vport->phba;
816         struct sli4_sge *sgl;
817         union lpfc_wqe128 *wqe;
818         uint32_t *wptr, *dptr;
819
820         /*
821          * Get a local pointer to the built-in wqe and correct
822          * the cmd size to match NVME's 96 bytes and fix
823          * the dma address.
824          */
825
826         wqe = &lpfc_ncmd->cur_iocbq.wqe;
827
828         /*
829          * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
830          * match NVME.  NVME sends 96 bytes. Also, use the
831          * nvme commands command and response dma addresses
832          * rather than the virtual memory to ease the restore
833          * operation.
834          */
835         sgl = lpfc_ncmd->dma_sgl;
836         sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
837         if (phba->cfg_nvme_embed_cmd) {
838                 sgl->addr_hi = 0;
839                 sgl->addr_lo = 0;
840
841                 /* Word 0-2 - NVME CMND IU (embedded payload) */
842                 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
843                 wqe->generic.bde.tus.f.bdeSize = 56;
844                 wqe->generic.bde.addrHigh = 0;
845                 wqe->generic.bde.addrLow =  64;  /* Word 16 */
846
847                 /* Word 10  - dbde is 0, wqes is 1 in template */
848
849                 /*
850                  * Embed the payload in the last half of the WQE
851                  * WQE words 16-30 get the NVME CMD IU payload
852                  *
853                  * WQE words 16-19 get payload Words 1-4
854                  * WQE words 20-21 get payload Words 6-7
855                  * WQE words 22-29 get payload Words 16-23
856                  */
857                 wptr = &wqe->words[16];  /* WQE ptr */
858                 dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
859                 dptr++;                 /* Skip Word 0 in payload */
860
861                 *wptr++ = *dptr++;      /* Word 1 */
862                 *wptr++ = *dptr++;      /* Word 2 */
863                 *wptr++ = *dptr++;      /* Word 3 */
864                 *wptr++ = *dptr++;      /* Word 4 */
865                 dptr++;                 /* Skip Word 5 in payload */
866                 *wptr++ = *dptr++;      /* Word 6 */
867                 *wptr++ = *dptr++;      /* Word 7 */
868                 dptr += 8;              /* Skip Words 8-15 in payload */
869                 *wptr++ = *dptr++;      /* Word 16 */
870                 *wptr++ = *dptr++;      /* Word 17 */
871                 *wptr++ = *dptr++;      /* Word 18 */
872                 *wptr++ = *dptr++;      /* Word 19 */
873                 *wptr++ = *dptr++;      /* Word 20 */
874                 *wptr++ = *dptr++;      /* Word 21 */
875                 *wptr++ = *dptr++;      /* Word 22 */
876                 *wptr   = *dptr;        /* Word 23 */
877         } else {
878                 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma));
879                 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma));
880
881                 /* Word 0-2 - NVME CMND IU Inline BDE */
882                 wqe->generic.bde.tus.f.bdeFlags =  BUFF_TYPE_BDE_64;
883                 wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen;
884                 wqe->generic.bde.addrHigh = sgl->addr_hi;
885                 wqe->generic.bde.addrLow =  sgl->addr_lo;
886
887                 /* Word 10 */
888                 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
889                 bf_set(wqe_wqes, &wqe->generic.wqe_com, 0);
890         }
891
892         sgl++;
893
894         /* Setup the physical region for the FCP RSP */
895         sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
896         sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
897         sgl->word2 = le32_to_cpu(sgl->word2);
898         if (nCmd->sg_cnt)
899                 bf_set(lpfc_sli4_sge_last, sgl, 0);
900         else
901                 bf_set(lpfc_sli4_sge_last, sgl, 1);
902         sgl->word2 = cpu_to_le32(sgl->word2);
903         sgl->sge_len = cpu_to_le32(nCmd->rsplen);
904 }
905
906
907 /*
908  * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO
909  *
910  * Driver registers this routine as it io request handler.  This
911  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
912  * data structure to the rport indicated in @lpfc_nvme_rport.
913  *
914  * Return value :
915  *   0 - Success
916  *   TODO: What are the failure codes.
917  **/
918 static void
919 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
920                           struct lpfc_wcqe_complete *wcqe)
921 {
922         struct lpfc_io_buf *lpfc_ncmd =
923                 (struct lpfc_io_buf *)pwqeIn->context1;
924         struct lpfc_vport *vport = pwqeIn->vport;
925         struct nvmefc_fcp_req *nCmd;
926         struct nvme_fc_ersp_iu *ep;
927         struct nvme_fc_cmd_iu *cp;
928         struct lpfc_nodelist *ndlp;
929         struct lpfc_nvme_fcpreq_priv *freqpriv;
930         struct lpfc_nvme_lport *lport;
931         uint32_t code, status, idx;
932         uint16_t cid, sqhd, data;
933         uint32_t *ptr;
934 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
935         int cpu;
936 #endif
937
938         /* Sanity check on return of outstanding command */
939         if (!lpfc_ncmd) {
940                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
941                                  "6071 Null lpfc_ncmd pointer. No "
942                                  "release, skip completion\n");
943                 return;
944         }
945
946         /* Guard against abort handler being called at same time */
947         spin_lock(&lpfc_ncmd->buf_lock);
948
949         if (!lpfc_ncmd->nvmeCmd) {
950                 spin_unlock(&lpfc_ncmd->buf_lock);
951                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
952                                  "6066 Missing cmpl ptrs: lpfc_ncmd x%px, "
953                                  "nvmeCmd x%px\n",
954                                  lpfc_ncmd, lpfc_ncmd->nvmeCmd);
955
956                 /* Release the lpfc_ncmd regardless of the missing elements. */
957                 lpfc_release_nvme_buf(phba, lpfc_ncmd);
958                 return;
959         }
960         nCmd = lpfc_ncmd->nvmeCmd;
961         status = bf_get(lpfc_wcqe_c_status, wcqe);
962
963         idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
964         phba->sli4_hba.hdwq[idx].nvme_cstat.io_cmpls++;
965
966         if (unlikely(status && vport->localport)) {
967                 lport = (struct lpfc_nvme_lport *)vport->localport->private;
968                 if (lport) {
969                         if (bf_get(lpfc_wcqe_c_xb, wcqe))
970                                 atomic_inc(&lport->cmpl_fcp_xb);
971                         atomic_inc(&lport->cmpl_fcp_err);
972                 }
973         }
974
975         lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
976                          lpfc_ncmd->cur_iocbq.sli4_xritag,
977                          status, wcqe->parameter);
978         /*
979          * Catch race where our node has transitioned, but the
980          * transport is still transitioning.
981          */
982         ndlp = lpfc_ncmd->ndlp;
983         if (!ndlp) {
984                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
985                                  "6062 Ignoring NVME cmpl.  No ndlp\n");
986                 goto out_err;
987         }
988
989         code = bf_get(lpfc_wcqe_c_code, wcqe);
990         if (code == CQE_CODE_NVME_ERSP) {
991                 /* For this type of CQE, we need to rebuild the rsp */
992                 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
993
994                 /*
995                  * Get Command Id from cmd to plug into response. This
996                  * code is not needed in the next NVME Transport drop.
997                  */
998                 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
999                 cid = cp->sqe.common.command_id;
1000
1001                 /*
1002                  * RSN is in CQE word 2
1003                  * SQHD is in CQE Word 3 bits 15:0
1004                  * Cmd Specific info is in CQE Word 1
1005                  * and in CQE Word 0 bits 15:0
1006                  */
1007                 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
1008
1009                 /* Now lets build the NVME ERSP IU */
1010                 ep->iu_len = cpu_to_be16(8);
1011                 ep->rsn = wcqe->parameter;
1012                 ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
1013                 ep->rsvd12 = 0;
1014                 ptr = (uint32_t *)&ep->cqe.result.u64;
1015                 *ptr++ = wcqe->total_data_placed;
1016                 data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
1017                 *ptr = (uint32_t)data;
1018                 ep->cqe.sq_head = sqhd;
1019                 ep->cqe.sq_id =  nCmd->sqid;
1020                 ep->cqe.command_id = cid;
1021                 ep->cqe.status = 0;
1022
1023                 lpfc_ncmd->status = IOSTAT_SUCCESS;
1024                 lpfc_ncmd->result = 0;
1025                 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
1026                 nCmd->transferred_length = nCmd->payload_length;
1027         } else {
1028                 lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK);
1029                 lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK);
1030
1031                 /* For NVME, the only failure path that results in an
1032                  * IO error is when the adapter rejects it.  All other
1033                  * conditions are a success case and resolved by the
1034                  * transport.
1035                  * IOSTAT_FCP_RSP_ERROR means:
1036                  * 1. Length of data received doesn't match total
1037                  *    transfer length in WQE
1038                  * 2. If the RSP payload does NOT match these cases:
1039                  *    a. RSP length 12/24 bytes and all zeros
1040                  *    b. NVME ERSP
1041                  */
1042                 switch (lpfc_ncmd->status) {
1043                 case IOSTAT_SUCCESS:
1044                         nCmd->transferred_length = wcqe->total_data_placed;
1045                         nCmd->rcv_rsplen = 0;
1046                         nCmd->status = 0;
1047                         break;
1048                 case IOSTAT_FCP_RSP_ERROR:
1049                         nCmd->transferred_length = wcqe->total_data_placed;
1050                         nCmd->rcv_rsplen = wcqe->parameter;
1051                         nCmd->status = 0;
1052
1053                         /* Check if this is really an ERSP */
1054                         if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN) {
1055                                 lpfc_ncmd->status = IOSTAT_SUCCESS;
1056                                 lpfc_ncmd->result = 0;
1057
1058                                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
1059                                          "6084 NVME Completion ERSP: "
1060                                          "xri %x placed x%x\n",
1061                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1062                                          wcqe->total_data_placed);
1063                                 break;
1064                         }
1065                         lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1066                                          "6081 NVME Completion Protocol Error: "
1067                                          "xri %x status x%x result x%x "
1068                                          "placed x%x\n",
1069                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1070                                          lpfc_ncmd->status, lpfc_ncmd->result,
1071                                          wcqe->total_data_placed);
1072                         break;
1073                 case IOSTAT_LOCAL_REJECT:
1074                         /* Let fall through to set command final state. */
1075                         if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED)
1076                                 lpfc_printf_vlog(vport, KERN_INFO,
1077                                          LOG_NVME_IOERR,
1078                                          "6032 Delay Aborted cmd x%px "
1079                                          "nvme cmd x%px, xri x%x, "
1080                                          "xb %d\n",
1081                                          lpfc_ncmd, nCmd,
1082                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1083                                          bf_get(lpfc_wcqe_c_xb, wcqe));
1084                         fallthrough;
1085                 default:
1086 out_err:
1087                         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1088                                          "6072 NVME Completion Error: xri %x "
1089                                          "status x%x result x%x [x%x] "
1090                                          "placed x%x\n",
1091                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1092                                          lpfc_ncmd->status, lpfc_ncmd->result,
1093                                          wcqe->parameter,
1094                                          wcqe->total_data_placed);
1095                         nCmd->transferred_length = 0;
1096                         nCmd->rcv_rsplen = 0;
1097                         nCmd->status = NVME_SC_INTERNAL;
1098                 }
1099         }
1100
1101         /* pick up SLI4 exhange busy condition */
1102         if (bf_get(lpfc_wcqe_c_xb, wcqe))
1103                 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1104         else
1105                 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1106
1107         /* Update stats and complete the IO.  There is
1108          * no need for dma unprep because the nvme_transport
1109          * owns the dma address.
1110          */
1111 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1112         if (lpfc_ncmd->ts_cmd_start) {
1113                 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
1114                 lpfc_ncmd->ts_data_io = ktime_get_ns();
1115                 phba->ktime_last_cmd = lpfc_ncmd->ts_data_io;
1116                 lpfc_io_ktime(phba, lpfc_ncmd);
1117         }
1118         if (unlikely(phba->hdwqstat_on & LPFC_CHECK_NVME_IO)) {
1119                 cpu = raw_smp_processor_id();
1120                 this_cpu_inc(phba->sli4_hba.c_stat->cmpl_io);
1121                 if (lpfc_ncmd->cpu != cpu)
1122                         lpfc_printf_vlog(vport,
1123                                          KERN_INFO, LOG_NVME_IOERR,
1124                                          "6701 CPU Check cmpl: "
1125                                          "cpu %d expect %d\n",
1126                                          cpu, lpfc_ncmd->cpu);
1127         }
1128 #endif
1129
1130         /* NVME targets need completion held off until the abort exchange
1131          * completes unless the NVME Rport is getting unregistered.
1132          */
1133
1134         if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1135                 freqpriv = nCmd->private;
1136                 freqpriv->nvme_buf = NULL;
1137                 lpfc_ncmd->nvmeCmd = NULL;
1138                 spin_unlock(&lpfc_ncmd->buf_lock);
1139                 nCmd->done(nCmd);
1140         } else
1141                 spin_unlock(&lpfc_ncmd->buf_lock);
1142
1143         /* Call release with XB=1 to queue the IO into the abort list. */
1144         lpfc_release_nvme_buf(phba, lpfc_ncmd);
1145 }
1146
1147
1148 /**
1149  * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
1150  * @vport: pointer to a host virtual N_Port data structure
1151  * @lpfc_ncmd: Pointer to lpfc scsi command
1152  * @pnode: pointer to a node-list data structure
1153  * @cstat: pointer to the control status structure
1154  *
1155  * Driver registers this routine as it io request handler.  This
1156  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1157  * data structure to the rport indicated in @lpfc_nvme_rport.
1158  *
1159  * Return value :
1160  *   0 - Success
1161  *   TODO: What are the failure codes.
1162  **/
1163 static int
1164 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
1165                       struct lpfc_io_buf *lpfc_ncmd,
1166                       struct lpfc_nodelist *pnode,
1167                       struct lpfc_fc4_ctrl_stat *cstat)
1168 {
1169         struct lpfc_hba *phba = vport->phba;
1170         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1171         struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
1172         union lpfc_wqe128 *wqe = &pwqeq->wqe;
1173         uint32_t req_len;
1174
1175         /*
1176          * There are three possibilities here - use scatter-gather segment, use
1177          * the single mapping, or neither.
1178          */
1179         if (nCmd->sg_cnt) {
1180                 if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
1181                         /* From the iwrite template, initialize words 7 - 11 */
1182                         memcpy(&wqe->words[7],
1183                                &lpfc_iwrite_cmd_template.words[7],
1184                                sizeof(uint32_t) * 5);
1185
1186                         /* Word 4 */
1187                         wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length;
1188
1189                         /* Word 5 */
1190                         if ((phba->cfg_nvme_enable_fb) &&
1191                             (pnode->nlp_flag & NLP_FIRSTBURST)) {
1192                                 req_len = lpfc_ncmd->nvmeCmd->payload_length;
1193                                 if (req_len < pnode->nvme_fb_size)
1194                                         wqe->fcp_iwrite.initial_xfer_len =
1195                                                 req_len;
1196                                 else
1197                                         wqe->fcp_iwrite.initial_xfer_len =
1198                                                 pnode->nvme_fb_size;
1199                         } else {
1200                                 wqe->fcp_iwrite.initial_xfer_len = 0;
1201                         }
1202                         cstat->output_requests++;
1203                 } else {
1204                         /* From the iread template, initialize words 7 - 11 */
1205                         memcpy(&wqe->words[7],
1206                                &lpfc_iread_cmd_template.words[7],
1207                                sizeof(uint32_t) * 5);
1208
1209                         /* Word 4 */
1210                         wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1211
1212                         /* Word 5 */
1213                         wqe->fcp_iread.rsrvd5 = 0;
1214
1215                         cstat->input_requests++;
1216                 }
1217         } else {
1218                 /* From the icmnd template, initialize words 4 - 11 */
1219                 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4],
1220                        sizeof(uint32_t) * 8);
1221                 cstat->control_requests++;
1222         }
1223
1224         if (pnode->nlp_nvme_info & NLP_NVME_NSLER)
1225                 bf_set(wqe_erp, &wqe->generic.wqe_com, 1);
1226         /*
1227          * Finish initializing those WQE fields that are independent
1228          * of the nvme_cmnd request_buffer
1229          */
1230
1231         /* Word 3 */
1232         bf_set(payload_offset_len, &wqe->fcp_icmd,
1233                (nCmd->rsplen + nCmd->cmdlen));
1234
1235         /* Word 6 */
1236         bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1237                phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1238         bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1239
1240         /* Word 8 */
1241         wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1242
1243         /* Word 9 */
1244         bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1245
1246         /* Word 10 */
1247         bf_set(wqe_xchg, &wqe->fcp_iwrite.wqe_com, LPFC_NVME_XCHG);
1248
1249         /* Words 13 14 15 are for PBDE support */
1250
1251         pwqeq->vport = vport;
1252         return 0;
1253 }
1254
1255
1256 /**
1257  * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1258  * @vport: pointer to a host virtual N_Port data structure
1259  * @lpfc_ncmd: Pointer to lpfc scsi command
1260  *
1261  * Driver registers this routine as it io request handler.  This
1262  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1263  * data structure to the rport indicated in @lpfc_nvme_rport.
1264  *
1265  * Return value :
1266  *   0 - Success
1267  *   TODO: What are the failure codes.
1268  **/
1269 static int
1270 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1271                       struct lpfc_io_buf *lpfc_ncmd)
1272 {
1273         struct lpfc_hba *phba = vport->phba;
1274         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1275         union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe;
1276         struct sli4_sge *sgl = lpfc_ncmd->dma_sgl;
1277         struct sli4_hybrid_sgl *sgl_xtra = NULL;
1278         struct scatterlist *data_sg;
1279         struct sli4_sge *first_data_sgl;
1280         struct ulp_bde64 *bde;
1281         dma_addr_t physaddr = 0;
1282         uint32_t num_bde = 0;
1283         uint32_t dma_len = 0;
1284         uint32_t dma_offset = 0;
1285         int nseg, i, j;
1286         bool lsp_just_set = false;
1287
1288         /* Fix up the command and response DMA stuff. */
1289         lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1290
1291         /*
1292          * There are three possibilities here - use scatter-gather segment, use
1293          * the single mapping, or neither.
1294          */
1295         if (nCmd->sg_cnt) {
1296                 /*
1297                  * Jump over the cmd and rsp SGEs.  The fix routine
1298                  * has already adjusted for this.
1299                  */
1300                 sgl += 2;
1301
1302                 first_data_sgl = sgl;
1303                 lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1304                 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) {
1305                         lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1306                                         "6058 Too many sg segments from "
1307                                         "NVME Transport.  Max %d, "
1308                                         "nvmeIO sg_cnt %d\n",
1309                                         phba->cfg_nvme_seg_cnt + 1,
1310                                         lpfc_ncmd->seg_cnt);
1311                         lpfc_ncmd->seg_cnt = 0;
1312                         return 1;
1313                 }
1314
1315                 /*
1316                  * The driver established a maximum scatter-gather segment count
1317                  * during probe that limits the number of sg elements in any
1318                  * single nvme command.  Just run through the seg_cnt and format
1319                  * the sge's.
1320                  */
1321                 nseg = nCmd->sg_cnt;
1322                 data_sg = nCmd->first_sgl;
1323
1324                 /* for tracking the segment boundaries */
1325                 j = 2;
1326                 for (i = 0; i < nseg; i++) {
1327                         if (data_sg == NULL) {
1328                                 lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1329                                                 "6059 dptr err %d, nseg %d\n",
1330                                                 i, nseg);
1331                                 lpfc_ncmd->seg_cnt = 0;
1332                                 return 1;
1333                         }
1334
1335                         sgl->word2 = 0;
1336                         if ((num_bde + 1) == nseg) {
1337                                 bf_set(lpfc_sli4_sge_last, sgl, 1);
1338                                 bf_set(lpfc_sli4_sge_type, sgl,
1339                                        LPFC_SGE_TYPE_DATA);
1340                         } else {
1341                                 bf_set(lpfc_sli4_sge_last, sgl, 0);
1342
1343                                 /* expand the segment */
1344                                 if (!lsp_just_set &&
1345                                     !((j + 1) % phba->border_sge_num) &&
1346                                     ((nseg - 1) != i)) {
1347                                         /* set LSP type */
1348                                         bf_set(lpfc_sli4_sge_type, sgl,
1349                                                LPFC_SGE_TYPE_LSP);
1350
1351                                         sgl_xtra = lpfc_get_sgl_per_hdwq(
1352                                                         phba, lpfc_ncmd);
1353
1354                                         if (unlikely(!sgl_xtra)) {
1355                                                 lpfc_ncmd->seg_cnt = 0;
1356                                                 return 1;
1357                                         }
1358                                         sgl->addr_lo = cpu_to_le32(putPaddrLow(
1359                                                        sgl_xtra->dma_phys_sgl));
1360                                         sgl->addr_hi = cpu_to_le32(putPaddrHigh(
1361                                                        sgl_xtra->dma_phys_sgl));
1362
1363                                 } else {
1364                                         bf_set(lpfc_sli4_sge_type, sgl,
1365                                                LPFC_SGE_TYPE_DATA);
1366                                 }
1367                         }
1368
1369                         if (!(bf_get(lpfc_sli4_sge_type, sgl) &
1370                                      LPFC_SGE_TYPE_LSP)) {
1371                                 if ((nseg - 1) == i)
1372                                         bf_set(lpfc_sli4_sge_last, sgl, 1);
1373
1374                                 physaddr = data_sg->dma_address;
1375                                 dma_len = data_sg->length;
1376                                 sgl->addr_lo = cpu_to_le32(
1377                                                          putPaddrLow(physaddr));
1378                                 sgl->addr_hi = cpu_to_le32(
1379                                                         putPaddrHigh(physaddr));
1380
1381                                 bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1382                                 sgl->word2 = cpu_to_le32(sgl->word2);
1383                                 sgl->sge_len = cpu_to_le32(dma_len);
1384
1385                                 dma_offset += dma_len;
1386                                 data_sg = sg_next(data_sg);
1387
1388                                 sgl++;
1389
1390                                 lsp_just_set = false;
1391                         } else {
1392                                 sgl->word2 = cpu_to_le32(sgl->word2);
1393
1394                                 sgl->sge_len = cpu_to_le32(
1395                                                      phba->cfg_sg_dma_buf_size);
1396
1397                                 sgl = (struct sli4_sge *)sgl_xtra->dma_sgl;
1398                                 i = i - 1;
1399
1400                                 lsp_just_set = true;
1401                         }
1402
1403                         j++;
1404                 }
1405                 if (phba->cfg_enable_pbde) {
1406                         /* Use PBDE support for first SGL only, offset == 0 */
1407                         /* Words 13-15 */
1408                         bde = (struct ulp_bde64 *)
1409                                 &wqe->words[13];
1410                         bde->addrLow = first_data_sgl->addr_lo;
1411                         bde->addrHigh = first_data_sgl->addr_hi;
1412                         bde->tus.f.bdeSize =
1413                                 le32_to_cpu(first_data_sgl->sge_len);
1414                         bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
1415                         bde->tus.w = cpu_to_le32(bde->tus.w);
1416
1417                         /* Word 11 */
1418                         bf_set(wqe_pbde, &wqe->generic.wqe_com, 1);
1419                 } else {
1420                         memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3));
1421                         bf_set(wqe_pbde, &wqe->generic.wqe_com, 0);
1422                 }
1423
1424         } else {
1425                 lpfc_ncmd->seg_cnt = 0;
1426
1427                 /* For this clause to be valid, the payload_length
1428                  * and sg_cnt must zero.
1429                  */
1430                 if (nCmd->payload_length != 0) {
1431                         lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1432                                         "6063 NVME DMA Prep Err: sg_cnt %d "
1433                                         "payload_length x%x\n",
1434                                         nCmd->sg_cnt, nCmd->payload_length);
1435                         return 1;
1436                 }
1437         }
1438         return 0;
1439 }
1440
1441 /**
1442  * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1443  * @pnvme_lport: Pointer to the driver's local port data
1444  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1445  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1446  * @pnvme_fcreq: IO request from nvme fc to driver.
1447  *
1448  * Driver registers this routine as it io request handler.  This
1449  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1450  * data structure to the rport indicated in @lpfc_nvme_rport.
1451  *
1452  * Return value :
1453  *   0 - Success
1454  *   TODO: What are the failure codes.
1455  **/
1456 static int
1457 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1458                         struct nvme_fc_remote_port *pnvme_rport,
1459                         void *hw_queue_handle,
1460                         struct nvmefc_fcp_req *pnvme_fcreq)
1461 {
1462         int ret = 0;
1463         int expedite = 0;
1464         int idx, cpu;
1465         struct lpfc_nvme_lport *lport;
1466         struct lpfc_fc4_ctrl_stat *cstat;
1467         struct lpfc_vport *vport;
1468         struct lpfc_hba *phba;
1469         struct lpfc_nodelist *ndlp;
1470         struct lpfc_io_buf *lpfc_ncmd;
1471         struct lpfc_nvme_rport *rport;
1472         struct lpfc_nvme_qhandle *lpfc_queue_info;
1473         struct lpfc_nvme_fcpreq_priv *freqpriv;
1474         struct nvme_common_command *sqe;
1475 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1476         uint64_t start = 0;
1477 #endif
1478
1479         /* Validate pointers. LLDD fault handling with transport does
1480          * have timing races.
1481          */
1482         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1483         if (unlikely(!lport)) {
1484                 ret = -EINVAL;
1485                 goto out_fail;
1486         }
1487
1488         vport = lport->vport;
1489
1490         if (unlikely(!hw_queue_handle)) {
1491                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1492                                  "6117 Fail IO, NULL hw_queue_handle\n");
1493                 atomic_inc(&lport->xmt_fcp_err);
1494                 ret = -EBUSY;
1495                 goto out_fail;
1496         }
1497
1498         phba = vport->phba;
1499
1500         if (unlikely(vport->load_flag & FC_UNLOADING)) {
1501                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1502                                  "6124 Fail IO, Driver unload\n");
1503                 atomic_inc(&lport->xmt_fcp_err);
1504                 ret = -ENODEV;
1505                 goto out_fail;
1506         }
1507
1508         freqpriv = pnvme_fcreq->private;
1509         if (unlikely(!freqpriv)) {
1510                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1511                                  "6158 Fail IO, NULL request data\n");
1512                 atomic_inc(&lport->xmt_fcp_err);
1513                 ret = -EINVAL;
1514                 goto out_fail;
1515         }
1516
1517 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1518         if (phba->ktime_on)
1519                 start = ktime_get_ns();
1520 #endif
1521         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1522         lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1523
1524         /*
1525          * Catch race where our node has transitioned, but the
1526          * transport is still transitioning.
1527          */
1528         ndlp = rport->ndlp;
1529         if (!ndlp) {
1530                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1531                                  "6053 Busy IO, ndlp not ready: rport x%px "
1532                                   "ndlp x%px, DID x%06x\n",
1533                                  rport, ndlp, pnvme_rport->port_id);
1534                 atomic_inc(&lport->xmt_fcp_err);
1535                 ret = -EBUSY;
1536                 goto out_fail;
1537         }
1538
1539         /* The remote node has to be a mapped target or it's an error. */
1540         if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1541             (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1542                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1543                                  "6036 Fail IO, DID x%06x not ready for "
1544                                  "IO. State x%x, Type x%x Flg x%x\n",
1545                                  pnvme_rport->port_id,
1546                                  ndlp->nlp_state, ndlp->nlp_type,
1547                                  ndlp->fc4_xpt_flags);
1548                 atomic_inc(&lport->xmt_fcp_bad_ndlp);
1549                 ret = -EBUSY;
1550                 goto out_fail;
1551
1552         }
1553
1554         /* Currently only NVME Keep alive commands should be expedited
1555          * if the driver runs out of a resource. These should only be
1556          * issued on the admin queue, qidx 0
1557          */
1558         if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
1559                 sqe = &((struct nvme_fc_cmd_iu *)
1560                         pnvme_fcreq->cmdaddr)->sqe.common;
1561                 if (sqe->opcode == nvme_admin_keep_alive)
1562                         expedite = 1;
1563         }
1564
1565         /* The node is shared with FCP IO, make sure the IO pending count does
1566          * not exceed the programmed depth.
1567          */
1568         if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1569                 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
1570                     !expedite) {
1571                         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1572                                          "6174 Fail IO, ndlp qdepth exceeded: "
1573                                          "idx %d DID %x pend %d qdepth %d\n",
1574                                          lpfc_queue_info->index, ndlp->nlp_DID,
1575                                          atomic_read(&ndlp->cmd_pending),
1576                                          ndlp->cmd_qdepth);
1577                         atomic_inc(&lport->xmt_fcp_qdepth);
1578                         ret = -EBUSY;
1579                         goto out_fail;
1580                 }
1581         }
1582
1583         /* Lookup Hardware Queue index based on fcp_io_sched module parameter */
1584         if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) {
1585                 idx = lpfc_queue_info->index;
1586         } else {
1587                 cpu = raw_smp_processor_id();
1588                 idx = phba->sli4_hba.cpu_map[cpu].hdwq;
1589         }
1590
1591         lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, idx, expedite);
1592         if (lpfc_ncmd == NULL) {
1593                 atomic_inc(&lport->xmt_fcp_noxri);
1594                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1595                                  "6065 Fail IO, driver buffer pool is empty: "
1596                                  "idx %d DID %x\n",
1597                                  lpfc_queue_info->index, ndlp->nlp_DID);
1598                 ret = -EBUSY;
1599                 goto out_fail;
1600         }
1601 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1602         if (start) {
1603                 lpfc_ncmd->ts_cmd_start = start;
1604                 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1605         } else {
1606                 lpfc_ncmd->ts_cmd_start = 0;
1607         }
1608 #endif
1609
1610         /*
1611          * Store the data needed by the driver to issue, abort, and complete
1612          * an IO.
1613          * Do not let the IO hang out forever.  There is no midlayer issuing
1614          * an abort so inform the FW of the maximum IO pending time.
1615          */
1616         freqpriv->nvme_buf = lpfc_ncmd;
1617         lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1618         lpfc_ncmd->ndlp = ndlp;
1619         lpfc_ncmd->qidx = lpfc_queue_info->qidx;
1620
1621         /*
1622          * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1623          * This identfier was create in our hardware queue create callback
1624          * routine. The driver now is dependent on the IO queue steering from
1625          * the transport.  We are trusting the upper NVME layers know which
1626          * index to use and that they have affinitized a CPU to this hardware
1627          * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1628          */
1629         lpfc_ncmd->cur_iocbq.hba_wqidx = idx;
1630         cstat = &phba->sli4_hba.hdwq[idx].nvme_cstat;
1631
1632         lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat);
1633         ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1634         if (ret) {
1635                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1636                                  "6175 Fail IO, Prep DMA: "
1637                                  "idx %d DID %x\n",
1638                                  lpfc_queue_info->index, ndlp->nlp_DID);
1639                 atomic_inc(&lport->xmt_fcp_err);
1640                 ret = -ENOMEM;
1641                 goto out_free_nvme_buf;
1642         }
1643
1644         lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1645                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1646                          lpfc_queue_info->index, ndlp->nlp_DID);
1647
1648         ret = lpfc_sli4_issue_wqe(phba, lpfc_ncmd->hdwq, &lpfc_ncmd->cur_iocbq);
1649         if (ret) {
1650                 atomic_inc(&lport->xmt_fcp_wqerr);
1651                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1652                                  "6113 Fail IO, Could not issue WQE err %x "
1653                                  "sid: x%x did: x%x oxid: x%x\n",
1654                                  ret, vport->fc_myDID, ndlp->nlp_DID,
1655                                  lpfc_ncmd->cur_iocbq.sli4_xritag);
1656                 goto out_free_nvme_buf;
1657         }
1658
1659         if (phba->cfg_xri_rebalancing)
1660                 lpfc_keep_pvt_pool_above_lowwm(phba, lpfc_ncmd->hdwq_no);
1661
1662 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1663         if (lpfc_ncmd->ts_cmd_start)
1664                 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1665
1666         if (phba->hdwqstat_on & LPFC_CHECK_NVME_IO) {
1667                 cpu = raw_smp_processor_id();
1668                 this_cpu_inc(phba->sli4_hba.c_stat->xmt_io);
1669                 lpfc_ncmd->cpu = cpu;
1670                 if (idx != cpu)
1671                         lpfc_printf_vlog(vport,
1672                                          KERN_INFO, LOG_NVME_IOERR,
1673                                         "6702 CPU Check cmd: "
1674                                         "cpu %d wq %d\n",
1675                                         lpfc_ncmd->cpu,
1676                                         lpfc_queue_info->index);
1677         }
1678 #endif
1679         return 0;
1680
1681  out_free_nvme_buf:
1682         if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1683                 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1684                         cstat->output_requests--;
1685                 else
1686                         cstat->input_requests--;
1687         } else
1688                 cstat->control_requests--;
1689         lpfc_release_nvme_buf(phba, lpfc_ncmd);
1690  out_fail:
1691         return ret;
1692 }
1693
1694 /**
1695  * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1696  * @phba: Pointer to HBA context object
1697  * @cmdiocb: Pointer to command iocb object.
1698  * @abts_cmpl: Pointer to wcqe complete object.
1699  *
1700  * This is the callback function for any NVME FCP IO that was aborted.
1701  *
1702  * Return value:
1703  *   None
1704  **/
1705 void
1706 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1707                            struct lpfc_wcqe_complete *abts_cmpl)
1708 {
1709         lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
1710                         "6145 ABORT_XRI_CN completing on rpi x%x "
1711                         "original iotag x%x, abort cmd iotag x%x "
1712                         "req_tag x%x, status x%x, hwstatus x%x\n",
1713                         cmdiocb->iocb.un.acxri.abortContextTag,
1714                         cmdiocb->iocb.un.acxri.abortIoTag,
1715                         cmdiocb->iotag,
1716                         bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1717                         bf_get(lpfc_wcqe_c_status, abts_cmpl),
1718                         bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1719         lpfc_sli_release_iocbq(phba, cmdiocb);
1720 }
1721
1722 /**
1723  * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1724  * @pnvme_lport: Pointer to the driver's local port data
1725  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1726  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1727  * @pnvme_fcreq: IO request from nvme fc to driver.
1728  *
1729  * Driver registers this routine as its nvme request io abort handler.  This
1730  * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1731  * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
1732  * is executed asynchronously - one the target is validated as "MAPPED" and
1733  * ready for IO, the driver issues the abort request and returns.
1734  *
1735  * Return value:
1736  *   None
1737  **/
1738 static void
1739 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1740                     struct nvme_fc_remote_port *pnvme_rport,
1741                     void *hw_queue_handle,
1742                     struct nvmefc_fcp_req *pnvme_fcreq)
1743 {
1744         struct lpfc_nvme_lport *lport;
1745         struct lpfc_vport *vport;
1746         struct lpfc_hba *phba;
1747         struct lpfc_io_buf *lpfc_nbuf;
1748         struct lpfc_iocbq *nvmereq_wqe;
1749         struct lpfc_nvme_fcpreq_priv *freqpriv;
1750         unsigned long flags;
1751         int ret_val;
1752
1753         /* Validate pointers. LLDD fault handling with transport does
1754          * have timing races.
1755          */
1756         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1757         if (unlikely(!lport))
1758                 return;
1759
1760         vport = lport->vport;
1761
1762         if (unlikely(!hw_queue_handle)) {
1763                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1764                                  "6129 Fail Abort, HW Queue Handle NULL.\n");
1765                 return;
1766         }
1767
1768         phba = vport->phba;
1769         freqpriv = pnvme_fcreq->private;
1770
1771         if (unlikely(!freqpriv))
1772                 return;
1773         if (vport->load_flag & FC_UNLOADING)
1774                 return;
1775
1776         /* Announce entry to new IO submit field. */
1777         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1778                          "6002 Abort Request to rport DID x%06x "
1779                          "for nvme_fc_req x%px\n",
1780                          pnvme_rport->port_id,
1781                          pnvme_fcreq);
1782
1783         /* If the hba is getting reset, this flag is set.  It is
1784          * cleared when the reset is complete and rings reestablished.
1785          */
1786         spin_lock_irqsave(&phba->hbalock, flags);
1787         /* driver queued commands are in process of being flushed */
1788         if (phba->hba_flag & HBA_IOQ_FLUSH) {
1789                 spin_unlock_irqrestore(&phba->hbalock, flags);
1790                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1791                                  "6139 Driver in reset cleanup - flushing "
1792                                  "NVME Req now.  hba_flag x%x\n",
1793                                  phba->hba_flag);
1794                 return;
1795         }
1796
1797         lpfc_nbuf = freqpriv->nvme_buf;
1798         if (!lpfc_nbuf) {
1799                 spin_unlock_irqrestore(&phba->hbalock, flags);
1800                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1801                                  "6140 NVME IO req has no matching lpfc nvme "
1802                                  "io buffer.  Skipping abort req.\n");
1803                 return;
1804         } else if (!lpfc_nbuf->nvmeCmd) {
1805                 spin_unlock_irqrestore(&phba->hbalock, flags);
1806                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1807                                  "6141 lpfc NVME IO req has no nvme_fcreq "
1808                                  "io buffer.  Skipping abort req.\n");
1809                 return;
1810         }
1811         nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1812
1813         /* Guard against IO completion being called at same time */
1814         spin_lock(&lpfc_nbuf->buf_lock);
1815
1816         /*
1817          * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1818          * state must match the nvme_fcreq passed by the nvme
1819          * transport.  If they don't match, it is likely the driver
1820          * has already completed the NVME IO and the nvme transport
1821          * has not seen it yet.
1822          */
1823         if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1824                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1825                                  "6143 NVME req mismatch: "
1826                                  "lpfc_nbuf x%px nvmeCmd x%px, "
1827                                  "pnvme_fcreq x%px.  Skipping Abort xri x%x\n",
1828                                  lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1829                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1830                 goto out_unlock;
1831         }
1832
1833         /* Don't abort IOs no longer on the pending queue. */
1834         if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) {
1835                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1836                                  "6142 NVME IO req x%px not queued - skipping "
1837                                  "abort req xri x%x\n",
1838                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1839                 goto out_unlock;
1840         }
1841
1842         atomic_inc(&lport->xmt_fcp_abort);
1843         lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1844                          nvmereq_wqe->sli4_xritag,
1845                          nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1846
1847         /* Outstanding abort is in progress */
1848         if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) {
1849                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1850                                  "6144 Outstanding NVME I/O Abort Request "
1851                                  "still pending on nvme_fcreq x%px, "
1852                                  "lpfc_ncmd x%px xri x%x\n",
1853                                  pnvme_fcreq, lpfc_nbuf,
1854                                  nvmereq_wqe->sli4_xritag);
1855                 goto out_unlock;
1856         }
1857
1858         ret_val = lpfc_sli4_issue_abort_iotag(phba, nvmereq_wqe,
1859                                               lpfc_nvme_abort_fcreq_cmpl);
1860
1861         spin_unlock(&lpfc_nbuf->buf_lock);
1862         spin_unlock_irqrestore(&phba->hbalock, flags);
1863
1864         /* Make sure HBA is alive */
1865         lpfc_issue_hb_tmo(phba);
1866
1867         if (ret_val != WQE_SUCCESS) {
1868                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1869                                  "6137 Failed abts issue_wqe with status x%x "
1870                                  "for nvme_fcreq x%px.\n",
1871                                  ret_val, pnvme_fcreq);
1872                 return;
1873         }
1874
1875         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1876                          "6138 Transport Abort NVME Request Issued for "
1877                          "ox_id x%x\n",
1878                          nvmereq_wqe->sli4_xritag);
1879         return;
1880
1881 out_unlock:
1882         spin_unlock(&lpfc_nbuf->buf_lock);
1883         spin_unlock_irqrestore(&phba->hbalock, flags);
1884         return;
1885 }
1886
1887 /* Declare and initialization an instance of the FC NVME template. */
1888 static struct nvme_fc_port_template lpfc_nvme_template = {
1889         /* initiator-based functions */
1890         .localport_delete  = lpfc_nvme_localport_delete,
1891         .remoteport_delete = lpfc_nvme_remoteport_delete,
1892         .create_queue = lpfc_nvme_create_queue,
1893         .delete_queue = lpfc_nvme_delete_queue,
1894         .ls_req       = lpfc_nvme_ls_req,
1895         .fcp_io       = lpfc_nvme_fcp_io_submit,
1896         .ls_abort     = lpfc_nvme_ls_abort,
1897         .fcp_abort    = lpfc_nvme_fcp_abort,
1898         .xmt_ls_rsp   = lpfc_nvme_xmt_ls_rsp,
1899
1900         .max_hw_queues = 1,
1901         .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1902         .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1903         .dma_boundary = 0xFFFFFFFF,
1904
1905         /* Sizes of additional private data for data structures.
1906          * No use for the last two sizes at this time.
1907          */
1908         .local_priv_sz = sizeof(struct lpfc_nvme_lport),
1909         .remote_priv_sz = sizeof(struct lpfc_nvme_rport),
1910         .lsrqst_priv_sz = 0,
1911         .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
1912 };
1913
1914 /*
1915  * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA
1916  *
1917  * This routine removes a nvme buffer from head of @hdwq io_buf_list
1918  * and returns to caller.
1919  *
1920  * Return codes:
1921  *   NULL - Error
1922  *   Pointer to lpfc_nvme_buf - Success
1923  **/
1924 static struct lpfc_io_buf *
1925 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
1926                   int idx, int expedite)
1927 {
1928         struct lpfc_io_buf *lpfc_ncmd;
1929         struct lpfc_sli4_hdw_queue *qp;
1930         struct sli4_sge *sgl;
1931         struct lpfc_iocbq *pwqeq;
1932         union lpfc_wqe128 *wqe;
1933
1934         lpfc_ncmd = lpfc_get_io_buf(phba, NULL, idx, expedite);
1935
1936         if (lpfc_ncmd) {
1937                 pwqeq = &(lpfc_ncmd->cur_iocbq);
1938                 wqe = &pwqeq->wqe;
1939
1940                 /* Setup key fields in buffer that may have been changed
1941                  * if other protocols used this buffer.
1942                  */
1943                 pwqeq->iocb_flag = LPFC_IO_NVME;
1944                 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl;
1945                 lpfc_ncmd->start_time = jiffies;
1946                 lpfc_ncmd->flags = 0;
1947
1948                 /* Rsp SGE will be filled in when we rcv an IO
1949                  * from the NVME Layer to be sent.
1950                  * The cmd is going to be embedded so we need a SKIP SGE.
1951                  */
1952                 sgl = lpfc_ncmd->dma_sgl;
1953                 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
1954                 bf_set(lpfc_sli4_sge_last, sgl, 0);
1955                 sgl->word2 = cpu_to_le32(sgl->word2);
1956                 /* Fill in word 3 / sgl_len during cmd submission */
1957
1958                 /* Initialize 64 bytes only */
1959                 memset(wqe, 0, sizeof(union lpfc_wqe));
1960
1961                 if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1962                         atomic_inc(&ndlp->cmd_pending);
1963                         lpfc_ncmd->flags |= LPFC_SBUF_BUMP_QDEPTH;
1964                 }
1965
1966         } else {
1967                 qp = &phba->sli4_hba.hdwq[idx];
1968                 qp->empty_io_bufs++;
1969         }
1970
1971         return  lpfc_ncmd;
1972 }
1973
1974 /**
1975  * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
1976  * @phba: The Hba for which this call is being executed.
1977  * @lpfc_ncmd: The nvme buffer which is being released.
1978  *
1979  * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
1980  * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer
1981  * and cannot be reused for at least RA_TOV amount of time if it was
1982  * aborted.
1983  **/
1984 static void
1985 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_io_buf *lpfc_ncmd)
1986 {
1987         struct lpfc_sli4_hdw_queue *qp;
1988         unsigned long iflag = 0;
1989
1990         if ((lpfc_ncmd->flags & LPFC_SBUF_BUMP_QDEPTH) && lpfc_ncmd->ndlp)
1991                 atomic_dec(&lpfc_ncmd->ndlp->cmd_pending);
1992
1993         lpfc_ncmd->ndlp = NULL;
1994         lpfc_ncmd->flags &= ~LPFC_SBUF_BUMP_QDEPTH;
1995
1996         qp = lpfc_ncmd->hdwq;
1997         if (unlikely(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1998                 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
1999                                 "6310 XB release deferred for "
2000                                 "ox_id x%x on reqtag x%x\n",
2001                                 lpfc_ncmd->cur_iocbq.sli4_xritag,
2002                                 lpfc_ncmd->cur_iocbq.iotag);
2003
2004                 spin_lock_irqsave(&qp->abts_io_buf_list_lock, iflag);
2005                 list_add_tail(&lpfc_ncmd->list,
2006                         &qp->lpfc_abts_io_buf_list);
2007                 qp->abts_nvme_io_bufs++;
2008                 spin_unlock_irqrestore(&qp->abts_io_buf_list_lock, iflag);
2009         } else
2010                 lpfc_release_io_buf(phba, (struct lpfc_io_buf *)lpfc_ncmd, qp);
2011 }
2012
2013 /**
2014  * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2015  * @vport: the lpfc_vport instance requesting a localport.
2016  *
2017  * This routine is invoked to create an nvme localport instance to bind
2018  * to the nvme_fc_transport.  It is called once during driver load
2019  * like lpfc_create_shost after all other services are initialized.
2020  * It requires a vport, vpi, and wwns at call time.  Other localport
2021  * parameters are modified as the driver's FCID and the Fabric WWN
2022  * are established.
2023  *
2024  * Return codes
2025  *      0 - successful
2026  *      -ENOMEM - no heap memory available
2027  *      other values - from nvme registration upcall
2028  **/
2029 int
2030 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2031 {
2032         int ret = 0;
2033         struct lpfc_hba  *phba = vport->phba;
2034         struct nvme_fc_port_info nfcp_info;
2035         struct nvme_fc_local_port *localport;
2036         struct lpfc_nvme_lport *lport;
2037
2038         /* Initialize this localport instance.  The vport wwn usage ensures
2039          * that NPIV is accounted for.
2040          */
2041         memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2042         nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2043         nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2044         nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2045
2046         /* We need to tell the transport layer + 1 because it takes page
2047          * alignment into account. When space for the SGL is allocated we
2048          * allocate + 3, one for cmd, one for rsp and one for this alignment
2049          */
2050         lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2051
2052         /* Advertise how many hw queues we support based on cfg_hdw_queue,
2053          * which will not exceed cpu count.
2054          */
2055         lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue;
2056
2057         if (!IS_ENABLED(CONFIG_NVME_FC))
2058                 return ret;
2059
2060         /* localport is allocated from the stack, but the registration
2061          * call allocates heap memory as well as the private area.
2062          */
2063
2064         ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2065                                          &vport->phba->pcidev->dev, &localport);
2066         if (!ret) {
2067                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2068                                  "6005 Successfully registered local "
2069                                  "NVME port num %d, localP x%px, private "
2070                                  "x%px, sg_seg %d\n",
2071                                  localport->port_num, localport,
2072                                  localport->private,
2073                                  lpfc_nvme_template.max_sgl_segments);
2074
2075                 /* Private is our lport size declared in the template. */
2076                 lport = (struct lpfc_nvme_lport *)localport->private;
2077                 vport->localport = localport;
2078                 lport->vport = vport;
2079                 vport->nvmei_support = 1;
2080
2081                 atomic_set(&lport->xmt_fcp_noxri, 0);
2082                 atomic_set(&lport->xmt_fcp_bad_ndlp, 0);
2083                 atomic_set(&lport->xmt_fcp_qdepth, 0);
2084                 atomic_set(&lport->xmt_fcp_err, 0);
2085                 atomic_set(&lport->xmt_fcp_wqerr, 0);
2086                 atomic_set(&lport->xmt_fcp_abort, 0);
2087                 atomic_set(&lport->xmt_ls_abort, 0);
2088                 atomic_set(&lport->xmt_ls_err, 0);
2089                 atomic_set(&lport->cmpl_fcp_xb, 0);
2090                 atomic_set(&lport->cmpl_fcp_err, 0);
2091                 atomic_set(&lport->cmpl_ls_xb, 0);
2092                 atomic_set(&lport->cmpl_ls_err, 0);
2093
2094                 atomic_set(&lport->fc4NvmeLsRequests, 0);
2095                 atomic_set(&lport->fc4NvmeLsCmpls, 0);
2096         }
2097
2098         return ret;
2099 }
2100
2101 #if (IS_ENABLED(CONFIG_NVME_FC))
2102 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
2103  *
2104  * The driver has to wait for the host nvme transport to callback
2105  * indicating the localport has successfully unregistered all
2106  * resources.  Since this is an uninterruptible wait, loop every ten
2107  * seconds and print a message indicating no progress.
2108  *
2109  * An uninterruptible wait is used because of the risk of transport-to-
2110  * driver state mismatch.
2111  */
2112 static void
2113 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
2114                            struct lpfc_nvme_lport *lport,
2115                            struct completion *lport_unreg_cmp)
2116 {
2117         u32 wait_tmo;
2118         int ret, i, pending = 0;
2119         struct lpfc_sli_ring  *pring;
2120         struct lpfc_hba  *phba = vport->phba;
2121         struct lpfc_sli4_hdw_queue *qp;
2122         int abts_scsi, abts_nvme;
2123
2124         /* Host transport has to clean up and confirm requiring an indefinite
2125          * wait. Print a message if a 10 second wait expires and renew the
2126          * wait. This is unexpected.
2127          */
2128         wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
2129         while (true) {
2130                 ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo);
2131                 if (unlikely(!ret)) {
2132                         pending = 0;
2133                         abts_scsi = 0;
2134                         abts_nvme = 0;
2135                         for (i = 0; i < phba->cfg_hdw_queue; i++) {
2136                                 qp = &phba->sli4_hba.hdwq[i];
2137                                 pring = qp->io_wq->pring;
2138                                 if (!pring)
2139                                         continue;
2140                                 pending += pring->txcmplq_cnt;
2141                                 abts_scsi += qp->abts_scsi_io_bufs;
2142                                 abts_nvme += qp->abts_nvme_io_bufs;
2143                         }
2144                         lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2145                                          "6176 Lport x%px Localport x%px wait "
2146                                          "timed out. Pending %d [%d:%d]. "
2147                                          "Renewing.\n",
2148                                          lport, vport->localport, pending,
2149                                          abts_scsi, abts_nvme);
2150                         continue;
2151                 }
2152                 break;
2153         }
2154         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
2155                          "6177 Lport x%px Localport x%px Complete Success\n",
2156                          lport, vport->localport);
2157 }
2158 #endif
2159
2160 /**
2161  * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2162  * @vport: pointer to a host virtual N_Port data structure
2163  *
2164  * This routine is invoked to destroy all lports bound to the phba.
2165  * The lport memory was allocated by the nvme fc transport and is
2166  * released there.  This routine ensures all rports bound to the
2167  * lport have been disconnected.
2168  *
2169  **/
2170 void
2171 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2172 {
2173 #if (IS_ENABLED(CONFIG_NVME_FC))
2174         struct nvme_fc_local_port *localport;
2175         struct lpfc_nvme_lport *lport;
2176         int ret;
2177         DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp);
2178
2179         if (vport->nvmei_support == 0)
2180                 return;
2181
2182         localport = vport->localport;
2183         lport = (struct lpfc_nvme_lport *)localport->private;
2184
2185         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2186                          "6011 Destroying NVME localport x%px\n",
2187                          localport);
2188
2189         /* lport's rport list is clear.  Unregister
2190          * lport and release resources.
2191          */
2192         lport->lport_unreg_cmp = &lport_unreg_cmp;
2193         ret = nvme_fc_unregister_localport(localport);
2194
2195         /* Wait for completion.  This either blocks
2196          * indefinitely or succeeds
2197          */
2198         lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp);
2199         vport->localport = NULL;
2200
2201         /* Regardless of the unregister upcall response, clear
2202          * nvmei_support.  All rports are unregistered and the
2203          * driver will clean up.
2204          */
2205         vport->nvmei_support = 0;
2206         if (ret == 0) {
2207                 lpfc_printf_vlog(vport,
2208                                  KERN_INFO, LOG_NVME_DISC,
2209                                  "6009 Unregistered lport Success\n");
2210         } else {
2211                 lpfc_printf_vlog(vport,
2212                                  KERN_INFO, LOG_NVME_DISC,
2213                                  "6010 Unregistered lport "
2214                                  "Failed, status x%x\n",
2215                                  ret);
2216         }
2217 #endif
2218 }
2219
2220 void
2221 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2222 {
2223 #if (IS_ENABLED(CONFIG_NVME_FC))
2224         struct nvme_fc_local_port *localport;
2225         struct lpfc_nvme_lport *lport;
2226
2227         localport = vport->localport;
2228         if (!localport) {
2229                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2230                                  "6710 Update NVME fail. No localport\n");
2231                 return;
2232         }
2233         lport = (struct lpfc_nvme_lport *)localport->private;
2234         if (!lport) {
2235                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2236                                  "6171 Update NVME fail. localP x%px, No lport\n",
2237                                  localport);
2238                 return;
2239         }
2240         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2241                          "6012 Update NVME lport x%px did x%x\n",
2242                          localport, vport->fc_myDID);
2243
2244         localport->port_id = vport->fc_myDID;
2245         if (localport->port_id == 0)
2246                 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2247         else
2248                 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2249
2250         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2251                          "6030 bound lport x%px to DID x%06x\n",
2252                          lport, localport->port_id);
2253 #endif
2254 }
2255
2256 int
2257 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2258 {
2259 #if (IS_ENABLED(CONFIG_NVME_FC))
2260         int ret = 0;
2261         struct nvme_fc_local_port *localport;
2262         struct lpfc_nvme_lport *lport;
2263         struct lpfc_nvme_rport *rport;
2264         struct lpfc_nvme_rport *oldrport;
2265         struct nvme_fc_remote_port *remote_port;
2266         struct nvme_fc_port_info rpinfo;
2267         struct lpfc_nodelist *prev_ndlp = NULL;
2268         struct fc_rport *srport = ndlp->rport;
2269
2270         lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2271                          "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2272                          ndlp->nlp_DID, ndlp->nlp_type);
2273
2274         localport = vport->localport;
2275         if (!localport)
2276                 return 0;
2277
2278         lport = (struct lpfc_nvme_lport *)localport->private;
2279
2280         /* NVME rports are not preserved across devloss.
2281          * Just register this instance.  Note, rpinfo->dev_loss_tmo
2282          * is left 0 to indicate accept transport defaults.  The
2283          * driver communicates port role capabilities consistent
2284          * with the PRLI response data.
2285          */
2286         memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2287         rpinfo.port_id = ndlp->nlp_DID;
2288         if (ndlp->nlp_type & NLP_NVME_TARGET)
2289                 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2290         if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2291                 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2292
2293         if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2294                 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2295
2296         rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2297         rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2298         if (srport)
2299                 rpinfo.dev_loss_tmo = srport->dev_loss_tmo;
2300         else
2301                 rpinfo.dev_loss_tmo = vport->cfg_devloss_tmo;
2302
2303         spin_lock_irq(&ndlp->lock);
2304         oldrport = lpfc_ndlp_get_nrport(ndlp);
2305         if (oldrport) {
2306                 prev_ndlp = oldrport->ndlp;
2307                 spin_unlock_irq(&ndlp->lock);
2308         } else {
2309                 spin_unlock_irq(&ndlp->lock);
2310                 if (!lpfc_nlp_get(ndlp)) {
2311                         dev_warn(&vport->phba->pcidev->dev,
2312                                  "Warning - No node ref - exit register\n");
2313                         return 0;
2314                 }
2315         }
2316
2317         ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2318         if (!ret) {
2319                 /* If the ndlp already has an nrport, this is just
2320                  * a resume of the existing rport.  Else this is a
2321                  * new rport.
2322                  */
2323                 /* Guard against an unregister/reregister
2324                  * race that leaves the WAIT flag set.
2325                  */
2326                 spin_lock_irq(&ndlp->lock);
2327                 ndlp->fc4_xpt_flags &= ~NLP_WAIT_FOR_UNREG;
2328                 ndlp->fc4_xpt_flags |= NVME_XPT_REGD;
2329                 spin_unlock_irq(&ndlp->lock);
2330                 rport = remote_port->private;
2331                 if (oldrport) {
2332
2333                         /* Sever the ndlp<->rport association
2334                          * before dropping the ndlp ref from
2335                          * register.
2336                          */
2337                         spin_lock_irq(&ndlp->lock);
2338                         ndlp->nrport = NULL;
2339                         ndlp->fc4_xpt_flags &= ~NLP_WAIT_FOR_UNREG;
2340                         spin_unlock_irq(&ndlp->lock);
2341                         rport->ndlp = NULL;
2342                         rport->remoteport = NULL;
2343
2344                         /* Reference only removed if previous NDLP is no longer
2345                          * active. It might be just a swap and removing the
2346                          * reference would cause a premature cleanup.
2347                          */
2348                         if (prev_ndlp && prev_ndlp != ndlp) {
2349                                 if (!prev_ndlp->nrport)
2350                                         lpfc_nlp_put(prev_ndlp);
2351                         }
2352                 }
2353
2354                 /* Clean bind the rport to the ndlp. */
2355                 rport->remoteport = remote_port;
2356                 rport->lport = lport;
2357                 rport->ndlp = ndlp;
2358                 spin_lock_irq(&ndlp->lock);
2359                 ndlp->nrport = rport;
2360                 spin_unlock_irq(&ndlp->lock);
2361                 lpfc_printf_vlog(vport, KERN_INFO,
2362                                  LOG_NVME_DISC | LOG_NODE,
2363                                  "6022 Bind lport x%px to remoteport x%px "
2364                                  "rport x%px WWNN 0x%llx, "
2365                                  "Rport WWPN 0x%llx DID "
2366                                  "x%06x Role x%x, ndlp %p prev_ndlp x%px\n",
2367                                  lport, remote_port, rport,
2368                                  rpinfo.node_name, rpinfo.port_name,
2369                                  rpinfo.port_id, rpinfo.port_role,
2370                                  ndlp, prev_ndlp);
2371         } else {
2372                 lpfc_printf_vlog(vport, KERN_ERR,
2373                                  LOG_TRACE_EVENT,
2374                                  "6031 RemotePort Registration failed "
2375                                  "err: %d, DID x%06x\n",
2376                                  ret, ndlp->nlp_DID);
2377         }
2378
2379         return ret;
2380 #else
2381         return 0;
2382 #endif
2383 }
2384
2385 /*
2386  * lpfc_nvme_rescan_port - Check to see if we should rescan this remoteport
2387  *
2388  * If the ndlp represents an NVME Target, that we are logged into,
2389  * ping the NVME FC Transport layer to initiate a device rescan
2390  * on this remote NPort.
2391  */
2392 void
2393 lpfc_nvme_rescan_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2394 {
2395 #if (IS_ENABLED(CONFIG_NVME_FC))
2396         struct lpfc_nvme_rport *nrport;
2397         struct nvme_fc_remote_port *remoteport = NULL;
2398
2399         spin_lock_irq(&ndlp->lock);
2400         nrport = lpfc_ndlp_get_nrport(ndlp);
2401         if (nrport)
2402                 remoteport = nrport->remoteport;
2403         spin_unlock_irq(&ndlp->lock);
2404
2405         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2406                          "6170 Rescan NPort DID x%06x type x%x "
2407                          "state x%x nrport x%px remoteport x%px\n",
2408                          ndlp->nlp_DID, ndlp->nlp_type, ndlp->nlp_state,
2409                          nrport, remoteport);
2410
2411         if (!nrport || !remoteport)
2412                 goto rescan_exit;
2413
2414         /* Only rescan if we are an NVME target in the MAPPED state */
2415         if (remoteport->port_role & FC_PORT_ROLE_NVME_DISCOVERY &&
2416             ndlp->nlp_state == NLP_STE_MAPPED_NODE) {
2417                 nvme_fc_rescan_remoteport(remoteport);
2418
2419                 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2420                                  "6172 NVME rescanned DID x%06x "
2421                                  "port_state x%x\n",
2422                                  ndlp->nlp_DID, remoteport->port_state);
2423         }
2424         return;
2425  rescan_exit:
2426         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2427                          "6169 Skip NVME Rport Rescan, NVME remoteport "
2428                          "unregistered\n");
2429 #endif
2430 }
2431
2432 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2433  *
2434  * There is no notion of Devloss or rport recovery from the current
2435  * nvme_transport perspective.  Loss of an rport just means IO cannot
2436  * be sent and recovery is completely up to the initator.
2437  * For now, the driver just unbinds the DID and port_role so that
2438  * no further IO can be issued.  Changes are planned for later.
2439  *
2440  * Notes - the ndlp reference count is not decremented here since
2441  * since there is no nvme_transport api for devloss.  Node ref count
2442  * is only adjusted in driver unload.
2443  */
2444 void
2445 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2446 {
2447 #if (IS_ENABLED(CONFIG_NVME_FC))
2448         int ret;
2449         struct nvme_fc_local_port *localport;
2450         struct lpfc_nvme_lport *lport;
2451         struct lpfc_nvme_rport *rport;
2452         struct nvme_fc_remote_port *remoteport = NULL;
2453
2454         localport = vport->localport;
2455
2456         /* This is fundamental error.  The localport is always
2457          * available until driver unload.  Just exit.
2458          */
2459         if (!localport)
2460                 return;
2461
2462         lport = (struct lpfc_nvme_lport *)localport->private;
2463         if (!lport)
2464                 goto input_err;
2465
2466         spin_lock_irq(&ndlp->lock);
2467         rport = lpfc_ndlp_get_nrport(ndlp);
2468         if (rport)
2469                 remoteport = rport->remoteport;
2470         spin_unlock_irq(&ndlp->lock);
2471         if (!remoteport)
2472                 goto input_err;
2473
2474         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2475                          "6033 Unreg nvme remoteport x%px, portname x%llx, "
2476                          "port_id x%06x, portstate x%x port type x%x "
2477                          "refcnt %d\n",
2478                          remoteport, remoteport->port_name,
2479                          remoteport->port_id, remoteport->port_state,
2480                          ndlp->nlp_type, kref_read(&ndlp->kref));
2481
2482         /* Sanity check ndlp type.  Only call for NVME ports. Don't
2483          * clear any rport state until the transport calls back.
2484          */
2485
2486         if (ndlp->nlp_type & NLP_NVME_TARGET) {
2487                 /* No concern about the role change on the nvme remoteport.
2488                  * The transport will update it.
2489                  */
2490                 spin_lock_irq(&vport->phba->hbalock);
2491                 ndlp->fc4_xpt_flags |= NLP_WAIT_FOR_UNREG;
2492                 spin_unlock_irq(&vport->phba->hbalock);
2493
2494                 /* Don't let the host nvme transport keep sending keep-alives
2495                  * on this remoteport. Vport is unloading, no recovery. The
2496                  * return values is ignored.  The upcall is a courtesy to the
2497                  * transport.
2498                  */
2499                 if (vport->load_flag & FC_UNLOADING)
2500                         (void)nvme_fc_set_remoteport_devloss(remoteport, 0);
2501
2502                 ret = nvme_fc_unregister_remoteport(remoteport);
2503
2504                 /* The driver no longer knows if the nrport memory is valid.
2505                  * because the controller teardown process has begun and
2506                  * is asynchronous.  Break the binding in the ndlp. Also
2507                  * remove the register ndlp reference to setup node release.
2508                  */
2509                 ndlp->nrport = NULL;
2510                 lpfc_nlp_put(ndlp);
2511                 if (ret != 0) {
2512                         lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2513                                          "6167 NVME unregister failed %d "
2514                                          "port_state x%x\n",
2515                                          ret, remoteport->port_state);
2516                 }
2517         }
2518         return;
2519
2520  input_err:
2521 #endif
2522         lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2523                          "6168 State error: lport x%px, rport x%px FCID x%06x\n",
2524                          vport->localport, ndlp->rport, ndlp->nlp_DID);
2525 }
2526
2527 /**
2528  * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2529  * @phba: pointer to lpfc hba data structure.
2530  * @axri: pointer to the fcp xri abort wcqe structure.
2531  * @lpfc_ncmd: The nvme job structure for the request being aborted.
2532  *
2533  * This routine is invoked by the worker thread to process a SLI4 fast-path
2534  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2535  * here.
2536  **/
2537 void
2538 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2539                            struct sli4_wcqe_xri_aborted *axri,
2540                            struct lpfc_io_buf *lpfc_ncmd)
2541 {
2542         uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2543         struct nvmefc_fcp_req *nvme_cmd = NULL;
2544         struct lpfc_nodelist *ndlp = lpfc_ncmd->ndlp;
2545
2546
2547         if (ndlp)
2548                 lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2549
2550         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2551                         "6311 nvme_cmd %p xri x%x tag x%x abort complete and "
2552                         "xri released\n",
2553                         lpfc_ncmd->nvmeCmd, xri,
2554                         lpfc_ncmd->cur_iocbq.iotag);
2555
2556         /* Aborted NVME commands are required to not complete
2557          * before the abort exchange command fully completes.
2558          * Once completed, it is available via the put list.
2559          */
2560         if (lpfc_ncmd->nvmeCmd) {
2561                 nvme_cmd = lpfc_ncmd->nvmeCmd;
2562                 nvme_cmd->done(nvme_cmd);
2563                 lpfc_ncmd->nvmeCmd = NULL;
2564         }
2565         lpfc_release_nvme_buf(phba, lpfc_ncmd);
2566 }
2567
2568 /**
2569  * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete
2570  * @phba: Pointer to HBA context object.
2571  *
2572  * This function flushes all wqes in the nvme rings and frees all resources
2573  * in the txcmplq. This function does not issue abort wqes for the IO
2574  * commands in txcmplq, they will just be returned with
2575  * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
2576  * slot has been permanently disabled.
2577  **/
2578 void
2579 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba)
2580 {
2581         struct lpfc_sli_ring  *pring;
2582         u32 i, wait_cnt = 0;
2583
2584         if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq)
2585                 return;
2586
2587         /* Cycle through all IO rings and make sure all outstanding
2588          * WQEs have been removed from the txcmplqs.
2589          */
2590         for (i = 0; i < phba->cfg_hdw_queue; i++) {
2591                 if (!phba->sli4_hba.hdwq[i].io_wq)
2592                         continue;
2593                 pring = phba->sli4_hba.hdwq[i].io_wq->pring;
2594
2595                 if (!pring)
2596                         continue;
2597
2598                 /* Retrieve everything on the txcmplq */
2599                 while (!list_empty(&pring->txcmplq)) {
2600                         msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1);
2601                         wait_cnt++;
2602
2603                         /* The sleep is 10mS.  Every ten seconds,
2604                          * dump a message.  Something is wrong.
2605                          */
2606                         if ((wait_cnt % 1000) == 0) {
2607                                 lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
2608                                                 "6178 NVME IO not empty, "
2609                                                 "cnt %d\n", wait_cnt);
2610                         }
2611                 }
2612         }
2613
2614         /* Make sure HBA is alive */
2615         lpfc_issue_hb_tmo(phba);
2616
2617 }
2618
2619 void
2620 lpfc_nvme_cancel_iocb(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
2621                       uint32_t stat, uint32_t param)
2622 {
2623 #if (IS_ENABLED(CONFIG_NVME_FC))
2624         struct lpfc_io_buf *lpfc_ncmd;
2625         struct nvmefc_fcp_req *nCmd;
2626         struct lpfc_wcqe_complete wcqe;
2627         struct lpfc_wcqe_complete *wcqep = &wcqe;
2628
2629         lpfc_ncmd = (struct lpfc_io_buf *)pwqeIn->context1;
2630         if (!lpfc_ncmd) {
2631                 lpfc_sli_release_iocbq(phba, pwqeIn);
2632                 return;
2633         }
2634         /* For abort iocb just return, IO iocb will do a done call */
2635         if (bf_get(wqe_cmnd, &pwqeIn->wqe.gen_req.wqe_com) ==
2636             CMD_ABORT_XRI_CX) {
2637                 lpfc_sli_release_iocbq(phba, pwqeIn);
2638                 return;
2639         }
2640
2641         spin_lock(&lpfc_ncmd->buf_lock);
2642         nCmd = lpfc_ncmd->nvmeCmd;
2643         if (!nCmd) {
2644                 spin_unlock(&lpfc_ncmd->buf_lock);
2645                 lpfc_release_nvme_buf(phba, lpfc_ncmd);
2646                 return;
2647         }
2648         spin_unlock(&lpfc_ncmd->buf_lock);
2649
2650         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_IOERR,
2651                         "6194 NVME Cancel xri %x\n",
2652                         lpfc_ncmd->cur_iocbq.sli4_xritag);
2653
2654         wcqep->word0 = 0;
2655         bf_set(lpfc_wcqe_c_status, wcqep, stat);
2656         wcqep->parameter = param;
2657         wcqep->word3 = 0; /* xb is 0 */
2658
2659         /* Call release with XB=1 to queue the IO into the abort list. */
2660         if (phba->sli.sli_flag & LPFC_SLI_ACTIVE)
2661                 bf_set(lpfc_wcqe_c_xb, wcqep, 1);
2662
2663         (pwqeIn->wqe_cmpl)(phba, pwqeIn, wcqep);
2664 #endif
2665 }